posix-acl: fixup extended ACL entries properlyv3.4.1rc1 v3.4.1

Typically when updating cached ACL from backend, we get both iatt and ACL xattrs (like lookup, readdirplus etc.) However in calls like setattr(), the mode would have updated but we receive only iatt and not the ACL xattrs. In such case we need to "spread" the effects of the changed mode properly into the cached ACL xattr ourselves. Change-Id: I23a7bc9c14722ff6848e175ed4bbe863a21ce2c9 BUG: 998967 Signed-off-by: Anand Avati <avati@redhat.com> Reviewed-on: http://review.gluster.org/6002 Tested-by: Gluster Build System <jenkins@build.gluster.com>
author: Anand Avati <avati@redhat.com> 2013-09-15 15:00:06 -0700
committer: Anand Avati <avati@redhat.com> 2013-09-25 14:09:15 -0700
commit: 56769c4db9e25f10c35b7f9b6ccf24dc27b35f07 (patch)
tree: 2746758353de959cfabb11221e2325c07bca4a0e /contrib/timer-wheel/timer-wheel.c
parent: 22e982e31bc232f3d9597a41f49e715428e010ce (diff)
0 files changed, 0 insertions, 0 deletions
diff --git a/xlators/Makefile.am b/xlators/Makefile.am
index 2abb52194..f60fa85ce 100644
--- a/xlators/Makefile.am
+++ b/xlators/Makefile.am
@@ -1,3 +1,4 @@
-SUBDIRS = cluster storage protocol performance debug features encryption mount
+SUBDIRS = cluster storage protocol performance debug features encryption mount nfs mgmt system \
+          playground
 
-CLEANFILES = 
+CLEANFILES =
diff --git a/xlators/bindings/python/src/Makefile.am b/xlators/bindings/python/src/Makefile.am
index c0b9141c6..90370d861 100644
--- a/xlators/bindings/python/src/Makefile.am
+++ b/xlators/bindings/python/src/Makefile.am
@@ -9,7 +9,7 @@ pythondir = $(xlatordir)/python
 
 python_so_SOURCES = python.c
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall \
+AM_CFLAGS = -fPIC $(GF_CPPFLAGS) -Wall \
 	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles \
 	$(PYTHON_CPPLAGS) -DGLUSTER_PYTHON_PATH=\"$(pythondir)\"
 
diff --git a/xlators/bindings/python/src/gluster.py b/xlators/bindings/python/src/gluster.py
index ee0eb1310..337c983ec 100644
--- a/xlators/bindings/python/src/gluster.py
+++ b/xlators/bindings/python/src/gluster.py
@@ -1,19 +1,12 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
+
+#  Copyright (c) 2007-2012 Red Hat, Inc. <http://www.redhat.com>
+#  This file is part of GlusterFS.
 #
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
+#  This file is licensed to you under your choice of the GNU Lesser
+#  General Public License, version 3 or any later version (LGPLv3 or
+#  later), or the GNU General Public License, version 2 (GPLv2), in all
+#  cases as published by the Free Software Foundation.
+
 from ctypes import *
 from glustertypes import *
 from glusterstack import *
diff --git a/xlators/bindings/python/src/glusterstack.py b/xlators/bindings/python/src/glusterstack.py
index ba24c8165..0c071ae98 100644
--- a/xlators/bindings/python/src/glusterstack.py
+++ b/xlators/bindings/python/src/glusterstack.py
@@ -1,19 +1,12 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
+
+#  Copyright (c) 2007-2012 Red Hat, Inc. <http://www.redhat.com>
+#  This file is part of GlusterFS.
 #
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
+#  This file is licensed to you under your choice of the GNU Lesser
+#  General Public License, version 3 or any later version (LGPLv3 or
+#  later), or the GNU General Public License, version 2 (GPLv2), in all
+#  cases as published by the Free Software Foundation.
+
 from ctypes import *
 from glustertypes import *
 
diff --git a/xlators/bindings/python/src/glustertypes.py b/xlators/bindings/python/src/glustertypes.py
index e9069d07c..98437d22e 100644
--- a/xlators/bindings/python/src/glustertypes.py
+++ b/xlators/bindings/python/src/glustertypes.py
@@ -1,19 +1,12 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
+
+#  Copyright (c) 2007-2012 Red Hat, Inc. <http://www.redhat.com>
+#  This file is part of GlusterFS.
 #
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
+#  This file is licensed to you under your choice of the GNU Lesser
+#  General Public License, version 3 or any later version (LGPLv3 or
+#  later), or the GNU General Public License, version 2 (GPLv2), in all
+#  cases as published by the Free Software Foundation.
+
 from ctypes import *
 import collections
 
diff --git a/xlators/bindings/python/src/python.c b/xlators/bindings/python/src/python.c
index 7f32d7f29..9b96790de 100644
--- a/xlators/bindings/python/src/python.c
+++ b/xlators/bindings/python/src/python.c
@@ -1,22 +1,12 @@
 /*
-   Copyright (c) 2007-2009 Chris AtLee <chris@atlee.ca>
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
-
 #include <Python.h>
 
 #ifndef _CONFIG_H
@@ -45,7 +35,7 @@ python_writev (call_frame_t *frame,
               xlator_t *this,
               fd_t *fd,
               struct iovec *vector,
-              int32_t count, 
+              int32_t count,
               off_t offset)
 {
   python_private_t *priv = (python_private_t *)this->private;
@@ -77,9 +67,6 @@ struct xlator_fops fops = {
     .writev       = python_writev
 };
 
-struct xlator_mops mops = {
-};
-
 static PyObject *
 AnonModule_FromFile (const char* fname)
 {
@@ -151,7 +138,7 @@ init (xlator_t *this)
   Py_InitializeEx(0);
 
   if (!this->children) {
-    gf_log ("python", GF_LOG_ERROR, 
+    gf_log ("python", GF_LOG_ERROR,
             "FATAL: python should have exactly one child");
     return -1;
   }
@@ -169,7 +156,7 @@ init (xlator_t *this)
   }
 
   priv->pInterp = Py_NewInterpreter();
-    
+
   // Adjust python's path
   PyObject *syspath = PySys_GetObject("path");
   PyObject *path = PyString_FromString(GLUSTER_PYTHON_PATH);
@@ -191,7 +178,7 @@ init (xlator_t *this)
   priv->pVectorType = PyObject_GetAttrString(priv->pGlusterModule, "iovec");
 
   gf_log("python", GF_LOG_DEBUG, "Loading script...%s", priv->scriptname);
-  
+
   priv->pScriptModule = AnonModule_FromFile(priv->scriptname);
   if (!priv->pScriptModule || PyErr_Occurred())
   {
@@ -220,7 +207,7 @@ init (xlator_t *this)
   return 0;
 }
 
-void 
+void
 fini (xlator_t *this)
 {
   python_private_t *priv = (python_private_t*)(this->private);
diff --git a/xlators/bindings/python/src/testxlator.py b/xlators/bindings/python/src/testxlator.py
index 507455c85..59a991dca 100644
--- a/xlators/bindings/python/src/testxlator.py
+++ b/xlators/bindings/python/src/testxlator.py
@@ -1,19 +1,12 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
+"""
+  Copyright (c) 2007-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+"""
 
 """
 This is a test translator written in python.
diff --git a/xlators/cluster/Makefile.am b/xlators/cluster/Makefile.am
index a6ddb3564..0990822a7 100644
--- a/xlators/cluster/Makefile.am
+++ b/xlators/cluster/Makefile.am
@@ -1,3 +1,3 @@
-SUBDIRS = unify stripe afr dht ha map
+SUBDIRS = stripe afr dht
 
 CLEANFILES = 
diff --git a/xlators/cluster/afr/src/Makefile.am b/xlators/cluster/afr/src/Makefile.am
index 1bde9e5ba..35d18a6c0 100644
--- a/xlators/cluster/afr/src/Makefile.am
+++ b/xlators/cluster/afr/src/Makefile.am
@@ -1,20 +1,37 @@
-xlator_LTLIBRARIES = afr.la
+xlator_LTLIBRARIES = afr.la pump.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
 
-afr_la_LDFLAGS = -module -avoidversion 
+afr_common_source = afr-dir-read.c afr-dir-write.c afr-inode-read.c \
+	afr-inode-write.c afr-open.c afr-transaction.c afr-self-heal-data.c \
+	afr-self-heal-common.c afr-self-heal-metadata.c afr-self-heal-entry.c \
+	afr-self-heal-algorithm.c afr-lk-common.c afr-self-heald.c \
+	$(top_builddir)/xlators/lib/src/libxlator.c
 
-afr_la_SOURCES = afr.c afr-dir-read.c afr-dir-write.c afr-inode-read.c afr-inode-write.c afr-transaction.c afr-self-heal-data.c afr-self-heal-common.c afr-self-heal-metadata.c afr-self-heal-entry.c
+afr_la_LDFLAGS = -module -avoid-version
+afr_la_SOURCES = $(afr_common_source) afr.c
 afr_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-noinst_HEADERS = afr.h afr-transaction.h afr-inode-write.h afr-inode-read.h afr-dir-read.h afr-dir-write.h afr-self-heal.h afr-self-heal-common.h
+pump_la_LDFLAGS = -module -avoid-version
+pump_la_SOURCES =  $(afr_common_source) pump.c
+pump_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	    -I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+noinst_HEADERS = afr.h afr-transaction.h afr-inode-write.h afr-inode-read.h \
+	afr-dir-read.h afr-dir-write.h afr-self-heal.h afr-self-heal-common.h \
+	afr-self-heal-algorithm.h pump.h afr-mem-types.h afr-common.c \
+	afr-self-heald.h $(top_builddir)/xlators/lib/src/libxlator.h \
+	$(top_builddir)/glusterfsd/src/glusterfsd.h
 
-CLEANFILES = 
+AM_CPPFLAGS = $(GF_CPPFLAGS) \
+	-I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/xlators/lib/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
 
 uninstall-local:
 	rm -f $(DESTDIR)$(xlatordir)/replicate.so
+	rm -f $(DESTDIR)$(xlatordir)/pump.so
 
 install-data-hook:
-	ln -sf afr.so $(DESTDIR)$(xlatordir)/replicate.so
-\ No newline at end of file
+	ln -sf afr.so $(DESTDIR)$(xlatordir)/replicate.so
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
new file mode 100644
index 000000000..af01f2ef2
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -0,0 +1,4591 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <libgen.h>
+#include <unistd.h>
+#include <fnmatch.h>
+#include <sys/time.h>
+#include <stdlib.h>
+#include <signal.h>
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "glusterfs.h"
+#include "afr.h"
+#include "dict.h"
+#include "xlator.h"
+#include "hashfn.h"
+#include "logging.h"
+#include "stack.h"
+#include "list.h"
+#include "call-stub.h"
+#include "defaults.h"
+#include "common-utils.h"
+#include "compat-errno.h"
+#include "compat.h"
+#include "byte-order.h"
+#include "statedump.h"
+#include "inode.h"
+
+#include "fd.h"
+
+#include "afr-inode-read.h"
+#include "afr-inode-write.h"
+#include "afr-dir-read.h"
+#include "afr-dir-write.h"
+#include "afr-transaction.h"
+#include "afr-self-heal.h"
+#include "afr-self-heal-common.h"
+#include "afr-self-heald.h"
+#include "pump.h"
+
+#define AFR_ICTX_OPENDIR_DONE_MASK     0x0000000100000000ULL
+#define AFR_ICTX_READ_CHILD_MASK       0x00000000FFFFFFFFULL
+#define AFR_STATISTICS_HISTORY_SIZE    50
+int
+afr_lookup_done_success_action (call_frame_t *frame, xlator_t *this,
+                                gf_boolean_t fail_conflict);
+void
+afr_children_copy (int32_t *dst, int32_t *src, unsigned int child_count)
+{
+        int     i = 0;
+
+        for (i = 0; i < child_count; i++)
+                dst[i] = src[i];
+}
+
+void
+afr_xattr_req_prepare (xlator_t *this, dict_t *xattr_req, const char *path)
+{
+        int             i           = 0;
+        afr_private_t   *priv       = NULL;
+        int             ret         = 0;
+
+        priv   = this->private;
+
+        for (i = 0; i < priv->child_count; i++) {
+                ret = dict_set_uint64 (xattr_req, priv->pending_key[i],
+                                       3 * sizeof(int32_t));
+                if (ret < 0)
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: Unable to set dict value for %s",
+                                path, priv->pending_key[i]);
+                /* 3 = data+metadata+entry */
+        }
+        ret = dict_set_int32 (xattr_req, GF_GFIDLESS_LOOKUP, 1);
+        if (ret) {
+                gf_log (this->name, GF_LOG_DEBUG, "%s: failed to set gfidless "
+                        "lookup", path);
+        }
+}
+
+int
+afr_lookup_xattr_req_prepare (afr_local_t *local, xlator_t *this,
+                              dict_t *xattr_req, loc_t *loc, void **gfid_req)
+{
+        int     ret = -ENOMEM;
+
+        GF_ASSERT (gfid_req);
+
+        *gfid_req = NULL;
+        local->xattr_req = dict_new ();
+        if (!local->xattr_req)
+                goto out;
+        if (xattr_req)
+                dict_copy (xattr_req, local->xattr_req);
+
+        afr_xattr_req_prepare (this, local->xattr_req, loc->path);
+        ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_INODELK_COUNT, 0);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: Unable to set dict value for %s",
+                        loc->path, GLUSTERFS_INODELK_COUNT);
+        }
+        ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_ENTRYLK_COUNT, 0);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: Unable to set dict value for %s",
+                        loc->path, GLUSTERFS_ENTRYLK_COUNT);
+        }
+
+        ret = dict_set_uint32 (local->xattr_req, GLUSTERFS_PARENT_ENTRYLK, 0);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: Unable to set dict value for %s",
+                        loc->path, GLUSTERFS_PARENT_ENTRYLK);
+        }
+
+        ret = dict_get_ptr (local->xattr_req, "gfid-req", gfid_req);
+        if (ret) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "%s: failed to get the gfid from dict", loc->path);
+                *gfid_req = NULL;
+        } else {
+                if (loc->parent != NULL)
+                        dict_del (local->xattr_req, "gfid-req");
+        }
+        ret = 0;
+out:
+        return ret;
+}
+
+void
+afr_lookup_save_gfid (uuid_t dst, void* new, const loc_t *loc)
+{
+        inode_t  *inode = NULL;
+
+        inode = loc->inode;
+        if (inode && !uuid_is_null (inode->gfid))
+                uuid_copy (dst, inode->gfid);
+        else if (!uuid_is_null (loc->gfid))
+                uuid_copy (dst, loc->gfid);
+        else if (new && !uuid_is_null (new))
+                uuid_copy (dst, new);
+}
+
+int
+afr_errno_count (int32_t *children, int *child_errno,
+                 unsigned int child_count, int32_t op_errno)
+{
+        int i = 0;
+        int errno_count = 0;
+        int child = 0;
+
+        for (i = 0; i < child_count; i++) {
+                if (children) {
+                        child = children[i];
+                        if (child == -1)
+                                break;
+                } else {
+                        child = i;
+                }
+                if (child_errno[child] == op_errno)
+                        errno_count++;
+        }
+        return errno_count;
+}
+
+int32_t
+afr_set_dict_gfid (dict_t *dict, uuid_t gfid)
+{
+        int ret       = 0;
+        uuid_t *pgfid = NULL;
+
+        GF_ASSERT (gfid);
+
+        pgfid = GF_CALLOC (1, sizeof (uuid_t), gf_common_mt_char);
+        if (!pgfid) {
+                ret = -1;
+                goto out;
+        }
+
+        uuid_copy (*pgfid, gfid);
+
+        ret = dict_set_dynptr (dict, "gfid-req", pgfid, sizeof (uuid_t));
+        if (ret)
+                gf_log (THIS->name, GF_LOG_ERROR, "gfid set failed");
+
+out:
+        if (ret && pgfid)
+                GF_FREE (pgfid);
+
+        return ret;
+}
+
+void
+afr_inode_ctx_destroy (afr_inode_ctx_t *ctx)
+{
+        if (!ctx)
+                return;
+        GF_FREE (ctx->fresh_children);
+        GF_FREE (ctx);
+}
+
+afr_inode_ctx_t*
+__afr_inode_ctx_get (inode_t *inode, xlator_t *this)
+{
+        int             ret      = 0;
+        uint64_t        ctx_addr = 0;
+        afr_inode_ctx_t *ctx     = NULL;
+        afr_private_t   *priv    = NULL;
+
+        priv = this->private;
+        ret = __inode_ctx_get (inode, this, &ctx_addr);
+        if (ret < 0)
+                ctx_addr = 0;
+        if (ctx_addr != 0) {
+                ctx = (afr_inode_ctx_t*) (long) ctx_addr;
+                goto out;
+        }
+        ctx = GF_CALLOC (1, sizeof (*ctx),
+                         gf_afr_mt_inode_ctx_t);
+        if (!ctx)
+                goto fail;
+        ctx->fresh_children = GF_CALLOC (priv->child_count,
+                                         sizeof (*ctx->fresh_children),
+                                         gf_afr_mt_int32_t);
+        if (!ctx->fresh_children)
+                goto fail;
+        ret = __inode_ctx_put (inode, this, (uint64_t)ctx);
+        if (ret) {
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "failed to "
+                                  "set the inode ctx (%s)",
+                                  uuid_utoa (inode->gfid));
+                goto fail;
+        }
+
+out:
+        return ctx;
+
+fail:
+        afr_inode_ctx_destroy (ctx);
+        return NULL;
+}
+
+afr_inode_ctx_t*
+afr_inode_ctx_get (inode_t *inode, xlator_t *this)
+{
+        afr_inode_ctx_t *ctx = NULL;
+
+        LOCK (&inode->lock);
+        {
+                ctx = __afr_inode_ctx_get (inode, this);
+        }
+        UNLOCK (&inode->lock);
+        return ctx;
+}
+
+void
+afr_inode_get_ctx_params (xlator_t *this, inode_t *inode,
+                          afr_inode_params_t *params)
+{
+        GF_ASSERT (inode);
+        GF_ASSERT (params);
+
+        afr_inode_ctx_t *ctx = NULL;
+        afr_private_t   *priv = NULL;
+        int             i = 0;
+        int32_t         read_child = -1;
+        int32_t         *fresh_children = NULL;
+
+        priv = this->private;
+        LOCK (&inode->lock);
+        {
+                ctx = __afr_inode_ctx_get (inode, this);
+                if (!ctx)
+                        goto unlock;
+                switch (params->op) {
+                case AFR_INODE_GET_READ_CTX:
+                        fresh_children = params->u.read_ctx.children;
+                        read_child = (int32_t)(ctx->masks &
+                                               AFR_ICTX_READ_CHILD_MASK);
+                        params->u.read_ctx.read_child = read_child;
+                        if (!fresh_children)
+                                goto unlock;
+                        for (i = 0; i < priv->child_count; i++)
+                                fresh_children[i] = ctx->fresh_children[i];
+                        break;
+                case AFR_INODE_GET_OPENDIR_DONE:
+                        params->u.value = _gf_false;
+                        if (ctx->masks & AFR_ICTX_OPENDIR_DONE_MASK)
+                                params->u.value = _gf_true;
+                        break;
+                default:
+                        GF_ASSERT (0);
+                        break;
+                }
+        }
+unlock:
+        UNLOCK (&inode->lock);
+}
+
+gf_boolean_t
+afr_is_split_brain (xlator_t *this, inode_t *inode)
+{
+        afr_inode_ctx_t *ctx = NULL;
+        gf_boolean_t    spb  = _gf_false;
+
+        ctx = afr_inode_ctx_get (inode, this);
+        if (!ctx)
+                goto out;
+        if ((ctx->mdata_spb == SPB) || (ctx->data_spb == SPB))
+                spb = _gf_true;
+out:
+        return spb;
+}
+
+gf_boolean_t
+afr_is_opendir_done (xlator_t *this, inode_t *inode)
+{
+        afr_inode_params_t params = {0};
+
+        params.op = AFR_INODE_GET_OPENDIR_DONE;
+        afr_inode_get_ctx_params (this, inode, &params);
+        return params.u.value;
+}
+
+int32_t
+afr_inode_get_read_ctx (xlator_t *this, inode_t *inode, int32_t *fresh_children)
+{
+        afr_inode_params_t      params = {0};
+
+        params.op = AFR_INODE_GET_READ_CTX;
+        params.u.read_ctx.children = fresh_children;
+        afr_inode_get_ctx_params (this, inode, &params);
+        return params.u.read_ctx.read_child;
+}
+
+void
+afr_inode_ctx_set_read_child (afr_inode_ctx_t *ctx, int32_t read_child)
+{
+        uint64_t        remaining_mask = 0;
+        uint64_t        mask         = 0;
+
+        remaining_mask = (~AFR_ICTX_READ_CHILD_MASK & ctx->masks);
+        mask = (AFR_ICTX_READ_CHILD_MASK & read_child);
+        ctx->masks = remaining_mask | mask;
+}
+
+void
+afr_inode_ctx_set_read_ctx (afr_inode_ctx_t *ctx, int32_t read_child,
+                            int32_t *fresh_children, int32_t child_count)
+{
+        int             i            = 0;
+
+        afr_inode_ctx_set_read_child (ctx, read_child);
+        for (i = 0; i < child_count; i++) {
+                if (fresh_children)
+                        ctx->fresh_children[i] = fresh_children[i];
+                else
+                        ctx->fresh_children[i] = -1;
+        }
+}
+
+void
+afr_inode_ctx_rm_stale_children (afr_inode_ctx_t *ctx, int32_t *stale_children,
+                                 int32_t child_count)
+{
+        int             i            = 0;
+        int32_t         read_child   = -1;
+
+        GF_ASSERT (stale_children);
+        for (i = 0; i < child_count; i++) {
+                if (stale_children[i] == -1)
+                        break;
+                afr_children_rm_child (ctx->fresh_children,
+                                       stale_children[i], child_count);
+        }
+        read_child = (int32_t)(ctx->masks & AFR_ICTX_READ_CHILD_MASK);
+        if (!afr_is_child_present (ctx->fresh_children, child_count,
+                                   read_child))
+                afr_inode_ctx_set_read_child (ctx, ctx->fresh_children[0]);
+}
+
+void
+afr_inode_ctx_set_opendir_done (afr_inode_ctx_t *ctx)
+{
+        uint64_t        remaining_mask = 0;
+        uint64_t        mask = 0;
+
+        remaining_mask = (~AFR_ICTX_OPENDIR_DONE_MASK & ctx->masks);
+        mask = (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_OPENDIR_DONE_MASK);
+        ctx->masks = remaining_mask | mask;
+}
+
+void
+afr_inode_set_ctx_params (xlator_t *this, inode_t *inode,
+                          afr_inode_params_t *params)
+{
+        GF_ASSERT (inode);
+        GF_ASSERT (params);
+
+        afr_inode_ctx_t *ctx            = NULL;
+        afr_private_t   *priv           = NULL;
+        int32_t         read_child      = -1;
+        int32_t         *fresh_children = NULL;
+        int32_t         *stale_children = NULL;
+
+        priv = this->private;
+        LOCK (&inode->lock);
+        {
+                ctx = __afr_inode_ctx_get (inode, this);
+                if (!ctx)
+                        goto unlock;
+                switch (params->op) {
+                case AFR_INODE_SET_READ_CTX:
+                        read_child = params->u.read_ctx.read_child;
+                        fresh_children = params->u.read_ctx.children;
+                        afr_inode_ctx_set_read_ctx (ctx, read_child,
+                                                    fresh_children,
+                                                    priv->child_count);
+                        break;
+                case AFR_INODE_RM_STALE_CHILDREN:
+                        stale_children = params->u.read_ctx.children;
+                        afr_inode_ctx_rm_stale_children (ctx,
+                                                         stale_children,
+                                                         priv->child_count);
+                        break;
+                case AFR_INODE_SET_OPENDIR_DONE:
+                        afr_inode_ctx_set_opendir_done (ctx);
+                        break;
+                default:
+                        GF_ASSERT (0);
+                        break;
+                }
+        }
+unlock:
+        UNLOCK (&inode->lock);
+}
+
+void
+afr_set_split_brain (xlator_t *this, inode_t *inode, afr_spb_state_t mdata_spb,
+                     afr_spb_state_t data_spb)
+{
+        afr_inode_ctx_t *ctx = NULL;
+
+        ctx = afr_inode_ctx_get (inode, this);
+        if (mdata_spb != DONT_KNOW)
+                ctx->mdata_spb = mdata_spb;
+        if (data_spb != DONT_KNOW)
+                ctx->data_spb = data_spb;
+}
+
+void
+afr_set_opendir_done (xlator_t *this, inode_t *inode)
+{
+        afr_inode_params_t params = {0};
+
+        params.op = AFR_INODE_SET_OPENDIR_DONE;
+        afr_inode_set_ctx_params (this, inode, &params);
+}
+
+void
+afr_inode_set_read_ctx (xlator_t *this, inode_t *inode, int32_t read_child,
+                        int32_t *fresh_children)
+{
+        afr_inode_params_t params = {0};
+        afr_private_t      *priv  = NULL;
+
+        priv = this->private;
+        GF_ASSERT (read_child >= 0);
+        GF_ASSERT (fresh_children);
+        GF_ASSERT (afr_is_child_present (fresh_children, priv->child_count,
+                                         read_child));
+
+        params.op = AFR_INODE_SET_READ_CTX;
+        params.u.read_ctx.read_child     = read_child;
+        params.u.read_ctx.children = fresh_children;
+        afr_inode_set_ctx_params (this, inode, &params);
+}
+
+void
+afr_inode_rm_stale_children (xlator_t *this, inode_t *inode,
+                             int32_t *stale_children)
+{
+        afr_inode_params_t params = {0};
+
+        GF_ASSERT (stale_children);
+
+        params.op = AFR_INODE_RM_STALE_CHILDREN;
+        params.u.read_ctx.children = stale_children;
+        afr_inode_set_ctx_params (this, inode, &params);
+}
+
+gf_boolean_t
+afr_is_source_child (int32_t *sources, int32_t child_count, int32_t child)
+{
+        gf_boolean_t             source_xattrs = _gf_false;
+
+        GF_ASSERT (child < child_count);
+
+        if ((child >= 0) && (child < child_count) &&
+             sources[child]) {
+                source_xattrs = _gf_true;
+        }
+        return source_xattrs;
+}
+
+gf_boolean_t
+afr_is_child_present (int32_t *success_children, int32_t child_count,
+                      int32_t child)
+{
+        gf_boolean_t             success_child = _gf_false;
+        int                      i = 0;
+
+        GF_ASSERT (child < child_count);
+
+        for (i = 0; i < child_count; i++) {
+                if (success_children[i] == -1)
+                        break;
+                if (child == success_children[i]) {
+                        success_child = _gf_true;
+                        break;
+                }
+        }
+        return success_child;
+}
+
+gf_boolean_t
+afr_is_read_child (int32_t *success_children, int32_t *sources,
+                   int32_t child_count, int32_t child)
+{
+        gf_boolean_t             success_child = _gf_false;
+        gf_boolean_t             source        = _gf_false;
+
+        if (child < 0) {
+                return _gf_false;
+        }
+
+        GF_ASSERT (success_children);
+        GF_ASSERT (child_count > 0);
+
+        success_child = afr_is_child_present (success_children, child_count,
+                                              child);
+        if (!success_child)
+                goto out;
+        if (NULL == sources) {
+                source = _gf_true;
+                goto out;
+        }
+        source = afr_is_source_child (sources, child_count, child);
+out:
+        return (success_child && source);
+}
+
+int32_t
+afr_hash_child (int32_t *success_children, int32_t child_count,
+                unsigned int hmode, uuid_t gfid)
+{
+        uuid_t  gfid_copy = {0,};
+        pid_t pid;
+
+        if (!hmode) {
+                return -1;
+        }
+
+        if (gfid) {
+               uuid_copy(gfid_copy,gfid);
+        }
+        if (hmode > 1) {
+                /*
+                 * Why getpid?  Because it's one of the cheapest calls
+                 * available - faster than gethostname etc. - and returns a
+                 * constant-length value that's sure to be shorter than a UUID.
+                 * It's still very unlikely to be the same across clients, so
+                 * it still provides good mixing.  We're not trying for
+                 * perfection here.  All we need is a low probability that
+                 * multiple clients won't converge on the same subvolume.
+                 */
+                pid = getpid();
+                memcpy (gfid_copy, &pid, sizeof(pid));
+        }
+
+        return SuperFastHash((char *)gfid_copy,
+                             sizeof(gfid_copy)) % child_count;
+}
+
+/* If sources is NULL the xattrs are assumed to be of source for all
+ * success_children.
+ */
+int
+afr_select_read_child_from_policy (int32_t *success_children,
+                                   int32_t child_count, int32_t prev_read_child,
+                                   int32_t config_read_child, int32_t *sources,
+                                   unsigned int hmode, uuid_t gfid)
+{
+        int32_t                  read_child   = -1;
+        int                      i            = 0;
+
+        GF_ASSERT (success_children);
+
+        read_child = config_read_child;
+        if (afr_is_read_child (success_children, sources, child_count,
+                               read_child))
+                goto out;
+
+        read_child = prev_read_child;
+        if (afr_is_read_child (success_children, sources, child_count,
+                               read_child))
+                goto out;
+
+        read_child = afr_hash_child (success_children, child_count,
+                                     hmode, gfid);
+        if (afr_is_read_child (success_children, sources, child_count,
+                               read_child)) {
+                goto out;
+        }
+
+        for (i = 0; i < child_count; i++) {
+                read_child = success_children[i];
+                if (read_child < 0)
+                        break;
+                if (afr_is_read_child (success_children, sources, child_count,
+                                       read_child))
+                        goto out;
+        }
+        read_child = -1;
+
+out:
+        return read_child;
+}
+
+/* This function should be used when all the success_children are sources
+ */
+void
+afr_set_read_ctx_from_policy (xlator_t *this, inode_t *inode,
+                              int32_t *fresh_children, int32_t prev_read_child,
+                              int32_t config_read_child, uuid_t gfid)
+{
+        int                      read_child = -1;
+        afr_private_t            *priv = NULL;
+
+        priv = this->private;
+        read_child = afr_select_read_child_from_policy (fresh_children,
+                                                        priv->child_count,
+                                                        prev_read_child,
+                                                        config_read_child,
+                                                        NULL,
+                                                        priv->hash_mode, gfid);
+        if (read_child >= 0)
+                afr_inode_set_read_ctx (this, inode, read_child,
+                                        fresh_children);
+}
+
+/* afr_next_call_child ()
+ * This is a common function used by all the read-type fops
+ * This function should not be called with the inode's read_children array.
+ * The fop's handler should make a copy of the inode's read_children,
+ * preferred read_child into the local vars, because while this function is
+ * in execution there is a chance for inode's read_ctx to change.
+ */
+int32_t
+afr_next_call_child (int32_t *fresh_children, unsigned char *child_up,
+                     size_t child_count, int32_t *last_index,
+                     int32_t read_child)
+{
+        int             next_index      = 0;
+        int32_t         next_call_child = -1;
+
+        GF_ASSERT (last_index);
+
+        next_index = *last_index;
+retry:
+        next_index++;
+        if ((next_index >= child_count) ||
+           (fresh_children[next_index] == -1))
+                goto out;
+        if ((fresh_children[next_index] == read_child) ||
+           (!child_up[fresh_children[next_index]]))
+                goto retry;
+        *last_index = next_index;
+        next_call_child = fresh_children[next_index];
+out:
+        return next_call_child;
+}
+
+ /* This function should not be called with the inode's read_children array.
+ * The fop's handler should make a copy of the inode's read_children,
+ * preferred read_child into the local vars, because while this function is
+ * in execution there is a chance for inode's read_ctx to change.
+ */
+int32_t
+afr_get_call_child (xlator_t *this, unsigned char *child_up, int32_t read_child,
+                    int32_t *fresh_children,
+                    int32_t *call_child, int32_t *last_index)
+{
+        int             ret   = 0;
+        afr_private_t   *priv = NULL;
+        int             i     = 0;
+
+        GF_ASSERT (child_up);
+        GF_ASSERT (call_child);
+        GF_ASSERT (last_index);
+        GF_ASSERT (fresh_children);
+
+        if (read_child < 0) {
+                ret = -EIO;
+                goto out;
+        }
+        priv = this->private;
+        *call_child = -1;
+        *last_index = -1;
+
+        if (child_up[read_child]) {
+                *call_child = read_child;
+        } else {
+                for (i = 0; i < priv->child_count; i++) {
+                        if (fresh_children[i] == -1)
+                                break;
+                        if (child_up[fresh_children[i]]) {
+                                *call_child = fresh_children[i];
+                                ret = 0;
+                                break;
+                        }
+                }
+
+                if (*call_child == -1) {
+                        ret = -ENOTCONN;
+                        goto out;
+                }
+
+                *last_index = i;
+        }
+out:
+        gf_log (this->name, GF_LOG_DEBUG, "Returning %d, call_child: %d, "
+                "last_index: %d", ret, *call_child, *last_index);
+        return ret;
+}
+
+void
+afr_reset_xattr (dict_t **xattr, unsigned int child_count)
+{
+        unsigned int i = 0;
+
+        if (!xattr)
+                goto out;
+        for (i = 0; i < child_count; i++) {
+                if (xattr[i]) {
+                        dict_unref (xattr[i]);
+                        xattr[i] = NULL;
+                }
+        }
+out:
+        return;
+}
+
+void
+afr_xattr_array_destroy (dict_t **xattr, unsigned int child_count)
+{
+        afr_reset_xattr (xattr, child_count);
+        GF_FREE (xattr);
+}
+
+void
+afr_local_sh_cleanup (afr_local_t *local, xlator_t *this)
+{
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+
+        sh = &local->self_heal;
+        priv = this->private;
+
+        if (sh->data_sh_info && strcmp (sh->data_sh_info, ""))
+                GF_FREE (sh->data_sh_info);
+
+        if (sh->metadata_sh_info && strcmp (sh->metadata_sh_info, ""))
+                GF_FREE (sh->metadata_sh_info);
+
+        GF_FREE (sh->buf);
+
+        GF_FREE (sh->parentbufs);
+
+        if (sh->inode)
+                inode_unref (sh->inode);
+
+        afr_xattr_array_destroy (sh->xattr, priv->child_count);
+
+        GF_FREE (sh->child_errno);
+
+        afr_matrix_cleanup (sh->pending_matrix, priv->child_count);
+        afr_matrix_cleanup (sh->delta_matrix, priv->child_count);
+
+        GF_FREE (sh->sources);
+
+        GF_FREE (sh->success);
+
+        GF_FREE (sh->locked_nodes);
+
+        if (sh->healing_fd) {
+                fd_unref (sh->healing_fd);
+                sh->healing_fd = NULL;
+        }
+
+        GF_FREE ((char *)sh->linkname);
+
+        GF_FREE (sh->success_children);
+
+        GF_FREE (sh->fresh_children);
+
+        GF_FREE (sh->fresh_parent_dirs);
+
+        loc_wipe (&sh->parent_loc);
+        loc_wipe (&sh->lookup_loc);
+
+        GF_FREE (sh->checksum);
+
+        GF_FREE (sh->write_needed);
+        if (sh->healing_fd)
+                fd_unref (sh->healing_fd);
+}
+
+
+void
+afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this)
+{
+        afr_private_t *priv    = NULL;
+        int           i        = 0;
+
+        priv = this->private;
+
+        afr_matrix_cleanup (local->pending, priv->child_count);
+        afr_matrix_cleanup (local->transaction.txn_changelog,
+                            priv->child_count);
+
+        GF_FREE (local->internal_lock.locked_nodes);
+
+        for (i = 0; local->internal_lock.inodelk[i].domain; i++) {
+                GF_FREE (local->internal_lock.inodelk[i].locked_nodes);
+        }
+
+        GF_FREE (local->internal_lock.lower_locked_nodes);
+
+        afr_entry_lockee_cleanup (&local->internal_lock);
+
+        GF_FREE (local->transaction.pre_op);
+        GF_FREE (local->transaction.eager_lock);
+
+        GF_FREE (local->transaction.basename);
+        GF_FREE (local->transaction.new_basename);
+
+        loc_wipe (&local->transaction.parent_loc);
+        loc_wipe (&local->transaction.new_parent_loc);
+
+        GF_FREE (local->transaction.postop_piggybacked);
+}
+
+
+void
+afr_local_cleanup (afr_local_t *local, xlator_t *this)
+{
+        afr_private_t * priv = NULL;
+
+        if (!local)
+                return;
+
+        afr_local_sh_cleanup (local, this);
+
+        afr_local_transaction_cleanup (local, this);
+
+        priv = this->private;
+
+        loc_wipe (&local->loc);
+        loc_wipe (&local->newloc);
+
+        if (local->fd)
+                fd_unref (local->fd);
+
+        if (local->xattr_req)
+                dict_unref (local->xattr_req);
+
+        if (local->dict)
+                dict_unref (local->dict);
+
+	GF_FREE(local->replies);
+
+        GF_FREE (local->child_up);
+
+        GF_FREE (local->child_errno);
+
+        GF_FREE (local->fresh_children);
+
+        { /* lookup */
+                if (local->cont.lookup.xattrs) {
+                        afr_reset_xattr (local->cont.lookup.xattrs,
+                                         priv->child_count);
+                        GF_FREE (local->cont.lookup.xattrs);
+                        local->cont.lookup.xattrs = NULL;
+                }
+
+                if (local->cont.lookup.xattr) {
+                        dict_unref (local->cont.lookup.xattr);
+                }
+
+                if (local->cont.lookup.inode) {
+                        inode_unref (local->cont.lookup.inode);
+                }
+
+                GF_FREE (local->cont.lookup.postparents);
+
+                GF_FREE (local->cont.lookup.bufs);
+
+                GF_FREE (local->cont.lookup.success_children);
+
+                GF_FREE (local->cont.lookup.sources);
+                afr_matrix_cleanup (local->cont.lookup.pending_matrix,
+                                    priv->child_count);
+        }
+
+        { /* getxattr */
+                GF_FREE (local->cont.getxattr.name);
+        }
+
+        { /* lk */
+                GF_FREE (local->cont.lk.locked_nodes);
+        }
+
+        { /* create */
+                if (local->cont.create.fd)
+                        fd_unref (local->cont.create.fd);
+                if (local->cont.create.params)
+                        dict_unref (local->cont.create.params);
+        }
+
+        { /* mknod */
+                if (local->cont.mknod.params)
+                        dict_unref (local->cont.mknod.params);
+        }
+
+        { /* mkdir */
+                if (local->cont.mkdir.params)
+                        dict_unref (local->cont.mkdir.params);
+        }
+
+        { /* symlink */
+                if (local->cont.symlink.params)
+                        dict_unref (local->cont.symlink.params);
+        }
+
+        { /* writev */
+                GF_FREE (local->cont.writev.vector);
+        }
+
+        { /* setxattr */
+                if (local->cont.setxattr.dict)
+                        dict_unref (local->cont.setxattr.dict);
+        }
+
+        { /* fsetxattr */
+                if (local->cont.fsetxattr.dict)
+                        dict_unref (local->cont.fsetxattr.dict);
+        }
+
+        { /* removexattr */
+                GF_FREE (local->cont.removexattr.name);
+        }
+        { /* xattrop */
+                if (local->cont.xattrop.xattr)
+                        dict_unref (local->cont.xattrop.xattr);
+        }
+        { /* fxattrop */
+                if (local->cont.fxattrop.xattr)
+                        dict_unref (local->cont.fxattrop.xattr);
+        }
+        { /* symlink */
+                GF_FREE (local->cont.symlink.linkpath);
+        }
+
+        { /* opendir */
+                GF_FREE (local->cont.opendir.checksum);
+        }
+
+        { /* readdirp */
+                if (local->cont.readdir.dict)
+                        dict_unref (local->cont.readdir.dict);
+        }
+
+        if (local->xdata_req)
+                dict_unref (local->xdata_req);
+
+        if (local->xdata_rsp)
+                dict_unref (local->xdata_rsp);
+}
+
+
+int
+afr_frame_return (call_frame_t *frame)
+{
+        afr_local_t *local = NULL;
+        int          call_count = 0;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                call_count = --local->call_count;
+        }
+        UNLOCK (&frame->lock);
+
+        return call_count;
+}
+
+int
+afr_set_elem_count_get (unsigned char *elems, int child_count)
+{
+        int i   = 0;
+        int ret = 0;
+
+        for (i = 0; i < child_count; i++)
+                if (elems[i])
+                        ret++;
+        return ret;
+}
+
+/**
+ * up_children_count - return the number of children that are up
+ */
+
+unsigned int
+afr_up_children_count (unsigned char *child_up, unsigned int child_count)
+{
+        return afr_set_elem_count_get (child_up, child_count);
+}
+
+unsigned int
+afr_locked_children_count (unsigned char *children, unsigned int child_count)
+{
+        return afr_set_elem_count_get (children, child_count);
+}
+
+unsigned int
+afr_pre_op_done_children_count (unsigned char *pre_op,
+                                unsigned int child_count)
+{
+        return afr_set_elem_count_get (pre_op, child_count);
+}
+
+gf_boolean_t
+afr_is_fresh_lookup (loc_t *loc, xlator_t *this)
+{
+        uint64_t          ctx = 0;
+        int32_t           ret = 0;
+
+        GF_ASSERT (loc);
+        GF_ASSERT (this);
+        GF_ASSERT (loc->inode);
+
+        ret = inode_ctx_get (loc->inode, this, &ctx);
+        if (0 == ret)
+                return _gf_false;
+        return _gf_true;
+}
+
+void
+afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *postparent)
+{
+        GF_ASSERT (loc);
+        GF_ASSERT (buf);
+
+        uuid_copy (loc->gfid, buf->ia_gfid);
+        if (postparent)
+                uuid_copy (loc->pargfid, postparent->ia_gfid);
+}
+
+/*
+ * Quota size xattrs are not maintained by afr. There is a
+ * possibility that they differ even when both the directory changelog xattrs
+ * suggest everything is fine. So if there is at least one 'source' check among
+ * the sources which has the maximum quota size. Otherwise check among all the
+ * available ones for maximum quota size. This way if there is a source and
+ * stale copies it always votes for the 'source'.
+ * */
+
+static void
+afr_handle_quota_size (afr_local_t *local, xlator_t *this,
+                       dict_t *rsp_dict)
+{
+        int32_t       *sources       = NULL;
+        dict_t        *xattr         = NULL;
+        data_t        *max_data      = NULL;
+        int64_t       max_quota_size = -1;
+        data_t        *data          = NULL;
+        int64_t       *size          = NULL;
+        int64_t       quota_size     = -1;
+        afr_private_t *priv          = NULL;
+        int           i              = 0;
+        int           ret            = -1;
+        gf_boolean_t  source_present = _gf_false;
+
+        priv    = this->private;
+        sources = local->cont.lookup.sources;
+
+        if (rsp_dict == NULL) {
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "%s: Invalid "
+                                  "response dictionary", local->loc.path);
+                return;
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (sources[i]) {
+                        source_present = _gf_true;
+                        break;
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                /*
+                 * If there is at least one source lets check
+                 * for maximum quota sizes among sources, otherwise take the
+                 * maximum of the ones present to be on the safer side.
+                 */
+                if (source_present && !sources[i])
+                        continue;
+
+                xattr = local->cont.lookup.xattrs[i];
+                if (!xattr)
+                        continue;
+
+                data = dict_get (xattr, QUOTA_SIZE_KEY);
+                if (!data)
+                        continue;
+
+                size = (int64_t*)data->data;
+                quota_size = ntoh64(*size);
+                gf_log (this->name, GF_LOG_DEBUG, "%s: %d, size: %"PRId64,
+                        local->loc.path, i, quota_size);
+                if (quota_size > max_quota_size) {
+                        if (max_data)
+                                data_unref (max_data);
+
+                        max_quota_size = quota_size;
+                        max_data = data_ref (data);
+                }
+        }
+
+        if (max_data) {
+                ret = dict_set (rsp_dict, QUOTA_SIZE_KEY, max_data);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR, "%s: Failed to set "
+                                "quota size", local->loc.path);
+                }
+
+                data_unref (max_data);
+        }
+}
+
+int
+afr_lookup_build_response_params (afr_local_t *local, xlator_t *this)
+{
+        struct iatt     *buf = NULL;
+        struct iatt     *postparent = NULL;
+        dict_t          **xattr = NULL;
+        int32_t         *success_children = NULL;
+        int32_t         *sources = NULL;
+        afr_private_t   *priv = NULL;
+        int32_t         read_child = -1;
+        int             ret = 0;
+        int             i = 0;
+
+        GF_ASSERT (local);
+
+        buf = &local->cont.lookup.buf;
+        postparent = &local->cont.lookup.postparent;
+        xattr = &local->cont.lookup.xattr;
+        priv = this->private;
+
+        read_child = afr_inode_get_read_ctx (this, local->cont.lookup.inode,
+                                             local->fresh_children);
+        if (read_child < 0) {
+                ret = -1;
+                goto out;
+        }
+        success_children = local->cont.lookup.success_children;
+        sources = local->cont.lookup.sources;
+        memset (sources, 0, sizeof (*sources) * priv->child_count);
+        afr_children_intersection_get (local->fresh_children, success_children,
+                                       sources, priv->child_count);
+        if (!sources[read_child]) {
+                read_child = -1;
+                for (i = 0; i < priv->child_count; i++) {
+                        if (sources[i]) {
+                                read_child = i;
+                                break;
+                        }
+                }
+        }
+        if (read_child < 0) {
+                ret = -1;
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG, "Building lookup response from %d",
+                read_child);
+        if (!*xattr)
+                *xattr = dict_ref (local->cont.lookup.xattrs[read_child]);
+
+        *buf = local->cont.lookup.bufs[read_child];
+        *postparent = local->cont.lookup.postparents[read_child];
+
+        if (dict_get (local->xattr_req, QUOTA_SIZE_KEY))
+                afr_handle_quota_size (local, this, *xattr);
+
+        if (IA_INVAL == local->cont.lookup.inode->ia_type) {
+                /* fix for RT #602 */
+                local->cont.lookup.inode->ia_type = buf->ia_type;
+        }
+out:
+        return ret;
+}
+
+static void
+afr_lookup_update_lk_counts (afr_local_t *local, xlator_t *this,
+                            int child_index, dict_t *xattr)
+{
+        uint32_t inodelk_count = 0;
+        uint32_t entrylk_count = 0;
+        int      ret           = -1;
+        uint32_t parent_entrylk = 0;
+
+        GF_ASSERT (local);
+        GF_ASSERT (this);
+        GF_ASSERT (xattr);
+        GF_ASSERT (child_index >= 0);
+
+        ret = dict_get_uint32 (xattr, GLUSTERFS_INODELK_COUNT,
+                               &inodelk_count);
+        if (ret == 0)
+                local->inodelk_count += inodelk_count;
+
+        ret = dict_get_uint32 (xattr, GLUSTERFS_ENTRYLK_COUNT,
+                               &entrylk_count);
+        if (ret == 0)
+                local->entrylk_count += entrylk_count;
+        ret = dict_get_uint32 (xattr, GLUSTERFS_PARENT_ENTRYLK,
+                               &parent_entrylk);
+        if (!ret)
+                local->cont.lookup.parent_entrylk += parent_entrylk;
+}
+
+/*
+ * It's important to maintain a commutative property on do_*_self_heal and
+ * found*; once set, they must not be cleared by a subsequent iteration or
+ * call, so that they represent a logical OR of all iterations and calls
+ * regardless of child/key order.  That allows the caller to call us multiple
+ * times without having to use a separate variable as a "reduce" accumulator.
+ */
+static void
+afr_lookup_set_self_heal_params_by_xattr (afr_local_t *local, xlator_t *this,
+                                          dict_t *xattr)
+{
+        afr_private_t *priv        = NULL;
+        int            i           = 0;
+        int            ret         = -1;
+        void          *pending_raw = NULL;
+        int32_t       *pending     = NULL;
+
+        GF_ASSERT (local);
+        GF_ASSERT (this);
+        GF_ASSERT (xattr);
+
+        priv = this->private;
+
+        for (i = 0; i < priv->child_count; i++) {
+                ret = dict_get_ptr (xattr, priv->pending_key[i],
+                                    &pending_raw);
+                if (ret != 0) {
+                        continue;
+                }
+                pending = pending_raw;
+
+                if (pending[AFR_METADATA_TRANSACTION]) {
+                        gf_log(this->name, GF_LOG_DEBUG,
+                               "metadata self-heal is pending for %s.",
+                               local->loc.path);
+                        local->self_heal.do_metadata_self_heal = _gf_true;
+                }
+
+                if (pending[AFR_ENTRY_TRANSACTION]) {
+                        gf_log(this->name, GF_LOG_DEBUG,
+                               "entry self-heal is pending for %s.",
+                               local->loc.path);
+                        local->self_heal.do_entry_self_heal = _gf_true;
+                }
+
+                if (pending[AFR_DATA_TRANSACTION]) {
+                        gf_log(this->name, GF_LOG_DEBUG,
+                               "data self-heal is pending for %s.",
+                               local->loc.path);
+                        local->self_heal.do_data_self_heal = _gf_true;
+                }
+        }
+}
+
+void
+afr_lookup_check_set_metadata_split_brain (afr_local_t *local, xlator_t *this)
+{
+        int32_t                  *sources = NULL;
+        afr_private_t            *priv = NULL;
+        int32_t                  subvol_status = 0;
+        int32_t                  *success_children   = NULL;
+        dict_t                   **xattrs = NULL;
+        struct iatt              *bufs = NULL;
+        int32_t                  **pending_matrix = NULL;
+
+        priv = this->private;
+
+        sources = GF_CALLOC (priv->child_count, sizeof (*sources),
+                             gf_afr_mt_int32_t);
+        if (NULL == sources)
+                goto out;
+        success_children = local->cont.lookup.success_children;
+        xattrs = local->cont.lookup.xattrs;
+        bufs = local->cont.lookup.bufs;
+        pending_matrix = local->cont.lookup.pending_matrix;
+        afr_build_sources (this, xattrs, bufs, pending_matrix,
+                           sources, success_children, AFR_METADATA_TRANSACTION,
+                           &subvol_status, _gf_false);
+        if (subvol_status & SPLIT_BRAIN)
+                local->cont.lookup.possible_spb = _gf_true;
+out:
+        GF_FREE (sources);
+}
+
+static void
+afr_detect_self_heal_by_iatt (afr_local_t *local, xlator_t *this,
+                            struct iatt *buf, struct iatt *lookup_buf)
+{
+        if (PERMISSION_DIFFERS (buf, lookup_buf)) {
+                /* mismatching permissions */
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "permissions differ for %s ", local->loc.path);
+                local->self_heal.do_metadata_self_heal = _gf_true;
+        }
+
+        if (OWNERSHIP_DIFFERS (buf, lookup_buf)) {
+                /* mismatching permissions */
+                local->self_heal.do_metadata_self_heal = _gf_true;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "ownership differs for %s ", local->loc.path);
+        }
+
+        if (SIZE_DIFFERS (buf, lookup_buf)
+            && IA_ISREG (buf->ia_type)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "size differs for %s ", local->loc.path);
+                local->self_heal.do_data_self_heal = _gf_true;
+        }
+
+        if (uuid_compare (buf->ia_gfid, lookup_buf->ia_gfid)) {
+                /* mismatching gfid */
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "%s: gfid different on subvolume", local->loc.path);
+        }
+}
+
+static void
+afr_detect_self_heal_by_split_brain_status (afr_local_t *local, xlator_t *this)
+{
+        gf_boolean_t split_brain = _gf_false;
+        afr_self_heal_t *sh = NULL;
+
+        sh = &local->self_heal;
+
+        split_brain = afr_is_split_brain (this, local->cont.lookup.inode);
+        split_brain = split_brain || local->cont.lookup.possible_spb;
+        if ((local->success_count > 0) && split_brain &&
+            IA_ISREG (local->cont.lookup.inode->ia_type)) {
+                sh->force_confirm_spb = _gf_true;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "split brain detected during lookup of %s.",
+                        local->loc.path);
+        }
+}
+
+static void
+afr_detect_self_heal_by_lookup_status (afr_local_t *local, xlator_t *this)
+{
+        GF_ASSERT (local);
+        GF_ASSERT (this);
+
+        if ((local->success_count > 0) && (local->enoent_count > 0)) {
+                local->self_heal.do_metadata_self_heal = _gf_true;
+                local->self_heal.do_data_self_heal     = _gf_true;
+                local->self_heal.do_entry_self_heal    = _gf_true;
+                local->self_heal.do_gfid_self_heal    = _gf_true;
+                local->self_heal.do_missing_entry_self_heal    = _gf_true;
+                gf_log(this->name, GF_LOG_DEBUG,
+                       "entries are missing in lookup of %s.",
+                       local->loc.path);
+        }
+
+        return;
+}
+
+gf_boolean_t
+afr_can_self_heal_proceed (afr_self_heal_t *sh, afr_private_t *priv)
+{
+        GF_ASSERT (sh);
+        GF_ASSERT (priv);
+
+        if (sh->force_confirm_spb)
+                return _gf_true;
+        return (sh->do_gfid_self_heal
+                || sh->do_missing_entry_self_heal
+                || (afr_data_self_heal_enabled (priv->data_self_heal) &&
+                    sh->do_data_self_heal)
+                || (priv->metadata_self_heal && sh->do_metadata_self_heal)
+                || (priv->entry_self_heal && sh->do_entry_self_heal));
+}
+
+afr_transaction_type
+afr_transaction_type_get (ia_type_t ia_type)
+{
+        afr_transaction_type    type = AFR_METADATA_TRANSACTION;
+
+        GF_ASSERT (ia_type != IA_INVAL);
+
+        if (IA_ISDIR (ia_type)) {
+                type = AFR_ENTRY_TRANSACTION;
+        } else if (IA_ISREG (ia_type)) {
+                type = AFR_DATA_TRANSACTION;
+        }
+        return type;
+}
+
+int
+afr_lookup_select_read_child (afr_local_t *local, xlator_t *this,
+                              int32_t *read_child)
+{
+        ia_type_t               ia_type        = IA_INVAL;
+        int32_t                 source         = -1;
+        int                     ret            = -1;
+        dict_t                  **xattrs       = NULL;
+        int32_t                 *success_children = NULL;
+        afr_transaction_type    type           = AFR_METADATA_TRANSACTION;
+        uuid_t                  *gfid          = NULL;
+
+        GF_ASSERT (local);
+        GF_ASSERT (this);
+        GF_ASSERT (local->success_count > 0);
+
+        success_children = local->cont.lookup.success_children;
+        /*We can take the success_children[0] only because we already
+         *handle the conflicting children other wise, we could select the
+         *read_child based on wrong file type
+         */
+        ia_type = local->cont.lookup.bufs[success_children[0]].ia_type;
+        type = afr_transaction_type_get (ia_type);
+        xattrs = local->cont.lookup.xattrs;
+        gfid = &local->cont.lookup.buf.ia_gfid;
+        source = afr_lookup_select_read_child_by_txn_type (this, local, xattrs,
+                                                           type, *gfid);
+        if (source < 0) {
+                gf_log (this->name, GF_LOG_DEBUG, "failed to select source "
+                        "for %s", local->loc.path);
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG, "Source selected as %d for %s",
+                source, local->loc.path);
+        *read_child = source;
+        ret = 0;
+out:
+        return ret;
+}
+
+static inline gf_boolean_t
+afr_is_transaction_running (afr_local_t *local)
+{
+        GF_ASSERT (local->fop == GF_FOP_LOOKUP);
+        return ((local->inodelk_count > 0) || (local->entrylk_count > 0));
+}
+
+void
+afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode,
+                      gf_boolean_t background, ia_type_t ia_type, char *reason,
+                      void (*gfid_sh_success_cbk) (call_frame_t *sh_frame,
+                                                   xlator_t *this),
+                      int (*unwind) (call_frame_t *frame, xlator_t *this,
+                                     int32_t op_ret, int32_t op_errno,
+                                     int32_t sh_failed))
+{
+        afr_local_t             *local = NULL;
+        char                    sh_type_str[256] = {0,};
+        char                    *bg = "";
+
+        GF_ASSERT (frame);
+        GF_ASSERT (this);
+        GF_ASSERT (inode);
+        GF_ASSERT (ia_type != IA_INVAL);
+
+        local = frame->local;
+        local->self_heal.background = background;
+        local->self_heal.type       = ia_type;
+        local->self_heal.unwind     = unwind;
+        local->self_heal.gfid_sh_success_cbk     = gfid_sh_success_cbk;
+
+        afr_self_heal_type_str_get (&local->self_heal,
+                                    sh_type_str,
+                                    sizeof (sh_type_str));
+
+        if (background)
+                bg = "background";
+        gf_log (this->name, GF_LOG_DEBUG,
+                "%s %s self-heal triggered. path: %s, reason: %s", bg,
+                sh_type_str, local->loc.path, reason);
+
+        afr_self_heal (frame, this, inode);
+}
+
+unsigned int
+afr_gfid_missing_count (const char *xlator_name, int32_t *success_children,
+                        struct iatt *bufs, unsigned int child_count,
+                        const char *path)
+{
+        unsigned int    gfid_miss_count   = 0;
+        int             i              = 0;
+        struct iatt     *child1        = NULL;
+
+        for (i = 0; i < child_count; i++) {
+                if (success_children[i] == -1)
+                        break;
+                child1 = &bufs[success_children[i]];
+                if (uuid_is_null (child1->ia_gfid)) {
+                        gf_log (xlator_name, GF_LOG_DEBUG, "%s: gfid is null"
+                                " on subvolume %d", path, success_children[i]);
+                        gfid_miss_count++;
+                }
+        }
+
+        return gfid_miss_count;
+}
+
+static int
+afr_lookup_gfid_missing_count (afr_local_t *local, xlator_t *this)
+{
+        int32_t         *success_children = NULL;
+        afr_private_t   *priv          = NULL;
+        struct iatt     *bufs          = NULL;
+        int             miss_count     = 0;
+
+        priv = this->private;
+        bufs = local->cont.lookup.bufs;
+        success_children = local->cont.lookup.success_children;
+
+        miss_count =  afr_gfid_missing_count (this->name, success_children,
+                                              bufs, priv->child_count,
+                                              local->loc.path);
+        return miss_count;
+}
+
+gf_boolean_t
+afr_conflicting_iattrs (struct iatt *bufs, int32_t *success_children,
+                        unsigned int child_count, const char *path,
+                        const char *xlator_name)
+{
+        gf_boolean_t    conflicting    = _gf_false;
+        int             i              = 0;
+        struct iatt     *child1        = NULL;
+        struct iatt     *child2        = NULL;
+        uuid_t          *gfid          = NULL;
+
+        for (i = 0; i < child_count; i++) {
+                if (success_children[i] == -1)
+                        break;
+                child1 = &bufs[success_children[i]];
+                if ((!gfid) && (!uuid_is_null (child1->ia_gfid)))
+                        gfid = &child1->ia_gfid;
+
+                if (i == 0)
+                        continue;
+
+                child2 = &bufs[success_children[i-1]];
+                if (FILETYPE_DIFFERS (child1, child2)) {
+                        gf_log (xlator_name, GF_LOG_DEBUG, "%s: filetype "
+                                "differs on subvolumes (%d, %d)", path,
+                                success_children[i-1], success_children[i]);
+                        conflicting = _gf_true;
+                        goto out;
+                }
+                if (!gfid || uuid_is_null (child1->ia_gfid))
+                        continue;
+                if (uuid_compare (*gfid, child1->ia_gfid)) {
+                       gf_log (xlator_name, GF_LOG_DEBUG, "%s: gfid differs"
+                               " on subvolume %d", path, success_children[i]);
+                       conflicting = _gf_true;
+                       goto out;
+                }
+        }
+out:
+        return conflicting;
+}
+
+/* afr_update_gfid_from_iatts: This function should be called only if the
+ * iatts are not conflicting.
+ */
+void
+afr_update_gfid_from_iatts (uuid_t uuid, struct iatt *bufs,
+                            int32_t *success_children, unsigned int child_count)
+{
+        uuid_t          *gfid = NULL;
+        int             i = 0;
+        int             child = 0;
+
+        for (i = 0; i < child_count; i++) {
+                child = success_children[i];
+                if (child == -1)
+                        break;
+                if ((!gfid) && (!uuid_is_null (bufs[child].ia_gfid))) {
+                        gfid = &bufs[child].ia_gfid;
+                } else if (gfid && (!uuid_is_null (bufs[child].ia_gfid))) {
+                        if (uuid_compare (*gfid, bufs[child].ia_gfid)) {
+                                GF_ASSERT (0);
+                                goto out;
+                        }
+                }
+        }
+        if (gfid && (!uuid_is_null (*gfid)))
+                uuid_copy (uuid, *gfid);
+out:
+        return;
+}
+
+static gf_boolean_t
+afr_lookup_conflicting_entries (afr_local_t *local, xlator_t *this)
+{
+        afr_private_t           *priv = NULL;
+        gf_boolean_t            conflict = _gf_false;
+
+        priv = this->private;
+        conflict =  afr_conflicting_iattrs (local->cont.lookup.bufs,
+                                            local->cont.lookup.success_children,
+                                            priv->child_count, local->loc.path,
+                                            this->name);
+        return conflict;
+}
+
+gf_boolean_t
+afr_open_only_data_self_heal (char *data_self_heal)
+{
+        return !strcmp (data_self_heal, "open");
+}
+
+gf_boolean_t
+afr_data_self_heal_enabled (char *data_self_heal)
+{
+        gf_boolean_t    enabled = _gf_false;
+
+        if (gf_string2boolean (data_self_heal, &enabled) == -1) {
+                enabled = !strcmp (data_self_heal, "open");
+                GF_ASSERT (enabled);
+        }
+
+        return enabled;
+}
+
+static void
+afr_lookup_set_self_heal_params (afr_local_t *local, xlator_t *this)
+{
+        int                     i = 0;
+        struct iatt             *bufs = NULL;
+        dict_t                  **xattr = NULL;
+        afr_private_t           *priv = NULL;
+        int32_t                 child1 = -1;
+        int32_t                 child2 = -1;
+        afr_self_heal_t         *sh = NULL;
+
+        priv  = this->private;
+        sh = &local->self_heal;
+
+        afr_detect_self_heal_by_lookup_status (local, this);
+
+        if (afr_lookup_gfid_missing_count (local, this))
+                local->self_heal.do_gfid_self_heal    = _gf_true;
+
+        if (_gf_true == afr_lookup_conflicting_entries (local, this))
+                local->self_heal.do_missing_entry_self_heal    = _gf_true;
+        else
+                afr_update_gfid_from_iatts (local->self_heal.sh_gfid_req,
+                                            local->cont.lookup.bufs,
+                                            local->cont.lookup.success_children,
+                                            priv->child_count);
+
+        bufs = local->cont.lookup.bufs;
+        for (i = 1; i < local->success_count; i++) {
+                child1 = local->cont.lookup.success_children[i-1];
+                child2 = local->cont.lookup.success_children[i];
+                afr_detect_self_heal_by_iatt (local, this,
+                                              &bufs[child1], &bufs[child2]);
+        }
+
+        xattr = local->cont.lookup.xattrs;
+        for (i = 0; i < local->success_count; i++) {
+                child1 = local->cont.lookup.success_children[i];
+                afr_lookup_set_self_heal_params_by_xattr (local, this,
+                                                          xattr[child1]);
+        }
+        if (afr_open_only_data_self_heal (priv->data_self_heal))
+                sh->do_data_self_heal = _gf_false;
+        if (sh->do_metadata_self_heal)
+                afr_lookup_check_set_metadata_split_brain (local, this);
+        afr_detect_self_heal_by_split_brain_status (local, this);
+}
+
+int
+afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno,
+                             int32_t sh_failed)
+{
+        afr_local_t *local = NULL;
+        int         ret    = -1;
+        dict_t      *xattr = NULL;
+
+        local = frame->local;
+
+        if (op_ret == -1) {
+                local->op_ret = -1;
+		local->op_errno = afr_most_important_error(local->op_errno,
+							   op_errno, _gf_true);
+
+                goto out;
+        } else {
+                local->op_ret = 0;
+        }
+
+        afr_lookup_done_success_action (frame, this, _gf_true);
+        xattr = local->cont.lookup.xattr;
+        if (xattr) {
+                ret = dict_set_int32 (xattr, "sh-failed", sh_failed);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR, "%s: Failed to set "
+                                "sh-failed to %d", local->loc.path, sh_failed);
+
+                if (local->self_heal.actual_sh_started == _gf_true &&
+                    sh_failed == 0) {
+                        ret = dict_set_int32 (xattr, "actual-sh-done", 1);
+                        if (ret)
+                                gf_log(this->name, GF_LOG_ERROR, "%s: Failed to"
+                                       " set actual-sh-done to %d",
+                                       local->loc.path,
+                                       local->self_heal.actual_sh_started);
+                }
+        }
+out:
+        AFR_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
+                          local->cont.lookup.inode, &local->cont.lookup.buf,
+                          local->cont.lookup.xattr,
+                          &local->cont.lookup.postparent);
+
+        return 0;
+}
+
+//TODO: At the moment only lookup needs this, so not doing any checks, in the
+// future we will have to do fop specific operations
+void
+afr_post_gfid_sh_success (call_frame_t *sh_frame, xlator_t *this)
+{
+        afr_local_t             *local = NULL;
+        afr_local_t             *sh_local = NULL;
+        afr_private_t           *priv = NULL;
+        afr_self_heal_t         *sh = NULL;
+        int                     i = 0;
+        struct iatt             *lookup_bufs = NULL;
+        struct iatt             *lookup_parentbufs = NULL;
+
+        sh_local = sh_frame->local;
+        sh       = &sh_local->self_heal;
+        local = sh->orig_frame->local;
+        lookup_bufs = local->cont.lookup.bufs;
+        lookup_parentbufs = local->cont.lookup.postparents;
+        priv = this->private;
+
+        memcpy (lookup_bufs, sh->buf, priv->child_count * sizeof (*sh->buf));
+        memcpy (lookup_parentbufs, sh->parentbufs,
+                priv->child_count * sizeof (*sh->parentbufs));
+
+        afr_reset_xattr (local->cont.lookup.xattrs, priv->child_count);
+        if (local->cont.lookup.xattr) {
+                dict_unref (local->cont.lookup.xattr);
+                local->cont.lookup.xattr = NULL;
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (sh->xattr[i])
+                        local->cont.lookup.xattrs[i] = dict_ref (sh->xattr[i]);
+        }
+
+        afr_reset_children (local->cont.lookup.success_children,
+                            priv->child_count);
+        afr_children_copy (local->cont.lookup.success_children,
+                           sh->fresh_children, priv->child_count);
+}
+
+static void
+afr_lookup_perform_self_heal (call_frame_t *frame, xlator_t *this,
+                              gf_boolean_t *sh_launched)
+{
+        unsigned int         up_count = 0;
+        afr_private_t       *priv    = NULL;
+        afr_local_t         *local   = NULL;
+        char                *reason  = NULL;
+
+        GF_ASSERT (sh_launched);
+        *sh_launched = _gf_false;
+        priv         = this->private;
+        local        = frame->local;
+
+        up_count  = afr_up_children_count (local->child_up, priv->child_count);
+        if (up_count == 1) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Only 1 child up - do not attempt to detect self heal");
+                goto out;
+        }
+
+        afr_lookup_set_self_heal_params (local, this);
+        if (afr_can_self_heal_proceed (&local->self_heal, priv)) {
+                if  (afr_is_transaction_running (local) &&
+                     (!local->allow_sh_for_running_transaction))
+                        goto out;
+
+                reason = "lookup detected pending operations";
+                afr_launch_self_heal (frame, this, local->cont.lookup.inode,
+                                      _gf_true, local->cont.lookup.buf.ia_type,
+                                      reason, afr_post_gfid_sh_success,
+                                      afr_self_heal_lookup_unwind);
+                *sh_launched = _gf_true;
+        }
+out:
+        return;
+}
+
+void
+afr_get_fresh_children (int32_t *success_children, int32_t *sources,
+                        int32_t *fresh_children, unsigned int child_count)
+{
+        unsigned int i = 0;
+        unsigned int j = 0;
+
+        GF_ASSERT (success_children);
+        GF_ASSERT (sources);
+        GF_ASSERT (fresh_children);
+
+        afr_reset_children (fresh_children, child_count);
+        for (i = 0; i < child_count; i++) {
+                if (success_children[i] == -1)
+                        break;
+                if (afr_is_read_child (success_children, sources, child_count,
+                                       success_children[i])) {
+                        fresh_children[j] = success_children[i];
+                        j++;
+                }
+        }
+}
+
+static int
+afr_lookup_set_read_ctx (afr_local_t *local, xlator_t *this, int32_t read_child)
+{
+        afr_private_t           *priv = NULL;
+
+        GF_ASSERT (read_child >= 0);
+
+        priv = this->private;
+        afr_get_fresh_children (local->cont.lookup.success_children,
+                                local->cont.lookup.sources,
+                                local->fresh_children, priv->child_count);
+        afr_inode_set_read_ctx (this, local->cont.lookup.inode, read_child,
+                                local->fresh_children);
+
+        return 0;
+}
+
+int
+afr_lookup_done_success_action (call_frame_t *frame, xlator_t *this,
+                                gf_boolean_t fail_conflict)
+{
+        int32_t             read_child = -1;
+        int32_t             ret        = -1;
+        afr_local_t         *local     = NULL;
+        gf_boolean_t        fresh_lookup = _gf_false;
+
+        local   = frame->local;
+        fresh_lookup = local->cont.lookup.fresh_lookup;
+
+        if (local->loc.parent == NULL)
+                fail_conflict = _gf_true;
+
+        if (afr_lookup_conflicting_entries (local, this)) {
+                if (fail_conflict == _gf_false)
+                        ret = 0;
+                goto out;
+        }
+
+        ret = afr_lookup_select_read_child (local, this, &read_child);
+        if (!afr_is_transaction_running (local) || fresh_lookup) {
+                if (read_child < 0)
+                        goto out;
+
+                ret = afr_lookup_set_read_ctx (local, this, read_child);
+                if (ret)
+                        goto out;
+        }
+
+        ret = afr_lookup_build_response_params (local, this);
+        if (ret)
+                goto out;
+        afr_update_loc_gfids (&local->loc,
+                              &local->cont.lookup.buf,
+                              &local->cont.lookup.postparent);
+
+        ret = 0;
+out:
+        if (ret) {
+                local->op_ret = -1;
+                local->op_errno = EIO;
+        }
+        return ret;
+}
+
+int
+afr_lookup_get_latest_subvol (afr_local_t *local, xlator_t *this)
+{
+        afr_private_t *priv = NULL;
+        int32_t       *success_children = NULL;
+        struct iatt   *bufs = NULL;
+        int           i = 0;
+        int           child = 0;
+        int           lsubvol = -1;
+
+        priv = this->private;
+        success_children = local->cont.lookup.success_children;
+        bufs = local->cont.lookup.bufs;
+        for (i = 0; i < priv->child_count; i++) {
+                child = success_children[i];
+                if (child == -1)
+                        break;
+                if (uuid_is_null (bufs[child].ia_gfid))
+                        continue;
+                if (lsubvol < 0) {
+                        lsubvol = child;
+                } else if (bufs[lsubvol].ia_ctime < bufs[child].ia_ctime) {
+                        lsubvol = child;
+                } else if ((bufs[lsubvol].ia_ctime == bufs[child].ia_ctime) &&
+                  (bufs[lsubvol].ia_ctime_nsec < bufs[child].ia_ctime_nsec)) {
+                        lsubvol = child;
+                }
+        }
+        return lsubvol;
+}
+
+void
+afr_lookup_mark_other_entries_stale (afr_local_t *local, xlator_t *this,
+                                     int subvol)
+{
+        afr_private_t *priv = NULL;
+        int32_t       *success_children = NULL;
+        struct iatt   *bufs = NULL;
+        int           i = 0;
+        int           child = 0;
+
+        priv = this->private;
+        success_children = local->cont.lookup.success_children;
+        bufs = local->cont.lookup.bufs;
+        memcpy (local->fresh_children, success_children,
+                sizeof (*success_children) * priv->child_count);
+        for (i = 0; i < priv->child_count; i++) {
+                child = local->fresh_children[i];
+                if (child == -1)
+                        break;
+                if (child == subvol)
+                        continue;
+                if (uuid_is_null (bufs[child].ia_gfid) &&
+                    (bufs[child].ia_type == bufs[subvol].ia_type))
+                        continue;
+                afr_children_rm_child (success_children, child,
+                                       priv->child_count);
+                local->success_count--;
+        }
+        afr_reset_children (local->fresh_children, priv->child_count);
+}
+
+void
+afr_succeed_lookup_on_latest_iatt (afr_local_t *local, xlator_t *this)
+{
+        int    lsubvol = 0;
+
+        if (!afr_lookup_conflicting_entries (local, this))
+                goto out;
+
+        lsubvol = afr_lookup_get_latest_subvol (local, this);
+        if (lsubvol < 0)
+                goto out;
+        afr_lookup_mark_other_entries_stale (local, this, lsubvol);
+out:
+        return;
+}
+
+gf_boolean_t
+afr_is_entry_possibly_under_creation (afr_local_t *local, xlator_t *this)
+{
+        /*
+         * We need to perform this test in lookup done and treat on going
+         * create/DELETE as ENOENT.
+         * Reason:
+        Multiple clients A, B and C are attempting 'mkdir -p /mnt/a/b/c'
+
+        1 Client A is in the middle of mkdir(/a). It has acquired lock.
+          It has performed mkdir(/a) on one subvol, and second one is still
+          in progress
+        2 Client B performs a lookup, sees directory /a on one,
+          ENOENT on the other, succeeds lookup.
+        3 Client B performs lookup on /a/b on both subvols, both return ENOENT
+          (one subvol because /a/b does not exist, another because /a
+          itself does not exist)
+        4 Client B proceeds to mkdir /a/b. It obtains entrylk on inode=/a with
+          basename=b on one subvol, but fails on other subvol as /a is yet to
+          be created by Client A.
+        5 Client A finishes mkdir of /a on other subvol
+        6 Client C also attempts to create /a/b, lookup returns ENOENT on
+          both subvols.
+        7 Client C tries to obtain entrylk on on inode=/a with basename=b,
+          obtains on one subvol (where B had failed), and waits for B to unlock
+          on other subvol.
+        8 Client B finishes mkdir() on one subvol with GFID-1 and completes
+          transaction and unlocks
+        9 Client C gets the lock on the second subvol, At this stage second
+          subvol already has /a/b created from Client B, but Client C does not
+          check that in the middle of mkdir transaction
+        10 Client C attempts mkdir /a/b on both subvols. It succeeds on
+           ONLY ONE (where Client B could not get lock because of
+           missing parent /a dir) with GFID-2, and gets EEXIST from ONE subvol.
+        This way we have /a/b in GFID mismatch. One subvol got GFID-1 because
+        Client B performed transaction on only one subvol (because entrylk()
+        could not be obtained on second subvol because of missing parent dir --
+        caused by premature/speculative succeeding of lookup() on /a when locks
+        are detected). Other subvol gets GFID-2 from Client C because while
+        it was waiting for entrylk() on both subvols, Client B was in the
+        middle of creating mkdir() on only one subvol, and Client C does not
+        "expect" this when it is between lock() and pre-op()/op() phase of the
+        transaction.
+         */
+	if (local->cont.lookup.parent_entrylk && local->enoent_count)
+		return _gf_true;
+
+	return _gf_false;
+}
+
+
+static void
+afr_lookup_done (call_frame_t *frame, xlator_t *this)
+{
+        int                 unwind = 1;
+        afr_private_t       *priv  = NULL;
+        afr_local_t         *local = NULL;
+        int                 ret = -1;
+        gf_boolean_t        sh_launched = _gf_false;
+        gf_boolean_t        fail_conflict = _gf_false;
+        int                 gfid_miss_count = 0;
+        int                 enotconn_count = 0;
+        int                 up_children_count = 0;
+
+        priv  = this->private;
+        local = frame->local;
+
+	if (afr_is_entry_possibly_under_creation (local, this)) {
+		local->op_ret = -1;
+		local->op_errno = ENOENT;
+		goto unwind;
+	}
+
+        if (local->op_ret < 0)
+                goto unwind;
+
+        if (local->cont.lookup.parent_entrylk && local->success_count > 1)
+                afr_succeed_lookup_on_latest_iatt (local, this);
+
+        gfid_miss_count = afr_lookup_gfid_missing_count (local, this);
+        up_children_count = afr_up_children_count (local->child_up,
+                                                   priv->child_count);
+        enotconn_count = priv->child_count - up_children_count;
+        if ((gfid_miss_count == local->success_count) &&
+            (enotconn_count > 0)) {
+                local->op_ret = -1;
+                local->op_errno = EIO;
+                gf_log (this->name, GF_LOG_ERROR, "Failing lookup for %s, "
+                        "LOOKUP on a file without gfid is not allowed when "
+                        "some of the children are down", local->loc.path);
+                goto unwind;
+        }
+
+        if ((gfid_miss_count == local->success_count) &&
+            uuid_is_null (local->cont.lookup.gfid_req)) {
+                local->op_ret = -1;
+                local->op_errno = ENODATA;
+                gf_log (this->name, GF_LOG_ERROR, "%s: No gfid present",
+                        local->loc.path);
+                goto unwind;
+        }
+
+        if (gfid_miss_count && uuid_is_null (local->cont.lookup.gfid_req))
+                fail_conflict = _gf_true;
+        ret = afr_lookup_done_success_action (frame, this, fail_conflict);
+        if (ret)
+                goto unwind;
+        uuid_copy (local->self_heal.sh_gfid_req, local->cont.lookup.gfid_req);
+
+        afr_lookup_perform_self_heal (frame, this, &sh_launched);
+        if (sh_launched) {
+                unwind = 0;
+                goto unwind;
+        }
+
+ unwind:
+         if (unwind) {
+                 AFR_STACK_UNWIND (lookup, frame, local->op_ret,
+                                   local->op_errno, local->cont.lookup.inode,
+                                   &local->cont.lookup.buf,
+                                   local->cont.lookup.xattr,
+                                   &local->cont.lookup.postparent);
+        }
+}
+
+/*
+ * During a lookup, some errors are more "important" than
+ * others in that they must be given higher priority while
+ * returning to the user.
+ *
+ * The hierarchy is ESTALE > EIO > ENOENT > others
+ */
+int32_t
+afr_most_important_error(int32_t old_errno, int32_t new_errno,
+			 gf_boolean_t eio)
+{
+	if (old_errno == ESTALE || new_errno == ESTALE)
+		return ESTALE;
+	if (eio && (old_errno == EIO || new_errno == EIO))
+		return EIO;
+	if (old_errno == ENOENT || new_errno == ENOENT)
+		return ENOENT;
+
+	return new_errno;
+}
+
+int32_t
+afr_resultant_errno_get (int32_t *children,
+                         int *child_errno, unsigned int child_count)
+{
+        int     i = 0;
+        int32_t op_errno = 0;
+        int     child = 0;
+
+        for (i = 0; i < child_count; i++) {
+                if (children) {
+                        child = children[i];
+                        if (child == -1)
+                                break;
+                } else {
+                        child = i;
+                }
+		op_errno = afr_most_important_error(op_errno,
+						    child_errno[child],
+						    _gf_false);
+        }
+        return op_errno;
+}
+
+static void
+afr_lookup_handle_error (afr_local_t *local, int32_t op_ret,  int32_t op_errno)
+{
+        GF_ASSERT (local);
+        if (op_errno == ENOENT)
+                local->enoent_count++;
+
+	local->op_errno = afr_most_important_error(local->op_errno, op_errno,
+						   _gf_false);
+
+        if (local->op_errno == ESTALE) {
+                local->op_ret = -1;
+        }
+}
+
+static void
+afr_set_root_inode_on_first_lookup (afr_local_t *local, xlator_t *this,
+                                    inode_t *inode)
+{
+        afr_private_t           *priv = NULL;
+        GF_ASSERT (inode);
+
+        if (!__is_root_gfid (inode->gfid))
+                goto out;
+        if (!afr_is_fresh_lookup (&local->loc, this))
+                goto out;
+        priv = this->private;
+        if ((priv->first_lookup)) {
+                gf_log (this->name, GF_LOG_INFO, "added root inode");
+                priv->root_inode = inode_ref (inode);
+                priv->first_lookup = 0;
+        }
+out:
+        return;
+}
+
+static void
+afr_lookup_cache_args (afr_local_t *local, int child_index, dict_t *xattr,
+                       struct iatt *buf, struct iatt *postparent)
+{
+        GF_ASSERT (child_index >= 0);
+        local->cont.lookup.xattrs[child_index] = dict_ref (xattr);
+        local->cont.lookup.postparents[child_index] = *postparent;
+        local->cont.lookup.bufs[child_index] = *buf;
+}
+
+static void
+afr_lookup_handle_first_success (afr_local_t *local, xlator_t *this,
+                                 inode_t *inode, struct iatt *buf)
+{
+        local->cont.lookup.inode      = inode_ref (inode);
+        local->cont.lookup.buf        = *buf;
+        afr_set_root_inode_on_first_lookup (local, this, inode);
+}
+
+static int32_t
+afr_discovery_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *dict,
+                   dict_t *xdata)
+{
+        int              ret            = 0;
+        char            *pathinfo       = NULL;
+        gf_boolean_t     is_local        = _gf_false;
+        afr_private_t   *priv           = NULL;
+        int32_t          child_index    = -1;
+
+        if (op_ret != 0) {
+                goto out;
+        }
+
+        ret = dict_get_str (dict, GF_XATTR_PATHINFO_KEY, &pathinfo);
+        if (ret != 0) {
+                goto out;
+        }
+
+        ret = afr_local_pathinfo (pathinfo, &is_local);
+        if (ret) {
+                goto out;
+        }
+
+        priv = this->private;
+        /*
+         * Note that one local subvolume will override another here.  The only
+         * way to avoid that would be to retain extra information about whether
+         * the previous read_child is local, and it's just not worth it.  Even
+         * the slowest local subvolume is far preferable to a remote one.
+         */
+        if (is_local) {
+                child_index = (int32_t)(long)cookie;
+                gf_log (this->name, GF_LOG_INFO,
+                        "selecting local read_child %s",
+                        priv->children[child_index]->name);
+                priv->read_child = child_index;
+        }
+
+out:
+        STACK_DESTROY(frame->root);
+        return 0;
+}
+
+static void
+afr_attempt_local_discovery (xlator_t *this, int32_t child_index)
+{
+        call_frame_t    *newframe = NULL;
+        loc_t            tmploc = {0,};
+        afr_private_t   *priv = this->private;
+
+        newframe = create_frame(this,this->ctx->pool);
+        if (!newframe) {
+                return;
+        }
+
+        tmploc.gfid[sizeof(tmploc.gfid)-1] = 1;
+        STACK_WIND_COOKIE (newframe, afr_discovery_cbk,
+                           (void *)(long)child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->getxattr,
+                           &tmploc, GF_XATTR_PATHINFO_KEY, NULL);
+}
+
+static void
+afr_lookup_handle_success (afr_local_t *local, xlator_t *this, int32_t child_index,
+                           int32_t op_ret, int32_t op_errno, inode_t *inode,
+                           struct iatt *buf, dict_t *xattr,
+                           struct iatt *postparent)
+{
+        afr_private_t   *priv   = this->private;
+
+        if (local->success_count == 0) {
+                if (local->op_errno != ESTALE) {
+                        local->op_ret = op_ret;
+                        local->op_errno = 0;
+                }
+                afr_lookup_handle_first_success (local, this, inode, buf);
+        }
+        afr_lookup_update_lk_counts (local, this,
+                                     child_index, xattr);
+
+        afr_lookup_cache_args (local, child_index, xattr,
+                               buf, postparent);
+
+        if (local->do_discovery && (priv->read_child == (-1))) {
+                afr_attempt_local_discovery(this,child_index);
+        }
+
+        local->cont.lookup.success_children[local->success_count] = child_index;
+        local->success_count++;
+}
+
+int
+afr_lookup_cbk (call_frame_t *frame, void *cookie,
+                xlator_t *this,  int32_t op_ret,  int32_t op_errno,
+                inode_t *inode,   struct iatt *buf, dict_t *xattr,
+                struct iatt *postparent)
+{
+        afr_local_t *   local = NULL;
+        int             call_count      = -1;
+        int             child_index     = -1;
+
+         child_index = (long) cookie;
+
+        LOCK (&frame->lock);
+        {
+                local = frame->local;
+
+                if (op_ret == -1) {
+                        afr_lookup_handle_error (local, op_ret, op_errno);
+                        goto unlock;
+                }
+                afr_lookup_handle_success (local, this, child_index, op_ret,
+                                           op_errno, inode, buf, xattr,
+                                           postparent);
+
+         }
+unlock:
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+        if (call_count == 0) {
+               afr_lookup_done (frame, this);
+        }
+
+         return 0;
+}
+
+int
+afr_lookup_cont_init (afr_local_t *local, unsigned int child_count)
+{
+        int               ret            = -ENOMEM;
+        struct iatt       *iatts         = NULL;
+        int32_t           *success_children = NULL;
+        int32_t           *sources       = NULL;
+        int32_t           **pending_matrix = NULL;
+
+        GF_ASSERT (local);
+        local->cont.lookup.xattrs = GF_CALLOC (child_count,
+                                               sizeof (*local->cont.lookup.xattr),
+                                               gf_afr_mt_dict_t);
+        if (NULL == local->cont.lookup.xattrs)
+                goto out;
+
+        iatts = GF_CALLOC (child_count, sizeof (*iatts), gf_afr_mt_iatt);
+        if (NULL == iatts)
+                goto out;
+        local->cont.lookup.postparents = iatts;
+
+        iatts = GF_CALLOC (child_count, sizeof (*iatts), gf_afr_mt_iatt);
+        if (NULL == iatts)
+                goto out;
+        local->cont.lookup.bufs = iatts;
+
+        success_children = afr_children_create (child_count);
+        if (NULL == success_children)
+                goto out;
+        local->cont.lookup.success_children = success_children;
+
+        local->fresh_children = afr_children_create (child_count);
+        if (NULL == local->fresh_children)
+                goto out;
+
+        sources = GF_CALLOC (sizeof (*sources), child_count, gf_afr_mt_int32_t);
+        if (NULL == sources)
+                goto out;
+        local->cont.lookup.sources = sources;
+
+        pending_matrix = afr_matrix_create (child_count, child_count);
+        if (NULL == pending_matrix)
+                goto out;
+        local->cont.lookup.pending_matrix = pending_matrix;
+
+        ret = 0;
+out:
+        return ret;
+}
+
+int
+afr_lookup (call_frame_t *frame, xlator_t *this,
+            loc_t *loc, dict_t *xattr_req)
+{
+        afr_private_t    *priv           = NULL;
+        afr_local_t      *local          = NULL;
+        void              *gfid_req      = NULL;
+        int               ret            = -1;
+        int               i              = 0;
+        int               call_count     = 0;
+        uint64_t          ctx            = 0;
+        int32_t           op_errno       = 0;
+        int               allow_sh       = 0;
+        priv = this->private;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (local, out);
+
+        local->op_ret = -1;
+
+        frame->local = local;
+        local->fop = GF_FOP_LOOKUP;
+
+        loc_copy (&local->loc, loc);
+        ret = loc_path (&local->loc, NULL);
+        if (ret < 0) {
+                op_errno = EINVAL;
+                goto out;
+        }
+
+        if (local->loc.path &&
+            (strcmp (local->loc.path, "/" GF_REPLICATE_TRASH_DIR) == 0)) {
+                op_errno = EPERM;
+                ret = -1;
+                goto out;
+        }
+
+        ret = inode_ctx_get (local->loc.inode, this, &ctx);
+        if (ret == 0) {
+                /* lookup is a revalidate */
+
+                local->read_child_index = afr_inode_get_read_ctx (this,
+                                                               local->loc.inode,
+                                                               NULL);
+        } else {
+                LOCK (&priv->read_child_lock);
+                {
+                        if (priv->hash_mode) {
+                                local->read_child_index = -1;
+                        }
+                        else {
+                                local->read_child_index =
+                                        (++priv->read_child_rr) %
+                                        (priv->child_count);
+                        }
+                }
+                UNLOCK (&priv->read_child_lock);
+                local->cont.lookup.fresh_lookup = _gf_true;
+        }
+
+        local->child_up = memdup (priv->child_up,
+                                  sizeof (*local->child_up) * priv->child_count);
+        if (NULL == local->child_up) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        ret = afr_lookup_cont_init (local, priv->child_count);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
+
+        local->call_count = afr_up_children_count (local->child_up,
+                                                   priv->child_count);
+        call_count = local->call_count;
+        if (local->call_count == 0) {
+                ret      = -1;
+                op_errno = ENOTCONN;
+                goto out;
+        }
+
+        /* By default assume ENOTCONN. On success it will be set to 0. */
+        local->op_errno = ENOTCONN;
+
+        ret = dict_get_int32 (xattr_req, "allow-sh-for-running-transaction",
+                              &allow_sh);
+        dict_del (xattr_req, "allow-sh-for-running-transaction");
+        local->allow_sh_for_running_transaction = allow_sh;
+
+        ret = afr_lookup_xattr_req_prepare (local, this, xattr_req, &local->loc,
+                                            &gfid_req);
+        if (ret) {
+                local->op_errno = -ret;
+                goto out;
+        }
+        afr_lookup_save_gfid (local->cont.lookup.gfid_req, gfid_req,
+                              &local->loc);
+        local->fop = GF_FOP_LOOKUP;
+        if (priv->choose_local && !priv->did_discovery) {
+                if (gfid_req && __is_root_gfid(gfid_req)) {
+                        local->do_discovery = _gf_true;
+                        priv->did_discovery = _gf_true;
+                }
+        }
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND_COOKIE (frame, afr_lookup_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->lookup,
+                                           &local->loc, local->xattr_req);
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret)
+                AFR_STACK_UNWIND (lookup, frame, -1, op_errno,
+                                  NULL, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+/* {{{ open */
+
+int
+__afr_fd_ctx_set (xlator_t *this, fd_t *fd)
+{
+        afr_private_t * priv   = NULL;
+        int             ret    = -1;
+        uint64_t        ctx    = 0;
+        afr_fd_ctx_t *  fd_ctx = NULL;
+
+        VALIDATE_OR_GOTO (this->private, out);
+        VALIDATE_OR_GOTO (fd, out);
+
+        priv = this->private;
+
+        ret = __fd_ctx_get (fd, this, &ctx);
+
+        if (ret == 0)
+                goto out;
+
+        fd_ctx = GF_CALLOC (1, sizeof (afr_fd_ctx_t),
+                            gf_afr_mt_afr_fd_ctx_t);
+        if (!fd_ctx) {
+                ret = -ENOMEM;
+                goto out;
+        }
+
+        fd_ctx->pre_op_done = GF_CALLOC (sizeof (*fd_ctx->pre_op_done),
+                                         priv->child_count,
+                                         gf_afr_mt_char);
+        if (!fd_ctx->pre_op_done) {
+                ret = -ENOMEM;
+                goto out;
+        }
+
+        fd_ctx->pre_op_piggyback = GF_CALLOC (sizeof (*fd_ctx->pre_op_piggyback),
+                                              priv->child_count,
+                                              gf_afr_mt_char);
+        if (!fd_ctx->pre_op_piggyback) {
+                ret = -ENOMEM;
+                goto out;
+        }
+
+        fd_ctx->opened_on = GF_CALLOC (sizeof (*fd_ctx->opened_on),
+                                       priv->child_count,
+                                       gf_afr_mt_int32_t);
+        if (!fd_ctx->opened_on) {
+                ret = -ENOMEM;
+                goto out;
+        }
+
+        fd_ctx->lock_piggyback = GF_CALLOC (sizeof (*fd_ctx->lock_piggyback),
+                                            priv->child_count,
+                                            gf_afr_mt_char);
+        if (!fd_ctx->lock_piggyback) {
+                ret = -ENOMEM;
+                goto out;
+        }
+
+        fd_ctx->lock_acquired = GF_CALLOC (sizeof (*fd_ctx->lock_acquired),
+                                           priv->child_count,
+                                           gf_afr_mt_char);
+        if (!fd_ctx->lock_acquired) {
+                ret = -ENOMEM;
+                goto out;
+        }
+
+        fd_ctx->up_count   = priv->up_count;
+        fd_ctx->down_count = priv->down_count;
+
+        fd_ctx->locked_on = GF_CALLOC (sizeof (*fd_ctx->locked_on),
+                                       priv->child_count,
+                                       gf_afr_mt_char);
+        if (!fd_ctx->locked_on) {
+                ret = -ENOMEM;
+                goto out;
+        }
+
+	pthread_mutex_init (&fd_ctx->delay_lock, NULL);
+        INIT_LIST_HEAD (&fd_ctx->entries);
+        fd_ctx->call_child = -1;
+
+        INIT_LIST_HEAD (&fd_ctx->eager_locked);
+
+        ret = __fd_ctx_set (fd, this, (uint64_t)(long) fd_ctx);
+        if (ret)
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "failed to set fd ctx (%p)", fd);
+out:
+        return ret;
+}
+
+
+int
+afr_fd_ctx_set (xlator_t *this, fd_t *fd)
+{
+        int ret = -1;
+
+        LOCK (&fd->lock);
+        {
+                ret = __afr_fd_ctx_set (this, fd);
+        }
+        UNLOCK (&fd->lock);
+
+        return ret;
+}
+
+/* {{{ flush */
+
+int
+afr_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+              int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_local_t *   local = NULL;
+        int call_count  = -1;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret != -1) {
+                        if (local->success_count == 0) {
+                                local->op_ret = op_ret;
+                        }
+                        local->success_count++;
+                }
+
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+	call_count = afr_frame_return (frame);
+
+	if (call_count == 0)
+		AFR_STACK_UNWIND(flush, frame, local->op_ret,
+				 local->op_errno, NULL);
+
+        return 0;
+}
+
+static int
+afr_flush_wrapper (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+        int           i      = 0;
+        afr_local_t   *local = NULL;
+        afr_private_t *priv  = NULL;
+        int call_count       = -1;
+
+        priv = this->private;
+        local = frame->local;
+        call_count = local->call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND_COOKIE (frame, afr_flush_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->flush,
+                                           local->fd, NULL);
+                        if (!--call_count)
+                                break;
+
+                }
+        }
+
+        return 0;
+}
+
+int
+afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+        afr_local_t   *local = NULL;
+        call_stub_t   *stub = NULL;
+        int            ret        = -1;
+        int            op_errno   = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+	AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+	local = frame->local;
+
+	ret = afr_local_init(local, priv, &op_errno);
+	if (ret < 0)
+		goto out;
+
+	local->fd = fd_ref(fd);
+        stub = fop_flush_stub (frame, afr_flush_wrapper, fd, xdata);
+        if (!stub) {
+                ret = -1;
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        afr_delayed_changelog_wake_resume (this, fd, stub);
+	ret = 0;
+
+out:
+	if (ret < 0)
+		AFR_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+
+        return 0;
+}
+
+/* }}} */
+
+
+int
+afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd)
+{
+        uint64_t        ctx = 0;
+        afr_fd_ctx_t    *fd_ctx = NULL;
+        int             ret = 0;
+
+        ret = fd_ctx_get (fd, this, &ctx);
+        if (ret < 0)
+                goto out;
+
+        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+
+        if (fd_ctx) {
+                GF_FREE (fd_ctx->pre_op_done);
+
+                GF_FREE (fd_ctx->opened_on);
+
+                GF_FREE (fd_ctx->locked_on);
+
+                GF_FREE (fd_ctx->pre_op_piggyback);
+                GF_FREE (fd_ctx->lock_piggyback);
+
+                GF_FREE (fd_ctx->lock_acquired);
+
+		pthread_mutex_destroy (&fd_ctx->delay_lock);
+
+                GF_FREE (fd_ctx);
+        }
+
+out:
+        return 0;
+}
+
+
+int
+afr_release (xlator_t *this, fd_t *fd)
+{
+        afr_locked_fd_t *locked_fd = NULL;
+        afr_locked_fd_t *tmp       = NULL;
+        afr_private_t   *priv      = NULL;
+
+        priv = this->private;
+
+        afr_cleanup_fd_ctx (this, fd);
+
+        list_for_each_entry_safe (locked_fd, tmp, &priv->saved_fds,
+                                  list) {
+
+                if (locked_fd->fd == fd) {
+                        list_del_init (&locked_fd->list);
+                        GF_FREE (locked_fd);
+                }
+
+        }
+
+        return 0;
+}
+
+
+/* {{{ fsync */
+
+int
+afr_fsync_unwind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+        AFR_STACK_UNWIND (fsync, frame, op_ret, op_errno, prebuf, postbuf,
+                          xdata);
+        return 0;
+}
+
+int
+afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+               struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t *local = NULL;
+        int call_count = -1;
+        int child_index = (long) cookie;
+        int read_child  = 0;
+	call_stub_t *stub = NULL;
+
+        local = frame->local;
+
+        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
+
+                if (op_ret == 0) {
+                        local->op_ret = 0;
+
+                        if (local->success_count == 0) {
+                                local->cont.inode_wfop.prebuf  = *prebuf;
+                                local->cont.inode_wfop.postbuf = *postbuf;
+                        }
+
+                        if (child_index == read_child) {
+                                local->cont.inode_wfop.prebuf  = *prebuf;
+                                local->cont.inode_wfop.postbuf = *postbuf;
+                        }
+
+                        local->success_count++;
+                }
+
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+		/* Make a stub out of the frame, and register it
+		   with the waking up post-op. When the call-stub resumes,
+		   we are guaranteed that there was no post-op pending
+		   (i.e changelogs were unset in the server). This is an
+		   essential "guarantee", that fsync() returns only after
+		   completely finishing EVERYTHING, including the delayed
+		   post-op. This guarantee is expected by FUSE graph switching
+		   for example.
+		*/
+		stub = fop_fsync_cbk_stub (frame, afr_fsync_unwind_cbk,
+                                           local->op_ret, local->op_errno,
+                                           &local->cont.inode_wfop.prebuf,
+                                           &local->cont.inode_wfop.postbuf,
+                                           xdata);
+		if (!stub) {
+			AFR_STACK_UNWIND (fsync, frame, -1, ENOMEM, 0, 0, 0);
+			return 0;
+		}
+
+		/* If no new unstable writes happened between the
+		   time we cleared the unstable write witness flag in afr_fsync
+		   and now, calling afr_delayed_changelog_wake_up() should
+		   wake up and skip over the fsync phase and go straight to
+		   afr_changelog_post_op_now()
+		*/
+		afr_delayed_changelog_wake_resume (this, local->fd, stub);
+        }
+
+        return 0;
+}
+
+
+int
+afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd,
+           int32_t datasync, dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local = NULL;
+        int ret = -1;
+        int i = 0;
+        int32_t call_count = 0;
+        int32_t op_errno = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        call_count = local->call_count;
+
+        local->fd             = fd_ref (fd);
+
+	if (afr_fd_has_witnessed_unstable_write (this, fd)) {
+		/* don't care. we only wanted to CLEAR the bit */
+	}
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND_COOKIE (frame, afr_fsync_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->fsync,
+                                           fd, datasync, xdata);
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
+/* }}} */
+
+/* {{{ fsync */
+
+int32_t
+afr_fsyncdir_cbk (call_frame_t *frame, void *cookie,
+                  xlator_t *this, int32_t op_ret, int32_t op_errno,
+                  dict_t *xdata)
+{
+        afr_local_t *local = NULL;
+        int call_count = -1;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == 0)
+                        local->op_ret = 0;
+
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0)
+                AFR_STACK_UNWIND (fsyncdir, frame, local->op_ret,
+                                  local->op_errno, xdata);
+
+        return 0;
+}
+
+
+int32_t
+afr_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
+              int32_t datasync, dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local = NULL;
+        int ret = -1;
+        int i = 0;
+        int32_t call_count = 0;
+        int32_t op_errno = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        call_count = local->call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND (frame, afr_fsyncdir_cbk,
+                                    priv->children[i],
+                                    priv->children[i]->fops->fsyncdir,
+                                    fd, datasync, xdata);
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (fsyncdir, frame, -1, op_errno, NULL);
+        return 0;
+}
+
+/* }}} */
+
+/* {{{ xattrop */
+
+int32_t
+afr_xattrop_cbk (call_frame_t *frame, void *cookie,
+                 xlator_t *this, int32_t op_ret, int32_t op_errno,
+                 dict_t *xattr, dict_t *xdata)
+{
+        afr_local_t *local = NULL;
+        int call_count = -1;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == 0) {
+                        if (!local->cont.xattrop.xattr)
+                                local->cont.xattrop.xattr = dict_ref (xattr);
+                        local->op_ret = 0;
+                }
+
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0)
+                AFR_STACK_UNWIND (xattrop, frame, local->op_ret, local->op_errno,
+                local->cont.xattrop.xattr, xdata);
+
+        return 0;
+}
+
+
+int32_t
+afr_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
+             gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local  = NULL;
+        int ret = -1;
+        int i = 0;
+        int32_t call_count = 0;
+        int32_t op_errno = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        call_count = local->call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND (frame, afr_xattrop_cbk,
+                                    priv->children[i],
+                                    priv->children[i]->fops->xattrop,
+                                    loc, optype, xattr, xdata);
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (xattrop, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
+
+/* }}} */
+
+/* {{{ fxattrop */
+
+int32_t
+afr_fxattrop_cbk (call_frame_t *frame, void *cookie,
+                  xlator_t *this, int32_t op_ret, int32_t op_errno,
+                  dict_t *xattr, dict_t *xdata)
+{
+        afr_local_t *local = NULL;
+
+        int call_count = -1;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == 0) {
+                        if (!local->cont.fxattrop.xattr)
+                                local->cont.fxattrop.xattr = dict_ref (xattr);
+
+                        local->op_ret = 0;
+                }
+
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0)
+                AFR_STACK_UNWIND (fxattrop, frame, local->op_ret, local->op_errno,
+                                  local->cont.fxattrop.xattr, xdata);
+
+        return 0;
+}
+
+
+int32_t
+afr_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
+              gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local  = NULL;
+        int ret = -1;
+        int i = 0;
+        int32_t call_count = 0;
+        int32_t op_errno = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        call_count = local->call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND (frame, afr_fxattrop_cbk,
+                                    priv->children[i],
+                                    priv->children[i]->fops->fxattrop,
+                                    fd, optype, xattr, xdata);
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (fxattrop, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
+
+/* }}} */
+
+
+int32_t
+afr_inodelk_cbk (call_frame_t *frame, void *cookie,
+                 xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+        afr_local_t *local = NULL;
+        int call_count = -1;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == 0)
+                        local->op_ret = 0;
+
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0)
+                AFR_STACK_UNWIND (inodelk, frame, local->op_ret,
+                                  local->op_errno, xdata);
+
+        return 0;
+}
+
+
+int32_t
+afr_inodelk (call_frame_t *frame, xlator_t *this,
+             const char *volume, loc_t *loc, int32_t cmd,
+             struct gf_flock *flock, dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local  = NULL;
+        int ret = -1;
+        int i = 0;
+        int32_t call_count = 0;
+        int32_t op_errno = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        call_count = local->call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND (frame, afr_inodelk_cbk,
+                                    priv->children[i],
+                                    priv->children[i]->fops->inodelk,
+                                    volume, loc, cmd, flock, xdata);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (inodelk, frame, -1, op_errno, NULL);
+        return 0;
+}
+
+
+int32_t
+afr_finodelk_cbk (call_frame_t *frame, void *cookie,
+                  xlator_t *this, int32_t op_ret, int32_t op_errno,
+                  dict_t *xdata)
+
+{
+        afr_local_t *local = NULL;
+        int call_count = -1;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == 0)
+                        local->op_ret = 0;
+
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0)
+                AFR_STACK_UNWIND (finodelk, frame, local->op_ret,
+                                  local->op_errno, xdata);
+
+        return 0;
+}
+
+
+int32_t
+afr_finodelk (call_frame_t *frame, xlator_t *this,
+              const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock,
+              dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local  = NULL;
+        int ret = -1;
+        int i = 0;
+        int32_t call_count = 0;
+        int32_t op_errno = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        call_count = local->call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND (frame, afr_finodelk_cbk,
+                                    priv->children[i],
+                                    priv->children[i]->fops->finodelk,
+                                    volume, fd, cmd, flock, xdata);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (finodelk, frame, -1, op_errno, NULL);
+        return 0;
+}
+
+
+int32_t
+afr_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_local_t *local = NULL;
+        int call_count = -1;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == 0)
+                        local->op_ret = 0;
+
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0)
+                AFR_STACK_UNWIND (entrylk, frame, local->op_ret,
+                                  local->op_errno, xdata);
+
+        return 0;
+}
+
+
+int32_t
+afr_entrylk (call_frame_t *frame, xlator_t *this,
+             const char *volume, loc_t *loc,
+             const char *basename, entrylk_cmd cmd, entrylk_type type,
+             dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local  = NULL;
+        int ret = -1;
+        int i = 0;
+        int32_t call_count = 0;
+        int32_t op_errno = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        call_count = local->call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND (frame, afr_entrylk_cbk,
+                                    priv->children[i],
+                                    priv->children[i]->fops->entrylk,
+                                    volume, loc, basename, cmd, type, xdata);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (entrylk, frame, -1, op_errno, NULL);
+        return 0;
+}
+
+
+
+int32_t
+afr_fentrylk_cbk (call_frame_t *frame, void *cookie,
+                  xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+        afr_local_t *local = NULL;
+        int call_count = -1;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == 0)
+                        local->op_ret = 0;
+
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0)
+                AFR_STACK_UNWIND (fentrylk, frame, local->op_ret,
+                                  local->op_errno, xdata);
+
+        return 0;
+}
+
+
+int32_t
+afr_fentrylk (call_frame_t *frame, xlator_t *this,
+              const char *volume, fd_t *fd,
+              const char *basename, entrylk_cmd cmd,
+              entrylk_type type, dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local  = NULL;
+        int ret = -1;
+        int i = 0;
+        int32_t call_count = 0;
+        int32_t op_errno = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        call_count = local->call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND (frame, afr_fentrylk_cbk,
+                                    priv->children[i],
+                                    priv->children[i]->fops->fentrylk,
+                                    volume, fd, basename, cmd, type, xdata);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (fentrylk, frame, -1, op_errno, NULL);
+        return 0;
+}
+
+int32_t
+afr_statfs_cbk (call_frame_t *frame, void *cookie,
+                xlator_t *this, int32_t op_ret, int32_t op_errno,
+                struct statvfs *statvfs, dict_t *xdata)
+{
+        afr_local_t *local = NULL;
+        int call_count = 0;
+
+        LOCK (&frame->lock);
+        {
+                local = frame->local;
+
+                if (op_ret == 0) {
+                        local->op_ret   = op_ret;
+
+                        if (local->cont.statfs.buf_set) {
+                                if (statvfs->f_bavail < local->cont.statfs.buf.f_bavail)
+                                        local->cont.statfs.buf = *statvfs;
+                        } else {
+                                local->cont.statfs.buf = *statvfs;
+                                local->cont.statfs.buf_set = 1;
+                        }
+                }
+
+                if (op_ret == -1)
+                        local->op_errno = op_errno;
+
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0)
+                AFR_STACK_UNWIND (statfs, frame, local->op_ret, local->op_errno,
+                                  &local->cont.statfs.buf, xdata);
+
+        return 0;
+}
+
+
+int32_t
+afr_statfs (call_frame_t *frame, xlator_t *this,
+            loc_t *loc, dict_t *xdata)
+{
+        afr_private_t *  priv        = NULL;
+        int              child_count = 0;
+        afr_local_t   *  local       = NULL;
+        int              i           = 0;
+        int              ret = -1;
+        int              call_count = 0;
+        int32_t          op_errno    = 0;
+
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+        VALIDATE_OR_GOTO (loc, out);
+
+        priv = this->private;
+        child_count = priv->child_count;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        call_count = local->call_count;
+
+        for (i = 0; i < child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND (frame, afr_statfs_cbk,
+                                    priv->children[i],
+                                    priv->children[i]->fops->statfs,
+                                    loc, xdata);
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (statfs, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
+
+
+int32_t
+afr_lk_unlock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                   dict_t *xdata)
+{
+        afr_local_t * local = NULL;
+        int call_count = -1;
+
+        local = frame->local;
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0)
+                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
+                                  lock, xdata);
+
+        return 0;
+}
+
+
+int32_t
+afr_lk_unlock (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   * local = NULL;
+        afr_private_t * priv  = NULL;
+        int i = 0;
+        int call_count = 0;
+
+        local = frame->local;
+        priv  = this->private;
+
+        call_count = afr_locked_nodes_count (local->cont.lk.locked_nodes,
+                                             priv->child_count);
+
+        if (call_count == 0) {
+                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
+                                  &local->cont.lk.ret_flock, NULL);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        local->cont.lk.user_flock.l_type = F_UNLCK;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->cont.lk.locked_nodes[i]) {
+                        STACK_WIND (frame, afr_lk_unlock_cbk,
+                                    priv->children[i],
+                                    priv->children[i]->fops->lk,
+                                    local->fd, F_SETLK,
+                                    &local->cont.lk.user_flock, NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
+
+
+int32_t
+afr_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+            int32_t op_ret, int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int child_index = -1;
+/*        int            ret  = 0; */
+
+
+        local = frame->local;
+        priv  = this->private;
+
+        child_index = (long) cookie;
+
+        if (!child_went_down (op_ret, op_errno) && (op_ret == -1)) {
+                local->op_ret   = -1;
+                local->op_errno = op_errno;
+
+                afr_lk_unlock (frame, this);
+                return 0;
+        }
+
+        if (op_ret == 0) {
+                local->op_ret        = 0;
+                local->op_errno      = 0;
+                local->cont.lk.locked_nodes[child_index] = 1;
+                local->cont.lk.ret_flock = *lock;
+        }
+
+        child_index++;
+
+        if (child_index < priv->child_count) {
+                STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) child_index,
+                                   priv->children[child_index],
+                                   priv->children[child_index]->fops->lk,
+                                   local->fd, local->cont.lk.cmd,
+                                   &local->cont.lk.user_flock, xdata);
+        } else if (local->op_ret == -1) {
+                /* all nodes have gone down */
+
+                AFR_STACK_UNWIND (lk, frame, -1, ENOTCONN,
+                                  &local->cont.lk.ret_flock, NULL);
+        } else {
+                /* locking has succeeded on all nodes that are up */
+
+                /* temporarily
+                   ret = afr_mark_locked_nodes (this, local->fd,
+                   local->cont.lk.locked_nodes);
+                   if (ret)
+                   gf_log (this->name, GF_LOG_DEBUG,
+                   "Could not save locked nodes info in fdctx");
+
+                   ret = afr_save_locked_fd (this, local->fd);
+                   if (ret)
+                   gf_log (this->name, GF_LOG_DEBUG,
+                   "Could not save locked fd");
+
+                */
+                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
+                                  &local->cont.lk.ret_flock, NULL);
+        }
+
+        return 0;
+}
+
+
+int
+afr_lk (call_frame_t *frame, xlator_t *this,
+        fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local = NULL;
+        int i = 0;
+        int32_t op_errno = 0;
+        int     ret      = -1;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        local->cont.lk.locked_nodes = GF_CALLOC (priv->child_count,
+                                                 sizeof (*local->cont.lk.locked_nodes),
+                                                 gf_afr_mt_char);
+
+        if (!local->cont.lk.locked_nodes) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        local->fd            = fd_ref (fd);
+        local->cont.lk.cmd   = cmd;
+        local->cont.lk.user_flock = *flock;
+        local->cont.lk.ret_flock = *flock;
+
+        STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) 0,
+                           priv->children[i],
+                           priv->children[i]->fops->lk,
+                           fd, cmd, flock, xdata);
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (lk, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
+
+int
+afr_forget (xlator_t *this, inode_t *inode)
+{
+        uint64_t        ctx_addr = 0;
+        afr_inode_ctx_t *ctx     = NULL;
+
+        inode_ctx_get (inode, this, &ctx_addr);
+
+        if (!ctx_addr)
+                goto out;
+
+        ctx = (afr_inode_ctx_t *)(long)ctx_addr;
+        GF_FREE (ctx->fresh_children);
+        GF_FREE (ctx);
+out:
+        return 0;
+}
+
+int
+afr_priv_dump (xlator_t *this)
+{
+        afr_private_t *priv = NULL;
+        char  key_prefix[GF_DUMP_MAX_BUF_LEN];
+        char  key[GF_DUMP_MAX_BUF_LEN];
+        int   i = 0;
+
+
+        GF_ASSERT (this);
+        priv = this->private;
+
+        GF_ASSERT (priv);
+        snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s", this->type, this->name);
+        gf_proc_dump_add_section(key_prefix);
+        gf_proc_dump_write("child_count", "%u", priv->child_count);
+        gf_proc_dump_write("read_child_rr", "%u", priv->read_child_rr);
+        for (i = 0; i < priv->child_count; i++) {
+                sprintf (key, "child_up[%d]", i);
+                gf_proc_dump_write(key, "%d", priv->child_up[i]);
+                sprintf (key, "pending_key[%d]", i);
+                gf_proc_dump_write(key, "%s", priv->pending_key[i]);
+        }
+        gf_proc_dump_write("data_self_heal", "%s", priv->data_self_heal);
+        gf_proc_dump_write("metadata_self_heal", "%d", priv->metadata_self_heal);
+        gf_proc_dump_write("entry_self_heal", "%d", priv->entry_self_heal);
+        gf_proc_dump_write("data_change_log", "%d", priv->data_change_log);
+        gf_proc_dump_write("metadata_change_log", "%d", priv->metadata_change_log);
+        gf_proc_dump_write("entry-change_log", "%d", priv->entry_change_log);
+        gf_proc_dump_write("read_child", "%d", priv->read_child);
+        gf_proc_dump_write("favorite_child", "%d", priv->favorite_child);
+        gf_proc_dump_write("wait_count", "%u", priv->wait_count);
+
+        return 0;
+}
+
+
+/**
+ * find_child_index - find the child's index in the array of subvolumes
+ * @this: AFR
+ * @child: child
+ */
+
+static int
+find_child_index (xlator_t *this, xlator_t *child)
+{
+        afr_private_t *priv = NULL;
+        int i = -1;
+
+        priv = this->private;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if ((xlator_t *) child == priv->children[i])
+                        break;
+        }
+
+        return i;
+}
+
+int32_t
+afr_notify (xlator_t *this, int32_t event,
+            void *data, void *data2)
+{
+        afr_private_t   *priv               = NULL;
+        int             i                   = -1;
+        int             up_children         = 0;
+        int             down_children       = 0;
+        int             propagate           = 0;
+        int             had_heard_from_all  = 0;
+        int             have_heard_from_all = 0;
+        int             idx                 = -1;
+        int             ret                 = -1;
+        int             call_psh            = 0;
+        int             up_child            = AFR_ALL_CHILDREN;
+        dict_t          *input              = NULL;
+        dict_t          *output             = NULL;
+
+        priv = this->private;
+
+        if (!priv)
+                return 0;
+
+        /*
+         * We need to reset this in case children come up in "staggered"
+         * fashion, so that we discover a late-arriving local subvolume.  Note
+         * that we could end up issuing N lookups to the first subvolume, and
+         * O(N^2) overall, but N is small for AFR so it shouldn't be an issue.
+         */
+        priv->did_discovery = _gf_false;
+
+        had_heard_from_all = 1;
+        for (i = 0; i < priv->child_count; i++) {
+                if (!priv->last_event[i]) {
+                        had_heard_from_all = 0;
+                }
+        }
+
+        /* parent xlators dont need to know about every child_up, child_down
+         * because of afr ha. If all subvolumes go down, child_down has
+         * to be triggered. In that state when 1 subvolume comes up child_up
+         * needs to be triggered. dht optimizes revalidate lookup by sending
+         * it only to one of its subvolumes. When child up/down happens
+         * for afr's subvolumes dht should be notified by child_modified. The
+         * subsequent revalidate lookup happens on all the dht's subvolumes
+         * which triggers afr self-heals if any.
+         */
+        idx = find_child_index (this, data);
+        if (idx < 0) {
+                gf_log (this->name, GF_LOG_ERROR, "Received child_up "
+                        "from invalid subvolume");
+                goto out;
+        }
+
+        switch (event) {
+        case GF_EVENT_CHILD_UP:
+                LOCK (&priv->lock);
+                {
+                        /*
+                         * This only really counts if the child was never up
+                         * (value = -1) or had been down (value = 0).  See
+                         * comment at GF_EVENT_CHILD_DOWN for a more detailed
+                         * explanation.
+                         */
+                        if (priv->child_up[idx] != 1) {
+                                priv->up_count++;
+                        }
+                        priv->child_up[idx] = 1;
+
+                        call_psh = 1;
+                        up_child = idx;
+                        for (i = 0; i < priv->child_count; i++)
+                                if (priv->child_up[i] == 1)
+                                        up_children++;
+                        if (up_children == 1) {
+                                gf_log (this->name, GF_LOG_INFO,
+                                        "Subvolume '%s' came back up; "
+                                        "going online.", ((xlator_t *)data)->name);
+                        } else {
+                                event = GF_EVENT_CHILD_MODIFIED;
+                        }
+
+                        priv->last_event[idx] = event;
+                }
+                UNLOCK (&priv->lock);
+
+                break;
+
+        case GF_EVENT_CHILD_DOWN:
+                LOCK (&priv->lock);
+                {
+                        /*
+                         * If a brick is down when we start, we'll get a
+                         * CHILD_DOWN to indicate its initial state.  There
+                         * was never a CHILD_UP in this case, so if we
+                         * increment "down_count" the difference between than
+                         * and "up_count" will no longer be the number of
+                         * children that are currently up.  This has serious
+                         * implications e.g. for quorum enforcement, so we
+                         * don't increment these values unless the event
+                         * represents an actual state transition between "up"
+                         * (value = 1) and anything else.
+                         */
+                        if (priv->child_up[idx] == 1) {
+                                priv->down_count++;
+                        }
+                        priv->child_up[idx] = 0;
+
+                        for (i = 0; i < priv->child_count; i++)
+                                if (priv->child_up[i] == 0)
+                                        down_children++;
+                        if (down_children == priv->child_count) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "All subvolumes are down. Going offline "
+                                        "until atleast one of them comes back up.");
+                        } else {
+                                event = GF_EVENT_CHILD_MODIFIED;
+                        }
+
+                        priv->last_event[idx] = event;
+                }
+                UNLOCK (&priv->lock);
+
+                break;
+
+        case GF_EVENT_CHILD_CONNECTING:
+                LOCK (&priv->lock);
+                {
+                        priv->last_event[idx] = event;
+                }
+                UNLOCK (&priv->lock);
+
+                break;
+
+        case GF_EVENT_TRANSLATOR_OP:
+                input = data;
+                output = data2;
+                ret = afr_xl_op (this, input, output);
+                goto out;
+                break;
+
+        default:
+                propagate = 1;
+                break;
+        }
+
+        /* have all subvolumes reported status once by now? */
+        have_heard_from_all = 1;
+        for (i = 0; i < priv->child_count; i++) {
+                if (!priv->last_event[i])
+                        have_heard_from_all = 0;
+        }
+
+        /* if all subvols have reported status, no need to hide anything
+           or wait for anything else. Just propagate blindly */
+        if (have_heard_from_all)
+                propagate = 1;
+
+        if (!had_heard_from_all && have_heard_from_all) {
+                /* This is the first event which completes aggregation
+                   of events from all subvolumes. If at least one subvol
+                   had come up, propagate CHILD_UP, but only this time
+                */
+                event = GF_EVENT_CHILD_DOWN;
+
+                LOCK (&priv->lock);
+                {
+                        up_children = afr_up_children_count (priv->child_up,
+                                                             priv->child_count);
+                        for (i = 0; i < priv->child_count; i++) {
+                                if (priv->last_event[i] == GF_EVENT_CHILD_UP) {
+                                        event = GF_EVENT_CHILD_UP;
+                                        break;
+                                }
+
+                                if (priv->last_event[i] ==
+                                                GF_EVENT_CHILD_CONNECTING) {
+                                        event = GF_EVENT_CHILD_CONNECTING;
+                                        /* continue to check other events for CHILD_UP */
+                                }
+                        }
+                }
+                UNLOCK (&priv->lock);
+        }
+
+        ret = 0;
+        if (propagate)
+                ret = default_notify (this, event, data);
+        if (call_psh && priv->shd.iamshd)
+                afr_proactive_self_heal ((void*) (long) up_child);
+
+out:
+        return ret;
+}
+
+int
+afr_first_up_child (unsigned char *child_up, size_t child_count)
+{
+        int         ret      = -1;
+        int         i        = 0;
+
+        GF_ASSERT (child_up);
+
+        for (i = 0; i < child_count; i++) {
+                if (child_up[i]) {
+                        ret = i;
+                        break;
+                }
+        }
+
+        return ret;
+}
+
+int
+afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno)
+{
+        int     ret = -1;
+
+        local->op_ret = -1;
+        local->op_errno = EUCLEAN;
+
+        local->child_up = GF_CALLOC (priv->child_count,
+                                     sizeof (*local->child_up),
+                                     gf_afr_mt_char);
+        if (!local->child_up) {
+                if (op_errno)
+                        *op_errno = ENOMEM;
+                goto out;
+        }
+
+        memcpy (local->child_up, priv->child_up,
+                sizeof (*local->child_up) * priv->child_count);
+        local->call_count = afr_up_children_count (local->child_up,
+                                                   priv->child_count);
+        if (local->call_count == 0) {
+                gf_log (THIS->name, GF_LOG_INFO, "no subvolumes up");
+                if (op_errno)
+                        *op_errno = ENOTCONN;
+                goto out;
+        }
+
+        local->child_errno = GF_CALLOC (priv->child_count,
+                                        sizeof (*local->child_errno),
+                                        gf_afr_mt_int32_t);
+        if (!local->child_errno) {
+                if (op_errno)
+                        *op_errno = ENOMEM;
+                goto out;
+        }
+
+        local->transaction.postop_piggybacked = GF_CALLOC (priv->child_count,
+							   sizeof (int),
+							   gf_afr_mt_int32_t);
+        if (!local->transaction.postop_piggybacked) {
+                if (op_errno)
+                        *op_errno = ENOMEM;
+                goto out;
+        }
+
+	local->append_write = _gf_false;
+
+        ret = 0;
+out:
+        return ret;
+}
+
+int
+afr_internal_lock_init (afr_internal_lock_t *lk, size_t child_count,
+                        transaction_lk_type_t lk_type)
+{
+        int             ret = -ENOMEM;
+
+        lk->locked_nodes = GF_CALLOC (sizeof (*lk->locked_nodes),
+                                      child_count, gf_afr_mt_char);
+        if (NULL == lk->locked_nodes)
+                goto out;
+
+        lk->lower_locked_nodes = GF_CALLOC (sizeof (*lk->lower_locked_nodes),
+                                            child_count, gf_afr_mt_char);
+        if (NULL == lk->lower_locked_nodes)
+                goto out;
+
+        lk->lock_op_ret   = -1;
+        lk->lock_op_errno = EUCLEAN;
+        lk->transaction_lk_type = lk_type;
+
+        ret = 0;
+out:
+        return ret;
+}
+
+void
+afr_matrix_cleanup (int32_t **matrix, unsigned int m)
+{
+        int             i         = 0;
+
+        if (!matrix)
+                goto out;
+        for (i = 0; i < m; i++) {
+                GF_FREE (matrix[i]);
+        }
+
+        GF_FREE (matrix);
+out:
+        return;
+}
+
+int32_t**
+afr_matrix_create (unsigned int m, unsigned int n)
+{
+        int32_t         **matrix = NULL;
+        int             i       = 0;
+
+        matrix = GF_CALLOC (sizeof (*matrix), m, gf_afr_mt_int32_t);
+        if (!matrix)
+                goto out;
+
+        for (i = 0; i < m; i++) {
+                matrix[i] = GF_CALLOC (sizeof (*matrix[i]), n,
+                                       gf_afr_mt_int32_t);
+                if (!matrix[i])
+                        goto out;
+        }
+        return matrix;
+out:
+        afr_matrix_cleanup (matrix, m);
+        return NULL;
+}
+
+int
+afr_inodelk_init (afr_inodelk_t *lk, char *dom, size_t child_count)
+{
+        int             ret = -ENOMEM;
+
+        lk->domain = dom;
+        lk->locked_nodes = GF_CALLOC (sizeof (*lk->locked_nodes),
+                                      child_count, gf_afr_mt_char);
+        if (NULL == lk->locked_nodes)
+                goto out;
+        ret = 0;
+out:
+        return ret;
+}
+
+int
+afr_transaction_local_init (afr_local_t *local, xlator_t *this)
+{
+        int            child_up_count = 0;
+        int            ret = -ENOMEM;
+        afr_private_t *priv = NULL;
+
+        priv = this->private;
+        ret = afr_internal_lock_init (&local->internal_lock, priv->child_count,
+                                      AFR_TRANSACTION_LK);
+        if (ret < 0)
+                goto out;
+
+        if ((local->transaction.type == AFR_DATA_TRANSACTION) ||
+            (local->transaction.type == AFR_METADATA_TRANSACTION)) {
+                ret = afr_inodelk_init (&local->internal_lock.inodelk[0],
+                                        this->name, priv->child_count);
+                if (ret < 0)
+                        goto out;
+        }
+
+        ret = -ENOMEM;
+        child_up_count = afr_up_children_count (local->child_up,
+                                                priv->child_count);
+        if (priv->optimistic_change_log && child_up_count == priv->child_count)
+                local->optimistic_change_log = 1;
+
+        local->first_up_child = afr_first_up_child (local->child_up,
+                                                    priv->child_count);
+
+        local->transaction.eager_lock =
+                GF_CALLOC (sizeof (*local->transaction.eager_lock),
+                           priv->child_count,
+                           gf_afr_mt_int32_t);
+
+        if (!local->transaction.eager_lock)
+                goto out;
+
+        local->fresh_children = afr_children_create (priv->child_count);
+        if (!local->fresh_children)
+                goto out;
+
+        local->transaction.pre_op = GF_CALLOC (sizeof (*local->transaction.pre_op),
+                                               priv->child_count,
+                                               gf_afr_mt_char);
+        if (!local->transaction.pre_op)
+                goto out;
+
+        local->pending = afr_matrix_create (priv->child_count,
+                                            AFR_NUM_CHANGE_LOGS);
+        if (!local->pending)
+                goto out;
+
+        local->transaction.txn_changelog = afr_matrix_create (priv->child_count,
+                                                           AFR_NUM_CHANGE_LOGS);
+        if (!local->transaction.txn_changelog)
+                goto out;
+
+	INIT_LIST_HEAD (&local->transaction.eager_locked);
+
+        ret = 0;
+out:
+        return ret;
+}
+
+void
+afr_reset_children (int32_t *fresh_children, int32_t child_count)
+{
+        unsigned int i = 0;
+        for (i = 0; i < child_count; i++)
+                fresh_children[i] = -1;
+}
+
+int32_t*
+afr_children_create (int32_t child_count)
+{
+        int32_t           *children = NULL;
+        int               i               = 0;
+
+        GF_ASSERT (child_count > 0);
+
+        children = GF_CALLOC (child_count, sizeof (*children),
+                              gf_afr_mt_int32_t);
+        if (NULL == children)
+                goto out;
+        for (i = 0; i < child_count; i++)
+                children[i] = -1;
+out:
+        return children;
+}
+
+void
+afr_children_add_child (int32_t *children, int32_t child,
+                        int32_t child_count)
+{
+        gf_boolean_t child_found = _gf_false;
+        int          i               = 0;
+
+        for (i = 0; i < child_count; i++) {
+                if (children[i] == -1)
+                        break;
+                if (children[i] == child) {
+                        child_found = _gf_true;
+                        break;
+                }
+        }
+
+        if (!child_found) {
+                GF_ASSERT (i < child_count);
+                children[i] = child;
+        }
+}
+
+void
+afr_children_rm_child (int32_t *children, int32_t child, int32_t child_count)
+{
+        int          i = 0;
+
+        GF_ASSERT ((child >= 0) && (child < child_count));
+        for (i = 0; i < child_count; i++) {
+                if (children[i] == -1)
+                        break;
+                if (children[i] == child) {
+                        if (i != (child_count - 1))
+                                memmove (children + i, children + i + 1,
+                                         sizeof (*children)*(child_count - i - 1));
+                        children[child_count - 1] = -1;
+                        break;
+                }
+        }
+}
+
+int
+afr_get_children_count (int32_t *children, unsigned int child_count)
+{
+        int count = 0;
+        int i = 0;
+
+        for (i = 0; i < child_count; i++) {
+                if (children[i] == -1)
+                        break;
+                count++;
+        }
+        return count;
+}
+
+void
+afr_set_low_priority (call_frame_t *frame)
+{
+        frame->root->pid = LOW_PRIO_PROC_PID;
+}
+
+int
+afr_child_fd_ctx_set (xlator_t *this, fd_t *fd, int32_t child,
+                      int flags)
+{
+        int             ret = 0;
+        uint64_t        ctx = 0;
+        afr_fd_ctx_t    *fd_ctx      = NULL;
+
+        GF_ASSERT (fd && fd->inode);
+        ret = afr_fd_ctx_set (this, fd);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "could not set fd ctx for fd=%p", fd);
+                goto out;
+        }
+
+        ret = fd_ctx_get (fd, this, &ctx);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "could not get fd ctx for fd=%p", fd);
+                goto out;
+        }
+
+        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+        fd_ctx->opened_on[child] = AFR_FD_OPENED;
+        if (!IA_ISDIR (fd->inode->ia_type)) {
+                fd_ctx->flags            = flags;
+        }
+        ret = 0;
+out:
+        return ret;
+}
+
+gf_boolean_t
+afr_have_quorum (char *logname, afr_private_t *priv)
+{
+        unsigned int        quorum = 0;
+
+        GF_VALIDATE_OR_GOTO(logname,priv,out);
+
+        quorum = priv->quorum_count;
+        if (quorum != AFR_QUORUM_AUTO) {
+                return (priv->up_count >= (priv->down_count + quorum));
+        }
+
+        quorum = priv->child_count / 2 + 1;
+        if (priv->up_count >= (priv->down_count + quorum)) {
+                return _gf_true;
+        }
+
+        /*
+         * Special case for even numbers of nodes: if we have exactly half
+         * and that includes the first ("senior-most") node, then that counts
+         * as quorum even if it wouldn't otherwise.  This supports e.g. N=2
+         * while preserving the critical property that there can only be one
+         * such group.
+         */
+        if ((priv->child_count % 2) == 0) {
+                quorum = priv->child_count / 2;
+                if (priv->up_count >= (priv->down_count + quorum)) {
+                        if (priv->child_up[0]) {
+                                return _gf_true;
+                        }
+                }
+        }
+
+out:
+        return _gf_false;
+}
+
+void
+afr_priv_destroy (afr_private_t *priv)
+{
+        int            i           = 0;
+
+        if (!priv)
+                goto out;
+        inode_unref (priv->root_inode);
+        GF_FREE (priv->shd.pos);
+        GF_FREE (priv->shd.pending);
+        GF_FREE (priv->shd.inprogress);
+//        for (i = 0; i < priv->child_count; i++)
+//                if (priv->shd.timer && priv->shd.timer[i])
+//                        gf_timer_call_cancel (this->ctx, priv->shd.timer[i]);
+        GF_FREE (priv->shd.timer);
+
+        if (priv->shd.healed)
+                eh_destroy (priv->shd.healed);
+
+        if (priv->shd.heal_failed)
+                eh_destroy (priv->shd.heal_failed);
+
+        if (priv->shd.split_brain)
+                eh_destroy (priv->shd.split_brain);
+
+        for (i = 0; i < priv->child_count; i++)
+        {
+                if (priv->shd.statistics[i])
+                        eh_destroy (priv->shd.statistics[i]);
+        }
+
+        GF_FREE (priv->shd.statistics);
+
+        GF_FREE (priv->shd.crawl_events);
+
+        GF_FREE (priv->last_event);
+        if (priv->pending_key) {
+                for (i = 0; i < priv->child_count; i++)
+                        GF_FREE (priv->pending_key[i]);
+        }
+        GF_FREE (priv->pending_key);
+        GF_FREE (priv->children);
+        GF_FREE (priv->child_up);
+        LOCK_DESTROY (&priv->lock);
+        LOCK_DESTROY (&priv->read_child_lock);
+        pthread_mutex_destroy (&priv->mutex);
+        GF_FREE (priv);
+out:
+        return;
+}
+
+int
+xlator_subvolume_count (xlator_t *this)
+{
+        int i = 0;
+        xlator_list_t *list = NULL;
+
+        for (list = this->children; list; list = list->next)
+                i++;
+        return i;
+}
+
+inline gf_boolean_t
+afr_is_errno_set (int *child_errno, int child)
+{
+        return child_errno[child];
+}
+
+inline gf_boolean_t
+afr_is_errno_unset (int *child_errno, int child)
+{
+        return !afr_is_errno_set (child_errno, child);
+}
+
+void
+afr_prepare_new_entry_pending_matrix (int32_t **pending,
+                                      gf_boolean_t (*is_pending) (int *, int),
+                                      int *ctx, struct iatt *buf,
+                                      unsigned int child_count)
+{
+        int midx = 0;
+        int idx  = 0;
+        int i    = 0;
+
+        midx = afr_index_for_transaction_type (AFR_METADATA_TRANSACTION);
+        if (IA_ISDIR (buf->ia_type))
+                idx = afr_index_for_transaction_type (AFR_ENTRY_TRANSACTION);
+        else if (IA_ISREG (buf->ia_type))
+                idx = afr_index_for_transaction_type (AFR_DATA_TRANSACTION);
+        else
+                idx = -1;
+        for (i = 0; i < child_count; i++) {
+                if (is_pending (ctx, i)) {
+                        pending[i][midx] = hton32 (1);
+                        if (idx == -1)
+                                continue;
+                        pending[i][idx] = hton32 (1);
+                }
+        }
+}
+
+gf_boolean_t
+afr_is_fd_fixable (fd_t *fd)
+{
+        if (!fd || !fd->inode)
+                return _gf_false;
+        else if (fd_is_anonymous (fd))
+                return _gf_false;
+        else if (uuid_is_null (fd->inode->gfid))
+                return _gf_false;
+
+        return _gf_true;
+}
+
+void
+afr_handle_open_fd_count (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        inode_t         *inode = NULL;
+        afr_inode_ctx_t *ctx   = NULL;
+
+        local = frame->local;
+
+        if (local->fd)
+                inode = local->fd->inode;
+        else
+                inode = local->loc.inode;
+
+        if (!inode)
+                return;
+
+        LOCK (&inode->lock);
+        {
+                ctx = __afr_inode_ctx_get (inode, this);
+                ctx->open_fd_count = local->open_fd_count;
+        }
+        UNLOCK (&inode->lock);
+}
+
+int
+afr_initialise_statistics (xlator_t *this)
+{
+        afr_private_t       *priv = NULL;
+        int                 ret = -1;
+        int                 i = 0;
+        int                 child_count = 0;
+        eh_t                *stats_per_brick = NULL;
+        shd_crawl_event_t   ***shd_crawl_events = NULL;
+        priv = this->private;
+
+        priv->shd.statistics = GF_CALLOC (sizeof(eh_t *), priv->child_count,
+                                          gf_common_mt_eh_t);
+        if (!priv->shd.statistics) {
+                ret = -1;
+                goto out;
+        }
+        child_count = priv->child_count;
+        for (i=0; i < child_count ; i++) {
+                stats_per_brick = eh_new (AFR_STATISTICS_HISTORY_SIZE,
+                                          _gf_false,
+                                          _destroy_crawl_event_data);
+                if (!stats_per_brick) {
+                        ret = -1;
+                        goto out;
+                }
+                priv->shd.statistics[i] = stats_per_brick;
+
+        }
+
+        shd_crawl_events = (shd_crawl_event_t***)(&priv->shd.crawl_events);
+        *shd_crawl_events  = GF_CALLOC (sizeof(shd_crawl_event_t*),
+                                        priv->child_count,
+                                        gf_afr_mt_shd_crawl_event_t);
+
+        if (!priv->shd.crawl_events) {
+                ret = -1;
+                goto out;
+        }
+        ret = 0;
+out:
+        return ret;
+
+}
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index fd3fc1ba7..689dd84e6 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 
@@ -24,6 +15,7 @@
 #include <sys/time.h>
 #include <stdlib.h>
 #include <signal.h>
+#include <string.h>
 
 #ifndef _CONFIG_H
 #define _CONFIG_H
@@ -42,304 +34,512 @@
 #include "common-utils.h"
 #include "compat-errno.h"
 #include "compat.h"
+#include "checksum.h"
 
 #include "afr.h"
+#include "afr-self-heal.h"
+#include "afr-self-heal-common.h"
 
+int
+afr_examine_dir_sh_unwind (call_frame_t *frame, xlator_t *this, int32_t op_ret,
+                           int32_t op_errno, int32_t sh_failed)
+{
+        afr_local_t *local  = NULL;
 
-int32_t
-afr_opendir_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno,
-		 fd_t *fd)
+        local = frame->local;
+
+        afr_set_opendir_done (this, local->fd->inode);
+
+        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
+                          local->op_errno, local->fd, NULL);
+
+        return 0;
+}
+
+
+gf_boolean_t
+__checksums_differ (uint32_t *checksum, int child_count,
+                    unsigned char *child_up)
 {
-	afr_local_t * local  = NULL;
+        int          ret            = _gf_false;
+        int          i              = 0;
+        uint32_t     cksum          = 0;
+        gf_boolean_t activate_check = _gf_false;
+
+        for (i = 0; i < child_count; i++) {
+                if (!child_up[i])
+                        continue;
+                if (_gf_false == activate_check) {
+                        cksum          = checksum[i];
+                        activate_check = _gf_true;
+                        continue;
+                }
+
+                if (cksum != checksum[i]) {
+                        ret = _gf_true;
+                        break;
+                }
+
+                cksum = checksum[i];
+        }
+
+        return ret;
+}
 
-	int call_count = -1;
 
-	LOCK (&frame->lock);
-	{
-		local = frame->local;
+int32_t
+afr_examine_dir_readdir_cbk (call_frame_t *frame, void *cookie,
+                             xlator_t *this, int32_t op_ret, int32_t op_errno,
+                             gf_dirent_t *entries, dict_t *xdata)
+{
+        afr_private_t *   priv        = NULL;
+        afr_local_t *     local       = NULL;
+        afr_self_heal_t * sh          = NULL;
+        gf_dirent_t *     entry       = NULL;
+        gf_dirent_t *     tmp         = NULL;
+        char              *reason     = NULL;
+        int               child_index = 0;
+        uint32_t          entry_cksum = 0;
+        int               call_count  = 0;
+        off_t             last_offset = 0;
+        inode_t           *inode      = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+        sh    = &local->self_heal;
+        inode = local->fd->inode;
+
+        child_index = (long) cookie;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "%s: failed to do opendir on %s",
+                        local->loc.path, priv->children[child_index]->name);
+                local->op_ret = -1;
+                local->op_ret = op_errno;
+                goto out;
+        }
+
+        if (op_ret == 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "%s: no entries found in %s",
+                        local->loc.path, priv->children[child_index]->name);
+                goto out;
+        }
+
+        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
+                entry_cksum = gf_rsync_weak_checksum ((unsigned char *)entry->d_name,
+                                                      strlen (entry->d_name));
+                local->cont.opendir.checksum[child_index] ^= entry_cksum;
+        }
+
+        list_for_each_entry (entry, &entries->list, list) {
+                last_offset = entry->d_off;
+        }
+
+        /* read more entries */
+
+        STACK_WIND_COOKIE (frame, afr_examine_dir_readdir_cbk,
+                           (void *) (long) child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->readdir,
+                           local->fd, 131072, last_offset, NULL);
+
+        return 0;
+
+out:
+        call_count = afr_frame_return (frame);
 
-		if (op_ret == 0)
-			local->op_ret = 0;
+        if (call_count == 0) {
+                if (__checksums_differ (local->cont.opendir.checksum,
+                                        priv->child_count,
+                                        local->child_up)) {
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+                        sh->do_entry_self_heal  = _gf_true;
+                        sh->forced_merge          = _gf_true;
 
-	call_count = afr_frame_return (frame);
+                        reason = "checksums of directory differ";
+                        afr_launch_self_heal (frame, this, inode, _gf_false,
+                                              inode->ia_type, reason, NULL,
+                                              afr_examine_dir_sh_unwind);
+                } else {
+                        afr_set_opendir_done (this, inode);
 
-	if (call_count == 0) {
-		AFR_STACK_UNWIND (frame, local->op_ret,
-				  local->op_errno, local->fd);
-	}
+                        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
+                                          local->op_errno, local->fd, NULL);
+                }
+        }
 
-	return 0;
+        return 0;
 }
 
 
-int32_t 
-afr_opendir (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, fd_t *fd)
+int
+afr_examine_dir (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv        = NULL;
-	afr_local_t   * local       = NULL;
-
-	int             child_count = 0;
-	int             i           = 0;
+        afr_private_t * priv       = NULL;
+        afr_local_t *   local      = NULL;
+        int             i          = 0;
+        int             call_count = 0;
 
-	int ret = -1;
-	int call_count = -1;
+        local = frame->local;
+        priv  = this->private;
 
-	int32_t         op_ret   = -1;
-	int32_t         op_errno = 0;
+        local->cont.opendir.checksum = GF_CALLOC (priv->child_count,
+                                                  sizeof (*local->cont.opendir.checksum),
+                                                  gf_afr_mt_int32_t);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        call_count = afr_up_children_count (local->child_up, priv->child_count);
 
-	priv = this->private;
+        local->call_count = call_count;
 
-	child_count = priv->child_count;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND_COOKIE (frame, afr_examine_dir_readdir_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->readdir,
+                                           local->fd, 131072, 0, NULL);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+                        if (!--call_count)
+                                break;
+                }
+        }
 
-	frame->local = local;
-	local->fd    = fd_ref (fd);
+        return 0;
+}
 
-	call_count = local->call_count;
-	
-	for (i = 0; i < child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_opendir_cbk, 
-				    priv->children[i],
-				    priv->children[i]->fops->opendir,
-				    loc, fd);
 
-			if (!--call_count)
-				break;
-		}
-	}
+int32_t
+afr_opendir_cbk (call_frame_t *frame, void *cookie,
+                 xlator_t *this, int32_t op_ret, int32_t op_errno,
+                 fd_t *fd, dict_t *xdata)
+{
+        afr_private_t *priv              = NULL;
+        afr_local_t   *local             = NULL;
+        int32_t        up_children_count = 0;
+        int            ret               = -1;
+        int            call_count        = -1;
+        int32_t        child_index       = 0;
+
+        priv  = this->private;
+        local = frame->local;
+        child_index = (long) cookie;
+
+        up_children_count = afr_up_children_count (local->child_up,
+                                                   priv->child_count);
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret >= 0) {
+                        local->op_ret = op_ret;
+                        ret = afr_child_fd_ctx_set (this, fd, child_index, 0);
+                        if (ret) {
+                                local->op_ret = -1;
+                                local->op_errno = -ret;
+                                goto unlock;
+                        }
+                }
+
+                local->op_errno = op_errno;
+        }
+unlock:
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                if (local->op_ret != 0)
+                        goto out;
+
+                if (!afr_is_opendir_done (this, local->fd->inode) &&
+                    up_children_count > 1 && priv->entry_self_heal) {
+
+                        /*
+                         * This is the first opendir on this inode. We need
+                         * to check if the directory's entries are the same
+                         * on all subvolumes. This is needed in addition
+                         * to regular entry self-heal because the readdir
+                         * call is sent only to the first subvolume, and
+                         * thus files that exist only there will never be healed
+                         * otherwise (assuming changelog shows no anomalies).
+                         */
+
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "reading contents of directory %s looking for mismatch",
+                                local->loc.path);
+
+                        afr_examine_dir (frame, this);
+
+                } else {
+                        /* do the unwind */
+                        goto out;
+                }
+        }
+
+        return 0;
 
-	op_ret = 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, fd);
-	}
+        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
+                          local->op_errno, local->fd, NULL);
 
-	return 0;
+        return 0;
 }
 
 
-/**
- * Common algorithm for directory read calls:
- * 
- * - Try the fop on the first child that is up
- * - if we have failed due to ENOTCONN:
- *     try the next child
- *
- * Applicable to: readdir
- */
-
 int32_t
-afr_readdir_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno,
-		 gf_dirent_t *buf)
+afr_opendir (call_frame_t *frame, xlator_t *this,
+             loc_t *loc, fd_t *fd)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+        afr_private_t * priv        = NULL;
+        afr_local_t   * local       = NULL;
+        int             child_count = 0;
+        int             i           = 0;
+        int             ret         = -1;
+        int             call_count  = -1;
+        int32_t         op_errno    = 0;
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	priv     = this->private;
-	children = priv->children;
+        priv = this->private;
 
-	local = frame->local;
+        child_count = priv->child_count;
 
-	if (op_ret == -1) {
-		last_tried = local->cont.readdir.last_tried;
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-		this_try = ++local->cont.readdir.last_tried;
-		unwind = 0;
+        loc_copy (&local->loc, loc);
 
-		STACK_WIND (frame, afr_readdir_cbk,
-			    children[this_try],
-			    children[this_try]->fops->readdir,
-			    local->fd, local->cont.readdir.size,
-			    local->cont.readdir.offset);
-	}
+        local->fd    = fd_ref (fd);
 
-out:
-	if (unwind) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, buf);
-	}
+        call_count = local->call_count;
 
-	return 0;
-}
+        for (i = 0; i < child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND_COOKIE (frame, afr_opendir_cbk,
+                                           (void*) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->opendir,
+                                           loc, fd, NULL);
 
+                        if (!--call_count)
+                                break;
+                }
+        }
 
-int32_t
-afr_readdir (call_frame_t *frame, xlator_t *this,
-	     fd_t *fd, size_t size, off_t offset)
-{
-	afr_private_t * priv       = NULL;
-	xlator_t **     children   = NULL;
-	int             call_child = 0;
-	afr_local_t     *local     = NULL;
-
-	int ret = -1;
-
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv     = this->private;
-	children = priv->children;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-						
-	frame->local = local;
-
-	call_child = afr_first_up_child (priv);
-	if (call_child == -1) {
-		op_errno = ENOTCONN;
-		gf_log (this->name, GF_LOG_ERROR,
-			"no child is up :(");
-		goto out;
-	}
-
-	local->cont.readdir.last_tried = call_child;
-
-	local->fd                  = fd_ref (fd);
-	local->cont.readdir.size   = size;
-	local->cont.readdir.offset = offset;
-
-	STACK_WIND (frame, afr_readdir_cbk,
-		    children[call_child], children[call_child]->fops->readdir,
-		    fd, size, offset);
-
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
-	return 0;
+        if (ret < 0)
+                AFR_STACK_UNWIND (opendir, frame, -1, op_errno, fd, NULL);
+
+        return 0;
 }
 
 
-int32_t
-afr_getdents_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno,
-		  dir_entry_t *entry, int32_t count)
-{
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+/**
+ * Common algorithm for directory read calls:
+ *
+ * - Try the fop on the first child that is up
+ * - if we have failed due to ENOTCONN:
+ *     try the next child
+ *
+ * Applicable to: readdir
+ */
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
 
-	priv     = this->private;
-	children = priv->children;
+struct entry_name {
+        char *name;
+        struct list_head list;
+};
 
-	local = frame->local;
+static void
+afr_forget_entries (fd_t *fd)
+{
+        struct entry_name *entry  = NULL;
+        struct entry_name *tmp    = NULL;
+        int                ret    = 0;
+        uint64_t           ctx    = 0;
+        afr_fd_ctx_t      *fd_ctx = NULL;
+
+        ret = fd_ctx_get (fd, THIS, &ctx);
+        if (ret < 0) {
+                gf_log (THIS->name, GF_LOG_INFO,
+                        "could not get fd ctx for fd=%p", fd);
+                return;
+        }
+
+        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+
+        list_for_each_entry_safe (entry, tmp, &fd_ctx->entries, list) {
+                GF_FREE (entry->name);
+                list_del (&entry->list);
+                GF_FREE (entry);
+        }
+}
 
-	if (op_ret == -1) {
-		last_tried = local->cont.getdents.last_tried;
+static void
+afr_readdir_filter_trash_dir (gf_dirent_t *entries, fd_t *fd)
+{
+        gf_dirent_t *   entry       = NULL;
+        gf_dirent_t *   tmp         = NULL;
+
+        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
+                if (__is_root_gfid (fd->inode->gfid) &&
+                    !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) {
+                        list_del_init (&entry->list);
+                        GF_FREE (entry);
+                }
+        }
+}
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
+int32_t
+afr_readdir_cbk (call_frame_t *frame, void *cookie,
+                 xlator_t *this, int32_t op_ret, int32_t op_errno,
+                 gf_dirent_t *entries, dict_t *xdata)
+{
+        afr_local_t     *local = NULL;
 
-		this_try = ++local->cont.getdents.last_tried;
-		unwind = 0;
+        if (op_ret == -1)
+                goto out;
 
-		STACK_WIND (frame, afr_getdents_cbk,
-			    children[this_try],
-			    children[this_try]->fops->getdents,
-			    local->fd, local->cont.getdents.size,
-			    local->cont.getdents.offset, local->cont.getdents.flag);
-	}
+        local = frame->local;
+        afr_readdir_filter_trash_dir (entries, local->fd);
 
 out:
-	if (unwind) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, entry, count);
-	}
-
-	return 0;
+        AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries, NULL);
+        return 0;
 }
 
 
 int32_t
-afr_getdents (call_frame_t *frame, xlator_t *this,
-	      fd_t *fd, size_t size, off_t offset, int32_t flag)
+afr_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                  dict_t *xdata)
 {
-	afr_private_t * priv       = NULL;
-	xlator_t **     children   = NULL;
-	int             call_child = 0;
-	afr_local_t     *local     = NULL;
+        afr_local_t     *local = NULL;
 
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
+        if (op_ret == -1)
+                goto out;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        local = frame->local;
+        afr_readdir_filter_trash_dir (entries, local->fd);
 
-	priv     = this->private;
-	children = priv->children;
+out:
+        AFR_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries, NULL);
+        return 0;
+}
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+int32_t
+afr_do_readdir (call_frame_t *frame, xlator_t *this,
+                fd_t *fd, size_t size, off_t offset, int whichop, dict_t *dict)
+{
+        afr_private_t *priv      = NULL;
+        xlator_t      **children = NULL;
+        int           call_child = 0;
+        afr_local_t   *local     = NULL;
+        afr_fd_ctx_t  *fd_ctx    = NULL;
+        int           ret        = -1;
+        int32_t       op_errno   = 0;
+        uint64_t      read_child = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv     = this->private;
+        children = priv->children;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        local->fresh_children = afr_children_create (priv->child_count);
+        if (!local->fresh_children) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        read_child = afr_inode_get_read_ctx (this, fd->inode,
+                                             local->fresh_children);
+        ret = afr_get_call_child (this, local->child_up, read_child,
+                                  local->fresh_children,
+                                  &call_child,
+                                  &local->cont.readdir.last_index);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
+
+        fd_ctx  = afr_fd_ctx_get (fd, this);
+        if (!fd_ctx) {
+                op_errno = EBADF;
+                goto out;
+        }
+
+        if ((offset == 0) || (fd_ctx->call_child == -1)) {
+                fd_ctx->call_child = call_child;
+        } else if ((priv->readdir_failover == _gf_false) &&
+                   (call_child != fd_ctx->call_child)) {
+                op_errno = EBADF;
+                goto out;
+        }
+
+        local->fd                  = fd_ref (fd);
+        local->cont.readdir.size   = size;
+        local->cont.readdir.dict   = (dict)? dict_ref (dict) : NULL;
+
+        if (whichop == GF_FOP_READDIR)
+                STACK_WIND_COOKIE (frame, afr_readdir_cbk,
+                                   (void *) (long) call_child,
+                                   children[call_child],
+                                   children[call_child]->fops->readdir, fd,
+                                   size, offset, dict);
+        else
+                STACK_WIND_COOKIE (frame, afr_readdirp_cbk,
+                                   (void *) (long) call_child,
+                                   children[call_child],
+                                   children[call_child]->fops->readdirp, fd,
+                                   size, offset, dict);
+
+        return 0;
+out:
+        AFR_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
 
-	call_child = afr_first_up_child (priv);
-	if (call_child == -1) {
-		op_errno = ENOTCONN;
-		gf_log (this->name, GF_LOG_ERROR,
-			"no child is up :(");
-		goto out;
-	}
 
-	local->cont.getdents.last_tried = call_child;
+int32_t
+afr_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t offset, dict_t *xdata)
+{
+        afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIR, xdata);
+        return 0;
+}
 
-	local->fd                   = fd_ref (fd);
 
-	local->cont.getdents.size   = size;
-	local->cont.getdents.offset = offset;
-	local->cont.getdents.flag   = flag;
-	
-	frame->local = local;
+int32_t
+afr_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, dict_t *dict)
+{
+        afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIRP, dict);
+        return 0;
+}
 
-	STACK_WIND (frame, afr_getdents_cbk,
-		    children[call_child], children[call_child]->fops->getdents,
-		    fd, size, offset, flag);
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+int32_t
+afr_releasedir (xlator_t *this, fd_t *fd)
+{
+        afr_forget_entries (fd);
+        afr_cleanup_fd_ctx (this, fd);
 
-	return 0;
+        return 0;
 }
-
-
diff --git a/xlators/cluster/afr/src/afr-dir-read.h b/xlators/cluster/afr/src/afr-dir-read.h
index 6d981fdfd..09456d159 100644
--- a/xlators/cluster/afr/src/afr-dir-read.h
+++ b/xlators/cluster/afr/src/afr-dir-read.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __DIR_READ_H__
@@ -23,25 +14,23 @@
 
 int32_t
 afr_opendir (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, fd_t *fd);
+	     loc_t *loc, fd_t *fd, dict_t *xdata);
 
 int32_t
-afr_closedir (call_frame_t *frame, xlator_t *this,
-	      fd_t *fd);
+afr_releasedir (xlator_t *this, fd_t *fd);
 
 int32_t
 afr_readdir (call_frame_t *frame, xlator_t *this,
-	     fd_t *fd, size_t size, off_t offset);
+	     fd_t *fd, size_t size, off_t offset, dict_t *xdata);
 
 
 int32_t
-afr_getdents (call_frame_t *frame, xlator_t *this,
-	      fd_t *fd, size_t size, off_t offset, int32_t flag);
-
+afr_readdirp (call_frame_t *frame, xlator_t *this,
+              fd_t *fd, size_t size, off_t offset, dict_t *dict);
 
 int32_t
 afr_checksum (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, int32_t flags);
+	      loc_t *loc, int32_t flags, dict_t *xdata);
 
 
 #endif /* __DIR_READ_H__ */
diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c
index 5c6c14736..1943b719b 100644
--- a/xlators/cluster/afr/src/afr-dir-write.c
+++ b/xlators/cluster/afr/src/afr-dir-write.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  General Public License for more details.
-
-  You should have received a copy of the GNU General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 
@@ -47,28 +38,222 @@
 #include "afr.h"
 #include "afr-transaction.h"
 
+int
+afr_build_parent_loc (loc_t *parent, loc_t *child, int32_t *op_errno)
+{
+        int     ret = -1;
+        char    *child_path = NULL;
+
+        if (!child->parent) {
+                if (op_errno)
+                        *op_errno = EINVAL;
+                goto out;
+        }
+
+        child_path = gf_strdup (child->path);
+        if (!child_path) {
+                if (op_errno)
+                        *op_errno = ENOMEM;
+                goto out;
+        }
+        parent->path = gf_strdup( dirname (child_path) );
+	if (!parent->path) {
+                if (op_errno)
+                        *op_errno = ENOMEM;
+                goto out;
+        }
+        parent->inode  = inode_ref (child->parent);
+        uuid_copy (parent->gfid, child->pargfid);
+
+        ret = 0;
+out:
+	GF_FREE(child_path);
+
+        return ret;
+}
+
+void
+__dir_entry_fop_common_cbk (call_frame_t *frame, int child_index,
+                            xlator_t *this, int32_t op_ret,
+                            int32_t op_errno, inode_t *inode,
+                            struct iatt *buf, struct iatt *preparent,
+                            struct iatt *postparent, struct iatt *prenewparent,
+                            struct iatt *postnewparent)
+{
+        afr_local_t     *local          = NULL;
+
+        local = frame->local;
+
+        if (afr_fop_failed (op_ret, op_errno))
+                afr_transaction_fop_failed (frame, this, child_index);
+
+        if (op_ret > -1) {
+                local->op_ret = op_ret;
+
+                if ((local->success_count == 0) ||
+                    (child_index == local->read_child_index)) {
+                        local->cont.dir_fop.preparent      = *preparent;
+                        local->cont.dir_fop.postparent     = *postparent;
+                        if (buf)
+                                local->cont.dir_fop.buf            = *buf;
+                        if (prenewparent)
+                             local->cont.dir_fop.prenewparent  = *prenewparent;
+                        if (postnewparent)
+                             local->cont.dir_fop.postnewparent = *postnewparent;
+                }
+
+                local->cont.dir_fop.inode = inode;
+
+                local->fresh_children[local->success_count] = child_index;
+                local->success_count++;
+                local->child_errno[child_index] = 0;
+        } else {
+                local->child_errno[child_index] = op_errno;
+        }
+
+        local->op_errno = op_errno;
+}
+
+int
+afr_mark_new_entry_changelog_cbk (call_frame_t *frame, void *cookie,
+                                  xlator_t *this,
+                                  int32_t op_ret, int32_t op_errno,
+                                  dict_t *xattr, dict_t *xdata)
+{
+        int     call_count = 0;
+
+        call_count = afr_frame_return (frame);
+        if (call_count == 0) {
+                AFR_STACK_DESTROY (frame);
+        }
+        return 0;
+}
 
 void
-afr_build_parent_loc (loc_t *parent, loc_t *child)
+afr_mark_new_entry_changelog (call_frame_t *frame, xlator_t *this)
 {
-	char *tmp = NULL;
+        call_frame_t  *new_frame  = NULL;
+        afr_local_t   *local      = NULL;
+        afr_local_t   *new_local  = NULL;
+        afr_private_t *priv       = NULL;
+        dict_t        **xattr     = NULL;
+        int32_t       **changelog = NULL;
+        int           i           = 0;
+        GF_UNUSED int op_errno    = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        new_frame = copy_frame (frame);
+        if (!new_frame) {
+                goto out;
+        }
 
-	if (!child->parent) {
-		loc_copy (parent, child);
-		return;
-	}
+        AFR_LOCAL_ALLOC_OR_GOTO (new_frame->local, out);
+        new_local = new_frame->local;
+        changelog = afr_matrix_create (priv->child_count, AFR_NUM_CHANGE_LOGS);
+        if (!changelog)
+                goto out;
+
+        xattr = GF_CALLOC (priv->child_count, sizeof (*xattr),
+                           gf_afr_mt_dict_t);
+        if (!xattr)
+                goto out;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_errno[i])
+                        continue;
+                xattr[i] = dict_new ();
+                if (!xattr[i])
+                        goto out;
+        }
 
-	tmp = strdup (child->path);
-	parent->path   = strdup (dirname (tmp));
-	FREE (tmp);
+        afr_prepare_new_entry_pending_matrix (changelog,
+                                              afr_is_errno_set,
+                                              local->child_errno,
+                                              &local->cont.dir_fop.buf,
+                                              priv->child_count);
+
+        new_local->pending = changelog;
+        uuid_copy (new_local->loc.gfid, local->cont.dir_fop.buf.ia_gfid);
+        new_local->loc.inode = inode_ref (local->cont.dir_fop.inode);
+        new_local->call_count = local->success_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_errno[i])
+                        continue;
+
+                afr_set_pending_dict (priv, xattr[i], changelog, i, LOCAL_LAST);
+                STACK_WIND_COOKIE (new_frame, afr_mark_new_entry_changelog_cbk,
+                                   (void *) (long) i, priv->children[i],
+                                   priv->children[i]->fops->xattrop,
+                                   &new_local->loc, GF_XATTROP_ADD_ARRAY,
+                                   xattr[i], NULL);
+        }
+        new_frame = NULL;
+out:
+        if (new_frame)
+                AFR_STACK_DESTROY (new_frame);
+        afr_xattr_array_destroy (xattr, priv->child_count);
+        return;
+}
 
-        parent->name   = strrchr (parent->path, '/');
-	if (parent->name)
-		parent->name++;
+gf_boolean_t
+afr_is_new_entry_changelog_needed (glusterfs_fop_t fop)
+{
+        glusterfs_fop_t fops[]   = {GF_FOP_CREATE, GF_FOP_MKNOD, GF_FOP_NULL};
+        int             i        = 0;
 
-	parent->inode  = inode_ref (child->parent);
-	parent->parent = inode_parent (parent->inode, 0, NULL);
-	parent->ino    = parent->inode->ino;
+        for (i = 0; fops[i] != GF_FOP_NULL; i++) {
+                if (fop == fops[i])
+                        return _gf_true;
+        }
+        return _gf_false;
+}
+
+void
+afr_dir_fop_mark_entry_pending_changelog (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local      = NULL;
+        afr_private_t *priv       = NULL;
+
+        local = frame->local;
+        priv  = this->private;
+
+        if (local->op_ret < 0)
+                goto out;
+
+        if (local->success_count == priv->child_count)
+                goto out;
+
+        if (!afr_is_new_entry_changelog_needed (local->op))
+                goto out;
+
+        afr_mark_new_entry_changelog (frame, this);
+
+out:
+        return;
+}
+
+void
+afr_dir_fop_done (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local          = NULL;
+        afr_private_t   *priv           = NULL;
+
+        local = frame->local;
+        priv  = this->private;
+
+        if (local->cont.dir_fop.inode == NULL)
+                goto done;
+        afr_set_read_ctx_from_policy (this, local->cont.dir_fop.inode,
+                                      local->fresh_children,
+                                      local->read_child_index,
+                                      priv->read_child,
+                                      local->cont.dir_fop.buf.ia_gfid);
+done:
+        local->transaction.unwind (frame, this);
+        afr_dir_fop_mark_entry_pending_changelog (frame, this);
+        local->transaction.resume (frame, this);
 }
 
 /* {{{ create */
@@ -76,227 +261,247 @@ afr_build_parent_loc (loc_t *parent, loc_t *child)
 int
 afr_create_unwind (call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  local->cont.create.fd,
-				  local->cont.create.inode,
-				  &local->cont.create.buf);
-        }
-        
-	return 0;
+        call_frame_t *main_frame = NULL;
+        afr_local_t  *local = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame) {
+                        main_frame = local->transaction.main_frame;
+                }
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (create, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  local->cont.create.fd,
+                                  local->cont.dir_fop.inode,
+                                  &local->cont.dir_fop.buf,
+                                  &local->cont.dir_fop.preparent,
+                                  &local->cont.dir_fop.postparent,
+                                  local->xdata_rsp);
+        }
+
+        return 0;
 }
 
 
 int
-afr_create_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		     int32_t op_ret, int32_t op_errno, 
-		     fd_t *fd, inode_t *inode, struct stat *buf)
+afr_create_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno,
+                     fd_t *fd, inode_t *inode, struct iatt *buf,
+                     struct iatt *preparent, struct iatt *postparent,
+                     dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-        
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
-
-		if (op_ret != -1) {
-			local->op_ret = op_ret;
-
-			if (local->success_count == 0) {
-				local->cont.create.buf        = *buf;
-				local->cont.create.buf.st_ino = 
-					afr_itransform (buf->st_ino,
-							priv->child_count,
-							child_index);
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode, 
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode, 
-                                                            local->read_child_index);
-                                }
-			}
-                        
-                        if (child_index == local->read_child_index) {
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode, 
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
+        afr_local_t     *local = NULL;
+        uint64_t        ctx = 0;
+        afr_fd_ctx_t    *fd_ctx = NULL;
+        int             ret = 0;
+        int             call_count = -1;
+        int             child_index = -1;
+
+        local = frame->local;
+
+        child_index = (long) cookie;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret > -1) {
+                        ret = afr_fd_ctx_set (this, fd);
+                        if (ret < 0) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "could not set ctx on fd=%p", fd);
+
+                                local->op_ret   = -1;
+                                local->op_errno = -ret;
+                                goto unlock;
                         }
 
-			local->cont.create.inode = inode;
+                        ret = fd_ctx_get (fd, this, &ctx);
+                        if (ret < 0) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "could not get fd ctx for fd=%p", fd);
+                                local->op_ret   = -1;
+                                local->op_errno = -ret;
+                                goto unlock;
+                        }
 
-			local->success_count++;
-		}
+                        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+                        fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+                        fd_ctx->flags                  = local->cont.create.flags;
 
-	call_count = afr_frame_return (frame);
+                        if (local->success_count == 0) {
+				if (xdata)
+					local->xdata_rsp = dict_ref(xdata);
+			}
+                }
+                __dir_entry_fop_common_cbk (frame, child_index, this,
+                                            op_ret, op_errno, inode, buf,
+                                            preparent, postparent, NULL, NULL);
+        }
+
+unlock:
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
 
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
+        if (call_count == 0)
+                afr_dir_fop_done (frame, this);
 
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        return 0;
 }
 
 
 int
 afr_create_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_create_wind_cbk,
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->create,
-					   &local->loc, 
-					   local->cont.create.flags, 
-					   local->cont.create.mode, 
-					   local->cont.create.fd);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_create_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->create,
+                                           &local->loc,
+                                           local->cont.create.flags,
+                                           local->cont.create.mode,
+                                           local->umask,
+                                           local->cont.create.fd,
+                                           local->xdata_req);
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
 }
 
 
 int
 afr_create_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t * local = NULL;
+        afr_local_t * local = NULL;
 
-	local = frame->local;
+        local = frame->local;
 
-	local->transaction.unwind (frame, this);
+        local->transaction.unwind (frame, this);
 
-	AFR_STACK_DESTROY (frame);
+        AFR_STACK_DESTROY (frame);
 
-	return 0;
+        return 0;
 }
 
 
 int
 afr_create (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t flags, mode_t mode, fd_t *fd)
+            loc_t *loc, int32_t flags, mode_t mode,
+            mode_t umask, fd_t *fd, dict_t *params)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
+        afr_private_t           *priv                   = NULL;
+        afr_local_t             *local                  = NULL;
+        afr_internal_lock_t     *int_lock               = NULL;
+        call_frame_t            *transaction_frame      = NULL;
+        int                     ret                     = -1;
+        int                     op_errno                = 0;
 
-	int ret = -1;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        priv = this->private;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        QUORUM_CHECK(create,out);
 
-	priv = this->private;
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
 
-	transaction_frame->local = local;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	loc_copy (&local->loc, loc);
+        loc_copy (&local->loc, loc);
 
         LOCK (&priv->read_child_lock);
         {
-                local->read_child_index = (++priv->read_child_rr) 
+                local->read_child_index = (++priv->read_child_rr)
                         % (priv->child_count);
         }
         UNLOCK (&priv->read_child_lock);
 
-	local->cont.create.flags = flags;
-	local->cont.create.mode  = mode;
-	local->cont.create.fd    = fd_ref (fd);
-
-	local->transaction.fop    = afr_create_wind;
-	local->transaction.done   = afr_create_done;
-	local->transaction.unwind = afr_create_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
-
-	local->transaction.main_frame = frame;
-	local->transaction.basename = AFR_BASENAME (loc->path);
-	local->transaction.pending  = AFR_ENTRY_PENDING;
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        local->op                = GF_FOP_CREATE;
+        local->cont.create.flags = flags;
+        local->cont.create.mode  = mode;
+        local->cont.create.fd    = fd_ref (fd);
+        local->umask  = umask;
+        if (params)
+                local->xdata_req = dict_ref (params);
+
+        local->transaction.fop    = afr_create_wind;
+        local->transaction.done   = afr_create_done;
+        local->transaction.unwind = afr_create_unwind;
+
+        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
+                                    &op_errno);
+        if (ret)
+                goto out;
+
+        local->transaction.main_frame = frame;
+        local->transaction.basename = AFR_BASENAME (loc->path);
+        int_lock = &local->internal_lock;
+
+        int_lock->lockee_count = 0;
+        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
+                                     &local->transaction.parent_loc,
+                                     local->transaction.basename,
+                                     priv->child_count);
+        if (ret)
+                goto out;
+
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (create, frame, -1, op_errno,
+                                  NULL, NULL, NULL, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 /* }}} */
@@ -306,221 +511,201 @@ out:
 int
 afr_mknod_unwind (call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
+        call_frame_t *main_frame = NULL;
+        afr_local_t  *local = NULL;
 
-	local = frame->local;
+        local = frame->local;
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  local->cont.mknod.inode,
-				  &local->cont.mknod.buf);
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame) {
+                        main_frame = local->transaction.main_frame;
+                }
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (mknod, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  local->cont.dir_fop.inode,
+                                  &local->cont.dir_fop.buf,
+                                  &local->cont.dir_fop.preparent,
+                                  &local->cont.dir_fop.postparent,
+                                  NULL);
         }
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_mknod_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		    int32_t op_ret, int32_t op_errno, 
-		    inode_t *inode, struct stat *buf)
+afr_mknod_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *buf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
-		
-		if (op_ret != -1) {
-			local->op_ret = op_ret;
-
-			if (local->success_count == 0){
-				local->cont.mknod.buf   = *buf;
-				local->cont.mknod.buf.st_ino = 
-					afr_itransform (buf->st_ino,
-							priv->child_count,
-							child_index);
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-			}
-
-                        if (child_index == local->read_child_index) {
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-                        }
-                        
-			local->cont.mknod.inode = inode;
+        int             call_count      = -1;
+        int             child_index     = -1;
 
-			local->success_count++;
-		}
+        child_index = (long) cookie;
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        LOCK (&frame->lock);
+        {
+                __dir_entry_fop_common_cbk (frame, child_index, this,
+                                            op_ret, op_errno, inode, buf,
+                                            preparent, postparent, NULL, NULL);
+        }
+        UNLOCK (&frame->lock);
 
-	call_count = afr_frame_return (frame);
+        call_count = afr_frame_return (frame);
 
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
+        if (call_count == 0)
+                afr_dir_fop_done (frame, this);
 
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        return 0;
 }
 
 
 int32_t
 afr_mknod_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_mknod_wind_cbk, (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->mknod,
-					   &local->loc, local->cont.mknod.mode,
-					   local->cont.mknod.dev);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv  = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_mknod_wind_cbk, (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->mknod,
+                                           &local->loc, local->cont.mknod.mode,
+                                           local->cont.mknod.dev,
+                                           local->umask,
+                                           local->xdata_req);
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
 }
 
 
 int
 afr_mknod_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t * local = NULL;
+        afr_local_t * local = NULL;
 
-	local = frame->local;
+        local = frame->local;
 
-	local->transaction.unwind (frame, this);
-	AFR_STACK_DESTROY (frame);
+        local->transaction.unwind (frame, this);
+        AFR_STACK_DESTROY (frame);
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_mknod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dev_t dev)
+afr_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+           dev_t dev, mode_t umask, dict_t *params)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
+        afr_private_t           *priv                   = NULL;
+        afr_local_t             *local                  = NULL;
+        afr_internal_lock_t     *int_lock               = NULL;
+        call_frame_t            *transaction_frame      = NULL;
+        int                     ret                     = -1;
+        int                     op_errno                = 0;
 
-	int ret = -1;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        priv = this->private;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        QUORUM_CHECK(mknod,out);
 
-	priv = this->private;
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
 
-	transaction_frame->local = local;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	loc_copy (&local->loc, loc);
+        loc_copy (&local->loc, loc);
 
         LOCK (&priv->read_child_lock);
         {
-                local->read_child_index = (++priv->read_child_rr) 
+                local->read_child_index = (++priv->read_child_rr)
                         % (priv->child_count);
         }
         UNLOCK (&priv->read_child_lock);
 
-	local->cont.mknod.mode  = mode;
-	local->cont.mknod.dev   = dev;
-
-	local->transaction.fop    = afr_mknod_wind;
-	local->transaction.done   = afr_mknod_done;
-	local->transaction.unwind = afr_mknod_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
-
-	local->transaction.main_frame = frame;
-	local->transaction.basename = AFR_BASENAME (loc->path);
-	local->transaction.pending  = AFR_ENTRY_PENDING;
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        local->op               = GF_FOP_MKNOD;
+        local->cont.mknod.mode  = mode;
+        local->cont.mknod.dev   = dev;
+        local->umask = umask;
+        if (params)
+                local->xdata_req = dict_ref (params);
+
+        local->transaction.fop    = afr_mknod_wind;
+        local->transaction.done   = afr_mknod_done;
+        local->transaction.unwind = afr_mknod_unwind;
+
+        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
+                                    &op_errno);
+        if (ret)
+                goto out;
+
+        local->transaction.main_frame = frame;
+        local->transaction.basename = AFR_BASENAME (loc->path);
+        int_lock = &local->internal_lock;
+
+        int_lock->lockee_count = 0;
+        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
+                                     &local->transaction.parent_loc,
+                                     local->transaction.basename,
+                                     priv->child_count);
+        if (ret)
+                goto out;
+
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (mknod, frame, -1, op_errno,
+                                  NULL, NULL, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 /* }}} */
@@ -531,221 +716,201 @@ out:
 int
 afr_mkdir_unwind (call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-	local = frame->local;
+        call_frame_t *main_frame = NULL;
+        afr_local_t  *local = NULL;
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
+        local = frame->local;
 
-	if (main_frame) {
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  local->cont.mkdir.inode,
-				  &local->cont.mkdir.buf);
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame) {
+                        main_frame = local->transaction.main_frame;
+                }
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (mkdir, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  local->cont.dir_fop.inode,
+                                  &local->cont.dir_fop.buf,
+                                  &local->cont.dir_fop.preparent,
+                                  &local->cont.dir_fop.postparent,
+                                  NULL);
         }
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_mkdir_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		    int32_t op_ret, int32_t op_errno, 
-		    inode_t *inode, struct stat *buf)
+afr_mkdir_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *buf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
-
-		if (op_ret != -1) {
-			local->op_ret           = op_ret;
-
-			if (local->success_count == 0) {
-				local->cont.mkdir.buf   = *buf;
-				local->cont.mkdir.buf.st_ino = 
-					afr_itransform (buf->st_ino, priv->child_count,
-							child_index);
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-			}
-                        
-                        if (child_index == local->read_child_index) {
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-                        }
-
-			local->cont.mkdir.inode = inode;
+        int             call_count      = -1;
+        int             child_index     = -1;
 
-			local->success_count++;
-		}
+        child_index = (long) cookie;
 
-		local->op_errno         = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        LOCK (&frame->lock);
+        {
+                __dir_entry_fop_common_cbk (frame, child_index, this,
+                                            op_ret, op_errno, inode, buf,
+                                            preparent, postparent, NULL, NULL);
+        }
+        UNLOCK (&frame->lock);
 
-	call_count = afr_frame_return (frame);
+        call_count = afr_frame_return (frame);
 
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
+        if (call_count == 0)
+                afr_dir_fop_done (frame, this);
 
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        return 0;
 }
 
 
 int
 afr_mkdir_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-	
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_mkdir_wind_cbk,
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->mkdir,
-					   &local->loc, local->cont.mkdir.mode);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv  = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_mkdir_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->mkdir,
+                                           &local->loc, local->cont.mkdir.mode,
+                                           local->umask,
+                                           local->xdata_req);
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
 }
 
 
 int
 afr_mkdir_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t * local = NULL;
+        afr_local_t * local = NULL;
 
-	local = frame->local;
+        local = frame->local;
 
-	local->transaction.unwind (frame, this);
+        local->transaction.unwind (frame, this);
 
-	AFR_STACK_DESTROY (frame);
+        AFR_STACK_DESTROY (frame);
 
-	return 0;
+        return 0;
 }
 
-
 int
 afr_mkdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode)
+           loc_t *loc, mode_t mode, mode_t umask, dict_t *params)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
+        afr_private_t           *priv                   = NULL;
+        afr_local_t             *local                  = NULL;
+        afr_internal_lock_t     *int_lock               = NULL;
+        call_frame_t            *transaction_frame      = NULL;
+        int                     ret                     = -1;
+        int                     op_errno                = 0;
 
-	int ret = -1;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        priv = this->private;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        QUORUM_CHECK(mkdir,out);
 
-	priv = this->private;
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
 
-	transaction_frame->local = local;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	loc_copy (&local->loc, loc);
+        loc_copy (&local->loc, loc);
 
         LOCK (&priv->read_child_lock);
         {
-                local->read_child_index = (++priv->read_child_rr) 
+                local->read_child_index = (++priv->read_child_rr)
                         % (priv->child_count);
         }
         UNLOCK (&priv->read_child_lock);
 
-	local->cont.mkdir.mode  = mode;
-
-	local->transaction.fop    = afr_mkdir_wind;
-	local->transaction.done   = afr_mkdir_done;
-	local->transaction.unwind = afr_mkdir_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
-
-	local->transaction.main_frame = frame;
-	local->transaction.basename = AFR_BASENAME (loc->path);
-	local->transaction.pending  = AFR_ENTRY_PENDING;
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        local->cont.mkdir.mode  = mode;
+        local->umask = umask;
+        if (params)
+                local->xdata_req = dict_ref (params);
+
+        local->op = GF_FOP_MKDIR;
+        local->transaction.fop    = afr_mkdir_wind;
+        local->transaction.done   = afr_mkdir_done;
+        local->transaction.unwind = afr_mkdir_unwind;
+
+        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
+                                    &op_errno);
+        if (ret)
+                goto out;
+
+        local->transaction.main_frame = frame;
+        local->transaction.basename = AFR_BASENAME (loc->path);
+        int_lock = &local->internal_lock;
+
+        int_lock->lockee_count = 0;
+        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
+                                     &local->transaction.parent_loc,
+                                     local->transaction.basename,
+                                     priv->child_count);
+        if (ret)
+                goto out;
+
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
 
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+                AFR_STACK_UNWIND (mkdir, frame, -1, op_errno,
+                                  NULL, NULL, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 /* }}} */
@@ -756,223 +921,196 @@ out:
 int
 afr_link_unwind (call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-	local = frame->local;
+        call_frame_t *main_frame = NULL;
+        afr_local_t  *local = NULL;
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
+        local = frame->local;
 
-	if (main_frame) {
-		local->cont.link.buf.st_ino = local->cont.link.ino;
-
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno, 
-				  local->cont.link.inode,
-				  &local->cont.link.buf);
-	}
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame) {
+                        main_frame = local->transaction.main_frame;
+                }
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (link, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  local->cont.dir_fop.inode,
+                                  &local->cont.dir_fop.buf,
+                                  &local->cont.dir_fop.preparent,
+                                  &local->cont.dir_fop.postparent,
+                                  NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_link_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		   int32_t op_ret, int32_t op_errno, inode_t *inode,
-		   struct stat *buf)
+afr_link_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
-
-		if (op_ret != -1) {
-			local->op_ret   = op_ret;
-
-			if (local->success_count == 0) {
-				local->cont.link.buf        = *buf;
-				local->cont.link.buf.st_ino = 
-					afr_itransform (buf->st_ino, priv->child_count,
-							child_index);
-                                
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-			}
-                        
-                        if (child_index == local->read_child_index) {
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-                        }
-
-			local->cont.link.inode    = inode;
+        int             call_count      = -1;
+        int             child_index     = -1;
 
-			local->success_count++;
-		}
+        child_index = (long) cookie;
 
-		local->op_errno = op_errno;		
-	}
-	UNLOCK (&frame->lock);
+        LOCK (&frame->lock);
+        {
+                __dir_entry_fop_common_cbk (frame, child_index, this,
+                                            op_ret, op_errno, inode, buf,
+                                            preparent, postparent, NULL, NULL);
+        }
+        UNLOCK (&frame->lock);
 
-	call_count = afr_frame_return (frame);
+        call_count = afr_frame_return (frame);
 
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
+        if (call_count == 0)
+                afr_dir_fop_done (frame, this);
 
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        return 0;
 }
 
 
 int
 afr_link_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_link_wind_cbk, (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->link,
-					   &local->loc,
-					   &local->newloc);
-			
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv  = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_link_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->link,
+                                           &local->loc,
+                                           &local->newloc, local->xdata_req);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
 }
 
 
 int
 afr_link_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t * local = frame->local;
+        afr_local_t * local = frame->local;
 
-	local->transaction.unwind (frame, this);
+        local->transaction.unwind (frame, this);
 
-	AFR_STACK_DESTROY (frame);
+        AFR_STACK_DESTROY (frame);
 
-	return 0;
+        return 0;
 }
 
 
 int
 afr_link (call_frame_t *frame, xlator_t *this,
-	  loc_t *oldloc, loc_t *newloc)
+          loc_t *oldloc, loc_t *newloc, dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-
-	int ret = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
+        afr_private_t           *priv                   = NULL;
+        afr_local_t             *local                  = NULL;
+        afr_internal_lock_t     *int_lock               = NULL;
+        call_frame_t            *transaction_frame      = NULL;
+        int                     ret                     = -1;
+        int                     op_errno                = 0;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	priv = this->private;
+        priv = this->private;
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        QUORUM_CHECK(link,out);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
 
-	transaction_frame->local = local;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	loc_copy (&local->loc,    oldloc);
-	loc_copy (&local->newloc, newloc);
+        loc_copy (&local->loc,    oldloc);
+        loc_copy (&local->newloc, newloc);
+        if (xdata)
+                local->xdata_req = dict_ref (xdata);
 
         LOCK (&priv->read_child_lock);
         {
-                local->read_child_index = (++priv->read_child_rr) 
+                local->read_child_index = (++priv->read_child_rr)
                         % (priv->child_count);
         }
         UNLOCK (&priv->read_child_lock);
 
-	local->cont.link.ino = oldloc->inode->ino;
-
-	local->transaction.fop    = afr_link_wind;
-	local->transaction.done   = afr_link_done;
-	local->transaction.unwind = afr_link_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, oldloc);
-
-	local->transaction.main_frame   = frame;
-	local->transaction.basename     = AFR_BASENAME (oldloc->path);
-	local->transaction.new_basename = AFR_BASENAME (newloc->path);
-	local->transaction.pending      = AFR_ENTRY_PENDING;
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-
-	op_ret = 0;
+        local->op = GF_FOP_LINK;
+        local->transaction.fop    = afr_link_wind;
+        local->transaction.done   = afr_link_done;
+        local->transaction.unwind = afr_link_unwind;
+
+        ret = afr_build_parent_loc (&local->transaction.parent_loc, newloc,
+                                    &op_errno);
+        if (ret)
+                goto out;
+
+        local->transaction.main_frame   = frame;
+        local->transaction.basename     = AFR_BASENAME (newloc->path);
+        int_lock = &local->internal_lock;
+
+        int_lock->lockee_count = 0;
+        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
+                                     &local->transaction.parent_loc,
+                                     local->transaction.basename,
+                                     priv->child_count);
+        if (ret)
+                goto out;
+
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (link, frame, -1, op_errno,
+                                  NULL, NULL, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 /* }}} */
@@ -983,222 +1121,202 @@ out:
 int
 afr_symlink_unwind (call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-	local = frame->local;
+        call_frame_t *main_frame = NULL;
+        afr_local_t  *local = NULL;
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
+        local = frame->local;
 
-	if (main_frame) {
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  local->cont.symlink.inode,
-				  &local->cont.symlink.buf);
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame) {
+                        main_frame = local->transaction.main_frame;
+                }
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (symlink, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  local->cont.dir_fop.inode,
+                                  &local->cont.dir_fop.buf,
+                                  &local->cont.dir_fop.preparent,
+                                  &local->cont.dir_fop.postparent,
+                                  NULL);
         }
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_symlink_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		      int32_t op_ret, int32_t op_errno, inode_t *inode,
-		      struct stat *buf)
+afr_symlink_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, inode_t *inode,
+                      struct iatt *buf, struct iatt *preparent,
+                      struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
-		
-		if (op_ret != -1) {
-			local->op_ret   = op_ret;
-
-			if (local->success_count == 0) {
-				local->cont.symlink.buf        = *buf;
-				local->cont.symlink.buf.st_ino = 
-					afr_itransform (buf->st_ino, priv->child_count,
-							child_index);
-                                
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-			}
-
-                        if (child_index == local->read_child_index) {
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-                        }
+        int             call_count      = -1;
+        int             child_index     = -1;
 
-			local->cont.symlink.inode    = inode;
+        child_index = (long) cookie;
 
-			local->success_count++;
-		}
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        LOCK (&frame->lock);
+        {
+                __dir_entry_fop_common_cbk (frame, child_index, this,
+                                            op_ret, op_errno, inode, buf,
+                                            preparent, postparent, NULL, NULL);
+        }
+        UNLOCK (&frame->lock);
 
-	call_count = afr_frame_return (frame);
+        call_count = afr_frame_return (frame);
 
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
+        if (call_count == 0)
+                afr_dir_fop_done (frame, this);
 
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        return 0;
 }
 
 
 int
 afr_symlink_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
 
-	int call_count = -1;
-	int i = 0;
+        local = frame->local;
+        priv = this->private;
 
-	local = frame->local;
-	priv = this->private;
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
 
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
+        local->call_count = call_count;
 
-	local->call_count = call_count;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_symlink_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->symlink,
+                                           local->cont.symlink.linkpath,
+                                           &local->loc,
+                                           local->umask,
+                                           local->xdata_req);
 
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_symlink_wind_cbk,
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->symlink,
-					   local->cont.symlink.linkpath,
-					   &local->loc);
+                        if (!--call_count)
+                                break;
 
-			if (!--call_count)
-				break;
+                }
+        }
 
-		}
-	}
-	
-	return 0;
+        return 0;
 }
 
 
 int
 afr_symlink_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t * local = frame->local;
+        afr_local_t * local = frame->local;
 
-	local->transaction.unwind (frame, this);
+        local->transaction.unwind (frame, this);
 
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
+        AFR_STACK_DESTROY (frame);
+
+        return 0;
 }
 
 
 int
 afr_symlink (call_frame_t *frame, xlator_t *this,
-	     const char *linkpath, loc_t *loc)
+             const char *linkpath, loc_t *loc, mode_t umask, dict_t *params)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-
-	int ret = -1;
+        afr_private_t           *priv                   = NULL;
+        afr_local_t             *local                  = NULL;
+        afr_internal_lock_t     *int_lock               = NULL;
+        call_frame_t            *transaction_frame      = NULL;
+        int                     ret                     = -1;
+        int                     op_errno                = 0;
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        priv = this->private;
 
-	priv = this->private;
+        QUORUM_CHECK(symlink,out);
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	transaction_frame->local = local;
-	
-	loc_copy (&local->loc, loc);
+        loc_copy (&local->loc, loc);
 
         LOCK (&priv->read_child_lock);
         {
-                local->read_child_index = (++priv->read_child_rr) 
+                local->read_child_index = (++priv->read_child_rr)
                         % (priv->child_count);
         }
         UNLOCK (&priv->read_child_lock);
 
-	local->cont.symlink.ino      = loc->inode->ino;
-	local->cont.symlink.linkpath = strdup (linkpath);
-
-	local->transaction.fop    = afr_symlink_wind;
-	local->transaction.done   = afr_symlink_done;
-	local->transaction.unwind = afr_symlink_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
-
-	local->transaction.main_frame   = frame;
-	local->transaction.basename     = AFR_BASENAME (loc->path);
-	local->transaction.pending      = AFR_ENTRY_PENDING;
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        local->cont.symlink.linkpath = gf_strdup (linkpath);
+        local->umask = umask;
+        if (params)
+                local->xdata_req = dict_ref (params);
+
+        local->op = GF_FOP_SYMLINK;
+        local->transaction.fop    = afr_symlink_wind;
+        local->transaction.done   = afr_symlink_done;
+        local->transaction.unwind = afr_symlink_unwind;
+
+        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
+                                    &op_errno);
+        if (ret)
+                goto out;
+
+        local->transaction.main_frame   = frame;
+        local->transaction.basename     = AFR_BASENAME (loc->path);
+        int_lock = &local->internal_lock;
+
+        int_lock->lockee_count = 0;
+        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
+                                     &local->transaction.parent_loc,
+                                     local->transaction.basename,
+                                     priv->child_count);
+        if (ret)
+                goto out;
+
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL, NULL);
-	}
+        if (ret  < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (symlink, frame, -1, op_errno,
+                                  NULL, NULL, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 /* }}} */
@@ -1208,195 +1326,231 @@ out:
 int
 afr_rename_unwind (call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-	local = frame->local;
+        call_frame_t *main_frame = NULL;
+        afr_local_t  *local = NULL;
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
+        local = frame->local;
 
-	if (main_frame) {
-		local->cont.rename.buf.st_ino = local->cont.rename.ino;
-
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno, 
-				  &local->cont.rename.buf);
-	}
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame) {
+                        main_frame = local->transaction.main_frame;
+                }
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (rename, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  &local->cont.dir_fop.buf,
+                                  &local->cont.dir_fop.preparent,
+                                  &local->cont.dir_fop.postparent,
+                                  &local->cont.dir_fop.prenewparent,
+                                  &local->cont.dir_fop.postnewparent,
+                                  NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_rename_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		     int32_t op_ret, int32_t op_errno, struct stat *buf)
+afr_rename_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                     struct iatt *preoldparent, struct iatt *postoldparent,
+                     struct iatt *prenewparent, struct iatt *postnewparent,
+                     dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv  = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
-
-		if ((op_ret != -1) && (local->success_count == 0)) {
-			local->op_ret = op_ret;
-		
-			if (buf) {
-				local->cont.rename.buf = *buf;
-				local->cont.rename.buf.st_ino = 
-					afr_itransform (buf->st_ino, priv->child_count,
-							child_index);
-			}
-			local->success_count++;
-		}
+        afr_local_t *   local = NULL;
+        int call_count = -1;
+        int child_index = -1;
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        local = frame->local;
 
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
-
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
-}
+        child_index = (long) cookie;
 
+        LOCK (&frame->lock);
+        {
+                if (afr_fop_failed (op_ret, op_errno) && op_errno != ENOTEMPTY)
+                        afr_transaction_fop_failed (frame, this, child_index);
+                local->op_errno = op_errno;
+                local->child_errno[child_index] = op_errno;
 
-int32_t
-afr_rename_wind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_rename_wind_cbk, 
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->rename,
-					   &local->loc,
-					   &local->newloc);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
-}
+                if (op_ret > -1)
+                        __dir_entry_fop_common_cbk (frame, child_index, this,
+                                                   op_ret, op_errno, NULL, buf,
+                                                   preoldparent, postoldparent,
+                                                   prenewparent, postnewparent);
 
+        }
+        UNLOCK (&frame->lock);
 
-int
-afr_rename_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t * local = frame->local;
+        call_count = afr_frame_return (frame);
 
-	local->transaction.unwind (frame, this);
+        if (call_count == 0)
+                afr_dir_fop_done (frame, this);
 
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
+        return 0;
 }
 
 
-int
-afr_rename (call_frame_t *frame, xlator_t *this,
-	    loc_t *oldloc, loc_t *newloc)
+int32_t
+afr_rename_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-
-	int ret = -1;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        local = frame->local;
+        priv = this->private;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
 
-	priv = this->private;
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_rename_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->rename,
+                                           &local->loc,
+                                           &local->newloc, NULL);
+                        if (!--call_count)
+                                break;
+                }
+        }
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        return 0;
+}
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
 
-	transaction_frame->local = local;
+int
+afr_rename_done (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t * local = frame->local;
 
-	loc_copy (&local->loc,    oldloc);
-	loc_copy (&local->newloc, newloc);
+        local->transaction.unwind (frame, this);
 
-	local->cont.rename.ino = oldloc->inode->ino;
+        AFR_STACK_DESTROY (frame);
 
-	local->transaction.fop    = afr_rename_wind;
-	local->transaction.done   = afr_rename_done;
-	local->transaction.unwind = afr_rename_unwind;
+        return 0;
+}
 
-	afr_build_parent_loc (&local->transaction.parent_loc, oldloc);
-	afr_build_parent_loc (&local->transaction.new_parent_loc, newloc);
 
-	local->transaction.main_frame   = frame;
-	local->transaction.basename     = AFR_BASENAME (oldloc->path);
-	local->transaction.new_basename = AFR_BASENAME (newloc->path);
-	local->transaction.pending      = AFR_ENTRY_PENDING;
+int
+afr_rename (call_frame_t *frame, xlator_t *this,
+            loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+{
+        afr_private_t           *priv                   = NULL;
+        afr_local_t             *local                  = NULL;
+        afr_internal_lock_t     *int_lock               = NULL;
+        call_frame_t            *transaction_frame      = NULL;
+        int                     ret                     = -1;
+        int                     op_errno                = 0;
+        int                     nlockee                 = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        QUORUM_CHECK(rename,out);
+
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	afr_transaction (transaction_frame, this, AFR_ENTRY_RENAME_TRANSACTION);
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        loc_copy (&local->loc,    oldloc);
+        loc_copy (&local->newloc, newloc);
+
+        local->read_child_index = afr_inode_get_read_ctx (this, oldloc->inode, NULL);
+
+        local->op = GF_FOP_RENAME;
+        local->transaction.fop    = afr_rename_wind;
+        local->transaction.done   = afr_rename_done;
+        local->transaction.unwind = afr_rename_unwind;
+
+        ret = afr_build_parent_loc (&local->transaction.parent_loc, oldloc,
+                                    &op_errno);
+        if (ret)
+                goto out;
+        ret = afr_build_parent_loc (&local->transaction.new_parent_loc, newloc,
+                                    &op_errno);
+        if (ret)
+                goto out;
+
+        local->transaction.main_frame   = frame;
+        local->transaction.basename     = AFR_BASENAME (oldloc->path);
+        local->transaction.new_basename = AFR_BASENAME (newloc->path);
+        int_lock = &local->internal_lock;
+
+        int_lock->lockee_count = nlockee = 0;
+        ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
+                                     &local->transaction.new_parent_loc,
+                                     local->transaction.new_basename,
+                                     priv->child_count);
+        if (ret)
+                goto out;
+
+        nlockee++;
+        ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
+                                     &local->transaction.parent_loc,
+                                     local->transaction.basename,
+                                     priv->child_count);
+        if (ret)
+                goto out;
+
+        nlockee++;
+        if (local->newloc.inode && IA_ISDIR (local->newloc.inode->ia_type)) {
+                ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
+                                             &local->newloc,
+                                             NULL,
+                                             priv->child_count);
+                if (ret)
+                        goto out;
+
+                nlockee++;
+        }
+        qsort (int_lock->lockee, nlockee, sizeof (*int_lock->lockee),
+               afr_entry_lockee_cmp);
+        int_lock->lockee_count = nlockee;
+
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_RENAME_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
 
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+                AFR_STACK_UNWIND (rename, frame, -1, op_errno,
+                                  NULL, NULL, NULL, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 /* }}} */
@@ -1406,183 +1560,190 @@ out:
 int
 afr_unlink_unwind (call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-	local = frame->local;
+        call_frame_t *main_frame = NULL;
+        afr_local_t  *local = NULL;
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
+        local = frame->local;
 
-	if (main_frame)
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno);
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame) {
+                        main_frame = local->transaction.main_frame;
+                }
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (unlink, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  &local->cont.dir_fop.preparent,
+                                  &local->cont.dir_fop.postparent,
+                                  NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_unlink_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		     int32_t op_ret, int32_t op_errno)
+afr_unlink_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count  = -1;
-	int child_index = (long) cookie;
-	int need_unwind = 0;
-
-	local = frame->local;
-	priv  = this->private;
-	
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
-
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret   = op_ret;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
-		}
+        afr_local_t *   local = NULL;
+        int call_count  = -1;
+        int child_index = (long) cookie;
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        local = frame->local;
 
-	call_count = afr_frame_return (frame);
+        LOCK (&frame->lock);
+        {
+                if (child_index == local->read_child_index) {
+                        local->read_child_returned = _gf_true;
+                }
+                __dir_entry_fop_common_cbk (frame, child_index, this,
+                                            op_ret, op_errno, NULL, NULL,
+                                            preparent, postparent, NULL, NULL);
+        }
+        UNLOCK (&frame->lock);
 
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
+        call_count = afr_frame_return (frame);
+        if (call_count == 0)
+                afr_dir_fop_done (frame, this);
 
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        return 0;
 }
 
 
 int32_t
 afr_unlink_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_unlink_wind_cbk,	
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->unlink,
-					   &local->loc);
-			
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
-}
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
 
+        local = frame->local;
+        priv  = this->private;
 
-int32_t
-afr_unlink_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t * local = frame->local;
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
 
-	local->transaction.unwind (frame, this);
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
 
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
-}
+        local->call_count = call_count;
 
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_unlink_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->unlink,
+                                           &local->loc, local->xflag,
+                                           local->xdata_req);
 
-int32_t
-afr_unlink (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc)
-{
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
+                        if (!--call_count)
+                                break;
+                }
+        }
 
-	int ret = -1;
+        return 0;
+}
 
-	int op_ret   = -1;
-	int op_errno = 0;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+int32_t
+afr_unlink_done (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t * local = frame->local;
 
-	priv = this->private;
+        local->transaction.unwind (frame, this);
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        AFR_STACK_DESTROY (frame);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        return 0;
+}
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
 
-	transaction_frame->local = local;
+int32_t
+afr_unlink (call_frame_t *frame, xlator_t *this,
+            loc_t *loc, int xflag, dict_t *xdata)
+{
+        afr_private_t           *priv                   = NULL;
+        afr_local_t             *local                  = NULL;
+        afr_internal_lock_t     *int_lock               = NULL;
+        call_frame_t            *transaction_frame      = NULL;
+        int                     ret                     = -1;
+        int                     op_errno                = 0;
 
-	loc_copy (&local->loc, loc);
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	local->transaction.fop    = afr_unlink_wind;
-	local->transaction.done   = afr_unlink_done;
-	local->transaction.unwind = afr_unlink_unwind;
+        priv = this->private;
 
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
+        QUORUM_CHECK(unlink,out);
 
-	local->transaction.main_frame = frame;
-	local->transaction.basename = AFR_BASENAME (loc->path);
-	local->transaction.pending  = AFR_ENTRY_PENDING;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        loc_copy (&local->loc, loc);
+        local->xflag = xflag;
+        if (xdata)
+                local->xdata_req = dict_ref (xdata);
+
+        local->op = GF_FOP_UNLINK;
+        local->transaction.fop    = afr_unlink_wind;
+        local->transaction.done   = afr_unlink_done;
+        local->transaction.unwind = afr_unlink_unwind;
+
+        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
+                                    &op_errno);
+        if (ret)
+                goto out;
+
+        local->transaction.main_frame = frame;
+        local->transaction.basename = AFR_BASENAME (loc->path);
+        int_lock = &local->internal_lock;
+
+        int_lock->lockee_count = 0;
+        ret = afr_init_entry_lockee (&int_lock->lockee[0], local,
+                                     &local->transaction.parent_loc,
+                                     local->transaction.basename,
+                                     priv->child_count);
+        if (ret)
+                goto out;
+
+        int_lock->lockee_count++;
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (unlink, frame, -1, op_errno,
+                                  NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 /* }}} */
@@ -1594,325 +1755,208 @@ out:
 int
 afr_rmdir_unwind (call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-	local = frame->local;
+        call_frame_t *main_frame = NULL;
+        afr_local_t  *local = NULL;
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
+        local = frame->local;
 
-	if (main_frame)
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno);
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame) {
+                        main_frame = local->transaction.main_frame;
+                }
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (rmdir, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  &local->cont.dir_fop.preparent,
+                                  &local->cont.dir_fop.postparent,
+                                  NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_rmdir_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		    int32_t op_ret, int32_t op_errno)
+afr_rmdir_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count  = -1;
-	int child_index = (long) cookie;
-	int need_unwind = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
-
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-			}
-			local->success_count++;
+        afr_local_t *   local = NULL;
+        int call_count  = -1;
+        int child_index = (long) cookie;
+        int read_child  = 0;
 
-			if (local->success_count == priv->wait_count)
-				need_unwind = 1;
-		}
+        local = frame->local;
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
+                if (afr_fop_failed (op_ret, op_errno) && (op_errno != ENOTEMPTY))
+                        afr_transaction_fop_failed (frame, this, child_index);
+                local->op_errno = op_errno;
+                local->child_errno[child_index] = op_errno;
+                if (op_ret > -1)
+                        __dir_entry_fop_common_cbk (frame, child_index, this,
+                                                   op_ret, op_errno, NULL, NULL,
+                                                   preparent, postparent, NULL,
+                                                   NULL);
 
-	call_count = afr_frame_return (frame);
+        }
+        UNLOCK (&frame->lock);
 
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
+        call_count = afr_frame_return (frame);
+        if (call_count == 0)
+                afr_dir_fop_done (frame, this);
 
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        return 0;
 }
 
 
 int
 afr_rmdir_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_rmdir_wind_cbk,	
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->rmdir,
-					   &local->loc);
-
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
-}
-
-
-int
-afr_rmdir_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t * local = frame->local;
-
-	local->transaction.unwind (frame, this);
-
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
-}
-
-
-int
-afr_rmdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc)
-{
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-	
-	int ret = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	transaction_frame->local = local;
-
-	loc_copy (&local->loc, loc);
-
-	local->transaction.fop    = afr_rmdir_wind;
-	local->transaction.done   = afr_rmdir_done;
-	local->transaction.unwind = afr_rmdir_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
-
-	local->transaction.main_frame = frame;
-	local->transaction.basename = AFR_BASENAME (loc->path);
-	local->transaction.pending  = AFR_ENTRY_PENDING;
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-
-	return 0;
-}
-
-/* }}} */
-
-/* {{{ setdents */
-
-int32_t
-afr_setdents_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		       int32_t op_ret, int32_t op_errno)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
 
-	int call_count  = -1;
-	int child_index = (long) cookie;
+        local = frame->local;
+        priv  = this->private;
 
-	local = frame->local;
-	priv = this->private;
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
 
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
 
-		if ((op_ret != -1) && (local->success_count == 0)) {
-			local->op_ret = op_ret;
-			local->success_count++;
-		}
+        local->call_count = call_count;
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_rmdir_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->rmdir,
+                                           &local->loc, local->cont.rmdir.flags,
+                                           NULL);
 
-	call_count = afr_frame_return (frame);
+                        if (!--call_count)
+                                break;
+                }
+        }
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        return 0;
 }
 
 
-int32_t
-afr_setdents_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_rmdir_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_setdents_wind_cbk,	
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->setdents,
-					   local->fd, local->cont.setdents.flags,
-					   local->cont.setdents.entries, 
-					   local->cont.setdents.count);
-			
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
-}
+        afr_local_t * local = frame->local;
 
+        local->transaction.unwind (frame, this);
 
-int32_t
-afr_setdents_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t * local = frame->local;
+        AFR_STACK_DESTROY (frame);
 
-	AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno);
-	
-	return 0;
+        return 0;
 }
 
 
-int32_t
-afr_setdents (call_frame_t *frame, xlator_t *this,
-	      fd_t *fd, int32_t flags, dir_entry_t *entries, int32_t count)
+int
+afr_rmdir (call_frame_t *frame, xlator_t *this,
+           loc_t *loc, int flags, dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	
-	int ret = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	frame->local = local;
-
-	local->fd = fd_ref (fd);
-
-	local->cont.setdents.flags   = flags;
-	local->cont.setdents.entries = entries;
-	local->cont.setdents.count   = count;
-
-	local->transaction.fop  = afr_setdents_wind;
-	local->transaction.done = afr_setdents_done;
-
-	local->transaction.basename = NULL;
-	local->transaction.pending  = AFR_ENTRY_PENDING;
+        afr_private_t           *priv                   = NULL;
+        afr_local_t             *local                  = NULL;
+        afr_internal_lock_t     *int_lock               = NULL;
+        call_frame_t            *transaction_frame      = NULL;
+        int                     ret                     = -1;
+        int                     op_errno                = 0;
+        int                     nlockee                 = 0;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv = this->private;
+
+        QUORUM_CHECK(rmdir,out);
+
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	afr_transaction (frame, this, AFR_ENTRY_TRANSACTION);
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        local->cont.rmdir.flags = flags;
+        loc_copy (&local->loc, loc);
+
+        local->op = GF_FOP_RMDIR;
+        local->transaction.fop    = afr_rmdir_wind;
+        local->transaction.done   = afr_rmdir_done;
+        local->transaction.unwind = afr_rmdir_unwind;
+
+        ret = afr_build_parent_loc (&local->transaction.parent_loc, loc,
+                                    &op_errno);
+        if (ret)
+                goto out;
+
+        local->transaction.main_frame = frame;
+        local->transaction.basename = AFR_BASENAME (loc->path);
+        int_lock = &local->internal_lock;
+
+        int_lock->lockee_count = nlockee = 0;
+        ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
+                                     &local->transaction.parent_loc,
+                                     local->transaction.basename,
+                                     priv->child_count);
+        if (ret)
+                goto out;
+
+        nlockee++;
+        ret = afr_init_entry_lockee (&int_lock->lockee[nlockee], local,
+                                     &local->loc,
+                                     NULL,
+                                     priv->child_count);
+        if (ret)
+                goto out;
+
+        nlockee++;
+        qsort (int_lock->lockee, nlockee, sizeof (*int_lock->lockee),
+               afr_entry_lockee_cmp);
+        int_lock->lockee_count = nlockee;
+
+        ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (rmdir, frame, -1, op_errno, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 /* }}} */
diff --git a/xlators/cluster/afr/src/afr-dir-write.h b/xlators/cluster/afr/src/afr-dir-write.h
index 76b4b60fa..02f0a3682 100644
--- a/xlators/cluster/afr/src/afr-dir-write.h
+++ b/xlators/cluster/afr/src/afr-dir-write.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __DIR_WRITE_H__
@@ -22,38 +13,35 @@
 
 int32_t
 afr_create (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t flags, mode_t mode, fd_t *fd);
+            loc_t *loc, int32_t flags, mode_t mode,
+            mode_t umask, fd_t *fd, dict_t *xdata);
 
 int32_t
 afr_mknod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dev_t dev);
+	   loc_t *loc, mode_t mode, dev_t dev, mode_t umask, dict_t *xdata);
 
 int32_t
 afr_mkdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode);
+	   loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata);
 
 int32_t
 afr_unlink (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc);
+	    loc_t *loc, int xflag, dict_t *xdata);
 
 int32_t
 afr_rmdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc);
+	   loc_t *loc, int flags, dict_t *xdata);
 
 int32_t
 afr_link (call_frame_t *frame, xlator_t *this,
-	  loc_t *oldloc, loc_t *newloc);
+	  loc_t *oldloc, loc_t *newloc, dict_t *xdata);
 
 int32_t
 afr_rename (call_frame_t *frame, xlator_t *this,
-	    loc_t *oldloc, loc_t *newloc);
+	    loc_t *oldloc, loc_t *newloc, dict_t *xdata);
 
-int32_t
+int
 afr_symlink (call_frame_t *frame, xlator_t *this,
-	     const char *linkpath, loc_t *oldloc);
-
-int32_t
-afr_setdents (call_frame_t *frame, xlator_t *this,
-	      fd_t *fd, int32_t flags, dir_entry_t *entries, int32_t count);
+	     const char *linkpath, loc_t *oldloc, mode_t umask, dict_t *params);
 
 #endif /* __DIR_WRITE_H__ */
diff --git a/xlators/cluster/afr/src/afr-inode-read.c b/xlators/cluster/afr/src/afr-inode-read.c
index c7a7e9987..e06e3b2f2 100644
--- a/xlators/cluster/afr/src/afr-inode-read.c
+++ b/xlators/cluster/afr/src/afr-inode-read.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 
@@ -44,9 +35,6 @@
 #include "compat-errno.h"
 #include "compat.h"
 
-#include "afr.h"
-
-
 /**
  * Common algorithm for inode read calls:
  *
@@ -61,114 +49,115 @@
 
 int32_t
 afr_access_cbk (call_frame_t *frame, void *cookie,
-		xlator_t *this, int32_t op_ret, int32_t op_errno)
+                xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
-
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-        int read_child = -1;
+        afr_private_t * priv            = NULL;
+        afr_local_t *   local           = NULL;
+        xlator_t **     children        = NULL;
+        int             unwind          = 1;
+        int32_t         *last_index     = NULL;
+        int32_t         next_call_child = -1;
+        int32_t         read_child      = -1;
+        int32_t         *fresh_children  = NULL;
 
-	priv     = this->private;
-	children = priv->children;
+        priv     = this->private;
+        children = priv->children;
 
-	local = frame->local;
+        local = frame->local;
 
         read_child = (long) cookie;
 
-	if (op_ret == -1) {
-        retry:
-		last_tried = local->cont.access.last_tried;
-
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try    = ++local->cont.access.last_tried;
-
-                if (this_try == read_child) {
-                        goto retry;
-                }
+        if (op_ret == -1) {
+                last_index = &local->cont.access.last_index;
+                fresh_children = local->fresh_children;
+                next_call_child = afr_next_call_child (fresh_children,
+                                                       local->child_up,
+                                                       priv->child_count,
+                                                       last_index, read_child);
+                if (next_call_child < 0)
+                        goto out;
 
-		unwind = 0;
+                unwind = 0;
 
-		STACK_WIND_COOKIE (frame, afr_access_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->access,
-				   &local->loc, local->cont.access.mask);
-	}
+                STACK_WIND_COOKIE (frame, afr_access_cbk,
+                                   (void *) (long) read_child,
+                                   children[next_call_child],
+                                   children[next_call_child]->fops->access,
+                                   &local->loc, local->cont.access.mask,
+                                   NULL);
+        }
 
 out:
-	if (unwind) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
+        if (unwind) {
+                AFR_STACK_UNWIND (access, frame, op_ret, op_errno, xdata);
+        }
 
-	return 0;
+        return 0;
 }
 
 
 int32_t
-afr_access (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t mask)
+afr_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+            dict_t *xdata)
 {
-	afr_private_t * priv       = NULL;
-	xlator_t **     children   = NULL;
-	int             call_child = 0;
-	afr_local_t     *local     = NULL;
-
+        afr_private_t   *priv      = NULL;
+        xlator_t        **children = NULL;
+        int             call_child = 0;
+        afr_local_t     *local     = NULL;
+        int32_t         op_errno   = 0;
         int32_t         read_child = -1;
+        int             ret        = -1;
 
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
+        priv     = this->private;
+        VALIDATE_OR_GOTO (priv->children, out);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        children = priv->children;
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+        AFR_SBRAIN_CHECK_LOC (loc, out);
 
-	children = priv->children;
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-        read_child = afr_read_child (this, loc->inode);
-        
-        if (read_child >= 0) {
-                call_child = read_child;
+        local->fresh_children = afr_children_create (priv->child_count);
+        if (!local->fresh_children) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-                local->cont.access.last_tried = -1;
 
-        } else {
-                call_child = afr_first_up_child (priv);
-                if (call_child == -1) {
-                        op_errno = ENOTCONN;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "no child is up :(");
-                        goto out;
-                }
-
-                local->cont.access.last_tried = call_child;
+        read_child = afr_inode_get_read_ctx (this, loc->inode,
+                                             local->fresh_children);
+        ret = afr_get_call_child (this, local->child_up, read_child,
+                                     local->fresh_children,
+                                     &call_child,
+                                     &local->cont.access.last_index);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
         }
 
-	loc_copy (&local->loc, loc);
-	local->cont.access.mask       = mask;
+        loc_copy (&local->loc, loc);
+        local->cont.access.mask = mask;
 
-	STACK_WIND_COOKIE (frame, afr_access_cbk,
-			   (void *) (long) call_child,
-			   children[call_child], children[call_child]->fops->access,
-			   loc, mask);
+        STACK_WIND_COOKIE (frame, afr_access_cbk,
+                           (void *) (long) call_child,
+                           children[call_child],
+                           children[call_child]->fops->access,
+                           loc, mask, xdata);
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-	return 0;
+        if (ret < 0)
+                AFR_STACK_UNWIND (access, frame, -1, op_errno, NULL);
+        return 0;
 }
 
 
@@ -178,485 +167,1629 @@ out:
 
 int32_t
 afr_stat_cbk (call_frame_t *frame, void *cookie,
-	      xlator_t *this, int32_t op_ret, int32_t op_errno,
-	      struct stat *buf)
+              xlator_t *this, int32_t op_ret, int32_t op_errno,
+              struct iatt *buf, dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+        afr_private_t * priv            = NULL;
+        afr_local_t *   local           = NULL;
+        xlator_t **     children        = NULL;
+        int             unwind          = 1;
+        int32_t         *last_index     = NULL;
+        int32_t         next_call_child = -1;
+        int32_t         read_child      = -1;
+        int32_t         *fresh_children  = NULL;
+
+        priv     = this->private;
+        children = priv->children;
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-	int read_child = -1;
+        read_child = (long) cookie;
 
-	priv     = this->private;
-	children = priv->children;
+        local = frame->local;
 
-	read_child = (long) cookie;
+        if (op_ret == -1) {
+                last_index = &local->cont.stat.last_index;
+                fresh_children = local->fresh_children;
+                next_call_child = afr_next_call_child (fresh_children,
+                                                       local->child_up,
+                                                       priv->child_count,
+                                                       last_index, read_child);
+                if (next_call_child < 0)
+                        goto out;
 
-	local = frame->local;
+                unwind = 0;
 
-	if (op_ret == -1) {
-	retry:
-		last_tried = local->cont.stat.last_tried;
+                STACK_WIND_COOKIE (frame, afr_stat_cbk,
+                                   (void *) (long) read_child,
+                                   children[next_call_child],
+                                   children[next_call_child]->fops->stat,
+                                   &local->loc, NULL);
+        }
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try = ++local->cont.stat.last_tried;
+out:
+        if (unwind) {
+                AFR_STACK_UNWIND (stat, frame, op_ret, op_errno, buf, xdata);
+        }
+
+        return 0;
+}
+
+
+int32_t
+afr_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+        afr_private_t   *priv      = NULL;
+        afr_local_t     *local     = NULL;
+        xlator_t        **children = NULL;
+        int             call_child = 0;
+        int32_t         op_errno   = 0;
+        int32_t         read_child = -1;
+        int             ret        = -1;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv     = this->private;
+        VALIDATE_OR_GOTO (priv->children, out);
+
+        children = priv->children;
 
-		if (this_try == read_child) {
-			goto retry;
-		}
+        AFR_SBRAIN_CHECK_LOC (loc, out);
 
-		unwind = 0;
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-		STACK_WIND_COOKIE (frame, afr_stat_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->stat,
-				   &local->loc);
-	}
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
+        local->fresh_children = afr_children_create (priv->child_count);
+        if (!local->fresh_children) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        read_child = afr_inode_get_read_ctx (this, loc->inode,
+                                             local->fresh_children);
+        ret = afr_get_call_child (this, local->child_up, read_child,
+                                     local->fresh_children,
+                                     &call_child,
+                                     &local->cont.stat.last_index);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
+        loc_copy (&local->loc, loc);
+
+        STACK_WIND_COOKIE (frame, afr_stat_cbk, (void *) (long) call_child,
+                           children[call_child],
+                           children[call_child]->fops->stat,
+                           loc, xdata);
+
+        ret = 0;
 out:
-	if (unwind) {
-		if (op_ret != -1)
-			buf->st_ino = local->cont.stat.ino;
+        if (ret < 0)
+                AFR_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+
+/* }}} */
 
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, buf);
-	}
+/* {{{ fstat */
+
+int32_t
+afr_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *buf,
+               dict_t *xdata)
+{
+        afr_private_t   *priv           = NULL;
+        afr_local_t     *local          = NULL;
+        xlator_t        **children      = NULL;
+        int             unwind          = 1;
+        int32_t         *last_index     = NULL;
+        int32_t         next_call_child = -1;
+        int32_t         read_child      = -1;
+        int32_t         *fresh_children  = NULL;
+
+        priv     = this->private;
+        children = priv->children;
+
+        local = frame->local;
+
+        read_child = (long) cookie;
+
+        if (op_ret == -1) {
+                last_index = &local->cont.fstat.last_index;
+                fresh_children = local->fresh_children;
+                next_call_child = afr_next_call_child (fresh_children,
+                                                       local->child_up,
+                                                       priv->child_count,
+                                                       last_index, read_child);
+                if (next_call_child < 0)
+                        goto out;
+
+                unwind = 0;
 
-	return 0;
+                STACK_WIND_COOKIE (frame, afr_fstat_cbk,
+                                   (void *) (long) read_child,
+                                   children[next_call_child],
+                                   children[next_call_child]->fops->fstat,
+                                   local->fd, NULL);
+        }
+
+out:
+        if (unwind) {
+                AFR_STACK_UNWIND (fstat, frame, op_ret, op_errno, buf, xdata);
+        }
+
+        return 0;
 }
 
 
 int32_t
-afr_stat (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc)
+afr_fstat (call_frame_t *frame, xlator_t *this,
+           fd_t *fd, dict_t *xdata)
 {
-	afr_private_t * priv       = NULL;
-	afr_local_t   * local      = NULL;
-	xlator_t **     children   = NULL;
+        afr_private_t   *priv      = NULL;
+        afr_local_t     *local     = NULL;
+        xlator_t        **children = NULL;
+        int             call_child = 0;
+        int32_t         op_errno   = 0;
+        int32_t         read_child = 0;
+        int             ret        = -1;
 
-        int32_t         read_child = -1;
-	int             call_child = 0;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (fd, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	int32_t         op_ret     = -1;
-	int32_t         op_errno   = 0;
+        priv     = this->private;
+        VALIDATE_OR_GOTO (priv->children, out);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        children = priv->children;
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+        VALIDATE_OR_GOTO (fd->inode, out);
 
-	children = priv->children;
+        AFR_SBRAIN_CHECK_FD (fd, out);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-	frame->local = local;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-        read_child = afr_read_child (this, loc->inode);
+        local->fresh_children = afr_children_create (priv->child_count);
+        if (!local->fresh_children) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-        if (read_child >= 0) {
-                call_child = read_child;
+        read_child = afr_inode_get_read_ctx (this, fd->inode,
+                                             local->fresh_children);
 
-                local->cont.stat.last_tried = -1;
 
-        } else {
-		call_child = afr_first_up_child (priv);
-		if (call_child == -1) {
-			op_errno = ENOTCONN;
-			gf_log (this->name, GF_LOG_ERROR,
-				"no child is up :(");
-			goto out;
-		}
 
-		local->cont.stat.last_tried = call_child;
-	}
+        ret = afr_get_call_child (this, local->child_up, read_child,
+                                     local->fresh_children,
+                                     &call_child,
+                                     &local->cont.fstat.last_index);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
 
-	loc_copy (&local->loc, loc);
+        local->fd = fd_ref (fd);
 
-	local->cont.stat.ino = loc->inode->ino;
+        afr_open_fd_fix (fd, this);
 
-	STACK_WIND_COOKIE (frame, afr_stat_cbk, (void *) (long) call_child,
-			   children[call_child],
-			   children[call_child]->fops->stat,
-			   loc);
+        STACK_WIND_COOKIE (frame, afr_fstat_cbk, (void *) (long) call_child,
+                           children[call_child],
+                           children[call_child]->fops->fstat,
+                           fd, xdata);
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+        if (ret < 0)
+                AFR_STACK_UNWIND (fstat, frame, -1, op_errno, NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
-
 /* }}} */
 
-/* {{{ fstat */
+/* {{{ readlink */
 
 int32_t
-afr_fstat_cbk (call_frame_t *frame, void *cookie,
-	       xlator_t *this, int32_t op_ret, int32_t op_errno,
-	       struct stat *buf)
+afr_readlink_cbk (call_frame_t *frame, void *cookie,
+                  xlator_t *this, int32_t op_ret, int32_t op_errno,
+                  const char *buf, struct iatt *sbuf, dict_t *xdata)
+{
+        afr_private_t * priv                  = NULL;
+        afr_local_t *   local                 = NULL;
+        xlator_t **     children              = NULL;
+        int             unwind                = 1;
+        int32_t         *last_index           = NULL;
+        int32_t         next_call_child       = -1;
+        int32_t         read_child            = -1;
+        int32_t         *fresh_children        = NULL;
+
+        priv     = this->private;
+        children = priv->children;
+
+        local = frame->local;
+
+        read_child = (long) cookie;
+
+        if (op_ret == -1) {
+                last_index = &local->cont.readlink.last_index;
+                fresh_children = local->fresh_children;
+                next_call_child = afr_next_call_child (fresh_children,
+                                                       local->child_up,
+                                                       priv->child_count,
+                                                       last_index, read_child);
+                if (next_call_child < 0)
+                        goto out;
+
+                unwind = 0;
+                STACK_WIND_COOKIE (frame, afr_readlink_cbk,
+                                   (void *) (long) read_child,
+                                   children[next_call_child],
+                                   children[next_call_child]->fops->readlink,
+                                   &local->loc,
+                                   local->cont.readlink.size, NULL);
+        }
+
+out:
+        if (unwind) {
+                AFR_STACK_UNWIND (readlink, frame, op_ret, op_errno, buf, sbuf,
+                                  xdata);
+        }
+
+        return 0;
+}
+
+
+int32_t
+afr_readlink (call_frame_t *frame, xlator_t *this,
+              loc_t *loc, size_t size, dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+        afr_private_t   *priv      = NULL;
+        xlator_t        **children = NULL;
+        int             call_child = 0;
+        afr_local_t     *local     = NULL;
+        int32_t         op_errno   = 0;
+        int32_t         read_child = -1;
+        int             ret        = -1;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-        int read_child = -1;
+        priv     = this->private;
+        VALIDATE_OR_GOTO (priv->children, out);
 
-	priv     = this->private;
-	children = priv->children;
+        children = priv->children;
 
-	local = frame->local;
+        AFR_SBRAIN_CHECK_LOC (loc, out);
 
-	read_child = (long) cookie;
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-	if (op_ret == -1) {
-	retry:
-		last_tried = local->cont.fstat.last_tried;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try   = ++local->cont.fstat.last_tried;
+        local->fresh_children = afr_children_create (priv->child_count);
+        if (!local->fresh_children) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+        read_child = afr_inode_get_read_ctx (this, loc->inode,
+                                             local->fresh_children);
+        ret = afr_get_call_child (this, local->child_up, read_child,
+                                     local->fresh_children,
+                                     &call_child,
+                                     &local->cont.readlink.last_index);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
 
-		if (this_try == read_child) {
-			goto retry;
-		}
+        loc_copy (&local->loc, loc);
 
-		unwind = 0;
+        local->cont.readlink.size       = size;
 
-		STACK_WIND_COOKIE (frame, afr_fstat_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->fstat,
-				   local->fd);
-	}
+        STACK_WIND_COOKIE (frame, afr_readlink_cbk,
+                           (void *) (long) call_child,
+                           children[call_child],
+                           children[call_child]->fops->readlink,
+                           loc, size, xdata);
 
+        ret = 0;
 out:
-	if (unwind) {
-		if (op_ret != -1)
-			buf->st_ino = local->cont.fstat.ino;
+        if (ret < 0)
+                AFR_STACK_UNWIND (readlink, frame, -1, op_errno, NULL, NULL, NULL);
+        return 0;
+}
+
+
+/* }}} */
+
+/* {{{ getxattr */
+
+struct _xattr_key {
+        char *key;
+        struct list_head list;
+};
 
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, buf);
-	}
 
-	return 0;
+int
+__gather_xattr_keys (dict_t *dict, char *key, data_t *value,
+                     void *data)
+{
+        struct list_head *  list  = data;
+        struct _xattr_key * xkey  = NULL;
+
+        if (!strncmp (key, AFR_XATTR_PREFIX,
+                      strlen (AFR_XATTR_PREFIX))) {
+
+                xkey = GF_CALLOC (1, sizeof (*xkey), gf_afr_mt_xattr_key);
+                if (!xkey)
+                        return -1;
+
+                xkey->key = key;
+                INIT_LIST_HEAD (&xkey->list);
+
+                list_add_tail (&xkey->list, list);
+        }
+        return 0;
 }
 
 
-int32_t
-afr_fstat (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd)
+void
+__filter_xattrs (dict_t *dict)
 {
-	afr_private_t * priv       = NULL;
-	afr_local_t   * local      = NULL;
-	xlator_t **     children   = NULL;
+        struct list_head   keys = {0,};
+        struct _xattr_key *key  = NULL;
+        struct _xattr_key *tmp  = NULL;
 
-	int             call_child = 0;
-        int32_t         read_child = -1;
+        INIT_LIST_HEAD (&keys);
 
-	int32_t         op_ret     = -1;
-	int32_t         op_errno   = 0;
+        dict_foreach (dict, __gather_xattr_keys,
+                      (void *) &keys);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (fd, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        list_for_each_entry_safe (key, tmp, &keys, list) {
+                dict_del (dict, key->key);
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+                list_del_init (&key->list);
 
-	children = priv->children;
+                GF_FREE (key);
+        }
+}
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
 
-	frame->local = local;
 
-	VALIDATE_OR_GOTO (fd->inode, out);
+int32_t
+afr_getxattr_cbk (call_frame_t *frame, void *cookie,
+                  xlator_t *this, int32_t op_ret, int32_t op_errno,
+                  dict_t *dict, dict_t *xdata)
+{
+        afr_private_t * priv            = NULL;
+        afr_local_t *   local           = NULL;
+        xlator_t **     children        = NULL;
+        int             unwind          = 1;
+        int32_t         *last_index     = NULL;
+        int32_t         next_call_child = -1;
+        int32_t         read_child      = -1;
+        int32_t         *fresh_children  = NULL;
 
-        read_child = afr_read_child (this, fd->inode);
+        priv     = this->private;
+        children = priv->children;
 
-        if (read_child >= 0) {
-                call_child = read_child;
+        local = frame->local;
 
-                local->cont.fstat.last_tried = -1;
-        } else {
-                call_child = afr_first_up_child (priv);
+        read_child = (long) cookie;
+
+        if (op_ret == -1) {
+                last_index = &local->cont.getxattr.last_index;
+                fresh_children = local->fresh_children;
+                next_call_child = afr_next_call_child (fresh_children,
+                                                       local->child_up,
+                                                       priv->child_count,
+                                                       last_index, read_child);
+                if (next_call_child < 0)
+                        goto out;
+
+                unwind = 0;
+                STACK_WIND_COOKIE (frame, afr_getxattr_cbk,
+                                   (void *) (long) read_child,
+                                   children[next_call_child],
+                                   children[next_call_child]->fops->getxattr,
+                                   &local->loc,
+                                   local->cont.getxattr.name,
+                                   NULL);
+        }
+
+out:
+        if (unwind) {
+                if (op_ret >= 0 && dict)
+                        __filter_xattrs (dict);
 
-                if (call_child == -1) {
-                        op_errno = ENOTCONN;
+                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, xdata);
+        }
+
+        return 0;
+}
+
+int32_t
+afr_getxattr_unwind (call_frame_t *frame, int op_ret, int op_errno,
+                     dict_t *dict, dict_t *xdata)
+
+{
+        AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, xdata);
+        return 0;
+}
+
+int32_t
+afr_fgetxattr_clrlk_cbk (call_frame_t *frame, void *cookie,
+                         xlator_t *this, int32_t op_ret, int32_t op_errno,
+                         dict_t *dict, dict_t *xdata)
+{
+        afr_local_t     *local                  = NULL;
+        afr_private_t   *priv                   = NULL;
+        xlator_t        **children              = NULL;
+        dict_t          *xattr                  = NULL;
+        char            *tmp_report             = NULL;
+        char            lk_summary[1024]        = {0,};
+        int             serz_len                = 0;
+        int32_t         callcnt                 = 0;
+        long int        cky                     = 0;
+        int             ret                     = 0;
+
+        priv     = this->private;
+        children = priv->children;
+
+        local = frame->local;
+        cky = (long) cookie;
+
+        LOCK (&frame->lock);
+        {
+                callcnt = --local->call_count;
+                if (op_ret == -1)
+                        local->child_errno[cky] = op_errno;
+
+                if (!local->dict)
+                        local->dict = dict_new ();
+                if (local->dict) {
+                        ret = dict_get_str (dict, local->cont.getxattr.name,
+                                            &tmp_report);
+                        if (ret)
+                                goto unlock;
+                        ret = dict_set_dynstr (local->dict,
+                                               children[cky]->name,
+                                               gf_strdup (tmp_report));
+                        if (ret)
+                                goto unlock;
+                }
+        }
+unlock:
+        UNLOCK (&frame->lock);
+
+        if (!callcnt) {
+                xattr = dict_new ();
+                if (!xattr) {
+                        op_ret = -1;
+                        op_errno = ENOMEM;
+                        goto unwind;
+                }
+                ret = dict_serialize_value_with_delim (local->dict,
+                                                       lk_summary,
+                                                       &serz_len, '\n');
+                if (ret) {
+                        op_ret = -1;
+                        op_errno = ENOMEM;
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Error serializing dictionary");
+                        goto unwind;
+                }
+                if (serz_len == -1)
+                        snprintf (lk_summary, sizeof (lk_summary),
+                                  "No locks cleared.");
+                ret = dict_set_dynstr (xattr, local->cont.getxattr.name,
+                                       gf_strdup (lk_summary));
+                if (ret) {
+                        op_ret = -1;
+                        op_errno = ENOMEM;
                         gf_log (this->name, GF_LOG_ERROR,
-				"no child is up :(");
-			goto out;
-		}
+                                "Error setting dictionary");
+                        goto unwind;
+                }
+
+        unwind:
+                // Updating child_errno with more recent 'events'
+                local->child_errno[cky] = op_errno;
+                op_errno = afr_resultant_errno_get (NULL, local->child_errno,
+                                                    priv->child_count);
+                AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, xattr,
+                                  xdata);
 
-                local->cont.fstat.last_tried = call_child;
+                if (xattr)
+                        dict_unref (xattr);
         }
 
-	local->cont.fstat.ino = fd->inode->ino;
-	local->fd = fd_ref (fd);
+        return ret;
+}
+
+int32_t
+afr_getxattr_clrlk_cbk (call_frame_t *frame, void *cookie,
+                        xlator_t *this, int32_t op_ret, int32_t op_errno,
+                        dict_t *dict, dict_t *xdata)
+{
+        afr_local_t     *local                  = NULL;
+        afr_private_t   *priv                   = NULL;
+        xlator_t        **children              = NULL;
+        dict_t          *xattr                  = NULL;
+        char            *tmp_report             = NULL;
+        char            lk_summary[1024]        = {0,};
+        int             serz_len                = 0;
+        int32_t         callcnt                 = 0;
+        long int        cky                     = 0;
+        int             ret                     = 0;
+
+        priv     = this->private;
+        children = priv->children;
+
+        local = frame->local;
+        cky = (long) cookie;
+
+        LOCK (&frame->lock);
+        {
+                callcnt = --local->call_count;
+                if (op_ret == -1)
+                        local->child_errno[cky] = op_errno;
+
+                if (!local->dict)
+                        local->dict = dict_new ();
+                if (local->dict) {
+                        ret = dict_get_str (dict, local->cont.getxattr.name,
+                                            &tmp_report);
+                        if (ret)
+                                goto unlock;
+                        ret = dict_set_dynstr (local->dict,
+                                               children[cky]->name,
+                                               gf_strdup (tmp_report));
+                        if (ret)
+                                goto unlock;
+                }
+        }
+unlock:
+        UNLOCK (&frame->lock);
+
+        if (!callcnt) {
+                xattr = dict_new ();
+                if (!xattr) {
+                        op_ret = -1;
+                        op_errno = ENOMEM;
+                        goto unwind;
+                }
+                ret = dict_serialize_value_with_delim (local->dict,
+                                                       lk_summary,
+                                                       &serz_len, '\n');
+                if (ret) {
+                        op_ret = -1;
+                        op_errno = ENOMEM;
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Error serializing dictionary");
+                        goto unwind;
+                }
+                if (serz_len == -1)
+                        snprintf (lk_summary, sizeof (lk_summary),
+                                  "No locks cleared.");
+                ret = dict_set_dynstr (xattr, local->cont.getxattr.name,
+                                       gf_strdup (lk_summary));
+                if (ret) {
+                        op_ret = -1;
+                        op_errno = ENOMEM;
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Error setting dictionary");
+                        goto unwind;
+                }
 
-	STACK_WIND_COOKIE (frame, afr_fstat_cbk, (void *) (long) call_child,
-			   children[call_child],
-			   children[call_child]->fops->fstat,
-			   fd);
+        unwind:
+                // Updating child_errno with more recent 'events'
+                local->child_errno[cky] = op_errno;
+                op_errno = afr_resultant_errno_get (NULL, local->child_errno,
+                                                    priv->child_count);
+                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr, xdata);
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+                if (xattr)
+                        dict_unref (xattr);
+        }
 
-	return 0;
+        return ret;
 }
 
-/* }}} */
+/**
+ * node-uuid cbk uses next child querying mechanism
+ */
+int32_t
+afr_getxattr_node_uuid_cbk (call_frame_t *frame, void *cookie,
+                            xlator_t *this, int32_t op_ret, int32_t op_errno,
+                            dict_t *dict, dict_t *xdata)
+{
+        afr_private_t  *priv            = NULL;
+        afr_local_t    *local           = NULL;
+        xlator_t      **children        = NULL;
+        int             unwind          = 1;
+        int             curr_call_child = 0;
+
+        priv = this->private;
+        children = priv->children;
+
+        local = frame->local;
+
+        if (op_ret == -1) { /** query the _next_ child */
+
+                /**
+                 * _current_ becomes _next_
+                 * If done with all childs and yet no success; give up !
+                 */
+                curr_call_child = (int) ((long)cookie);
+                if (++curr_call_child == priv->child_count)
+                        goto unwind;
+
+                gf_log (this->name, GF_LOG_WARNING,
+                        "op_ret (-1): Re-querying afr-child (%d/%d)",
+                        curr_call_child, priv->child_count);
+
+                unwind = 0;
+                STACK_WIND_COOKIE (frame, afr_getxattr_node_uuid_cbk,
+                                   (void *) (long) curr_call_child,
+                                   children[curr_call_child],
+                                   children[curr_call_child]->fops->getxattr,
+                                   &local->loc,
+                                   local->cont.getxattr.name,
+                                   NULL);
+        }
 
-/* {{{ readlink */
+ unwind:
+        if (unwind)
+                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict,
+                                  NULL);
+
+        return 0;
+}
 
 int32_t
-afr_readlink_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno,
-		  const char *buf)
+afr_getxattr_lockinfo_cbk (call_frame_t *frame, void *cookie,
+                           xlator_t *this, int32_t op_ret, int32_t op_errno,
+                           dict_t *dict, dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+        int          call_cnt     = 0, len = 0;
+        char        *lockinfo_buf = NULL;
+        dict_t      *lockinfo     = NULL, *newdict = NULL;
+        afr_local_t *local        = NULL;
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-        int read_child = -1;
+        LOCK (&frame->lock);
+        {
+                local = frame->local;
 
-	priv     = this->private;
-	children = priv->children;
+                call_cnt = --local->call_count;
 
-	local = frame->local;
+                if ((op_ret < 0) || (!dict && !xdata)) {
+                        goto unlock;
+                }
 
-        read_child = (long) cookie;
+                if (xdata) {
+                        if (!local->xdata_rsp) {
+                                local->xdata_rsp = dict_new ();
+                                if (!local->xdata_rsp) {
+                                        local->op_ret = -1;
+                                        local->op_errno = ENOMEM;
+                                        goto unlock;
+                                }
+                        }
+                }
 
-	if (op_ret == -1) {
-        retry:
-		last_tried = local->cont.readlink.last_tried;
+                if (!dict) {
+                        goto unlock;
+                }
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try = ++local->cont.readlink.last_tried;
+                op_ret = dict_get_ptr_and_len (dict, GF_XATTR_LOCKINFO_KEY,
+                                               (void **)&lockinfo_buf, &len);
 
-                if (this_try == read_child) {
-                        goto retry;
+                if (!lockinfo_buf) {
+                        goto unlock;
                 }
 
-		unwind = 0;
-		STACK_WIND_COOKIE (frame, afr_readlink_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->readlink,
-				   &local->loc,
-				   local->cont.readlink.size);
-	}
+                if (!local->dict) {
+                        local->dict = dict_new ();
+                        if (!local->dict) {
+                                local->op_ret = -1;
+                                local->op_errno = ENOMEM;
+                                goto unlock;
+                        }
+                }
+        }
+unlock:
+        UNLOCK (&frame->lock);
+
+        if (lockinfo_buf != NULL) {
+                lockinfo = dict_new ();
+                if (lockinfo == NULL) {
+                        local->op_ret = -1;
+                        local->op_errno = ENOMEM;
+                } else {
+                        op_ret = dict_unserialize (lockinfo_buf, len,
+                                                   &lockinfo);
+
+                        if (lockinfo && local->dict) {
+                                dict_copy (lockinfo, local->dict);
+                        }
+                }
+        }
 
-out:
-	if (unwind) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, buf);
-	}
+        if (xdata && local->xdata_rsp) {
+                dict_copy (xdata, local->xdata_rsp);
+        }
 
-	return 0;
-}
+        if (!call_cnt) {
+                newdict = dict_new ();
+                if (!newdict) {
+                        local->op_ret = -1;
+                        local->op_errno = ENOMEM;
+                        goto unwind;
+                }
 
+                len = dict_serialized_length (local->dict);
+                if (len == 0) {
+                        goto unwind;
+                }
+
+                lockinfo_buf = GF_CALLOC (1, len, gf_common_mt_char);
+                if (!lockinfo_buf) {
+                        local->op_ret = -1;
+                        local->op_errno = ENOMEM;
+                        goto unwind;
+                }
+
+                op_ret = dict_serialize (local->dict, lockinfo_buf);
+                if (op_ret < 0) {
+                        local->op_ret = -1;
+                        local->op_errno = -op_ret;
+                }
+
+                op_ret = dict_set_dynptr (newdict, GF_XATTR_LOCKINFO_KEY,
+                                          (void *)lockinfo_buf, len);
+                if (op_ret < 0) {
+                        local->op_ret = -1;
+                        local->op_errno = -op_ret;
+                        goto unwind;
+                }
+
+        unwind:
+                AFR_STACK_UNWIND (getxattr, frame, op_ret,
+                                  op_errno, newdict,
+                                  local->xdata_rsp);
+        }
+
+        dict_unref (lockinfo);
+
+        return 0;
+}
 
 int32_t
-afr_readlink (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, size_t size)
+afr_fgetxattr_lockinfo_cbk (call_frame_t *frame, void *cookie,
+                            xlator_t *this, int32_t op_ret, int32_t op_errno,
+                            dict_t *dict, dict_t *xdata)
 {
-	afr_private_t * priv       = NULL;
-	xlator_t **     children   = NULL;
-	int             call_child = 0;
-	afr_local_t     *local     = NULL;
+        int          call_cnt     = 0, len = 0;
+        char        *lockinfo_buf = NULL;
+        dict_t      *lockinfo     = NULL, *newdict = NULL;
+        afr_local_t *local        = NULL;
 
-        int32_t         read_child = -1;
+        LOCK (&frame->lock);
+        {
+                local = frame->local;
 
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
+                call_cnt = --local->call_count;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+                if ((op_ret < 0) || (!dict && !xdata)) {
+                        goto unlock;
+                }
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+                if (xdata) {
+                        if (!local->xdata_rsp) {
+                                local->xdata_rsp = dict_new ();
+                                if (!local->xdata_rsp) {
+                                        local->op_ret = -1;
+                                        local->op_errno = ENOMEM;
+                                        goto unlock;
+                                }
+                        }
+                }
 
-	children = priv->children;
+                if (!dict) {
+                        goto unlock;
+                }
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+                op_ret = dict_get_ptr_and_len (dict, GF_XATTR_LOCKINFO_KEY,
+                                               (void **)&lockinfo_buf, &len);
 
-	frame->local = local;
+                if (!lockinfo_buf) {
+                        goto unlock;
+                }
 
-        read_child = afr_read_child (this, loc->inode);
+                if (!local->dict) {
+                        local->dict = dict_new ();
+                        if (!local->dict) {
+                                local->op_ret = -1;
+                                local->op_errno = ENOMEM;
+                                goto unlock;
+                        }
+                }
+        }
+unlock:
+        UNLOCK (&frame->lock);
+
+        if (lockinfo_buf != NULL) {
+                lockinfo = dict_new ();
+                if (lockinfo == NULL) {
+                        local->op_ret = -1;
+                        local->op_errno = ENOMEM;
+                } else {
+                        op_ret = dict_unserialize (lockinfo_buf, len,
+                                                   &lockinfo);
+
+                        if (lockinfo && local->dict) {
+                                dict_copy (lockinfo, local->dict);
+                        }
+                }
+        }
 
-        if (read_child >= 0) {
-                call_child = read_child;
+        if (xdata && local->xdata_rsp) {
+                dict_copy (xdata, local->xdata_rsp);
+        }
 
-                local->cont.readlink.last_tried = -1;
+        if (!call_cnt) {
+                newdict = dict_new ();
+                if (!newdict) {
+                        local->op_ret = -1;
+                        local->op_errno = ENOMEM;
+                        goto unwind;
+                }
 
-        } else {
-                call_child = afr_first_up_child (priv);
+                len = dict_serialized_length (local->dict);
+                if (len <= 0) {
+                        goto unwind;
+                }
 
-                if (call_child == -1) {
-                        op_errno = ENOTCONN;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "no child is up :(");
-                        goto out;
+                lockinfo_buf = GF_CALLOC (1, len, gf_common_mt_char);
+                if (!lockinfo_buf) {
+                        local->op_ret = -1;
+                        local->op_errno = ENOMEM;
+                        goto unwind;
                 }
 
-                local->cont.readlink.last_tried = call_child;
+                op_ret = dict_serialize (local->dict, lockinfo_buf);
+                if (op_ret < 0) {
+                        local->op_ret = -1;
+                        local->op_errno = -op_ret;
+                }
+
+                op_ret = dict_set_dynptr (newdict, GF_XATTR_LOCKINFO_KEY,
+                                          (void *)lockinfo_buf, len);
+                if (op_ret < 0) {
+                        local->op_ret = -1;
+                        local->op_errno = -op_ret;
+                        goto unwind;
+                }
+
+        unwind:
+                AFR_STACK_UNWIND (fgetxattr, frame, op_ret,
+                                  op_errno, newdict,
+                                  local->xdata_rsp);
         }
 
-	loc_copy (&local->loc, loc);
-	local->cont.readlink.size       = size;
+        dict_unref (lockinfo);
 
-	STACK_WIND_COOKIE (frame, afr_readlink_cbk,
-			   (void *) (long) call_child,
-			   children[call_child], children[call_child]->fops->readlink,
-			   loc, size);
+        return 0;
+}
 
-	op_ret = 0;
+int32_t
+afr_fgetxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
+                            xlator_t *this, int32_t op_ret, int32_t op_errno,
+                            dict_t *dict, dict_t *xdata)
+{
+        afr_local_t *local          = NULL;
+        int32_t      callcnt        = 0;
+        int          ret            = 0;
+        char        *xattr          = NULL;
+        char        *xattr_serz     = NULL;
+        char        xattr_cky[1024] = {0,};
+        dict_t      *nxattr         = NULL;
+        long         cky            = 0;
+        int32_t      padding        = 0;
+        int32_t      tlen           = 0;
+
+        if (!frame || !frame->local || !this) {
+                gf_log ("", GF_LOG_ERROR, "possible NULL deref");
+                goto out;
+        }
+
+        local = frame->local;
+        cky = (long) cookie;
+
+        LOCK (&frame->lock);
+        {
+                callcnt = --local->call_count;
+
+                if (!dict || (op_ret < 0))
+                        goto out;
+
+                if (!local->dict)
+                        local->dict = dict_new ();
+
+                if (local->dict) {
+                        ret = dict_get_str (dict,
+                                            local->cont.getxattr.name,
+                                            &xattr);
+                        if (ret)
+                                goto out;
+
+                        xattr = gf_strdup (xattr);
+
+                        (void)snprintf (xattr_cky, 1024, "%s-%ld",
+                                        local->cont.getxattr.name, cky);
+                        ret = dict_set_dynstr (local->dict,
+                                               xattr_cky, xattr);
+                        if (ret) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "Cannot set xattr cookie key");
+                                goto out;
+                        }
+
+                        local->cont.getxattr.xattr_len
+                                += strlen (xattr) + 1;
+                }
+        }
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
-	return 0;
-}
+        UNLOCK (&frame->lock);
+
+        if (!callcnt) {
+                if (!local->cont.getxattr.xattr_len)
+                        goto unwind;
+
+                nxattr = dict_new ();
+                if (!nxattr)
+                        goto unwind;
+
+                /* extra bytes for decorations (brackets and <>'s) */
+                padding += strlen (this->name)
+                        + strlen (AFR_PATHINFO_HEADER) + 4;
+                local->cont.getxattr.xattr_len += (padding + 2);
+
+                xattr_serz = GF_CALLOC (local->cont.getxattr.xattr_len,
+                                        sizeof (char), gf_common_mt_char);
+
+                if (!xattr_serz)
+                        goto unwind;
+
+                /* the xlator info */
+                (void) sprintf (xattr_serz, "(<"AFR_PATHINFO_HEADER"%s> ",
+                                this->name);
+
+                /* actual series of pathinfo */
+                ret = dict_serialize_value_with_delim (local->dict,
+                                                       xattr_serz
+                                                       + strlen (xattr_serz),
+                                                       &tlen, ' ');
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR, "Error serializing"
+                                " dictionary");
+                        goto unwind;
+                }
 
+                /* closing part */
+                *(xattr_serz + padding + tlen) = ')';
+                *(xattr_serz + padding + tlen + 1) = '\0';
 
-/* }}} */
+                ret = dict_set_dynstr (nxattr, local->cont.getxattr.name,
+                                       xattr_serz);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR, "Cannot set pathinfo"
+                                " key in dict");
 
-/* {{{ getxattr */
+        unwind:
+                AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, nxattr,
+                                  xdata);
+
+                if (nxattr)
+                        dict_unref (nxattr);
+        }
+
+        return ret;
+}
 
 int32_t
-afr_getxattr_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno,
-		  dict_t *dict)
+afr_getxattr_pathinfo_cbk (call_frame_t *frame, void *cookie,
+                           xlator_t *this, int32_t op_ret, int32_t op_errno,
+                           dict_t *dict, dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+        afr_local_t *local          = NULL;
+        int32_t      callcnt        = 0;
+        int          ret            = 0;
+        char        *xattr          = NULL;
+        char        *xattr_serz     = NULL;
+        char        xattr_cky[1024] = {0,};
+        dict_t      *nxattr         = NULL;
+        long         cky            = 0;
+        int32_t      padding        = 0;
+        int32_t      tlen           = 0;
+
+        if (!frame || !frame->local || !this) {
+                gf_log ("", GF_LOG_ERROR, "possible NULL deref");
+                goto out;
+        }
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-        int read_child = -1;
+        local = frame->local;
+        cky = (long) cookie;
 
-	priv     = this->private;
-	children = priv->children;
+        LOCK (&frame->lock);
+                {
+                        callcnt = --local->call_count;
 
-	local = frame->local;
+                        if (!dict || (op_ret < 0))
+                                goto out;
 
-        read_child = (long) cookie;
+                        if (!local->dict)
+                                local->dict = dict_new ();
 
-	if (op_ret == -1) {
-        retry:
-		last_tried = local->cont.getxattr.last_tried;
+                        if (local->dict) {
+                                ret = dict_get_str (dict,
+                                                    local->cont.getxattr.name,
+                                                    &xattr);
+                                if (ret)
+                                        goto out;
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try = ++local->cont.getxattr.last_tried;
+                                xattr = gf_strdup (xattr);
 
-                if (this_try == read_child) {
-                        goto retry;
+                                (void)snprintf (xattr_cky, 1024, "%s-%ld",
+                                                local->cont.getxattr.name, cky);
+                                ret = dict_set_dynstr (local->dict,
+                                                       xattr_cky, xattr);
+                                if (ret) {
+                                        gf_log (this->name, GF_LOG_ERROR,
+                                                "Cannot set xattr cookie key");
+                                        goto out;
+                                }
+
+                                local->cont.getxattr.xattr_len += strlen (xattr) + 1;
+                        }
+                }
+ out:
+        UNLOCK (&frame->lock);
+
+        if (!callcnt) {
+                if (!local->cont.getxattr.xattr_len)
+                        goto unwind;
+
+                nxattr = dict_new ();
+                if (!nxattr)
+                        goto unwind;
+
+                /* extra bytes for decorations (brackets and <>'s) */
+                padding += strlen (this->name) + strlen (AFR_PATHINFO_HEADER) + 4;
+                local->cont.getxattr.xattr_len += (padding + 2);
+
+                xattr_serz = GF_CALLOC (local->cont.getxattr.xattr_len,
+                                        sizeof (char), gf_common_mt_char);
+
+                if (!xattr_serz)
+                        goto unwind;
+
+                /* the xlator info */
+                (void) sprintf (xattr_serz, "(<"AFR_PATHINFO_HEADER"%s> ",
+                                this->name);
+
+                /* actual series of pathinfo */
+                ret = dict_serialize_value_with_delim (local->dict,
+                                                       xattr_serz + strlen (xattr_serz),
+                                                       &tlen, ' ');
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR, "Error serializing"
+                                " dictionary");
+                        goto unwind;
                 }
 
-		unwind = 0;
-		STACK_WIND_COOKIE (frame, afr_getxattr_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->getxattr,
-				   &local->loc,
-				   local->cont.getxattr.name);
-	}
+                /* closing part */
+                *(xattr_serz + padding + tlen) = ')';
+                *(xattr_serz + padding + tlen + 1) = '\0';
 
-out:
-	if (unwind) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, dict);
-	}
+                ret = dict_set_dynstr (nxattr, local->cont.getxattr.name,
+                                       xattr_serz);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR, "Cannot set pathinfo"
+                                " key in dict");
+
+        unwind:
+                AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, nxattr,
+                                  xdata);
+
+                if (nxattr)
+                        dict_unref (nxattr);
+        }
 
-	return 0;
+        return ret;
 }
 
+static int
+afr_aggregate_stime_xattr (dict_t *this, char *key, data_t *value, void *data)
+{
+        int ret = 0;
+
+        if (fnmatch (GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0)
+                ret = gf_get_min_stime (THIS, data, key, value);
+
+        return ret;
+}
 
 int32_t
-afr_getxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, const char *name)
+afr_common_getxattr_stime_cbk (call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               dict_t *dict, dict_t *xdata)
 {
-	afr_private_t *   priv       = NULL;
-	xlator_t **       children   = NULL;
-	int               call_child = 0;
-	afr_local_t     * local      = NULL;
+        afr_local_t *local          = NULL;
+        int32_t      callcnt        = 0;
 
-        int               read_child = -1;
+        if (!frame || !frame->local || !this) {
+                gf_log ("", GF_LOG_ERROR, "possible NULL deref");
+                goto out;
+        }
 
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
+        local = frame->local;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        LOCK (&frame->lock);
+        {
+                callcnt = --local->call_count;
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+                if (!dict || (op_ret < 0)) {
+                        local->op_errno = op_errno;
+                        goto cleanup;
+                }
 
-	children = priv->children;
+                if (!local->dict)
+                        local->dict = dict_copy_with_ref (dict, NULL);
+                else
+                        dict_foreach (dict, afr_aggregate_stime_xattr,
+                                      local->dict);
+                local->op_ret = 0;
+        }
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	frame->local = local;
+cleanup:
+        UNLOCK (&frame->lock);
 
-        read_child = afr_read_child (this, loc->inode);
+        if (!callcnt) {
+                AFR_STACK_UNWIND (getxattr, frame, local->op_ret,
+                                  local->op_errno, local->dict, xdata);
+        }
 
-        if (read_child >= 0) {
-                call_child = read_child;
+out:
+        return 0;
+}
+
+
+static gf_boolean_t
+afr_is_special_xattr (const char *name, fop_getxattr_cbk_t *cbk,
+                      gf_boolean_t is_fgetxattr)
+{
+        gf_boolean_t    is_spl = _gf_true;
+
+        GF_ASSERT (cbk);
+        if (!cbk) {
+                is_spl = _gf_false;
+                goto out;
+        }
 
-                local->cont.getxattr.last_tried = -1;
+        if (!strcmp (name, GF_XATTR_PATHINFO_KEY)) {
+                if (is_fgetxattr) {
+                        *cbk = afr_fgetxattr_pathinfo_cbk;
+                } else {
+                        *cbk = afr_getxattr_pathinfo_cbk;
+                }
+        } else if (!strncmp (name, GF_XATTR_CLRLK_CMD,
+                             strlen (GF_XATTR_CLRLK_CMD))) {
+                if (is_fgetxattr) {
+                        *cbk = afr_fgetxattr_clrlk_cbk;
+                } else {
+                        *cbk = afr_getxattr_clrlk_cbk;
+                }
+        } else if (!strncmp (name, GF_XATTR_LOCKINFO_KEY,
+                             strlen (GF_XATTR_LOCKINFO_KEY))) {
+                if (is_fgetxattr) {
+                        *cbk = afr_fgetxattr_lockinfo_cbk;
+                } else {
+                        *cbk = afr_getxattr_lockinfo_cbk;
+                }
+        } else if (fnmatch (GF_XATTR_STIME_PATTERN, name, FNM_NOESCAPE) == 0) {
+                *cbk = afr_common_getxattr_stime_cbk;
         } else {
-                call_child = afr_first_up_child (priv);
+                is_spl = _gf_false;
+        }
 
-                if (call_child == -1) {
-                        op_errno = ENOTCONN;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "no child is up :(");
+out:
+        return is_spl;
+}
+
+static void
+afr_getxattr_frm_all_children (xlator_t *this, call_frame_t *frame,
+                               const char *name, loc_t *loc,
+                               fop_getxattr_cbk_t cbk)
+{
+        afr_private_t   *priv           = NULL;
+        afr_local_t     *local          = NULL;
+        xlator_t        **children      = NULL;
+        int             i               = 0;
+
+        priv     = this->private;
+        children = priv->children;
+
+        local = frame->local;
+        local->call_count = priv->child_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                STACK_WIND_COOKIE (frame, cbk,
+                                   (void *) (long) i,
+                                   children[i], children[i]->fops->getxattr,
+                                   loc, name, NULL);
+        }
+        return;
+}
+
+int32_t
+afr_getxattr (call_frame_t *frame, xlator_t *this,
+              loc_t *loc, const char *name, dict_t *xdata)
+{
+        afr_private_t           *priv         = NULL;
+        xlator_t                **children    = NULL;
+        int                     call_child    = 0;
+        afr_local_t             *local        = NULL;
+        xlator_list_t           *trav         = NULL;
+        xlator_t                **sub_volumes = NULL;
+        int                     i             = 0;
+        int32_t                 op_errno      = 0;
+        int32_t                 read_child    = -1;
+        int                     ret           = -1;
+        fop_getxattr_cbk_t      cbk           = NULL;
+        int                     afr_xtime_gauge[MCNT_MAX] = {0,};
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv     = this->private;
+        VALIDATE_OR_GOTO (priv->children, out);
+
+        children = priv->children;
+
+        AFR_SBRAIN_CHECK_LOC (loc, out);
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        loc_copy (&local->loc, loc);
+        if (!name)
+                goto no_name;
+
+        local->cont.getxattr.name = gf_strdup (name);
+
+        if (!strncmp (name, AFR_XATTR_PREFIX,
+                      strlen (AFR_XATTR_PREFIX))) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "%s: no data present for key %s",
+                        loc->path, name);
+                op_errno = ENODATA;
+                goto out;
+        }
+        if ((strcmp (GF_XATTR_MARKER_KEY, name) == 0)
+            && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
+
+                local->marker.call_count = priv->child_count;
+
+                sub_volumes = alloca ( priv->child_count * sizeof (xlator_t *));
+                for (i = 0, trav = this->children; trav ;
+                     trav = trav->next, i++) {
+
+                        *(sub_volumes + i)  = trav->xlator;
+                }
+
+                if (cluster_getmarkerattr (frame, this, loc, name,
+                                           local, afr_getxattr_unwind,
+                                           sub_volumes,
+                                           priv->child_count,
+                                           MARKER_UUID_TYPE,
+                                           marker_uuid_default_gauge,
+                                           priv->vol_uuid)) {
+
+                        gf_log (this->name, GF_LOG_INFO,
+                                "%s: failed to get marker attr (%s)",
+                                loc->path, name);
+                        op_errno = EINVAL;
                         goto out;
                 }
 
-                local->cont.getxattr.last_tried = call_child;
+                return 0;
         }
 
-	loc_copy (&local->loc, loc);
-	if (name)
-	  local->cont.getxattr.name       = strdup (name);
+        /*
+         * if we are doing getxattr with pathinfo as the key then we
+         * collect information from all childs
+         */
+        if (afr_is_special_xattr (name, &cbk, 0)) {
+                afr_getxattr_frm_all_children (this, frame, name,
+                                               loc, cbk);
+                return 0;
+        }
+
+        if (XATTR_IS_NODE_UUID (name)) {
+                i = 0;
+                STACK_WIND_COOKIE (frame, afr_getxattr_node_uuid_cbk,
+                                   (void *) (long) i,
+                                   children[i],
+                                   children[i]->fops->getxattr,
+                                   loc, name, xdata);
+                return 0;
+        }
 
-	STACK_WIND_COOKIE (frame, afr_getxattr_cbk,
-			   (void *) (long) call_child,
-			   children[call_child], children[call_child]->fops->getxattr,
-			   loc, name);
+        if (*priv->vol_uuid) {
+                if ((match_uuid_local (name, priv->vol_uuid) == 0)
+                    && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
+                        local->marker.call_count = priv->child_count;
+
+                        sub_volumes = alloca ( priv->child_count
+                                               * sizeof (xlator_t *));
+                        for (i = 0, trav = this->children; trav ;
+                             trav = trav->next, i++) {
+
+                                *(sub_volumes + i)  = trav->xlator;
+
+                        }
+
+                        /* don't err out on getting ENOTCONN (brick down)
+                         * from a subset of the bricks
+                         */
+                        memcpy (afr_xtime_gauge, marker_xtime_default_gauge,
+                                sizeof (afr_xtime_gauge));
+                        afr_xtime_gauge[MCNT_NOTFOUND] = 0;
+                        afr_xtime_gauge[MCNT_ENOTCONN] = 0;
+                        if (cluster_getmarkerattr (frame, this, loc,
+                                                   name, local,
+                                                   afr_getxattr_unwind,
+                                                   sub_volumes,
+                                                   priv->child_count,
+                                                   MARKER_XTIME_TYPE,
+                                                   afr_xtime_gauge,
+                                                   priv->vol_uuid)) {
+                                gf_log (this->name, GF_LOG_INFO,
+                                        "%s: failed to get marker attr (%s)",
+                                        loc->path, name);
+                                op_errno = EINVAL;
+                                goto out;
+                        }
+
+                        return 0;
+                }
+        }
 
-	op_ret = 0;
+no_name:
+        local->fresh_children = afr_children_create (priv->child_count);
+        if (!local->fresh_children) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        read_child = afr_inode_get_read_ctx (this, loc->inode,
+                                             local->fresh_children);
+        ret = afr_get_call_child (this, local->child_up, read_child,
+                                     local->fresh_children,
+                                     &call_child,
+                                     &local->cont.getxattr.last_index);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
+
+        STACK_WIND_COOKIE (frame, afr_getxattr_cbk,
+                           (void *) (long) call_child,
+                           children[call_child],
+                           children[call_child]->fops->getxattr,
+                           loc, name, xdata);
+
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
-	return 0;
+        if (ret < 0)
+                AFR_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
+
+/* {{{ fgetxattr */
+
+
+int32_t
+afr_fgetxattr_cbk (call_frame_t *frame, void *cookie,
+                   xlator_t *this, int32_t op_ret, int32_t op_errno,
+                   dict_t *dict, dict_t *xdata)
+{
+        afr_private_t * priv            = NULL;
+        afr_local_t *   local           = NULL;
+        xlator_t **     children        = NULL;
+        int             unwind          = 1;
+        int32_t         *last_index     = NULL;
+        int32_t         next_call_child = -1;
+        int32_t         read_child      = -1;
+        int32_t         *fresh_children  = NULL;
+
+        priv     = this->private;
+        children = priv->children;
+
+        local = frame->local;
+
+        read_child = (long) cookie;
+
+        if (op_ret == -1) {
+                last_index = &local->cont.getxattr.last_index;
+                fresh_children = local->fresh_children;
+                next_call_child = afr_next_call_child (fresh_children,
+                                                       local->child_up,
+                                                       priv->child_count,
+                                                       last_index, read_child);
+                if (next_call_child < 0)
+                        goto out;
+
+                unwind = 0;
+                STACK_WIND_COOKIE (frame, afr_fgetxattr_cbk,
+                                   (void *) (long) read_child,
+                                   children[next_call_child],
+                                   children[next_call_child]->fops->fgetxattr,
+                                   local->fd,
+                                   local->cont.getxattr.name,
+                                   NULL);
+        }
+
+out:
+        if (unwind) {
+                if (op_ret >= 0 && dict)
+                        __filter_xattrs (dict);
+
+                AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, dict,
+                                  xdata);
+        }
+
+        return 0;
+}
+
+int32_t
+afr_fgetxattr_unwind (call_frame_t *frame,
+                      int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
+
+{
+        AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, dict, xdata);
+        return 0;
+}
+
+static void
+afr_fgetxattr_frm_all_children (xlator_t *this, call_frame_t *frame,
+                                const char *name, fd_t *fd,
+                                fop_fgetxattr_cbk_t cbk)
+{
+        afr_private_t   *priv           = NULL;
+        afr_local_t     *local          = NULL;
+        xlator_t        **children      = NULL;
+        int             i               = 0;
+
+        priv     = this->private;
+        children = priv->children;
+
+        local = frame->local;
+        local->call_count = priv->child_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                STACK_WIND_COOKIE (frame, cbk,
+                                   (void *) (long) i,
+                                   children[i], children[i]->fops->fgetxattr,
+                                   fd, name, NULL);
+        }
+
+        return;
+}
+
+int32_t
+afr_fgetxattr (call_frame_t *frame, xlator_t *this,
+               fd_t *fd, const char *name, dict_t *xdata)
+{
+        afr_private_t        *priv       = NULL;
+        xlator_t            **children   = NULL;
+        int                   call_child = 0;
+        afr_local_t          *local      = NULL;
+        int32_t               op_ret     = -1;
+        int32_t               op_errno   = 0;
+        int32_t               read_child = -1;
+        fop_fgetxattr_cbk_t   cbk        = NULL;
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+
+        priv     = this->private;
+        VALIDATE_OR_GOTO (priv->children, out);
+
+        children = priv->children;
+
+        AFR_SBRAIN_CHECK_FD (fd, out);
+
+        AFR_LOCAL_ALLOC_OR_GOTO (local, out);
+        frame->local = local;
+
+        op_ret = afr_local_init (local, priv, &op_errno);
+        if (op_ret < 0) {
+                op_errno = -op_ret;
+                goto out;
+        }
+
+        local->fd = fd_ref (fd);
+        if (name)
+                local->cont.getxattr.name = gf_strdup (name);
+
+        /* pathinfo gets handled only in getxattr(), but we need to handle
+         * lockinfo.
+         * If we are doing fgetxattr with lockinfo as the key then we
+         * collect information from all children.
+         */
+        if (afr_is_special_xattr (name, &cbk, 1)) {
+                afr_fgetxattr_frm_all_children (this, frame, name,
+                                                fd, cbk);
+                return 0;
+        }
+
+
+        local->fresh_children = afr_children_create (priv->child_count);
+        if (!local->fresh_children) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        read_child = afr_inode_get_read_ctx (this, fd->inode,
+                                             local->fresh_children);
+        op_ret = afr_get_call_child (this, local->child_up, read_child,
+                                     local->fresh_children,
+                                     &call_child,
+                                     &local->cont.getxattr.last_index);
+        if (op_ret < 0) {
+                op_errno = -op_ret;
+                op_ret = -1;
+                goto out;
+        }
+
+        STACK_WIND_COOKIE (frame, afr_fgetxattr_cbk,
+                           (void *) (long) call_child,
+                           children[call_child],
+                           children[call_child]->fops->fgetxattr,
+                           fd, name, xdata);
+
+        op_ret = 0;
+out:
+        if (op_ret == -1) {
+                AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, NULL,
+                                  NULL);
+        }
+        return 0;
 }
 
 
@@ -678,133 +1811,130 @@ out:
 
 int32_t
 afr_readv_cbk (call_frame_t *frame, void *cookie,
-	       xlator_t *this, int32_t op_ret, int32_t op_errno,
-	       struct iovec *vector, int32_t count, struct stat *buf)
+               xlator_t *this, int32_t op_ret, int32_t op_errno,
+               struct iovec *vector, int32_t count, struct iatt *buf,
+               struct iobref *iobref, dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
-
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try   = -1;
-        int read_child = -1;
+        afr_private_t * priv            = NULL;
+        afr_local_t *   local           = NULL;
+        xlator_t **     children        = NULL;
+        int             unwind          = 1;
+        int32_t         *last_index     = NULL;
+        int32_t         next_call_child = -1;
+        int32_t         *fresh_children  = NULL;
+        int32_t         read_child      = -1;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+        priv     = this->private;
+        VALIDATE_OR_GOTO (priv->children, out);
 
-	children = priv->children;
+        children = priv->children;
 
-	local = frame->local;
+        local = frame->local;
 
         read_child = (long) cookie;
 
-	if (op_ret == -1) {
-	retry:
-		last_tried = local->cont.readv.last_tried;
-
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try = ++local->cont.readv.last_tried;
-
-		if (this_try == read_child) {
-			/*
-			   skip the read child since if we are here
-			   we must have already tried that child
-			*/
-			goto retry;
-		}
-
-		unwind = 0;
-
-		STACK_WIND_COOKIE (frame, afr_readv_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->readv,
-				   local->fd, local->cont.readv.size,
-				   local->cont.readv.offset);
-	}
+        if (op_ret == -1) {
+                last_index = &local->cont.readv.last_index;
+                fresh_children = local->fresh_children;
+                next_call_child = afr_next_call_child (fresh_children,
+                                                       local->child_up,
+                                                       priv->child_count,
+                                                       last_index, read_child);
+                if (next_call_child < 0)
+                        goto out;
+
+                unwind = 0;
+
+                STACK_WIND_COOKIE (frame, afr_readv_cbk,
+                                   (void *) (long) read_child,
+                                   children[next_call_child],
+                                   children[next_call_child]->fops->readv,
+                                   local->fd, local->cont.readv.size,
+                                   local->cont.readv.offset,
+                                   local->cont.readv.flags,
+                                   NULL);
+        }
 
 out:
-	if (unwind) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, vector, count, buf);
-	}
+        if (unwind) {
+                AFR_STACK_UNWIND (readv, frame, op_ret, op_errno,
+                                  vector, count, buf, iobref, xdata);
+        }
 
-	return 0;
+        return 0;
 }
 
 
 int32_t
 afr_readv (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd, size_t size, off_t offset)
+           fd_t *fd, size_t size, off_t offset, uint32_t flags, dict_t *xdata)
 {
-	afr_private_t * priv       = NULL;
-	afr_local_t   * local      = NULL;
-	xlator_t **     children   = NULL;
-
+        afr_private_t * priv       = NULL;
+        afr_local_t   * local      = NULL;
+        xlator_t **     children   = NULL;
+        int             call_child = 0;
+        int32_t         op_errno   = 0;
         int32_t         read_child = -1;
-	int             call_child = 0;
-
-	int32_t         op_ret     = -1;
-	int32_t         op_errno   = 0;
+        int             ret        = -1;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-	VALIDATE_OR_GOTO (fd, out);
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+        VALIDATE_OR_GOTO (fd, out);
 
-	priv     = this->private;
-	children = priv->children;
+        priv     = this->private;
+        children = priv->children;
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        AFR_SBRAIN_CHECK_FD (fd, out);
 
-	frame->local = local;
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-        read_child = afr_read_child (this, fd->inode);
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-        if (read_child >= 0) {
-                call_child = read_child;
-
-		/*
-		   if read fails from the read child, we try
-		   all children starting with the first one
-		*/
-                local->cont.readv.last_tried = -1;
+        local->fresh_children = afr_children_create (priv->child_count);
+        if (!local->fresh_children) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-        } else {
-		call_child = afr_first_up_child (priv);
-		if (call_child == -1) {
-			op_errno = ENOTCONN;
-			gf_log (this->name, GF_LOG_ERROR,
-				"no child is up :(");
-			goto out;
-		}
+        read_child = afr_inode_get_read_ctx (this, fd->inode, local->fresh_children);
+        ret = afr_get_call_child (this, local->child_up, read_child,
+                                     local->fresh_children,
+                                     &call_child,
+                                     &local->cont.readv.last_index);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
 
-		local->cont.readv.last_tried = call_child;
-	}
+        local->fd                    = fd_ref (fd);
 
-	local->fd                    = fd_ref (fd);
+        local->cont.readv.size       = size;
+        local->cont.readv.offset     = offset;
+        local->cont.readv.flags      = flags;
 
-	local->cont.readv.size       = size;
-	local->cont.readv.offset     = offset;
+        afr_open_fd_fix (fd, this);
 
-	STACK_WIND_COOKIE (frame, afr_readv_cbk,
-			   (void *) (long) call_child,
-			   children[call_child],
-			   children[call_child]->fops->readv,
-			   fd, size, offset);
+        STACK_WIND_COOKIE (frame, afr_readv_cbk,
+                           (void *) (long) call_child,
+                           children[call_child],
+                           children[call_child]->fops->readv,
+                           fd, size, offset, flags, xdata);
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL, 0, NULL);
-	}
-	return 0;
+        if (ret < 0) {
+                AFR_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL,
+                                  NULL, NULL);
+        }
+        return 0;
 }
 
 /* }}} */
diff --git a/xlators/cluster/afr/src/afr-inode-read.h b/xlators/cluster/afr/src/afr-inode-read.h
index 1127cb652..e4091a793 100644
--- a/xlators/cluster/afr/src/afr-inode-read.h
+++ b/xlators/cluster/afr/src/afr-inode-read.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __INODE_READ_H__
@@ -22,26 +13,30 @@
 
 int32_t
 afr_access (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t mask);
+	    loc_t *loc, int32_t mask, dict_t *xdata);
 
 int32_t
 afr_stat (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc);
+	  loc_t *loc, dict_t *xdata);
 
 int32_t
 afr_fstat (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd);
+	   fd_t *fd, dict_t *xdata);
 
 int32_t
 afr_readlink (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, size_t size);
+	      loc_t *loc, size_t size, dict_t *xdata);
 
 int32_t
 afr_readv (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd, size_t size, off_t offset);
+	   fd_t *fd, size_t size, off_t offset, uint32_t flags, dict_t *xdata);
 
 int32_t
 afr_getxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, const char *name);
+	      loc_t *loc, const char *name, dict_t *xdata);
+
+int32_t
+afr_fgetxattr (call_frame_t *frame, xlator_t *this,
+               fd_t *fd, const char *name, dict_t *xdata);
 
 #endif /* __INODE_READ_H__ */
diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index cf4d493a4..c1ec69a55 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 
@@ -46,1979 +37,2825 @@
 
 #include "afr.h"
 #include "afr-transaction.h"
+#include "afr-self-heal-common.h"
 
+void
+__inode_write_fop_cbk (call_frame_t *frame, int child_index, int read_child,
+                       xlator_t *this, int32_t *op_ret, int32_t *op_errno,
+                       struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t     *local = NULL;
+
+        local = frame->local;
+
+        if (afr_fop_failed (*op_ret, *op_errno)) {
+                local->child_errno[child_index] = *op_errno;
+
+                switch (local->op) {
+                case GF_FOP_TRUNCATE:
+                case GF_FOP_FTRUNCATE:
+                        if (*op_errno != EFBIG)
+                                afr_transaction_fop_failed (frame, this,
+                                                            child_index);
+                break;
+                default:
+                        afr_transaction_fop_failed (frame, this, child_index);
+                break;
+                }
+                local->op_errno = *op_errno;
+                goto out;
+        }
+
+        if ((local->success_count == 0) || (read_child == child_index)) {
+                local->op_ret              = *op_ret;
+                if (prebuf)
+                        local->cont.inode_wfop.prebuf  = *prebuf;
+                if (postbuf)
+                        local->cont.inode_wfop.postbuf = *postbuf;
+        }
+
+        local->success_count++;
+out:
+        return;
+}
 
-/* {{{ chmod */
-
+/* {{{ writev */
 
-int
-afr_chmod_unwind (call_frame_t *frame, xlator_t *this)
+void
+afr_writev_copy_outvars (call_frame_t *src_frame, call_frame_t *dst_frame)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
+        afr_local_t *src_local = NULL;
+        afr_local_t *dst_local = NULL;
 
-	local = frame->local;
-	priv  = this->private;
+        src_local = src_frame->local;
+        dst_local = dst_frame->local;
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
+        dst_local->op_ret = src_local->op_ret;
+        dst_local->op_errno = src_local->op_errno;
+        dst_local->cont.inode_wfop.prebuf = src_local->cont.inode_wfop.prebuf;
+        dst_local->cont.inode_wfop.postbuf = src_local->cont.inode_wfop.postbuf;
+}
 
-	if (main_frame) {
-		local->cont.chmod.buf.st_ino = local->cont.chmod.ino;
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  &local->cont.chmod.buf);
-	}
-	return 0;
+void
+afr_writev_unwind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *   local = NULL;
+        local = frame->local;
+
+        AFR_STACK_UNWIND (writev, frame,
+                          local->op_ret, local->op_errno,
+                          &local->cont.inode_wfop.prebuf,
+                          &local->cont.inode_wfop.postbuf,
+                          NULL);
 }
 
+call_frame_t*
+afr_transaction_detach_fop_frame (call_frame_t *frame)
+{
+        afr_local_t *   local = NULL;
+        call_frame_t   *fop_frame = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                fop_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        return fop_frame;
+}
 
 int
-afr_chmod_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		    int32_t op_ret, int32_t op_errno, struct stat *buf)
+afr_transaction_writev_unwind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        call_frame_t *fop_frame = NULL;
 
-	int call_count  = -1;
-	int child_index = (long) cookie;
-	int need_unwind = 0;
+        fop_frame = afr_transaction_detach_fop_frame (frame);
 
-	local = frame->local;
-	priv  = this->private;
+        if (fop_frame) {
+                afr_writev_copy_outvars (frame, fop_frame);
+                afr_writev_unwind (fop_frame, this);
+        }
+        return 0;
+}
 
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
+static void
+afr_writev_handle_short_writes (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local = NULL;
+        afr_private_t *priv  = NULL;
+        int           i      = 0;
+
+        local = frame->local;
+        priv = this->private;
+        /*
+         * We already have the best case result of the writev calls staged
+         * as the return value. Any writev that returns some value less
+         * than the best case is now out of sync, so mark the fop as
+         * failed. Note that fops that have returned with errors have
+         * already been marked as failed.
+         */
+        for (i = 0; i < priv->child_count; i++) {
+                if ((!local->replies[i].valid) ||
+                    (local->replies[i].op_ret == -1))
+                        continue;
+
+                if (local->replies[i].op_ret < local->op_ret)
+                        afr_transaction_fop_failed(frame, this, i);
+        }
+}
 
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-				local->cont.chmod.buf = *buf;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
+int
+afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t *   local = NULL;
+        afr_private_t  *priv  = NULL;
+        call_frame_t    *fop_frame = NULL;
+        int child_index = (long) cookie;
+        int call_count  = -1;
+        int read_child  = 0;
+        int      ret = 0;
+        uint32_t open_fd_count = 0;
+        uint32_t write_is_append = 0;
+
+        local = frame->local;
+        priv  = this->private;
+
+        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
+
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, prebuf, postbuf,
+                                       xdata);
+
+		local->replies[child_index].valid = 1;
+		local->replies[child_index].op_ret = op_ret;
+		local->replies[child_index].op_errno = op_errno;
+
+
+		/* stage the best case return value for unwind */
+                if ((local->success_count == 0) || (op_ret > local->op_ret)) {
+                        local->op_ret              = op_ret;
+			local->op_errno		   = op_errno;
 		}
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	if (need_unwind)
-		afr_chmod_unwind (frame, this);
+		if (op_ret != -1) {
+                        if (xdata) {
+                                ret = dict_get_uint32 (xdata,
+                                                       GLUSTERFS_OPEN_FD_COUNT,
+                                                       &open_fd_count);
+                                if ((ret == 0) &&
+                                    (open_fd_count > local->open_fd_count)) {
+                                        local->open_fd_count = open_fd_count;
+                                        local->update_open_fd_count = _gf_true;
+                                }
+
+				write_is_append = 0;
+                                ret = dict_get_uint32 (xdata,
+                                                       GLUSTERFS_WRITE_IS_APPEND,
+                                                       &write_is_append);
+                                if (ret || !write_is_append)
+					local->append_write = _gf_false;
+                        }
 
-	call_count = afr_frame_return (frame);
+		}
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+
+                if (local->update_open_fd_count)
+                        afr_handle_open_fd_count (frame, this);
+
+                if (!local->stable_write && !local->append_write)
+			/* An appended write removes the necessity to
+			   fsync() the file. This is because self-heal
+			   has the logic to check for larger file when
+			   the xattrs are not reliably pointing at
+			   a stale file.
+			*/
+                        afr_fd_report_unstable_write (this, local->fd);
+
+                afr_writev_handle_short_writes (frame, this);
+                if (afr_any_fops_failed (local, priv)) {
+                        //Don't unwind until post-op is complete
+                        local->transaction.resume (frame, this);
+                } else {
+                /*
+                 * Generally inode-write fops do transaction.unwind then
+                 * transaction.resume, but writev needs to make sure that
+                 * delayed post-op frame is placed in fdctx before unwind
+                 * happens. This prevents the race of flush doing the
+                 * changelog wakeup first in fuse thread and then this
+                 * writev placing its delayed post-op frame in fdctx.
+                 * This helps flush make sure all the delayed post-ops are
+                 * completed.
+                 */
+
+                        fop_frame = afr_transaction_detach_fop_frame (frame);
+                        afr_writev_copy_outvars (frame, fop_frame);
+                        local->transaction.resume (frame, this);
+                        afr_writev_unwind (fop_frame, this);
+                }
+        }
+        return 0;
+}
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
+int
+afr_writev_wind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int i = 0;
+        int call_count = -1;
+        dict_t *xdata = NULL;
+        GF_UNUSED int     ret = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+	local->replies = GF_CALLOC(priv->child_count, sizeof(*local->replies),
+				   gf_afr_mt_reply_t);
+	if (!local->replies) {
+		local->op_ret = -1;
+		local->op_errno = ENOMEM;
+		local->transaction.unwind(frame, this);
+		local->transaction.resume(frame, this);
+		return 0;
 	}
-	
-	return 0;
+
+        xdata = dict_new ();
+        if (xdata) {
+                ret = dict_set_uint32 (xdata, GLUSTERFS_OPEN_FD_COUNT,
+                                       sizeof (uint32_t));
+		ret = dict_set_uint32 (xdata, GLUSTERFS_WRITE_IS_APPEND,
+				       0);
+		/* Set append_write to be true speculatively. If on any
+		   server it turns not be true, we unset it in the
+		   callback.
+		*/
+		local->append_write = _gf_true;
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_writev_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->writev,
+                                           local->fd,
+                                           local->cont.writev.vector,
+                                           local->cont.writev.count,
+                                           local->cont.writev.offset,
+                                           local->cont.writev.flags,
+                                           local->cont.writev.iobref,
+                                           xdata);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        if (xdata)
+                dict_unref (xdata);
+
+        return 0;
 }
 
 
 int
-afr_chmod_wind (call_frame_t *frame, xlator_t *this)
+afr_writev_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	
-	int i = 0;
-	int call_count = -1;
+        afr_local_t *local = NULL;
 
-	local = frame->local;
-	priv  = this->private;
+        local = frame->local;
 
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
+        iobref_unref (local->cont.writev.iobref);
+        local->cont.writev.iobref = NULL;
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
+        local->transaction.unwind (frame, this);
 
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_chmod_wind_cbk, (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->chmod,
-					   &local->loc, 
-					   local->cont.chmod.mode); 
-		
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+        AFR_STACK_DESTROY (frame);
+
+        return 0;
 }
 
 
 int
-afr_chmod_done (call_frame_t *frame, xlator_t *this)
+afr_do_writev (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t * local = NULL;
-
-	local = frame->local;
+        call_frame_t    *transaction_frame = NULL;
+        afr_local_t     *local             = NULL;
+        int             op_ret   = -1;
+        int             op_errno = 0;
+
+        local = frame->local;
+
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        transaction_frame->local = local;
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+
+        local->op = GF_FOP_WRITE;
+
+        local->success_count      = 0;
+
+        local->transaction.fop    = afr_writev_wind;
+        local->transaction.done   = afr_writev_done;
+        local->transaction.unwind = afr_transaction_writev_unwind;
+
+        local->transaction.main_frame = frame;
+        if (local->fd->flags & O_APPEND) {
+               /*
+                * Backend vfs ignores the 'offset' for append mode fd so
+                * locking just the region provided for the writev does not
+                * give consistency gurantee. The actual write may happen at a
+                * completely different range than the one provided by the
+                * offset, len in the fop. So lock the entire file.
+                */
+                local->transaction.start   = 0;
+                local->transaction.len     = 0;
+        } else {
+                local->transaction.start   = local->cont.writev.offset;
+                local->transaction.len     = iov_length (local->cont.writev.vector,
+                                                         local->cont.writev.count);
+        }
+
+        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (op_ret < 0) {
+            op_errno = -op_ret;
+            goto out;
+        }
+
+        op_ret = 0;
+out:
+        if (op_ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (writev, frame, op_ret, op_errno, NULL, NULL, NULL);
+        }
 
-	local->transaction.unwind (frame, this);
+        return 0;
+}
 
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
+static void
+afr_trigger_open_fd_self_heal (fd_t *fd, xlator_t *this)
+{
+        call_frame_t    *frame   = NULL;
+        afr_local_t     *local   = NULL;
+        afr_self_heal_t *sh      = NULL;
+        char            *reason  = NULL;
+        int32_t         op_errno = 0;
+        int             ret      = 0;
+
+        if (!fd || !fd->inode || uuid_is_null (fd->inode->gfid)) {
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid args: "
+                                  "fd: %p, inode: %p", fd,
+                                  fd ? fd->inode : NULL);
+                goto out;
+        }
+
+        frame = create_frame (this, this->ctx->pool);
+        if (!frame)
+                goto out;
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+        ret = afr_local_init (local, this->private, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        local->loc.inode = inode_ref (fd->inode);
+        ret = loc_path (&local->loc, NULL);
+        if (ret < 0)
+                goto out;
+
+        sh    = &local->self_heal;
+        sh->do_metadata_self_heal = _gf_true;
+        if (fd->inode->ia_type == IA_IFREG)
+                sh->do_data_self_heal = _gf_true;
+        else if (fd->inode->ia_type == IA_IFDIR)
+                sh->do_entry_self_heal = _gf_true;
+
+        reason = "subvolume came online";
+        afr_launch_self_heal (frame, this, fd->inode, _gf_true,
+                              fd->inode->ia_type, reason, NULL, NULL);
+        return;
+out:
+        AFR_STACK_DESTROY (frame);
 }
 
+void
+afr_open_fd_fix (fd_t *fd, xlator_t *this)
+{
+        int           ret             = 0;
+        int           i               = 0;
+        afr_fd_ctx_t  *fd_ctx         = NULL;
+        gf_boolean_t  need_self_heal  = _gf_false;
+        int           *need_open      = NULL;
+        size_t        need_open_count = 0;
+        afr_private_t *priv           = NULL;
+
+        priv  = this->private;
+
+        if (!afr_is_fd_fixable (fd))
+                goto out;
+
+        fd_ctx = afr_fd_ctx_get (fd, this);
+        if (!fd_ctx)
+                goto out;
+
+        LOCK (&fd->lock);
+        {
+                if (fd_ctx->up_count < priv->up_count) {
+                        need_self_heal = _gf_true;
+                        fd_ctx->up_count   = priv->up_count;
+                        fd_ctx->down_count = priv->down_count;
+                }
+
+                need_open = alloca (priv->child_count * sizeof (*need_open));
+                for (i = 0; i < priv->child_count; i++) {
+                        need_open[i] = 0;
+                        if (fd_ctx->opened_on[i] != AFR_FD_NOT_OPENED)
+                                continue;
+
+                        if (!priv->child_up[i])
+                                continue;
+
+                        fd_ctx->opened_on[i] = AFR_FD_OPENING;
+
+                        need_open[i] = 1;
+                        need_open_count++;
+                }
+        }
+        UNLOCK (&fd->lock);
+        if (ret)
+                goto out;
+
+        if (need_self_heal)
+                afr_trigger_open_fd_self_heal (fd, this);
+
+        if (!need_open_count)
+                goto out;
+
+        afr_fix_open (this, fd, need_open_count, need_open);
+out:
+        return;
+}
 
-int32_t
-afr_chmod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode)
+int
+afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
+            struct iovec *vector, int32_t count, off_t offset,
+            uint32_t flags, struct iobref *iobref, dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
+        afr_private_t * priv  = NULL;
+        afr_local_t   * local = NULL;
+        int ret = -1;
+        int op_errno = 0;
 
-	int ret = -1;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        priv = this->private;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
 
-	priv = this->private;
+        QUORUM_CHECK(writev,out);
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	transaction_frame->local = local;
+        local->cont.writev.vector     = iov_dup (vector, count);
+        local->cont.writev.count      = count;
+        local->cont.writev.offset     = offset;
+        local->cont.writev.flags      = flags;
+        local->cont.writev.iobref     = iobref_ref (iobref);
 
-	local->cont.chmod.mode = mode;
-	local->cont.chmod.ino  = loc->inode->ino;
+        local->fd                = fd_ref (fd);
 
-	local->transaction.fop    = afr_chmod_wind;
-	local->transaction.done   = afr_chmod_done;
-	local->transaction.unwind = afr_chmod_unwind;
+	/* detect here, but set it in writev_wind_cbk *after* the unstable
+	   write is performed
+	*/
+	local->stable_write = !!((fd->flags|flags)&(O_SYNC|O_DSYNC));
 
-	loc_copy (&local->loc, loc);
-	
-	local->transaction.main_frame = frame;
-	local->transaction.start   = 0;
-	local->transaction.len     = 0;
-	local->transaction.pending = AFR_METADATA_PENDING;
+        afr_open_fd_fix (fd, this);
 
-	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        afr_do_writev (frame, this);
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+        if (ret < 0)
+                AFR_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
-/* }}} */
 
+/* }}} */
 
-/* {{{ fchmod */
+/* {{{ truncate */
 
 int
-afr_fchmod_unwind (call_frame_t *frame, xlator_t *this)
+afr_truncate_unwind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
-
-	local = frame->local;
-	priv  = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-		local->cont.fchmod.buf.st_ino = local->cont.fchmod.ino;
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  &local->cont.fchmod.buf);
-	}
-	return 0;
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame)
+                        main_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (truncate, main_frame, local->op_ret,
+                                  local->op_errno,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
+                                  NULL);
+        }
+
+        return 0;
 }
 
 
 int
-afr_fchmod_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		    int32_t op_ret, int32_t op_errno, struct stat *buf)
+afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        afr_local_t *   local = NULL;
+        int child_index = (long) cookie;
+        int read_child  = 0;
+        int call_count  = -1;
 
-	int call_count  = -1;
-	int child_index = (long) cookie;
-	int need_unwind = 0;
+        local = frame->local;
 
-	local = frame->local;
-	priv  = this->private;
+        read_child = afr_inode_get_read_ctx (this, local->loc.inode, NULL);
 
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
 
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-				local->cont.fchmod.buf = *buf;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
-		}
+                if (op_ret != -1) {
+			if (prebuf->ia_size != postbuf->ia_size)
+				local->stable_write = _gf_false;
+                }
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, prebuf, postbuf,
+                                       xdata);
+        }
+        UNLOCK (&frame->lock);
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        call_count = afr_frame_return (frame);
 
-	if (need_unwind)
-		afr_fchmod_unwind (frame, this);
+        if (call_count == 0) {
+		if (local->stable_write && afr_txn_nothing_failed (frame, this))
+			local->transaction.unwind (frame, this);
 
-	call_count = afr_frame_return (frame);
+                local->transaction.resume (frame, this);
+        }
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        return 0;
 }
 
 
-int
-afr_fchmod_wind (call_frame_t *frame, xlator_t *this)
+int32_t
+afr_truncate_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	
-	int i = 0;
-	int call_count = -1;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_fchmod_wind_cbk, (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->fchmod,
-					   local->fd, 
-					   local->cont.fchmod.mode); 
-		
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+	local->stable_write = _gf_true;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_truncate_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->truncate,
+                                           &local->loc,
+                                           local->cont.truncate.offset,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
 }
 
 
 int
-afr_fchmod_done (call_frame_t *frame, xlator_t *this)
+afr_truncate_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t * local = NULL;
+        afr_local_t *local = NULL;
+
+        local = frame->local;
 
-	local = frame->local;
+        local->transaction.unwind (frame, this);
 
-	local->transaction.unwind (frame, this);
+        AFR_STACK_DESTROY (frame);
 
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
+        return 0;
 }
 
 
-int32_t
-afr_fchmod (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, mode_t mode)
+int
+afr_truncate (call_frame_t *frame, xlator_t *this,
+              loc_t *loc, off_t offset, dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
+        afr_private_t * priv  = NULL;
+        afr_local_t   * local = NULL;
+        call_frame_t   *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = 0;
 
-	int ret = -1;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        priv = this->private;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        QUORUM_CHECK(truncate,out);
 
-	priv = this->private;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	transaction_frame->local = local;
+        local->cont.truncate.offset  = offset;
 
-	local->cont.fchmod.mode = mode;
-	local->cont.fchmod.ino  = fd->inode->ino;
+        local->transaction.fop    = afr_truncate_wind;
+        local->transaction.done   = afr_truncate_done;
+        local->transaction.unwind = afr_truncate_unwind;
 
-	local->transaction.fop    = afr_fchmod_wind;
-	local->transaction.done   = afr_fchmod_done;
-	local->transaction.unwind = afr_fchmod_unwind;
+        loc_copy (&local->loc, loc);
 
-	local->fd = fd_ref (fd);
-	
-	local->transaction.main_frame = frame;
-	local->transaction.start   = 0;
-	local->transaction.len     = 0;
-	local->transaction.pending = AFR_METADATA_PENDING;
+        local->transaction.main_frame = frame;
+        local->transaction.start   = offset;
+        local->transaction.len     = 0;
 
-	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
+
 /* }}} */
 
-/* {{{ chown */
+/* {{{ ftruncate */
+
 
 int
-afr_chown_unwind (call_frame_t *frame, xlator_t *this)
+afr_ftruncate_unwind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
-
-	local = frame->local;
-	priv  = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-		local->cont.chown.buf.st_ino = local->cont.chown.ino;
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  &local->cont.chown.buf);
-	}
-	return 0;
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame)
+                        main_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (ftruncate, main_frame, local->op_ret,
+                                  local->op_errno,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
+                                  NULL);
+        }
+        return 0;
 }
 
 
 int
-afr_chown_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		    int32_t op_ret, int32_t op_errno, struct stat *buf)
+afr_ftruncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                        struct iatt *postbuf, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
- 
-	int call_count  = -1;
-	int child_index = (long) cookie;
-	int need_unwind = 0;
+        afr_local_t *   local = NULL;
+        int child_index = (long) cookie;
+        int call_count  = -1;
+        int read_child  = 0;
 
-	local = frame->local;
-	priv = this->private;
+        local = frame->local;
 
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
+        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
 
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-				local->cont.chown.buf = *buf;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
-		}
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+                if (op_ret != -1) {
+			if (prebuf->ia_size != postbuf->ia_size)
+				local->stable_write = _gf_false;
+                }
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, prebuf, postbuf,
+                                       xdata);
+        }
+        UNLOCK (&frame->lock);
 
-	if (need_unwind) {
-		local->transaction.unwind (frame, this);
-	}
+        call_count = afr_frame_return (frame);
 
-	call_count = afr_frame_return (frame);
+        if (call_count == 0) {
+		if (local->stable_write && afr_txn_nothing_failed (frame, this))
+			local->transaction.unwind (frame, this);
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+                local->transaction.resume (frame, this);
+        }
+
+        return 0;
 }
 
 
 int
-afr_chown_wind (call_frame_t *frame, xlator_t *this)
+afr_ftruncate_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int i = 0;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+	local->stable_write = _gf_true;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_ftruncate_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->ftruncate,
+                                           local->fd,
+                                           local->cont.ftruncate.offset,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
 
-	local = frame->local;
-	priv  = this->private;
 
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
+int
+afr_ftruncate_done (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
+        local = frame->local;
 
-	local->call_count = call_count;
+        local->transaction.unwind (frame, this);
 
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_chown_wind_cbk, (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->chown,
-					   &local->loc, local->cont.chown.uid,
-					   local->cont.chown.gid); 
+        AFR_STACK_DESTROY (frame);
 
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+        return 0;
 }
 
 
 int
-afr_chown_done (call_frame_t *frame, xlator_t *this)
+afr_do_ftruncate (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
+        call_frame_t * transaction_frame = NULL;
+        afr_local_t *  local             = NULL;
+        int op_ret   = -1;
+        int op_errno = 0;
 
-	local = frame->local;
+        local = frame->local;
 
-	local->transaction.unwind (frame, this);
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                goto out;
+        }
 
-	AFR_STACK_DESTROY (frame);
+        transaction_frame->local = local;
+        frame->local = NULL;
 
-	return 0;
-}
+        local->op = GF_FOP_FTRUNCATE;
 
+        local->transaction.fop    = afr_ftruncate_wind;
+        local->transaction.done   = afr_ftruncate_done;
+        local->transaction.unwind = afr_ftruncate_unwind;
 
-int
-afr_chown (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, uid_t uid, gid_t gid)
-{
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t   *transaction_frame = NULL;
+        local->transaction.main_frame = frame;
+
+        local->transaction.start   = local->cont.ftruncate.offset;
+        local->transaction.len     = 0;
 
-	int ret = -1;
+        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (op_ret < 0) {
+            op_errno = -op_ret;
+            goto out;
+        }
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        op_ret = 0;
+out:
+        if (op_ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (ftruncate, frame, op_ret, op_errno, NULL,
+                                  NULL, NULL);
+        }
+
+        return 0;
+}
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
 
-	priv = this->private;
+int
+afr_ftruncate (call_frame_t *frame, xlator_t *this,
+               fd_t *fd, off_t offset, dict_t *xdata)
+{
+        afr_private_t * priv  = NULL;
+        afr_local_t   * local = NULL;
+        call_frame_t   *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = 0;
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        priv = this->private;
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
+        QUORUM_CHECK(ftruncate,out);
 
-	transaction_frame->local = local;
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-	local->cont.chown.uid  = uid;
-	local->cont.chown.gid  = gid;
-	local->cont.chown.ino  = loc->inode->ino;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	local->transaction.fop    = afr_chown_wind;
-	local->transaction.done   = afr_chown_done;
-	local->transaction.unwind = afr_chown_unwind;
+        local->cont.ftruncate.offset  = offset;
 
-	loc_copy (&local->loc, loc);
+        local->fd = fd_ref (fd);
 
-	local->transaction.main_frame = frame;
-	local->transaction.start   = 0;
-	local->transaction.len     = 0;
-	local->transaction.pending = AFR_METADATA_PENDING;
+        afr_open_fd_fix (fd, this);
 
-	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        afr_do_ftruncate (frame, this);
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
-
 /* }}} */
 
-/* {{{ chown */
+/* {{{ setattr */
 
 int
-afr_fchown_unwind (call_frame_t *frame, xlator_t *this)
+afr_setattr_unwind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
-
-	local = frame->local;
-	priv  = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-		local->cont.fchown.buf.st_ino = local->cont.fchown.ino;
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  &local->cont.fchown.buf);
-	}
-	return 0;
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame)
+                        main_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (setattr, main_frame, local->op_ret,
+                                  local->op_errno,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
+                                  NULL);
+        }
+
+        return 0;
 }
 
 
 int
-afr_fchown_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		    int32_t op_ret, int32_t op_errno, struct stat *buf)
+afr_setattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno,
+                      struct iatt *preop, struct iatt *postop, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
- 
-	int call_count  = -1;
-	int child_index = (long) cookie;
-	int need_unwind = 0;
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = NULL;
+        int child_index = (long) cookie;
+        int read_child  = 0;
+        int call_count  = -1;
+        int need_unwind = 0;
 
-	local = frame->local;
-	priv = this->private;
+        local = frame->local;
+        priv  = this->private;
 
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
+        read_child = afr_inode_get_read_ctx (this, local->loc.inode, NULL);
 
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-				local->cont.fchown.buf = *buf;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
-		}
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, preop, postop,
+                                       xdata);
 
-	if (need_unwind) {
-		local->transaction.unwind (frame, this);
-	}
+                if ((local->success_count >= priv->wait_count)
+                    && local->read_child_returned) {
+                        need_unwind = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
 
-	call_count = afr_frame_return (frame);
+        if (need_unwind)
+                local->transaction.unwind (frame, this);
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
-}
+        call_count = afr_frame_return (frame);
 
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+        }
 
-int
-afr_fchown_wind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
+        return 0;
+}
 
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_fchown_wind_cbk, (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->fchown,
-					   local->fd, local->cont.fchown.uid,
-					   local->cont.fchown.gid); 
 
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+int32_t
+afr_setattr_wind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_setattr_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->setattr,
+                                           &local->loc,
+                                           &local->cont.setattr.in_buf,
+                                           local->cont.setattr.valid,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
 }
 
 
 int
-afr_fchown_done (call_frame_t *frame, xlator_t *this)
+afr_setattr_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
+        afr_local_t *local = NULL;
 
-	local = frame->local;
+        local = frame->local;
 
-	local->transaction.unwind (frame, this);
+        local->transaction.unwind (frame, this);
 
-	AFR_STACK_DESTROY (frame);
+        AFR_STACK_DESTROY (frame);
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_fchown (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, uid_t uid, gid_t gid)
+afr_setattr (call_frame_t *frame, xlator_t *this,
+             loc_t *loc, struct iatt *buf, int32_t valid, dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t   *transaction_frame = NULL;
+        afr_private_t * priv  = NULL;
+        afr_local_t   * local = NULL;
+        call_frame_t   *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = 0;
 
-	int ret = -1;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        priv = this->private;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        QUORUM_CHECK(setattr,out);
 
-	priv = this->private;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
 
-	transaction_frame->local = local;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	local->cont.fchown.uid  = uid;
-	local->cont.fchown.gid  = gid;
-	local->cont.fchown.ino  = fd->inode->ino;
+        local->cont.setattr.in_buf = *buf;
+        local->cont.setattr.valid  = valid;
 
-	local->transaction.fop    = afr_fchown_wind;
-	local->transaction.done   = afr_fchown_done;
-	local->transaction.unwind = afr_fchown_unwind;
+        local->transaction.fop    = afr_setattr_wind;
+        local->transaction.done   = afr_setattr_done;
+        local->transaction.unwind = afr_setattr_unwind;
 
-	local->fd = fd_ref (fd);
+        loc_copy (&local->loc, loc);
 
-	local->transaction.main_frame = frame;
-	local->transaction.start   = 0;
-	local->transaction.len     = 0;
-	local->transaction.pending = AFR_METADATA_PENDING;
+        local->transaction.main_frame = frame;
+        local->transaction.start   = LLONG_MAX - 1;
+        local->transaction.len     = 0;
 
-	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
-/* }}} */
-
-/* {{{ writev */
+/* {{{ fsetattr */
 
 int
-afr_writev_unwind (call_frame_t *frame, xlator_t *this)
+afr_fsetattr_unwind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
-
-	local = frame->local;
-	priv  = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-		local->cont.writev.buf.st_ino = local->cont.writev.ino;
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  &local->cont.writev.buf);
-	}
-	return 0;
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame)
+                        main_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (fsetattr, main_frame, local->op_ret,
+                                  local->op_errno,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
+                                  NULL);
+        }
+
+        return 0;
 }
 
 
 int
-afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		     int32_t op_ret, int32_t op_errno, struct stat *buf)
+afr_fsetattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno,
+                       struct iatt *preop, struct iatt *postop, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = NULL;
+        int child_index = (long) cookie;
+        int read_child  = 0;
+        int call_count  = -1;
+        int need_unwind = 0;
 
-	int child_index = (long) cookie;
-	int call_count  = -1;
-	int need_unwind = 0;
+        local = frame->local;
+        priv  = this->private;
 
-	local = frame->local;
-	priv = this->private;
+        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
 
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
 
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret   = op_ret;
-				local->cont.writev.buf = *buf;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
-		}
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, preop, postop,
+                                       xdata);
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+                if ((local->success_count >= priv->wait_count)
+                    && local->read_child_returned) {
+                        need_unwind = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
 
-	call_count = afr_frame_return (frame);
+        if (need_unwind)
+                local->transaction.unwind (frame, this);
 
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
+        call_count = afr_frame_return (frame);
 
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+        }
+
+        return 0;
 }
 
 
-int
-afr_writev_wind (call_frame_t *frame, xlator_t *this)
+int32_t
+afr_fsetattr_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-	
-	int i = 0;
-	int call_count = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_writev_wind_cbk, 
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->writev,
-					   local->fd, 
-					   local->cont.writev.vector,
-					   local->cont.writev.count, 
-					   local->cont.writev.offset); 
-		
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_fsetattr_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->fsetattr,
+                                           local->fd,
+                                           &local->cont.fsetattr.in_buf,
+                                           local->cont.fsetattr.valid,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
 }
 
 
 int
-afr_writev_done (call_frame_t *frame, xlator_t *this)
+afr_fsetattr_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
+        afr_local_t *local = NULL;
 
-	local = frame->local;
+        local = frame->local;
 
-	if (local->cont.writev.refs)
-		dict_unref (local->cont.writev.refs);
-	local->cont.writev.refs = NULL;
+        local->transaction.unwind (frame, this);
 
-	local->transaction.unwind (frame, this);
+        AFR_STACK_DESTROY (frame);
 
-	AFR_STACK_DESTROY (frame);
-
-	return 0;
+        return 0;
 }
 
-
 int
-afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, 
-	    struct iovec *vector, int32_t count, off_t offset)
+afr_fsetattr (call_frame_t *frame, xlator_t *this,
+              fd_t *fd, struct iatt *buf, int32_t valid, dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t   *transaction_frame = NULL;
+        afr_private_t * priv  = NULL;
+        afr_local_t   * local = NULL;
+        call_frame_t   *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = 0;
 
-	int ret = -1;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        priv = this->private;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
 
-	priv = this->private;
+        QUORUM_CHECK(fsetattr,out);
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	transaction_frame->local = local;
-
-	local->op = GF_FOP_WRITE;
-	local->cont.writev.vector  = iov_dup (vector, count);
-	local->cont.writev.count   = count;
-	local->cont.writev.offset  = offset;
-	local->cont.writev.ino     = fd->inode->ino;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	if (frame->root->req_refs)
-		local->cont.writev.refs = dict_ref (frame->root->req_refs);
+        local->cont.fsetattr.in_buf = *buf;
+        local->cont.fsetattr.valid  = valid;
 
-	local->transaction.fop    = afr_writev_wind;
-	local->transaction.done   = afr_writev_done;
-	local->transaction.unwind = afr_writev_unwind;
+        local->transaction.fop    = afr_fsetattr_wind;
+        local->transaction.done   = afr_fsetattr_done;
+        local->transaction.unwind = afr_fsetattr_unwind;
 
-	local->fd                = fd_ref (fd);
+        local->fd                 = fd_ref (fd);
 
-	local->transaction.main_frame = frame;
-	if (fd->flags & O_APPEND) {
-		local->transaction.start   = 0;
-		local->transaction.len     = 0;
-	} else {
-		local->transaction.start   = offset;
-		local->transaction.len     = iov_length (vector, count);
-	}
+        afr_open_fd_fix (fd, this);
 
-	local->transaction.pending = AFR_DATA_PENDING;
+        local->transaction.main_frame = frame;
+        local->transaction.start   = LLONG_MAX - 1;
+        local->transaction.len     = 0;
 
-	afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
 
-/* }}} */
+/* {{{ setxattr */
 
-/* {{{ truncate */
 
 int
-afr_truncate_unwind (call_frame_t *frame, xlator_t *this)
+afr_setxattr_unwind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
-
-	local = frame->local;
-	priv  = this->private;
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame)
+                        main_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (setxattr, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  NULL);
+        }
+        return 0;
+}
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
 
-	if (main_frame) {
-		local->cont.truncate.buf.st_ino = local->cont.truncate.ino;
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  &local->cont.truncate.buf);
-	}
-	return 0;
+int
+afr_setxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_local_t   *local      = NULL;
+        afr_private_t *priv       = NULL;
+        int           call_count  = -1;
+        int           need_unwind = 0;
+        int           child_index = (long) cookie;
+
+        local = frame->local;
+        priv = this->private;
+
+        LOCK (&frame->lock);
+        {
+                __inode_write_fop_cbk (frame, child_index, -1, this,
+                                       &op_ret, &op_errno, NULL, NULL,
+                                       xdata);
+                if (local->success_count == priv->child_count) {
+                        need_unwind = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        if (need_unwind)
+                local->transaction.unwind (frame, this);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+        }
+
+        return 0;
 }
 
 
 int
-afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		       int32_t op_ret, int32_t op_errno, struct stat *buf)
+afr_setxattr_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        afr_local_t      *local         = NULL;
+        afr_private_t    *priv          = NULL;
+        int               call_count    = -1;
+        int               i             = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_setxattr_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->setxattr,
+                                           &local->loc,
+                                           local->cont.setxattr.dict,
+                                           local->cont.setxattr.flags,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
 
-	int child_index = (long) cookie;
-	int call_count  = -1;
-	int need_unwind = 0;
 
-	local = frame->local;
-	priv  = this->private;
+int
+afr_setxattr_done (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local    = frame->local;
 
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
+        local->transaction.unwind (frame, this);
 
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-				local->cont.truncate.buf = *buf;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
-		}
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        AFR_STACK_DESTROY (frame);
 
-	if (need_unwind)
-		local->transaction.unwind (frame, this);
+        return 0;
+}
 
-	call_count = afr_frame_return (frame);
+int
+afr_setxattr (call_frame_t *frame, xlator_t *this,
+              loc_t *loc, dict_t *dict, int32_t flags, dict_t *xdata)
+{
+        afr_private_t  *priv              = NULL;
+        afr_local_t    *local             = NULL;
+        call_frame_t   *transaction_frame = NULL;
+        int             ret               = -1;
+        int             op_errno          = EINVAL;
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
-}
+        VALIDATE_OR_GOTO (this, out);
 
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.afr.*", dict,
+                                   op_errno, out);
 
-int32_t
-afr_truncate_wind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-	
-	int call_count = -1;
-	int i = 0;
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.afr.*", dict,
+                                   op_errno, out);
 
-	local = frame->local;
-	priv = this->private;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
+        priv = this->private;
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
+        QUORUM_CHECK(setxattr,out);
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	local->call_count = call_count;
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
 
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_truncate_wind_cbk,
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->truncate,
-					   &local->loc, 
-					   local->cont.truncate.offset);
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
-}
+        local->cont.setxattr.dict  = dict_ref (dict);
+        local->cont.setxattr.flags = flags;
 
+        local->transaction.fop    = afr_setxattr_wind;
+        local->transaction.done   = afr_setxattr_done;
+        local->transaction.unwind = afr_setxattr_unwind;
 
-int
-afr_truncate_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *local = NULL;
+        loc_copy (&local->loc, loc);
 
-	local = frame->local;
+        local->transaction.main_frame = frame;
+        local->transaction.start   = LLONG_MAX - 1;
+        local->transaction.len     = 0;
 
-	local->transaction.unwind (frame, this);
+        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	AFR_STACK_DESTROY (frame);
+        ret = 0;
+out:
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
+/* {{{ fsetxattr */
+
 
 int
-afr_truncate (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, off_t offset)
+afr_fsetxattr_unwind (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t   *transaction_frame = NULL;
+        afr_local_t    *local         = NULL;
+        call_frame_t   *main_frame    = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame)
+                        main_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (fsetxattr, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  NULL);
+        }
+        return 0;
+}
 
-	int ret = -1;
 
-	int op_ret   = -1;
-	int op_errno = 0;
+int
+afr_fsetxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_local_t   *local      = NULL;
+        afr_private_t *priv       = NULL;
+        int           call_count  = -1;
+        int           need_unwind = 0;
+        int           child_index = (long) cookie;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        local = frame->local;
+        priv = this->private;
 
-	priv = this->private;
+        LOCK (&frame->lock);
+        {
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+                __inode_write_fop_cbk (frame, child_index, -1, this,
+                                       &op_ret, &op_errno, NULL, NULL,
+                                       xdata);
+                if (local->success_count == priv->child_count) {
+                        need_unwind = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        if (need_unwind)
+                local->transaction.unwind (frame, this);
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        call_count = afr_frame_return (frame);
 
-	transaction_frame->local = local;
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+        }
 
-	local->op_ret = -1;
+        return 0;
+}
 
-	local->cont.truncate.offset  = offset;
-	local->cont.truncate.ino     = loc->inode->ino;
 
-	local->transaction.fop    = afr_truncate_wind;
-	local->transaction.done   = afr_truncate_done;
-	local->transaction.unwind = afr_truncate_unwind;
+int
+afr_fsetxattr_wind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t        *local       = NULL;
+        afr_private_t      *priv        = NULL;
+        int                 call_count  = -1;
+        int                 i           = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_fsetxattr_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->fsetxattr,
+                                           local->fd,
+                                           local->cont.fsetxattr.dict,
+                                           local->cont.fsetxattr.flags,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
 
-	loc_copy (&local->loc, loc);
 
-	local->transaction.main_frame = frame;
-	local->transaction.start   = 0;
-	local->transaction.len     = offset;
-	local->transaction.pending = AFR_DATA_PENDING;
+int
+afr_fsetxattr_done (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local   = frame->local;
 
-	afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        local->transaction.unwind (frame, this);
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+        AFR_STACK_DESTROY (frame);
 
-	return 0;
+        return 0;
 }
 
+int
+afr_fsetxattr (call_frame_t *frame, xlator_t *this,
+               fd_t *fd, dict_t *dict, int32_t flags, dict_t *xdata)
+{
+        afr_private_t    *priv              = NULL;
+        afr_local_t      *local             = NULL;
+        call_frame_t     *transaction_frame = NULL;
+        int               ret               = -1;
+        int               op_errno          = EINVAL;
 
-/* }}} */
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-/* {{{ ftruncate */
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.afr.*", dict,
+                                   op_errno, out);
 
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.afr.*", dict,
+                                   op_errno, out);
 
-int
-afr_ftruncate_unwind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
+        priv = this->private;
 
-	local = frame->local;
-	priv  = this->private;
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
+        QUORUM_CHECK(fsetxattr,out);
 
-	if (main_frame) {
-		local->cont.ftruncate.buf.st_ino = local->cont.ftruncate.ino;
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  &local->cont.ftruncate.buf);
-	}
-	return 0;
-}
+        AFR_LOCAL_ALLOC_OR_GOTO (local, out);
 
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-int
-afr_ftruncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-			int32_t op_ret, int32_t op_errno, struct stat *buf)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                goto out;
+        }
 
-	int child_index = (long) cookie;
-	int call_count  = -1;
-	int need_unwind = 0;
+        transaction_frame->local = local;
 
-	local = frame->local;
-	priv  = this->private;
+        local->op_ret = -1;
 
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
+        local->cont.fsetxattr.dict  = dict_ref (dict);
+        local->cont.fsetxattr.flags = flags;
 
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-				local->cont.ftruncate.buf = *buf;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
-		}
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        local->transaction.fop    = afr_fsetxattr_wind;
+        local->transaction.done   = afr_fsetxattr_done;
+        local->transaction.unwind = afr_fsetxattr_unwind;
 
-	if (need_unwind)
-		local->transaction.unwind (frame, this);
+        local->fd                 = fd_ref (fd);
 
-	call_count = afr_frame_return (frame);
+        local->transaction.main_frame = frame;
+        local->transaction.start  = LLONG_MAX - 1;
+        local->transaction.len    = 0;
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
+
+        ret = 0;
+out:
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (fsetxattr, frame, -1, op_errno, NULL);
+        }
+
+        return 0;
 }
 
+/* }}} */
 
-int
-afr_ftruncate_wind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-	
-	int call_count = -1;
-	int i = 0;
 
-	local = frame->local;
-	priv = this->private;
+/* {{{ removexattr */
 
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
+int
+afr_removexattr_unwind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame)
+                        main_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (removexattr, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  NULL);
+        }
+        return 0;
+}
 
-	local->call_count = call_count;
 
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_ftruncate_wind_cbk,
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->ftruncate,
-					   local->fd, local->cont.ftruncate.offset);
+int
+afr_removexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_local_t   *local      = NULL;
+        afr_private_t *priv       = NULL;
+        int           call_count  = -1;
+        int           need_unwind = 0;
+        int           child_index = (long) cookie;
+
+        local = frame->local;
+        priv = this->private;
+
+        LOCK (&frame->lock);
+        {
+                __inode_write_fop_cbk (frame, child_index, -1, this,
+                                       &op_ret, &op_errno, NULL, NULL,
+                                       xdata);
+                if (local->success_count == priv->wait_count) {
+                        need_unwind = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        if (need_unwind)
+                local->transaction.unwind (frame, this);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+        }
+
+        return 0;
+}
 
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+
+int32_t
+afr_removexattr_wind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_removexattr_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->removexattr,
+                                           &local->loc,
+                                           local->cont.removexattr.name,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
 }
 
 
 int
-afr_ftruncate_done (call_frame_t *frame, xlator_t *this)
+afr_removexattr_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-
-	local = frame->local;
+        afr_local_t * local = frame->local;
 
-	local->transaction.unwind (frame, this);
+        local->transaction.unwind (frame, this);
 
-	AFR_STACK_DESTROY (frame);
+        AFR_STACK_DESTROY (frame);
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_ftruncate (call_frame_t *frame, xlator_t *this,
-	       fd_t *fd, off_t offset)
+afr_removexattr (call_frame_t *frame, xlator_t *this,
+                 loc_t *loc, const char *name, dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t   *transaction_frame = NULL;
+        afr_private_t   *priv              = NULL;
+        afr_local_t     *local             = NULL;
+        call_frame_t    *transaction_frame = NULL;
+        int              ret               = -1;
+        int              op_errno          = 0;
 
-	int ret = -1;
+        VALIDATE_OR_GOTO (this, out);
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.afr.*",
+                                 name, op_errno, out);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.afr.*",
+                                 name, op_errno, out);
 
-	priv = this->private;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this->private, out);
+        VALIDATE_OR_GOTO (loc, out);
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        priv = this->private;
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        QUORUM_CHECK(removexattr,out);
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
 
-	transaction_frame->local = local;
+        AFR_LOCAL_ALLOC_OR_GOTO (transaction_frame->local, out);
+        local = transaction_frame->local;
 
-	local->op = GF_FOP_FTRUNCATE;
-	local->op_ret = -1;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	local->cont.ftruncate.offset  = offset;
-	local->cont.ftruncate.ino     = fd->inode->ino;
+        local->cont.removexattr.name = gf_strdup (name);
 
-	local->transaction.fop    = afr_ftruncate_wind;
-	local->transaction.done   = afr_ftruncate_done;
-	local->transaction.unwind = afr_ftruncate_unwind;
+        local->transaction.fop    = afr_removexattr_wind;
+        local->transaction.done   = afr_removexattr_done;
+        local->transaction.unwind = afr_removexattr_unwind;
 
-	local->fd = fd_ref (fd);
+        loc_copy (&local->loc, loc);
 
-	local->transaction.main_frame = frame;
-	local->transaction.start   = 0;
-	local->transaction.len     = offset;
-	local->transaction.pending = AFR_DATA_PENDING;
+        local->transaction.main_frame = frame;
+        local->transaction.start   = LLONG_MAX - 1;
+        local->transaction.len     = 0;
 
-	afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        if (ret < 0) {
+            op_errno = -ret;
+            goto out;
+        }
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (removexattr, frame, -1, op_errno, NULL);
+        }
 
-	return 0;
+        return 0;
 }
 
-/* }}} */
-
-/* {{{ utimens */
-
-
+/* ffremovexattr */
 int
-afr_utimens_unwind (call_frame_t *frame, xlator_t *this)
+afr_fremovexattr_unwind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
-
-	local = frame->local;
-	priv  = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-		local->cont.utimens.buf.st_ino = local->cont.utimens.ino;
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno,
-				  &local->cont.utimens.buf);
-	}
-	return 0;
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame)
+                        main_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (fremovexattr, main_frame,
+                                  local->op_ret, local->op_errno,
+                                  NULL);
+        }
+        return 0;
 }
 
 
 int
-afr_utimens_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		      int32_t op_ret, int32_t op_errno, struct stat *buf)
+afr_fremovexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        afr_local_t *   local       = NULL;
+        afr_private_t * priv        = NULL;
+        int             call_count  = -1;
+        int             need_unwind = 0;
+        int             child_index = (long) cookie;
 
-	int child_index = (long) cookie;
-	int call_count  = -1;
-	int need_unwind = 1;
+        local = frame->local;
+        priv = this->private;
 
-	local = frame->local;
-	priv = this->private;
+        LOCK (&frame->lock);
+        {
+                __inode_write_fop_cbk (frame, child_index, -1, this,
+                                       &op_ret, &op_errno, NULL, NULL,
+                                       xdata);
 
-	LOCK (&frame->lock);
-	{
-		if (child_went_down (op_ret, op_errno))
-			afr_transaction_child_died (frame, this, child_index);
+                if (local->success_count == priv->wait_count) {
+                        need_unwind = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
 
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-				local->cont.utimens.buf = *buf;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
-		}
+        if (need_unwind)
+                local->transaction.unwind (frame, this);
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        call_count = afr_frame_return (frame);
 
-	if (need_unwind)
-		local->transaction.unwind (frame, this);
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+        }
 
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        return 0;
 }
 
 
-int
-afr_utimens_wind (call_frame_t *frame, xlator_t *this)
+int32_t
+afr_fremovexattr_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-	
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv = this->private;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_fremovexattr_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->fremovexattr,
+                                           local->fd,
+                                           local->cont.removexattr.name,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
 
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
+int
+afr_fremovexattr_done (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t * local = frame->local;
 
-	local->call_count = call_count;
+        local->transaction.unwind (frame, this);
 
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_utimens_wind_cbk,
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->utimens,
-					   &local->loc, 
-					   local->cont.utimens.tv); 
+        AFR_STACK_DESTROY (frame);
 
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+        return 0;
 }
 
 
 int
-afr_utimens_done (call_frame_t *frame, xlator_t *this)
+afr_fremovexattr (call_frame_t *frame, xlator_t *this,
+                  fd_t *fd, const char *name, dict_t *xdata)
 {
-	afr_local_t * local = NULL;
+        afr_private_t * priv  = NULL;
+        afr_local_t   * local = NULL;
+        call_frame_t   *transaction_frame = NULL;
+        int ret = -1;
+        int op_ret   = -1;
+        int op_errno = 0;
 
-	local = frame->local;
+        VALIDATE_OR_GOTO (this, out);
 
-	local->transaction.unwind (frame, this);
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.afr.*",
+                                 name, op_errno, out);
 
-	AFR_STACK_DESTROY (frame);
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.afr.*",
+                                 name, op_errno, out);
 
-	return 0;
-}
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
+        priv = this->private;
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
 
-int
-afr_utimens (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, struct timespec tv[2])
-{
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t   *transaction_frame = NULL;
+        QUORUM_CHECK(fremovexattr, out);
 
-	int ret = -1;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                goto out;
+        }
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        AFR_LOCAL_ALLOC_OR_GOTO (local, out);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
 
-	priv = this->private;
+        transaction_frame->local = local;
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        local->op_ret = -1;
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        local->cont.removexattr.name = gf_strdup (name);
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        local->transaction.fop    = afr_fremovexattr_wind;
+        local->transaction.done   = afr_fremovexattr_done;
+        local->transaction.unwind = afr_fremovexattr_unwind;
+
+        local->fd = fd_ref (fd);
+
+        local->transaction.main_frame = frame;
+        local->transaction.start   = LLONG_MAX - 1;
+        local->transaction.len     = 0;
+
+        op_ret = afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        if (op_ret < 0) {
+            op_errno = -op_ret;
+            goto out;
+        }
+
+        op_ret = 0;
+out:
+        if (op_ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (fremovexattr, frame, op_ret, op_errno, NULL);
+        }
 
-	transaction_frame->local = local;
+        return 0;
+}
 
-	local->op_ret = -1;
+static int
+afr_fallocate_unwind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame)
+                        main_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (fallocate, main_frame, local->op_ret,
+                                  local->op_errno,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
+                                  NULL);
+        }
+        return 0;
+}
 
-	local->cont.utimens.tv[0] = tv[0];
-	local->cont.utimens.tv[1] = tv[1];
+static int
+afr_fallocate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                        struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = NULL;
+        int child_index = (long) cookie;
+        int call_count  = -1;
+        int need_unwind = 0;
+        int read_child  = 0;
 
-	local->cont.utimens.ino  = loc->inode->ino;
+        local = frame->local;
+        priv  = this->private;
 
-	local->transaction.fop    = afr_utimens_wind;
-	local->transaction.done   = afr_utimens_done;
-	local->transaction.unwind = afr_utimens_unwind;
+        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
 
-	loc_copy (&local->loc, loc);
-	
-	local->transaction.main_frame = frame;
-	local->transaction.start   = 0;
-	local->transaction.len     = 0;
-	local->transaction.pending = AFR_METADATA_PENDING;
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
 
-	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, prebuf, postbuf,
+                                       xdata);
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
+                if ((local->success_count >= priv->wait_count)
+                    && local->read_child_returned) {
+                        need_unwind = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
 
-	return 0;
-}
+        if (need_unwind)
+                local->transaction.unwind (frame, this);
 
-/* }}} */
+        call_count = afr_frame_return (frame);
 
-/* {{{ setxattr */
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+        }
 
+        return 0;
+}
 
-int
-afr_setxattr_unwind (call_frame_t *frame, xlator_t *this)
+static int
+afr_fallocate_wind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_fallocate_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->fallocate,
+                                           local->fd,
+                                           local->cont.fallocate.mode,
+                                           local->cont.fallocate.offset,
+                                           local->cont.fallocate.len,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
 
-	local = frame->local;
-	priv  = this->private;
+static int
+afr_fallocate_done (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
+        local = frame->local;
 
-	if (main_frame) {
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno)
-	}
-	return 0;
-}
+        local->transaction.unwind (frame, this);
 
+        AFR_STACK_DESTROY (frame);
 
-int
-afr_setxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		       int32_t op_ret, int32_t op_errno)
+        return 0;
+}
+
+static int
+afr_do_fallocate (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        call_frame_t * transaction_frame = NULL;
+        afr_local_t *  local             = NULL;
+        int op_ret   = -1;
+        int op_errno = 0;
 
-	int call_count  = -1;
-	int need_unwind = 0;
+        local = frame->local;
 
-	local = frame->local;
-	priv = this->private;
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                goto out;
+        }
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
-		}
+        transaction_frame->local = local;
+        frame->local = NULL;
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        local->op = GF_FOP_FALLOCATE;
 
-	if (need_unwind)
-		local->transaction.unwind (frame, this);
+        local->transaction.fop    = afr_fallocate_wind;
+        local->transaction.done   = afr_fallocate_done;
+        local->transaction.unwind = afr_fallocate_unwind;
 
-	call_count = afr_frame_return (frame);
+        local->transaction.main_frame = frame;
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
-}
+        local->transaction.start   = local->cont.fallocate.offset;
+        local->transaction.len     = 0;
 
+        /* fallocate can modify the file size */
+        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (op_ret < 0) {
+            op_errno = -op_ret;
+            goto out;
+        }
+
+        op_ret = 0;
+out:
+        if (op_ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (fallocate, frame, op_ret, op_errno, NULL,
+                                  NULL, NULL);
+        }
+
+        return 0;
+}
 
 int
-afr_setxattr_wind (call_frame_t *frame, xlator_t *this)
+afr_fallocate (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+               off_t offset, size_t len, dict_t *xdata)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
+        afr_private_t * priv  = NULL;
+        afr_local_t   * local = NULL;
+        call_frame_t   *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = 0;
 
-	int call_count = -1;
-	int i = 0;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	local = frame->local;
-	priv = this->private;
+        priv = this->private;
 
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
+        QUORUM_CHECK(fallocate,out);
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-	local->call_count = call_count;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_setxattr_wind_cbk,
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->setxattr,
-					   &local->loc, 
-					   local->cont.setxattr.dict,
-					   local->cont.setxattr.flags); 
+        local->cont.fallocate.mode = mode;
+        local->cont.fallocate.offset  = offset;
+        local->cont.fallocate.len = len;
 
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
-}
+        local->fd = fd_ref (fd);
 
+        afr_open_fd_fix (fd, this);
 
-int
-afr_setxattr_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t * local = frame->local;
+        afr_do_fallocate (frame, this);
 
-	local->transaction.unwind (frame, this);
+        ret = 0;
+out:
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+        }
 
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
+        return 0;
 }
 
+/* }}} */
 
-int
-afr_setxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, dict_t *dict, int32_t flags)
+/* {{{ discard */
+
+static int
+afr_discard_unwind (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t   *transaction_frame = NULL;
+        afr_local_t *   local = NULL;
+        call_frame_t   *main_frame = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame)
+                        main_frame = local->transaction.main_frame;
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (discard, main_frame, local->op_ret,
+                                  local->op_errno,
+                                  &local->cont.inode_wfop.prebuf,
+                                  &local->cont.inode_wfop.postbuf,
+                                  NULL);
+        }
+        return 0;
+}
 
-	int ret = -1;
+static int
+afr_discard_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = NULL;
+        int child_index = (long) cookie;
+        int call_count  = -1;
+        int need_unwind = 0;
+        int read_child  = 0;
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        local = frame->local;
+        priv  = this->private;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
 
-	priv = this->private;
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+                __inode_write_fop_cbk (frame, child_index, read_child, this,
+                                       &op_ret, &op_errno, prebuf, postbuf,
+                                       xdata);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+                if ((local->success_count >= priv->wait_count)
+                    && local->read_child_returned) {
+                        need_unwind = 1;
+                }
+        }
+        UNLOCK (&frame->lock);
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        if (need_unwind)
+                local->transaction.unwind (frame, this);
 
-	transaction_frame->local = local;
+        call_count = afr_frame_return (frame);
 
-	local->op_ret = -1;
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+        }
 
-	local->cont.setxattr.dict  = dict_ref (dict);
-	local->cont.setxattr.flags = flags;
+        return 0;
+}
 
-	local->transaction.fop    = afr_setxattr_wind;
-	local->transaction.done   = afr_setxattr_done;
-	local->transaction.unwind = afr_setxattr_unwind;
+static int
+afr_discard_wind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+        afr_private_t *priv = NULL;
+        int call_count = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_discard_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->discard,
+                                           local->fd,
+                                           local->cont.discard.offset,
+                                           local->cont.discard.len,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
 
-	loc_copy (&local->loc, loc);
+static int
+afr_discard_done (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
 
-	local->transaction.main_frame = frame;
-	local->transaction.start   = 0;
-	local->transaction.len     = 0;
-	local->transaction.pending = AFR_METADATA_PENDING;
+        local = frame->local;
 
-	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        local->transaction.unwind (frame, this);
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
+        AFR_STACK_DESTROY (frame);
 
-	return 0;
+        return 0;
 }
 
-/* }}} */
+static int
+afr_do_discard (call_frame_t *frame, xlator_t *this)
+{
+        call_frame_t * transaction_frame = NULL;
+        afr_local_t *  local             = NULL;
+        int op_ret   = -1;
+        int op_errno = 0;
 
-/* {{{ removexattr */
+        local = frame->local;
 
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                goto out;
+        }
 
-int
-afr_removexattr_unwind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
+        transaction_frame->local = local;
+        frame->local = NULL;
 
-	local = frame->local;
-	priv  = this->private;
+        local->op = GF_FOP_DISCARD;
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
+        local->transaction.fop    = afr_discard_wind;
+        local->transaction.done   = afr_discard_done;
+        local->transaction.unwind = afr_discard_unwind;
 
-	if (main_frame) {
-		AFR_STACK_UNWIND (main_frame, local->op_ret, local->op_errno)
-	}
-	return 0;
-}
+        local->transaction.main_frame = frame;
+
+        local->transaction.start   = local->cont.discard.offset;
+        local->transaction.len     = 0;
 
+        op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+        if (op_ret < 0) {
+            op_errno = -op_ret;
+            goto out;
+        }
+
+        op_ret = 0;
+out:
+        if (op_ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (discard, frame, op_ret, op_errno, NULL,
+                                  NULL, NULL);
+        }
+
+        return 0;
+}
 
 int
-afr_removexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-			  int32_t op_ret, int32_t op_errno)
+afr_discard (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             size_t len, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        afr_private_t * priv  = NULL;
+        afr_local_t   * local = NULL;
+        call_frame_t   *transaction_frame = NULL;
+        int ret = -1;
+        int op_errno = 0;
 
-	int call_count  = -1;
-	int need_unwind = 0;
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	local = frame->local;
-	priv = this->private;
+        priv = this->private;
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->wait_count) {
-				need_unwind = 1;
-			}
-		}
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
+        QUORUM_CHECK(discard, out);
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-	if (need_unwind)
-		local->transaction.unwind (frame, this);
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
 
-	call_count = afr_frame_return (frame);
+        local->cont.discard.offset  = offset;
+        local->cont.discard.len = len;
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+        local->fd = fd_ref (fd);
+
+        afr_open_fd_fix (fd, this);
+
+        afr_do_discard(frame, this);
+
+        ret = 0;
+out:
+        if (ret < 0) {
+                if (transaction_frame)
+                        AFR_STACK_DESTROY (transaction_frame);
+                AFR_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL);
+        }
+
+        return 0;
 }
 
 
-int32_t
-afr_removexattr_wind (call_frame_t *frame, xlator_t *this)
+/* {{{ zerofill */
+
+static int
+afr_zerofill_unwind (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
+        afr_local_t     *local            = NULL;
+        call_frame_t    *main_frame       = NULL;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (local->transaction.main_frame) {
+                        main_frame = local->transaction.main_frame;
+                }
+                local->transaction.main_frame = NULL;
+        }
+        UNLOCK (&frame->lock);
+
+        if (main_frame) {
+                AFR_STACK_UNWIND (zerofill, main_frame, local->op_ret,
+                                  local->op_errno,
+                                  &local->cont.zerofill.prebuf,
+                                  &local->cont.zerofill.postbuf,
+                                  NULL);
+        }
+        return 0;
+}
 
-	int call_count = -1;
-	int i = 0;
+static int
+afr_zerofill_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t       *local             = NULL;
+        afr_private_t     *priv              = NULL;
+        int                child_index       = (long) cookie;
+        int                call_count        = -1;
+        int                need_unwind       = 0;
+        int                read_child        = 0;
+
+        local = frame->local;
+        priv  = this->private;
+
+        read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL);
+
+        LOCK (&frame->lock);
+        {
+                if (child_index == read_child) {
+                        local->read_child_returned = _gf_true;
+                }
+
+                if (afr_fop_failed (op_ret, op_errno)) {
+                        afr_transaction_fop_failed (frame, this, child_index);
+                }
+
+                if (op_ret != -1) {
+                        if (local->success_count == 0) {
+                                local->op_ret = op_ret;
+                                local->cont.zerofill.prebuf  = *prebuf;
+                                local->cont.zerofill.postbuf = *postbuf;
+                        }
+
+                        if (child_index == read_child) {
+                                local->cont.zerofill.prebuf  = *prebuf;
+                                local->cont.zerofill.postbuf = *postbuf;
+                        }
+
+                        local->success_count++;
+
+                        if ((local->success_count >= priv->wait_count)
+                            && local->read_child_returned) {
+                                need_unwind = 1;
+                        }
+                }
+                local->op_errno = op_errno;
+        }
+        UNLOCK (&frame->lock);
+
+        if (need_unwind) {
+                local->transaction.unwind (frame, this);
+        }
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+        }
+
+        return 0;
+}
 
-	local = frame->local;
-	priv = this->private;
+static int
+afr_zerofill_wind (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t    *local         = NULL;
+        afr_private_t  *priv          = NULL;
+        int             call_count    = -1;
+        int             i             = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (call_count == 0) {
+                local->transaction.resume (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i]) {
+                        STACK_WIND_COOKIE (frame, afr_zerofill_wind_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->zerofill,
+                                           local->fd,
+                                           local->cont.zerofill.offset,
+                                           local->cont.zerofill.len,
+                                           NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        return 0;
+}
 
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
+static int
+afr_zerofill_done (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
 
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
+        local = frame->local;
 
-	local->call_count = call_count;
+        local->transaction.unwind (frame, this);
 
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_removexattr_wind_cbk, 
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->removexattr,
-					   &local->loc, 
-					   local->cont.removexattr.name);
+        AFR_STACK_DESTROY (frame);
 
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+        return 0;
 }
 
-
-int
-afr_removexattr_done (call_frame_t *frame, xlator_t *this)
+static int
+afr_do_zerofill(call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t * local = frame->local;
+        call_frame_t  *transaction_frame = NULL;
+        afr_local_t   *local             = NULL;
+        int            op_ret            = -1;
+        int            op_errno          = 0;
 
-	local->transaction.unwind (frame, this);
+        local = frame->local;
 
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
-}
+        transaction_frame = copy_frame (frame);
+        if (!transaction_frame) {
+                goto out;
+        }
 
+        transaction_frame->local = local;
+        frame->local = NULL;
 
-int
-afr_removexattr (call_frame_t *frame, xlator_t *this,
-		 loc_t *loc, const char *name)
-{
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t   *transaction_frame = NULL;
+        local->op = GF_FOP_ZEROFILL;
 
-	int ret = -1;
+        local->transaction.fop    = afr_zerofill_wind;
+        local->transaction.done   = afr_zerofill_done;
+        local->transaction.unwind = afr_zerofill_unwind;
 
-	int op_ret   = -1;
-	int op_errno = 0;
+        local->transaction.main_frame = frame;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-	VALIDATE_OR_GOTO (loc, out);
+        local->transaction.start   = local->cont.zerofill.offset;
+        local->transaction.len     = 0;
 
-	priv = this->private;
+        op_ret = afr_transaction (transaction_frame, this,
+                                  AFR_DATA_TRANSACTION);
+        if (op_ret < 0) {
+                op_errno = -op_ret;
+                goto out;
+        }
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
+        op_ret = 0;
+out:
+        if (op_ret < 0) {
+                if (transaction_frame) {
+                        AFR_STACK_DESTROY (transaction_frame);
+                }
+                AFR_STACK_UNWIND (zerofill, frame, op_ret, op_errno, NULL,
+                                  NULL, NULL);
+        }
+
+        return 0;
+}
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+int
+afr_zerofill (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              size_t len, dict_t *xdata)
+{
+        afr_private_t   *priv               = NULL;
+        afr_local_t     *local              = NULL;
+        call_frame_t    *transaction_frame  = NULL;
+        int              ret                = -1;
+        int              op_errno           = 0;
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	transaction_frame->local = local;
+        priv = this->private;
 
-	local->op_ret = -1;
+        if (afr_is_split_brain (this, fd->inode)) {
+                op_errno = EIO;
+                goto out;
+        }
+        QUORUM_CHECK(zerofill, out);
 
-	local->cont.removexattr.name = strdup (name);
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
 
-	local->transaction.fop    = afr_removexattr_wind;
-	local->transaction.done   = afr_removexattr_done;
-	local->transaction.unwind = afr_removexattr_unwind;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0) {
+                goto out;
+        }
+        local->cont.zerofill.offset  = offset;
+        local->cont.zerofill.len = len;
 
-	loc_copy (&local->loc, loc);
+        local->fd = fd_ref (fd);
 
-	local->transaction.main_frame = frame;
-	local->transaction.start   = 0;
-	local->transaction.len     = 0;
-	local->transaction.pending = AFR_METADATA_PENDING;
+        afr_open_fd_fix (fd, this);
 
-	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+        afr_do_zerofill(frame, this);
 
-	op_ret = 0;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-
-	return 0;
+        if (ret < 0) {
+                if (transaction_frame) {
+                        AFR_STACK_DESTROY (transaction_frame);
+                }
+                AFR_STACK_UNWIND (zerofill, frame, -1, op_errno, NULL,
+                                  NULL, NULL);
+        }
+
+        return 0;
 }
+
+/* }}} */
+
+
diff --git a/xlators/cluster/afr/src/afr-inode-write.h b/xlators/cluster/afr/src/afr-inode-write.h
index 1f1a3a35a..8e93ca44a 100644
--- a/xlators/cluster/afr/src/afr-inode-write.h
+++ b/xlators/cluster/afr/src/afr-inode-write.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __INODE_WRITE_H__
@@ -22,42 +13,70 @@
 
 int32_t
 afr_chmod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode);
+	   loc_t *loc, mode_t mode, dict_t *xdata);
 
 int32_t
 afr_chown (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, uid_t uid, gid_t gid);
+	   loc_t *loc, uid_t uid, gid_t gid, dict_t *xdata);
 
 int
 afr_fchown (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, uid_t uid, gid_t gid);
+	    fd_t *fd, uid_t uid, gid_t gid, dict_t *xdata);
 
 int32_t
 afr_fchmod (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, mode_t mode);
+	    fd_t *fd, mode_t mode, dict_t *xdata);
 
 int32_t
-afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, 
-	    struct iovec *vector, int32_t count, off_t offset);
+afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
+	    struct iovec *vector, int32_t count, off_t offset,
+            uint32_t flags, struct iobref *iobref, dict_t *xdata);
 
 int32_t
 afr_truncate (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, off_t offset);
+	      loc_t *loc, off_t offset, dict_t *xdata);
 
 int32_t
 afr_ftruncate (call_frame_t *frame, xlator_t *this,
-	       fd_t *fd, off_t offset);
+	       fd_t *fd, off_t offset, dict_t *xdata);
 
 int32_t
 afr_utimens (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, struct timespec tv[2]);
+	     loc_t *loc, struct timespec tv[2], dict_t *xdata);
+
+int
+afr_setattr (call_frame_t *frame, xlator_t *this,
+             loc_t *loc, struct iatt *buf, int32_t valid, dict_t *xdata);
+
+int
+afr_fsetattr (call_frame_t *frame, xlator_t *this,
+              fd_t *fd, struct iatt *buf, int32_t valid, dict_t *xdata);
 
 int32_t
 afr_setxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, dict_t *dict, int32_t flags);
+              loc_t *loc, dict_t *dict, int32_t flags, dict_t *xdata);
+
+int32_t
+afr_fsetxattr (call_frame_t *frame, xlator_t *this,
+               fd_t *fd, dict_t *dict, int32_t flags, dict_t *xdata);
 
 int32_t
 afr_removexattr (call_frame_t *frame, xlator_t *this,
-		 loc_t *loc, const char *name);
+		 loc_t *loc, const char *name, dict_t *xdata);
 
+int32_t
+afr_fremovexattr (call_frame_t *frame, xlator_t *this,
+                  fd_t *fd, const char *name, dict_t *xdata);
+
+int
+afr_discard (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             size_t len, dict_t *xdata);
+
+int
+afr_fallocate (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+               off_t offset, size_t len, dict_t *xdata);
+
+int
+afr_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             size_t len, dict_t *xdata);
 #endif /* __INODE_WRITE_H__ */
diff --git a/xlators/cluster/afr/src/afr-lk-common.c b/xlators/cluster/afr/src/afr-lk-common.c
new file mode 100644
index 000000000..060d78f35
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-lk-common.c
@@ -0,0 +1,2174 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "dict.h"
+#include "byte-order.h"
+#include "common-utils.h"
+
+#include "afr.h"
+#include "afr-transaction.h"
+
+#include <signal.h>
+
+
+#define LOCKED_NO       0x0        /* no lock held */
+#define LOCKED_YES      0x1        /* for DATA, METADATA, ENTRY and higher_path */
+#define LOCKED_LOWER    0x2        /* for lower path */
+
+#define AFR_TRACE_INODELK_IN(frame, this, params ...)           \
+        do {                                                    \
+                afr_private_t *_priv = this->private;           \
+                if (!_priv->inodelk_trace)                      \
+                        break;                                  \
+                afr_trace_inodelk_in (frame, this, params);     \
+        } while (0);
+
+#define AFR_TRACE_INODELK_OUT(frame, this, params ...)          \
+        do {                                                    \
+                afr_private_t *_priv = this->private;           \
+                if (!_priv->inodelk_trace)                      \
+                        break;                                  \
+                afr_trace_inodelk_out (frame, this, params);    \
+        } while (0);
+
+#define AFR_TRACE_ENTRYLK_IN(frame, this, params ...)           \
+        do {                                                    \
+                afr_private_t *_priv = this->private;           \
+                if (!_priv->entrylk_trace)                      \
+                        break;                                  \
+                afr_trace_entrylk_in (frame, this, params);     \
+        } while (0);
+
+#define AFR_TRACE_ENTRYLK_OUT(frame, this, params ...)          \
+        do {                                                    \
+                afr_private_t *_priv = this->private;           \
+                if (!_priv->entrylk_trace)                      \
+                        break;                                  \
+                afr_trace_entrylk_out (frame, this, params);    \
+        } while (0);
+
+int
+afr_entry_lockee_cmp (const void *l1, const void *l2)
+{
+        const afr_entry_lockee_t       *r1 = l1;
+        const afr_entry_lockee_t       *r2 = l2;
+        int                            ret = 0;
+        uuid_t                         gfid1 = {0};
+        uuid_t                         gfid2 = {0};
+
+        loc_gfid ((loc_t*)&r1->loc, gfid1);
+        loc_gfid ((loc_t*)&r2->loc, gfid2);
+        ret = uuid_compare (gfid1, gfid2);
+        /*Entrylks with NULL basename are the 'smallest'*/
+        if (ret == 0) {
+                if (!r1->basename)
+                        return -1;
+                if (!r2->basename)
+                        return 1;
+                ret = strcmp (r1->basename, r2->basename);
+        }
+
+        if (ret <= 0)
+                return -1;
+        else
+                return 1;
+}
+
+int afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index);
+
+static int
+afr_copy_locked_nodes (call_frame_t *frame, xlator_t *this);
+
+static uint64_t afr_lock_number = 1;
+
+static uint64_t
+get_afr_lock_number ()
+{
+        return (++afr_lock_number);
+}
+
+int
+afr_set_lock_number (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t         *local    = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        int_lock->lock_number = get_afr_lock_number ();
+
+        return 0;
+}
+
+void
+afr_set_lk_owner (call_frame_t *frame, xlator_t *this, void *lk_owner)
+{
+        gf_log (this->name, GF_LOG_TRACE,
+                "Setting lk-owner=%llu",
+                (unsigned long long) (unsigned long)lk_owner);
+
+        set_lk_owner_from_ptr (&frame->root->lk_owner, lk_owner);
+}
+
+static int
+is_afr_lock_selfheal (afr_local_t *local)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        int                  ret      = -1;
+
+        int_lock = &local->internal_lock;
+
+        switch (int_lock->selfheal_lk_type) {
+        case AFR_DATA_SELF_HEAL_LK:
+        case AFR_METADATA_SELF_HEAL_LK:
+                ret = 1;
+                break;
+        case AFR_ENTRY_SELF_HEAL_LK:
+                ret = 0;
+                break;
+        }
+
+        return ret;
+
+}
+
+int32_t
+internal_lock_count (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local = NULL;
+        afr_private_t *priv  = NULL;
+        int32_t call_count = 0;
+        int i = 0;
+
+        local = frame->local;
+        priv  = this->private;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i])
+                        ++call_count;
+        }
+
+        return call_count;
+}
+
+static void
+afr_print_inodelk (char *str, int size, int cmd,
+                   struct gf_flock *flock, gf_lkowner_t *owner)
+{
+        char *cmd_str = NULL;
+        char *type_str = NULL;
+
+        switch (cmd) {
+#if F_GETLK != F_GETLK64
+        case F_GETLK64:
+#endif
+        case F_GETLK:
+                cmd_str = "GETLK";
+                break;
+
+#if F_SETLK != F_SETLK64
+        case F_SETLK64:
+#endif
+        case F_SETLK:
+                cmd_str = "SETLK";
+                break;
+
+#if F_SETLKW != F_SETLKW64
+        case F_SETLKW64:
+#endif
+        case F_SETLKW:
+                cmd_str = "SETLKW";
+                break;
+
+        default:
+                cmd_str = "<null>";
+                break;
+        }
+
+        switch (flock->l_type) {
+        case F_RDLCK:
+                type_str = "READ";
+                break;
+        case F_WRLCK:
+                type_str = "WRITE";
+                break;
+        case F_UNLCK:
+                type_str = "UNLOCK";
+                break;
+        default:
+                type_str = "UNKNOWN";
+                break;
+        }
+
+        snprintf (str, size, "lock=INODELK, cmd=%s, type=%s, "
+                  "start=%llu, len=%llu, pid=%llu, lk-owner=%s",
+                  cmd_str, type_str, (unsigned long long) flock->l_start,
+                  (unsigned long long) flock->l_len,
+                  (unsigned long long) flock->l_pid,
+                  lkowner_utoa (owner));
+
+}
+
+static void
+afr_print_lockee (char *str, int size, loc_t *loc, fd_t *fd,
+                  int child_index)
+{
+        snprintf (str, size, "path=%s, fd=%p, child=%d",
+                  loc->path ? loc->path : "<nul>",
+                  fd ? fd : NULL,
+                  child_index);
+}
+
+void
+afr_print_entrylk (char *str, int size, const char *basename,
+                   gf_lkowner_t *owner)
+{
+        snprintf (str, size, "Basename=%s, lk-owner=%s",
+                  basename ? basename : "<nul>",
+                  lkowner_utoa (owner));
+}
+
+static void
+afr_print_verdict (int op_ret, int op_errno, char *str)
+{
+        if (op_ret < 0) {
+                if (op_errno == EAGAIN)
+                        strcpy (str, "EAGAIN");
+                else
+                        strcpy (str, "FAILED");
+        }
+        else
+                strcpy (str, "GRANTED");
+}
+
+static void
+afr_set_lock_call_type (afr_lock_call_type_t lock_call_type,
+                        char *lock_call_type_str,
+                        afr_internal_lock_t *int_lock)
+{
+        switch (lock_call_type) {
+        case AFR_INODELK_TRANSACTION:
+                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
+                        strcpy (lock_call_type_str, "AFR_INODELK_TRANSACTION");
+                else
+                        strcpy (lock_call_type_str, "AFR_INODELK_SELFHEAL");
+                break;
+        case AFR_INODELK_NB_TRANSACTION:
+                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
+                        strcpy (lock_call_type_str, "AFR_INODELK_NB_TRANSACTION");
+                else
+                        strcpy (lock_call_type_str, "AFR_INODELK_NB_SELFHEAL");
+                break;
+        case AFR_ENTRYLK_TRANSACTION:
+                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
+                        strcpy (lock_call_type_str, "AFR_ENTRYLK_TRANSACTION");
+                else
+                        strcpy (lock_call_type_str, "AFR_ENTRYLK_SELFHEAL");
+                break;
+        case AFR_ENTRYLK_NB_TRANSACTION:
+                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
+                        strcpy (lock_call_type_str, "AFR_ENTRYLK_NB_TRANSACTION");
+                else
+                        strcpy (lock_call_type_str, "AFR_ENTRYLK_NB_SELFHEAL");
+                break;
+        default:
+                strcpy (lock_call_type_str, "UNKNOWN");
+                break;
+        }
+
+}
+
+static void
+afr_trace_inodelk_out (call_frame_t *frame, xlator_t *this,
+                       afr_lock_call_type_t lock_call_type,
+                       afr_lock_op_type_t lk_op_type, struct gf_flock *flock,
+                       int op_ret, int op_errno, int32_t child_index)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+
+        char lockee[256];
+        char lock_call_type_str[256];
+        char verdict[16];
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index);
+
+        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
+
+        afr_print_verdict (op_ret, op_errno, verdict);
+
+        gf_log (this->name, GF_LOG_INFO,
+                "[%s %s] [%s] lk-owner=%s Lockee={%s} Number={%llu}",
+                lock_call_type_str,
+                lk_op_type == AFR_LOCK_OP ? "LOCK REPLY" : "UNLOCK REPLY",
+                verdict, lkowner_utoa (&frame->root->lk_owner), lockee,
+                (unsigned long long) int_lock->lock_number);
+
+}
+
+static void
+afr_trace_inodelk_in (call_frame_t *frame, xlator_t *this,
+                      afr_lock_call_type_t lock_call_type,
+                      afr_lock_op_type_t lk_op_type, struct gf_flock *flock,
+                      int32_t cmd, int32_t child_index)
+{
+        afr_local_t         *local    = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+
+        char lock[256];
+        char lockee[256];
+        char lock_call_type_str[256];
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        afr_print_inodelk (lock, 256, cmd, flock, &frame->root->lk_owner);
+        afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index);
+
+        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
+
+        gf_log (this->name, GF_LOG_INFO,
+                "[%s %s] Lock={%s} Lockee={%s} Number={%llu}",
+                lock_call_type_str,
+                lk_op_type == AFR_LOCK_OP ? "LOCK REQUEST" : "UNLOCK REQUEST",
+                lock, lockee,
+                (unsigned long long) int_lock->lock_number);
+
+}
+
+static void
+afr_trace_entrylk_in (call_frame_t *frame, xlator_t *this,
+                      afr_lock_call_type_t lock_call_type,
+                      afr_lock_op_type_t lk_op_type, const char *basename,
+                      int32_t cookie)
+{
+        afr_local_t         *local    = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_private_t       *priv     = NULL;
+        int                 child_index = 0;
+        int                 lockee_no = 0;
+
+        char lock[256];
+        char lockee[256];
+        char lock_call_type_str[256];
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        priv     = this->private;
+
+        if (!priv->entrylk_trace) {
+                return;
+        }
+        lockee_no = cookie / priv->child_count;
+        child_index = cookie % priv->child_count;
+
+        afr_print_entrylk (lock, 256, basename, &frame->root->lk_owner);
+        afr_print_lockee (lockee, 256, &int_lock->lockee[lockee_no].loc, local->fd,
+                          child_index);
+
+        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
+
+        gf_log (this->name, GF_LOG_INFO,
+                "[%s %s] Lock={%s} Lockee={%s} Number={%llu}, Cookie={%d}",
+                lock_call_type_str,
+                lk_op_type == AFR_LOCK_OP ? "LOCK REQUEST" : "UNLOCK REQUEST",
+                lock, lockee,
+                (unsigned long long) int_lock->lock_number,
+                cookie);
+}
+
+static void
+afr_trace_entrylk_out (call_frame_t *frame, xlator_t *this,
+                       afr_lock_call_type_t lock_call_type,
+                       afr_lock_op_type_t lk_op_type, const char *basename,
+                       int op_ret, int op_errno, int32_t cookie)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+        afr_private_t       *priv     = NULL;
+        int                 lockee_no = 0;
+        int                 child_index = 0;
+
+        char lock[256];
+        char lockee[256];
+        char lock_call_type_str[256];
+        char verdict[16];
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        priv     = this->private;
+
+        if (!priv->entrylk_trace) {
+                return;
+        }
+        lockee_no = cookie / priv->child_count;
+        child_index = cookie % priv->child_count;
+
+        afr_print_entrylk (lock, 256, basename, &frame->root->lk_owner);
+        afr_print_lockee (lockee, 256, &int_lock->lockee[lockee_no].loc, local->fd,
+                          child_index);
+
+        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
+
+        afr_print_verdict (op_ret, op_errno, verdict);
+
+        gf_log (this->name, GF_LOG_INFO,
+                "[%s %s] [%s] Lock={%s} Lockee={%s} Number={%llu} Cookie={%d}",
+                lock_call_type_str,
+                lk_op_type == AFR_LOCK_OP ? "LOCK REPLY" : "UNLOCK REPLY",
+                verdict,
+                lock, lockee,
+                (unsigned long long) int_lock->lock_number,
+                cookie);
+
+}
+
+static int
+transaction_lk_op (afr_local_t *local)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        int ret = -1;
+
+        int_lock = &local->internal_lock;
+
+        if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK) {
+                gf_log (THIS->name, GF_LOG_DEBUG,
+                        "lk op is for a transaction");
+                ret = 1;
+        }
+        else if (int_lock->transaction_lk_type == AFR_SELFHEAL_LK) {
+                gf_log (THIS->name, GF_LOG_DEBUG,
+                        "lk op is for a self heal");
+
+                ret = 0;
+        }
+
+        if (ret == -1)
+                gf_log (THIS->name, GF_LOG_DEBUG,
+                        "lk op is not set");
+
+        return ret;
+
+}
+
+static int
+is_afr_lock_transaction (afr_local_t *local)
+{
+        int ret = 0;
+
+        switch (local->transaction.type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+                ret = 1;
+                break;
+
+        case AFR_ENTRY_RENAME_TRANSACTION:
+        case AFR_ENTRY_TRANSACTION:
+                ret = 0;
+                break;
+
+        }
+
+        return ret;
+}
+
+int
+afr_init_entry_lockee (afr_entry_lockee_t *lockee, afr_local_t *local,
+                       loc_t *loc, char *basename, int child_count)
+{
+        int     ret     = -1;
+
+        loc_copy (&lockee->loc, loc);
+        lockee->basename        = (basename)? gf_strdup (basename): NULL;
+        if (basename && !lockee->basename)
+                goto out;
+
+        lockee->locked_count    = 0;
+        lockee->locked_nodes    = GF_CALLOC (child_count,
+                                             sizeof (*lockee->locked_nodes),
+                                             gf_afr_mt_afr_node_character);
+
+        if (!lockee->locked_nodes)
+                goto out;
+
+        ret = 0;
+out:
+        return ret;
+
+}
+
+void
+afr_entry_lockee_cleanup (afr_internal_lock_t *int_lock)
+{
+        int     i   = 0;
+
+        for (i = 0; i < int_lock->lockee_count; i++) {
+                loc_wipe (&int_lock->lockee[i].loc);
+                if (int_lock->lockee[i].basename)
+                        GF_FREE (int_lock->lockee[i].basename);
+                if (int_lock->lockee[i].locked_nodes)
+                       GF_FREE (int_lock->lockee[i].locked_nodes);
+        }
+
+        return;
+}
+
+static int
+initialize_entrylk_variables (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t         *local    = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_private_t       *priv     = NULL;
+
+        int i = 0;
+
+        priv     = this->private;
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        int_lock->entrylk_lock_count = 0;
+        int_lock->lock_op_ret        = -1;
+        int_lock->lock_op_errno      = 0;
+
+        for (i = 0; i < AFR_LOCKEE_COUNT_MAX; i++) {
+                if (!int_lock->lockee[i].locked_nodes)
+                        break;
+                int_lock->lockee[i].locked_count = 0;
+                memset (int_lock->lockee[i].locked_nodes, 0,
+                        sizeof (*int_lock->lockee[i].locked_nodes) *
+                        priv->child_count);
+        }
+
+        return 0;
+}
+
+static int
+initialize_inodelk_variables (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t         *local    = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_private_t       *priv     = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
+
+        priv     = this->private;
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+
+        inodelk->lock_count    = 0;
+        int_lock->lk_attempted_count = 0;
+        int_lock->lock_op_ret   = -1;
+        int_lock->lock_op_errno = 0;
+
+        memset (inodelk->locked_nodes, 0,
+                sizeof (*inodelk->locked_nodes) * priv->child_count);
+        memset (int_lock->locked_nodes, 0,
+                sizeof (*int_lock->locked_nodes) * priv->child_count);
+
+        return 0;
+}
+
+loc_t *
+lower_path (loc_t *l1, const char *b1, loc_t *l2, const char *b2)
+{
+        int ret = 0;
+
+        ret = uuid_compare (l1->inode->gfid, l2->inode->gfid);
+
+        if (ret == 0)
+                ret = strcmp (b1, b2);
+
+        if (ret <= 0)
+                return l1;
+        else
+                return l2;
+}
+
+int
+afr_lockee_locked_nodes_count (afr_internal_lock_t *int_lock)
+{
+        int call_count  = 0;
+        int i           = 0;
+
+        for (i = 0; i < int_lock->lockee_count; i++)
+                call_count += int_lock->lockee[i].locked_count;
+
+        return call_count;
+}
+
+int
+afr_locked_nodes_count (unsigned char *locked_nodes, int child_count)
+
+{
+        int i = 0;
+        int call_count = 0;
+
+        for (i = 0; i < child_count; i++) {
+                if (locked_nodes[i] & LOCKED_YES)
+                        call_count++;
+        }
+
+        return call_count;
+}
+
+/* FIXME: What if UNLOCK fails */
+static int32_t
+afr_unlock_common_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_local_t         *local    = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+        int call_count = 0;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        LOCK (&frame->lock);
+        {
+                call_count = --int_lock->lk_call_count;
+        }
+        UNLOCK (&frame->lock);
+
+        if (call_count == 0) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "All internal locks unlocked");
+                int_lock->lock_cbk (frame, this);
+        }
+
+        return 0;
+}
+
+static int32_t
+afr_unlock_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_local_t         *local = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk = NULL;
+        int32_t             child_index = (long)cookie;
+        afr_private_t       *priv = NULL;
+
+        local = frame->local;
+        int_lock = &local->internal_lock;
+
+        AFR_TRACE_INODELK_OUT (frame, this, AFR_INODELK_TRANSACTION,
+                               AFR_UNLOCK_OP, NULL, op_ret,
+                               op_errno, child_index);
+
+        priv = this->private;
+
+        if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
+                gf_log (this->name, GF_LOG_INFO, "%s: unlock failed on subvolume %s "
+                        "with lock owner %s", local->loc.path,
+                        priv->children[child_index]->name,
+                        lkowner_utoa (&frame->root->lk_owner));
+        }
+
+
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+        inodelk->locked_nodes[child_index] &= LOCKED_NO;
+        if (local->transaction.eager_lock)
+                local->transaction.eager_lock[child_index] = 0;
+
+        afr_unlock_common_cbk (frame, cookie, this, op_ret, op_errno, xdata);
+
+        return 0;
+
+}
+
+static int
+afr_unlock_inodelk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
+        afr_local_t         *local    = NULL;
+        afr_private_t       *priv     = NULL;
+        struct gf_flock flock = {0,};
+        struct gf_flock full_flock = {0,};
+        struct gf_flock *flock_use = NULL;
+        int call_count = 0;
+        int i = 0;
+        int piggyback = 0;
+        afr_fd_ctx_t        *fd_ctx      = NULL;
+
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        priv     = this->private;
+
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+
+        flock.l_start = inodelk->flock.l_start;
+        flock.l_len   = inodelk->flock.l_len;
+        flock.l_type  = F_UNLCK;
+
+        full_flock.l_type = F_UNLCK;
+        call_count = afr_locked_nodes_count (inodelk->locked_nodes,
+                                             priv->child_count);
+
+        int_lock->lk_call_count = call_count;
+
+        if (!call_count) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "No internal locks unlocked");
+                int_lock->lock_cbk (frame, this);
+                goto out;
+        }
+
+        if (local->fd)
+                fd_ctx = afr_fd_ctx_get (local->fd, this);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if ((inodelk->locked_nodes[i] & LOCKED_YES) != LOCKED_YES)
+                        continue;
+
+                if (local->fd) {
+                        flock_use = &flock;
+                        if (!local->transaction.eager_lock[i]) {
+                                goto wind;
+                        }
+
+                        piggyback = 0;
+
+                        LOCK (&local->fd->lock);
+                        {
+                                if (fd_ctx->lock_piggyback[i]) {
+                                        fd_ctx->lock_piggyback[i]--;
+                                        piggyback = 1;
+                                } else {
+                                        fd_ctx->lock_acquired[i]--;
+                                }
+                        }
+                        UNLOCK (&local->fd->lock);
+
+                        if (piggyback) {
+                                afr_unlock_inodelk_cbk (frame, (void *) (long) i,
+                                                        this, 1, 0, NULL);
+                                if (!--call_count)
+                                        break;
+                                continue;
+                        }
+
+                        flock_use = &full_flock;
+                wind:
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_TRANSACTION,
+                                              AFR_UNLOCK_OP, flock_use, F_SETLK,
+                                              i);
+
+                        STACK_WIND_COOKIE (frame, afr_unlock_inodelk_cbk,
+                                           (void *) (long)i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->finodelk,
+                                           int_lock->domain, local->fd,
+                                           F_SETLK, flock_use, NULL);
+
+                        if (!--call_count)
+                                break;
+
+                } else {
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_TRANSACTION,
+                                              AFR_UNLOCK_OP, &flock, F_SETLK, i);
+
+                        STACK_WIND_COOKIE (frame, afr_unlock_inodelk_cbk,
+                                           (void *) (long)i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->inodelk,
+                                           int_lock->domain, &local->loc,
+                                           F_SETLK, &flock, NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+out:
+        return 0;
+}
+
+static int32_t
+afr_unlock_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_local_t         *local = NULL;
+        afr_private_t       *priv  = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+        int32_t             child_index = 0;
+        int                 lockee_no   = 0;
+
+        priv = this->private;
+        lockee_no = (int)((long) cookie) / priv->child_count;
+        child_index = (int) ((long) cookie) % priv->child_count;
+
+        local = frame->local;
+        int_lock = &local->internal_lock;
+
+        AFR_TRACE_ENTRYLK_OUT (frame, this, AFR_ENTRYLK_TRANSACTION,
+                               AFR_UNLOCK_OP,
+                               int_lock->lockee[lockee_no].basename, op_ret,
+                               op_errno, (int) ((long)cookie));
+
+        if (op_ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: unlock failed on %d, reason: %s",
+                        local->loc.path, child_index, strerror (op_errno));
+        }
+
+        int_lock->lockee[lockee_no].locked_nodes[child_index] &= LOCKED_NO;
+        afr_unlock_common_cbk (frame, cookie, this, op_ret, op_errno, NULL);
+
+        return 0;
+}
+
+static int
+afr_unlock_entrylk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t     *int_lock       = NULL;
+        afr_local_t             *local          = NULL;
+        afr_private_t           *priv           = NULL;
+        int                     call_count      = 0;
+        int                     index           = 0;
+        int                     lockee_no       = 0;
+        int                     copies          = 0;
+        int                     i               = -1;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        priv     = this->private;
+        copies   = priv->child_count;
+
+        call_count = afr_lockee_locked_nodes_count (int_lock);
+
+        int_lock->lk_call_count = call_count;
+
+        if (!call_count){
+                gf_log (this->name, GF_LOG_TRACE,
+                        "No internal locks unlocked");
+                int_lock->lock_cbk (frame, this);
+                goto out;
+        }
+
+        for (i = 0; i < int_lock->lockee_count * priv->child_count; i++) {
+                lockee_no = i / copies;
+                index     = i % copies;
+                if (int_lock->lockee[lockee_no].locked_nodes[index] & LOCKED_YES) {
+                        AFR_TRACE_ENTRYLK_IN (frame, this, AFR_ENTRYLK_NB_TRANSACTION,
+                                              AFR_UNLOCK_OP,
+                                              int_lock->lockee[lockee_no].basename,
+                                              i);
+
+                        STACK_WIND_COOKIE (frame, afr_unlock_entrylk_cbk,
+                                           (void *) (long) i,
+                                           priv->children[index],
+                                           priv->children[index]->fops->entrylk,
+                                           int_lock->domain,
+                                           &int_lock->lockee[lockee_no].loc,
+                                           int_lock->lockee[lockee_no].basename,
+                                           ENTRYLK_UNLOCK, ENTRYLK_WRLCK, NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+out:
+        return 0;
+
+}
+
+static int32_t
+afr_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+              int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_internal_lock_t     *int_lock       = NULL;
+        afr_local_t             *local          = NULL;
+        afr_private_t           *priv           = NULL;
+        int                     cky             = (long) cookie;
+        int                     child_index     = 0;
+        int                     lockee_no       = 0;
+
+        priv     = this->private;
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        child_index = ((int)cky) % priv->child_count;
+        lockee_no   = ((int)cky) / priv->child_count;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        if (op_errno == ENOSYS) {
+                                /* return ENOTSUP */
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "subvolume does not support locking. "
+                                        "please load features/locks xlator on server");
+                                local->op_ret = op_ret;
+                                int_lock->lock_op_ret = op_ret;
+                        }
+
+                        local->op_errno              = op_errno;
+                        int_lock->lock_op_errno      = op_errno;
+                }
+
+		int_lock->lk_attempted_count++;
+        }
+        UNLOCK (&frame->lock);
+
+        if ((op_ret == -1) &&
+            (op_errno == ENOSYS)) {
+                afr_unlock (frame, this);
+        } else {
+                if (op_ret == 0) {
+                        if (local->transaction.type == AFR_ENTRY_TRANSACTION ||
+                            local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
+                                int_lock->lockee[lockee_no].locked_nodes[child_index] |= LOCKED_YES;
+                                int_lock->lockee[lockee_no].locked_count++;
+                                int_lock->entrylk_lock_count++;
+                        } else {
+                                int_lock->locked_nodes[child_index] |= LOCKED_YES;
+                                int_lock->lock_count++;
+                        }
+                }
+                afr_lock_blocking (frame, this, cky + 1);
+        }
+
+        return 0;
+}
+
+static int32_t
+afr_blocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        AFR_TRACE_INODELK_OUT (frame, this, AFR_INODELK_TRANSACTION,
+                               AFR_LOCK_OP, NULL, op_ret,
+                               op_errno, (long) cookie);
+
+        afr_lock_cbk (frame, cookie, this, op_ret, op_errno, xdata);
+        return 0;
+
+}
+
+static int32_t
+afr_blocking_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        AFR_TRACE_ENTRYLK_OUT (frame, this, AFR_ENTRYLK_TRANSACTION,
+                               AFR_LOCK_OP, NULL, op_ret,
+                               op_errno, (long)cookie);
+
+        afr_lock_cbk (frame, cookie, this, op_ret, op_errno, xdata);
+        return 0;
+}
+
+static int
+afr_copy_locked_nodes (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
+        afr_local_t         *local    = NULL;
+        afr_private_t       *priv     = NULL;
+
+        priv     = this->private;
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        switch (local->transaction.type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+                inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+                memcpy (inodelk->locked_nodes, int_lock->locked_nodes,
+                        sizeof (*inodelk->locked_nodes) * priv->child_count);
+                inodelk->lock_count = int_lock->lock_count;
+                break;
+
+        case AFR_ENTRY_RENAME_TRANSACTION:
+        case AFR_ENTRY_TRANSACTION:
+                /*entrylk_count is being used in both non-blocking and blocking
+                 * modes */
+                break;
+        }
+
+        return 0;
+
+}
+
+static inline gf_boolean_t
+afr_is_entrylk (afr_internal_lock_t *int_lock,
+                afr_transaction_type trans_type)
+{
+        gf_boolean_t is_entrylk = _gf_false;
+
+        if ((int_lock->transaction_lk_type == AFR_SELFHEAL_LK) &&
+            int_lock->selfheal_lk_type == AFR_ENTRY_SELF_HEAL_LK) {
+
+                is_entrylk = _gf_true;
+
+        } else if ((int_lock->transaction_lk_type == AFR_TRANSACTION_LK) &&
+                 (trans_type == AFR_ENTRY_TRANSACTION ||
+                  trans_type == AFR_ENTRY_RENAME_TRANSACTION)) {
+
+                is_entrylk = _gf_true;
+
+        } else {
+                is_entrylk = _gf_false;
+        }
+
+        return is_entrylk;
+}
+
+static gf_boolean_t
+_is_lock_wind_needed (afr_local_t *local, int child_index)
+{
+        if (!local->child_up[child_index])
+                return _gf_false;
+
+        return _gf_true;
+}
+
+int
+afr_lock_blocking (call_frame_t *frame, xlator_t *this, int cookie)
+{
+        afr_internal_lock_t *int_lock    = NULL;
+        afr_inodelk_t       *inodelk     = NULL;
+        afr_local_t         *local       = NULL;
+        afr_private_t       *priv        = NULL;
+        struct gf_flock flock = {0,};
+        uint64_t ctx = 0;
+        int ret = 0;
+        int child_index = 0;
+        int lockee_no   = 0;
+        gf_boolean_t is_entrylk = _gf_false;
+
+        local         = frame->local;
+        int_lock      = &local->internal_lock;
+        priv          = this->private;
+        child_index   = cookie % priv->child_count;
+        lockee_no     = cookie / priv->child_count;
+        is_entrylk    = afr_is_entrylk (int_lock, local->transaction.type);
+
+
+        if (!is_entrylk) {
+                inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+                flock.l_start = inodelk->flock.l_start;
+                flock.l_len   = inodelk->flock.l_len;
+                flock.l_type  = inodelk->flock.l_type;
+        }
+
+        if (local->fd) {
+                ret = fd_ctx_get (local->fd, this, &ctx);
+
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "unable to get fd ctx for fd=%p",
+                                local->fd);
+
+                        local->op_ret           = -1;
+                        int_lock->lock_op_ret   = -1;
+
+                        afr_copy_locked_nodes (frame, this);
+
+                        afr_unlock (frame, this);
+
+                        return 0;
+                }
+        }
+
+        if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
+                if ((is_entrylk && int_lock->entrylk_lock_count == 0) ||
+                    (!is_entrylk && int_lock->lock_count == 0)) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "unable to lock on even one child");
+
+                        local->op_ret           = -1;
+                        int_lock->lock_op_ret   = -1;
+
+                        afr_copy_locked_nodes (frame, this);
+
+                        afr_unlock(frame, this);
+
+                        return 0;
+                }
+        }
+
+        if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
+                /* we're done locking */
+
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "we're done locking");
+
+                afr_copy_locked_nodes (frame, this);
+
+                int_lock->lock_op_ret = 0;
+                int_lock->lock_cbk (frame, this);
+                return 0;
+        }
+
+        if (!_is_lock_wind_needed (local, child_index)) {
+                afr_lock_blocking (frame, this, cookie + 1);
+                return 0;
+        }
+
+        switch (local->transaction.type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+
+                if (local->fd) {
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_TRANSACTION,
+                                              AFR_LOCK_OP, &flock, F_SETLKW,
+                                              child_index);
+
+                        STACK_WIND_COOKIE (frame, afr_blocking_inodelk_cbk,
+                                           (void *) (long) child_index,
+                                           priv->children[child_index],
+                                           priv->children[child_index]->fops->finodelk,
+                                           int_lock->domain, local->fd,
+                                           F_SETLKW, &flock, NULL);
+
+                } else {
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_TRANSACTION,
+                                              AFR_LOCK_OP, &flock, F_SETLKW,
+                                              child_index);
+
+                        STACK_WIND_COOKIE (frame, afr_blocking_inodelk_cbk,
+                                           (void *) (long) child_index,
+                                           priv->children[child_index],
+                                           priv->children[child_index]->fops->inodelk,
+                                           int_lock->domain, &local->loc,
+                                           F_SETLKW, &flock, NULL);
+                }
+
+                break;
+
+        case AFR_ENTRY_RENAME_TRANSACTION:
+        case AFR_ENTRY_TRANSACTION:
+                /*Accounting for child_index increments on 'down'
+                 *and 'fd-less' children */
+
+                if (local->fd) {
+                        AFR_TRACE_ENTRYLK_IN (frame, this, AFR_ENTRYLK_TRANSACTION,
+                                              AFR_LOCK_OP,
+                                              int_lock->lockee[lockee_no].basename,
+                                              cookie);
+
+                        STACK_WIND_COOKIE (frame, afr_blocking_entrylk_cbk,
+                                           (void *) (long) cookie,
+                                           priv->children[child_index],
+                                           priv->children[child_index]->fops->fentrylk,
+                                           int_lock->domain, local->fd,
+                                           int_lock->lockee[lockee_no].basename,
+                                           ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
+                } else {
+                        AFR_TRACE_ENTRYLK_IN (frame, this,
+                                              AFR_ENTRYLK_TRANSACTION,
+                                              AFR_LOCK_OP, local->transaction.basename,
+                                              child_index);
+
+                        STACK_WIND_COOKIE (frame, afr_blocking_entrylk_cbk,
+                                           (void *) (long) cookie,
+                                           priv->children[child_index],
+                                           priv->children[child_index]->fops->entrylk,
+                                           int_lock->domain,
+                                           &int_lock->lockee[lockee_no].loc,
+                                           int_lock->lockee[lockee_no].basename,
+                                           ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
+                }
+
+                break;
+        }
+
+        return 0;
+}
+
+int32_t
+afr_blocking_lock (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+        afr_private_t       *priv     = NULL;
+        int                  up_count = 0;
+
+        priv     = this->private;
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        switch (local->transaction.type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+                initialize_inodelk_variables (frame, this);
+                break;
+
+        case AFR_ENTRY_RENAME_TRANSACTION:
+        case AFR_ENTRY_TRANSACTION:
+                up_count = afr_up_children_count (local->child_up,
+                                                  priv->child_count);
+                int_lock->lk_call_count = int_lock->lk_expected_count
+                                        = (int_lock->lockee_count *
+                                           up_count);
+                initialize_entrylk_variables (frame, this);
+                break;
+        }
+
+        afr_lock_blocking (frame, this, 0);
+
+        return 0;
+}
+
+static int32_t
+afr_nonblocking_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+        int call_count          = 0;
+        int child_index         = (long) cookie;
+        int copies              = 0;
+        int index               = 0;
+        int lockee_no           = 0;
+        afr_private_t       *priv = NULL;
+
+        priv = this->private;
+
+        copies = priv->child_count;
+        index = child_index % copies;
+        lockee_no = child_index / copies;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        AFR_TRACE_ENTRYLK_OUT (frame, this, AFR_ENTRYLK_TRANSACTION,
+                               AFR_LOCK_OP,
+                               int_lock->lockee[lockee_no].basename, op_ret,
+                               op_errno, (long) cookie);
+
+	LOCK (&frame->lock);
+	{
+		if (op_ret < 0 ) {
+			if (op_errno == ENOSYS) {
+                        /* return ENOTSUP */
+				gf_log (this->name, GF_LOG_ERROR,
+					"subvolume does not support locking. "
+					"please load features/locks xlator on server");
+				local->op_ret         = op_ret;
+				int_lock->lock_op_ret = op_ret;
+
+				int_lock->lock_op_errno      = op_errno;
+				local->op_errno              = op_errno;
+			}
+		} else if (op_ret == 0) {
+			int_lock->lockee[lockee_no].locked_nodes[index] |= \
+				LOCKED_YES;
+			int_lock->lockee[lockee_no].locked_count++;
+			int_lock->entrylk_lock_count++;
+		}
+
+                call_count = --int_lock->lk_call_count;
+        }
+        UNLOCK (&frame->lock);
+
+        if (call_count == 0) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "Last locking reply received");
+                /* all locks successful. Proceed to call FOP */
+                if (int_lock->entrylk_lock_count ==
+                                int_lock->lk_expected_count) {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "All servers locked. Calling the cbk");
+                        int_lock->lock_op_ret = 0;
+                        int_lock->lock_cbk (frame, this);
+                }
+                /* Not all locks were successful. Unlock and try locking
+                   again, this time with serially blocking locks */
+                else {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "%d servers locked. Trying again with blocking calls",
+                                int_lock->lock_count);
+
+                        afr_unlock(frame, this);
+                }
+        }
+
+        return 0;
+}
+
+int
+afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock   = NULL;
+        afr_local_t         *local      = NULL;
+        afr_private_t       *priv       = NULL;
+        afr_fd_ctx_t        *fd_ctx     = NULL;
+        int                 copies      = 0;
+        int                 index       = 0;
+        int                 lockee_no   = 0;
+        int32_t             call_count  = 0;
+        int i = 0;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        priv     = this->private;
+
+        copies = priv->child_count;
+        initialize_entrylk_variables (frame, this);
+
+        if (local->fd) {
+                fd_ctx = afr_fd_ctx_get (local->fd, this);
+                if (!fd_ctx) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "unable to get fd ctx for fd=%p",
+                                local->fd);
+
+                        local->op_ret           = -1;
+                        int_lock->lock_op_ret   = -1;
+                        local->op_errno         = EINVAL;
+                        int_lock->lock_op_errno = EINVAL;
+
+			afr_unlock (frame, this);
+                        return -1;
+                }
+
+                call_count = int_lock->lockee_count * internal_lock_count (frame, this);
+                int_lock->lk_call_count = call_count;
+                int_lock->lk_expected_count = call_count;
+
+                if (!call_count) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "fd not open on any subvolumes. aborting.");
+                        afr_unlock (frame, this);
+                        goto out;
+                }
+
+                /* Send non-blocking entrylk calls only on up children
+                   and where the fd has been opened */
+                for (i = 0; i < int_lock->lockee_count*priv->child_count; i++) {
+                        index = i%copies;
+                        lockee_no = i/copies;
+                        if (local->child_up[index]) {
+                                AFR_TRACE_ENTRYLK_IN (frame, this, AFR_ENTRYLK_NB_TRANSACTION,
+                                                      AFR_LOCK_OP,
+                                                      int_lock->lockee[lockee_no].basename,
+                                                      i);
+
+                                STACK_WIND_COOKIE (frame, afr_nonblocking_entrylk_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[index],
+                                                   priv->children[index]->fops->fentrylk,
+                                                   this->name, local->fd,
+                                                   int_lock->lockee[lockee_no].basename,
+                                                   ENTRYLK_LOCK_NB, ENTRYLK_WRLCK,
+                                                   NULL);
+                                if (!--call_count)
+                                        break;
+                        }
+                }
+        } else {
+                call_count = int_lock->lockee_count * internal_lock_count (frame, this);
+                int_lock->lk_call_count = call_count;
+                int_lock->lk_expected_count = call_count;
+
+                for (i = 0; i < int_lock->lockee_count*priv->child_count; i++) {
+                        index = i%copies;
+                        lockee_no = i/copies;
+                        if (local->child_up[index]) {
+                                AFR_TRACE_ENTRYLK_IN (frame, this, AFR_ENTRYLK_NB_TRANSACTION,
+                                                      AFR_LOCK_OP,
+                                                      int_lock->lockee[lockee_no].basename,
+                                                      i);
+
+                                STACK_WIND_COOKIE (frame, afr_nonblocking_entrylk_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[index],
+                                                   priv->children[index]->fops->entrylk,
+                                                   this->name, &int_lock->lockee[lockee_no].loc,
+                                                   int_lock->lockee[lockee_no].basename,
+                                                   ENTRYLK_LOCK_NB, ENTRYLK_WRLCK,
+                                                   NULL);
+
+                                if (!--call_count)
+                                        break;
+                        }
+                }
+        }
+out:
+        return 0;
+}
+
+int32_t
+afr_nonblocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
+        afr_local_t         *local    = NULL;
+        int call_count  = 0;
+        int child_index = (long) cookie;
+        afr_fd_ctx_t        *fd_ctx = NULL;
+
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+
+        AFR_TRACE_INODELK_OUT (frame, this, AFR_INODELK_NB_TRANSACTION,
+                               AFR_LOCK_OP, NULL, op_ret,
+                               op_errno, (long) cookie);
+
+	if (local->fd)
+		fd_ctx = afr_fd_ctx_get (local->fd, this);
+
+        LOCK (&frame->lock);
+        {
+		if (op_ret < 0) {
+			if (op_errno == ENOSYS) {
+				/* return ENOTSUP */
+				gf_log (this->name, GF_LOG_ERROR,
+					"subvolume does not support locking. "
+					"please load features/locks xlator on "
+					"server");
+				local->op_ret                = op_ret;
+				int_lock->lock_op_ret        = op_ret;
+				int_lock->lock_op_errno      = op_errno;
+				local->op_errno              = op_errno;
+			}
+			if (local->transaction.eager_lock)
+				local->transaction.eager_lock[child_index] = 0;
+		} else {
+			inodelk->locked_nodes[child_index] |= LOCKED_YES;
+			inodelk->lock_count++;
+
+			if (local->transaction.eager_lock &&
+			    local->transaction.eager_lock[child_index] &&
+			    local->fd) {
+				/* piggybacked */
+				if (op_ret == 1) {
+					/* piggybacked */
+				} else if (op_ret == 0) {
+					/* lock acquired from server */
+                                        fd_ctx->lock_acquired[child_index]++;
+				}
+			}
+		}
+
+                call_count = --int_lock->lk_call_count;
+        }
+        UNLOCK (&frame->lock);
+
+        if (call_count == 0) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "Last inode locking reply received");
+                /* all locks successful. Proceed to call FOP */
+                if (inodelk->lock_count == int_lock->lk_expected_count) {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "All servers locked. Calling the cbk");
+                        int_lock->lock_op_ret = 0;
+                        int_lock->lock_cbk (frame, this);
+                }
+                /* Not all locks were successful. Unlock and try locking
+                   again, this time with serially blocking locks */
+                else {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "%d servers locked. Trying again with blocking calls",
+                                int_lock->lock_count);
+
+                        afr_unlock(frame, this);
+                }
+        }
+
+        return 0;
+}
+
+int
+afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
+        afr_local_t         *local    = NULL;
+        afr_private_t       *priv     = NULL;
+        afr_fd_ctx_t        *fd_ctx   = NULL;
+        int32_t             call_count = 0;
+        int                 i          = 0;
+        int                 ret        = 0;
+        struct              gf_flock flock = {0,};
+        struct              gf_flock full_flock = {0,};
+        struct              gf_flock *flock_use = NULL;
+        int                 piggyback = 0;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        priv     = this->private;
+
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+
+        flock.l_start = inodelk->flock.l_start;
+        flock.l_len   = inodelk->flock.l_len;
+        flock.l_type  = inodelk->flock.l_type;
+
+        full_flock.l_type = inodelk->flock.l_type;
+
+        initialize_inodelk_variables (frame, this);
+
+        if (local->fd) {
+                fd_ctx = afr_fd_ctx_get (local->fd, this);
+                if (!fd_ctx) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "unable to get fd ctx for fd=%p",
+                                local->fd);
+
+                        local->op_ret           = -1;
+                        int_lock->lock_op_ret   = -1;
+                        local->op_errno         = EINVAL;
+                        int_lock->lock_op_errno = EINVAL;
+
+			afr_unlock (frame, this);
+                        ret = -1;
+                        goto out;
+                }
+
+                call_count = internal_lock_count (frame, this);
+                int_lock->lk_call_count = call_count;
+                int_lock->lk_expected_count = call_count;
+
+                if (!call_count) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "fd not open on any subvolumes. aborting.");
+                        afr_unlock (frame, this);
+                        goto out;
+                }
+
+                /* Send non-blocking inodelk calls only on up children
+                   and where the fd has been opened */
+                for (i = 0; i < priv->child_count; i++) {
+                        if (!local->child_up[i])
+                                continue;
+
+                        flock_use = &flock;
+                        if (!local->transaction.eager_lock_on) {
+                                goto wind;
+                        }
+
+                        piggyback = 0;
+                        local->transaction.eager_lock[i] = 1;
+
+			afr_set_delayed_post_op (frame, this);
+
+                        LOCK (&local->fd->lock);
+                        {
+                                if (fd_ctx->lock_acquired[i]) {
+                                        fd_ctx->lock_piggyback[i]++;
+                                        piggyback = 1;
+                                }
+                        }
+                        UNLOCK (&local->fd->lock);
+
+                        if (piggyback) {
+                                /* (op_ret == 1) => indicate piggybacked lock */
+                                afr_nonblocking_inodelk_cbk (frame, (void *) (long) i,
+                                                             this, 1, 0, NULL);
+                                if (!--call_count)
+                                        break;
+                                continue;
+                        }
+                        flock_use = &full_flock;
+                wind:
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_NB_TRANSACTION,
+                                              AFR_LOCK_OP, flock_use, F_SETLK, i);
+
+                        STACK_WIND_COOKIE (frame, afr_nonblocking_inodelk_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->finodelk,
+                                           int_lock->domain, local->fd,
+                                           F_SETLK, flock_use, NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        } else {
+                call_count = internal_lock_count (frame, this);
+                int_lock->lk_call_count = call_count;
+                int_lock->lk_expected_count = call_count;
+
+                for (i = 0; i < priv->child_count; i++) {
+                        if (!local->child_up[i])
+                                continue;
+                        AFR_TRACE_INODELK_IN (frame, this,
+                                              AFR_INODELK_NB_TRANSACTION,
+                                              AFR_LOCK_OP, &flock, F_SETLK, i);
+
+                        STACK_WIND_COOKIE (frame, afr_nonblocking_inodelk_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->inodelk,
+                                           int_lock->domain, &local->loc,
+                                           F_SETLK, &flock, NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+out:
+        return ret;
+}
+
+int32_t
+afr_unlock (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+
+        local = frame->local;
+
+        if (transaction_lk_op (local)) {
+                if (is_afr_lock_transaction (local))
+                        afr_unlock_inodelk (frame, this);
+                else
+                        afr_unlock_entrylk (frame, this);
+
+        } else {
+                if (is_afr_lock_selfheal (local))
+                        afr_unlock_inodelk (frame, this);
+                else
+                        afr_unlock_entrylk (frame, this);
+        }
+
+        return 0;
+}
+
+int
+afr_mark_locked_nodes (xlator_t *this, fd_t *fd,
+                       unsigned char *locked_nodes)
+{
+        afr_private_t *priv  = NULL;
+        afr_fd_ctx_t  *fdctx = NULL;
+        uint64_t       tmp   = 0;
+        int            ret   = 0;
+
+        priv = this->private;
+
+        ret = afr_fd_ctx_set (this, fd);
+        if (ret)
+                goto out;
+
+        ret = fd_ctx_get (fd, this, &tmp);
+        if (ret) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "failed to get the fd ctx");
+                goto out;
+        }
+        fdctx = (afr_fd_ctx_t *) (long) tmp;
+
+        GF_ASSERT (fdctx->locked_on);
+
+        memcpy (fdctx->locked_on, locked_nodes,
+                priv->child_count);
+
+out:
+        return ret;
+}
+
+static int
+__is_fd_saved (xlator_t *this, fd_t *fd)
+{
+        afr_locked_fd_t *locked_fd = NULL;
+        afr_private_t   *priv      = NULL;
+        int              found     = 0;
+
+        priv = this->private;
+
+        list_for_each_entry (locked_fd, &priv->saved_fds, list) {
+                if (locked_fd->fd == fd) {
+                        found = 1;
+                        break;
+                }
+        }
+
+        return found;
+}
+
+static int
+__afr_save_locked_fd (xlator_t *this, fd_t *fd)
+{
+        afr_private_t   *priv      = NULL;
+        afr_locked_fd_t *locked_fd = NULL;
+        int              ret       = 0;
+
+        priv = this->private;
+
+        locked_fd = GF_CALLOC (1, sizeof (*locked_fd),
+                               gf_afr_mt_locked_fd);
+        if (!locked_fd) {
+                ret = -1;
+                goto out;
+        }
+
+        locked_fd->fd = fd;
+        INIT_LIST_HEAD (&locked_fd->list);
+
+        list_add_tail (&locked_fd->list, &priv->saved_fds);
+
+out:
+        return ret;
+}
+
+int
+afr_save_locked_fd (xlator_t *this, fd_t *fd)
+{
+        afr_private_t   *priv      = NULL;
+        int              ret       = 0;
+
+        priv = this->private;
+
+        pthread_mutex_lock (&priv->mutex);
+        {
+                if (__is_fd_saved (this, fd)) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "fd=%p already saved", fd);
+                        goto unlock;
+                }
+
+                ret = __afr_save_locked_fd (this, fd);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "fd=%p could not be saved", fd);
+                        goto unlock;
+                }
+        }
+unlock:
+        pthread_mutex_unlock (&priv->mutex);
+
+        return ret;
+}
+
+static int
+afr_lock_recovery_cleanup (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local     = NULL;
+        afr_locked_fd_t *locked_fd = NULL;
+
+        local = frame->local;
+
+        locked_fd = local->locked_fd;
+
+        STACK_DESTROY (frame->root);
+        afr_local_cleanup (local, this);
+
+        afr_save_locked_fd (this, locked_fd->fd);
+
+        return 0;
+
+}
+
+static int
+afr_get_source_lock_recovery (xlator_t *this, fd_t *fd)
+{
+        afr_fd_ctx_t  *fdctx        = NULL;
+        afr_private_t *priv         = NULL;
+        uint64_t      tmp           = 0;
+        int           i             = 0;
+        int           source_child  = -1;
+        int           ret           = 0;
+
+        priv = this->private;
+
+        ret = fd_ctx_get (fd, this, &tmp);
+        if (ret)
+                goto out;
+
+        fdctx = (afr_fd_ctx_t *) (long) tmp;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (fdctx->locked_on[i]) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "Found lock recovery source=%d", i);
+                        source_child = i;
+                        break;
+                }
+        }
+
+out:
+        return source_child;
+
+}
+
+int32_t
+afr_get_locks_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                      dict_t *xdata);
+int32_t
+afr_recover_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                      dict_t *xdata)
+{
+        afr_local_t   *local = NULL;
+        afr_private_t *priv  = NULL;
+        int32_t source_child = 0;
+        struct gf_flock flock   = {0,};
+
+        local = frame->local;
+        priv  = this->private;
+
+        if (op_ret) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "lock recovery failed");
+                goto cleanup;
+        }
+
+        source_child = local->source_child;
+
+        memcpy (&flock, lock, sizeof (*lock));
+
+        STACK_WIND_COOKIE (frame, afr_get_locks_fd_cbk,
+                           (void *) (long) source_child,
+                           priv->children[source_child],
+                           priv->children[source_child]->fops->lk,
+                           local->fd, F_GETLK_FD, &flock, NULL);
+
+        return 0;
+
+cleanup:
+        afr_lock_recovery_cleanup (frame, this);
+        return 0;
+}
+
+int
+afr_recover_lock (call_frame_t *frame, xlator_t *this,
+                  struct gf_flock *flock)
+{
+        afr_local_t   *local             = NULL;
+        afr_private_t *priv              = NULL;
+        int32_t      lock_recovery_child = 0;
+
+        priv  = this->private;
+        local = frame->local;
+
+        lock_recovery_child = local->lock_recovery_child;
+
+        frame->root->lk_owner = flock->l_owner;
+
+        STACK_WIND_COOKIE (frame, afr_recover_lock_cbk,
+                           (void *) (long) lock_recovery_child,
+                           priv->children[lock_recovery_child],
+                           priv->children[lock_recovery_child]->fops->lk,
+                           local->fd, F_SETLK, flock, NULL);
+
+        return 0;
+}
+
+static int
+is_afr_lock_eol (struct gf_flock *lock)
+{
+        int ret = 0;
+
+        if ((lock->l_type == GF_LK_EOL))
+                ret = 1;
+
+        return ret;
+}
+
+int32_t
+afr_get_locks_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                      dict_t *xdata)
+{
+        if (op_ret) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Failed to get locks on fd");
+                goto cleanup;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Got a lock on fd");
+
+        if (is_afr_lock_eol (lock)) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Reached EOL on locks on fd");
+                goto cleanup;
+        }
+
+        afr_recover_lock (frame, this, lock);
+
+        return 0;
+
+cleanup:
+        afr_lock_recovery_cleanup (frame, this);
+
+        return 0;
+}
+
+static int
+afr_lock_recovery (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local        = NULL;
+        afr_private_t *priv         = NULL;
+        fd_t          *fd           = NULL;
+        int            ret          = 0;
+        int32_t        source_child = 0;
+        struct gf_flock   flock        = {0,};
+
+        priv  = this->private;
+        local = frame->local;
+
+        fd = local->fd;
+
+        source_child = afr_get_source_lock_recovery (this, fd);
+        if (source_child < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Could not recover locks due to lock "
+                        "split brain");
+                ret = -1;
+                goto out;
+        }
+
+        local->source_child = source_child;
+
+        /* the flock can be zero filled as we're querying incrementally
+           the locks held on the fd.
+        */
+        STACK_WIND_COOKIE (frame, afr_get_locks_fd_cbk,
+                           (void *) (long) source_child,
+                           priv->children[source_child],
+                           priv->children[source_child]->fops->lk,
+                           local->fd, F_GETLK_FD, &flock, NULL);
+
+out:
+        return ret;
+}
+
+
+static int
+afr_mark_fd_opened (xlator_t *this, fd_t *fd, int32_t child_index)
+{
+        afr_fd_ctx_t *fdctx = NULL;
+        uint64_t      tmp   = 0;
+        int           ret   = 0;
+
+        ret = fd_ctx_get (fd, this, &tmp);
+        if (ret)
+                goto out;
+
+        fdctx = (afr_fd_ctx_t *) (long) tmp;
+
+        fdctx->opened_on[child_index] = AFR_FD_OPENED;
+
+out:
+        return ret;
+}
+
+int32_t
+afr_lock_recovery_preopen_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                               int32_t op_ret, int32_t op_errno, fd_t *fd,
+                               dict_t *xdata)
+{
+        int32_t child_index = (long )cookie;
+        int ret = 0;
+
+        if (op_ret) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Reopen during lock-recovery failed");
+                goto cleanup;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Open succeeded => proceed to recover locks");
+
+        ret = afr_lock_recovery (frame, this);
+        if (ret) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Lock recovery failed");
+                goto cleanup;
+        }
+
+        ret = afr_mark_fd_opened (this, fd, child_index);
+        if (ret) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Marking fd open failed");
+                goto cleanup;
+        }
+
+        return 0;
+
+cleanup:
+        afr_lock_recovery_cleanup (frame, this);
+        return 0;
+}
+
+static int
+afr_lock_recovery_preopen (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t *priv  = NULL;
+        afr_local_t   *local = NULL;
+        uint64_t       tmp   = 0;
+        afr_fd_ctx_t  *fdctx = NULL;
+        loc_t          loc   = {0,};
+        int32_t        child_index = 0;
+        int            ret   = 0;
+
+        priv  = this->private;
+        local = frame->local;
+
+        GF_ASSERT (local && local->fd);
+
+        ret = fd_ctx_get (local->fd, this, &tmp);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to get the context of fd",
+                        uuid_utoa (local->fd->inode->gfid));
+        fdctx = (afr_fd_ctx_t *) (long) tmp;
+        /* TODO: instead we should return from the function */
+        GF_ASSERT (fdctx);
+
+        child_index = local->lock_recovery_child;
+
+        inode_path (local->fd->inode, NULL, (char **)&loc.path);
+        loc.name   = strrchr (loc.path, '/');
+        loc.inode  = inode_ref (local->fd->inode);
+        loc.parent = inode_parent (local->fd->inode, 0, NULL);
+
+
+        STACK_WIND_COOKIE (frame, afr_lock_recovery_preopen_cbk,
+                           (void *)(long) child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->open,
+                           &loc, fdctx->flags, local->fd, NULL);
+
+        return 0;
+}
+
+static int
+is_fd_opened (fd_t *fd, int32_t child_index)
+{
+        afr_fd_ctx_t *fdctx = NULL;
+        uint64_t      tmp = 0;
+        int           ret = 0;
+
+        ret = fd_ctx_get (fd, THIS, &tmp);
+        if (ret)
+                goto out;
+
+        fdctx = (afr_fd_ctx_t *) (long) tmp;
+
+        if (fdctx->opened_on[child_index] == AFR_FD_OPENED)
+                ret = 1;
+
+out:
+        return ret;
+}
+
+int
+afr_attempt_lock_recovery (xlator_t *this, int32_t child_index)
+{
+        call_frame_t    *frame      = NULL;
+        afr_private_t   *priv       = NULL;
+        afr_local_t     *local      = NULL;
+        afr_locked_fd_t *locked_fd  = NULL;
+        afr_locked_fd_t  *tmp       = NULL;
+        int              ret        = -1;
+        struct list_head locks_list = {0,};
+        int32_t          op_errno   = 0;
+
+
+        priv = this->private;
+
+        if (list_empty (&priv->saved_fds))
+                goto out;
+
+        frame = create_frame (this, this->ctx->pool);
+        if (!frame) {
+                ret = -1;
+                goto out;
+        }
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0) {
+                ret = -1;
+                goto out;
+        }
+
+        frame->local = local;
+
+        INIT_LIST_HEAD (&locks_list);
+
+        pthread_mutex_lock (&priv->mutex);
+        {
+                list_splice_init (&priv->saved_fds, &locks_list);
+        }
+        pthread_mutex_unlock (&priv->mutex);
+
+        list_for_each_entry_safe (locked_fd, tmp,
+                                  &locks_list, list) {
+
+                list_del_init (&locked_fd->list);
+
+                local->fd                  = fd_ref (locked_fd->fd);
+                local->lock_recovery_child = child_index;
+                local->locked_fd           = locked_fd;
+
+                if (!is_fd_opened (locked_fd->fd, child_index)) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "attempting open before lock "
+                                "recovery");
+                        afr_lock_recovery_preopen (frame, this);
+                } else {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "attempting lock recovery "
+                                "without a preopen");
+                        afr_lock_recovery (frame, this);
+                }
+        }
+
+out:
+        if ((ret < 0) && frame)
+                AFR_STACK_DESTROY (frame);
+        return ret;
+}
+
+int
+afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src, char *dom,
+                          unsigned int child_count)
+{
+        afr_local_t         *dst_local   = NULL;
+        afr_local_t         *src_local   = NULL;
+        afr_internal_lock_t *dst_lock    = NULL;
+        afr_internal_lock_t *src_lock    = NULL;
+        afr_inodelk_t       *dst_inodelk = NULL;
+        afr_inodelk_t       *src_inodelk = NULL;
+        int                 ret = -1;
+
+        src_local = src->local;
+        src_lock  = &src_local->internal_lock;
+        src_inodelk = afr_get_inodelk (src_lock, dom);
+        dst_local = dst->local;
+        dst_lock  = &dst_local->internal_lock;
+        dst_inodelk = afr_get_inodelk (dst_lock, dom);
+        if (!dst_inodelk || !src_inodelk)
+                goto out;
+        if (src_inodelk->locked_nodes) {
+                memcpy (dst_inodelk->locked_nodes, src_inodelk->locked_nodes,
+                        sizeof (*dst_inodelk->locked_nodes) * child_count);
+                memset (src_inodelk->locked_nodes, 0,
+                        sizeof (*src_inodelk->locked_nodes) * child_count);
+        }
+
+        dst_lock->transaction_lk_type = src_lock->transaction_lk_type;
+        dst_lock->selfheal_lk_type    = src_lock->selfheal_lk_type;
+        dst_inodelk->lock_count = src_inodelk->lock_count;
+        src_inodelk->lock_count = 0;
+        ret = 0;
+out:
+        return ret;
+}
diff --git a/xlators/cluster/afr/src/afr-mem-types.h b/xlators/cluster/afr/src/afr-mem-types.h
new file mode 100644
index 000000000..73594f265
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-mem-types.h
@@ -0,0 +1,51 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+
+#ifndef __AFR_MEM_TYPES_H__
+#define __AFR_MEM_TYPES_H__
+
+#include "mem-types.h"
+
+enum gf_afr_mem_types_ {
+        gf_afr_mt_iovec  = gf_common_mt_end + 1,
+        gf_afr_mt_afr_fd_ctx_t,
+        gf_afr_mt_afr_private_t,
+        gf_afr_mt_int32_t,
+        gf_afr_mt_char,
+        gf_afr_mt_xattr_key,
+        gf_afr_mt_dict_t,
+        gf_afr_mt_xlator_t,
+        gf_afr_mt_iatt,
+        gf_afr_mt_int,
+        gf_afr_mt_afr_node_character,
+        gf_afr_mt_sh_diff_loop_state,
+        gf_afr_mt_uint8_t,
+        gf_afr_mt_loc_t,
+        gf_afr_mt_entry_name,
+        gf_afr_mt_pump_priv,
+        gf_afr_mt_locked_fd,
+        gf_afr_mt_inode_ctx_t,
+        gf_afr_fd_paused_call_t,
+        gf_afr_mt_crawl_data_t,
+        gf_afr_mt_brick_pos_t,
+        gf_afr_mt_shd_bool_t,
+        gf_afr_mt_shd_timer_t,
+        gf_afr_mt_shd_event_t,
+        gf_afr_mt_time_t,
+        gf_afr_mt_pos_data_t,
+        gf_afr_mt_reply_t,
+        gf_afr_mt_stats_t,
+        gf_afr_mt_shd_crawl_event_t,
+        gf_afr_mt_uint64_t,
+        gf_afr_mt_end
+};
+#endif
+
diff --git a/xlators/cluster/afr/src/afr-open.c b/xlators/cluster/afr/src/afr-open.c
new file mode 100644
index 000000000..643a5d692
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-open.c
@@ -0,0 +1,382 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <libgen.h>
+#include <unistd.h>
+#include <fnmatch.h>
+#include <sys/time.h>
+#include <stdlib.h>
+#include <signal.h>
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "glusterfs.h"
+#include "afr.h"
+#include "dict.h"
+#include "xlator.h"
+#include "hashfn.h"
+#include "logging.h"
+#include "stack.h"
+#include "list.h"
+#include "call-stub.h"
+#include "defaults.h"
+#include "common-utils.h"
+#include "compat-errno.h"
+#include "compat.h"
+#include "byte-order.h"
+#include "statedump.h"
+
+#include "fd.h"
+
+#include "afr-inode-read.h"
+#include "afr-inode-write.h"
+#include "afr-dir-read.h"
+#include "afr-dir-write.h"
+#include "afr-transaction.h"
+#include "afr-self-heal.h"
+#include "afr-self-heal-common.h"
+
+int
+afr_stale_child_up (afr_local_t *local, xlator_t *this)
+{
+        int             i = 0;
+        afr_private_t   *priv = NULL;
+        int             up = -1;
+
+        priv = this->private;
+
+        if (!local->fresh_children)
+                local->fresh_children = afr_children_create (priv->child_count);
+        if (!local->fresh_children)
+                goto out;
+
+        afr_inode_get_read_ctx (this, local->fd->inode, local->fresh_children);
+        if (priv->child_count == afr_get_children_count (local->fresh_children,
+                                                         priv->child_count))
+                goto out;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (!local->child_up[i])
+                        continue;
+                if (afr_is_child_present (local->fresh_children,
+                                          priv->child_count, i))
+                        continue;
+                up = i;
+                break;
+        }
+out:
+        return up;
+}
+
+void
+afr_perform_data_self_heal (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        inode_t         *inode = NULL;
+        int             st_child = -1;
+        char            reason[64] = {0};
+
+        local = frame->local;
+        sh = &local->self_heal;
+        inode = local->fd->inode;
+
+        if (!IA_ISREG (inode->ia_type))
+                goto out;
+
+        st_child = afr_stale_child_up (local, this);
+        if (st_child < 0)
+                goto out;
+
+        sh->do_data_self_heal          = _gf_true;
+        sh->do_metadata_self_heal      = _gf_true;
+        sh->do_gfid_self_heal          = _gf_true;
+        sh->do_missing_entry_self_heal = _gf_true;
+
+        snprintf (reason, sizeof (reason), "stale subvolume %d detected",
+                  st_child);
+        afr_launch_self_heal (frame, this, inode, _gf_true, inode->ia_type,
+                              reason, NULL, NULL);
+out:
+        return;
+}
+
+int
+afr_open_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                        struct iatt *postbuf, dict_t *xdata)
+{
+        afr_local_t * local = frame->local;
+        afr_private_t *priv = NULL;
+
+        priv = this->private;
+        if (afr_open_only_data_self_heal (priv->data_self_heal))
+                afr_perform_data_self_heal (frame, this);
+        AFR_STACK_UNWIND (open, frame, local->op_ret, local->op_errno,
+                          local->fd, xdata);
+        return 0;
+}
+
+
+int
+afr_open_cbk (call_frame_t *frame, void *cookie,
+              xlator_t *this, int32_t op_ret, int32_t op_errno,
+              fd_t *fd, dict_t *xdata)
+{
+        afr_local_t *  local       = NULL;
+        int            ret         = 0;
+        int            call_count  = -1;
+        int            child_index = (long) cookie;
+        afr_private_t *priv        = NULL;
+
+        priv = this->private;
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                }
+
+                if (op_ret >= 0) {
+                        local->op_ret = op_ret;
+                        local->success_count++;
+
+                        ret = afr_child_fd_ctx_set (this, fd, child_index,
+                                                    local->cont.open.flags);
+                        if (ret) {
+                                local->op_ret = -1;
+                                local->op_errno = -ret;
+                                goto unlock;
+                        }
+                }
+        }
+unlock:
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                if ((local->cont.open.flags & O_TRUNC)
+                    && (local->op_ret >= 0)) {
+                        STACK_WIND (frame, afr_open_ftruncate_cbk,
+                                    this, this->fops->ftruncate,
+                                    fd, 0, NULL);
+                } else {
+                        if (afr_open_only_data_self_heal (priv->data_self_heal))
+                                afr_perform_data_self_heal (frame, this);
+                        AFR_STACK_UNWIND (open, frame, local->op_ret,
+                                          local->op_errno, local->fd, xdata);
+                }
+        }
+
+        return 0;
+}
+
+int
+afr_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          fd_t *fd, dict_t *xdata)
+{
+        afr_private_t * priv       = NULL;
+        afr_local_t *   local      = NULL;
+        int             i          = 0;
+        int             ret        = -1;
+        int32_t         call_count = 0;
+        int32_t         op_errno   = 0;
+        int32_t         wind_flags = flags & (~O_TRUNC);
+        //We can't let truncation to happen outside transaction.
+
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (this, out);
+        VALIDATE_OR_GOTO (this->private, out);
+        VALIDATE_OR_GOTO (loc, out);
+
+        priv = this->private;
+
+        if (flags & (O_CREAT|O_TRUNC)) {
+                QUORUM_CHECK(open,out);
+        }
+
+        if (afr_is_split_brain (this, loc->inode)) {
+                /* self-heal failed */
+                gf_log (this->name, GF_LOG_WARNING,
+                        "failed to open as split brain seen, returning EIO");
+                op_errno = EIO;
+                goto out;
+        }
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        call_count   = local->call_count;
+        loc_copy (&local->loc, loc);
+
+        local->cont.open.flags   = flags;
+
+        local->fd = fd_ref (fd);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND_COOKIE (frame, afr_open_cbk, (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->open,
+                                           loc, wind_flags, fd, xdata);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
+
+        ret = 0;
+out:
+        if (ret < 0)
+                AFR_STACK_UNWIND (open, frame, -1, op_errno, fd, xdata);
+
+        return 0;
+}
+
+int
+afr_openfd_fix_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, fd_t *fd,
+                         dict_t *xdata)
+{
+        afr_local_t   *local      = NULL;
+        afr_private_t *priv       = NULL;
+        afr_fd_ctx_t  *fd_ctx     = NULL;
+        int           call_count  = 0;
+        int           child_index = (long) cookie;
+
+        priv     = this->private;
+        local    = frame->local;
+
+        if (op_ret >= 0) {
+                gf_log (this->name, GF_LOG_DEBUG, "fd for %s opened "
+                        "successfully on subvolume %s", local->loc.path,
+                        priv->children[child_index]->name);
+        } else {
+                gf_log (this->name, GF_LOG_ERROR, "Failed to open %s "
+                        "on subvolume %s", local->loc.path,
+                        priv->children[child_index]->name);
+        }
+
+        fd_ctx = afr_fd_ctx_get (local->fd, this);
+        if (!fd_ctx) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "failed to get fd context, %p", local->fd);
+                goto out;
+        }
+
+        LOCK (&local->fd->lock);
+        {
+                if (op_ret >= 0) {
+                        fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+                } else {
+                        fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
+                }
+        }
+        UNLOCK (&local->fd->lock);
+out:
+        call_count = afr_frame_return (frame);
+        if (call_count == 0)
+                AFR_STACK_DESTROY (frame);
+
+        return 0;
+}
+
+void
+afr_fix_open (xlator_t *this, fd_t *fd, size_t need_open_count, int *need_open)
+{
+        afr_private_t *priv    = NULL;
+        int           i        = 0;
+        call_frame_t  *frame   = NULL;
+        afr_local_t   *local   = NULL;
+        int           ret      = -1;
+        int32_t       op_errno = 0;
+        afr_fd_ctx_t  *fd_ctx  = NULL;
+
+        priv  = this->private;
+
+        if (!afr_is_fd_fixable (fd) || !need_open || !need_open_count)
+                goto out;
+
+        fd_ctx = afr_fd_ctx_get (fd, this);
+        if (!fd_ctx) {
+                ret = -1;
+                goto out;
+        }
+
+        frame = create_frame (this, this->ctx->pool);
+        if (!frame) {
+                ret = -1;
+                goto out;
+        }
+
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+        local = frame->local;
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        local->loc.inode = inode_ref (fd->inode);
+        ret = loc_path (&local->loc, NULL);
+        if (ret < 0)
+                goto out;
+
+        local->fd = fd_ref (fd);
+        local->call_count = need_open_count;
+
+        gf_log (this->name, GF_LOG_DEBUG, "need open count: %zd",
+                need_open_count);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (!need_open[i])
+                        continue;
+
+                if (IA_IFDIR == fd->inode->ia_type) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "opening fd for dir %s on subvolume %s",
+                                local->loc.path, priv->children[i]->name);
+
+                        STACK_WIND_COOKIE (frame, afr_openfd_fix_open_cbk,
+                                           (void*) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->opendir,
+                                           &local->loc, local->fd,
+                                           NULL);
+                } else {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "opening fd for file %s on subvolume %s",
+                                local->loc.path, priv->children[i]->name);
+
+                        STACK_WIND_COOKIE (frame, afr_openfd_fix_open_cbk,
+                                           (void *)(long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->open,
+                                           &local->loc,
+                                           fd_ctx->flags & (~O_TRUNC),
+                                           local->fd, NULL);
+                }
+
+        }
+        op_errno = 0;
+        ret = 0;
+out:
+        if (op_errno)
+                ret = -1; //For handling ALLOC_OR_GOTO
+        if (ret && frame)
+                AFR_STACK_DESTROY (frame);
+}
diff --git a/xlators/cluster/afr/src/afr-self-heal-algorithm.c b/xlators/cluster/afr/src/afr-self-heal-algorithm.c
new file mode 100644
index 000000000..83846f152
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heal-algorithm.c
@@ -0,0 +1,837 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+
+#include <openssl/md5.h>
+#include "glusterfs.h"
+#include "afr.h"
+#include "xlator.h"
+#include "dict.h"
+#include "xlator.h"
+#include "hashfn.h"
+#include "logging.h"
+#include "stack.h"
+#include "list.h"
+#include "call-stub.h"
+#include "defaults.h"
+#include "common-utils.h"
+#include "compat-errno.h"
+#include "compat.h"
+#include "byte-order.h"
+
+#include "afr-transaction.h"
+#include "afr-self-heal.h"
+#include "afr-self-heal-common.h"
+#include "afr-self-heal-algorithm.h"
+
+/*
+  This file contains the various self-heal algorithms
+*/
+
+static int
+sh_loop_driver (call_frame_t *sh_frame, xlator_t *this,
+                gf_boolean_t is_first_call, call_frame_t *old_loop_frame);
+static int
+sh_loop_return (call_frame_t *sh_frame, xlator_t *this, call_frame_t *loop_frame,
+                int32_t op_ret, int32_t op_errno);
+static int
+sh_destroy_frame (call_frame_t *frame, xlator_t *this)
+{
+        if (!frame)
+                goto out;
+
+        AFR_STACK_DESTROY (frame);
+out:
+        return 0;
+}
+
+static void
+sh_private_cleanup (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t             *local   = NULL;
+        afr_self_heal_t         *sh      = NULL;
+        afr_sh_algo_private_t   *sh_priv = NULL;
+
+        local = frame->local;
+        sh    = &local->self_heal;
+
+        sh_priv = sh->private;
+        GF_FREE (sh_priv);
+}
+
+static int
+sh_number_of_writes_needed (unsigned char *write_needed, int child_count)
+{
+        int writes = 0;
+        int i      = 0;
+
+        for (i = 0; i < child_count; i++) {
+                if (write_needed[i])
+                        writes++;
+        }
+
+        return writes;
+}
+
+
+static int
+sh_loop_driver_done (call_frame_t *sh_frame, xlator_t *this,
+                     call_frame_t *last_loop_frame)
+{
+        afr_local_t             *local        = NULL;
+        afr_self_heal_t         *sh           = NULL;
+        afr_sh_algo_private_t   *sh_priv      = NULL;
+        int32_t                 total_blocks = 0;
+        int32_t                 diff_blocks  = 0;
+
+        local        = sh_frame->local;
+        sh           = &local->self_heal;
+        sh_priv      = sh->private;
+        if (sh_priv) {
+                total_blocks = sh_priv->total_blocks;
+                diff_blocks  = sh_priv->diff_blocks;
+        }
+
+        sh_private_cleanup (sh_frame, this);
+        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
+                GF_ASSERT (!last_loop_frame);
+                //loop_finish should have happened and the old_loop should be NULL
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "self-heal aborting on %s",
+                        local->loc.path);
+
+                local->self_heal.algo_abort_cbk (sh_frame, this);
+        } else {
+                GF_ASSERT (last_loop_frame);
+                if (diff_blocks == total_blocks) {
+                        gf_log (this->name, GF_LOG_DEBUG, "full self-heal "
+                                "completed on %s",local->loc.path);
+                } else {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "diff self-heal on %s: completed. "
+                                "(%d blocks of %d were different (%.2f%%))",
+                                local->loc.path, diff_blocks, total_blocks,
+                                ((diff_blocks * 1.0)/total_blocks) * 100);
+                }
+
+                sh->old_loop_frame = last_loop_frame;
+                local->self_heal.algo_completion_cbk (sh_frame, this);
+        }
+
+        return 0;
+}
+
+int
+sh_loop_finish (call_frame_t *loop_frame, xlator_t *this)
+{
+        afr_local_t             *loop_local = NULL;
+        afr_self_heal_t         *loop_sh = NULL;
+
+        if (!loop_frame)
+                goto out;
+
+        loop_local = loop_frame->local;
+        if (loop_local) {
+                loop_sh = &loop_local->self_heal;
+        }
+
+        if (loop_sh && loop_sh->data_lock_held) {
+                afr_sh_data_unlock (loop_frame, this, this->name,
+                                    sh_destroy_frame);
+        } else {
+                sh_destroy_frame (loop_frame, this);
+        }
+out:
+        return 0;
+}
+
+static int
+sh_loop_lock_success (call_frame_t *loop_frame, xlator_t *this)
+{
+        afr_local_t                 *loop_local = NULL;
+        afr_self_heal_t             *loop_sh    = NULL;
+
+        loop_local = loop_frame->local;
+        loop_sh = &loop_local->self_heal;
+
+        sh_loop_finish (loop_sh->old_loop_frame, this);
+        loop_sh->old_loop_frame = NULL;
+
+        gf_log (this->name, GF_LOG_DEBUG, "Acquired lock for range %"PRIu64
+                " %"PRIu64, loop_sh->offset, loop_sh->block_size);
+        loop_sh->data_lock_held = _gf_true;
+        loop_sh->sh_data_algo_start (loop_frame, this);
+        return 0;
+}
+
+static int
+sh_loop_lock_failure (call_frame_t *loop_frame, xlator_t *this)
+{
+        call_frame_t                *sh_frame = NULL;
+        afr_local_t                 *loop_local = NULL;
+        afr_self_heal_t             *loop_sh    = NULL;
+
+        loop_local = loop_frame->local;
+        loop_sh = &loop_local->self_heal;
+        sh_frame = loop_sh->sh_frame;
+
+        gf_log (this->name, GF_LOG_ERROR, "failed lock for range %"PRIu64
+                " %"PRIu64, loop_sh->offset, loop_sh->block_size);
+        sh_loop_finish (loop_sh->old_loop_frame, this);
+        loop_sh->old_loop_frame = NULL;
+        sh_loop_return (sh_frame, this, loop_frame, -1, ENOTCONN);
+        return 0;
+}
+
+static int
+sh_loop_frame_create (call_frame_t *sh_frame, xlator_t *this,
+                      call_frame_t *old_loop_frame, call_frame_t **loop_frame)
+{
+        call_frame_t                *new_loop_frame = NULL;
+        afr_local_t                 *local          = NULL;
+        afr_self_heal_t             *sh             = NULL;
+        afr_local_t                 *new_loop_local = NULL;
+        afr_self_heal_t             *new_loop_sh    = NULL;
+        afr_private_t               *priv           = NULL;
+
+        GF_ASSERT (sh_frame);
+        GF_ASSERT (loop_frame);
+
+        *loop_frame = NULL;
+        local   = sh_frame->local;
+        sh      = &local->self_heal;
+        priv    = this->private;
+
+        new_loop_frame = copy_frame (sh_frame);
+        if (!new_loop_frame)
+                goto out;
+        //We want the frame to have same lk_owner as sh_frame
+        //so that locks translator allows conflicting locks
+        new_loop_local = afr_self_heal_local_init (local, this);
+        if (!new_loop_local)
+                goto out;
+        new_loop_frame->local = new_loop_local;
+
+        new_loop_sh = &new_loop_local->self_heal;
+        new_loop_sh->sources = memdup (sh->sources,
+                                       priv->child_count * sizeof (*sh->sources));
+        if (!new_loop_sh->sources)
+                goto out;
+        new_loop_sh->write_needed = GF_CALLOC (priv->child_count,
+                                               sizeof (*new_loop_sh->write_needed),
+                                               gf_afr_mt_char);
+        if (!new_loop_sh->write_needed)
+                goto out;
+        new_loop_sh->checksum = GF_CALLOC (priv->child_count, MD5_DIGEST_LENGTH,
+                                           gf_afr_mt_uint8_t);
+        if (!new_loop_sh->checksum)
+                goto out;
+        new_loop_sh->inode      = inode_ref (sh->inode);
+        new_loop_sh->sh_data_algo_start = sh->sh_data_algo_start;
+        new_loop_sh->source = sh->source;
+        new_loop_sh->active_sinks = sh->active_sinks;
+        new_loop_sh->healing_fd = fd_ref (sh->healing_fd);
+        new_loop_sh->file_has_holes = sh->file_has_holes;
+        new_loop_sh->old_loop_frame = old_loop_frame;
+        new_loop_sh->sh_frame = sh_frame;
+        *loop_frame = new_loop_frame;
+        return 0;
+out:
+        sh_destroy_frame (new_loop_frame, this);
+        return -ENOMEM;
+}
+
+static int
+sh_loop_start (call_frame_t *sh_frame, xlator_t *this, off_t offset,
+               call_frame_t *old_loop_frame)
+{
+        call_frame_t                *new_loop_frame = NULL;
+        afr_local_t                 *local          = NULL;
+        afr_self_heal_t             *sh             = NULL;
+        afr_local_t                 *new_loop_local = NULL;
+        afr_self_heal_t             *new_loop_sh    = NULL;
+        int                         ret             = 0;
+
+        GF_ASSERT (sh_frame);
+
+        local   = sh_frame->local;
+        sh      = &local->self_heal;
+
+        ret = sh_loop_frame_create (sh_frame, this, old_loop_frame,
+                                    &new_loop_frame);
+        if (ret)
+                goto out;
+        new_loop_local = new_loop_frame->local;
+        new_loop_sh = &new_loop_local->self_heal;
+        new_loop_sh->offset = offset;
+        new_loop_sh->block_size = sh->block_size;
+        afr_sh_data_lock (new_loop_frame, this, offset, new_loop_sh->block_size,
+                          _gf_true, this->name, sh_loop_lock_success, sh_loop_lock_failure);
+        return 0;
+out:
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        if (old_loop_frame)
+                sh_loop_finish (old_loop_frame, this);
+        sh_loop_return (sh_frame, this, new_loop_frame, -1, ENOMEM);
+        return 0;
+}
+
+static int
+sh_loop_driver (call_frame_t *sh_frame, xlator_t *this,
+                gf_boolean_t is_first_call, call_frame_t *old_loop_frame)
+{
+        afr_local_t *               local          = NULL;
+        afr_self_heal_t *           sh             = NULL;
+        afr_sh_algo_private_t       *sh_priv        = NULL;
+        gf_boolean_t                is_driver_done = _gf_false;
+        blksize_t                   block_size     = 0;
+        int                         loop           = 0;
+        off_t                       offset         = 0;
+        afr_private_t               *priv          = NULL;
+
+        priv    = this->private;
+        local   = sh_frame->local;
+        sh      = &local->self_heal;
+        sh_priv = sh->private;
+
+        LOCK (&sh_priv->lock);
+        {
+                if (!is_first_call)
+                        sh_priv->loops_running--;
+                offset = sh_priv->offset;
+                block_size = sh->block_size;
+                while ((!sh->eof_reached) &&
+                       (!is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) &&
+                     (sh_priv->loops_running < priv->data_self_heal_window_size)
+                       && (sh_priv->offset < sh->file_size)) {
+
+                        loop++;
+                        sh_priv->offset += block_size;
+                        sh_priv->loops_running++;
+
+                        if (!is_first_call)
+                                break;
+                }
+                if (0 == sh_priv->loops_running) {
+                        is_driver_done = _gf_true;
+                }
+        }
+        UNLOCK (&sh_priv->lock);
+
+        if (0 == loop) {
+                //loop finish does unlock, but the erasing of the pending
+                //xattrs needs to happen before that so do not finish the loop
+                if (is_driver_done &&
+                    !is_self_heal_failed (sh, AFR_CHECK_SPECIFIC))
+                        goto driver_done;
+                if (old_loop_frame) {
+                        sh_loop_finish (old_loop_frame, this);
+                        old_loop_frame = NULL;
+                }
+        }
+
+        //If we have more loops to form we should finish previous loop after
+        //the next loop lock
+        while (loop--) {
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
+                        // op failed in other loop, stop spawning more loops
+                        if (old_loop_frame) {
+                                sh_loop_finish (old_loop_frame, this);
+                                old_loop_frame = NULL;
+                        }
+                        sh_loop_driver (sh_frame, this, _gf_false, NULL);
+                } else {
+                        gf_log (this->name, GF_LOG_TRACE, "spawning a loop "
+                                "for offset %"PRId64, offset);
+
+                        sh_loop_start (sh_frame, this, offset, old_loop_frame);
+                        old_loop_frame = NULL;
+                        offset += block_size;
+                }
+        }
+
+driver_done:
+        if (is_driver_done) {
+                sh_loop_driver_done (sh_frame, this, old_loop_frame);
+        }
+        return 0;
+}
+
+static int
+sh_loop_return (call_frame_t *sh_frame, xlator_t *this, call_frame_t *loop_frame,
+                int32_t op_ret, int32_t op_errno)
+{
+        afr_local_t *               loop_local = NULL;
+        afr_self_heal_t *           loop_sh    = NULL;
+        afr_local_t *               sh_local = NULL;
+        afr_self_heal_t            *sh       = NULL;
+
+        sh_local = sh_frame->local;
+        sh       = &sh_local->self_heal;
+
+        if (loop_frame) {
+                loop_local = loop_frame->local;
+                if (loop_local)
+                        loop_sh    = &loop_local->self_heal;
+                if (loop_sh)
+                        gf_log (this->name, GF_LOG_TRACE, "loop for offset "
+                                "%"PRId64" returned", loop_sh->offset);
+        }
+
+        if (op_ret == -1) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                afr_sh_set_error (sh, op_errno);
+                if (loop_frame) {
+                        sh_loop_finish (loop_frame, this);
+                        loop_frame = NULL;
+                }
+        }
+
+        sh_loop_driver (sh_frame, this, _gf_false, loop_frame);
+
+        return 0;
+}
+
+static int
+sh_loop_write_cbk (call_frame_t *loop_frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+        afr_private_t *             priv        = NULL;
+        afr_local_t *               loop_local    = NULL;
+        afr_self_heal_t *           loop_sh       = NULL;
+        call_frame_t               *sh_frame    = NULL;
+        afr_local_t *               sh_local    = NULL;
+        afr_self_heal_t            *sh          = NULL;
+        int                         call_count  = 0;
+        int                         child_index = 0;
+
+        priv     = this->private;
+        loop_local = loop_frame->local;
+        loop_sh    = &loop_local->self_heal;
+
+        sh_frame = loop_sh->sh_frame;
+        sh_local = sh_frame->local;
+        sh       = &sh_local->self_heal;
+
+        child_index =  (long) cookie;
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "wrote %d bytes of data from %s to child %d, offset %"PRId64"",
+                op_ret, sh_local->loc.path, child_index, loop_sh->offset);
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "write to %s failed on subvolume %s (%s)",
+                        sh_local->loc.path,
+                        priv->children[child_index]->name,
+                        strerror (op_errno));
+
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                afr_sh_set_error (loop_sh, op_errno);
+        } else if (op_ret < loop_local->cont.writev.vector->iov_len) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "incomplete write to %s on subvolume %s "
+                        "(expected %lu, returned %d)", sh_local->loc.path,
+                        priv->children[child_index]->name,
+                        loop_local->cont.writev.vector->iov_len, op_ret);
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        }
+
+        call_count = afr_frame_return (loop_frame);
+
+        if (call_count == 0) {
+		iobref_unref(loop_local->cont.writev.iobref);
+
+                sh_loop_return (sh_frame, this, loop_frame,
+                                loop_sh->op_ret, loop_sh->op_errno);
+        }
+
+        return 0;
+}
+
+static void
+sh_prune_writes_needed (call_frame_t *sh_frame, call_frame_t *loop_frame,
+                        afr_private_t *priv)
+{
+        afr_local_t     *sh_local     = NULL;
+        afr_self_heal_t *sh           = NULL;
+        afr_local_t     *loop_local   = NULL;
+        afr_self_heal_t *loop_sh      = NULL;
+        int             i             = 0;
+
+        sh_local   = sh_frame->local;
+        sh         = &sh_local->self_heal;
+
+        if (!strcmp (sh->algo->name, "diff"))
+                return;
+
+        loop_local = loop_frame->local;
+        loop_sh    = &loop_local->self_heal;
+
+        /* full self-heal guarantees there exists atleast 1 file with size 0
+         * That means for other files we can preserve holes that come after
+         * its size before 'trim'
+         */
+        for (i = 0; i < priv->child_count; i++) {
+                if (loop_sh->write_needed[i] &&
+                    ((loop_sh->offset + 1) > sh->buf[i].ia_size))
+                        loop_sh->write_needed[i] = 0;
+        }
+}
+
+static int
+sh_loop_read_cbk (call_frame_t *loop_frame, void *cookie,
+                  xlator_t *this, int32_t op_ret, int32_t op_errno,
+                  struct iovec *vector, int32_t count, struct iatt *buf,
+                  struct iobref *iobref, dict_t *xdata)
+{
+        afr_private_t *               priv       = NULL;
+        afr_local_t *                 loop_local   = NULL;
+        afr_self_heal_t *             loop_sh      = NULL;
+        call_frame_t                 *sh_frame   = NULL;
+        int                           i          = 0;
+        int                           call_count = 0;
+        afr_local_t *                 sh_local   = NULL;
+        afr_self_heal_t *             sh      = NULL;
+
+        priv       = this->private;
+        loop_local = loop_frame->local;
+        loop_sh    = &loop_local->self_heal;
+
+        sh_frame = loop_sh->sh_frame;
+        sh_local = sh_frame->local;
+        sh       = &sh_local->self_heal;
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "read %d bytes of data from %s, offset %"PRId64"",
+                op_ret, loop_local->loc.path, loop_sh->offset);
+
+        if (op_ret <= 0) {
+                if (op_ret < 0) {
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                        gf_log (this->name, GF_LOG_ERROR, "read failed on %d "
+                                "for %s reason :%s", sh->source,
+                                sh_local->loc.path, strerror (errno));
+                } else {
+                        sh->eof_reached = _gf_true;
+                        gf_log (this->name, GF_LOG_DEBUG, "Eof reached for %s",
+                                sh_local->loc.path);
+                }
+                sh_loop_return (sh_frame, this, loop_frame, op_ret, op_errno);
+                goto out;
+        }
+
+        if (loop_sh->file_has_holes && iov_0filled (vector, count) == 0)
+                sh_prune_writes_needed (sh_frame, loop_frame, priv);
+
+        call_count = sh_number_of_writes_needed (loop_sh->write_needed,
+                                                 priv->child_count);
+        if (call_count == 0) {
+                sh_loop_return (sh_frame, this, loop_frame, 0, 0);
+                goto out;
+        }
+
+        loop_local->call_count = call_count;
+
+	/*
+	 * We only really need the request size at the moment, but the buffer
+	 * is required if we want to issue a retry in the event of a short write.
+	 * Therefore, we duplicate the vector and ref the iobref here...
+	 */
+	loop_local->cont.writev.vector = iov_dup(vector, count);
+	loop_local->cont.writev.iobref = iobref_ref(iobref);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (!loop_sh->write_needed[i])
+                        continue;
+                STACK_WIND_COOKIE (loop_frame, sh_loop_write_cbk,
+                                   (void *) (long) i,
+                                   priv->children[i],
+                                   priv->children[i]->fops->writev,
+                                   loop_sh->healing_fd, vector, count,
+                                   loop_sh->offset, 0, iobref, NULL);
+
+                if (!--call_count)
+                        break;
+        }
+
+out:
+        return 0;
+}
+
+
+static int
+sh_loop_read (call_frame_t *loop_frame, xlator_t *this)
+{
+        afr_private_t           *priv       = NULL;
+        afr_local_t             *loop_local   = NULL;
+        afr_self_heal_t         *loop_sh      = NULL;
+
+        priv     = this->private;
+        loop_local = loop_frame->local;
+        loop_sh    = &loop_local->self_heal;
+
+        STACK_WIND_COOKIE (loop_frame, sh_loop_read_cbk,
+                           (void *) (long) loop_sh->source,
+                           priv->children[loop_sh->source],
+                           priv->children[loop_sh->source]->fops->readv,
+                           loop_sh->healing_fd, loop_sh->block_size,
+                           loop_sh->offset, 0, NULL);
+
+        return 0;
+}
+
+
+static int
+sh_diff_checksum_cbk (call_frame_t *loop_frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno,
+                      uint32_t weak_checksum, uint8_t *strong_checksum,
+                      dict_t *xdata)
+{
+        afr_private_t                 *priv         = NULL;
+        afr_local_t                   *loop_local   = NULL;
+        afr_self_heal_t               *loop_sh      = NULL;
+        call_frame_t                  *sh_frame     = NULL;
+        afr_local_t                   *sh_local     = NULL;
+        afr_self_heal_t               *sh           = NULL;
+        afr_sh_algo_private_t         *sh_priv      = NULL;
+        int                           child_index  = 0;
+        int                           call_count   = 0;
+        int                           i            = 0;
+        int                           write_needed = 0;
+
+        priv  = this->private;
+
+        loop_local = loop_frame->local;
+        loop_sh    = &loop_local->self_heal;
+
+        sh_frame = loop_sh->sh_frame;
+        sh_local = sh_frame->local;
+        sh       = &sh_local->self_heal;
+
+        sh_priv = sh->private;
+
+        child_index = (long) cookie;
+
+        if (op_ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "checksum on %s failed on subvolume %s (%s)",
+                        sh_local->loc.path, priv->children[child_index]->name,
+                        strerror (op_errno));
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        } else {
+                memcpy (loop_sh->checksum + child_index * MD5_DIGEST_LENGTH,
+                        strong_checksum, MD5_DIGEST_LENGTH);
+        }
+
+        call_count = afr_frame_return (loop_frame);
+
+        if (call_count == 0) {
+                for (i = 0; i < priv->child_count; i++) {
+                        if (sh->sources[i] || !sh_local->child_up[i])
+                                continue;
+
+                        if (memcmp (loop_sh->checksum + (i * MD5_DIGEST_LENGTH),
+                                    loop_sh->checksum + (sh->source * MD5_DIGEST_LENGTH),
+                                    MD5_DIGEST_LENGTH)) {
+                                /*
+                                  Checksums differ, so this block
+                                  must be written to this sink
+                                */
+
+                                gf_log (this->name, GF_LOG_DEBUG,
+                                        "checksum on subvolume %s at offset %"
+                                        PRId64" differs from that on source",
+                                        priv->children[i]->name, loop_sh->offset);
+
+                                write_needed = loop_sh->write_needed[i] = 1;
+                        }
+                }
+
+                LOCK (&sh_priv->lock);
+                {
+                        sh_priv->total_blocks++;
+                        if (write_needed)
+                                sh_priv->diff_blocks++;
+                }
+                UNLOCK (&sh_priv->lock);
+
+                if (write_needed &&
+                    !is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
+                        sh_loop_read (loop_frame, this);
+                } else {
+                        sh_loop_return (sh_frame, this, loop_frame,
+                                        op_ret, op_errno);
+                }
+        }
+
+        return 0;
+}
+
+static int
+sh_diff_checksum (call_frame_t *loop_frame, xlator_t *this)
+{
+        afr_private_t           *priv         = NULL;
+        afr_local_t             *loop_local   = NULL;
+        afr_self_heal_t         *loop_sh      = NULL;
+        int                     call_count    = 0;
+        int                     i             = 0;
+
+        priv         = this->private;
+        loop_local   = loop_frame->local;
+        loop_sh      = &loop_local->self_heal;
+
+        call_count = loop_sh->active_sinks + 1;  /* sinks and source */
+
+        loop_local->call_count = call_count;
+
+        STACK_WIND_COOKIE (loop_frame, sh_diff_checksum_cbk,
+                           (void *) (long) loop_sh->source,
+                           priv->children[loop_sh->source],
+                           priv->children[loop_sh->source]->fops->rchecksum,
+                           loop_sh->healing_fd,
+                           loop_sh->offset, loop_sh->block_size, NULL);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (loop_sh->sources[i] || !loop_local->child_up[i])
+                        continue;
+
+                STACK_WIND_COOKIE (loop_frame, sh_diff_checksum_cbk,
+                                   (void *) (long) i,
+                                   priv->children[i],
+                                   priv->children[i]->fops->rchecksum,
+                                   loop_sh->healing_fd,
+                                   loop_sh->offset, loop_sh->block_size, NULL);
+
+                if (!--call_count)
+                        break;
+        }
+
+        return 0;
+}
+
+static int
+sh_full_read_write_to_sinks (call_frame_t *loop_frame, xlator_t *this)
+{
+        afr_private_t           *priv         = NULL;
+        afr_local_t             *loop_local   = NULL;
+        afr_self_heal_t         *loop_sh      = NULL;
+        int                     i             = 0;
+
+        priv         = this->private;
+        loop_local   = loop_frame->local;
+        loop_sh      = &loop_local->self_heal;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (loop_sh->sources[i] || !loop_local->child_up[i])
+                        continue;
+                loop_sh->write_needed[i] = 1;
+        }
+        sh_loop_read (loop_frame, this);
+        return 0;
+}
+
+afr_sh_algo_private_t*
+afr_sh_priv_init ()
+{
+        afr_sh_algo_private_t   *sh_priv = NULL;
+
+        sh_priv = GF_CALLOC (1, sizeof (*sh_priv),
+                             gf_afr_mt_afr_private_t);
+        if (!sh_priv)
+                goto out;
+
+        LOCK_INIT (&sh_priv->lock);
+out:
+        return sh_priv;
+}
+
+int
+afr_sh_transfer_lock (call_frame_t *dst, call_frame_t *src, char *dom,
+                      unsigned int child_count)
+{
+        afr_local_t             *dst_local   = NULL;
+        afr_self_heal_t         *dst_sh      = NULL;
+        afr_local_t             *src_local   = NULL;
+        afr_self_heal_t         *src_sh      = NULL;
+        int                     ret          = -1;
+
+        dst_local = dst->local;
+        dst_sh = &dst_local->self_heal;
+        src_local = src->local;
+        src_sh = &src_local->self_heal;
+        GF_ASSERT (src_sh->data_lock_held);
+        GF_ASSERT (!dst_sh->data_lock_held);
+        ret = afr_lk_transfer_datalock (dst, src, dom, child_count);
+        if (ret)
+                return ret;
+        src_sh->data_lock_held = _gf_false;
+        dst_sh->data_lock_held = _gf_true;
+        return 0;
+}
+
+int
+afr_sh_start_loops (call_frame_t *sh_frame, xlator_t *this,
+                    afr_sh_algo_fn sh_data_algo_start)
+{
+        call_frame_t            *first_loop_frame = NULL;
+        afr_local_t             *local   = NULL;
+        afr_self_heal_t         *sh      = NULL;
+        int                     ret      = 0;
+        afr_private_t           *priv    = NULL;
+
+        local = sh_frame->local;
+        sh    = &local->self_heal;
+        priv  = this->private;
+
+        sh->sh_data_algo_start = sh_data_algo_start;
+        local->call_count = 0;
+        ret = sh_loop_frame_create (sh_frame, this, NULL, &first_loop_frame);
+        if (ret)
+                goto out;
+        ret = afr_sh_transfer_lock (first_loop_frame, sh_frame, this->name,
+                                    priv->child_count);
+        if (ret)
+                goto out;
+        sh->private = afr_sh_priv_init ();
+        if (!sh->private) {
+                ret = -1;
+                goto out;
+        }
+        sh_loop_driver (sh_frame, this, _gf_true, first_loop_frame);
+        ret = 0;
+out:
+        if (ret) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                sh_loop_driver_done (sh_frame, this, NULL);
+        }
+        return 0;
+}
+
+int
+afr_sh_algo_diff (call_frame_t *sh_frame, xlator_t *this)
+{
+        afr_sh_start_loops (sh_frame, this, sh_diff_checksum);
+        return 0;
+}
+
+int
+afr_sh_algo_full (call_frame_t *sh_frame, xlator_t *this)
+{
+        afr_sh_start_loops (sh_frame, this, sh_full_read_write_to_sinks);
+        return 0;
+}
+
+struct afr_sh_algorithm afr_self_heal_algorithms[] = {
+        {.name = "full",  .fn = afr_sh_algo_full},
+        {.name = "diff",  .fn = afr_sh_algo_diff},
+        {0, 0},
+};
diff --git a/xlators/cluster/afr/src/afr-self-heal-algorithm.h b/xlators/cluster/afr/src/afr-self-heal-algorithm.h
new file mode 100644
index 000000000..6b20789b1
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heal-algorithm.h
@@ -0,0 +1,32 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __AFR_SELF_HEAL_ALGORITHM_H__
+#define __AFR_SELF_HEAL_ALGORITHM_H__
+
+typedef int (*afr_sh_algo_fn) (call_frame_t *frame,
+                               xlator_t *this);
+
+struct afr_sh_algorithm {
+        const char *name;
+        afr_sh_algo_fn fn;
+};
+
+extern struct afr_sh_algorithm afr_self_heal_algorithms[3];
+typedef struct {
+        gf_lock_t lock;
+        unsigned int loops_running;
+        off_t offset;
+
+        int32_t total_blocks;
+        int32_t diff_blocks;
+} afr_sh_algo_private_t;
+
+#endif /* __AFR_SELF_HEAL_ALGORITHM_H__ */
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c
index 04530ccb0..ef92b4205 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.c
+++ b/xlators/cluster/afr/src/afr-self-heal-common.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  General Public License for more details.
-
-  You should have received a copy of the GNU General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include "glusterfs.h"
@@ -25,217 +16,312 @@
 #include "afr-transaction.h"
 #include "afr-self-heal-common.h"
 #include "afr-self-heal.h"
+#include "pump.h"
+
+#define ADD_FMT_STRING(msg, off, sh_str, status, print_log)                 \
+        do {                                                                \
+                if (AFR_SELF_HEAL_NOT_ATTEMPTED != status) {                \
+                        off += snprintf (msg + off, sizeof (msg) - off,     \
+                                         " "sh_str" self heal %s,",         \
+                                         get_sh_completion_status (status));\
+                        print_log = 1;                                      \
+                }                                                           \
+        } while (0)
+
+#define ADD_FMT_STRING_SYNC(msg, off, sh_str, status, print_log)            \
+        do {                                                                \
+                if (AFR_SELF_HEAL_SYNC_BEGIN == status ||                   \
+                    AFR_SELF_HEAL_FAILED == status)  {                      \
+                        off += snprintf (msg + off, sizeof (msg) - off,     \
+                                         " "sh_str" self heal %s,",         \
+                                         get_sh_completion_status (status));\
+                        print_log = 1;                                      \
+                }                                                           \
+        } while (0)
+
+
+void
+afr_sh_reset (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        memset (sh->child_errno, 0,
+                sizeof (*sh->child_errno) * priv->child_count);
+        memset (sh->buf, 0, sizeof (*sh->buf) * priv->child_count);
+        memset (sh->parentbufs, 0,
+                sizeof (*sh->parentbufs) * priv->child_count);
+        memset (sh->success, 0, sizeof (*sh->success) * priv->child_count);
+        memset (sh->locked_nodes, 0,
+                sizeof (*sh->locked_nodes) * priv->child_count);
+        sh->active_sinks = 0;
+
+        afr_reset_xattr (sh->xattr, priv->child_count);
+}
 
+//Intersection[child]=1 if child is part of intersection
+void
+afr_children_intersection_get (int32_t *set1, int32_t *set2,
+                               int *intersection, unsigned int child_count)
+{
+        int                      i = 0;
+
+        memset (intersection, 0, sizeof (*intersection) * child_count);
+        for (i = 0; i < child_count; i++) {
+                intersection[i] = afr_is_child_present (set1, child_count, i)
+                                     && afr_is_child_present (set2, child_count,
+                                                              i);
+        }
+}
 
 /**
  * select_source - select a source and return it
- * TODO: take into account option 'favorite-child'
  */
 
 int
 afr_sh_select_source (int sources[], int child_count)
 {
-	int i;
-	for (i = 0; i < child_count; i++)
-		if (sources[i])
-			return i;
+        int i = 0;
+        for (i = 0; i < child_count; i++)
+                if (sources[i])
+                        return i;
 
-	return -1;
+        return -1;
 }
 
-
-/**
- * sink_count - return number of sinks in sources array
- */
-
-int
-afr_sh_sink_count (int sources[], int child_count)
+void
+afr_sh_mark_source_sinks (call_frame_t *frame, xlator_t *this)
 {
-	int i;
-	int sinks = 0;
-	for (i = 0; i < child_count; i++)
-		if (!sources[i])
-			sinks++;
-	return sinks;
+        int              i = 0;
+        afr_local_t     *local      = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              active_sinks = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (sh->sources[i] == 0 && local->child_up[i] == 1) {
+                        active_sinks++;
+                        sh->success[i] = 1;
+                } else if (sh->sources[i] == 1 && local->child_up[i] == 1) {
+                        sh->success[i] = 1;
+                }
+        }
+        sh->active_sinks = active_sinks;
 }
 
 int
 afr_sh_source_count (int sources[], int child_count)
 {
-	int i;
-	int nsource = 0;
+        int i = 0;
+        int nsource = 0;
 
-	for (i = 0; i < child_count; i++)
-		if (sources[i])
-			nsource++;
-	return nsource;
+        for (i = 0; i < child_count; i++)
+                if (sources[i])
+                        nsource++;
+        return nsource;
 }
 
+void
+afr_sh_set_error (afr_self_heal_t *sh, int32_t op_errno)
+{
+        sh->op_ret = -1;
+	sh->op_errno = afr_most_important_error(sh->op_errno, op_errno,
+						_gf_false);
+}
 
-int
-afr_sh_supress_errenous_children (int sources[], int child_errno[],
-				  int child_count)
+void
+afr_sh_print_pending_matrix (int32_t *pending_matrix[], xlator_t *this)
 {
-	int i = 0;
+        afr_private_t *  priv = this->private;
+        char            *buf  = NULL;
+        char            *ptr  = NULL;
+        int              i    = 0;
+        int              j    = 0;
 
-	for (i = 0; i < child_count; i++) {
-		if (child_errno[i] && sources[i]) {
-			sources[i] = 0;
-		}
-	}
+        /* 10 digits per entry + 1 space + '[' and ']' */
+        buf = GF_MALLOC (priv->child_count * 11 + 8, gf_afr_mt_char);
 
-	return 0;
-}
+        for (i = 0; i < priv->child_count; i++) {
+                ptr = buf;
+                ptr += sprintf (ptr, "[ ");
+                for (j = 0; j < priv->child_count; j++) {
+                        ptr += sprintf (ptr, "%d ", pending_matrix[i][j]);
+                }
+                sprintf (ptr, "]");
+                gf_log (this->name, GF_LOG_DEBUG, "pending_matrix: %s", buf);
+        }
 
+        GF_FREE (buf);
+}
 
-int
-afr_sh_supress_empty_children (int sources[], dict_t *xattr[],
-			       struct stat *buf,
-			       int child_count, const char *key)
+char*
+afr_get_pending_matrix_str (int32_t *pending_matrix[], xlator_t *this)
 {
-	int      i = 0;
-	int32_t *pending = NULL;
-	int      ret = 0;
-	int      all_xattr_missing = 1;
-
-	/* if the file was created by afr with xattrs */
-	for (i = 0; i < child_count; i++) {
-		if (!xattr[i])
-			continue;
+        afr_private_t *  priv = this->private;
+        char            *buf  = NULL;
+        char            *ptr  = NULL;
+        int              i    = 0;
+        int              j    = 0;
+        int             child_count = priv->child_count;
+        char            *matrix_begin = "[ [ ";
+        char            *matrix_end = "] ]";
+        char            *seperator = "] [ ";
+        int             pending_entry_strlen = 12; //Including space after entry
+        int             matrix_begin_strlen = 0;
+        int             matrix_end_strlen = 0;
+        int             seperator_strlen = 0;
+        int             string_length = 0;
+        char            *msg = "- Pending matrix:  ";
+
+        /*
+         *  for a list of lists of [ [ a b ] [ c d ] ]
+         * */
+
+        matrix_begin_strlen = strlen (matrix_begin);
+        matrix_end_strlen = strlen (matrix_end);
+        seperator_strlen = strlen (seperator);
+        string_length = matrix_begin_strlen + matrix_end_strlen
+                        + (child_count -1) * seperator_strlen
+                        + (child_count * child_count * pending_entry_strlen);
+
+        buf = GF_CALLOC (1, 1 + strlen (msg) + string_length , gf_afr_mt_char);
+        if (!buf)
+                goto out;
 
-		ret = dict_get_ptr (xattr[i], (char *)key, VOID(&pending));
-		if (ret != 0) {
-			continue;
-		}
+        ptr = buf;
+        ptr += sprintf (ptr, "%s", msg);
+        ptr += sprintf (ptr, "%s", matrix_begin);
+        for (i = 0; i < priv->child_count; i++) {
+                for (j = 0; j < priv->child_count; j++) {
+                        ptr += sprintf (ptr, "%d ", pending_matrix[i][j]);
+                }
+                if (i < priv->child_count -1)
+                        ptr += sprintf (ptr, "%s", seperator);
+        }
 
-		all_xattr_missing = 0;
-		break;
-	}
-
-	if (all_xattr_missing) {
-		/* supress 0byte files.. this avoids empty file created
-		   by dir selfheal to overwrite the 'good' file */
-		for (i = 0; i < child_count; i++) {
-			if (!buf[i].st_size)
-				sources[i] = 0;
-		}
-		goto out;
-	}
+        ptr += sprintf (ptr, "%s", matrix_end);
 
+out:
+        return buf;
+}
 
-	for (i = 0; i < child_count; i++) {
-		if (!xattr[i]) {
-			sources[i] = 0;
-			continue;
-		}
+void
+afr_sh_print_split_brain_log (int32_t *pending_matrix[], xlator_t *this,
+                              const char *loc)
+{
+        char *buf      = NULL;
+        char *free_ptr = NULL;
 
-		ret = dict_get_ptr (xattr[i], (char *)key, VOID(&pending));
-		if (ret != 0) {
-			sources[i] = 0;
-			continue;
-		}
+        buf = afr_get_pending_matrix_str (pending_matrix, this);
+        if (buf)
+                free_ptr = buf;
+        else
+                buf = "";
 
-		if (!pending) {
-			sources[i] = 0;
-			continue;
-		}
-	}
 
-out:
-	return 0;
+        gf_log (this->name, GF_LOG_ERROR, "Unable to self-heal contents of '%s'"
+                " (possible split-brain). Please delete the file from all but "
+                "the preferred subvolume.%s", loc, buf);
+        GF_FREE (free_ptr);
+        return;
 }
 
 
 void
-afr_sh_print_pending_matrix (int32_t *pending_matrix[], xlator_t *this)
+afr_init_pending_matrix (int32_t **pending_matrix, size_t child_count)
 {
-	afr_private_t * priv = this->private;
+        int             i   = 0;
+        int             j   = 0;
 
-	char *buf = NULL;
-	char *ptr = NULL;
+        GF_ASSERT (pending_matrix);
 
-	int i, j;
+        for (i = 0; i < child_count; i++) {
+                for (j = 0; j < child_count; j++) {
+                        pending_matrix[i][j] = 0;
+                }
+        }
+}
 
-        /* 10 digits per entry + 1 space + '[' and ']' */
-	buf = MALLOC (priv->child_count * 11 + 8); 
+void
+afr_mark_ignorant_subvols_as_pending (int32_t **pending_matrix,
+                                      unsigned char *ignorant_subvols,
+                                      size_t  child_count)
+{
+        int            i                = 0;
+        int            j                = 0;
 
-	for (i = 0; i < priv->child_count; i++) {
-		ptr = buf;
-		ptr += sprintf (ptr, "[ ");
-		for (j = 0; j < priv->child_count; j++) {
-			ptr += sprintf (ptr, "%d ", pending_matrix[i][j]);
-		}
-		ptr += sprintf (ptr, "]");
-		gf_log (this->name, GF_LOG_DEBUG,
-			"pending_matrix: %s", buf);
-	}
+        GF_ASSERT (pending_matrix);
+        GF_ASSERT (ignorant_subvols);
 
-	FREE (buf);
+        for (i = 0; i < child_count; i++) {
+                if (ignorant_subvols[i]) {
+                        for (j = 0; j < child_count; j++) {
+                                if (!ignorant_subvols[j])
+                                        pending_matrix[j][i] += 1;
+                        }
+                }
+        }
 }
 
-
-void
-afr_sh_build_pending_matrix (int32_t *pending_matrix[], dict_t *xattr[],
-			     int child_count, const char *key)
+int
+afr_build_pending_matrix (char **pending_key, int32_t **pending_matrix,
+                          unsigned char *ignorant_subvols,
+                          dict_t *xattr[], afr_transaction_type type,
+                          size_t child_count)
 {
-	int i = 0;
-	int j = 0;
-	int32_t *pending = NULL;
-	int ret = -1;
+        /* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
+        int32_t        pending[3]       = {0,};
+        void          *pending_raw      = NULL;
+        int            ret              = -1;
+        int            i                = 0;
+        int            j                = 0;
+        int            k                = 0;
 
-	/* start clean */
-	for (i = 0; i < child_count; i++) {
-		for (j = 0; j < child_count; j++) {
-			pending_matrix[i][j] = 0;
-		}
-	}
+        afr_init_pending_matrix (pending_matrix, child_count);
 
-	for (i = 0; i < child_count; i++) {
-		if (!xattr[i])
-			continue;
+        for (i = 0; i < child_count; i++) {
+                pending_raw = NULL;
+
+                for (j = 0; j < child_count; j++) {
+                        ret = dict_get_ptr (xattr[i], pending_key[j],
+                                            &pending_raw);
+
+                        if (ret != 0) {
+                                /*
+                                 * There is no xattr present. This means this
+                                 * subvolume should be considered an 'ignorant'
+                                 * subvolume.
+                                 */
+
+                                if (ignorant_subvols)
+                                        ignorant_subvols[i] = 1;
+                                continue;
+                        }
 
-		pending = NULL;
+                        memcpy (pending, pending_raw, sizeof(pending));
+                        k = afr_index_for_transaction_type (type);
 
-		ret = dict_get_ptr (xattr[i], (char *) key,
-				    VOID(&pending));
-		if (ret != 0)
-			continue;
+                        pending_matrix[i][j] = ntoh32 (pending[k]);
+                }
+        }
 
-		for (j = 0; j < child_count; j++) {
-			pending_matrix[i][j] = ntoh32 (pending[j]);
-		}
-	}
+        return ret;
 }
 
-
-/**
- * mark_sources: Mark all 'source' nodes and return number of source
- * nodes found
- *
- * A node (a row in the pending matrix) belongs to one of
- * three categories:
- *
- * M is the pending matrix.
- *
- * 'innocent' - M[i] is all zeroes
- * 'fool'     - M[i] has i'th element = 1 (self-reference)
- * 'wise'     - M[i] has i'th element = 0, others are 1 or 0.
- *
- * All 'innocent' nodes are sinks. If all nodes are innocent, no self-heal is
- * needed.
- *
- * A 'wise' node can be a source. If two 'wise' nodes conflict, it is 
- * a split-brain. If one wise node refers to the other but the other doesn't
- * refer back, the referrer is a source.
- *
- * All fools are sinks, unless there are no 'wise' nodes. In that case,
- * one of the fools is made a source.
- */
-
 typedef enum {
+        AFR_NODE_INVALID,
         AFR_NODE_INNOCENT,
         AFR_NODE_FOOL,
-        AFR_NODE_WISE
+        AFR_NODE_WISE,
 } afr_node_type;
 
 typedef struct {
@@ -276,7 +362,7 @@ afr_sh_is_wise (int32_t *array, int i, int child_count)
 
 
 static int
-afr_sh_all_nodes_innocent (afr_node_character *characters, 
+afr_sh_all_nodes_innocent (afr_node_character *characters,
                            int child_count)
 {
         int i   = 0;
@@ -311,11 +397,11 @@ afr_sh_wise_nodes_exist (afr_node_character *characters, int child_count)
 
 
 /*
- * The 'wisdom' of a wise node is 0 if any other wise node accuses to it.
- * It is 1 if no other wise node accuses it. 
+ * The 'wisdom' of a wise node is 0 if any other wise node accuses it.
+ * It is 1 if no other wise node accuses it.
  * Only wise nodes with wisdom 1 are sources.
  *
- * If no nodes with wisdom 1 exist, a split-brain has occured.
+ * If no nodes with wisdom 1 exist, a split-brain has occurred.
  */
 
 static void
@@ -332,7 +418,7 @@ afr_sh_compute_wisdom (int32_t *pending_matrix[],
                         for (j = 0; j < child_count; j++) {
                                 if ((characters[j].type == AFR_NODE_WISE)
                                     && pending_matrix[j][i]) {
-                                        
+
                                         characters[i].wisdom = 0;
                                 }
                         }
@@ -342,7 +428,7 @@ afr_sh_compute_wisdom (int32_t *pending_matrix[],
 
 
 static int
-afr_sh_wise_nodes_conflict (afr_node_character *characters, 
+afr_sh_wise_nodes_conflict (afr_node_character *characters,
                             int child_count)
 {
         int i   = 0;
@@ -363,13 +449,12 @@ afr_sh_wise_nodes_conflict (afr_node_character *characters,
 
 
 static int
-afr_sh_mark_wisest_as_sources (int sources[], 
-                               afr_node_character *characters, 
+afr_sh_mark_wisest_as_sources (int sources[],
+                               afr_node_character *characters,
                                int child_count)
 {
         int nsources = 0;
-        
-        int i = 0;
+        int i        = 0;
 
         for (i = 0; i < child_count; i++) {
                 if (characters[i].wisdom == 1) {
@@ -381,906 +466,2347 @@ afr_sh_mark_wisest_as_sources (int sources[],
         return nsources;
 }
 
-        
-static int
-afr_sh_mark_a_fool_as_source (int sources[], afr_node_character *characters, 
-                              int child_count)
+static void
+afr_compute_witness_of_fools (int32_t *witnesses, int32_t **pending_matrix,
+                              afr_node_character *characters,
+                              int32_t child_count)
 {
-        int i = 0;
-        
+        int i       = 0;
+        int j       = 0;
+        int witness = 0;
+
+        GF_ASSERT (witnesses);
+        GF_ASSERT (pending_matrix);
+        GF_ASSERT (characters);
+        GF_ASSERT (child_count > 0);
+
+        for (i = 0; i < child_count; i++) {
+                if (characters[i].type != AFR_NODE_FOOL)
+                        continue;
+
+                witness = 0;
+                for (j = 0; j < child_count; j++) {
+                        if (i == j)
+                                continue;
+                        witness += pending_matrix[i][j];
+                }
+                witnesses[i] = witness;
+        }
+}
+
+static int32_t
+afr_find_biggest_witness_among_fools (int32_t *witnesses,
+                                      afr_node_character *characters,
+                                      int32_t child_count)
+{
+        int i               = 0;
+        int biggest_witness = -1;
+        int biggest_witness_idx = -1;
+        int biggest_witness_cnt = -1;
+
+        GF_ASSERT (witnesses);
+        GF_ASSERT (characters);
+        GF_ASSERT (child_count > 0);
+
+        for (i = 0; i < child_count; i++) {
+                if (characters[i].type != AFR_NODE_FOOL)
+                        continue;
+
+                if (biggest_witness < witnesses[i]) {
+                        biggest_witness = witnesses[i];
+			biggest_witness_idx = i;
+			biggest_witness_cnt = 1;
+			continue;
+		}
+
+		if (biggest_witness == witnesses[i])
+			biggest_witness_cnt++;
+        }
+
+	if (biggest_witness_cnt != 1)
+		return -1;
+
+        return biggest_witness_idx;
+}
+
+int
+afr_mark_fool_as_source_by_witness (int32_t *sources, int32_t *witnesses,
+                                    afr_node_character *characters,
+                                    int32_t child_count, int32_t witness)
+{
+        int i        = 0;
         int nsources = 0;
-        
+
+        GF_ASSERT (sources);
+        GF_ASSERT (witnesses);
+        GF_ASSERT (characters);
+        GF_ASSERT (child_count > 0);
+
         for (i = 0; i < child_count; i++) {
-                if (characters[i].type == AFR_NODE_FOOL) {
+                if (characters[i].type != AFR_NODE_FOOL)
+                        continue;
+
+                if (witness == witnesses[i]) {
                         sources[i] = 1;
                         nsources++;
-                        break;
                 }
         }
-
         return nsources;
 }
 
-        
+
 int
-afr_sh_mark_sources (int32_t *pending_matrix[], int sources[], int child_count)
+afr_mark_fool_as_source_by_idx (int32_t *sources, int child_count, int idx)
 {
-	int i = 0;
+	if (idx >= 0 && idx < child_count) {
+		sources[idx] = 1;
+		return 1;
+	}
+	return 0;
+}
 
-	int nsources = 0;
 
-        /* stores the 'characters' (innocent, fool, wise) of the nodes */
-        afr_node_character *
-                characters = CALLOC (sizeof (afr_node_character), 
-                                     child_count);
+static int
+afr_find_largest_file_size (struct iatt *bufs, int32_t *success_children,
+			    int child_count)
+{
+	int idx = -1;
+	int i = -1;
+	int child = -1;
+	uint64_t max_size = 0;
+        uint64_t min_size = 0;
+        int      num_children = 0;
+
+	for (i = 0; i < child_count; i++) {
+		if (success_children[i] == -1)
+			break;
+
+		child = success_children[i];
+		if (bufs[child].ia_size > max_size) {
+			max_size = bufs[child].ia_size;
+			idx = child;
+		}
+
+                if ((num_children == 0) || (bufs[child].ia_size < min_size)) {
+                        min_size = bufs[child].ia_size;
+                }
+
+                num_children++;
+	}
+
+        /* If sizes are same for all of them, finding sources will have to
+         * happen with pending changelog. So return -1
+         */
+        if ((num_children > 1) && (min_size == max_size))
+                return -1;
+	return idx;
+}
+
+
+static int
+afr_find_newest_file (struct iatt *bufs, int32_t *success_children,
+		      int child_count)
+{
+	int idx = -1;
+	int i = -1;
+	int child = -1;
+	uint64_t max_ctime = 0;
 
-	/* start clean */
 	for (i = 0; i < child_count; i++) {
-		sources[i] = 0;
+		if (success_children[i] == -1)
+			break;
+
+		child = success_children[i];
+		if (bufs[child].ia_ctime > max_ctime) {
+			max_ctime = bufs[child].ia_ctime;
+			idx = child;
+		}
 	}
-        
+
+	return idx;
+}
+
+
+static int
+afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix,
+                                     afr_node_character *characters,
+				     int32_t *success_children,
+                                     int child_count, struct iatt *bufs)
+{
+        int32_t       biggest_witness = 0;
+        int           nsources        = 0;
+        int32_t       *witnesses      = NULL;
+
+        GF_ASSERT (child_count > 0);
+
+	biggest_witness = afr_find_largest_file_size (bufs, success_children,
+						      child_count);
+	if (biggest_witness != -1)
+		goto found;
+
+        witnesses = GF_CALLOC (child_count, sizeof (*witnesses),
+                               gf_afr_mt_int32_t);
+        if (NULL == witnesses) {
+                nsources = -1;
+                goto out;
+        }
+
+        afr_compute_witness_of_fools (witnesses, pending_matrix, characters,
+                                      child_count);
+        biggest_witness = afr_find_biggest_witness_among_fools (witnesses,
+                                                                characters,
+                                                                child_count);
+	if (biggest_witness != -1)
+		goto found;
+
+	biggest_witness = afr_find_newest_file (bufs, success_children,
+						child_count);
+
+found:
+	nsources = afr_mark_fool_as_source_by_idx (sources, child_count,
+						   biggest_witness);
+out:
+        GF_FREE (witnesses);
+        return nsources;
+}
+
+int
+afr_mark_child_as_source_by_uid (int32_t *sources, struct iatt *bufs,
+                                 int32_t *success_children,
+                                 unsigned int child_count, uint32_t uid)
+{
+        int     i        = 0;
+        int     nsources = 0;
+        int     child    = 0;
+
         for (i = 0; i < child_count; i++) {
-                if (afr_sh_is_innocent (pending_matrix[i], child_count)) {
-                        characters[i].type = AFR_NODE_INNOCENT;
+                if (-1 == success_children[i])
+                        break;
 
-                } else if (afr_sh_is_fool (pending_matrix[i], i, child_count)) {
-                        characters[i].type = AFR_NODE_FOOL;
+                child = success_children[i];
+                if (uid == bufs[child].ia_uid) {
+                        sources[child] = 1;
+                        nsources++;
+                }
+        }
+        return nsources;
+}
 
-                } else if (afr_sh_is_wise (pending_matrix[i], i, child_count)) {
-                        characters[i].type = AFR_NODE_WISE;
+int
+afr_get_child_with_lowest_uid (struct iatt *bufs, int32_t *success_children,
+                               unsigned int child_count)
+{
+        int     i        = 0;
+        int     smallest = -1;
+        int     child    = 0;
 
-                } else {
-                        gf_log ("[module:afr]", GF_LOG_ERROR,
-                                "node %d is diabolical! "
-                                "(This message should never appear."
-                                " Please file a bug report.)", i);
+        for (i = 0; i < child_count; i++) {
+                if (-1 == success_children[i])
+                        break;
+                child = success_children[i];
+                if ((smallest == -1) ||
+                    (bufs[child].ia_uid < bufs[smallest].ia_uid)) {
+                        smallest = child;
                 }
         }
+        return smallest;
+}
+
+static int
+afr_sh_mark_lowest_uid_as_source (struct iatt *bufs, int32_t *success_children,
+                                  int child_count, int32_t *sources)
+{
+        int   nsources              = 0;
+        int   smallest              = 0;
+
+        smallest = afr_get_child_with_lowest_uid (bufs, success_children,
+                                                  child_count);
+        if (smallest < 0) {
+                nsources = -1;
+                goto out;
+        }
+        nsources = afr_mark_child_as_source_by_uid (sources, bufs,
+                                                    success_children, child_count,
+                                                    bufs[smallest].ia_uid);
+out:
+        return nsources;
+}
 
+int
+afr_get_no_xattr_dir_read_child (xlator_t *this, int32_t *success_children,
+                                 struct iatt *bufs)
+{
+        afr_private_t *priv = NULL;
+        int            i = 0;
+        int            child = -1;
+        int            read_child = -1;
+
+        priv = this->private;
+        for (i = 0; i < priv->child_count; i++) {
+                child = success_children[i];
+                if (child < 0)
+                        break;
+                if (read_child < 0)
+                        read_child = child;
+                else if (bufs[read_child].ia_size < bufs[child].ia_size)
+                        read_child = child;
+        }
+        return read_child;
+}
+
+int
+afr_sh_mark_zero_size_file_as_sink (struct iatt *bufs, int32_t *success_children,
+                                    int child_count, int32_t *sources)
+{
+        int             nsources = 0;
+        int             i = 0;
+        int             child = 0;
+        gf_boolean_t    sink_exists = _gf_false;
+        gf_boolean_t    source_exists = _gf_false;
+        int             source = -1;
+
+        for (i = 0; i < child_count; i++) {
+                child = success_children[i];
+                if (child < 0)
+                        break;
+                if (!bufs[child].ia_size) {
+                        sink_exists = _gf_true;
+                        continue;
+                }
+                if (!source_exists) {
+                        source_exists = _gf_true;
+                        source = child;
+                        continue;
+                }
+                if (bufs[source].ia_size != bufs[child].ia_size) {
+                        nsources = -1;
+                        goto out;
+                }
+        }
+        if (!source_exists && !sink_exists) {
+                nsources = -1;
+                goto out;
+        }
+
+        if (!source_exists || !sink_exists)
+                goto out;
+
+        for (i = 0; i < child_count; i++) {
+                child = success_children[i];
+                if (child < 0)
+                        break;
+                if (bufs[child].ia_size) {
+                        sources[child] = 1;
+                        nsources++;
+                }
+        }
+out:
+        return nsources;
+}
+
+char *
+afr_get_character_str (afr_node_type type)
+{
+        char *character = NULL;
+
+        switch (type) {
+        case AFR_NODE_INNOCENT:
+                character = "innocent";
+                break;
+        case AFR_NODE_FOOL:
+                character = "fool";
+                break;
+        case AFR_NODE_WISE:
+                character = "wise";
+                break;
+        default:
+                character = "invalid";
+                break;
+        }
+        return character;
+}
+
+afr_node_type
+afr_find_child_character_type (int32_t *pending_row, int32_t child,
+                               unsigned int child_count)
+{
+        afr_node_type type = AFR_NODE_INVALID;
+
+        GF_ASSERT ((child >= 0) && (child < child_count));
+
+        if (afr_sh_is_innocent (pending_row, child_count))
+                type = AFR_NODE_INNOCENT;
+        else if (afr_sh_is_fool (pending_row, child, child_count))
+                type = AFR_NODE_FOOL;
+        else if (afr_sh_is_wise (pending_row, child, child_count))
+                type = AFR_NODE_WISE;
+        return type;
+}
+
+int
+afr_build_sources (xlator_t *this, dict_t **xattr, struct iatt *bufs,
+                   int32_t **pending_matrix, int32_t *sources,
+                   int32_t *success_children, afr_transaction_type type,
+                   int32_t *subvol_status, gf_boolean_t ignore_ignorant)
+{
+        afr_private_t           *priv = NULL;
+        afr_self_heal_type      sh_type    = AFR_SELF_HEAL_INVALID;
+        int                     nsources   = -1;
+        unsigned char           *ignorant_subvols = NULL;
+        unsigned int            child_count = 0;
+
+        priv = this->private;
+        child_count = priv->child_count;
+
+        if (afr_get_children_count (success_children, priv->child_count) == 0)
+                goto out;
+
+        if (!ignore_ignorant) {
+                ignorant_subvols = GF_CALLOC (sizeof (*ignorant_subvols),
+                                              child_count, gf_afr_mt_char);
+                if (NULL == ignorant_subvols)
+                        goto out;
+        }
+
+        afr_build_pending_matrix (priv->pending_key, pending_matrix,
+                                  ignorant_subvols, xattr, type,
+                                  priv->child_count);
+
+        if (!ignore_ignorant)
+                afr_mark_ignorant_subvols_as_pending (pending_matrix,
+                                                      ignorant_subvols,
+                                                      priv->child_count);
+        sh_type = afr_self_heal_type_for_transaction (type);
+        if (AFR_SELF_HEAL_INVALID == sh_type)
+                goto out;
+
+        afr_sh_print_pending_matrix (pending_matrix, this);
+
+        nsources = afr_mark_sources (this, sources, pending_matrix, bufs,
+                                     sh_type, success_children, subvol_status);
+out:
+        GF_FREE (ignorant_subvols);
+        return nsources;
+}
+
+void
+afr_find_character_types (afr_node_character *characters,
+                          int32_t **pending_matrix, int32_t *success_children,
+                          unsigned int child_count)
+{
+        afr_node_type type  = AFR_NODE_INVALID;
+        int           child = 0;
+        int           i     = 0;
+
+        for (i = 0; i < child_count; i++) {
+                child = success_children[i];
+                if (child == -1)
+                        break;
+                type = afr_find_child_character_type (pending_matrix[child],
+                                                      child, child_count);
+                characters[child].type = type;
+        }
+}
+
+void
+afr_mark_success_children_sources (int32_t *sources, int32_t *success_children,
+                                   unsigned int child_count)
+{
+        int i = 0;
+        for (i = 0; i < child_count; i++) {
+                if (success_children[i] == -1)
+                        break;
+                sources[success_children[i]] = 1;
+        }
+}
+/**
+ * mark_sources: Mark all 'source' nodes and return number of source
+ * nodes found
+ *
+ * A node (a row in the pending matrix) belongs to one of
+ * three categories:
+ *
+ * M is the pending matrix.
+ *
+ * 'innocent' - M[i] is all zeroes
+ * 'fool'     - M[i] has i'th element = 1 (self-reference)
+ * 'wise'     - M[i] has i'th element = 0, others are 1 or 0.
+ *
+ * All 'innocent' nodes are sinks. If all nodes are innocent, no self-heal is
+ * needed.
+ *
+ * A 'wise' node can be a source. If two 'wise' nodes conflict, it is
+ * a split-brain. If one wise node refers to the other but the other doesn't
+ * refer back, the referrer is a source.
+ *
+ * All fools are sinks, unless there are no 'wise' nodes. In that case,
+ * one of the fools is made a source.
+ */
+
+int
+afr_mark_sources (xlator_t *this, int32_t *sources, int32_t **pending_matrix,
+                  struct iatt *bufs, afr_self_heal_type type,
+                  int32_t *success_children, int32_t *subvol_status)
+{
+        /* stores the 'characters' (innocent, fool, wise) of the nodes */
+        afr_node_character *characters =  NULL;
+        int                nsources    = -1;
+        unsigned int       child_count = 0;
+        afr_private_t      *priv       = NULL;
+
+        priv = this->private;
+        child_count = priv->child_count;
+        characters = GF_CALLOC (sizeof (afr_node_character),
+                                child_count, gf_afr_mt_afr_node_character);
+        if (!characters)
+                goto out;
+
+        this = THIS;
+
+        /* start clean */
+        memset (sources, 0, sizeof (*sources) * child_count);
+        nsources = 0;
+        afr_find_character_types (characters, pending_matrix, success_children,
+                                  child_count);
         if (afr_sh_all_nodes_innocent (characters, child_count)) {
-                /* no self-heal needed */
+                switch (type) {
+                case AFR_SELF_HEAL_METADATA:
+                        nsources = afr_sh_mark_lowest_uid_as_source (bufs,
+                                                             success_children,
+                                                             child_count,
+                                                             sources);
+                        break;
+                case AFR_SELF_HEAL_DATA:
+                        nsources = afr_sh_mark_zero_size_file_as_sink (bufs,
+                                                             success_children,
+                                                             child_count,
+                                                             sources);
+                        if ((nsources < 0) && subvol_status)
+                                *subvol_status |= SPLIT_BRAIN;
+                        break;
+                default:
+                        break;
+                }
                 goto out;
+        }
 
-        } else if (afr_sh_wise_nodes_exist (characters, child_count)) {
+        if (afr_sh_wise_nodes_exist (characters, child_count)) {
                 afr_sh_compute_wisdom (pending_matrix, characters, child_count);
 
                 if (afr_sh_wise_nodes_conflict (characters, child_count)) {
-                        /* split-brain */
-                        goto out;
-
+                        if (subvol_status)
+                                *subvol_status |= SPLIT_BRAIN;
+                        nsources = -1;
                 } else {
-                        nsources = afr_sh_mark_wisest_as_sources (sources, 
+                        nsources = afr_sh_mark_wisest_as_sources (sources,
                                                                   characters,
                                                                   child_count);
                 }
         } else {
-                nsources = afr_sh_mark_a_fool_as_source (sources, characters, 
-                                                         child_count);
+                if (subvol_status)
+                        *subvol_status |= ALL_FOOLS;
+                nsources = afr_mark_biggest_of_fools_as_source (sources,
+                                                                pending_matrix,
+                                                                characters,
+								success_children,
+                                                                child_count, bufs);
         }
 
 out:
-	return nsources;
-}
+        if (nsources == 0)
+                afr_mark_success_children_sources (sources, success_children,
+                                                   child_count);
+        GF_FREE (characters);
 
+        gf_log (this->name, GF_LOG_DEBUG, "Number of sources: %d", nsources);
+        return nsources;
+}
 
 void
-afr_sh_pending_to_delta (int32_t *pending_matrix[], int32_t *delta_matrix[],
-			 int success[], int child_count)
+afr_sh_pending_to_delta (afr_private_t *priv, dict_t **xattr,
+                         int32_t *delta_matrix[], unsigned char success[],
+                         int child_count, afr_transaction_type type)
 {
-	int i = 0;
-	int j = 0;
+        int     tgt     = 0;
+        int     src     = 0;
+        int     value   = 0;
+
+        afr_build_pending_matrix (priv->pending_key, delta_matrix, NULL,
+                                  xattr, type, priv->child_count);
+
+        /*
+         * The algorithm here has two parts.  First, for each subvol indexed
+         * as tgt, we try to figure out what count everyone should have for it.
+         * If the self-heal succeeded, that's easy; the value is zero.
+         * Otherwise, the value is the maximum of the succeeding nodes' counts.
+         * Once we know the value, we loop through (possibly for a second time)
+         * setting each count to the difference so that when we're done all
+         * succeeding nodes will have the same count for tgt.
+         */
+        for (tgt = 0; tgt < priv->child_count; ++tgt) {
+                value = 0;
+                if (!success[tgt]) {
+                        /* Find the maximum. */
+                        for (src = 0; src < priv->child_count; ++src) {
+                                if (!success[src]) {
+                                        continue;
+                                }
+                                if (delta_matrix[src][tgt] > value) {
+                                        value = delta_matrix[src][tgt];
+                                }
+                        }
+                }
+                /* Force everyone who succeeded to the chosen value. */
+                for (src = 0; src < priv->child_count; ++src) {
+                        if (success[src]) {
+                                delta_matrix[src][tgt] = value
+                                                       - delta_matrix[src][tgt];
+                        }
+                        else {
+                                delta_matrix[src][tgt] = 0;
+                        }
+                }
+        }
+}
 
-	/* start clean */
-	for (i = 0; i < child_count; i++) {
-		for (j = 0; j < child_count; j++) {
-			delta_matrix[i][j] = 0;
-		}
-	}
 
-	for (i = 0; i < child_count; i++) {
-		for (j = 0; j < child_count; j++) {
-			if (!success[j])
-				continue;
-			delta_matrix[i][j] = -pending_matrix[i][j];
-		}
-	}
+int
+afr_sh_delta_to_xattr (xlator_t *this,
+                       int32_t *delta_matrix[], dict_t *xattr[],
+                       int child_count, afr_transaction_type type)
+{
+        int              i       = 0;
+        int              j       = 0;
+        int              k       = 0;
+        int              ret     = 0;
+        int32_t         *pending = NULL;
+        int32_t         *local_pending = NULL;
+        afr_private_t   *priv = NULL;
+
+        priv = this->private;
+        for (i = 0; i < child_count; i++) {
+                if (!xattr[i])
+                        continue;
+
+                local_pending = NULL;
+                for (j = 0; j < child_count; j++) {
+                        pending = GF_CALLOC (sizeof (int32_t), 3,
+                                             gf_afr_mt_int32_t);
+
+                        if (!pending) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "failed to allocate pending entry "
+                                        "for %s[%d] on %s",
+                                        priv->pending_key[j], type,
+                                        priv->children[i]->name);
+                                continue;
+                        }
+                        /* 3 = data+metadata+entry */
+
+                        k = afr_index_for_transaction_type (type);
+
+                        pending[k] = hton32 (delta_matrix[i][j]);
+
+                        if (j == i) {
+                                local_pending = pending;
+                                continue;
+                        }
+                        ret = dict_set_bin (xattr[i], priv->pending_key[j],
+                                            pending,
+                                        AFR_NUM_CHANGE_LOGS * sizeof (int32_t));
+                        if (ret < 0) {
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "Unable to set dict value.");
+                                GF_FREE (pending);
+                        }
+                }
+                if (local_pending) {
+                        ret = dict_set_bin (xattr[i], priv->pending_key[i],
+                                            local_pending,
+                                        AFR_NUM_CHANGE_LOGS * sizeof (int32_t));
+                        if (ret < 0) {
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "Unable to set dict value.");
+                                GF_FREE (local_pending);
+                        }
+                }
+        }
+        return 0;
 }
 
 
 int
-afr_sh_delta_to_xattr (int32_t *delta_matrix[], dict_t *xattr[],
-		       int child_count, const char *key)
+afr_sh_missing_entries_done (call_frame_t *frame, xlator_t *this)
 {
-	int i = 0;
-	int j = 0;
-
-	int ret = 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
 
-	int32_t *pending = 0;
+        local = frame->local;
+        sh = &local->self_heal;
 
-	for (i = 0; i < child_count; i++) {
-		if (!xattr[i])
-			continue;
+        afr_sh_reset (frame, this);
 
-		pending = CALLOC (sizeof (int32_t), child_count);
-		for (j = 0; j < child_count; j++) {
-			pending[j] = hton32 (delta_matrix[i][j]);
-		}
+        if (local->unhealable) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "split brain found, aborting selfheal of %s",
+                        local->loc.path);
+        }
 
-		ret = dict_set_bin (xattr[i], (char *) key, pending,
-				    child_count * sizeof (int32_t));
-	}
+        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
+                sh->completion_cbk (frame, this);
+        } else {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "proceeding to metadata check on %s",
+                        local->loc.path);
+                afr_self_heal_metadata (frame, this);
+        }
 
-	return 0;
+        return 0;
 }
 
 
+static int
+afr_sh_missing_entries_finish (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+
+        local = frame->local;
+        int_lock = &local->internal_lock;
+
+        int_lock->lock_cbk = afr_sh_missing_entries_done;
+        afr_unlock (frame, this);
+
+        return 0;
+}
+
 int
-afr_sh_has_metadata_pending (dict_t *xattr, int child_count, xlator_t *this)
+afr_sh_common_create (afr_self_heal_t *sh, unsigned int child_count)
 {
-	afr_private_t *priv = NULL;
-	int32_t       *pending = NULL;
-	void          *tmp_pending = NULL; /* This is required to remove 'type-punned' warnings from gcc */
+        int     ret = -ENOMEM;
+        sh->buf = GF_CALLOC (child_count, sizeof (*sh->buf),
+                             gf_afr_mt_iatt);
+        if (!sh->buf)
+                goto out;
+        sh->parentbufs = GF_CALLOC (child_count, sizeof (*sh->parentbufs),
+                                    gf_afr_mt_iatt);
+        if (!sh->parentbufs)
+                goto out;
+        sh->child_errno = GF_CALLOC (child_count, sizeof (*sh->child_errno),
+                                     gf_afr_mt_int);
+        if (!sh->child_errno)
+                goto out;
+        sh->success_children = afr_children_create (child_count);
+        if (!sh->success_children)
+                goto out;
+        sh->fresh_children = afr_children_create (child_count);
+        if (!sh->fresh_children)
+                goto out;
+        sh->xattr = GF_CALLOC (child_count, sizeof (*sh->xattr),
+                               gf_afr_mt_dict_t);
+        if (!sh->xattr)
+                goto out;
+        ret = 0;
+out:
+        return ret;
+}
 
-	int           ret = -1;
-	int            i  = 0;
+void
+afr_sh_common_lookup_resp_handler (call_frame_t *frame, void *cookie,
+                                   xlator_t *this,
+                                   int32_t op_ret, int32_t op_errno,
+                                   inode_t *inode, struct iatt *buf,
+                                   dict_t *xattr, struct iatt *postparent,
+                                   loc_t *loc)
+{
+        int              child_index = 0;
+        afr_local_t     *local = NULL;
+        afr_private_t   *priv = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        priv = this->private;
+        sh   = &local->self_heal;
+        child_index = (long) cookie;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == 0) {
+                        sh->buf[child_index] = *buf;
+                        sh->parentbufs[child_index] = *postparent;
+                        sh->success_children[sh->success_count] = child_index;
+                        sh->success_count++;
+                        sh->xattr[child_index] = dict_ref (xattr);
+                } else {
+                        gf_log (this->name, GF_LOG_DEBUG, "path %s on subvolume"
+                                " %s => -1 (%s)", loc->path,
+                                priv->children[child_index]->name,
+                                strerror (op_errno));
+                        local->self_heal.child_errno[child_index] = op_errno;
+                }
+        }
+        UNLOCK (&frame->lock);
+        return;
+}
 
-	priv = this->private;
+gf_boolean_t
+afr_valid_ia_type (ia_type_t ia_type)
+{
+        switch (ia_type) {
+        case IA_IFSOCK:
+        case IA_IFREG:
+        case IA_IFBLK:
+        case IA_IFCHR:
+        case IA_IFIFO:
+        case IA_IFLNK:
+        case IA_IFDIR:
+                return _gf_true;
+        default:
+                return _gf_false;
+        }
+        return _gf_false;
+}
 
-	ret = dict_get_ptr (xattr, AFR_METADATA_PENDING, &tmp_pending);
+int
+afr_impunge_frame_create (call_frame_t *frame, xlator_t *this,
+                          int active_source, call_frame_t **impunge_frame)
+{
+        afr_local_t     *local         = NULL;
+        afr_local_t     *impunge_local = NULL;
+        afr_self_heal_t *impunge_sh    = NULL;
+        int32_t         op_errno       = 0;
+        afr_private_t   *priv          = NULL;
+        int             ret            = 0;
+        call_frame_t    *new_frame     = NULL;
+
+        op_errno = ENOMEM;
+        priv = this->private;
+        new_frame = copy_frame (frame);
+        if (!new_frame) {
+                goto out;
+        }
 
-	if (ret != 0)
-		return 0;
+        AFR_LOCAL_ALLOC_OR_GOTO (impunge_local, out);
+
+        local = frame->local;
+        new_frame->local = impunge_local;
+        impunge_sh = &impunge_local->self_heal;
+        impunge_sh->sh_frame = frame;
+        impunge_sh->active_source = active_source;
+        impunge_local->child_up  = memdup (local->child_up,
+                                           sizeof (*local->child_up) *
+                                           priv->child_count);
+        if (!impunge_local->child_up)
+                goto out;
 
-	pending = tmp_pending;
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == child_count)
-			continue;
-		if (pending[i])
-			return 1;
-	}
+        impunge_local->pending = afr_matrix_create (priv->child_count,
+                                                    AFR_NUM_CHANGE_LOGS);
+        if (!impunge_local->pending)
+                goto out;
 
-	return 0;
+        ret = afr_sh_common_create (impunge_sh, priv->child_count);
+        if (ret) {
+                op_errno = -ret;
+                goto out;
+        }
+        op_errno = 0;
+        *impunge_frame = new_frame;
+out:
+        if (op_errno && new_frame)
+                AFR_STACK_DESTROY (new_frame);
+        return -op_errno;
 }
 
+void
+afr_sh_missing_entry_call_impunge_recreate (call_frame_t *frame, xlator_t *this,
+                                            struct iatt *buf,
+                                            struct iatt *postparent,
+                                            afr_impunge_done_cbk_t impunge_done)
+{
+        call_frame_t    *impunge_frame = NULL;
+        afr_local_t     *local = NULL;
+        afr_local_t     *impunge_local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_self_heal_t *impunge_sh = NULL;
+        int             ret = 0;
+        unsigned int    enoent_count = 0;
+        afr_private_t   *priv = NULL;
+        int             i = 0;
+        int32_t         op_errno = 0;
+
+        local = frame->local;
+        sh    = &local->self_heal;
+        priv  = this->private;
+
+        enoent_count = afr_errno_count (NULL, sh->child_errno,
+                                        priv->child_count, ENOENT);
+        if (!enoent_count) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "no missing files - %s. proceeding to metadata check",
+                        local->loc.path);
+                goto out;
+        }
+        sh->impunge_done = impunge_done;
+        ret = afr_impunge_frame_create (frame, this, sh->source, &impunge_frame);
+        if (ret)
+                goto out;
+        impunge_local = impunge_frame->local;
+        impunge_sh    = &impunge_local->self_heal;
+        loc_copy (&impunge_local->loc, &local->loc);
+        ret = afr_build_parent_loc (&impunge_sh->parent_loc,
+                                    &impunge_local->loc, &op_errno);
+        if (ret) {
+                ret = -op_errno;
+                goto out;
+        }
+        impunge_local->call_count = enoent_count;
+        impunge_sh->entrybuf = sh->buf[sh->source];
+        impunge_sh->parentbuf = sh->parentbufs[sh->source];
+        for (i = 0; i < priv->child_count; i++) {
+                if (!impunge_local->child_up[i]) {
+                        impunge_sh->child_errno[i] = ENOTCONN;
+                        continue;
+                }
+                if (sh->child_errno[i] != ENOENT) {
+                        impunge_sh->child_errno[i] = EEXIST;
+                        continue;
+                }
+        }
+        for (i = 0; i < priv->child_count; i++) {
+                if (sh->child_errno[i] != ENOENT)
+                        continue;
+                afr_sh_entry_impunge_create (impunge_frame, this, i);
+                enoent_count--;
+        }
+        GF_ASSERT (!enoent_count);
+        return;
+out:
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "impunge of %s failed, "
+                        "reason: %s", local->loc.path, strerror (-ret));
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        }
+        afr_sh_missing_entries_finish (frame, this);
+}
 
 int
-afr_sh_has_data_pending (dict_t *xattr, int child_count, xlator_t *this)
+afr_sh_create_entry_cbk (call_frame_t *frame, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno)
 {
-	afr_private_t *priv = NULL;
-	int32_t       *pending = NULL;
-	void          *tmp_pending = NULL; /* This is required to remove 'type-punned' warnings from gcc */
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        if (op_ret < 0)
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        afr_sh_missing_entries_finish (frame, this);
+        return 0;
+}
 
-	int          ret = -1;
-	int            i = 0;
+static int
+sh_missing_entries_create (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        int              type = 0;
+        struct iatt     *buf = NULL;
+        struct iatt     *postparent = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+
+        buf = &sh->buf[sh->source];
+        postparent = &sh->parentbufs[sh->source];
+
+        type = buf->ia_type;
+        if (!afr_valid_ia_type (type)) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: unknown file type: 0%o", local->loc.path, type);
+                afr_set_local_for_unhealable (local);
+                afr_sh_missing_entries_finish (frame, this);
+                goto out;
+        }
 
-	priv = this->private;
+        afr_sh_missing_entry_call_impunge_recreate (frame, this,
+                                                    buf, postparent,
+                                                    afr_sh_create_entry_cbk);
+out:
+        return 0;
+}
 
-	ret = dict_get_ptr (xattr, AFR_DATA_PENDING, &tmp_pending);
+void
+afr_sh_missing_entries_lookup_done (call_frame_t *frame, xlator_t *this,
+                                    int32_t op_ret, int32_t op_errno)
+{
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        ia_type_t       ia_type = IA_INVAL;
+        int32_t         nsources = 0;
+        loc_t           *loc = NULL;
+        int32_t         subvol_status = 0;
+        afr_transaction_type txn_type = AFR_DATA_TRANSACTION;
+        gf_boolean_t    split_brain = _gf_false;
+        int             read_child = -1;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+        loc = &local->loc;
+
+        if (op_ret < 0) {
+                if (op_errno == EIO) {
+                        afr_set_local_for_unhealable (local);
+                }
+                // EIO can happen if finding the fresh parent dir failed
+                goto out;
+        }
 
-	if (ret != 0)
-		return 0;
+        //now No chance for the ia_type to conflict
+        ia_type = sh->buf[sh->success_children[0]].ia_type;
+        txn_type = afr_transaction_type_get (ia_type);
+        nsources = afr_build_sources (this, sh->xattr, sh->buf,
+                                      sh->pending_matrix, sh->sources,
+                                      sh->success_children, txn_type,
+                                      &subvol_status, _gf_false);
+        if (nsources < 0) {
+                gf_log (this->name, GF_LOG_INFO, "No sources for dir of %s,"
+                        " in missing entry self-heal, continuing with the rest"
+                        " of the self-heals", local->loc.path);
+                if (subvol_status & SPLIT_BRAIN) {
+                        split_brain = _gf_true;
+                        switch (txn_type) {
+                        case AFR_DATA_TRANSACTION:
+                                nsources = 1;
+                                sh->sources[sh->success_children[0]] = 1;
+                                break;
+                        case AFR_ENTRY_TRANSACTION:
+                                read_child = afr_get_no_xattr_dir_read_child
+                                                          (this,
+                                                           sh->success_children,
+                                                           sh->buf);
+                                sh->sources[read_child] = 1;
+                                nsources = 1;
+                                break;
+                        default:
+                                op_errno = EIO;
+                                goto out;
+                        }
+                } else {
+                        op_errno = EIO;
+                        goto out;
+                }
+        }
 
-	pending = tmp_pending;
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == child_count)
-			continue;
-		if (pending[i])
-			return 1;
-	}
+        afr_get_fresh_children (sh->success_children, sh->sources,
+                                sh->fresh_children, priv->child_count);
+        sh->source = sh->fresh_children[0];
+        if (sh->source == -1) {
+                gf_log (this->name, GF_LOG_DEBUG, "No active sources found.");
+                op_errno = EIO;
+                goto out;
+        }
 
-	return 0;
+        if (sh->gfid_sh_success_cbk)
+                sh->gfid_sh_success_cbk (frame, this);
+        sh->type = sh->buf[sh->source].ia_type;
+        if (uuid_is_null (loc->inode->gfid))
+                uuid_copy (loc->gfid, sh->buf[sh->source].ia_gfid);
+        if (split_brain) {
+                afr_sh_missing_entries_finish (frame, this);
+        } else {
+                sh_missing_entries_create (frame, this);
+        }
+        return;
+out:
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        afr_sh_set_error (sh, op_errno);
+        afr_sh_missing_entries_finish (frame, this);
+        return;
 }
 
-
-int
-afr_sh_has_entry_pending (dict_t *xattr, int child_count, xlator_t *this)
+static int
+afr_sh_common_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, inode_t *inode,
+                          struct iatt *buf, dict_t *xattr,
+                          struct iatt *postparent)
 {
-	afr_private_t *priv = NULL;
-	int32_t       *pending = NULL;
-	void          *tmp_pending = NULL; /* This is required to remove 'type-punned' warnings from gcc */
-	
-	int          ret = -1;
-	int            i = 0;
+        int                     call_count = 0;
+        afr_local_t             *local = NULL;
+        afr_self_heal_t         *sh    = NULL;
+        afr_private_t           *priv  = NULL;
 
-	priv = this->private;
+        local = frame->local;
+        sh    = &local->self_heal;
+        priv  = this->private;
 
-	ret = dict_get_ptr (xattr, AFR_ENTRY_PENDING, &tmp_pending);
+        afr_sh_common_lookup_resp_handler (frame, cookie, this, op_ret,
+                                           op_errno, inode, buf, xattr,
+                                           postparent, &sh->lookup_loc);
+        call_count = afr_frame_return (frame);
 
-	if (ret != 0)
-		return 0;
+        if (call_count)
+                goto out;
+        op_ret = -1;
+        if (!sh->success_count) {
+                op_errno = afr_resultant_errno_get (NULL, sh->child_errno,
+                                                    priv->child_count);
+                gf_log (this->name, GF_LOG_ERROR, "Failed to lookup %s, "
+                        "reason %s", sh->lookup_loc.path,
+                        strerror (op_errno));
+                goto done;
+        }
 
-	pending = tmp_pending;
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == child_count)
-			continue;
-		if (pending[i])
-			return 1;
-	}
+        if ((sh->lookup_flags & AFR_LOOKUP_FAIL_CONFLICTS) &&
+            (afr_conflicting_iattrs (sh->buf, sh->success_children,
+                                     priv->child_count,
+                                     sh->lookup_loc.path, this->name))) {
+                op_errno = EIO;
+                gf_log (this->name, GF_LOG_ERROR, "Conflicting entries "
+                        "for %s", sh->lookup_loc.path);
+                goto done;
+        }
 
-	return 0;
+        if ((sh->lookup_flags & AFR_LOOKUP_FAIL_MISSING_GFIDS) &&
+            (afr_gfid_missing_count (this->name, sh->success_children,
+                                     sh->buf, priv->child_count,
+                                     sh->lookup_loc.path))) {
+                op_errno = ENODATA;
+                gf_log (this->name, GF_LOG_ERROR, "Missing Gfids "
+                        "for %s", sh->lookup_loc.path);
+                goto done;
+        }
+        op_ret = 0;
+
+done:
+        sh->lookup_done (frame, this, op_ret, op_errno);
+out:
+        return 0;
 }
 
+int
+afr_sh_remove_entry_cbk (call_frame_t *frame, xlator_t *this, int child,
+                         int32_t op_ret, int32_t op_errno)
+{
+        int             call_count = 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+
+        GF_ASSERT (sh->post_remove_call);
+        if ((op_ret == -1) && (op_errno != ENOENT)) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "purge entry %s failed, on child %d reason, %s",
+                        local->loc.path, child, strerror (op_errno));
+                LOCK (&frame->lock);
+                {
+                        afr_sh_set_error (sh, EIO);
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                }
+                UNLOCK (&frame->lock);
+        }
+        call_count = afr_frame_return (frame);
+        if (call_count == 0)
+                sh->post_remove_call (frame, this);
+        return 0;
+}
 
+void
+afr_sh_call_entry_expunge_remove (call_frame_t *frame, xlator_t *this,
+                                  int child_index, struct iatt *buf,
+                                  struct iatt *parentbuf,
+                                  afr_expunge_done_cbk_t expunge_done)
+{
+        call_frame_t    *expunge_frame = NULL;
+        afr_local_t     *local = NULL;
+        afr_local_t     *expunge_local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_self_heal_t *expunge_sh = NULL;
+        int32_t         op_errno = 0;
+        int             ret = 0;
+
+        expunge_frame = copy_frame (frame);
+        if (!expunge_frame) {
+                goto out;
+        }
 
-/**
- * is_matrix_zero - return true if pending matrix is all zeroes
- */
+        AFR_LOCAL_ALLOC_OR_GOTO (expunge_local, out);
+
+        local = frame->local;
+        sh = &local->self_heal;
+        expunge_frame->local = expunge_local;
+        expunge_sh = &expunge_local->self_heal;
+        expunge_sh->sh_frame = frame;
+        loc_copy (&expunge_local->loc, &local->loc);
+        ret = afr_build_parent_loc (&expunge_sh->parent_loc,
+                                    &expunge_local->loc, &op_errno);
+        if (ret) {
+                ret = -op_errno;
+                goto out;
+        }
+        sh->expunge_done = expunge_done;
+        afr_sh_entry_expunge_remove (expunge_frame, this, child_index, buf,
+                                     parentbuf);
+        return;
+out:
+        gf_log (this->name, GF_LOG_ERROR, "Expunge of %s failed, reason: %s",
+                local->loc.path, strerror (op_errno));
+        expunge_done (frame, this, child_index, -1, op_errno);
+}
 
-int
-afr_sh_is_matrix_zero (int32_t *pending_matrix[], int child_count)
+void
+afr_sh_remove_stale_lookup_info (afr_self_heal_t *sh, int32_t *success_children,
+                                 int32_t *fresh_children,
+                                 unsigned int child_count)
 {
-	int i, j;
+        int     i = 0;
 
-	for (i = 0; i < child_count; i++) 
-		for (j = 0; j < child_count; j++) 
-			if (pending_matrix[i][j]) 
-				return 0;
-	return 1;
+        for (i = 0; i < child_count; i++) {
+                if (afr_is_child_present (success_children, child_count, i) &&
+                    !afr_is_child_present (fresh_children, child_count, i)) {
+                        sh->child_errno[i] = ENOENT;
+                        GF_ASSERT (sh->xattr[i]);
+                        dict_unref (sh->xattr[i]);
+                        sh->xattr[i] = NULL;
+                }
+        }
 }
 
-
 int
-afr_sh_missing_entries_done (call_frame_t *frame, xlator_t *this)
+afr_sh_purge_stale_entries_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-//	memset (sh->child_errno, 0, sizeof (int) * priv->child_count);
-	memset (sh->buf, 0, sizeof (struct stat) * priv->child_count);
-	
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->xattr[i])
-			dict_unref (sh->xattr[i]);
-		sh->xattr[i] = NULL;
-	}
+        afr_local_t         *local    = NULL;
+        afr_self_heal_t     *sh       = NULL;
+        afr_private_t       *priv     = NULL;
 
-	if (local->govinda_gOvinda) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"aborting selfheal of %s",
-			local->loc.path);
-		sh->completion_cbk (frame, this);
-	} else {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"proceeding to metadata check on %s",
-			local->loc.path);
-		afr_self_heal_metadata (frame, this);
-	}
+        local    = frame->local;
+        sh       = &local->self_heal;
+        priv     = this->private;
 
-	return 0;
+        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
+                afr_sh_missing_entries_finish (frame, this);
+        } else {
+                if (afr_gfid_missing_count (this->name, sh->fresh_children,
+                                            sh->buf, priv->child_count,
+                                            local->loc.path)) {
+                        afr_sh_common_lookup (frame, this, &local->loc,
+                                              afr_sh_missing_entries_lookup_done,
+                                              sh->sh_gfid_req,
+                                              AFR_LOOKUP_FAIL_CONFLICTS|
+                                              AFR_LOOKUP_FAIL_MISSING_GFIDS,
+                                              NULL);
+                } else {
+                        //No need to set gfid so goto missing entries lookup done
+                        //Behave as if you have done the lookup
+                        afr_sh_remove_stale_lookup_info (sh,
+                                                         sh->success_children,
+                                                         sh->fresh_children,
+                                                         priv->child_count);
+                        afr_children_copy (sh->success_children,
+                                           sh->fresh_children,
+                                           priv->child_count);
+                        afr_sh_missing_entries_lookup_done (frame, this, 0, 0);
+                }
+        }
+        return 0;
 }
 
+gf_boolean_t
+afr_sh_purge_entry_condition (afr_local_t *local, afr_private_t *priv,
+                              int child)
+{
+        afr_self_heal_t *sh = NULL;
 
-int
-sh_missing_entries_unlck_cbk (call_frame_t *frame, void *cookie,
-			      xlator_t *this,
-			      int32_t op_ret, int32_t op_errno)
+        sh = &local->self_heal;
+
+        if (local->child_up[child] &&
+            (!afr_is_child_present (sh->fresh_parent_dirs, priv->child_count,
+                                    child))
+            && (sh->child_errno[child] != ENOENT))
+                return _gf_true;
+
+        return _gf_false;
+}
+
+gf_boolean_t
+afr_sh_purge_stale_entry_condition (afr_local_t *local, afr_private_t *priv,
+                                    int child)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
+        afr_self_heal_t *sh = NULL;
 
+        sh = &local->self_heal;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        if (local->child_up[child] &&
+            (!afr_is_child_present (sh->fresh_children, priv->child_count,
+                                    child))
+             && (sh->child_errno[child] != ENOENT))
+                return _gf_true;
 
-	LOCK (&frame->lock);
-	{
-	}
-	UNLOCK (&frame->lock);
+        return _gf_false;
+}
 
-	call_count = afr_frame_return (frame);
+void
+afr_sh_purge_entry_common (call_frame_t *frame, xlator_t *this,
+                           gf_boolean_t purge_condition (afr_local_t *local,
+                                                         afr_private_t *priv,
+                                                         int child))
+{
+        afr_local_t     *local = NULL;
+        afr_private_t   *priv = NULL;
+        afr_self_heal_t *sh = NULL;
+        int             i = 0;
+        int             call_count = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (purge_condition (local, priv, i))
+                        call_count++;
+        }
 
-	if (call_count == 0) {
-		afr_sh_missing_entries_done (frame, this);
-	}
+        if (call_count == 0) {
+                sh->post_remove_call (frame, this);
+                goto out;
+        }
 
-	return 0;
+        local->call_count = call_count;
+        for (i = 0; i < priv->child_count; i++) {
+                if (!purge_condition (local, priv, i))
+                        continue;
+                gf_log (this->name, GF_LOG_INFO, "purging the stale entry %s "
+                        "on %s", local->loc.path, priv->children[i]->name);
+                afr_sh_call_entry_expunge_remove (frame, this,
+                                                  (long) i, &sh->buf[i],
+                                                  &sh->parentbufs[i],
+                                                  afr_sh_remove_entry_cbk);
+        }
+out:
+        return;
 }
-			      
 
-static int
-sh_missing_entries_finish (call_frame_t *frame, xlator_t *this)
+void
+afr_sh_purge_entry (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t      *priv = NULL;
-	afr_local_t        *local = NULL;
-	int                 i = 0;
-	int                 call_count = 0;
-	afr_self_heal_t    *sh = NULL;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
 
+        local = frame->local;
+        sh = &local->self_heal;
+        sh->post_remove_call = afr_sh_missing_entries_finish;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        afr_sh_purge_entry_common (frame, this, afr_sh_purge_entry_condition);
+}
 
-	call_count = local->child_count;
+void
+afr_sh_purge_stale_entry (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int             i = 0;
 
-	local->call_count = call_count;
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"unlocking %"PRId64"/%s on subvolume %s",
-				sh->parent_loc.inode->ino, local->loc.name,
-				priv->children[i]->name);
+        sh->post_remove_call = afr_sh_purge_stale_entries_done;
 
-			STACK_WIND (frame, sh_missing_entries_unlck_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->entrylk,
-                                    this->name,
-				    &sh->parent_loc, local->loc.name,
-				    ENTRYLK_UNLOCK, ENTRYLK_WRLCK);
+        for (i = 0; i < priv->child_count; i++) {
+                if (afr_is_child_present (sh->fresh_children,
+                                          priv->child_count, i))
+                        continue;
 
-			if (!--call_count)
-				break;
-		}
-	}
-	return 0;
+                if ((!local->child_up[i]) || sh->child_errno[i] != 0)
+                        continue;
+
+                GF_ASSERT (!uuid_is_null (sh->entrybuf.ia_gfid) ||
+                           uuid_is_null (sh->buf[i].ia_gfid));
+
+                if ((sh->entrybuf.ia_type != sh->buf[i].ia_type) ||
+                    (uuid_compare (sh->buf[i].ia_gfid,
+                                   sh->entrybuf.ia_gfid)))
+                        continue;
+
+                afr_children_add_child (sh->fresh_children, i,
+                                        priv->child_count);
+
+        }
+        afr_sh_purge_entry_common (frame, this,
+                                   afr_sh_purge_stale_entry_condition);
 }
 
+void
+afr_sh_save_child_iatts_from_policy (int32_t *children, struct iatt *bufs,
+                                     struct iatt *save,
+                                     unsigned int child_count)
+{
+        int             i = 0;
+        int             child = 0;
+        gf_boolean_t    saved = _gf_false;
 
-static int
-sh_destroy_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		int32_t op_ret, int op_errno, struct stat *stbuf)
+        GF_ASSERT (save);
+        //if iatt buf with gfid exists sets it
+        for (i = 0; i < child_count; i++) {
+                child = children[i];
+                if (child == -1)
+                        break;
+                *save = bufs[child];
+                saved = _gf_true;
+                if (!uuid_is_null (save->ia_gfid))
+                        break;
+        }
+        GF_ASSERT (saved);
+}
+
+void
+afr_get_children_of_fresh_parent_dirs (afr_self_heal_t *sh,
+                                       unsigned int child_count)
 {
-	STACK_DESTROY (frame->root);
-	return 0;
+        afr_children_intersection_get (sh->success_children,
+                                       sh->fresh_parent_dirs,
+                                       sh->sources, child_count);
+        afr_get_fresh_children (sh->success_children, sh->sources,
+                                sh->fresh_children, child_count);
+        memset (sh->sources, 0, sizeof (*sh->sources) * child_count);
 }
 
+void
+afr_sh_children_lookup_done (call_frame_t *frame, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno)
+{
+        afr_local_t      *local = NULL;
+        afr_self_heal_t  *sh = NULL;
+        afr_private_t    *priv = NULL;
+        int32_t          fresh_child_enoents = 0;
+        int32_t          fresh_parent_count = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        if (op_ret < 0)
+                goto fail;
+        afr_get_children_of_fresh_parent_dirs (sh, priv->child_count);
+        fresh_parent_count = afr_get_children_count (sh->fresh_parent_dirs,
+                                                     priv->child_count);
+        //we need the enoent count of the subvols present in fresh_parent_dirs
+        fresh_child_enoents = afr_errno_count (sh->fresh_parent_dirs,
+                                               sh->child_errno,
+                                               priv->child_count, ENOENT);
+        if (fresh_child_enoents == fresh_parent_count) {
+                afr_sh_set_error (sh, ENOENT);
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                afr_sh_purge_entry (frame, this);
+        } else if (!afr_conflicting_iattrs (sh->buf, sh->fresh_children,
+                                            priv->child_count, local->loc.path,
+                                            this->name)) {
+                afr_sh_save_child_iatts_from_policy (sh->fresh_children,
+                                                     sh->buf, &sh->entrybuf,
+                                                     priv->child_count);
+                afr_update_gfid_from_iatts (sh->sh_gfid_req, sh->buf,
+                                            sh->fresh_children,
+                                            priv->child_count);
+                afr_sh_purge_stale_entry (frame, this);
+        } else {
+                op_errno = EIO;
+                afr_set_local_for_unhealable (local);
+                goto fail;
+        }
 
-static int
-sh_missing_entries_newentry_cbk (call_frame_t *frame, void *cookie,
-				 xlator_t *this,
-				 int32_t op_ret, int32_t op_errno,
-				 inode_t *inode, struct stat *stbuf)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	call_frame_t    *chown_frame = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-	struct stat     *buf = NULL;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	buf = &sh->buf[sh->source];
-	child_index = (long) cookie;
-
-	if (op_ret == 0) {
-		chown_frame = copy_frame (frame);
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"chown %s to %d %d on subvolume %s",
-			local->loc.path, buf->st_uid, buf->st_gid,
-			priv->children[child_index]->name);
-
-		STACK_WIND (chown_frame, sh_destroy_cbk,
-			    priv->children[child_index],
-			    priv->children[child_index]->fops->chown,
-			    &local->loc,
-			    buf->st_uid, buf->st_gid);
-	}
+        return;
 
-	LOCK (&frame->lock);
-	{
-	}
-	UNLOCK (&frame->lock);
+fail:
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        afr_sh_set_error (sh, op_errno);
+        afr_sh_missing_entries_finish (frame, this);
+        return;
+}
 
-	call_count = afr_frame_return (frame);
+static void
+afr_sh_find_fresh_parents (call_frame_t *frame, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno)
+{
+        afr_self_heal_t *sh  = NULL;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local = NULL;
+        int             enoent_count = 0;
+        int             nsources = 0;
+        int             source  = -1;
+        int32_t         subvol_status = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        if (op_ret < 0)
+                goto out;
+        enoent_count = afr_errno_count (NULL, sh->child_errno,
+                                        priv->child_count, ENOENT);
+        if (enoent_count > 0) {
+                gf_log (this->name, GF_LOG_INFO, "Parent dir missing for %s,"
+                        " in missing entry self-heal, aborting missing-entry "
+                        "self-heal",
+                        local->loc.path);
+                afr_sh_missing_entries_finish (frame, this);
+                return;
+        }
 
-	if (call_count == 0) {
-		sh_missing_entries_finish (frame, this);
-	}
+        nsources = afr_build_sources (this, sh->xattr, sh->buf,
+                                      sh->pending_matrix, sh->sources,
+                                      sh->success_children,
+                                      AFR_ENTRY_TRANSACTION, &subvol_status,
+                                      _gf_true);
+        if ((subvol_status & ALL_FOOLS) ||
+            (subvol_status & SPLIT_BRAIN)) {
+                gf_log (this->name, GF_LOG_INFO, "%s: Performing conservative "
+                        "merge", sh->parent_loc.path);
+                afr_mark_success_children_sources (sh->sources,
+                                                   sh->success_children,
+                                                   priv->child_count);
+        } else if (nsources < 0) {
+                gf_log (this->name, GF_LOG_ERROR, "No sources for dir "
+                        "of %s, in missing entry self-heal, aborting "
+                        "self-heal", local->loc.path);
+                op_errno = EIO;
+                goto out;
+        }
 
-	return 0;
-}
+        source = afr_sh_select_source (sh->sources, priv->child_count);
+        if (source == -1) {
+                GF_ASSERT (0);
+                gf_log (this->name, GF_LOG_DEBUG, "No active sources found.");
+                op_errno = EIO;
+                goto out;
+        }
+        afr_get_fresh_children (sh->success_children, sh->sources,
+                                sh->fresh_parent_dirs, priv->child_count);
+        afr_sh_common_lookup (frame, this, &local->loc,
+                              afr_sh_children_lookup_done, NULL, 0,
+                              NULL);
+        return;
 
+out:
+        afr_sh_set_error (sh, op_errno);
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+	afr_sh_missing_entries_finish (frame, this);
+        return;
+}
 
-static int
-sh_missing_entries_mknod (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-	int              enoent_count = 0;
-	int              call_count = 0;
-	mode_t           st_mode = 0;
-	dev_t            st_dev = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++)
-		if (sh->child_errno[i] == ENOENT)
-			enoent_count++;
-
-	call_count = enoent_count;
-	local->call_count = call_count;
-
-	st_mode = sh->buf[sh->source].st_mode;
-	st_dev  = sh->buf[sh->source].st_dev;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"mknod %s mode 0%o on %d subvolumes",
-		local->loc.path, st_mode, enoent_count);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->child_errno[i] == ENOENT) {
-			STACK_WIND_COOKIE (frame,
-					   sh_missing_entries_newentry_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->mknod,
-					   &local->loc, st_mode, st_dev);
-			if (!--call_count)
-				break;
-		}
-	}
+void
+afr_sh_common_reset (afr_self_heal_t *sh, unsigned int child_count)
+{
+        int             i = 0;
 
-	return 0;
+        for (i = 0; i < child_count; i++) {
+                memset (&sh->buf[i], 0, sizeof (sh->buf[i]));
+                memset (&sh->parentbufs[i], 0, sizeof (sh->parentbufs[i]));
+                sh->child_errno[i] = 0;
+        }
+        memset (&sh->parentbuf, 0, sizeof (sh->parentbuf));
+        sh->success_count = 0;
+        afr_reset_children (sh->success_children, child_count);
+        afr_reset_children (sh->fresh_children, child_count);
+        afr_reset_xattr (sh->xattr, child_count);
+        loc_wipe (&sh->lookup_loc);
 }
 
+/* afr self-heal state will be lost if this call is made
+ * please check the afr_sh_common_reset that is called in this function
+ */
+int
+afr_sh_common_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      afr_lookup_done_cbk_t lookup_done , uuid_t gfid,
+                      int32_t flags, dict_t *xdata)
+{
+        afr_local_t    *local = NULL;
+        int             i = 0;
+        int             call_count = 0;
+        afr_private_t  *priv = NULL;
+        dict_t         *xattr_req = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        priv  = this->private;
+        sh    = &local->self_heal;
+
+        call_count = afr_up_children_count (local->child_up, priv->child_count);
+
+        local->call_count = call_count;
+
+        xattr_req = dict_new();
+
+        if (xattr_req) {
+                afr_xattr_req_prepare (this, xattr_req, loc->path);
+                if (gfid) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "looking up %s with gfid: %s",
+                                loc->path, uuid_utoa (gfid));
+                        GF_ASSERT (!uuid_is_null (gfid));
+                        afr_set_dict_gfid (xattr_req, gfid);
+                }
+        }
+
+        afr_sh_common_reset (sh, priv->child_count);
+        sh->lookup_done = lookup_done;
+        loc_copy (&sh->lookup_loc, loc);
+        sh->lookup_flags = flags;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "looking up %s on subvolume %s",
+                                loc->path, priv->children[i]->name);
+
+                        STACK_WIND_COOKIE (frame,
+                                           afr_sh_common_lookup_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->lookup,
+                                           loc, xattr_req);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
 
-static int
-sh_missing_entries_mkdir (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-	int              enoent_count = 0;
-	int              call_count = 0;
-	mode_t           st_mode = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++)
-		if (sh->child_errno[i] == ENOENT)
-			enoent_count++;
-
-	call_count = enoent_count;
-	local->call_count = call_count;
-
-	st_mode = sh->buf[sh->source].st_mode;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"mkdir %s mode 0%o on %d subvolumes",
-		local->loc.path, st_mode, enoent_count);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->child_errno[i] == ENOENT) {
-			STACK_WIND_COOKIE (frame,
-					   sh_missing_entries_newentry_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->mkdir,
-					   &local->loc, st_mode);
-			if (!--call_count)
-				break;
-		}
-	}
+        if (xattr_req)
+                dict_unref (xattr_req);
 
-	return 0;
+        return 0;
 }
 
 
-static int
-sh_missing_entries_symlink (call_frame_t *frame, xlator_t *this,
-			    const char *link)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-	int              enoent_count = 0;
-	int              call_count = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++)
-		if (sh->child_errno[i] == ENOENT)
-			enoent_count++;
-
-	call_count = enoent_count;
-	local->call_count = call_count;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"symlink %s -> %s on %d subvolumes",
-		local->loc.path, link, enoent_count);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->child_errno[i] == ENOENT) {
-			STACK_WIND_COOKIE (frame,
-					   sh_missing_entries_newentry_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->symlink,
-					   link, &local->loc);
-			if (!--call_count)
-				break;
-		}
-	}
 
-	return 0;
+int
+afr_sh_post_nb_entrylk_missing_entry_sh_cbk (call_frame_t *frame,
+                                             xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+        afr_self_heal_t     *sh       = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        sh       = &local->self_heal;
+
+        if (int_lock->lock_op_ret < 0) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Non blocking entrylks failed.");
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                afr_sh_missing_entries_done (frame, this);
+        } else {
+
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Non blocking entrylks done. Proceeding to FOP");
+                afr_sh_common_lookup (frame, this, &sh->parent_loc,
+                                      afr_sh_find_fresh_parents,
+                                      NULL, AFR_LOOKUP_FAIL_CONFLICTS,
+                                      NULL);
+        }
+
+        return 0;
 }
 
+int
+afr_sh_entrylk (call_frame_t *frame, xlator_t *this, loc_t *loc,
+                char *base_name, afr_lock_cbk_t lock_cbk)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+        afr_private_t       *priv     = NULL;
+
+        priv     = this->private;
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
+        int_lock->selfheal_lk_type    = AFR_ENTRY_SELF_HEAL_LK;
+
+        afr_set_lock_number (frame, this);
+
+        int_lock->lk_basename = base_name;
+        int_lock->lk_loc      = loc;
+        int_lock->lock_cbk    = lock_cbk;
+        int_lock->domain      = this->name;
+
+        int_lock->lockee_count = 0;
+        afr_init_entry_lockee (&int_lock->lockee[0], local, loc,
+                               base_name, priv->child_count);
+        int_lock->lockee_count++;
+        afr_nonblocking_entrylk (frame, this);
+
+        return 0;
+}
 
 static int
-sh_missing_entries_readlink_cbk (call_frame_t *frame, void *cookie,
-				 xlator_t *this,
-				 int32_t op_ret, int32_t op_errno,
-				 const char *link)
+afr_self_heal_parent_entrylk (call_frame_t *frame, xlator_t *this,
+                              afr_lock_cbk_t lock_cbk)
 {
-	if (op_ret > 0)
-		sh_missing_entries_symlink (frame, this, link);
-	else
-		sh_missing_entries_finish (frame, this);
+        afr_local_t         *local    = NULL;
+        afr_self_heal_t     *sh       = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+        int                 ret       = -1;
+        int32_t             op_errno  = 0;
 
-	return 0;
-}
+        local    = frame->local;
+        sh       = &local->self_heal;
 
+        gf_log (this->name, GF_LOG_TRACE,
+                "attempting to recreate missing entries for path=%s",
+                local->loc.path);
+
+        ret = afr_build_parent_loc (&sh->parent_loc, &local->loc, &op_errno);
+        if (ret)
+                goto out;
+
+        afr_sh_entrylk (frame, this, &sh->parent_loc, NULL,
+                        lock_cbk);
+        return 0;
+out:
+        int_lock = &local->internal_lock;
+        int_lock->lock_op_ret = -1;
+        lock_cbk (frame, this);
+        return 0;
+}
 
 static int
-sh_missing_entries_readlink (call_frame_t *frame, xlator_t *this)
+afr_self_heal_missing_entries (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
 
+        local = frame->local;
+        sh = &local->self_heal;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        sh->sh_type_in_action  = AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY;
 
-	STACK_WIND (frame, sh_missing_entries_readlink_cbk,
-		    priv->children[sh->source],
-		    priv->children[sh->source]->fops->readlink,
-		    &local->loc, 4096);
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
 
-	return 0;
+        afr_self_heal_parent_entrylk (frame, this,
+                                      afr_sh_post_nb_entrylk_missing_entry_sh_cbk);
+        return 0;
 }
 
-
-static int
-sh_missing_entries_create (call_frame_t *frame, xlator_t *this)
+afr_local_t*
+afr_self_heal_local_init (afr_local_t *l, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	int              type = 0;
-	int              i = 0;
-	afr_private_t   *priv = NULL;
-	int              enoent_count = 0;
-	int              govinda_gOvinda = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->child_errno[i]) {
-			if (sh->child_errno[i] == ENOENT)
-				enoent_count++;
-		} else {
-			if (type) {
-				if (type != (sh->buf[i].st_mode & S_IFMT))
-					govinda_gOvinda = 1;
-			} else {
-				sh->source = i;
-				type = sh->buf[i].st_mode & S_IFMT;
-			}
-		}
-	}
+        afr_private_t   *priv  = NULL;
+        afr_local_t     *lc    = NULL;
+        afr_self_heal_t *sh    = NULL;
+        afr_self_heal_t *shc   = NULL;
+        int             ret    = 0;
 
-	if (govinda_gOvinda) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"conflicing filetypes exist for path %s. returning.",
-			local->loc.path);
+        priv = this->private;
 
-		local->govinda_gOvinda = 1;
-		sh_missing_entries_finish (frame, this);
-		return 0;
-	}
+        sh = &l->self_heal;
 
-	if (!type) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no source found for %s. all nodes down?. returning.",
-			local->loc.path);
-		/* subvolumes down and/or file does not exist */
-		sh_missing_entries_finish (frame, this);
-		return 0;
-	}
+        lc = mem_get0 (this->local_pool);
+        if (!lc)
+                goto out;
 
-	if (enoent_count == 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no missing files - %s. proceeding to metadata check",
-			local->loc.path);
-		/* proceed to next step - metadata self-heal */
-		sh_missing_entries_finish (frame, this);
-		return 0;
-	}
+        shc = &lc->self_heal;
+
+        shc->unwind = sh->unwind;
+        shc->gfid_sh_success_cbk = sh->gfid_sh_success_cbk;
+        shc->do_missing_entry_self_heal = sh->do_missing_entry_self_heal;
+        shc->do_gfid_self_heal = sh->do_gfid_self_heal;
+        shc->do_data_self_heal = sh->do_data_self_heal;
+        shc->do_metadata_self_heal = sh->do_metadata_self_heal;
+        shc->do_entry_self_heal = sh->do_entry_self_heal;
+        shc->force_confirm_spb = sh->force_confirm_spb;
+        shc->forced_merge = sh->forced_merge;
+        shc->background = sh->background;
+        shc->type = sh->type;
+        shc->data_sh_info = "";
+        shc->metadata_sh_info =  "";
+
+        uuid_copy (shc->sh_gfid_req, sh->sh_gfid_req);
+        if (l->loc.path) {
+                ret = loc_copy (&lc->loc, &l->loc);
+                if (ret < 0)
+                        goto out;
+        }
 
-	switch (type) {
-	case S_IFSOCK:
-	case S_IFREG:
-	case S_IFBLK:
-	case S_IFCHR:
-	case S_IFIFO:
-		sh_missing_entries_mknod (frame, this);
-		break;
-	case S_IFLNK:
-		sh_missing_entries_readlink (frame, this);
-		break;
-	case S_IFDIR:
-		sh_missing_entries_mkdir (frame, this);
-		break;
-	default:
-		gf_log (this->name, GF_LOG_ERROR,
-			"unknown file type: 0%o", type);
-		local->govinda_gOvinda = 1;
-		sh_missing_entries_finish (frame, this);
-	}
+        lc->child_up  = memdup (l->child_up,
+                                sizeof (*lc->child_up) * priv->child_count);
+        if (!lc->child_up) {
+                ret = -1;
+                goto out;
+        }
 
-	return 0;
+        if (l->xattr_req)
+                lc->xattr_req = dict_ref (l->xattr_req);
+
+        if (l->cont.lookup.inode)
+                lc->cont.lookup.inode = inode_ref (l->cont.lookup.inode);
+        if (l->cont.lookup.xattr)
+                lc->cont.lookup.xattr = dict_ref (l->cont.lookup.xattr);
+
+        lc->internal_lock.locked_nodes =
+                             GF_CALLOC (sizeof (*l->internal_lock.locked_nodes),
+                                        priv->child_count, gf_afr_mt_char);
+        if (!lc->internal_lock.locked_nodes) {
+                ret = -1;
+                goto out;
+        }
+
+        ret = afr_inodelk_init (&lc->internal_lock.inodelk[0],
+                                this->name, priv->child_count);
+        if (ret)
+                goto out;
+
+out:
+        if (ret) {
+                afr_local_cleanup (lc, this);
+                lc = NULL;
+        }
+        return lc;
 }
 
+int
+afr_self_heal_completion_cbk (call_frame_t *bgsh_frame, xlator_t *this)
+{
+        afr_private_t *   priv  = NULL;
+        afr_local_t *     local = NULL;
+        afr_self_heal_t * sh    = NULL;
+        afr_local_t *     orig_frame_local = NULL;
+        afr_self_heal_t * orig_frame_sh = NULL;
+        char              sh_type_str[256] = {0,};
+        gf_loglevel_t     loglevel = 0;
+
+        priv  = this->private;
+        local = bgsh_frame->local;
+        sh    = &local->self_heal;
+
+        if (local->unhealable) {
+                afr_set_split_brain (this, sh->inode, SPB, SPB);
+        }
 
-static int
-sh_missing_entries_lookup_cbk (call_frame_t *frame, void *cookie,
-			       xlator_t *this,
-			       int32_t op_ret, int32_t op_errno,
-			       inode_t *inode, struct stat *buf, dict_t *xattr)
-{
-	int              child_index = 0;
-	afr_local_t     *local = NULL;
-	int              call_count = 0;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"path %s on subvolume %s is of mode 0%o",
-				local->loc.path,
-				priv->children[child_index]->name,
-				buf->st_mode);
-
-			local->self_heal.buf[child_index] = *buf;
-		} else {
-			gf_log (this->name, GF_LOG_WARNING,
-				"path %s on subvolume %s => -1 (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-
-			local->self_heal.child_errno[child_index] = op_errno;
-		}
+        afr_self_heal_type_str_get (sh, sh_type_str,
+                                    sizeof(sh_type_str));
+        if (is_self_heal_failed (sh, AFR_CHECK_ALL) && !priv->shd.iamshd) {
+                loglevel = GF_LOG_ERROR;
+        } else if (!is_self_heal_failed (sh, AFR_CHECK_ALL)) {
+                loglevel = GF_LOG_INFO;
+        } else {
+                loglevel = GF_LOG_DEBUG;
+        }
 
-	}
-	UNLOCK (&frame->lock);
+        afr_log_self_heal_completion_status (local, loglevel);
 
-	call_count = afr_frame_return (frame);
+        FRAME_SU_UNDO (bgsh_frame, afr_local_t);
 
-	if (call_count == 0) {
-		sh_missing_entries_create (frame, this);
-	}
+        if (!sh->unwound && sh->unwind) {
+                orig_frame_local = sh->orig_frame->local;
+                orig_frame_sh = &orig_frame_local->self_heal;
+                orig_frame_sh->actual_sh_started = _gf_true;
+                sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno,
+                            is_self_heal_failed (sh, AFR_CHECK_ALL));
+        }
 
-	return 0;
+        if (sh->background) {
+                LOCK (&priv->lock);
+                {
+                        priv->background_self_heals_started--;
+                }
+                UNLOCK (&priv->lock);
+        }
+
+        AFR_STACK_DESTROY (bgsh_frame);
+
+        return 0;
 }
 
+int
+afr_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode)
+{
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int32_t          op_errno = 0;
+        int              ret = 0;
+        afr_self_heal_t *orig_sh = NULL;
+        call_frame_t    *sh_frame = NULL;
+        afr_local_t     *sh_local = NULL;
+        loc_t           *loc   = NULL;
+
+        local = frame->local;
+        orig_sh = &local->self_heal;
+        priv  = this->private;
+
+        GF_ASSERT (local->loc.path);
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "performing self heal on %s (metadata=%d data=%d entry=%d)",
+                local->loc.path,
+                local->self_heal.do_metadata_self_heal,
+                local->self_heal.do_data_self_heal,
+                local->self_heal.do_entry_self_heal);
+
+        op_errno        = ENOMEM;
+        sh_frame        = copy_frame (frame);
+        if (!sh_frame)
+                goto out;
+        afr_set_lk_owner (sh_frame, this, sh_frame->root);
+        afr_set_low_priority (sh_frame);
+
+        sh_local        = afr_self_heal_local_init (local, this);
+        if (!sh_local)
+                goto out;
+        sh_frame->local = sh_local;
+        sh              = &sh_local->self_heal;
 
-static int
-sh_missing_entries_lookup (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t    *local = NULL;
-	int             i = 0;
-	int             call_count = 0;
-	afr_private_t  *priv = NULL;
-	dict_t         *xattr_req = NULL;
-	int             ret = -1;
-
-	local = frame->local;
-	call_count = local->child_count;
-	priv = this->private;
-
-	local->call_count = call_count;
-	
-	xattr_req = dict_new();
-	
-	if (xattr_req)
-		ret = dict_set_uint64 (xattr_req, AFR_ENTRY_PENDING,
-				       priv->child_count * sizeof(int32_t));
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"looking up %s on subvolume %s",
-				local->loc.path, priv->children[i]->name);
-
-			STACK_WIND_COOKIE (frame,
-					   sh_missing_entries_lookup_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->lookup,
-					   &local->loc, xattr_req);
-
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	if (xattr_req)
-		dict_unref (xattr_req);
+        sh->inode       = inode_ref (inode);
+        sh->orig_frame  = frame;
 
-	return 0;
+        sh->completion_cbk = afr_self_heal_completion_cbk;
+
+        sh->success = GF_CALLOC (priv->child_count, sizeof (*sh->success),
+                                 gf_afr_mt_char);
+        if (!sh->success)
+                goto out;
+        sh->sources = GF_CALLOC (sizeof (*sh->sources), priv->child_count,
+                                 gf_afr_mt_int);
+        if (!sh->sources)
+                goto out;
+        sh->locked_nodes = GF_CALLOC (sizeof (*sh->locked_nodes),
+                                      priv->child_count,
+                                      gf_afr_mt_int);
+        if (!sh->locked_nodes)
+                goto out;
+
+        sh->pending_matrix = afr_matrix_create (priv->child_count,
+                                                priv->child_count);
+        if (!sh->pending_matrix)
+                goto out;
+
+        sh->delta_matrix = afr_matrix_create (priv->child_count,
+                                              priv->child_count);
+        if (!sh->delta_matrix)
+                goto out;
+
+        sh->fresh_parent_dirs = afr_children_create (priv->child_count);
+        if (!sh->fresh_parent_dirs)
+                goto out;
+        ret = afr_sh_common_create (sh, priv->child_count);
+        if (ret) {
+                op_errno = -ret;
+                goto out;
+        }
+
+        if (local->self_heal.background) {
+                LOCK (&priv->lock);
+                {
+                        if (priv->background_self_heals_started
+                            < priv->background_self_heal_count) {
+                                priv->background_self_heals_started++;
+
+
+                        } else {
+                                local->self_heal.background = _gf_false;
+                                sh->background = _gf_false;
+                        }
+                }
+                UNLOCK (&priv->lock);
+        }
+
+        if (!local->loc.parent) {
+                sh->do_missing_entry_self_heal = _gf_false;
+                sh->do_gfid_self_heal = _gf_false;
+        }
+
+        sh->sh_type_in_action = AFR_SELF_HEAL_INVALID;
+
+        FRAME_SU_DO (sh_frame, afr_local_t);
+        if (sh->do_missing_entry_self_heal || sh->do_gfid_self_heal) {
+                afr_self_heal_missing_entries (sh_frame, this);
+        } else {
+                loc = &sh_local->loc;
+                if (uuid_is_null (loc->inode->gfid) && uuid_is_null (loc->gfid)) {
+                        if (!uuid_is_null (inode->gfid))
+                                GF_ASSERT (!uuid_compare (inode->gfid,
+                                           sh->sh_gfid_req));
+                        uuid_copy (loc->gfid, sh->sh_gfid_req);
+                }
+                gf_log (this->name, GF_LOG_TRACE,
+                        "proceeding to metadata check on %s",
+                        local->loc.path);
+
+                afr_sh_missing_entries_done (sh_frame, this);
+        }
+        op_errno = 0;
+
+out:
+        if (op_errno) {
+                orig_sh->unwind (frame, this, -1, op_errno, 1);
+                if (sh_frame)
+                        AFR_STACK_DESTROY (sh_frame);
+        }
+        return 0;
 }
 
+void
+afr_self_heal_type_str_get (afr_self_heal_t *self_heal_p, char *str,
+                            size_t size)
+{
+        GF_ASSERT (str && (size > strlen (" missing-entry gfid "
+                                          "meta-data data entry")));
 
-static int
-sh_missing_entries_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	int              call_count = 0;
-	int              child_index = (long) cookie;
-
-
-	local = frame->local;
-	sh    = &local->self_heal;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			sh->op_failed = 1;
-
-			gf_log (this->name,
-				(op_errno == EAGAIN ? GF_LOG_DEBUG : GF_LOG_ERROR),
-				"locking inode of %s on child %d failed: %s",
-				local->loc.path, child_index,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"inode of %s on child %d locked",
-				local->loc.path, child_index);
-		}
-	}
-	UNLOCK (&frame->lock);
+        if (self_heal_p->do_metadata_self_heal) {
+                snprintf (str, size, " meta-data");
+        }
 
-	call_count = afr_frame_return (frame);
+        if (self_heal_p->do_data_self_heal) {
+                snprintf (str + strlen(str), size - strlen(str), " data");
+        }
 
-	if (call_count == 0) {
-		if (sh->op_failed == 1) {
-			sh_missing_entries_finish (frame, this);
-			return 0;
-		}
+        if (self_heal_p->do_entry_self_heal) {
+                snprintf (str + strlen(str), size - strlen(str), " entry");
+        }
 
-		sh_missing_entries_lookup (frame, this);
-	}
+        if (self_heal_p->do_missing_entry_self_heal) {
+                snprintf (str + strlen(str), size - strlen(str),
+                         " missing-entry");
+        }
 
-	return 0;
+        if (self_heal_p->do_gfid_self_heal) {
+                snprintf (str + strlen(str), size - strlen(str), " gfid");
+        }
 }
 
+afr_self_heal_type
+afr_self_heal_type_for_transaction (afr_transaction_type type)
+{
+        afr_self_heal_type sh_type = AFR_SELF_HEAL_INVALID;
+
+        switch (type) {
+        case AFR_DATA_TRANSACTION:
+                sh_type = AFR_SELF_HEAL_DATA;
+                break;
+        case AFR_METADATA_TRANSACTION:
+                sh_type = AFR_SELF_HEAL_METADATA;
+                break;
+        case AFR_ENTRY_TRANSACTION:
+                sh_type = AFR_SELF_HEAL_ENTRY;
+                break;
+        case AFR_ENTRY_RENAME_TRANSACTION:
+                GF_ASSERT (0);
+                break;
+        }
+        return sh_type;
+}
 
-static int
-afr_self_heal_missing_entries (call_frame_t *frame, xlator_t *this)
+int
+afr_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-	int              call_count = 0;
+        int   ret = -1;
+        uuid_t pargfid = {0};
 
+        if (!child)
+                goto out;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        if (!uuid_is_null (parent->inode->gfid))
+                uuid_copy (pargfid, parent->inode->gfid);
+        else if (!uuid_is_null (parent->gfid))
+                uuid_copy (pargfid, parent->gfid);
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"attempting to recreate missing entries for path=%s",
-		local->loc.path);
+        if (uuid_is_null (pargfid))
+                goto out;
 
-	afr_build_parent_loc (&sh->parent_loc, &local->loc);
+        if (strcmp (parent->path, "/") == 0)
+                ret = gf_asprintf ((char **)&child->path, "/%s", name);
+        else
+                ret = gf_asprintf ((char **)&child->path, "%s/%s", parent->path,
+                                   name);
 
-	call_count = local->child_count;
+        if (-1 == ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "asprintf failed while setting child path");
+        }
 
-	local->call_count = call_count;
+        child->name = strrchr (child->path, '/');
+        if (child->name)
+                child->name++;
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, sh_missing_entries_lk_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->entrylk,
-                                    this->name,
-				    &sh->parent_loc, local->loc.name,
-				    ENTRYLK_LOCK_NB, ENTRYLK_WRLCK);
-			if (!--call_count)
-				break;
-		}
- 	}
+        child->parent = inode_ref (parent->inode);
+        child->inode = inode_new (parent->inode->table);
+        uuid_copy (child->pargfid, pargfid);
 
-	return 0;
+        if (!child->inode) {
+                ret = -1;
+                goto out;
+        }
+
+        ret = 0;
+out:
+        if ((ret == -1) && child)
+                loc_wipe (child);
+
+        return ret;
+}
+
+int
+afr_sh_erase_pending (call_frame_t *frame, xlator_t *this,
+                      afr_transaction_type type, afr_fxattrop_cbk_t cbk,
+                      int (*finish)(call_frame_t *frame, xlator_t *this))
+{
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              call_count = 0;
+        int              i = 0;
+        dict_t          **erase_xattr = NULL;
+        int             ret = -1;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        afr_sh_pending_to_delta (priv, sh->xattr, sh->delta_matrix,
+                                 sh->success, priv->child_count, type);
+
+        erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count,
+                                 gf_afr_mt_dict_t);
+        if (!erase_xattr)
+                goto out;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (sh->xattr[i]) {
+                        call_count++;
+                        erase_xattr[i] = dict_new ();
+                        if (!erase_xattr[i])
+                                goto out;
+                }
+        }
+
+        afr_sh_delta_to_xattr (this, sh->delta_matrix, erase_xattr,
+                               priv->child_count, type);
+
+        gf_log (this->name, GF_LOG_DEBUG, "Delta matrix for: %s",
+                lkowner_utoa (&frame->root->lk_owner));
+        afr_sh_print_pending_matrix (sh->delta_matrix, this);
+        local->call_count = call_count;
+        if (call_count == 0) {
+                ret = 0;
+                finish (frame, this);
+                goto out;
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (!erase_xattr[i])
+                        continue;
+
+                if (sh->healing_fd) {//true for ENTRY, reg file DATA transaction
+                        STACK_WIND_COOKIE (frame, cbk, (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->fxattrop,
+                                           sh->healing_fd,
+                                           GF_XATTROP_ADD_ARRAY, erase_xattr[i],
+                                           NULL);
+                } else {
+                        STACK_WIND_COOKIE (frame, cbk, (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->xattrop,
+                                           &local->loc,
+                                           GF_XATTROP_ADD_ARRAY, erase_xattr[i],
+                                           NULL);
+                }
+        }
+
+        ret = 0;
+out:
+        if (erase_xattr) {
+                for (i = 0; i < priv->child_count; i++) {
+                        if (erase_xattr[i]) {
+                                dict_unref (erase_xattr[i]);
+                        }
+                }
+        }
+
+        GF_FREE (erase_xattr);
+
+        if (ret < 0) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                finish (frame, this);
+        }
+
+        return 0;
 }
 
+void
+afr_set_self_heal_status(afr_self_heal_t *sh, afr_self_heal_status status)
+{
+        xlator_t                *this = NULL;
+        afr_sh_status_for_all_type *sh_status = &(sh->afr_all_sh_status);
+        afr_self_heal_type  sh_type_in_action = sh->sh_type_in_action;
+        this = THIS;
+
+        if (!sh) {
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "Null self heal"
+                                  "Structure");
+                goto out;
+        }
+
+        switch (sh_type_in_action) {
+                case AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY:
+                       sh_status->gfid_or_missing_entry_self_heal = status;
+                        break;
+                case AFR_SELF_HEAL_METADATA:
+                        sh_status->metadata_self_heal = status;
+                        break;
+                case AFR_SELF_HEAL_DATA:
+                        sh_status->data_self_heal = status;
+                        break;
+                case AFR_SELF_HEAL_ENTRY:
+                        sh_status->entry_self_heal = status;
+                        break;
+                case AFR_SELF_HEAL_INVALID:
+                        gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid"
+                                          "self heal type in action");
+                        break;
+        }
+out:
+        return;
+}
+
+void
+afr_set_local_for_unhealable (afr_local_t *local)
+{
+        afr_self_heal_t  *sh = NULL;
+
+        sh = &local->self_heal;
+
+        local->unhealable = 1;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+}
 
 int
-afr_self_heal (call_frame_t *frame, xlator_t *this,
-	       int (*completion_cbk) (call_frame_t *, xlator_t *))
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"performing self heal on %s (metadata=%d data=%d entry=%d)",
-		local->loc.path,
-		local->need_metadata_self_heal,
-		local->need_data_self_heal,
-		local->need_entry_self_heal);
-
-	sh->completion_cbk = completion_cbk;
-
-	sh->buf = CALLOC (priv->child_count, sizeof (struct stat));
-	sh->child_errno = CALLOC (priv->child_count, sizeof (int));
-	sh->success = CALLOC (priv->child_count, sizeof (int));
-	sh->xattr = CALLOC (priv->child_count, sizeof (dict_t *));
-	sh->sources = CALLOC (sizeof (*sh->sources), priv->child_count);
-
-	sh->pending_matrix = CALLOC (sizeof (int32_t *), priv->child_count);
-	for (i = 0; i < priv->child_count; i++) {
-		sh->pending_matrix[i] = CALLOC (sizeof (int32_t),
-						priv->child_count);
-	}
+is_self_heal_failed (afr_self_heal_t *sh, afr_sh_fail_check_type type)
+{
+        afr_sh_status_for_all_type      sh_status = sh->afr_all_sh_status;
+        afr_self_heal_type   sh_type_in_action =  AFR_SELF_HEAL_INVALID;
+        afr_self_heal_status    status = AFR_SELF_HEAL_FAILED;
+        xlator_t                *this = NULL;
+        int                     sh_failed = 0;
+
+        this = THIS;
+
+        if (!sh) {
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "Null self heal "
+                                  "structure");
+                sh_failed = 1;
+                goto out;
+        }
 
-	sh->delta_matrix = CALLOC (sizeof (int32_t *), priv->child_count);
-	for (i = 0; i < priv->child_count; i++) {
-		sh->delta_matrix[i] = CALLOC (sizeof (int32_t),
-					      priv->child_count);
-	}
+        if (type == AFR_CHECK_ALL) {
+                if ((sh_status.gfid_or_missing_entry_self_heal == AFR_SELF_HEAL_FAILED)
+                    || (sh_status.metadata_self_heal == AFR_SELF_HEAL_FAILED)
+                    || (sh_status.data_self_heal == AFR_SELF_HEAL_FAILED)
+                    || (sh_status.entry_self_heal == AFR_SELF_HEAL_FAILED))
+                sh_failed = 1;
+        } else if (type == AFR_CHECK_SPECIFIC) {
+                sh_type_in_action = sh->sh_type_in_action;
+                switch (sh_type_in_action) {
+                        case AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY:
+                             status = sh_status.gfid_or_missing_entry_self_heal;
+                                break;
+                        case AFR_SELF_HEAL_METADATA:
+                                status = sh_status.metadata_self_heal;
+                                break;
+                        case AFR_SELF_HEAL_ENTRY:
+                                status = sh_status.entry_self_heal;
+                                break;
+                        case AFR_SELF_HEAL_DATA:
+                                status = sh_status.data_self_heal;
+                                break;
+                        case AFR_SELF_HEAL_INVALID:
+                                status = AFR_SELF_HEAL_NOT_ATTEMPTED;
+                                break;
+                }
+                if (status == AFR_SELF_HEAL_FAILED)
+                        sh_failed = 1;
 
-	if (local->success_count && local->enoent_count) {
-		afr_self_heal_missing_entries (frame, this);
-	} else {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"proceeding to metadata check on %s",
-			local->loc.path);
-		afr_sh_missing_entries_done (frame, this);
-	}
+        }
 
-	return 0;
+out:
+        return sh_failed;
+}
+
+char *
+get_sh_completion_status (afr_self_heal_status status)
+{
+
+        char *not_attempted       = " is not attempted";
+        char *failed              = " failed";
+        char *started             = " is started";
+        char *sync_begin          = " is successfully completed";
+        char *result              = " has unknown status";
+
+        switch (status)
+        {
+                case AFR_SELF_HEAL_NOT_ATTEMPTED:
+                        result = not_attempted;
+                        break;
+                case AFR_SELF_HEAL_FAILED:
+                        result = failed;
+                        break;
+                case AFR_SELF_HEAL_STARTED:
+                        result = started;
+                        break;
+                case AFR_SELF_HEAL_SYNC_BEGIN:
+                        result = sync_begin;
+                        break;
+        }
+
+        return result;
+
+}
+
+void
+afr_log_self_heal_completion_status (afr_local_t *local, gf_loglevel_t loglvl)
+{
+
+        char sh_log[4096]              = {0};
+        afr_self_heal_t *sh            = &local->self_heal;
+        afr_sh_status_for_all_type   all_status = sh->afr_all_sh_status;
+        xlator_t      *this            = NULL;
+        size_t        off              = 0;
+        int           data_sh          = 0;
+        int           metadata_sh      = 0;
+        int           print_log        = 0;
+
+        this = THIS;
+
+        ADD_FMT_STRING (sh_log, off, "gfid or missing entry",
+                        all_status.gfid_or_missing_entry_self_heal, print_log);
+        ADD_FMT_STRING_SYNC (sh_log, off, "metadata",
+                             all_status.metadata_self_heal, print_log);
+        if (sh->background) {
+                ADD_FMT_STRING_SYNC (sh_log, off, "backgroung data",
+                                all_status.data_self_heal, print_log);
+        } else {
+                ADD_FMT_STRING_SYNC (sh_log, off, "foreground data",
+                                all_status.data_self_heal, print_log);
+        }
+        ADD_FMT_STRING_SYNC (sh_log, off, "entry", all_status.entry_self_heal,
+                             print_log);
+
+        if (AFR_SELF_HEAL_SYNC_BEGIN == all_status.data_self_heal &&
+	    strcmp (sh->data_sh_info, "") && sh->data_sh_info )
+                data_sh = 1;
+        if (AFR_SELF_HEAL_SYNC_BEGIN == all_status.metadata_self_heal &&
+	    strcmp (sh->metadata_sh_info, "") && sh->metadata_sh_info)
+                metadata_sh = 1;
+
+        if (!print_log)
+                return;
+
+        gf_log (this->name, loglvl, "%s %s %s on %s", sh_log,
+                ((data_sh == 1) ? sh->data_sh_info : ""),
+                ((metadata_sh == 1) ? sh->metadata_sh_info : ""),
+                local->loc.path);
 }
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.h b/xlators/cluster/afr/src/afr-self-heal-common.h
index f3beff71f..473264776 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.h
+++ b/xlators/cluster/afr/src/afr-self-heal-common.h
@@ -1,66 +1,144 @@
 /*
-  Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  General Public License for more details.
-
-  You should have received a copy of the GNU General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __AFR_SELF_HEAL_COMMON_H__
 #define __AFR_SELF_HEAL_COMMON_H__
 
-#define FILE_HAS_HOLES(buf) (((buf)->st_size) > ((buf)->st_blocks * 512))
+#define FILE_HAS_HOLES(buf) (((buf)->ia_size) > ((buf)->ia_blocks * 512))
+#define AFR_SH_MIN_PARTICIPANTS 2
+
+typedef enum {
+        AFR_LOOKUP_FAIL_CONFLICTS = 1,
+        AFR_LOOKUP_FAIL_MISSING_GFIDS = 2,
+} afr_lookup_flags_t;
 
 int
 afr_sh_select_source (int sources[], int child_count);
 
 int
-afr_sh_sink_count (int sources[], int child_count);
+afr_sh_source_count (int sources[], int child_count);
+
+void
+afr_sh_print_pending_matrix (int32_t *pending_matrix[], xlator_t *this);
+
+void
+afr_sh_print_split_brain_log (int32_t *pending_matrix[], xlator_t *this,
+                              const char *loc);
 
 int
-afr_sh_source_count (int sources[], int child_count);
+afr_build_pending_matrix (char **pending_key, int32_t **pending_matrix,
+                          unsigned char *ignorant_subvols,
+                          dict_t *xattr[], afr_transaction_type type,
+                          size_t child_count);
+
+void
+afr_sh_pending_to_delta (afr_private_t *priv, dict_t **xattr,
+                         int32_t *delta_matrix[], unsigned char success[],
+                         int child_count, afr_transaction_type type);
 
 int
-afr_sh_supress_errenous_children (int sources[], int child_errno[],
-				  int child_count);
+afr_mark_sources (xlator_t *this, int32_t *sources, int32_t **pending_matrix,
+                  struct iatt *bufs, afr_self_heal_type type,
+                  int32_t *success_children, int32_t *subvol_status);
 
 int
-afr_sh_supress_empty_children (int sources[], dict_t *xattr[],
-			       struct stat *buf,
-			       int child_count, const char *key);
+afr_sh_delta_to_xattr (xlator_t *this,
+                       int32_t *delta_matrix[], dict_t *xattr[],
+		       int child_count, afr_transaction_type type);
 
 void
-afr_sh_print_pending_matrix (int32_t *pending_matrix[], xlator_t *this);
+afr_self_heal_type_str_get (afr_self_heal_t *self_heal_p, char *str,
+                            size_t size);
+
+afr_self_heal_type
+afr_self_heal_type_for_transaction (afr_transaction_type type);
 
+int
+afr_build_sources (xlator_t *this, dict_t **xattr, struct iatt *bufs,
+                   int32_t **pending_matrix, int32_t *sources,
+                   int32_t *success_children, afr_transaction_type type,
+                   int32_t *subvol_status, gf_boolean_t ignore_ignorant);
 void
-afr_sh_build_pending_matrix (int32_t *pending_matrix[], dict_t *xattr[],
-			     int child_count, const char *key);
+afr_sh_common_reset (afr_self_heal_t *sh, unsigned int child_count);
 
 void
-afr_sh_pending_to_delta (int32_t *pending_matrix[], int32_t *delta_matrix[],
-			 int32_t success[], int child_count);
+afr_sh_common_lookup_resp_handler (call_frame_t *frame, void *cookie,
+                                   xlator_t *this,
+                                   int32_t op_ret, int32_t op_errno,
+                                   inode_t *inode, struct iatt *buf,
+                                   dict_t *xattr, struct iatt *postparent,
+                                   loc_t *loc);
 
 int
-afr_sh_mark_sources (int32_t *pending_matrix[], int sources[],
-		     int child_count);
+afr_sh_common_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      afr_lookup_done_cbk_t lookup_cbk, uuid_t uuid,
+                      int32_t flags, dict_t *xdata);
+int
+afr_sh_entry_expunge_remove (call_frame_t *expunge_frame, xlator_t *this,
+                             int active_src, struct iatt *buf,
+                             struct iatt *parentbuf);
+int
+afr_sh_entrylk (call_frame_t *frame, xlator_t *this, loc_t *loc,
+                char *base_name, afr_lock_cbk_t lock_cbk);
+int
+afr_sh_entry_impunge_create (call_frame_t *impunge_frame, xlator_t *this,
+                             int child_index);
+int
+afr_sh_data_unlock (call_frame_t *frame, xlator_t *this, char *dom,
+                    afr_lock_cbk_t lock_cbk);
+afr_local_t *
+afr_self_heal_local_init (afr_local_t *l, xlator_t *this);
+int
+afr_sh_data_lock (call_frame_t *frame, xlator_t *this,
+                  off_t start, off_t len, gf_boolean_t block, char *dom,
+                  afr_lock_cbk_t success_handler,
+                  afr_lock_cbk_t failure_handler);
+void
+afr_sh_set_error (afr_self_heal_t *sh, int32_t op_errno);
+void
+afr_sh_mark_source_sinks (call_frame_t *frame, xlator_t *this);
+typedef int
+(*afr_fxattrop_cbk_t) (call_frame_t *frame, void *cookie,
+                       xlator_t *this, int32_t op_ret, int32_t op_errno,
+                       dict_t *xattr, dict_t *xdata);
+int
+afr_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name);
+int
+afr_impunge_frame_create (call_frame_t *frame, xlator_t *this,
+                          int active_source, call_frame_t **impunge_frame);
+void
+afr_sh_reset (call_frame_t *frame, xlator_t *this);
 
+void
+afr_children_intersection_get (int32_t *set1, int32_t *set2,
+                               int *intersection, unsigned int child_count);
 int
-afr_sh_delta_to_xattr (int32_t *delta_matrix[], dict_t *xattr[],
-		       int child_count, const char *key);
+afr_get_no_xattr_dir_read_child (xlator_t *this, int32_t *success_children,
+                                 struct iatt *bufs);
+int
+afr_sh_erase_pending (call_frame_t *frame, xlator_t *this,
+                      afr_transaction_type type, afr_fxattrop_cbk_t cbk,
+                      int (*finish)(call_frame_t *frame, xlator_t *this));
+
+void
+afr_set_local_for_unhealable (afr_local_t *local);
 
 int
-afr_sh_is_matrix_zero (int32_t *pending_matrix[], int child_count);
+is_self_heal_failed (afr_self_heal_t *sh, afr_sh_fail_check_type type);
+
+void
+afr_set_self_heal_status (afr_self_heal_t *sh, afr_self_heal_status status);
 
+void
+afr_log_self_heal_completion_status (afr_local_t *local, gf_loglevel_t  logl);
 
+char*
+afr_get_pending_matrix_str (int32_t *pending_matrix[], xlator_t *this);
 #endif /* __AFR_SELF_HEAL_COMMON_H__ */
diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c
index 775845fe6..9de26ee56 100644
--- a/xlators/cluster/afr/src/afr-self-heal-data.c
+++ b/xlators/cluster/afr/src/afr-self-heal-data.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include <libgen.h>
@@ -47,986 +38,1717 @@
 #include "afr-transaction.h"
 #include "afr-self-heal.h"
 #include "afr-self-heal-common.h"
-
-
+#include "afr-self-heal-algorithm.h"
 
 int
-afr_sh_data_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+afr_sh_data_fail (call_frame_t *frame, xlator_t *this);
 
-	/* 
-	   TODO: cleanup sh->* 
-	 */
+static inline gf_boolean_t
+afr_sh_data_proceed (unsigned int success_count)
+{
+        return (success_count >= AFR_SH_MIN_PARTICIPANTS);
+}
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"self heal of %s completed",
-		local->loc.path);
+extern int
+sh_loop_finish (call_frame_t *loop_frame, xlator_t *this);
 
-	sh->completion_cbk (frame, this);
+int
+afr_post_sh_big_lock_success (call_frame_t *frame, xlator_t *this);
 
-	return 0;
-}
+int
+afr_post_sh_big_lock_failure (call_frame_t *frame, xlator_t *this);
 
+int
+afr_sh_data_finish (call_frame_t *frame, xlator_t *this);
 
 int
-afr_sh_data_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno)
+afr_sh_data_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_private_t   *priv  = NULL;
-	afr_self_heal_t *sh = NULL;
-	int              call_count = 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
 
-	LOCK (&frame->lock);
-	{
-	}
-	UNLOCK (&frame->lock);
+        sh->completion_cbk (frame, this);
 
-	call_count = afr_frame_return (frame);
+        return 0;
+}
 
-	if (call_count == 0) {
-		fd_unref (sh->healing_fd);
-		sh->healing_fd = NULL;
-		afr_sh_data_done (frame, this);
-	}
 
-	return 0;
+int
+afr_sh_data_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_local_t   *local       = NULL;
+        afr_private_t *priv        = NULL;
+        int            call_count  = 0;
+        int            child_index = (long) cookie;
+
+        local = frame->local;
+        priv = this->private;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "flush failed on %s on subvolume %s: %s",
+                                local->loc.path, priv->children[child_index]->name,
+                                strerror (op_errno));
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                afr_sh_data_done (frame, this);
+        }
+
+        return 0;
 }
 
-
 int
 afr_sh_data_close (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_private_t   *priv  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              i = 0;
-	int              call_count = 0;
-
+        afr_local_t     *local      = NULL;
+        afr_private_t   *priv       = NULL;
+        afr_self_heal_t *sh         = NULL;
+        int              i          = 0;
+        int              call_count = 0;
+
+        local = frame->local;
+        sh    = &local->self_heal;
+        priv  = this->private;
+
+        if (!sh->healing_fd) {
+                //This happens when file is non-reg
+                afr_sh_data_done (frame, this);
+                return 0;
+        }
+        call_count        = afr_set_elem_count_get (sh->success,
+                                                    priv->child_count);
+        local->call_count = call_count;
+
+        if (call_count == 0) {
+                afr_sh_data_done (frame, this);
+                return 0;
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (!sh->success[i])
+                        continue;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "closing fd of %s on %s",
+                        local->loc.path, priv->children[i]->name);
+
+                STACK_WIND_COOKIE (frame, afr_sh_data_flush_cbk,
+                                   (void *) (long) i,
+                                   priv->children[i],
+                                   priv->children[i]->fops->flush,
+                                   sh->healing_fd, NULL);
+
+                if (!--call_count)
+                        break;
+        }
+
+        return 0;
+}
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+int
+afr_sh_dom_unlock (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh    = NULL;
+        afr_private_t   *priv  = NULL;
+
+        local = frame->local;
+        sh    = &local->self_heal;
+        priv  = this->private;
+
+        if (sh->sh_dom_lock_held)
+                afr_sh_data_unlock (frame, this, priv->sh_domain,
+                                    afr_sh_data_close);
+        else
+                afr_sh_data_close (frame, this);
+        return 0;
+}
 
-	if (!sh->healing_fd) {
-		afr_sh_data_done (frame, this);
-		return 0;
-	}
+int
+afr_sh_data_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                         struct iatt *statpost, dict_t *xdata)
+{
 
-	call_count = sh->active_sinks + 1;
-	local->call_count = call_count;
+        afr_local_t   *local       = NULL;
+        afr_private_t *priv        = NULL;
+        int            call_count  = 0;
+        int            child_index = (long) cookie;
 
+        local = frame->local;
+        priv = this->private;
 
-	/* closed source */
-	gf_log (this->name, GF_LOG_DEBUG,
-		"closing fd of %s on %s",
-		local->loc.path, priv->children[sh->source]->name);
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "setattr failed on %s on subvolume %s: %s",
+                                local->loc.path, priv->children[child_index]->name,
+                                strerror (op_errno));
+                }
+        }
+        UNLOCK (&frame->lock);
 
-	STACK_WIND_COOKIE (frame, afr_sh_data_flush_cbk,
-			   (void *) (long) sh->source,
-			   priv->children[sh->source],
-			   priv->children[sh->source]->fops->flush,
-			   sh->healing_fd);
-	call_count--;
+        call_count = afr_frame_return (frame);
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->sources[i] || !local->child_up[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"closing fd of %s on %s",
-			local->loc.path, priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_data_flush_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->flush,
-				   sh->healing_fd);
-		if (!--call_count)
-			break;
-	}
+        if (call_count == 0) {
+                afr_sh_data_finish (frame, this);
+        }
 
-	return 0;
+        return 0;
 }
 
+int
+afr_sh_data_setattr (call_frame_t *frame, xlator_t *this, struct iatt* stbuf)
+{
+        afr_local_t     *local      = NULL;
+        afr_private_t   *priv       = NULL;
+        afr_self_heal_t *sh         = NULL;
+        int              i          = 0;
+        int              call_count = 0;
+        int32_t          valid      = 0;
+
+        local = frame->local;
+        sh    = &local->self_heal;
+        priv  = this->private;
+
+        valid = (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME);
+
+        call_count        = afr_set_elem_count_get (sh->success,
+                                                    priv->child_count);
+        local->call_count = call_count;
+
+        if (call_count == 0) {
+                GF_ASSERT (0);
+                afr_sh_data_finish (frame, this);
+                return 0;
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (!sh->success[i])
+                        continue;
+
+                STACK_WIND_COOKIE (frame, afr_sh_data_setattr_cbk,
+                                   (void *) (long) i,
+                                   priv->children[i],
+                                   priv->children[i]->fops->setattr,
+                                   &local->loc, stbuf, valid, NULL);
+
+                if (!--call_count)
+                        break;
+        }
+
+        return 0;
+}
 
 int
-afr_sh_data_unlck_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno)
+afr_sh_data_setattr_fstat_cbk (call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               struct iatt *buf, dict_t *xdata)
 {
-	afr_local_t * local = NULL;
-	int           call_count = 0;
-	int           child_index = (long) cookie;
-
-	
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"locking inode of %s on child %d failed: %s",
-				local->loc.path, child_index,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"inode of %s on child %d locked",
-				local->loc.path, child_index);
-		}
-	}
-	UNLOCK (&frame->lock);
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        int child_index = (long) cookie;
+
+        local = frame->local;
+        sh = &local->self_heal;
+
+        GF_ASSERT (sh->source == child_index);
+        if (op_ret != -1) {
+                sh->buf[child_index] = *buf;
+                afr_sh_data_setattr (frame, this, buf);
+        } else {
+                gf_log (this->name, GF_LOG_ERROR, "%s: Failed to set "
+                        "time-stamps after self-heal", local->loc.path);
+                afr_sh_data_fail (frame, this);
+        }
+
+        return 0;
+}
 
-	call_count = afr_frame_return (frame);
+/*
+ * If there are any writes after the self-heal is triggered then the
+ * stbuf stored in local->self_heal.buf[] will be invalid so we do one more
+ * stat on the source and then set the [am]times
+ */
+int
+afr_sh_set_timestamps (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local      = NULL;
+        afr_private_t   *priv       = NULL;
+        afr_self_heal_t *sh         = NULL;
+
+        local = frame->local;
+        sh    = &local->self_heal;
+        priv  = this->private;
+
+        STACK_WIND_COOKIE (frame, afr_sh_data_setattr_fstat_cbk,
+                           (void *) (long) sh->source,
+                           priv->children[sh->source],
+                           priv->children[sh->source]->fops->fstat,
+                           sh->healing_fd, NULL);
+        return 0;
+}
 
-	if (call_count == 0) {
-		afr_sh_data_close (frame, this);
-	}
+//Fun fact, lock_cbk is being used for both lock & unlock
+int
+afr_sh_data_unlock (call_frame_t *frame, xlator_t *this, char *dom,
+                    afr_lock_cbk_t lock_cbk)
+{
+        afr_local_t         *local    = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_self_heal_t     *sh       = NULL;
+        afr_private_t       *priv     = NULL;
+        int                 ret       = 0;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        sh       = &local->self_heal;
+        priv     = this->private;
+
+        if (strcmp (dom, this->name) == 0) {
+                sh->data_lock_held = _gf_false;
+        } else if (strcmp (dom, priv->sh_domain) == 0) {
+                sh->sh_dom_lock_held = _gf_false;
+        } else {
+                ret = -1;
+                goto out;
+        }
+        int_lock->lock_cbk = lock_cbk;
+        int_lock->domain = dom;
+        afr_unlock (frame, this);
 
-	return 0;
+out:
+        if (ret) {
+                int_lock->lock_op_ret = -1;
+                int_lock->lock_cbk (frame, this);
+        }
+        return 0;
 }
 
-
 int
-afr_sh_data_unlock (call_frame_t *frame, xlator_t *this)
+afr_sh_data_finish (call_frame_t *frame, xlator_t *this)
 {
-	struct flock flock;			
-	int i = 0;				
-	int call_count = 0;		     
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh    = NULL;
 
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	afr_self_heal_t * sh  = NULL;
+        local = frame->local;
+        sh = &local->self_heal;
 
+        gf_log (this->name, GF_LOG_DEBUG,
+                "finishing data selfheal of %s", local->loc.path);
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        if (sh->data_lock_held)
+                afr_sh_data_unlock (frame, this, this->name, afr_sh_dom_unlock);
+        else
+                afr_sh_dom_unlock (frame, this);
 
-	call_count = local->child_count;
+        return 0;
+}
 
-	local->call_count = call_count;		
+int
+afr_sh_data_fail (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local = NULL;
+        afr_self_heal_t *sh = NULL;
 
-	flock.l_start = 0;
-	flock.l_len   = 0;
-	flock.l_type  = F_UNLCK;
+        local = frame->local;
+        sh = &local->self_heal;
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"unlocking %s on subvolume %s",
-				local->loc.path, priv->children[i]->name);
-
-			STACK_WIND_COOKIE (frame, afr_sh_data_unlck_cbk,
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->inodelk,
-                                           this->name,
-					   &local->loc, F_SETLK, &flock); 
-			if (!--call_count)
-				break;
-		}
-	}
+        gf_log (this->name, GF_LOG_DEBUG,
+                "finishing failed data selfheal of %s", local->loc.path);
 
-	return 0;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        afr_sh_data_finish (frame, this);
+        return 0;
 }
 
-
 int
-afr_sh_data_finish (call_frame_t *frame, xlator_t *this)
+afr_sh_data_erase_pending_cbk (call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret,
+                               int32_t op_errno, dict_t *xattr, dict_t *xdata)
 {
-	afr_local_t   *local = NULL;
-
-	local = frame->local;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"finishing data selfheal of %s", local->loc.path);
-
-	afr_sh_data_unlock (frame, this);
+        int             call_count = 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int32_t         child_index = (long) cookie;
+
+        priv  = this->private;
+        local = frame->local;
+        sh    = &local->self_heal;
+        if (op_ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR, "Erasing of pending change "
+                        "log failed on %s for subvol %s, reason: %s",
+                        local->loc.path, priv->children[child_index]->name,
+                        strerror (op_errno));
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        }
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
+                        if (sh->old_loop_frame)
+                                sh_loop_finish (sh->old_loop_frame, this);
+                        sh->old_loop_frame = NULL;
+                        afr_sh_data_fail (frame, this);
+                        goto out;
+                }
+                if (!IA_ISREG (sh->type)) {
+                        afr_sh_data_finish (frame, this);
+                        goto out;
+                }
+                GF_ASSERT (sh->old_loop_frame);
+                afr_sh_data_lock (frame, this, 0, 0, _gf_true, this->name,
+                                  afr_post_sh_big_lock_success,
+                                  afr_post_sh_big_lock_failure);
+        }
+out:
+        return 0;
+}
 
-	return 0;
+int
+afr_sh_data_erase_pending (call_frame_t *frame, xlator_t *this)
+{
+        afr_sh_erase_pending (frame, this, AFR_DATA_TRANSACTION,
+                              afr_sh_data_erase_pending_cbk,
+                              afr_sh_data_finish);
+        return 0;
 }
 
+int
+afr_sh_data_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, struct iatt *pre,
+                       struct iatt *post, dict_t *xdata)
+{
+        afr_local_t     *local      = NULL;
+        afr_private_t   *priv       = NULL;
+        afr_self_heal_t *sh         = NULL;
+        int             call_count  = 0;
+        int             child_index = (long) cookie;
+
+        local = frame->local;
+        priv = this->private;
+        sh   = &local->self_heal;
+
+        if (op_ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR, "%s: Failed to fsync on "
+                        "%s - %s", local->loc.path,
+                        priv->children[child_index]->name, strerror (op_errno));
+                LOCK (&frame->lock);
+                {
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                }
+                UNLOCK (&frame->lock);
+                if (sh->old_loop_frame)
+                        sh_loop_finish (sh->old_loop_frame, this);
+                sh->old_loop_frame = NULL;
+        }
+
+        call_count = afr_frame_return (frame);
+        if (call_count == 0) {
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC))
+                        afr_sh_data_fail (frame, this);
+                else
+                        afr_sh_data_erase_pending (frame, this);
+        }
+        return 0;
+}
 
+/*
+ * Before erasing xattrs, make sure the data is written to disk
+ */
 int
-afr_sh_data_erase_pending_cbk (call_frame_t *frame, void *cookie,
-			       xlator_t *this, int32_t op_ret,
-			       int32_t op_errno, dict_t *xattr)
+afr_sh_data_fsync (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int             call_count = 0;
+        afr_local_t     *local     = NULL;
+        afr_private_t   *priv      = NULL;
+        afr_self_heal_t *sh        = NULL;
+        int             i          = 0;
+        int             call_count = 0;
+
+        local = frame->local;
+        priv = this->private;
+        sh   = &local->self_heal;
+
+        call_count        = sh->active_sinks;
+        if (call_count == 0) {
+                afr_sh_data_erase_pending (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+        for (i = 0; i < priv->child_count; i++) {
+                if (!sh->success[i] || sh->sources[i])
+                        continue;
+
+                STACK_WIND_COOKIE (frame, afr_sh_data_fsync_cbk,
+                                   (void *) (long) i, priv->children[i],
+                                   priv->children[i]->fops->fsync,
+                                   sh->healing_fd, 1, NULL);
+        }
+
+        return 0;
+}
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+static struct afr_sh_algorithm *
+sh_algo_from_name (xlator_t *this, char *name)
+{
+        int i = 0;
 
-	LOCK (&frame->lock);
-	{
-	}
-	UNLOCK (&frame->lock);
+        if (name == NULL)
+                goto out;
 
-	call_count = afr_frame_return (frame);
+        while (afr_self_heal_algorithms[i].name) {
+                if (!strcmp (name, afr_self_heal_algorithms[i].name)) {
+                        return &afr_self_heal_algorithms[i];
+                }
 
-	if (call_count == 0)
-		afr_sh_data_finish (frame, this);
+                i++;
+        }
 
-	return 0;
+out:
+        return NULL;
 }
 
 
-int
-afr_sh_data_erase_pending (call_frame_t *frame, xlator_t *this)
+static int
+sh_zero_byte_files_exist (afr_local_t *local, int child_count)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              i = 0;
-	dict_t          **erase_xattr = NULL;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        int             i = 0;
+        int             ret = 0;
+        afr_self_heal_t *sh = NULL;
+
+        sh = &local->self_heal;
+        for (i = 0; i < child_count; i++) {
+                if (!local->child_up[i] || sh->child_errno[i])
+                        continue;
+                if (sh->buf[i].ia_size == 0) {
+                        ret = 1;
+                        break;
+                }
+        }
+
+        return ret;
+}
 
 
-	afr_sh_pending_to_delta (sh->pending_matrix, sh->delta_matrix,
-				 sh->success, priv->child_count);
+struct afr_sh_algorithm *
+afr_sh_data_pick_algo (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t *           priv  = NULL;
+        struct afr_sh_algorithm * algo  = NULL;
+        afr_local_t *             local = NULL;
+        afr_self_heal_t *         sh    = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+        sh    = &local->self_heal;
+        algo  = sh_algo_from_name (this, priv->data_self_heal_algorithm);
+
+        if (algo == NULL) {
+                /* option not set, so fall back on heuristics */
+
+                if (sh_zero_byte_files_exist (local, priv->child_count)
+                    || (sh->file_size <= (priv->data_self_heal_window_size *
+                                          this->ctx->page_size))) {
+
+                        /*
+                         * If the file does not exist on one of the subvolumes,
+                         * or a zero-byte file exists (created by entry self-heal)
+                         * the entire content has to be copied anyway, so there
+                         * is no benefit from using the "diff" algorithm.
+                         *
+                         * If the file size is about the same as page size,
+                         * the entire file can be read and written with a few
+                         * (pipelined) STACK_WINDs, which will be faster
+                         * than "diff" which has to read checksums and then
+                         * read and write.
+                         */
+
+                        algo = sh_algo_from_name (this, "full");
+
+                } else {
+                        algo = sh_algo_from_name (this, "diff");
+                }
+        }
+
+        return algo;
+}
 
-	erase_xattr = CALLOC (sizeof (*erase_xattr), priv->child_count);
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->xattr[i]) {
-			call_count++;
+int
+afr_sh_data_sync_prepare (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        struct afr_sh_algorithm *sh_algo = NULL;
 
-			erase_xattr[i] = get_new_dict();
-			dict_ref (erase_xattr[i]);
-		}
-	}
+        local = frame->local;
+        sh = &local->self_heal;
 
-	afr_sh_delta_to_xattr (sh->delta_matrix, erase_xattr,
-			       priv->child_count, AFR_DATA_PENDING);
+        sh->algo_completion_cbk = afr_sh_data_fsync;
+        sh->algo_abort_cbk      = afr_sh_data_fail;
 
-	local->call_count = call_count;
-	for (i = 0; i < priv->child_count; i++) {
-		if (!erase_xattr[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"erasing pending flags from %s on %s",
-			local->loc.path, priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_data_erase_pending_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->xattrop,
-				   &local->loc,
-				   GF_XATTROP_ADD_ARRAY, erase_xattr[i]);
-		if (!--call_count)
-			break;
-	}
+        sh_algo = afr_sh_data_pick_algo (frame, this);
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (erase_xattr[i]) {
-			dict_unref (erase_xattr[i]);
-		}
-	}
-	FREE (erase_xattr);
+        sh->algo = sh_algo;
+        sh_algo->fn (frame, this);
 
-	return 0;
+        return 0;
 }
 
-
 int
 afr_sh_data_trim_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, struct stat *buf)
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
 {
-	afr_private_t * priv = NULL;
-	afr_local_t * local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1)
-			gf_log (this->name, GF_LOG_ERROR,
-				"ftruncate of %s on subvolume %s failed (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-		else
-			gf_log (this->name, GF_LOG_DEBUG,
-				"ftruncate of %s on subvolume %s completed",
-				local->loc.path,
-				priv->children[child_index]->name);
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		afr_sh_data_erase_pending (frame, this);
-	}
-
-	return 0;
+        int              call_count = 0;
+        int              child_index = 0;
+        afr_private_t    *priv = NULL;
+        afr_local_t      *local  = NULL;
+        afr_self_heal_t  *sh = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+        sh    = &local->self_heal;
+
+        child_index = (long) cookie;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "ftruncate of %s on subvolume %s failed (%s)",
+                                local->loc.path,
+                                priv->children[child_index]->name,
+                                strerror (op_errno));
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                } else {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "ftruncate of %s on subvolume %s completed",
+                                local->loc.path,
+                                priv->children[child_index]->name);
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC))
+                        afr_sh_data_fail (frame, this);
+                else
+                        afr_sh_data_sync_prepare (frame, this);
+        }
+
+        return 0;
 }
 
 
 int
 afr_sh_data_trim_sinks (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv = NULL;
-	afr_local_t * local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int             *sources = NULL;
-	int              call_count = 0;
-	int              i = 0;
+        afr_private_t * priv = NULL;
+        afr_local_t * local  = NULL;
+        afr_self_heal_t *sh  = NULL;
+        int             *sources = NULL;
+        int              call_count = 0;
+        int              i = 0;
 
 
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
 
-	sources = sh->sources;
-	call_count = sh->active_sinks;
+        sources = sh->sources;
+        call_count = sh->active_sinks;
 
-	local->call_count = call_count;
+        local->call_count = call_count;
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (sources[i] || !local->child_up[i])
-			continue;
+        for (i = 0; i < priv->child_count; i++) {
+                if (sources[i] || !local->child_up[i])
+                        continue;
 
-		STACK_WIND_COOKIE (frame, afr_sh_data_trim_cbk,
-				   (void *) (long) i,
-				   priv->children[i], 
-				   priv->children[i]->fops->ftruncate,
-				   sh->healing_fd, sh->file_size); 
+                STACK_WIND_COOKIE (frame, afr_sh_data_trim_cbk,
+                                   (void *) (long) i,
+                                   priv->children[i],
+                                   priv->children[i]->fops->ftruncate,
+                                   sh->healing_fd, sh->file_size,
+                                   NULL);
 
-		if (!--call_count)
-			break;
-	}
+                if (!--call_count)
+                        break;
+        }
 
-	return 0;
+        return 0;
 }
 
-
 int
-afr_sh_data_read_write_iter (call_frame_t *frame, xlator_t *this);
+afr_sh_inode_set_read_ctx (afr_self_heal_t *sh, xlator_t *this)
+{
+        afr_private_t   *priv = NULL;
+        int             ret = 0;
+        int             i = 0;
+
+        priv = this->private;
+        sh->source = afr_sh_select_source (sh->sources, priv->child_count);
+        if (sh->source < 0) {
+                ret = -1;
+                goto out;
+        }
+
+        /* detect changes not visible through pending flags -- JIC */
+        for (i = 0; i < priv->child_count; i++) {
+                if (i == sh->source || sh->child_errno[i])
+                        continue;
+
+                if (SIZE_DIFFERS (&sh->buf[i], &sh->buf[sh->source]))
+                        sh->sources[i] = 0;
+        }
+
+        afr_reset_children (sh->fresh_children, priv->child_count);
+        afr_get_fresh_children (sh->success_children, sh->sources,
+                                sh->fresh_children, priv->child_count);
+        afr_inode_set_read_ctx (this, sh->inode, sh->source,
+                                sh->fresh_children);
+out:
+        return ret;
+}
 
-int
-afr_sh_data_write_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		       int32_t op_ret, int32_t op_errno, struct stat *buf)
+char*
+afr_get_sizes_str (afr_local_t *local, struct iatt *bufs, xlator_t *this)
 {
-	afr_private_t * priv = NULL;
-	afr_local_t * local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-
-	int child_index = (long) cookie;
-	int call_count = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	gf_log (this->name, GF_LOG_DEBUG, 
-		"wrote %d bytes of data from %s to child %d, offset %"PRId64"", 
-		op_ret, local->loc.path, child_index, sh->offset - op_ret);
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"write to %s failed on subvolume %s (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-			sh->op_failed = 1;
-		}
-	}
-	UNLOCK (&frame->lock);
+        afr_private_t *priv = NULL;
+        int           i     = 0;
+        char          num[1024] = {0};
+        size_t        len = 0;
+        char          *sizes_str = NULL;
+        size_t        off = 0;
+        char          *fmt_str = "%llu bytes on %s, ";
+        char          *child_down =  " %s,";
+        char          *child_unknown = " %s,";
+        int           down_child_present = 0;
+        int           down_count = 0;
+        int           unknown_count = 0;
+        int           unknown_child_present = 0;
+        char          *down_subvol_1 = " down subvolume is ";
+        char          *unknown_subvol_1 = " unknown subvolume is ";
+        char          *down_subvol_2 = " down subvolumes are ";
+        char          *unknown_subvol_2 = " unknown subvolumes are ";
+
+        priv = this->private;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == 1) {
+                        len += snprintf (num, sizeof (num), fmt_str,
+                                         (unsigned long long) bufs[i].ia_size,
+                                         priv->children[i]->name);
+                } else if (local->child_up[i] == 0) {
+                        len += snprintf (num, sizeof (num), child_down,
+                                         priv->children[i]->name);
+                        if (!down_child_present)
+                                down_child_present = 1;
+                        down_count ++;
+                } else if (local->child_up[i] == -1) {
+                        len += snprintf (num, sizeof (num), child_unknown,
+                                         priv->children[i]->name);
+                        if (!unknown_child_present)
+                                unknown_child_present = 1;
+                        unknown_count++;
+                }
+
+        }
+
+        if (down_child_present) {
+                if (down_count > 1)
+                        len += snprintf (num, sizeof (num), "%s",
+                                         down_subvol_2);
+                else
+                        len += snprintf (num, sizeof (num), "%s",
+                                        down_subvol_1);
+        }
+        if (unknown_child_present) {
+                if (unknown_count > 1)
+                        len += snprintf (num, sizeof (num), "%s",
+                                         unknown_subvol_2);
+                else
+                        len += snprintf (num, sizeof (num), "%s",
+                                         unknown_subvol_1);
+        }
+
+        len++;//for '\0'
+
+        sizes_str = GF_CALLOC (len, sizeof (char), gf_common_mt_char);
+
+        if (!sizes_str)
+                return NULL;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == 1) {
+                        off += snprintf (sizes_str + off, len - off, fmt_str,
+                                         (unsigned long long) bufs[i].ia_size,
+                                         priv->children[i]->name);
+                }
+        }
+
+        if (down_child_present) {
+                if (down_count > 1) {
+                        off += snprintf (sizes_str + off, len - off, "%s",
+                                         down_subvol_2);
+                } else {
+                        off += snprintf (sizes_str + off, len - off, "%s",
+                                         down_subvol_1);
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == 0) {
+                        off += snprintf (sizes_str + off, len - off, child_down,
+                                         priv->children[i]->name);
+                }
+        }
+
+        if (unknown_child_present) {
+                if (unknown_count > 1) {
+                        off += snprintf (sizes_str + off, len - off, "%s",
+                                        unknown_subvol_2);
+                } else {
+                        off += snprintf (sizes_str + off, len - off, "%s",
+                                         unknown_subvol_1);
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == -1) {
+                        off += snprintf (sizes_str + off, len - off,
+                                         child_unknown,
+                                         priv->children[i]->name);
+
+                }
+        }
+
+        return sizes_str;
+}
 
-	call_count = afr_frame_return (frame);
+char*
+afr_get_sinks_str (xlator_t *this, afr_local_t *local, afr_self_heal_t *sh)
+{
+        afr_private_t   *priv = NULL;
+        int             i = 0;
+        char            num[1024] = {0};
+        size_t          len = 0;
+        char            *sinks_str = NULL;
+        char            *temp_str = " to sinks ";
+        char            *str_format = " %s,";
+        char            off = 0;
 
-	if (call_count == 0) {
-		afr_sh_data_read_write_iter (frame, this);
-	}
+        priv = this->private;
 
-	return 0;
-}
+        len += snprintf (num, sizeof (num), "%s", temp_str);
+        for (i = 0; i < priv->child_count; i++) {
+                if ((sh->sources[i] == 0) && (local->child_up[i] == 1)) {
+                        len += snprintf (num, sizeof (num), str_format,
+                                         priv->children[i]->name);
+                }
+        }
 
+        len ++;
 
-int
-afr_sh_data_read_cbk (call_frame_t *frame, void *cookie,
-		      xlator_t *this, int32_t op_ret, int32_t op_errno,
-		      struct iovec *vector, int32_t count, struct stat *buf)
-{
-	afr_private_t * priv = NULL;
-	afr_local_t * local  = NULL;
-	afr_self_heal_t *sh  = NULL;
+        sinks_str = GF_CALLOC (len, sizeof (char), gf_common_mt_char);
 
-	int child_index = (long) cookie;
-	int i = 0;
-	int call_count = 0;
+        if (!sinks_str)
+                return NULL;
 
-	off_t offset;
+        off += snprintf (sinks_str + off, len - off, "%s", temp_str);
 
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
+        for (i = 0; i < priv->child_count; i++) {
+                if ((sh->sources[i] == 0) && (local->child_up[i] == 1)) {
+                        off += snprintf (sinks_str + off, len - off,
+                                         str_format,
+                                         priv->children[i]->name);
+                }
+        }
 
-	call_count = sh->active_sinks;
+        return sinks_str;
 
-	local->call_count = call_count;
+}
 
-	gf_log (this->name, GF_LOG_DEBUG, 
-		"read %d bytes of data from %s on child %d, offset %"PRId64"",
-		op_ret, local->loc.path, child_index, sh->offset);
 
-	if (op_ret <= 0) {
-		afr_sh_data_trim_sinks (frame, this);
-		return 0;
-	}
+void
+afr_set_data_sh_info_str (afr_local_t *local, afr_self_heal_t *sh, xlator_t *this)
+{
+        char            *pending_matrix_str = NULL;
+        char            *sizes_str = NULL;
+        char            *sinks_str = NULL;
+        afr_private_t   *priv = NULL;
 
-	/* what if we read less than block size? */
-	offset = sh->offset;
-	sh->offset += op_ret;
+        priv = this->private;
 
-	frame->root->req_refs = frame->root->rsp_refs;
+        pending_matrix_str = afr_get_pending_matrix_str (sh->pending_matrix,
+                                                         this);
+        if (!pending_matrix_str)
+                pending_matrix_str = "";
 
-	if (sh->file_has_holes) {
-		if (iov_0filled (vector, count) == 0) {
-			/* the iter function depends on the
-			   sh->offset already being updated 
-			   above
-			*/
-			afr_sh_data_read_write_iter (frame, this);
-			goto out;
-		}
-	}
+        sizes_str = afr_get_sizes_str (local, sh->buf, this);
+        if (!sizes_str)
+                sizes_str = "";
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->sources[i] || !local->child_up[i])
-			continue;
-
-		/* this is a sink, so write to it */
-		STACK_WIND_COOKIE (frame, afr_sh_data_write_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->writev,
-				   sh->healing_fd, vector, count, offset);
-
-		if (!--call_count)
-			break;
-	}
+        sinks_str = afr_get_sinks_str (this, local, sh);
+        if (!sinks_str)
+                sinks_str = "";
 
-out:
-	return 0;
+        gf_asprintf (&sh->data_sh_info, " data self heal from %s %s with "
+                     "%s data %s", priv->children[sh->source]->name, sinks_str,
+                     sizes_str, pending_matrix_str);
+
+        if (pending_matrix_str && strcmp (pending_matrix_str, ""))
+                GF_FREE (pending_matrix_str);
+
+        if (sizes_str && strcmp (sizes_str, ""))
+                GF_FREE (sizes_str);
 }
 
+void
+afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
+{
+        int              source = 0;
+        afr_local_t     *local      = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        source     = sh->source;
+        sh->block_size = this->ctx->page_size;
+        sh->file_size  = sh->buf[source].ia_size;
+
+        if (FILE_HAS_HOLES (&sh->buf[source]))
+                sh->file_has_holes = 1;
+
+        if (sh->background && sh->unwind && !sh->unwound) {
+                sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno,
+                            is_self_heal_failed (sh, AFR_CHECK_SPECIFIC));
+                sh->unwound = _gf_true;
+        }
+
+        afr_sh_mark_source_sinks (frame, this);
+        if (sh->active_sinks == 0) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "no active sinks for performing self-heal on file %s",
+                        local->loc.path);
+                afr_sh_data_finish (frame, this);
+                return;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "self-healing file %s from subvolume %s to %d other",
+                local->loc.path, priv->children[sh->source]->name,
+                sh->active_sinks);
+
+        sh->actual_sh_started = _gf_true;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN);
+        afr_sh_data_trim_sinks (frame, this);
+}
 
 int
-afr_sh_data_read_write (call_frame_t *frame, xlator_t *this)
+afr_sh_data_fxattrop_fstat_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv = NULL;
-	afr_local_t * local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	STACK_WIND_COOKIE (frame, afr_sh_data_read_cbk,
-			   (void *) (long) sh->source,
-			   priv->children[sh->source],
-			   priv->children[sh->source]->fops->readv,
-			   sh->healing_fd, sh->block_size,
-			   sh->offset);
-
-	return 0;
+        afr_local_t     *local      = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              nsources = 0;
+        int              ret = 0;
+        int             *old_sources = NULL;
+        int             tstamp_source = 0;
+        int             i = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        gf_log (this->name, GF_LOG_DEBUG, "Pending matrix for: %s",
+                lkowner_utoa (&frame->root->lk_owner));
+        if (sh->sync_done) {
+                //store sources before sync so that mtime can be set using the
+                //iatt buf from one of them.
+                old_sources = alloca (priv->child_count*sizeof (*old_sources));
+                memcpy (old_sources, sh->sources,
+                        priv->child_count * sizeof (*old_sources));
+        }
+
+        nsources = afr_build_sources (this, sh->xattr, sh->buf, sh->pending_matrix,
+                                      sh->sources, sh->success_children,
+                                      AFR_DATA_TRANSACTION, NULL, _gf_true);
+        if ((nsources == -1)
+            && (priv->favorite_child != -1)
+            && (sh->child_errno[priv->favorite_child] == 0)) {
+
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Picking favorite child %s as authentic source to "
+                        "resolve conflicting data of %s",
+                        priv->children[priv->favorite_child]->name,
+                        local->loc.path);
+
+                sh->sources[priv->favorite_child] = 1;
+
+                nsources = afr_sh_source_count (sh->sources,
+                                                priv->child_count);
+        }
+
+        if (nsources == -1) {
+                afr_sh_print_split_brain_log (sh->pending_matrix, this,
+                                              local->loc.path);
+                afr_set_split_brain (this, sh->inode, DONT_KNOW, SPB);
+
+                afr_sh_data_fail (frame, this);
+                return 0;
+        }
+
+        afr_set_split_brain (this, sh->inode, DONT_KNOW, NO_SPB);
+
+        ret = afr_sh_inode_set_read_ctx (sh, this);
+        if (ret) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "No active sources found.");
+
+                afr_sh_data_fail (frame, this);
+                return 0;
+        }
+
+        if (sh->sync_done) {
+                /* Perform setattr from one of the old_sources if possible
+                 * Because only they have the correct mtime, the new sources
+                 * (i.e. old sinks) have mtime from last writev in sync.
+                 */
+                tstamp_source = sh->source;
+                for (i = 0; i < priv->child_count; i++) {
+                        if (old_sources[i] && sh->sources[i])
+                                tstamp_source = i;
+                }
+                afr_sh_data_setattr (frame, this, &sh->buf[tstamp_source]);
+        } else {
+                afr_set_data_sh_info_str (local, sh, this);
+                if (nsources == 0) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "No self-heal needed for %s",
+                                local->loc.path);
+
+                        afr_sh_data_finish (frame, this);
+                        return 0;
+                }
+
+                if (sh->do_data_self_heal &&
+                    afr_data_self_heal_enabled (priv->data_self_heal))
+                        afr_sh_data_fix (frame, this);
+                else
+                        afr_sh_data_finish (frame, this);
+        }
+        return 0;
 }
 
-
 int
-afr_sh_data_read_write_iter (call_frame_t *frame, xlator_t *this)
+afr_lookup_select_read_child_by_txn_type (xlator_t *this, afr_local_t *local,
+                                          dict_t **xattr,
+                                          afr_transaction_type txn_type,
+                                          uuid_t gfid)
 {
-	afr_private_t * priv = NULL;
-	afr_local_t * local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
+        afr_private_t            *priv      = NULL;
+        int                      read_child = -1;
+        int32_t                  **pending_matrix = NULL;
+        int32_t                  *sources         = NULL;
+        int32_t                  *success_children   = NULL;
+        struct iatt              *bufs            = NULL;
+        int32_t                  nsources         = 0;
+        int32_t                  prev_read_child  = -1;
+        int32_t                  config_read_child = -1;
+        int32_t                  subvol_status = 0;
+
+        priv = this->private;
+        bufs = local->cont.lookup.bufs;
+        success_children = local->cont.lookup.success_children;
+
+        pending_matrix = local->cont.lookup.pending_matrix;
+        sources = local->cont.lookup.sources;
+        memset (sources, 0, sizeof (*sources) * priv->child_count);
+
+        nsources = afr_build_sources (this, xattr, bufs, pending_matrix,
+                                      sources, success_children, txn_type,
+                                      &subvol_status, _gf_false);
+        if (subvol_status & SPLIT_BRAIN) {
+                gf_log (this->name, GF_LOG_DEBUG, "%s: Possible split-brain",
+                        local->loc.path);
+                switch (txn_type) {
+                case AFR_DATA_TRANSACTION:
+                        local->cont.lookup.possible_spb = _gf_true;
+                        nsources = 1;
+                        sources[success_children[0]] = 1;
+                        break;
+                case AFR_ENTRY_TRANSACTION:
+                        read_child = afr_get_no_xattr_dir_read_child (this,
+                                                             success_children,
+                                                             bufs);
+                        sources[read_child] = 1;
+                        nsources = 1;
+                        break;
+                default:
+                        break;
+                }
+        }
+        if (nsources < 0)
+                goto out;
+
+        prev_read_child = local->read_child_index;
+        config_read_child = priv->read_child;
+        read_child = afr_select_read_child_from_policy (success_children,
+                                                        priv->child_count,
+                                                        prev_read_child,
+                                                        config_read_child,
+                                                        sources,
+                                                        priv->hash_mode, gfid);
+out:
+        gf_log (this->name, GF_LOG_DEBUG, "returning read_child: %d",
+                read_child);
+        return read_child;
+}
 
-	if (sh->op_failed) {
-		afr_sh_data_finish (frame, this);
-		goto out;
-	}
+int
+afr_sh_data_fstat_cbk (call_frame_t *frame, void *cookie,
+                       xlator_t *this, int32_t op_ret, int32_t op_errno,
+                       struct iatt *buf, dict_t *xdata)
+{
+        afr_private_t   *priv  = NULL;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        int call_count  = -1;
+        int child_index = (long) cookie;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret != -1) {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "fstat of %s on %s succeeded",
+                                local->loc.path,
+                                priv->children[child_index]->name);
+
+                        sh->buf[child_index] = *buf;
+                        sh->success_children[sh->success_count] = child_index;
+                        sh->success_count++;
+                } else {
+                        gf_log (this->name, GF_LOG_ERROR, "%s: fstat failed "
+                                "on %s, reason %s", local->loc.path,
+                                priv->children[child_index]->name,
+                                strerror (op_errno));
+                        sh->child_errno[child_index] = op_errno;
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                /* Previous versions of glusterfs might have set
+                 * the pending data xattrs which need to be erased
+                 */
+                if (!afr_sh_data_proceed (sh->success_count)) {
+                        gf_log (this->name, GF_LOG_ERROR, "inspecting metadata "
+                                "succeeded on < %d children, aborting "
+                                "self-heal for %s", AFR_SH_MIN_PARTICIPANTS,
+                                local->loc.path);
+                        afr_sh_data_fail (frame, this);
+                        goto out;
+                }
+                afr_sh_data_fxattrop_fstat_done (frame, this);
+        }
+out:
+        return 0;
+}
 
-	if (sh->offset >= sh->file_size) {
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"closing fd's of %s",
-			local->loc.path);
-		afr_sh_data_trim_sinks (frame, this);
 
-		goto out;
-	}
+int
+afr_sh_data_fstat (call_frame_t *frame, xlator_t *this)
+{
+        afr_self_heal_t *sh    = NULL;
+        afr_local_t     *local = NULL;
+        afr_private_t   *priv  = NULL;
+        int             call_count = 0;
+        int             i = 0;
+        int             child = 0;
+        int32_t         *fstat_children = NULL;
+
+        priv  = this->private;
+        local = frame->local;
+        sh    = &local->self_heal;
+
+        fstat_children = memdup (sh->success_children,
+                                 sizeof (*fstat_children) * priv->child_count);
+        if (!fstat_children) {
+                afr_sh_data_fail (frame, this);
+                goto out;
+        }
+        call_count = sh->success_count;
+        local->call_count = call_count;
+
+        memset (sh->buf, 0, sizeof (*sh->buf) * priv->child_count);
+        afr_reset_children (sh->success_children, priv->child_count);
+        sh->success_count = 0;
+        for (i = 0; i < priv->child_count; i++) {
+                child = fstat_children[i];
+                if (child == -1)
+                        break;
+                STACK_WIND_COOKIE (frame, afr_sh_data_fstat_cbk,
+                                   (void *) (long) child,
+                                   priv->children[child],
+                                   priv->children[child]->fops->fstat,
+                                   sh->healing_fd, NULL);
+                --call_count;
+        }
+        GF_ASSERT (!call_count);
+out:
+        GF_FREE (fstat_children);
+        return 0;
+}
 
-	afr_sh_data_read_write (frame, this);
+void
+afr_sh_common_fxattrop_resp_handler (call_frame_t *frame, void *cookie,
+                                     xlator_t *this, int32_t op_ret,
+                                     int32_t op_errno, dict_t *xattr)
+{
+        afr_private_t   *priv  = NULL;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        int child_index = (long) cookie;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret != -1) {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "fxattrop of %s on %s succeeded",
+                                local->loc.path,
+                                priv->children[child_index]->name);
+
+                        sh->xattr[child_index] = dict_ref (xattr);
+                        sh->success_children[sh->success_count] = child_index;
+                        sh->success_count++;
+                } else {
+                        gf_log (this->name, GF_LOG_ERROR, "fxattrop of %s "
+                                "failed on %s, reason %s", local->loc.path,
+                                priv->children[child_index]->name,
+                                strerror (op_errno));
+                        sh->child_errno[child_index] = op_errno;
+                }
+        }
+        UNLOCK (&frame->lock);
+}
 
+int
+afr_sh_data_fxattrop_cbk (call_frame_t *frame, void *cookie,
+                          xlator_t *this, int32_t op_ret, int32_t op_errno,
+                          dict_t *xattr, dict_t *xdata)
+{
+        int             call_count  = -1;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh     = NULL;
+
+        local = frame->local;
+        sh    = &local->self_heal;
+
+        afr_sh_common_fxattrop_resp_handler (frame, cookie, this, op_ret,
+                                             op_errno, xattr);
+
+        call_count = afr_frame_return (frame);
+        if (call_count == 0) {
+                if (!afr_sh_data_proceed (sh->success_count)) {
+                        gf_log (this->name, GF_LOG_ERROR, "%s, inspecting "
+                                "change log succeeded on < %d children",
+                                local->loc.path, AFR_SH_MIN_PARTICIPANTS);
+                        afr_sh_data_fail (frame, this);
+                        goto out;
+                }
+                afr_sh_data_fstat (frame, this);
+        }
 out:
-	return 0;
+        return 0;
 }
 
 
 int
-afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, fd_t *fd)
+afr_sh_data_fxattrop (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	/* TODO: some of the open's might fail.
-	   In that case, modify cleanup fn to send flush on those 
-	   fd's which are already open */
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"open of %s failed on child %s (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-			sh->op_failed = 1;
-		}
+        afr_self_heal_t *sh    = NULL;
+        afr_local_t     *local = NULL;
+        afr_private_t   *priv  = NULL;
+        dict_t          **xattr_req;
+        int32_t         *zero_pending = NULL;
+        int call_count = 0;
+        int i = 0;
+        int ret = 0;
+	int j;
+
+        priv  = this->private;
+        local = frame->local;
+        sh    = &local->self_heal;
+
+        call_count = afr_up_children_count (local->child_up,
+                                            priv->child_count);
+
+        local->call_count = call_count;
+
+	xattr_req = GF_CALLOC(priv->child_count, sizeof(struct dict_t *),
+			      gf_afr_mt_dict_t);
+	if (!xattr_req)
+		goto out;
 
+	for (i = 0; i < priv->child_count; i++) {
+		xattr_req[i] = dict_new();
+		if (!xattr_req[i]) {
+			ret = -1;
+			goto out;
+		}
 	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
 
-	if (call_count == 0) {
-		if (sh->op_failed) {
-			afr_sh_data_finish (frame, this);
-			return 0;
+	for (i = 0; i < priv->child_count; i++) {
+		for (j = 0; j < priv->child_count; j++) {
+			zero_pending = GF_CALLOC (3, sizeof (*zero_pending),
+						  gf_afr_mt_int32_t);
+			if (!zero_pending) {
+				ret = -1;
+				goto out;
+			}
+			ret = dict_set_dynptr (xattr_req[i], priv->pending_key[j],
+					       zero_pending,
+					       3 * sizeof (*zero_pending));
+			if (ret < 0) {
+				gf_log (this->name, GF_LOG_WARNING,
+					"Unable to set dict value");
+				goto out;
+			} else {
+				zero_pending = NULL;
+			}
 		}
-		gf_log (this->name, GF_LOG_DEBUG,
-			"fd for %s opened, commencing sync",
-			local->loc.path);
+	}
 
-		gf_log (this->name, GF_LOG_WARNING,
-			"sourcing file %s from %s to other sinks",
-			local->loc.path, priv->children[sh->source]->name);
+        afr_reset_xattr (sh->xattr, priv->child_count);
+        afr_reset_children (sh->success_children, priv->child_count);
+        memset (sh->child_errno, 0,
+                sizeof (*sh->child_errno) * priv->child_count);
+        sh->success_count = 0;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i]) {
+                        STACK_WIND_COOKIE (frame, afr_sh_data_fxattrop_cbk,
+                                           (void *) (long) i,
+                                           priv->children[i],
+                                           priv->children[i]->fops->fxattrop,
+                                           sh->healing_fd, GF_XATTROP_ADD_ARRAY,
+                                           xattr_req[i], NULL);
+
+                        if (!--call_count)
+                                break;
+                }
+        }
 
-		afr_sh_data_read_write (frame, this);
+out:
+	if (xattr_req) {
+		for (i = 0; i < priv->child_count; i++)
+			if (xattr_req[i])
+				dict_unref(xattr_req[i]);
+		GF_FREE(xattr_req);
 	}
 
-	return 0;
-}
+        if (ret) {
+                GF_FREE (zero_pending);
+                afr_sh_data_fail (frame, this);
+        }
 
+        return 0;
+}
 
 int
-afr_sh_data_open (call_frame_t *frame, xlator_t *this)
+afr_sh_data_big_lock_success (call_frame_t *frame, xlator_t *this)
 {
-	int i = 0;				
-	int call_count = 0;		     
+        afr_local_t   *local = NULL;
+        afr_self_heal_t *sh = NULL;
 
-	int source = -1;
-	int *sources = NULL;
+        local = frame->local;
+        sh = &local->self_heal;
 
-	fd_t *fd = NULL;
+        sh->data_lock_held = _gf_true;
+        afr_sh_data_fxattrop (frame, this);
+        return 0;
+}
 
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	afr_self_heal_t *sh = NULL;
+int
+afr_sh_dom_lock_success (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local = NULL;
+        afr_self_heal_t *sh = NULL;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
 
-	call_count = sh->active_sinks + 1;
-	local->call_count = call_count;
+        sh->sh_dom_lock_held = _gf_true;
+        afr_sh_data_lock (frame, this, 0, 0, _gf_true, this->name,
+                          afr_sh_data_big_lock_success,
+                          afr_sh_data_fail);
+        return 0;
+}
 
-	fd = fd_create (local->loc.inode, frame->root->pid);
-	sh->healing_fd = fd;
+int
+afr_sh_data_post_blocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+        afr_self_heal_t     *sh       = NULL;
 
-	source  = local->self_heal.source;
-	sources = local->self_heal.sources;
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        sh       = &local->self_heal;
 
-	sh->block_size = 65536;
-	sh->file_size  = sh->buf[source].st_size;
+        if (int_lock->lock_op_ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR, "Blocking data inodelks "
+                        "failed for %s. by %s",
+                        local->loc.path, lkowner_utoa (&frame->root->lk_owner));
 
-	if (FILE_HAS_HOLES (&sh->buf[source]))
-		sh->file_has_holes = 1;
+                sh->data_lock_failure_handler (frame, this);
+        } else {
 
-	/* open source */
-	STACK_WIND_COOKIE (frame, afr_sh_data_open_cbk,
-			   (void *) (long) source,
-			   priv->children[source],
-			   priv->children[source]->fops->open,
-			   &local->loc, O_RDONLY|O_LARGEFILE, fd);
-	call_count--;
+                gf_log (this->name, GF_LOG_DEBUG, "Blocking data inodelks "
+                        "done for %s by %s. Proceding to self-heal",
+                        local->loc.path, lkowner_utoa (&frame->root->lk_owner));
 
-	/* open sinks */
-	for (i = 0; i < priv->child_count; i++) {
-		if(sources[i] || !local->child_up[i])
-			continue;
-
-		STACK_WIND_COOKIE (frame, afr_sh_data_open_cbk,
-				   (void *) (long) i,
-				   priv->children[i], 
-				   priv->children[i]->fops->open,
-				   &local->loc, 
-				   O_WRONLY|O_LARGEFILE, fd); 
-
-		if (!--call_count)
-			break;
-	}
+                sh->data_lock_success_handler (frame, this);
+        }
 
-	return 0;
+        return 0;
 }
 
-
 int
-afr_sh_data_sync_prepare (call_frame_t *frame, xlator_t *this)
+afr_sh_data_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              active_sinks = 0;
-	int              source = 0;
-	int              i = 0;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+        afr_self_heal_t     *sh       = NULL;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        sh       = &local->self_heal;
 
-	source = sh->source;
+        if (int_lock->lock_op_ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking data inodelks "
+                        "failed for %s. by %s",
+                        local->loc.path, lkowner_utoa (&frame->root->lk_owner));
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->sources[i] == 0 && local->child_up[i] == 1) {
-			active_sinks++;
-			sh->success[i] = 1;
+		if (!sh->data_lock_block) {
+			sh->data_lock_failure_handler(frame, this);
+		} else {
+			int_lock->lock_cbk =
+				afr_sh_data_post_blocking_inodelk_cbk;
+			afr_blocking_lock (frame, this);
 		}
-	}
-	sh->success[source] = 1;
-
-	if (active_sinks == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no active sinks for performing self-heal on file %s",
-			local->loc.path);
-		afr_sh_data_finish (frame, this);
-		return 0;
-	}
-	sh->active_sinks = active_sinks;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"syncing data of %s from subvolume %s to %d active sinks",
-		local->loc.path, priv->children[source]->name, active_sinks);
+        } else {
 
-	afr_sh_data_open (frame, this);
+                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking data inodelks "
+                        "done for %s by %s. Proceeding to self-heal",
+                        local->loc.path, lkowner_utoa (&frame->root->lk_owner));
+                sh->data_lock_success_handler (frame, this);
+        }
 
-	return 0;
+        return 0;
 }
 
-
 int
-afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
+afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this, char *dom,
+                      off_t start, off_t len)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              nsources = 0;
-	int              source = 0;
-	int              i = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	afr_sh_build_pending_matrix (sh->pending_matrix, sh->xattr, 
-				     priv->child_count, AFR_DATA_PENDING);
-
-	afr_sh_print_pending_matrix (sh->pending_matrix, this);
-
-
-	afr_sh_mark_sources (sh->pending_matrix, sh->sources, 
-			     priv->child_count);
+        afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
+        afr_local_t         *local    = NULL;
 
-	afr_sh_supress_empty_children (sh->sources, sh->xattr, sh->buf,
-				       priv->child_count, AFR_DATA_PENDING);
+        local    = frame->local;
+        int_lock = &local->internal_lock;
 
-	afr_sh_supress_errenous_children (sh->sources, sh->child_errno,
-					  priv->child_count);
+        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
+        int_lock->selfheal_lk_type    = AFR_DATA_SELF_HEAL_LK;
 
-	nsources = afr_sh_source_count (sh->sources, priv->child_count);
+        afr_set_lock_number (frame, this);
 
-	if ((nsources == 0)
-	    && (priv->favorite_child != -1)
-	    && (sh->child_errno[priv->favorite_child] == 0)) {
+        int_lock->lock_cbk         = afr_sh_data_post_nonblocking_inodelk_cbk;
 
-		gf_log (this->name, GF_LOG_WARNING,
-			"Picking favorite child %s as authentic source to resolve conflicting data of %s",
-			priv->children[priv->favorite_child]->name,
-			local->loc.path);
+        int_lock->domain = dom;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+        inodelk->flock.l_start = start;
+        inodelk->flock.l_len   = len;
+        inodelk->flock.l_type  = F_WRLCK;
 
-		sh->sources[priv->favorite_child] = 1;
+        afr_nonblocking_inodelk (frame, this);
 
-		nsources = afr_sh_source_count (sh->sources,
-						priv->child_count);
-	}
-
-	if (nsources == 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Unable to resolve conflicting data of %s. "
-			"Please resolve manually by deleting the file %s "
-			"from all but the preferred subvolume. "
-			"Please consider 'option favorite-child <>'",
-			local->loc.path, local->loc.path);
-
-		local->govinda_gOvinda = 1;
-
-		afr_sh_data_finish (frame, this);
-		return 0;
-	}
-
-	source = afr_sh_select_source (sh->sources, priv->child_count);
-	sh->source = source;
-
-	/* detect changes not visible through pending flags -- JIC */
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == source || sh->child_errno[i])
-			continue;
-
-		if (SIZE_DIFFERS (&sh->buf[i], &sh->buf[source]))
-			sh->sources[i] = 0;
-	}
-
-	afr_sh_data_sync_prepare (frame, this);
-
-	return 0;
+        return 0;
 }
 
-
 int
-afr_sh_data_lookup_cbk (call_frame_t *frame, void *cookie,
-			xlator_t *this, int32_t op_ret, int32_t op_errno,
-			inode_t *inode, struct stat *buf, dict_t *xattr)
+afr_post_sh_big_lock_success (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t   *priv  = NULL;
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-
-	int call_count  = -1;
-	int child_index = (long) cookie;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret != -1) {
-			sh->xattr[child_index] = dict_ref (xattr);
-			sh->buf[child_index] = *buf;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		afr_sh_data_fix (frame, this);
-	}
-
-	return 0;
+        afr_local_t     *local     = NULL;
+        afr_self_heal_t *sh        = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+
+        GF_ASSERT (sh->old_loop_frame);
+        sh_loop_finish (sh->old_loop_frame, this);
+        sh->old_loop_frame = NULL;
+        sh->data_lock_held = _gf_true;
+        sh->sync_done = _gf_true;
+        afr_sh_data_fxattrop (frame, this);
+        return 0;
 }
 
-
 int
-afr_sh_data_lookup (call_frame_t *frame, xlator_t *this)
+afr_post_sh_big_lock_failure (call_frame_t *frame, xlator_t *this)
 {
-	afr_self_heal_t *sh    = NULL; 
-	afr_local_t     *local = NULL;
-	afr_private_t   *priv  = NULL;
-	dict_t          *xattr_req = NULL;
-
-	int call_count = 0;
-	int i = 0;
-	int ret = 0;
-
-	priv  = this->private;
-	local = frame->local;
-	sh    = &local->self_heal;
-
-	call_count = local->child_count;
+        afr_local_t     *local     = NULL;
+        afr_self_heal_t *sh        = NULL;
 
-	local->call_count = call_count;
-	
-	xattr_req = dict_new();
-	if (xattr_req)
-		ret = dict_set_uint64 (xattr_req, AFR_DATA_PENDING,
-				       priv->child_count * sizeof(int32_t));
+        local = frame->local;
+        sh = &local->self_heal;
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_sh_data_lookup_cbk,
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->lookup,
-					   &local->loc, xattr_req);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	if (xattr_req)
-		dict_unref (xattr_req);
-
-	return 0;
+        GF_ASSERT (sh->old_loop_frame);
+        sh_loop_finish (sh->old_loop_frame, this);
+        sh->old_loop_frame = NULL;
+        afr_sh_set_timestamps (frame, this);
+        return 0;
 }
 
 
 int
-afr_sh_data_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno)
+afr_sh_data_lock (call_frame_t *frame, xlator_t *this,
+                  off_t start, off_t len, gf_boolean_t block,
+                  char *dom, afr_lock_cbk_t success_handler,
+                  afr_lock_cbk_t failure_handler)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	int              call_count = 0;
-	int              child_index = (long) cookie;
-
-	/* TODO: what if lock fails? */
-	
-	local = frame->local;
-	sh = &local->self_heal;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			sh->op_failed = 1;
-
-			gf_log (this->name,
-				(op_errno == EAGAIN ? GF_LOG_DEBUG : GF_LOG_ERROR),
-				"locking of %s on child %d failed: %s",
-				local->loc.path, child_index,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"inode of %s on child %d locked",
-				local->loc.path, child_index);
-		}
-	}
-	UNLOCK (&frame->lock);
+        afr_local_t *   local = NULL;
+        afr_self_heal_t * sh  = NULL;
 
-	call_count = afr_frame_return (frame);
+        local = frame->local;
+        sh    = &local->self_heal;
 
-	if (call_count == 0) {
-		if (sh->op_failed) {
-			afr_sh_data_finish (frame, this);
-			return 0;
-		}
-
-		afr_sh_data_lookup (frame, this);
-	}
-
-	return 0;
+        sh->data_lock_success_handler = success_handler;
+        sh->data_lock_failure_handler = failure_handler;
+	sh->data_lock_block = block;
+        return afr_sh_data_lock_rec (frame, this, dom, start, len);
 }
 
-
 int
-afr_sh_data_lock (call_frame_t *frame, xlator_t *this)
+afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-	struct flock flock;			
-	int i = 0;				
-	int call_count = 0;		     
-
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	afr_self_heal_t * sh  = NULL;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	call_count = local->child_count;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              call_count = 0;
+        int              child_index = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        child_index = (long) cookie;
+
+        /* TODO: some of the open's might fail.
+           In that case, modify cleanup fn to send flush on those
+           fd's which are already open */
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "open of %s failed on child %s (%s)",
+                                local->loc.path,
+                                priv->children[child_index]->name,
+                                strerror (op_errno));
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                } else {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "open of %s succeeded on child %s",
+                                local->loc.path,
+                                priv->children[child_index]->name);
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
+                        afr_sh_data_fail (frame, this);
+                        return 0;
+                }
+
+                gf_log (this->name, GF_LOG_TRACE,
+                        "fd for %s opened, commencing sync",
+                        local->loc.path);
+
+                afr_sh_data_lock (frame, this, 0, 0, _gf_true, priv->sh_domain,
+                                  afr_sh_dom_lock_success, afr_sh_data_fail);
+        }
+
+        return 0;
+}
 
-	local->call_count = call_count;		
 
-	flock.l_start = 0;
-	flock.l_len   = 0;
-	flock.l_type  = F_WRLCK;			
+int
+afr_sh_data_open (call_frame_t *frame, xlator_t *this)
+{
+        int i = 0;
+        int call_count = 0;
+        fd_t *fd = NULL;
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        call_count = afr_up_children_count (local->child_up, priv->child_count);
+        local->call_count = call_count;
+
+        fd = fd_create (local->loc.inode, frame->root->pid);
+        sh->healing_fd = fd;
+
+        /* open sinks */
+        for (i = 0; i < priv->child_count; i++) {
+                if(!local->child_up[i])
+                        continue;
+
+                STACK_WIND_COOKIE (frame, afr_sh_data_open_cbk,
+                                   (void *) (long) i,
+                                   priv->children[i],
+                                   priv->children[i]->fops->open,
+                                   &local->loc,
+                                   O_RDWR|O_LARGEFILE, fd, NULL);
+
+                if (!--call_count)
+                        break;
+        }
+
+        return 0;
+}
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"locking %s on subvolume %s",
-				local->loc.path, priv->children[i]->name);
-
-			STACK_WIND_COOKIE (frame, afr_sh_data_lock_cbk,
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->inodelk,
-                                           this->name,
-					   &local->loc, F_SETLK, &flock); 
-			if (!--call_count)
-				break;
-		}
-	}
+void
+afr_sh_non_reg_fix (call_frame_t *frame, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno)
+{
+        afr_private_t   *priv = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_local_t     *local = NULL;
+        int             i = 0;
+
+        if (op_ret < 0) {
+                afr_sh_data_fail (frame, this);
+                return;
+        }
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        for (i = 0; i < priv->child_count ; i++) {
+                if (1 == local->child_up[i])
+                        sh->success[i] = 1;
+        }
+
+        afr_sh_erase_pending (frame, this, AFR_DATA_TRANSACTION,
+                              afr_sh_data_erase_pending_cbk,
+                              afr_sh_data_finish);
+}
 
-	return 0;
+int
+afr_sh_non_reg_lock_success (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        sh->data_lock_held = _gf_true;
+        afr_sh_common_lookup (frame, this, &local->loc,
+                              afr_sh_non_reg_fix, NULL,
+                              AFR_LOOKUP_FAIL_CONFLICTS |
+                              AFR_LOOKUP_FAIL_MISSING_GFIDS,
+                              NULL);
+        return 0;
 }
 
+gf_boolean_t
+afr_can_start_data_self_heal (afr_self_heal_t *sh, afr_private_t *priv)
+{
+        if (sh->force_confirm_spb)
+                return _gf_true;
+        if (sh->do_data_self_heal &&
+            afr_data_self_heal_enabled (priv->data_self_heal))
+                return _gf_true;
+        return _gf_false;
+}
 
 int
 afr_self_heal_data (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t   *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t *priv = this->private;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-
-	if (local->need_data_self_heal && priv->data_self_heal) {
-		afr_sh_data_lock (frame, this);
-	} else {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"not doing data self heal on %s",
-			local->loc.path);
-		afr_sh_data_done (frame, this);
-	}
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh    = NULL;
+        afr_private_t   *priv  = this->private;
+        int             ret    = -1;
+
+        local = frame->local;
+        sh = &local->self_heal;
+
+        sh->sh_type_in_action = AFR_SELF_HEAL_DATA;
+
+        if (afr_can_start_data_self_heal (sh, priv)) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
+                ret = afr_inodelk_init (&local->internal_lock.inodelk[1],
+                                        priv->sh_domain, priv->child_count);
+                if (ret < 0) {
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                        afr_sh_data_done (frame, this);
+                        return 0;
+                }
+
+                if (IA_ISREG (sh->type)) {
+                        afr_sh_data_open (frame, this);
+                } else {
+                        afr_sh_data_lock (frame, this, 0, 0, _gf_true,
+                                          this->name,
+                                          afr_sh_non_reg_lock_success,
+                                          afr_sh_data_fail);
+                }
+        } else {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "not doing data self heal on %s",
+                        local->loc.path);
+                afr_sh_data_done (frame, this);
+        }
+
+        return 0;
 }
-
diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c
index e7431249f..53491a1d7 100644
--- a/xlators/cluster/afr/src/afr-self-heal-entry.c
+++ b/xlators/cluster/afr/src/afr-self-heal-entry.c
@@ -1,20 +1,11 @@
 /*
-  Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  General Public License for more details.
-
-  You should have received a copy of the GNU General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include <libgen.h>
@@ -30,6 +21,7 @@
 #endif
 
 #include "glusterfs.h"
+#include "inode.h"
 #include "afr.h"
 #include "dict.h"
 #include "xlator.h"
@@ -48,1993 +40,2331 @@
 #include "afr-self-heal.h"
 #include "afr-self-heal-common.h"
 
-
+#define AFR_INIT_SH_FRAME_VALS(_frame, _local, _sh, _sh_frame, _sh_local, _sh_sh)\
+        do {\
+                _local = _frame->local;\
+                _sh = &_local->self_heal;\
+                _sh_frame = _sh->sh_frame;\
+                _sh_local = _sh_frame->local;\
+                _sh_sh    = &_sh_local->self_heal;\
+        } while (0);
 
 int
+afr_sh_entry_impunge_create_file (call_frame_t *impunge_frame, xlator_t *this,
+                                  int child_index);
+int
 afr_sh_entry_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	/* 
-	   TODO: cleanup sh->* 
-	*/
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"self heal of %s completed",
-		local->loc.path);
+        local = frame->local;
+        sh = &local->self_heal;
 
-	sh->completion_cbk (frame, this);
+        sh->completion_cbk (frame, this);
 
-	return 0;
-}
-
-
-int
-afr_sh_entry_unlck_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	int           call_count = 0;
-	int           child_index = (long) cookie;
-
-	/* TODO: what if lock fails? */
-	
-	local = frame->local;
-	sh = &local->self_heal;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"unlocking inode of %s on child %d failed: %s",
-				local->loc.path, child_index,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"unlocked inode of %s on child %d",
-				local->loc.path, child_index);
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		if (sh->healing_fd)
-			fd_unref (sh->healing_fd);
-		sh->healing_fd = NULL;
-		afr_sh_entry_done (frame, this);
-	}
-
-	return 0;
+        return 0;
 }
 
 
 int
 afr_sh_entry_unlock (call_frame_t *frame, xlator_t *this)
 {
-	int i = 0;				
-	int call_count = 0;		     
-
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	afr_self_heal_t * sh  = NULL;
+        afr_local_t         *local    = NULL;
+        afr_internal_lock_t *int_lock = NULL;
 
+        local    = frame->local;
+        int_lock = &local->internal_lock;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        int_lock->lock_cbk = afr_sh_entry_done;
+        afr_unlock (frame, this);
 
-	call_count = local->child_count;
-
-	local->call_count = call_count;		
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"unlocking %s on subvolume %s",
-				local->loc.path, priv->children[i]->name);
-
-			STACK_WIND_COOKIE (frame, afr_sh_entry_unlck_cbk,
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->entrylk,
-                                           this->name,
-					   &local->loc, NULL,
-					   ENTRYLK_UNLOCK, ENTRYLK_WRLCK);
-			if (!--call_count)
-				break;
-		}
-	}
-
-	return 0;
+        return 0;
 }
 
 
 int
 afr_sh_entry_finish (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t   *local = NULL;
+        afr_local_t   *local = NULL;
 
-	local = frame->local;
+        local = frame->local;
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"finishing entry selfheal of %s", local->loc.path);
+        gf_log (this->name, GF_LOG_TRACE,
+                "finishing entry selfheal of %s", local->loc.path);
 
-	afr_sh_entry_unlock (frame, this);
+        afr_sh_entry_unlock (frame, this);
 
-	return 0;
+        return 0;
 }
 
 
 int
 afr_sh_entry_erase_pending_cbk (call_frame_t *frame, void *cookie,
-				xlator_t *this, int32_t op_ret,
-				int32_t op_errno, dict_t *xattr)
+                                xlator_t *this, int32_t op_ret,
+                                int32_t op_errno, dict_t *xattr, dict_t *xdata)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int             call_count = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	LOCK (&frame->lock);
-	{
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		afr_sh_entry_finish (frame, this);
-
-	return 0;
+        long                 i          = 0;
+        int                  call_count = 0;
+        afr_local_t         *local      = NULL;
+        afr_self_heal_t     *sh         = NULL;
+        afr_local_t         *orig_local = NULL;
+        call_frame_t        *orig_frame = NULL;
+        afr_private_t       *priv       = NULL;
+        int32_t             read_child  = -1;
+
+        local = frame->local;
+        priv  = this->private;
+        sh = &local->self_heal;
+        i = (long)cookie;
+
+
+        afr_children_add_child (sh->fresh_children, i, priv->child_count);
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "%s: failed to erase pending xattrs on %s (%s)",
+                        local->loc.path, priv->children[i]->name,
+                        strerror (op_errno));
+        }
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                if (sh->source == -1) {
+                        //this happens if the forced merge option is set
+                        read_child = sh->fresh_children[0];
+                } else {
+                        read_child = sh->source;
+                }
+                afr_inode_set_read_ctx (this, sh->inode, read_child,
+                                        sh->fresh_children);
+                orig_frame = sh->orig_frame;
+                orig_local = orig_frame->local;
+
+                if (sh->source != -1) {
+                        orig_local->cont.lookup.buf.ia_nlink = sh->buf[sh->source].ia_nlink;
+                }
+
+                afr_sh_entry_finish (frame, this);
+        }
+
+        return 0;
 }
 
 
 int
 afr_sh_entry_erase_pending (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              i = 0;
-	dict_t          **erase_xattr = NULL;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-
-	afr_sh_pending_to_delta (sh->pending_matrix, sh->delta_matrix,
-				 sh->success, priv->child_count);
-
-	erase_xattr = CALLOC (sizeof (*erase_xattr), priv->child_count);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->xattr[i]) {
-			call_count++;
-
-			erase_xattr[i] = get_new_dict();
-			dict_ref (erase_xattr[i]);
-		}
-	}
-
-	afr_sh_delta_to_xattr (sh->delta_matrix, erase_xattr,
-			       priv->child_count, AFR_ENTRY_PENDING);
-
-	local->call_count = call_count;
-	for (i = 0; i < priv->child_count; i++) {
-		if (!erase_xattr[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"erasing pending flags from %s on %s",
-			local->loc.path, priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_entry_erase_pending_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->xattrop,
-				   &local->loc,
-				   GF_XATTROP_ADD_ARRAY, erase_xattr[i]);
-		if (!--call_count)
-			break;
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (erase_xattr[i]) {
-			dict_unref (erase_xattr[i]);
-		}
-	}
-	FREE (erase_xattr);
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+
+        if (sh->entries_skipped) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                goto out;
+        }
+        afr_sh_erase_pending (frame, this, AFR_ENTRY_TRANSACTION,
+                              afr_sh_entry_erase_pending_cbk,
+                              afr_sh_entry_finish);
+        return 0;
+out:
+        afr_sh_entry_finish (frame, this);
+        return 0;
 }
 
 
 
 static int
 next_active_source (call_frame_t *frame, xlator_t *this,
-		    int current_active_source)
+                    int current_active_source)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              source = -1;
-	int              next_active_source = -1;
-	int              i = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	source = sh->source;
-
-	if (source != -1) {
-		if (current_active_source != source)
-			next_active_source = source;
-		goto out;
-	}
-
-	/*
-	  the next active sink becomes the source for the
-	  'conservative decision' of merging all entries
-	*/
-
-	for (i = 0; i < priv->child_count; i++) {
-		if ((sh->sources[i] == 0)
-		    && (local->child_up[i] == 1)
-		    && (i > current_active_source)) {
-
-			next_active_source = i;
-			break;
-		}
-	}
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh  = NULL;
+        int              source = -1;
+        int              next_active_source = -1;
+        int              i = 0;
+
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
+
+        source = sh->source;
+
+        if (source != -1) {
+                if (current_active_source != source)
+                        next_active_source = source;
+                goto out;
+        }
+
+        /*
+          the next active sink becomes the source for the
+          'conservative decision' of merging all entries
+        */
+
+        for (i = 0; i < priv->child_count; i++) {
+                if ((sh->sources[i] == 0)
+                    && (local->child_up[i] == 1)
+                    && (i > current_active_source)) {
+
+                        next_active_source = i;
+                        break;
+                }
+        }
 out:
-	return next_active_source;
+        return next_active_source;
 }
 
 
 
 static int
 next_active_sink (call_frame_t *frame, xlator_t *this,
-		  int current_active_sink)
+                  int current_active_sink)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              next_active_sink = -1;
-	int              i = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	/*
-	  the next active sink becomes the source for the
-	  'conservative decision' of merging all entries
-	*/
-
-	for (i = 0; i < priv->child_count; i++) {
-		if ((sh->sources[i] == 0)
-		    && (local->child_up[i] == 1)
-		    && (i > current_active_sink)) {
-
-			next_active_sink = i;
-			break;
-		}
-	}
-
-	return next_active_sink;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh  = NULL;
+        int              next_active_sink = -1;
+        int              i = 0;
+
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
+
+        /*
+          the next active sink becomes the source for the
+          'conservative decision' of merging all entries
+        */
+
+        for (i = 0; i < priv->child_count; i++) {
+                if ((sh->sources[i] == 0)
+                    && (local->child_up[i] == 1)
+                    && (i > current_active_sink)) {
+
+                        next_active_sink = i;
+                        break;
+                }
+        }
+
+        return next_active_sink;
 }
 
-
 int
-build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name)
-{
-	int   ret = -1;
-
-	if (!child) {
-		goto out;
-	}
-
-	if (strcmp (parent->path, "/") == 0)
-		asprintf ((char **)&child->path, "/%s", name);
-	else
-		asprintf ((char **)&child->path, "%s/%s", parent->path, name);
-
-	if (!child->path) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
-
-	child->name = strrchr (child->path, '/');
-	if (child->name)
-		child->name++;
-
-	child->parent = inode_ref (parent->inode);
-	child->inode = inode_new (parent->inode->table);
-
-	if (!child->inode) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
-
-	ret = 0;
-out:
-	if (ret == -1)
-		loc_wipe (child);
-
-	return ret;
-}
+afr_sh_entry_impunge_all (call_frame_t *frame, xlator_t *this);
 
+int
+afr_sh_entry_impunge_subvol (call_frame_t *frame, xlator_t *this);
 
 int
 afr_sh_entry_expunge_all (call_frame_t *frame, xlator_t *this);
 
 int
 afr_sh_entry_expunge_subvol (call_frame_t *frame, xlator_t *this,
-			     int active_src);
+                             int active_src);
 
 int
 afr_sh_entry_expunge_entry_done (call_frame_t *frame, xlator_t *this,
-				 int active_src)
+                                 int active_src, int32_t op_ret,
+                                 int32_t op_errno)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              call_count = 0;
+        int              call_count = 0;
 
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
+        call_count = afr_frame_return (frame);
 
-	LOCK (&frame->lock);
-	{
-	}
-	UNLOCK (&frame->lock);
+        if (call_count == 0)
+                afr_sh_entry_expunge_subvol (frame, this, active_src);
 
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		afr_sh_entry_expunge_subvol (frame, this, active_src);
+        return 0;
+}
 
-	return 0;
+int
+afr_sh_entry_expunge_parent_setattr_cbk (call_frame_t *expunge_frame,
+                                         void *cookie, xlator_t *this,
+                                         int32_t op_ret, int32_t op_errno,
+                                         struct iatt *preop, struct iatt *postop,
+                                         dict_t *xdata)
+{
+        afr_private_t   *priv          = NULL;
+        afr_local_t     *expunge_local = NULL;
+        afr_self_heal_t *expunge_sh    = NULL;
+        call_frame_t    *frame         = NULL;
+        int              active_src    = (long) cookie;
+        afr_self_heal_t *sh            = NULL;
+        afr_local_t     *local         = NULL;
+
+        priv          = this->private;
+        expunge_local = expunge_frame->local;
+        expunge_sh    = &expunge_local->self_heal;
+        frame         = expunge_sh->sh_frame;
+        local         = frame->local;
+        sh            = &local->self_heal;
+
+        if (op_ret != 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "setattr on parent directory of %s on subvolume %s failed: %s",
+                        expunge_local->loc.path,
+                        priv->children[active_src]->name, strerror (op_errno));
+        }
+
+        AFR_STACK_DESTROY (expunge_frame);
+        sh->expunge_done (frame, this, active_src, op_ret, op_errno);
+
+        return 0;
 }
 
 
 int
 afr_sh_entry_expunge_remove_cbk (call_frame_t *expunge_frame, void *cookie,
-				 xlator_t *this,
-				 int32_t op_ret, int32_t op_errno)
+                                 xlator_t *this,
+                                 int32_t op_ret, int32_t op_errno,
+                                 struct iatt *preparent,
+                                 struct iatt *postparent, dict_t *xdata)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh = NULL;
-	int              active_src = 0;
-	call_frame_t    *frame = NULL;
-
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-	expunge_sh = &expunge_local->self_heal;
-	frame = expunge_sh->sh_frame;
-
-	active_src = (long) cookie;
-
-	if (op_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"removed %s on %s",
-			expunge_local->loc.path,
-			priv->children[active_src]->name);
-	} else {
-		gf_log (this->name, GF_LOG_ERROR,
-			"removing %s on %s failed (%s)",
-			expunge_local->loc.path,
-			priv->children[active_src]->name,
-			strerror (op_errno));
-	}
-
-	AFR_STACK_DESTROY (expunge_frame);
-	afr_sh_entry_expunge_entry_done (frame, this, active_src);
-
-	return 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *expunge_local = NULL;
+        afr_self_heal_t *expunge_sh = NULL;
+        int              active_src = 0;
+        int32_t          valid = 0;
+
+        priv = this->private;
+        expunge_local = expunge_frame->local;
+        expunge_sh = &expunge_local->self_heal;
+
+        active_src = (long) cookie;
+
+        if (op_ret == 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "removed %s on %s",
+                        expunge_local->loc.path,
+                        priv->children[active_src]->name);
+        } else {
+                gf_log (this->name, GF_LOG_INFO,
+                        "removing %s on %s failed (%s)",
+                        expunge_local->loc.path,
+                        priv->children[active_src]->name,
+                        strerror (op_errno));
+        }
+
+        valid = GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
+
+        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_parent_setattr_cbk,
+                           (void *) (long) active_src,
+                           priv->children[active_src],
+                           priv->children[active_src]->fops->setattr,
+                           &expunge_sh->parent_loc,
+                           &expunge_sh->parentbuf,
+                           valid, NULL);
+
+        return 0;
 }
 
 
 int
-afr_sh_entry_expunge_rmdir (call_frame_t *expunge_frame, xlator_t *this,
-			     int active_src)
+afr_sh_entry_expunge_unlink (call_frame_t *expunge_frame, xlator_t *this,
+                             int active_src)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *expunge_local = NULL;
 
-	priv = this->private;
-	expunge_local = expunge_frame->local;
+        priv          = this->private;
+        expunge_local = expunge_frame->local;
 
-	gf_log (this->name, GF_LOG_WARNING,
-		"removing directory %s on %s",
-		expunge_local->loc.path, priv->children[active_src]->name);
+        gf_log (this->name, GF_LOG_TRACE,
+                "expunging file %s on %s",
+                expunge_local->loc.path, priv->children[active_src]->name);
 
-	STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_remove_cbk,
-			   (void *) (long) active_src,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->rmdir,
-			   &expunge_local->loc);
+        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_remove_cbk,
+                           (void *) (long) active_src,
+                           priv->children[active_src],
+                           priv->children[active_src]->fops->unlink,
+                           &expunge_local->loc, 0, NULL);
 
-	return 0;
+        return 0;
 }
 
 
+
 int
-afr_sh_entry_expunge_unlink (call_frame_t *expunge_frame, xlator_t *this,
-			     int active_src)
+afr_sh_entry_expunge_rmdir (call_frame_t *expunge_frame, xlator_t *this,
+                            int active_src)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-
-	gf_log (this->name, GF_LOG_WARNING,
-		"unlinking file %s on %s",
-		expunge_local->loc.path, priv->children[active_src]->name);
-	
-	STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_remove_cbk,
-			   (void *) (long) active_src,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->unlink,
-			   &expunge_local->loc);
-
-	return 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *expunge_local = NULL;
+
+        priv          = this->private;
+        expunge_local = expunge_frame->local;
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "expunging directory %s on %s",
+                expunge_local->loc.path, priv->children[active_src]->name);
+
+        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_remove_cbk,
+                           (void *) (long) active_src,
+                           priv->children[active_src],
+                           priv->children[active_src]->fops->rmdir,
+                           &expunge_local->loc, 1, NULL);
+
+        return 0;
 }
 
 
 int
 afr_sh_entry_expunge_remove (call_frame_t *expunge_frame, xlator_t *this,
-			     int active_src, struct stat *buf)
+                             int active_src, struct iatt *buf,
+                             struct iatt *parentbuf)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh = NULL;
-	int              source = 0;
-	call_frame_t    *frame = NULL;
-	int              type = 0;
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-	expunge_sh = &expunge_local->self_heal;
-	frame = expunge_sh->sh_frame;
-	source = expunge_sh->source;
-
-	type = (buf->st_mode & S_IFMT);
-
-	switch (type) {
-	case S_IFSOCK:
-	case S_IFREG:
-	case S_IFBLK:
-	case S_IFCHR:
-	case S_IFIFO:
-	case S_IFLNK:
-		afr_sh_entry_expunge_unlink (expunge_frame, this, active_src);
-
-		break;
-	case S_IFDIR:
-		afr_sh_entry_expunge_rmdir (expunge_frame, this, active_src);
-		break;
-	default:
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s has unknown file type on %s: 0%o",
-			expunge_local->loc.path,
-			priv->children[source]->name, type);
-		goto out;
-		break;
-	}
-
-	return 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *expunge_local = NULL;
+        afr_self_heal_t *expunge_sh = NULL;
+        call_frame_t    *frame = NULL;
+        int              type = 0;
+        afr_self_heal_t *sh            = NULL;
+        afr_local_t     *local         = NULL;
+        loc_t           *loc           = NULL;
+
+        priv = this->private;
+        expunge_local = expunge_frame->local;
+        expunge_sh = &expunge_local->self_heal;
+        frame = expunge_sh->sh_frame;
+        local         = frame->local;
+        sh            = &local->self_heal;
+        loc           = &expunge_local->loc;
+
+        type = buf->ia_type;
+        if (loc->parent && uuid_is_null (loc->parent->gfid))
+                uuid_copy (loc->pargfid, parentbuf->ia_gfid);
+
+        switch (type) {
+        case IA_IFSOCK:
+        case IA_IFREG:
+        case IA_IFBLK:
+        case IA_IFCHR:
+        case IA_IFIFO:
+        case IA_IFLNK:
+                afr_sh_entry_expunge_unlink (expunge_frame, this, active_src);
+                break;
+        case IA_IFDIR:
+                afr_sh_entry_expunge_rmdir (expunge_frame, this, active_src);
+                break;
+        default:
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s has unknown file type on %s: 0%o",
+                        expunge_local->loc.path,
+                        priv->children[active_src]->name, type);
+                goto out;
+                break;
+        }
+
+        return 0;
 out:
-	AFR_STACK_DESTROY (expunge_frame);
-	afr_sh_entry_expunge_entry_done (frame, this, active_src);
+        AFR_STACK_DESTROY (expunge_frame);
+        sh->expunge_done (frame, this, active_src, -1, EINVAL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 afr_sh_entry_expunge_lookup_cbk (call_frame_t *expunge_frame, void *cookie,
-				xlator_t *this,
-				int32_t op_ret,	int32_t op_errno,
-				inode_t *inode, struct stat *buf, dict_t *x)
+                                 xlator_t *this,
+                                 int32_t op_ret, int32_t op_errno,
+                                 inode_t *inode, struct iatt *buf, dict_t *x,
+                                 struct iatt *postparent)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh = NULL;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-	expunge_sh = &expunge_local->self_heal;
-	frame = expunge_sh->sh_frame;
-	active_src = (long) cookie;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"lookup of %s on %s failed (%s)",
-			expunge_local->loc.path,
-			priv->children[active_src]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-	afr_sh_entry_expunge_remove (expunge_frame, this, active_src, buf);
-
-	return 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *expunge_local = NULL;
+        afr_self_heal_t *expunge_sh = NULL;
+        call_frame_t    *frame = NULL;
+        int              active_src = 0;
+        afr_self_heal_t *sh            = NULL;
+        afr_local_t     *local         = NULL;
+
+        priv = this->private;
+        expunge_local = expunge_frame->local;
+        expunge_sh = &expunge_local->self_heal;
+        frame = expunge_sh->sh_frame;
+        active_src = (long) cookie;
+        local         = frame->local;
+        sh            = &local->self_heal;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "lookup of %s on %s failed (%s)",
+                        expunge_local->loc.path,
+                        priv->children[active_src]->name,
+                        strerror (op_errno));
+                goto out;
+        }
+
+        afr_sh_entry_expunge_remove (expunge_frame, this, active_src, buf,
+                                     postparent);
+
+        return 0;
 out:
-	AFR_STACK_DESTROY (expunge_frame);
-	afr_sh_entry_expunge_entry_done (frame, this, active_src);
+        AFR_STACK_DESTROY (expunge_frame);
+        sh->expunge_done (frame, this, active_src, op_ret, op_errno);
 
-	return 0;
+        return 0;
 }
 
 
 int
 afr_sh_entry_expunge_purge (call_frame_t *expunge_frame, xlator_t *this,
-			    int active_src)
+                            int active_src)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"looking up %s on %s",
-		expunge_local->loc.path, priv->children[active_src]->name);
-	
-	STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_lookup_cbk,
-			   (void *) (long) active_src,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->lookup,
-			   &expunge_local->loc, 0);
-
-	return 0;
-}
+        afr_private_t   *priv = NULL;
+        afr_local_t     *expunge_local = NULL;
+
+        priv = this->private;
+        expunge_local = expunge_frame->local;
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "looking up %s on %s",
+                expunge_local->loc.path, priv->children[active_src]->name);
+
+        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_lookup_cbk,
+                           (void *) (long) active_src,
+                           priv->children[active_src],
+                           priv->children[active_src]->fops->lookup,
+                           &expunge_local->loc, NULL);
 
+        return 0;
+}
 
 int
 afr_sh_entry_expunge_entry_cbk (call_frame_t *expunge_frame, void *cookie,
-				xlator_t *this,
-				int32_t op_ret,	int32_t op_errno,
-				inode_t *inode, struct stat *buf, dict_t *x)
+                                xlator_t *this,
+                                int32_t op_ret, int32_t op_errno,
+                                inode_t *inode, struct iatt *buf, dict_t *x,
+                                struct iatt *postparent)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh = NULL;
-	int              source = 0;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-	expunge_sh = &expunge_local->self_heal;
-	frame = expunge_sh->sh_frame;
-	active_src = expunge_sh->active_source;
-	source = (long) cookie;
-
-	if (op_ret == -1 && op_errno == ENOENT) {
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"missing entry %s on %s",
-			expunge_local->loc.path,
-			priv->children[source]->name);
-
-		afr_sh_entry_expunge_purge (expunge_frame, this, active_src);
-
-		return 0;
-	}
-
-	if (op_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"%s exists under %s",
-			expunge_local->loc.path,
-			priv->children[source]->name);
-	} else {
-		gf_log (this->name, GF_LOG_ERROR,
-			"looking up %s under %s failed (%s)",
-			expunge_local->loc.path,
-			priv->children[source]->name,
-			strerror (op_errno));
-	}
-
-	AFR_STACK_DESTROY (expunge_frame);
-	afr_sh_entry_expunge_entry_done (frame, this, active_src);
-
-	return 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *expunge_local = NULL;
+        afr_self_heal_t *expunge_sh = NULL;
+        int              source = 0;
+        call_frame_t    *frame = NULL;
+        int              active_src = 0;
+        int              need_expunge = 0;
+        afr_self_heal_t *sh            = NULL;
+        afr_local_t     *local         = NULL;
+
+        priv = this->private;
+        expunge_local = expunge_frame->local;
+        expunge_sh = &expunge_local->self_heal;
+        frame = expunge_sh->sh_frame;
+        active_src = expunge_sh->active_source;
+        source = (long) cookie;
+        local         = frame->local;
+        sh            = &local->self_heal;
+
+        if (op_ret == -1 && op_errno == ENOENT)
+                need_expunge = 1;
+        else if (op_ret == -1)
+                goto out;
+
+        if (!uuid_is_null (expunge_sh->entrybuf.ia_gfid) &&
+            !uuid_is_null (buf->ia_gfid) &&
+            (uuid_compare (expunge_sh->entrybuf.ia_gfid, buf->ia_gfid) != 0)) {
+                char uuidbuf1[64];
+                char uuidbuf2[64];
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "entry %s found on %s with mismatching gfid (%s/%s)",
+                        expunge_local->loc.path,
+                        priv->children[source]->name,
+                        uuid_utoa_r (expunge_sh->entrybuf.ia_gfid, uuidbuf1),
+                        uuid_utoa_r (buf->ia_gfid, uuidbuf2));
+                need_expunge = 1;
+        }
+
+        if (need_expunge) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Entry %s is missing on %s and deleting from "
+                        "replica's other bricks",
+                        expunge_local->loc.path,
+                        priv->children[source]->name);
+
+                if (postparent)
+                        expunge_sh->parentbuf = *postparent;
+
+                afr_sh_entry_expunge_purge (expunge_frame, this, active_src);
+
+                return 0;
+        }
+
+out:
+        if (op_ret == 0) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "%s exists under %s",
+                        expunge_local->loc.path,
+                        priv->children[source]->name);
+        } else {
+                gf_log (this->name, GF_LOG_INFO,
+                        "looking up %s under %s failed (%s)",
+                        expunge_local->loc.path,
+                        priv->children[source]->name,
+                        strerror (op_errno));
+        }
+
+        AFR_STACK_DESTROY (expunge_frame);
+        sh->expunge_done (frame, this, active_src, op_ret, op_errno);
+
+        return 0;
 }
 
+static gf_boolean_t
+can_skip_entry_self_heal (char *name, loc_t *parent_loc)
+{
+        if (strcmp (name, ".") == 0) {
+                return _gf_true;
+        } else if (strcmp (name, "..") == 0) {
+                return _gf_true;
+        } else if (loc_is_root (parent_loc) &&
+                   (strcmp (name, GF_REPLICATE_TRASH_DIR) == 0)) {
+                return _gf_true;
+        }
+        return _gf_false;
+}
 
 int
 afr_sh_entry_expunge_entry (call_frame_t *frame, xlator_t *this,
-			    char *name)
+                            gf_dirent_t *entry)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              ret = -1;
-	call_frame_t    *expunge_frame = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh = NULL;
-	int              active_src = 0;
-	int              source = 0;
-	int              op_errno = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	active_src = sh->active_source;
-	source = sh->source;
-
-	if ((strcmp (name, ".") == 0)
-	    || (strcmp (name, "..") == 0)) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"skipping inspection of %s under %s",
-			name, local->loc.path);
-		goto out;
-	}
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"inspecting existance of %s under %s",
-		name, local->loc.path);
-
-	expunge_frame = copy_frame (frame);
-	if (!expunge_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (expunge_local, afr_local_t, out);
-
-	expunge_frame->local = expunge_local;
-	expunge_sh = &expunge_local->self_heal;
-	expunge_sh->sh_frame = frame;
-	expunge_sh->active_source = active_src;
-
-	ret = build_child_loc (this, &expunge_local->loc, &local->loc, name);
-	if (ret != 0) {
-		goto out;
-	}
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"looking up %s on %s", expunge_local->loc.path,
-		priv->children[source]->name);
-
-	STACK_WIND_COOKIE (expunge_frame,
-			   afr_sh_entry_expunge_entry_cbk,
-			   (void *) (long) source,
-			   priv->children[source],
-			   priv->children[source]->fops->lookup,
-			   &expunge_local->loc, 0);
-
-	ret = 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh  = NULL;
+        int              ret = -1;
+        call_frame_t    *expunge_frame = NULL;
+        afr_local_t     *expunge_local = NULL;
+        afr_self_heal_t *expunge_sh = NULL;
+        int              active_src = 0;
+        int              source = 0;
+        int              op_errno = 0;
+        char            *name = NULL;
+        int             op_ret = -1;
+
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
+
+        active_src = sh->active_source;
+        source = sh->source;
+        sh->expunge_done = afr_sh_entry_expunge_entry_done;
+
+        name = entry->d_name;
+        if (can_skip_entry_self_heal (name, &local->loc)) {
+                op_ret = 0;
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "inspecting existence of %s under %s",
+                name, local->loc.path);
+
+        expunge_frame = copy_frame (frame);
+        if (!expunge_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        AFR_LOCAL_ALLOC_OR_GOTO (expunge_local, out);
+
+        expunge_frame->local = expunge_local;
+        expunge_sh = &expunge_local->self_heal;
+        expunge_sh->sh_frame = frame;
+        expunge_sh->active_source = active_src;
+        expunge_sh->entrybuf = entry->d_stat;
+        loc_copy (&expunge_sh->parent_loc, &local->loc);
+
+        ret = afr_build_child_loc (this, &expunge_local->loc, &local->loc,
+                                   name);
+        if (ret != 0) {
+                op_errno = EINVAL;
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "looking up %s on %s", expunge_local->loc.path,
+                priv->children[source]->name);
+
+        STACK_WIND_COOKIE (expunge_frame,
+                           afr_sh_entry_expunge_entry_cbk,
+                           (void *) (long) source,
+                           priv->children[source],
+                           priv->children[source]->fops->lookup,
+                           &expunge_local->loc, NULL);
+
+        ret = 0;
 out:
-	if (ret == -1)
-		afr_sh_entry_expunge_entry_done (frame, this, active_src);
+        if (ret == -1)
+                sh->expunge_done (frame, this, active_src, op_ret, op_errno);
 
-	return 0;
+        return 0;
 }
 
 
 int
 afr_sh_entry_expunge_readdir_cbk (call_frame_t *frame, void *cookie,
-				  xlator_t *this,
-				  int32_t op_ret, int32_t op_errno,
-				  gf_dirent_t *entries)
+                                  xlator_t *this,
+                                  int32_t op_ret, int32_t op_errno,
+                                  gf_dirent_t *entries, dict_t *xdata)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	gf_dirent_t     *entry = NULL;
-	off_t            last_offset = 0;
-	int              active_src = 0;
-	int              entry_count = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	active_src = sh->active_source;
-
-	if (op_ret <= 0) {
-		if (op_ret < 0) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"readdir of %s on subvolume %s failed (%s)",
-				local->loc.path,
-				priv->children[active_src]->name,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"readdir of %s on subvolume %s complete",
-				local->loc.path,
-				priv->children[active_src]->name);
-		}
-
-		afr_sh_entry_expunge_all (frame, this);
-		return 0;
-	}
-
-	list_for_each_entry (entry, &entries->list, list) {
-		last_offset = entry->d_off;
-		entry_count++;
-	}
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"readdir'ed %d entries from %s",
-		entry_count, priv->children[active_src]->name);
-
-	sh->offset = last_offset;
-	local->call_count = entry_count;
-
-	list_for_each_entry (entry, &entries->list, list) {
-		afr_sh_entry_expunge_entry (frame, this, entry->d_name);
-	}
-
-	return 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh  = NULL;
+        gf_dirent_t     *entry = NULL;
+        off_t            last_offset = 0;
+        int              active_src = 0;
+        int              entry_count = 0;
+
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
+
+        active_src = sh->active_source;
+
+        if (op_ret <= 0) {
+                if (op_ret < 0) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "readdir of %s on subvolume %s failed (%s)",
+                                local->loc.path,
+                                priv->children[active_src]->name,
+                                strerror (op_errno));
+                } else {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "readdir of %s on subvolume %s complete",
+                                local->loc.path,
+                                priv->children[active_src]->name);
+                }
+
+                afr_sh_entry_expunge_all (frame, this);
+                return 0;
+        }
+
+        list_for_each_entry (entry, &entries->list, list) {
+                last_offset = entry->d_off;
+                entry_count++;
+        }
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "readdir'ed %d entries from %s",
+                entry_count, priv->children[active_src]->name);
+
+        sh->offset = last_offset;
+        local->call_count = entry_count;
+
+        list_for_each_entry (entry, &entries->list, list) {
+                afr_sh_entry_expunge_entry (frame, this, entry);
+        }
+
+        return 0;
 }
 
 int
 afr_sh_entry_expunge_subvol (call_frame_t *frame, xlator_t *this,
-			     int active_src)
+                             int active_src)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh  = NULL;
 
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
 
-	STACK_WIND (frame, afr_sh_entry_expunge_readdir_cbk,
-		    priv->children[active_src],
-		    priv->children[active_src]->fops->readdir,
-		    sh->healing_fd, sh->block_size, sh->offset);
+        STACK_WIND (frame, afr_sh_entry_expunge_readdir_cbk,
+                    priv->children[active_src],
+                    priv->children[active_src]->fops->readdirp,
+                    sh->healing_fd, sh->block_size, sh->offset, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 afr_sh_entry_expunge_all (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              active_src = -1;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh  = NULL;
+        int              active_src = -1;
 
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
 
-	sh->offset = 0;
+        sh->offset = 0;
 
-	if (sh->source == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no active sources for %s to expunge entries",
-			local->loc.path);
-		goto out;
-	}
+        if (sh->source == -1) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no active sources for %s to expunge entries",
+                        local->loc.path);
+                goto out;
+        }
 
-	active_src = next_active_sink (frame, this, sh->active_source);
-	sh->active_source = active_src;
+        active_src = next_active_sink (frame, this, sh->active_source);
+        sh->active_source = active_src;
 
-	if (sh->op_failed) {
-		goto out;
-	}
+        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
+                goto out;
+        }
 
-	if (active_src == -1) {
-		/* completed creating missing files on all subvolumes */
-		goto out;
-	}
+        if (active_src == -1) {
+                /* completed creating missing files on all subvolumes */
+                goto out;
+        }
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"expunging entries of %s on %s to other sinks",
-		local->loc.path, priv->children[active_src]->name);
+        gf_log (this->name, GF_LOG_TRACE,
+                "expunging entries of %s on %s to other sinks",
+                local->loc.path, priv->children[active_src]->name);
 
-	afr_sh_entry_expunge_subvol (frame, this, active_src);
+        afr_sh_entry_expunge_subvol (frame, this, active_src);
 
-	return 0;
+        return 0;
 out:
-	afr_sh_entry_erase_pending (frame, this);
-	return 0;
+        afr_sh_entry_impunge_all (frame, this);
+        return 0;
 
 }
 
 
 int
-afr_sh_entry_impunge_all (call_frame_t *frame, xlator_t *this);
-
-int
-afr_sh_entry_impunge_subvol (call_frame_t *frame, xlator_t *this,
-			     int active_src);
-
-int
 afr_sh_entry_impunge_entry_done (call_frame_t *frame, xlator_t *this,
-				 int active_src)
+                                 int32_t op_ret, int32_t op_errno)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              call_count = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	LOCK (&frame->lock);
-	{
-	}
-	UNLOCK (&frame->lock);
+        int              call_count = 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        if (op_ret < 0)
+                sh->entries_skipped = _gf_true;
+        call_count = afr_frame_return (frame);
+        if (call_count == 0)
+                afr_sh_entry_impunge_subvol (frame, this);
+
+        return 0;
+}
 
-	call_count = afr_frame_return (frame);
+void
+afr_sh_entry_call_impunge_done (call_frame_t *impunge_frame, xlator_t *this,
+                                int32_t op_ret, int32_t op_errno)
+{
+        afr_local_t     *impunge_local = NULL;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_self_heal_t *impunge_sh = NULL;
+        call_frame_t    *frame = NULL;
 
-	if (call_count == 0)
-		afr_sh_entry_impunge_subvol (frame, this, active_src);
+        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
+                                frame, local, sh);
 
-	return 0;
+        AFR_STACK_DESTROY (impunge_frame);
+        sh->impunge_done (frame, this, op_ret, op_errno);
 }
 
-
 int
-afr_sh_entry_impunge_utimens_cbk (call_frame_t *impunge_frame, void *cookie,
-				  xlator_t *this, int32_t op_ret,
-				  int32_t op_errno, struct stat *stbuf)
+afr_sh_entry_impunge_setattr_cbk (call_frame_t *impunge_frame, void *cookie,
+                                  xlator_t *this,
+                                  int32_t op_ret, int32_t op_errno,
+                                  struct iatt *preop, struct iatt *postop,
+                                  dict_t *xdata)
 {
-	int              call_count = 0;
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-	int              child_index = 0;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-	child_index = (long) cookie;
-
-	if (op_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"utimes set for %s on %s",
-			impunge_local->loc.path,
-			priv->children[child_index]->name);
-	} else {
-		gf_log (this->name, GF_LOG_ERROR,
-			"setting utimes of %s on %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[child_index]->name,
-			strerror (op_errno));
-	}
-
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
+        int              call_count = 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
+        int              child_index = 0;
+
+        priv = this->private;
+        impunge_local = impunge_frame->local;
+        child_index = (long) cookie;
+
+        if (op_ret == 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "setattr done for %s on %s",
+                        impunge_local->loc.path,
+                        priv->children[child_index]->name);
+        } else {
+                gf_log (this->name, GF_LOG_INFO,
+                        "setattr (%s) on %s failed (%s)",
+                        impunge_local->loc.path,
+                        priv->children[child_index]->name,
+                        strerror (op_errno));
+        }
+
+        call_count = afr_frame_return (impunge_frame);
+        if (call_count == 0) {
+                afr_sh_entry_call_impunge_done (impunge_frame, this,
+                                                0, op_errno);
+        }
+
+        return 0;
 }
 
-
 int
-afr_sh_entry_impunge_chown_cbk (call_frame_t *impunge_frame, void *cookie,
-				xlator_t *this, int32_t op_ret,
-				int32_t op_errno, struct stat *stbuf)
+afr_sh_entry_impunge_parent_setattr_cbk (call_frame_t *setattr_frame,
+                                         void *cookie, xlator_t *this,
+                                         int32_t op_ret, int32_t op_errno,
+                                         struct iatt *preop, struct iatt *postop,
+                                         dict_t *xdata)
 {
-	int              call_count = 0;
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-	int              child_index = 0;
-	struct timespec  ts[2];
-
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-	child_index = (long) cookie;
-
-	if (op_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"ownership of %s on %s changed",
-			impunge_local->loc.path,
-			priv->children[child_index]->name);
-	} else {
-		gf_log (this->name, GF_LOG_ERROR,
-			"setting ownership of %s on %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[child_index]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-#ifdef HAVE_STRUCT_STAT_ST_ATIM_TV_NSEC
-	ts[0] = impunge_local->cont.lookup.buf.st_atim;
-	ts[1] = impunge_local->cont.lookup.buf.st_mtim;
-#elif HAVE_STRUCT_STAT_ST_ATIMESPEC_TV_NSEC
-	ts[0] = impunge_local->cont.lookup.buf.st_atimespec;
-	ts[1] = impunge_local->cont.lookup.buf.st_mtimespec;
-#else
-	ts[0].tv_sec = impunge_local->cont.lookup.buf.st_atime;
-	ts[1].tv_sec = impunge_local->cont.lookup.buf.st_mtime;
-#endif
-	STACK_WIND_COOKIE (impunge_frame,
-			   afr_sh_entry_impunge_utimens_cbk,
-			   (void *) (long) child_index,
-			   priv->children[child_index],
-			   priv->children[child_index]->fops->utimens,
-			   &impunge_local->loc, ts);
+        int             call_count = 0;
+        afr_local_t     *setattr_local = NULL;
+
+        setattr_local = setattr_frame->local;
+        if (op_ret != 0) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "setattr on parent directory (%s) failed: %s",
+                        setattr_local->loc.path, strerror (op_errno));
+        }
+
+        call_count = afr_frame_return (setattr_frame);
+        if (call_count == 0)
+                AFR_STACK_DESTROY (setattr_frame);
+        return 0;
+}
 
-	return 0;
+int
+afr_sh_entry_impunge_setattr (call_frame_t *impunge_frame, xlator_t *this)
+{
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
+        afr_local_t     *setattr_local = NULL;
+        afr_self_heal_t *impunge_sh = NULL;
+        call_frame_t    *setattr_frame = NULL;
+        int32_t          valid = 0;
+        int32_t          op_errno = 0;
+        int              child_index = 0;
+        int              call_count = 0;
+        int              i = 0;
+
+        priv          = this->private;
+        impunge_local = impunge_frame->local;
+        impunge_sh    = &impunge_local->self_heal;
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "setting ownership of %s on %s to %d/%d",
+                impunge_local->loc.path,
+                priv->children[child_index]->name,
+                impunge_sh->entrybuf.ia_uid,
+                impunge_sh->entrybuf.ia_gid);
+
+        setattr_frame = copy_frame (impunge_frame);
+        if (!setattr_frame) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+        AFR_LOCAL_ALLOC_OR_GOTO (setattr_frame->local, out);
+        setattr_local = setattr_frame->local;
+        call_count = afr_errno_count (NULL, impunge_sh->child_errno,
+                                      priv->child_count, 0);
+        loc_copy (&setattr_local->loc, &impunge_sh->parent_loc);
+        impunge_local->call_count = call_count;
+        setattr_local->call_count = call_count;
+        for (i = 0; i < priv->child_count; i++) {
+                if (impunge_sh->child_errno[i])
+                        continue;
+                valid         = GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
+                STACK_WIND_COOKIE (setattr_frame,
+                                   afr_sh_entry_impunge_parent_setattr_cbk,
+                                   (void *) (long) i, priv->children[i],
+                                   priv->children[i]->fops->setattr,
+                                   &setattr_local->loc,
+                                   &impunge_sh->parentbuf, valid, NULL);
+
+                valid = GF_SET_ATTR_UID   | GF_SET_ATTR_GID |
+                        GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
+                STACK_WIND_COOKIE (impunge_frame,
+                                   afr_sh_entry_impunge_setattr_cbk,
+                                   (void *) (long) i, priv->children[i],
+                                   priv->children[i]->fops->setattr,
+                                   &impunge_local->loc,
+                                   &impunge_sh->entrybuf, valid, NULL);
+                call_count--;
+        }
+        GF_ASSERT (!call_count);
+        return 0;
+out:
+        if (setattr_frame)
+                AFR_STACK_DESTROY (setattr_frame);
+        afr_sh_entry_call_impunge_done (impunge_frame, this, 0, op_errno);
+        return 0;
+}
 
+int
+afr_sh_entry_impunge_xattrop_cbk (call_frame_t *impunge_frame, void *cookie,
+                                  xlator_t *this,
+                                  int32_t op_ret, int32_t op_errno,
+                                  dict_t *xattr, dict_t *xdata)
+{
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
+        int              child_index = 0;
+
+        priv          = this->private;
+        impunge_local = impunge_frame->local;
+
+        child_index = (long) cookie;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "%s: failed to perform xattrop on %s (%s)",
+                        impunge_local->loc.path,
+                        priv->children[child_index]->name,
+                        strerror (op_errno));
+                goto out;
+        }
+
+        afr_sh_entry_impunge_setattr (impunge_frame, this);
+        return 0;
 out:
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
+        afr_sh_entry_call_impunge_done (impunge_frame, this,
+                                        -1, op_errno);
+        return 0;
 }
 
+int
+afr_sh_entry_impunge_perform_xattrop (call_frame_t *impunge_frame,
+                                      xlator_t *this)
+{
+        int              active_src       = 0;
+        dict_t          *xattr            = NULL;
+        afr_private_t   *priv             = NULL;
+        afr_local_t     *impunge_local    = NULL;
+        afr_self_heal_t *impunge_sh       = NULL;
+        int32_t         op_errno          = 0;
+
+        priv = this->private;
+        impunge_local = impunge_frame->local;
+        impunge_sh = &impunge_local->self_heal;
+        active_src = impunge_sh->active_source;
+
+        afr_prepare_new_entry_pending_matrix (impunge_local->pending,
+                                              afr_is_errno_unset,
+                                              impunge_sh->child_errno,
+                                              &impunge_sh->entrybuf,
+                                              priv->child_count);
+        xattr = dict_new ();
+        if (!xattr) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        afr_set_pending_dict (priv, xattr, impunge_local->pending, active_src,
+                              LOCAL_LAST);
+
+        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_xattrop_cbk,
+                           (void *) (long) active_src,
+                           priv->children[active_src],
+                           priv->children[active_src]->fops->xattrop,
+                           &impunge_local->loc, GF_XATTROP_ADD_ARRAY, xattr, NULL);
+
+        if (xattr)
+                dict_unref (xattr);
+        return 0;
+out:
+        afr_sh_entry_call_impunge_done (impunge_frame, this,
+                                        -1, op_errno);
+        return 0;
+}
 
 int
 afr_sh_entry_impunge_newfile_cbk (call_frame_t *impunge_frame, void *cookie,
-				  xlator_t *this,
-				  int32_t op_ret, int32_t op_errno,
-				  inode_t *inode, struct stat *stbuf)
+                                  xlator_t *this,
+                                  int32_t op_ret, int32_t op_errno,
+                                  inode_t *inode, struct iatt *stbuf,
+                                  struct iatt *preparent,
+                                  struct iatt *postparent, dict_t *xdata)
 {
-	int              call_count = 0;
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-	int              child_index = 0;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-
-	child_index = (long) cookie;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"creation of %s on %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[child_index]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"setting ownership of %s on %s to %d/%d",
-		impunge_local->loc.path,
-		priv->children[child_index]->name,
-		impunge_local->cont.lookup.buf.st_uid,
-		impunge_local->cont.lookup.buf.st_gid);
-
-	inode->st_mode = stbuf->st_mode;
-
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_chown_cbk,
-			   (void *) (long) child_index,
-			   priv->children[child_index],
-			   priv->children[child_index]->fops->chown,
-			   &impunge_local->loc,
-			   impunge_local->cont.lookup.buf.st_uid,
-			   impunge_local->cont.lookup.buf.st_gid);
-	return 0;
-
+        int              call_count       = 0;
+        afr_private_t   *priv             = NULL;
+        afr_local_t     *impunge_local    = NULL;
+        afr_self_heal_t *impunge_sh       = NULL;
+        int              child_index      = 0;
+
+        priv = this->private;
+        impunge_local = impunge_frame->local;
+        impunge_sh = &impunge_local->self_heal;
+
+        child_index = (long) cookie;
+
+        if (op_ret == -1) {
+                impunge_sh->child_errno[child_index] = op_errno;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "creation of %s on %s failed (%s)",
+                        impunge_local->loc.path,
+                        priv->children[child_index]->name,
+                        strerror (op_errno));
+        } else {
+                impunge_sh->child_errno[child_index] = 0;
+        }
+
+        call_count = afr_frame_return (impunge_frame);
+        if (call_count == 0) {
+                if (!afr_errno_count (NULL, impunge_sh->child_errno,
+                                      priv->child_count, 0)) {
+                        // new_file creation failed every where
+                        afr_sh_entry_call_impunge_done (impunge_frame, this,
+                                                        -1, op_errno);
+                        goto out;
+                }
+                afr_sh_entry_impunge_perform_xattrop (impunge_frame, this);
+        }
 out:
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
+        return 0;
 }
 
-
 int
-afr_sh_entry_impunge_mknod (call_frame_t *impunge_frame, xlator_t *this,
-			    int child_index, struct stat *stbuf)
+afr_sh_entry_impunge_hardlink_cbk (call_frame_t *impunge_frame, void *cookie,
+                                   xlator_t *this, int32_t op_ret,
+                                   int32_t op_errno, inode_t *inode,
+                                   struct iatt *buf, struct iatt *preparent,
+                                   struct iatt *postparent, dict_t *xdata)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-
+        int              call_count        = 0;
+        afr_local_t     *impunge_local = NULL;
+        afr_self_heal_t *impunge_sh  = NULL;
+
+        impunge_local = impunge_frame->local;
+        impunge_sh = &impunge_local->self_heal;
+
+        if (IA_IFLNK == impunge_sh->entrybuf.ia_type) {
+                //For symlinks impunge is attempted un-conditionally
+                //So the file can already exist.
+                if ((op_ret < 0) && (op_errno == EEXIST))
+                        op_ret = 0;
+        }
+
+        call_count = afr_frame_return (impunge_frame);
+        if (call_count == 0)
+                afr_sh_entry_call_impunge_done (impunge_frame, this,
+                                                op_ret, op_errno);
+
+        return 0;
+}
 
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
+int
+afr_sh_entry_impunge_hardlink (call_frame_t *impunge_frame, xlator_t *this,
+                               int child_index)
+{
+        afr_private_t   *priv          = NULL;
+        afr_local_t     *impunge_local = NULL;
+        afr_self_heal_t *impunge_sh  = NULL;
+        loc_t           *loc           = NULL;
+        struct iatt     *buf           = NULL;
+        loc_t            oldloc        = {0};
+
+        priv = this->private;
+        impunge_local = impunge_frame->local;
+        impunge_sh = &impunge_local->self_heal;
+        loc = &impunge_local->loc;
+        buf = &impunge_sh->entrybuf;
+
+        oldloc.inode = inode_ref (loc->inode);
+        uuid_copy (oldloc.gfid, buf->ia_gfid);
+        gf_log (this->name, GF_LOG_DEBUG, "linking missing file %s on %s",
+                loc->path, priv->children[child_index]->name);
+
+        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_hardlink_cbk,
+                           (void *) (long) child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->link,
+                           &oldloc, loc, NULL);
+        loc_wipe (&oldloc);
+
+        return 0;
+}
 
-	gf_log (this->name, GF_LOG_WARNING,
-		"creating file %s mode=0%o dev=0x%"GF_PRI_DEV" on %s",
-		impunge_local->loc.path,
-		stbuf->st_mode, stbuf->st_rdev,
-		priv->children[child_index]->name);
+int
+afr_sh_nameless_lookup_cbk (call_frame_t *impunge_frame, void *cookie,
+                            xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, inode_t *inode,
+                            struct iatt *buf, dict_t *xattr,
+                            struct iatt *postparent)
+{
+        if (op_ret < 0) {
+                 afr_sh_entry_impunge_create_file (impunge_frame, this,
+                                                   (long)cookie);
+        } else {
+                afr_sh_entry_impunge_hardlink (impunge_frame, this,
+                                               (long)cookie);
+        }
+        return 0;
+}
 
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
-			   (void *) (long) child_index,
-			   priv->children[child_index],
-			   priv->children[child_index]->fops->mknod,
-			   &impunge_local->loc,
-			   stbuf->st_mode, stbuf->st_rdev);
+int
+afr_sh_entry_impunge_check_hardlink (call_frame_t *impunge_frame,
+                                     xlator_t *this,
+                                     int child_index, struct iatt *stbuf)
+{
+        afr_private_t   *priv          = NULL;
+        call_frame_t    *frame             = NULL;
+        afr_local_t     *impunge_local     = NULL;
+        afr_local_t     *local             = NULL;
+        afr_self_heal_t *impunge_sh        = NULL;
+        afr_self_heal_t *sh                = NULL;
+        loc_t           *loc           = NULL;
+        dict_t          *xattr_req     = NULL;
+        loc_t            oldloc        = {0};
+        int              ret           = -1;
+
+        priv = this->private;
+        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
+                                frame, local, sh);
+        loc = &impunge_local->loc;
+
+        xattr_req = dict_new ();
+        if (!xattr_req)
+                goto out;
+        oldloc.inode = inode_ref (loc->inode);
+        uuid_copy (oldloc.gfid, stbuf->ia_gfid);
+
+        STACK_WIND_COOKIE (impunge_frame, afr_sh_nameless_lookup_cbk,
+                           (void *) (long) child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->lookup,
+                           &oldloc, xattr_req);
+        ret = 0;
+out:
+        if (xattr_req)
+                dict_unref (xattr_req);
+        loc_wipe (&oldloc);
+        if (ret)
+                sh->impunge_done (frame, this, -1, ENOMEM);
+        return 0;
+}
 
-	return 0;
+int
+afr_sh_entry_impunge_mknod (call_frame_t *impunge_frame, xlator_t *this,
+                            int child_index, struct iatt *stbuf)
+{
+        afr_private_t *priv          = NULL;
+        afr_local_t   *impunge_local = NULL;
+        dict_t        *dict          = NULL;
+        int            ret           = 0;
+
+        priv = this->private;
+        impunge_local = impunge_frame->local;
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "creating missing file %s on %s",
+                impunge_local->loc.path,
+                priv->children[child_index]->name);
+
+        dict = dict_new ();
+        if (!dict)
+                gf_log (this->name, GF_LOG_ERROR, "Out of memory");
+
+        GF_ASSERT (!uuid_is_null (stbuf->ia_gfid));
+        ret = afr_set_dict_gfid (dict, stbuf->ia_gfid);
+        if (ret)
+                gf_log (this->name, GF_LOG_INFO, "%s: gfid set failed",
+                        impunge_local->loc.path);
+
+        /*
+         * Reason for adding GLUSTERFS_INTERNAL_FOP_KEY :
+         *
+         * Problem:
+         * While a brick is down in a replica pair, lets say the user creates
+         * one file(file-A) and a hard link to that file(h-file-A). After the
+         * brick comes back up, entry self-heal is attempted on parent dir of
+         * these two files. As part of readdir in self-heal it reads both the
+         * entries file-A and h-file-A for both of them it does name less lookup
+         * to check if there are any hardlinks already present in the
+         * destination brick. It finds that there are no hard links already
+         * present for files file-A, h-file-A. Self-heal does mknods for both
+         * file-A and h-file-A. This leads to file-A and h-file-A not being
+         * hardlinks anymore.
+         *
+         * Fix: (More like shrinking of race-window, the race itself is still
+         * present in posix-mknod).
+         * If mknod comes with the presence of GLUSTERFS_INTERNAL_FOP_KEY then
+         * posix_mknod checks if there are already any gfid-links and does
+         * link() instead of mknod. There still can be a race where two
+         * posix_mknods same gfid see that
+         * gfid-link file is not present and proceeds with mknods and result in
+         * two different files with same gfid.
+         */
+        ret = dict_set_str (dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+        if (ret)
+                gf_log (this->name, GF_LOG_INFO, "%s: %s set failed",
+                        impunge_local->loc.path, GLUSTERFS_INTERNAL_FOP_KEY);
+
+        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
+                           (void *) (long) child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->mknod,
+                           &impunge_local->loc,
+                           st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type),
+                           makedev (ia_major (stbuf->ia_rdev),
+                                    ia_minor (stbuf->ia_rdev)), 0, dict);
+
+        if (dict)
+                dict_unref (dict);
+
+        return 0;
 }
 
 
 
 int
 afr_sh_entry_impunge_mkdir (call_frame_t *impunge_frame, xlator_t *this,
-			    int child_index, struct stat *stbuf)
+                            int child_index, struct iatt *stbuf)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-
-	gf_log (this->name, GF_LOG_WARNING,
-		"creating directory %s mode=0%o on %s",
-		impunge_local->loc.path,
-		stbuf->st_mode,
-		priv->children[child_index]->name);
-
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
-			   (void *) (long) child_index,
-			   priv->children[child_index],
-			   priv->children[child_index]->fops->mkdir,
-			   &impunge_local->loc, stbuf->st_mode);
-
-	return 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
+        dict_t          *dict = NULL;
+
+        int ret = 0;
+
+        priv = this->private;
+        impunge_local = impunge_frame->local;
+
+        dict = dict_new ();
+        if (!dict) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Out of memory");
+                return 0;
+        }
+
+        GF_ASSERT (!uuid_is_null (stbuf->ia_gfid));
+        ret = afr_set_dict_gfid (dict, stbuf->ia_gfid);
+        if (ret)
+                gf_log (this->name, GF_LOG_INFO, "%s: gfid set failed",
+                        impunge_local->loc.path);
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "creating missing directory %s on %s",
+                impunge_local->loc.path,
+                priv->children[child_index]->name);
+
+        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
+                           (void *) (long) child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->mkdir,
+                           &impunge_local->loc,
+                           st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type),
+                           0, dict);
+
+        if (dict)
+                dict_unref (dict);
+
+        return 0;
 }
 
 
 int
 afr_sh_entry_impunge_symlink (call_frame_t *impunge_frame, xlator_t *this,
-			      int child_index, const char *linkname)
+                              int child_index, const char *linkname)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-
+        afr_private_t   *priv          = NULL;
+        afr_local_t     *impunge_local = NULL;
+        dict_t          *dict          = NULL;
+        struct iatt     *buf           = NULL;
+        int              ret           = 0;
+
+        priv = this->private;
+        impunge_local = impunge_frame->local;
+
+        buf = &impunge_local->cont.dir_fop.buf;
+
+        dict = dict_new ();
+        if (!dict) {
+                afr_sh_entry_call_impunge_done (impunge_frame, this,
+                                                -1, ENOMEM);
+                goto out;
+        }
+
+        GF_ASSERT (!uuid_is_null (buf->ia_gfid));
+        ret = afr_set_dict_gfid (dict, buf->ia_gfid);
+        if (ret)
+                gf_log (this->name, GF_LOG_INFO,
+                        "%s: dict set gfid failed",
+                        impunge_local->loc.path);
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "creating missing symlink %s -> %s on %s",
+                impunge_local->loc.path, linkname,
+                priv->children[child_index]->name);
+
+        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
+                           (void *) (long) child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->symlink,
+                           linkname, &impunge_local->loc, 0, dict);
+
+        if (dict)
+                dict_unref (dict);
+out:
+        return 0;
+}
 
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
 
-	gf_log (this->name, GF_LOG_WARNING,
-		"creating symlink %s -> %s on %s",
-		impunge_local->loc.path, linkname,
-		priv->children[child_index]->name);
+int
+afr_sh_entry_impunge_symlink_unlink_cbk (call_frame_t *impunge_frame,
+                                         void *cookie, xlator_t *this,
+                                         int32_t op_ret, int32_t op_errno,
+                                         struct iatt *preparent,
+                                         struct iatt *postparent, dict_t *xdata)
+{
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
+        afr_self_heal_t *impunge_sh = NULL;
+        int              child_index = -1;
+        int              call_count = -1;
+
+        priv          = this->private;
+        impunge_local = impunge_frame->local;
+        impunge_sh    = &impunge_local->self_heal;
+
+        child_index = (long) cookie;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "unlink of %s on %s failed (%s)",
+                        impunge_local->loc.path,
+                        priv->children[child_index]->name,
+                        strerror (op_errno));
+                goto out;
+        }
+
+        afr_sh_entry_impunge_symlink (impunge_frame, this, child_index,
+                                      impunge_sh->linkname);
+
+        return 0;
+out:
+        LOCK (&impunge_frame->lock);
+        {
+                call_count = --impunge_local->call_count;
+        }
+        UNLOCK (&impunge_frame->lock);
 
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
-			   (void *) (long) child_index,
-			   priv->children[child_index],
-			   priv->children[child_index]->fops->symlink,
-			   linkname, &impunge_local->loc);
+        if (call_count == 0)
+                afr_sh_entry_call_impunge_done (impunge_frame, this,
+                                                op_ret, op_errno);
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_sh_entry_impunge_readlink_cbk (call_frame_t *impunge_frame, void *cookie,
-				   xlator_t *this,
-				   int32_t op_ret, int32_t op_errno,
-				   const char *linkname)
+afr_sh_entry_impunge_symlink_unlink (call_frame_t *impunge_frame, xlator_t *this,
+                                     int child_index)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              child_index = -1;
-	call_frame_t    *frame = NULL;
-	int              call_count = -1;
-	int              active_src = -1;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-	active_src = impunge_sh->active_source;
-
-	child_index = (long) cookie;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"readlink of %s on %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[active_src]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-	afr_sh_entry_impunge_symlink (impunge_frame, this, child_index,
-				      linkname);
-	return 0;
+        afr_private_t   *priv          = NULL;
+        afr_local_t     *impunge_local = NULL;
 
-out:
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
-}
+        priv          = this->private;
+        impunge_local = impunge_frame->local;
 
+        gf_log (this->name, GF_LOG_DEBUG,
+                "unlinking symlink %s with wrong target on %s",
+                impunge_local->loc.path,
+                priv->children[child_index]->name);
 
-int
-afr_sh_entry_impunge_readlink (call_frame_t *impunge_frame, xlator_t *this,
-			       int child_index, struct stat *stbuf)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              active_src = -1;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	active_src = impunge_sh->active_source;
-
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_readlink_cbk,
-			   (void *) (long) child_index,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->readlink,
-			   &impunge_local->loc, 4096);
-
-	return 0;
+        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_symlink_unlink_cbk,
+                           (void *) (long) child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->unlink,
+                           &impunge_local->loc, 0, NULL);
+
+        return 0;
 }
 
 
 int
-afr_sh_entry_impunge_recreate_lookup_cbk (call_frame_t *impunge_frame,
-					  void *cookie, xlator_t *this,
-					  int32_t op_ret, int32_t op_errno,
-					  inode_t *inode, struct stat *buf,
-					  dict_t *xattr)
+afr_sh_entry_impunge_readlink_sink_cbk (call_frame_t *impunge_frame, void *cookie,
+                                        xlator_t *this,
+                                        int32_t op_ret, int32_t op_errno,
+                                        const char *linkname, struct iatt *sbuf, dict_t *xdata)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              active_src = 0;
-	int              type = 0;
-	int              child_index = 0;
-	call_frame_t    *frame = NULL;
-	int              call_count = 0;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-
-	child_index = (long) cookie;
-
-	active_src = impunge_sh->active_source;
-
-	if (op_ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"looking up %s on %s (for %s) failed (%s)",
-			impunge_local->loc.path,
-			priv->children[active_src]->name,
-			priv->children[child_index]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-	impunge_local->cont.lookup.buf = *buf;
-	type = (buf->st_mode & S_IFMT);
-
-	switch (type) {
-	case S_IFSOCK:
-	case S_IFREG:
-	case S_IFBLK:
-	case S_IFCHR:
-	case S_IFIFO:
-		afr_sh_entry_impunge_mknod (impunge_frame, this,
-					    child_index, buf);
-		break;
-	case S_IFLNK:
-		afr_sh_entry_impunge_readlink (impunge_frame, this,
-					       child_index, buf);
-		break;
-	case S_IFDIR:
-		afr_sh_entry_impunge_mkdir (impunge_frame, this,
-					    child_index, buf);
-		break;
-	default:
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s has unknown file type on %s: 0%o",
-			impunge_local->loc.path,
-			priv->children[active_src]->name, type);
-		goto out;
-		break;
-	}
-
-	return 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
+        afr_self_heal_t *impunge_sh = NULL;
+        int              child_index = -1;
+        int              call_count = -1;
+        int              active_src = -1;
+
+        priv          = this->private;
+        impunge_local = impunge_frame->local;
+        impunge_sh    = &impunge_local->self_heal;
+        active_src    = impunge_sh->active_source;
+
+        child_index = (long) cookie;
+
+        if ((op_ret == -1) && (op_errno != ENOENT)) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "readlink of %s on %s failed (%s)",
+                        impunge_local->loc.path,
+                        priv->children[active_src]->name,
+                        strerror (op_errno));
+                goto out;
+        }
+
+        /* symlink doesn't exist on the sink */
+
+        if ((op_ret == -1) && (op_errno == ENOENT)) {
+                afr_sh_entry_impunge_symlink (impunge_frame, this,
+                                              child_index, impunge_sh->linkname);
+                return 0;
+        }
+
+
+        /* symlink exists on the sink, so check if targets match */
+
+        if (strcmp (linkname, impunge_sh->linkname) == 0) {
+                /* targets match, nothing to do */
+
+                goto out;
+        } else {
+                /*
+                 * Hah! Sneaky wolf in sheep's clothing!
+                 */
+                afr_sh_entry_impunge_symlink_unlink (impunge_frame, this,
+                                                     child_index);
+                return 0;
+        }
 
 out:
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
+        LOCK (&impunge_frame->lock);
+        {
+                call_count = --impunge_local->call_count;
+        }
+        UNLOCK (&impunge_frame->lock);
+
+        if (call_count == 0)
+                afr_sh_entry_call_impunge_done (impunge_frame, this,
+                                                op_ret, op_errno);
+
+        return 0;
 }
 
 
 int
-afr_sh_entry_impunge_recreate (call_frame_t *impunge_frame, xlator_t *this,
-			       int child_index)
+afr_sh_entry_impunge_readlink_sink (call_frame_t *impunge_frame, xlator_t *this,
+                                    int child_index)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              active_src = 0;
-
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
 
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
+        priv = this->private;
+        impunge_local = impunge_frame->local;
 
-	active_src = impunge_sh->active_source;
+        gf_log (this->name, GF_LOG_DEBUG,
+                "checking symlink target of %s on %s",
+                impunge_local->loc.path, priv->children[child_index]->name);
 
-	STACK_WIND_COOKIE (impunge_frame,
-			   afr_sh_entry_impunge_recreate_lookup_cbk,
-			   (void *) (long) child_index,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->lookup,
-			   &impunge_local->loc, 0);
+        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_readlink_sink_cbk,
+                           (void *) (long) child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->readlink,
+                           &impunge_local->loc, 4096, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_sh_entry_impunge_entry_cbk (call_frame_t *impunge_frame, void *cookie,
-				xlator_t *this,
-				int32_t op_ret,	int32_t op_errno,
-				inode_t *inode, struct stat *buf, dict_t *x)
+afr_sh_entry_impunge_readlink_cbk (call_frame_t *impunge_frame, void *cookie,
+                                   xlator_t *this,
+                                   int32_t op_ret, int32_t op_errno,
+                                   const char *linkname, struct iatt *sbuf, dict_t *xdata)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-	child_index = (long) cookie;
-	active_src = impunge_sh->active_source;
-
-	if (op_ret == -1 && op_errno == ENOENT) {
-		/* decrease call_count in recreate-callback */
-		gf_log (this->name, GF_LOG_DEBUG,
-			"missing entry %s on %s",
-			impunge_local->loc.path,
-			priv->children[child_index]->name);
-
-		afr_sh_entry_impunge_recreate (impunge_frame, this,
-					       child_index);
-		return 0;
-	}
-
-	if (op_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"%s exists under %s",
-			impunge_local->loc.path,
-			priv->children[child_index]->name);
-	} else {
-		gf_log (this->name, GF_LOG_ERROR,
-			"looking up %s under %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[child_index]->name,
-			strerror (op_errno));
-	}
-
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
-}
-
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
+        afr_self_heal_t *impunge_sh = NULL;
+        int              child_index = -1;
+        int              call_count = -1;
+        int              active_src = -1;
+
+        priv = this->private;
+        impunge_local = impunge_frame->local;
+        impunge_sh = &impunge_local->self_heal;
+        active_src = impunge_sh->active_source;
+
+        child_index = (long) cookie;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "readlink of %s on %s failed (%s)",
+                        impunge_local->loc.path,
+                        priv->children[active_src]->name,
+                        strerror (op_errno));
+                goto out;
+        }
+
+        impunge_sh->linkname = gf_strdup (linkname);
+        afr_sh_entry_impunge_readlink_sink (impunge_frame, this, child_index);
+
+        return 0;
 
-int
-afr_sh_entry_impunge_entry (call_frame_t *frame, xlator_t *this,
-			    char *name)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              ret = -1;
-	call_frame_t    *impunge_frame = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              active_src = 0;
-	int              i = 0;
-	int              call_count = 0;
-	int              op_errno = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	active_src = sh->active_source;
-
-	if ((strcmp (name, ".") == 0)
-	    || (strcmp (name, "..") == 0)) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"skipping inspection of %s under %s",
-			name, local->loc.path);
-		goto out;
-	}
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"inspecting existance of %s under %s",
-		name, local->loc.path);
-
-	impunge_frame = copy_frame (frame);
-	if (!impunge_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"out of memory :(");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (impunge_local, afr_local_t, out);
-
-	impunge_frame->local = impunge_local;
-	impunge_sh = &impunge_local->self_heal;
-	impunge_sh->sh_frame = frame;
-	impunge_sh->active_source = active_src;
-
-	ret = build_child_loc (this, &impunge_local->loc, &local->loc, name);
-	if (ret != 0) {
-		goto out;
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == active_src)
-			continue;
-		if (local->child_up[i] == 0)
-			continue;
-		if (sh->sources[i] == 1)
-			continue;
-		call_count++;
-	}
-
-	impunge_local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == active_src)
-			continue;
-		if (local->child_up[i] == 0)
-			continue;
-		if (sh->sources[i] == 1)
-			continue;
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"looking up %s on %s", impunge_local->loc.path,
-			priv->children[i]->name);
-
-		STACK_WIND_COOKIE (impunge_frame,
-				   afr_sh_entry_impunge_entry_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->lookup,
-				   &impunge_local->loc, 0);
-
-		if (!--call_count)
-			break;
-	}
-
-	ret = 0;
 out:
-	if (ret == -1)
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	
-	return 0;
+        LOCK (&impunge_frame->lock);
+        {
+                call_count = --impunge_local->call_count;
+        }
+        UNLOCK (&impunge_frame->lock);
+
+        if (call_count == 0)
+                afr_sh_entry_call_impunge_done (impunge_frame, this,
+                                                op_ret, op_errno);
+
+        return 0;
 }
 
 
 int
-afr_sh_entry_impunge_readdir_cbk (call_frame_t *frame, void *cookie,
-				  xlator_t *this,
-				  int32_t op_ret, int32_t op_errno,
-				  gf_dirent_t *entries)
+afr_sh_entry_impunge_readlink (call_frame_t *impunge_frame, xlator_t *this,
+                               int child_index, struct iatt *stbuf)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	gf_dirent_t     *entry = NULL;
-	off_t            last_offset = 0;
-	int              active_src = 0;
-	int              entry_count = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	active_src = sh->active_source;
-
-	if (op_ret <= 0) {
-		if (op_ret < 0) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"readdir of %s on subvolume %s failed (%s)",
-				local->loc.path,
-				priv->children[active_src]->name,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"readdir of %s on subvolume %s complete",
-				local->loc.path,
-				priv->children[active_src]->name);
-		}
-
-		afr_sh_entry_impunge_all (frame, this);
-		return 0;
-	}
-
-	list_for_each_entry (entry, &entries->list, list) {
-		last_offset = entry->d_off;
-		entry_count++;
-	}
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"readdir'ed %d entries from %s",
-		entry_count, priv->children[active_src]->name);
-
-	sh->offset = last_offset;
-	local->call_count = entry_count;
-
-	list_for_each_entry (entry, &entries->list, list) {
-		afr_sh_entry_impunge_entry (frame, this, entry->d_name);
-	}
-
-	return 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
+        afr_self_heal_t *impunge_sh = NULL;
+        int              active_src = -1;
+
+        priv = this->private;
+        impunge_local = impunge_frame->local;
+        impunge_sh = &impunge_local->self_heal;
+        active_src = impunge_sh->active_source;
+        impunge_local->cont.dir_fop.buf = *stbuf;
+
+        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_readlink_cbk,
+                           (void *) (long) child_index,
+                           priv->children[active_src],
+                           priv->children[active_src]->fops->readlink,
+                           &impunge_local->loc, 4096, NULL);
+
+        return 0;
 }
-				  
 
 int
-afr_sh_entry_impunge_subvol (call_frame_t *frame, xlator_t *this,
-			     int active_src)
+afr_sh_entry_impunge_create (call_frame_t *impunge_frame, xlator_t *this,
+                             int child_index)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	STACK_WIND (frame, afr_sh_entry_impunge_readdir_cbk,
-		    priv->children[active_src],
-		    priv->children[active_src]->fops->readdir,
-		    sh->healing_fd, sh->block_size, sh->offset);
-
-	return 0;
+        call_frame_t    *frame             = NULL;
+        afr_local_t     *impunge_local     = NULL;
+        afr_local_t     *local             = NULL;
+        afr_self_heal_t *impunge_sh        = NULL;
+        afr_self_heal_t *sh                = NULL;
+        afr_private_t   *priv = NULL;
+        ia_type_t       type = IA_INVAL;
+        int             active_src = 0;
+        struct iatt     *buf = NULL;
+
+        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
+                                frame, local, sh);
+        active_src = impunge_sh->active_source;
+        afr_update_loc_gfids (&impunge_local->loc, &impunge_sh->entrybuf,
+                              &impunge_sh->parentbuf);
+
+        buf = &impunge_sh->entrybuf;
+        type = buf->ia_type;
+
+        switch (type) {
+        case IA_IFSOCK:
+        case IA_IFREG:
+        case IA_IFBLK:
+        case IA_IFCHR:
+        case IA_IFIFO:
+        case IA_IFLNK:
+                afr_sh_entry_impunge_check_hardlink (impunge_frame, this,
+                                                     child_index, buf);
+                break;
+        case IA_IFDIR:
+                afr_sh_entry_impunge_mkdir (impunge_frame, this,
+                                            child_index, buf);
+                break;
+        default:
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s has unknown file type on %s: 0%o",
+                        impunge_local->loc.path,
+                        priv->children[active_src]->name, type);
+                sh->impunge_done (frame, this, -1, EINVAL);
+                break;
+        }
+
+        return 0;
 }
 
-
 int
-afr_sh_entry_impunge_all (call_frame_t *frame, xlator_t *this)
+afr_sh_entry_impunge_create_file (call_frame_t *impunge_frame, xlator_t *this,
+                                  int child_index)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              active_src = -1;
+        call_frame_t    *frame             = NULL;
+        afr_local_t     *impunge_local     = NULL;
+        afr_local_t     *local             = NULL;
+        afr_self_heal_t *impunge_sh        = NULL;
+        afr_self_heal_t *sh                = NULL;
+        afr_private_t   *priv = NULL;
+        ia_type_t       type = IA_INVAL;
+        int             active_src = 0;
+        struct iatt     *buf = NULL;
+
+        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
+                                frame, local, sh);
+        active_src = impunge_sh->active_source;
+        buf = &impunge_sh->entrybuf;
+        type = buf->ia_type;
+
+        switch (type) {
+        case IA_IFSOCK:
+        case IA_IFREG:
+        case IA_IFBLK:
+        case IA_IFCHR:
+        case IA_IFIFO:
+                afr_sh_entry_impunge_mknod (impunge_frame, this,
+                                            child_index, buf);
+                break;
+        case IA_IFLNK:
+                afr_sh_entry_impunge_readlink (impunge_frame, this,
+                                               child_index, buf);
+                break;
+        default:
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s has unknown file type on %s: 0%o",
+                        impunge_local->loc.path,
+                        priv->children[active_src]->name, type);
+                sh->impunge_done (frame, this, -1, EINVAL);
+                break;
+        }
+
+        return 0;
+}
 
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
+gf_boolean_t
+afr_sh_need_recreate (afr_self_heal_t *impunge_sh, unsigned int child,
+                      unsigned int child_count)
+{
+        gf_boolean_t    recreate = _gf_false;
 
-	sh->offset = 0;
+        GF_ASSERT (impunge_sh->child_errno);
 
-	active_src = next_active_source (frame, this, sh->active_source);
-	sh->active_source = active_src;
+        if (child == impunge_sh->active_source)
+                goto out;
 
-	if (sh->op_failed) {
-		afr_sh_entry_finish (frame, this);
-		return 0;
-	}
+        if (IA_IFLNK == impunge_sh->entrybuf.ia_type) {
+                recreate = _gf_true;
+                goto out;
+        }
 
-	if (active_src == -1) {
-		/* completed creating missing files on all subvolumes */
-		afr_sh_entry_expunge_all (frame, this);
-		return 0;
-	}
+        if (impunge_sh->child_errno[child] == ENOENT)
+                recreate = _gf_true;
+out:
+        return recreate;
+}
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"impunging entries of %s on %s to other sinks",
-		local->loc.path, priv->children[active_src]->name);
+unsigned int
+afr_sh_recreate_count (afr_self_heal_t *impunge_sh, int *sources,
+                       unsigned int child_count)
+{
+        int             count = 0;
+        int             i = 0;
 
-	afr_sh_entry_impunge_subvol (frame, this, active_src);
+        for (i = 0; i < child_count; i++) {
+                if (afr_sh_need_recreate (impunge_sh, i, child_count))
+                        count++;
+        }
 
-	return 0;
+        return count;
 }
 
-
 int
-afr_sh_entry_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			  int32_t op_ret, int32_t op_errno, fd_t *fd)
+afr_sh_entry_call_impunge_recreate (call_frame_t *impunge_frame,
+                                    xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	/* TODO: some of the open's might fail.
-	   In that case, modify cleanup fn to send flush on those 
-	   fd's which are already open */
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"opendir of %s failed on child %s (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-			sh->op_failed = 1;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		if (sh->op_failed) {
-			afr_sh_entry_finish (frame, this);
-			return 0;
-		}
-		gf_log (this->name, GF_LOG_DEBUG,
-			"fd for %s opened, commencing sync",
-			local->loc.path);
-
-		sh->active_source = -1;
-		afr_sh_entry_impunge_all (frame, this);
-	}
-
-	return 0;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
+        afr_self_heal_t *impunge_sh = NULL;
+        call_frame_t    *frame = NULL;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        unsigned int     recreate_count = 0;
+        int              i = 0;
+        int              active_src = 0;
+
+        priv          = this->private;
+        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
+                                frame, local, sh);
+        active_src = impunge_sh->active_source;
+        impunge_sh->entrybuf = impunge_sh->buf[active_src];
+        impunge_sh->parentbuf = impunge_sh->parentbufs[active_src];
+        recreate_count = afr_sh_recreate_count (impunge_sh, sh->sources,
+                                                priv->child_count);
+        if (!recreate_count) {
+                afr_sh_entry_call_impunge_done (impunge_frame, this, 0, 0);
+                goto out;
+        }
+        impunge_local->call_count = recreate_count;
+        for (i = 0; i < priv->child_count; i++) {
+                if (!impunge_local->child_up[i]) {
+                        impunge_sh->child_errno[i] = ENOTCONN;
+                        continue;
+                }
+                if (!afr_sh_need_recreate (impunge_sh, i, priv->child_count)) {
+                        impunge_sh->child_errno[i] = EEXIST;
+                        continue;
+                }
+        }
+        for (i = 0; i < priv->child_count; i++) {
+                if (!afr_sh_need_recreate (impunge_sh, i, priv->child_count))
+                        continue;
+                (void)afr_sh_entry_impunge_create (impunge_frame, this, i);
+                recreate_count--;
+        }
+        GF_ASSERT (!recreate_count);
+out:
+        return 0;
 }
 
+void
+afr_sh_entry_common_lookup_done (call_frame_t *impunge_frame, xlator_t *this,
+                                 int32_t op_ret, int32_t op_errno)
+{
+        afr_private_t   *priv = NULL;
+        afr_local_t     *impunge_local = NULL;
+        afr_self_heal_t *impunge_sh = NULL;
+        call_frame_t    *frame = NULL;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        unsigned int     gfid_miss_count = 0;
+        unsigned int     children_up_count = 0;
+        uuid_t           gfid = {0};
+        int              active_src = 0;
+
+        priv          = this->private;
+        AFR_INIT_SH_FRAME_VALS (impunge_frame, impunge_local, impunge_sh,
+                                frame, local, sh);
+        active_src    = impunge_sh->active_source;
+
+        if (op_ret < 0)
+                goto done;
+        if (impunge_sh->child_errno[active_src]) {
+                op_ret = -1;
+                op_errno = impunge_sh->child_errno[active_src];
+                goto done;
+        }
+
+        gfid_miss_count = afr_gfid_missing_count (this->name,
+                                                  impunge_sh->success_children,
+                                                  impunge_sh->buf, priv->child_count,
+                                                  impunge_local->loc.path);
+        children_up_count = afr_up_children_count (impunge_local->child_up,
+                                                   priv->child_count);
+        if ((gfid_miss_count == children_up_count) &&
+            (children_up_count < priv->child_count)) {
+                op_ret = -1;
+                op_errno = ENODATA;
+                gf_log (this->name, GF_LOG_ERROR, "Not all children are up, "
+                        "gfid should not be assigned in this state for %s",
+                        impunge_local->loc.path);
+                goto done;
+        }
+
+        if (gfid_miss_count) {
+                afr_update_gfid_from_iatts (gfid, impunge_sh->buf,
+                                            impunge_sh->success_children,
+                                            priv->child_count);
+                if (uuid_is_null (gfid)) {
+                        sh->entries_skipped = _gf_true;
+                        gf_log (this->name, GF_LOG_INFO, "%s: Skipping entry "
+                                "self-heal because of gfid absence",
+                                impunge_local->loc.path);
+                        goto done;
+                }
+                afr_sh_common_lookup (impunge_frame, this, &impunge_local->loc,
+                                      afr_sh_entry_common_lookup_done, gfid,
+                                      AFR_LOOKUP_FAIL_CONFLICTS |
+                                      AFR_LOOKUP_FAIL_MISSING_GFIDS,
+                                      NULL);
+        } else {
+                afr_sh_entry_call_impunge_recreate (impunge_frame, this);
+        }
+        return;
+done:
+        afr_sh_entry_call_impunge_done (impunge_frame, this,
+                                        op_ret, op_errno);
+        return;
+}
 
 int
-afr_sh_entry_open (call_frame_t *frame, xlator_t *this)
+afr_sh_entry_impunge_entry (call_frame_t *frame, xlator_t *this,
+                            gf_dirent_t *entry)
 {
-	int i = 0;				
-	int call_count = 0;		     
-
-	int source = -1;
-	int *sources = NULL;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh  = NULL;
+        afr_self_heal_t *impunge_sh  = NULL;
+        int              ret = -1;
+        call_frame_t    *impunge_frame = NULL;
+        afr_local_t     *impunge_local = NULL;
+        int              active_src = 0;
+        int              op_errno = 0;
+        int              op_ret = -1;
+
+        local = frame->local;
+        sh = &local->self_heal;
+
+        active_src = sh->active_source;
+        sh->impunge_done = afr_sh_entry_impunge_entry_done;
+
+        if (can_skip_entry_self_heal (entry->d_name, &local->loc)) {
+                op_ret = 0;
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "inspecting existence of %s under %s",
+                entry->d_name, local->loc.path);
+
+        ret = afr_impunge_frame_create (frame, this, active_src,
+                                        &impunge_frame);
+        if (ret) {
+                op_errno = -ret;
+                goto out;
+        }
+
+        impunge_local = impunge_frame->local;
+        impunge_sh = &impunge_local->self_heal;
+        ret = afr_build_child_loc (this, &impunge_local->loc, &local->loc,
+                                   entry->d_name);
+        loc_copy (&impunge_sh->parent_loc, &local->loc);
+        if (ret != 0) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        afr_sh_common_lookup (impunge_frame, this, &impunge_local->loc,
+                              afr_sh_entry_common_lookup_done, NULL,
+                              AFR_LOOKUP_FAIL_CONFLICTS, NULL);
+
+        op_ret = 0;
+out:
+        if (ret) {
+                if (impunge_frame)
+                        AFR_STACK_DESTROY (impunge_frame);
+                sh->impunge_done (frame, this, op_ret, op_errno);
+        }
 
-	fd_t *fd = NULL;
+        return 0;
+}
 
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	afr_self_heal_t *sh = NULL;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+int
+afr_sh_entry_impunge_readdir_cbk (call_frame_t *frame, void *cookie,
+                                  xlator_t *this,
+                                  int32_t op_ret, int32_t op_errno,
+                                  gf_dirent_t *entries, dict_t *xdata)
+{
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh  = NULL;
+        gf_dirent_t     *entry = NULL;
+        off_t            last_offset = 0;
+        int              active_src = 0;
+        int              entry_count = 0;
+
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
+
+        active_src = sh->active_source;
+
+        if (op_ret <= 0) {
+                if (op_ret < 0) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "readdir of %s on subvolume %s failed (%s)",
+                                local->loc.path,
+                                priv->children[active_src]->name,
+                                strerror (op_errno));
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                } else {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "readdir of %s on subvolume %s complete",
+                                local->loc.path,
+                                priv->children[active_src]->name);
+                }
+
+                afr_sh_entry_impunge_all (frame, this);
+                return 0;
+        }
+
+        list_for_each_entry (entry, &entries->list, list) {
+                last_offset = entry->d_off;
+                entry_count++;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "readdir'ed %d entries from %s",
+                entry_count, priv->children[active_src]->name);
+
+        sh->offset = last_offset;
+        local->call_count = entry_count;
+
+        list_for_each_entry (entry, &entries->list, list) {
+                afr_sh_entry_impunge_entry (frame, this, entry);
+        }
+
+        return 0;
+}
 
-	source  = local->self_heal.source;
-	sources = local->self_heal.sources;
 
-	sh->block_size = 131072;
-	sh->offset = 0;
+int
+afr_sh_entry_impunge_subvol (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh  = NULL;
+        int32_t         active_src = 0;
+
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
+        active_src = sh->active_source;
+        gf_log (this->name, GF_LOG_DEBUG, "%s: readdir from offset %zd",
+                local->loc.path, sh->offset);
+
+        STACK_WIND (frame, afr_sh_entry_impunge_readdir_cbk,
+                    priv->children[active_src],
+                    priv->children[active_src]->fops->readdirp,
+                    sh->healing_fd, sh->block_size, sh->offset, NULL);
+
+        return 0;
+}
 
-	call_count = sh->active_sinks;
-	if (source != -1)
-		call_count++;
 
-	local->call_count = call_count;
+int
+afr_sh_entry_impunge_all (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local  = NULL;
+        afr_self_heal_t *sh  = NULL;
+        int              active_src = -1;
 
-	fd = fd_create (local->loc.inode, frame->root->pid);
-	sh->healing_fd = fd;
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
 
-	if (source != -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"opening directory %s on subvolume %s (source)",
-			local->loc.path, priv->children[source]->name);
+        sh->offset = 0;
 
-		/* open source */
-		STACK_WIND_COOKIE (frame, afr_sh_entry_opendir_cbk,
-				   (void *) (long) source,
-				   priv->children[source],
-				   priv->children[source]->fops->opendir,
-				   &local->loc, fd);
-		call_count--;
-	}
+        active_src = next_active_source (frame, this, sh->active_source);
+        sh->active_source = active_src;
 
-	/* open sinks */
-	for (i = 0; i < priv->child_count; i++) {
-		if (sources[i] || !local->child_up[i])
-			continue;
+        if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
+                afr_sh_entry_finish (frame, this);
+                return 0;
+        }
 
-		gf_log (this->name, GF_LOG_DEBUG,
-			"opening directory %s on subvolume %s (sink)",
-			local->loc.path, priv->children[i]->name);
+        if (active_src == -1) {
+                /* completed creating missing files on all subvolumes */
+                afr_sh_entry_erase_pending (frame, this);
+                return 0;
+        }
 
-		STACK_WIND_COOKIE (frame, afr_sh_entry_opendir_cbk,
-				   (void *) (long) i,
-				   priv->children[i], 
-				   priv->children[i]->fops->opendir,
-				   &local->loc, fd);
+        gf_log (this->name, GF_LOG_TRACE,
+                "impunging entries of %s on %s to other sinks",
+                local->loc.path, priv->children[active_src]->name);
 
-		if (!--call_count)
-			break;
-	}
+        afr_sh_entry_impunge_subvol (frame, this);
 
-	return 0;
+        return 0;
 }
 
 
 int
-afr_sh_entry_sync_prepare (call_frame_t *frame, xlator_t *this)
+afr_sh_entry_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              active_sinks = 0;
-	int              source = 0;
-	int              i = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	source = sh->source;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->sources[i] == 0 && local->child_up[i] == 1) {
-			active_sinks++;
-			sh->success[i] = 1;
-		}
-	}
-	if (source != -1)
-		sh->success[source] = 1;
-
-	if (active_sinks == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no active sinks for self-heal on dir %s",
-			local->loc.path);
-		afr_sh_entry_finish (frame, this);
-		return 0;
-	}
-	if (source == -1 && active_sinks < 2) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"cannot sync with 0 sources and 1 sink on dir %s",
-			local->loc.path);
-		afr_sh_entry_finish (frame, this);
-		return 0;
-	}
-	sh->active_sinks = active_sinks;
-
-	if (source != -1)
-		gf_log (this->name, GF_LOG_DEBUG,
-			"syncing %s from subvolume %s to %d active sinks",
-			local->loc.path, priv->children[source]->name,
-			active_sinks);
-	else
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no active sources for %s found. "
-			"merging all entries as a conservative decision",
-			local->loc.path);
-
-	afr_sh_entry_open (frame, this);
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              call_count = 0;
+        int              child_index = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        child_index = (long) cookie;
+
+        /* TODO: some of the open's might fail.
+           In that case, modify cleanup fn to send flush on those
+           fd's which are already open */
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "opendir of %s failed on child %s (%s)",
+                                local->loc.path,
+                                priv->children[child_index]->name,
+                                strerror (op_errno));
+                        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) {
+                        afr_sh_entry_finish (frame, this);
+                        return 0;
+                }
+                gf_log (this->name, GF_LOG_TRACE,
+                        "fd for %s opened, commencing sync",
+                        local->loc.path);
+
+                sh->active_source = -1;
+                afr_sh_entry_expunge_all (frame, this);
+        }
+
+        return 0;
 }
 
 
 int
-afr_sh_entry_fix (call_frame_t *frame, xlator_t *this)
+afr_sh_entry_open (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              source = 0;
+        int i = 0;
+        int call_count = 0;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        int source = -1;
+        int *sources = NULL;
 
-	afr_sh_build_pending_matrix (sh->pending_matrix, sh->xattr, 
-				     priv->child_count, AFR_ENTRY_PENDING);
+        fd_t *fd = NULL;
 
-	afr_sh_print_pending_matrix (sh->pending_matrix, this);
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = NULL;
+        afr_self_heal_t *sh = NULL;
 
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
 
-	afr_sh_mark_sources (sh->pending_matrix, sh->sources, 
-			     priv->child_count);
+        source  = local->self_heal.source;
+        sources = local->self_heal.sources;
 
-	afr_sh_supress_errenous_children (sh->sources, sh->child_errno,
-					  priv->child_count);
+        sh->block_size = priv->sh_readdir_size;
+        sh->offset = 0;
 
-	source = afr_sh_select_source (sh->sources, priv->child_count);
-	sh->source = source;
+        call_count = sh->active_sinks;
+        if (source != -1)
+                call_count++;
 
-	afr_sh_entry_sync_prepare (frame, this);
+        local->call_count = call_count;
 
-	return 0;
-}
+        fd = fd_create (local->loc.inode, frame->root->pid);
+        sh->healing_fd = fd;
 
+        if (source != -1) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "opening directory %s on subvolume %s (source)",
+                        local->loc.path, priv->children[source]->name);
 
+                /* open source */
+                STACK_WIND_COOKIE (frame, afr_sh_entry_opendir_cbk,
+                                   (void *) (long) source,
+                                   priv->children[source],
+                                   priv->children[source]->fops->opendir,
+                                   &local->loc, fd, NULL);
+                call_count--;
+        }
 
-int
-afr_sh_entry_lookup_cbk (call_frame_t *frame, void *cookie,
-			 xlator_t *this, int32_t op_ret, int32_t op_errno,
-			 inode_t *inode, struct stat *buf, dict_t *xattr)
-{
-	afr_private_t   *priv  = NULL;
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-
-	int call_count  = -1;
-	int child_index = (long) cookie;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+        /* open sinks */
+        for (i = 0; i < priv->child_count; i++) {
+                if (sources[i] || !local->child_up[i])
+                        continue;
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret != -1) {
-			sh->xattr[child_index] = dict_ref (xattr);
-			sh->buf[child_index] = *buf;
-		}
-	}
-	UNLOCK (&frame->lock);
+                gf_log (this->name, GF_LOG_TRACE,
+                        "opening directory %s on subvolume %s (sink)",
+                        local->loc.path, priv->children[i]->name);
 
-	call_count = afr_frame_return (frame);
+                STACK_WIND_COOKIE (frame, afr_sh_entry_opendir_cbk,
+                                   (void *) (long) i,
+                                   priv->children[i],
+                                   priv->children[i]->fops->opendir,
+                                   &local->loc, fd, NULL);
 
-	if (call_count == 0) {
-		afr_sh_entry_fix (frame, this);
-	}
+                if (!--call_count)
+                        break;
+        }
 
-	return 0;
+        return 0;
 }
 
 
-
 int
-afr_sh_entry_lookup (call_frame_t *frame, xlator_t *this)
+afr_sh_entry_sync_prepare (call_frame_t *frame, xlator_t *this)
 {
-	afr_self_heal_t * sh    = NULL; 
-	afr_local_t    *  local = NULL;
-	afr_private_t  *  priv  = NULL;
-	dict_t         *xattr_req = NULL;
-	int ret = 0;
-	int call_count = 0;
-	int i = 0;
-
-	priv  = this->private;
-	local = frame->local;
-	sh    = &local->self_heal;
-
-	call_count = local->child_count;
-
-	local->call_count = call_count;
-	
-	xattr_req = dict_new();
-	if (xattr_req)
-		ret = dict_set_uint64 (xattr_req, AFR_ENTRY_PENDING,
-				       priv->child_count * sizeof(int32_t));
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame,
-					   afr_sh_entry_lookup_cbk,
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->lookup,
-					   &local->loc, xattr_req);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	if (xattr_req)
-		dict_unref (xattr_req);
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              source = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        source = sh->source;
+
+        afr_sh_mark_source_sinks (frame, this);
+        if (source != -1)
+                sh->success[source] = 1;
+
+        if (sh->active_sinks == 0) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "no active sinks for self-heal on dir %s",
+                        local->loc.path);
+                afr_sh_entry_finish (frame, this);
+                return 0;
+        }
+        if (source == -1 && sh->active_sinks < 2) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "cannot sync with 0 sources and 1 sink on dir %s",
+                        local->loc.path);
+                afr_sh_entry_finish (frame, this);
+                return 0;
+        }
+
+        if (source != -1)
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "self-healing directory %s from subvolume %s to "
+                        "%d other",
+                        local->loc.path, priv->children[source]->name,
+                        sh->active_sinks);
+        else
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no active sources for %s found. "
+                        "merging all entries as a conservative decision",
+                        local->loc.path);
+
+        sh->actual_sh_started = _gf_true;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN);
+        afr_sh_entry_open (frame, this);
+
+        return 0;
 }
 
 
-
-int
-afr_sh_entry_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno)
+void
+afr_sh_entry_fix (call_frame_t *frame, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	int              call_count = 0;
-	int              child_index = (long) cookie;
-
-	/* TODO: what if lock fails? */
-	
-	local = frame->local;
-	sh    = &local->self_heal;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			sh->op_failed = 1;
-
-			gf_log (this->name,
-				(op_errno == EAGAIN ? GF_LOG_DEBUG : GF_LOG_ERROR),
-				"locking inode of %s on child %d failed: %s",
-				local->loc.path, child_index,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"inode of %s on child %d locked",
-				local->loc.path, child_index);
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		if (sh->op_failed == 1) {
-			afr_sh_entry_finish (frame, this);
-			return 0;
-		}
-
-		afr_sh_entry_lookup (frame, this);
-	}
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              source = 0;
+        int              nsources = 0;
+        int32_t          subvol_status = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        if (op_ret < 0) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                afr_sh_set_error (sh, op_errno);
+                afr_sh_entry_finish (frame, this);
+                goto out;
+        }
+
+        if (sh->forced_merge) {
+                sh->source = -1;
+                goto heal;
+        }
+
+        nsources = afr_build_sources (this, sh->xattr, sh->buf,
+                                      sh->pending_matrix, sh->sources,
+                                      sh->success_children,
+                                      AFR_ENTRY_TRANSACTION, &subvol_status,
+                                      _gf_true);
+        if ((subvol_status & ALL_FOOLS) ||
+            (subvol_status & SPLIT_BRAIN)) {
+                gf_log (this->name, GF_LOG_INFO, "%s: Performing conservative "
+                        "merge", local->loc.path);
+                source = -1;
+                memset (sh->sources, 0,
+                        sizeof (*sh->sources) * priv->child_count);
+        } else if (nsources == 0) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "No self-heal needed for %s",
+                        local->loc.path);
+
+                afr_sh_entry_finish (frame, this);
+                return;
+        } else {
+                source = afr_sh_select_source (sh->sources, priv->child_count);
+        }
+
+        sh->source = source;
+
+        afr_reset_children (sh->fresh_children, priv->child_count);
+        afr_get_fresh_children (sh->success_children, sh->sources,
+                                sh->fresh_children, priv->child_count);
+        if (sh->source >= 0)
+                afr_inode_set_read_ctx (this, sh->inode, sh->source,
+                                        sh->fresh_children);
+
+heal:
+        afr_sh_entry_sync_prepare (frame, this);
+out:
+        return;
 }
 
-
 int
-afr_sh_entry_lock (call_frame_t *frame, xlator_t *this)
+afr_sh_post_nonblocking_entry_cbk (call_frame_t *frame, xlator_t *this)
 {
-	int i = 0;				
-	int call_count = 0;		     
-
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	afr_self_heal_t * sh  = NULL;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	call_count = local->child_count;
-
-	local->call_count = call_count;		
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"locking %s on subvolume %s",
-				local->loc.path, priv->children[i]->name);
-
-			STACK_WIND_COOKIE (frame, afr_sh_entry_lock_cbk,
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->entrylk,
-                                           this->name,
-					   &local->loc, NULL,
-					   ENTRYLK_LOCK_NB, ENTRYLK_WRLCK);
-			if (!--call_count)
-				break;
-		}
-	}
-
-	return 0;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+        afr_self_heal_t     *sh       = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        sh       = &local->self_heal;
+
+        if (int_lock->lock_op_ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR, "Non Blocking entrylks "
+                        "failed for %s.", local->loc.path);
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                afr_sh_entry_done (frame, this);
+        } else {
+
+                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking entrylks done "
+                        "for %s. Proceeding to FOP", local->loc.path);
+                afr_sh_common_lookup (frame, this, &local->loc,
+                                      afr_sh_entry_fix, NULL,
+                                      AFR_LOOKUP_FAIL_CONFLICTS |
+                                      AFR_LOOKUP_FAIL_MISSING_GFIDS,
+                                      NULL);
+        }
+
+        return 0;
 }
 
-
 int
 afr_self_heal_entry (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t   *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	if (local->need_entry_self_heal && priv->entry_self_heal) {
-		afr_sh_entry_lock (frame, this);
-	} else {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"proceeding to completion on %s",
-			local->loc.path);
-		afr_sh_entry_done (frame, this);
-	}
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_private_t   *priv = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        priv = this->private;
+        local = frame->local;
+        sh = &local->self_heal;
+
+        sh->sh_type_in_action = AFR_SELF_HEAL_ENTRY;
+
+        if (local->self_heal.do_entry_self_heal && priv->entry_self_heal) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
+                afr_sh_entrylk (frame, this, &local->loc, NULL,
+                                afr_sh_post_nonblocking_entry_cbk);
+        } else {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "proceeding to completion on %s",
+                        local->loc.path);
+                afr_sh_entry_done (frame, this);
+        }
+
+        return 0;
 }
-
diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c
index de8fa1a39..fd5da6cfd 100644
--- a/xlators/cluster/afr/src/afr-self-heal-metadata.c
+++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include <libgen.h>
@@ -52,742 +43,728 @@
 int
 afr_sh_metadata_done (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-//	memset (sh->child_errno, 0, sizeof (int) * priv->child_count);
-	memset (sh->buf, 0, sizeof (struct stat) * priv->child_count);
-	memset (sh->success, 0, sizeof (int) * priv->child_count);
-	
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->xattr[i])
-			dict_unref (sh->xattr[i]);
-		sh->xattr[i] = NULL;
-	}
-
-	if (local->govinda_gOvinda) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"aborting selfheal of %s",
-			local->loc.path);
-		sh->completion_cbk (frame, this);
-	} else {
-		if (S_ISREG (local->cont.lookup.buf.st_mode)) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"proceeding to data check on %s",
-				local->loc.path);
-			afr_self_heal_data (frame, this);
-			return 0;
-		}
-
-		if (S_ISDIR (local->cont.lookup.buf.st_mode)) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"proceeding to entry check on %s",
-				local->loc.path);
-			afr_self_heal_entry (frame, this);
-			return 0;
-		}
-		gf_log (this->name, GF_LOG_DEBUG,
-			"completed self heal of %s",
-			local->loc.path);
-
-		sh->completion_cbk (frame, this);
-	}
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+
+        local = frame->local;
+        sh = &local->self_heal;
+
+        afr_sh_reset (frame, this);
+        if (IA_ISDIR (sh->type)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "proceeding to entry check on %s",
+                        local->loc.path);
+                afr_self_heal_entry (frame, this);
+        } else {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "proceeding to data check on %s",
+                        local->loc.path);
+                afr_self_heal_data (frame, this);
+        }
+
+        return 0;
 }
 
-
 int
-afr_sh_metadata_unlck_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno)
+afr_sh_inode_unlock (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t      *local = NULL;
-	int               call_count = 0;
-
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
 
-	local = frame->local;
+        local    = frame->local;
+        int_lock = &local->internal_lock;
 
-	LOCK (&frame->lock);
-	{
-	}
-	UNLOCK (&frame->lock);
+        int_lock->lock_cbk = afr_sh_metadata_done;
+        afr_unlock (frame, this);
 
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		afr_sh_metadata_done (frame, this);
-
-	return 0;
+        return 0;
 }
 
-
 int
 afr_sh_metadata_finish (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-	int              call_count = 0;
-	struct flock     flock = {0, };
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	call_count = local->child_count;
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {
-		flock.l_start   = 0;
-		flock.l_len     = 0;
-		flock.l_type    = F_UNLCK;
-
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"unlocking %s on subvolume %s",
-				local->loc.path, priv->children[i]->name);
-
-			STACK_WIND (frame, afr_sh_metadata_unlck_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->inodelk,
-                                    this->name,
-				    &local->loc, F_SETLK, &flock);
-
-			if (!--call_count)
-				break;
-		}
-	}
-
-	return 0;
-}
+        afr_sh_inode_unlock (frame, this);
 
+        return 0;
+}
 
 int
-afr_sh_metadata_erase_pending_cbk (call_frame_t *frame, void *cookie,
-				   xlator_t *this, int32_t op_ret,
-				   int32_t op_errno, dict_t *xattr)
+afr_sh_metadata_fail (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int             call_count = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	LOCK (&frame->lock);
-	{
-	}
-	UNLOCK (&frame->lock);
+        afr_local_t         *local    = NULL;
+        afr_self_heal_t     *sh       = NULL;
 
-	call_count = afr_frame_return (frame);
+        local    = frame->local;
+        sh       = &local->self_heal;
 
-	if (call_count == 0)
-		afr_sh_metadata_finish (frame, this);
-
-	return 0;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+        afr_sh_metadata_finish (frame, this);
+        return 0;
 }
 
+int
+afr_sh_metadata_erase_pending_cbk (call_frame_t *frame, void *cookie,
+                                   xlator_t *this, int32_t op_ret,
+                                   int32_t op_errno, dict_t *xattr, dict_t *xdata)
+{
+        afr_local_t     *local     = NULL;
+        int             call_count = 0;
+        long            i          = 0;
+        afr_self_heal_t *sh        = NULL;
+        afr_private_t   *priv      = NULL;
+
+        local = frame->local;
+        priv  = this->private;
+        sh = &local->self_heal;
+        i = (long)cookie;
+
+        if ((!IA_ISREG (sh->buf[sh->source].ia_type)) &&
+            (!IA_ISDIR (sh->buf[sh->source].ia_type))) {
+                afr_children_add_child (sh->fresh_children, i,
+                                        priv->child_count);
+        }
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                if ((!IA_ISREG (sh->buf[sh->source].ia_type)) &&
+                    (!IA_ISDIR (sh->buf[sh->source].ia_type))) {
+                        afr_inode_set_read_ctx (this, sh->inode, sh->source,
+                                                sh->fresh_children);
+                }
+                afr_sh_metadata_finish (frame, this);
+        }
+
+        return 0;
+}
 
 int
 afr_sh_metadata_erase_pending (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              i = 0;
-	dict_t          **erase_xattr = NULL;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-
-	afr_sh_pending_to_delta (sh->pending_matrix, sh->delta_matrix,
-				 sh->success, priv->child_count);
-
-	erase_xattr = CALLOC (sizeof (*erase_xattr), priv->child_count);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->xattr[i]) {
-			call_count++;
-
-			erase_xattr[i] = get_new_dict();
-			dict_ref (erase_xattr[i]);
-		}
-	}
-
-	afr_sh_delta_to_xattr (sh->delta_matrix, erase_xattr,
-			       priv->child_count, AFR_METADATA_PENDING);
-
-	local->call_count = call_count;
-
-	if (call_count == 0) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"metadata of %s not healed on any subvolume",
-			local->loc.path);
-
-		afr_sh_metadata_finish (frame, this);
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (!erase_xattr[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"erasing pending flags from %s on %s",
-			local->loc.path, priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_metadata_erase_pending_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->xattrop,
-				   &local->loc,
-				   GF_XATTROP_ADD_ARRAY, erase_xattr[i]);
-		if (!--call_count)
-			break;
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (erase_xattr[i]) {
-			dict_unref (erase_xattr[i]);
-		}
-	}
-	FREE (erase_xattr);
-
-	return 0;
+         afr_sh_erase_pending (frame, this, AFR_METADATA_TRANSACTION,
+                               afr_sh_metadata_erase_pending_cbk,
+                               afr_sh_metadata_finish);
+         return 0;
 }
 
 
 int
 afr_sh_metadata_sync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			  int32_t op_ret, int32_t op_errno)
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"setting attributes failed for %s on %s (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-
-			sh->success[child_index] = 0;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		afr_sh_metadata_erase_pending (frame, this);
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              call_count = 0;
+        int              child_index = 0;
+
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        child_index = (long) cookie;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "setting attributes failed for %s on %s (%s)",
+                                local->loc.path,
+                                priv->children[child_index]->name,
+                                strerror (op_errno));
+
+                        sh->success[child_index] = 0;
+                }
+        }
+        UNLOCK (&frame->lock);
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0) {
+                if (local->xattr_req) {
+                        dict_unref (local->xattr_req);
+                        local->xattr_req = NULL;
+                }
+                afr_sh_metadata_erase_pending (frame, this);
+        }
+
+        return 0;
 }
 
 
 int
-afr_sh_metadata_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			  int32_t op_ret, int32_t op_errno, struct stat *buf)
+afr_sh_metadata_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno,
+                             struct iatt *preop, struct iatt *postop, dict_t *xdata)
 {
-	afr_sh_metadata_sync_cbk (frame, cookie, this, op_ret, op_errno);
+        afr_sh_metadata_sync_cbk (frame, cookie, this, op_ret, op_errno, xdata);
 
-	return 0;
+        return 0;
 }
 
 
 int
 afr_sh_metadata_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno)
+                           int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	afr_sh_metadata_sync_cbk (frame, cookie, this, op_ret, op_errno);
+        afr_sh_metadata_sync_cbk (frame, cookie, this, op_ret, op_errno, xdata);
 
-	return 0;
+        return 0;
 }
 
-
 int
-afr_sh_metadata_sync (call_frame_t *frame, xlator_t *this, dict_t *xattr)
+afr_sh_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno,
+                        dict_t *xdata)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              source = 0;
-	int              active_sinks = 0;
-	int              call_count = 0;
-	int              i = 0;
-	struct timespec  ts[2];
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	source = sh->source;
-	active_sinks = sh->active_sinks;
-
-	/*
-	 * 4 calls per sink - chown, chmod, utimes, setxattr
-	 */
-	if (xattr)
-		call_count = active_sinks * 4;
-	else
-		call_count = active_sinks * 3;
-
-	local->call_count = call_count;
-
-#ifdef HAVE_STRUCT_STAT_ST_ATIM_TV_NSEC
-	ts[0] = sh->buf[source].st_atim;
-	ts[1] = sh->buf[source].st_mtim;
-#elif HAVE_STRUCT_STAT_ST_ATIMESPEC_TV_NSEC
-	ts[0] = sh->buf[source].st_atimespec;
-	ts[1] = sh->buf[source].st_mtimespec;
-#else
-	ts[0].tv_sec = sh->buf[source].st_atime;
-	ts[1].tv_sec = sh->buf[source].st_mtime;
-#endif
+        int            i     = 0;
+        afr_private_t *priv  = NULL;
+        afr_local_t   *local = NULL;
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (call_count == 0) {
-			break;
-		}
-		if (sh->sources[i] || !local->child_up[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"syncing metadata of %s from %s to %s",
-			local->loc.path, priv->children[source]->name,
-			priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_metadata_attr_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->chown,
-				   &local->loc,
-				   sh->buf[source].st_uid,
-				   sh->buf[source].st_gid);
-
-		STACK_WIND_COOKIE (frame, afr_sh_metadata_attr_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->chmod,
-				   &local->loc, sh->buf[source].st_mode);
-
-		STACK_WIND_COOKIE (frame, afr_sh_metadata_attr_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->utimens,
-				   &local->loc, ts);
-
-		call_count = call_count - 3;
-
-		if (!xattr)
-			continue;
-
-		STACK_WIND_COOKIE (frame, afr_sh_metadata_xattr_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->setxattr,
-				   &local->loc, xattr, 0);
-		call_count--;
-	}
-
-	return 0;
-}
+        priv = this->private;
+        local = frame->local;
 
+        if (op_ret < 0) {
+                afr_sh_metadata_sync_cbk (frame, cookie,
+                                          this, -1, op_errno, xdata);
+                goto out;
+        }
 
-int
-afr_sh_metadata_getxattr_cbk (call_frame_t *frame, void *cookie,
-			      xlator_t *this,
-			      int32_t op_ret, int32_t op_errno, dict_t *xattr)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              source = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	source = sh->source;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"getxattr of %s failed on subvolume %s (%s). proceeding without xattr",
-			local->loc.path, priv->children[source]->name,
-			strerror (op_errno));
-
-		afr_sh_metadata_sync (frame, this, NULL);
-	} else {
-		dict_del (xattr, AFR_DATA_PENDING);
-		dict_del (xattr, AFR_METADATA_PENDING);
-		dict_del (xattr, AFR_ENTRY_PENDING);
-		afr_sh_metadata_sync (frame, this, xattr);
-	}
-
-	return 0;
-}
+        i = (long) cookie;
 
+        STACK_WIND_COOKIE (frame, afr_sh_metadata_xattr_cbk,
+                           (void *) (long) i,
+                           priv->children[i],
+                           priv->children[i]->fops->setxattr,
+                           &local->loc, local->xattr_req, 0, NULL);
 
-int
-afr_sh_metadata_sync_prepare (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              active_sinks = 0;
-	int              source = 0;
-	int              i = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	source = sh->source;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->sources[i] == 0 && local->child_up[i] == 1) {
-			active_sinks++;
-			sh->success[i] = 1;
-		}
-	}
-	sh->success[source] = 1;
-
-	if (active_sinks == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no active sinks for performing self-heal on file %s",
-			local->loc.path);
-		afr_sh_metadata_finish (frame, this);
-		return 0;
-	}
-	sh->active_sinks = active_sinks;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"syncing metadata of %s from subvolume %s to %d active sinks",
-		local->loc.path, priv->children[source]->name, active_sinks);
-
-	STACK_WIND (frame, afr_sh_metadata_getxattr_cbk,
-		    priv->children[source],
-		    priv->children[source]->fops->getxattr,
-		    &local->loc, NULL);
-
-	return 0;
+ out:
+        return 0;
 }
 
-
-int
-afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this)
+inline void
+afr_prune_special_keys (dict_t *xattr_dict)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              nsources = 0;
-	int              source = 0;
-	int              i = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	afr_sh_build_pending_matrix (sh->pending_matrix, sh->xattr, 
-				     priv->child_count, AFR_METADATA_PENDING);
-
-	afr_sh_print_pending_matrix (sh->pending_matrix, this);
-
-	afr_sh_mark_sources (sh->pending_matrix, sh->sources, 
-			     priv->child_count);
-
-	afr_sh_supress_errenous_children (sh->sources, sh->child_errno,
-					  priv->child_count);
-
-	nsources = afr_sh_source_count (sh->sources, priv->child_count);
-
-	if ((nsources == 0)
-	    && (priv->favorite_child != -1)
-	    && (sh->child_errno[priv->favorite_child] == 0)) {
-
-		gf_log (this->name, GF_LOG_WARNING,
-			"Picking favorite child %s as authentic source to resolve conflicting metadata of %s",
-			priv->children[priv->favorite_child]->name,
-			local->loc.path);
-
-		sh->sources[priv->favorite_child] = 1;
+        dict_del (xattr_dict, GF_SELINUX_XATTR_KEY);
+}
 
-		nsources = afr_sh_source_count (sh->sources,
-						priv->child_count);
-	}
+inline void
+afr_prune_pending_keys (dict_t *xattr_dict, afr_private_t *priv)
+{
+        int i = 0;
 
-	if (nsources == 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Unable to resolve conflicting metadata of %s. "
-			"Please resolve manually by fixing the "
-			"permissions/ownership of %s on your subvolumes. "
-			"You can also consider 'option favorite-child <>'",
-			local->loc.path, local->loc.path);
+        for (; i < priv->child_count; i++) {
+                dict_del (xattr_dict, priv->pending_key[i]);
+        }
+}
 
-		local->govinda_gOvinda = 1;
+int
+afr_sh_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                     dict_t *xdata)
+{
+        int            i     = 0;
+        afr_private_t *priv  = NULL;
+        afr_local_t   *local = NULL;
 
-		afr_sh_metadata_finish (frame, this);
-		return 0;
-	}
+        priv = this->private;
+        local = frame->local;
 
-	source = afr_sh_select_source (sh->sources, priv->child_count);
-	sh->source = source;
+        if (op_ret < 0) {
+                afr_sh_metadata_sync_cbk (frame, cookie,
+                                          this, -1, op_errno, xdata);
+                goto out;
+        }
 
-	/* detect changes not visible through pending flags -- JIC */
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == source || sh->child_errno[i])
-			continue;
+        afr_prune_pending_keys (xattr, priv);
 
-		if (PERMISSION_DIFFERS (&sh->buf[i], &sh->buf[source]))
-			sh->sources[i] = 0;
+        afr_prune_special_keys (xattr);
 
-		if (OWNERSHIP_DIFFERS (&sh->buf[i], &sh->buf[source]))
-			sh->sources[i] = 0;
-	}
+        i = (long) cookie;
 
-	afr_sh_metadata_sync_prepare (frame, this);
+        /* send removexattr in bulk via xdata */
+        STACK_WIND_COOKIE (frame, afr_sh_removexattr_cbk,
+                           cookie,
+                           priv->children[i],
+                           priv->children[i]->fops->removexattr,
+                           &local->loc, "", xattr);
 
-	return 0;
+ out:
+        return 0;
 }
 
-
 int
-afr_sh_metadata_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			    int32_t op_ret, int32_t op_errno,
-			    inode_t *inode, struct stat *buf, dict_t *xattr)
+afr_sh_metadata_sync (call_frame_t *frame, xlator_t *this, dict_t *xattr)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"path %s on subvolume %s is of mode 0%o",
-				local->loc.path,
-				priv->children[child_index]->name,
-				buf->st_mode);
-
-			sh->buf[child_index] = *buf;
-			if (xattr)
-				sh->xattr[child_index] = dict_ref (xattr);
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"path %s on subvolume %s => -1 (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-
-			sh->child_errno[child_index] = op_errno;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		afr_sh_metadata_fix (frame, this);
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              source = 0;
+        int              active_sinks = 0;
+        int              call_count = 0;
+        int              i = 0;
+
+        struct iatt      stbuf = {0,};
+        int32_t          valid = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        source = sh->source;
+        active_sinks = sh->active_sinks;
+
+        /*
+         * 2 calls per sink - setattr, setxattr
+         */
+        if (xattr) {
+                call_count = active_sinks * 2;
+                local->xattr_req = dict_ref (xattr);
+        } else
+                call_count = active_sinks;
+
+        local->call_count = call_count;
+
+        stbuf.ia_atime = sh->buf[source].ia_atime;
+        stbuf.ia_atime_nsec = sh->buf[source].ia_atime_nsec;
+        stbuf.ia_mtime = sh->buf[source].ia_mtime;
+        stbuf.ia_mtime_nsec = sh->buf[source].ia_mtime_nsec;
+
+        stbuf.ia_uid = sh->buf[source].ia_uid;
+        stbuf.ia_gid = sh->buf[source].ia_gid;
+
+        stbuf.ia_type = sh->buf[source].ia_type;
+        stbuf.ia_prot = sh->buf[source].ia_prot;
+
+        valid = GF_SET_ATTR_MODE  |
+                GF_SET_ATTR_UID   | GF_SET_ATTR_GID |
+                GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (call_count == 0) {
+                        break;
+                }
+                if (sh->sources[i] || !local->child_up[i])
+                        continue;
+
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "self-healing metadata of %s from %s to %s",
+                        local->loc.path, priv->children[source]->name,
+                        priv->children[i]->name);
+
+                STACK_WIND_COOKIE (frame, afr_sh_metadata_setattr_cbk,
+                                   (void *) (long) i,
+                                   priv->children[i],
+                                   priv->children[i]->fops->setattr,
+                                   &local->loc, &stbuf, valid, NULL);
+
+                call_count--;
+
+                if (!xattr)
+                        continue;
+
+                STACK_WIND_COOKIE (frame, afr_sh_getxattr_cbk,
+                                   (void *) (long) i,
+                                   priv->children[i],
+                                   priv->children[i]->fops->getxattr,
+                                   &local->loc, NULL, NULL);
+                call_count--;
+        }
+
+        return 0;
 }
 
 
 int
-afr_sh_metadata_lookup (call_frame_t *frame, xlator_t *this)
+afr_sh_metadata_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                              dict_t *xdata)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-	int              call_count = 0;
-	dict_t          *xattr_req = NULL;
-	int              ret = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	call_count = local->child_count;
-	local->call_count = call_count;
-	
-	xattr_req = dict_new();
-	
-	if (xattr_req)
-		ret = dict_set_uint64 (xattr_req, AFR_METADATA_PENDING,
-				       priv->child_count * sizeof(int32_t));
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"looking up %s on %s",
-				local->loc.path, priv->children[i]->name);
-
-			STACK_WIND_COOKIE (frame, afr_sh_metadata_lookup_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->lookup,
-					   &local->loc, xattr_req);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	if (xattr_req)
-		dict_unref (xattr_req);
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              source = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        source = sh->source;
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "getxattr of %s failed on subvolume %s (%s). proceeding without xattr",
+                        local->loc.path, priv->children[source]->name,
+                        strerror (op_errno));
+
+                afr_sh_metadata_sync (frame, this, NULL);
+        } else {
+                afr_prune_pending_keys (xattr, priv);
+                afr_sh_metadata_sync (frame, this, xattr);
+        }
+
+        return 0;
 }
 
+static void
+afr_set_metadata_sh_info_str (afr_local_t *local, afr_self_heal_t *sh,
+                              xlator_t *this)
+{
+        afr_private_t    *priv = NULL;
+        int              i = 0;
+        char             num[1024] = {0};
+        size_t           len = 0;
+        char             *string = NULL;
+        size_t           off = 0;
+        char             *source_child =  " from source %s to";
+        char             *format = " %s, ";
+        char             *string_msg = " metadata self heal";
+        char             *pending_matrix_str = NULL;
+        int              down_child_present = 0;
+        int              unknown_child_present = 0;
+        char             *down_subvol_1 = " down subvolume is ";
+        char             *unknown_subvol_1 = " unknown subvolume is";
+        char             *down_subvol_2 = " down subvolumes are ";
+        char             *unknown_subvol_2 = " unknown subvolumes are ";
+        int              down_count = 0;
+        int              unknown_count = 0;
+
+        priv = this->private;
+
+        pending_matrix_str = afr_get_pending_matrix_str (sh->pending_matrix,
+                                                         this);
+
+        if (!pending_matrix_str)
+                pending_matrix_str = "";
+
+        len += snprintf (num, sizeof (num), "%s", string_msg);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if ((sh->source == i) && (local->child_up[i] == 1)) {
+                        len += snprintf (num, sizeof (num), source_child,
+                                         priv->children[i]->name);
+                } else if ((local->child_up[i] == 1) && (sh->sources[i] == 0)) {
+                        len += snprintf (num, sizeof (num), format,
+                                         priv->children[i]->name);
+                } else if (local->child_up[i] == 0) {
+                        len += snprintf (num, sizeof (num), format,
+                                         priv->children[i]->name);
+                        if (!down_child_present)
+                                down_child_present = 1;
+                        down_count++;
+                } else if (local->child_up[i] == -1) {
+                        len += snprintf (num, sizeof (num), format,
+                                         priv->children[i]->name);
+                        if (!unknown_child_present)
+                                unknown_child_present = 1;
+                        unknown_count++;
+                }
+        }
+
+        if (down_child_present) {
+                if (down_count > 1) {
+                        len += snprintf (num, sizeof (num), "%s",
+                                         down_subvol_2);
+                } else {
+                        len += snprintf (num, sizeof (num), "%s",
+                                         down_subvol_1);
+                }
+        }
+        if (unknown_child_present) {
+                if (unknown_count > 1) {
+                        len += snprintf (num, sizeof (num), "%s",
+                                         unknown_subvol_2);
+                } else {
+                        len += snprintf (num, sizeof (num), "%s",
+                                         unknown_subvol_1);
+                }
+        }
+
+        len ++;
+
+        string = GF_CALLOC (len, sizeof (char), gf_common_mt_char);
+        if (!string)
+                return;
+
+        off += snprintf (string + off, len - off, "%s", string_msg);
+        for (i=0; i < priv->child_count; i++) {
+                if ((sh->source == i) && (local->child_up[i] == 1))
+                        off += snprintf (string + off, len - off, source_child,
+                                         priv->children[i]->name);
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if ((local->child_up[i] == 1)&& (sh->sources[i] == 0))
+                        off += snprintf (string + off, len - off, format,
+                                         priv->children[i]->name);
+        }
+
+        if (down_child_present) {
+                if (down_count > 1) {
+                        off += snprintf (string + off, len - off, "%s",
+                                         down_subvol_2);
+                } else {
+                        off += snprintf (string + off, len - off, "%s",
+                                         down_subvol_1);
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == 0)
+                        off += snprintf (string + off, len - off, format,
+                                         priv->children[i]->name);
+        }
+
+        if (unknown_child_present) {
+                if (unknown_count > 1) {
+                        off += snprintf (string + off, len - off, "%s",
+                                 unknown_subvol_2);
+                } else {
+                        off += snprintf (string + off, len - off, "%s",
+                                         unknown_subvol_1);
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_up[i] == -1)
+                        off += snprintf (string + off, len - off, format,
+                                         priv->children[i]->name);
+        }
+
+        gf_asprintf (&sh->metadata_sh_info, "%s metadata %s,", string,
+                     pending_matrix_str);
+
+        if (pending_matrix_str && strcmp (pending_matrix_str, ""))
+                GF_FREE (pending_matrix_str);
+
+        if (string && strcmp (string, ""))
+                GF_FREE (string);
+}
 
 int
-afr_sh_metadata_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno)
+afr_sh_metadata_sync_prepare (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              child_index = (long) cookie;
-
-	/* TODO: what if lock fails? */
-	
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			sh->op_failed = 1;
-
-			gf_log (this->name,
-				(op_errno == EAGAIN ? GF_LOG_DEBUG : GF_LOG_ERROR),
-				"locking of %s on child %d failed: %s",
-				local->loc.path, child_index,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"inode of %s on child %d locked",
-				local->loc.path, child_index);
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		if (sh->op_failed) {
-			afr_sh_metadata_finish (frame, this);
-			return 0;
-		}
-
-		afr_sh_metadata_lookup (frame, this);
-	}
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              source = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        source = sh->source;
+
+        afr_sh_mark_source_sinks (frame, this);
+        if (sh->active_sinks == 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no active sinks for performing self-heal on file %s",
+                        local->loc.path);
+                afr_sh_metadata_finish (frame, this);
+                return 0;
+        }
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "syncing metadata of %s from subvolume %s to %d active sinks",
+                local->loc.path, priv->children[source]->name,
+                sh->active_sinks);
+
+        sh->actual_sh_started = _gf_true;
+        afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN);
+        afr_set_metadata_sh_info_str (local, sh, this);
+        STACK_WIND (frame, afr_sh_metadata_getxattr_cbk,
+                    priv->children[source],
+                    priv->children[source]->fops->getxattr,
+                    &local->loc, NULL, NULL);
+
+        return 0;
 }
 
 
-int
-afr_sh_metadata_lock (call_frame_t *frame, xlator_t *this)
+void
+afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-	int              call_count = 0;
-	struct flock     flock = {0, };
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	call_count = local->child_count;
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {
-		flock.l_start   = 0;
-		flock.l_len     = 0;
-		flock.l_type    = F_WRLCK;
-
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"locking %s on subvolume %s",
-				local->loc.path, priv->children[i]->name);
-
-			STACK_WIND_COOKIE (frame, afr_sh_metadata_lk_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->inodelk,
-                                           this->name,
-					   &local->loc, F_SETLK, &flock);
-
-			if (!--call_count)
-				break;
-		}
-	}
-
-	return 0;
+        afr_local_t     *local = NULL;
+        afr_self_heal_t *sh = NULL;
+        afr_private_t   *priv = NULL;
+        int              nsources = 0;
+        int              source = 0;
+        int              i = 0;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        priv = this->private;
+
+        if (op_ret < 0) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED);
+                afr_sh_set_error (sh, op_errno);
+                afr_sh_metadata_finish (frame, this);
+                goto out;
+        }
+        nsources = afr_build_sources (this, sh->xattr, sh->buf,
+                                      sh->pending_matrix, sh->sources,
+                                      sh->success_children,
+                                      AFR_METADATA_TRANSACTION, NULL, _gf_false);
+        if ((nsources == -1)
+            && (priv->favorite_child != -1)
+            && (sh->child_errno[priv->favorite_child] == 0)) {
+
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Picking favorite child %s as authentic source to resolve conflicting metadata of %s",
+                        priv->children[priv->favorite_child]->name,
+                        local->loc.path);
+
+                sh->sources[priv->favorite_child] = 1;
+
+                nsources = afr_sh_source_count (sh->sources,
+                                                priv->child_count);
+        }
+
+        if (nsources == -1) {
+                afr_sh_print_split_brain_log (sh->pending_matrix, this,
+                                              local->loc.path);
+                afr_set_split_brain (this, sh->inode, SPB, DONT_KNOW);
+                afr_sh_metadata_fail (frame, this);
+                goto out;
+        }
+
+        afr_set_split_brain (this, sh->inode, NO_SPB, DONT_KNOW);
+        if (nsources == 0) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "No self-heal needed for %s",
+                        local->loc.path);
+
+                afr_sh_metadata_finish (frame, this);
+                goto out;
+        }
+
+        source = afr_sh_select_source (sh->sources, priv->child_count);
+
+        if (source == -1) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "No active sources found.");
+
+                afr_sh_metadata_finish (frame, this);
+                goto out;
+        }
+
+        sh->source = source;
+
+        /* detect changes not visible through pending flags -- JIC */
+        for (i = 0; i < priv->child_count; i++) {
+                if (i == source || sh->child_errno[i])
+                        continue;
+
+                if (PERMISSION_DIFFERS (&sh->buf[i], &sh->buf[source]))
+                        sh->sources[i] = 0;
+
+                if (OWNERSHIP_DIFFERS (&sh->buf[i], &sh->buf[source]))
+                        sh->sources[i] = 0;
+        }
+
+        if ((!IA_ISREG (sh->buf[source].ia_type)) &&
+            (!IA_ISDIR (sh->buf[source].ia_type))) {
+                afr_reset_children (sh->fresh_children, priv->child_count);
+                afr_get_fresh_children (sh->success_children, sh->sources,
+                                        sh->fresh_children, priv->child_count);
+                afr_inode_set_read_ctx (this, sh->inode, sh->source,
+                                        sh->fresh_children);
+        }
+
+        if (sh->do_metadata_self_heal && priv->metadata_self_heal)
+                afr_sh_metadata_sync_prepare (frame, this);
+        else
+                afr_sh_metadata_finish (frame, this);
+out:
+        return;
 }
 
+int
+afr_sh_metadata_post_nonblocking_inodelk_cbk (call_frame_t *frame,
+                                              xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        if (int_lock->lock_op_ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking metadata "
+                        "inodelks failed for %s.", local->loc.path);
+                gf_log (this->name, GF_LOG_DEBUG, "Metadata self-heal "
+                        "failed for %s.", local->loc.path);
+                afr_sh_metadata_done (frame, this);
+        } else {
+
+                gf_log (this->name, GF_LOG_DEBUG, "Non Blocking metadata "
+                        "inodelks done for %s. Proceeding to FOP",
+                        local->loc.path);
+                afr_sh_common_lookup (frame, this, &local->loc,
+                                      afr_sh_metadata_fix, NULL,
+                                      AFR_LOOKUP_FAIL_CONFLICTS |
+                                      AFR_LOOKUP_FAIL_MISSING_GFIDS,
+                                      NULL);
+        }
+
+        return 0;
+}
 
 int
-afr_self_heal_metadata (call_frame_t *frame, xlator_t *this)
+afr_sh_metadata_lock (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t   *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t *priv = this->private;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
+        afr_local_t         *local    = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        int_lock->domain           = this->name;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
+        int_lock->selfheal_lk_type    = AFR_METADATA_SELF_HEAL_LK;
 
+        afr_set_lock_number (frame, this);
 
-	local = frame->local;
-	sh = &local->self_heal;
+        inodelk->flock.l_start = LLONG_MAX - 1;
+        inodelk->flock.l_len   = 0;
+        inodelk->flock.l_type  = F_WRLCK;
+        int_lock->lock_cbk         = afr_sh_metadata_post_nonblocking_inodelk_cbk;
 
-	if (local->need_metadata_self_heal && priv->metadata_self_heal) {
-		afr_sh_metadata_lock (frame, this);
-	} else {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"proceeding to data check on %s",
-			local->loc.path);
-		afr_sh_metadata_done (frame, this);
-	}
+        afr_nonblocking_inodelk (frame, this);
+
+        return 0;
+}
 
-	return 0;
+gf_boolean_t
+afr_can_start_metadata_self_heal (afr_self_heal_t *sh, afr_private_t *priv)
+{
+        if (sh->force_confirm_spb)
+                return _gf_true;
+        if (sh->do_metadata_self_heal && priv->metadata_self_heal)
+                return _gf_true;
+        return _gf_false;
 }
 
+int
+afr_self_heal_metadata (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local = NULL;
+        afr_private_t *priv = this->private;
+        afr_self_heal_t *sh = &local->self_heal;
+
+        local = frame->local;
+        sh = &local->self_heal;
+        sh->sh_type_in_action = AFR_SELF_HEAL_METADATA;
+
+        if (afr_can_start_metadata_self_heal (sh, priv)) {
+                afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED);
+                afr_sh_metadata_lock (frame, this);
+        } else {
+                afr_sh_metadata_done (frame, this);
+        }
+
+        return 0;
+}
diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h
index c98831b8b..7c9bc8111 100644
--- a/xlators/cluster/afr/src/afr-self-heal.h
+++ b/xlators/cluster/afr/src/afr-self-heal.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __AFR_SELF_HEAL_H__
@@ -22,19 +13,12 @@
 
 #include <sys/stat.h>
 
-#define FILETYPE_DIFFERS(buf1,buf2) ((S_IFMT & ((struct stat *)buf1)->st_mode) != (S_IFMT & ((struct stat *)buf2)->st_mode))
-#define PERMISSION_DIFFERS(buf1,buf2) ((((struct stat *)buf1)->st_mode) != (((struct stat *)buf2)->st_mode))
-#define OWNERSHIP_DIFFERS(buf1,buf2) ((((struct stat *)buf1)->st_uid) != (((struct stat *)buf2)->st_uid) || (((struct stat *)buf1)->st_gid != (((struct stat *)buf2)->st_gid)))
-#define SIZE_DIFFERS(buf1,buf2) ((((struct stat *)buf1)->st_size) != (((struct stat *)buf2)->st_size))
-
+#define FILETYPE_DIFFERS(buf1,buf2) ((buf1)->ia_type != (buf2)->ia_type)
+#define PERMISSION_DIFFERS(buf1,buf2) (st_mode_from_ia ((buf1)->ia_prot, (buf1)->ia_type) != st_mode_from_ia ((buf2)->ia_prot, (buf2)->ia_type))
+#define OWNERSHIP_DIFFERS(buf1,buf2) (((buf1)->ia_uid != (buf2)->ia_uid) || ((buf1)->ia_gid != (buf2)->ia_gid))
+#define SIZE_DIFFERS(buf1,buf2) ((buf1)->ia_size != (buf2)->ia_size)
 
-
-int
-afr_sh_has_metadata_pending (dict_t *xattr, int child_count, xlator_t *this);
-int
-afr_sh_has_entry_pending (dict_t *xattr, int child_count, xlator_t *this);
-int
-afr_sh_has_data_pending (dict_t *xattr, int child_count, xlator_t *this);
+#define SIZE_GREATER(buf1,buf2) ((buf1)->ia_size > (buf2)->ia_size)
 
 int
 afr_self_heal_entry (call_frame_t *frame, xlator_t *this);
@@ -46,7 +30,14 @@ int
 afr_self_heal_metadata (call_frame_t *frame, xlator_t *this);
 
 int
-afr_self_heal (call_frame_t *frame, xlator_t *this,
-	       int (*completion_cbk) (call_frame_t *, xlator_t *));
+afr_self_heal_get_source (xlator_t *this, afr_local_t *local, dict_t **xattr);
+
+int
+afr_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode);
 
+int
+afr_lookup_select_read_child_by_txn_type (xlator_t *this, afr_local_t *local,
+                                          dict_t **xattr,
+                                          afr_transaction_type txn_type,
+                                          uuid_t gfid);
 #endif /* __AFR_SELF_HEAL_H__ */
diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c
new file mode 100644
index 000000000..1b48a1bca
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heald.c
@@ -0,0 +1,1787 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+#include "afr.h"
+#include "syncop.h"
+#include "afr-self-heald.h"
+#include "afr-self-heal-common.h"
+#include "protocol-common.h"
+#include "event-history.h"
+
+typedef enum {
+        STOP_CRAWL_ON_SINGLE_SUBVOL = 1
+} afr_crawl_flags_t;
+
+typedef enum {
+        HEAL = 1,
+        INFO,
+        STATISTICS_TO_BE_HEALED,
+} shd_crawl_op;
+
+typedef struct shd_dump {
+        dict_t   *dict;
+        xlator_t *this;
+        int      child;
+} shd_dump_t;
+
+typedef struct shd_event_ {
+        int     child;
+        char    *path;
+} shd_event_t;
+
+typedef struct shd_pos_ {
+        int     child;
+        xlator_t *this;
+        afr_child_pos_t pos;
+} shd_pos_t;
+
+typedef int
+(*afr_crawl_done_cbk_t)  (int ret, call_frame_t *sync_frame, void *crawl_data);
+
+void
+afr_start_crawl (xlator_t *this, int idx, afr_crawl_type_t crawl,
+                 process_entry_cbk_t process_entry, void *op_data,
+                 gf_boolean_t exclusive, int crawl_flags,
+                 afr_crawl_done_cbk_t crawl_done);
+
+static int
+_crawl_directory (fd_t *fd, loc_t *loc, afr_crawl_data_t *crawl_data);
+
+/* For calling straight through (e.g. already in a synctask). */
+int
+afr_find_child_position (xlator_t *this, int child, afr_child_pos_t *pos);
+
+/* For deferring through a new synctask. */
+int
+afr_syncop_find_child_position (void *data);
+
+static int
+_loc_assign_gfid_path (loc_t *loc)
+{
+        int  ret = -1;
+        char gfid_path[64] = {0};
+
+        if (loc->inode && !uuid_is_null (loc->inode->gfid)) {
+                ret = inode_path (loc->inode, NULL, (char**)&loc->path);
+        } else if (!uuid_is_null (loc->gfid)) {
+                snprintf (gfid_path, sizeof (gfid_path), "<gfid:%s>",
+                          uuid_utoa (loc->gfid));
+                loc->path = gf_strdup (gfid_path);
+                if (loc->path)
+                        ret = 0;
+        }
+        return ret;
+}
+
+void
+_destroy_crawl_event_data (void *data)
+{
+        shd_crawl_event_t        *crawl_event = NULL;
+
+        if (!data)
+                goto out;
+
+        crawl_event = (shd_crawl_event_t *)data;
+        GF_FREE (crawl_event->start_time_str);
+        GF_FREE (crawl_event->end_time_str);
+
+out:
+        return;
+}
+
+void
+_destroy_shd_event_data (void *data)
+{
+        shd_event_t             *event = NULL;
+        if (!data)
+                goto out;
+        event = (shd_event_t*)data;
+        GF_FREE (event->path);
+out:
+        return;
+}
+void
+shd_cleanup_event (void *event)
+{
+        shd_event_t *shd_event = event;
+
+        if (!shd_event)
+                goto out;
+        GF_FREE (shd_event->path);
+        GF_FREE (shd_event);
+out:
+        return;
+}
+
+int
+afr_get_local_child (afr_self_heald_t *shd, unsigned int child_count)
+{
+        int i = 0;
+        int ret = -1;
+        for (i = 0; i < child_count; i++) {
+                if (shd->pos[i] == AFR_POS_LOCAL) {
+                        ret = i;
+                        break;
+                }
+        }
+        return ret;
+}
+
+static int
+_build_index_loc (xlator_t *this, loc_t *loc, char *name, loc_t *parent)
+{
+        int             ret = 0;
+
+        uuid_copy (loc->pargfid, parent->inode->gfid);
+        loc->path = "";
+        loc->name = name;
+        loc->parent = inode_ref (parent->inode);
+        if (!loc->parent) {
+                loc->path = NULL;
+                loc_wipe (loc);
+                ret = -1;
+        }
+        return ret;
+}
+
+int
+_add_crawl_stats_to_dict (xlator_t *this, dict_t *output, int child,
+                          shd_crawl_event_t *shd_event, struct timeval *tv)
+{
+        int             ret = 0;
+        uint64_t        count = 0;
+        char            key[256] = {0};
+        int             xl_id = 0;
+        uint64_t        healed_count = 0;
+        uint64_t        split_brain_count = 0;
+        uint64_t        heal_failed_count = 0;
+        char            *start_time_str = NULL;
+        char            *end_time_str = NULL;
+        char            *crawl_type = NULL;
+        int             progress = -1;
+
+        healed_count = shd_event->healed_count;
+        split_brain_count = shd_event->split_brain_count;
+        heal_failed_count = shd_event->heal_failed_count;
+        start_time_str = shd_event->start_time_str;
+        end_time_str = shd_event->end_time_str;
+        crawl_type = shd_event->crawl_type;
+
+        if (!start_time_str) {
+                ret = -1;
+                goto out;
+        }
+
+
+        ret = dict_get_int32 (output, this->name, &xl_id);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "xl does not have id");
+                goto out;
+        }
+
+        snprintf (key, sizeof (key), "statistics-%d-%d-count", xl_id, child);
+        ret = dict_get_uint64 (output, key, &count);
+
+        snprintf (key, sizeof (key), "statistics_healed_cnt-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        ret = dict_set_uint64(output, key, healed_count);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_"
+                        "healed_count to outout");
+                goto out;
+         }
+        snprintf (key, sizeof (key), "statistics_sb_cnt-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        ret = dict_set_uint64 (output, key, split_brain_count);
+         if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_"
+                        "split_brain_count to outout");
+                goto out;
+        }
+        snprintf (key, sizeof (key), "statistics_crawl_type-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        ret = dict_set_dynstr (output, key, gf_strdup (crawl_type));
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_"
+                        "crawl_type to output");
+                goto out;
+        }
+        snprintf (key, sizeof (key), "statistics_heal_failed_cnt-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        ret = dict_set_uint64 (output, key, heal_failed_count);
+         if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_"
+                        "healed_failed_count to outout");
+                goto out;
+        }
+        snprintf (key, sizeof (key), "statistics_strt_time-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+        ret = dict_set_dynstr (output, key, gf_strdup(start_time_str));
+
+         if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_"
+                        "crawl_start_time to outout");
+                goto out;
+        }
+
+        snprintf (key, sizeof (key), "statistics_end_time-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+
+        if (!end_time_str)
+                end_time_str = "Could not determine the end time";
+        ret = dict_set_dynstr (output, key, gf_strdup(end_time_str));
+         if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_"
+                        "crawl_end_time to outout");
+                goto out;
+        }
+        snprintf (key, sizeof (key), "statistics_inprogress-%d-%d-%"PRIu64,
+                  xl_id, child, count);
+
+        if (shd_event->crawl_inprogress == _gf_true)
+                progress = 1;
+        else
+                progress = 0;
+
+        ret = dict_set_int32 (output, key, progress);
+         if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_"
+                        "inprogress to outout");
+                goto out;
+        }
+
+         snprintf (key, sizeof (key), "statistics-%d-%d-count",xl_id, child);
+         ret = dict_set_uint64 (output, key, count + 1);
+         if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not increment the "
+                        "counter.");
+                goto out;
+         }
+out:
+        return ret;
+}
+
+int
+_add_path_to_dict (xlator_t *this, dict_t *output, int child, char *path,
+                   struct timeval *tv, gf_boolean_t dyn)
+{
+        //subkey not used for now
+        int             ret = -1;
+        uint64_t        count = 0;
+        char            key[256] = {0};
+        int             xl_id = 0;
+
+        ret = dict_get_int32 (output, this->name, &xl_id);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "xl does not have id");
+                goto out;
+        }
+
+        snprintf (key, sizeof (key), "%d-%d-count", xl_id, child);
+        ret = dict_get_uint64 (output, key, &count);
+
+        snprintf (key, sizeof (key), "%d-%d-%"PRIu64, xl_id, child, count);
+        if (dyn)
+                ret = dict_set_dynstr (output, key, path);
+        else
+                ret = dict_set_str (output, key, path);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "%s: Could not add to output",
+                        path);
+                goto out;
+        }
+
+        if (!tv)
+                goto inc_count;
+        snprintf (key, sizeof (key), "%d-%d-%"PRIu64"-time", xl_id,
+                  child, count);
+        ret = dict_set_uint32 (output, key, tv->tv_sec);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "%s: Could not set time",
+                        path);
+                goto out;
+        }
+
+inc_count:
+        snprintf (key, sizeof (key), "%d-%d-count", xl_id, child);
+        ret = dict_set_uint64 (output, key, count + 1);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Could not increment count");
+                goto out;
+        }
+        ret = 0;
+out:
+        return ret;
+}
+
+int
+_get_path_from_gfid_loc (xlator_t *this, xlator_t *readdir_xl, loc_t *child,
+                         char **fpath, gf_boolean_t *missing)
+{
+        dict_t          *xattr = NULL;
+        char            *path = NULL;
+        int             ret = -1;
+
+        ret = syncop_getxattr (readdir_xl, child, &xattr, GFID_TO_PATH_KEY);
+        if (ret < 0) {
+                if ((errno == ENOENT) && missing)
+                        *missing = _gf_true;
+                goto out;
+        }
+        ret = dict_get_str (xattr, GFID_TO_PATH_KEY, &path);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Failed to get path for "
+                        "gfid %s", uuid_utoa (child->gfid));
+                goto out;
+        }
+        path = gf_strdup (path);
+        if (!path) {
+                ret = -1;
+                goto out;
+        }
+        ret = 0;
+out:
+        if (!ret)
+                *fpath = path;
+        if (xattr)
+                dict_unref (xattr);
+        return ret;
+}
+
+int
+_add_event_to_dict (circular_buffer_t *cb, void *data)
+{
+        int               ret = 0;
+        shd_dump_t        *dump_data = NULL;
+        shd_event_t       *shd_event = NULL;
+
+        dump_data = data;
+        shd_event = cb->data;
+        if (shd_event->child != dump_data->child)
+                goto out;
+        ret = _add_path_to_dict (dump_data->this, dump_data->dict,
+                                 dump_data->child, shd_event->path, &cb->tv,
+                                 _gf_false);
+out:
+        return ret;
+}
+
+int
+_add_crawl_event_statistics_to_dict (circular_buffer_t *cb, void *data)
+{
+        int               ret = 0;
+        shd_dump_t        *dump_data = NULL;
+        shd_crawl_event_t *shd_event = NULL;
+
+        dump_data = data;
+        shd_event = cb->data;
+        ret = _add_crawl_stats_to_dict (dump_data->this, dump_data->dict,
+                                        dump_data->child, shd_event, &cb->tv);
+        return ret;
+}
+
+int
+_add_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict, int child)
+{
+        shd_dump_t dump_data = {0};
+
+        dump_data.this = this;
+        dump_data.dict = dict;
+        dump_data.child = child;
+        eh_dump (eh, &dump_data, _add_event_to_dict);
+        return 0;
+}
+
+
+int
+_add_statistics_to_dict (xlator_t *this, dict_t *dict, int child)
+{
+        shd_dump_t              dump_data = {0};
+        afr_private_t           *priv  = NULL;
+        afr_self_heald_t        *shd = NULL;
+
+        priv = this->private;
+        shd = &priv->shd;
+
+        dump_data.this = this;
+        dump_data.dict = dict;
+        dump_data.child = child;
+        eh_dump (shd->statistics[child], &dump_data,
+                 _add_crawl_event_statistics_to_dict);
+        return 0;
+
+}
+
+void
+_remove_stale_index (xlator_t *this, xlator_t *readdir_xl,
+                     loc_t *parent, char *fname)
+{
+        int              ret = 0;
+        loc_t            index_loc = {0};
+
+        ret = _build_index_loc (this, &index_loc, fname, parent);
+        if (ret)
+                goto out;
+        gf_log (this->name, GF_LOG_DEBUG, "Removing stale index "
+                "for %s on %s", index_loc.name, readdir_xl->name);
+        ret = syncop_unlink (readdir_xl, &index_loc);
+        if(ret && (errno != ENOENT)) {
+                gf_log(this->name, GF_LOG_ERROR, "%s: Failed to remove index "
+                       "on %s - %s",index_loc.name, readdir_xl->name,
+                       strerror (errno));
+        }
+        index_loc.path = NULL;
+        loc_wipe (&index_loc);
+out:
+        return;
+}
+
+int
+_count_hard_links_under_base_indices_dir (xlator_t *this,
+                                           afr_crawl_data_t *crawl_data,
+                                           gf_dirent_t *entry, loc_t *childloc,
+                                           loc_t *parentloc, struct iatt *iattr)
+{
+        xlator_t                *readdir_xl = crawl_data->readdir_xl;
+        struct iatt             parent = {0};
+        int                     ret = 0;
+        dict_t                  *output = NULL;
+        int                     xl_id =  0;
+        char                    key[256] = {0};
+        int                     child  = -1;
+        uint64_t                     hardlinks = 0;
+
+        output = crawl_data->op_data;
+        child = crawl_data->child;
+
+        ret = syncop_lookup (readdir_xl, childloc, NULL, iattr, NULL, &parent);
+        if (ret)
+                goto out;
+
+        ret = dict_get_int32 (output, this->name, &xl_id);
+        if (ret)
+                goto out;
+
+        snprintf (key, sizeof (key), "%d-%d-hardlinks", xl_id, child);
+        ret =  dict_get_uint64 (output, key, &hardlinks);
+
+        /*Removing the count of base_entry under indices/base_indicies and
+         * entry under indices/xattrop */
+        hardlinks = hardlinks + iattr->ia_nlink - 2;
+        ret = dict_set_uint64 (output, key, hardlinks);
+        if (ret)
+                goto out;
+
+out:
+        return ret;
+}
+
+int
+_add_summary_to_dict (xlator_t *this, afr_crawl_data_t *crawl_data,
+                      gf_dirent_t *entry,
+                      loc_t *childloc, loc_t *parentloc, struct iatt *iattr)
+{
+        dict_t          *output = NULL;
+        xlator_t        *readdir_xl = NULL;
+        int             ret = -1;
+        char            *path = NULL;
+        gf_boolean_t    missing = _gf_false;
+        char            gfid_str[64] = {0};
+
+        if (uuid_is_null (childloc->gfid))
+                goto out;
+
+        output = crawl_data->op_data;
+        readdir_xl = crawl_data->readdir_xl;
+
+        ret = _get_path_from_gfid_loc (this, readdir_xl, childloc, &path,
+                                       &missing);
+        if (ret == 0) {
+                ret = _add_path_to_dict (this, output, crawl_data->child, path,
+                                         NULL, _gf_true);
+        } else if (missing) {
+                _remove_stale_index (this, readdir_xl, parentloc,
+                                     uuid_utoa_r (childloc->gfid, gfid_str));
+        }
+
+out:
+        if (ret && path)
+                GF_FREE (path);
+        return ret;
+}
+
+void
+_crawl_post_sh_action (xlator_t *this, loc_t *parent, loc_t *child,
+                       int32_t op_ret, int32_t op_errno, dict_t *xattr_rsp,
+                       afr_crawl_data_t *crawl_data)
+{
+        int                ret = 0;
+        afr_private_t      *priv = NULL;
+        afr_self_heald_t   *shd = NULL;
+        eh_t               *eh = NULL;
+        char               *path = NULL;
+        char               gfid_str[64] = {0};
+        shd_event_t        *event = NULL;
+        int32_t            sh_failed = 0;
+        gf_boolean_t       split_brain = 0;
+        int32_t            actual_sh_done = 0;
+        shd_crawl_event_t  **shd_crawl_event = NULL;
+
+        priv = this->private;
+        shd  = &priv->shd;
+        if (crawl_data->crawl == INDEX) {
+                if ((op_ret < 0) && (op_errno == ENOENT)) {
+                        _remove_stale_index (this, crawl_data->readdir_xl,
+                                             parent, uuid_utoa_r (child->gfid,
+                                                                  gfid_str));
+                        goto out;
+                }
+                ret = _get_path_from_gfid_loc (this, crawl_data->readdir_xl,
+                                               child, &path, NULL);
+                if (ret)
+                        goto out;
+        } else {
+                path = gf_strdup (child->path);
+                if (!path) {
+                        ret = -1;
+                        goto out;
+                }
+        }
+
+        if (xattr_rsp) {
+                ret = dict_get_int32 (xattr_rsp, "sh-failed", &sh_failed);
+                ret = dict_get_int32 (xattr_rsp, "actual-sh-done", &actual_sh_done);
+        }
+
+        shd_crawl_event = (shd_crawl_event_t**)(shd->crawl_events);
+
+        split_brain = afr_is_split_brain (this, child->inode);
+        if ((op_ret < 0 && op_errno == EIO) || split_brain) {
+                eh = shd->split_brain;
+                shd_crawl_event[crawl_data->child]->split_brain_count += 1;
+        } else if ((op_ret < 0) || sh_failed) {
+                eh = shd->heal_failed;
+                shd_crawl_event[crawl_data->child]->heal_failed_count += 1;
+        } else if (actual_sh_done == 1) {
+                eh = shd->healed;
+                shd_crawl_event[crawl_data->child]->healed_count += 1;
+        }
+        ret = -1;
+
+        if (eh != NULL) {
+                event = GF_CALLOC (1, sizeof (*event), gf_afr_mt_shd_event_t);
+                if (!event)
+                        goto out;
+                event->child = crawl_data->child;
+                event->path = path;
+
+                ret = eh_save_history (eh, event);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR, "%s:Failed to save "
+                                "to event history, (%d, %s)", path, op_ret,
+                                strerror (op_errno));
+
+                        goto out;
+                }
+        } else {
+                gf_log (this->name, GF_LOG_DEBUG, "%s:Self heal already done ",
+                        path);
+
+        }
+        ret = 0;
+out:
+        if (ret && path)
+                GF_FREE (path);
+        return;
+}
+
+int
+_link_inode_update_loc (xlator_t *this, loc_t *loc, struct iatt *iattr)
+{
+        inode_t       *link_inode = NULL;
+        int           ret = -1;
+
+        link_inode = inode_link (loc->inode, NULL, NULL, iattr);
+        if (link_inode == NULL) {
+                gf_log (this->name, GF_LOG_ERROR, "inode link failed "
+                        "on the inode (%s)", uuid_utoa (iattr->ia_gfid));
+                goto out;
+        }
+        inode_unref (loc->inode);
+        loc->inode = link_inode;
+        ret = 0;
+out:
+        return ret;
+}
+
+int
+_self_heal_entry (xlator_t *this, afr_crawl_data_t *crawl_data, gf_dirent_t *entry,
+                  loc_t *child, loc_t *parent, struct iatt *iattr)
+{
+        struct iatt      parentbuf = {0};
+        int              ret = 0;
+        dict_t           *xattr_rsp = NULL;
+        dict_t           *xattr_req = NULL;
+
+        xattr_req = dict_new ();
+        if (!xattr_req) {
+                errno = ENOMEM;
+                ret = -1;
+                goto out;
+        }
+
+        ret = dict_set_int32 (xattr_req, "allow-sh-for-running-transaction", 1);
+
+        gf_log (this->name, GF_LOG_DEBUG, "lookup %s", child->path);
+
+        ret = syncop_lookup (this, child, xattr_req,
+                             iattr, &xattr_rsp, &parentbuf);
+        _crawl_post_sh_action (this, parent, child, ret, errno, xattr_rsp,
+                               crawl_data);
+        if (xattr_rsp)
+                dict_unref (xattr_rsp);
+        if (ret == 0)
+                ret = _link_inode_update_loc (this, child, iattr);
+
+out:
+        if (xattr_req)
+                dict_unref(xattr_req);
+        return ret;
+}
+
+static int
+afr_crawl_done  (int ret, call_frame_t *sync_frame, void *data)
+{
+        GF_FREE (data);
+        STACK_DESTROY (sync_frame->root);
+        return 0;
+}
+
+void
+_do_self_heal_on_subvol (xlator_t *this, int child, afr_crawl_type_t crawl)
+{
+        afr_start_crawl (this, child, crawl, _self_heal_entry,
+                         NULL, _gf_true, STOP_CRAWL_ON_SINGLE_SUBVOL,
+                         afr_crawl_done);
+}
+
+gf_boolean_t
+_crawl_proceed (xlator_t *this, int child, int crawl_flags, char **reason)
+{
+        afr_private_t           *priv = NULL;
+        afr_self_heald_t        *shd = NULL;
+        gf_boolean_t            proceed = _gf_false;
+        char                    *msg = NULL;
+
+        priv = this->private;
+        shd  = &priv->shd;
+        if (!shd->enabled) {
+                msg = "Self-heal daemon is not enabled";
+                gf_log (this->name, GF_LOG_DEBUG, "%s", msg);
+                goto out;
+        }
+        if (!priv->child_up[child]) {
+                gf_log (this->name, GF_LOG_DEBUG, "Stopping crawl for %s , "
+                        "subvol went down", priv->children[child]->name);
+                msg = "Brick is Not connected";
+                goto out;
+        }
+
+        if (crawl_flags & STOP_CRAWL_ON_SINGLE_SUBVOL) {
+                if (afr_up_children_count (priv->child_up,
+                                           priv->child_count) < 2) {
+                        gf_log (this->name, GF_LOG_DEBUG, "Stopping crawl as "
+                                "< 2 children are up");
+                        msg = "< 2 bricks in replica are running";
+                        goto out;
+                }
+        }
+        proceed = _gf_true;
+out:
+        if (reason)
+                *reason = msg;
+        return proceed;
+}
+
+int
+_do_crawl_op_on_local_subvols (xlator_t *this, afr_crawl_type_t crawl,
+                               shd_crawl_op op, dict_t *output)
+{
+        afr_private_t       *priv = NULL;
+        char                *status = NULL;
+        char                *subkey = NULL;
+        char                key[256] = {0};
+        shd_pos_t           pos_data = {0};
+        int                 op_ret = -1;
+        int                 xl_id = -1;
+        int                 i = 0;
+        int                 ret = 0;
+        int                 crawl_flags = 0;
+
+        priv = this->private;
+        if (op == HEAL)
+                crawl_flags |= STOP_CRAWL_ON_SINGLE_SUBVOL;
+
+        if (output) {
+                ret = dict_get_int32 (output, this->name, &xl_id);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR, "Invalid input, "
+                                "translator-id is not available");
+                        goto out;
+                }
+        }
+        pos_data.this = this;
+        subkey = "status";
+        for (i = 0; i < priv->child_count; i++) {
+                if (_crawl_proceed (this, i, crawl_flags, &status)) {
+                        pos_data.child = i;
+                        /*
+                         * We're already in a synctask in this case, so we
+                         * don't need to defer through a second (and in fact
+                         * that can cause deadlock).  Just call straight
+                         * through instead.
+                         */
+                        ret = afr_find_child_position(pos_data.this,
+                                                      pos_data.child,
+                                                      &pos_data.pos);
+                        if (ret) {
+                                status = "Not able to find brick location";
+                        } else if (pos_data.pos == AFR_POS_REMOTE) {
+                                status = "brick is remote";
+                        } else {
+                                op_ret = 0;
+                                if (op == HEAL) {
+                                        status = "Started self-heal";
+                                        _do_self_heal_on_subvol (this, i,
+                                                                 crawl);
+                                } else if (output && (op == INFO)) {
+                                        status = "";
+                                        afr_start_crawl (this, i, INDEX,
+                                                         _add_summary_to_dict,
+                                                         output, _gf_false, 0,
+                                                         NULL);
+                                } else if (output &&
+                                           (op == STATISTICS_TO_BE_HEALED)) {
+                                            status = "";
+                                            afr_start_crawl (this, i,
+                                                             INDEX_TO_BE_HEALED,
+                                       _count_hard_links_under_base_indices_dir,
+                                                             output, _gf_false,
+                                                             0, NULL);
+                                }
+                        }
+                        if (output) {
+                                snprintf (key, sizeof (key), "%d-%d-%s", xl_id,
+                                          i, subkey);
+                                ret = dict_set_str (output, key, status);
+                        }
+                        if (!op_ret && (crawl == FULL))
+                                break;
+                }
+                if (output) {
+                        snprintf (key, sizeof (key), "%d-%d-%s", xl_id, i,
+                                  subkey);
+                        ret = dict_set_str (output, key, status);
+                }
+        }
+out:
+        return op_ret;
+}
+
+int
+_do_self_heal_on_local_subvols (xlator_t *this, afr_crawl_type_t crawl,
+                                dict_t *output)
+{
+        return _do_crawl_op_on_local_subvols (this, crawl, HEAL, output);
+}
+
+int
+_get_index_summary_on_local_subvols (xlator_t *this, dict_t *output)
+{
+        return _do_crawl_op_on_local_subvols (this, INDEX, INFO, output);
+}
+
+void
+afr_fill_completed_crawl_statistics_to_dict (xlator_t *this, dict_t *dict)
+{
+        afr_private_t           *priv  = NULL;
+        afr_self_heald_t        *shd = NULL;
+        int                     i = 0;
+        priv = this->private;
+        shd= &priv->shd;
+        for (i = 0; i < priv->child_count; i++) {
+                if (shd->pos[i] != AFR_POS_LOCAL)
+                        continue;
+                _add_statistics_to_dict (this, dict, i);
+        }
+
+        return ;
+}
+
+static void
+reset_crawl_event (shd_crawl_event_t *crawl_event)
+{
+    crawl_event->healed_count = 0;
+    crawl_event->split_brain_count = 0;
+    crawl_event->heal_failed_count = 0;
+    GF_FREE (crawl_event->start_time_str);
+    crawl_event->start_time_str = NULL;
+    crawl_event->end_time_str = NULL;
+    crawl_event->crawl_type = NULL;
+    crawl_event->crawl_inprogress = _gf_false;
+    return;
+}
+
+static void
+afr_copy_crawl_event_struct (shd_crawl_event_t *src, shd_crawl_event_t *dst)
+{
+        dst->healed_count = src->healed_count;
+        dst->split_brain_count = src->split_brain_count;
+        dst->heal_failed_count = src->heal_failed_count;
+        dst->start_time_str = gf_strdup (src->start_time_str);
+        dst->end_time_str = "Crawl is already in progress";
+        dst->crawl_type = src->crawl_type;
+        dst->crawl_inprogress = _gf_true;
+        return;
+}
+
+static int
+afr_fill_crawl_statistics_of_running_crawl(xlator_t *this, dict_t *dict)
+{
+        shd_crawl_event_t       *evnt = NULL;
+        int                     ret = 0;
+        afr_private_t           *priv = NULL;
+        afr_self_heald_t        *shd = NULL;
+        int                     i = 0;
+        priv = this->private;
+        shd = &priv->shd;
+
+        evnt = GF_CALLOC (1, sizeof (shd_crawl_event_t),
+                          gf_afr_mt_shd_crawl_event_t);
+        if (!evnt) {
+                ret = -1;
+                goto out;
+        }
+        LOCK (&priv->lock);
+        {
+                for (i = 0; i < priv->child_count; i++) {
+                        if (shd->pos[i] != AFR_POS_LOCAL)
+                                continue;
+
+                        reset_crawl_event (evnt);
+
+                        if (!shd->crawl_events[i]) {
+                                continue;
+                        }
+
+                        afr_copy_crawl_event_struct (shd->crawl_events[i],
+                                                     evnt);
+                        _add_crawl_stats_to_dict (this, dict, i, evnt, NULL);
+
+                }
+        }
+        UNLOCK (&priv->lock);
+        reset_crawl_event (evnt);
+        GF_FREE (evnt);
+
+out:
+        return ret;
+}
+
+static int
+_add_local_subvols_crawl_statistics_to_dict (xlator_t *this, dict_t *dict)
+{
+        int ret = 0;
+        afr_fill_completed_crawl_statistics_to_dict (this, dict);
+        ret = afr_fill_crawl_statistics_of_running_crawl (this, dict);
+        return ret;
+}
+int
+_add_local_subvols_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict)
+{
+        afr_private_t           *priv = NULL;
+        afr_self_heald_t        *shd = NULL;
+        int                     i = 0;
+
+        priv = this->private;
+        shd = &priv->shd;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (shd->pos[i] != AFR_POS_LOCAL)
+                        continue;
+                _add_eh_to_dict (this, eh, dict, i);
+        }
+        return 0;
+}
+
+int
+afr_xl_op (xlator_t *this, dict_t *input, dict_t *output)
+{
+        gf_xl_afr_op_t   op = GF_AFR_OP_INVALID;
+        int              ret = 0;
+        afr_private_t    *priv = NULL;
+        afr_self_heald_t *shd = NULL;
+        int              xl_id = 0;
+
+        priv = this->private;
+        shd = &priv->shd;
+
+        ret = dict_get_int32 (input, "xl-op", (int32_t*)&op);
+        if (ret)
+                goto out;
+        ret = dict_get_int32 (input, this->name, &xl_id);
+        if (ret)
+                goto out;
+        ret = dict_set_int32 (output, this->name, xl_id);
+        if (ret)
+                goto out;
+        switch (op) {
+        case GF_AFR_OP_HEAL_INDEX:
+                ret = _do_self_heal_on_local_subvols (this, INDEX, output);
+                break;
+        case GF_AFR_OP_HEAL_FULL:
+                ret = _do_self_heal_on_local_subvols (this, FULL, output);
+                break;
+        case GF_AFR_OP_INDEX_SUMMARY:
+                (void)_get_index_summary_on_local_subvols (this, output);
+                ret = 0;
+                break;
+        case GF_AFR_OP_HEALED_FILES:
+                ret = _add_local_subvols_eh_to_dict (this, shd->healed, output);
+                break;
+        case GF_AFR_OP_HEAL_FAILED_FILES:
+                ret = _add_local_subvols_eh_to_dict (this, shd->heal_failed,
+                                                   output);
+                break;
+        case GF_AFR_OP_SPLIT_BRAIN_FILES:
+                ret = _add_local_subvols_eh_to_dict (this, shd->split_brain,
+                                                   output);
+                break;
+        case GF_AFR_OP_STATISTICS:
+                ret = _add_local_subvols_crawl_statistics_to_dict (this, output);
+                break;
+        case GF_AFR_OP_STATISTICS_HEAL_COUNT:
+        case GF_AFR_OP_STATISTICS_HEAL_COUNT_PER_REPLICA:
+                ret = _do_crawl_op_on_local_subvols (this, INDEX_TO_BE_HEALED,
+                                                     STATISTICS_TO_BE_HEALED,
+                                                     output);
+                break;
+        default:
+                gf_log (this->name, GF_LOG_ERROR, "Unknown set op %d", op);
+                break;
+        }
+out:
+        dict_del (output, this->name);
+        return ret;
+}
+
+void
+afr_poll_self_heal (void *data)
+{
+        afr_private_t    *priv = NULL;
+        afr_self_heald_t *shd = NULL;
+        struct timespec  timeout = {0};
+        xlator_t         *this = NULL;
+        long             child = (long)data;
+        gf_timer_t       *old_timer = NULL;
+        gf_timer_t       *new_timer = NULL;
+        shd_pos_t        pos_data = {0};
+        int              ret = 0;
+
+        this = THIS;
+        priv = this->private;
+        shd = &priv->shd;
+
+        if (shd->pos[child] == AFR_POS_UNKNOWN) {
+                pos_data.this = this;
+                pos_data.child = child;
+                ret = synctask_new (this->ctx->env,
+                                    afr_syncop_find_child_position,
+                                    NULL, NULL, &pos_data);
+                if (!ret)
+                        shd->pos[child] = pos_data.pos;
+        }
+        if (shd->enabled && (shd->pos[child] == AFR_POS_LOCAL))
+                _do_self_heal_on_subvol (this, child, INDEX);
+        timeout.tv_sec = shd->timeout;
+        timeout.tv_nsec = 0;
+        //notify and previous timer should be synchronized.
+        LOCK (&priv->lock);
+        {
+                old_timer = shd->timer[child];
+                if (shd->pos[child] == AFR_POS_REMOTE)
+                        goto unlock;
+                shd->timer[child] = gf_timer_call_after (this->ctx, timeout,
+                                                         afr_poll_self_heal,
+                                                         data);
+                new_timer = shd->timer[child];
+        }
+unlock:
+        UNLOCK (&priv->lock);
+
+        if (old_timer)
+                gf_timer_call_cancel (this->ctx, old_timer);
+        if (!new_timer && (shd->pos[child] != AFR_POS_REMOTE)) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Could not create self-heal polling timer for %s",
+                        priv->children[child]->name);
+        }
+        return;
+}
+
+static int
+afr_handle_child_up  (int ret, call_frame_t *sync_frame, void *data)
+{
+        afr_self_heald_t *shd = NULL;
+        shd_pos_t        *pos_data = data;
+        afr_private_t    *priv = NULL;
+
+        if (ret)
+                goto out;
+
+        priv = pos_data->this->private;
+        shd = &priv->shd;
+        shd->pos[pos_data->child] = pos_data->pos;
+        if (pos_data->pos != AFR_POS_REMOTE)
+                afr_poll_self_heal ((void*)(long)pos_data->child);
+        _do_self_heal_on_local_subvols (THIS, INDEX, NULL);
+out:
+        GF_FREE (data);
+        return 0;
+}
+
+void
+afr_proactive_self_heal (void *data)
+{
+        xlator_t         *this = NULL;
+        long             child = (long)data;
+        shd_pos_t        *pos_data = NULL;
+        int              ret = 0;
+
+        this = THIS;
+
+        //Position of brick could have changed and it could be local now.
+        //Compute the position again
+        pos_data = GF_CALLOC (1, sizeof (*pos_data), gf_afr_mt_pos_data_t);
+        if (!pos_data)
+                goto out;
+        pos_data->this = this;
+        pos_data->child = child;
+        ret = synctask_new (this->ctx->env, afr_syncop_find_child_position,
+                            afr_handle_child_up, NULL, pos_data);
+        if (ret)
+                goto out;
+out:
+        return;
+}
+
+static int
+get_pathinfo_host (char *pathinfo, char *hostname, size_t size)
+{
+        char    *start = NULL;
+        char    *end = NULL;
+        int     ret  = -1;
+        int     i    = 0;
+
+        if (!pathinfo)
+                goto out;
+
+        start = strchr (pathinfo, ':');
+        if (!start)
+                goto out;
+        end = strrchr (pathinfo, ':');
+        if (start == end)
+                goto out;
+
+        memset (hostname, 0, size);
+        i = 0;
+        while (++start != end)
+                hostname[i++] = *start;
+        ret = 0;
+out:
+        return ret;
+}
+
+int
+afr_local_pathinfo (char *pathinfo, gf_boolean_t *local)
+{
+        int             ret   = 0;
+        char            pathinfohost[1024] = {0};
+        char            localhost[1024] = {0};
+        xlator_t        *this = THIS;
+
+        *local = _gf_false;
+        ret = get_pathinfo_host (pathinfo, pathinfohost, sizeof (pathinfohost));
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Invalid pathinfo: %s",
+                        pathinfo);
+                goto out;
+        }
+
+        ret = gethostname (localhost, sizeof (localhost));
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "gethostname() failed, "
+                        "reason: %s", strerror (errno));
+                goto out;
+        }
+
+        if (!strcmp (localhost, pathinfohost))
+                *local = _gf_true;
+out:
+        return ret;
+}
+
+int
+afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data,
+                           loc_t *dirloc)
+{
+        afr_private_t *priv = NULL;
+        dict_t        *xattr = NULL;
+        void          *index_gfid = NULL;
+        void          *base_indices_holder_vgfid = NULL;
+        loc_t         rootloc = {0};
+        struct iatt   iattr = {0};
+        struct iatt   parent = {0};
+        int           ret = 0;
+        xlator_t      *readdir_xl = crawl_data->readdir_xl;
+
+        priv = this->private;
+        if (crawl_data->crawl == FULL) {
+                afr_build_root_loc (this, dirloc);
+        } else if (crawl_data->crawl == INDEX) {
+                afr_build_root_loc (this, &rootloc);
+                ret = syncop_getxattr (readdir_xl, &rootloc, &xattr,
+                                       GF_XATTROP_INDEX_GFID);
+                if (ret < 0)
+                        goto out;
+                ret = dict_get_ptr (xattr, GF_XATTROP_INDEX_GFID, &index_gfid);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR, "failed to get index "
+                                "dir gfid on %s", readdir_xl->name);
+                        goto out;
+                }
+                if (!index_gfid) {
+                        gf_log (this->name, GF_LOG_ERROR, "index gfid empty "
+                                "on %s", readdir_xl->name);
+                        ret = -1;
+                        goto out;
+                }
+                uuid_copy (dirloc->gfid, index_gfid);
+                dirloc->path = "";
+                dirloc->inode = inode_new (priv->root_inode->table);
+                ret = syncop_lookup (readdir_xl, dirloc, NULL,
+                                     &iattr, NULL, &parent);
+                if (ret < 0) {
+                        if (errno != ENOENT) {
+                                gf_log (this->name, GF_LOG_ERROR, "lookup "
+                                        "failed on index dir on %s - (%s)",
+                                        readdir_xl->name, strerror (errno));
+                        }
+                        goto out;
+                }
+                ret = _link_inode_update_loc (this, dirloc, &iattr);
+                if (ret)
+                        goto out;
+        } else if (crawl_data->crawl == INDEX_TO_BE_HEALED) {
+                afr_build_root_loc (this, &rootloc);
+                ret = syncop_getxattr (readdir_xl, &rootloc, &xattr,
+                                       GF_BASE_INDICES_HOLDER_GFID);
+                if (ret < 0)
+                        goto out;
+                ret = dict_get_ptr (xattr, GF_BASE_INDICES_HOLDER_GFID,
+                                    &base_indices_holder_vgfid);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR, "index gfid empty "
+                                "on %s", readdir_xl->name);
+                        ret = -1;
+                        goto out;
+                }
+                if (!base_indices_holder_vgfid) {
+                        gf_log (this->name, GF_LOG_ERROR, "Base indices holder"
+                                "virtual gfid is null on %s", readdir_xl->name);
+                        ret = -1;
+                        goto out;
+                }
+                uuid_copy (dirloc->gfid,  base_indices_holder_vgfid);
+                dirloc->path = "";
+                dirloc->inode = inode_new (priv->root_inode->table);
+                ret = syncop_lookup (readdir_xl, dirloc, NULL, &iattr, NULL,
+                                     &parent);
+                if (ret < 0) {
+                        if (errno != ENOENT) {
+                                gf_log (this->name, GF_LOG_ERROR, "lookup "
+                                        "failed for base_indices_holder dir"
+                                        " on %s - (%s)", readdir_xl->name,
+                                        strerror (errno));
+
+                        } else {
+                                gf_log (this->name, GF_LOG_ERROR, "base_indices"
+                                        "_holder is not yet created.");
+                        }
+                        goto out;
+                }
+                ret = _link_inode_update_loc (this, dirloc, &iattr);
+                if (ret)
+                        goto out;
+        }
+        ret = 0;
+out:
+        if (xattr)
+                dict_unref (xattr);
+        loc_wipe (&rootloc);
+        return ret;
+}
+
+int
+afr_crawl_opendir (xlator_t *this, afr_crawl_data_t *crawl_data, fd_t **dirfd,
+                   loc_t *dirloc)
+{
+        fd_t          *fd   = NULL;
+        int           ret = 0;
+
+        if (crawl_data->crawl == FULL) {
+                fd = fd_create (dirloc->inode, crawl_data->pid);
+                if (!fd) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "Failed to create fd for %s", dirloc->path);
+                        ret = -1;
+                        goto out;
+                }
+
+                ret = syncop_opendir (crawl_data->readdir_xl, dirloc, fd);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "opendir failed on %s", dirloc->path);
+                        goto out;
+                }
+        } else {
+                fd = fd_anonymous (dirloc->inode);
+        }
+        ret = 0;
+out:
+        if (!ret)
+                *dirfd = fd;
+        return ret;
+}
+
+xlator_t*
+afr_crawl_readdir_xl_get (xlator_t *this, afr_crawl_data_t *crawl_data)
+{
+        afr_private_t *priv = this->private;
+
+        if (crawl_data->crawl == FULL) {
+                return this;
+        } else {
+                return priv->children[crawl_data->child];
+        }
+        return NULL;
+}
+
+int
+afr_crawl_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent,
+                           gf_dirent_t *entry, afr_crawl_data_t *crawl_data)
+{
+        int           ret = -1;
+        afr_private_t *priv = NULL;
+
+        priv = this->private;
+        if (crawl_data->crawl == FULL) {
+                ret = afr_build_child_loc (this, child, parent, entry->d_name);
+        } else if (crawl_data->crawl == INDEX_TO_BE_HEALED) {
+                ret = _build_index_loc (this, child, entry->d_name, parent);
+                if (ret)
+                        goto out;
+                child->inode = inode_new (priv->root_inode->table);
+                if (!child->inode) {
+                        ret = -1;
+                        goto out;
+                }
+                child->path = NULL;
+        } else {
+                child->inode = inode_new (priv->root_inode->table);
+                if (!child->inode)
+                        goto out;
+                uuid_parse (entry->d_name, child->gfid);
+                ret = _loc_assign_gfid_path (child);
+        }
+out:
+        return ret;
+}
+
+static int
+_process_entries (xlator_t *this, loc_t *parentloc, gf_dirent_t *entries,
+                  off_t *offset, afr_crawl_data_t *crawl_data)
+{
+        gf_dirent_t      *entry = NULL;
+        gf_dirent_t      *tmp = NULL;
+        int              ret = 0;
+        loc_t            entry_loc = {0};
+        fd_t             *fd = NULL;
+        struct iatt      iattr = {0};
+
+        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
+                if (!_crawl_proceed (this, crawl_data->child,
+                                     crawl_data->crawl_flags, NULL)) {
+                        ret = -1;
+                        goto out;
+                }
+                *offset = entry->d_off;
+                if (IS_ENTRY_CWD (entry->d_name) ||
+                    IS_ENTRY_PARENT (entry->d_name))
+                        continue;
+                if ((crawl_data->crawl == FULL) &&
+                     uuid_is_null (entry->d_stat.ia_gfid)) {
+                        gf_log (this->name, GF_LOG_WARNING, "%s/%s: No "
+                                "gfid present skipping",
+                                parentloc->path, entry->d_name);
+                        continue;
+                }
+
+                loc_wipe (&entry_loc);
+                ret = afr_crawl_build_child_loc (this, &entry_loc, parentloc,
+                                                 entry, crawl_data);
+                if (ret)
+                        goto out;
+
+                ret = crawl_data->process_entry (this, crawl_data, entry,
+                                                 &entry_loc, parentloc, &iattr);
+
+                if (crawl_data->crawl == INDEX_TO_BE_HEALED && ret) {
+                       goto out;
+                } else if (ret) {
+                        continue;
+                }
+
+                if ((crawl_data->crawl == INDEX) ||
+                    (crawl_data->crawl == INDEX_TO_BE_HEALED))
+                        continue;
+
+                if (!IA_ISDIR (iattr.ia_type))
+                        continue;
+                fd = NULL;
+                ret = afr_crawl_opendir (this, crawl_data, &fd, &entry_loc);
+                if (ret)
+                        continue;
+                ret = _crawl_directory (fd, &entry_loc, crawl_data);
+                if (fd)
+                        fd_unref (fd);
+        }
+        ret = 0;
+out:
+        if ((crawl_data->crawl == INDEX_TO_BE_HEALED)  && ret) {
+                gf_log (this->name, GF_LOG_ERROR,"Failed to get the hardlink "
+                        "count");
+        }
+        loc_wipe (&entry_loc);
+        return ret;
+}
+
+static int
+_crawl_directory (fd_t *fd, loc_t *loc, afr_crawl_data_t *crawl_data)
+{
+        xlator_t        *this = NULL;
+        off_t           offset   = 0;
+        gf_dirent_t     entries;
+        int             ret = 0;
+        gf_boolean_t    free_entries = _gf_false;
+        xlator_t        *readdir_xl = crawl_data->readdir_xl;
+
+        INIT_LIST_HEAD (&entries.list);
+        this = THIS;
+
+        GF_ASSERT (loc->inode);
+
+        if (crawl_data->crawl == FULL)
+                gf_log (this->name, GF_LOG_DEBUG, "crawling %s", loc->path);
+        else
+                gf_log (this->name, GF_LOG_DEBUG, "crawling INDEX %s",
+                        uuid_utoa (loc->gfid));
+
+        while (1) {
+                if (crawl_data->crawl == FULL)
+                        ret = syncop_readdirp (readdir_xl, fd, 131072, offset,
+                                               NULL, &entries);
+                else
+                        ret = syncop_readdir (readdir_xl, fd, 131072, offset,
+                                              &entries);
+                if (ret <= 0)
+                        break;
+                ret = 0;
+                free_entries = _gf_true;
+
+                if (!_crawl_proceed (this, crawl_data->child,
+                                     crawl_data->crawl_flags, NULL)) {
+                        ret = -1;
+                        goto out;
+                }
+                if (list_empty (&entries.list))
+                        goto out;
+
+                ret = _process_entries (this, loc, &entries, &offset,
+                                        crawl_data);
+                if ((ret < 0) && (crawl_data->crawl == INDEX_TO_BE_HEALED)) {
+                        goto out;
+                }
+                gf_dirent_free (&entries);
+                free_entries = _gf_false;
+        }
+        ret = 0;
+out:
+        if (free_entries)
+                gf_dirent_free (&entries);
+        return ret;
+}
+
+static char*
+position_str_get (afr_child_pos_t pos)
+{
+        switch (pos) {
+        case AFR_POS_UNKNOWN:
+                return "unknown";
+        case AFR_POS_LOCAL:
+                return "local";
+        case AFR_POS_REMOTE:
+                return "remote";
+        }
+        return NULL;
+}
+
+int
+afr_find_child_position (xlator_t *this, int child, afr_child_pos_t *pos)
+{
+        afr_private_t    *priv = NULL;
+        afr_self_heald_t *shd  = NULL;
+        dict_t           *xattr_rsp = NULL;
+        loc_t            loc = {0};
+        int              ret = 0;
+        char             *node_uuid = NULL;
+
+        priv = this->private;
+        shd  = &priv->shd;
+
+        afr_build_root_loc (this, &loc);
+
+        ret = syncop_getxattr (priv->children[child], &loc, &xattr_rsp,
+                               GF_XATTR_NODE_UUID_KEY);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR, "getxattr failed on %s - "
+                        "(%s)", priv->children[child]->name, strerror (errno));
+                goto out;
+        }
+
+        ret = dict_get_str (xattr_rsp, GF_XATTR_NODE_UUID_KEY, &node_uuid);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "node-uuid key not found on "
+                        "child %s", priv->children[child]->name);
+                goto out;
+        }
+
+        if (!strcmp (node_uuid, shd->node_uuid))
+                *pos = AFR_POS_LOCAL;
+        else
+                *pos = AFR_POS_REMOTE;
+
+        gf_log (this->name, GF_LOG_DEBUG, "child %s is %s",
+                priv->children[child]->name, position_str_get (*pos));
+out:
+        if (ret)
+                *pos = AFR_POS_UNKNOWN;
+        loc_wipe (&loc);
+        return ret;
+}
+
+int
+afr_syncop_find_child_position (void *data)
+{
+        shd_pos_t *pos_data = data;
+        int       ret = 0;
+
+        ret = afr_find_child_position (pos_data->this, pos_data->child,
+                                       &pos_data->pos);
+        return ret;
+}
+
+static int
+afr_dir_crawl (void *data)
+{
+        xlator_t            *this = NULL;
+        int                 ret = -1;
+        xlator_t            *readdir_xl = NULL;
+        fd_t                *fd = NULL;
+        loc_t               dirloc = {0};
+        afr_crawl_data_t    *crawl_data = data;
+
+        this = THIS;
+
+        if (!_crawl_proceed (this, crawl_data->child, crawl_data->crawl_flags,
+                             NULL))
+                goto out;
+
+        readdir_xl = afr_crawl_readdir_xl_get (this, crawl_data);
+        if (!readdir_xl)
+                goto out;
+        crawl_data->readdir_xl = readdir_xl;
+
+        ret = afr_crawl_build_start_loc (this, crawl_data, &dirloc);
+        if (ret)
+                goto out;
+
+        ret = afr_crawl_opendir (this, crawl_data, &fd, &dirloc);
+        if (ret) {
+                if (crawl_data->crawl == INDEX_TO_BE_HEALED) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to open base_"
+                                "indices_holder");
+                }
+                goto out;
+        }
+
+        ret = _crawl_directory (fd, &dirloc, crawl_data);
+        if (ret)
+                gf_log (this->name, GF_LOG_ERROR, "Crawl failed on %s",
+                        readdir_xl->name);
+        else
+                gf_log (this->name, GF_LOG_DEBUG, "Crawl completed "
+                        "on %s", readdir_xl->name);
+        if (crawl_data->crawl == INDEX)
+                dirloc.path = NULL;
+out:
+        if (fd)
+                fd_unref (fd);
+        if ((crawl_data->crawl == INDEX) ||
+            (crawl_data->crawl == INDEX_TO_BE_HEALED ))
+                dirloc.path = NULL;
+        loc_wipe (&dirloc);
+        return ret;
+}
+
+char *
+get_crawl_type_in_string (afr_crawl_type_t crawl)
+{
+        char    *index = "INDEX";
+        char    *full  = "FULL";
+        char    *crawl_type = NULL;
+
+        if (crawl == INDEX){
+                crawl_type = index;
+        } else if (crawl == FULL) {
+                crawl_type = full;
+        }
+
+        return  crawl_type;
+}
+
+static int
+afr_allocate_crawl_event (xlator_t *this, int child, afr_crawl_type_t crawl)
+{
+        afr_private_t           *priv = NULL;
+        afr_self_heald_t        *shd = NULL;
+        int                     ret = 0;
+        shd_crawl_event_t       *crawl_event = NULL;
+        time_t                  get_time = 0;
+
+        priv = this->private;
+        shd = &priv->shd;
+
+        crawl_event = GF_CALLOC (sizeof (shd_crawl_event_t), 1,
+                                 gf_afr_mt_shd_crawl_event_t);
+        if (!crawl_event) {
+                ret = -1;
+                goto out;
+        }
+
+        get_time =  time(NULL);
+        if (get_time == ((time_t)-1)) {
+                 ret = -1;
+                goto out;
+        }
+
+        crawl_event->start_time_str = gf_strdup (ctime(&get_time));
+
+        crawl_event->crawl_type = get_crawl_type_in_string (crawl);
+        if (!crawl_event->crawl_type) {
+                ret = -1;
+                goto out;
+        }
+        LOCK (&priv->lock);
+        {
+                shd->crawl_events[child] = crawl_event;
+        }
+        UNLOCK (&priv->lock);
+        ret = 0;
+out:
+        return ret;
+
+}
+
+static int
+afr_put_crawl_event_in_eh (xlator_t *this, int child)
+{
+        afr_private_t           *priv = NULL;
+        afr_self_heald_t        *shd = NULL;
+        int                     ret = 0;
+        time_t                  get_time = 0;
+        shd_crawl_event_t       **crawl_event = NULL;
+
+        priv = this->private;
+        shd = &priv->shd;
+
+        get_time = time(NULL);
+        if (get_time == ((time_t)-1)) {
+                ret = -1;
+                goto out;
+        }
+        crawl_event = (shd_crawl_event_t**)shd->crawl_events;
+        LOCK (&priv->lock);
+        {
+                crawl_event[child]->end_time_str = gf_strdup (ctime(&get_time));
+                ret = eh_save_history (shd->statistics[child],
+                                       crawl_event[child]);
+                crawl_event[child] = NULL;
+        }
+        UNLOCK (&priv->lock);
+out:
+        return ret;
+}
+
+static int
+afr_dir_exclusive_crawl (void *data)
+{
+        afr_private_t    *priv = NULL;
+        afr_self_heald_t *shd = NULL;
+        gf_boolean_t     crawl = _gf_false;
+        int              ret = 0;
+        int              child = -1;
+        xlator_t         *this = NULL;
+        afr_crawl_data_t *crawl_data = data;
+
+        this = THIS;
+        priv = this->private;
+        shd = &priv->shd;
+        child = crawl_data->child;
+
+        LOCK (&priv->lock);
+        {
+                if (shd->inprogress[child]) {
+                        if (shd->pending[child] != FULL)
+                                shd->pending[child] = crawl_data->crawl;
+                } else {
+                        shd->inprogress[child] = _gf_true;
+                        crawl = _gf_true;
+                }
+        }
+        UNLOCK (&priv->lock);
+
+        if (!crawl) {
+                gf_log (this->name, GF_LOG_INFO, "Another crawl is in progress "
+                        "for %s", priv->children[child]->name);
+                goto out;
+        }
+
+        do {
+                ret = afr_allocate_crawl_event (this, child, crawl_data->crawl);
+                if (ret)
+                        goto out;
+                afr_dir_crawl (data);
+
+                ret = afr_put_crawl_event_in_eh (this, child);
+                if (ret < 0)
+                        goto out;
+
+                LOCK (&priv->lock);
+                {
+                        if (shd->pending[child] != NONE) {
+                                crawl_data->crawl = shd->pending[child];
+                                shd->pending[child] = NONE;
+                        } else {
+                                shd->inprogress[child] = _gf_false;
+                                crawl = _gf_false;
+                        }
+                }
+                UNLOCK (&priv->lock);
+        } while (crawl);
+out:
+        return ret;
+}
+
+void
+afr_start_crawl (xlator_t *this, int idx, afr_crawl_type_t crawl,
+                 process_entry_cbk_t process_entry, void *op_data,
+                 gf_boolean_t exclusive, int crawl_flags,
+                 afr_crawl_done_cbk_t crawl_done)
+{
+        afr_private_t              *priv = NULL;
+        call_frame_t               *frame = NULL;
+        afr_crawl_data_t           *crawl_data = NULL;
+        int                        ret = 0;
+        int (*crawler) (void*) = NULL;
+
+        priv = this->private;
+
+        frame = create_frame (this, this->ctx->pool);
+        if (!frame)
+                goto out;
+
+        afr_set_lk_owner (frame, this, frame->root);
+        afr_set_low_priority (frame);
+        crawl_data = GF_CALLOC (1, sizeof (*crawl_data),
+                                gf_afr_mt_crawl_data_t);
+        if (!crawl_data)
+                goto out;
+        crawl_data->process_entry = process_entry;
+        crawl_data->child = idx;
+        crawl_data->pid = frame->root->pid;
+        crawl_data->crawl = crawl;
+        crawl_data->op_data = op_data;
+        crawl_data->crawl_flags = crawl_flags;
+        gf_log (this->name, GF_LOG_DEBUG, "starting crawl %d for %s",
+                crawl_data->crawl, priv->children[idx]->name);
+
+        if (exclusive)
+                crawler = afr_dir_exclusive_crawl;
+        else
+                crawler = afr_dir_crawl;
+        ret = synctask_new (this->ctx->env, crawler,
+                            crawl_done, frame, crawl_data);
+        if (ret)
+                gf_log (this->name, GF_LOG_ERROR, "afr crawl failed for child"
+                        " %d with ret %d", idx, ret);
+out:
+        return;
+}
+
+void
+afr_build_root_loc (xlator_t *this, loc_t *loc)
+{
+        afr_private_t   *priv = NULL;
+
+        priv = this->private;
+        loc->path = gf_strdup ("/");
+        loc->name = "";
+        loc->inode = inode_ref (priv->root_inode);
+        uuid_copy (loc->gfid, loc->inode->gfid);
+}
+
+int
+afr_set_root_gfid (dict_t *dict)
+{
+        uuid_t gfid;
+        int ret = 0;
+
+        memset (gfid, 0, 16);
+        gfid[15] = 1;
+
+        ret = afr_set_dict_gfid (dict, gfid);
+
+        return ret;
+}
diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h
new file mode 100644
index 000000000..e0c083754
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heald.h
@@ -0,0 +1,65 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __AFR_SELF_HEALD_H__
+#define __AFR_SELF_HEALD_H__
+#include "xlator.h"
+
+#define IS_ROOT_PATH(path) (!strcmp (path, "/"))
+#define IS_ENTRY_CWD(entry) (!strcmp (entry, "."))
+#define IS_ENTRY_PARENT(entry) (!strcmp (entry, ".."))
+#define AFR_ALL_CHILDREN -1
+
+typedef struct afr_crawl_data_ {
+        int                 child;
+        pid_t               pid;
+        afr_crawl_type_t    crawl;
+        xlator_t            *readdir_xl;
+        void                *op_data;
+        int                 crawl_flags;
+        int (*process_entry) (xlator_t *this, struct afr_crawl_data_ *crawl_data,
+                              gf_dirent_t *entry, loc_t *child, loc_t *parent,
+                              struct iatt *iattr);
+} afr_crawl_data_t;
+
+typedef struct crawl_event_stats_ {
+        uint64_t healed_count;
+        uint64_t split_brain_count;
+        uint64_t heal_failed_count;
+        char     *start_time_str;
+        char     *end_time_str;
+        char     *crawl_type;
+        gf_boolean_t crawl_inprogress;
+} shd_crawl_event_t;
+
+void _destroy_crawl_event_data (void *data);
+void _destroy_shd_event_data (void *data);
+
+typedef int (*process_entry_cbk_t) (xlator_t *this, afr_crawl_data_t *crawl_data,
+                              gf_dirent_t *entry, loc_t *child, loc_t *parent,
+                              struct iatt *iattr);
+
+void afr_build_root_loc (xlator_t *this, loc_t *loc);
+
+int afr_set_root_gfid (dict_t *dict);
+
+void
+afr_proactive_self_heal (void *data);
+
+int
+afr_xl_op (xlator_t *this, dict_t *input, dict_t *output);
+
+/*
+ * In addition to its self-heal use, this is used to find a local default
+ * read_child.
+ */
+int
+afr_local_pathinfo (char *pathinfo, gf_boolean_t *local);
+#endif /* __AFR_SELF_HEALD_H__ */
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
index 0590b5984..20306e469 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -1,24 +1,17 @@
 /*
-  Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  General Public License for more details.
-
-  You should have received a copy of the GNU General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include "dict.h"
 #include "byte-order.h"
+#include "common-utils.h"
+#include "timer.h"
 
 #include "afr.h"
 #include "afr-transaction.h"
@@ -26,187 +19,262 @@
 #include <signal.h>
 
 
-static void
-__mark_all_pending (int32_t *pending, int child_count)
-{	
-	int i;
-	
-	for (i = 0; i < child_count; i++)
-		pending[i] = hton32 (1);
+#define LOCKED_NO       0x0        /* no lock held */
+#define LOCKED_YES      0x1        /* for DATA, METADATA, ENTRY and higher_path
+                                      of RENAME */
+#define LOCKED_LOWER    0x2        /* for lower_path of RENAME */
+
+afr_fd_ctx_t *
+__afr_fd_ctx_get (fd_t *fd, xlator_t *this)
+{
+        uint64_t       ctx = 0;
+        int            ret = 0;
+        afr_fd_ctx_t  *fd_ctx = NULL;
+        int            i = 0;
+        afr_private_t *priv = NULL;
+
+        priv = this->private;
+
+        ret = __fd_ctx_get (fd, this, &ctx);
+
+        if (ret < 0 && fd_is_anonymous (fd)) {
+                ret = __afr_fd_ctx_set (this, fd);
+                if (ret < 0)
+                        goto out;
+
+                ret = __fd_ctx_get (fd, this, &ctx);
+                if (ret < 0)
+                        goto out;
+
+                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+                for (i = 0; i < priv->child_count; i++)
+                        fd_ctx->opened_on[i] = AFR_FD_OPENED;
+        }
+
+        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+out:
+        return fd_ctx;
 }
 
 
-static void
-__mark_child_dead (int32_t *pending, int child_count, int child)
+afr_fd_ctx_t *
+afr_fd_ctx_get (fd_t *fd, xlator_t *this)
 {
-	pending[child] = 0;
+        afr_fd_ctx_t  *fd_ctx = NULL;
+
+        LOCK(&fd->lock);
+        {
+                fd_ctx = __afr_fd_ctx_get (fd, this);
+        }
+        UNLOCK(&fd->lock);
+
+        return fd_ctx;
 }
 
 
 static void
-__mark_down_children (int32_t *pending, int child_count, unsigned char *child_up)
+afr_save_lk_owner (call_frame_t *frame)
 {
-	int i;
-	
-	for (i = 0; i < child_count; i++)
-		if (!child_up[i])
-			pending[i] = 0;
+        afr_local_t * local = NULL;
+
+        local = frame->local;
+
+        local->saved_lk_owner = frame->root->lk_owner;
 }
 
 
 static void
-__mark_all_success (int32_t *pending, int child_count)
+afr_restore_lk_owner (call_frame_t *frame)
 {
-	int i;
-	
-	for (i = 0; i < child_count; i++)
-		pending[i] = hton32 (-1);
-}
+        afr_local_t * local = NULL;
 
+        local = frame->local;
 
-static int
-__is_first_write_on_fd (xlator_t *this, fd_t *fd)
-{
-        int op_ret     = 0;
-        int _ret       = -1;
+        frame->root->lk_owner = local->saved_lk_owner;
+}
 
-        LOCK (&fd->inode->lock);
-        {
-                _ret = fd_ctx_get (fd, this, NULL);
-                if (_ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "first writev() on fd=%p, writing changelog",
-                                fd);
+static void
+__mark_all_pending (int32_t *pending[], int child_count,
+                    afr_transaction_type type)
+{
+        int i = 0;
+        int j = 0;
 
-                        _ret = fd_ctx_set (fd, this, 0xaf1);
-                        op_ret = 1;
-                }
+        for (i = 0; i < child_count; i++) {
+                j = afr_index_for_transaction_type (type);
+                pending[i][j] = hton32 (1);
         }
-        UNLOCK (&fd->inode->lock);
+}
 
-        return op_ret;
+
+static void
+__mark_child_dead (int32_t *pending[], int child_count, int child,
+                   afr_transaction_type type)
+{
+        int j = 0;
+
+        j = afr_index_for_transaction_type (type);
+
+        pending[child][j] = 0;
 }
 
 
-static int
-__unset_fd_ctx_if_set (xlator_t *this, fd_t *fd)
+static void
+__mark_pre_op_done_on_fd (call_frame_t *frame, xlator_t *this, int child_index)
 {
-        int op_ret = 0;
-        int _ret   = -1;
+        afr_local_t   *local = NULL;
+        afr_fd_ctx_t  *fd_ctx = NULL;
+
+        local = frame->local;
+
+        if (!local->fd)
+                return;
+
+        fd_ctx = afr_fd_ctx_get (local->fd, this);
+
+        if (!fd_ctx)
+                goto out;
 
-        LOCK (&fd->inode->lock);
+        LOCK (&local->fd->lock);
         {
-                _ret = fd_ctx_get (fd, this, NULL);
-                if (_ret == 0) {
-                        fd_ctx_del (fd, this, NULL);
-                        op_ret = 1;
-                }
+                if (local->transaction.type == AFR_DATA_TRANSACTION)
+                        fd_ctx->pre_op_done[child_index]++;
         }
-        UNLOCK (&fd->inode->lock);
-
-        return op_ret;
+        UNLOCK (&local->fd->lock);
+out:
+        return;
 }
 
-
-static int
-__changelog_enabled (afr_private_t *priv, afr_transaction_type type)
+static void
+__mark_non_participant_children (int32_t *pending[], int child_count,
+                                 unsigned char *participants,
+                                 afr_transaction_type type)
 {
-	int ret = 0;
+        int i = 0;
+        int j = 0;
 
-	switch (type) {
-	case AFR_DATA_TRANSACTION:
-		if (priv->data_change_log)
-			ret = 1;
-		
-		break;
+        j = afr_index_for_transaction_type (type);
+        for (i = 0; i < child_count; i++) {
+                if (!participants[i])
+                        pending[i][j] = 0;
+        }
+}
 
-	case AFR_METADATA_TRANSACTION:
-		if (priv->metadata_change_log)
-			ret = 1;
 
-		break;
+void
+__mark_all_success (int32_t *pending[], int child_count,
+                    afr_transaction_type type)
+{
+        int i;
+        int j;
 
-	case AFR_ENTRY_TRANSACTION:
-	case AFR_ENTRY_RENAME_TRANSACTION:
-		if (priv->entry_change_log)
-			ret = 1;
+        for (i = 0; i < child_count; i++) {
+                j = afr_index_for_transaction_type (type);
+                pending[i][j] = hton32 (-1);
+        }
+}
 
-		break;
-		
-	case AFR_FLUSH_TRANSACTION:
-		ret = 1;
-	}
+void
+_set_all_child_errno (int *child_errno, unsigned int child_count)
+{
+        int     i = 0;
 
-	return ret;
+        for (i = 0; i < child_count; i++)
+                if (child_errno[i] == 0)
+                        child_errno[i] = ENOTCONN;
+}
+
+void
+afr_transaction_perform_fop (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        afr_private_t   *priv = NULL;
+        fd_t            *fd   = NULL;
+
+        local = frame->local;
+        priv  = this->private;
+        fd    = local->fd;
+
+        __mark_all_success (local->pending, priv->child_count,
+                            local->transaction.type);
+
+        _set_all_child_errno (local->child_errno, priv->child_count);
+
+        /*  Perform fops with the lk-owner from top xlator.
+         *  Eg: lk-owner of posix-lk and flush should be same,
+         *  flush cant clear the  posix-lks without that lk-owner.
+         */
+        afr_save_lk_owner (frame);
+        frame->root->lk_owner =
+                local->transaction.main_frame->root->lk_owner;
+
+
+        /* The wake up needs to happen independent of
+           what type of fop arrives here. If it was
+           a write, then it has already inherited the
+           lock and changelog. If it was not a write,
+           then the presumption of the optimization (of
+           optimizing for successive write operations)
+           fails.
+        */
+        if (fd)
+                afr_delayed_changelog_wake_up (this, fd);
+        local->transaction.fop (frame, this);
 }
 
 
 static int
-__changelog_needed_pre_op (call_frame_t *frame, xlator_t *this)
+__changelog_enabled (afr_private_t *priv, afr_transaction_type type)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	fd_t *          fd    = NULL;
+        int ret = 0;
 
-	int op_ret   = 0;
+        switch (type) {
+        case AFR_DATA_TRANSACTION:
+                if (priv->data_change_log)
+                        ret = 1;
 
-	priv  = this->private;
-	local = frame->local;
-	
-	if (__changelog_enabled (priv, local->transaction.type)) {
-		switch (local->op) {
-
-		case GF_FOP_WRITE:
-		case GF_FOP_FTRUNCATE:
-			/* 
-			   if it's a data transaction, we write the changelog
-			   only on the first write on an fd 
-			*/
-			
-			fd = local->fd;
-			if (!fd || __is_first_write_on_fd (this, fd))
-				op_ret = 1;
-
-			break;
-
-		case GF_FOP_FLUSH:
-			/* only do post-op on flush() */
-
-			op_ret = 0;
-			break;
-
-		default:
-			op_ret = 1;
-		}
-	}
+                break;
+
+        case AFR_METADATA_TRANSACTION:
+                if (priv->metadata_change_log)
+                        ret = 1;
+
+                break;
+
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+                if (priv->entry_change_log)
+                        ret = 1;
 
-	return op_ret;
+                break;
+        }
+
+        return ret;
 }
 
 
 static int
-__changelog_needed_post_op (call_frame_t *frame, xlator_t *this)
+__fop_changelog_needed (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
+        afr_private_t * priv  = NULL;
+        afr_local_t   * local = NULL;
+        int op_ret = 0;
+        afr_transaction_type type = -1;
 
-	int op_ret = 0;
-	afr_transaction_type type = -1;
+        priv  = this->private;
+        local = frame->local;
+        type  = local->transaction.type;
 
-	priv  = this->private;
-	local = frame->local;
-	type  = local->transaction.type;
-
-	if (__changelog_enabled (priv, type)) {
+        if (__changelog_enabled (priv, type)) {
                 switch (local->op) {
 
                 case GF_FOP_WRITE:
                 case GF_FOP_FTRUNCATE:
-                        op_ret = 0;
+                        op_ret = 1;
                         break;
 
                 case GF_FOP_FLUSH:
-                        op_ret = __unset_fd_ctx_if_set (this, local->fd);
+                        op_ret = 0;
                         break;
 
                 default:
@@ -214,789 +282,1682 @@ __changelog_needed_post_op (call_frame_t *frame, xlator_t *this)
                 }
         }
 
-	return op_ret;
+        return op_ret;
 }
 
+int
+afr_set_pending_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending,
+                      int child, afr_xattrop_type_t op)
+{
+        int i = 0;
+        int ret = 0;
+
+        if (op == LOCAL_FIRST) {
+                ret = dict_set_static_bin (xattr, priv->pending_key[child],
+                                           pending[child],
+                                   AFR_NUM_CHANGE_LOGS * sizeof (int32_t));
+                if (ret)
+                        goto out;
+        }
+        for (i = 0; i < priv->child_count; i++) {
+                if (i == child)
+                        continue;
+                ret = dict_set_static_bin (xattr, priv->pending_key[i],
+                                           pending[i],
+                                   AFR_NUM_CHANGE_LOGS * sizeof (int32_t));
+                /* 3 = data+metadata+entry */
+
+                if (ret < 0)
+                        goto out;
+        }
+        if (op == LOCAL_LAST) {
+                ret = dict_set_static_bin (xattr, priv->pending_key[child],
+                                           pending[child],
+                                   AFR_NUM_CHANGE_LOGS * sizeof (int32_t));
+                if (ret)
+                        goto out;
+        }
+out:
+        return ret;
+}
 
-static int
+int
 afr_lock_server_count (afr_private_t *priv, afr_transaction_type type)
 {
-	int ret = 0;
+        int ret = 0;
 
-	switch (type) {
-	case AFR_FLUSH_TRANSACTION:
-	case AFR_DATA_TRANSACTION:
-		ret = priv->data_lock_server_count;
-		break;
+        switch (type) {
+        case AFR_DATA_TRANSACTION:
+                ret = priv->child_count;
+                break;
 
-	case AFR_METADATA_TRANSACTION:
-		ret = priv->metadata_lock_server_count;
-		break;
+        case AFR_METADATA_TRANSACTION:
+                ret = priv->child_count;
+                break;
 
-	case AFR_ENTRY_TRANSACTION:
-	case AFR_ENTRY_RENAME_TRANSACTION:
-		ret = priv->entry_lock_server_count;
-		break;
-	}
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+                ret = priv->child_count;
+                break;
+        }
 
-	return ret;
+        return ret;
 }
 
-
-/* {{{ unlock */
+/* {{{ pending */
 
 int32_t
-afr_unlock_common_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno)
+afr_changelog_post_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                           dict_t *xdata)
 {
-	afr_local_t *local;
-	int call_count = 0;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_private_t       *priv     = NULL;
+        afr_local_t         *local    = NULL;
+        int                  call_count = -1;
 
-	local = frame->local;
+        priv     = this->private;
+        local    = frame->local;
+        int_lock = &local->internal_lock;
 
-	LOCK (&frame->lock);
-	{
-		call_count = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
+        LOCK (&frame->lock);
+        {
+                call_count = --local->call_count;
+        }
+        UNLOCK (&frame->lock);
 
-	if (call_count == 0) {
-		local->transaction.done (frame, this);
-	}
-	
-	return 0;
+        if (call_count == 0) {
+                if (local->transaction.resume_stub) {
+			call_resume (local->transaction.resume_stub);
+                        local->transaction.resume_stub = NULL;
+                }
+
+                if (afr_lock_server_count (priv, local->transaction.type) == 0) {
+                        local->transaction.done (frame, this);
+                } else {
+                        int_lock->lock_cbk = local->transaction.done;
+                        afr_unlock (frame, this);
+                }
+        }
+
+        return 0;
 }
 
 
+void
+afr_transaction_rm_stale_children (call_frame_t *frame, xlator_t *this,
+                                   inode_t *inode, afr_transaction_type type)
+{
+        int             i = -1;
+        int             count = 0;
+        int             read_child = -1;
+        afr_private_t   *priv = NULL;
+        afr_local_t     *local = NULL;
+        int             **pending = NULL;
+        int             idx = 0;
+        int32_t         *stale_children = NULL;
+        int32_t         *fresh_children = NULL;
+        gf_boolean_t    rm_stale_children = _gf_false;
+
+        idx = afr_index_for_transaction_type (type);
+
+        priv = this->private;
+        local = frame->local;
+        pending = local->pending;
+
+        if (local->op_ret < 0)
+                goto out;
+        fresh_children = local->fresh_children;
+        read_child = afr_inode_get_read_ctx (this, inode, fresh_children);
+        if (read_child < 0) {
+                gf_log (this->name, GF_LOG_DEBUG, "Possible split-brain "
+                        "for %s", uuid_utoa (inode->gfid));
+                goto out;
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (!afr_is_child_present (fresh_children,
+                                           priv->child_count, i))
+                        continue;
+                if (pending[i][idx])
+                        continue;
+                /* child is down or op failed on it */
+                if (!stale_children)
+                        stale_children = afr_children_create (priv->child_count);
+                if (!stale_children)
+                        goto out;
+
+                rm_stale_children = _gf_true;
+                stale_children[count++] = i;
+                gf_log (this->name, GF_LOG_DEBUG, "Removing stale child "
+                        "%d for %s", i, uuid_utoa (inode->gfid));
+        }
+
+        if (!rm_stale_children)
+                goto out;
+
+        afr_inode_rm_stale_children (this, inode, stale_children);
+out:
+        GF_FREE (stale_children);
+        return;
+}
+
+afr_inodelk_t*
+afr_get_inodelk (afr_internal_lock_t *int_lock, char *dom)
+{
+        afr_inodelk_t *inodelk = NULL;
+        int           i = 0;
+
+        for (i = 0; int_lock->inodelk[i].domain; i++) {
+                inodelk = &int_lock->inodelk[i];
+                if (strcmp (dom, inodelk->domain) == 0)
+                        return inodelk;
+        }
+        return NULL;
+}
+
+unsigned char*
+afr_locked_nodes_get (afr_transaction_type type, afr_internal_lock_t *int_lock)
+{
+        unsigned char *locked_nodes = NULL;
+        afr_inodelk_t *inodelk = NULL;
+        switch (type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+                inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+                locked_nodes = inodelk->locked_nodes;
+        break;
+
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+                /*Because same set of subvols participate in all lockee
+                 * entities*/
+                locked_nodes = int_lock->lockee[0].locked_nodes;
+        break;
+        }
+        return locked_nodes;
+}
+
 int
-afr_unlock (call_frame_t *frame, xlator_t *this)
+afr_changelog_pre_op_call_count (afr_transaction_type type,
+                                 afr_internal_lock_t *int_lock,
+                                 unsigned int child_count)
 {
-	struct flock flock;			
+        int           call_count = 0;
+        unsigned char *locked_nodes = NULL;
 
-	int i = 0;				
-	int call_count = 0;		     
+        locked_nodes = afr_locked_nodes_get (type, int_lock);
+        GF_ASSERT (locked_nodes);
 
-	afr_local_t *local = NULL;
-	afr_private_t * priv = this->private;
+        call_count = afr_locked_children_count (locked_nodes, child_count);
+        if (type == AFR_ENTRY_RENAME_TRANSACTION)
+                call_count *= 2;
 
-	local = frame->local;
-	
-	call_count = afr_locked_nodes_count (local->transaction.locked_nodes, 
-					     priv->child_count);
-	
-	if (call_count == 0) {
-		local->transaction.done (frame, this);
-		return 0;
-	}
+        return call_count;
+}
 
-	if (local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) 
-		call_count *= 2;
-
-	local->call_count = call_count;		
-
-	for (i = 0; i < priv->child_count; i++) {				
-		flock.l_start = local->transaction.start;			
-		flock.l_len   = local->transaction.len;
-		flock.l_type  = F_UNLCK;			
-
-		if (local->transaction.locked_nodes[i]) {
-			switch (local->transaction.type) {
-			case AFR_DATA_TRANSACTION:
-			case AFR_METADATA_TRANSACTION:
-			case AFR_FLUSH_TRANSACTION:
-
-				if (local->fd) {
-					STACK_WIND (frame, afr_unlock_common_cbk,	
-						    priv->children[i], 
-						    priv->children[i]->fops->finodelk, 
-						    this->name, local->fd, 
-                                                    F_SETLK, &flock); 
-				} else {
-					STACK_WIND (frame, afr_unlock_common_cbk,	
-						    priv->children[i], 
-						    priv->children[i]->fops->inodelk, 
-						    this->name, &local->loc,
-                                                    F_SETLK, &flock); 
-				}
-				
-				break;
-
-			case AFR_ENTRY_RENAME_TRANSACTION:
-				
-				STACK_WIND (frame, afr_unlock_common_cbk,	
-					    priv->children[i], 
-					    priv->children[i]->fops->entrylk, 
-                                            this->name,
-					    &local->transaction.new_parent_loc, 
-					    local->transaction.new_basename,
-					    ENTRYLK_UNLOCK, ENTRYLK_WRLCK);
-
-				call_count--;
-
-				/* fall through */
-
-			case AFR_ENTRY_TRANSACTION:
-				if (local->fd) {
-					STACK_WIND (frame, afr_unlock_common_cbk,	
-						    priv->children[i], 
-						    priv->children[i]->fops->fentrylk, 
-						    this->name, local->fd, 
-						    local->transaction.basename,
-						    ENTRYLK_UNLOCK, ENTRYLK_WRLCK);
-				} else {
-					STACK_WIND (frame, afr_unlock_common_cbk,	
-						    priv->children[i], 
-						    priv->children[i]->fops->entrylk, 
-						    this->name,
-                                                    &local->transaction.parent_loc, 
-						    local->transaction.basename,
-						    ENTRYLK_UNLOCK, ENTRYLK_WRLCK);
-
-				}
-				break;
-			}
-			
-			if (!--call_count)
-				break;
-		}
-	}
+int
+afr_changelog_post_op_call_count (afr_transaction_type type,
+                                  unsigned char *pre_op,
+                                  unsigned int child_count)
+{
+        int           call_count = 0;
+
+        call_count = afr_pre_op_done_children_count (pre_op, child_count);
+        if (type == AFR_ENTRY_RENAME_TRANSACTION)
+                call_count *= 2;
 
-	return 0;
+        return call_count;
 }
 
-/* }}} */
+void
+afr_compute_txn_changelog (afr_local_t *local, afr_private_t *priv)
+{
+        int             i = 0;
+        int             index = 0;
+        int32_t         postop = 0;
+        int32_t         preop = 1;
+        int32_t         **txn_changelog = NULL;
+
+        txn_changelog = local->transaction.txn_changelog;
+        index = afr_index_for_transaction_type (local->transaction.type);
+        for (i = 0; i < priv->child_count; i++) {
+                postop = ntoh32 (local->pending[i][index]);
+                txn_changelog[i][index] = hton32 (postop + preop);
+        }
+}
 
+afr_xattrop_type_t
+afr_get_postop_xattrop_type (int32_t **pending, int optimized, int child,
+                             afr_transaction_type type)
+{
+        int                     index = 0;
+        afr_xattrop_type_t      op = LOCAL_LAST;
 
-/* {{{ pending */
+        index = afr_index_for_transaction_type (type);
+        if (optimized && !pending[child][index])
+                op = LOCAL_FIRST;
+        return op;
+}
 
-int32_t
-afr_changelog_post_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno, dict_t *xattr)
+void
+afr_set_postop_dict (afr_local_t *local, xlator_t *this, dict_t *xattr,
+                     int optimized, int child)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t *   local = NULL;
-	
-	int call_count = -1;
+        int32_t                 **txn_changelog = NULL;
+        int32_t                 **changelog = NULL;
+        afr_private_t           *priv = NULL;
+        int                     ret = 0;
+        afr_xattrop_type_t      op = LOCAL_LAST;
+
+        priv = this->private;
+        txn_changelog = local->transaction.txn_changelog;
+        op = afr_get_postop_xattrop_type (local->pending, optimized, child,
+                                          local->transaction.type);
+        if (optimized)
+                changelog = txn_changelog;
+        else
+                changelog = local->pending;
+        ret = afr_set_pending_dict (priv, xattr, changelog, child, op);
+        if (ret < 0)
+                gf_log (this->name, GF_LOG_INFO,
+                        "failed to set pending entry");
+}
 
-	priv  = this->private;
-	local = frame->local;
 
-	LOCK (&frame->lock);
-	{
-		call_count = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-
-	if (call_count == 0) {
-		if (afr_lock_server_count (priv, local->transaction.type) == 0) {
-			local->transaction.done (frame, this);
-		} else {
-			afr_unlock (frame, this);
-		}
-	}
+gf_boolean_t
+afr_txn_nothing_failed (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t *local = NULL;
+        int index = -1;
+        int i = 0;
+
+        local = frame->local;
+        priv = this->private;
 
-	return 0;	
+        index = afr_index_for_transaction_type (local->transaction.type);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->pending[i][index] == 0)
+                        return _gf_false;
+        }
+
+        return _gf_true;
 }
 
+static void
+afr_dir_fop_handle_all_fop_failures (call_frame_t *frame)
+{
+        xlator_t        *this = NULL;
+        afr_local_t     *local = NULL;
+        afr_private_t   *priv = NULL;
 
-int 
-afr_changelog_post_op (call_frame_t *frame, xlator_t *this)
+        this = frame->this;
+        local = frame->local;
+        priv = this->private;
+
+        if ((local->transaction.type != AFR_ENTRY_TRANSACTION) &&
+            (local->transaction.type != AFR_ENTRY_RENAME_TRANSACTION))
+                return;
+
+        if (local->op_ret >= 0)
+                goto out;
+
+        __mark_all_success (local->pending, priv->child_count,
+                            local->transaction.type);
+out:
+        return;
+}
+
+static void
+afr_data_handle_quota_errors (call_frame_t *frame, xlator_t *this)
+{
+        int     i = 0;
+        afr_private_t *priv = NULL;
+        afr_local_t   *local = NULL;
+        gf_boolean_t  all_quota_failures = _gf_false;
+
+        local = frame->local;
+        priv  = this->private;
+        if (local->transaction.type != AFR_DATA_TRANSACTION)
+                return;
+        /*
+         * Idea is to not leave the file in FOOL-FOOL scenario in case on
+         * all the bricks data transaction failed with EDQUOT to avoid
+         * increasing un-necessary load of self-heals in the system.
+         */
+        all_quota_failures = _gf_true;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->transaction.pre_op[i] &&
+                    (local->child_errno[i] != EDQUOT)) {
+                        all_quota_failures = _gf_false;
+                        break;
+                }
+        }
+        if (all_quota_failures)
+                __mark_all_success (local->pending, priv->child_count,
+                                    local->transaction.type);
+}
+
+int
+afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv = this->private;
+        afr_private_t * priv = this->private;
+        afr_internal_lock_t *int_lock = NULL;
+        int i          = 0;
+        int call_count = 0;
+
+        afr_local_t *  local = NULL;
+        afr_fd_ctx_t  *fdctx = NULL;
+        dict_t        **xattr = NULL;
+        int            piggyback = 0;
+        int            nothing_failed = 1;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        __mark_non_participant_children (local->pending, priv->child_count,
+                                         local->transaction.pre_op,
+                                         local->transaction.type);
+
+        afr_data_handle_quota_errors (frame, this);
+        afr_dir_fop_handle_all_fop_failures (frame);
+
+        if (local->fd)
+                afr_transaction_rm_stale_children (frame, this,
+                                                   local->fd->inode,
+                                                   local->transaction.type);
+
+        xattr = alloca (priv->child_count * sizeof (*xattr));
+        memset (xattr, 0, (priv->child_count * sizeof (*xattr)));
+        for (i = 0; i < priv->child_count; i++) {
+                xattr[i] = dict_new ();
+        }
 
-	int ret        = 0;
-	int i          = 0;				
-	int call_count = 0;
-	
-	afr_local_t *  local = NULL;	
-	dict_t *       xattr = dict_ref (get_new_dict ());
+        call_count = afr_changelog_post_op_call_count (local->transaction.type,
+                                                       local->transaction.pre_op,
+                                                       priv->child_count);
+        local->call_count = call_count;
 
-	local = frame->local;
+        if (local->fd)
+                fdctx = afr_fd_ctx_get (local->fd, this);
 
-	__mark_all_success (local->pending_array, priv->child_count);
-	__mark_down_children (local->pending_array, priv->child_count, local->child_up);
+        if (call_count == 0) {
+                /* no child is up */
+                int_lock->lock_cbk = local->transaction.done;
+                afr_unlock (frame, this);
+                goto out;
+        }
 
-	call_count = afr_up_children_count (priv->child_count, local->child_up); 
+        nothing_failed = afr_txn_nothing_failed (frame, this);
+
+        afr_compute_txn_changelog (local , priv);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (!local->transaction.pre_op[i])
+                        continue;
+
+                if (local->transaction.type != AFR_DATA_TRANSACTION)
+                        afr_set_postop_dict (local, this, xattr[i],
+                                             local->optimistic_change_log, i);
+                switch (local->transaction.type) {
+                case AFR_DATA_TRANSACTION:
+                {
+                        if (!fdctx) {
+                                afr_set_postop_dict (local, this, xattr[i],
+                                                     0, i);
+                                STACK_WIND (frame, afr_changelog_post_op_cbk,
+                                            priv->children[i],
+                                            priv->children[i]->fops->xattrop,
+                                            &local->loc,
+                                            GF_XATTROP_ADD_ARRAY, xattr[i],
+                                            NULL);
+                                break;
+                        }
+
+                        /* local->transaction.postop_piggybacked[] was
+                           precomputed in is_piggyback_postop() when called from
+                           afr_changelog_post_op_safe()
+                        */
+
+                        piggyback = 0;
+                        if (local->transaction.postop_piggybacked[i])
+                                piggyback = 1;
+
+                        afr_set_postop_dict (local, this, xattr[i],
+                                             piggyback, i);
+
+                        if (nothing_failed && piggyback) {
+                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
+                                                           this, 1, 0, xattr[i], NULL);
+                        } else {
+                                STACK_WIND_COOKIE (frame,
+                                                   afr_changelog_post_op_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[i],
+                                                   priv->children[i]->fops->fxattrop,
+                                                   local->fd,
+                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
+                                                   NULL);
+                        }
+                }
+                break;
+                case AFR_METADATA_TRANSACTION:
+                {
+                        if (nothing_failed && local->optimistic_change_log) {
+                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
+                                                           this, 1, 0, xattr[i],
+                                                           NULL);
+                                break;
+                        }
+
+                        if (local->fd)
+                                STACK_WIND (frame, afr_changelog_post_op_cbk,
+                                            priv->children[i],
+                                            priv->children[i]->fops->fxattrop,
+                                            local->fd,
+                                            GF_XATTROP_ADD_ARRAY, xattr[i],
+                                            NULL);
+                        else
+                                STACK_WIND (frame, afr_changelog_post_op_cbk,
+                                            priv->children[i],
+                                            priv->children[i]->fops->xattrop,
+                                            &local->loc,
+                                            GF_XATTROP_ADD_ARRAY, xattr[i],
+                                            NULL);
+                }
+                break;
+
+                case AFR_ENTRY_RENAME_TRANSACTION:
+                {
+                        if (nothing_failed && local->optimistic_change_log) {
+                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
+                                                           this, 1, 0, xattr[i],
+                                                           NULL);
+                        } else {
+                                STACK_WIND_COOKIE (frame, afr_changelog_post_op_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[i],
+                                                   priv->children[i]->fops->xattrop,
+                                                   &local->transaction.new_parent_loc,
+                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
+                                                   NULL);
+                        }
+                        call_count--;
+                }
 
-	if (local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
-		call_count *= 2;
-	}
+                /*
+                  set it again because previous stack_wind
+                  might have already returned (think of case
+                  where subvolume is posix) and would have
+                  used the dict as placeholder for return
+                  value
+                */
+
+                afr_set_postop_dict (local, this, xattr[i],
+                                     local->optimistic_change_log, i);
+
+                /* fall through */
+
+                case AFR_ENTRY_TRANSACTION:
+                {
+                        if (nothing_failed && local->optimistic_change_log) {
+                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
+                                                           this, 1, 0, xattr[i],
+                                                           NULL);
+                                break;
+                        }
+
+                        if (local->fd)
+                                STACK_WIND (frame, afr_changelog_post_op_cbk,
+                                            priv->children[i],
+                                            priv->children[i]->fops->fxattrop,
+                                            local->fd,
+                                            GF_XATTROP_ADD_ARRAY, xattr[i],
+                                            NULL);
+                        else
+                                STACK_WIND (frame, afr_changelog_post_op_cbk,
+                                            priv->children[i],
+                                            priv->children[i]->fops->xattrop,
+                                            &local->transaction.parent_loc,
+                                            GF_XATTROP_ADD_ARRAY, xattr[i],
+                                            NULL);
+                }
+                break;
+                }
 
-	local->call_count = call_count;		
+                if (!--call_count)
+                        break;
+        }
 
-	if (call_count == 0) {
-		/* no child is up */
-		dict_unref (xattr);
-		afr_unlock (frame, this);
-		return 0;
-	}
+out:
+        for (i = 0; i < priv->child_count; i++) {
+                dict_unref (xattr[i]);
+        }
 
-	for (i = 0; i < priv->child_count; i++) {					
-		if (local->child_up[i]) {
-			ret = dict_set_static_bin (xattr, local->transaction.pending, 
-						   local->pending_array, 
-						   priv->child_count * sizeof (int32_t));
-			if (ret < 0)
-				gf_log (this->name, GF_LOG_ERROR, 
-					"failed to set pending entry");
-
-
-			switch (local->transaction.type) {
-			case AFR_DATA_TRANSACTION:
-			case AFR_METADATA_TRANSACTION:
-			case AFR_FLUSH_TRANSACTION:
-			{
-				if (local->fd)
-					STACK_WIND (frame, afr_changelog_post_op_cbk,
-						    priv->children[i], 
-						    priv->children[i]->fops->fxattrop,
-						    local->fd, 
-						    GF_XATTROP_ADD_ARRAY, xattr);
-				else 
-					STACK_WIND (frame, afr_changelog_post_op_cbk,
-						    priv->children[i], 
-						    priv->children[i]->fops->xattrop,
-						    &local->loc, 
-						    GF_XATTROP_ADD_ARRAY, xattr);
-			}
-			break;
-
-			case AFR_ENTRY_RENAME_TRANSACTION:
-			{
-				STACK_WIND_COOKIE (frame, afr_changelog_post_op_cbk,
-						   (void *) (long) i,
-						   priv->children[i],
-						   priv->children[i]->fops->xattrop,
-						   &local->transaction.new_parent_loc,
-						   GF_XATTROP_ADD_ARRAY, xattr);
-				
-				call_count--;
-			}
-
-			/* 
-			   set it again because previous stack_wind
-			   might have already returned (think of case
-			   where subvolume is posix) and would have
-			   used the dict as placeholder for return
-			   value
-			*/
-			ret = dict_set_static_bin (xattr, local->transaction.pending, 
-						   local->pending_array, 
-						   priv->child_count * sizeof (int32_t));
-			if (ret < 0)
-				gf_log (this->name, GF_LOG_ERROR, 
-					"failed to set pending entry");
-
-			/* fall through */
-
-			case AFR_ENTRY_TRANSACTION:
-			{
-				if (local->fd)
-					STACK_WIND (frame, afr_changelog_post_op_cbk,
-						    priv->children[i], 
-						    priv->children[i]->fops->fxattrop,
-						    local->fd, 
-						    GF_XATTROP_ADD_ARRAY, xattr);
-				else 
-					STACK_WIND (frame, afr_changelog_post_op_cbk,
-						    priv->children[i], 
-						    priv->children[i]->fops->xattrop,
-						    &local->transaction.parent_loc, 
-						    GF_XATTROP_ADD_ARRAY, xattr);
-			}
-			break;
-			}
-
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	dict_unref (xattr);
-	return 0;
+        return 0;
 }
 
 
 int32_t
 afr_changelog_pre_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			      int32_t op_ret, int32_t op_errno, dict_t *xattr)
+                          int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                          dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = this->private;
-	loc_t       *   loc   = NULL;
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = this->private;
+        int call_count  = -1;
+        int child_index = (long) cookie;
 
-	int call_count  = -1;
-	int child_index = (long) cookie;
+        local = frame->local;
 
-	local = frame->local;
-	loc   = &local->loc;
+        LOCK (&frame->lock);
+        {
+                switch (op_ret) {
+                case 0:
+                        __mark_pre_op_done_on_fd (frame, this, child_index);
+                        //fallthrough we need to mark the pre_op
+                case 1:
+                        local->transaction.pre_op[child_index] = 1;
+                        /* special op_ret for piggyback */
+                        break;
+                case -1:
+                        if (op_errno == ENOTSUP) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "xattrop not supported by %s",
+                                        priv->children[child_index]->name);
+                                local->op_ret = -1;
+
+                        } else if (!child_went_down (op_ret, op_errno)) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "xattrop failed on child %s: %s",
+                                        priv->children[child_index]->name,
+                                        strerror (op_errno));
+                        }
+                        local->op_errno = op_errno;
+                        break;
+                }
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->child_up[child_index] = 0;
-			
-			if (op_errno == ENOTSUP) {
-				gf_log (this->name, GF_LOG_ERROR,
-					"xattrop not supported by %s",
-					priv->children[child_index]->name);
-				local->op_ret = -1;
-			} else if (!child_went_down (op_ret, op_errno)) {
-				gf_log (this->name, GF_LOG_ERROR,
-					"xattrop failed on child %s: %s",
-					priv->children[child_index]->name, 
-					strerror (op_errno));
-			}
-			local->op_errno = op_errno;
-		}
-
-		call_count = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-
-	if (call_count == 0) {
-		if ((local->op_ret == -1) && 
-		    (local->op_errno == ENOTSUP)) {
-			local->transaction.resume (frame, this);
-		} else {
-			local->transaction.fop (frame, this);
-		}
-	}
+                call_count = --local->call_count;
+        }
+        UNLOCK (&frame->lock);
+
+        if (call_count == 0) {
+                if ((local->op_ret == -1) &&
+                    (local->op_errno == ENOTSUP)) {
+                        local->transaction.resume (frame, this);
+                } else {
+                        afr_transaction_perform_fop (frame, this);
+                }
+        }
 
-	return 0;	
+        return 0;
 }
 
-
-int 
+int
 afr_changelog_pre_op (call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv = this->private;
+        afr_private_t * priv = this->private;
+        int i = 0;
+        int ret = 0;
+        int call_count = 0;
+        dict_t **xattr = NULL;
+        afr_fd_ctx_t *fdctx = NULL;
+        afr_local_t *local = NULL;
+        int          piggyback = 0;
+        afr_internal_lock_t *int_lock = NULL;
+        unsigned char       *locked_nodes = NULL;
+
+        local = frame->local;
+        int_lock = &local->internal_lock;
+
+        xattr = alloca (priv->child_count * sizeof (*xattr));
+        memset (xattr, 0, (priv->child_count * sizeof (*xattr)));
+
+        for (i = 0; i < priv->child_count; i++) {
+                xattr[i] = dict_new ();
+        }
 
-	int i = 0;				
-	int ret = 0;
-	int call_count = 0;		     
-	dict_t *xattr = NULL;
+        call_count = afr_changelog_pre_op_call_count (local->transaction.type,
+                                                      int_lock,
+                                                      priv->child_count);
+        if (call_count == 0) {
+                local->internal_lock.lock_cbk =
+                        local->transaction.done;
+                afr_unlock (frame, this);
+                goto out;
+        }
 
-	afr_local_t *local = NULL;
+        local->call_count = call_count;
+
+        __mark_all_pending (local->pending, priv->child_count,
+                            local->transaction.type);
+
+        if (local->fd)
+                fdctx = afr_fd_ctx_get (local->fd, this);
+
+        locked_nodes = afr_locked_nodes_get (local->transaction.type, int_lock);
+        for (i = 0; i < priv->child_count; i++) {
+                if (!locked_nodes[i])
+                        continue;
+                ret = afr_set_pending_dict (priv, xattr[i], local->pending,
+                                            i, LOCAL_FIRST);
+
+                if (ret < 0)
+                        gf_log (this->name, GF_LOG_INFO,
+                                "failed to set pending entry");
+
+
+                switch (local->transaction.type) {
+                case AFR_DATA_TRANSACTION:
+                {
+                        if (!fdctx) {
+                                STACK_WIND_COOKIE (frame,
+                                                   afr_changelog_pre_op_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[i],
+                                                   priv->children[i]->fops->xattrop,
+                                                   &(local->loc),
+                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
+                                                   NULL);
+                                break;
+                        }
+
+                        LOCK (&local->fd->lock);
+                        {
+                                piggyback = 0;
+                                if (fdctx->pre_op_done[i]) {
+                                        fdctx->pre_op_piggyback[i]++;
+                                        piggyback = 1;
+                                        fdctx->hit++;
+                                } else {
+                                        fdctx->miss++;
+                                }
+                        }
+                        UNLOCK (&local->fd->lock);
+
+                        afr_set_delayed_post_op (frame, this);
+
+                        if (piggyback)
+                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
+                                                          this, 1, 0, xattr[i],
+                                                          NULL);
+                        else
+                                STACK_WIND_COOKIE (frame,
+                                                   afr_changelog_pre_op_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[i],
+                                                   priv->children[i]->fops->fxattrop,
+                                                   local->fd,
+                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
+                                                   NULL);
+                }
+                break;
+                case AFR_METADATA_TRANSACTION:
+                {
+                        if (local->optimistic_change_log) {
+                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
+                                                          this, 1, 0, xattr[i],
+                                                          NULL);
+                                break;
+                        }
+
+                        if (local->fd)
+                                STACK_WIND_COOKIE (frame,
+                                                   afr_changelog_pre_op_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[i],
+                                                   priv->children[i]->fops->fxattrop,
+                                                   local->fd,
+                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
+                                                   NULL);
+                        else
+                                STACK_WIND_COOKIE (frame,
+                                                   afr_changelog_pre_op_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[i],
+                                                   priv->children[i]->fops->xattrop,
+                                                   &(local->loc),
+                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
+                                                   NULL);
+                }
+                break;
+
+                case AFR_ENTRY_RENAME_TRANSACTION:
+                {
+                        if (local->optimistic_change_log) {
+                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
+                                                          this, 1, 0, xattr[i],
+                                                          NULL);
+                        } else {
+                                STACK_WIND_COOKIE (frame,
+                                                   afr_changelog_pre_op_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[i],
+                                                   priv->children[i]->fops->xattrop,
+                                                   &local->transaction.new_parent_loc,
+                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
+                                                   NULL);
+                        }
+
+                        call_count--;
+                }
 
-	local = frame->local;
-	xattr = get_new_dict ();
-	dict_ref (xattr);
 
-	call_count = afr_up_children_count (priv->child_count, 
-					    local->child_up); 
+                /*
+                  set it again because previous stack_wind
+                  might have already returned (think of case
+                  where subvolume is posix) and would have
+                  used the dict as placeholder for return
+                  value
+                */
+
+                ret = afr_set_pending_dict (priv, xattr[i], local->pending,
+                                            i, LOCAL_FIRST);
+
+                if (ret < 0)
+                        gf_log (this->name, GF_LOG_INFO,
+                                "failed to set pending entry");
+
+                /* fall through */
+
+                case AFR_ENTRY_TRANSACTION:
+                {
+                        if (local->optimistic_change_log) {
+                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
+                                                          this, 1, 0, xattr[i],
+                                                          NULL);
+                                break;
+                        }
+
+                        if (local->fd)
+                                STACK_WIND_COOKIE (frame,
+                                                   afr_changelog_pre_op_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[i],
+                                                   priv->children[i]->fops->fxattrop,
+                                                   local->fd,
+                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
+                                                   NULL);
+                        else
+                                STACK_WIND_COOKIE (frame,
+                                                   afr_changelog_pre_op_cbk,
+                                                   (void *) (long) i,
+                                                   priv->children[i],
+                                                   priv->children[i]->fops->xattrop,
+                                                   &local->transaction.parent_loc,
+                                                   GF_XATTROP_ADD_ARRAY, xattr[i],
+                                                   NULL);
+                }
+                break;
+                }
 
-	if (local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
-		call_count *= 2;
-	}
+                if (!--call_count)
+                        break;
+        }
+out:
+        for (i = 0; i < priv->child_count; i++) {
+                dict_unref (xattr[i]);
+        }
 
-	if (call_count == 0) {
-		/* no child is up */
-		dict_unref (xattr);
-		afr_unlock (frame, this);
-		return 0;
-	}
+        return 0;
+}
 
-	local->call_count = call_count;		
-
-	__mark_all_pending (local->pending_array, priv->child_count);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			ret = dict_set_static_bin (xattr, 
-						   local->transaction.pending, 
-						   local->pending_array, 
-						   (priv->child_count * 
-						    sizeof (int32_t)));
-			if (ret < 0)
-				gf_log (this->name, GF_LOG_ERROR, 
-					"failed to set pending entry");
-
-
-			switch (local->transaction.type) {
-			case AFR_DATA_TRANSACTION:
-			case AFR_METADATA_TRANSACTION:
-			case AFR_FLUSH_TRANSACTION:
-			{
-				if (local->fd)
-					STACK_WIND_COOKIE (frame, 
-							   afr_changelog_pre_op_cbk,
-							   (void *) (long) i,
-							   priv->children[i], 
-							   priv->children[i]->fops->fxattrop,
-							   local->fd,
-							   GF_XATTROP_ADD_ARRAY, xattr);
-				else
-					STACK_WIND_COOKIE (frame, 
-							   afr_changelog_pre_op_cbk,
-							   (void *) (long) i,
-							   priv->children[i], 
-							   priv->children[i]->fops->xattrop,
-							   &(local->loc), 
-							   GF_XATTROP_ADD_ARRAY, xattr);
-			}
-			break;
-				
-			case AFR_ENTRY_RENAME_TRANSACTION: 
-			{
-				STACK_WIND_COOKIE (frame, 
-						   afr_changelog_pre_op_cbk,
-						   (void *) (long) i,
-						   priv->children[i], 
-						   priv->children[i]->fops->xattrop,
-						   &local->transaction.new_parent_loc, 
-						   GF_XATTROP_ADD_ARRAY, xattr);
-
-				call_count--;
-			}
-
-
-			/* 
-			   set it again because previous stack_wind
-			   might have already returned (think of case
-			   where subvolume is posix) and would have
-			   used the dict as placeholder for return
-			   value
-			*/
-
-			ret = dict_set_static_bin (xattr, local->transaction.pending, 
-						   local->pending_array, 
-						   priv->child_count * sizeof (int32_t));
-			if (ret < 0)
-				gf_log (this->name, GF_LOG_ERROR, 
-					"failed to set pending entry");
-
-			/* fall through */
-				
-			case AFR_ENTRY_TRANSACTION:
-			{
-				if (local->fd)
-					STACK_WIND_COOKIE (frame, 
-							   afr_changelog_pre_op_cbk,
-							   (void *) (long) i,
-							   priv->children[i], 
-							   priv->children[i]->fops->fxattrop,
-							   local->fd, 
-							   GF_XATTROP_ADD_ARRAY, xattr);
-				else
-					STACK_WIND_COOKIE (frame, 
-							   afr_changelog_pre_op_cbk,
-							   (void *) (long) i,
-							   priv->children[i], 
-							   priv->children[i]->fops->xattrop,
-							   &local->transaction.parent_loc, 
-							   GF_XATTROP_ADD_ARRAY, xattr);
-			}
-
-			break;
-			}
-
-			if (!--call_count)
-				break;
-		}
-	}
 
-	dict_unref (xattr);
-	return 0;
+int
+afr_post_blocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        if (int_lock->lock_op_ret < 0) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Blocking inodelks failed.");
+                local->transaction.done (frame, this);
+        } else {
+
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Blocking inodelks done. Proceeding to FOP");
+                afr_internal_lock_finish (frame, this);
+        }
+
+        return 0;
 }
 
+
+int
+afr_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        /* Initiate blocking locks if non-blocking has failed */
+        if (int_lock->lock_op_ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Non blocking inodelks failed. Proceeding to blocking");
+                int_lock->lock_cbk = afr_post_blocking_inodelk_cbk;
+                afr_blocking_lock (frame, this);
+        } else {
+
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Non blocking inodelks done. Proceeding to FOP");
+                afr_internal_lock_finish (frame, this);
+        }
+
+        return 0;
+}
+
+
+int
+afr_post_blocking_entrylk_cbk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        if (int_lock->lock_op_ret < 0) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Blocking entrylks failed.");
+                local->transaction.done (frame, this);
+        } else {
+
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Blocking entrylks done. Proceeding to FOP");
+                afr_internal_lock_finish (frame, this);
+        }
+
+        return 0;
+}
+
+
+int
+afr_post_nonblocking_entrylk_cbk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+
+        local = frame->local;
+        int_lock = &local->internal_lock;
+
+        /* Initiate blocking locks if non-blocking has failed */
+        if (int_lock->lock_op_ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Non blocking entrylks failed. Proceeding to blocking");
+                int_lock->lock_cbk = afr_post_blocking_entrylk_cbk;
+                afr_blocking_lock (frame, this);
+        } else {
+
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Non blocking entrylks done. Proceeding to FOP");
+                afr_internal_lock_finish (frame, this);
+        }
+
+        return 0;
+}
+
+
+int
+afr_post_blocking_rename_cbk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        if (int_lock->lock_op_ret < 0) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Blocking entrylks failed.");
+                local->transaction.done (frame, this);
+        } else {
+
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Blocking entrylks done. Proceeding to FOP");
+                afr_internal_lock_finish (frame, this);
+        }
+        return 0;
+}
+
+
+int
+afr_post_lower_unlock_cbk (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        GF_ASSERT (!int_lock->higher_locked);
+
+        int_lock->lock_cbk = afr_post_blocking_rename_cbk;
+        afr_blocking_lock (frame, this);
+
+        return 0;
+}
+
+
+int
+afr_set_transaction_flock (afr_local_t *local)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_inodelk_t       *inodelk  = NULL;
+
+        int_lock = &local->internal_lock;
+        inodelk = afr_get_inodelk (int_lock, int_lock->domain);
+
+        inodelk->flock.l_len   = local->transaction.len;
+        inodelk->flock.l_start = local->transaction.start;
+        inodelk->flock.l_type  = F_WRLCK;
+
+        return 0;
+}
+
+int
+afr_lock_rec (call_frame_t *frame, xlator_t *this)
+{
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+
+        int_lock->transaction_lk_type = AFR_TRANSACTION_LK;
+        int_lock->domain = this->name;
+
+        switch (local->transaction.type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+                afr_set_transaction_flock (local);
+
+                int_lock->lock_cbk = afr_post_nonblocking_inodelk_cbk;
+
+                afr_nonblocking_inodelk (frame, this);
+                break;
+
+        case AFR_ENTRY_RENAME_TRANSACTION:
+
+                int_lock->lock_cbk = afr_post_nonblocking_entrylk_cbk;
+                afr_nonblocking_entrylk (frame, this);
+                break;
+
+        case AFR_ENTRY_TRANSACTION:
+                int_lock->lk_basename = local->transaction.basename;
+                if (&local->transaction.parent_loc)
+                        int_lock->lk_loc = &local->transaction.parent_loc;
+                else
+                        GF_ASSERT (local->fd);
+
+                int_lock->lock_cbk = afr_post_nonblocking_entrylk_cbk;
+                afr_nonblocking_entrylk (frame, this);
+                break;
+        }
+
+        return 0;
+}
+
+
+int
+afr_lock (call_frame_t *frame, xlator_t *this)
+{
+        afr_set_lock_number (frame, this);
+
+        return afr_lock_rec (frame, this);
+}
+
+
 /* }}} */
 
-/* {{{ lock */
+int
+afr_internal_lock_finish (call_frame_t *frame, xlator_t *this)
+{
+        if (__fop_changelog_needed (frame, this)) {
+                afr_changelog_pre_op (frame, this);
+        } else {
+                afr_transaction_perform_fop (frame, this);
+        }
 
-static
-int afr_lock_rec (call_frame_t *frame, xlator_t *this, int child_index);
+        return 0;
+}
 
-int32_t
-afr_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	      int32_t op_ret, int32_t op_errno)
+
+void
+afr_set_delayed_post_op (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv = NULL;
-	int done = 0;
-	int child_index = (long) cookie;
+        afr_local_t    *local = NULL;
+        afr_private_t  *priv = NULL;
 
-	int call_count = 0;
+        /* call this function from any of the related optimizations
+           which benefit from delaying post op are enabled, namely:
 
-	local = frame->local;
-	priv  = this->private;
+           - changelog piggybacking
+           - eager locking
+        */
 
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
-			/* wait for the other lock to return */
-			call_count = --local->call_count;
-		}
-
-		if (op_ret == -1) {
-			if (op_errno == ENOSYS) {
-				/* return ENOTSUP */
-				gf_log (this->name, GF_LOG_ERROR,
-					"subvolume does not support locking. "
-					"please load features/posix-locks xlator on server");
-				local->op_ret   = op_ret;
-				done = 1;
-			}
-
-			local->child_up[child_index] = 0;
-			local->op_errno = op_errno;
-		}
-	}
-	UNLOCK (&frame->lock);
-	
-	if (call_count == 0) {
-		if ((local->op_ret == -1) &&
-		    (local->op_errno == ENOSYS)) {
-			afr_unlock (frame, this);
-		} else {
-			local->transaction.locked_nodes[child_index] = 1;
-			local->transaction.lock_count++;
-			afr_lock_rec (frame, this, child_index + 1);
-		}
-	}
+        priv = this->private;
+        if (!priv)
+                return;
+
+        if (!priv->post_op_delay_secs)
+                return;
+
+        local = frame->local;
+        if (!local->transaction.eager_lock_on)
+                return;
 
-	return 0;
+        if (!local)
+                return;
+
+        if (!local->fd)
+                return;
+
+        if (local->op == GF_FOP_WRITE)
+                local->delayed_post_op = _gf_true;
+}
+
+gf_boolean_t
+afr_are_multiple_fds_opened (inode_t *inode, xlator_t *this)
+{
+        afr_inode_ctx_t *ictx = NULL;
+
+        if (!inode) {
+                /* If false is returned, it may keep on taking eager-lock
+                 * which may lead to starvation, so return true to avoid that.
+                 */
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid inode");
+                return _gf_true;
+        }
+        /* Lets say mount1 has eager-lock(full-lock) and after the eager-lock
+         * is taken mount2 opened the same file, it won't be able to
+         * perform any data operations until mount1 releases eager-lock.
+         * To avoid such scenario do not enable eager-lock for this transaction
+         * if open-fd-count is > 1
+         */
+
+        ictx = afr_inode_ctx_get (inode, this);
+        if (!ictx)
+                return _gf_true;
+
+        if (ictx->open_fd_count > 1)
+                return _gf_true;
+
+        return _gf_false;
 }
 
+gf_boolean_t
+afr_any_fops_failed (afr_local_t *local, afr_private_t *priv)
+{
+        if (local->success_count != priv->child_count)
+                return _gf_true;
+        return _gf_false;
+}
 
-static loc_t *
-lower_path (loc_t *l1, const char *b1, loc_t *l2, const char *b2)
+gf_boolean_t
+is_afr_delayed_changelog_post_op_needed (call_frame_t *frame, xlator_t *this)
 {
-	int ret = 0;
+        afr_local_t      *local = NULL;
+        gf_boolean_t      res = _gf_false;
+        afr_private_t    *priv  = NULL;
+
+        priv  = this->private;
+
+        local = frame->local;
+        if (!local)
+                goto out;
+
+        if (!local->delayed_post_op)
+                goto out;
 
-	ret = strcmp (l1->path, l2->path);
-	
-	if (ret == 0) 
-		ret = strcmp (b1, b2);
+        //Mark pending changelog ASAP
+        if (afr_any_fops_failed (local, priv))
+                goto out;
 
-	if (ret <= 0)
-		return l1;
-	else
-		return l2;
+        if (local->fd && afr_are_multiple_fds_opened (local->fd->inode, this))
+                goto out;
+
+        res = _gf_true;
+out:
+        return res;
 }
 
 
-static
-int afr_lock_rec (call_frame_t *frame, xlator_t *this, int child_index)
+void
+afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
+                               call_stub_t *stub);
+
+void
+afr_delayed_changelog_wake_up_cbk (void *data)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        fd_t           *fd = NULL;
 
-	struct flock flock;
+        fd = data;
 
-	loc_t * lower  = NULL;
-	loc_t * higher = NULL;
+        afr_delayed_changelog_wake_up (THIS, fd);
+}
 
-	const char *lower_name  = NULL;
-	const char *higher_name = NULL;
 
-	local = frame->local;
-	priv  = this->private;
+/*
+  Check if the frame is destined to get optimized away
+  with changelog piggybacking
+*/
+static gf_boolean_t
+is_piggyback_post_op (call_frame_t *frame, fd_t *fd)
+{
+        afr_fd_ctx_t *fdctx = NULL;
+        afr_local_t *local = NULL;
+        gf_boolean_t piggyback = _gf_true;
+        afr_private_t *priv = NULL;
+        int i = 0;
+
+        priv = frame->this->private;
+        local = frame->local;
+        fdctx = afr_fd_ctx_get (fd, frame->this);
+
+        LOCK(&fd->lock);
+        {
+                piggyback = _gf_true;
+
+                for (i = 0; i < priv->child_count; i++) {
+                        if (!local->transaction.pre_op[i])
+                                continue;
+                        if (fdctx->pre_op_piggyback[i]) {
+                                fdctx->pre_op_piggyback[i]--;
+                                local->transaction.postop_piggybacked[i] = 1;
+                        } else {
+                                /* For at least _one_ subvolume we cannot
+                                   piggyback on the changelog, and have to
+                                   perform a hard POST-OP and therefore fsync
+                                   if necesssary
+                                */
+                                piggyback = _gf_false;
+                                GF_ASSERT (fdctx->pre_op_done[i]);
+                                fdctx->pre_op_done[i]--;
+                        }
+                }
+        }
+        UNLOCK(&fd->lock);
 
-	flock.l_start = local->transaction.start;
-	flock.l_len   = local->transaction.len;
-	flock.l_type  = F_WRLCK;
+        if (!afr_txn_nothing_failed (frame, frame->this)) {
+                /* something failed in this transaction,
+                   we will be performing a hard post-op
+                */
+                return _gf_false;
+        }
 
-	/* skip over children that are down */
-	while ((child_index < priv->child_count)
-	       && !local->child_up[child_index])
-		child_index++;
+        return piggyback;
+}
 
-	if ((child_index == priv->child_count) &&
-	    local->transaction.lock_count == 0) {
 
-		gf_log (this->name, GF_LOG_DEBUG,
-			"unable to lock on even one child");
+/* SET operation */
+int
+afr_fd_report_unstable_write (xlator_t *this, fd_t *fd)
+{
+        afr_fd_ctx_t *fdctx = NULL;
 
-		local->op_ret   = -1;
-		local->op_errno = EAGAIN;
+        fdctx = afr_fd_ctx_get (fd, this);
 
-		local->transaction.done (frame, this);
-		
-		return 0;
+        LOCK(&fd->lock);
+        {
+                fdctx->witnessed_unstable_write = _gf_true;
+        }
+        UNLOCK(&fd->lock);
 
-	}
+        return 0;
+}
 
-	if ((child_index == priv->child_count) 
-	    || (local->transaction.lock_count == 
-		afr_lock_server_count (priv, local->transaction.type))) {
+/* TEST and CLEAR operation */
+gf_boolean_t
+afr_fd_has_witnessed_unstable_write (xlator_t *this, fd_t *fd)
+{
+        afr_fd_ctx_t *fdctx = NULL;
+        gf_boolean_t witness = _gf_false;
 
-		/* we're done locking */
+	fdctx = afr_fd_ctx_get (fd, this);
+        if (!fdctx)
+                return _gf_true;
 
-		if (__changelog_needed_pre_op (frame, this)) {
-			afr_changelog_pre_op (frame, this);
-		} else {
-			local->transaction.fop (frame, this);
-		}
+        LOCK(&fd->lock);
+        {
+                if (fdctx->witnessed_unstable_write) {
+                        witness = _gf_true;
+                        fdctx->witnessed_unstable_write = _gf_false;
+                }
+        }
+        UNLOCK (&fd->lock);
+
+        return witness;
+}
+
+
+int
+afr_changelog_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, struct iatt *pre,
+                         struct iatt *post, dict_t *xdata)
+{
+        afr_private_t *priv = NULL;
+        int child_index = (long) cookie;
+        int call_count = -1;
+        afr_local_t *local = NULL;
+
+        priv = this->private;
+        local = frame->local;
+
+        if (afr_fop_failed (op_ret, op_errno)) {
+                /* Failure of fsync() is as good as failure of previous
+                   write(). So treat it like one.
+                */
+                gf_log (this->name, GF_LOG_WARNING,
+                        "fsync(%s) failed on subvolume %s. Transaction was %s",
+                        uuid_utoa (local->fd->inode->gfid),
+                        priv->children[child_index]->name,
+                        gf_fop_list[local->op]);
+
+                afr_transaction_fop_failed (frame, this, child_index);
+        }
+
+        call_count = afr_frame_return (frame);
+
+        if (call_count == 0)
+                afr_changelog_post_op_now (frame, this);
+
+        return 0;
+}
+
+
+int
+afr_changelog_fsync (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+        int i = 0;
+        int call_count = 0;
+        afr_private_t *priv = NULL;
+ 	dict_t *xdata = NULL;
+ 	GF_UNUSED int ret = -1;
+
+        local = frame->local;
+        priv = this->private;
+
+        call_count = afr_pre_op_done_children_count (local->transaction.pre_op,
+                                                     priv->child_count);
+
+        if (!call_count) {
+                /* will go straight to unlock */
+                afr_changelog_post_op_now (frame, this);
+                return 0;
+        }
+
+        local->call_count = call_count;
+
+	xdata = dict_new();
+	if (xdata)
+		ret = dict_set_int32 (xdata, "batch-fsync", 1);
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (!local->transaction.pre_op[i])
+                        continue;
+
+                STACK_WIND_COOKIE (frame, afr_changelog_fsync_cbk,
+                                (void *) (long) i, priv->children[i],
+                                priv->children[i]->fops->fsync, local->fd,
+                                1, xdata);
+                if (!--call_count)
+                        break;
+        }
+
+	if (xdata)
+		dict_unref (xdata);
+
+        return 0;
+}
+
+
+        int
+afr_changelog_post_op_safe (call_frame_t *frame, xlator_t *this)
+{
+	afr_local_t    *local = NULL;
+        afr_private_t  *priv = NULL;
+
+	local = frame->local;
+        priv = this->private;
+
+        if (!local->fd || local->transaction.type != AFR_DATA_TRANSACTION) {
+                afr_changelog_post_op_now (frame, this);
+                return 0;
+        }
+
+        if (is_piggyback_post_op (frame, local->fd)) {
+                /* just detected that this post-op is about to
+                   be optimized away as a new write() has
+                   already piggybacked on this frame's changelog.
+                   */
+                afr_changelog_post_op_now (frame, this);
+                return 0;
+        }
+
+        /* Calling afr_changelog_post_op_now() now will result in
+           issuing ->[f]xattrop().
+
+           Performing a hard POST-OP (->[f]xattrop() FOP) is a more
+           responsible operation that what it might appear on the surface.
+
+           The changelog of a file (in the xattr of the file on the server)
+           stores information (pending count) about the state of the file
+           on the OTHER server. This changelog is blindly trusted, and must
+           therefore be updated in such a way it remains trustworthy. This
+           implies that decrementing the pending count (essentially "clearing
+           the dirty flag") must be done STRICTLY after we are sure that the
+           operation on the other server has reached stable storage.
+
+           While the backend filesystem on that server will eventually flush
+           it to stable storage, we (being in userspace) have no mechanism
+           to get notified when the write became "stable".
+
+           This means we need take matter into our own hands and issue an
+           fsync() EVEN IF THE APPLICATION WAS PERFORMING UNSTABLE WRITES,
+           and get an acknowledgement for it. And we need to wait for the
+           fsync() acknowledgement before initiating the hard POST-OP.
+
+           However if the FD itself was opened in O_SYNC or O_DSYNC then
+           we are already guaranteed that the writes were made stable as
+           part of the FOP itself. The same holds true for NFS stable
+           writes which happen on an anonymous FD with O_DSYNC or O_SYNC
+           flag set in the writev() @flags param. For all other write types,
+           mark a flag in the fdctx whenever an unstable write is witnessed.
+           */
+
+        if (!afr_fd_has_witnessed_unstable_write (this, local->fd)) {
+                afr_changelog_post_op_now (frame, this);
+                return 0;
+        }
+
+        /* Check whether users want durability and perform fsync/post-op
+         * accordingly.
+         */
+        if (priv->ensure_durability) {
+                /* Time to fsync() */
+                afr_changelog_fsync (frame, this);
+        } else {
+                afr_changelog_post_op_now (frame, this);
+        }
+
+        return 0;
+}
 
-		return 0;
-	}
 
-	switch (local->transaction.type) {
-	case AFR_DATA_TRANSACTION:		
-	case AFR_METADATA_TRANSACTION:
-	case AFR_FLUSH_TRANSACTION:
-
-		if (local->fd) {
-			STACK_WIND_COOKIE (frame, afr_lock_cbk,
-					   (void *) (long) child_index,
-					   priv->children[child_index], 
-					   priv->children[child_index]->fops->finodelk,
-					   this->name, local->fd, 
-                                           F_SETLKW, &flock);
-			
-		} else {
-			STACK_WIND_COOKIE (frame, afr_lock_cbk,
-					   (void *) (long) child_index,
-					   priv->children[child_index], 
-					   priv->children[child_index]->fops->inodelk,
-					   this->name, &local->loc, 
-                                           F_SETLKW, &flock);
-		}
-		
-		break;
-		
-	case AFR_ENTRY_RENAME_TRANSACTION:
+void
+afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd,
+                               call_stub_t *stub)
+{
+	afr_fd_ctx_t      *fd_ctx = NULL;
+	call_frame_t      *prev_frame = NULL;
+	struct timespec    delta = {0, };
+	afr_private_t     *priv = NULL;
+	afr_local_t       *local = NULL;
+
+	priv = this->private;
+
+	fd_ctx = afr_fd_ctx_get (fd, this);
+	if (!fd_ctx)
+                goto out;
+
+	delta.tv_sec = priv->post_op_delay_secs;
+	delta.tv_nsec = 0;
+
+	pthread_mutex_lock (&fd_ctx->delay_lock);
 	{
-		local->call_count = 2;
-
-		lower = lower_path (&local->transaction.parent_loc, 
-				    local->transaction.basename,
-				    &local->transaction.new_parent_loc,
-				    local->transaction.new_basename);
-		
-		lower_name = (lower == &local->transaction.parent_loc ? 
-			      local->transaction.basename :
-			      local->transaction.new_basename);
-
-		higher = (lower == &local->transaction.parent_loc ? 
-			  &local->transaction.new_parent_loc :
-			  &local->transaction.parent_loc);
-
-		higher_name = (higher == &local->transaction.parent_loc ? 
-			       local->transaction.basename :
-			       local->transaction.new_basename);
-
-
-		/* TODO: these locks should be blocking */
-
-		STACK_WIND_COOKIE (frame, afr_lock_cbk,
-				   (void *) (long) child_index,
-				   priv->children[child_index], 
-				   priv->children[child_index]->fops->entrylk, 
-				   this->name, lower, lower_name,
-				   ENTRYLK_LOCK, ENTRYLK_WRLCK);
-
-		STACK_WIND_COOKIE (frame, afr_lock_cbk,
-				   (void *) (long) child_index,
-				   priv->children[child_index], 
-				   priv->children[child_index]->fops->entrylk, 
-				   this->name, higher, higher_name,
-				   ENTRYLK_LOCK, ENTRYLK_WRLCK);
-
-		break;
+		prev_frame = fd_ctx->delay_frame;
+		fd_ctx->delay_frame = NULL;
+		if (fd_ctx->delay_timer)
+			gf_timer_call_cancel (this->ctx, fd_ctx->delay_timer);
+		fd_ctx->delay_timer = NULL;
+		if (!frame)
+			goto unlock;
+		fd_ctx->delay_timer = gf_timer_call_after (this->ctx, delta,
+							   afr_delayed_changelog_wake_up_cbk,
+							   fd);
+		fd_ctx->delay_frame = frame;
 	}
-		
-	case AFR_ENTRY_TRANSACTION:
-		if (local->fd) {
-			STACK_WIND_COOKIE (frame, afr_lock_cbk,
-					   (void *) (long) child_index,	
-					   priv->children[child_index], 
-					   priv->children[child_index]->fops->fentrylk, 
-					   this->name, local->fd, 
-					   local->transaction.basename,
-					   ENTRYLK_LOCK, ENTRYLK_WRLCK);
-		} else {
-			STACK_WIND_COOKIE (frame, afr_lock_cbk,
-					   (void *) (long) child_index,	
-					   priv->children[child_index], 
-					   priv->children[child_index]->fops->entrylk, 
-					   this->name, 
-                                           &local->transaction.parent_loc, 
-					   local->transaction.basename,
-					   ENTRYLK_LOCK, ENTRYLK_WRLCK);
-		}
-
-		break;
+unlock:
+	pthread_mutex_unlock (&fd_ctx->delay_lock);
+
+out:
+	if (prev_frame) {
+		local = prev_frame->local;
+		local->transaction.resume_stub = stub;
+		afr_changelog_post_op_safe (prev_frame, this);
+	} else if (stub) {
+		call_resume (stub);
 	}
-
-	return 0;
 }
 
 
-int32_t afr_lock (call_frame_t *frame, xlator_t *this)
+void
+afr_changelog_post_op (call_frame_t *frame, xlator_t *this)
 {
-        frame->root->pid = (long) frame->root;
+        afr_local_t  *local = NULL;
 
-	return afr_lock_rec (frame, this, 0);
+        local = frame->local;
+
+        if (is_afr_delayed_changelog_post_op_needed (frame, this))
+                afr_delayed_changelog_post_op (this, frame, local->fd, NULL);
+        else
+                afr_changelog_post_op_safe (frame, this);
 }
 
 
-/* }}} */
 
-int32_t
+/* Wake up the sleeping/delayed post-op, and also register
+   a stub to have it resumed after this transaction
+   completely finishes.
+
+   The @stub gets saved in @local and gets resumed in
+   afr_local_cleanup()
+   */
+void
+afr_delayed_changelog_wake_resume (xlator_t *this, fd_t *fd, call_stub_t *stub)
+{
+        afr_delayed_changelog_post_op (this, NULL, fd, stub);
+}
+
+
+void
+afr_delayed_changelog_wake_up (xlator_t *this, fd_t *fd)
+{
+        afr_delayed_changelog_post_op (this, NULL, fd, NULL);
+}
+
+
+        int
 afr_transaction_resume (call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        afr_internal_lock_t *int_lock = NULL;
+        afr_local_t         *local    = NULL;
+        afr_private_t       *priv     = NULL;
+
+        local    = frame->local;
+        int_lock = &local->internal_lock;
+        priv     = this->private;
+
+        if (local->transaction.eager_lock_on) {
+                /* We don't need to retain "local" in the
+                   fd list anymore, writes to all subvols
+                   are finished by now */
+                LOCK (&local->fd->lock);
+                {
+                        list_del_init (&local->transaction.eager_locked);
+                }
+                UNLOCK (&local->fd->lock);
+        }
 
-	local = frame->local;
-	priv  = this->private;
-
-	if (__changelog_needed_post_op (frame, this)) {
-		afr_changelog_post_op (frame, this);
-	} else {
-		if (afr_lock_server_count (priv, local->transaction.type) == 0) {
-			local->transaction.done (frame, this);
-		} else {
-			afr_unlock (frame, this);
-		}
-	}
+        afr_restore_lk_owner (frame);
 
-	return 0;
+        if (__fop_changelog_needed (frame, this)) {
+                afr_changelog_post_op (frame, this);
+        } else {
+                if (afr_lock_server_count (priv, local->transaction.type) == 0) {
+                        local->transaction.done (frame, this);
+                } else {
+                        int_lock->lock_cbk = local->transaction.done;
+                        afr_unlock (frame, this);
+                }
+        }
+
+        return 0;
 }
 
 
 /**
- * afr_transaction_child_died - inform that a child died during an fop
+ * afr_transaction_fop_failed - inform that an fop failed
  */
 
 void
-afr_transaction_child_died (call_frame_t *frame, xlator_t *this, int child_index)
+afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this,
+                            int child_index)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = NULL;
 
-	local = frame->local;
-	priv  = this->private;
+        local = frame->local;
+        priv  = this->private;
 
-	__mark_child_dead (local->pending_array, priv->child_count, child_index);
+        __mark_child_dead (local->pending, priv->child_count,
+                        child_index, local->transaction.type);
 }
 
 
-int32_t
+
+        static gf_boolean_t
+afr_locals_overlap (afr_local_t *local1, afr_local_t *local2)
+{
+        uint64_t start1 = local1->transaction.start;
+        uint64_t start2 = local2->transaction.start;
+        uint64_t end1 = 0;
+        uint64_t end2 = 0;
+
+        if (local1->transaction.len)
+                end1 = start1 + local1->transaction.len - 1;
+        else
+                end1 = ULLONG_MAX;
+
+        if (local2->transaction.len)
+                end2 = start2 + local2->transaction.len - 1;
+        else
+                end2 = ULLONG_MAX;
+
+        return ((end1 >= start2) && (end2 >= start1));
+}
+
+void
+afr_transaction_eager_lock_init (afr_local_t *local, xlator_t *this)
+{
+        afr_private_t *priv = NULL;
+        afr_fd_ctx_t  *fdctx = NULL;
+        afr_local_t   *each = NULL;
+
+        priv = this->private;
+
+        if (!local->fd)
+                return;
+
+        if (local->transaction.type != AFR_DATA_TRANSACTION)
+                return;
+
+        if (!priv->eager_lock)
+                return;
+
+        fdctx = afr_fd_ctx_get (local->fd, this);
+        if (!fdctx)
+                return;
+
+        if (afr_are_multiple_fds_opened (local->fd->inode, this))
+                return;
+        /*
+         * Once full file lock is acquired in eager-lock phase, overlapping
+         * writes do not compete for inode-locks, instead are transferred to the
+         * next writes. Because of this overlapping writes are not ordered.
+         * This can cause inconsistencies in replication.
+         * Example:
+         * Two overlapping writes w1, w2 are sent in parallel on same fd
+         * in two threads t1, t2.
+         * Both threads can execute afr_writev_wind in the following manner.
+         * t1 winds w1 on brick-0
+         * t2 winds w2 on brick-0
+         * t2 winds w2 on brick-1
+         * t1 winds w1 on brick-1
+         *
+         * This check makes sure the locks are not transferred for
+         * overlapping writes.
+         */
+        LOCK (&local->fd->lock);
+        {
+                list_for_each_entry (each, &fdctx->eager_locked,
+                                     transaction.eager_locked) {
+                        if (afr_locals_overlap (each, local)) {
+                                local->transaction.eager_lock_on = _gf_false;
+                                goto unlock;
+                        }
+                }
+
+                local->transaction.eager_lock_on = _gf_true;
+                list_add_tail (&local->transaction.eager_locked,
+                               &fdctx->eager_locked);
+        }
+unlock:
+        UNLOCK (&local->fd->lock);
+}
+
+
+int
 afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+        afr_local_t *   local = NULL;
+        afr_private_t * priv  = NULL;
+        fd_t            *fd   = NULL;
+        int             ret   = -1;
 
-	local = frame->local;
-	priv  = this->private;
+        local = frame->local;
+        priv  = this->private;
 
-	afr_transaction_local_init (local, priv);
+        local->transaction.resume = afr_transaction_resume;
+        local->transaction.type   = type;
 
-	local->transaction.resume = afr_transaction_resume;
-	local->transaction.type   = type;
+        ret = afr_transaction_local_init (local, this);
+        if (ret < 0)
+            goto out;
 
-	if (afr_lock_server_count (priv, local->transaction.type) == 0) {
-		if (__changelog_needed_pre_op (frame, this)) {
-			afr_changelog_pre_op (frame, this);
-		} else {
-			local->transaction.fop (frame, this);
-		}
-	} else {
-		afr_lock (frame, this);
-	}
+        afr_transaction_eager_lock_init (local, this);
 
-	return 0;
+        if (local->fd && local->transaction.eager_lock_on)
+                afr_set_lk_owner (frame, this, local->fd);
+        else
+                afr_set_lk_owner (frame, this, frame->root);
+
+        if (!local->transaction.eager_lock_on && local->loc.inode) {
+                fd = fd_lookup (local->loc.inode, frame->root->pid);
+                if (fd == NULL)
+                        fd = fd_lookup_anonymous (local->loc.inode);
+
+                if (fd) {
+                        afr_delayed_changelog_wake_up (this, fd);
+                        fd_unref (fd);
+                }
+        }
+
+        if (afr_lock_server_count (priv, local->transaction.type) == 0) {
+                afr_internal_lock_finish (frame, this);
+        } else {
+                afr_lock (frame, this);
+        }
+        ret = 0;
+out:
+        return ret;
 }
diff --git a/xlators/cluster/afr/src/afr-transaction.h b/xlators/cluster/afr/src/afr-transaction.h
index 77d7a813c..fa626fd0d 100644
--- a/xlators/cluster/afr/src/afr-transaction.h
+++ b/xlators/cluster/afr/src/afr-transaction.h
@@ -1,36 +1,51 @@
 /*
-   Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __TRANSACTION_H__
 #define __TRANSACTION_H__
 
-#define AFR_METADATA_PENDING "trusted.glusterfs.afr.metadata-pending"
-
-#define AFR_DATA_PENDING "trusted.glusterfs.afr.data-pending"
-
-#define AFR_ENTRY_PENDING "trusted.glusterfs.afr.entry-pending"
+typedef enum {
+        LOCAL_FIRST = 1,
+        LOCAL_LAST = 2
+} afr_xattrop_type_t;
 
 void
-afr_transaction_child_died (call_frame_t *frame, xlator_t *this,
+afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this,
 			    int child_index);
 
+int
+afr_lock_server_count (afr_private_t *priv, afr_transaction_type type);
+
+afr_inodelk_t*
+afr_get_inodelk (afr_internal_lock_t *int_lock, char *dom);
+
 int32_t
 afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type);
 
+afr_fd_ctx_t *
+afr_fd_ctx_get (fd_t *fd, xlator_t *this);
+int
+afr_set_pending_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending,
+                      int child, afr_xattrop_type_t op);
+void
+afr_set_delayed_post_op (call_frame_t *frame, xlator_t *this);
+
+void
+afr_delayed_changelog_wake_up (xlator_t *this, fd_t *fd);
+
+void
+__mark_all_success (int32_t *pending[], int child_count,
+                    afr_transaction_type type);
+gf_boolean_t
+afr_any_fops_failed (afr_local_t *local, afr_private_t *priv);
+
+gf_boolean_t
+afr_txn_nothing_failed (call_frame_t *frame, xlator_t *this);
 #endif /* __TRANSACTION_H__ */
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index 76e28cc4c..c724eb2ae 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include <libgen.h>
@@ -28,2392 +19,775 @@
 #define _CONFIG_H
 #include "config.h"
 #endif
+#include "afr-common.c"
 
-#include "glusterfs.h"
-#include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-
-#include "afr-inode-read.h"
-#include "afr-inode-write.h"
-#include "afr-dir-read.h"
-#include "afr-dir-write.h"
-#include "afr-transaction.h"
-
-#include "afr-self-heal.h"
-
-
-uint64_t
-afr_is_split_brain (xlator_t *this, inode_t *inode)
-{
-        int ret = 0;
-
-        uint64_t ctx         = 0;
-        uint64_t split_brain = 0;
-        
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx);
-
-                if (ret < 0)
-                        goto unlock;
-                
-                split_brain = ctx & 0xFFFFFFFF00000000ULL;
-        }
-unlock:
-        UNLOCK (&inode->lock);
-
-        return split_brain;
-}
+#define SHD_INODE_LRU_LIMIT          2048
+#define AFR_EH_HEALED_LIMIT          1024
+#define AFR_EH_HEAL_FAIL_LIMIT       1024
+#define AFR_EH_SPLIT_BRAIN_LIMIT     1024
 
+struct volume_options options[];
 
-void
-afr_set_split_brain (xlator_t *this, inode_t *inode, int32_t split_brain)
+int32_t
+notify (xlator_t *this, int32_t event,
+        void *data, ...)
 {
-        uint64_t ctx = 0;
-        int      ret = 0;
+        int ret = -1;
+        va_list         ap;
+        void *data2 = NULL;
 
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx);
+        va_start (ap, data);
+        data2 = va_arg (ap, dict_t*);
+        va_end (ap);
+        ret = afr_notify (this, event, data, data2);
 
-                if (ret < 0) {
-                        ctx = 0;
-                }
-
-                ctx = (0x00000000FFFFFFFFULL & ctx)
-                        | (split_brain & 0xFFFFFFFF00000000ULL);
-
-                __inode_ctx_put (inode, this, ctx);
-        }
-        UNLOCK (&inode->lock);
+        return ret;
 }
 
-
-uint64_t
-afr_read_child (xlator_t *this, inode_t *inode)
-{
-        int ret = 0;
-
-        uint64_t ctx         = 0;
-        uint64_t read_child  = 0;
-        
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx);
-
-                if (ret < 0)
-                        goto unlock;
-                
-                read_child = ctx & 0x00000000FFFFFFFFULL;
-        }
-unlock:
-        UNLOCK (&inode->lock);
-
-        return read_child;
-}
-
-
-void
-afr_set_read_child (xlator_t *this, inode_t *inode, int32_t read_child)
+int32_t
+mem_acct_init (xlator_t *this)
 {
-        uint64_t ctx = 0;
-        int      ret = 0;
-
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx);
+        int     ret = -1;
 
-                if (ret < 0) {
-                        ctx = 0;
-                }
+        if (!this)
+                return ret;
 
-                ctx = (0xFFFFFFFF00000000ULL & ctx) 
-                        | (0x00000000FFFFFFFFULL & read_child);
+        ret = xlator_mem_acct_init (this, gf_afr_mt_end + 1);
 
-                __inode_ctx_put (inode, this, ctx);
+        if (ret != 0) {
+                gf_log(this->name, GF_LOG_ERROR, "Memory accounting init"
+                       "failed");
+                return ret;
         }
-        UNLOCK (&inode->lock);
-}
-
-
-/**
- * afr_local_cleanup - cleanup everything in frame->local
- */
-
-void
-afr_local_sh_cleanup (afr_local_t *local, xlator_t *this)
-{
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-
-
-	sh = &local->self_heal;
-	priv = this->private;
-
-	if (sh->buf)
-		FREE (sh->buf);
-
-	if (sh->xattr) {
-		for (i = 0; i < priv->child_count; i++) {
-			if (sh->xattr[i]) {
-				dict_unref (sh->xattr[i]);
-				sh->xattr[i] = NULL;
-			}
-		}
-		FREE (sh->xattr);
-	}
-
-	if (sh->child_errno)
-		FREE (sh->child_errno);
-
-	if (sh->pending_matrix) {
-		for (i = 0; i < priv->child_count; i++) {
-			FREE (sh->pending_matrix[i]);
-		}
-		FREE (sh->pending_matrix);
-	}
-
-	if (sh->delta_matrix) {
-		for (i = 0; i < priv->child_count; i++) {
-			FREE (sh->delta_matrix[i]);
-		}
-		FREE (sh->delta_matrix);
-	}
-
-	if (sh->sources)
-		FREE (sh->sources);
-
-	if (sh->success)
-		FREE (sh->success);
-
-	if (sh->healing_fd) {
-		fd_unref (sh->healing_fd);
-		sh->healing_fd = NULL;
-	}
-
-	loc_wipe (&sh->parent_loc);
-}
-
 
-void 
-afr_local_cleanup (afr_local_t *local, xlator_t *this)
-{
-	if (!local)
-		return;
-
-	afr_local_sh_cleanup (local, this);
-
-	FREE (local->child_errno);
-	FREE (local->pending_array);
-
-	loc_wipe (&local->loc);
-	loc_wipe (&local->newloc);
-
-	FREE (local->transaction.locked_nodes);
-	FREE (local->transaction.child_errno);
-
-	FREE (local->transaction.basename);
-	FREE (local->transaction.new_basename);
-
-	loc_wipe (&local->transaction.parent_loc);	
-	loc_wipe (&local->transaction.new_parent_loc);
-
-	if (local->fd)
-		fd_unref (local->fd);
-	
-	if (local->xattr_req)
-		dict_unref (local->xattr_req);
-
-	FREE (local->child_up);
-
-	{ /* lookup */
-		if (local->cont.lookup.xattr)
-			dict_unref (local->cont.lookup.xattr);
-	}
-
-	{ /* getxattr */
-		if (local->cont.getxattr.name)
-			FREE (local->cont.getxattr.name);
-	}
-
-	{ /* lk */
-		if (local->cont.lk.locked_nodes)
-			FREE (local->cont.lk.locked_nodes);
-	}
-
-	{ /* checksum */
-		if (local->cont.checksum.file_checksum)
-			FREE (local->cont.checksum.file_checksum);
-		if (local->cont.checksum.dir_checksum)
-			FREE (local->cont.checksum.dir_checksum);
-	}
-
-	{ /* create */
-		if (local->cont.create.fd)
-			fd_unref (local->cont.create.fd);
-	}
-
-	{ /* writev */
-		FREE (local->cont.writev.vector);
-	}
-
-	{ /* setxattr */
-		if (local->cont.setxattr.dict)
-			dict_unref (local->cont.setxattr.dict);
-	}
-
-	{ /* removexattr */
-		FREE (local->cont.removexattr.name);
-	}
-
-	{ /* symlink */
-		FREE (local->cont.symlink.linkpath);
-	}
+        return ret;
 }
 
 
 int
-afr_frame_return (call_frame_t *frame)
+xlator_subvolume_index (xlator_t *this, xlator_t *subvol)
 {
-	afr_local_t *local = NULL;
-	int          call_count = 0;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		call_count = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
+        int index = -1;
+        int i = 0;
+        xlator_list_t *list = NULL;
 
-	return call_count;
-}
-
-/**
- * first_up_child - return the index of the first child that is up
- */
-
-int
-afr_first_up_child (afr_private_t *priv)
-{
-	xlator_t ** children = NULL;
-	int         ret      = -1;
-	int         i        = 0;
-
-	LOCK (&priv->lock);
-	{
-		children = priv->children;
-		for (i = 0; i < priv->child_count; i++) {
-			if (priv->child_up[i]) {
-				ret = i;
-				break;
-			}
-		}
-	}
-	UNLOCK (&priv->lock);
-
-	return ret;
-}
-
-
-/**
- * up_children_count - return the number of children that are up
- */
-
-int
-afr_up_children_count (int child_count, unsigned char *child_up)
-{
-	int i   = 0;
-	int ret = 0;
+        list = this->children;
 
-	for (i = 0; i < child_count; i++)
-		if (child_up[i])
-			ret++;
-	return ret;
-}
-
-
-int
-afr_locked_nodes_count (unsigned char *locked_nodes, int child_count)
-{
-	int ret = 0;
-	int i;
-
-	for (i = 0; i < child_count; i++)
-		if (locked_nodes[i])
-			ret++;
-
-	return ret;
-}
-
-
-ino64_t
-afr_itransform (ino64_t ino, int child_count, int child_index)
-{
-	ino64_t scaled_ino = -1;
-
-	if (ino == ((uint64_t) -1)) {
-		scaled_ino = ((uint64_t) -1);
-		goto out;
-	}
-
-	scaled_ino = (ino * child_count) + child_index;
-
-out:
-	return scaled_ino;
-}
-
-
-int
-afr_deitransform_orig (ino64_t ino, int child_count)
-{
-	int index = -1;
-
-	index = ino % child_count;
-
-	return index;
-}
-
-
-int
-afr_deitransform (ino64_t ino, int child_count)
-{
-	return 0;
-}
-
-
-int
-afr_self_heal_cbk (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *local = NULL;
-	int ret = -1;
-
-	local = frame->local;
-
-	if (local->govinda_gOvinda) {
-                afr_set_split_brain (this, local->cont.lookup.inode, 1);
-
-		if (ret < 0) {
-			local->op_ret   = -1;
-			local->op_errno = -ret;
-		}
-	} else {
-                afr_set_split_brain (this, local->cont.lookup.inode, 0);
-	}
-
-	AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-			  local->cont.lookup.inode,
-			  &local->cont.lookup.buf,
-			  local->cont.lookup.xattr);
-
-	return 0;
-}
-
-
-int
-afr_lookup_cbk (call_frame_t *frame, void *cookie,
-		xlator_t *this,	int32_t op_ret,	int32_t op_errno,
-		inode_t *inode,	struct stat *buf, dict_t *xattr)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	struct stat *   lookup_buf = NULL;
-	int             call_count = -1;
-	int             child_index = -1;
-
-	uint32_t        open_fd_count = 0;
-	int             ret = 0;
-
-	child_index = (long) cookie;
-	priv = this->private;
-
-	LOCK (&frame->lock);
-	{
-		local = frame->local;
-
-		lookup_buf = &local->cont.lookup.buf;
-
-		if (op_ret == -1) {
-			if (op_errno == ENOENT)
-				local->enoent_count++;
-			
-			if (op_errno != ENOTCONN) {
-                                if (local->op_errno != ESTALE)
-                                        local->op_errno = op_errno;
-                        }
-
-                        if (op_errno == ESTALE) {
-                                /* no matter what other subvolumes return for
-                                 * this call, ESTALE _must_ be sent to parent
-                                 */
-                                local->op_ret = -1;
-                                local->op_errno = ESTALE;
-                        }
-			goto unlock;
-		}
-
-		if (afr_sh_has_metadata_pending (xattr, child_index, this))
-			local->need_metadata_self_heal = 1;
-
-		if (afr_sh_has_entry_pending (xattr, child_index, this))
-			local->need_entry_self_heal = 1;
-
-		if (afr_sh_has_data_pending (xattr, child_index, this))
-			local->need_data_self_heal = 1;
-
-		ret = dict_get_uint32 (xattr, GLUSTERFS_OPEN_FD_COUNT,
-				       &open_fd_count);
-		local->open_fd_count += open_fd_count;
-
-		/* in case of revalidate, we need to send stat of the
-		 * child whose stat was sent during the first lookup.
-		 * (so that time stamp does not vary with revalidate.
-		 * in case it is down, stat of the fist success will
-		 * be replied */
-
-		/* inode number should be preserved across revalidates */
-
-		if (local->success_count == 0) {
-                        if (local->op_errno != ESTALE)
-                                local->op_ret   = op_ret;
-				
-			local->cont.lookup.inode = inode;
-			local->cont.lookup.xattr = dict_ref (xattr);
-
-			*lookup_buf = *buf;
-			lookup_buf->st_ino = afr_itransform (buf->st_ino,
-							     priv->child_count,
-							     child_index);
-                        
-                        if (priv->read_child >= 0) {
-                                afr_set_read_child (this,
-                                                    local->cont.lookup.inode, 
-                                                    priv->read_child);
-                        } else {
-                                afr_set_read_child (this,
-                                                    local->cont.lookup.inode, 
-                                                    child_index);
-                        }
-
-		} else {
-                        if ((local->op_ret == 0)
-                            && (child_index == local->read_child_index)) {
-                                
-                                /* 
-                                   lookup has succeeded on the read child.
-                                   So use its inode number
-                                */
-
-                                local->op_ret = op_ret;
-
-                                if (local->cont.lookup.xattr)
-                                        dict_unref (local->cont.lookup.xattr);
-                                
-                                local->cont.lookup.inode = inode;
-                                local->cont.lookup.xattr = dict_ref (xattr);
-
-                                *lookup_buf = *buf;
-                                lookup_buf->st_ino = afr_itransform (buf->st_ino,
-                                                                     priv->child_count,
-                                                                     child_index);
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this,
-                                                            local->cont.lookup.inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this,
-                                                            local->cont.lookup.inode,
-                                                            local->read_child_index);
-                                }
-                        }
-
-			if (FILETYPE_DIFFERS (buf, lookup_buf)) {
-				/* mismatching filetypes with same name
-				   -- Govinda !! GOvinda !!!
-				*/
-				local->govinda_gOvinda = 1;
-			}
-
-			if (PERMISSION_DIFFERS (buf, lookup_buf)) {
-				/* mismatching permissions */
-				local->need_metadata_self_heal = 1;
-			}
-
-			if (OWNERSHIP_DIFFERS (buf, lookup_buf)) {
-				/* mismatching permissions */
-				local->need_metadata_self_heal = 1;
-			}
-
-			if (SIZE_DIFFERS (buf, lookup_buf)
-			    && S_ISREG (buf->st_mode)) {
-				local->need_data_self_heal = 1;
-			}
-		}
-
-		local->success_count++;
-	}
-unlock:
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		if (local->op_ret == 0) {
-			/* KLUDGE: assuming DHT will not itransform in 
-			   revalidate */
-			if (local->cont.lookup.inode->ino)
-				lookup_buf->st_ino = 
-					local->cont.lookup.inode->ino;
-		}
-
-		if (local->success_count && local->enoent_count) {
-			local->need_metadata_self_heal = 1;
-			local->need_data_self_heal = 1;
-			local->need_entry_self_heal = 1;
-		}
-
-		if (local->success_count) {
-			/* check for split-brain case in previous lookup */
-			if (afr_is_split_brain (this,
-                                                local->cont.lookup.inode))
-				local->need_data_self_heal = 1;
-		}
-
-		if ((local->need_metadata_self_heal
-		     || local->need_data_self_heal
-		     || local->need_entry_self_heal)
-		    && (!local->open_fd_count)) {
-
-			if (!local->cont.lookup.inode->st_mode) {
-				/* fix for RT #602 */
-				local->cont.lookup.inode->st_mode =
-					lookup_buf->st_mode;
-			}
-
-			afr_self_heal (frame, this, afr_self_heal_cbk);
-		} else {
-			AFR_STACK_UNWIND (frame, local->op_ret,
-					  local->op_errno,
-					  local->cont.lookup.inode, 
-					  &local->cont.lookup.buf,
-					  local->cont.lookup.xattr);
-		}
-	}
-
-	return 0;
-}
-
-
-int
-afr_lookup (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, dict_t *xattr_req)
-{
-	afr_private_t *priv = NULL;
-	afr_local_t   *local = NULL;
-	int            ret = -1;
-	int            i = 0;
-
-	int32_t        op_errno = 0;
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	local->op_ret = -1;
-
-	frame->local = local;
-
-	loc_copy (&local->loc, loc);
-
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr) 
-                        % (priv->child_count);
+        while (list) {
+                if (subvol == list->xlator ||
+                    strcmp (subvol->name, list->xlator->name) == 0) {
+                        index = i;
+                        break;
+                }
+                list = list->next;
+                i++;
         }
-        UNLOCK (&priv->read_child_lock);
-
-	local->call_count = priv->child_count;
-
-	local->child_up = memdup (priv->child_up, priv->child_count);
-	local->child_count = afr_up_children_count (priv->child_count,
-						    local->child_up);
-
-	/* By default assume ENOTCONN. On success it will be set to 0. */
-	local->op_errno = ENOTCONN;
-	
-	if ((xattr_req == NULL)
-	    && (priv->metadata_self_heal
-		|| priv->data_self_heal
-		|| priv->entry_self_heal))
-		local->xattr_req = dict_new ();
-	else
-		local->xattr_req = dict_ref (xattr_req);
-
-	if (priv->metadata_self_heal) {
-		ret = dict_set_uint64 (local->xattr_req, AFR_METADATA_PENDING,
-				       priv->child_count * sizeof(int32_t));
-	}
-	
-	if (priv->data_self_heal) {
-		ret = dict_set_uint64 (local->xattr_req, AFR_DATA_PENDING,
-				       priv->child_count * sizeof(int32_t));
-	}
-	
-	if (priv->entry_self_heal) {
-		ret = dict_set_uint64 (local->xattr_req, AFR_ENTRY_PENDING,
-				       priv->child_count * sizeof(int32_t));
-	}
-
-	ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_OPEN_FD_COUNT, 0);
-
-	for (i = 0; i < priv->child_count; i++) {
-		STACK_WIND_COOKIE (frame, afr_lookup_cbk, (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->lookup,
-				   loc, local->xattr_req);
-	}
-
-	ret = 0;
-out:
-	if (ret == -1)
-		AFR_STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL);
-
-	return 0;
-}
-
-
-/* {{{ open */
-
-int
-afr_open_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-			int32_t op_ret, int32_t op_errno, struct stat *buf)
-{
-	afr_local_t * local = frame->local;
-
-	AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-			  local->fd);
-	return 0;
-}
-
-
-int
-afr_open_cbk (call_frame_t *frame, void *cookie,
-	      xlator_t *this, int32_t op_ret, int32_t op_errno,
-	      fd_t *fd)
-{
-	afr_local_t *  local = NULL;
-	afr_private_t * priv = NULL;
-
-	int call_count = -1;
-	
-	priv  = this->private;
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-		}
-
-		if (op_ret >= 0) {
-			local->op_ret = op_ret;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		if ((local->cont.open.flags & O_TRUNC)
-		    && (local->op_ret >= 0)) {
-			STACK_WIND (frame, afr_open_ftruncate_cbk,
-				    this, this->fops->ftruncate,
-				    fd, 0);
-		} else {
-			AFR_STACK_UNWIND (frame, local->op_ret,
-					  local->op_errno, local->fd);
-		}
-	}
-
-	return 0;
-}
-
-
-int
-afr_open (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc, int32_t flags, fd_t *fd)
-{
-	afr_private_t * priv  = NULL;
-	afr_local_t *   local = NULL;
-
-	int     i = 0;
-	int   ret = -1;
-
-	int32_t call_count = 0;	
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-	int32_t wind_flags = flags & (~O_TRUNC);
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-	VALIDATE_OR_GOTO (loc, out);
-
-	priv = this->private;
-
-        if (afr_is_split_brain (this, loc->inode)) {
-		/* self-heal failed */
-
-		gf_log (this->name, GF_LOG_WARNING,
-			"returning EIO, file has to be manually corrected "
-			"in the backend");
-
-		op_errno = EIO;
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	frame->local = local;
-	call_count   = local->call_count;
-
-	local->cont.open.flags = flags;
-	local->fd = fd_ref (fd);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_open_cbk, (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->open,
-					   loc, wind_flags, fd);
-			
-			if (!--call_count)
-				break;
-		}
-	}
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, fd);
-	}
-
-	return 0;
-}
-
-/* }}} */
-
-/* {{{ flush */
-
-int
-afr_flush_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		      int32_t op_ret, int32_t op_errno)
-{
-	afr_local_t *   local = NULL;
-
-	int call_count  = -1;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0)
-			local->op_ret = 0;
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
-}
-
-
-int
-afr_flush_wind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-	
-	int i = 0;
-	int call_count = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_flush_wind_cbk, 
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->flush,
-					   local->fd);
-		
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
-}
-
-
-int
-afr_flush_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *local = NULL;
-
-	local = frame->local;
-
-	AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno);
 
-	return 0;
+        return index;
 }
 
-
-int
-afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
-{
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-
-	int ret        = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	frame->local = local;
-
-        local->op = GF_FOP_FLUSH;
-        local->transaction.fop    = afr_flush_wind;
-        local->transaction.done   = afr_flush_done;
-
-        local->fd                 = fd_ref (fd);
-
-        local->transaction.start  = 0;
-        local->transaction.len    = 0;
-
-        local->transaction.pending = AFR_DATA_PENDING;
-
-        afr_transaction (frame, this, AFR_FLUSH_TRANSACTION);
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
-
-	return 0;
-}
-
-/* }}} */
-
-/* {{{ fsync */
-
-int
-afr_fsync_cbk (call_frame_t *frame, void *cookie,
-	       xlator_t *this, int32_t op_ret, int32_t op_errno)
+void
+fix_quorum_options (xlator_t *this, afr_private_t *priv, char *qtype)
 {
-	afr_local_t *local = NULL;
-	
-	int call_count = -1;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0)
-			local->op_ret = 0;
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno);
-
-	return 0;
+        if (priv->quorum_count && strcmp(qtype,"fixed")) {
+                gf_log(this->name,GF_LOG_WARNING,
+                       "quorum-type %s overriding quorum-count %u",
+                       qtype, priv->quorum_count);
+        }
+        if (!strcmp(qtype,"none")) {
+                priv->quorum_count = 0;
+        }
+        else if (!strcmp(qtype,"auto")) {
+                priv->quorum_count = AFR_QUORUM_AUTO;
+        }
 }
 
-
 int
-afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd,
-	   int32_t datasync)
-{
-	afr_private_t *priv = NULL;
-	afr_local_t *local = NULL;
-
-	int ret = -1;
-
-	int i = 0;
-	int32_t call_count = 0;
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	call_count = local->call_count;
-	frame->local = local;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_fsync_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->fsync,
-				    fd, datasync);
-			if (!--call_count)
-				break;
-		}
-	}
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-	return 0;
-}
-
-/* }}} */
-
-/* {{{ fsync */
-
-int32_t
-afr_fsyncdir_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno)
-{
-	afr_local_t *local = NULL;
-	
-	int call_count = -1;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0)
-			local->op_ret = 0;
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno);
-
-	return 0;
-}
-
-
-int32_t
-afr_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-	      int32_t datasync)
-{
-	afr_private_t *priv = NULL;
-	afr_local_t *local = NULL;
-
-	int ret = -1;
-
-	int i = 0;
-	int32_t call_count = 0;
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	call_count = local->call_count;
-	frame->local = local;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_fsync_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->fsyncdir,
-				    fd, datasync);
-			if (!--call_count)
-				break;
-		}
-	}
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-	return 0;
-}
-
-/* }}} */
-
-/* {{{ xattrop */
-
-int32_t
-afr_xattrop_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno,
-		 dict_t *xattr)
-{
-	afr_local_t *local = NULL;
-	
-	int call_count = -1;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0)
-			local->op_ret = 0;
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno, xattr);
-
-	return 0;
-}
-
-
-int32_t
-afr_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-	     gf_xattrop_flags_t optype, dict_t *xattr)
-{
-	afr_private_t *priv = NULL;
-	afr_local_t *local  = NULL;
-
-	int ret = -1;
-
-	int i = 0;
-	int32_t call_count = 0;
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	call_count = local->call_count;
-	frame->local = local;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_xattrop_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->xattrop,
-				    loc, optype, xattr);
-			if (!--call_count)
-				break;
-		}
-	}
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-	return 0;
-}
-
-/* }}} */
-
-/* {{{ fxattrop */
-
-int32_t
-afr_fxattrop_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno,
-		  dict_t *xattr)
-{
-	afr_local_t *local = NULL;
-	
-	int call_count = -1;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0)
-			local->op_ret = 0;
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno, xattr);
-
-	return 0;
-}
-
-
-int32_t
-afr_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
-	      gf_xattrop_flags_t optype, dict_t *xattr)
+reconfigure (xlator_t *this, dict_t *options)
 {
-	afr_private_t *priv = NULL;
-	afr_local_t *local  = NULL;
+        afr_private_t *priv        = NULL;
+        xlator_t      *read_subvol = NULL;
+        int            read_subvol_index = -1;
+        int            ret         = -1;
+        int            index       = -1;
+        char          *qtype       = NULL;
 
-	int ret = -1;
+        priv = this->private;
 
-	int i = 0;
-	int32_t call_count = 0;
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
+        GF_OPTION_RECONF ("background-self-heal-count",
+                          priv->background_self_heal_count, options, uint32,
+                          out);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        GF_OPTION_RECONF ("metadata-self-heal",
+                          priv->metadata_self_heal, options, bool, out);
 
-	priv = this->private;
+        GF_OPTION_RECONF ("data-self-heal", priv->data_self_heal, options, str,
+                          out);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        GF_OPTION_RECONF ("entry-self-heal", priv->entry_self_heal, options,
+                          bool, out);
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        GF_OPTION_RECONF ("strict-readdir", priv->strict_readdir, options, bool,
+                          out);
 
-	call_count = local->call_count;
-	frame->local = local;
+        GF_OPTION_RECONF ("data-self-heal-window-size",
+                          priv->data_self_heal_window_size, options,
+                          uint32, out);
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_fxattrop_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->fxattrop,
-				    fd, optype, xattr);
-			if (!--call_count)
-				break;
-		}
-	}
+        GF_OPTION_RECONF ("data-change-log", priv->data_change_log, options,
+                          bool, out);
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-	return 0;
-}
-
-/* }}} */
-
-
-int32_t
-afr_inodelk_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno)
-		
-{
-	afr_local_t *local = NULL;
-	
-	int call_count = -1;
-
-	local = frame->local;
+        GF_OPTION_RECONF ("metadata-change-log",
+                          priv->metadata_change_log, options, bool, out);
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0)
-			local->op_ret = 0;
+        GF_OPTION_RECONF ("entry-change-log", priv->entry_change_log, options,
+                          bool, out);
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno);
-
-	return 0;
-}
-
-
-int32_t
-afr_inodelk (call_frame_t *frame, xlator_t *this, 
-             const char *volume, loc_t *loc, int32_t cmd, struct flock *flock)
-{
-	afr_private_t *priv = NULL;
-	afr_local_t *local  = NULL;
+        GF_OPTION_RECONF ("data-self-heal-algorithm",
+                          priv->data_self_heal_algorithm, options, str, out);
 
-	int ret = -1;
-
-	int i = 0;
-	int32_t call_count = 0;
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	call_count = local->call_count;
-	frame->local = local;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_inodelk_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->inodelk,
-				    volume, loc, cmd, flock);
-
-			if (!--call_count)
-				break;
-		}
-	}
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-	return 0;
-}
-
-
-int32_t
-afr_finodelk_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno)
-		
-{
-	afr_local_t *local = NULL;
-	
-	int call_count = -1;
+        GF_OPTION_RECONF ("self-heal-daemon", priv->shd.enabled, options, bool, out);
 
-	local = frame->local;
+        GF_OPTION_RECONF ("read-subvolume", read_subvol, options, xlator, out);
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0)
-			local->op_ret = 0;
+        GF_OPTION_RECONF ("read-hash-mode", priv->hash_mode,
+                          options, uint32, out);
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno);
-
-	return 0;
-}
-
-
-int32_t
-afr_finodelk (call_frame_t *frame, xlator_t *this, 
-              const char *volume, fd_t *fd, int32_t cmd, struct flock *flock)
-{
-	afr_private_t *priv = NULL;
-	afr_local_t *local  = NULL;
-
-	int ret = -1;
-
-	int i = 0;
-	int32_t call_count = 0;
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	call_count = local->call_count;
-	frame->local = local;
+        if (read_subvol) {
+                index = xlator_subvolume_index (this, read_subvol);
+                if (index == -1) {
+                        gf_log (this->name, GF_LOG_ERROR, "%s not a subvolume",
+                                read_subvol->name);
+                        goto out;
+                }
+                priv->read_child = index;
+        }
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_finodelk_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->finodelk,
-				    volume, fd, cmd, flock);
+        GF_OPTION_RECONF ("read-subvolume-index",read_subvol_index, options,int32,out);
 
-			if (!--call_count)
-				break;
-		}
-	}
+        if (read_subvol_index >-1) {
+                index=read_subvol_index;
+                if (index >= priv->child_count) {
+                        gf_log (this->name, GF_LOG_ERROR, "%d not a subvolume-index",
+                                index);
+                        goto out;
+                }
+                priv->read_child = index;
+        }
 
-	op_ret = 0;
+        GF_OPTION_RECONF ("eager-lock", priv->eager_lock, options, bool, out);
+        GF_OPTION_RECONF ("quorum-type", qtype, options, str, out);
+        GF_OPTION_RECONF ("quorum-count", priv->quorum_count, options,
+                          uint32, out);
+        fix_quorum_options(this,priv,qtype);
+        GF_OPTION_RECONF ("heal-timeout", priv->shd.timeout, options,
+                          int32, out);
+
+	GF_OPTION_RECONF ("post-op-delay-secs", priv->post_op_delay_secs, options,
+			  uint32, out);
+
+        GF_OPTION_RECONF (AFR_SH_READDIR_SIZE_KEY, priv->sh_readdir_size,
+                          options, size, out);
+        /* Reset this so we re-discover in case the topology changed.  */
+        GF_OPTION_RECONF ("readdir-failover", priv->readdir_failover, options,
+                          bool, out);
+        GF_OPTION_RECONF ("ensure-durability", priv->ensure_durability, options,
+                          bool, out);
+        priv->did_discovery = _gf_false;
+
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-	return 0;
-}
-
-
-int32_t
-afr_entrylk_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno)
-		
-{
-	afr_local_t *local = NULL;
-	
-	int call_count = -1;
+        return ret;
 
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0)
-			local->op_ret = 0;
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno);
-
-	return 0;
 }
 
 
-int32_t
-afr_entrylk (call_frame_t *frame, xlator_t *this, 
-             const char *volume, loc_t *loc,
-	     const char *basename, entrylk_cmd cmd, entrylk_type type)
-{
-	afr_private_t *priv = NULL;
-	afr_local_t *local  = NULL;
-
-	int ret = -1;
-
-	int i = 0;
-	int32_t call_count = 0;
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	call_count = local->call_count;
-	frame->local = local;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_entrylk_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->entrylk,
-				    volume, loc, basename, cmd, type);
-
-			if (!--call_count)
-				break;
-		}
-	}
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-	return 0;
-}
-
-
-
-int32_t
-afr_fentrylk_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno)
-		
-{
-	afr_local_t *local = NULL;
-	
-	int call_count = -1;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0)
-			local->op_ret = 0;
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno);
-
-	return 0;
-}
+static const char *favorite_child_warning_str = "You have specified subvolume '%s' "
+        "as the 'favorite child'. This means that if a discrepancy in the content "
+        "or attributes (ownership, permission, etc.) of a file is detected among "
+        "the subvolumes, the file on '%s' will be considered the definitive "
+        "version and its contents will OVERWRITE the contents of the file on other "
+        "subvolumes. All versions of the file except that on '%s' "
+        "WILL BE LOST.";
 
 
 int32_t
-afr_fentrylk (call_frame_t *frame, xlator_t *this, 
-              const char *volume, fd_t *fd,
-	      const char *basename, entrylk_cmd cmd, entrylk_type type)
+init (xlator_t *this)
 {
-	afr_private_t *priv = NULL;
-	afr_local_t *local  = NULL;
-
-	int ret = -1;
-
-	int i = 0;
-	int32_t call_count = 0;
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	call_count = local->call_count;
-	frame->local = local;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_fentrylk_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->fentrylk,
-				    volume, fd, basename, cmd, type);
+        afr_private_t *priv        = NULL;
+        int            child_count = 0;
+        xlator_list_t *trav        = NULL;
+        int            i           = 0;
+        int            ret         = -1;
+        GF_UNUSED int  op_errno    = 0;
+        xlator_t      *read_subvol = NULL;
+        int            read_subvol_index = -1;
+        xlator_t      *fav_child   = NULL;
+        char          *qtype       = NULL;
+
+        if (!this->children) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "replicate translator needs more than one "
+                        "subvolume defined.");
+                return -1;
+        }
 
-			if (!--call_count)
-				break;
-		}
-	}
+        if (!this->parents) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Volume is dangling.");
+        }
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-	return 0;
-}
+	this->private = GF_CALLOC (1, sizeof (afr_private_t),
+                                   gf_afr_mt_afr_private_t);
+        if (!this->private)
+                goto out;
 
+        priv = this->private;
+        LOCK_INIT (&priv->lock);
+        LOCK_INIT (&priv->read_child_lock);
+        //lock recovery is not done in afr
+        pthread_mutex_init (&priv->mutex, NULL);
+        INIT_LIST_HEAD (&priv->saved_fds);
 
-int32_t
-afr_checksum_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno,
-		  uint8_t *file_checksum, uint8_t *dir_checksum)
-		
-{
-	afr_local_t *local = NULL;
-	
-	int call_count = -1;
+        child_count = xlator_subvolume_count (this);
 
-	local = frame->local;
+        priv->child_count = child_count;
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0 && (local->op_ret != 0)) {
-			local->op_ret = 0;
+        priv->read_child = -1;
 
-			local->cont.checksum.file_checksum = MALLOC (ZR_FILENAME_MAX);
-			memcpy (local->cont.checksum.file_checksum, file_checksum, 
-				ZR_FILENAME_MAX);
+        GF_OPTION_INIT ("read-subvolume", read_subvol, xlator, out);
+        if (read_subvol) {
+                priv->read_child = xlator_subvolume_index (this, read_subvol);
+                if (priv->read_child == -1) {
+                        gf_log (this->name, GF_LOG_ERROR, "%s not a subvolume",
+                                read_subvol->name);
+                        goto out;
+                }
+        }
+        GF_OPTION_INIT ("read-subvolume-index",read_subvol_index,int32,out);
+        if (read_subvol_index > -1) {
+                if (read_subvol_index >= priv->child_count) {
+                        gf_log (this->name, GF_LOG_ERROR, "%d not a subvolume-index",
+                                read_subvol_index);
+                        goto out;
+                }
+                priv->read_child = read_subvol_index;
+        }
+        GF_OPTION_INIT ("choose-local", priv->choose_local, bool, out);
+
+        GF_OPTION_INIT ("read-hash-mode", priv->hash_mode, uint32, out);
+
+        priv->favorite_child = -1;
+        GF_OPTION_INIT ("favorite-child", fav_child, xlator, out);
+        if (fav_child) {
+                priv->favorite_child = xlator_subvolume_index (this, fav_child);
+                if (priv->favorite_child == -1) {
+                        gf_log (this->name, GF_LOG_ERROR, "%s not a subvolume",
+                                fav_child->name);
+                        goto out;
+                }
+                gf_log (this->name, GF_LOG_WARNING,
+                        favorite_child_warning_str, fav_child->name,
+                        fav_child->name, fav_child->name);
+        }
 
-			local->cont.checksum.dir_checksum = MALLOC (ZR_FILENAME_MAX);
-			memcpy (local->cont.checksum.dir_checksum, dir_checksum, 
-				ZR_FILENAME_MAX);
 
-		}
+        GF_OPTION_INIT ("background-self-heal-count",
+                        priv->background_self_heal_count, uint32, out);
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
+        GF_OPTION_INIT ("data-self-heal", priv->data_self_heal, str, out);
 
-	call_count = afr_frame_return (frame);
+        GF_OPTION_INIT ("data-self-heal-algorithm",
+                        priv->data_self_heal_algorithm, str, out);
 
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-				  local->cont.checksum.file_checksum, 
-				  local->cont.checksum.dir_checksum);
+        GF_OPTION_INIT ("data-self-heal-window-size",
+                        priv->data_self_heal_window_size, uint32, out);
 
-	return 0;
-}
+        GF_OPTION_INIT ("metadata-self-heal", priv->metadata_self_heal, bool,
+                        out);
 
+        GF_OPTION_INIT ("entry-self-heal", priv->entry_self_heal, bool, out);
 
-int32_t
-afr_checksum (call_frame_t *frame, xlator_t *this, loc_t *loc,
-	      int32_t flag)
-{
-	afr_private_t *priv = NULL;
-	afr_local_t *local  = NULL;
+        GF_OPTION_INIT ("self-heal-daemon", priv->shd.enabled, bool, out);
 
-	int ret = -1;
+        GF_OPTION_INIT ("iam-self-heal-daemon", priv->shd.iamshd, bool, out);
 
-	int i = 0;
-	int32_t call_count = 0;
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
+        GF_OPTION_INIT ("data-change-log", priv->data_change_log, bool, out);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        GF_OPTION_INIT ("metadata-change-log", priv->metadata_change_log, bool,
+                        out);
 
-	priv = this->private;
+        GF_OPTION_INIT ("entry-change-log", priv->entry_change_log, bool, out);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        GF_OPTION_INIT ("optimistic-change-log", priv->optimistic_change_log,
+                        bool, out);
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+        GF_OPTION_INIT ("inodelk-trace", priv->inodelk_trace, bool, out);
 
-	call_count = local->call_count;
-	frame->local = local;
+        GF_OPTION_INIT ("entrylk-trace", priv->entrylk_trace, bool, out);
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_checksum_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->checksum,
-				    loc, flag);
+        GF_OPTION_INIT ("strict-readdir", priv->strict_readdir, bool, out);
 
-			if (!--call_count)
-				break;
-		}
-	}
+        GF_OPTION_INIT ("eager-lock", priv->eager_lock, bool, out);
+        GF_OPTION_INIT ("quorum-type", qtype, str, out);
+        GF_OPTION_INIT ("quorum-count", priv->quorum_count, uint32, out);
+        GF_OPTION_INIT (AFR_SH_READDIR_SIZE_KEY, priv->sh_readdir_size, size,
+                        out);
+        fix_quorum_options(this,priv,qtype);
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno);
-	}
-	return 0;
-}
+	GF_OPTION_INIT ("post-op-delay-secs", priv->post_op_delay_secs, uint32, out);
+        GF_OPTION_INIT ("readdir-failover", priv->readdir_failover, bool, out);
+        GF_OPTION_INIT ("ensure-durability", priv->ensure_durability, bool,
+                        out);
 
+        priv->wait_count = 1;
 
-int32_t
-afr_statfs_cbk (call_frame_t *frame, void *cookie,
-		xlator_t *this, int32_t op_ret, int32_t op_errno,
-		struct statvfs *statvfs)
-{
-	afr_local_t *local = NULL;
+        priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count,
+                                    gf_afr_mt_char);
+        if (!priv->child_up) {
+                ret = -ENOMEM;
+                goto out;
+        }
 
-	int call_count = 0;
+        for (i = 0; i < child_count; i++)
+                priv->child_up[i] = -1; /* start with unknown state.
+                                           this initialization needed
+                                           for afr_notify() to work
+                                           reliably
+                                        */
+
+        priv->children = GF_CALLOC (sizeof (xlator_t *), child_count,
+                                    gf_afr_mt_xlator_t);
+        if (!priv->children) {
+                ret = -ENOMEM;
+                goto out;
+        }
 
-	LOCK (&frame->lock);
-	{
-		local = frame->local;
+        priv->pending_key = GF_CALLOC (sizeof (*priv->pending_key),
+                                       child_count,
+                                       gf_afr_mt_char);
+        if (!priv->pending_key) {
+                ret = -ENOMEM;
+                goto out;
+        }
 
-		if (op_ret == 0) {
-			local->op_ret   = op_ret;
-			
-			if (local->cont.statfs.buf_set) {
-				if (statvfs->f_bavail < local->cont.statfs.buf.f_bavail)
-					local->cont.statfs.buf = *statvfs;
-			} else {
-				local->cont.statfs.buf = *statvfs;
-				local->cont.statfs.buf_set = 1;
-			}
-		}
+        trav = this->children;
+        i = 0;
+        while (i < child_count) {
+                priv->children[i] = trav->xlator;
+
+                ret = gf_asprintf (&priv->pending_key[i], "%s.%s",
+                                   AFR_XATTR_PREFIX,
+                                   trav->xlator->name);
+                if (-1 == ret) {
+                        ret = -ENOMEM;
+                        goto out;
+                }
 
-		if (op_ret == -1)
-			local->op_errno = op_errno;
+                trav = trav->next;
+                i++;
+        }
 
-	}
-	UNLOCK (&frame->lock);
+        ret = gf_asprintf (&priv->sh_domain, AFR_SH_DATA_DOMAIN_FMT,
+                           this->name);
+        if (-1 == ret) {
+                ret = -ENOMEM;
+                goto out;
+        }
 
-	call_count = afr_frame_return (frame);
+        priv->last_event = GF_CALLOC (child_count, sizeof (*priv->last_event),
+                                      gf_afr_mt_int32_t);
+        if (!priv->last_event) {
+                ret = -ENOMEM;
+                goto out;
+        }
 
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-				  &local->cont.statfs.buf);
+        /* keep more local here as we may need them for self-heal etc */
+        this->local_pool = mem_pool_new (afr_local_t, 512);
+        if (!this->local_pool) {
+                ret = -1;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to create local_t's memory pool");
+                goto out;
+        }
 
-	return 0;
-}
+        priv->first_lookup = 1;
+        priv->root_inode = NULL;
 
+        if (!priv->shd.iamshd) {
+                ret = 0;
+                goto out;
+        }
 
-int32_t
-afr_statfs (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc)
-{
-	afr_private_t *  priv        = NULL;
-	int              child_count = 0;
-	afr_local_t   *  local       = NULL;
-	int              i           = 0;
-
-	int ret = -1;
-	int              call_count = 0;
-	int32_t          op_ret      = -1;
-	int32_t          op_errno    = 0;
-
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-	VALIDATE_OR_GOTO (loc, out);
-
-	priv = this->private;
-	child_count = priv->child_count;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	frame->local = local;
-	call_count = local->call_count;
-
-	for (i = 0; i < child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_statfs_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->statfs, 
-				    loc);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	op_ret = 0;
+        ret = -ENOMEM;
+        priv->shd.pos = GF_CALLOC (sizeof (*priv->shd.pos), child_count,
+                                   gf_afr_mt_brick_pos_t);
+        if (!priv->shd.pos)
+                goto out;
+
+        priv->shd.pending = GF_CALLOC (sizeof (*priv->shd.pending), child_count,
+                                       gf_afr_mt_int32_t);
+        if (!priv->shd.pending)
+                goto out;
+
+        priv->shd.inprogress = GF_CALLOC (sizeof (*priv->shd.inprogress),
+                                          child_count, gf_afr_mt_shd_bool_t);
+        if (!priv->shd.inprogress)
+                goto out;
+        priv->shd.timer = GF_CALLOC (sizeof (*priv->shd.timer), child_count,
+                                     gf_afr_mt_shd_timer_t);
+        if (!priv->shd.timer)
+                goto out;
+
+        priv->shd.healed = eh_new (AFR_EH_HEALED_LIMIT, _gf_false,
+                                   _destroy_shd_event_data);
+        if (!priv->shd.healed)
+                goto out;
+
+        priv->shd.heal_failed = eh_new (AFR_EH_HEAL_FAIL_LIMIT, _gf_false,
+                                        _destroy_shd_event_data);
+        if (!priv->shd.heal_failed)
+                goto out;
+
+        priv->shd.split_brain = eh_new (AFR_EH_SPLIT_BRAIN_LIMIT, _gf_false,
+                                        _destroy_shd_event_data);
+        if (!priv->shd.split_brain)
+                goto out;
+
+        this->itable = inode_table_new (SHD_INODE_LRU_LIMIT, this);
+        if (!this->itable)
+                goto out;
+        priv->root_inode = inode_ref (this->itable->root);
+        GF_OPTION_INIT ("node-uuid", priv->shd.node_uuid, str, out);
+        GF_OPTION_INIT ("heal-timeout", priv->shd.timeout, int32, out);
+        ret = afr_initialise_statistics (this);
+        if (ret)
+                goto out;
+        ret = 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
-	return 0;
-}
-
-
-int32_t
-afr_lk_unlock_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		   int32_t op_ret, int32_t op_errno, struct flock *lock)
-{
-	afr_local_t * local = NULL;
-
-	int call_count = -1;
-
-	local = frame->local;
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-				  lock);
-
-	return 0;
-}
-
-
-int32_t 
-afr_lk_unlock (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t   * local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int i;
-	int call_count = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_locked_nodes_count (local->cont.lk.locked_nodes, 
-					     priv->child_count);
-
-	if (call_count == 0) {
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-				  &local->cont.lk.flock);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	local->cont.lk.flock.l_type = F_UNLCK;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->cont.lk.locked_nodes[i]) {
-			STACK_WIND (frame, afr_lk_unlock_cbk,
-				    priv->children[i],
-				    priv->children[i]->fops->lk,
-				    local->fd, F_SETLK, 
-				    &local->cont.lk.flock);
-
-			if (!--call_count)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-
-int32_t
-afr_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-	    int32_t op_ret, int32_t op_errno, struct flock *lock)
-{
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count  = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv  = this->private;
-
-	child_index = (long) cookie;
-
-	call_count = --local->call_count;
-
-	if (!child_went_down (op_ret, op_errno) && (op_ret == -1)) {
-		local->op_ret   = -1;
-		local->op_errno = op_errno;
-
-		afr_lk_unlock (frame, this);
-		return 0;
-	}
-
-	if (op_ret == 0) {
-		local->op_ret        = 0;
-		local->op_errno      = 0;
-		local->cont.lk.flock = *lock;
-		local->cont.lk.locked_nodes[child_index] = 1;
-	}
-
-	child_index++;
-
-	if (child_index < priv->child_count) {
-		STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) child_index,
-				   priv->children[child_index],
-				   priv->children[child_index]->fops->lk,
-				   local->fd, local->cont.lk.cmd, 
-				   &local->cont.lk.flock);
-	} else if (local->op_ret == -1) {
-		/* all nodes have gone down */
-		
-		AFR_STACK_UNWIND (frame, -1, ENOTCONN, &local->cont.lk.flock);
-	} else {
-		/* locking has succeeded on all nodes that are up */
-		
-		AFR_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-			      &local->cont.lk.flock);
-	}
-
-	return 0;
+        return ret;
 }
 
 
 int
-afr_lk (call_frame_t *frame, xlator_t *this,
-	fd_t *fd, int32_t cmd,
-	struct flock *flock)
-{
-	afr_private_t *priv = NULL;
-	afr_local_t *local = NULL;
-
-	int i = 0;
-
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	AFR_LOCAL_INIT (local, priv);
-
-	frame->local  = local;
-
-	local->cont.lk.locked_nodes = CALLOC (priv->child_count, 
-					      sizeof (*local->cont.lk.locked_nodes));
-	
-	if (!local->cont.lk.locked_nodes) {
-		gf_log (this->name, GF_LOG_ERROR, "out of memory :(");
-		op_errno = ENOMEM;
-		goto out;
-	}
-
-	local->fd            = fd_ref (fd);
-	local->cont.lk.cmd   = cmd;
-	local->cont.lk.flock = *flock;
-
-	STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) 0,
-			   priv->children[i],
-			   priv->children[i]->fops->lk,
-			   fd, cmd, flock);
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (frame, op_ret, op_errno, NULL);
-	}
-	return 0;
-}
-
-
-/**
- * find_child_index - find the child's index in the array of subvolumes
- * @this: AFR
- * @child: child
- */
-
-static int
-find_child_index (xlator_t *this, xlator_t *child)
-{
-	afr_private_t *priv = NULL;
-
-	int i = -1;
-
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if ((xlator_t *) child == priv->children[i])
-			break;
-	}
-
-	return i;
-}
-
-
-int32_t
-notify (xlator_t *this, int32_t event,
-	void *data, ...)
-{
-	afr_private_t *     priv     = NULL;
-	unsigned char *     child_up = NULL;
-
-	int i           = -1;
-	int up_children = 0;
-
-	priv = this->private;
-
-	if (!priv)
-		return 0;
-
-	child_up = priv->child_up;
-
-	switch (event) {
-	case GF_EVENT_CHILD_UP:
-		i = find_child_index (this, data);
-
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "subvolume %s came up", ((xlator_t *) data)->name);
-
-		child_up[i] = 1;
-
-		/* 
-		   if all the children were down, and one child came up, 
-		   send notify to parent
-		*/
-
-		for (i = 0; i < priv->child_count; i++)
-			if (child_up[i])
-				up_children++;
-
-		if (up_children == 1)
-			default_notify (this, event, data);
-
-		break;
-
-	case GF_EVENT_CHILD_DOWN:
-		i = find_child_index (this, data);
-
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "subvolume %s went down", ((xlator_t *) data)->name);
-
-		child_up[i] = 0;
-		
-		/* 
-		   if all children are down, and this was the last to go down,
-		   send notify to parent
-		*/
-
-		for (i = 0; i < priv->child_count; i++)
-			if (child_up[i])
-				up_children++;
-
-		if (up_children == 0)
-			default_notify (this, event, data);
-
-		break;
-
-	default:
-		default_notify (this, event, data);
-	}
-
-	return 0;
-}
-
-
-static const char *favorite_child_warning_str = "You have specified subvolume '%s' "
-	"as the 'favorite child'. This means that if a discrepancy in the content "
-	"or attributes (ownership, permission, etc.) of a file is detected among "
-	"the subvolumes, the file on '%s' will be considered the definitive "
-	"version and its contents will OVERWRITE the contents of the file on other "
-	"subvolumes. All versions of the file except that on '%s' "
-	"WILL BE LOST.";
-
-static const char *no_lock_servers_warning_str = "You have set lock-server-count = 0. "
-	"This means correctness is NO LONGER GUARANTEED in all cases. If two or more "
-	"applications write to the same region of a file, there is a possibility that "
-	"its copies will be INCONSISTENT. Set it to a value greater than 0 unless you "
-	"are ABSOLUTELY SURE of what you are doing and WILL NOT HOLD GlusterFS "
-	"RESPONSIBLE for inconsistent data. If you are in doubt, set it to a value "
-	"greater than 0.";
-
-int32_t 
-init (xlator_t *this)
+fini (xlator_t *this)
 {
-	afr_private_t * priv        = NULL;
-	int             child_count = 0;
-	xlator_list_t * trav        = NULL;
-	int             i           = 0;
-	int             ret         = -1;
-	int             op_errno    = 0;
-
-	char * read_subvol = NULL;
-	char * fav_child   = NULL;
-	char * self_heal   = NULL;
-	char * change_log  = NULL;
-
-	int32_t lock_server_count = 1;
-
-	int    fav_ret       = -1;
-	int    read_ret      = -1;
-	int    dict_ret      = -1;
-
-	if (!this->children) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"AFR needs more than one child defined");
-		return -1;
-	}
-  
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-
-	ALLOC_OR_GOTO (this->private, afr_private_t, out);
-
-	priv = this->private;
-
-	read_ret = dict_get_str (this->options, "read-subvolume", &read_subvol);
-	priv->read_child = -1;
-
-	fav_ret = dict_get_str (this->options, "favorite-child", &fav_child);
-	priv->favorite_child = -1;
-
-	/* Default values */
-
-	priv->data_self_heal     = 1;
-	priv->metadata_self_heal = 1;
-	priv->entry_self_heal    = 1;
-
-	dict_ret = dict_get_str (this->options, "data-self-heal", &self_heal);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (self_heal, &priv->data_self_heal);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"invalid 'option data-self-heal %s' "
-				"defaulting to data-self-heal as 'on'",
-				self_heal);
-			priv->data_self_heal = 1;
-		} 
-	}
-
-	dict_ret = dict_get_str (this->options, "metadata-self-heal",
-				 &self_heal);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (self_heal, &priv->metadata_self_heal);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"invalid 'option metadata-self-heal %s' "
-				"defaulting to metadata-self-heal as 'on'", 
-				self_heal);
-			priv->metadata_self_heal = 1;
-		} 
-	}
-
-	dict_ret = dict_get_str (this->options, "entry-self-heal", &self_heal);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (self_heal, &priv->entry_self_heal);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"invalid 'option entry-self-heal %s' "
-				"defaulting to entry-self-heal as 'on'", 
-				self_heal);
-			priv->entry_self_heal = 1;
-		} 
-	}
-
-	/* Change log options */
-
-	priv->data_change_log     = 1;
-	priv->metadata_change_log = 0;
-	priv->entry_change_log    = 1;
-
-	dict_ret = dict_get_str (this->options, "data-change-log",
-				 &change_log);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (change_log, &priv->data_change_log);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"invalid 'option data-change-log %s'. "
-				"defaulting to data-change-log as 'on'", 
-				change_log);
-			priv->data_change_log = 1;
-		} 
-	}
-
-	dict_ret = dict_get_str (this->options, "metadata-change-log",
-				 &change_log);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (change_log,
-					 &priv->metadata_change_log);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"invalid 'option metadata-change-log %s'. "
-				"defaulting to metadata-change-log as 'off'",
-				change_log);
-			priv->metadata_change_log = 0;
-		} 
-	}
-
-	dict_ret = dict_get_str (this->options, "entry-change-log",
-				 &change_log);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (change_log, &priv->entry_change_log);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"invalid 'option entry-change-log %s'. "
-				"defaulting to entry-change-log as 'on'", 
-				change_log);
-			priv->entry_change_log = 1;
-		} 
-	}
-
-	/* Locking options */
-
-	priv->data_lock_server_count = 1;
-	priv->metadata_lock_server_count = 0;
-	priv->entry_lock_server_count = 1;
-
-	dict_ret = dict_get_int32 (this->options, "data-lock-server-count", 
-				   &lock_server_count);
-	if (dict_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"setting data lock server count to %d",
-			lock_server_count);
-
-		if (lock_server_count == 0) 
-			gf_log (this->name, GF_LOG_WARNING,
-				no_lock_servers_warning_str);
-
-		priv->data_lock_server_count = lock_server_count;
-	}
-
-
-	dict_ret = dict_get_int32 (this->options,
-				   "metadata-lock-server-count", 
-				   &lock_server_count);
-	if (dict_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"setting metadata lock server count to %d",
-			lock_server_count);
-		priv->metadata_lock_server_count = lock_server_count;
-	}
-
-
-	dict_ret = dict_get_int32 (this->options, "entry-lock-server-count", 
-				   &lock_server_count);
-	if (dict_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"setting entry lock server count to %d",
-			lock_server_count);
-
-		priv->entry_lock_server_count = lock_server_count;
-	}
-
-
-	trav = this->children;
-	while (trav) {
-		if (!read_ret && !strcmp (read_subvol, trav->xlator->name)) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"subvolume '%s' specified as read child",
-				trav->xlator->name);
-
-			priv->read_child = child_count;
-		}
-
-		if (fav_ret == 0 && !strcmp (fav_child, trav->xlator->name)) {
-			gf_log (this->name, GF_LOG_WARNING,
-				favorite_child_warning_str, trav->xlator->name,
-				trav->xlator->name, trav->xlator->name);
-			priv->favorite_child = child_count;
-		}
-
-		child_count++;
-		trav = trav->next;
-	}
-
-	priv->wait_count = 1;
-
-	priv->child_count = child_count;
-
-	LOCK_INIT (&priv->lock);
-        LOCK_INIT (&priv->read_child_lock);
+        afr_private_t *priv = NULL;
 
-	priv->child_up = CALLOC (sizeof (unsigned char), child_count);
-	if (!priv->child_up) {
-		gf_log (this->name, GF_LOG_ERROR,	
-			"out of memory :(");		
-		op_errno = ENOMEM;			
-		goto out;
-	}
-
-	priv->children = CALLOC (sizeof (xlator_t *), child_count);
-	if (!priv->children) {
-		gf_log (this->name, GF_LOG_ERROR,	
-			"out of memory :(");		
-		op_errno = ENOMEM;			
-		goto out;
-	}
-
-	trav = this->children;
-	i = 0;
-	while (i < child_count) {
-		priv->children[i] = trav->xlator;
-
-		trav = trav->next;
-		i++;
-	}
-
-	ret = 0;
-out:
-	return ret;
-}
+        priv = this->private;
+        this->private = NULL;
+        afr_priv_destroy (priv);
+        if (this->itable);//I dont see any destroy func
 
-
-int
-fini (xlator_t *this)
-{
-	return 0;
+        return 0;
 }
 
 
 struct xlator_fops fops = {
-	.lookup      = afr_lookup,
-	.open        = afr_open,
-	.lk          = afr_lk,
-	.flush       = afr_flush,
-	.statfs      = afr_statfs,
-	.fsync       = afr_fsync,
-	.fsyncdir    = afr_fsyncdir,
-	.xattrop     = afr_xattrop,
-	.fxattrop    = afr_fxattrop,
-	.inodelk     = afr_inodelk,
-	.finodelk    = afr_finodelk,
-	.entrylk     = afr_entrylk,
-	.fentrylk    = afr_fentrylk,
-	.checksum    = afr_checksum,
-
-	/* inode read */
-	.access      = afr_access,
-	.stat        = afr_stat,
-	.fstat       = afr_fstat,
-	.readlink    = afr_readlink,
-	.getxattr    = afr_getxattr,
-	.readv       = afr_readv,
-
-	/* inode write */
-	.chmod       = afr_chmod,
-	.chown       = afr_chown,
-	.fchmod      = afr_fchmod,
-	.fchown      = afr_fchown,
-	.writev      = afr_writev,
-	.truncate    = afr_truncate,
-	.ftruncate   = afr_ftruncate,
-	.utimens     = afr_utimens,
-	.setxattr    = afr_setxattr,
-	.removexattr = afr_removexattr,
-
-	/* dir read */
-	.opendir     = afr_opendir,
-	.readdir     = afr_readdir,
-	.getdents    = afr_getdents,
-
-	/* dir write */
-	.create      = afr_create,
-	.mknod       = afr_mknod,
-	.mkdir       = afr_mkdir,
-	.unlink      = afr_unlink,
-	.rmdir       = afr_rmdir,
-	.link        = afr_link,
-	.symlink     = afr_symlink,
-	.rename      = afr_rename,
-	.setdents    = afr_setdents,
+        .lookup      = afr_lookup,
+        .open        = afr_open,
+        .lk          = afr_lk,
+        .flush       = afr_flush,
+        .statfs      = afr_statfs,
+        .fsync       = afr_fsync,
+        .fsyncdir    = afr_fsyncdir,
+        .xattrop     = afr_xattrop,
+        .fxattrop    = afr_fxattrop,
+        .inodelk     = afr_inodelk,
+        .finodelk    = afr_finodelk,
+        .entrylk     = afr_entrylk,
+        .fentrylk    = afr_fentrylk,
+	.fallocate   = afr_fallocate,
+	.discard     = afr_discard,
+        .zerofill    = afr_zerofill,
+
+        /* inode read */
+        .access      = afr_access,
+        .stat        = afr_stat,
+        .fstat       = afr_fstat,
+        .readlink    = afr_readlink,
+        .getxattr    = afr_getxattr,
+        .fgetxattr   = afr_fgetxattr,
+        .readv       = afr_readv,
+
+        /* inode write */
+        .writev      = afr_writev,
+        .truncate    = afr_truncate,
+        .ftruncate   = afr_ftruncate,
+        .setxattr    = afr_setxattr,
+        .fsetxattr   = afr_fsetxattr,
+        .setattr     = afr_setattr,
+        .fsetattr    = afr_fsetattr,
+        .removexattr = afr_removexattr,
+        .fremovexattr = afr_fremovexattr,
+
+        /* dir read */
+        .opendir     = afr_opendir,
+        .readdir     = afr_readdir,
+        .readdirp    = afr_readdirp,
+
+        /* dir write */
+        .create      = afr_create,
+        .mknod       = afr_mknod,
+        .mkdir       = afr_mkdir,
+        .unlink      = afr_unlink,
+        .rmdir       = afr_rmdir,
+        .link        = afr_link,
+        .symlink     = afr_symlink,
+        .rename      = afr_rename,
 };
 
 
-struct xlator_mops mops = {
+struct xlator_dumpops dumpops = {
+        .priv       = afr_priv_dump,
 };
 
 
 struct xlator_cbks cbks = {
+        .release     = afr_release,
+        .releasedir  = afr_releasedir,
+        .forget      = afr_forget,
 };
 
+
 struct volume_options options[] = {
-	{ .key  = {"read-subvolume" }, 
-	  .type = GF_OPTION_TYPE_XLATOR
-	},
-	{ .key  = {"favorite-child"}, 
-	  .type = GF_OPTION_TYPE_XLATOR
-	},
-	{ .key  = {"data-self-heal"},  
-	  .type = GF_OPTION_TYPE_BOOL 
-	},
-	{ .key  = {"metadata-self-heal"},  
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-	{ .key  = {"entry-self-heal"},  
-	  .type = GF_OPTION_TYPE_BOOL 
-	},
-	{ .key  = {"data-change-log"},  
-	  .type = GF_OPTION_TYPE_BOOL 
-	},
-	{ .key  = {"metadata-change-log"},  
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-	{ .key  = {"entry-change-log"},  
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-	{ .key  = {"data-lock-server-count"},  
-	  .type = GF_OPTION_TYPE_INT, 
-	  .min  = 0
-	},
-	{ .key  = {"metadata-lock-server-count"},  
-	  .type = GF_OPTION_TYPE_INT, 
-	  .min  = 0
-	},
-	{ .key  = {"entry-lock-server-count"},  
-	  .type = GF_OPTION_TYPE_INT,
-	  .min  = 0
-	},
-	{ .key  = {NULL} },
+        { .key  = {"read-subvolume" },
+          .type = GF_OPTION_TYPE_XLATOR,
+          .description = "inode-read fops happen only on one of the bricks in "
+                         "replicate. Afr will prefer the one specified using "
+                         "this option if it is not stale. Option value must be "
+                         "one of the xlator names of the children. "
+                         "Ex: <volname>-client-0 till "
+                         "<volname>-client-<number-of-bricks - 1>"
+        },
+        { .key  = {"read-subvolume-index" },
+          .type = GF_OPTION_TYPE_INT,
+          .default_value = "-1",
+          .description = "inode-read fops happen only on one of the bricks in "
+                         "replicate. AFR will prefer the one specified using "
+                         "this option if it is not stale. allowed options"
+                         " include -1 till replica-count - 1"
+        },
+        { .key = {"read-hash-mode" },
+          .type = GF_OPTION_TYPE_INT,
+          .min = 0,
+          .max = 2,
+          .default_value = "0",
+          .description = "inode-read fops happen only on one of the bricks in "
+                         "replicate. AFR will prefer the one computed using "
+                         "the method specified using this option"
+                         "0 = first responder, "
+                         "1 = hash by GFID of file (all clients use "
+                                                    "same subvolume), "
+                         "2 = hash by GFID of file and client PID",
+        },
+        { .key  = {"choose-local" },
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "true",
+          .description = "Choose a local subvolume(i.e. Brick) to read from if "
+                         "read-subvolume is not explicitly set.",
+        },
+        { .key  = {"favorite-child"},
+          .type = GF_OPTION_TYPE_XLATOR,
+          .description = "If a split-brain happens choose subvol/brick set by "
+                         "this option as source."
+        },
+        { .key  = {"background-self-heal-count"},
+          .type = GF_OPTION_TYPE_INT,
+          .min  = 0,
+          .default_value = "16",
+          .validate = GF_OPT_VALIDATE_MIN,
+          .description = "This specifies the number of self-heals that can be "
+                         " performed in background without blocking the fop"
+        },
+        { .key  = {"data-self-heal"},
+          .type = GF_OPTION_TYPE_STR,
+          .value = {"1", "on", "yes", "true", "enable",
+                    "0", "off", "no", "false", "disable",
+                    "open"},
+          .default_value = "on",
+          .description   = "Using this option we can enable/disable data "
+                           "self-heal on the file. \"open\" means data "
+                           "self-heal action will only be triggered by file "
+                           "open operations."
+        },
+        { .key  = {"data-self-heal-algorithm"},
+          .type = GF_OPTION_TYPE_STR,
+          .description   = "Select between \"full\", \"diff\". The "
+                           "\"full\" algorithm copies the entire file from "
+                           "source to sink. The \"diff\" algorithm copies to "
+                           "sink only those blocks whose checksums don't match "
+                           "with those of source. If no option is configured "
+                           "the option is chosen dynamically as follows: "
+                           "If the file does not exist on one of the sinks "
+                           "or empty file exists or if the source file size is "
+                           "about the same as page size the entire file will "
+                           "be read and written i.e \"full\" algo, "
+                           "otherwise \"diff\" algo is chosen.",
+          .value = { "diff", "full"}
+        },
+        { .key  = {"data-self-heal-window-size"},
+          .type = GF_OPTION_TYPE_INT,
+          .min  = 1,
+          .max  = 1024,
+          .default_value = "1",
+          .description = "Maximum number blocks per file for which self-heal "
+                         "process would be applied simultaneously."
+        },
+        { .key  = {"metadata-self-heal"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "on",
+          .description = "Using this option we can enable/disable metadata "
+                         "i.e. Permissions, ownerships, xattrs self-heal on "
+                         "the file/directory."
+        },
+        { .key  = {"entry-self-heal"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "on",
+          .description = "Using this option we can enable/disable entry "
+                         "self-heal on the directory."
+        },
+        { .key  = {"data-change-log"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "on",
+          .description = "Data fops like write/truncate will not perform "
+                         "pre/post fop changelog operations in afr transaction "
+                         "if this option is disabled"
+        },
+        { .key  = {"metadata-change-log"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "on",
+          .description = "Metadata fops like setattr/setxattr will not perform "
+                         "pre/post fop changelog operations in afr transaction "
+                         "if this option is disabled"
+        },
+        { .key  = {"entry-change-log"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "on",
+          .description = "Entry fops like create/unlink will not perform "
+                         "pre/post fop changelog operations in afr transaction "
+                         "if this option is disabled"
+        },
+        { .key  = {"optimistic-change-log"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "on",
+          .description = "Entry/Metadata fops will not perform "
+                         "pre fop changelog operations in afr transaction "
+                         "if this option is enabled."
+        },
+        { .key  = {"strict-readdir"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+        },
+        { .key = {"inodelk-trace"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+          .description = "Enabling this option logs inode lock/unlocks"
+        },
+        { .key = {"entrylk-trace"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+          .description = "Enabling this option logs entry lock/unlocks"
+        },
+        { .key = {"eager-lock"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "on",
+          .description = "Lock phase of a transaction has two sub-phases. "
+                         "First is an attempt to acquire locks in parallel by "
+                         "broadcasting non-blocking lock requests. If lock "
+                         "aquistion fails on any server, then the held locks "
+                         "are unlocked and revert to a blocking locked mode "
+                         "sequentially on one server after another.  If this "
+                         "option is enabled the initial broadcasting lock "
+                         "request attempt to acquire lock on the entire file. "
+                         "If this fails, we revert back to the sequential "
+                         "\"regional\" blocking lock as before. In the case "
+                         "where such an \"eager\" lock is granted in the "
+                         "non-blocking phase, it gives rise to an opportunity "
+                         "for optimization. i.e, if the next write transaction "
+                         "on the same FD arrives before the unlock phase of "
+                         "the first transaction, it \"takes over\" the full "
+                         "file lock. Similarly if yet another data transaction "
+                         "arrives before the unlock phase of the \"optimized\" "
+                         "transaction, that in turn \"takes over\" the lock as "
+                         "well. The actual unlock now happens at the end of "
+                         "the last \"optimzed\" transaction."
+
+        },
+        { .key = {"self-heal-daemon"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+          .description = "This option applies to only self-heal-daemon. "
+                         "Index directory crawl and automatic healing of files"
+                         "will not be performed if this option is turned off."
+        },
+        { .key = {"iam-self-heal-daemon"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+          .description = "This option differentiates if the replicate "
+                         "translator is running as part of self-heal-daemon "
+                         "or not."
+        },
+        { .key = {"quorum-type"},
+          .type = GF_OPTION_TYPE_STR,
+          .value = { "none", "auto", "fixed"},
+          .default_value = "none",
+          .description = "If value is \"fixed\" only allow writes if "
+                         "quorum-count bricks are present.  If value is "
+                         "\"auto\" only allow writes if more than half of "
+                         "bricks, or exactly half including the first, are "
+                         "present.",
+        },
+        { .key = {"quorum-count"},
+          .type = GF_OPTION_TYPE_INT,
+          .min = 1,
+          .max = INT_MAX,
+          .default_value = 0,
+          .description = "If quorum-type is \"fixed\" only allow writes if "
+                         "this many bricks or present.  Other quorum types "
+                         "will OVERWRITE this value.",
+        },
+        { .key  = {"node-uuid"},
+          .type = GF_OPTION_TYPE_STR,
+          .description = "Local glusterd uuid string, used in starting "
+                         "self-heal-daemon so that it can crawl only on "
+                         "local index directories.",
+        },
+        { .key  = {"heal-timeout"},
+          .type = GF_OPTION_TYPE_INT,
+          .min  = 60,
+          .max  = INT_MAX,
+          .default_value = "600",
+          .description = "time interval for checking the need to self-heal "
+                         "in self-heal-daemon"
+        },
+        { .key  = {"post-op-delay-secs"},
+          .type = GF_OPTION_TYPE_INT,
+          .min  = 0,
+          .max  = INT_MAX,
+          .default_value = "1",
+          .description = "Time interval induced artificially before "
+	                 "post-operation phase of the transaction to "
+                         "enhance overlap of adjacent write operations.",
+        },
+        { .key = {AFR_SH_READDIR_SIZE_KEY},
+          .type = GF_OPTION_TYPE_SIZET,
+          .description = "readdirp size for performing entry self-heal",
+          .min = 1024,
+          .max = 131072,
+          .default_value = "1KB",
+        },
+        { .key = {"readdir-failover"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .description = "readdir(p) will not failover if this option is off",
+          .default_value = "on",
+        },
+        { .key = {"ensure-durability"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .description = "Afr performs fsyncs for transactions if this "
+                         "option is on to make sure the changelogs/data is "
+                         "written to the disk",
+          .default_value = "on",
+        },
+        { .key  = {NULL} },
 };
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index ac1a5f6ad..21064db58 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 
@@ -26,417 +17,919 @@
 #include "config.h"
 #endif
 
-#include "scheduler.h"
 #include "call-stub.h"
 #include "compat-errno.h"
+#include "afr-mem-types.h"
+#include "afr-self-heal-algorithm.h"
+
+#include "libxlator.h"
+#include "timer.h"
+
+#define AFR_XATTR_PREFIX "trusted.afr"
+#define AFR_PATHINFO_HEADER "REPLICATE:"
+#define AFR_SH_READDIR_SIZE_KEY "self-heal-readdir-size"
+#define AFR_SH_DATA_DOMAIN_FMT "%s:self-heal"
+
+#define AFR_LOCKEE_COUNT_MAX    3
+#define AFR_DOM_COUNT_MAX    3
+
+struct _pump_private;
+
+typedef int (*afr_expunge_done_cbk_t) (call_frame_t *frame, xlator_t *this,
+                                       int child, int32_t op_error,
+                                       int32_t op_errno);
+
+typedef int (*afr_impunge_done_cbk_t) (call_frame_t *frame, xlator_t *this,
+                                       int32_t op_error, int32_t op_errno);
+typedef int (*afr_post_remove_call_t) (call_frame_t *frame, xlator_t *this);
+
+typedef int (*afr_lock_cbk_t) (call_frame_t *frame, xlator_t *this);
+typedef void (*afr_lookup_done_cbk_t) (call_frame_t *frame, xlator_t *this,
+                                      int32_t op_ret, int32_t op_errno);
+
+typedef enum {
+        AFR_POS_UNKNOWN,
+        AFR_POS_LOCAL,
+        AFR_POS_REMOTE
+} afr_child_pos_t;
+
+typedef enum {
+        SPLIT_BRAIN = 1,
+        ALL_FOOLS = 2
+} afr_subvol_status_t;
+
+typedef enum {
+        AFR_INODE_SET_READ_CTX = 1,
+        AFR_INODE_RM_STALE_CHILDREN,
+        AFR_INODE_SET_OPENDIR_DONE,
+        AFR_INODE_GET_READ_CTX,
+        AFR_INODE_GET_OPENDIR_DONE,
+} afr_inode_op_t;
+
+typedef struct afr_inode_params_ {
+        afr_inode_op_t op;
+        union {
+                gf_boolean_t value;
+                struct {
+                        int32_t read_child;
+                        int32_t *children;
+                } read_ctx;
+        } u;
+} afr_inode_params_t;
+
+typedef enum afr_spb_state {
+        DONT_KNOW,
+        SPB,
+        NO_SPB
+} afr_spb_state_t;
+
+typedef struct afr_inode_ctx_ {
+        uint64_t masks;
+        int32_t  *fresh_children;//increasing order of latency
+        afr_spb_state_t mdata_spb;
+        afr_spb_state_t data_spb;
+        uint32_t        open_fd_count;
+} afr_inode_ctx_t;
 
+typedef enum {
+        NONE,
+        INDEX,
+        INDEX_TO_BE_HEALED,
+        FULL,
+} afr_crawl_type_t;
+
+typedef struct afr_self_heald_ {
+        gf_boolean_t            enabled;
+        gf_boolean_t            iamshd;
+        afr_crawl_type_t        *pending;
+        gf_boolean_t            *inprogress;
+        afr_child_pos_t         *pos;
+        gf_timer_t              **timer;
+        eh_t                    *healed;
+        eh_t                    *heal_failed;
+        eh_t                    *split_brain;
+        eh_t                    **statistics;
+        void                    **crawl_events;
+        char                    *node_uuid;
+        int                     timeout;
+} afr_self_heald_t;
 
 typedef struct _afr_private {
-	gf_lock_t lock;               /* to guard access to child_count, etc */
-	unsigned int child_count;     /* total number of children   */
+        gf_lock_t lock;               /* to guard access to child_count, etc */
+        unsigned int child_count;     /* total number of children   */
 
         unsigned int read_child_rr;   /* round-robin index of the read_child */
         gf_lock_t read_child_lock;    /* lock to protect above */
-        
-	xlator_t **children;
 
-	unsigned char *child_up;
+        xlator_t **children;
 
-	gf_boolean_t data_self_heal;       /* on/off */
-	gf_boolean_t metadata_self_heal;   /* on/off */
-	gf_boolean_t entry_self_heal;      /* on/off */
+        int first_lookup;
+        inode_t *root_inode;
 
+        unsigned char *child_up;
 
-	gf_boolean_t data_change_log;       /* on/off */
-	gf_boolean_t metadata_change_log;   /* on/off */
-	gf_boolean_t entry_change_log;      /* on/off */
+        char **pending_key;
 
-	int read_child;               /* read-subvolume */
-	unsigned int favorite_child;  /* subvolume to be preferred in resolving
-					 split-brain cases */
+        char         *data_self_heal;              /* on/off/open */
+        char *       data_self_heal_algorithm;    /* name of algorithm */
+        unsigned int data_self_heal_window_size;  /* max number of pipelined
+                                                     read/writes */
 
-	unsigned int data_lock_server_count;
-	unsigned int metadata_lock_server_count;
-	unsigned int entry_lock_server_count;
+        unsigned int background_self_heal_count;
+        unsigned int background_self_heals_started;
+        gf_boolean_t metadata_self_heal;   /* on/off */
+        gf_boolean_t entry_self_heal;      /* on/off */
 
-	unsigned int wait_count;      /* # of servers to wait for success */
-} afr_private_t;
+        gf_boolean_t data_change_log;       /* on/off */
+        gf_boolean_t metadata_change_log;   /* on/off */
+        gf_boolean_t entry_change_log;      /* on/off */
 
-typedef struct {
-	/* array of stat's, one for each child */
-	struct stat *buf;
+        int read_child;               /* read-subvolume */
+        unsigned int hash_mode;       /* for when read_child is not set */
+        int favorite_child;  /* subvolume to be preferred in resolving
+                                         split-brain cases */
 
-	/* array of xattr's, one for each child */
-	dict_t **xattr;
+        gf_boolean_t inodelk_trace;
+        gf_boolean_t entrylk_trace;
 
-	/* array of errno's, one for each child */
-	int *child_errno;
+        gf_boolean_t strict_readdir;
 
-	int32_t **pending_matrix;
-	int32_t **delta_matrix;
+        unsigned int wait_count;      /* # of servers to wait for success */
 
-	int *sources;
-	int source;
-	int active_source;
-	int active_sinks;
-	int *success;
+        uint64_t up_count;      /* number of CHILD_UPs we have seen */
+        uint64_t down_count;    /* number of CHILD_DOWNs we have seen */
 
-	fd_t *healing_fd;
-	int   op_failed;
+        struct _pump_private *pump_private; /* Set if we are loaded as pump */
+        int                   use_afr_in_pump;
 
-	int   file_has_holes;
-	blksize_t block_size;
-	off_t file_size;
-	off_t offset;
+        pthread_mutex_t  mutex;
+        struct list_head saved_fds;   /* list of fds on which locks have succeeded */
+        gf_boolean_t      optimistic_change_log;
+        gf_boolean_t      eager_lock;
+	uint32_t          post_op_delay_secs;
+        unsigned int      quorum_count;
 
-	loc_t parent_loc;
-	int (*completion_cbk) (call_frame_t *frame, xlator_t *this);
-	call_frame_t *sh_frame;
-} afr_self_heal_t;
+        char                   vol_uuid[UUID_SIZE + 1];
+        int32_t                *last_event;
+        afr_self_heald_t       shd;
+        gf_boolean_t           choose_local;
+        gf_boolean_t           did_discovery;
+        gf_boolean_t           readdir_failover;
+        uint64_t               sh_readdir_size;
+        gf_boolean_t           ensure_durability;
+        char                   *sh_domain;
+} afr_private_t;
 
+typedef enum {
+        AFR_SELF_HEAL_NOT_ATTEMPTED,
+        AFR_SELF_HEAL_STARTED,
+        AFR_SELF_HEAL_FAILED,
+        AFR_SELF_HEAL_SYNC_BEGIN,
+} afr_self_heal_status;
+
+typedef struct {
+        afr_self_heal_status gfid_or_missing_entry_self_heal;
+        afr_self_heal_status metadata_self_heal;
+        afr_self_heal_status data_self_heal;
+        afr_self_heal_status entry_self_heal;
+} afr_sh_status_for_all_type;
+
+typedef enum {
+        AFR_SELF_HEAL_ENTRY,
+        AFR_SELF_HEAL_METADATA,
+        AFR_SELF_HEAL_DATA,
+        AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY,
+        AFR_SELF_HEAL_INVALID = -1,
+} afr_self_heal_type;
 
 typedef enum {
-	AFR_DATA_TRANSACTION,          /* truncate, write, ... */
-	AFR_METADATA_TRANSACTION,      /* chmod, chown, ... */
-	AFR_ENTRY_TRANSACTION,         /* create, rmdir, ... */
-	AFR_ENTRY_RENAME_TRANSACTION,  /* rename */
-	AFR_FLUSH_TRANSACTION,         /* flush */
+        AFR_CHECK_ALL,
+        AFR_CHECK_SPECIFIC,
+} afr_sh_fail_check_type;
+
+struct afr_self_heal_ {
+        /* External interface: These are variables (some optional) that
+           are set by whoever has triggered self-heal */
+
+        gf_boolean_t do_data_self_heal;
+        gf_boolean_t do_metadata_self_heal;
+        gf_boolean_t do_entry_self_heal;
+        gf_boolean_t do_gfid_self_heal;
+        gf_boolean_t do_missing_entry_self_heal;
+        gf_boolean_t force_confirm_spb; /* Check for split-brains even when
+                                           self-heal is turned off */
+
+        gf_boolean_t forced_merge;        /* Is this a self-heal triggered to
+                                             forcibly merge the directories? */
+
+        gf_boolean_t background;          /* do self-heal in background
+                                             if possible */
+        ia_type_t type;                   /* st_mode of the entry we're doing
+                                             self-heal on */
+        inode_t   *inode;                 /* inode on which the self-heal is
+                                             performed on */
+        uuid_t  sh_gfid_req;                 /* gfid self-heal needs to be done
+                                             with this gfid if it is not null */
+
+        /* Function to call to unwind. If self-heal is being done in the
+           background, this function will be called as soon as possible. */
+
+        int (*unwind) (call_frame_t *frame, xlator_t *this, int32_t op_ret,
+                       int32_t op_errno, int32_t sh_failed);
+
+        /* End of external interface members */
+
+
+        /* array of stat's, one for each child */
+        struct iatt *buf;
+        struct iatt *parentbufs;
+        struct iatt parentbuf;
+        struct iatt entrybuf;
+
+        afr_expunge_done_cbk_t expunge_done;
+        afr_impunge_done_cbk_t impunge_done;
+
+        /* array of xattr's, one for each child */
+        dict_t **xattr;
+
+        /* array containing if the lookups succeeded in the order of response
+         */
+        int32_t *success_children;
+        int     success_count;
+        /* array containing the fresh children found in the self-heal process */
+        int32_t *fresh_children;
+        /* array containing the fresh children found in the parent lookup */
+        int32_t *fresh_parent_dirs;
+        /* array of errno's, one for each child */
+        int *child_errno;
+        /*loc used for lookup*/
+        loc_t lookup_loc;
+        int32_t lookup_flags;
+        afr_lookup_done_cbk_t lookup_done;
+
+        int32_t **pending_matrix;
+        int32_t **delta_matrix;
+
+        int32_t op_ret;
+        int32_t op_errno;
+
+        int *sources;
+        int source;
+        int active_source;
+        int active_sinks;
+        unsigned char *success;
+        unsigned char *locked_nodes;
+        int lock_count;
+
+        const char *linkname;
+        gf_boolean_t entries_skipped;
+
+        gf_boolean_t actual_sh_started;
+        gf_boolean_t sync_done;
+        gf_boolean_t data_lock_held;
+        gf_boolean_t sh_dom_lock_held;
+        gf_boolean_t eof_reached;
+        fd_t  *healing_fd;
+        int   file_has_holes;
+        blksize_t block_size;
+        off_t file_size;
+        off_t offset;
+        unsigned char *write_needed;
+        uint8_t *checksum;
+        afr_post_remove_call_t post_remove_call;
+
+        char    *data_sh_info;
+        char    *metadata_sh_info;
+
+        loc_t parent_loc;
+        call_frame_t *orig_frame;
+        call_frame_t *old_loop_frame;
+        gf_boolean_t unwound;
+
+        afr_sh_algo_private_t *private;
+        afr_sh_status_for_all_type  afr_all_sh_status;
+        afr_self_heal_type       sh_type_in_action;
+
+        struct afr_sh_algorithm  *algo;
+        afr_lock_cbk_t data_lock_success_handler;
+        afr_lock_cbk_t data_lock_failure_handler;
+	gf_boolean_t data_lock_block;
+        int (*completion_cbk) (call_frame_t *frame, xlator_t *this);
+        int (*sh_data_algo_start) (call_frame_t *frame, xlator_t *this);
+        int (*algo_completion_cbk) (call_frame_t *frame, xlator_t *this);
+        int (*algo_abort_cbk) (call_frame_t *frame, xlator_t *this);
+        void (*gfid_sh_success_cbk) (call_frame_t *sh_frame, xlator_t *this);
+
+        call_frame_t *sh_frame;
+};
+
+typedef struct afr_self_heal_ afr_self_heal_t;
+
+typedef enum {
+        AFR_DATA_TRANSACTION,          /* truncate, write, ... */
+        AFR_METADATA_TRANSACTION,      /* chmod, chown, ... */
+        AFR_ENTRY_TRANSACTION,         /* create, rmdir, ... */
+        AFR_ENTRY_RENAME_TRANSACTION,  /* rename */
 } afr_transaction_type;
 
-typedef struct _afr_local {
-	unsigned int call_count;
-	unsigned int success_count;
-	unsigned int enoent_count;
+typedef enum {
+        AFR_TRANSACTION_LK,
+        AFR_SELFHEAL_LK,
+} transaction_lk_type_t;
 
-	unsigned int need_metadata_self_heal;
-	unsigned int need_entry_self_heal;
-	unsigned int need_data_self_heal;
-	unsigned int govinda_gOvinda;
+typedef enum {
+        AFR_LOCK_OP,
+        AFR_UNLOCK_OP,
+} afr_lock_op_type_t;
 
-	unsigned int read_child_index;
-        
-	int32_t op_ret;
-	int32_t op_errno;
+typedef enum {
+        AFR_DATA_SELF_HEAL_LK,
+        AFR_METADATA_SELF_HEAL_LK,
+        AFR_ENTRY_SELF_HEAL_LK,
+}selfheal_lk_type_t;
 
-	int32_t *pending_array;
+typedef enum {
+        AFR_INODELK_TRANSACTION,
+        AFR_INODELK_NB_TRANSACTION,
+        AFR_ENTRYLK_TRANSACTION,
+        AFR_ENTRYLK_NB_TRANSACTION,
+        AFR_INODELK_SELFHEAL,
+        AFR_INODELK_NB_SELFHEAL,
+        AFR_ENTRYLK_SELFHEAL,
+        AFR_ENTRYLK_NB_SELFHEAL,
+} afr_lock_call_type_t;
 
-	loc_t loc;
-	loc_t newloc;
+/*
+  xattr format: trusted.afr.volume = [x y z]
+  x - data pending
+  y - metadata pending
+  z - entry pending
+*/
 
-	fd_t *fd;
+static inline int
+afr_index_for_transaction_type (afr_transaction_type type)
+{
+        switch (type) {
 
-	glusterfs_fop_t fop;
+        case AFR_DATA_TRANSACTION:
+                return 0;
 
-	unsigned char *child_up; 
-	int            child_count;
+        case AFR_METADATA_TRANSACTION:
+                return 1;
 
-	int32_t *child_errno;
-	
-	dict_t  *xattr_req;
-	int      open_fd_count;
-	/* 
-	   This struct contains the arguments for the "continuation"
-	   (scheme-like) of fops
-	*/
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+                return 2;
+        }
 
-	int   op;
-	struct {
-		struct {
-			unsigned char buf_set;
-			struct statvfs buf;
-		} statfs;
+        return -1;  /* make gcc happy */
+}
 
-		struct {
-			inode_t *inode;
-			struct stat buf;
-			dict_t *xattr;
-		} lookup;
+typedef struct {
+        loc_t                   loc;
+        char                    *basename;
+        unsigned char           *locked_nodes;
+        int                     locked_count;
 
-		struct {
-			int32_t flags;
-		} open;
+} afr_entry_lockee_t;
 
-		struct {
-			int32_t cmd;
-			struct flock flock;
-			unsigned char *locked_nodes;
-		} lk;
+int
+afr_entry_lockee_cmp (const void *l1, const void *l2);
 
-		struct {
-			uint8_t *file_checksum;
-			uint8_t *dir_checksum;
-		} checksum;
+typedef struct {
+        char    *domain; /* Domain on which inodelk is taken */
+        struct gf_flock flock;
+        unsigned char *locked_nodes;
+        int32_t lock_count;
+} afr_inodelk_t;
 
-		/* inode read */
+typedef struct {
+        loc_t *lk_loc;
 
-		struct {
-			int32_t mask;
-			int last_tried;  /* index of the child we tried previously */
-		} access;
+        int                     lockee_count;
+        afr_entry_lockee_t      lockee[AFR_LOCKEE_COUNT_MAX];
 
-		struct {
-			int last_tried;
-			ino_t ino;
-		} stat;
+        afr_inodelk_t       inodelk[AFR_DOM_COUNT_MAX];
+        const char *lk_basename;
+        const char *lower_basename;
+        const char *higher_basename;
+        char lower_locked;
+        char higher_locked;
 
-		struct {
-			int last_tried;
-			ino_t ino;
-		} fstat;
+        unsigned char *locked_nodes;
+        unsigned char *lower_locked_nodes;
 
-		struct {
-			size_t size;
-			int last_tried;
-		} readlink;
+        selfheal_lk_type_t selfheal_lk_type;
+        transaction_lk_type_t transaction_lk_type;
 
-		struct {
-			const char *name;
-			int last_tried;
-		} getxattr;
+        int32_t lock_count;
+        int32_t entrylk_lock_count;
 
-		struct {
-			size_t size;
-			off_t offset;
-			int last_tried;
-		} readv;
+        uint64_t lock_number;
+        int32_t lk_call_count;
+        int32_t lk_expected_count;
+        int32_t lk_attempted_count;
 
-		/* dir read */
+        int32_t lock_op_ret;
+        int32_t lock_op_errno;
+        afr_lock_cbk_t lock_cbk;
+        char *domain; /* Domain on which inode/entry lock/unlock in progress.*/
+} afr_internal_lock_t;
 
-		struct {
-			int success_count;
-			int32_t op_ret;
-			int32_t op_errno;
-		} opendir;
+typedef struct _afr_locked_fd {
+        fd_t  *fd;
+        struct list_head list;
+} afr_locked_fd_t;
 
-		struct {
-			int32_t op_ret;
-			int32_t op_errno;
-			size_t size;
-			off_t offset;
+struct afr_reply {
+	int	valid;
+	int32_t	op_ret;
+	int32_t	op_errno;
+};
 
-			int last_tried;
-		} readdir;
+typedef struct _afr_local {
+        int     uid;
+        int     gid;
+        unsigned int call_count;
+        unsigned int success_count;
+        unsigned int enoent_count;
+        uint32_t     open_fd_count;
+        gf_boolean_t update_open_fd_count;
 
-		struct {
-			int32_t op_ret;
-			int32_t op_errno;
 
-			size_t size;
-			off_t offset;
-			int32_t flag;
+        unsigned int unhealable;
 
-			int last_tried;
-		} getdents;
+        unsigned int read_child_index;
+        unsigned char read_child_returned;
+        unsigned int first_up_child;
 
-		/* inode write */
+	gf_lkowner_t  saved_lk_owner;
 
-		struct {
-			ino_t ino;
-			mode_t mode;
-			struct stat buf;
-		} chmod;
+        int32_t op_ret;
+        int32_t op_errno;
 
-		struct {
-			ino_t ino;
-			mode_t mode;
-			struct stat buf;
-		} fchmod;
+        int32_t **pending;
 
-		struct {
-			ino_t ino;
-			uid_t uid;
-			gid_t gid;
-			struct stat buf;
-		} chown;
+        loc_t loc;
+        loc_t newloc;
 
-		struct {
-			ino_t ino;
-			uid_t uid;
-			gid_t gid;
-			struct stat buf;
-		} fchown;
-		
-		struct {
-			ino_t ino;
-			struct stat buf;
+        fd_t *fd;
 
-			int32_t op_ret;
+        glusterfs_fop_t fop;
 
-			struct iovec *vector;
-			dict_t *refs;
-			int32_t count;
-			off_t offset;
-		} writev;
+        unsigned char *child_up;
+        int32_t       *fresh_children; //in the order of response
 
-		struct {
-			ino_t ino;
-			off_t offset;
-			struct stat buf;
-		} truncate;
+        int32_t *child_errno;
 
-		struct {
-			ino_t ino;
-			off_t offset;
-			struct stat buf;
-		} ftruncate;
+        dict_t  *xattr_req;
 
-		struct {
-			ino_t ino;
-			struct timespec tv[2];
-			struct stat buf;
-		} utimens;
+        int32_t  inodelk_count;
+        int32_t  entrylk_count;
 
-		struct {
-			dict_t *dict;
-			int32_t flags;
-		} setxattr;
+        afr_internal_lock_t internal_lock;
 
-		struct {
-			const char *name;
-		} removexattr;
+        afr_locked_fd_t *locked_fd;
+        int32_t          source_child;
+        int32_t          lock_recovery_child;
 
-		/* dir write */
-		
-		struct {
-			ino_t ino;
-			fd_t *fd;
-			int32_t flags;
-			mode_t mode;
-			inode_t *inode;
-			struct stat buf;
-		} create;
+        dict_t  *dict;
+        int      optimistic_change_log;
+	gf_boolean_t      delayed_post_op;
 
-		struct {
-			ino_t ino;
-			dev_t dev;
-			mode_t mode;
-			inode_t *inode;
-			struct stat buf;
-		} mknod;
+
+	/* Is the current writev() going to perform a stable write?
+	   i.e, is fd->flags or @flags writev param have O_SYNC or
+	   O_DSYNC?
+	*/
+        gf_boolean_t      stable_write;
+
+        /* This write appended to the file. Nnot necessarily O_APPEND,
+           just means the offset of write was at the end of file.
+        */
+        gf_boolean_t      append_write;
+
+        int allow_sh_for_running_transaction;
+
+
+        /* This struct contains the arguments for the "continuation"
+           (scheme-like) of fops
+        */
+
+        int   op;
+        struct {
+                struct {
+                        unsigned char buf_set;
+                        struct statvfs buf;
+                } statfs;
+
+                struct {
+                        uint32_t parent_entrylk;
+                        uuid_t  gfid_req;
+                        inode_t *inode;
+                        struct iatt buf;
+                        struct iatt postparent;
+                        dict_t **xattrs;
+                        dict_t *xattr;
+                        struct iatt *postparents;
+                        struct iatt *bufs;
+                        int32_t read_child;
+                        int32_t *sources;
+                        int32_t *success_children;
+                        int32_t **pending_matrix;
+                        gf_boolean_t fresh_lookup;
+                        gf_boolean_t possible_spb;
+                } lookup;
+
+                struct {
+                        int32_t flags;
+                } open;
+
+                struct {
+                        int32_t cmd;
+                        struct gf_flock user_flock;
+                        struct gf_flock ret_flock;
+                        unsigned char *locked_nodes;
+                } lk;
+
+                /* inode read */
+
+                struct {
+                        int32_t mask;
+                        int last_index;  /* index of the child we tried previously */
+                } access;
+
+                struct {
+                        int last_index;
+                } stat;
+
+                struct {
+                        int last_index;
+                } fstat;
+
+                struct {
+                        size_t size;
+                        int last_index;
+                } readlink;
+
+                struct {
+                        char *name;
+                        int last_index;
+                        long xattr_len;
+                } getxattr;
+
+                struct {
+                        size_t size;
+                        off_t offset;
+                        int last_index;
+                        uint32_t flags;
+                } readv;
+
+                /* dir read */
+
+                struct {
+                        int success_count;
+                        int32_t op_ret;
+                        int32_t op_errno;
+
+                        uint32_t *checksum;
+                } opendir;
+
+                struct {
+                        int32_t op_ret;
+                        int32_t op_errno;
+                        size_t size;
+                        off_t offset;
+                        dict_t *dict;
+                        gf_boolean_t failed;
+                        int last_index;
+                } readdir;
+                /* inode write */
+
+                struct {
+                        struct iatt prebuf;
+                        struct iatt postbuf;
+                } inode_wfop; //common structure for all inode-write-fops
+
+                struct {
+                        int32_t op_ret;
+
+                        struct iovec *vector;
+                        struct iobref *iobref;
+                        int32_t count;
+                        off_t offset;
+                        uint32_t flags;
+                } writev;
+
+                struct {
+                        off_t offset;
+                } truncate;
+
+                struct {
+                        off_t offset;
+                } ftruncate;
+
+                struct {
+                        struct iatt in_buf;
+                        int32_t valid;
+                } setattr;
+
+                struct {
+                        struct iatt in_buf;
+                        int32_t valid;
+                } fsetattr;
+
+                struct {
+                        dict_t *dict;
+                        int32_t flags;
+                } setxattr;
+
+                struct {
+                        dict_t *dict;
+                        int32_t flags;
+                } fsetxattr;
+
+                struct {
+                        char *name;
+                } removexattr;
+
+                struct {
+                        dict_t *xattr;
+                } xattrop;
+
+                struct {
+                        dict_t *xattr;
+                } fxattrop;
+
+                /* dir write */
+
+                struct {
+                        inode_t *inode;
+                        struct iatt buf;
+                        struct iatt preparent;
+                        struct iatt postparent;
+                        struct iatt prenewparent;
+                        struct iatt postnewparent;
+                } dir_fop; //common structure for all dir fops
+
+                struct {
+                        fd_t *fd;
+                        dict_t *params;
+                        int32_t flags;
+                        mode_t mode;
+                } create;
+
+                struct {
+                        dev_t dev;
+                        mode_t mode;
+                        dict_t *params;
+                } mknod;
+
+                struct {
+                        int32_t mode;
+                        dict_t *params;
+                } mkdir;
+
+                struct {
+                        int flags;
+                } rmdir;
+
+                struct {
+                        dict_t *params;
+                        char *linkpath;
+                } symlink;
 
 		struct {
-			ino_t ino;
 			int32_t mode;
-			inode_t *inode;
-			struct stat buf;
-		} mkdir;
+			off_t offset;
+			size_t len;
+		} fallocate;
 
 		struct {
-			int32_t op_ret;
-			int32_t op_errno;
-		} unlink;
+			off_t offset;
+			size_t len;
+		} discard;
 
-		struct {
-			int32_t op_ret;
-			int32_t op_errno;
-		} rmdir;
+                struct {
+                        off_t offset;
+                        size_t len;
+                        struct iatt prebuf;
+                        struct iatt postbuf;
+                } zerofill;
 
-		struct {
-			ino_t ino;
-			struct stat buf;
-		} rename;
 
-		struct {
-			ino_t ino;
-			inode_t *inode;
-			struct stat buf;
-		} link;
+        } cont;
 
-		struct {
-			ino_t ino;
-			inode_t *inode;
-			struct stat buf;
-			char *linkpath;
-		} symlink;
+        struct {
+                off_t start, len;
 
-		struct {
-			int32_t flags;
-			dir_entry_t *entries;
-			int32_t count;
-		} setdents;
-	} cont;
-	
-	struct {
-		off_t start, len;
+                gf_boolean_t    eager_lock_on;
+                int *eager_lock;
+
+                char *basename;
+                char *new_basename;
 
-		unsigned char *locked_nodes;
-		int lock_count;
+                loc_t parent_loc;
+                loc_t new_parent_loc;
 
-		const char *basename;
-		const char *new_basename;
+                afr_transaction_type type;
 
-		char *pending;
+		/* pre-compute the post piggyback status before
+		   entering POST-OP phase
+		*/
+		int              *postop_piggybacked;
 
-		loc_t parent_loc;
-		loc_t new_parent_loc;
+		/* stub to resume on destruction
+		   of the transaction frame */
+		call_stub_t      *resume_stub;
 
-		afr_transaction_type type;
+		struct list_head  eager_locked;
 
-		int success_count;
-		int erase_pending;
-		int failure_count;
+                int32_t         **txn_changelog;//changelog after pre+post ops
+                unsigned char   *pre_op;
 
-		int last_tried;
-		int32_t *child_errno;
+                call_frame_t *main_frame;
 
-		call_frame_t *main_frame;
+                int (*fop) (call_frame_t *frame, xlator_t *this);
 
-		int (*fop) (call_frame_t *frame, xlator_t *this);
+                int (*done) (call_frame_t *frame, xlator_t *this);
 
-		int (*done) (call_frame_t *frame, xlator_t *this);
+                int (*resume) (call_frame_t *frame, xlator_t *this);
 
-		int (*resume) (call_frame_t *frame, xlator_t *this);
+                int (*unwind) (call_frame_t *frame, xlator_t *this);
 
-		int (*unwind) (call_frame_t *frame, xlator_t *this);
-	} transaction;
+                /* post-op hook */
+        } transaction;
 
-	afr_self_heal_t self_heal;
+        afr_self_heal_t self_heal;
+
+        struct marker_str     marker;
+
+        /* extra data for fops */
+        dict_t         *xdata_req;
+        dict_t         *xdata_rsp;
+
+        mode_t          umask;
+        int             xflag;
+        gf_boolean_t    do_discovery;
+	struct afr_reply *replies;
 } afr_local_t;
 
+typedef enum {
+        AFR_FD_NOT_OPENED,
+        AFR_FD_OPENED,
+        AFR_FD_OPENING
+} afr_fd_open_status_t;
+
+typedef struct {
+        unsigned int *pre_op_done;
+        afr_fd_open_status_t *opened_on; /* which subvolumes the fd is open on */
+        unsigned int *pre_op_piggyback;
+
+        unsigned int *lock_piggyback;
+        unsigned int *lock_acquired;
+
+        int flags;
+        uint64_t up_count;   /* number of CHILD_UPs this fd has seen */
+        uint64_t down_count; /* number of CHILD_DOWNs this fd has seen */
+
+        int32_t last_tried;
+
+        int  hit, miss;
+        gf_boolean_t failed_over;
+        struct list_head entries; /* needed for readdir failover */
+
+        unsigned char *locked_on; /* which subvolumes locks have been successful */
+
+	/* used for delayed-post-op optimization */
+	pthread_mutex_t    delay_lock;
+	gf_timer_t        *delay_timer;
+	call_frame_t      *delay_frame;
+        int               call_child;
+
+	/* set if any write on this fd was a non stable write
+	   (i.e, without O_SYNC or O_DSYNC)
+	*/
+	gf_boolean_t      witnessed_unstable_write;
+
+	/* list of frames currently in progress */
+	struct list_head  eager_locked;
+} afr_fd_ctx_t;
+
 
 /* try alloc and if it fails, goto label */
-#define ALLOC_OR_GOTO(var, type, label) do {			\
-		var = CALLOC (sizeof (type), 1);		\
-		if (!var) {					\
-			gf_log (this->name, GF_LOG_ERROR,	\
-				"out of memory :(");		\
-			op_errno = ENOMEM;			\
-			goto label;				\
-		}						\
-	} while (0);
+#define AFR_LOCAL_ALLOC_OR_GOTO(var, label) do {                    \
+                var = mem_get0 (THIS->local_pool);                  \
+                if (!var) {                                         \
+                        gf_log (this->name, GF_LOG_ERROR,           \
+                                "out of memory :(");                \
+                        op_errno = ENOMEM;                          \
+                        goto label;                                 \
+                }                                                   \
+        } while (0);
 
 
 /* did a call fail due to a child failing? */
-#define child_went_down(op_ret, op_errno) (((op_ret) < 0) &&	      \
-					   ((op_errno == ENOTCONN) || \
-					    (op_errno == EBADFD)))
+#define child_went_down(op_ret, op_errno) (((op_ret) < 0) &&            \
+                                           ((op_errno == ENOTCONN) ||   \
+                                            (op_errno == EBADFD)))
+
+#define afr_fop_failed(op_ret, op_errno) ((op_ret) == -1)
 
 /* have we tried all children? */
 #define all_tried(i, count)  ((i) == (count) - 1)
 
-uint64_t
-afr_read_child (xlator_t *this, inode_t *inode);
+int32_t
+afr_set_dict_gfid (dict_t *dict, uuid_t gfid);
+
+int
+pump_command_reply (call_frame_t *frame, xlator_t *this);
+
+int32_t
+afr_notify (xlator_t *this, int32_t event, void *data, void *data2);
+
+int
+afr_init_entry_lockee (afr_entry_lockee_t *lockee, afr_local_t *local,
+                       loc_t *loc, char *basename, int child_count);
 
 void
-afr_set_read_child (xlator_t *this, inode_t *inode, int32_t read_child);
+afr_entry_lockee_cleanup (afr_internal_lock_t *int_lock);
+
+int
+afr_attempt_lock_recovery (xlator_t *this, int32_t child_index);
+
+int
+afr_save_locked_fd (xlator_t *this, fd_t *fd);
+
+int
+afr_mark_locked_nodes (xlator_t *this, fd_t *fd,
+                       unsigned char *locked_nodes);
 
 void
-afr_build_parent_loc (loc_t *parent, loc_t *child);
+afr_set_lk_owner (call_frame_t *frame, xlator_t *this, void *lk_owner);
 
 int
-afr_up_children_count (int child_count, unsigned char *child_up);
+afr_set_lock_number (call_frame_t *frame, xlator_t *this);
+
+
+loc_t *
+lower_path (loc_t *l1, const char *b1, loc_t *l2, const char *b2);
+
+int32_t
+afr_unlock (call_frame_t *frame, xlator_t *this);
 
 int
-afr_locked_nodes_count (unsigned char *locked_nodes, int child_count);
+afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this);
+
+int
+afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this);
+
+int
+afr_blocking_lock (call_frame_t *frame, xlator_t *this);
+
+int
+afr_internal_lock_finish (call_frame_t *frame, xlator_t *this);
+
+int
+afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src, char *dom,
+                          unsigned int child_count);
+
+int pump_start (call_frame_t *frame, xlator_t *this);
+
+int
+__afr_fd_ctx_set (xlator_t *this, fd_t *fd);
 
 int
-afr_first_up_child (afr_private_t *priv);
+afr_fd_ctx_set (xlator_t *this, fd_t *fd);
 
-ino64_t
-afr_itransform (ino64_t ino, int child_count, int child_index);
+int32_t
+afr_inode_get_read_ctx (xlator_t *this, inode_t *inode, int32_t *fresh_children);
+
+void
+afr_inode_set_read_ctx (xlator_t *this, inode_t *inode, int32_t read_child,
+                        int32_t *fresh_children);
 
 int
-afr_deitransform (ino64_t ino, int child_count);
+afr_build_parent_loc (loc_t *parent, loc_t *child, int32_t *op_errno);
+
+unsigned int
+afr_up_children_count (unsigned char *child_up, unsigned int child_count);
+
+unsigned int
+afr_locked_children_count (unsigned char *children, unsigned int child_count);
+
+unsigned int
+afr_pre_op_done_children_count (unsigned char *pre_op,
+                                unsigned int child_count);
+
+gf_boolean_t
+afr_is_fresh_lookup (loc_t *loc, xlator_t *this);
+
+void
+afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *postparent);
+
+int
+afr_locked_nodes_count (unsigned char *locked_nodes, int child_count);
 
 void
 afr_local_cleanup (afr_local_t *local, xlator_t *this);
@@ -444,91 +937,276 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this);
 int
 afr_frame_return (call_frame_t *frame);
 
-#define AFR_STACK_UNWIND(frame, params ...)		\
-	do {						\
-		afr_local_t *__local = NULL;		\
-		xlator_t    *__this = NULL;		\
-		__local = frame->local;			\
-		__this = frame->this;			\
-		frame->local = NULL;                    \
-		STACK_UNWIND (frame, params);		\
-		afr_local_cleanup (__local, __this);	\
-		free (__local);				\
-} while (0);					
-
-#define AFR_STACK_DESTROY(frame)			\
-	do {						\
-		afr_local_t *__local = NULL;		\
-		xlator_t    *__this = NULL;		\
-		__local = frame->local;			\
-		__this = frame->this;			\
-		frame->local = NULL;                    \
-		STACK_DESTROY (frame->root);		\
-		afr_local_cleanup (__local, __this);	\
-		free (__local);				\
-} while (0);					
+gf_boolean_t
+afr_is_split_brain (xlator_t *this, inode_t *inode);
+
+void
+afr_set_split_brain (xlator_t *this, inode_t *inode, afr_spb_state_t mdata_spb,
+                     afr_spb_state_t data_spb);
 
+int
+afr_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          fd_t *fd, dict_t *xdata);
+
+void
+afr_set_opendir_done (xlator_t *this, inode_t *inode);
+
+gf_boolean_t
+afr_is_opendir_done (xlator_t *this, inode_t *inode);
+
+void
+afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this);
+
+int
+afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd);
+
+int
+afr_launch_openfd_self_heal (call_frame_t *frame, xlator_t *this, fd_t *fd);
+
+#define AFR_STACK_UNWIND(fop, frame, params ...)                \
+        do {                                                    \
+                afr_local_t *__local = NULL;                    \
+                xlator_t    *__this = NULL;                     \
+                if (frame) {                                    \
+                        __local = frame->local;                 \
+                        __this = frame->this;                   \
+                        frame->local = NULL;                    \
+                }                                               \
+                STACK_UNWIND_STRICT (fop, frame, params);       \
+                if (__local) {                                  \
+                        afr_local_cleanup (__local, __this);    \
+                        mem_put (__local);                      \
+                }                                               \
+        } while (0)
+
+#define AFR_STACK_DESTROY(frame)                                \
+        do {                                                    \
+                afr_local_t *__local = NULL;                    \
+                xlator_t    *__this = NULL;                     \
+                __local = frame->local;                         \
+                __this = frame->this;                           \
+                frame->local = NULL;                            \
+                STACK_DESTROY (frame->root);                    \
+                if (__local) {                                  \
+                        afr_local_cleanup (__local, __this);    \
+                        mem_put (__local);                      \
+                }                                               \
+        } while (0);
+
+#define AFR_NUM_CHANGE_LOGS            3 /*data + metadata + entry*/
 /* allocate and return a string that is the basename of argument */
-static inline char * 
-AFR_BASENAME (const char *str)						
+static inline char *
+AFR_BASENAME (const char *str)
 {
-	char *__tmp_str = NULL;				
-	char *__basename_str = NULL;			
-	__tmp_str = strdup (str);			
-	__basename_str = strdup (basename (__tmp_str));	
-	FREE (__tmp_str);
-	return __basename_str;
+        char *__tmp_str = NULL;
+        char *__basename_str = NULL;
+        __tmp_str = gf_strdup (str);
+        __basename_str = gf_strdup (basename (__tmp_str));
+        GF_FREE (__tmp_str);
+        return __basename_str;
 }
 
-/* initialize local_t */
-static inline int
-AFR_LOCAL_INIT (afr_local_t *local, afr_private_t *priv)
-{
-	local->child_up = CALLOC (sizeof (*local->child_up),
-				  priv->child_count);
-	if (!local->child_up) {
-		return -ENOMEM;
-	}
+int
+afr_transaction_local_init (afr_local_t *local, xlator_t *this);
 
-	memcpy (local->child_up, priv->child_up, 
-		sizeof (*local->child_up) * priv->child_count);
+int32_t
+afr_marker_getxattr (call_frame_t *frame, xlator_t *this,
+                     loc_t *loc, const char *name,afr_local_t *local, afr_private_t *priv );
 
+int32_t *
+afr_children_create (int32_t child_count);
 
-	local->call_count = afr_up_children_count (priv->child_count, local->child_up);
-	if (local->call_count == 0)
-		return -ENOTCONN;
+int
+afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno);
 
-	local->transaction.erase_pending = 1;
+int
+afr_internal_lock_init (afr_internal_lock_t *lk, size_t child_count,
+                        transaction_lk_type_t lk_type);
 
-	local->op_ret = -1;
-	local->op_errno = EUCLEAN;
+int
+afr_first_up_child (unsigned char *child_up, size_t child_count);
 
-	return 0;
-}
+int
+afr_select_read_child_from_policy (int32_t *fresh_children, int32_t child_count,
+                                   int32_t prev_read_child,
+                                   int32_t config_read_child, int32_t *sources,
+                                   unsigned int hmode, uuid_t gfid);
 
+void
+afr_set_read_ctx_from_policy (xlator_t *this, inode_t *inode,
+                              int32_t *fresh_children, int32_t prev_read_child,
+                              int32_t config_read_child, uuid_t gfid);
+
+int32_t
+afr_get_call_child (xlator_t *this, unsigned char *child_up, int32_t read_child,
+                    int32_t *fresh_children,
+                    int32_t *call_child, int32_t *last_index);
+
+int32_t
+afr_next_call_child (int32_t *fresh_children, unsigned char *child_up,
+                     size_t child_count, int32_t *last_index,
+                     int32_t read_child);
+void
+afr_get_fresh_children (int32_t *success_children, int32_t *sources,
+                        int32_t *children, unsigned int child_count);
+void
+afr_children_add_child (int32_t *children, int32_t child,
+                              int32_t child_count);
+void
+afr_children_rm_child (int32_t *children, int32_t child,
+                             int32_t child_count);
+void
+afr_reset_children (int32_t *children, int32_t child_count);
+int32_t
+afr_most_important_error(int32_t old_errno, int32_t new_errno,
+			 gf_boolean_t eio);
+int
+afr_errno_count (int32_t *children, int *child_errno,
+                 unsigned int child_count, int32_t op_errno);
+int
+afr_get_children_count (int32_t *children, unsigned int child_count);
+gf_boolean_t
+afr_is_child_present (int32_t *success_children, int32_t child_count,
+                      int32_t child);
+void
+afr_update_gfid_from_iatts (uuid_t uuid, struct iatt *bufs,
+                            int32_t *success_children,
+                            unsigned int child_count);
+void
+afr_reset_xattr (dict_t **xattr, unsigned int child_count);
+gf_boolean_t
+afr_conflicting_iattrs (struct iatt *bufs, int32_t *success_children,
+                        unsigned int child_count, const char *path,
+                        const char *xlator_name);
+unsigned int
+afr_gfid_missing_count (const char *xlator_name, int32_t *children,
+                        struct iatt *bufs, unsigned int child_count,
+                        const char *path);
+void
+afr_xattr_req_prepare (xlator_t *this, dict_t *xattr_req, const char *path);
+void
+afr_children_copy (int32_t *dst, int32_t *src, unsigned int child_count);
+afr_transaction_type
+afr_transaction_type_get (ia_type_t ia_type);
+int32_t
+afr_resultant_errno_get (int32_t *children,
+                         int *child_errno, unsigned int child_count);
+void
+afr_inode_rm_stale_children (xlator_t *this, inode_t *inode,
+                             int32_t *stale_children);
+void
+afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode,
+                      gf_boolean_t background, ia_type_t ia_type, char *reason,
+                      void (*gfid_sh_success_cbk) (call_frame_t *sh_frame,
+                                                   xlator_t *this),
+                      int (*unwind) (call_frame_t *frame, xlator_t *this,
+                                     int32_t op_ret, int32_t op_errno,
+                                     int32_t sh_failed));
+void
+afr_fix_open (xlator_t *this, fd_t *fd, size_t need_open_count, int *need_open);
 
-static inline int
-afr_transaction_local_init (afr_local_t *local, afr_private_t *priv)
-{
-	local->child_errno = CALLOC (sizeof (*local->child_errno),
-				     priv->child_count);
-	if (!local->child_errno) {
-		return -ENOMEM;
-	}
+void
+afr_open_fd_fix (fd_t *fd, xlator_t *this);
+int
+afr_set_elem_count_get (unsigned char *elems, int child_count);
 
-	local->pending_array = CALLOC (sizeof (*local->pending_array),
-				       priv->child_count);
-	if (!local->pending_array) {
-		return -ENOMEM;
-	}
+afr_fd_ctx_t *
+afr_fd_ctx_get (fd_t *fd, xlator_t *this);
 
-	local->transaction.locked_nodes = CALLOC (sizeof (*local->transaction.locked_nodes),
-						  priv->child_count);
+gf_boolean_t
+afr_open_only_data_self_heal (char *data_self_heal);
 
-	local->transaction.child_errno = CALLOC (sizeof (*local->transaction.child_errno),
-						  priv->child_count);
+gf_boolean_t
+afr_data_self_heal_enabled (char *data_self_heal);
 
-	return 0;
-}
+void
+afr_set_low_priority (call_frame_t *frame);
+int
+afr_child_fd_ctx_set (xlator_t *this, fd_t *fd, int32_t child,
+                      int flags);
+
+gf_boolean_t
+afr_have_quorum (char *logname, afr_private_t *priv);
+
+void
+afr_matrix_cleanup (int32_t **pending, unsigned int m);
+
+int32_t**
+afr_matrix_create (unsigned int m, unsigned int n);
+
+gf_boolean_t
+afr_is_errno_set (int *child_errno, int child);
+
+gf_boolean_t
+afr_is_errno_unset (int *child_errno, int child);
+
+gf_boolean_t
+afr_is_fd_fixable (fd_t *fd);
+
+void
+afr_prepare_new_entry_pending_matrix (int32_t **pending,
+                                      gf_boolean_t (*is_pending) (int *, int),
+                                      int *ctx, struct iatt *buf,
+                                      unsigned int child_count);
+void
+afr_xattr_array_destroy (dict_t **xattr, unsigned int child_count);
+/*
+ * Special value indicating we should use the "auto" quorum method instead of
+ * a fixed value (including zero to turn off quorum enforcement).
+ */
+#define AFR_QUORUM_AUTO INT_MAX
+
+/*
+ * Having this as a macro will make debugging a bit weirder, but does reduce
+ * the probability of functions handling this check inconsistently.
+ */
+#define QUORUM_CHECK(_func,_label) do {                                  \
+        if (priv->quorum_count && !afr_have_quorum(this->name,priv)) { \
+                gf_log(this->name,GF_LOG_WARNING,                        \
+                       "failing "#_func" due to lack of quorum");        \
+                op_errno = EROFS;                                        \
+                goto _label;                                             \
+        }                                                                \
+} while (0);
+
+
+#define AFR_SBRAIN_MSG "Failed on %s as split-brain is seen. Returning EIO."
+
+#define AFR_SBRAIN_CHECK_FD(fd, label) do {                              \
+        if (fd->inode && afr_is_split_brain (this, fd->inode)) {        \
+                op_errno = EIO;                                         \
+                gf_log (this->name, GF_LOG_WARNING,                     \
+                        AFR_SBRAIN_MSG ,uuid_utoa (fd->inode->gfid));   \
+                goto label;                                             \
+        }                                                               \
+} while (0)
+
+#define AFR_SBRAIN_CHECK_LOC(loc, label) do {                           \
+        if (loc->inode && afr_is_split_brain (this, loc->inode)) {      \
+                op_errno = EIO;                                         \
+                loc_path (loc, NULL);                                   \
+                gf_log (this->name, GF_LOG_WARNING,                     \
+                        AFR_SBRAIN_MSG , loc->path);                    \
+                goto label;                                             \
+        }                                                               \
+} while (0)
+
+int
+afr_fd_report_unstable_write (xlator_t *this, fd_t *fd);
+
+gf_boolean_t
+afr_fd_has_witnessed_unstable_write (xlator_t *this, fd_t *fd);
+
+void
+afr_delayed_changelog_wake_resume (xlator_t *this, fd_t *fd, call_stub_t *stub);
+
+int
+afr_inodelk_init (afr_inodelk_t *lk, char *dom, size_t child_count);
+
+void
+afr_handle_open_fd_count (call_frame_t *frame, xlator_t *this);
+
+afr_inode_ctx_t*
+afr_inode_ctx_get (inode_t *inode, xlator_t *this);
 
 #endif /* __AFR_H__ */
diff --git a/xlators/cluster/afr/src/pump.c b/xlators/cluster/afr/src/pump.c
new file mode 100644
index 000000000..a7f72fb30
--- /dev/null
+++ b/xlators/cluster/afr/src/pump.c
@@ -0,0 +1,2641 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <unistd.h>
+#include <sys/time.h>
+#include <stdlib.h>
+#include <fnmatch.h>
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "afr-common.c"
+#include "defaults.c"
+#include "glusterfs.h"
+
+static uint64_t pump_pid = 0;
+static inline void
+pump_fill_loc_info (loc_t *loc, struct iatt *iatt, struct iatt *parent)
+{
+        afr_update_loc_gfids (loc, iatt, parent);
+        uuid_copy (loc->inode->gfid, iatt->ia_gfid);
+}
+
+static int
+pump_mark_start_pending (xlator_t *this)
+{
+        afr_private_t  *priv      = NULL;
+        pump_private_t *pump_priv = NULL;
+
+        priv      = this->private;
+        pump_priv = priv->pump_private;
+
+        pump_priv->pump_start_pending = 1;
+
+        return 0;
+}
+
+static int
+is_pump_start_pending (xlator_t *this)
+{
+        afr_private_t  *priv      = NULL;
+        pump_private_t *pump_priv = NULL;
+
+        priv      = this->private;
+        pump_priv = priv->pump_private;
+
+        return (pump_priv->pump_start_pending);
+}
+
+static int
+pump_remove_start_pending (xlator_t *this)
+{
+        afr_private_t  *priv      = NULL;
+        pump_private_t *pump_priv = NULL;
+
+        priv      = this->private;
+        pump_priv = priv->pump_private;
+
+        pump_priv->pump_start_pending = 0;
+
+        return 0;
+}
+
+static pump_state_t
+pump_get_state ()
+{
+        xlator_t *this = NULL;
+        afr_private_t *priv = NULL;
+        pump_private_t *pump_priv = NULL;
+
+        pump_state_t ret;
+
+        this = THIS;
+        priv = this->private;
+        pump_priv = priv->pump_private;
+
+        LOCK (&pump_priv->pump_state_lock);
+        {
+                ret = pump_priv->pump_state;
+        }
+        UNLOCK (&pump_priv->pump_state_lock);
+
+        return ret;
+}
+
+int
+pump_change_state (xlator_t *this, pump_state_t state)
+{
+        afr_private_t *priv = NULL;
+        pump_private_t *pump_priv = NULL;
+
+        pump_state_t state_old;
+        pump_state_t state_new;
+
+
+        priv = this->private;
+        pump_priv = priv->pump_private;
+
+        GF_ASSERT (pump_priv);
+
+        LOCK (&pump_priv->pump_state_lock);
+        {
+                state_old = pump_priv->pump_state;
+                state_new = state;
+
+                pump_priv->pump_state = state;
+
+        }
+        UNLOCK (&pump_priv->pump_state_lock);
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Pump changing state from %d to %d",
+                state_old,
+                state_new);
+
+        return  0;
+}
+
+static int
+pump_set_resume_path (xlator_t *this, const char *path)
+{
+        int ret = 0;
+
+        afr_private_t *priv = NULL;
+        pump_private_t *pump_priv = NULL;
+
+        priv = this->private;
+        pump_priv = priv->pump_private;
+
+        GF_ASSERT (pump_priv);
+
+        LOCK (&pump_priv->resume_path_lock);
+        {
+                strncpy (pump_priv->resume_path, path, strlen (path) + 1);
+        }
+        UNLOCK (&pump_priv->resume_path_lock);
+
+        return ret;
+}
+
+static int
+pump_save_path (xlator_t *this, const char *path)
+{
+        afr_private_t *priv = NULL;
+        pump_state_t state;
+        dict_t *dict = NULL;
+        loc_t  loc = {0};
+        int dict_ret = 0;
+        int ret = -1;
+
+        state = pump_get_state ();
+        if (state == PUMP_STATE_RESUME)
+                return 0;
+
+        priv = this->private;
+
+        GF_ASSERT (priv->root_inode);
+
+        afr_build_root_loc (this, &loc);
+
+        dict = dict_new ();
+        dict_ret = dict_set_str (dict, PUMP_PATH, (char *)path);
+        if (dict_ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to set the key %s", path, PUMP_PATH);
+
+        ret = syncop_setxattr (PUMP_SOURCE_CHILD (this), &loc, dict, 0);
+
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "setxattr failed - could not save path=%s", path);
+        } else {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "setxattr succeeded - saved path=%s", path);
+        }
+
+        dict_unref (dict);
+
+        loc_wipe (&loc);
+        return 0;
+}
+
+static int
+pump_check_and_update_status (xlator_t *this)
+{
+        pump_state_t state;
+        int ret = -1;
+
+        state = pump_get_state ();
+
+        switch (state) {
+
+        case PUMP_STATE_RESUME:
+        case PUMP_STATE_RUNNING:
+        {
+                ret = 0;
+                break;
+        }
+        case PUMP_STATE_PAUSE:
+        {
+                ret = -1;
+                break;
+        }
+        case PUMP_STATE_ABORT:
+        {
+                pump_save_path (this, "/");
+                ret = -1;
+                break;
+        }
+        default:
+        {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Unknown pump state");
+                ret = -1;
+                break;
+        }
+
+        }
+
+        return ret;
+}
+
+static const char *
+pump_get_resume_path (xlator_t *this)
+{
+        afr_private_t *priv = NULL;
+        pump_private_t *pump_priv = NULL;
+
+        const char *resume_path = NULL;
+
+        priv = this->private;
+        pump_priv = priv->pump_private;
+
+        resume_path = pump_priv->resume_path;
+
+        return resume_path;
+}
+
+static int
+pump_update_resume_state (xlator_t *this, const char *path)
+{
+        pump_state_t state;
+        const char *resume_path = NULL;
+
+        state = pump_get_state ();
+
+        if (state == PUMP_STATE_RESUME) {
+                resume_path = pump_get_resume_path (this);
+                if (strcmp (resume_path, "/") == 0) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "Reached the resume path (/). Proceeding to change state"
+                                " to running");
+                        pump_change_state (this, PUMP_STATE_RUNNING);
+                } else if (strcmp (resume_path, path) == 0) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "Reached the resume path. Proceeding to change state"
+                                " to running");
+                        pump_change_state (this, PUMP_STATE_RUNNING);
+                } else {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "Not yet hit the resume path:res-path=%s,path=%s",
+                                resume_path, path);
+                }
+        }
+
+        return 0;
+}
+
+static gf_boolean_t
+is_pump_traversal_allowed (xlator_t *this, const char *path)
+{
+        pump_state_t state;
+        const char *resume_path = NULL;
+        gf_boolean_t ret = _gf_true;
+
+        state = pump_get_state ();
+
+        if (state == PUMP_STATE_RESUME) {
+                resume_path = pump_get_resume_path (this);
+                if (strstr (resume_path, path)) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "On the right path to resumption path");
+                        ret = _gf_true;
+                } else {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "Not the right path to resuming=> ignoring traverse");
+                        ret = _gf_false;
+                }
+        }
+
+        return ret;
+}
+
+static int
+pump_save_file_stats (xlator_t *this, const char *path)
+{
+        afr_private_t  *priv        = NULL;
+        pump_private_t *pump_priv   = NULL;
+
+        priv      = this->private;
+        pump_priv = priv->pump_private;
+
+        LOCK (&pump_priv->resume_path_lock);
+        {
+                pump_priv->number_files_pumped++;
+
+                strncpy (pump_priv->current_file, path,
+                         PATH_MAX);
+        }
+        UNLOCK (&pump_priv->resume_path_lock);
+
+        return 0;
+}
+
+static int
+gf_pump_traverse_directory (loc_t *loc)
+{
+        xlator_t        *this              = NULL;
+        fd_t            *fd                = NULL;
+        off_t           offset             = 0;
+        loc_t           entry_loc          = {0};
+        gf_dirent_t     *entry             = NULL;
+        gf_dirent_t     *tmp               = NULL;
+        gf_dirent_t     entries;
+	struct iatt     iatt               = {0};
+        struct iatt     parent             = {0};
+	dict_t          *xattr_rsp         = NULL;
+        int             ret                = 0;
+        gf_boolean_t    is_directory_empty = _gf_true;
+        gf_boolean_t    free_entries       = _gf_false;
+
+        INIT_LIST_HEAD (&entries.list);
+        this = THIS;
+
+        GF_ASSERT (loc->inode);
+
+	fd = fd_create (loc->inode, pump_pid);
+        if (!fd) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Failed to create fd for %s", loc->path);
+                goto out;
+        }
+
+        ret = syncop_opendir (this, loc, fd);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "opendir failed on %s", loc->path);
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "pump opendir on %s returned=%d",
+                loc->path, ret);
+
+        while (syncop_readdirp (this, fd, 131072, offset, NULL, &entries)) {
+                free_entries = _gf_true;
+
+                if (list_empty (&entries.list)) {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "no more entries in directory");
+                        goto out;
+                }
+
+                list_for_each_entry_safe (entry, tmp, &entries.list, list) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "found readdir entry=%s", entry->d_name);
+
+                        offset = entry->d_off;
+                        if (uuid_is_null (entry->d_stat.ia_gfid)) {
+                                gf_log (this->name, GF_LOG_WARNING, "%s/%s: No "
+                                        "gfid present skipping",
+                                        loc->path, entry->d_name);
+                                continue;
+                        }
+                        loc_wipe (&entry_loc);
+                        ret = afr_build_child_loc (this, &entry_loc, loc,
+                                                   entry->d_name);
+                        if (ret)
+                                goto out;
+
+                        if (!IS_ENTRY_CWD (entry->d_name) &&
+                            !IS_ENTRY_PARENT (entry->d_name)) {
+
+                                    is_directory_empty = _gf_false;
+                                    gf_log (this->name, GF_LOG_DEBUG,
+                                            "lookup %s => %"PRId64,
+                                            entry_loc.path,
+                                            iatt.ia_ino);
+
+                                    ret = syncop_lookup (this, &entry_loc, NULL,
+                                                         &iatt, &xattr_rsp, &parent);
+
+                                    if (ret) {
+                                            gf_log (this->name, GF_LOG_ERROR,
+                                                    "%s: lookup failed",
+                                                    entry_loc.path);
+                                            continue;
+                                    }
+                                    pump_fill_loc_info (&entry_loc, &iatt,
+                                                       &parent);
+
+                                    pump_update_resume_state (this, entry_loc.path);
+
+                                    pump_save_path (this, entry_loc.path);
+                                    pump_save_file_stats (this, entry_loc.path);
+
+                                    ret = pump_check_and_update_status (this);
+                                    if (ret < 0) {
+                                            gf_log (this->name, GF_LOG_DEBUG,
+                                                    "Pump beginning to exit out");
+                                            goto out;
+                                    }
+
+                                    if (IA_ISDIR (iatt.ia_type)) {
+                                            if (is_pump_traversal_allowed (this, entry_loc.path)) {
+                                                    gf_log (this->name, GF_LOG_TRACE,
+                                                            "entering dir=%s",
+                                                            entry->d_name);
+                                                    gf_pump_traverse_directory (&entry_loc);
+                                            }
+                                    }
+                        }
+                }
+
+                gf_dirent_free (&entries);
+                free_entries = _gf_false;
+                gf_log (this->name, GF_LOG_TRACE,
+                        "offset incremented to %d",
+                        (int32_t ) offset);
+
+        }
+
+        ret = syncop_close (fd);
+        if (ret < 0)
+                gf_log (this->name, GF_LOG_DEBUG, "closing the fd failed");
+
+        if (is_directory_empty && IS_ROOT_PATH (loc->path)) {
+               pump_change_state (this, PUMP_STATE_RUNNING);
+               gf_log (this->name, GF_LOG_INFO, "Empty source brick. "
+                                "Nothing to be done.");
+        }
+
+out:
+        if (entry_loc.path)
+                loc_wipe (&entry_loc);
+        if (free_entries)
+                gf_dirent_free (&entries);
+        return 0;
+}
+
+static int
+pump_update_resume_path (xlator_t *this)
+{
+        const char *resume_path = NULL;
+
+        resume_path = pump_get_resume_path (this);
+
+        if (resume_path) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Found a path to resume from: %s",
+                        resume_path);
+
+        }else {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Did not find a path=> setting to '/'");
+                pump_set_resume_path (this, "/");
+        }
+
+        pump_change_state (this, PUMP_STATE_RESUME);
+
+        return 0;
+}
+
+static int32_t
+pump_xattr_cleaner (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+        afr_private_t  *priv      = NULL;
+        loc_t           loc       = {0};
+        int             i         = 0;
+        int             ret       = 0;
+        int             source    = 0;
+        int             sink      = 1;
+
+        priv      = this->private;
+
+        afr_build_root_loc (this, &loc);
+
+        ret = syncop_removexattr (priv->children[source], &loc,
+                                          PUMP_PATH);
+
+        ret = syncop_removexattr (priv->children[sink], &loc,
+                                  PUMP_SINK_COMPLETE);
+
+        for (i = 0; i < priv->child_count; i++) {
+                ret = syncop_removexattr (priv->children[i], &loc,
+                                          PUMP_SOURCE_COMPLETE);
+                if (ret)
+                        gf_log (this->name, GF_LOG_DEBUG, "removexattr "
+                                "failed with %s", strerror (errno));
+        }
+
+        loc_wipe (&loc);
+        return pump_command_reply (frame, this);
+}
+
+static int
+pump_complete_migration (xlator_t *this)
+{
+        afr_private_t *priv = NULL;
+        pump_private_t *pump_priv = NULL;
+        dict_t *dict = NULL;
+        pump_state_t state;
+        loc_t  loc = {0};
+        int dict_ret = 0;
+        int ret = -1;
+
+        priv = this->private;
+        pump_priv = priv->pump_private;
+
+        GF_ASSERT (priv->root_inode);
+
+        afr_build_root_loc (this, &loc);
+
+        dict = dict_new ();
+
+        state = pump_get_state ();
+        if (state == PUMP_STATE_RUNNING) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Pump finished pumping");
+
+                pump_priv->pump_finished = _gf_true;
+
+                dict_ret = dict_set_str (dict, PUMP_SOURCE_COMPLETE, "jargon");
+                if (dict_ret)
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: failed to set the key %s",
+                                loc.path, PUMP_SOURCE_COMPLETE);
+
+                ret = syncop_setxattr (PUMP_SOURCE_CHILD (this), &loc, dict, 0);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "setxattr failed - while  notifying source complete");
+                }
+                dict_ret = dict_set_str (dict, PUMP_SINK_COMPLETE, "jargon");
+                if (dict_ret)
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: failed to set the key %s",
+                                loc.path, PUMP_SINK_COMPLETE);
+
+                ret = syncop_setxattr (PUMP_SINK_CHILD (this), &loc, dict, 0);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "setxattr failed - while notifying sink complete");
+                }
+
+                pump_save_path (this, "/");
+
+        } else if (state == PUMP_STATE_ABORT) {
+                gf_log (this->name, GF_LOG_DEBUG, "Starting cleanup "
+                        "of pump internal xattrs");
+                call_resume (pump_priv->cleaner);
+        }
+
+        loc_wipe (&loc);
+        return 0;
+}
+
+static int
+pump_lookup_sink (loc_t *loc)
+{
+        xlator_t *this = NULL;
+	struct iatt iatt, parent;
+	dict_t *xattr_rsp;
+        dict_t *xattr_req = NULL;
+        int ret = 0;
+
+        this = THIS;
+
+        xattr_req = dict_new ();
+
+        ret = afr_set_root_gfid (xattr_req);
+        if (ret)
+                goto out;
+
+        ret = syncop_lookup (PUMP_SINK_CHILD (this), loc,
+                             xattr_req, &iatt, &xattr_rsp, &parent);
+
+        if (ret) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Lookup on sink child failed");
+                goto out;
+        }
+
+out:
+        if (xattr_req)
+                dict_unref (xattr_req);
+
+        return ret;
+}
+
+static int
+pump_task (void *data)
+{
+	xlator_t *this = NULL;
+        afr_private_t *priv = NULL;
+
+
+        loc_t loc = {0};
+	struct iatt iatt, parent;
+	dict_t *xattr_rsp = NULL;
+        dict_t *xattr_req = NULL;
+
+        int ret = -1;
+
+        this = THIS;
+        priv = this->private;
+
+        GF_ASSERT (priv->root_inode);
+
+        afr_build_root_loc (this, &loc);
+        xattr_req = dict_new ();
+        if (!xattr_req) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Out of memory");
+                ret = -1;
+                goto out;
+        }
+
+        afr_set_root_gfid (xattr_req);
+        ret = syncop_lookup (this, &loc, xattr_req,
+                             &iatt, &xattr_rsp, &parent);
+
+        gf_log (this->name, GF_LOG_TRACE,
+                "lookup: path=%s gfid=%s",
+                loc.path, uuid_utoa (loc.inode->gfid));
+
+        ret = pump_check_and_update_status (this);
+        if (ret < 0) {
+                goto out;
+        }
+
+        pump_update_resume_path (this);
+
+        afr_set_root_gfid (xattr_req);
+        ret = pump_lookup_sink (&loc);
+        if (ret) {
+                pump_update_resume_path (this);
+                goto out;
+        }
+
+        gf_pump_traverse_directory (&loc);
+
+        pump_complete_migration (this);
+out:
+        if (xattr_req)
+                dict_unref (xattr_req);
+
+        loc_wipe (&loc);
+	return 0;
+}
+
+
+static int
+pump_task_completion (int ret, call_frame_t *sync_frame, void *data)
+{
+        xlator_t *this = NULL;
+        afr_private_t *priv = NULL;
+
+        this = THIS;
+
+        priv = this->private;
+
+        inode_unref (priv->root_inode);
+        STACK_DESTROY (sync_frame->root);
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Pump xlator exiting");
+	return 0;
+}
+
+int
+pump_start (call_frame_t *pump_frame, xlator_t *this)
+{
+	afr_private_t *priv = NULL;
+	pump_private_t *pump_priv = NULL;
+
+	int ret = -1;
+
+	priv = this->private;
+        pump_priv = priv->pump_private;
+
+        afr_set_lk_owner (pump_frame, this, pump_frame->root);
+	pump_pid = (uint64_t) (unsigned long)pump_frame->root;
+
+	ret = synctask_new (pump_priv->env, pump_task,
+                            pump_task_completion,
+                            pump_frame, NULL);
+        if (ret == -1) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "starting pump failed");
+                pump_change_state (this, PUMP_STATE_ABORT);
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "setting pump as started lk_owner: %s %"PRIu64,
+                lkowner_utoa (&pump_frame->root->lk_owner), pump_pid);
+
+        priv->use_afr_in_pump = 1;
+out:
+	return ret;
+}
+
+static int
+pump_start_synctask (xlator_t *this)
+{
+        call_frame_t *frame = NULL;
+        int ret = 0;
+
+        frame = create_frame (this, this->ctx->pool);
+        if (!frame) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Out of memory");
+                ret = -1;
+                goto out;
+        }
+
+        pump_change_state (this, PUMP_STATE_RUNNING);
+
+        ret = pump_start (frame, this);
+
+out:
+        return ret;
+}
+
+int32_t
+pump_cmd_start_setxattr_cbk (call_frame_t *frame,
+                             void *cookie,
+                             xlator_t *this,
+                             int32_t op_ret,
+                             int32_t op_errno, dict_t *xdata)
+
+{
+        call_frame_t *prev = NULL;
+        afr_local_t *local = NULL;
+        int ret = 0;
+
+        local = frame->local;
+
+        if (op_ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Could not initiate destination "
+                        "brick connect");
+                ret = op_ret;
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Successfully initiated destination "
+                "brick connect");
+
+        pump_mark_start_pending (this);
+
+        /* send the PARENT_UP as pump is ready now */
+        prev = cookie;
+        if (prev && prev->this)
+                prev->this->notify (prev->this, GF_EVENT_PARENT_UP, this);
+
+out:
+        local->op_ret = ret;
+        pump_command_reply (frame, this);
+
+        return 0;
+}
+
+static int
+pump_initiate_sink_connect (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local     = NULL;
+        afr_private_t *priv      = NULL;
+        dict_t        *dict      = NULL;
+        data_t        *data      = NULL;
+        char          *clnt_cmd  = NULL;
+        loc_t loc = {0};
+
+        int ret = 0;
+
+        priv  = this->private;
+        local = frame->local;
+
+        GF_ASSERT (priv->root_inode);
+
+        afr_build_root_loc (this, &loc);
+
+        data = data_ref (dict_get (local->dict, RB_PUMP_CMD_START));
+        if (!data) {
+                ret = -1;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Could not get destination brick value");
+                goto out;
+        }
+
+        dict = dict_new ();
+        if (!dict) {
+                ret = -1;
+                goto out;
+        }
+
+        clnt_cmd = GF_CALLOC (1, data->len+1, gf_common_mt_char);
+        if (!clnt_cmd) {
+                ret = -1;
+                goto out;
+        }
+
+        memcpy (clnt_cmd, data->data, data->len);
+        clnt_cmd[data->len] = '\0';
+        gf_log (this->name, GF_LOG_DEBUG, "Got destination brick %s\n",
+                        clnt_cmd);
+
+        ret = dict_set_dynstr (dict, CLIENT_CMD_CONNECT, clnt_cmd);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Could not inititiate destination brick "
+                        "connect");
+                goto out;
+        }
+
+	STACK_WIND (frame,
+		    pump_cmd_start_setxattr_cbk,
+		    PUMP_SINK_CHILD(this),
+		    PUMP_SINK_CHILD(this)->fops->setxattr,
+		    &loc,
+		    dict,
+		    0, NULL);
+
+        ret = 0;
+
+out:
+        if (dict)
+                dict_unref (dict);
+
+        if (data)
+                data_unref (data);
+
+        if (ret && clnt_cmd)
+                GF_FREE (clnt_cmd);
+
+        loc_wipe (&loc);
+        return ret;
+}
+
+static int
+is_pump_aborted (xlator_t *this)
+{
+        pump_state_t state;
+
+        state = pump_get_state ();
+
+        return ((state == PUMP_STATE_ABORT));
+}
+
+int32_t
+pump_cmd_start_getxattr_cbk (call_frame_t *frame,
+                             void *cookie,
+                             xlator_t *this,
+                             int32_t op_ret,
+                             int32_t op_errno,
+                             dict_t *dict, dict_t *xdata)
+{
+        afr_local_t *local = NULL;
+        char *path = NULL;
+
+        pump_state_t state;
+        int ret = 0;
+        int need_unwind = 0;
+        int dict_ret = -1;
+
+        local = frame->local;
+
+        if (op_ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "getxattr failed - changing pump "
+                        "state to RUNNING with '/'");
+                path = "/";
+                ret = op_ret;
+        } else {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "getxattr succeeded");
+
+                dict_ret =  dict_get_str (dict, PUMP_PATH, &path);
+                if (dict_ret < 0)
+                        path = "/";
+        }
+
+        state = pump_get_state ();
+        if ((state == PUMP_STATE_RUNNING) ||
+            (state == PUMP_STATE_RESUME)) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Pump is already started");
+                ret = -1;
+                goto out;
+        }
+
+        pump_set_resume_path (this, path);
+
+        if (is_pump_aborted (this))
+                /* We're re-starting pump afresh */
+                ret = pump_initiate_sink_connect (frame, this);
+        else {
+                /* We're re-starting pump from a previous
+                   pause */
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "about to start synctask");
+                ret = pump_start_synctask (this);
+                need_unwind = 1;
+        }
+
+out:
+        if ((ret < 0) || (need_unwind == 1)) {
+                local->op_ret = ret;
+                pump_command_reply (frame, this);
+        }
+	return 0;
+}
+
+int
+pump_execute_status (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t *priv = NULL;
+        pump_private_t *pump_priv = NULL;
+
+        uint64_t number_files = 0;
+
+        char filename[PATH_MAX];
+        char summary[PATH_MAX+256];
+        char *dict_str = NULL;
+
+        int32_t op_ret = 0;
+        int32_t op_errno = 0;
+
+        dict_t *dict = NULL;
+        int ret = -1;
+
+        priv = this->private;
+        pump_priv = priv->pump_private;
+
+        LOCK (&pump_priv->resume_path_lock);
+        {
+                number_files  = pump_priv->number_files_pumped;
+                strncpy (filename, pump_priv->current_file, PATH_MAX);
+        }
+        UNLOCK (&pump_priv->resume_path_lock);
+
+        dict_str     = GF_CALLOC (1, PATH_MAX + 256, gf_afr_mt_char);
+        if (!dict_str) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Out of memory");
+                op_ret = -1;
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        if (pump_priv->pump_finished) {
+                snprintf (summary, PATH_MAX+256,
+                          "no_of_files=%"PRIu64, number_files);
+        } else {
+                snprintf (summary, PATH_MAX+256,
+                          "no_of_files=%"PRIu64":current_file=%s",
+                          number_files, filename);
+        }
+        snprintf (dict_str, PATH_MAX+256, "status=%d:%s",
+                  (pump_priv->pump_finished)?1:0, summary);
+
+        dict = dict_new ();
+
+        ret = dict_set_dynstr (dict, RB_PUMP_CMD_STATUS, dict_str);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "dict_set_dynstr returned negative value");
+        } else {
+                dict_str = NULL;
+        }
+
+        op_ret = 0;
+
+out:
+
+        AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, NULL);
+
+        if (dict)
+                dict_unref (dict);
+
+        GF_FREE (dict_str);
+
+        return 0;
+}
+
+int
+pump_execute_pause (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+
+        local = frame->local;
+
+        pump_change_state (this, PUMP_STATE_PAUSE);
+
+        local->op_ret = 0;
+        pump_command_reply (frame, this);
+
+        return 0;
+}
+
+int
+pump_execute_start (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t *priv = NULL;
+        afr_local_t   *local = NULL;
+
+        int ret = 0;
+        loc_t loc = {0};
+
+        priv = this->private;
+        local = frame->local;
+
+        if (!priv->root_inode) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Pump xlator cannot be started without an initial "
+                        "lookup");
+                ret = -1;
+                goto out;
+        }
+
+        GF_ASSERT (priv->root_inode);
+
+        afr_build_root_loc (this, &loc);
+
+	STACK_WIND (frame,
+		    pump_cmd_start_getxattr_cbk,
+		    PUMP_SOURCE_CHILD(this),
+		    PUMP_SOURCE_CHILD(this)->fops->getxattr,
+		    &loc,
+		    PUMP_PATH, NULL);
+
+        ret = 0;
+
+out:
+        if (ret < 0) {
+                local->op_ret = ret;
+                pump_command_reply (frame, this);
+        }
+
+        loc_wipe (&loc);
+	return 0;
+}
+
+static int
+pump_cleanup_helper (void *data) {
+        call_frame_t *frame = data;
+
+        pump_xattr_cleaner (frame, 0, frame->this, 0, 0, NULL);
+
+        return 0;
+}
+
+static int
+pump_cleanup_done (int ret, call_frame_t *sync_frame, void *data)
+{
+        STACK_DESTROY (sync_frame->root);
+
+        return 0;
+}
+
+int
+pump_execute_commit (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t  *priv       = NULL;
+        pump_private_t *pump_priv  = NULL;
+        afr_local_t    *local      = NULL;
+        call_frame_t   *sync_frame = NULL;
+        int             ret        = 0;
+
+        priv      = this->private;
+        pump_priv = priv->pump_private;
+        local     = frame->local;
+
+        local->op_ret = 0;
+        if (pump_priv->pump_finished) {
+                pump_change_state (this, PUMP_STATE_COMMIT);
+                sync_frame = create_frame (this, this->ctx->pool);
+                ret = synctask_new (pump_priv->env, pump_cleanup_helper,
+                                    pump_cleanup_done, sync_frame, frame);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_DEBUG, "Couldn't create "
+                                "synctask for cleaning up xattrs.");
+                }
+
+        } else {
+                gf_log (this->name, GF_LOG_ERROR, "Commit can't proceed. "
+                        "Migration in progress");
+                local->op_ret = -1;
+                local->op_errno = EINPROGRESS;
+                pump_command_reply (frame, this);
+        }
+
+        return 0;
+}
+int
+pump_execute_abort (call_frame_t *frame, xlator_t *this)
+{
+        afr_private_t  *priv       = NULL;
+        pump_private_t *pump_priv  = NULL;
+        afr_local_t    *local      = NULL;
+        call_frame_t   *sync_frame = NULL;
+        int             ret        = 0;
+
+        priv      = this->private;
+        pump_priv = priv->pump_private;
+        local     = frame->local;
+
+        pump_change_state (this, PUMP_STATE_ABORT);
+
+        LOCK (&pump_priv->resume_path_lock);
+        {
+                pump_priv->number_files_pumped = 0;
+                pump_priv->current_file[0] = '\0';
+        }
+        UNLOCK (&pump_priv->resume_path_lock);
+
+        local->op_ret = 0;
+        if (pump_priv->pump_finished) {
+                sync_frame = create_frame (this, this->ctx->pool);
+                ret = synctask_new (pump_priv->env, pump_cleanup_helper,
+                                    pump_cleanup_done, sync_frame, frame);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_DEBUG, "Couldn't create "
+                                "synctask for cleaning up xattrs.");
+                }
+
+        } else {
+                pump_priv->cleaner = fop_setxattr_cbk_stub (frame,
+                                                            pump_xattr_cleaner,
+                                                            0, 0, NULL);
+        }
+
+        return 0;
+}
+
+gf_boolean_t
+pump_command_status (xlator_t *this, dict_t *dict)
+{
+        char *cmd = NULL;
+        int dict_ret = -1;
+        int ret = _gf_true;
+
+        dict_ret = dict_get_str (dict, RB_PUMP_CMD_STATUS, &cmd);
+        if (dict_ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Not a pump status command");
+                ret = _gf_false;
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Hit a pump command - status");
+        ret = _gf_true;
+
+out:
+        return ret;
+
+}
+
+gf_boolean_t
+pump_command_pause (xlator_t *this, dict_t *dict)
+{
+        char *cmd = NULL;
+        int dict_ret = -1;
+        int ret = _gf_true;
+
+        dict_ret = dict_get_str (dict, RB_PUMP_CMD_PAUSE, &cmd);
+        if (dict_ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Not a pump pause command");
+                ret = _gf_false;
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Hit a pump command - pause");
+        ret = _gf_true;
+
+out:
+        return ret;
+
+}
+
+gf_boolean_t
+pump_command_commit (xlator_t *this, dict_t *dict)
+{
+        char *cmd = NULL;
+        int dict_ret = -1;
+        int ret = _gf_true;
+
+        dict_ret = dict_get_str (dict, RB_PUMP_CMD_COMMIT, &cmd);
+        if (dict_ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Not a pump commit command");
+                ret = _gf_false;
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Hit a pump command - commit");
+        ret = _gf_true;
+
+out:
+        return ret;
+
+}
+
+gf_boolean_t
+pump_command_abort (xlator_t *this, dict_t *dict)
+{
+        char *cmd = NULL;
+        int dict_ret = -1;
+        int ret = _gf_true;
+
+        dict_ret = dict_get_str (dict, RB_PUMP_CMD_ABORT, &cmd);
+        if (dict_ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Not a pump abort command");
+                ret = _gf_false;
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Hit a pump command - abort");
+        ret = _gf_true;
+
+out:
+        return ret;
+
+}
+
+gf_boolean_t
+pump_command_start (xlator_t *this, dict_t *dict)
+{
+        char *cmd = NULL;
+        int dict_ret = -1;
+        int ret = _gf_true;
+
+        dict_ret = dict_get_str (dict, RB_PUMP_CMD_START, &cmd);
+        if (dict_ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "Not a pump start command");
+                ret = _gf_false;
+                goto out;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "Hit a pump command - start");
+        ret = _gf_true;
+
+out:
+        return ret;
+
+}
+
+struct _xattr_key {
+        char *key;
+        struct list_head list;
+};
+
+static int
+__gather_xattr_keys (dict_t *dict, char *key, data_t *value,
+                     void *data)
+{
+        struct list_head *  list  = data;
+        struct _xattr_key * xkey  = NULL;
+
+        if (!strncmp (key, AFR_XATTR_PREFIX,
+                      strlen (AFR_XATTR_PREFIX))) {
+
+                xkey = GF_CALLOC (1, sizeof (*xkey), gf_afr_mt_xattr_key);
+                if (!xkey)
+                        return -1;
+
+                xkey->key = key;
+                INIT_LIST_HEAD (&xkey->list);
+
+                list_add_tail (&xkey->list, list);
+        }
+        return 0;
+}
+
+static void
+__filter_xattrs (dict_t *dict)
+{
+        struct list_head keys;
+
+        struct _xattr_key *key;
+        struct _xattr_key *tmp;
+
+        INIT_LIST_HEAD (&keys);
+
+        dict_foreach (dict, __gather_xattr_keys,
+                      (void *) &keys);
+
+        list_for_each_entry_safe (key, tmp, &keys, list) {
+                dict_del (dict, key->key);
+
+                list_del_init (&key->list);
+
+                GF_FREE (key);
+        }
+}
+
+int32_t
+pump_getxattr_cbk (call_frame_t *frame, void *cookie,
+		  xlator_t *this, int32_t op_ret, int32_t op_errno,
+		  dict_t *dict, dict_t *xdata)
+{
+	afr_private_t   *priv           = NULL;
+	afr_local_t     *local          = NULL;
+	xlator_t        **children      = NULL;
+	int             unwind          = 1;
+        int32_t         *last_index     = NULL;
+        int32_t         next_call_child = -1;
+        int32_t         read_child      = -1;
+        int32_t         *fresh_children = NULL;
+
+
+	priv     = this->private;
+	children = priv->children;
+
+	local = frame->local;
+
+        read_child = (long) cookie;
+
+	if (op_ret == -1) {
+		last_index = &local->cont.getxattr.last_index;
+                fresh_children = local->fresh_children;
+                next_call_child = afr_next_call_child (fresh_children,
+                                                       local->child_up,
+                                                       priv->child_count,
+                                                       last_index, read_child);
+                if (next_call_child < 0)
+                        goto out;
+
+		unwind = 0;
+		STACK_WIND_COOKIE (frame, pump_getxattr_cbk,
+				   (void *) (long) read_child,
+				   children[next_call_child],
+				   children[next_call_child]->fops->getxattr,
+				   &local->loc,
+				   local->cont.getxattr.name, NULL);
+	}
+
+out:
+	if (unwind) {
+                if (op_ret >= 0 && dict)
+                        __filter_xattrs (dict);
+
+		AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, NULL);
+	}
+
+	return 0;
+}
+
+int32_t
+pump_getxattr (call_frame_t *frame, xlator_t *this,
+	      loc_t *loc, const char *name, dict_t *xdata)
+{
+	afr_private_t *   priv       = NULL;
+	xlator_t **       children   = NULL;
+	int               call_child = 0;
+	afr_local_t       *local     = NULL;
+	int32_t           ret     = -1;
+	int32_t           op_errno   = 0;
+        uint64_t          read_child = 0;
+
+
+	VALIDATE_OR_GOTO (frame, out);
+	VALIDATE_OR_GOTO (this, out);
+	VALIDATE_OR_GOTO (this->private, out);
+
+	priv     = this->private;
+	VALIDATE_OR_GOTO (priv->children, out);
+
+	children = priv->children;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame, default_getxattr_cbk,
+                            FIRST_CHILD (this),
+                            (FIRST_CHILD (this))->fops->getxattr,
+                            loc, name, xdata);
+                return 0;
+        }
+
+
+	AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
+	local = frame->local;
+
+        ret = afr_local_init (local, priv, &op_errno);
+        if (ret < 0)
+                goto out;
+
+        if (name) {
+                if (!strncmp (name, AFR_XATTR_PREFIX,
+                              strlen (AFR_XATTR_PREFIX))) {
+
+                        op_errno = ENODATA;
+                        goto out;
+                }
+
+                if (!strcmp (name, RB_PUMP_CMD_STATUS)) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "Hit pump command - status");
+                        pump_execute_status (frame, this);
+                        ret = 0;
+                        goto out;
+                }
+        }
+
+        local->fresh_children = GF_CALLOC (priv->child_count,
+                                          sizeof (*local->fresh_children),
+                                          gf_afr_mt_int32_t);
+        if (!local->fresh_children) {
+                ret = -1;
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        read_child = afr_inode_get_read_ctx (this, loc->inode, local->fresh_children);
+        ret = afr_get_call_child (this, local->child_up, read_child,
+                                     local->fresh_children,
+                                     &call_child,
+                                     &local->cont.getxattr.last_index);
+        if (ret < 0) {
+                op_errno = -ret;
+                goto out;
+        }
+	loc_copy (&local->loc, loc);
+	if (name)
+	  local->cont.getxattr.name       = gf_strdup (name);
+
+	STACK_WIND_COOKIE (frame, pump_getxattr_cbk,
+			   (void *) (long) call_child,
+			   children[call_child], children[call_child]->fops->getxattr,
+			   loc, name, xdata);
+
+	ret = 0;
+out:
+	if (ret < 0)
+		AFR_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
+	return 0;
+}
+
+static int
+afr_setxattr_unwind (call_frame_t *frame, xlator_t *this)
+{
+	afr_local_t *   local = NULL;
+	call_frame_t   *main_frame = NULL;
+
+	local = frame->local;
+
+	LOCK (&frame->lock);
+	{
+		if (local->transaction.main_frame)
+			main_frame = local->transaction.main_frame;
+		local->transaction.main_frame = NULL;
+	}
+	UNLOCK (&frame->lock);
+
+	if (main_frame) {
+		AFR_STACK_UNWIND (setxattr, main_frame,
+                                  local->op_ret, local->op_errno, NULL);
+	}
+	return 0;
+}
+
+static int
+afr_setxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+		       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+	afr_local_t *   local = NULL;
+	afr_private_t * priv  = NULL;
+
+	int call_count  = -1;
+	int need_unwind = 0;
+
+	local = frame->local;
+	priv = this->private;
+
+	LOCK (&frame->lock);
+	{
+		if (op_ret != -1) {
+			if (local->success_count == 0) {
+				local->op_ret = op_ret;
+			}
+			local->success_count++;
+
+			if (local->success_count == priv->child_count) {
+				need_unwind = 1;
+			}
+		}
+
+		local->op_errno = op_errno;
+	}
+	UNLOCK (&frame->lock);
+
+	if (need_unwind)
+		local->transaction.unwind (frame, this);
+
+	call_count = afr_frame_return (frame);
+
+	if (call_count == 0) {
+		local->transaction.resume (frame, this);
+	}
+
+	return 0;
+}
+
+static int
+afr_setxattr_wind (call_frame_t *frame, xlator_t *this)
+{
+	afr_local_t *local = NULL;
+	afr_private_t *priv = NULL;
+
+	int call_count = -1;
+	int i = 0;
+
+	local = frame->local;
+	priv = this->private;
+
+	call_count = afr_up_children_count (local->child_up, priv->child_count);
+
+	if (call_count == 0) {
+		local->transaction.resume (frame, this);
+		return 0;
+	}
+
+	local->call_count = call_count;
+
+	for (i = 0; i < priv->child_count; i++) {
+		if (local->child_up[i]) {
+			STACK_WIND_COOKIE (frame, afr_setxattr_wind_cbk,
+					   (void *) (long) i,
+					   priv->children[i],
+					   priv->children[i]->fops->setxattr,
+					   &local->loc,
+					   local->cont.setxattr.dict,
+					   local->cont.setxattr.flags, NULL);
+
+			if (!--call_count)
+				break;
+		}
+	}
+
+	return 0;
+}
+
+
+static int
+afr_setxattr_done (call_frame_t *frame, xlator_t *this)
+{
+	afr_local_t * local = frame->local;
+
+	local->transaction.unwind (frame, this);
+
+	AFR_STACK_DESTROY (frame);
+
+	return 0;
+}
+
+int32_t
+pump_setxattr_cbk (call_frame_t *frame,
+		      void *cookie,
+		      xlator_t *this,
+		      int32_t op_ret,
+		      int32_t op_errno, dict_t *xdata)
+{
+	AFR_STACK_UNWIND (setxattr, frame, op_ret, op_errno, xdata);
+	return 0;
+}
+
+int
+pump_command_reply (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t *local = NULL;
+
+        local = frame->local;
+
+        if (local->op_ret < 0)
+                gf_log (this->name, GF_LOG_INFO,
+                        "Command failed");
+        else
+                gf_log (this->name, GF_LOG_INFO,
+                        "Command succeeded");
+
+        AFR_STACK_UNWIND (setxattr,
+                          frame,
+                          local->op_ret,
+                          local->op_errno, NULL);
+
+        return 0;
+}
+
+int
+pump_parse_command (call_frame_t *frame, xlator_t *this,
+                    afr_local_t *local, dict_t *dict)
+{
+
+        int ret = -1;
+
+        if (pump_command_start (this, dict)) {
+                frame->local = local;
+                local->dict = dict_ref (dict);
+                ret = pump_execute_start (frame, this);
+
+        } else if (pump_command_pause (this, dict)) {
+                frame->local = local;
+                local->dict = dict_ref (dict);
+                ret = pump_execute_pause (frame, this);
+
+        } else if (pump_command_abort (this, dict)) {
+                frame->local = local;
+                local->dict = dict_ref (dict);
+                ret = pump_execute_abort (frame, this);
+
+        } else if (pump_command_commit (this, dict)) {
+                frame->local = local;
+                local->dict = dict_ref (dict);
+                ret = pump_execute_commit (frame, this);
+        }
+        return ret;
+}
+
+int
+pump_setxattr (call_frame_t *frame, xlator_t *this,
+               loc_t *loc, dict_t *dict, int32_t flags, dict_t *xdata)
+{
+	afr_private_t * priv  = NULL;
+	afr_local_t   * local = NULL;
+	call_frame_t   *transaction_frame = NULL;
+	int ret = -1;
+	int op_errno = 0;
+
+	VALIDATE_OR_GOTO (frame, out);
+	VALIDATE_OR_GOTO (this, out);
+	VALIDATE_OR_GOTO (this->private, out);
+
+        GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.pump*", dict,
+                                   op_errno, out);
+
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame, default_setxattr_cbk,
+                            FIRST_CHILD (this),
+                            (FIRST_CHILD (this))->fops->setxattr,
+                            loc, dict, flags, xdata);
+                return 0;
+        }
+
+
+	AFR_LOCAL_ALLOC_OR_GOTO (local, out);
+
+	ret = afr_local_init (local, priv, &op_errno);
+	if (ret < 0) {
+                afr_local_cleanup (local, this);
+		goto out;
+        }
+
+        ret = pump_parse_command (frame, this,
+                                  local, dict);
+        if (ret >= 0) {
+                ret = 0;
+                goto out;
+        }
+
+	transaction_frame = copy_frame (frame);
+	if (!transaction_frame) {
+		gf_log (this->name, GF_LOG_ERROR,
+			"Out of memory.");
+                op_errno = ENOMEM;
+                ret = -1;
+                afr_local_cleanup (local, this);
+		goto out;
+	}
+
+	transaction_frame->local = local;
+
+	local->op_ret = -1;
+
+	local->cont.setxattr.dict  = dict_ref (dict);
+	local->cont.setxattr.flags = flags;
+
+	local->transaction.fop    = afr_setxattr_wind;
+	local->transaction.done   = afr_setxattr_done;
+	local->transaction.unwind = afr_setxattr_unwind;
+
+	loc_copy (&local->loc, loc);
+
+	local->transaction.main_frame = frame;
+	local->transaction.start   = LLONG_MAX - 1;
+	local->transaction.len     = 0;
+
+	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+
+	ret = 0;
+out:
+	if (ret < 0) {
+		if (transaction_frame)
+			AFR_STACK_DESTROY (transaction_frame);
+		AFR_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL);
+	}
+
+	return 0;
+}
+
+/* Defaults */
+static int32_t
+pump_lookup (call_frame_t *frame,
+             xlator_t *this,
+             loc_t *loc,
+             dict_t *xattr_req)
+{
+	afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_lookup_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->lookup,
+                            loc,
+                            xattr_req);
+                return 0;
+        }
+
+        afr_lookup (frame, this, loc, xattr_req);
+        return 0;
+}
+
+
+static int32_t
+pump_truncate (call_frame_t *frame,
+               xlator_t *this,
+               loc_t *loc,
+               off_t offset, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_truncate_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->truncate,
+                            loc,
+                            offset, xdata);
+                return 0;
+        }
+
+        afr_truncate (frame, this, loc, offset, xdata);
+        return 0;
+}
+
+
+static int32_t
+pump_ftruncate (call_frame_t *frame,
+                xlator_t *this,
+                fd_t *fd,
+                off_t offset, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_ftruncate_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->ftruncate,
+                            fd,
+                            offset, xdata);
+                return 0;
+        }
+
+        afr_ftruncate (frame, this, fd, offset, xdata);
+        return 0;
+}
+
+
+
+
+int
+pump_mknod (call_frame_t *frame, xlator_t *this,
+            loc_t *loc, mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame, default_mknod_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->mknod,
+                            loc, mode, rdev, umask, xdata);
+                return 0;
+        }
+        afr_mknod (frame, this, loc, mode, rdev, umask, xdata);
+        return 0;
+
+}
+
+
+
+int
+pump_mkdir (call_frame_t *frame, xlator_t *this,
+            loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame, default_mkdir_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->mkdir,
+                            loc, mode, umask, xdata);
+                return 0;
+        }
+        afr_mkdir (frame, this, loc, mode, umask, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_unlink (call_frame_t *frame,
+             xlator_t *this,
+             loc_t *loc, int xflag, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_unlink_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->unlink,
+                            loc, xflag, xdata);
+                return 0;
+        }
+        afr_unlink (frame, this, loc, xflag, xdata);
+        return 0;
+
+}
+
+
+static int
+pump_rmdir (call_frame_t *frame, xlator_t *this,
+            loc_t *loc, int flags, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+
+	priv = this->private;
+
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame, default_rmdir_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->rmdir,
+                            loc, flags, xdata);
+                return 0;
+        }
+
+        afr_rmdir (frame, this, loc, flags, xdata);
+        return 0;
+
+}
+
+
+
+int
+pump_symlink (call_frame_t *frame, xlator_t *this,
+              const char *linkpath, loc_t *loc, mode_t umask, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame, default_symlink_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->symlink,
+                            linkpath, loc, umask, xdata);
+                return 0;
+        }
+        afr_symlink (frame, this, linkpath, loc, umask, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_rename (call_frame_t *frame,
+             xlator_t *this,
+             loc_t *oldloc,
+             loc_t *newloc, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_rename_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->rename,
+                            oldloc, newloc, xdata);
+                return 0;
+        }
+        afr_rename (frame, this, oldloc, newloc, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_link (call_frame_t *frame,
+           xlator_t *this,
+           loc_t *oldloc,
+           loc_t *newloc, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_link_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->link,
+                            oldloc, newloc, xdata);
+                return 0;
+        }
+        afr_link (frame, this, oldloc, newloc, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_create (call_frame_t *frame, xlator_t *this,
+             loc_t *loc, int32_t flags, mode_t mode,
+             mode_t umask, fd_t *fd, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame, default_create_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->create,
+                            loc, flags, mode, umask, fd, xdata);
+                return 0;
+        }
+        afr_create (frame, this, loc, flags, mode, umask, fd, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_open (call_frame_t *frame,
+           xlator_t *this,
+           loc_t *loc,
+           int32_t flags, fd_t *fd, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_open_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->open,
+                            loc, flags, fd, xdata);
+                return 0;
+        }
+        afr_open (frame, this, loc, flags, fd, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_writev (call_frame_t *frame,
+             xlator_t *this,
+             fd_t *fd,
+             struct iovec *vector,
+             int32_t count,
+             off_t off, uint32_t flags,
+             struct iobref *iobref, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_writev_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->writev,
+                            fd,
+                            vector,
+                            count,
+                            off, flags,
+                            iobref, xdata);
+                return 0;
+        }
+
+        afr_writev (frame, this, fd, vector, count, off, flags, iobref, xdata);
+        return 0;
+}
+
+
+static int32_t
+pump_flush (call_frame_t *frame,
+            xlator_t *this,
+            fd_t *fd, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_flush_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->flush,
+                            fd, xdata);
+                return 0;
+        }
+        afr_flush (frame, this, fd, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_fsync (call_frame_t *frame,
+            xlator_t *this,
+            fd_t *fd,
+            int32_t flags, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_fsync_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->fsync,
+                            fd,
+                            flags, xdata);
+                return 0;
+        }
+        afr_fsync (frame, this, fd, flags, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_opendir (call_frame_t *frame,
+              xlator_t *this,
+              loc_t *loc, fd_t *fd, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_opendir_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->opendir,
+                            loc, fd, xdata);
+                return 0;
+        }
+        afr_opendir (frame, this, loc, fd, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_fsyncdir (call_frame_t *frame,
+               xlator_t *this,
+               fd_t *fd,
+               int32_t flags, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_fsyncdir_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->fsyncdir,
+                            fd,
+                            flags, xdata);
+                return 0;
+        }
+        afr_fsyncdir (frame, this, fd, flags, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_xattrop (call_frame_t *frame,
+              xlator_t *this,
+              loc_t *loc,
+              gf_xattrop_flags_t flags,
+              dict_t *dict, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_xattrop_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->xattrop,
+                            loc,
+                            flags,
+                            dict, xdata);
+                return 0;
+        }
+        afr_xattrop (frame, this, loc, flags, dict, xdata);
+        return 0;
+
+}
+
+static int32_t
+pump_fxattrop (call_frame_t *frame,
+               xlator_t *this,
+               fd_t *fd,
+               gf_xattrop_flags_t flags,
+               dict_t *dict, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_fxattrop_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->fxattrop,
+                            fd,
+                            flags,
+                            dict, xdata);
+                return 0;
+        }
+        afr_fxattrop (frame, this, fd, flags, dict, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_removexattr (call_frame_t *frame,
+                  xlator_t *this,
+                  loc_t *loc,
+                  const char *name, dict_t *xdata)
+{
+        afr_private_t *priv     = NULL;
+        int            op_errno = -1;
+
+        VALIDATE_OR_GOTO (this, out);
+
+        GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.pump*",
+                                 name, op_errno, out);
+
+        op_errno = 0;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_removexattr_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->removexattr,
+                            loc,
+                            name, xdata);
+                return 0;
+        }
+        afr_removexattr (frame, this, loc, name, xdata);
+
+ out:
+        if (op_errno)
+                AFR_STACK_UNWIND (removexattr, frame, -1, op_errno, NULL);
+        return 0;
+
+}
+
+
+
+static int32_t
+pump_readdir (call_frame_t *frame,
+              xlator_t *this,
+              fd_t *fd,
+              size_t size,
+              off_t off, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_readdir_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->readdir,
+                            fd, size, off, xdata);
+                return 0;
+        }
+        afr_readdir (frame, this, fd, size, off, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd,
+               size_t size, off_t off, dict_t *dict)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_readdirp_cbk,
+                            FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->readdirp,
+                            fd, size, off, dict);
+                return 0;
+        }
+        afr_readdirp (frame, this, fd, size, off, dict);
+        return 0;
+
+}
+
+
+
+static int32_t
+pump_releasedir (xlator_t *this,
+                 fd_t *fd)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (priv->use_afr_in_pump)
+                afr_releasedir (this, fd);
+	return 0;
+
+}
+
+static int32_t
+pump_release (xlator_t *this,
+              fd_t *fd)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (priv->use_afr_in_pump)
+                afr_release (this, fd);
+	return 0;
+
+}
+
+static int32_t
+pump_forget (xlator_t *this, inode_t *inode)
+{
+        afr_private_t  *priv  = NULL;
+
+        priv = this->private;
+        if (priv->use_afr_in_pump)
+                afr_forget (this, inode);
+
+        return 0;
+}
+
+static int32_t
+pump_setattr (call_frame_t *frame,
+              xlator_t *this,
+              loc_t *loc,
+              struct iatt *stbuf,
+              int32_t valid, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_setattr_cbk,
+                            FIRST_CHILD (this),
+                            FIRST_CHILD (this)->fops->setattr,
+                            loc, stbuf, valid, xdata);
+                return 0;
+        }
+        afr_setattr (frame, this, loc, stbuf, valid, xdata);
+        return 0;
+
+}
+
+
+static int32_t
+pump_fsetattr (call_frame_t *frame,
+               xlator_t *this,
+               fd_t *fd,
+               struct iatt *stbuf,
+               int32_t valid, dict_t *xdata)
+{
+        afr_private_t *priv  = NULL;
+	priv = this->private;
+        if (!priv->use_afr_in_pump) {
+                STACK_WIND (frame,
+                            default_fsetattr_cbk,
+                            FIRST_CHILD (this),
+                            FIRST_CHILD (this)->fops->fsetattr,
+                            fd, stbuf, valid, xdata);
+                return 0;
+        }
+        afr_fsetattr (frame, this, fd, stbuf, valid, xdata);
+        return 0;
+
+}
+
+
+/* End of defaults */
+
+
+int32_t
+mem_acct_init (xlator_t *this)
+{
+        int     ret = -1;
+
+        if (!this)
+                return ret;
+
+        ret = xlator_mem_acct_init (this, gf_afr_mt_end + 1);
+
+        if (ret != 0) {
+                gf_log(this->name, GF_LOG_ERROR, "Memory accounting init"
+                                "failed");
+                return ret;
+        }
+
+        return ret;
+}
+
+static int
+is_xlator_pump_sink (xlator_t *child)
+{
+        return (child == PUMP_SINK_CHILD(THIS));
+}
+
+static int
+is_xlator_pump_source (xlator_t *child)
+{
+        return (child == PUMP_SOURCE_CHILD(THIS));
+}
+
+int32_t
+notify (xlator_t *this, int32_t event,
+	void *data, ...)
+{
+        int ret = -1;
+        xlator_t *child_xl = NULL;
+
+        child_xl = (xlator_t *) data;
+
+        ret = afr_notify (this, event, data, NULL);
+
+	switch (event) {
+	case GF_EVENT_CHILD_DOWN:
+                if (is_xlator_pump_source (child_xl))
+                        pump_change_state (this, PUMP_STATE_ABORT);
+                break;
+
+        case GF_EVENT_CHILD_UP:
+                if (is_xlator_pump_sink (child_xl))
+                        if (is_pump_start_pending (this)) {
+                                gf_log (this->name, GF_LOG_DEBUG,
+                                        "about to start synctask");
+                                ret = pump_start_synctask (this);
+                                if (ret < 0)
+                                        gf_log (this->name, GF_LOG_DEBUG,
+                                                "Could not start pump "
+                                                "synctask");
+                                else
+                                        pump_remove_start_pending (this);
+                        }
+        }
+
+        return ret;
+}
+
+int32_t
+init (xlator_t *this)
+{
+	afr_private_t * priv        = NULL;
+        pump_private_t *pump_priv = NULL;
+	int             child_count = 0;
+	xlator_list_t * trav        = NULL;
+	int             i           = 0;
+	int             ret         = -1;
+	GF_UNUSED int   op_errno    = 0;
+
+        int source_child = 0;
+
+	if (!this->children) {
+		gf_log (this->name, GF_LOG_ERROR,
+			"pump translator needs a source and sink"
+                        "subvolumes defined.");
+		return -1;
+	}
+
+	if (!this->parents) {
+		gf_log (this->name, GF_LOG_WARNING,
+			"Volume is dangling.");
+	}
+
+	this->private = GF_CALLOC (1, sizeof (afr_private_t),
+                                   gf_afr_mt_afr_private_t);
+        if (!this->private)
+                goto out;
+
+	priv = this->private;
+        LOCK_INIT (&priv->lock);
+        LOCK_INIT (&priv->read_child_lock);
+        //lock recovery is not done in afr
+        pthread_mutex_init (&priv->mutex, NULL);
+        INIT_LIST_HEAD (&priv->saved_fds);
+
+        child_count = xlator_subvolume_count (this);
+        if (child_count != 2) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "There should be exactly 2 children - one source "
+                        "and one sink");
+                return -1;
+        }
+	priv->child_count = child_count;
+
+        priv->read_child = source_child;
+        priv->favorite_child = source_child;
+        priv->background_self_heal_count = 0;
+
+	priv->data_self_heal     = "on";
+	priv->metadata_self_heal = 1;
+	priv->entry_self_heal    = 1;
+
+        priv->data_self_heal_window_size = 16;
+
+	priv->data_change_log     = 1;
+	priv->metadata_change_log = 1;
+	priv->entry_change_log    = 1;
+        priv->use_afr_in_pump = 1;
+        priv->sh_readdir_size = 65536;
+
+	/* Locking options */
+
+        /* Lock server count infact does not matter. Locks are held
+           on all subvolumes, in this case being the source
+           and the sink.
+        */
+
+	priv->strict_readdir = _gf_false;
+
+	priv->wait_count = 1;
+	priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count,
+                                 gf_afr_mt_char);
+	if (!priv->child_up) {
+		gf_log (this->name, GF_LOG_ERROR,
+			"Out of memory.");
+		op_errno = ENOMEM;
+		goto out;
+	}
+
+	priv->children = GF_CALLOC (sizeof (xlator_t *), child_count,
+                                 gf_afr_mt_xlator_t);
+	if (!priv->children) {
+		gf_log (this->name, GF_LOG_ERROR,
+			"Out of memory.");
+		op_errno = ENOMEM;
+		goto out;
+	}
+
+        priv->pending_key = GF_CALLOC (sizeof (*priv->pending_key),
+                                       child_count,
+                                       gf_afr_mt_char);
+        if (!priv->pending_key) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Out of memory.");
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+	trav = this->children;
+	i = 0;
+	while (i < child_count) {
+		priv->children[i] = trav->xlator;
+
+                ret = gf_asprintf (&priv->pending_key[i], "%s.%s", AFR_XATTR_PREFIX,
+                                   trav->xlator->name);
+                if (-1 == ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "asprintf failed to set pending key");
+                        op_errno = ENOMEM;
+                        goto out;
+                }
+
+		trav = trav->next;
+		i++;
+	}
+
+        ret = gf_asprintf (&priv->sh_domain, "%s-self-heal", this->name);
+        if (-1 == ret) {
+                op_errno = ENOMEM;
+                goto out;
+        }
+
+        priv->first_lookup = 1;
+        priv->root_inode = NULL;
+
+        priv->last_event = GF_CALLOC (child_count, sizeof (*priv->last_event),
+                                      gf_afr_mt_int32_t);
+        if (!priv->last_event) {
+                ret = -ENOMEM;
+                goto out;
+        }
+
+	pump_priv = GF_CALLOC (1, sizeof (*pump_priv),
+                            gf_afr_mt_pump_priv);
+	if (!pump_priv) {
+		gf_log (this->name, GF_LOG_ERROR,
+			"Out of memory");
+                op_errno = ENOMEM;
+		goto out;
+	}
+
+        LOCK_INIT (&pump_priv->resume_path_lock);
+        LOCK_INIT (&pump_priv->pump_state_lock);
+
+        pump_priv->resume_path = GF_CALLOC (1, PATH_MAX,
+                                            gf_afr_mt_char);
+        if (!pump_priv->resume_path) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Out of memory");
+                ret = -1;
+                goto out;
+        }
+
+	pump_priv->env = this->ctx->env;
+        if (!pump_priv->env) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Could not create new sync-environment");
+                ret = -1;
+                goto out;
+        }
+
+        /* keep more local here as we may need them for self-heal etc */
+        this->local_pool = mem_pool_new (afr_local_t, 128);
+        if (!this->local_pool) {
+                ret = -1;
+                gf_log (this->name, GF_LOG_ERROR,
+                        "failed to create local_t's memory pool");
+                goto out;
+        }
+
+	priv->pump_private = pump_priv;
+
+        pump_change_state (this, PUMP_STATE_ABORT);
+
+	ret = 0;
+out:
+	return ret;
+}
+
+int
+fini (xlator_t *this)
+{
+        afr_private_t * priv        = NULL;
+        pump_private_t *pump_priv = NULL;
+
+        priv      = this->private;
+        this->private = NULL;
+        if (!priv)
+                goto out;
+
+        pump_priv = priv->pump_private;
+        if (!pump_priv)
+                goto afr_priv;
+
+        GF_FREE (pump_priv->resume_path);
+        LOCK_DESTROY (&pump_priv->resume_path_lock);
+        LOCK_DESTROY (&pump_priv->pump_state_lock);
+        GF_FREE (pump_priv);
+afr_priv:
+        afr_priv_destroy (priv);
+out:
+	return 0;
+}
+
+
+struct xlator_fops fops = {
+	.lookup      = pump_lookup,
+	.open        = pump_open,
+	.flush       = pump_flush,
+	.fsync       = pump_fsync,
+	.fsyncdir    = pump_fsyncdir,
+	.xattrop     = pump_xattrop,
+	.fxattrop    = pump_fxattrop,
+        .getxattr    = pump_getxattr,
+
+	/* inode write */
+	.writev      = pump_writev,
+	.truncate    = pump_truncate,
+	.ftruncate   = pump_ftruncate,
+	.setxattr    = pump_setxattr,
+        .setattr     = pump_setattr,
+	.fsetattr    = pump_fsetattr,
+	.removexattr = pump_removexattr,
+
+	/* dir read */
+	.opendir     = pump_opendir,
+	.readdir     = pump_readdir,
+	.readdirp    = pump_readdirp,
+
+	/* dir write */
+	.create      = pump_create,
+	.mknod       = pump_mknod,
+	.mkdir       = pump_mkdir,
+	.unlink      = pump_unlink,
+	.rmdir       = pump_rmdir,
+	.link        = pump_link,
+	.symlink     = pump_symlink,
+	.rename      = pump_rename,
+};
+
+struct xlator_dumpops dumpops = {
+        .priv       = afr_priv_dump,
+};
+
+
+struct xlator_cbks cbks = {
+	.release     = pump_release,
+	.releasedir  = pump_releasedir,
+        .forget      = pump_forget,
+};
+
+struct volume_options options[] = {
+	{ .key  = {NULL} },
+};
diff --git a/xlators/cluster/afr/src/pump.h b/xlators/cluster/afr/src/pump.h
new file mode 100644
index 000000000..bc4c31a78
--- /dev/null
+++ b/xlators/cluster/afr/src/pump.h
@@ -0,0 +1,78 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __PUMP_H__
+#define __PUMP_H__
+
+#include "syncop.h"
+
+/* FIXME: Needs to be defined in a common file */
+#define CLIENT_CMD_CONNECT "trusted.glusterfs.client-connect"
+#define CLIENT_CMD_DISCONNECT "trusted.glusterfs.client-disconnect"
+
+#define PUMP_SOURCE_COMPLETE "trusted.glusterfs.pump-source-complete"
+#define PUMP_SINK_COMPLETE "trusted.glusterfs.pump-sink-complete"
+
+#define PUMP_PATH "trusted.glusterfs.pump-path"
+
+#define PUMP_SOURCE_CHILD(xl) (xl->children->xlator)
+#define PUMP_SINK_CHILD(xl) (xl->children->next->xlator)
+
+typedef enum {
+        PUMP_STATE_RUNNING,             /* Pump is running and migrating files */
+        PUMP_STATE_RESUME,              /* Pump is resuming from a previous pause */
+        PUMP_STATE_PAUSE,               /* Pump is paused */
+        PUMP_STATE_ABORT,               /* Pump is aborted */
+        PUMP_STATE_COMMIT,              /* Pump is commited */
+} pump_state_t;
+
+typedef struct _pump_private {
+	struct syncenv *env;            /* The env pointer to the pump synctask */
+        char *resume_path;              /* path to resume from the last pause */
+        gf_lock_t resume_path_lock;     /* Synchronize resume_path changes */
+        gf_lock_t pump_state_lock;      /* Synchronize pump_state changes */
+        pump_state_t pump_state;        /* State of pump */
+        char current_file[PATH_MAX];    /* Current file being pumped */
+        uint64_t number_files_pumped;   /* Number of files pumped */
+        gf_boolean_t pump_finished;     /* Boolean to indicate pump termination */
+        char pump_start_pending;        /* Boolean to mark start pending until
+                                           CHILD_UP */
+        call_stub_t *cleaner;
+} pump_private_t;
+
+void
+build_root_loc (inode_t *inode, loc_t *loc);
+int pump_start (call_frame_t *frame, xlator_t *this);
+
+gf_boolean_t
+pump_command_start (xlator_t *this, dict_t *dict);
+
+int
+pump_execute_start (call_frame_t *frame, xlator_t *this);
+
+gf_boolean_t
+pump_command_pause (xlator_t *this, dict_t *dict);
+
+int
+pump_execute_pause (call_frame_t *frame, xlator_t *this);
+
+gf_boolean_t
+pump_command_abort (xlator_t *this, dict_t *dict);
+
+int
+pump_execute_abort (call_frame_t *frame, xlator_t *this);
+
+gf_boolean_t
+pump_command_status (xlator_t *this, dict_t *dict);
+
+int
+pump_execute_status (call_frame_t *frame, xlator_t *this);
+
+#endif /* __PUMP_H__ */
diff --git a/xlators/cluster/dht/src/Makefile.am b/xlators/cluster/dht/src/Makefile.am
index d4e0752a5..174bea841 100644
--- a/xlators/cluster/dht/src/Makefile.am
+++ b/xlators/cluster/dht/src/Makefile.am
@@ -1,30 +1,38 @@
 
-xlator_LTLIBRARIES = dht.la nufa.la
+xlator_LTLIBRARIES = dht.la nufa.la switch.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
 
+dht_common_source = dht-layout.c dht-helper.c dht-linkfile.c dht-rebalance.c \
+	dht-selfheal.c dht-rename.c dht-hashfn.c dht-diskusage.c \
+	dht-common.c dht-inode-write.c dht-inode-read.c dht-shared.c \
+	$(top_builddir)/xlators/lib/src/libxlator.c
 
-dht_common_source = dht-layout.c dht-helper.c dht-linkfile.c \
-		dht-selfheal.c dht-rename.c dht-hashfn.c
-
-dht_la_SOURCES = $(dht_common_source) dht.c 
+dht_la_SOURCES = $(dht_common_source) dht.c
 
 nufa_la_SOURCES = $(dht_common_source) nufa.c
+switch_la_SOURCES = $(dht_common_source) switch.c
 
-dht_la_LDFLAGS = -module -avoidversion
+dht_la_LDFLAGS = -module -avoid-version
 dht_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-nufa_la_LDFLAGS = -module -avoidversion
+nufa_la_LDFLAGS = -module -avoid-version
 nufa_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-noinst_HEADERS = dht-common.h dht-common.c
+switch_la_LDFLAGS = -module -avoid-version
+switch_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = dht-common.h dht-mem-types.h \
+	$(top_builddir)/xlators/lib/src/libxlator.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/xlators/lib/src
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
-CLEANFILES = 
+CLEANFILES =
 
 uninstall-local:
 	rm -f $(DESTDIR)$(xlatordir)/distribute.so
 
 install-data-hook:
-	ln -sf dht.so $(DESTDIR)$(xlatordir)/distribute.so
-\ No newline at end of file
+	ln -sf dht.so $(DESTDIR)$(xlatordir)/distribute.so
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index 41daac677..8f61339e6 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2009-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 
@@ -27,9 +18,87 @@
 
 #include "glusterfs.h"
 #include "xlator.h"
+#include "libxlator.h"
 #include "dht-common.h"
 #include "defaults.h"
+#include "byte-order.h"
+#include "glusterfs-acl.h"
+
+#include <sys/time.h>
+#include <libgen.h>
+
+int
+dht_aggregate (dict_t *this, char *key, data_t *value, void *data)
+{
+        dict_t  *dst  = NULL;
+        int64_t *ptr  = 0, *size = NULL;
+        int32_t  ret  = -1;
+        data_t  *dict_data = NULL;
+
+        dst = data;
+
+        if (strcmp (key, GF_XATTR_QUOTA_SIZE_KEY) == 0) {
+                ret = dict_get_bin (dst, key, (void **)&size);
+                if (ret < 0) {
+                        size = GF_CALLOC (1, sizeof (int64_t),
+                                          gf_common_mt_char);
+                        if (size == NULL) {
+                                gf_log ("dht", GF_LOG_WARNING,
+                                        "memory allocation failed");
+                                return -1;
+                        }
+                        ret = dict_set_bin (dst, key, size, sizeof (int64_t));
+                        if (ret < 0) {
+                                gf_log ("dht", GF_LOG_WARNING,
+                                        "dht aggregate dict set failed");
+                                GF_FREE (size);
+                                return -1;
+                        }
+                }
 
+                ptr = data_to_bin (value);
+                if (ptr == NULL) {
+                        gf_log ("dht", GF_LOG_WARNING, "data to bin failed");
+                        return -1;
+                }
+
+                *size = hton64 (ntoh64 (*size) + ntoh64 (*ptr));
+
+        } else if (fnmatch (GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0) {
+                ret = gf_get_min_stime (THIS, dst, key, value);
+                if (ret < 0)
+                        return ret;
+        } else {
+                /* compare user xattrs only */
+                if (!strncmp (key, "user.", strlen ("user."))) {
+                        ret = dict_lookup (dst, key, &dict_data);
+                        if (!ret && dict_data && value) {
+                                ret = is_data_equal (dict_data, value);
+                                if (!ret)
+                                        gf_log ("dht", GF_LOG_DEBUG,
+                                                "xattr mismatch for %s", key);
+                        }
+                }
+                ret = dict_set (dst, key, value);
+                if (ret)
+                        gf_log ("dht", GF_LOG_WARNING, "xattr dict set failed");
+        }
+
+        return 0;
+}
+
+
+void
+dht_aggregate_xattr (dict_t *dst, dict_t *src)
+{
+        if ((dst == NULL) || (src == NULL)) {
+                goto out;
+        }
+
+        dict_foreach (src, dht_aggregate, dst);
+out:
+        return;
+}
 
 /* TODO:
    - use volumename in xattr instead of "dht"
@@ -38,101 +107,372 @@
    - complete linkfile selfheal
 */
 
+
 int
 dht_lookup_selfheal_cbk (call_frame_t *frame, void *cookie,
-			 xlator_t *this,
-			 int op_ret, int op_errno)
+                         xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	dht_layout_t *layout = NULL;
-	int           ret = 0;
+        dht_local_t  *local = NULL;
+        dht_layout_t *layout = NULL;
+	int           ret = -1;
 
-	local = frame->local;
-	ret = op_ret;
-
-	if (ret == 0) {
-		layout = local->selfheal.layout;
-		ret = inode_ctx_put (local->inode, this, (uint64_t)(long)layout);
-
-		if (ret == 0)
-			local->selfheal.layout = NULL;
-		
-		if (local->st_ino) {
-			local->stbuf.st_ino = local->st_ino;
-		} else {
-			gf_log (this->name, GF_LOG_WARNING,
-				"could not find hashed subvolume for %s",
-				local->loc.path);
-		}
-	}
+        GF_VALIDATE_OR_GOTO ("dht", frame, out);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+
+        local = frame->local;
+        ret = op_ret;
 
-	DHT_STACK_UNWIND (frame, ret, local->op_errno, local->inode,
-			  &local->stbuf, local->xattr);
+        FRAME_SU_UNDO (frame, dht_local_t);
 
-	return 0;
+        if (ret == 0) {
+                layout = local->selfheal.layout;
+                ret = dht_layout_set (this, local->inode, layout);
+        }
+
+        if (local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           &local->postparent, 1);
+        }
+
+        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
+
+        DHT_STACK_UNWIND (lookup, frame, ret, local->op_errno, local->inode,
+                          &local->stbuf, local->xattr, &local->postparent);
+
+out:
+        return ret;
+}
+
+
+int
+dht_discover_complete (xlator_t *this, call_frame_t *discover_frame)
+{
+        dht_local_t     *local = NULL;
+        call_frame_t    *main_frame = NULL;
+        int              op_errno = 0;
+        int              ret = -1;
+        dht_layout_t    *layout = NULL;
+        dht_conf_t      *conf = NULL;
+
+        local = discover_frame->local;
+        layout = local->layout;
+        conf = this->private;
+
+        LOCK(&discover_frame->lock);
+        {
+                main_frame = local->main_frame;
+                local->main_frame = NULL;
+        }
+        UNLOCK(&discover_frame->lock);
+
+        if (!main_frame)
+                return 0;
+
+        if (local->file_count && local->dir_count) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "path %s exists as a file on one subvolume "
+                        "and directory on another. "
+                        "Please fix it manually",
+                        local->loc.path);
+                op_errno = EIO;
+                goto out;
+        }
+
+        if (local->cached_subvol) {
+                ret = dht_layout_preset (this, local->cached_subvol,
+                                         local->inode);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "failed to set layout for subvolume %s",
+                                local->cached_subvol ? local->cached_subvol->name : "<nil>");
+                        op_errno = EINVAL;
+                        goto out;
+                }
+        } else {
+                ret = dht_layout_normalize (this, &local->loc, layout);
+                if ((ret < 0) || ((ret > 0) && (local->op_ret != 0))) {
+                        /* either the layout is incorrect or the directory is
+                         * not found even in one subvolume.
+                         */
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "normalizing failed on %s "
+                                "(overlaps/holes present: %s, "
+                                "ENOENT errors: %d)", local->loc.path,
+                                (ret < 0) ? "yes" : "no", (ret > 0) ? ret : 0);
+                        if ((ret > 0) && (ret == conf->subvolume_cnt)) {
+                                op_errno = ESTALE;
+                                goto out;
+                        }
+                }
+
+                if (local->inode)
+                        dht_layout_set (this, local->inode, layout);
+        }
+
+        DHT_STACK_UNWIND (lookup, main_frame, local->op_ret, local->op_errno,
+                          local->inode, &local->stbuf, local->xattr,
+                          &local->postparent);
+        return 0;
+out:
+        DHT_STACK_UNWIND (lookup, main_frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL);
+
+        return ret;
+}
+
+
+int
+dht_discover_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int op_ret, int op_errno,
+                  inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                  struct iatt *postparent)
+{
+        dht_local_t  *local                   = NULL;
+        int           this_call_cnt           = 0;
+        call_frame_t *prev                    = NULL;
+        dht_layout_t *layout                  = NULL;
+        int           ret                     = -1;
+        int           is_dir                  = 0;
+        int           is_linkfile             = 0;
+        int           attempt_unwind          = 0;
+        dht_conf_t   *conf                    = 0;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, out);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev  = cookie;
+        conf = this->private;
+
+        layout = local->layout;
+
+        /* Check if the gfid is different for file from other node */
+        if (!op_ret && uuid_compare (local->gfid, stbuf->ia_gfid)) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: gfid different on %s",
+                        local->loc.path, prev->this->name);
+        }
+
+
+        LOCK (&frame->lock);
+        {
+                /* TODO: assert equal mode on stbuf->st_mode and
+                   local->stbuf->st_mode
+
+                   else mkdir/chmod/chown and fix
+                */
+                ret = dht_layout_merge (this, layout, prev->this,
+                                        op_ret, op_errno, xattr);
+                if (ret)
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: failed to merge layouts", local->loc.path);
+
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "lookup of %s on %s returned error (%s)",
+                                local->loc.path, prev->this->name,
+                                strerror (op_errno));
+
+                        goto unlock;
+                }
+
+                is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+                                                 conf->link_xattr_name);
+                is_dir = check_is_dir (inode, stbuf, xattr);
+
+                if (is_dir) {
+                        local->dir_count ++;
+                } else {
+                        local->file_count ++;
+
+                        if (!is_linkfile) {
+                                /* real file */
+                                local->cached_subvol = prev->this;
+                                attempt_unwind = 1;
+                        } else {
+                                goto unlock;
+                        }
+                }
+
+                local->op_ret = 0;
+
+                if (local->xattr == NULL) {
+                        local->xattr = dict_ref (xattr);
+                } else {
+                        dht_aggregate_xattr (local->xattr, xattr);
+                }
+
+                if (local->inode == NULL)
+                        local->inode = inode_ref (inode);
+
+                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+                dht_iatt_merge (this, &local->postparent, postparent,
+                                prev->this);
+        }
+unlock:
+        UNLOCK (&frame->lock);
+out:
+        this_call_cnt = dht_frame_return (frame);
+
+        if (is_last_call (this_call_cnt) || attempt_unwind) {
+                dht_discover_complete (this, frame);
+        }
+
+        if (is_last_call (this_call_cnt))
+                DHT_STACK_DESTROY (frame);
+
+        return 0;
+}
+
+
+int
+dht_discover (call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+        int          ret;
+        dht_local_t *local = NULL;
+        dht_conf_t  *conf = NULL;
+        int          call_cnt = 0;
+        int          op_errno = EINVAL;
+        int          i = 0;
+        call_frame_t *discover_frame = NULL;
+
+        conf = this->private;
+        local = frame->local;
+
+        ret = dict_set_uint32 (local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to set '%s' key",
+                        loc->path, conf->xattr_name);
+
+        ret = dict_set_uint32 (local->xattr_req, conf->link_xattr_name, 256);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to set '%s' key",
+                        loc->path, conf->link_xattr_name);
+
+        call_cnt        = conf->subvolume_cnt;
+        local->call_cnt = call_cnt;
+
+        local->layout = dht_layout_new (this, conf->subvolume_cnt);
+
+        if (!local->layout) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        uuid_copy (local->gfid, loc->gfid);
+
+        discover_frame = copy_frame (frame);
+        if (!discover_frame) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        discover_frame->local = local;
+        frame->local = NULL;
+        local->main_frame = frame;
+
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (discover_frame, dht_discover_cbk,
+                            conf->subvolumes[i],
+                            conf->subvolumes[i]->fops->lookup,
+                            &local->loc, local->xattr_req);
+        }
+
+        return 0;
+
+err:
+        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL);
+
+        return 0;
 }
 
 
 int
 dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                     int op_ret, int op_errno,
-                    inode_t *inode, struct stat *stbuf, dict_t *xattr)
+                    inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                    struct iatt *postparent)
 {
-	dht_conf_t   *conf          = NULL;
-        dht_local_t  *local         = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev          = NULL;
-	dht_layout_t *layout        = NULL;
-	int           ret           = 0;
-	int           is_dir        = 0;
+        dht_local_t  *local                   = NULL;
+        int           this_call_cnt           = 0;
+        call_frame_t *prev                    = NULL;
+        dht_layout_t *layout                  = NULL;
+        int           ret                     = -1;
+        int           is_dir                  = 0;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, out);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
 
-	conf  = this->private;
         local = frame->local;
         prev  = cookie;
 
-	layout = local->layout;
+        layout = local->layout;
+
+        if (!op_ret && uuid_is_null (local->gfid))
+                memcpy (local->gfid, stbuf->ia_gfid, 16);
+
+        /* Check if the gfid is different for file from other node */
+        if (!op_ret && uuid_compare (local->gfid, stbuf->ia_gfid)) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: gfid different on %s",
+                        local->loc.path, prev->this->name);
+        }
 
         LOCK (&frame->lock);
         {
                 /* TODO: assert equal mode on stbuf->st_mode and
-		   local->stbuf->st_mode
+                   local->stbuf->st_mode
 
-		   else mkdir/chmod/chown and fix
-		*/
-		ret = dht_layout_merge (this, layout, prev->this,
-					op_ret, op_errno, xattr);
+                   else mkdir/chmod/chown and fix
+                */
+                ret = dht_layout_merge (this, layout, prev->this,
+                                        op_ret, op_errno, xattr);
 
-		if (op_ret == -1) {
-			local->op_errno = ENOENT;
-			gf_log (this->name, GF_LOG_WARNING,
-				"lookup of %s on %s returned error (%s)",
-				local->loc.path, prev->this->name,
-				strerror (op_errno));
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "lookup of %s on %s returned error (%s)",
+                                local->loc.path, prev->this->name,
+                                strerror (op_errno));
 
-			goto unlock;
-		}
+                        goto unlock;
+                }
 
- 		is_dir = check_is_dir (inode, stbuf, xattr);
- 		if (!is_dir) {
-                        gf_log (this->name, GF_LOG_WARNING,
+                is_dir = check_is_dir (inode, stbuf, xattr);
+                if (!is_dir) {
+                        gf_log (this->name, GF_LOG_DEBUG,
                                 "lookup of %s on %s returned non dir 0%o",
                                 local->loc.path, prev->this->name,
-                                stbuf->st_mode);
+                                stbuf->ia_type);
                         local->need_selfheal = 1;
- 			goto unlock;
+                        goto unlock;
                 }
 
- 		local->op_ret = 0;
- 		if (local->xattr == NULL)
- 			local->xattr = dict_ref (xattr);
- 		if (local->inode == NULL)
- 			local->inode = inode_ref (inode);
-
-		dht_stat_merge (this, &local->stbuf, stbuf, prev->this);
+                local->op_ret = 0;
+                if (local->xattr == NULL) {
+                        local->xattr = dict_ref (xattr);
+                } else {
+                        dht_aggregate_xattr (local->xattr, xattr);
+                }
 
-		if (prev->this == local->hashed_subvol)
-			local->st_ino = local->stbuf.st_ino;
+                if (local->inode == NULL)
+                        local->inode = inode_ref (inode);
 
+                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+                dht_iatt_merge (this, &local->postparent, postparent,
+                                prev->this);
         }
 unlock:
         UNLOCK (&frame->lock);
@@ -147,243 +487,510 @@ unlock:
                         return 0;
                 }
 
-		if (local->op_ret == 0) {
-			ret = dht_layout_normalize (this, &local->loc, layout);
+                if (local->op_ret == 0) {
+                        ret = dht_layout_normalize (this, &local->loc, layout);
 
-			local->layout = NULL;
+                        if (ret != 0) {
+                                gf_log (this->name, GF_LOG_DEBUG,
+                                        "fixing assignment on %s",
+                                        local->loc.path);
+                                goto selfheal;
+                        }
 
-			if (ret != 0) {
-				layout->gen = conf->gen;
+                        dht_layout_set (this, local->inode, layout);
+                }
 
-				gf_log (this->name, GF_LOG_WARNING,
-					"fixing assignment on %s",
-					local->loc.path);
-				goto selfheal;
-			}
-			
-			inode_ctx_put (local->inode, this,
-                                       (uint64_t)(long)layout);
-			
-			if (local->st_ino) {
-				local->stbuf.st_ino = local->st_ino;
-			} else {
-				gf_log (this->name, GF_LOG_WARNING,
-					"could not find hashed subvol for %s",
-					local->loc.path);
-			}
-		}
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
 
-		DHT_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-				  local->inode, &local->stbuf, local->xattr);
+                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
+                DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
+                                  local->inode, &local->stbuf, local->xattr,
+                                  &local->postparent);
         }
 
-	return 0;
+        return 0;
 
 selfheal:
-	ret = dht_selfheal_directory (frame, dht_lookup_selfheal_cbk,
-				      &local->loc, layout);
-
-	return 0;
+        FRAME_SU_DO (frame, dht_local_t);
+        uuid_copy (local->loc.gfid, local->gfid);
+        ret = dht_selfheal_directory (frame, dht_lookup_selfheal_cbk,
+                                      &local->loc, layout);
+out:
+        return ret;
 }
 
 int
 dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                     int op_ret, int op_errno,
-                    inode_t *inode, struct stat *stbuf, dict_t *xattr)
+                    inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                    struct iatt *postparent)
 {
         dht_local_t  *local         = NULL;
         int           this_call_cnt = 0;
         call_frame_t *prev          = NULL;
-	dht_layout_t *layout        = NULL;
-	int           ret  = -1;
-	int           is_dir = 0;
-	int           is_linkfile = 0;
+        dht_layout_t *layout        = NULL;
+        dht_conf_t   *conf          = NULL;
+        int           ret  = -1;
+        int           is_dir = 0;
+        int           is_linkfile = 0;
+        call_frame_t *copy          = NULL;
+        dht_local_t  *copy_local    = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, err);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, err);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, err);
 
         local = frame->local;
         prev  = cookie;
+        conf = this->private;
+        if (!conf)
+                goto out;
 
         LOCK (&frame->lock);
         {
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
 
-			if ((op_errno != ENOTCONN) 
+                        if ((op_errno != ENOTCONN)
                             && (op_errno != ENOENT)
                             && (op_errno != ESTALE)) {
-				gf_log (this->name, GF_LOG_WARNING,
-					"subvolume %s returned -1 (%s)",
-					prev->this->name, strerror (op_errno));
+				gf_log (this->name, GF_LOG_INFO,
+					"subvolume %s for %s returned -1 (%s)",
+					prev->this->name, local->loc.path,
+                                        strerror (op_errno));
 			}
-                        
                         if (op_errno == ESTALE) {
-                                /* propogate the ESTALE to parent. 
-                                 * setting local->layout_mismatch would send
+                                /* propagate the ESTALE to parent.
+                                 * setting local->return_estale would send
                                  * ESTALE to parent. */
-                                local->layout_mismatch = 1;
+                                local->return_estale = 1;
                         }
 
-			goto unlock;
-		}
+                        /* if it is ENOENT, we may have to do a
+                         * 'lookup_everywhere()' to make sure
+                         * the file is not migrated */
+                        if (op_errno == ENOENT) {
+                                if (IA_ISREG (local->loc.inode->ia_type)) {
+                                        local->need_lookup_everywhere = 1;
+                                }
+                        }
+                        goto unlock;
+                }
 
-		if (S_IFMT & (stbuf->st_mode ^ local->inode->st_mode)) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"mismatching filetypes 0%o v/s 0%o for %s",
-				(stbuf->st_mode & S_IFMT),
-				(local->inode->st_mode & S_IFMT),
-				local->loc.path);
+                if (stbuf->ia_type != local->inode->ia_type) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "mismatching filetypes 0%o v/s 0%o for %s",
+                                (stbuf->ia_type), (local->inode->ia_type),
+                                local->loc.path);
 
-			local->op_ret = -1;
-			local->op_errno = EINVAL;
+                        local->op_ret = -1;
+                        local->op_errno = EINVAL;
 
-			goto unlock;
-		}
+                        goto unlock;
+                }
 
-		layout = dht_layout_get (this, inode);
-		
-		is_dir = check_is_dir (inode, stbuf, xattr);
-		is_linkfile = check_is_linkfile (inode, stbuf, xattr);
-		
-		if (is_linkfile) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"linkfile found in revalidate for %s",
-				local->loc.path);
-			local->layout_mismatch = 1;
-
-			goto unlock;
-		}
+                layout = local->layout;
 
-		if (is_dir) {
-			ret = dht_layout_dir_mismatch (this, layout,
-						       prev->this, &local->loc,
-						       xattr);
-			if (ret != 0) {
-				gf_log (this->name, GF_LOG_WARNING,
-					"mismatching layouts for %s", 
-					local->loc.path);
-			
-				local->layout_mismatch = 1;
-
-				goto unlock;
-			}
-		} 
-		
-		dht_stat_merge (this, &local->stbuf, stbuf, prev->this);
-		
-		local->op_ret = 0;
-		local->stbuf.st_ino = local->st_ino;
-
-		if (!local->xattr)
-			local->xattr = dict_ref (xattr);
-	}
-unlock:
-	UNLOCK (&frame->lock);
+                is_dir = check_is_dir (inode, stbuf, xattr);
+                is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+                                                 conf->link_xattr_name);
 
+                if (is_linkfile) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "linkfile found in revalidate for %s",
+                                local->loc.path);
+                        local->return_estale = 1;
+
+                        goto unlock;
+                }
+
+                if (is_dir) {
+                        ret = dht_dir_has_layout (xattr, conf->xattr_name);
+                        if (ret >= 0) {
+                                if (is_greater_time(local->stbuf.ia_ctime,
+                                                    local->stbuf.ia_ctime_nsec,
+                                                    stbuf->ia_ctime,
+                                                    stbuf->ia_ctime_nsec)) {
+                                        local->prebuf.ia_gid = stbuf->ia_gid;
+                                        local->prebuf.ia_uid = stbuf->ia_uid;
+                                }
+                        }
+                        if (local->stbuf.ia_type != IA_INVAL)
+                        {
+                                if ((local->stbuf.ia_gid != stbuf->ia_gid) ||
+                                    (local->stbuf.ia_uid != stbuf->ia_uid)) {
+                                        local->need_selfheal = 1;
+                                }
+                        }
+                        ret = dht_layout_dir_mismatch (this, layout,
+                                                       prev->this, &local->loc,
+                                                       xattr);
+                        if (ret != 0) {
+                                gf_log (this->name, GF_LOG_INFO,
+                                        "mismatching layouts for %s",
+                                        local->loc.path);
+
+                                local->layout_mismatch = 1;
+
+                                goto unlock;
+                        }
+                }
+
+                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+                dht_iatt_merge (this, &local->postparent, postparent,
+                                prev->this);
+
+                local->op_ret = 0;
+
+                if (!local->xattr) {
+                        local->xattr = dict_ref (xattr);
+                } else if (is_dir) {
+                        dht_aggregate_xattr (local->xattr, xattr);
+                }
+        }
+unlock:
+        UNLOCK (&frame->lock);
+out:
         this_call_cnt = dht_frame_return (frame);
 
         if (is_last_call (this_call_cnt)) {
-		if (!S_ISDIR (local->stbuf.st_mode)
-		    && (local->hashed_subvol != local->cached_subvol)
-		    && (local->stbuf.st_nlink == 1))
-			local->stbuf.st_mode |= S_ISVTX;
-		
+                if (!IA_ISDIR (local->stbuf.ia_type)
+                    && (local->hashed_subvol != local->cached_subvol)
+                    && (local->stbuf.ia_nlink == 1)
+                    && (conf && conf->unhashed_sticky_bit)) {
+                        local->stbuf.ia_prot.sticky = 1;
+                }
+                if (local->need_selfheal) {
+                        local->need_selfheal = 0;
+                        uuid_copy (local->gfid, local->stbuf.ia_gfid);
+                        local->stbuf.ia_gid = local->prebuf.ia_gid;
+                        local->stbuf.ia_uid = local->prebuf.ia_uid;
+                        copy = create_frame (this, this->ctx->pool);
+                        if (copy) {
+                                copy_local = dht_local_init (copy, &local->loc,
+                                                             NULL, 0);
+                                if (!copy_local)
+                                        goto cont;
+                                copy_local->stbuf = local->stbuf;
+                                copy->local = copy_local;
+                                FRAME_SU_DO (copy, dht_local_t);
+                                ret = synctask_new (this->ctx->env,
+                                                    dht_dir_attr_heal,
+                                                    dht_dir_attr_heal_done,
+                                                    copy, copy);
+                        }
+                }
+cont:
 		if (local->layout_mismatch) {
-			local->op_ret = -1;
-			local->op_errno = ESTALE;
-		}
-			
-		DHT_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-				  local->inode, &local->stbuf, local->xattr);
-	}
+                        /* Found layout mismatch in the directory, need to
+                           fix this in the inode context */
+                        dht_layout_unref (this, local->layout);
+                        local->layout = NULL;
+                        dht_lookup_directory (frame, this, &local->loc);
+                        return 0;
+                }
 
-        return 0;
+                if (local->need_lookup_everywhere) {
+                        /* As the current layout gave ENOENT error, we would
+                           need a new layout */
+                        dht_layout_unref (this, local->layout);
+                        local->layout = NULL;
+
+                        /* We know that current cached subvol is no more
+                           valid, get the new one */
+                        local->cached_subvol = NULL;
+                        dht_lookup_everywhere (frame, this, &local->loc);
+                        return 0;
+                }
+                if (local->return_estale) {
+                        local->op_ret = -1;
+                        local->op_errno = ESTALE;
+                }
+
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
+
+                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
+                DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
+                                  local->inode, &local->stbuf, local->xattr,
+                                  &local->postparent);
+        }
+
+err:
+        return ret;
 }
 
 
 int
 dht_lookup_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-				xlator_t *this,
-				int32_t op_ret, int32_t op_errno,
-				inode_t *inode, struct stat *stbuf)
+                                xlator_t *this,
+                                int32_t op_ret, int32_t op_errno,
+                                inode_t *inode, struct iatt *stbuf,
+                                struct iatt *preparent, struct iatt *postparent,
+                                dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	dht_layout_t *layout = NULL;
-	xlator_t     *cached_subvol = NULL;
+        dht_local_t  *local = NULL;
+        xlator_t     *cached_subvol = NULL;
+        dht_conf_t   *conf = NULL;
+        int           ret = -1;
 
-	local = frame->local;
-	cached_subvol = local->cached_subvol;
-
-	layout = dht_layout_for_subvol (this, local->cached_subvol);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no pre-set layout for subvolume %s",
-			cached_subvol ? cached_subvol->name : "<nil>");
-		local->op_ret = -1;
-		local->op_errno = EINVAL;
-		goto unwind;
-	}
+        GF_VALIDATE_OR_GOTO ("dht", frame, out);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        cached_subvol = local->cached_subvol;
+        conf = this->private;
+
+        ret = dht_layout_preset (this, local->cached_subvol, inode);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "failed to set layout for subvolume %s",
+                        cached_subvol ? cached_subvol->name : "<nil>");
+                local->op_ret = -1;
+                local->op_errno = EINVAL;
+                goto unwind;
+        }
+
+        local->op_ret = 0;
+        if ((local->stbuf.ia_nlink == 1)
+            && (conf && conf->unhashed_sticky_bit)) {
+                local->stbuf.ia_prot.sticky = 1;
+        }
 
-	inode_ctx_put (local->inode, this, (uint64_t)(long)layout);
-	local->op_ret = 0;
-	if (local->stbuf.st_nlink == 1)
-		local->stbuf.st_mode |= S_ISVTX;
+        if (local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
+        }
 
 unwind:
-	DHT_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-			  local->inode, &local->stbuf, local->xattr);
-	return 0;
+        if (local->linked == _gf_true)
+                dht_linkfile_attr_heal (frame, this);
+
+        DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
+        DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
+                          local->inode, &local->stbuf, local->xattr,
+                          &local->postparent);
+out:
+        return ret;
+}
+
+
+int
+dht_lookup_everywhere_done (call_frame_t *frame, xlator_t *this)
+{
+        int           ret = 0;
+        dht_local_t  *local = NULL;
+        xlator_t     *hashed_subvol = NULL;
+        xlator_t     *cached_subvol = NULL;
+        dht_layout_t *layout = NULL;
+
+        local = frame->local;
+        hashed_subvol = local->hashed_subvol;
+        cached_subvol = local->cached_subvol;
+
+        if (local->file_count && local->dir_count) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "path %s exists as a file on one subvolume "
+                        "and directory on another. "
+                        "Please fix it manually",
+                        local->loc.path);
+                DHT_STACK_UNWIND (lookup, frame, -1, EIO, NULL, NULL, NULL,
+                                  NULL);
+                return 0;
+        }
+
+        if (local->dir_count) {
+                dht_lookup_directory (frame, this, &local->loc);
+                return 0;
+        }
+
+        if (!cached_subvol) {
+                DHT_STACK_UNWIND (lookup, frame, -1, ENOENT, NULL, NULL, NULL,
+                                  NULL);
+                return 0;
+        }
+
+        if (local->need_lookup_everywhere) {
+                if (uuid_compare (local->gfid, local->inode->gfid)) {
+                        /* GFID different, return error */
+                        DHT_STACK_UNWIND (lookup, frame, -1, ENOENT, NULL,
+                                          NULL, NULL, NULL);
+                        return 0;
+                }
+                local->op_ret = 0;
+                local->op_errno = 0;
+                layout = dht_layout_for_subvol (this, cached_subvol);
+                if (!layout) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "%s: no pre-set layout for subvolume %s",
+                                local->loc.path, (cached_subvol ?
+                                                  cached_subvol->name :
+                                                  "<nil>"));
+                }
+
+                ret = dht_layout_set (this, local->inode, layout);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "%s: failed to set layout for subvol %s",
+                                local->loc.path, (cached_subvol ?
+                                                  cached_subvol->name :
+                                                  "<nil>"));
+                }
+
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
+
+                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
+                DHT_STACK_UNWIND (lookup, frame, local->op_ret,
+                                  local->op_errno, local->inode,
+                                  &local->stbuf, local->xattr,
+                                  &local->postparent);
+                return 0;
+        }
+
+        if (!hashed_subvol) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "cannot create linkfile file for %s on %s: "
+                        "hashed subvolume cannot be found.",
+                        local->loc.path, cached_subvol->name);
+
+                local->op_ret = 0;
+                local->op_errno = 0;
+
+                ret = dht_layout_preset (frame->this, cached_subvol,
+                                         local->inode);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "failed to set layout for subvol %s",
+                                cached_subvol ? cached_subvol->name :
+                                "<nil>");
+                        local->op_ret = -1;
+                        local->op_errno = EINVAL;
+                }
+
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
+
+                DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
+                DHT_STACK_UNWIND (lookup, frame, local->op_ret,
+                                  local->op_errno, local->inode,
+                                  &local->stbuf, local->xattr,
+                                  &local->postparent);
+                return 0;
+        }
+
+        gf_log (this->name, GF_LOG_DEBUG,
+                "linking file %s existing on %s to %s (hash)",
+                local->loc.path, cached_subvol->name,
+                hashed_subvol->name);
+
+        ret = dht_linkfile_create (frame,
+                                   dht_lookup_linkfile_create_cbk, this,
+                                   cached_subvol, hashed_subvol, &local->loc);
+
+        return ret;
+}
+
+
+int
+dht_lookup_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno,
+                       struct iatt *preparent, struct iatt *postparent,
+                       dict_t *xdata)
+{
+        int  this_call_cnt = 0;
+
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+                dht_lookup_everywhere_done (frame, this);
+        }
+
+        return 0;
 }
 
 
 int
 dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno,
-			   inode_t *inode, struct stat *buf, dict_t *xattr)
+                           int32_t op_ret, int32_t op_errno,
+                           inode_t *inode, struct iatt *buf, dict_t *xattr,
+                           struct iatt *postparent)
 {
-	dht_conf_t   *conf          = NULL;
         dht_local_t  *local         = NULL;
         int           this_call_cnt = 0;
         call_frame_t *prev          = NULL;
-	int           is_linkfile   = 0;
-	int           is_dir        = 0;
-	xlator_t     *subvol        = NULL;
-	loc_t        *loc           = NULL;
-	xlator_t     *link_subvol   = NULL;
-	xlator_t     *hashed_subvol = NULL;
-	xlator_t     *cached_subvol = NULL;
-
-	conf   = this->private;
-
-	local  = frame->local;
-	loc    = &local->loc;
-
-	prev   = cookie;
-	subvol = prev->this;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			if (op_errno != ENOENT)
-				local->op_errno = op_errno;
-			goto unlock;
-		}
+        int           is_linkfile   = 0;
+        int           is_dir        = 0;
+        xlator_t     *subvol        = NULL;
+        loc_t        *loc           = NULL;
+        xlator_t     *link_subvol   = NULL;
+        int           ret           = -1;
+        int32_t       fd_count      = 0;
+        dht_conf_t   *conf          = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, out);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
+
+        local  = frame->local;
+        loc    = &local->loc;
+        conf   = this->private;
 
-		is_linkfile = check_is_linkfile (inode, buf, xattr);
-		is_dir = check_is_dir (inode, buf, xattr);
-
-		if (is_linkfile) {
-			link_subvol = dht_linkfile_subvol (this, inode, buf,
-							   xattr);
-			gf_log (this->name, GF_LOG_WARNING,
-				"found on %s linkfile %s (-> %s)",
-				subvol->name, loc->path,
-				link_subvol ? link_subvol->name : "''");
-			goto unlock;
-		}
+        prev   = cookie;
+        subvol = prev->this;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        if (op_errno != ENOENT)
+                                local->op_errno = op_errno;
+                        goto unlock;
+                }
+
+                if (uuid_is_null (local->gfid))
+                        uuid_copy (local->gfid, buf->ia_gfid);
+
+                if (uuid_compare (local->gfid, buf->ia_gfid)) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: gfid differs on subvolume %s",
+                                loc->path, prev->this->name);
+                }
+
+                is_linkfile = check_is_linkfile (inode, buf, xattr,
+                                                 conf->link_xattr_name);
+                is_dir = check_is_dir (inode, buf, xattr);
+
+                if (is_linkfile) {
+                        link_subvol = dht_linkfile_subvol (this, inode, buf,
+                                                           xattr);
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "found on %s linkfile %s (-> %s)",
+                                subvol->name, loc->path,
+                                link_subvol ? link_subvol->name : "''");
+                        goto unlock;
+                }
+
+                /* non linkfile GFID takes precedence */
+                uuid_copy (local->gfid, buf->ia_gfid);
 
                 if (is_dir) {
                         local->dir_count++;
 
-                        gf_log (this->name, GF_LOG_WARNING,
+                        gf_log (this->name, GF_LOG_DEBUG,
                                 "found on %s directory %s",
                                 subvol->name, loc->path);
                 } else {
@@ -391,157 +998,183 @@ dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
                         if (!local->cached_subvol) {
                                 /* found one file */
-                                dht_stat_merge (this, &local->stbuf, buf,
+                                dht_iatt_merge (this, &local->stbuf, buf,
                                                 subvol);
                                 local->xattr = dict_ref (xattr);
                                 local->cached_subvol = subvol;
                                 gf_log (this->name, GF_LOG_DEBUG,
                                         "found on %s file %s",
                                         subvol->name, loc->path);
+
+                                dht_iatt_merge (this, &local->postparent,
+                                                postparent, subvol);
                         } else {
+                                /* This is where we need 'rename' both entries logic */
                                 gf_log (this->name, GF_LOG_WARNING,
                                         "multiple subvolumes (%s and %s) have "
-                                        "file %s", local->cached_subvol->name,
+                                        "file %s (preferably rename the file "
+                                        "in the backend, and do a fresh lookup)",
+                                        local->cached_subvol->name,
                                         subvol->name, local->loc.path);
                         }
                 }
-	}
+        }
 unlock:
-	UNLOCK (&frame->lock);
-
-	if (is_linkfile) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"deleting stale linkfile %s on %s",
-			loc->path, subvol->name);
-		dht_linkfile_unlink (frame, this, subvol, loc);
-	}
-
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt)) {
-		hashed_subvol = local->hashed_subvol;
-		cached_subvol = local->cached_subvol;
-
-                if (local->file_count && local->dir_count) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "path %s is both file and directory at the " 
-                                "backend. Please fix it manually",
-                                loc->path);
-                        DHT_STACK_UNWIND (frame, -1, EIO, NULL, NULL, NULL);
-                        return 0;
-                }
+        UNLOCK (&frame->lock);
 
-                if (local->dir_count) {
-                        dht_lookup_directory (frame, this, &local->loc);
+        if (is_linkfile) {
+                ret = dict_get_int32 (xattr, GLUSTERFS_OPEN_FD_COUNT, &fd_count);
+                /* Delete the linkfile only if there are no open fds on it.
+                   if there is a open-fd, it may be in migration */
+                if (!ret && (fd_count == 0)) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "deleting stale linkfile %s on %s",
+                                loc->path, subvol->name);
+                        STACK_WIND (frame, dht_lookup_unlink_cbk,
+                                    subvol, subvol->fops->unlink, loc, 0, NULL);
                         return 0;
                 }
+        }
 
-		if (!cached_subvol) {
-			DHT_STACK_UNWIND (frame, -1, ENOENT, NULL, NULL, NULL);
-			return 0;
-		}
-
-		gf_log (this->name, GF_LOG_WARNING,
-			"linking file %s existing on %s to %s (hash)",
-			loc->path, cached_subvol->name, hashed_subvol->name);
-
-		dht_linkfile_create (frame, dht_lookup_linkfile_create_cbk,
-				     cached_subvol, hashed_subvol, loc);
-	}
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+                dht_lookup_everywhere_done (frame, this);
+        }
 
-	return 0;
+out:
+        return ret;
 }
 
 
 int
 dht_lookup_everywhere (call_frame_t *frame, xlator_t *this, loc_t *loc)
 {
-	dht_conf_t     *conf = NULL;
-	dht_local_t    *local = NULL;
-	int             i = 0;
-	int             call_cnt = 0;
+        dht_conf_t     *conf = NULL;
+        dht_local_t    *local = NULL;
+        int             i = 0;
+        int             call_cnt = 0;
 
-	conf = this->private;
-	local = frame->local;
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
+        GF_VALIDATE_OR_GOTO ("dht", loc, out);
 
-	call_cnt = conf->subvolume_cnt;
-	local->call_cnt = call_cnt;
+        conf = this->private;
+        local = frame->local;
 
-	if (!local->inode)
-		local->inode = inode_ref (loc->inode);
+        call_cnt = conf->subvolume_cnt;
+        local->call_cnt = call_cnt;
 
-	for (i = 0; i < call_cnt; i++) {
-		STACK_WIND (frame, dht_lookup_everywhere_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->lookup,
-			    loc, local->xattr_req);
-	}
+        if (!local->inode)
+                local->inode = inode_ref (loc->inode);
 
-	return 0;
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (frame, dht_lookup_everywhere_cbk,
+                            conf->subvolumes[i],
+                            conf->subvolumes[i]->fops->lookup,
+                            loc, local->xattr_req);
+        }
+
+        return 0;
+out:
+        DHT_STACK_UNWIND (lookup, frame, -1, EINVAL, NULL, NULL, NULL, NULL);
+err:
+        return -1;
 }
 
 
 int
 dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
                          xlator_t *this, int op_ret, int op_errno,
-                         inode_t *inode, struct stat *stbuf, dict_t *xattr)
+                         inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                         struct iatt *postparent)
 {
-        call_frame_t *prev = NULL;
-	dht_local_t  *local = NULL;
-	dht_layout_t *layout = NULL;
-	xlator_t     *subvol = NULL;
-	loc_t        *loc = NULL;
+        call_frame_t *prev          = NULL;
+        dht_local_t  *local         = NULL;
+        xlator_t     *subvol        = NULL;
+        loc_t        *loc           = NULL;
+        dht_conf_t   *conf          = NULL;
+        int           ret           = 0;
 
-        prev   = cookie;
-	subvol = prev->this;
+        GF_VALIDATE_OR_GOTO ("dht", frame, out);
+        GF_VALIDATE_OR_GOTO ("dht", this, unwind);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, unwind);
+        GF_VALIDATE_OR_GOTO ("dht", this->private, unwind);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, unwind);
 
-	local  = frame->local;
-	loc    = &local->loc;
+        prev   = cookie;
+        subvol = prev->this;
+        conf   = this->private;
+        local  = frame->local;
+        loc    = &local->loc;
 
         if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"lookup of %s on %s (following linkfile) failed (%s)",
-			local->loc.path, subvol->name, strerror (op_errno));
-                goto err;
-	}
+                gf_log (this->name, GF_LOG_INFO,
+                        "lookup of %s on %s (following linkfile) failed (%s)",
+                        local->loc.path, subvol->name, strerror (op_errno));
+
+                /* If cached subvol returned ENOTCONN, do not do
+                lookup_everywhere. We need to make sure linkfile does not get
+                removed, which can take away the namespace, and subvol is
+                anyways down. */
+
+                if (op_errno != ENOTCONN)
+                        goto err;
+                else
+                        goto unwind;
+        }
 
         if (check_is_dir (inode, stbuf, xattr)) {
-                gf_log (this->name, GF_LOG_WARNING,
+                gf_log (this->name, GF_LOG_INFO,
                         "lookup of %s on %s (following linkfile) reached dir",
                         local->loc.path, subvol->name);
                 goto err;
         }
 
-        if (check_is_linkfile (inode, stbuf, xattr)) {
-                gf_log (this->name, GF_LOG_WARNING,
+        if (check_is_linkfile (inode, stbuf, xattr, conf->link_xattr_name)) {
+                gf_log (this->name, GF_LOG_INFO,
                         "lookup of %s on %s (following linkfile) reached link",
                         local->loc.path, subvol->name);
                 goto err;
         }
 
-	if (stbuf->st_nlink == 1)
-		stbuf->st_mode |= S_ISVTX;
-        dht_itransform (this, prev->this, stbuf->st_ino, &stbuf->st_ino);
+        if (uuid_compare (local->gfid, stbuf->ia_gfid)) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: gfid different on data file on %s",
+                        local->loc.path, subvol->name);
+                goto err;
+        }
 
-	layout = dht_layout_for_subvol (this, prev->this);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no pre-set layout for subvolume %s",
-			prev->this->name);
-		op_ret   = -1;
-		op_errno = EINVAL;
-		goto out;
-	}
+        if ((stbuf->ia_nlink == 1)
+            && (conf && conf->unhashed_sticky_bit)) {
+                stbuf->ia_prot.sticky = 1;
+        }
 
-	inode_ctx_put (inode, this, (uint64_t)(long)layout);
+        ret = dht_layout_preset (this, prev->this, inode);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "failed to set layout for subvolume %s",
+                        prev->this->name);
+                op_ret   = -1;
+                op_errno = EINVAL;
+        }
 
-out:
-        DHT_STACK_UNWIND (frame, op_ret, op_errno, inode, stbuf, xattr);
+        if (local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
+        }
+
+unwind:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                          postparent);
 
         return 0;
 
 err:
         dht_lookup_everywhere (frame, this, loc);
-
+out:
         return 0;
 }
 
@@ -553,21 +1186,38 @@ dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc)
         int           i = 0;
         dht_conf_t   *conf = NULL;
         dht_local_t  *local = NULL;
+        int           ret = 0;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, out);
+        GF_VALIDATE_OR_GOTO ("dht", this, unwind);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, unwind);
+        GF_VALIDATE_OR_GOTO ("dht", this->private, unwind);
+        GF_VALIDATE_OR_GOTO ("dht", loc, unwind);
 
         conf = this->private;
         local = frame->local;
 
         call_cnt        = conf->subvolume_cnt;
         local->call_cnt = call_cnt;
-		
+
         local->layout = dht_layout_new (this, conf->subvolume_cnt);
         if (!local->layout) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "memory allocation failed :(");
-                DHT_STACK_UNWIND (frame, -1, ENOMEM, NULL, NULL, NULL);
-                return 0;
+                goto unwind;
+        }
+
+        if (local->xattr != NULL) {
+                dict_unref (local->xattr);
+                local->xattr = NULL;
+        }
+
+        if (!uuid_is_null (local->gfid)) {
+                ret = dict_set_static_bin (local->xattr_req, "gfid-req",
+                                           local->gfid, 16);
+                if (ret)
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: failed to set gfid", local->loc.path);
         }
-		
+
         for (i = 0; i < call_cnt; i++) {
                 STACK_WIND (frame, dht_lookup_dir_cbk,
                             conf->subvolumes[i],
@@ -575,23 +1225,35 @@ dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc)
                             &local->loc, local->xattr_req);
         }
         return 0;
+unwind:
+        DHT_STACK_UNWIND (lookup, frame, -1, ENOMEM, NULL, NULL, NULL, NULL);
+out:
+        return 0;
+
 }
 
 
 int
 dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                 int op_ret, int op_errno,
-                inode_t *inode, struct stat *stbuf, dict_t *xattr)
+                inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                struct iatt *postparent)
 {
-	dht_layout_t *layout      = NULL;
-        char          is_linkfile = 0;
-        char          is_dir      = 0;
-        xlator_t     *subvol      = NULL;
-        dht_conf_t   *conf        = NULL;
-        dht_local_t  *local       = NULL;
-        loc_t        *loc         = NULL;
-        call_frame_t *prev        = NULL;
+        char          is_linkfile   = 0;
+        char          is_dir        = 0;
+        xlator_t     *subvol        = NULL;
+        dht_conf_t   *conf          = NULL;
+        dht_local_t  *local         = NULL;
+        loc_t        *loc           = NULL;
+        call_frame_t *prev          = NULL;
+        int           ret           = 0;
+        dht_layout_t *parent_layout = NULL;
 
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+        GF_VALIDATE_OR_GOTO ("dht", this->private, out);
 
         conf  = this->private;
 
@@ -599,76 +1261,140 @@ dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         local = frame->local;
         loc   = &local->loc;
 
-	if (ENTRY_MISSING (op_ret, op_errno)) {
-		if (conf->search_unhashed) {
-			local->op_errno = ENOENT;
-			dht_lookup_everywhere (frame, this, loc);
-			return 0;
-		}
-	}
+        /* This is required for handling stale linkfile deletion,
+         * or any more call which happens from this 'loc'.
+         */
+        if (!op_ret && uuid_is_null (local->gfid))
+                memcpy (local->gfid, stbuf->ia_gfid, 16);
+
+        if (ENTRY_MISSING (op_ret, op_errno)) {
+                gf_log (this->name, GF_LOG_TRACE, "Entry %s missing on subvol"
+                        " %s", loc->path, prev->this->name);
+                if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_ON) {
+                        local->op_errno = ENOENT;
+                        dht_lookup_everywhere (frame, this, loc);
+                        return 0;
+                }
+                if ((conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) &&
+                    (loc->parent)) {
+                        ret = dht_inode_ctx_layout_get (loc->parent, this,
+                                                        &parent_layout);
+                        if (ret || !parent_layout)
+                                goto out;
+                        if (parent_layout->search_unhashed) {
+                                local->op_errno = ENOENT;
+                                dht_lookup_everywhere (frame, this, loc);
+                                return 0;
+                        }
+                }
+        }
 
- 	if (op_ret == 0) {
- 		is_dir      = check_is_dir (inode, stbuf, xattr);
- 		if (is_dir) {
- 			local->inode = inode_ref (inode);
- 			local->xattr = dict_ref (xattr);
- 		}
- 	}
+        if (op_ret == 0) {
+                is_dir      = check_is_dir (inode, stbuf, xattr);
+                if (is_dir) {
+                        local->inode = inode_ref (inode);
+                        local->xattr = dict_ref (xattr);
+                }
+        }
 
- 	if (is_dir || (op_ret == -1 && op_errno == ENOTCONN)) {
+        if (is_dir || (op_ret == -1 && op_errno == ENOTCONN)) {
                 dht_lookup_directory (frame, this, &local->loc);
                 return 0;
- 	}
- 
-        if (op_ret == -1)
+        }
+
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_DEBUG, "Lookup of %s for subvolume"
+                       " %s failed with error %s", loc->path, prev->this->name,
+                       strerror (op_errno));
                 goto out;
+        }
 
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr);
-        is_dir      = check_is_dir (inode, stbuf, xattr);
+        is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+                                         conf->link_xattr_name);
 
-        if (!is_dir && !is_linkfile) {
+        if (!is_linkfile) {
                 /* non-directory and not a linkfile */
 
-		dht_itransform (this, prev->this, stbuf->st_ino,
-				&stbuf->st_ino);
-
-		layout = dht_layout_for_subvol (this, prev->this);
-		if (!layout) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"no pre-set layout for subvolume %s",
-				prev->this->name);
-			op_ret   = -1;
-			op_errno = EINVAL;
-			goto out;
-		}
-
-                inode_ctx_put (inode, this, (uint64_t)(long)layout);
-                goto out; 
-	}
-
-        if (is_linkfile) {
-                subvol = dht_linkfile_subvol (this, inode, stbuf, xattr);
-
-                if (!subvol) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "linkfile not having link subvolume. path=%s",
-                                loc->path);
-			dht_lookup_everywhere (frame, this, loc);
-			return 0;
+                ret = dht_layout_preset (this, prev->this, inode);
+                if (ret < 0) {
+                        gf_log (this->name, GF_LOG_INFO,
+                                "could not set pre-set layout for subvolume %s",
+                                prev->this->name);
+                        op_ret   = -1;
+                        op_errno = EINVAL;
+                        goto out;
                 }
+                goto out;
+        }
 
-		STACK_WIND (frame, dht_lookup_linkfile_cbk,
-			    subvol, subvol->fops->lookup,
-			    &local->loc, local->xattr_req);
+        subvol = dht_linkfile_subvol (this, inode, stbuf, xattr);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "linkfile not having link subvolume. path=%s",
+                        loc->path);
+                dht_lookup_everywhere (frame, this, loc);
+                return 0;
         }
 
+        STACK_WIND (frame, dht_lookup_linkfile_cbk,
+                    subvol, subvol->fops->lookup,
+                    &local->loc, local->xattr_req);
+
         return 0;
 
 out:
-        DHT_STACK_UNWIND (frame, op_ret, op_errno, inode, stbuf, xattr);
+        /*
+         * FIXME: postparent->ia_size and postparent->st_blocks do not have
+         * correct values. since, postparent corresponds to a directory these
+         * two members should have values equal to sum of corresponding values
+         * from each of the subvolume. See dht_iatt_merge for reference.
+         */
+
+        if (!op_ret && local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
+        }
+
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                          postparent);
+err:
         return 0;
 }
 
+/* For directories, check if acl xattrs have been requested (by the acl xlator),
+ * if not, request for them. These xattrs are needed for dht dir self-heal to
+ * perform proper self-healing of dirs
+ */
+void
+dht_check_and_set_acl_xattr_req (inode_t *inode, dict_t *xattr_req)
+{
+        int     ret = 0;
+
+        GF_ASSERT (inode);
+        GF_ASSERT (xattr_req);
+
+        if (inode->ia_type != IA_IFDIR)
+                return;
+
+        if (!dict_get (xattr_req, POSIX_ACL_ACCESS_XATTR)) {
+                ret = dict_set_int8 (xattr_req, POSIX_ACL_ACCESS_XATTR, 0);
+                if (ret)
+                        gf_log (THIS->name, GF_LOG_WARNING,
+                                "failed to set key %s",
+                                POSIX_ACL_ACCESS_XATTR);
+        }
+
+        if (!dict_get (xattr_req, POSIX_ACL_DEFAULT_XATTR)) {
+                ret = dict_set_int8 (xattr_req, POSIX_ACL_DEFAULT_XATTR, 0);
+                if (ret)
+                        gf_log (THIS->name, GF_LOG_WARNING,
+                                "failed to set key %s",
+                                POSIX_ACL_DEFAULT_XATTR);
+        }
+
+        return;
+}
 
 int
 dht_lookup (call_frame_t *frame, xlator_t *this,
@@ -676,125 +1402,163 @@ dht_lookup (call_frame_t *frame, xlator_t *this,
 {
         xlator_t     *subvol = NULL;
         xlator_t     *hashed_subvol = NULL;
-        xlator_t     *cached_subvol = NULL;
         dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
+        dht_conf_t   *conf = NULL;
         int           ret    = -1;
         int           op_errno = -1;
-	dht_layout_t *layout = NULL;
-	int           i = 0;
-	int           call_cnt = 0;
-
+        dht_layout_t *layout = NULL;
+        int           i = 0;
+        int           call_cnt = 0;
+        loc_t         new_loc = {0,};
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
 
-	conf = this->private;
-
-        local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        conf = this->private;
+        if (!conf)
+                goto err;
 
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = errno;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "copying location failed for path=%s",
-                        loc->path);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_LOOKUP);
+        if (!local) {
+                op_errno = ENOMEM;
                 goto err;
         }
-	
-	if (xattr_req) {
-		local->xattr_req = dict_ref (xattr_req);
-	} else {
-		local->xattr_req = dict_new ();
-	}
 
-	hashed_subvol = dht_subvol_get_hashed (this, loc);
-	cached_subvol = dht_subvol_get_cached (this, loc->inode);
+        ret = dht_filter_loc_subvol_key (this, loc, &new_loc,
+                                         &hashed_subvol);
+        if (ret) {
+                loc_wipe (&local->loc);
+                ret = loc_dup (&new_loc, &local->loc);
 
-	local->cached_subvol = cached_subvol;
-	local->hashed_subvol = hashed_subvol;
+                /* we no more need 'new_loc' entries */
+                loc_wipe (&new_loc);
 
-        if (is_revalidate (loc)) {
-		layout = dht_layout_get (this, loc->inode);
+                /* check if loc_dup() is successful */
+                if (ret == -1) {
+                        op_errno = errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "copying location failed for path=%s",
+                                loc->path);
+                        goto err;
+                }
+        }
+
+        if (xattr_req) {
+                local->xattr_req = dict_ref (xattr_req);
+        } else {
+                local->xattr_req = dict_new ();
+        }
+
+        if (uuid_is_null (loc->pargfid) && !uuid_is_null (loc->gfid) &&
+            !__is_root_gfid (loc->inode->gfid)) {
+                local->cached_subvol = NULL;
+                dht_discover (frame, this, loc);
+                return 0;
+        }
 
+        if (!hashed_subvol)
+                hashed_subvol = dht_subvol_get_hashed (this, loc);
+        local->hashed_subvol = hashed_subvol;
+
+        if (is_revalidate (loc)) {
+                layout = local->layout;
                 if (!layout) {
-                        gf_log (this->name, GF_LOG_ERROR,
+                        gf_log (this->name, GF_LOG_DEBUG,
                                 "revalidate without cache. path=%s",
                                 loc->path);
                         op_errno = EINVAL;
                         goto err;
                 }
 
-		if (layout->gen && (layout->gen < conf->gen)) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"incomplete layout failure for path=%s",
-				loc->path);
-			op_errno = ESTALE;
-			goto err;
-		}
+                if (layout->gen && (layout->gen < conf->gen)) {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "incomplete layout failure for path=%s",
+                                loc->path);
+
+                        dht_layout_unref (this, local->layout);
+                        local->layout = NULL;
+                        local->cached_subvol = NULL;
+                        goto do_fresh_lookup;
+                }
+
+                local->inode = inode_ref (loc->inode);
+
+                /* NOTE: we don't require 'trusted.glusterfs.dht.linkto' attribute,
+                 *       revalidates directly go to the cached-subvolume.
+                 */
+                ret = dict_set_uint32 (local->xattr_req,
+                                       conf->xattr_name, 4 * 4);
+
+                if (IA_ISDIR (local->inode->ia_type)) {
+                        local->call_cnt = call_cnt = conf->subvolume_cnt;
+                        for (i = 0; i < call_cnt; i++) {
+                                STACK_WIND (frame, dht_revalidate_cbk,
+                                            conf->subvolumes[i],
+                                            conf->subvolumes[i]->fops->lookup,
+                                            loc, local->xattr_req);
+                        }
+                        return 0;
+                }
 
-		local->inode    = inode_ref (loc->inode);
-		local->st_ino   = loc->inode->ino;
-		
-		local->call_cnt = layout->cnt;
-		call_cnt = local->call_cnt;
-		
-		/* NOTE: we don't require 'trusted.glusterfs.dht.linkto' attribute,
-		 *       revalidates directly go to the cached-subvolume.
-		 */
-		ret = dict_set_uint32 (local->xattr_req, 
-				       "trusted.glusterfs.dht", 4 * 4);
-
-		for (i = 0; i < layout->cnt; i++) {
+                call_cnt = local->call_cnt = layout->cnt;
+
+                /* need it for self-healing linkfiles which is
+                   'in-migration' state */
+                ret = dict_set_uint32 (local->xattr_req,
+                                       GLUSTERFS_OPEN_FD_COUNT, 4);
+
+                /* need it for dir self-heal */
+                dht_check_and_set_acl_xattr_req (loc->inode, local->xattr_req);
+
+		for (i = 0; i < call_cnt; i++) {
 			subvol = layout->list[i].xlator;
-			
+
 			STACK_WIND (frame, dht_revalidate_cbk,
 				    subvol, subvol->fops->lookup,
-				    loc, local->xattr_req);
+				    &local->loc, local->xattr_req);
 
-			if (!--call_cnt)
-				break;
 		}
         } else {
-		/* TODO: remove the hard-coding */
-		ret = dict_set_uint32 (local->xattr_req, 
-				       "trusted.glusterfs.dht", 4 * 4);
+        do_fresh_lookup:
+                /* TODO: remove the hard-coding */
+                ret = dict_set_uint32 (local->xattr_req,
+                                       conf->xattr_name, 4 * 4);
 
-		ret = dict_set_uint32 (local->xattr_req, 
-				       "trusted.glusterfs.dht.linkto", 256);
+                ret = dict_set_uint32 (local->xattr_req,
+                                       conf->link_xattr_name, 256);
+
+                /* need it for self-healing linkfiles which is
+                   'in-migration' state */
+                ret = dict_set_uint32 (local->xattr_req,
+                                       GLUSTERFS_OPEN_FD_COUNT, 4);
+
+                /* need it for dir self-heal */
+                dht_check_and_set_acl_xattr_req (loc->inode, local->xattr_req);
 
                 if (!hashed_subvol) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"no subvolume in layout for path=%s, "
-				"checking on all the subvols to see if "
-				"it is a directory", loc->path);
- 			call_cnt        = conf->subvolume_cnt;
- 			local->call_cnt = call_cnt;
- 			
- 			local->layout = dht_layout_new (this, conf->subvolume_cnt);
- 			if (!local->layout) {
- 				op_errno = ENOMEM;
- 				gf_log (this->name, GF_LOG_ERROR,
- 					"memory allocation failed :(");
- 				goto err;
- 			}
-
-			for (i = 0; i < call_cnt; i++) {
- 				STACK_WIND (frame, dht_lookup_dir_cbk,
- 					    conf->subvolumes[i],
- 					    conf->subvolumes[i]->fops->lookup,
- 					    &local->loc, local->xattr_req);
- 			}
- 			return 0;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "no subvolume in layout for path=%s, "
+                                "checking on all the subvols to see if "
+                                "it is a directory", loc->path);
+                        call_cnt        = conf->subvolume_cnt;
+                        local->call_cnt = call_cnt;
+
+                        local->layout = dht_layout_new (this,
+                                                        conf->subvolume_cnt);
+                        if (!local->layout) {
+                                op_errno = ENOMEM;
+                                goto err;
+                        }
+
+                        for (i = 0; i < call_cnt; i++) {
+                                STACK_WIND (frame, dht_lookup_dir_cbk,
+                                            conf->subvolumes[i],
+                                            conf->subvolumes[i]->fops->lookup,
+                                            &local->loc, local->xattr_req);
+                        }
+                        return 0;
                 }
 
                 STACK_WIND (frame, dht_lookup_cbk,
@@ -805,1355 +1569,1660 @@ dht_lookup (call_frame_t *frame, xlator_t *this,
         return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL);
         return 0;
 }
 
 
 int
-dht_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	      int op_ret, int op_errno, struct stat *stbuf)
+dht_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                int op_ret, int op_errno, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
+        dht_local_t  *local = NULL;
+        call_frame_t *prev  = NULL;
 
+        local = frame->local;
+        prev  = cookie;
 
-	local = frame->local;
-	prev = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_ERROR,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_ret   = -1;
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "subvolume %s returned -1 (%s)",
+                                prev->this->name, strerror (op_errno));
+                        goto unlock;
+                }
 
-		dht_stat_merge (this, &local->stbuf, stbuf, prev->this);
-		
-		if (local->inode)
-			local->stbuf.st_ino = local->inode->ino;
-		local->op_ret = 0;
-	}
+                local->op_ret = 0;
+
+                local->postparent = *postparent;
+                local->preparent = *preparent;
+
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->preparent, 0);
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
+        }
 unlock:
-	UNLOCK (&frame->lock);
+        UNLOCK (&frame->lock);
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-				  &local->stbuf);
+        DHT_STACK_UNWIND (unlink, frame, local->op_ret, local->op_errno,
+                          &local->preparent, &local->postparent, NULL);
 
         return 0;
 }
 
 
 int
-dht_stat (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc)
+dht_unlink_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, struct iatt *preparent,
+                         struct iatt *postparent, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
-	dht_layout_t *layout = NULL;
-	int           i = 0;
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
 
+        xlator_t *cached_subvol = NULL;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        local = frame->local;
+        prev  = cookie;
 
-	layout = dht_layout_get (this, loc->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no layout for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        LOCK (&frame->lock);
+        {
+                if ((op_ret == -1) && !((op_errno == ENOENT) ||
+                                        (op_errno == ENOTCONN))) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "subvolume %s returned -1 (%s)",
+                                prev->this->name, strerror (op_errno));
+                        goto unlock;
+                }
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+                local->op_ret = 0;
+        }
+unlock:
+        UNLOCK (&frame->lock);
 
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = layout->cnt;
+        if (local->op_ret == -1)
+                goto err;
 
-	for (i = 0; i < layout->cnt; i++) {
-		subvol = layout->list[i].xlator;
+        cached_subvol = dht_subvol_get_cached (this, local->loc.inode);
+        if (!cached_subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s",
+                        local->loc.path);
+                local->op_errno = EINVAL;
+                goto err;
+        }
 
-		STACK_WIND (frame, dht_attr_cbk,
-			    subvol, subvol->fops->stat,
-			    loc);
-	}
+        STACK_WIND (frame, dht_unlink_cbk,
+                    cached_subvol, cached_subvol->fops->unlink,
+                    &local->loc, local->flags, NULL);
 
-	return 0;
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
-
-	return 0;
+        DHT_STACK_UNWIND (unlink, frame, -1, local->op_errno,
+                          NULL, NULL, NULL);
+        return 0;
 }
 
-
 int
-dht_fstat (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd)
+dht_err_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+             int op_ret, int op_errno, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
-	dht_layout_t *layout = NULL;
-	int           i = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-	layout = dht_layout_get (this, fd->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no layout for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"local allocation failed :(");
-		goto err;
-	}
+        local = frame->local;
+        prev = cookie;
 
-	local->inode    = inode_ref (fd->inode);
-	local->call_cnt = layout->cnt;;
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "subvolume %s returned -1 (%s)",
+                                prev->this->name, strerror (op_errno));
+                        goto unlock;
+                }
 
-	for (i = 0; i < layout->cnt; i++) {
-		subvol = layout->list[i].xlator;
-		STACK_WIND (frame, dht_attr_cbk,
-			    subvol, subvol->fops->fstat,
-			    fd);
-	}
+                local->op_ret = 0;
+        }
+unlock:
+        UNLOCK (&frame->lock);
 
-	return 0;
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+                DHT_STACK_UNWIND (setxattr, frame, local->op_ret,
+                                  local->op_errno, NULL);
+        }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        return 0;
+}
 
-	return 0;
+static void
+fill_layout_info (dht_layout_t *layout, char *buf)
+{
+        int i = 0;
+        char tmp_buf[128] = {0,};
+
+        for (i = 0; i < layout->cnt; i++) {
+                snprintf (tmp_buf, 128, "(%s %u %u)",
+                          layout->list[i].xlator->name,
+                          layout->list[i].start,
+                          layout->list[i].stop);
+                if (i)
+                        strcat (buf, " ");
+                strcat (buf, tmp_buf);
+        }
 }
 
+void
+dht_fill_pathinfo_xattr (xlator_t *this, dht_local_t *local,
+                         char *xattr_buf, int32_t alloc_len,
+                         int flag, char *layout_buf)
+{
+        if (flag && local->xattr_val)
+                snprintf (xattr_buf, alloc_len,
+                          "((<"DHT_PATHINFO_HEADER"%s> %s) (%s-layout %s))",
+                          this->name, local->xattr_val, this->name,
+                          layout_buf);
+        else if (local->xattr_val)
+                snprintf (xattr_buf, alloc_len,
+                          "(<"DHT_PATHINFO_HEADER"%s> %s)",
+                          this->name, local->xattr_val);
+        else if (flag)
+                snprintf (xattr_buf, alloc_len, "(%s-layout %s)",
+                          this->name, layout_buf);
+}
 
 int
-dht_chmod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode)
+dht_vgetxattr_alloc_and_fill (dht_local_t *local, dict_t *xattr, xlator_t *this,
+                              int op_errno)
 {
-	dht_layout_t *layout = NULL;
-	dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-	int           i = -1;
+        int      ret       = -1;
+        char    *value     = NULL;
+        int32_t  plen      = 0;
+
+        ret = dict_get_str (xattr, local->xsel, &value);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Subvolume %s returned -1 (%s)", this->name,
+                        strerror (op_errno));
+                local->op_ret = -1;
+                local->op_errno = op_errno;
+                goto out;
+        }
 
+        local->alloc_len += strlen(value);
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        if (!local->xattr_val) {
+                local->alloc_len += (strlen (DHT_PATHINFO_HEADER) + 10);
+                local->xattr_val = GF_CALLOC (local->alloc_len, sizeof (char),
+                                              gf_common_mt_char);
+                if (!local->xattr_val) {
+                        ret = -1;
+                        goto out;
+                }
+        }
 
-	layout = dht_layout_get (this, loc->inode);
+        if (local->xattr_val) {
+                plen = strlen (local->xattr_val);
+                if (plen) {
+                        /* extra byte(s) for \0 to be safe */
+                        local->alloc_len += (plen + 2);
+                        local->xattr_val = GF_REALLOC (local->xattr_val,
+                                                       local->alloc_len);
+                        if (!local->xattr_val) {
+                                ret = -1;
+                                goto out;
+                        }
+                }
 
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no layout for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+                (void) strcat (local->xattr_val, value);
+                (void) strcat (local->xattr_val, " ");
+                local->op_ret = 0;
+        }
 
-	if (!layout_is_sane (layout)) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"layout is not sane for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        ret = 0;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+ out:
+        return ret;
+}
 
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = layout->cnt;
+int
+dht_vgetxattr_fill_and_set (dht_local_t *local, dict_t **dict, xlator_t *this,
+                            gf_boolean_t flag)
+{
+        int   ret             = -1;
+        char *xattr_buf       = NULL;
+        char layout_buf[8192] = {0,};
 
-	for (i = 0; i < layout->cnt; i++) {
-		STACK_WIND (frame, dht_attr_cbk,
-			    layout->list[i].xlator,
-			    layout->list[i].xlator->fops->chmod,
-			    loc, mode);
-	}
+        if (flag)
+                fill_layout_info (local->layout, layout_buf);
 
-	return 0;
+        *dict = dict_new ();
+        if (!*dict)
+                goto out;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        local->xattr_val[strlen (local->xattr_val) - 1] = '\0';
+
+        /* we would need max this many bytes to create xattr string
+         * extra 40 bytes is just an estimated amount of additional
+         * space required as we include translator name and some
+         * spaces, brackets etc. when forming the pathinfo string.
+         *
+         * For node-uuid we just don't have all the pretty formatting,
+         * but since this is a generic routine for pathinfo & node-uuid
+         * we dont have conditional space allocation and try to be
+         * generic
+         */
+        local->alloc_len += (2 * strlen (this->name))
+                + strlen (layout_buf)
+                + 40;
+        xattr_buf = GF_CALLOC (local->alloc_len, sizeof (char),
+                               gf_common_mt_char);
+        if (!xattr_buf)
+                goto out;
 
-	return 0;
-}
+        if (XATTR_IS_PATHINFO (local->xsel)) {
+                (void) dht_fill_pathinfo_xattr (this, local, xattr_buf,
+                                                local->alloc_len, flag,
+                                                layout_buf);
+        } else if (XATTR_IS_NODE_UUID (local->xsel)) {
+                (void) snprintf (xattr_buf, local->alloc_len, "%s",
+                                 local->xattr_val);
+        } else {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Unknown local->xsel (%s)", local->xsel);
+                goto out;
+        }
 
+        ret = dict_set_dynstr (*dict, local->xsel, xattr_buf);
+        GF_FREE (local->xattr_val);
+
+ out:
+        return ret;
+}
 
 int
-dht_chown (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, uid_t uid, gid_t gid)
+dht_vgetxattr_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
 {
-	dht_layout_t *layout = NULL;
-	dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-	int           i = -1;
+        int          ret           = 0;
+        dht_local_t *local         = NULL;
+        int          this_call_cnt = 0;
+        dict_t      *dict          = NULL;
 
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (frame->local, out);
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        local = frame->local;
 
-	layout = dht_layout_get (this, loc->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no layout for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        LOCK (&frame->lock);
+        {
+                this_call_cnt = --local->call_cnt;
+                if (op_ret < 0) {
+                        if (op_errno != ENOTCONN) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "getxattr err (%s) for dir",
+                                        strerror (op_errno));
+				local->op_ret = -1;
+				local->op_errno = op_errno;
+                        }
 
-	if (!layout_is_sane (layout)) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"layout is not sane for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+                        goto unlock;
+                }
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+                ret = dht_vgetxattr_alloc_and_fill (local, xattr, this,
+                                                    op_errno);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "alloc or fill failure");
+        }
+ unlock:
+        UNLOCK (&frame->lock);
 
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = layout->cnt;
+        if (!is_last_call (this_call_cnt))
+                goto out;
 
-	for (i = 0; i < layout->cnt; i++) {
-		STACK_WIND (frame, dht_attr_cbk,
-			    layout->list[i].xlator,
-			    layout->list[i].xlator->fops->chown,
-			    loc, uid, gid);
-	}
+        /* -- last call: do patch ups -- */
 
-	return 0;
+        if (local->op_ret == -1) {
+                goto unwind;
+        }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        ret = dht_vgetxattr_fill_and_set (local, &dict, this, _gf_true);
+        if (ret)
+                goto unwind;
 
-	return 0;
-}
+        DHT_STACK_UNWIND (getxattr, frame, 0, 0, dict, xdata);
+        goto cleanup;
 
+ unwind:
+        DHT_STACK_UNWIND (getxattr, frame, -1, local->op_errno, NULL, NULL);
+ cleanup:
+        if (dict)
+                dict_unref (dict);
+ out:
+        return 0;
+}
 
 int
-dht_fchmod (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, mode_t mode)
+dht_vgetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
 {
-	dht_layout_t *layout = NULL;
-	dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-	int           i = -1;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
+        dht_local_t  *local         = NULL;
+        int           ret           = 0;
+        dict_t       *dict          = NULL;
+        call_frame_t *prev          = NULL;
+        gf_boolean_t  flag          = _gf_true;
 
-	layout = dht_layout_get (this, fd->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no layout for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        local = frame->local;
+        prev = cookie;
 
-	if (!layout_is_sane (layout)) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"layout is not sane for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        if (op_ret < 0) {
+                local->op_ret = -1;
+                local->op_errno = op_errno;
+                gf_log (this->name, GF_LOG_ERROR, "Subvolume %s returned -1 "
+                        "(%s)", prev->this->name, strerror (op_errno));
+                goto unwind;
+        }
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        ret = dht_vgetxattr_alloc_and_fill (local, xattr, this,
+                                            op_errno);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "alloc or fill failure");
+                goto unwind;
+        }
 
-	local->inode = inode_ref (fd->inode);
-	local->call_cnt = layout->cnt;
+        flag = (local->layout->cnt > 1) ? _gf_true : _gf_false;
 
-	for (i = 0; i < layout->cnt; i++) {
-		STACK_WIND (frame, dht_attr_cbk,
-			    layout->list[i].xlator,
-			    layout->list[i].xlator->fops->fchmod,
-			    fd, mode);
-	}
+        ret = dht_vgetxattr_fill_and_set (local, &dict, this, flag);
+        if (ret)
+                goto unwind;
 
-	return 0;
+        DHT_STACK_UNWIND (getxattr, frame, 0, 0, dict, xdata);
+        goto cleanup;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+ unwind:
+        DHT_STACK_UNWIND (getxattr, frame, -1, local->op_errno,
+                          NULL, NULL);
+ cleanup:
+        if (dict)
+                dict_unref (dict);
 
-	return 0;
+        return 0;
 }
 
-
 int
-dht_fchown (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, uid_t uid, gid_t gid)
+dht_linkinfo_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                           int op_ret, int op_errno, dict_t *xattr,
+                           dict_t *xdata)
 {
-	dht_layout_t *layout = NULL;
-	dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-	int           i = -1;
+        int   ret   = 0;
+        char *value = NULL;
+
+        if (op_ret != -1) {
+                ret = dict_get_str (xattr, GF_XATTR_PATHINFO_KEY, &value);
+                if (!ret) {
+                        ret = dict_set_str (xattr, GF_XATTR_LINKINFO_KEY, value);
+                        if (!ret)
+                                gf_log (this->name, GF_LOG_TRACE,
+                                        "failed to set linkinfo");
+                }
+        }
 
+        DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr, xdata);
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+        return 0;
+}
 
-	layout = dht_layout_get (this, fd->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no layout for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+int
+dht_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+        int             this_call_cnt = 0;
+        dht_local_t     *local = NULL;
+        dht_conf_t      *conf = NULL;
 
-	if (!layout_is_sane (layout)) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"layout is not sane for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        VALIDATE_OR_GOTO (frame, out);
+        VALIDATE_OR_GOTO (frame->local, out);
+        VALIDATE_OR_GOTO (this->private, out);
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        conf = this->private;
+        local = frame->local;
 
-	local->inode = inode_ref (fd->inode);
-	local->call_cnt = layout->cnt;
+        this_call_cnt = dht_frame_return (frame);
 
-	for (i = 0; i < layout->cnt; i++) {
-		STACK_WIND (frame, dht_attr_cbk,
-			    layout->list[i].xlator,
-			    layout->list[i].xlator->fops->fchown,
-			    fd, uid, gid);
-	}
+        if (!xattr || (op_ret == -1))
+                goto out;
 
-	return 0;
+        if (dict_get (xattr, conf->xattr_name)) {
+                dict_del (xattr, conf->xattr_name);
+        }
+        local->op_ret = 0;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        if (!local->xattr) {
+                local->xattr = dict_copy_with_ref (xattr, NULL);
+        } else {
+                /* first aggregate everything into xattr and then copy into
+                 * local->xattr. This is required as we want to have
+                 * 'local->xattr' as the proper final dictionary passed above
+                 * distribute xlator.
+                 */
+                dht_aggregate_xattr (xattr, local->xattr);
+                local->xattr = dict_copy (xattr, local->xattr);
+        }
+out:
+        if (is_last_call (this_call_cnt)) {
+                DHT_STACK_UNWIND (getxattr, frame, local->op_ret, op_errno,
+                                  local->xattr, NULL);
+        }
+        return 0;
+}
 
-	return 0;
+int32_t
+dht_getxattr_unwind (call_frame_t *frame,
+                     int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
+{
+        DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, xdata);
+        return 0;
 }
 
 
 int
-dht_utimens (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, struct timespec tv[2])
+dht_getxattr_get_real_filename_cbk (call_frame_t *frame, void *cookie,
+				    xlator_t *this, int op_ret, int op_errno,
+				    dict_t *xattr, dict_t *xdata)
 {
-	dht_layout_t *layout = NULL;
-	dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-	int           i = -1;
+        int             this_call_cnt = 0;
+        dht_local_t     *local = NULL;
 
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        local = frame->local;
 
-	layout = dht_layout_get (this, loc->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no layout for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+	if (op_ret != -1) {
+		if (local->xattr)
+			dict_unref (local->xattr);
+		local->xattr = dict_ref (xattr);
 
-	if (!layout_is_sane (layout)) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"layout is not sane for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
+		if (local->xattr_req)
+			dict_unref (local->xattr_req);
+		local->xattr_req = dict_ref (xdata);
 	}
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
+	this_call_cnt = dht_frame_return (frame);
+	if (is_last_call (this_call_cnt)) {
+		DHT_STACK_UNWIND (getxattr, frame, local->op_ret, op_errno,
+				  local->xattr, local->xattr_req);
 	}
 
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = layout->cnt;
+	return 0;
+}
 
-	for (i = 0; i < layout->cnt; i++) {
-		STACK_WIND (frame, dht_attr_cbk,
-			    layout->list[i].xlator,
-			    layout->list[i].xlator->fops->utimens,
-			    loc, tv);
-	}
 
-	return 0;
+int
+dht_getxattr_get_real_filename (call_frame_t *frame, xlator_t *this,
+				loc_t *loc, const char *key, dict_t *xdata)
+{
+	dht_local_t     *local = NULL;
+	int              i = 0;
+	dht_layout_t    *layout = NULL;
+	int              cnt = 0;
+	xlator_t        *subvol = NULL;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+
+	local = frame->local;
+	layout = local->layout;
+
+	cnt = local->call_cnt = layout->cnt;
+
+	local->op_ret = -1;
+	local->op_errno = ENODATA;
+
+	for (i = 0; i < cnt; i++) {
+		subvol = layout->list[i].xlator;
+		STACK_WIND (frame, dht_getxattr_get_real_filename_cbk,
+			    subvol, subvol->fops->getxattr,
+			    loc, key, xdata);
+	}
 
 	return 0;
 }
 
 
 int
-dht_truncate (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, off_t offset)
+dht_getxattr (call_frame_t *frame, xlator_t *this,
+              loc_t *loc, const char *key, dict_t *xdata)
+#define DHT_IS_DIR(layout)  (layout->cnt > 1)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
 
+        xlator_t     *subvol        = NULL;
+        xlator_t     *hashed_subvol = NULL;
+        xlator_t     *cached_subvol = NULL;
+        dht_conf_t   *conf          = NULL;
+        dht_local_t  *local         = NULL;
+        dht_layout_t *layout        = NULL;
+        xlator_t     **sub_volumes  = NULL;
+        int           op_errno      = -1;
+        int           i             = 0;
+        int           cnt           = 0;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        conf   = this->private;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        local = dht_local_init (frame, loc, NULL, GF_FOP_GETXATTR);
+        if (!local) {
+                op_errno = ENOMEM;
 
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = 1;
-
-	STACK_WIND (frame, dht_attr_cbk,
-		    subvol, subvol->fops->truncate,
-		    loc, offset);
-
-	return 0;
+                goto err;
+        }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "layout is NULL");
+                op_errno = ENOENT;
+                goto err;
+        }
 
-	return 0;
-}
+        if (key) {
+                local->key = gf_strdup (key);
+                if (!local->key) {
+                        op_errno = ENOMEM;
+                        goto err;
+                }
+        }
 
+	if (key &&
+	    (strncmp (key, GF_XATTR_GET_REAL_FILENAME_KEY,
+		      strlen (GF_XATTR_GET_REAL_FILENAME_KEY)) == 0)
+	    && DHT_IS_DIR(layout)) {
+		dht_getxattr_get_real_filename (frame, this, loc, key, xdata);
+		return 0;
+	}
 
-int
-dht_ftruncate (call_frame_t *frame, xlator_t *this,
-	       fd_t *fd, off_t offset)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+        /* for file use cached subvolume (obviously!): see if {}
+         * below
+         * for directory:
+         *  wind to all subvolumes and exclude subvolumes which
+         *  return ENOTCONN (in callback)
+         *
+         * NOTE: Don't trust inode here, as that may not be valid
+         *       (until inode_link() happens)
+         */
+        if (key && DHT_IS_DIR(layout) &&
+            ((strcmp (key, GF_XATTR_PATHINFO_KEY) == 0)
+             || (strcmp (key, GF_XATTR_NODE_UUID_KEY) == 0))) {
+                (void) strncpy (local->xsel, key, 256);
+                cnt = local->call_cnt = layout->cnt;
+                for (i = 0; i < cnt; i++) {
+                        subvol = layout->list[i].xlator;
+                        STACK_WIND (frame, dht_vgetxattr_dir_cbk,
+                                    subvol, subvol->fops->getxattr,
+                                    loc, key, NULL);
+                }
+                return 0;
+        }
 
+        /* node-uuid or pathinfo for files */
+        if (key && ((strcmp (key, GF_XATTR_NODE_UUID_KEY) == 0)
+                    || (strcmp (key, GF_XATTR_PATHINFO_KEY) == 0))) {
+                cached_subvol = local->cached_subvol;
+                (void) strncpy (local->xsel, key, 256);
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+                local->call_cnt = 1;
+                STACK_WIND (frame, dht_vgetxattr_cbk, cached_subvol,
+                            cached_subvol->fops->getxattr, loc, key, NULL);
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+                return 0;
+        }
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        if (key && (strcmp (key, GF_XATTR_LINKINFO_KEY) == 0)) {
+                hashed_subvol = dht_subvol_get_hashed (this, loc);
+                if (!hashed_subvol) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to get"
+                                "hashed subvol for %s", loc->path);
+                        op_errno = EINVAL;
+                        goto err;
+                }
 
-	local->inode = inode_ref (fd->inode);
-	local->call_cnt = 1;
+                cached_subvol = dht_subvol_get_cached (this, loc->inode);
+                if (!cached_subvol) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to get"
+                                "cached subvol for %s", loc->path);
+                        op_errno = EINVAL;
+                        goto err;
+                }
 
-	STACK_WIND (frame, dht_attr_cbk,
-		    subvol, subvol->fops->ftruncate,
-		    fd, offset);
+                if (hashed_subvol == cached_subvol) {
+                        op_errno = ENODATA;
+                        goto err;
+                }
+                if (hashed_subvol) {
+                        STACK_WIND (frame, dht_linkinfo_getxattr_cbk, hashed_subvol,
+                                    hashed_subvol->fops->getxattr, loc,
+                                    GF_XATTR_PATHINFO_KEY, NULL);
+                        return 0;
+                }
+                op_errno = ENODATA;
+                goto err;
+        }
 
-	return 0;
+        if (key && (!strcmp (GF_XATTR_MARKER_KEY, key))
+            && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
+                if (DHT_IS_DIR(layout)) {
+                        cnt = layout->cnt;
+                } else {
+                        cnt = 1;
+                }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+                sub_volumes = alloca ( cnt * sizeof (xlator_t *));
+                for (i = 0; i < cnt; i++)
+                        *(sub_volumes + i) = layout->list[i].xlator;
 
-	return 0;
-}
+                if (cluster_getmarkerattr (frame, this, loc, key,
+                                           local, dht_getxattr_unwind,
+                                           sub_volumes, cnt,
+                                           MARKER_UUID_TYPE, marker_uuid_default_gauge,
+                                           conf->vol_uuid)) {
+                        op_errno = EINVAL;
+                        goto err;
+                }
 
+                return 0;
+        }
 
-int
-dht_err_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	     int op_ret, int op_errno)
-{
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
+        if (key && *conf->vol_uuid) {
+                if ((match_uuid_local (key, conf->vol_uuid) == 0) &&
+                    (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
+                        if (DHT_IS_DIR(layout)) {
+                                cnt = layout->cnt;
+                        } else {
+                                cnt = 1;
+                        }
+                        sub_volumes = alloca ( cnt * sizeof (xlator_t *));
+                        for (i = 0; i < cnt; i++)
+                                sub_volumes[i] = layout->list[i].xlator;
+
+                        if (cluster_getmarkerattr (frame, this, loc, key,
+                                                   local, dht_getxattr_unwind,
+                                                   sub_volumes, cnt,
+                                                   MARKER_XTIME_TYPE,
+                                                   marker_xtime_default_gauge,
+                                                   conf->vol_uuid)) {
+                                op_errno = EINVAL;
+                                goto err;
+                        }
 
+                        return 0;
+                }
+        }
 
-	local = frame->local;
-	prev = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_ERROR,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
+        if (DHT_IS_DIR(layout)) {
+                cnt = local->call_cnt = layout->cnt;
+        } else {
+                cnt = local->call_cnt  = 1;
+        }
 
-		local->op_ret = 0;
-	}
-unlock:
-	UNLOCK (&frame->lock);
+        for (i = 0; i < cnt; i++) {
+                subvol = layout->list[i].xlator;
+                STACK_WIND (frame, dht_getxattr_cbk,
+                            subvol, subvol->fops->getxattr,
+                            loc, key, NULL);
+        }
+        return 0;
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (frame, local->op_ret, local->op_errno);
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL, NULL);
 
         return 0;
 }
-
+#undef DHT_IS_DIR
 
 int
-dht_access (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t mask)
+dht_fgetxattr (call_frame_t *frame, xlator_t *this,
+               fd_t *fd, const char *key, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
-
+        xlator_t     *subvol        = NULL;
+        dht_local_t  *local         = NULL;
+        dht_layout_t *layout        = NULL;
+        int           op_errno      = -1;
+        int           i             = 0;
+        int           cnt           = 0;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (fd->inode, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FGETXATTR);
+        if (!local) {
+                op_errno = ENOMEM;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+                goto err;
+        }
 
-	local->call_cnt = 1;
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "layout is NULL");
+                op_errno = ENOENT;
+                goto err;
+        }
 
-	STACK_WIND (frame, dht_err_cbk,
-		    subvol, subvol->fops->access,
-		    loc, mask);
+        if (key) {
+                local->key = gf_strdup (key);
+                if (!local->key) {
+                        op_errno = ENOMEM;
+                        goto err;
+                }
+        }
 
-	return 0;
+        if ((fd->inode->ia_type == IA_IFDIR)
+            && (strncmp (key, GF_XATTR_LOCKINFO_KEY,
+                         strlen (GF_XATTR_LOCKINFO_KEY) != 0))) {
+                cnt = local->call_cnt = layout->cnt;
+        } else {
+                cnt = local->call_cnt  = 1;
+        }
+
+        for (i = 0; i < cnt; i++) {
+                subvol = layout->list[i].xlator;
+                STACK_WIND (frame, dht_getxattr_cbk,
+                            subvol, subvol->fops->fgetxattr,
+                            fd, key, NULL);
+        }
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fgetxattr, frame, -1, op_errno, NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
-
 int
-dht_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int op_ret, int op_errno, const char *path)
+dht_fsetxattr (call_frame_t *frame, xlator_t *this,
+               fd_t *fd, dict_t *xattr, int flags, dict_t *xdata)
 {
-        DHT_STACK_UNWIND (frame, op_ret, op_errno, path);
+        xlator_t     *subvol   = NULL;
+        dht_local_t  *local    = NULL;
+        int           op_errno = EINVAL;
+        dht_conf_t   *conf     = NULL;
 
-        return 0;
-}
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (fd->inode, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
+        conf = this->private;
 
-int
-dht_readlink (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, size_t size)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
+        GF_IF_INTERNAL_XATTR_GOTO (conf->wild_xattr_name, xattr,
+                                   op_errno, err);
 
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FSETXATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        local->call_cnt = 1;
 
-	STACK_WIND (frame, dht_readlink_cbk,
-		    subvol, subvol->fops->readlink,
-		    loc, size);
+        STACK_WIND (frame, dht_err_cbk, subvol, subvol->fops->fsetxattr,
+                    fd, xattr, flags, NULL);
 
-	return 0;
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fsetxattr, frame, -1, op_errno, NULL);
 
-	return 0;
+        return 0;
 }
 
 
-int
-dht_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int op_ret, int op_errno, dict_t *xattr)
+static int
+dht_common_setxattr_cbk (call_frame_t *frame, void *cookie,
+                         xlator_t *this, int32_t op_ret, int32_t op_errno,
+                         dict_t *xdata)
 {
-        DHT_STACK_UNWIND (frame, op_ret, op_errno, xattr);
+        DHT_STACK_UNWIND (setxattr, frame, op_ret, op_errno, xdata);
 
         return 0;
 }
 
-
 int
-dht_getxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, const char *key)
+dht_checking_pathinfo_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                           int op_ret, int op_errno, dict_t *xattr,
+                           dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
+        int           i     = -1;
+        int           ret   = -1;
+        char         *value = NULL;
+        dht_local_t  *local = NULL;
+        dht_conf_t   *conf  = NULL;
+        call_frame_t *prev  = NULL;
+        int           this_call_cnt = 0;
 
+        local = frame->local;
+        prev = cookie;
+        conf = this->private;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        if (op_ret == -1)
+                goto out;
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
 
-	STACK_WIND (frame, dht_getxattr_cbk,
-		    subvol, subvol->fops->getxattr,
-		    loc, key);
+        ret = dict_get_str (xattr, GF_XATTR_PATHINFO_KEY, &value);
+        if (ret)
+                goto out;
 
-	return 0;
+        if (!strcmp (value, local->key)) {
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                        if (conf->subvolumes[i] == prev->this)
+                                conf->decommissioned_bricks[i] = prev->this;
+                }
+        }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+out:
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+                DHT_STACK_UNWIND (setxattr, frame, local->op_ret, ENOTSUP, NULL);
+        }
+        return 0;
 
-	return 0;
 }
 
-
 int
 dht_setxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, dict_t *xattr, int flags)
+              loc_t *loc, dict_t *xattr, int flags, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
-
+        xlator_t     *subvol   = NULL;
+        dht_local_t  *local    = NULL;
+        dht_conf_t   *conf     = NULL;
+        dht_layout_t *layout   = NULL;
+        int           i        = 0;
+        int           op_errno = EINVAL;
+        int           ret      = -1;
+        data_t       *tmp      = NULL;
+        uint32_t      dir_spread = 0;
+        char          value[4096] = {0,};
+        gf_dht_migrate_data_type_t forced_rebalance = GF_DHT_MIGRATE_DATA;
+        int           call_cnt = 0;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        conf   = this->private;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        GF_IF_INTERNAL_XATTR_GOTO (conf->wild_xattr_name, xattr,
+                                   op_errno, err);
 
-	local->call_cnt = 1;
+        local = dht_local_init (frame, loc, NULL, GF_FOP_SETXATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	STACK_WIND (frame, dht_err_cbk,
-		    subvol, subvol->fops->setxattr,
-		    loc, xattr, flags);
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-	return 0;
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        local->call_cnt = call_cnt = layout->cnt;
 
-	return 0;
-}
+        tmp = dict_get (xattr, "distribute.migrate-data");
+        if (tmp) {
+                if (IA_ISDIR (loc->inode->ia_type)) {
+                        op_errno = ENOTSUP;
+                        goto err;
+                }
 
+                /* TODO: need to interpret the 'value' for more meaning
+                   (ie, 'target' subvolume given there, etc) */
+                memcpy (value, tmp->data, tmp->len);
+                if (strcmp (value, "force") == 0)
+                       forced_rebalance =
+                                GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS;
 
-int
-dht_removexattr (call_frame_t *frame, xlator_t *this,
-		 loc_t *loc, const char *key)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+                if (conf->decommission_in_progress)
+                        forced_rebalance = GF_DHT_MIGRATE_HARDLINK;
 
+                local->rebalance.target_node = dht_subvol_get_hashed (this, loc);
+                if (!local->rebalance.target_node) {
+                        gf_log (this->name, GF_LOG_ERROR, "Failed to get "
+                                "hashed subvol for %s", loc->path);
+                        op_errno = EINVAL;
+                        goto err;
+                }
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+                local->rebalance.from_subvol = local->cached_subvol;
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+                if (local->rebalance.target_node == local->rebalance.from_subvol) {
+                        op_errno = EEXIST;
+                        goto err;
+                }
+                if (local->rebalance.target_node) {
+                        local->flags = forced_rebalance;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+                        ret = dht_start_rebalance_task (this, frame);
+                        if (!ret)
+                                return 0;
 
-	local->call_cnt = 1;
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: failed to create a new synctask",
+                                loc->path);
+                }
+                op_errno = EINVAL;
+                goto err;
 
-	STACK_WIND (frame, dht_err_cbk,
-		    subvol, subvol->fops->removexattr,
-		    loc, key);
+        }
 
-	return 0;
+        tmp = dict_get (xattr, "decommission-brick");
+        if (tmp) {
+                /* This operation should happen only on '/' */
+                if (!__is_root_gfid (loc->inode->gfid)) {
+                        op_errno = ENOTSUP;
+                        goto err;
+                }
+
+                memcpy (value, tmp->data, ((tmp->len < 4095) ? tmp->len : 4095));
+                local->key = gf_strdup (value);
+                local->call_cnt = conf->subvolume_cnt;
+
+                for (i = 0 ; i < conf->subvolume_cnt; i++) {
+                        /* Get the pathinfo, and then compare */
+                        STACK_WIND (frame, dht_checking_pathinfo_cbk,
+                                    conf->subvolumes[i],
+                                    conf->subvolumes[i]->fops->getxattr,
+                                    loc, GF_XATTR_PATHINFO_KEY, NULL);
+                }
+                return 0;
+        }
+
+        tmp = dict_get (xattr, GF_XATTR_FIX_LAYOUT_KEY);
+        if (tmp) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "fixing the layout of %s", loc->path);
+
+                ret = dht_fix_directory_layout (frame, dht_common_setxattr_cbk,
+                                                layout);
+                if (ret) {
+                        op_errno = ENOTCONN;
+                        goto err;
+                }
+                return ret;
+        }
+
+        tmp = dict_get (xattr, "distribute.directory-spread-count");
+        if (tmp) {
+                /* Setxattr value is packed as 'binary', not string */
+                memcpy (value, tmp->data, ((tmp->len < 4095)?tmp->len:4095));
+                ret = gf_string2uint32 (value, &dir_spread);
+                if (!ret && ((dir_spread <= conf->subvolume_cnt) &&
+                             (dir_spread > 0))) {
+                        layout->spread_cnt = dir_spread;
+
+                        ret = dht_fix_directory_layout (frame,
+                                                        dht_common_setxattr_cbk,
+                                                        layout);
+                        if (ret) {
+                                op_errno = ENOTCONN;
+                                goto err;
+                        }
+                        return ret;
+                }
+                gf_log (this->name, GF_LOG_ERROR,
+                        "wrong 'directory-spread-count' value (%s)", value);
+                op_errno = ENOTSUP;
+                goto err;
+        }
+
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (frame, dht_err_cbk,
+                            layout->list[i].xlator,
+                            layout->list[i].xlator->fops->setxattr,
+                            loc, xattr, flags, xdata);
+        }
+
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (setxattr, frame, -1, op_errno, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
-dht_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	      int op_ret, int op_errno, fd_t *fd)
+dht_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
 
+        local = frame->local;
+        prev = cookie;
 
-	local = frame->local;
-	prev = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_ERROR,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "subvolume %s returned -1 (%s)",
+                                prev->this->name, strerror (op_errno));
+                        goto unlock;
+                }
 
-		local->op_ret = 0;
-	}
+                local->op_ret = 0;
+        }
 unlock:
-	UNLOCK (&frame->lock);
+        UNLOCK (&frame->lock);
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-				  local->fd);
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+                DHT_STACK_UNWIND (removexattr, frame, local->op_ret,
+                                  local->op_errno, NULL);
+        }
 
         return 0;
 }
 
 
 int
-dht_open (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc, int flags, fd_t *fd)
+dht_removexattr (call_frame_t *frame, xlator_t *this,
+                 loc_t *loc, const char *key, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-	int           ret = -1;
+        xlator_t     *subvol = NULL;
         int           op_errno = -1;
-	dht_local_t  *local = NULL;
+        dht_local_t  *local = NULL;
+        dht_layout_t *layout = NULL;
+        int           call_cnt = 0;
+        dht_conf_t   *conf = NULL;
 
+        int i;
 
-        VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        conf = this->private;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        GF_IF_NATIVE_XATTR_GOTO (conf->wild_xattr_name, key, op_errno, err);
 
-	local->fd = fd_ref (fd);
-	ret = loc_dup (loc, &local->loc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
 
-	local->call_cnt = 1;
+        local = dht_local_init (frame, loc, NULL, GF_FOP_REMOVEXATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	STACK_WIND (frame, dht_fd_cbk,
-		    subvol, subvol->fops->open,
-		    loc, flags, fd);
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-	return 0;
+        layout = local->layout;
+        if (!local->layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        local->call_cnt = call_cnt = layout->cnt;
+        local->key = gf_strdup (key);
 
-	return 0;
-}
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (frame, dht_removexattr_cbk,
+                            layout->list[i].xlator,
+                            layout->list[i].xlator->fops->removexattr,
+                            loc, key, NULL);
+        }
 
+        return 0;
 
-int
-dht_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	       int op_ret, int op_errno,
-	       struct iovec *vector, int count, struct stat *stbuf)
-{
-        DHT_STACK_UNWIND (frame, op_ret, op_errno, vector, count, stbuf);
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (removexattr, frame, -1, op_errno, NULL);
 
         return 0;
 }
 
-
 int
-dht_readv (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd, size_t size, off_t off)
+dht_fremovexattr (call_frame_t *frame, xlator_t *this,
+                  fd_t *fd, const char *key, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
+        xlator_t     *subvol = NULL;
         int           op_errno = -1;
+        dht_local_t  *local = NULL;
+        dht_layout_t *layout = NULL;
+        int           call_cnt = 0;
+        dht_conf_t   *conf = 0;
 
+        int i;
 
-        VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, dht_readv_cbk,
-		    subvol, subvol->fops->readv,
-		    fd, size, off);
-
-	return 0;
-
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL, 0, NULL);
-
-	return 0;
-}
+        VALIDATE_OR_GOTO (this->private, err);
 
+        conf = this->private;
 
-int
-dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		int op_ret, int op_errno, struct stat *stbuf)
-{
-        DHT_STACK_UNWIND (frame, op_ret, op_errno, stbuf);
-
-        return 0;
-}
+        GF_IF_NATIVE_XATTR_GOTO (conf->wild_xattr_name, key, op_errno, err);
 
+        VALIDATE_OR_GOTO (frame, err);
 
-int
-dht_writev (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, struct iovec *vector, int count, off_t off)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FREMOVEXATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for inode=%s",
+                        uuid_utoa (fd->inode->gfid));
+                op_errno = EINVAL;
+                goto err;
+        }
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+        layout = local->layout;
+        if (!local->layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for inode=%s", uuid_utoa (fd->inode->gfid));
+                op_errno = EINVAL;
+                goto err;
+        }
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        local->call_cnt = call_cnt = layout->cnt;
+        local->key = gf_strdup (key);
 
-	STACK_WIND (frame, dht_writev_cbk,
-		    subvol, subvol->fops->writev,
-		    fd, vector, count, off);
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (frame, dht_removexattr_cbk,
+                            layout->list[i].xlator,
+                            layout->list[i].xlator->fops->fremovexattr,
+                            fd, key, NULL);
+        }
 
-	return 0;
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL, 0);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fremovexattr, frame, -1, op_errno, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
-dht_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
+dht_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+            int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
 
+        local = frame->local;
+        prev = cookie;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "subvolume %s returned -1 (%s)",
+                                prev->this->name, strerror (op_errno));
+                        goto unlock;
+                }
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+                local->op_ret = 0;
+        }
+unlock:
+        UNLOCK (&frame->lock);
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt))
+                DHT_STACK_UNWIND (open, frame, local->op_ret, local->op_errno,
+                                  local->fd, NULL);
 
-	local->fd = fd_ref (fd);
-	local->call_cnt = 1;
+        return 0;
+}
 
-	STACK_WIND (frame, dht_err_cbk,
-		    subvol, subvol->fops->flush, fd);
+/*
+ * dht_normalize_stats -
+ */
+static void
+dht_normalize_stats (struct statvfs *buf, unsigned long bsize,
+                     unsigned long frsize)
+{
+        double factor = 0;
 
-	return 0;
+        if (buf->f_bsize != bsize) {
+                buf->f_bsize = bsize;
+        }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno);
+        if (buf->f_frsize != frsize) {
+                factor = ((double) buf->f_frsize) / frsize;
+                buf->f_frsize = frsize;
+                buf->f_blocks = (fsblkcnt_t) (factor * buf->f_blocks);
+                buf->f_bfree  = (fsblkcnt_t) (factor * buf->f_bfree);
+                buf->f_bavail = (fsblkcnt_t) (factor * buf->f_bavail);
 
-	return 0;
+        }
 }
 
-
 int
-dht_fsync (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd, int datasync)
+dht_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                int op_ret, int op_errno, struct statvfs *statvfs, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+        dht_local_t *local         = NULL;
+        int          this_call_cnt = 0;
+        int          bsize         = 0;
+        int          frsize        = 0;
 
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocatoin failed :(");
-		goto err;
-	}
-	local->call_cnt = 1;
+        local = frame->local;
 
-	STACK_WIND (frame, dht_err_cbk,
-		    subvol, subvol->fops->fsync,
-		    fd, datasync);
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        goto unlock;
+                }
+                local->op_ret = 0;
 
-	return 0;
+                if (local->statvfs.f_bsize != 0) {
+                        bsize = max(local->statvfs.f_bsize, statvfs->f_bsize);
+                        frsize = max(local->statvfs.f_frsize, statvfs->f_frsize);
+                        dht_normalize_stats(&local->statvfs, bsize, frsize);
+                        dht_normalize_stats(statvfs, bsize, frsize);
+                } else {
+                        local->statvfs.f_bsize    = statvfs->f_bsize;
+                        local->statvfs.f_frsize   = statvfs->f_frsize;
+                }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno);
+                local->statvfs.f_blocks  += statvfs->f_blocks;
+                local->statvfs.f_bfree   += statvfs->f_bfree;
+                local->statvfs.f_bavail  += statvfs->f_bavail;
+                local->statvfs.f_files   += statvfs->f_files;
+                local->statvfs.f_ffree   += statvfs->f_ffree;
+                local->statvfs.f_favail  += statvfs->f_favail;
+                local->statvfs.f_fsid     = statvfs->f_fsid;
+                local->statvfs.f_flag     = statvfs->f_flag;
+                local->statvfs.f_namemax  = statvfs->f_namemax;
 
-	return 0;
-}
+        }
+unlock:
+        UNLOCK (&frame->lock);
 
 
-int
-dht_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	    int op_ret, int op_errno, struct flock *flock)
-{
-        DHT_STACK_UNWIND (frame, op_ret, op_errno, flock);
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt))
+                DHT_STACK_UNWIND (statfs, frame, local->op_ret, local->op_errno,
+                                  &local->statvfs, xdata);
 
         return 0;
 }
 
 
 int
-dht_lk (call_frame_t *frame, xlator_t *this,
-	fd_t *fd, int cmd, struct flock *flock)
+dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
+        xlator_t     *subvol = NULL;
+        dht_local_t  *local  = NULL;
+        dht_conf_t   *conf = NULL;
         int           op_errno = -1;
-
+        int           i = -1;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        conf = this->private;
 
-	STACK_WIND (frame, dht_lk_cbk,
-		    subvol, subvol->fops->lk,
-		    fd, cmd, flock);
+        local = dht_local_init (frame, NULL, NULL, GF_FOP_STATFS);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	return 0;
+        if (IA_ISDIR (loc->inode->ia_type)) {
+                local->call_cnt = conf->subvolume_cnt;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                        STACK_WIND (frame, dht_statfs_cbk,
+                                    conf->subvolumes[i],
+                                    conf->subvolumes[i]->fops->statfs, loc,
+                                    xdata);
+                }
+                return 0;
+        }
 
-	return 0;
-}
+        subvol = dht_subvol_get_cached (this, loc->inode);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-/* gf_lk no longer exists 
-int
-dht_gf_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	    int op_ret, int op_errno, struct flock *flock)
-{
-        DHT_STACK_UNWIND (frame, op_ret, op_errno, flock);
+        local->call_cnt = 1;
+
+        STACK_WIND (frame, dht_statfs_cbk,
+                    subvol, subvol->fops->statfs, loc, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (statfs, frame, -1, op_errno, NULL, NULL);
 
         return 0;
 }
 
 
 int
-dht_gf_lk (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, int cmd, struct flock *flock)
+dht_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+             dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
+        dht_local_t  *local  = NULL;
+        dht_conf_t   *conf = NULL;
         int           op_errno = -1;
-
+        int           i = -1;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, dht_gf_lk_cbk,
-		    subvol, subvol->fops->gf_lk,
-		    fd, cmd, flock);
-
-	return 0;
-
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
-
-	return 0;
-}
-*/
-
-int
-dht_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		int op_ret, int op_errno, struct statvfs *statvfs)
-{
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
+        conf = this->private;
 
+        local = dht_local_init (frame, loc, fd, GF_FOP_OPENDIR);
+        if (!local) {
+                op_errno = ENOMEM;
 
-	local = frame->local;
+                goto err;
+        }
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			goto unlock;
-		}
-		local->op_ret = 0;
-
-		/* TODO: normalize sizes */
-		local->statvfs.f_bsize    = statvfs->f_bsize;
-		local->statvfs.f_frsize   = statvfs->f_frsize;
-
-		local->statvfs.f_blocks  += statvfs->f_blocks;
-		local->statvfs.f_bfree   += statvfs->f_bfree;
-		local->statvfs.f_bavail  += statvfs->f_bavail;
-		local->statvfs.f_files   += statvfs->f_files;
-		local->statvfs.f_ffree   += statvfs->f_ffree;
-		local->statvfs.f_favail  += statvfs->f_favail;
-		local->statvfs.f_fsid     = statvfs->f_fsid;
-		local->statvfs.f_flag     = statvfs->f_flag;
-		local->statvfs.f_namemax  = statvfs->f_namemax;
+        local->call_cnt = conf->subvolume_cnt;
 
-	}
-unlock:
-	UNLOCK (&frame->lock);
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                STACK_WIND (frame, dht_fd_cbk,
+                            conf->subvolumes[i],
+                            conf->subvolumes[i]->fops->opendir,
+                            loc, fd, xdata);
+        }
 
+        return 0;
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (frame, local->op_ret, local->op_errno,
-				  &local->statvfs);
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (opendir, frame, -1, op_errno, NULL, NULL);
 
         return 0;
 }
 
 
 int
-dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
+dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, gf_dirent_t *orig_entries, dict_t *xdata)
 {
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-	int           i = -1;
+        dht_local_t  *local = NULL;
+        gf_dirent_t   entries;
+        gf_dirent_t  *orig_entry = NULL;
+        gf_dirent_t  *entry = NULL;
+        call_frame_t *prev = NULL;
+        xlator_t     *next_subvol = NULL;
+        off_t         next_offset = 0;
+        int           count = 0;
+        dht_layout_t *layout = 0;
+        dht_conf_t   *conf   = NULL;
+        xlator_t     *subvol = 0;
+        int           ret    = 0;
 
+        INIT_LIST_HEAD (&entries.list);
+        prev = cookie;
+        local = frame->local;
+        conf  = this->private;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        if (op_ret < 0)
+                goto done;
 
-	conf = this->private;
+        if (!local->layout)
+                local->layout = dht_layout_get (this, local->fd->inode);
 
-	local = dht_local_init (frame);
-	local->call_cnt = conf->subvolume_cnt;
+        layout = local->layout;
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_statfs_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->statfs, loc);
-	}
+        list_for_each_entry (orig_entry, (&orig_entries->list), list) {
+                next_offset = orig_entry->d_off;
+                if (check_is_dir (NULL, (&orig_entry->d_stat), NULL) &&
+                    (prev->this != local->first_up_subvol)) {
+                        continue;
+                }
+                if (check_is_linkfile (NULL, (&orig_entry->d_stat),
+                                       orig_entry->dict,
+                                       conf->link_xattr_name)) {
+                        continue;
+                }
 
-	return 0;
+                entry = gf_dirent_for_name (orig_entry->d_name);
+                if (!entry) {
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+                        goto unwind;
+                }
 
-	return 0;
-}
+                /* Do this if conf->search_unhashed is set to "auto" */
+                if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) {
+                        subvol = dht_layout_search (this, layout,
+                                                    orig_entry->d_name);
+                        if (!subvol || (subvol != prev->this)) {
+                                /* TODO: Count the number of entries which need
+                                   linkfile to prove its existence in fs */
+                                layout->search_unhashed++;
+                        }
+                }
 
+                dht_itransform (this, prev->this, orig_entry->d_off,
+                                &entry->d_off);
 
-int
-dht_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
-{
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-	int           ret = -1;
-        int           op_errno = -1;
-	int           i = -1;
+                entry->d_stat = orig_entry->d_stat;
+                entry->d_ino  = orig_entry->d_ino;
+                entry->d_type = orig_entry->d_type;
+                entry->d_len  = orig_entry->d_len;
 
+                if (orig_entry->dict)
+                        entry->dict = dict_ref (orig_entry->dict);
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+                /* making sure we set the inode ctx right with layout,
+                   currently possible only for non-directories, so for
+                   directories don't set entry inodes */
+                if (!IA_ISDIR(entry->d_stat.ia_type)) {
+                        ret = dht_layout_preset (this, prev->this,
+                                                 orig_entry->inode);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "failed to link the layout in inode");
+                        entry->inode = inode_ref (orig_entry->inode);
+                } else if (orig_entry->inode) {
+                        dht_inode_ctx_time_update (orig_entry->inode, this,
+                                                   &entry->d_stat, 1);
+                }
 
-	conf = this->private;
+                list_add_tail (&entry->list, &entries.list);
+                count++;
+        }
+        op_ret = count;
+        /* We need to ensure that only the last subvolume's end-of-directory
+         * notification is respected so that directory reading does not stop
+         * before all subvolumes have been read. That could happen because the
+         * posix for each subvolume sends a ENOENT on end-of-directory but in
+         * distribute we're not concerned only with a posix's view of the
+         * directory but the aggregated namespace' view of the directory.
+         */
+        if (prev->this != dht_last_up_subvol (this))
+                op_errno = 0;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+done:
+        if (count == 0) {
+                /* non-zero next_offset means that
+                   EOF is not yet hit on the current subvol
+                */
+                if (next_offset == 0) {
+                        next_subvol = dht_subvol_next (this, prev->this);
+                } else {
+                        next_subvol = prev->this;
+                }
 
-	local->fd = fd_ref (fd);
-	ret = loc_dup (loc, &local->loc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+                if (!next_subvol) {
+                        goto unwind;
+                }
 
-	local->call_cnt = conf->subvolume_cnt;
+		if (conf->readdir_optimize == _gf_true) {
+                        if (next_subvol != local->first_up_subvol) {
+                                ret = dict_set_int32 (local->xattr,
+                                                      GF_READDIR_SKIP_DIRS, 1);
+                                if (ret)
+                                        gf_log (this->name, GF_LOG_ERROR,
+					         "dict set failed");
+		        } else {
+                                 dict_del (local->xattr,
+                                           GF_READDIR_SKIP_DIRS);
+                        }
+                }
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_fd_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->opendir,
-			    loc, fd);
-	}
+                STACK_WIND (frame, dht_readdirp_cbk,
+                            next_subvol, next_subvol->fops->readdirp,
+                            local->fd, local->size, next_offset,
+                            local->xattr);
+                return 0;
+        }
 
-	return 0;
+unwind:
+        if (op_ret < 0)
+                op_ret = 0;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        DHT_STACK_UNWIND (readdirp, frame, op_ret, op_errno, &entries, NULL);
 
-	return 0;
+        gf_dirent_free (&entries);
+
+        return 0;
 }
 
 
+
 int
 dht_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int op_ret, int op_errno, gf_dirent_t *orig_entries)
-{
-	dht_local_t  *local = NULL;
-	gf_dirent_t   entries;
-	gf_dirent_t  *orig_entry = NULL;
-	gf_dirent_t  *entry = NULL;
-	call_frame_t *prev = NULL;
-	xlator_t     *subvol = NULL;
-	xlator_t     *next_subvol = NULL;
+                 int op_ret, int op_errno, gf_dirent_t *orig_entries,
+                 dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        gf_dirent_t   entries;
+        gf_dirent_t  *orig_entry = NULL;
+        gf_dirent_t  *entry = NULL;
+        call_frame_t *prev = NULL;
+        xlator_t     *next_subvol = NULL;
         off_t         next_offset = 0;
-	dht_layout_t *layout = NULL;
-	int           count = 0;
+        int           count = 0;
+        dht_layout_t *layout = 0;
+        xlator_t     *subvol = 0;
 
+        INIT_LIST_HEAD (&entries.list);
+        prev = cookie;
+        local = frame->local;
 
-	INIT_LIST_HEAD (&entries.list);
-	prev = cookie;
-	local = frame->local;
+        if (op_ret < 0)
+                goto done;
 
-	if (op_ret < 0)
-		goto done;
+        if (!local->layout)
+                local->layout = dht_layout_get (this, local->fd->inode);
 
-	layout = dht_layout_get (this, local->fd->inode);
+        layout = local->layout;
 
-	list_for_each_entry (orig_entry, &orig_entries->list, list) {
-		subvol = dht_layout_search (this, layout, orig_entry->d_name);
+        list_for_each_entry (orig_entry, (&orig_entries->list), list) {
+                next_offset = orig_entry->d_off;
 
-		if (!subvol || subvol == prev->this) {
-			entry = gf_dirent_for_name (orig_entry->d_name);
-			if (!entry) {
-				gf_log (this->name, GF_LOG_ERROR,
-					"memory allocation failed :(");
-				goto unwind;
-			}
+                subvol = dht_layout_search (this, layout, orig_entry->d_name);
+
+                if (!subvol || (subvol == prev->this)) {
+                        entry = gf_dirent_for_name (orig_entry->d_name);
+                        if (!entry) {
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "memory allocation failed :(");
+                                goto unwind;
+                        }
 
-			dht_itransform (this, prev->this, orig_entry->d_ino,
-					&entry->d_ino);
-			dht_itransform (this, prev->this, orig_entry->d_off,
-					&entry->d_off);
+                        dht_itransform (this, prev->this, orig_entry->d_off,
+                                        &entry->d_off);
 
-			entry->d_type = orig_entry->d_type;
-			entry->d_len  = orig_entry->d_len;
+                        entry->d_ino  = orig_entry->d_ino;
+                        entry->d_type = orig_entry->d_type;
+                        entry->d_len  = orig_entry->d_len;
 
-			list_add_tail (&entry->list, &entries.list);
-			count++;
-		}
-                next_offset = orig_entry->d_off;
-	}
-	op_ret = count;
+                        list_add_tail (&entry->list, &entries.list);
+                        count++;
+                }
+        }
+        op_ret = count;
+        /* We need to ensure that only the last subvolume's end-of-directory
+         * notification is respected so that directory reading does not stop
+         * before all subvolumes have been read. That could happen because the
+         * posix for each subvolume sends a ENOENT on end-of-directory but in
+         * distribute we're not concerned only with a posix's view of the
+         * directory but the aggregated namespace' view of the directory.
+         */
+        if (prev->this != dht_last_up_subvol (this))
+                op_errno = 0;
 
 done:
-	if (count == 0) {
+        if (count == 0) {
                 /* non-zero next_offset means that
                    EOF is not yet hit on the current subvol
                 */
@@ -2163,663 +3232,967 @@ done:
                         next_subvol = prev->this;
                 }
 
-		if (!next_subvol) {
-			goto unwind;
-		}
+                if (!next_subvol) {
+                        goto unwind;
+                }
 
-		STACK_WIND (frame, dht_readdir_cbk,
-			    next_subvol, next_subvol->fops->readdir,
-			    local->fd, local->size, next_offset);
-		return 0;
-	}
+                STACK_WIND (frame, dht_readdir_cbk,
+                            next_subvol, next_subvol->fops->readdir,
+                            local->fd, local->size, next_offset, NULL);
+                return 0;
+        }
 
 unwind:
-	if (op_ret < 0)
-		op_ret = 0;
+        if (op_ret < 0)
+                op_ret = 0;
 
-	DHT_STACK_UNWIND (frame, op_ret, op_errno, &entries);
+        DHT_STACK_UNWIND (readdir, frame, op_ret, op_errno, &entries, NULL);
 
-	gf_dirent_free (&entries);
+        gf_dirent_free (&entries);
 
         return 0;
 }
 
 
 int
-dht_readdir (call_frame_t *frame, xlator_t *this,
-	     fd_t *fd, size_t size, off_t yoff)
+dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                off_t yoff, int whichop, dict_t *dict)
 {
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
+        dht_local_t  *local  = NULL;
         int           op_errno = -1;
-	xlator_t     *xvol = NULL;
-	off_t         xoff = 0;
-
+        xlator_t     *xvol = NULL;
+        off_t         xoff = 0;
+        int           ret = 0;
+        dht_conf_t   *conf = NULL;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
-	conf = this->private;
+        conf = this->private;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		op_errno = ENOMEM;
-		goto err;
-	}
+        local = dht_local_init (frame, NULL, NULL, whichop);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	local->fd = fd_ref (fd);
-	local->size = size;
+        local->fd = fd_ref (fd);
+        local->size = size;
+        local->xattr_req = (dict)? dict_ref (dict) : NULL;
+        local->first_up_subvol = dht_first_up_subvol (this);
 
-	dht_deitransform (this, yoff, &xvol, (uint64_t *)&xoff);
+        dht_deitransform (this, yoff, &xvol, (uint64_t *)&xoff);
 
-	/* TODO: do proper readdir */
-	STACK_WIND (frame, dht_readdir_cbk,
-		    xvol, xvol->fops->readdir,
-		    fd, size, xoff);
+        /* TODO: do proper readdir */
+        if (whichop == GF_FOP_READDIRP) {
+                if (dict)
+                        local->xattr = dict_ref (dict);
+                else
+                        local->xattr = dict_new ();
 
-	return 0;
+                if (local->xattr) {
+                        ret = dict_set_uint32 (local->xattr,
+                                               conf->link_xattr_name, 256);
+                        if (ret)
+                                gf_log (this->name, GF_LOG_WARNING,
+                                        "failed to set '%s' key",
+                                        conf->link_xattr_name);
+			if (conf->readdir_optimize == _gf_true) {
+                                if (xvol != local->first_up_subvol) {
+				        ret = dict_set_int32 (local->xattr,
+			                               GF_READDIR_SKIP_DIRS, 1);
+				        if (ret)
+					        gf_log (this->name,
+                                                        GF_LOG_ERROR,
+						        "Dict set failed");
+                                } else {
+                                        dict_del (local->xattr,
+                                                  GF_READDIR_SKIP_DIRS);
+                                }
+			}
+                }
+
+                STACK_WIND (frame, dht_readdirp_cbk, xvol, xvol->fops->readdirp,
+                            fd, size, xoff, local->xattr);
+        } else {
+                STACK_WIND (frame, dht_readdir_cbk, xvol, xvol->fops->readdir,
+                            fd, size, xoff, local->xattr);
+        }
+
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
+dht_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t yoff, dict_t *xdata)
+{
+        int          op = GF_FOP_READDIR;
+        dht_conf_t  *conf = NULL;
+        int          i = 0;
+
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (!conf->subvolume_status[i]) {
+                        op = GF_FOP_READDIRP;
+                        break;
+                }
+        }
+
+        if (conf->use_readdirp)
+                op = GF_FOP_READDIRP;
+
+out:
+        dht_do_readdir (frame, this, fd, size, yoff, op, 0);
+        return 0;
+}
+
+int
+dht_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t yoff, dict_t *dict)
+{
+        dht_do_readdir (frame, this, fd, size, yoff, GF_FOP_READDIRP, dict);
+        return 0;
+}
+
+
+
+int
 dht_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int op_ret, int op_errno)
+                  int op_ret, int op_errno, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
 
 
-	local = frame->local;
+        local = frame->local;
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1)
-			local->op_errno = op_errno;
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1)
+                        local->op_errno = op_errno;
 
-		if (op_ret == 0)
-			local->op_ret = 0;
-	}
-	UNLOCK (&frame->lock);
+                if (op_ret == 0)
+                        local->op_ret = 0;
+        }
+        UNLOCK (&frame->lock);
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (frame, local->op_ret, local->op_errno);
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt))
+                DHT_STACK_UNWIND (fsyncdir, frame, local->op_ret,
+                                  local->op_errno, xdata);
 
         return 0;
 }
 
 
 int
-dht_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync)
+dht_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
+              int datasync, dict_t *xdata)
 {
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
+        dht_local_t  *local  = NULL;
+        dht_conf_t   *conf = NULL;
         int           op_errno = -1;
-	int           i = -1;
-
+        int           i = -1;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
-	conf = this->private;
+        conf = this->private;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        local = dht_local_init (frame, NULL, NULL, GF_FOP_FSYNCDIR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	local->fd = fd_ref (fd);
-	local->call_cnt = conf->subvolume_cnt;
+        local->fd = fd_ref (fd);
+        local->call_cnt = conf->subvolume_cnt;
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_fsyncdir_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->fsyncdir,
-			    fd, datasync);
-	}
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                STACK_WIND (frame, dht_fsyncdir_cbk,
+                            conf->subvolumes[i],
+                            conf->subvolumes[i]->fops->fsyncdir,
+                            fd, datasync, xdata);
+        }
 
-	return 0;
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fsyncdir, frame, -1, op_errno, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_newfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int op_ret, int op_errno,
-		 inode_t *inode, struct stat *stbuf)
+                 int op_ret, int op_errno,
+                 inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
 {
-	call_frame_t *prev = NULL;
-	dht_layout_t *layout = NULL;
-	int           ret = -1;
+        xlator_t     *prev = NULL;
+        int           ret = -1;
+        dht_local_t  *local = NULL;
 
 
-	if (op_ret == -1)
-		goto out;
+        if (op_ret == -1)
+                goto out;
 
-	prev = cookie;
+        local = frame->local;
+        if (!local) {
+                op_ret = -1;
+                op_errno = EINVAL;
+                goto out;
+        }
 
-	dht_itransform (this, prev->this, stbuf->st_ino, &stbuf->st_ino);
-	layout = dht_layout_for_subvol (this, prev->this);
+        prev = cookie;
 
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no pre-set layout for subvolume %s",
-			prev->this->name);
-		op_ret   = -1;
-		op_errno = EINVAL;
-		goto out;
-	}
+        if (local->loc.parent) {
 
-	ret = inode_ctx_put (inode, this, (uint64_t)(long)layout);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"could not set inode context");
-		op_ret   = -1;
-		op_errno = EINVAL;
-		goto out;
-	}
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           preparent, 0);
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
+        }
 
+        ret = dht_layout_preset (this, prev, inode);
+        if (ret < 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "could not set pre-set layout for subvolume %s",
+                        prev? prev->name: NULL);
+                op_ret   = -1;
+                op_errno = EINVAL;
+                goto out;
+        }
+        if (local->linked == _gf_true)
+                dht_linkfile_attr_heal (frame, this);
 out:
-	DHT_STACK_UNWIND (frame, op_ret, op_errno, inode, stbuf);
-	return 0;
+        /*
+         * FIXME: ia_size and st_blocks of preparent and postparent do not have
+         * correct values. since, preparent and postparent buffers correspond
+         * to a directory these two members should have values equal to sum of
+         * corresponding values from each of the subvolume.
+         * See dht_iatt_merge for reference.
+         */
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (mknod, frame, op_ret, op_errno, inode, stbuf,
+                          preparent, postparent, xdata);
+        return 0;
 }
 
+int
+dht_mknod_linkfile_create_cbk (call_frame_t *frame, void *cookie,
+                               xlator_t *this,
+                               int32_t op_ret, int32_t op_errno,
+                               inode_t *inode, struct iatt *stbuf,
+                               struct iatt *preparent, struct iatt *postparent,
+                               dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        xlator_t     *cached_subvol = NULL;
+
+        if (op_ret == -1)
+                goto err;
+
+        local = frame->local;
+        if (!local || !local->cached_subvol) {
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        cached_subvol = local->cached_subvol;
+
+        STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)cached_subvol,
+                           cached_subvol, cached_subvol->fops->mknod,
+                           &local->loc, local->mode, local->rdev, local->umask,
+                           local->params);
+
+        return 0;
+err:
+        DHT_STACK_UNWIND (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                          NULL);
+        return 0;
+}
 
 int
 dht_mknod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dev_t rdev)
+           loc_t *loc, mode_t mode, dev_t rdev, mode_t umask, dict_t *params)
 {
-	xlator_t  *subvol = NULL;
-	int        op_errno = -1;
+        xlator_t    *subvol = NULL;
+        int          op_errno = -1;
+        xlator_t    *avail_subvol = NULL;
+        dht_local_t *local = NULL;
 
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
 
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
+        dht_get_du_info (frame, this, loc);
 
-	subvol = dht_subvol_get_hashed (this, loc);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = ENOENT;
-		goto err;
-	}
+        local = dht_local_init (frame, loc, NULL, GF_FOP_MKNOD);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"creating %s on %s", loc->path, subvol->name);
+        subvol = dht_subvol_get_hashed (this, loc);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no subvolume in layout for path=%s",
+                        loc->path);
+                op_errno = ENOENT;
+                goto err;
+        }
 
-	STACK_WIND (frame, dht_newfile_cbk,
-		    subvol, subvol->fops->mknod,
-		    loc, mode, rdev);
+        if (!dht_is_subvol_filled (this, subvol)) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "creating %s on %s", loc->path, subvol->name);
 
-	return 0;
+                STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol,
+                                   subvol, subvol->fops->mknod, loc, mode,
+                                   rdev, umask, params);
+        } else {
+
+                avail_subvol = dht_free_disk_available_subvol (this, subvol,
+                                                               local);
+                if (avail_subvol != subvol) {
+                        /* Choose the minimum filled volume, and create the
+                           files there */
+
+                        local->params = dict_ref (params);
+                        local->cached_subvol = avail_subvol;
+                        local->mode = mode;
+                        local->rdev = rdev;
+                        local->umask = umask;
+                        dht_linkfile_create (frame,
+                                             dht_mknod_linkfile_create_cbk,
+                                             this, avail_subvol, subvol, loc);
+                } else {
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "creating %s on %s", loc->path, subvol->name);
+
+                        STACK_WIND_COOKIE (frame, dht_newfile_cbk,
+                                           (void *)subvol, subvol,
+                                           subvol->fops->mknod, loc, mode,
+                                           rdev, umask, params);
+                }
+        }
+
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
+                          NULL, NULL, NULL, NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_symlink (call_frame_t *frame, xlator_t *this,
-	     const char *linkname, loc_t *loc)
+             const char *linkname, loc_t *loc, mode_t umask, dict_t *params)
 {
-	xlator_t  *subvol = NULL;
-	int        op_errno = -1;
+        xlator_t    *subvol = NULL;
+        int          op_errno = -1;
+        dht_local_t *local = NULL;
 
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
 
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
+        local = dht_local_init (frame, loc, NULL, GF_FOP_SYMLINK);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	subvol = dht_subvol_get_hashed (this, loc);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = ENOENT;
-		goto err;
-	}
+        subvol = dht_subvol_get_hashed (this, loc);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no subvolume in layout for path=%s",
+                        loc->path);
+                op_errno = ENOENT;
+                goto err;
+        }
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"creating %s on %s", loc->path, subvol->name);
+        gf_log (this->name, GF_LOG_TRACE,
+                "creating %s on %s", loc->path, subvol->name);
 
-	STACK_WIND (frame, dht_newfile_cbk,
-		    subvol, subvol->fops->symlink,
-		    linkname, loc);
+        STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol, subvol,
+                           subvol->fops->symlink, linkname, loc, umask,
+                           params);
 
-	return 0;
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (link, frame, -1, op_errno,
+                          NULL, NULL, NULL, NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
-dht_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
+dht_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+            dict_t *xdata)
 {
-	xlator_t    *cached_subvol = NULL;
-	xlator_t    *hashed_subvol = NULL;
-	int          op_errno = -1;
-	dht_local_t *local = NULL;
-
+        xlator_t    *cached_subvol = NULL;
+        xlator_t    *hashed_subvol = NULL;
+        int          op_errno = -1;
+        dht_local_t *local = NULL;
 
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
-
-	cached_subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!cached_subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	hashed_subvol = dht_subvol_get_hashed (this, loc);
-	if (!hashed_subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        if (dht_filter_loc_subvol_key (this, loc, &local->loc,
+                                       &cached_subvol)) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "unlinking %s on %s (given path %s)",
+                        local->loc.path, cached_subvol->name, loc->path);
+                STACK_WIND (frame, dht_unlink_cbk,
+                            cached_subvol, cached_subvol->fops->unlink,
+                            &local->loc, xflag, xdata);
+                goto done;
+        }
 
-	local->call_cnt = 1;
-	if (hashed_subvol != cached_subvol)
-		local->call_cnt++;
+        local = dht_local_init (frame, loc, NULL, GF_FOP_UNLINK);
+        if (!local) {
+                op_errno = ENOMEM;
 
-	STACK_WIND (frame, dht_err_cbk,
-		    cached_subvol, cached_subvol->fops->unlink, loc);
+                goto err;
+        }
 
-	if (hashed_subvol != cached_subvol)
-		STACK_WIND (frame, dht_err_cbk,
-			    hashed_subvol, hashed_subvol->fops->unlink, loc);
+        hashed_subvol = dht_subvol_get_hashed (this, loc);
+        if (!hashed_subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no subvolume in layout for path=%s",
+                        loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-	return 0;
+        cached_subvol = local->cached_subvol;
+        if (!cached_subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
+        local->flags = xflag;
+        if (hashed_subvol != cached_subvol) {
+                STACK_WIND (frame, dht_unlink_linkfile_cbk,
+                            hashed_subvol, hashed_subvol->fops->unlink, loc,
+                            xflag, xdata);
+        } else {
+                STACK_WIND (frame, dht_unlink_cbk,
+                            cached_subvol, cached_subvol->fops->unlink, loc,
+                            xflag, xdata);
+        }
+done:
+        return 0;
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	      int op_ret, int op_errno,
-	      inode_t *inode, struct stat *stbuf)
+              int op_ret, int op_errno,
+              inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
+              struct iatt *postparent, dict_t *xdata)
 {
         call_frame_t *prev = NULL;
-	dht_layout_t *layout = NULL;
-	dht_local_t  *local = NULL;
+        dht_layout_t *layout = NULL;
+        dht_local_t  *local = NULL;
 
         prev = cookie;
-	local = frame->local;
+
+        local = frame->local;
 
         if (op_ret == -1)
                 goto out;
 
-	layout = dht_layout_for_subvol (this, prev->this);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no pre-set layout for subvolume %s",
-			prev->this->name);
-		op_ret   = -1;
-		op_errno = EINVAL;
-		goto out;
-	}
-
-	stbuf->st_ino = local->loc.inode->ino;
+        layout = dht_layout_for_subvol (this, prev->this);
+        if (!layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no pre-set layout for subvolume %s",
+                        prev->this->name);
+                op_ret   = -1;
+                op_errno = EINVAL;
+                goto out;
+        }
 
+        if (local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           preparent, 0);
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
+        }
+        if (local->linked == _gf_true) {
+                local->stbuf = *stbuf;
+                dht_linkfile_attr_heal (frame, this);
+        }
 out:
-        DHT_STACK_UNWIND (frame, op_ret, op_errno, inode, stbuf);
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                          postparent, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_link_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int op_ret, int op_errno,
-		       inode_t *inode, struct stat *stbuf)
+                       int op_ret, int op_errno,
+                       inode_t *inode, struct iatt *stbuf,
+                       struct iatt *preparent, struct iatt *postparent,
+                       dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	xlator_t     *srcvol = NULL;
-
+        dht_local_t  *local = NULL;
+        xlator_t     *srcvol = NULL;
 
-	if (op_ret == -1)
-		goto err;
+        if (op_ret == -1)
+                goto err;
 
-	local = frame->local;
-	srcvol = local->linkfile.srcvol;
+        local = frame->local;
+        srcvol = local->linkfile.srcvol;
 
-	STACK_WIND (frame, dht_link_cbk,
-		    srcvol, srcvol->fops->link,
-		    &local->loc, &local->loc2);
+        STACK_WIND (frame, dht_link_cbk, srcvol, srcvol->fops->link,
+                    &local->loc, &local->loc2, xdata);
 
-	return 0;
+        return 0;
 
 err:
-	DHT_STACK_UNWIND (frame, op_ret, op_errno, inode, stbuf);
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                          postparent, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_link (call_frame_t *frame, xlator_t *this,
-	  loc_t *oldloc, loc_t *newloc)
-{
-	xlator_t    *cached_subvol = NULL;
-	xlator_t    *hashed_subvol = NULL;
-	int          op_errno = -1;
-	int          ret = -1;
-	dht_local_t *local = NULL;
-
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (oldloc, err);
-	VALIDATE_OR_GOTO (newloc, err);
-
-	cached_subvol = dht_subvol_get_cached (this, oldloc->inode);
-	if (!cached_subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", oldloc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+          loc_t *oldloc, loc_t *newloc, dict_t *xdata)
+{
+        xlator_t    *cached_subvol = NULL;
+        xlator_t    *hashed_subvol = NULL;
+        int          op_errno = -1;
+        int          ret = -1;
+        dht_local_t *local = NULL;
 
-	hashed_subvol = dht_subvol_get_hashed (this, newloc);
-	if (!hashed_subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no subvolume in layout for path=%s",
-			newloc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (oldloc, err);
+        VALIDATE_OR_GOTO (newloc, err);
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        local = dht_local_init (frame, oldloc, NULL, GF_FOP_LINK);
+        if (!local) {
+                op_errno = ENOMEM;
 
-	ret = loc_copy (&local->loc, oldloc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+                goto err;
+        }
 
-	ret = loc_copy (&local->loc2, newloc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        cached_subvol = local->cached_subvol;
+        if (!cached_subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", oldloc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-	if (hashed_subvol != cached_subvol) {
-		dht_linkfile_create (frame, dht_link_linkfile_cbk,
-				     cached_subvol, hashed_subvol, newloc);
-	} else {
-		STACK_WIND (frame, dht_link_cbk,
-			    cached_subvol, cached_subvol->fops->link,
-			    oldloc, newloc);
-	}
+        hashed_subvol = dht_subvol_get_hashed (this, newloc);
+        if (!hashed_subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no subvolume in layout for path=%s",
+                        newloc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-	return 0;
+        ret = loc_copy (&local->loc2, newloc);
+        if (ret == -1) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        if (hashed_subvol != cached_subvol) {
+                uuid_copy (local->gfid, oldloc->inode->gfid);
+                dht_linkfile_create (frame, dht_link_linkfile_cbk, this,
+                                     cached_subvol, hashed_subvol, newloc);
+        } else {
+                STACK_WIND (frame, dht_link_cbk,
+                            cached_subvol, cached_subvol->fops->link,
+                            oldloc, newloc, xdata);
+        }
+
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int op_ret, int op_errno,
-		 fd_t *fd, inode_t *inode, struct stat *stbuf)
+                int op_ret, int op_errno,
+                fd_t *fd, inode_t *inode, struct iatt *stbuf,
+                struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
-	call_frame_t *prev = NULL;
-	dht_layout_t *layout = NULL;
-	int           ret = -1;
-
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        dht_local_t  *local = NULL;
 
-	if (op_ret == -1)
-		goto out;
+        if (op_ret == -1)
+                goto out;
 
-	prev = cookie;
+        local = frame->local;
+        if (!local) {
+                op_ret = -1;
+                op_errno = EINVAL;
+                goto out;
+        }
 
-	dht_itransform (this, prev->this, stbuf->st_ino, &stbuf->st_ino);
-	layout = dht_layout_for_subvol (this, prev->this);
+        prev = cookie;
 
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no pre-set layout for subvolume %s",
-			prev->this->name);
-		op_ret   = -1;
-		op_errno = EINVAL;
-		goto out;
-	}
+        if (local->loc.parent) {
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           preparent, 0);
 
-	ret = inode_ctx_put (inode, this, (uint64_t)(long)layout);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"could not set inode context");
-		op_ret   = -1;
-		op_errno = EINVAL;
-		goto out;
-	}
+                dht_inode_ctx_time_update (local->loc.parent, this,
+                                           postparent, 1);
+        }
 
+        ret = dht_layout_preset (this, prev->this, inode);
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "could not set preset layout for subvol %s",
+                        prev->this->name);
+                op_ret   = -1;
+                op_errno = EINVAL;
+                goto out;
+        }
+        if (local->linked == _gf_true) {
+                local->stbuf = *stbuf;
+                dht_linkfile_attr_heal (frame, this);
+        }
 out:
-	DHT_STACK_UNWIND (frame, op_ret, op_errno, fd, inode, stbuf);
-	return 0;
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (create, frame, op_ret, op_errno, fd, inode, stbuf, preparent,
+                          postparent, NULL);
+        return 0;
 }
 
 
 int
+dht_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
+                                xlator_t *this,
+                                int32_t op_ret, int32_t op_errno,
+                                inode_t *inode, struct iatt *stbuf,
+                                struct iatt *preparent, struct iatt *postparent,
+                                dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        xlator_t     *cached_subvol = NULL;
+
+        if (op_ret == -1)
+                goto err;
+
+        local = frame->local;
+        cached_subvol = local->cached_subvol;
+
+        STACK_WIND (frame, dht_create_cbk,
+                    cached_subvol, cached_subvol->fops->create,
+                    &local->loc, local->flags, local->mode,
+                    local->umask, local->fd, local->params);
+
+        return 0;
+err:
+        DHT_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL, NULL, NULL);
+        return 0;
+}
+
+int
 dht_create (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t flags, mode_t mode, fd_t *fd)
+            loc_t *loc, int32_t flags, mode_t mode,
+            mode_t umask, fd_t *fd, dict_t *params)
 {
-	xlator_t  *subvol = NULL;
-	int        op_errno = -1;
+        int          op_errno = -1;
+        xlator_t    *subvol = NULL;
+        dht_local_t *local = NULL;
+        xlator_t    *avail_subvol = NULL;
 
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
 
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
+        dht_get_du_info (frame, this, loc);
 
-	subvol = dht_subvol_get_hashed (this, loc);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = ENOENT;
-		goto err;
-	}
+        local = dht_local_init (frame, loc, fd, GF_FOP_CREATE);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"creating %s on %s", loc->path, subvol->name);
+        if (dht_filter_loc_subvol_key (this, loc, &local->loc,
+                                       &subvol)) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "creating %s on %s (got create on %s)",
+                        local->loc.path, subvol->name, loc->path);
+                STACK_WIND (frame, dht_create_cbk,
+                            subvol, subvol->fops->create,
+                            &local->loc, flags, mode, umask, fd, params);
+                goto done;
+        }
 
-	STACK_WIND (frame, dht_create_cbk,
-		    subvol, subvol->fops->create,
-		    loc, flags, mode, fd);
+        subvol = dht_subvol_get_hashed (this, loc);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no subvolume in layout for path=%s",
+                        loc->path);
+                op_errno = ENOENT;
+                goto err;
+        }
 
-	return 0;
+        if (!dht_is_subvol_filled (this, subvol)) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "creating %s on %s", loc->path, subvol->name);
+                STACK_WIND (frame, dht_create_cbk,
+                            subvol, subvol->fops->create,
+                            loc, flags, mode, umask, fd, params);
+                goto done;
+        }
+        /* Choose the minimum filled volume, and create the
+           files there */
+        avail_subvol = dht_free_disk_available_subvol (this, subvol, local);
+        if (avail_subvol != subvol) {
+                local->params = dict_ref (params);
+                local->flags = flags;
+                local->mode = mode;
+                local->umask = umask;
+                local->cached_subvol = avail_subvol;
+                local->hashed_subvol = subvol;
+                gf_log (this->name, GF_LOG_TRACE,
+                        "creating %s on %s (link at %s)", loc->path,
+                        avail_subvol->name, subvol->name);
+                dht_linkfile_create (frame, dht_create_linkfile_create_cbk,
+                                     this, avail_subvol, subvol, loc);
+                goto done;
+        }
+        gf_log (this->name, GF_LOG_TRACE,
+                "creating %s on %s", loc->path, subvol->name);
+        STACK_WIND (frame, dht_create_cbk,
+                    subvol, subvol->fops->create,
+                    loc, flags, mode, umask, fd, params);
+done:
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL, NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_mkdir_selfheal_cbk (call_frame_t *frame, void *cookie,
-			xlator_t *this,
-			int32_t op_ret, int32_t op_errno)
+                        xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	dht_local_t   *local = NULL;
-	dht_layout_t  *layout = NULL;
+        dht_local_t   *local = NULL;
+        dht_layout_t  *layout = NULL;
 
+        local = frame->local;
+        layout = local->selfheal.layout;
 
-	local = frame->local;
-	layout = local->selfheal.layout;
+        if (op_ret == 0) {
+                dht_layout_set (this, local->inode, layout);
+                if (local->loc.parent) {
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->preparent, 0);
 
-	if (op_ret == 0) {
-		inode_ctx_put (local->inode, this, (uint64_t)(long)layout);
-		local->selfheal.layout = NULL;
-		local->stbuf.st_ino = local->st_ino;
-	}
+                        dht_inode_ctx_time_update (local->loc.parent, this,
+                                                   &local->postparent, 1);
+                }
+        }
 
-	DHT_STACK_UNWIND (frame, op_ret, op_errno,
-			  local->inode, &local->stbuf);
+        DHT_STACK_UNWIND (mkdir, frame, op_ret, op_errno,
+                          local->inode, &local->stbuf, &local->preparent,
+                          &local->postparent, NULL);
 
-	return 0;
+        return 0;
 }
 
 int
 dht_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	       int op_ret, int op_errno, inode_t *inode, struct stat *stbuf)
+               int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
+               struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	int           ret = -1;
-	call_frame_t *prev = NULL;
-	dht_layout_t *layout = NULL;
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        int           ret = -1;
+        gf_boolean_t subvol_filled = _gf_false;
+        call_frame_t *prev = NULL;
+        dht_layout_t *layout = NULL;
 
-	local = frame->local;
-	prev  = cookie;
-	layout = local->layout;
+        local = frame->local;
+        prev  = cookie;
+        layout = local->layout;
 
-	LOCK (&frame->lock);
-	{
-		ret = dht_layout_merge (this, layout, prev->this,
-					op_ret, op_errno, NULL);
+        subvol_filled = dht_is_subvol_filled (this, prev->this);
 
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			goto unlock;
-		}
-		dht_stat_merge (this, &local->stbuf, stbuf, prev->this);
-	}
+        LOCK (&frame->lock);
+        {
+                if (subvol_filled && (op_ret != -1)) {
+                        ret = dht_layout_merge (this, layout, prev->this,
+                                                -1, ENOSPC, NULL);
+                } else {
+			if (op_ret == -1 && op_errno == EEXIST)
+				/* Very likely just a race between mkdir and
+				   self-heal (from lookup of a concurrent mkdir
+				   attempt).
+				   Ignore error for now. layout setting will
+				   anyways fail if this was a different (old)
+				   pre-existing different directory.
+				*/
+				op_ret = 0;
+                        ret = dht_layout_merge (this, layout, prev->this,
+                                                op_ret, op_errno, NULL);
+                }
+                if (ret)
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "%s: failed to merge layouts", local->loc.path);
+
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        goto unlock;
+                }
+                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+                dht_iatt_merge (this, &local->preparent, preparent, prev->this);
+                dht_iatt_merge (this, &local->postparent, postparent,
+                                prev->this);
+        }
 unlock:
-	UNLOCK (&frame->lock);
+        UNLOCK (&frame->lock);
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt)) {
-		local->layout = NULL;
-		dht_selfheal_new_directory (frame, dht_mkdir_selfheal_cbk,
-					    layout);
-	}
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+                dht_selfheal_new_directory (frame, dht_mkdir_selfheal_cbk,
+                                            layout);
+        }
 
         return 0;
 }
 
 int
-dht_mkdir_hashed_cbk (call_frame_t *frame, void *cookie, 
-		      xlator_t *this, int op_ret, int op_errno, 
-		      inode_t *inode, struct stat *stbuf)
+dht_mkdir_hashed_cbk (call_frame_t *frame, void *cookie,
+                      xlator_t *this, int op_ret, int op_errno,
+                      inode_t *inode, struct iatt *stbuf,
+                      struct iatt *preparent, struct iatt *postparent,
+                      dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	int           ret = -1;
-	call_frame_t *prev = NULL;
-	dht_layout_t *layout = NULL;
-	dht_conf_t   *conf = NULL;
-	int           i = 0;
-	xlator_t     *hashed_subvol = NULL;
+        dht_local_t  *local = NULL;
+        int           ret = -1;
+        call_frame_t *prev = NULL;
+        dht_layout_t *layout = NULL;
+        dht_conf_t   *conf = NULL;
+        int           i = 0;
+        xlator_t     *hashed_subvol = NULL;
 
-	local = frame->local;
-	prev  = cookie;
-	layout = local->layout;
-	conf = this->private;
-	hashed_subvol = local->hashed_subvol;
+        VALIDATE_OR_GOTO (this->private, err);
 
-	ret = dht_layout_merge (this, layout, prev->this,
-				op_ret, op_errno, NULL);
+        local = frame->local;
+        prev  = cookie;
+        layout = local->layout;
+        conf = this->private;
+        hashed_subvol = local->hashed_subvol;
 
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		goto err;
-	}
-	local->op_ret = 0;
+        if (uuid_is_null (local->loc.gfid) && !op_ret)
+                uuid_copy (local->loc.gfid, stbuf->ia_gfid);
 
-	dht_stat_merge (this, &local->stbuf, stbuf, prev->this);
+        if (dht_is_subvol_filled (this, hashed_subvol))
+                ret = dht_layout_merge (this, layout, prev->this,
+                                        -1, ENOSPC, NULL);
+        else
+                ret = dht_layout_merge (this, layout, prev->this,
+                                        op_ret, op_errno, NULL);
 
-	local->st_ino = local->stbuf.st_ino;
+        /* TODO: we may have to return from the function
+           if layout merge fails. For now, lets just log an error */
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to merge layouts", local->loc.path);
 
-	local->call_cnt = conf->subvolume_cnt - 1;
-	
-	if (local->call_cnt == 0) {
-		local->layout = NULL;
-		dht_selfheal_directory (frame, dht_mkdir_selfheal_cbk,
-					&local->loc, layout);
-	}
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		if (conf->subvolumes[i] == hashed_subvol)
-			continue;
-		STACK_WIND (frame, dht_mkdir_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->mkdir,
-			    &local->loc, local->mode);
-	}
-	return 0;
+        if (op_ret == -1) {
+                local->op_errno = op_errno;
+                goto err;
+        }
+        local->op_ret = 0;
+
+        dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+        dht_iatt_merge (this, &local->preparent, preparent, prev->this);
+        dht_iatt_merge (this, &local->postparent, postparent, prev->this);
+
+        local->call_cnt = conf->subvolume_cnt - 1;
+
+        if (uuid_is_null (local->loc.gfid))
+                uuid_copy (local->loc.gfid, stbuf->ia_gfid);
+        if (local->call_cnt == 0) {
+                dht_selfheal_directory (frame, dht_mkdir_selfheal_cbk,
+                                        &local->loc, layout);
+        }
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (conf->subvolumes[i] == hashed_subvol)
+                        continue;
+                STACK_WIND (frame, dht_mkdir_cbk,
+                            conf->subvolumes[i],
+                            conf->subvolumes[i]->fops->mkdir, &local->loc,
+                            local->mode, local->umask, local->params);
+        }
+        return 0;
 err:
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
+        DHT_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL, NULL);
         return 0;
 }
 
-int
+
+ int
 dht_mkdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode)
+           loc_t *loc, mode_t mode, mode_t umask, dict_t *params)
 {
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
+        dht_local_t  *local  = NULL;
+        dht_conf_t   *conf = NULL;
         int           op_errno = -1;
-	int           ret = -1;
-	xlator_t     *hashed_subvol = NULL;
+        xlator_t     *hashed_subvol = NULL;
 
 
         VALIDATE_OR_GOTO (frame, err);
@@ -2827,723 +4200,1013 @@ dht_mkdir (call_frame_t *frame, xlator_t *this,
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
         VALIDATE_OR_GOTO (loc->path, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
-	conf = this->private;
-
-	local = dht_local_init (frame);
-	if (!local) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		op_errno = ENOMEM;
-		goto err;
-	}
+        conf = this->private;
 
-	hashed_subvol = dht_subvol_get_hashed (this, loc);
+        dht_get_du_info (frame, this, loc);
 
-	if (hashed_subvol == NULL) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"hashed subvol not found");
-		op_errno = EINVAL;
-		goto err;
-	}
+        local = dht_local_init (frame, loc, NULL, GF_FOP_MKDIR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	local->hashed_subvol = hashed_subvol;
-	local->inode = inode_ref (loc->inode);
-	ret = loc_copy (&local->loc, loc);
-	local->mode = mode;
+        hashed_subvol = dht_subvol_get_hashed (this, loc);
+        if (hashed_subvol == NULL) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "hashed subvol not found for %s",
+                        loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-	if (ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		op_errno = ENOMEM;
-		goto err;
-	}
+        local->hashed_subvol = hashed_subvol;
+        local->mode = mode;
+        local->umask = umask;
+        local->params = dict_ref (params);
+        local->inode  = inode_ref (loc->inode);
 
-	local->layout = dht_layout_new (this, conf->subvolume_cnt);
-	if (!local->layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		op_errno = ENOMEM;
-		goto err;
-	}
+        local->layout = dht_layout_new (this, conf->subvolume_cnt);
+        if (!local->layout) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	STACK_WIND (frame, dht_mkdir_hashed_cbk,
-		    hashed_subvol,
-		    hashed_subvol->fops->mkdir,
-		    loc, mode);
+        STACK_WIND (frame, dht_mkdir_hashed_cbk,
+                    hashed_subvol,
+                    hashed_subvol->fops->mkdir,
+                    loc, mode, umask, params);
 
-	return 0;
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL,
+                          NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_rmdir_selfheal_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int op_ret, int op_errno)
+                        int op_ret, int op_errno, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
+        dht_local_t  *local = NULL;
 
-	local = frame->local;
-	local->layout = NULL;
+        local = frame->local;
 
-	DHT_STACK_UNWIND (frame, local->op_ret, local->op_errno);
+        DHT_STACK_UNWIND (rmdir, frame, local->op_ret, local->op_errno,
+                          &local->preparent, &local->postparent, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
-dht_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	       int op_ret, int op_errno)
+dht_rmdir_hashed_subvol_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int op_ret, int op_errno, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
 {
-	uint64_t      tmp_layout = 0;
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
-	dht_layout_t *layout = NULL;
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
 
-	local = frame->local;
-	prev  = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			local->op_ret   = -1;
-
-			if (op_errno != ENOENT)
-				local->need_selfheal = 1;
-
-			gf_log (this->name, GF_LOG_ERROR,
-				"rmdir on %s for %s failed (%s)",
-				prev->this->name, local->loc.path,
-				strerror (op_errno));
-			goto unlock;
-		}
-	}
+        local = frame->local;
+        prev  = cookie;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        local->op_ret   = -1;
+                        if (op_errno != ENOENT && op_errno != EACCES) {
+                                local->need_selfheal = 1;
+                        }
+
+
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "rmdir on %s for %s failed (%s)",
+                                prev->this->name, local->loc.path,
+                                strerror (op_errno));
+                        goto unlock;
+                }
+
+                dht_iatt_merge (this, &local->preparent, preparent, prev->this);
+                dht_iatt_merge (this, &local->postparent, postparent,
+                                prev->this);
+
+        }
 unlock:
-	UNLOCK (&frame->lock);
+        UNLOCK (&frame->lock);
 
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+               if (local->need_selfheal) {
+                        local->layout =
+                                dht_layout_get (this, local->loc.inode);
+
+                        /* TODO: neater interface needed below */
+                        local->stbuf.ia_type = local->loc.inode->ia_type;
+
+                        uuid_copy (local->gfid, local->loc.inode->gfid);
+                        dht_selfheal_restore (frame, dht_rmdir_selfheal_cbk,
+                                              &local->loc, local->layout);
+               } else {
+
+                        if (local->loc.parent) {
+                                dht_inode_ctx_time_update (local->loc.parent,
+                                                           this,
+                                                           &local->preparent,
+                                                           0);
+
+                                dht_inode_ctx_time_update (local->loc.parent,
+                                                           this,
+                                                           &local->postparent,
+                                                           1);
+                        }
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt)) {
-		if (local->need_selfheal) {
-			inode_ctx_get (local->loc.inode, this, 
-				       &tmp_layout);
-			layout = (dht_layout_t *)(long)tmp_layout;
-
-			/* TODO: neater interface needed below */
-			local->stbuf.st_mode = local->loc.inode->st_mode;
-
-			dht_selfheal_restore (frame, dht_rmdir_selfheal_cbk,
-					      &local->loc, layout);
-		} else {
-			DHT_STACK_UNWIND (frame, local->op_ret,
-					  local->op_errno);
-		}
-	}
+                        DHT_STACK_UNWIND (rmdir, frame, local->op_ret,
+                                          local->op_errno, &local->preparent,
+                                          &local->postparent, NULL);
+               }
+        }
 
         return 0;
 }
 
 
 int
-dht_rmdir_do (call_frame_t *frame, xlator_t *this)
+dht_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int op_ret, int op_errno, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	dht_conf_t   *conf = NULL;
-	int           i = 0;
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
+        int           done = 0;
 
-	conf = this->private;
-	local = frame->local;
+        local = frame->local;
+        prev  = cookie;
 
-	if (local->op_ret == -1)
-		goto err;
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        local->op_ret   = -1;
 
-	local->call_cnt = conf->subvolume_cnt;
+                        if (op_errno != ENOENT && op_errno != EACCES) {
+                                local->need_selfheal = 1;
+                        }
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_rmdir_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->rmdir,
-			    &local->loc);
-	}
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "rmdir on %s for %s failed (%s)",
+                                prev->this->name, local->loc.path,
+                                strerror (op_errno));
+                        goto unlock;
+                }
 
-	return 0;
+                /* Track if rmdir succeeded on atleast one subvol*/
+                local->fop_succeeded = 1;
+                dht_iatt_merge (this, &local->preparent, preparent, prev->this);
+                dht_iatt_merge (this, &local->postparent, postparent,
+                                prev->this);
+        }
+unlock:
+        UNLOCK (&frame->lock);
 
-err:
-	DHT_STACK_UNWIND (frame, local->op_ret, local->op_errno);
-	return 0;
-}
 
+        this_call_cnt = dht_frame_return (frame);
 
-int
-dht_rmdir_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int op_ret, int op_errno, gf_dirent_t *entries)
-{
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = -1;
-	call_frame_t *prev = NULL;
+        /* if local->hashed_subvol, we are yet to wind to hashed_subvol. */
+        if (local->hashed_subvol && (this_call_cnt == 1)) {
+                done = 1;
+        } else if (!local->hashed_subvol && !this_call_cnt) {
+                done = 1;
+        }
 
-	local = frame->local;
-	prev  = cookie;
-
-	if (op_ret > 2) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"readdir on %s for %s returned %d entries",
-			prev->this->name, local->loc.path, op_ret);
-		local->op_ret = -1;
-		local->op_errno = ENOTEMPTY;
-	}
 
-	this_call_cnt = dht_frame_return (frame);
+        if (done) {
+                if (local->need_selfheal && local->fop_succeeded) {
+                        local->layout =
+                                dht_layout_get (this, local->loc.inode);
 
-	if (is_last_call (this_call_cnt)) {
-		dht_rmdir_do (frame, this);
-	}
+                        /* TODO: neater interface needed below */
+                        local->stbuf.ia_type = local->loc.inode->ia_type;
 
-	return 0;
+                        uuid_copy (local->gfid, local->loc.inode->gfid);
+                        dht_selfheal_restore (frame, dht_rmdir_selfheal_cbk,
+                                              &local->loc, local->layout);
+                } else if (this_call_cnt) {
+                        /* If non-hashed subvol's have responded, proceed */
+
+                        local->need_selfheal = 0;
+                        STACK_WIND (frame, dht_rmdir_hashed_subvol_cbk,
+                                    local->hashed_subvol,
+                                    local->hashed_subvol->fops->rmdir,
+                                    &local->loc, local->flags, NULL);
+                } else if (!this_call_cnt) {
+                        /* All subvol's have responded, proceed */
+
+                        if (local->loc.parent) {
+
+                                dht_inode_ctx_time_update (local->loc.parent,
+                                                           this,
+                                                           &local->preparent,
+                                                           0);
+
+                                dht_inode_ctx_time_update (local->loc.parent,
+                                                           this,
+                                                           &local->postparent,
+                                                           1);
+
+                        }
+
+                        DHT_STACK_UNWIND (rmdir, frame, local->op_ret,
+                                          local->op_errno, &local->preparent,
+                                          &local->postparent, NULL);
+                }
+        }
+
+        return 0;
 }
 
 
 int
-dht_rmdir_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int op_ret, int op_errno, fd_t *fd)
+dht_rmdir_do (call_frame_t *frame, xlator_t *this)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = -1;
-	call_frame_t *prev = NULL;
+        dht_local_t  *local = NULL;
+        dht_conf_t   *conf = NULL;
+        int           i = 0;
+        xlator_t     *hashed_subvol = NULL;
 
+        VALIDATE_OR_GOTO (this->private, err);
 
-	local = frame->local;
-	prev  = cookie;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"opendir on %s for %s failed (%s)",
-			prev->this->name, local->loc.path,
-			strerror (op_errno));
-		goto err;
-	}
+        conf = this->private;
+        local = frame->local;
 
-	STACK_WIND (frame, dht_rmdir_readdir_cbk,
-		    prev->this, prev->this->fops->readdir,
-		    local->fd, 4096, 0);
+        if (local->op_ret == -1)
+                goto err;
 
-	return 0;
+        local->call_cnt = conf->subvolume_cnt;
 
-err:
-	this_call_cnt = dht_frame_return (frame);
+        /* first remove from non-hashed_subvol */
+        hashed_subvol = dht_subvol_get_hashed (this, &local->loc);
 
-	if (is_last_call (this_call_cnt)) {
-		dht_rmdir_do (frame, this);
-	}
+        if (!hashed_subvol) {
+                gf_log (this->name, GF_LOG_WARNING, "failed to get hashed "
+                        "subvol for %s",local->loc.path);
+        } else {
+                local->hashed_subvol = hashed_subvol;
+        }
 
-	return 0;
+        /* When DHT has only 1 child */
+        if (conf->subvolume_cnt == 1) {
+                STACK_WIND (frame, dht_rmdir_hashed_subvol_cbk,
+                            conf->subvolumes[0],
+                            conf->subvolumes[0]->fops->rmdir,
+                            &local->loc, local->flags, NULL);
+                return 0;
+        }
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (hashed_subvol &&
+                    (hashed_subvol == conf->subvolumes[i]))
+                        continue;
+
+                STACK_WIND (frame, dht_rmdir_cbk,
+                            conf->subvolumes[i],
+                            conf->subvolumes[i]->fops->rmdir,
+                            &local->loc, local->flags, NULL);
+        }
+
+        return 0;
+
+err:
+        DHT_STACK_UNWIND (rmdir, frame, local->op_ret, local->op_errno,
+                          &local->preparent, &local->postparent, NULL);
+        return 0;
 }
 
 
 int
-dht_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc)
+dht_rmdir_linkfile_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                               int op_ret, int op_errno, struct iatt *preparent,
+                               struct iatt *postparent, dict_t *xdata)
 {
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-	int           i = -1;
-	int           ret = -1;
+        dht_local_t    *local = NULL;
+        call_frame_t   *prev = NULL;
+        xlator_t       *src = NULL;
+        call_frame_t   *main_frame = NULL;
+        dht_local_t    *main_local = NULL;
+        int             this_call_cnt = 0;
+
+        local  = frame->local;
+        prev   = cookie;
+        src    = prev->this;
 
+        main_frame = local->main_frame;
+        main_local = main_frame->local;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        if (op_ret == 0) {
+                gf_log (this->name, GF_LOG_TRACE,
+                        "unlinked linkfile %s on %s",
+                        local->loc.path, src->name);
+        } else {
+                main_local->op_ret   = -1;
+                main_local->op_errno = op_errno;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "unlink of %s on %s failed (%s)",
+                        local->loc.path, src->name, strerror (op_errno));
+        }
 
-	conf = this->private;
+        this_call_cnt = dht_frame_return (main_frame);
+        if (is_last_call (this_call_cnt))
+                dht_rmdir_do (main_frame, this);
 
-	local = dht_local_init (frame);
-	if (!local) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		op_errno = ENOMEM;
-		goto err;
-	}
+        DHT_STACK_DESTROY (frame);
+        return 0;
+}
 
-	local->call_cnt = conf->subvolume_cnt;
-	local->op_ret   = 0;
 
-	ret = loc_copy (&local->loc, loc);
-	if (ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		op_errno = ENOMEM;
-		goto err;
-	}
+int
+dht_rmdir_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, inode_t *inode,
+                      struct iatt *stbuf, dict_t *xattr, struct iatt *parent)
+{
+        dht_local_t    *local = NULL;
+        call_frame_t   *prev = NULL;
+        xlator_t       *src = NULL;
+        call_frame_t   *main_frame = NULL;
+        dht_local_t    *main_local = NULL;
+        int             this_call_cnt = 0;
+        dht_conf_t     *conf = this->private;
 
-	local->fd = fd_create (local->loc.inode, frame->root->pid);
-	if (!local->fd) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		op_errno = ENOMEM;
-		goto err;
-	}
+        local = frame->local;
+        prev  = cookie;
+        src   = prev->this;
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_rmdir_opendir_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->opendir,
-			    loc, local->fd);
-	}
+        main_frame = local->main_frame;
+        main_local = main_frame->local;
 
-	return 0;
+        if (op_ret != 0)
+                goto err;
+
+        if (!check_is_linkfile (inode, stbuf, xattr, conf->link_xattr_name)) {
+                main_local->op_ret  = -1;
+                main_local->op_errno = ENOTEMPTY;
 
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s on %s found to be not a linkfile (type=0%o)",
+                        local->loc.path, src->name, stbuf->ia_type);
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_rmdir_linkfile_unlink_cbk,
+                    src, src->fops->unlink, &local->loc, 0, NULL);
+        return 0;
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno);
 
-	return 0;
+        this_call_cnt = dht_frame_return (main_frame);
+        if (is_last_call (this_call_cnt))
+                dht_rmdir_do (main_frame, this);
+
+        DHT_STACK_DESTROY (frame);
+        return 0;
 }
 
 
-static int32_t
-dht_xattrop_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 dict_t *dict)
+int
+dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
+                           gf_dirent_t *entries, xlator_t *src)
 {
-	DHT_STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
+        int                 ret = 0;
+        int                 build_ret = 0;
+        gf_dirent_t        *trav = NULL;
+        call_frame_t       *lookup_frame = NULL;
+        dht_local_t        *lookup_local = NULL;
+        dht_local_t        *local = NULL;
+        dict_t             *xattrs = NULL;
+        dht_conf_t         *conf = this->private;
 
-int32_t
-dht_xattrop (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     gf_xattrop_flags_t flags,
-	     dict_t *dict)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+        local = frame->local;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        list_for_each_entry (trav, &entries->list, list) {
+                if (strcmp (trav->d_name, ".") == 0)
+                        continue;
+                if (strcmp (trav->d_name, "..") == 0)
+                        continue;
+                if (check_is_linkfile (NULL, (&trav->d_stat), trav->dict,
+                                              conf->link_xattr_name)) {
+                        ret++;
+                        continue;
+                }
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+                /* this entry is either a directory which is neither "." nor "..",
+                   or a non directory which is not a linkfile. the directory is to
+                   be treated as non-empty
+                */
+                return 0;
+        }
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        xattrs = dict_new ();
+        if (!xattrs) {
+                gf_log (this->name, GF_LOG_ERROR, "dict_new failed");
+                return -1;
+        }
 
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = 1;
+        ret = dict_set_uint32 (xattrs, conf->link_xattr_name, 256);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "failed to set linkto key"
+                        " in dict");
+                if (xattrs)
+                        dict_unref (xattrs);
+                return -1;
+        }
 
-	STACK_WIND (frame,
-		    dht_xattrop_cbk,
-		    subvol, subvol->fops->xattrop,
-		    loc, flags, dict);
+        list_for_each_entry (trav, &entries->list, list) {
+                if (strcmp (trav->d_name, ".") == 0)
+                        continue;
+                if (strcmp (trav->d_name, "..") == 0)
+                        continue;
 
-	return 0;
+                lookup_frame = NULL;
+                lookup_local = NULL;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+                lookup_frame = copy_frame (frame);
+                if (!lookup_frame) {
+                        /* out of memory, let the rmdir fail
+                           (as non-empty, unfortunately) */
+                        goto err;
+                }
 
-	return 0;
-}
+                lookup_local = mem_get0 (this->local_pool);
+                if (!lookup_local) {
+                        goto err;
+                }
 
-static int32_t
-dht_fxattrop_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  dict_t *dict)
-{
-	DHT_STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
+                lookup_frame->local = lookup_local;
+                lookup_local->main_frame = frame;
 
-int32_t
-dht_fxattrop (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      gf_xattrop_flags_t flags,
-	      dict_t *dict)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
+                build_ret = dht_build_child_loc (this, &lookup_local->loc,
+                                                 &local->loc, trav->d_name);
+                if (build_ret != 0)
+                        goto err;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+                uuid_copy (lookup_local->loc.gfid, trav->d_stat.ia_gfid);
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+                gf_log (this->name, GF_LOG_TRACE,
+                        "looking up %s on %s",
+                        lookup_local->loc.path, src->name);
 
-	STACK_WIND (frame,
-		    dht_fxattrop_cbk,
-		    subvol, subvol->fops->fxattrop,
-		    fd, flags, dict);
+                LOCK (&frame->lock);
+                {
+                        local->call_cnt++;
+                }
+                UNLOCK (&frame->lock);
 
-	return 0;
+                STACK_WIND (lookup_frame, dht_rmdir_lookup_cbk,
+                            src, src->fops->lookup,
+                            &lookup_local->loc, xattrs);
+                ret++;
+        }
+
+        if (xattrs)
+                dict_unref (xattrs);
 
+        return ret;
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno, NULL);
+        if (xattrs)
+                dict_unref (xattrs);
 
-	return 0;
+        DHT_STACK_DESTROY (lookup_frame);
+        return 0;
 }
 
 
-static int32_t
-dht_inodelk_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno)
-
+int
+dht_rmdir_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, gf_dirent_t *entries,
+                        dict_t *xdata)
 {
-	DHT_STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = -1;
+        call_frame_t *prev = NULL;
+        xlator_t     *src = NULL;
+        int           ret = 0;
+
+        local = frame->local;
+        prev  = cookie;
+        src   = prev->this;
+
+        if (op_ret > 2) {
+                ret = dht_rmdir_is_subvol_empty (frame, this, entries, src);
+
+                switch (ret) {
+                case 0: /* non linkfiles exist */
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "readdir on %s for %s returned %d entries",
+                                prev->this->name, local->loc.path, op_ret);
+                        local->op_ret = -1;
+                        local->op_errno = ENOTEMPTY;
+                        break;
+                default:
+                        /* @ret number of linkfiles are getting unlinked */
+                        gf_log (this->name, GF_LOG_TRACE,
+                                "readdir on %s for %s found %d linkfiles",
+                                prev->this->name, local->loc.path, ret);
+                        break;
+                }
+        }
+
+        this_call_cnt = dht_frame_return (frame);
+
+        if (is_last_call (this_call_cnt)) {
+                dht_rmdir_do (frame, this);
+        }
+
+        return 0;
 }
 
 
-int32_t
-dht_inodelk (call_frame_t *frame, xlator_t *this,
-	     const char *volume, loc_t *loc, int32_t cmd, struct flock *lock)
+int
+dht_rmdir_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = -1;
+        call_frame_t *prev = NULL;
+        dict_t       *dict = NULL;
+        int           ret = 0;
+        dht_conf_t   *conf = this->private;
 
+        local = frame->local;
+        prev  = cookie;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+        if (op_ret == -1) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "opendir on %s for %s failed (%s)",
+                        prev->this->name, local->loc.path,
+                        strerror (op_errno));
+                if (op_errno != ENOENT) {
+                        local->op_ret = -1;
+                        local->op_errno = op_errno;
+                }
+                goto err;
+        }
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        dict = dict_new ();
+        if (!dict) {
+                local->op_ret = -1;
+                local->op_errno = ENOMEM;
+                goto err;
+        }
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        ret = dict_set_uint32 (dict, conf->link_xattr_name, 256);
+        if (ret)
+                gf_log (this->name, GF_LOG_WARNING,
+                        "%s: failed to set '%s' key",
+                        local->loc.path, conf->link_xattr_name);
 
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = 1;
+        STACK_WIND (frame, dht_rmdir_readdirp_cbk,
+                    prev->this, prev->this->fops->readdirp,
+                    local->fd, 4096, 0, dict);
 
-	STACK_WIND (frame,
-		    dht_inodelk_cbk,
-		    subvol, subvol->fops->inodelk,
-		    volume, loc, cmd, lock);
+        if (dict)
+                dict_unref (dict);
 
-	return 0;
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno);
-
-	return 0;
-}
-
+        this_call_cnt = dht_frame_return (frame);
 
-static int32_t
-dht_finodelk_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno)
+        if (is_last_call (this_call_cnt)) {
+                dht_rmdir_do (frame, this);
+        }
 
-{
-	DHT_STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
+        return 0;
 }
 
 
-int32_t
-dht_finodelk (call_frame_t *frame, xlator_t *this,
-	      const char *volume, fd_t *fd, int32_t cmd, struct flock *lock)
+int
+dht_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+           dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
+        dht_local_t  *local  = NULL;
+        dht_conf_t   *conf = NULL;
         int           op_errno = -1;
+        int           i = -1;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+        VALIDATE_OR_GOTO (this->private, err);
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        conf = this->private;
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_RMDIR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
+        local->call_cnt = conf->subvolume_cnt;
+        local->op_ret   = 0;
+        local->fop_succeeded = 0;
 
-	STACK_WIND (frame,
-		    dht_finodelk_cbk,
-		    subvol, subvol->fops->finodelk,
-		    this->name, fd, cmd, lock);
+        local->flags = flags;
 
-	return 0;
+        local->fd = fd_create (local->loc.inode, frame->root->pid);
+        if (!local->fd) {
+
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                STACK_WIND (frame, dht_rmdir_opendir_cbk,
+                            conf->subvolumes[i],
+                            conf->subvolumes[i]->fops->opendir,
+                            loc, local->fd, NULL);
+        }
+
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (rmdir, frame, -1, op_errno,
+                          NULL, NULL, NULL);
 
-	return 0;
+        return 0;
 }
 
-
-static int32_t
+int
 dht_entrylk_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno)
+                 xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
 {
-	DHT_STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
+        DHT_STACK_UNWIND (entrylk, frame, op_ret, op_errno, xdata);
+        return 0;
 }
 
-int32_t
+
+int
 dht_entrylk (call_frame_t *frame, xlator_t *this,
-	     const char *volume, loc_t *loc, const char *basename,
-	     entrylk_cmd cmd, entrylk_type type)
+             const char *volume, loc_t *loc, const char *basename,
+             entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
+        xlator_t     *subvol = NULL;
         int           op_errno = -1;
-	dht_local_t  *local = NULL;
+        dht_local_t  *local = NULL;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (loc, err);
         VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        local = dht_local_init (frame, loc, NULL, GF_FOP_ENTRYLK);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto err;
-	}
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = 1;
+        local->call_cnt = 1;
 
-	STACK_WIND (frame, dht_entrylk_cbk,
-		    subvol, subvol->fops->entrylk,
-		    volume, loc, basename, cmd, type);
+        STACK_WIND (frame, dht_entrylk_cbk,
+                    subvol, subvol->fops->entrylk,
+                    volume, loc, basename, cmd, type, xdata);
 
-	return 0;
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (entrylk, frame, -1, op_errno, NULL);
 
-	return 0;
+        return 0;
 }
 
-static int32_t
+
+int
 dht_fentrylk_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno)
+                  xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
 {
-	DHT_STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
+        DHT_STACK_UNWIND (fentrylk, frame, op_ret, op_errno, NULL);
+        return 0;
 }
 
-int32_t
+
+int
 dht_fentrylk (call_frame_t *frame, xlator_t *this,
-	      const char *volume, fd_t *fd, const char *basename,
-	      entrylk_cmd cmd, entrylk_type type)
+              const char *volume, fd_t *fd, const char *basename,
+              entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
+        xlator_t     *subvol = NULL;
         int           op_errno = -1;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        subvol = dht_subvol_get_cached (this, fd->inode);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
 
-	STACK_WIND (frame, dht_fentrylk_cbk,
-		    subvol, subvol->fops->fentrylk,
-		    volume, fd, basename, cmd, type);
+        STACK_WIND (frame, dht_fentrylk_cbk,
+                    subvol, subvol->fops->fentrylk,
+                    volume, fd, basename, cmd, type, xdata);
 
-	return 0;
+        return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (frame, -1, op_errno);
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fentrylk, frame, -1, op_errno, NULL);
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_forget (xlator_t *this, inode_t *inode)
 {
-	uint64_t      tmp_layout = 0;
-	dht_layout_t *layout = NULL;
+        uint64_t        ctx_int = 0;
+        dht_inode_ctx_t *ctx    = NULL;
+        dht_layout_t *layout = NULL;
 
-	inode_ctx_get (inode, this, &tmp_layout);
+        inode_ctx_del (inode, this, &ctx_int);
 
-	if (!tmp_layout)
-		return 0;
-	layout = (dht_layout_t *)(long)tmp_layout;
-	if (!layout->preset)
-		FREE (layout);
+        if (!ctx_int)
+                return 0;
 
-	return 0;
-}
+        ctx = (dht_inode_ctx_t *) (long) ctx_int;
 
+        layout = ctx->layout;
+        ctx->layout = NULL;
+        dht_layout_unref (this, layout);
+        GF_FREE (ctx);
 
+        return 0;
+}
 
-static int
-dht_init_subvolumes (xlator_t *this, dht_conf_t *conf)
-{
-        xlator_list_t *subvols = NULL;
-        int            cnt = 0;
 
+int
+dht_notify (xlator_t *this, int event, void *data, ...)
+{
+        xlator_t                *subvol = NULL;
+        int                      cnt    = -1;
+        int                      i      = -1;
+        dht_conf_t              *conf   = NULL;
+        int                      ret    = -1;
+        int                      propagate = 0;
+
+        int                      had_heard_from_all = 0;
+        int                      have_heard_from_all = 0;
+        struct timeval           time = {0,};
+        gf_defrag_info_t        *defrag = NULL;
+        dict_t                  *dict   = NULL;
+        gf_defrag_type           cmd    = 0;
+        dict_t                  *output = NULL;
+        va_list                  ap;
 
-        for (subvols = this->children; subvols; subvols = subvols->next)
-                cnt++;
 
-        conf->subvolumes = CALLOC (cnt, sizeof (xlator_t *));
-        if (!conf->subvolumes) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "memory allocation failed :(");
-                return -1;
+        conf = this->private;
+        if (!conf)
+                return ret;
+
+        /* had all subvolumes reported status once till now? */
+        had_heard_from_all = 1;
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (!conf->last_event[i]) {
+                        had_heard_from_all = 0;
+                }
         }
-        conf->subvolume_cnt = cnt;
 
-        cnt = 0;
-        for (subvols = this->children; subvols; subvols = subvols->next)
-                conf->subvolumes[cnt++] = subvols->xlator;
+        switch (event) {
+        case GF_EVENT_CHILD_UP:
+                subvol = data;
 
-	conf->subvolume_status = CALLOC (cnt, sizeof (char));
-	if (!conf->subvolume_status) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		return -1;
-	}
+                conf->gen++;
 
-        return 0;
-}
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                        if (subvol == conf->subvolumes[i]) {
+                                cnt = i;
+                                break;
+                        }
+                }
 
+                if (cnt == -1) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "got GF_EVENT_CHILD_UP bad subvolume %s",
+                                subvol->name);
+                        break;
+                }
 
-int
-dht_notify (xlator_t *this, int event, void *data, ...)
-{
-	xlator_t   *subvol = NULL;
-	int         cnt    = -1;
-	int         i      = -1;
-	dht_conf_t *conf   = NULL;
-	int         ret    = -1;
+                gettimeofday (&time, NULL);
+                LOCK (&conf->subvolume_lock);
+                {
+                        conf->subvolume_status[cnt] = 1;
+                        conf->last_event[cnt] = event;
+                        conf->subvol_up_time[cnt] = time.tv_sec;
+                }
+                UNLOCK (&conf->subvolume_lock);
 
+                /* one of the node came back up, do a stat update */
+                dht_get_du_info_for_subvol (this, cnt);
 
-	conf = this->private;
+                break;
 
-	switch (event) {
-	case GF_EVENT_CHILD_UP:
-		subvol = data;
+        case GF_EVENT_CHILD_MODIFIED:
+                subvol = data;
 
-		conf->gen++;
+                conf->gen++;
+                propagate = 1;
 
-		for (i = 0; i < conf->subvolume_cnt; i++) {
-			if (subvol == conf->subvolumes[i]) {
-				cnt = i;
-				break;
-			}
-		}
+                break;
 
-		if (cnt == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"got GF_EVENT_CHILD_UP bad subvolume %s",
-				subvol->name);
-			break;
-		}
+        case GF_EVENT_CHILD_DOWN:
+                subvol = data;
 
-		LOCK (&conf->subvolume_lock);
-		{
-			conf->subvolume_status[cnt] = 1;
-		}
-		UNLOCK (&conf->subvolume_lock);
+                if (conf->assert_no_child_down) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "Received CHILD_DOWN. Exiting");
+                        if (conf->defrag) {
+                                gf_defrag_stop (conf->defrag, NULL);
+                        } else {
+                                kill (getpid(), SIGTERM);
+                        }
+                }
 
-		break;
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                        if (subvol == conf->subvolumes[i]) {
+                                cnt = i;
+                                break;
+                        }
+                }
 
-	case GF_EVENT_CHILD_DOWN:
-		subvol = data;
+                if (cnt == -1) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "got GF_EVENT_CHILD_DOWN bad subvolume %s",
+                                subvol->name);
+                        break;
+                }
 
-		for (i = 0; i < conf->subvolume_cnt; i++) {
-			if (subvol == conf->subvolumes[i]) {
-				cnt = i;
-				break;
-			}
-		}
+                LOCK (&conf->subvolume_lock);
+                {
+                        conf->subvolume_status[cnt] = 0;
+                        conf->last_event[cnt] = event;
+                        conf->subvol_up_time[cnt] = 0;
+                }
+                UNLOCK (&conf->subvolume_lock);
 
-		if (cnt == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"got GF_EVENT_CHILD_DOWN bad subvolume %s",
-				subvol->name);
-			break;
-		}
+                break;
 
-		LOCK (&conf->subvolume_lock);
-		{
-			conf->subvolume_status[cnt] = 0;
-		}
-		UNLOCK (&conf->subvolume_lock);
+        case GF_EVENT_CHILD_CONNECTING:
+                subvol = data;
 
-		break;
-	}
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                        if (subvol == conf->subvolumes[i]) {
+                                cnt = i;
+                                break;
+                        }
+                }
+
+                if (cnt == -1) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "got GF_EVENT_CHILD_CONNECTING bad subvolume %s",
+                                subvol->name);
+                        break;
+                }
+
+                LOCK (&conf->subvolume_lock);
+                {
+                        conf->last_event[cnt] = event;
+                }
+                UNLOCK (&conf->subvolume_lock);
+
+                break;
+        case GF_EVENT_VOLUME_DEFRAG:
+        {
+                if (!conf->defrag) {
+                        return ret;
+                }
+                defrag = conf->defrag;
+
+                dict = data;
+                va_start (ap, data);
+                output = va_arg (ap, dict_t*);
+
+                ret = dict_get_int32 (dict, "rebalance-command",
+                                      (int32_t*)&cmd);
+                if (ret)
+                        return ret;
+                LOCK (&defrag->lock);
+                {
+                        if (defrag->is_exiting)
+                                goto unlock;
+                        if (cmd == GF_DEFRAG_CMD_STATUS)
+                                gf_defrag_status_get (defrag, output);
+                        else if (cmd == GF_DEFRAG_CMD_STOP)
+                                gf_defrag_stop (defrag, output);
+                }
+unlock:
+                UNLOCK (&defrag->lock);
+                return 0;
+                break;
+        }
+
+        default:
+                propagate = 1;
+                break;
+        }
+
+
+        /* have all subvolumes reported status once by now? */
+        have_heard_from_all = 1;
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (!conf->last_event[i])
+                        have_heard_from_all = 0;
+        }
+
+        /* if all subvols have reported status, no need to hide anything
+           or wait for anything else. Just propagate blindly */
+        if (have_heard_from_all) {
+                propagate = 1;
+
+        }
+
+
+        if (!had_heard_from_all && have_heard_from_all) {
+                /* This is the first event which completes aggregation
+                   of events from all subvolumes. If at least one subvol
+                   had come up, propagate CHILD_UP, but only this time
+                */
+                event = GF_EVENT_CHILD_DOWN;
+
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                        if (conf->last_event[i] == GF_EVENT_CHILD_UP) {
+                                event = GF_EVENT_CHILD_UP;
+                                break;
+                        }
+
+                        if (conf->last_event[i] == GF_EVENT_CHILD_CONNECTING) {
+                                event = GF_EVENT_CHILD_CONNECTING;
+                                /* continue to check other events for CHILD_UP */
+                        }
+                }
+
+                /* rebalance is started with assert_no_child_down. So we do
+                 * not need to handle CHILD_DOWN event here.
+                 */
+                if (conf->defrag) {
+                        ret = gf_thread_create (&conf->defrag->th, NULL,
+						gf_defrag_start, this);
+                        if (ret) {
+                                conf->defrag = NULL;
+                                GF_FREE (conf->defrag);
+                                kill (getpid(), SIGTERM);
+                        }
+                }
+        }
 
-	ret = default_notify (this, event, data);
+        ret = 0;
+        if (propagate)
+                ret = default_notify (this, event, data);
 
-	return ret;
+        return ret;
 }
 
+int
+dht_inode_ctx_layout_get (inode_t *inode, xlator_t *this, dht_layout_t **layout)
+{
+        dht_inode_ctx_t         *ctx            = NULL;
+        int                      ret            = -1;
+
+        ret = dht_inode_ctx_get (inode, this, &ctx);
+
+        if (!ret && ctx) {
+                if (ctx->layout) {
+                        if (layout)
+                                *layout = ctx->layout;
+                        ret = 0;
+                } else {
+                        ret = -1;
+                }
+        }
+
+        return ret;
+}
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index 0eb57a196..5ccd66799 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef _CONFIG_H
@@ -22,200 +13,775 @@
 #include "config.h"
 #endif
 
+#include <regex.h>
+
+#include "dht-mem-types.h"
+#include "libxlator.h"
+#include "syncop.h"
+
 #ifndef _DHT_H
 #define _DHT_H
 
+#define GF_XATTR_FIX_LAYOUT_KEY     "distribute.fix.layout"
+#define GF_DHT_LOOKUP_UNHASHED_ON   1
+#define GF_DHT_LOOKUP_UNHASHED_AUTO 2
+#define DHT_PATHINFO_HEADER         "DISTRIBUTE:"
+
+#include <fnmatch.h>
 
 typedef int (*dht_selfheal_dir_cbk_t) (call_frame_t *frame, void *cookie,
-				       xlator_t *this,
-				       int32_t op_ret, int32_t op_errno);
+                                       xlator_t     *this,
+                                       int32_t       op_ret, int32_t op_errno,
+                                       dict_t *xdata);
+typedef int (*dht_defrag_cbk_fn_t) (xlator_t        *this, call_frame_t *frame,
+                                    int              ret);
 
 
 struct dht_layout {
-        int               cnt;
-	int               preset;
-        int               gen;
-	int               type;
+        int                spread_cnt;  /* layout spread count per directory,
+                                           is controlled by 'setxattr()' with
+                                           special key */
+        int                cnt;
+        int                preset;
+        int                gen;
+        int                type;
+        int                ref; /* use with dht_conf_t->layout_lock */
+        int                search_unhashed;
         struct {
-		int       err;   /* 0 = normal
-				   -1 = dir exists and no xattr
-				   >0 = dir lookup failed with errno
-				 */
-                uint32_t  start;
-                uint32_t  stop;
-                xlator_t *xlator;
-        } list[0];
+                int        err;   /* 0 = normal
+                                     -1 = dir exists and no xattr
+                                     >0 = dir lookup failed with errno
+                                  */
+                uint32_t   start;
+                uint32_t   stop;
+                xlator_t  *xlator;
+        } list[];
+};
+typedef struct dht_layout  dht_layout_t;
+
+struct dht_stat_time {
+        uint32_t        atime;
+        uint32_t        atime_nsec;
+        uint32_t        ctime;
+        uint32_t        ctime_nsec;
+        uint32_t        mtime;
+        uint32_t        mtime_nsec;
+};
+
+typedef struct dht_stat_time dht_stat_time_t;
+
+struct dht_inode_ctx {
+        dht_layout_t    *layout;
+        dht_stat_time_t  time;
 };
-typedef struct dht_layout dht_layout_t;
 
+typedef struct dht_inode_ctx dht_inode_ctx_t;
+
+
+typedef enum {
+        DHT_HASH_TYPE_DM,
+        DHT_HASH_TYPE_DM_USER,
+} dht_hashfn_type_t;
+
+/* rebalance related */
+struct dht_rebalance_ {
+        xlator_t            *from_subvol;
+        xlator_t            *target_node;
+        off_t                offset;
+        size_t               size;
+        int32_t              flags;
+        int                  count;
+        struct iobref       *iobref;
+        struct iovec        *vector;
+        struct iatt          stbuf;
+        dht_defrag_cbk_fn_t  target_op_fn;
+        dict_t              *xdata;
+};
 
 struct dht_local {
-	int                      call_cnt;
-	loc_t                    loc;
-	loc_t                    loc2;
-	int                      op_ret;
-	int                      op_errno;
-	int                      layout_mismatch;
-	struct stat              stbuf;
-	struct statvfs           statvfs;
-	fd_t                    *fd;
-	inode_t                 *inode;
-	dict_t                  *xattr;
-	dict_t                  *xattr_req;
-	dht_layout_t            *layout;
-	size_t                   size;
-	ino_t                    st_ino;
-	xlator_t                *src_hashed, *src_cached;
-	xlator_t                *dst_hashed, *dst_cached;
-	xlator_t                *cached_subvol;
-	xlator_t                *hashed_subvol;
-	char                     need_selfheal;
+        int                      call_cnt;
+        loc_t                    loc;
+        loc_t                    loc2;
+        int                      op_ret;
+        int                      op_errno;
+        int                      layout_mismatch;
+        /* Use stbuf as the postbuf, when we require both
+         * pre and post attrs */
+        struct iatt              stbuf;
+        struct iatt              prebuf;
+        struct iatt              preoldparent;
+        struct iatt              postoldparent;
+        struct iatt              preparent;
+        struct iatt              postparent;
+        struct statvfs           statvfs;
+        fd_t                    *fd;
+        inode_t                 *inode;
+        dict_t                  *params;
+        dict_t                  *xattr;
+        dict_t                  *xattr_req;
+        dht_layout_t            *layout;
+        size_t                   size;
+        ino_t                    ia_ino;
+        xlator_t                *src_hashed, *src_cached;
+        xlator_t                *dst_hashed, *dst_cached;
+        xlator_t                *cached_subvol;
+        xlator_t                *hashed_subvol;
+        char                     need_selfheal;
         int                      file_count;
         int                      dir_count;
-	struct {
-		fop_mknod_cbk_t  linkfile_cbk;
-		struct stat      stbuf;
-		loc_t            loc;
-		inode_t         *inode;
-		dict_t          *xattr;
-		xlator_t        *srcvol;
-	} linkfile;
-	struct {
-		uint32_t         hole_cnt;
-		uint32_t         overlaps_cnt;
-		uint32_t         missing;
-		uint32_t         down;
-		uint32_t         misc;
-		dht_selfheal_dir_cbk_t   dir_cbk;
-		dht_layout_t    *layout;
-	} selfheal;
-
-	/* needed by nufa */
-	int32_t flags;
-	mode_t  mode;
-	dev_t   rdev;
+        call_frame_t            *main_frame;
+        int                      fop_succeeded;
+        struct {
+                fop_mknod_cbk_t  linkfile_cbk;
+                struct iatt      stbuf;
+                loc_t            loc;
+                inode_t         *inode;
+                dict_t          *xattr;
+                xlator_t        *srcvol;
+        } linkfile;
+        struct {
+                uint32_t         hole_cnt;
+                uint32_t         overlaps_cnt;
+                uint32_t         down;
+                uint32_t         misc;
+                dht_selfheal_dir_cbk_t   dir_cbk;
+                dht_layout_t    *layout;
+        } selfheal;
+        uint32_t                 uid;
+        uint32_t                 gid;
+
+        /* needed by nufa */
+        int32_t flags;
+        mode_t  mode;
+        dev_t   rdev;
+        mode_t  umask;
+
+        /* need for file-info */
+        char   *xattr_val;
+        char   *key;
+
+        /* which xattr request? */
+        char xsel[256];
+        int32_t alloc_len;
+
+        char   *newpath;
+
+        /* gfid related */
+        uuid_t  gfid;
+
+        /*Marker Related*/
+        struct marker_str    marker;
+
+        /* flag used to make sure we need to return estale in
+           {lookup,revalidate}_cbk */
+        char return_estale;
+        char need_lookup_everywhere;
+
+        glusterfs_fop_t      fop;
+
+        gf_boolean_t     linked;
+        xlator_t        *link_subvol;
+
+        struct dht_rebalance_ rebalance;
+        xlator_t        *first_up_subvol;
+
 };
 typedef struct dht_local dht_local_t;
 
+/* du - disk-usage */
+struct dht_du {
+        double   avail_percent;
+	double   avail_inodes;
+        uint64_t avail_space;
+        uint32_t log;
+};
+typedef struct dht_du dht_du_t;
+
+enum gf_defrag_type {
+        GF_DEFRAG_CMD_START = 1,
+        GF_DEFRAG_CMD_STOP = 1 + 1,
+        GF_DEFRAG_CMD_STATUS = 1 + 2,
+        GF_DEFRAG_CMD_START_LAYOUT_FIX = 1 + 3,
+        GF_DEFRAG_CMD_START_FORCE = 1 + 4,
+};
+typedef enum gf_defrag_type gf_defrag_type;
+
+enum gf_defrag_status_t {
+        GF_DEFRAG_STATUS_NOT_STARTED,
+        GF_DEFRAG_STATUS_STARTED,
+        GF_DEFRAG_STATUS_STOPPED,
+        GF_DEFRAG_STATUS_COMPLETE,
+        GF_DEFRAG_STATUS_FAILED,
+        GF_DEFRAG_STATUS_LAYOUT_FIX_STARTED,
+        GF_DEFRAG_STATUS_LAYOUT_FIX_STOPPED,
+        GF_DEFRAG_STATUS_LAYOUT_FIX_COMPLETE,
+        GF_DEFRAG_STATUS_LAYOUT_FIX_FAILED,
+};
+typedef enum gf_defrag_status_t gf_defrag_status_t;
+
+typedef struct gf_defrag_pattern_list gf_defrag_pattern_list_t;
+
+struct gf_defrag_pattern_list {
+        char                       path_pattern[256];
+        uint64_t                   size;
+        gf_defrag_pattern_list_t  *next;
+};
+
+struct gf_defrag_info_ {
+        uint64_t                     total_files;
+        uint64_t                     total_data;
+        uint64_t                     num_files_lookedup;
+        uint64_t                     total_failures;
+        uint64_t                     skipped;
+        gf_lock_t                    lock;
+        int                          cmd;
+        pthread_t                    th;
+        gf_defrag_status_t           defrag_status;
+        struct rpc_clnt             *rpc;
+        uint32_t                     connected;
+        uint32_t                     is_exiting;
+        pid_t                        pid;
+        inode_t                     *root_inode;
+        uuid_t                       node_uuid;
+        struct timeval               start_time;
+        gf_boolean_t                 stats;
+        gf_defrag_pattern_list_t    *defrag_pattern;
+};
+
+typedef struct gf_defrag_info_ gf_defrag_info_t;
 
 struct dht_conf {
-	gf_lock_t      subvolume_lock;
+        gf_lock_t      subvolume_lock;
         int            subvolume_cnt;
         xlator_t     **subvolumes;
-	xlator_t      *local_volume;     /* Needed by NUFA */
-	char          *subvolume_status;
-	dht_layout_t **file_layouts;
-	dht_layout_t **dir_layouts;
-	dht_layout_t  *default_dir_layout;
-	gf_boolean_t   search_unhashed;
-	int            gen;
+        char          *subvolume_status;
+        int           *last_event;
+        dht_layout_t **file_layouts;
+        dht_layout_t **dir_layouts;
+        gf_boolean_t   search_unhashed;
+        int            gen;
+        dht_du_t      *du_stats;
+        double         min_free_disk;
+	double         min_free_inodes;
+        char           disk_unit;
+        int32_t        refresh_interval;
+        gf_boolean_t   unhashed_sticky_bit;
+        struct timeval last_stat_fetch;
+        gf_lock_t      layout_lock;
+        void          *private;     /* Can be used by wrapper xlators over
+                                       dht */
+        gf_boolean_t   use_readdirp;
+        char           vol_uuid[UUID_SIZE + 1];
+        gf_boolean_t   assert_no_child_down;
+        time_t        *subvol_up_time;
+
+        /* This is the count used as the distribute layout for a directory */
+        /* Will be a global flag to control the layout spread count */
+        uint32_t       dir_spread_cnt;
+
+        /* to keep track of nodes which are decomissioned */
+        xlator_t     **decommissioned_bricks;
+        int            decommission_in_progress;
+        int            decommission_subvols_cnt;
+
+        /* defrag related */
+        gf_defrag_info_t *defrag;
+
+        /* Request to filter directory entries in readdir request */
+
+        gf_boolean_t    readdir_optimize;
+
+        /* Support regex-based name reinterpretation. */
+        regex_t         rsync_regex;
+        gf_boolean_t    rsync_regex_valid;
+        regex_t         extra_regex;
+        gf_boolean_t    extra_regex_valid;
+
+        /* Support variable xattr names. */
+        char            *xattr_name;
+        char            *link_xattr_name;
+        char            *wild_xattr_name;
 };
 typedef struct dht_conf dht_conf_t;
 
 
 struct dht_disk_layout {
-	uint32_t           cnt;
-	uint32_t           type;
-	struct {
-		uint32_t   start;
-		uint32_t   stop;
-	} list[1];
+        uint32_t           cnt;
+        uint32_t           type;
+        struct {
+                uint32_t   start;
+                uint32_t   stop;
+        } list[1];
 };
 typedef struct dht_disk_layout dht_disk_layout_t;
- 
-#define ENTRY_MISSING(op_ret, op_errno) (op_ret == -1 && op_errno == ENOENT)
 
-#define is_fs_root(loc) (strcmp (loc->path, "/") == 0)
+typedef enum {
+        GF_DHT_MIGRATE_DATA,
+        GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS,
+        GF_DHT_MIGRATE_HARDLINK,
+        GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS
+} gf_dht_migrate_data_type_t;
+
+#define ENTRY_MISSING(op_ret, op_errno) (op_ret == -1 && op_errno == ENOENT)
 
-#define is_revalidate(loc) (inode_ctx_get (loc->inode, this, NULL) == 0)
+#define is_revalidate(loc) (dht_inode_ctx_layout_get (loc->inode, this, NULL) == 0)
 
 #define is_last_call(cnt) (cnt == 0)
 
-#define DHT_LINKFILE_MODE (S_ISVTX)
-#define check_is_linkfile(i,s,x) ((s->st_mode & ~S_IFMT) == DHT_LINKFILE_MODE)
+#define DHT_MIGRATION_IN_PROGRESS 1
+#define DHT_MIGRATION_COMPLETED   2
+
+#define DHT_LINKFILE_MODE        (S_ISVTX)
+
+#define check_is_linkfile(i,s,x,n) (                                      \
+                ((st_mode_from_ia ((s)->ia_prot, (s)->ia_type) & ~S_IFMT) \
+                 == DHT_LINKFILE_MODE) &&                                 \
+                dict_get (x, n))
+
+#define IS_DHT_MIGRATION_PHASE2(buf)  (                                 \
+                IA_ISREG ((buf)->ia_type) &&                            \
+                ((st_mode_from_ia ((buf)->ia_prot, (buf)->ia_type) &    \
+                  ~S_IFMT) == DHT_LINKFILE_MODE))
+
+#define IS_DHT_MIGRATION_PHASE1(buf)  (                                 \
+                IA_ISREG ((buf)->ia_type) &&                            \
+                ((buf)->ia_prot.sticky == 1) &&                         \
+                ((buf)->ia_prot.sgid == 1))
+
+#define DHT_STRIP_PHASE1_FLAGS(buf)  do {                       \
+                if ((buf) && IS_DHT_MIGRATION_PHASE1(buf)) {    \
+                        (buf)->ia_prot.sticky = 0;              \
+                        (buf)->ia_prot.sgid = 0;                \
+                }                                               \
+        } while (0)
 
-#define check_is_dir(i,s,x) (S_ISDIR(s->st_mode))
+#define check_is_dir(i,s,x) (IA_ISDIR(s->ia_type))
 
 #define layout_is_sane(layout) ((layout) && (layout->cnt > 0))
 
-#define DHT_STACK_UNWIND(frame, params ...) do {       \
-		dht_local_t *__local = NULL;           \
-		__local = frame->local;                \
-		frame->local = NULL;		       \
-		STACK_UNWIND (frame, params);          \
-		dht_local_wipe (__local);	       \
-	} while (0)
-
-#define DHT_STACK_DESTROY(frame) do {		       \
-		dht_local_t *__local = NULL;           \
-		__local = frame->local;                \
-		frame->local = NULL;		       \
-		STACK_DESTROY (frame->root);	       \
-		dht_local_wipe (__local);	       \
-	} while (0)
-
-dht_layout_t *dht_layout_new (xlator_t *this, int cnt);
-dht_layout_t *dht_layout_get (xlator_t *this, inode_t *inode);
-dht_layout_t *dht_layout_for_subvol (xlator_t *this, xlator_t *subvol);
-xlator_t *dht_layout_search (xlator_t *this, dht_layout_t *layout,
-			     const char *name);
-int dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout);
-int dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
-			  uint32_t *holes_p, uint32_t *overlaps_p,
-			  uint32_t *missing_p, uint32_t *down_p,
-			  uint32_t *misc_p);
-int dht_layout_dir_mismatch (xlator_t *this, dht_layout_t *layout,
-			     xlator_t *subvol, loc_t *loc, dict_t *xattr);
+#define DHT_STACK_UNWIND(fop, frame, params ...) do {           \
+                dht_local_t *__local = NULL;                    \
+                xlator_t    *__xl    = NULL;                    \
+                if (frame) {                                    \
+                        __xl         = frame->this;             \
+                        __local      = frame->local;            \
+                        frame->local = NULL;                    \
+                }                                               \
+                STACK_UNWIND_STRICT (fop, frame, params);       \
+                dht_local_wipe (__xl, __local);                 \
+        } while (0)
+
+#define DHT_STACK_DESTROY(frame) do {                   \
+                dht_local_t *__local = NULL;            \
+                xlator_t    *__xl    = NULL;            \
+                __xl                 = frame->this;     \
+                __local              = frame->local;    \
+                frame->local         = NULL;            \
+                STACK_DESTROY (frame->root);            \
+                dht_local_wipe (__xl, __local);         \
+        } while (0)
+
+#define DHT_UPDATE_TIME(ctx_sec, ctx_nsec, new_sec, new_nsec, inode, post) do {\
+                int32_t sec = 0;                                        \
+                sec = new_sec;                                          \
+                LOCK (&inode->lock);                                    \
+                {                                                       \
+                        new_sec = max(new_sec, ctx_sec);                \
+                        if (sec < new_sec)                              \
+                                new_nsec = ctx_nsec;                    \
+                        if (sec == new_sec)                             \
+                                new_nsec = max (new_nsec, ctx_nsec);    \
+                        if (post) {                                     \
+                                ctx_sec = new_sec;                      \
+                                ctx_nsec = new_nsec;                    \
+                        }                                               \
+                }                                                       \
+                UNLOCK (&inode->lock);                                  \
+        } while (0)
+
+#define is_greater_time(a, an, b, bn) (((a) < (b)) || (((a) == (b)) && ((an) < (bn))))
+dht_layout_t                            *dht_layout_new (xlator_t *this, int cnt);
+dht_layout_t                            *dht_layout_get (xlator_t *this, inode_t *inode);
+dht_layout_t                            *dht_layout_for_subvol (xlator_t *this, xlator_t *subvol);
+xlator_t *dht_layout_search (xlator_t   *this, dht_layout_t *layout,
+                             const char *name);
+int                                      dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout);
+int dht_layout_anomalies (xlator_t      *this, loc_t *loc, dht_layout_t *layout,
+                          uint32_t      *holes_p, uint32_t *overlaps_p,
+                          uint32_t      *missing_p, uint32_t *down_p,
+                          uint32_t      *misc_p, uint32_t *no_space_p);
+int dht_layout_dir_mismatch (xlator_t   *this, dht_layout_t *layout,
+                             xlator_t   *subvol, loc_t *loc, dict_t *xattr);
 
 xlator_t *dht_linkfile_subvol (xlator_t *this, inode_t *inode,
-			       struct stat *buf, dict_t *xattr);
-int dht_linkfile_unlink (call_frame_t *frame, xlator_t *this,
-			 xlator_t *subvol, loc_t *loc);
+                               struct iatt *buf, dict_t *xattr);
+int dht_linkfile_unlink (call_frame_t      *frame, xlator_t *this,
+                         xlator_t          *subvol, loc_t *loc);
 
 int dht_layouts_init (xlator_t *this, dht_conf_t *conf);
 int dht_layout_merge (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
-		      int op_ret, int op_errno, dict_t *xattr);
+                      int       op_ret, int op_errno, dict_t *xattr);
 
 int dht_disk_layout_extract (xlator_t *this, dht_layout_t *layout,
-			     int pos, int32_t **disk_layout_p);
-int dht_disk_layout_merge (xlator_t *this, dht_layout_t *layout,
-			   int pos, int32_t *disk_layout);
+                             int       pos, int32_t **disk_layout_p);
+int dht_disk_layout_merge (xlator_t   *this, dht_layout_t *layout,
+                           int         pos, void *disk_layout_raw, int disk_layout_len);
 
 
 int dht_frame_return (call_frame_t *frame);
 
-int dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y);
+int                             dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y);
 int dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol,
-		      uint64_t *x);
+                      uint64_t *x);
 
-void dht_local_wipe (dht_local_t *local);
-dht_local_t *dht_local_init (call_frame_t *frame);
-int dht_stat_merge (xlator_t *this, struct stat *to, struct stat *from,
-		    xlator_t *subvol);
+void dht_local_wipe (xlator_t *this, dht_local_t *local);
+dht_local_t *dht_local_init (call_frame_t    *frame, loc_t *loc, fd_t *fd,
+                             glusterfs_fop_t  fop);
+int dht_iatt_merge (xlator_t                 *this, struct iatt *to, struct iatt *from,
+                    xlator_t                 *subvol);
 
 xlator_t *dht_subvol_get_hashed (xlator_t *this, loc_t *loc);
 xlator_t *dht_subvol_get_cached (xlator_t *this, inode_t *inode);
 xlator_t *dht_subvol_next (xlator_t *this, xlator_t *prev);
-int dht_subvol_cnt (xlator_t *this, xlator_t *subvol);
+xlator_t *dht_subvol_next_available (xlator_t *this, xlator_t *prev);
+int       dht_subvol_cnt (xlator_t *this, xlator_t *subvol);
 
-int dht_hash_compute (int type, const char *name, uint32_t *hash_p);
+int dht_hash_compute (xlator_t *this, int type, const char *name, uint32_t *hash_p);
 
-int dht_linkfile_create (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
-			 xlator_t *tovol, xlator_t *fromvol, loc_t *loc);
-int dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc);
-int dht_lookup_everywhere (call_frame_t *frame, xlator_t *this, loc_t *loc);
+int dht_linkfile_create (call_frame_t    *frame, fop_mknod_cbk_t linkfile_cbk,
+                         xlator_t        *this, xlator_t *tovol,
+                         xlator_t        *fromvol, loc_t *loc);
+int                                       dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc);
+int                                       dht_lookup_everywhere (call_frame_t *frame, xlator_t *this, loc_t *loc);
 int
-dht_selfheal_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
-			loc_t *loc, dht_layout_t *layout);
+dht_selfheal_directory (call_frame_t     *frame, dht_selfheal_dir_cbk_t cbk,
+                        loc_t            *loc, dht_layout_t *layout);
 int
 dht_selfheal_new_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
-			    dht_layout_t *layout);
+                            dht_layout_t *layout);
 int
-dht_selfheal_restore (call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
-		      loc_t *loc, dht_layout_t *layout);
+dht_selfheal_restore (call_frame_t       *frame, dht_selfheal_dir_cbk_t cbk,
+                      loc_t              *loc, dht_layout_t *layout);
 int
 dht_layout_sort_volname (dht_layout_t *layout);
 
-int dht_rename (call_frame_t *frame, xlator_t *this,
-		loc_t *oldloc, loc_t *newloc);
-#endif /* _DHT_H */
+int dht_get_du_info (call_frame_t *frame, xlator_t *this, loc_t *loc);
+
+gf_boolean_t dht_is_subvol_filled (xlator_t *this, xlator_t *subvol);
+xlator_t *dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol,
+                                          dht_local_t *layout);
+int       dht_get_du_info_for_subvol (xlator_t *this, int subvol_idx);
+
+int dht_layout_preset (xlator_t *this, xlator_t *subvol, inode_t *inode);
+int           dht_layout_set (xlator_t *this, inode_t *inode, dht_layout_t *layout);;
+void          dht_layout_unref (xlator_t *this, dht_layout_t *layout);
+dht_layout_t *dht_layout_ref (xlator_t *this, dht_layout_t *layout);
+xlator_t     *dht_first_up_subvol (xlator_t *this);
+xlator_t     *dht_last_up_subvol (xlator_t *this);
+
+int dht_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name);
+
+int dht_filter_loc_subvol_key (xlator_t  *this, loc_t *loc, loc_t *new_loc,
+                               xlator_t **subvol);
+
+int                                     dht_rename_cleanup (call_frame_t *frame);
+int dht_rename_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t           op_ret, int32_t op_errno,
+                          inode_t          *inode, struct iatt *stbuf,
+                          struct iatt      *preparent, struct iatt *postparent,
+                          dict_t *xdata);
+
+int dht_fix_directory_layout (call_frame_t *frame,
+                              dht_selfheal_dir_cbk_t  dir_cbk,
+                              dht_layout_t           *layout);
+
+int dht_init_subvolumes (xlator_t *this, dht_conf_t *conf);
+
+/* migration/rebalance */
+int dht_start_rebalance_task (xlator_t *this, call_frame_t *frame);
+
+int dht_rebalance_in_progress_check (xlator_t *this, call_frame_t *frame);
+int dht_rebalance_complete_check (xlator_t *this, call_frame_t *frame);
+
+
+/* FOPS */
+int32_t dht_lookup (call_frame_t *frame,
+                    xlator_t *this,
+                    loc_t    *loc,
+                    dict_t   *xattr_req);
+
+int32_t dht_stat (call_frame_t *frame,
+                  xlator_t *this,
+                  loc_t    *loc, dict_t *xdata);
+
+int32_t dht_fstat (call_frame_t *frame,
+                   xlator_t *this,
+                   fd_t     *fd, dict_t *xdata);
+
+int32_t dht_truncate (call_frame_t *frame,
+                      xlator_t *this,
+                      loc_t    *loc,
+                      off_t     offset, dict_t *xdata);
+
+int32_t dht_ftruncate (call_frame_t *frame,
+                       xlator_t *this,
+                       fd_t     *fd,
+                       off_t     offset, dict_t *xdata);
+
+int32_t dht_access (call_frame_t *frame,
+                    xlator_t *this,
+                    loc_t    *loc,
+                    int32_t   mask, dict_t *xdata);
+
+int32_t dht_readlink (call_frame_t *frame,
+                      xlator_t *this,
+                      loc_t    *loc,
+                      size_t    size, dict_t *xdata);
+
+int32_t dht_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata);
+
+int32_t dht_mkdir (call_frame_t *frame, xlator_t *this,
+                   loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata);
+
+int32_t dht_unlink (call_frame_t *frame,
+                    xlator_t *this,
+                    loc_t    *loc, int xflag, dict_t *xdata);
+
+int32_t dht_rmdir (call_frame_t *frame, xlator_t *this,
+                   loc_t    *loc, int flags, dict_t *xdata);
+
+int32_t dht_symlink (call_frame_t   *frame, xlator_t *this,
+                     const char *linkpath, loc_t *loc, mode_t umask,
+                     dict_t *xdata);
+
+int32_t dht_rename (call_frame_t *frame,
+                    xlator_t *this,
+                    loc_t    *oldloc,
+                    loc_t    *newloc, dict_t *xdata);
+
+int32_t dht_link (call_frame_t *frame,
+                  xlator_t *this,
+                  loc_t    *oldloc,
+                  loc_t    *newloc, dict_t *xdata);
+
+int32_t dht_create (call_frame_t *frame, xlator_t *this,
+                    loc_t    *loc, int32_t flags, mode_t mode,
+                    mode_t umask, fd_t     *fd, dict_t *params);
+
+int32_t dht_open (call_frame_t *frame,
+                  xlator_t *this,
+                  loc_t    *loc,
+                  int32_t   flags, fd_t *fd, dict_t *xdata);
+
+int32_t dht_readv (call_frame_t *frame,
+                   xlator_t *this,
+                   fd_t     *fd,
+                   size_t    size,
+                   off_t     offset, uint32_t flags, dict_t *xdata);
+
+int32_t dht_writev (call_frame_t      *frame,
+                    xlator_t      *this,
+                    fd_t          *fd,
+                    struct iovec  *vector,
+                    int32_t        count,
+                    off_t          offset,
+                    uint32_t       flags,
+                    struct iobref *iobref, dict_t *xdata);
+
+int32_t dht_flush (call_frame_t *frame,
+                   xlator_t *this,
+                   fd_t     *fd, dict_t *xdata);
+
+int32_t dht_fsync (call_frame_t *frame,
+                   xlator_t *this,
+                   fd_t     *fd,
+                   int32_t   datasync, dict_t *xdata);
+
+int32_t dht_opendir (call_frame_t *frame,
+                     xlator_t *this,
+                     loc_t    *loc, fd_t *fd, dict_t *xdata);
+
+int32_t dht_fsyncdir (call_frame_t *frame,
+                      xlator_t *this,
+                      fd_t     *fd,
+                      int32_t   datasync, dict_t *xdata);
+
+int32_t dht_statfs (call_frame_t *frame,
+                    xlator_t *this,
+                    loc_t    *loc, dict_t *xdata);
+
+int32_t dht_setxattr (call_frame_t *frame,
+                      xlator_t *this,
+                      loc_t    *loc,
+                      dict_t   *dict,
+                      int32_t   flags, dict_t *xdata);
+
+int32_t dht_getxattr (call_frame_t   *frame,
+                      xlator_t   *this,
+                      loc_t      *loc,
+                      const char *name, dict_t *xdata);
+
+int32_t dht_fsetxattr (call_frame_t *frame,
+                       xlator_t *this,
+                       fd_t     *fd,
+                       dict_t   *dict,
+                       int32_t   flags, dict_t *xdata);
+
+int32_t dht_fgetxattr (call_frame_t   *frame,
+                       xlator_t   *this,
+                       fd_t       *fd,
+                       const char *name, dict_t *xdata);
+
+int32_t dht_removexattr (call_frame_t   *frame,
+                         xlator_t   *this,
+                         loc_t      *loc,
+                         const char *name, dict_t *xdata);
+int32_t dht_fremovexattr (call_frame_t   *frame,
+                          xlator_t   *this,
+                          fd_t      *fd,
+                          const char *name, dict_t *xdata);
+
+int32_t dht_lk (call_frame_t        *frame,
+                xlator_t        *this,
+                fd_t            *fd,
+                int32_t          cmd,
+                struct gf_flock *flock, dict_t *xdata);
+
+int32_t dht_inodelk (call_frame_t *frame, xlator_t *this,
+                     const char      *volume, loc_t *loc, int32_t cmd,
+                     struct gf_flock *flock, dict_t *xdata);
+
+int32_t dht_finodelk (call_frame_t        *frame, xlator_t *this,
+                      const char      *volume, fd_t *fd, int32_t cmd,
+                      struct gf_flock *flock, dict_t *xdata);
+
+int32_t dht_entrylk (call_frame_t    *frame, xlator_t *this,
+                     const char  *volume, loc_t *loc, const char *basename,
+                     entrylk_cmd  cmd, entrylk_type type, dict_t *xdata);
+
+int32_t dht_fentrylk (call_frame_t    *frame, xlator_t *this,
+                      const char  *volume, fd_t *fd, const char *basename,
+                      entrylk_cmd  cmd, entrylk_type type, dict_t *xdata);
+
+int32_t dht_readdir (call_frame_t  *frame,
+                     xlator_t *this,
+                     fd_t     *fd,
+                     size_t    size, off_t off, dict_t *xdata);
+
+int32_t dht_readdirp (call_frame_t *frame,
+                      xlator_t *this,
+                      fd_t     *fd,
+                      size_t    size, off_t off, dict_t *dict);
+
+int32_t dht_xattrop (call_frame_t           *frame,
+                     xlator_t           *this,
+                     loc_t              *loc,
+                     gf_xattrop_flags_t  flags,
+                     dict_t             *dict, dict_t *xdata);
+
+int32_t dht_fxattrop (call_frame_t *frame,
+                      xlator_t           *this,
+                      fd_t               *fd,
+                      gf_xattrop_flags_t  flags,
+                      dict_t             *dict, dict_t *xdata);
+
+int32_t dht_forget (xlator_t *this, inode_t *inode);
+int32_t dht_setattr (call_frame_t  *frame, xlator_t *this, loc_t *loc,
+                     struct iatt   *stbuf, int32_t valid, dict_t *xdata);
+int32_t dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      struct iatt  *stbuf, int32_t valid, dict_t *xdata);
+int32_t dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd,
+		      int32_t mode, off_t offset, size_t len, dict_t *xdata);
+int32_t dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd,
+		    off_t offset, size_t len, dict_t *xdata);
+int32_t dht_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                    off_t offset, size_t len, dict_t *xdata);
+
+int32_t dht_init (xlator_t *this);
+void    dht_fini (xlator_t *this);
+int     dht_reconfigure (xlator_t *this, dict_t *options);
+int32_t dht_notify (xlator_t *this, int32_t event, void *data, ...);
+
+/* definitions for nufa/switch */
+int dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent);
+int dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent);
+int dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
+                             xlator_t *this, int op_ret, int op_errno,
+                             inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                             struct iatt *postparent);
+int dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno,
+                    inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+                    struct iatt *postparent);
+int dht_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno,
+                    fd_t *fd, inode_t *inode, struct iatt *stbuf,
+                    struct iatt *preparent, struct iatt *postparent,
+                    dict_t *xdata);
+int dht_newfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno,
+                     inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata);
+
+int
+gf_defrag_status_get (gf_defrag_info_t *defrag, dict_t *dict);
+
+int
+gf_defrag_stop (gf_defrag_info_t *defrag, dict_t *output);
+
+void*
+gf_defrag_start (void *this);
+
+int32_t
+gf_defrag_handle_hardlink (xlator_t *this, loc_t *loc, dict_t  *xattrs,
+                           struct iatt *stbuf);
+int
+dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
+                 int flag);
+int
+dht_inode_ctx_layout_get (inode_t *inode, xlator_t *this,
+                          dht_layout_t **layout_int);
+int
+dht_inode_ctx_layout_set (inode_t *inode, xlator_t *this,
+                          dht_layout_t* layout_int);
+int
+dht_inode_ctx_time_update (inode_t *inode, xlator_t *this, struct iatt *stat,
+                           int32_t update_ctx);
+
+int dht_inode_ctx_get (inode_t *inode, xlator_t *this, dht_inode_ctx_t **ctx);
+int dht_inode_ctx_set (inode_t *inode, xlator_t *this, dht_inode_ctx_t *ctx);
+int
+dht_dir_attr_heal (void *data);
+int
+dht_dir_attr_heal_done (int ret, call_frame_t *sync_frame, void *data);
+int
+dht_dir_has_layout (dict_t *xattr, char *name);
+gf_boolean_t
+dht_is_subvol_in_layout (dht_layout_t *layout, xlator_t *xlator);
+xlator_t *
+dht_subvol_with_free_space_inodes (xlator_t *this, xlator_t *subvol,
+                                   dht_layout_t *layout);
+xlator_t *
+dht_subvol_maxspace_nonzeroinode (xlator_t *this, xlator_t *subvol,
+                                  dht_layout_t *layout);
+int
+dht_linkfile_attr_heal (call_frame_t *frame, xlator_t *this);
+
+void
+dht_layout_dump (dht_layout_t  *layout, const char *prefix);
+int32_t
+dht_priv_dump (xlator_t *this);
+int32_t
+dht_inodectx_dump (xlator_t *this, inode_t *inode);
+
+int
+dht_inode_ctx_get1 (xlator_t *this, inode_t *inode, xlator_t **subvol);
+
+#endif/* _DHT_H */
diff --git a/xlators/cluster/dht/src/dht-diskusage.c b/xlators/cluster/dht/src/dht-diskusage.c
new file mode 100644
index 000000000..fe3955ecb
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-diskusage.c
@@ -0,0 +1,414 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+/* TODO: add NS locking */
+
+#include "glusterfs.h"
+#include "xlator.h"
+#include "dht-common.h"
+#include "defaults.h"
+
+#include <sys/time.h>
+
+
+int
+dht_du_info_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+		 int op_ret, int op_errno, struct statvfs *statvfs,
+                 dict_t *xdata)
+{
+	dht_conf_t    *conf         = NULL;
+	call_frame_t  *prev          = NULL;
+	int            this_call_cnt = 0;
+	int            i = 0;
+	double         percent = 0;
+	double         percent_inodes = 0;
+	uint64_t       bytes = 0;
+
+	conf = this->private;
+	prev = cookie;
+
+	if (op_ret == -1) {
+		gf_log (this->name, GF_LOG_WARNING,
+			"failed to get disk info from %s", prev->this->name);
+		goto out;
+	}
+
+	if (statvfs && statvfs->f_blocks) {
+		percent = (statvfs->f_bavail * 100) / statvfs->f_blocks;
+		bytes = (statvfs->f_bavail * statvfs->f_frsize);
+	}
+
+	if (statvfs && statvfs->f_files) {
+		percent_inodes = (statvfs->f_ffree * 100) / statvfs->f_files;
+	} else {
+		/* set percent inodes to 100 for dynamically allocated inode filesystems
+		   this logic holds good so that, distribute has nothing to worry about
+		   total inodes rather let the 'create()' to be scheduled on the hashed
+		   subvol regardless of the total inodes. since we have no awareness on
+		   loosing inodes this logic fits well
+		*/
+		percent_inodes = 100;
+	}
+
+	LOCK (&conf->subvolume_lock);
+	{
+		for (i = 0; i < conf->subvolume_cnt; i++)
+			if (prev->this == conf->subvolumes[i]) {
+				conf->du_stats[i].avail_percent = percent;
+				conf->du_stats[i].avail_space   = bytes;
+				conf->du_stats[i].avail_inodes  = percent_inodes;
+				gf_log (this->name, GF_LOG_DEBUG,
+					"on subvolume '%s': avail_percent is: "
+					"%.2f and avail_space is: %"PRIu64" "
+					"and avail_inodes is: %.2f",
+					prev->this->name,
+					conf->du_stats[i].avail_percent,
+					conf->du_stats[i].avail_space,
+					conf->du_stats[i].avail_inodes);
+			}
+	}
+	UNLOCK (&conf->subvolume_lock);
+
+out:
+	this_call_cnt = dht_frame_return (frame);
+	if (is_last_call (this_call_cnt))
+		DHT_STACK_DESTROY (frame);
+
+	return 0;
+}
+
+int
+dht_get_du_info_for_subvol (xlator_t *this, int subvol_idx)
+{
+	dht_conf_t    *conf         = NULL;
+	call_frame_t  *statfs_frame = NULL;
+	dht_local_t   *statfs_local = NULL;
+	call_pool_t   *pool         = NULL;
+	loc_t          tmp_loc      = {0,};
+
+	conf = this->private;
+	pool = this->ctx->pool;
+
+	statfs_frame = create_frame (this, pool);
+	if (!statfs_frame) {
+		goto err;
+	}
+
+	/* local->fop value is not used in this case */
+	statfs_local = dht_local_init (statfs_frame, NULL, NULL,
+				       GF_FOP_MAXVALUE);
+	if (!statfs_local) {
+		goto err;
+	}
+
+        /* make it root gfid, should be enough to get the proper info back */
+        tmp_loc.gfid[15] = 1;
+
+	statfs_local->call_cnt = 1;
+	STACK_WIND (statfs_frame, dht_du_info_cbk,
+		    conf->subvolumes[subvol_idx],
+		    conf->subvolumes[subvol_idx]->fops->statfs,
+		    &tmp_loc, NULL);
+
+	return 0;
+err:
+	if (statfs_frame)
+		DHT_STACK_DESTROY (statfs_frame);
+
+	return -1;
+}
+
+int
+dht_get_du_info (call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+	int            i            = 0;
+	dht_conf_t    *conf         = NULL;
+	call_frame_t  *statfs_frame = NULL;
+	dht_local_t   *statfs_local = NULL;
+	struct timeval tv           = {0,};
+        loc_t          tmp_loc      = {0,};
+
+	conf  = this->private;
+
+	gettimeofday (&tv, NULL);
+
+        /* make it root gfid, should be enough to get the proper
+           info back */
+        tmp_loc.gfid[15] = 1;
+
+	if (tv.tv_sec > (conf->refresh_interval
+			 + conf->last_stat_fetch.tv_sec)) {
+
+		statfs_frame = copy_frame (frame);
+		if (!statfs_frame) {
+			goto err;
+		}
+
+		/* In this case, 'local->fop' is not used */
+		statfs_local = dht_local_init (statfs_frame, loc, NULL,
+					       GF_FOP_MAXVALUE);
+		if (!statfs_local) {
+			goto err;
+		}
+
+		statfs_local->call_cnt = conf->subvolume_cnt;
+		for (i = 0; i < conf->subvolume_cnt; i++) {
+			STACK_WIND (statfs_frame, dht_du_info_cbk,
+				    conf->subvolumes[i],
+				    conf->subvolumes[i]->fops->statfs,
+				    &tmp_loc, NULL);
+		}
+
+		conf->last_stat_fetch.tv_sec = tv.tv_sec;
+	}
+	return 0;
+err:
+	if (statfs_frame)
+		DHT_STACK_DESTROY (statfs_frame);
+
+	return -1;
+}
+
+
+gf_boolean_t
+dht_is_subvol_filled (xlator_t *this, xlator_t *subvol)
+{
+	int         i = 0;
+	dht_conf_t *conf = NULL;
+	gf_boolean_t subvol_filled_inodes = _gf_false;
+	gf_boolean_t subvol_filled_space = _gf_false;
+	gf_boolean_t is_subvol_filled = _gf_false;
+
+	conf = this->private;
+
+	/* Check for values above specified percent or free disk */
+	LOCK (&conf->subvolume_lock);
+	{
+		for (i = 0; i < conf->subvolume_cnt; i++) {
+			if (subvol == conf->subvolumes[i]) {
+				if (conf->disk_unit == 'p') {
+					if (conf->du_stats[i].avail_percent <
+					    conf->min_free_disk) {
+						subvol_filled_space = _gf_true;
+						break;
+					}
+
+				} else {
+					if (conf->du_stats[i].avail_space <
+					    conf->min_free_disk) {
+						subvol_filled_space = _gf_true;
+						break;
+					}
+				}
+				if (conf->du_stats[i].avail_inodes <
+				    conf->min_free_inodes) {
+					subvol_filled_inodes = _gf_true;
+					break;
+				}
+			}
+		}
+	}
+	UNLOCK (&conf->subvolume_lock);
+
+	if (subvol_filled_space && conf->subvolume_status[i]) {
+		if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
+			gf_log (this->name, GF_LOG_WARNING,
+				"disk space on subvolume '%s' is getting "
+				"full (%.2f %%), consider adding more nodes",
+				subvol->name,
+				(100 - conf->du_stats[i].avail_percent));
+		}
+	}
+
+	if (subvol_filled_inodes && conf->subvolume_status[i]) {
+		if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
+			gf_log (this->name, GF_LOG_CRITICAL,
+				"inodes on subvolume '%s' are at "
+				"(%.2f %%), consider adding more nodes",
+				subvol->name,
+				(100 - conf->du_stats[i].avail_inodes));
+		}
+	}
+
+	is_subvol_filled = (subvol_filled_space || subvol_filled_inodes);
+
+	return is_subvol_filled;
+}
+
+
+/*Get the best subvolume to create the file in*/
+xlator_t *
+dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol,
+                                dht_local_t *local)
+{
+	xlator_t   *avail_subvol = NULL;
+	dht_conf_t *conf = NULL;
+        dht_layout_t *layout = NULL;
+        loc_t      *loc = NULL;
+
+	conf = this->private;
+        if (!local)
+                goto out;
+        loc = &local->loc;
+        if (!local->layout) {
+                layout = dht_layout_get (this, loc->parent);
+
+                if (!layout) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "layout missing path=%s parent=%s",
+                                loc->path, uuid_utoa (loc->parent->gfid));
+                        goto out;
+                }
+        } else {
+                layout = dht_layout_ref (this, local->layout);
+        }
+
+        LOCK (&conf->subvolume_lock);
+	{
+                avail_subvol = dht_subvol_with_free_space_inodes(this, subvol,
+                                                                 layout);
+                if(!avail_subvol)
+                {
+                        avail_subvol = dht_subvol_maxspace_nonzeroinode(this,
+                                                                        subvol,
+                                                                        layout);
+                }
+
+	}
+	UNLOCK (&conf->subvolume_lock);
+out:
+	if (!avail_subvol) {
+		gf_log (this->name,
+                        GF_LOG_DEBUG,
+			"no subvolume has enough free space and/or inodes\
+                         to create");
+                avail_subvol = subvol;
+	}
+
+        if (layout)
+                dht_layout_unref (this, layout);
+	return avail_subvol;
+}
+
+static inline
+int32_t dht_subvol_has_err (xlator_t *this, dht_layout_t *layout)
+{
+        int ret = -1;
+        int i   = 0;
+
+        if (!this || !layout)
+                goto out;
+
+        /* check if subvol has layout errors, before selecting it */
+        for (i = 0; i < layout->cnt; i++) {
+                if (!strcmp (layout->list[i].xlator->name, this->name) &&
+                     (layout->list[i].err != 0)) {
+                        ret = -1;
+                        goto out;
+                }
+        }
+        ret = 0;
+out:
+        return ret;
+}
+
+/*Get subvolume which has both space and inodes more than the min criteria*/
+xlator_t *
+dht_subvol_with_free_space_inodes(xlator_t *this, xlator_t *subvol,
+                                  dht_layout_t *layout)
+{
+        int i = 0;
+        double max = 0;
+        double max_inodes = 0;
+        int    ignore_subvol = 0;
+
+        xlator_t *avail_subvol = NULL;
+        dht_conf_t *conf = NULL;
+
+        conf = this->private;
+
+        for(i=0; i < conf->subvolume_cnt; i++) {
+                /* check if subvol has layout errors, before selecting it */
+                ignore_subvol = dht_subvol_has_err (conf->subvolumes[i],
+                                                    layout);
+                if (ignore_subvol)
+                        continue;
+
+                if ((conf->disk_unit == 'p') &&
+                    (conf->du_stats[i].avail_percent > conf->min_free_disk) &&
+                    (conf->du_stats[i].avail_inodes  > conf->min_free_inodes)) {
+                        if ((conf->du_stats[i].avail_inodes > max_inodes) ||
+                            (conf->du_stats[i].avail_percent > max)) {
+                                max = conf->du_stats[i].avail_percent;
+                                max_inodes = conf->du_stats[i].avail_inodes;
+                                avail_subvol = conf->subvolumes[i];
+                        }
+                }
+
+                if ((conf->disk_unit != 'p') &&
+                    (conf->du_stats[i].avail_space > conf->min_free_disk) &&
+                    (conf->du_stats[i].avail_inodes  > conf->min_free_inodes)) {
+                        if ((conf->du_stats[i].avail_inodes > max_inodes) ||
+                            (conf->du_stats[i].avail_space > max)) {
+                                max = conf->du_stats[i].avail_space;
+                                max_inodes = conf->du_stats[i].avail_inodes;
+                                avail_subvol = conf->subvolumes[i];
+                        }
+                }
+        }
+
+        return avail_subvol;
+}
+
+
+/* Get subvol which has atleast one inode and maximum space */
+xlator_t *
+dht_subvol_maxspace_nonzeroinode (xlator_t *this, xlator_t *subvol,
+                                  dht_layout_t *layout)
+{
+        int         i = 0;
+        double      max = 0;
+        int         ignore_subvol = 0;
+
+        xlator_t   *avail_subvol = NULL;
+        dht_conf_t *conf = NULL;
+
+        conf = this->private;
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                /* check if subvol has layout errors, before selecting it */
+                ignore_subvol = dht_subvol_has_err (conf->subvolumes[i],
+                                                    layout);
+                if (ignore_subvol)
+                        continue;
+
+                if (conf->disk_unit == 'p') {
+                        if ((conf->du_stats[i].avail_percent > max)
+                            && (conf->du_stats[i].avail_inodes > 0 )) {
+                                max = conf->du_stats[i].avail_percent;
+                                avail_subvol = conf->subvolumes[i];
+                        }
+               } else {
+                         if ((conf->du_stats[i].avail_space > max)
+                            && (conf->du_stats[i].avail_inodes > 0)) {
+                                 max = conf->du_stats[i].avail_space;
+                                 avail_subvol = conf->subvolumes[i];
+                         }
+               }
+        }
+
+        return avail_subvol;
+}
diff --git a/xlators/cluster/dht/src/dht-hashfn.c b/xlators/cluster/dht/src/dht-hashfn.c
index 31cb828a4..656cf23a0 100644
--- a/xlators/cluster/dht/src/dht-hashfn.c
+++ b/xlators/cluster/dht/src/dht-hashfn.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef _CONFIG_H
@@ -29,58 +20,92 @@
 #include "hashfn.h"
 
 
-typedef enum {
-	DHT_HASH_TYPE_DM,
-} dht_hashfn_type_t;
-
-
 int
 dht_hash_compute_internal (int type, const char *name, uint32_t *hash_p)
 {
-	int      ret = 0;
-	uint32_t hash = 0;
-
-	switch (type) {
-	case DHT_HASH_TYPE_DM:
-		hash = gf_dm_hashfn (name, strlen (name));
-		break;
-	default:
-		ret = -1;
-		break;
-	}
-
-	if (ret == 0) {
-		*hash_p = hash;
-	}
-
-	return ret;
+        int      ret = 0;
+        uint32_t hash = 0;
+
+        switch (type) {
+        case DHT_HASH_TYPE_DM:
+        case DHT_HASH_TYPE_DM_USER:
+                hash = gf_dm_hashfn (name, strlen (name));
+                break;
+        default:
+                ret = -1;
+                break;
+        }
+
+        if (ret == 0) {
+                *hash_p = hash;
+        }
+
+        return ret;
 }
 
 
-#define MAKE_RSYNC_FRIENDLY_NAME(rsync_frndly_name, name) do {          \
-                rsync_frndly_name = (char *) name;			\
-                if (name[0] == '.') {                                   \
-                        char *dot   = 0;                                \
-                        int namelen = 0;                                \
-                                                                        \
-                        dot = strrchr (name, '.');                      \
-                        if (dot && dot > (name + 1) && *(dot + 1)) {    \
-                                namelen = (dot - name);                 \
-                                rsync_frndly_name = alloca (namelen);   \
-                                strncpy (rsync_frndly_name, name + 1,   \
-                                         namelen);                      \
-                                rsync_frndly_name[namelen - 1] = 0;     \
-                        }                                               \
-                }                                                       \
-        } while (0);
-
+static inline
+gf_boolean_t
+dht_munge_name (const char *original, char *modified, size_t len, regex_t *re)
+{
+        regmatch_t      matches[2];
+        size_t          new_len;
+
+        if (regexec(re,original,2,matches,0) != REG_NOMATCH) {
+                if (matches[1].rm_so != -1) {
+                        new_len = matches[1].rm_eo - matches[1].rm_so;
+                        /* Equal would fail due to the NUL at the end. */
+                        if (new_len < len) {
+                                memcpy (modified,original+matches[1].rm_so,
+                                        new_len);
+                                modified[new_len] = '\0';
+                                return _gf_true;
+                        }
+                }
+        }
+
+        /* This is guaranteed safe because of how the dest was allocated. */
+        strcpy(modified,original);
+        return _gf_false;
+}
 
 int
-dht_hash_compute (int type, const char *name, uint32_t *hash_p)
+dht_hash_compute (xlator_t *this, int type, const char *name, uint32_t *hash_p)
 {
-	char     *rsync_friendly_name = NULL;
-
-	MAKE_RSYNC_FRIENDLY_NAME (rsync_friendly_name, name);
-
-	return dht_hash_compute_internal (type, rsync_friendly_name, hash_p);
+        char            *rsync_friendly_name    = NULL;
+        dht_conf_t      *priv                   = this->private;
+        size_t           len                    = 0;
+        gf_boolean_t     munged                 = _gf_false;
+
+        /*
+         * It wouldn't be safe to use alloca in an inline function that doesn't
+         * actually get inlined, and it wouldn't be efficient to do a real
+         * allocation, so we use alloca here (if needed) and pass that to the
+         * inline.
+         */
+
+        if (priv->extra_regex_valid) {
+                len = strlen(name) + 1;
+                rsync_friendly_name = alloca(len);
+                munged = dht_munge_name (name, rsync_friendly_name, len,
+                                         &priv->extra_regex);
+        }
+
+        if (!munged && priv->rsync_regex_valid) {
+                len = strlen(name) + 1;
+                rsync_friendly_name = alloca(len);
+                gf_log (this->name, GF_LOG_TRACE, "trying regex for %s", name);
+                munged = dht_munge_name (name, rsync_friendly_name, len,
+                                         &priv->rsync_regex);
+                if (munged) {
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "munged down to %s", rsync_friendly_name);
+                }
+        }
+
+        if (!munged) {
+                rsync_friendly_name = (char *)name;
+        }
+
+        return dht_hash_compute_internal (type, rsync_friendly_name, hash_p);
 }
diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c
index f7e89b263..311a48112 100644
--- a/xlators/cluster/dht/src/dht-helper.c
+++ b/xlators/cluster/dht/src/dht-helper.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef _CONFIG_H
@@ -27,177 +18,406 @@
 #include "xlator.h"
 #include "dht-common.h"
 
+static inline int
+dht_inode_ctx_set1 (xlator_t *this, inode_t *inode, xlator_t *subvol)
+{
+        uint64_t tmp_subvol = 0;
+
+        tmp_subvol = (long)subvol;
+        return inode_ctx_set1 (inode, this, &tmp_subvol);
+}
+
+int
+dht_inode_ctx_get1 (xlator_t *this, inode_t *inode, xlator_t **subvol)
+{
+        int ret = -1;
+        uint64_t tmp_subvol = 0;
+
+        ret =  inode_ctx_get1 (inode, this, &tmp_subvol);
+        if (tmp_subvol && subvol)
+                *subvol = (xlator_t *)tmp_subvol;
+
+        return ret;
+}
+
 
 int
 dht_frame_return (call_frame_t *frame)
 {
-	dht_local_t *local = NULL;
-	int          this_call_cnt = -1;
+        dht_local_t *local = NULL;
+        int          this_call_cnt = -1;
+
+        if (!frame)
+                return -1;
+
+        local = frame->local;
+
+        LOCK (&frame->lock);
+        {
+                this_call_cnt = --local->call_cnt;
+        }
+        UNLOCK (&frame->lock);
+
+        return this_call_cnt;
+}
 
-	if (!frame)
-		return -1;
 
-	local = frame->local;
+static uint64_t
+dht_bits_for (uint64_t num)
+{
+	uint64_t bits = 0, ctrl = 1;
 
-	LOCK (&frame->lock);
-	{
-		this_call_cnt = --local->call_cnt;
+	while (ctrl < num) {
+		ctrl *= 2;
+		bits ++;
 	}
-	UNLOCK (&frame->lock);
 
-	return this_call_cnt;
+	return bits;
 }
 
+/*
+ * A slightly "updated" version of the algorithm described in the commit log
+ * is used here.
+ *
+ * The only enhancement is that:
+ *
+ * - The number of bits used by the backend filesystem for HUGE d_off which
+ *   is described as 63, and
+ * - The number of bits used by the d_off presented by the transformation
+ *   upwards which is described as 64, are both made "configurable."
+ */
+
+
+#define BACKEND_D_OFF_BITS 63
+#define PRESENT_D_OFF_BITS 63
+
+#define ONE 1ULL
+#define MASK (~0ULL)
+#define PRESENT_MASK (MASK >> (64 - PRESENT_D_OFF_BITS))
+#define BACKEND_MASK (MASK >> (64 - BACKEND_D_OFF_BITS))
+
+#define TOP_BIT (ONE << (PRESENT_D_OFF_BITS - 1))
+#define SHIFT_BITS (max (0, (BACKEND_D_OFF_BITS - PRESENT_D_OFF_BITS + 1)))
 
 int
 dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
 {
-	dht_conf_t *conf = NULL;
-	int         cnt = 0;
-	int         max = 0;
-	uint64_t    y = 0;
+        dht_conf_t *conf = NULL;
+        int         cnt = 0;
+        int         max = 0;
+        uint64_t    y = 0;
+        uint64_t    hi_mask = 0;
+        uint64_t    off_mask = 0;
+        int         max_bits = 0;
+
+        if (x == ((uint64_t) -1)) {
+                y = (uint64_t) -1;
+                goto out;
+        }
 
+        conf = this->private;
+        if (!conf)
+                goto out;
 
-	if (x == ((uint64_t) -1)) {
-		y = (uint64_t) -1;
+        max = conf->subvolume_cnt;
+        cnt = dht_subvol_cnt (this, subvol);
+
+	if (max == 1) {
+		y = x;
 		goto out;
 	}
 
-	conf = this->private;
+        max_bits = dht_bits_for (max);
 
-	max = conf->subvolume_cnt;
-	cnt = dht_subvol_cnt (this, subvol);
+        hi_mask = ~(PRESENT_MASK >> (max_bits + 1));
 
-	y = ((x * max) + cnt);
+        if (x & hi_mask) {
+                /* HUGE d_off */
+                off_mask = MASK << max_bits;
+                y = TOP_BIT | ((x >> SHIFT_BITS) & off_mask) | cnt;
+        } else {
+                /* small d_off */
+                y = ((x * max) + cnt);
+        }
 
 out:
-	if (y_p)
-		*y_p = y;
+        if (y_p)
+                *y_p = y;
 
-	return 0;
+        return 0;
 }
 
+int
+dht_filter_loc_subvol_key (xlator_t *this, loc_t *loc, loc_t *new_loc,
+                           xlator_t **subvol)
+{
+        char          *new_name  = NULL;
+        char          *new_path  = NULL;
+        xlator_list_t *trav      = NULL;
+        char           key[1024] = {0,};
+        int            ret       = 0; /* not found */
+
+        /* Why do other tasks if first required 'char' itself is not there */
+        if (!new_loc || !loc || !loc->name || !strchr (loc->name, '@'))
+                goto out;
+
+        trav = this->children;
+        while (trav) {
+                snprintf (key, 1024, "*@%s:%s", this->name, trav->xlator->name);
+                if (fnmatch (key, loc->name, FNM_NOESCAPE) == 0) {
+                        new_name = GF_CALLOC(strlen (loc->name),
+                                             sizeof (char),
+                                             gf_common_mt_char);
+                        if (!new_name)
+                                goto out;
+                        if (fnmatch (key, loc->path, FNM_NOESCAPE) == 0) {
+                                new_path = GF_CALLOC(strlen (loc->path),
+                                                     sizeof (char),
+                                                     gf_common_mt_char);
+                                if (!new_path)
+                                        goto out;
+                                strncpy (new_path, loc->path, (strlen (loc->path) -
+                                                               strlen (key) + 1));
+                        }
+                        strncpy (new_name, loc->name, (strlen (loc->name) -
+                                                       strlen (key) + 1));
+
+                        if (new_loc) {
+                                new_loc->path   = ((new_path) ? new_path:
+                                                   gf_strdup (loc->path));
+                                new_loc->name   = new_name;
+                                new_loc->inode  = inode_ref (loc->inode);
+                                new_loc->parent = inode_ref (loc->parent);
+                        }
+                        *subvol         = trav->xlator;
+                        ret = 1;  /* success */
+                        goto out;
+                }
+                trav = trav->next;
+        }
+out:
+        if (!ret) {
+                /* !success */
+                GF_FREE (new_path);
+                GF_FREE (new_name);
+        }
+        return ret;
+}
 
 int
 dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol_p,
-		  uint64_t *x_p)
+                  uint64_t *x_p)
 {
-	dht_conf_t *conf = NULL;
-	int         cnt = 0;
-	int         max = 0;
-	uint64_t    x = 0;
-	xlator_t   *subvol = 0;
+        dht_conf_t *conf = NULL;
+        int         cnt = 0;
+        int         max = 0;
+        uint64_t    x = 0;
+        xlator_t   *subvol = 0;
+        int         max_bits = 0;
+        uint64_t    off_mask = 0;
+        uint64_t    host_mask = 0;
+
+        if (!this->private)
+                return -1;
+
+        conf = this->private;
+        max = conf->subvolume_cnt;
+
+	if (max == 1) {
+		x = y;
+		cnt = 0;
+		goto out;
+	}
 
+        if (y & TOP_BIT) {
+                /* HUGE d_off */
+                max_bits = dht_bits_for (max);
+                off_mask = (MASK << max_bits);
+                host_mask = ~(off_mask);
 
-	conf = this->private;
-	max = conf->subvolume_cnt;
+                x = ((y & ~TOP_BIT) & off_mask) << SHIFT_BITS;
 
-	cnt = y % max;
-	x   = y / max;
+                cnt = y & host_mask;
+	} else {
+                /* small d_off */
+                cnt = y % max;
+                x = y / max;
+        }
 
-	subvol = conf->subvolumes[cnt];
+out:
+        subvol = conf->subvolumes[cnt];
 
-	if (subvol_p)
-		*subvol_p = subvol;
+        if (subvol_p)
+                *subvol_p = subvol;
 
-	if (x_p)
-		*x_p = x;
+        if (x_p)
+                *x_p = x;
 
-	return 0;
+        return 0;
 }
 
 
 void
-dht_local_wipe (dht_local_t *local)
+dht_local_wipe (xlator_t *this, dht_local_t *local)
 {
-	if (!local)
-		return;
+        if (!local)
+                return;
 
-	loc_wipe (&local->loc);
-	loc_wipe (&local->loc2);
+        loc_wipe (&local->loc);
+        loc_wipe (&local->loc2);
 
-	if (local->xattr)
-		dict_unref (local->xattr);
+        if (local->xattr)
+                dict_unref (local->xattr);
 
-	if (local->inode)
-		inode_unref (local->inode);
+        if (local->inode)
+                inode_unref (local->inode);
 
-	if (local->layout)
-		FREE (local->layout);
+        if (local->layout) {
+                dht_layout_unref (this, local->layout);
+                local->layout = NULL;
+        }
 
-	loc_wipe (&local->linkfile.loc);
+        loc_wipe (&local->linkfile.loc);
 
-	if (local->linkfile.xattr)
-		dict_unref (local->linkfile.xattr);
+        if (local->linkfile.xattr)
+                dict_unref (local->linkfile.xattr);
 
-	if (local->linkfile.inode)
-		inode_unref (local->linkfile.inode);
+        if (local->linkfile.inode)
+                inode_unref (local->linkfile.inode);
 
-	if (local->fd) {
-		fd_unref (local->fd);
-		local->fd = NULL;
-	}
-	
-	if (local->xattr_req)
-		dict_unref (local->xattr_req);
+        if (local->fd) {
+                fd_unref (local->fd);
+                local->fd = NULL;
+        }
+
+        if (local->params) {
+                dict_unref (local->params);
+                local->params = NULL;
+        }
+
+        if (local->xattr_req)
+                dict_unref (local->xattr_req);
+
+        if (local->selfheal.layout) {
+                dht_layout_unref (this, local->selfheal.layout);
+                local->selfheal.layout = NULL;
+        }
+
+        GF_FREE (local->newpath);
+
+        GF_FREE (local->key);
+
+        GF_FREE (local->rebalance.vector);
 
-	FREE (local);
+        if (local->rebalance.iobref)
+                iobref_unref (local->rebalance.iobref);
+
+        mem_put (local);
 }
 
 
 dht_local_t *
-dht_local_init (call_frame_t *frame)
+dht_local_init (call_frame_t *frame, loc_t *loc, fd_t *fd, glusterfs_fop_t fop)
 {
-	dht_local_t *local = NULL;
+        dht_local_t *local = NULL;
+        inode_t     *inode = NULL;
+        int          ret   = 0;
+
+        local = mem_get0 (THIS->local_pool);
+        if (!local)
+                goto out;
+
+        if (loc) {
+                ret = loc_copy (&local->loc, loc);
+                if (ret)
+                        goto out;
 
-	/* TODO: use mem-pool */
-	local = CALLOC (1, sizeof (*local));
+                inode = loc->inode;
+        }
 
-	if (!local)
-		return NULL;
+        if (fd) {
+                local->fd = fd_ref (fd);
+                if (!inode)
+                        inode = fd->inode;
+        }
 
-	local->op_ret = -1;
-	local->op_errno = EUCLEAN;
+        local->op_ret   = -1;
+        local->op_errno = EUCLEAN;
+        local->fop      = fop;
 
-	frame->local = local;
+        if (inode) {
+                local->layout   = dht_layout_get (frame->this, inode);
+                local->cached_subvol = dht_subvol_get_cached (frame->this,
+                                                              inode);
+        }
 
-	return local;
-}
+        frame->local = local;
 
+out:
+        if (ret) {
+                if (local)
+                        mem_put (local);
+                local = NULL;
+        }
+        return local;
+}
 
-char *
-basestr (const char *str)
+xlator_t *
+dht_first_up_subvol (xlator_t *this)
 {
-        char *basestr = NULL;
+        dht_conf_t *conf = NULL;
+        xlator_t   *child = NULL;
+        int         i = 0;
+        time_t      time = 0;
 
-        basestr = strrchr (str, '/');
-        if (basestr)
-                basestr ++;
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        LOCK (&conf->subvolume_lock);
+        {
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                        if (conf->subvol_up_time[i]) {
+                                if (!time) {
+                                        time = conf->subvol_up_time[i];
+                                        child = conf->subvolumes[i];
+                                } else if (time > conf->subvol_up_time[i]) {
+                                        time  = conf->subvol_up_time[i];
+                                        child = conf->subvolumes[i];
+                                }
+                        }
+                }
+        }
+        UNLOCK (&conf->subvolume_lock);
 
-        return basestr;
+out:
+        return child;
 }
 
 xlator_t *
-dht_first_up_child (xlator_t *this)
+dht_last_up_subvol (xlator_t *this)
 {
-	dht_conf_t *conf = NULL;
-	xlator_t   *child = NULL;
-	int         i = 0;
-
-	conf = this->private;
-	
-	LOCK (&conf->subvolume_lock);
-	{
-		for (i = 0; i < conf->subvolume_cnt; i++) {
-			if (conf->subvolume_status[i]) {
-				child = conf->subvolumes[i];
-				break;
-			}
-		}
-	}
-	UNLOCK (&conf->subvolume_lock);
-	
-	return child;
+        dht_conf_t *conf = NULL;
+        xlator_t   *child = NULL;
+        int         i = 0;
+
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        LOCK (&conf->subvolume_lock);
+        {
+                for (i = conf->subvolume_cnt-1; i >= 0; i--) {
+                        if (conf->subvolume_status[i]) {
+                                child = conf->subvolumes[i];
+                                break;
+                        }
+                }
+        }
+        UNLOCK (&conf->subvolume_lock);
+
+out:
+        return child;
 }
 
 xlator_t *
@@ -206,30 +426,40 @@ dht_subvol_get_hashed (xlator_t *this, loc_t *loc)
         dht_layout_t *layout = NULL;
         xlator_t     *subvol = NULL;
 
-        if (is_fs_root (loc)) {
-                subvol = dht_first_up_child (this);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO (this->name, loc, out);
+
+        if (__is_root_gfid (loc->gfid)) {
+                subvol = dht_first_up_subvol (this);
                 goto out;
         }
 
+        GF_VALIDATE_OR_GOTO (this->name, loc->parent, out);
+        GF_VALIDATE_OR_GOTO (this->name, loc->name, out);
+
         layout = dht_layout_get (this, loc->parent);
 
         if (!layout) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "layout missing path=%s parent=%"PRId64,
-                        loc->path, loc->parent->ino);
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "layout missing path=%s parent=%s",
+                        loc->path, uuid_utoa (loc->parent->gfid));
                 goto out;
         }
 
         subvol = dht_layout_search (this, layout, loc->name);
 
         if (!subvol) {
-                gf_log (this->name, GF_LOG_ERROR,
+                gf_log (this->name, GF_LOG_DEBUG,
                         "could not find subvolume for path=%s",
                         loc->path);
                 goto out;
         }
 
 out:
+        if (layout) {
+                dht_layout_unref (this, layout);
+        }
+
         return subvol;
 }
 
@@ -240,6 +470,8 @@ dht_subvol_get_cached (xlator_t *this, inode_t *inode)
         dht_layout_t *layout = NULL;
         xlator_t     *subvol = NULL;
 
+        GF_VALIDATE_OR_GOTO (this->name, this, out);
+        GF_VALIDATE_OR_GOTO (this->name, inode, out);
 
         layout = dht_layout_get (this, inode);
 
@@ -247,9 +479,13 @@ dht_subvol_get_cached (xlator_t *this, inode_t *inode)
                 goto out;
         }
 
-	subvol = layout->list[0].xlator;
+        subvol = layout->list[0].xlator;
 
 out:
+        if (layout) {
+                dht_layout_unref (this, layout);
+        }
+
         return subvol;
 }
 
@@ -257,70 +493,655 @@ out:
 xlator_t *
 dht_subvol_next (xlator_t *this, xlator_t *prev)
 {
-	dht_conf_t *conf = NULL;
-	int         i = 0;
-	xlator_t   *next = NULL;
-
-	conf = this->private;
-
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		if (conf->subvolumes[i] == prev) {
-			if ((i + 1) < conf->subvolume_cnt)
-				next = conf->subvolumes[i + 1];
-			break;
-		}
-	}
+        dht_conf_t *conf = NULL;
+        int         i = 0;
+        xlator_t   *next = NULL;
+
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (conf->subvolumes[i] == prev) {
+                        if ((i + 1) < conf->subvolume_cnt)
+                                next = conf->subvolumes[i + 1];
+                        break;
+                }
+        }
 
-	return next;
+out:
+        return next;
 }
 
+/* This func wraps around, if prev is actually the last subvol.
+ */
+xlator_t *
+dht_subvol_next_available (xlator_t *this, xlator_t *prev)
+{
+        dht_conf_t *conf = NULL;
+        int         i = 0;
+        xlator_t   *next = NULL;
+
+        conf = this->private;
+        if (!conf)
+                goto out;
 
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (conf->subvolumes[i] == prev) {
+                        /* if prev is last in conf->subvolumes, then wrap
+                         * around.
+                         */
+                        if ((i + 1) < conf->subvolume_cnt) {
+                                next = conf->subvolumes[i + 1];
+                        } else {
+                                next = conf->subvolumes[0];
+                        }
+                        break;
+                }
+        }
+
+out:
+        return next;
+}
 int
 dht_subvol_cnt (xlator_t *this, xlator_t *subvol)
 {
-	int i = 0;
-	int ret = -1;
-	dht_conf_t *conf = NULL;
+        int i = 0;
+        int ret = -1;
+        dht_conf_t *conf = NULL;
 
+        conf = this->private;
+        if (!conf)
+                goto out;
 
-	conf = this->private;
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (subvol == conf->subvolumes[i]) {
+                        ret = i;
+                        break;
+                }
+        }
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		if (subvol == conf->subvolumes[i]) {
-			ret = i;
-			break;
-		}
-	}
+out:
+        return ret;
+}
+
+
+#define set_if_greater(a, b) do {               \
+                if ((a) < (b))                  \
+                        (a) = (b);              \
+        } while (0)
+
+
+#define set_if_greater_time(a, an, b, bn) do {                          \
+                if (((a) < (b)) || (((a) == (b)) && ((an) < (bn)))){    \
+                        (a) = (b);                                      \
+                        (an) = (bn);                                    \
+                }                                                       \
+        } while (0)                                                     \
+
+
+int
+dht_iatt_merge (xlator_t *this, struct iatt *to,
+                struct iatt *from, xlator_t *subvol)
+{
+        if (!from || !to)
+                return 0;
+
+        to->ia_dev      = from->ia_dev;
+
+        uuid_copy (to->ia_gfid, from->ia_gfid);
+
+        to->ia_ino      = from->ia_ino;
+        to->ia_prot     = from->ia_prot;
+        to->ia_type     = from->ia_type;
+        to->ia_nlink    = from->ia_nlink;
+        to->ia_rdev     = from->ia_rdev;
+        to->ia_size    += from->ia_size;
+        to->ia_blksize  = from->ia_blksize;
+        to->ia_blocks  += from->ia_blocks;
+
+        set_if_greater (to->ia_uid, from->ia_uid);
+        set_if_greater (to->ia_gid, from->ia_gid);
+
+        set_if_greater_time(to->ia_atime, to->ia_atime_nsec,
+                            from->ia_atime, from->ia_atime_nsec);
+        set_if_greater_time (to->ia_mtime, to->ia_mtime_nsec,
+                             from->ia_mtime, from->ia_mtime_nsec);
+        set_if_greater_time (to->ia_ctime, to->ia_ctime_nsec,
+                             from->ia_ctime, from->ia_ctime_nsec);
+
+        return 0;
+}
+
+int
+dht_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name)
+{
+        if (!child) {
+                goto err;
+        }
+
+        if (strcmp (parent->path, "/") == 0)
+                gf_asprintf ((char **)&child->path, "/%s", name);
+        else
+                gf_asprintf ((char **)&child->path, "%s/%s", parent->path, name);
+
+        if (!child->path) {
+                goto err;
+        }
+
+        child->name = strrchr (child->path, '/');
+        if (child->name)
+                child->name++;
+
+        child->parent = inode_ref (parent->inode);
+        child->inode = inode_new (parent->inode->table);
+
+        if (!child->inode) {
+                goto err;
+        }
+
+        return 0;
+err:
+        loc_wipe (child);
+        return -1;
+}
+
+
+
+int
+dht_init_subvolumes (xlator_t *this, dht_conf_t *conf)
+{
+        xlator_list_t *subvols = NULL;
+        int            cnt = 0;
 
-	return ret;
+        if (!conf)
+                return -1;
+
+        for (subvols = this->children; subvols; subvols = subvols->next)
+                cnt++;
+
+        conf->subvolumes = GF_CALLOC (cnt, sizeof (xlator_t *),
+                                      gf_dht_mt_xlator_t);
+        if (!conf->subvolumes) {
+                return -1;
+        }
+        conf->subvolume_cnt = cnt;
+
+        cnt = 0;
+        for (subvols = this->children; subvols; subvols = subvols->next)
+                conf->subvolumes[cnt++] = subvols->xlator;
+
+        conf->subvolume_status = GF_CALLOC (cnt, sizeof (char),
+                                            gf_dht_mt_char);
+        if (!conf->subvolume_status) {
+                return -1;
+        }
+
+        conf->last_event = GF_CALLOC (cnt, sizeof (int),
+                                      gf_dht_mt_char);
+        if (!conf->last_event) {
+                return -1;
+        }
+
+        conf->subvol_up_time = GF_CALLOC (cnt, sizeof (time_t),
+                                          gf_dht_mt_subvol_time);
+        if (!conf->subvol_up_time) {
+                return -1;
+        }
+
+        conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
+                                    gf_dht_mt_dht_du_t);
+        if (!conf->du_stats) {
+                return -1;
+        }
+
+        conf->decommissioned_bricks = GF_CALLOC (cnt, sizeof (xlator_t *),
+                                                 gf_dht_mt_xlator_t);
+        if (!conf->decommissioned_bricks) {
+                return -1;
+        }
+
+        return 0;
+}
+
+
+
+
+static int
+dht_migration_complete_check_done (int op_ret, call_frame_t *frame, void *data)
+{
+        dht_local_t *local = NULL;
+
+        local = frame->local;
+
+        local->rebalance.target_op_fn (THIS, frame, op_ret);
+
+        return 0;
+}
+
+
+int
+dht_migration_complete_check_task (void *data)
+{
+        int           ret      = -1;
+        xlator_t     *src_node = NULL;
+        xlator_t     *dst_node = NULL;
+        dht_local_t  *local    = NULL;
+        dict_t       *dict     = NULL;
+        dht_layout_t *layout   = NULL;
+        struct iatt   stbuf    = {0,};
+        xlator_t     *this     = NULL;
+        call_frame_t *frame    = NULL;
+        loc_t         tmp_loc  = {0,};
+        char         *path     = NULL;
+        dht_conf_t   *conf     = NULL;
+        inode_t      *inode    = NULL;
+        fd_t         *iter_fd  = NULL;
+        uint64_t      tmp_subvol = 0;
+        int           open_failed = 0;
+
+        this  = THIS;
+        frame = data;
+        local = frame->local;
+        conf = this->private;
+
+        src_node = local->cached_subvol;
+
+        if (!local->loc.inode && !local->fd)
+                goto out;
+
+        inode = (!local->fd) ? local->loc.inode : local->fd->inode;
+
+        /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
+         * as root:root. If a fd is already open, access check wont be done*/
+
+        if (!local->loc.inode) {
+                ret = syncop_fgetxattr (src_node, local->fd, &dict,
+                                        conf->link_xattr_name);
+        } else {
+                SYNCTASK_SETID (0, 0);
+                ret = syncop_getxattr (src_node, &local->loc, &dict,
+                                       conf->link_xattr_name);
+                SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+        }
+
+        if (!ret)
+                dst_node = dht_linkfile_subvol (this, NULL, NULL, dict);
+
+        if (ret) {
+                if ((errno != ENOENT) || (!local->loc.inode)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: failed to get the 'linkto' xattr %s",
+                                local->loc.path, strerror (errno));
+                        goto out;
+                }
+                /* Need to do lookup on hashed subvol, then get the file */
+                ret = syncop_lookup (this, &local->loc, NULL, &stbuf, NULL,
+                                     NULL);
+                if (ret)
+                        goto out;
+                dst_node = dht_subvol_get_cached (this, local->loc.inode);
+        }
+
+        if (!dst_node) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to get the destination node",
+                        local->loc.path);
+                ret = -1;
+                goto out;
+        }
+
+        /* lookup on dst */
+        if (local->loc.inode) {
+                ret = syncop_lookup (dst_node, &local->loc, NULL, &stbuf, NULL, NULL);
+
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: failed to lookup the file on %s",
+                                local->loc.path, dst_node->name);
+                        goto out;
+                }
+
+                if (uuid_compare (stbuf.ia_gfid, local->loc.inode->gfid)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: gfid different on the target file on %s",
+                                local->loc.path, dst_node->name);
+                        ret = -1;
+                        goto out;
+                }
+        }
+
+        /* update inode ctx (the layout) */
+        dht_layout_unref (this, local->layout);
+
+        ret = dht_layout_preset (this, dst_node, inode);
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "%s: could not set preset layout for subvol %s",
+                        local->loc.path, dst_node->name);
+                ret   = -1;
+                goto out;
+        }
+
+        layout = dht_layout_for_subvol (this, dst_node);
+        if (!layout) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "%s: no pre-set layout for subvolume %s",
+                        local->loc.path, dst_node ? dst_node->name : "<nil>");
+                ret = -1;
+                goto out;
+        }
+
+        ret = dht_layout_set (this, inode, layout);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to set the new layout",
+                        local->loc.path);
+                goto out;
+        }
+
+        local->cached_subvol = dst_node;
+        ret = 0;
+
+        /* once we detect the migration complete, the inode-ctx2 is no more
+           required.. delete the ctx and also, it means, open() already
+           done on all the fd of inode */
+        ret = inode_ctx_reset1 (inode, this, &tmp_subvol);
+        if (tmp_subvol)
+                goto out;
+
+        if (list_empty (&inode->fd_list))
+                goto out;
+
+        /* perform open as root:root. There is window between linkfile
+         * creation(root:root) and setattr with the correct uid/gid
+         */
+        SYNCTASK_SETID(0, 0);
+
+        /* perform 'open()' on all the fd's present on the inode */
+        tmp_loc.inode = inode;
+        inode_path (inode, NULL, &path);
+        if (path)
+                tmp_loc.path = path;
+        list_for_each_entry (iter_fd, &inode->fd_list, inode_list) {
+                if (fd_is_anonymous (iter_fd))
+                        continue;
+
+                ret = syncop_open (dst_node, &tmp_loc,
+                                   iter_fd->flags, iter_fd);
+                if (ret == -1) {
+                        gf_log (this->name, GF_LOG_ERROR, "failed to open "
+                                "the fd (%p, flags=0%o) on file %s @ %s",
+                                iter_fd, iter_fd->flags, path, dst_node->name);
+                        open_failed = 1;
+                }
+        }
+        GF_FREE (path);
+
+        SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+        if (open_failed) {
+                ret = -1;
+                goto out;
+        }
+        ret = 0;
+out:
+
+        return ret;
+}
+
+int
+dht_rebalance_complete_check (xlator_t *this, call_frame_t *frame)
+{
+        int         ret     = -1;
+
+        ret = synctask_new (this->ctx->env, dht_migration_complete_check_task,
+                            dht_migration_complete_check_done,
+                            frame, frame);
+        return ret;
 }
 
+/* During 'in-progress' state, both nodes should have the file */
+static int
+dht_inprogress_check_done (int op_ret, call_frame_t *sync_frame, void *data)
+{
+        dht_local_t *local = NULL;
+
+        local = sync_frame->local;
+
+        local->rebalance.target_op_fn (THIS, sync_frame, op_ret);
+
+        return 0;
+}
+
+static int
+dht_rebalance_inprogress_task (void *data)
+{
+        int           ret      = -1;
+        xlator_t     *src_node = NULL;
+        xlator_t     *dst_node = NULL;
+        dht_local_t  *local    = NULL;
+        dict_t       *dict     = NULL;
+        call_frame_t *frame    = NULL;
+        xlator_t     *this     = NULL;
+        char         *path     = NULL;
+        struct iatt   stbuf    = {0,};
+        loc_t         tmp_loc  = {0,};
+        dht_conf_t   *conf     = NULL;
+        inode_t      *inode    = NULL;
+        fd_t         *iter_fd  = NULL;
+        int           open_failed = 0;
+
+        this  = THIS;
+        frame = data;
+        local = frame->local;
+        conf = this->private;
+
+        src_node = local->cached_subvol;
+
+        if (!local->loc.inode && !local->fd)
+                goto out;
+
+        inode = (!local->fd) ? local->loc.inode : local->fd->inode;
+
+        /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
+         * as root:root. If a fd is already open, access check wont be done*/
+        if (local->loc.inode) {
+                SYNCTASK_SETID (0, 0);
+                ret = syncop_getxattr (src_node, &local->loc, &dict,
+                                       conf->link_xattr_name);
+                SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+        } else {
+                ret = syncop_fgetxattr (src_node, local->fd, &dict,
+                                        conf->link_xattr_name);
+        }
+
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to get the 'linkto' xattr %s",
+                        local->loc.path, strerror (errno));
+                        goto out;
+        }
+
+        dst_node = dht_linkfile_subvol (this, NULL, NULL, dict);
+        if (!dst_node) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to get the 'linkto' xattr from dict",
+                        local->loc.path);
+                ret = -1;
+                goto out;
+        }
 
-#define set_if_greater(a, b) do {		\
-		if ((a) < (b))			\
-			(a) = (b);		\
-	} while (0)
+        local->rebalance.target_node = dst_node;
+
+        if (local->loc.inode) {
+                /* lookup on dst */
+                ret = syncop_lookup (dst_node, &local->loc, NULL,
+                                     &stbuf, NULL, NULL);
+                if (ret) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: failed to lookup the file on %s",
+                                local->loc.path, dst_node->name);
+                        goto out;
+                }
+
+                if (uuid_compare (stbuf.ia_gfid, local->loc.inode->gfid)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "%s: gfid different on the target file on %s",
+                                local->loc.path, dst_node->name);
+                        ret = -1;
+                        goto out;
+                }
+        }
+
+        ret = 0;
+
+        if (list_empty (&inode->fd_list))
+                goto done;
+
+        /* perform open as root:root. There is window between linkfile
+         * creation(root:root) and setattr with the correct uid/gid
+         */
+        SYNCTASK_SETID (0, 0);
+
+        tmp_loc.inode = inode;
+        inode_path (inode, NULL, &path);
+        if (path)
+                tmp_loc.path = path;
+
+        list_for_each_entry (iter_fd, &inode->fd_list, inode_list) {
+                if (fd_is_anonymous (iter_fd))
+                        continue;
+
+                ret = syncop_open (dst_node, &tmp_loc,
+                                   iter_fd->flags, iter_fd);
+                if (ret == -1) {
+                        gf_log (this->name, GF_LOG_ERROR, "failed to send open "
+                                "the fd (%p, flags=0%o) on file %s @ %s",
+                                iter_fd, iter_fd->flags, path, dst_node->name);
+                        open_failed = 1;
+                }
+        }
+        GF_FREE (path);
+
+        SYNCTASK_SETID (frame->root->uid, frame->root->gid);
+
+        if (open_failed) {
+                ret = -1;
+                goto out;
+        }
+
+done:
+        ret = dht_inode_ctx_set1 (this, inode, dst_node);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "%s: failed to set inode-ctx target file at %s",
+                        local->loc.path, dst_node->name);
+                goto out;
+        }
+
+        ret = 0;
+out:
+        return ret;
+}
 
 int
-dht_stat_merge (xlator_t *this, struct stat *to,
-		struct stat *from, xlator_t *subvol)
+dht_rebalance_in_progress_check (xlator_t *this, call_frame_t *frame)
 {
-	to->st_dev      = from->st_dev;
 
-	dht_itransform (this, subvol, from->st_ino, &to->st_ino);
+        int         ret     = -1;
+
+        ret = synctask_new (this->ctx->env, dht_rebalance_inprogress_task,
+                            dht_inprogress_check_done,
+                            frame, frame);
+        return ret;
+}
+
+int
+dht_inode_ctx_layout_set (inode_t *inode, xlator_t *this,
+                          dht_layout_t *layout_int)
+{
+        dht_inode_ctx_t         *ctx            = NULL;
+        int                      ret            = -1;
+
+        ret = dht_inode_ctx_get (inode, this, &ctx);
+        if (!ret && ctx) {
+                ctx->layout = layout_int;
+        } else {
+                ctx = GF_CALLOC (1, sizeof (*ctx), gf_dht_mt_inode_ctx_t);
+                if (!ctx)
+                        return ret;
+                ctx->layout = layout_int;
+        }
+
+        ret = dht_inode_ctx_set (inode, this, ctx);
+
+        return ret;
+}
+
+int
+dht_inode_ctx_time_update (inode_t *inode, xlator_t *this, struct iatt *stat,
+                           int32_t post)
+{
+        dht_inode_ctx_t         *ctx            = NULL;
+        dht_stat_time_t         *time           = 0;
+        int                      ret            = -1;
+
+        GF_VALIDATE_OR_GOTO (this->name, stat, out);
+        GF_VALIDATE_OR_GOTO (this->name, inode, out);
+
+        ret = dht_inode_ctx_get (inode, this, &ctx);
+
+        if (ret) {
+                ctx = GF_CALLOC (1, sizeof (*ctx), gf_dht_mt_inode_ctx_t);
+                if (!ctx)
+                        return -1;
+        }
 
-	to->st_mode     = from->st_mode;
-	to->st_nlink    = from->st_nlink;
-	to->st_uid      = from->st_uid;
-	to->st_gid      = from->st_gid;
-	to->st_rdev     = from->st_rdev;
-	to->st_size    += from->st_size;
-	to->st_blksize  = from->st_blksize;
-	to->st_blocks  += from->st_blocks;
+        time = &ctx->time;
 
-	set_if_greater (to->st_atime, from->st_atime);
-	set_if_greater (to->st_mtime, from->st_mtime);
-	set_if_greater (to->st_ctime, from->st_ctime);
+        DHT_UPDATE_TIME(time->mtime, time->mtime_nsec,
+                        stat->ia_mtime, stat->ia_mtime_nsec, inode, post);
+        DHT_UPDATE_TIME(time->ctime, time->ctime_nsec,
+                        stat->ia_ctime, stat->ia_ctime_nsec, inode, post);
+        DHT_UPDATE_TIME(time->atime, time->atime_nsec,
+                        stat->ia_atime, stat->ia_atime_nsec, inode, post);
 
-	return 0;
+        ret = dht_inode_ctx_set (inode, this, ctx);
+out:
+        return 0;
+}
+
+int
+dht_inode_ctx_get (inode_t *inode, xlator_t *this, dht_inode_ctx_t **ctx)
+{
+        int             ret     = -1;
+        uint64_t        ctx_int = 0;
+
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO (this->name, inode, out);
+
+        ret = inode_ctx_get (inode, this, &ctx_int);
+
+        if (ret)
+                return ret;
+
+        if (ctx)
+                *ctx = (dht_inode_ctx_t *) ctx_int;
+out:
+        return ret;
+}
+
+int dht_inode_ctx_set (inode_t *inode, xlator_t *this, dht_inode_ctx_t *ctx)
+{
+        int             ret = -1;
+        uint64_t        ctx_int = 0;
+
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO (this->name, inode, out);
+        GF_VALIDATE_OR_GOTO (this->name, ctx, out);
+
+        ctx_int = (long)ctx;
+        ret = inode_ctx_set (inode, this, &ctx_int);
+out:
+        return ret;
 }
diff --git a/xlators/cluster/dht/src/dht-inode-read.c b/xlators/cluster/dht/src/dht-inode-read.c
new file mode 100644
index 000000000..ece84151a
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-inode-read.c
@@ -0,0 +1,1139 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "dht-common.h"
+
+int dht_access2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_readv2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_attr2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_open2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_flush2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_lk2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_fsync2 (xlator_t *this, call_frame_t *frame, int ret);
+
+int
+dht_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+              int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = 0;
+
+        local = frame->local;
+        prev = cookie;
+
+        local->op_errno = op_errno;
+        if ((op_ret == -1) && (op_errno != ENOENT)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        if (!op_ret || (local->call_cnt != 1))
+                goto out;
+
+        /* rebalance would have happened */
+        local->rebalance.target_op_fn = dht_open2;
+        ret = dht_rebalance_complete_check (this, frame);
+        if (!ret)
+                return 0;
+
+out:
+        DHT_STACK_UNWIND (open, frame, op_ret, op_errno, local->fd, xdata);
+
+        return 0;
+}
+
+int
+dht_open2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t *local  = NULL;
+        xlator_t    *subvol = NULL;
+        int          op_errno = EINVAL;
+
+        local = frame->local;
+        if (!local)
+                goto out;
+
+        op_errno = ENOENT;
+        if (op_ret)
+                goto out;
+
+        local->call_cnt = 2;
+        subvol = local->cached_subvol;
+
+        STACK_WIND (frame, dht_open_cbk, subvol, subvol->fops->open,
+                    &local->loc, local->rebalance.flags, local->fd,
+                    NULL);
+        return 0;
+
+out:
+        DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
+
+int
+dht_open (call_frame_t *frame, xlator_t *this,
+          loc_t *loc, int flags, fd_t *fd, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, loc, fd, GF_FOP_OPEN);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        local->rebalance.flags = flags;
+        local->call_cnt = 1;
+
+        STACK_WIND (frame, dht_open_cbk, subvol, subvol->fops->open,
+                    loc, flags, fd, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (open, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+int
+dht_file_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, struct iatt *stbuf, dict_t *xdata)
+{
+        xlator_t     *subvol = 0;
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        inode_t      *inode = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        if ((op_ret == -1) && (op_errno != ENOENT)) {
+                local->op_errno = op_errno;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        if (local->call_cnt != 1)
+                goto out;
+
+        local->op_errno = op_errno;
+        /* Check if the rebalance phase2 is true */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (stbuf)) {
+                inode = (local->fd) ? local->fd->inode : local->loc.inode;
+                ret = dht_inode_ctx_get1 (this, inode, &subvol);
+                if (!subvol) {
+                        /* Phase 2 of migration */
+                        local->rebalance.target_op_fn = dht_attr2;
+                        ret = dht_rebalance_complete_check (this, frame);
+                        if (!ret)
+                                return 0;
+                } else {
+                        /* value is already set in fd_ctx, that means no need
+                           to check for whether its complete or not. */
+                        dht_attr2 (this, frame, 0);
+                        return 0;
+                }
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (stat, frame, op_ret, op_errno, stbuf, xdata);
+err:
+        return 0;
+}
+
+int
+dht_attr2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t *local  = NULL;
+        xlator_t    *subvol = NULL;
+        int          op_errno = EINVAL;
+
+        local = frame->local;
+        if (!local)
+                goto out;
+
+        op_errno = local->op_errno;
+        if (op_ret == -1)
+                goto out;
+
+        subvol = local->cached_subvol;
+        local->call_cnt = 2;
+
+        if (local->fop == GF_FOP_FSTAT) {
+                STACK_WIND (frame, dht_file_attr_cbk, subvol,
+                            subvol->fops->fstat, local->fd, NULL);
+        } else {
+                STACK_WIND (frame, dht_file_attr_cbk, subvol,
+                            subvol->fops->stat, &local->loc, NULL);
+        }
+        return 0;
+
+out:
+        DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
+        return 0;
+}
+
+int
+dht_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+              int op_ret, int op_errno, struct iatt *stbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "subvolume %s returned -1 (%s)",
+                                prev->this->name, strerror (op_errno));
+
+                        goto unlock;
+                }
+
+                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+
+                local->op_ret = 0;
+        }
+unlock:
+        UNLOCK (&frame->lock);
+out:
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt)) {
+                DHT_STACK_UNWIND (stat, frame, local->op_ret, local->op_errno,
+                                  &local->stbuf, xdata);
+        }
+err:
+        return 0;
+}
+
+int
+dht_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+        dht_layout_t *layout = NULL;
+        int           i = 0;
+        int           call_cnt = 0;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_STAT);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (IA_ISREG (loc->inode->ia_type)) {
+                local->call_cnt = 1;
+
+                subvol = local->cached_subvol;
+
+                STACK_WIND (frame, dht_file_attr_cbk, subvol,
+                            subvol->fops->stat, loc, xdata);
+
+                return 0;
+        }
+
+        local->call_cnt = call_cnt = layout->cnt;
+
+        for (i = 0; i < call_cnt; i++) {
+                subvol = layout->list[i].xlator;
+
+                STACK_WIND (frame, dht_attr_cbk,
+                            subvol, subvol->fops->stat,
+                            loc, xdata);
+        }
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+        dht_layout_t *layout = NULL;
+        int           i = 0;
+        int           call_cnt = 0;
+
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FSTAT);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (IA_ISREG (fd->inode->ia_type)) {
+                local->call_cnt = 1;
+
+                subvol = local->cached_subvol;
+
+                STACK_WIND (frame, dht_file_attr_cbk, subvol,
+                            subvol->fops->fstat, fd, xdata);
+
+                return 0;
+        }
+
+        local->call_cnt = call_cnt = layout->cnt;
+
+        for (i = 0; i < call_cnt; i++) {
+                subvol = layout->list[i].xlator;
+                STACK_WIND (frame, dht_attr_cbk,
+                            subvol, subvol->fops->fstat,
+                            fd, xdata);
+        }
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fstat, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+int
+dht_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int op_ret, int op_errno,
+               struct iovec *vector, int count, struct iatt *stbuf,
+               struct iobref *iobref, dict_t *xdata)
+{
+        dht_local_t *local      = NULL;
+        int          ret        = 0;
+        inode_t     *inode      = NULL;
+        xlator_t    *subvol = 0;
+
+        local = frame->local;
+        if (!local) {
+                op_ret = -1;
+                op_errno = EINVAL;
+                goto out;
+        }
+
+        /* This is already second try, no need for re-check */
+        if (local->call_cnt != 1)
+                goto out;
+
+        if ((op_ret == -1) && (op_errno != ENOENT))
+                goto out;
+
+        local->op_errno = op_errno;
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (stbuf)) {
+                /* File would be migrated to other node */
+                ret = dht_inode_ctx_get1 (this, inode, &subvol);
+                if (!subvol) {
+                        local->rebalance.target_op_fn = dht_readv2;
+                        ret = dht_rebalance_complete_check (this, frame);
+                        if (!ret)
+                                return 0;
+                } else {
+                        /* value is already set in fd_ctx, that means no need
+                           to check for whether its complete or not. */
+                        dht_readv2 (this, frame, 0);
+                        return 0;
+                }
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (readv, frame, op_ret, op_errno, vector, count, stbuf,
+                          iobref, xdata);
+
+        return 0;
+}
+
+int
+dht_readv2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t *local  = NULL;
+        xlator_t    *subvol = NULL;
+        int          op_errno = EINVAL;
+
+        local = frame->local;
+        if (!local)
+                goto out;
+
+        op_errno = local->op_errno;
+        if (op_ret == -1)
+                goto out;
+
+        local->call_cnt = 2;
+        subvol = local->cached_subvol;
+
+        STACK_WIND (frame, dht_readv_cbk, subvol, subvol->fops->readv,
+                    local->fd, local->rebalance.size, local->rebalance.offset,
+                    local->rebalance.flags, NULL);
+
+        return 0;
+
+out:
+        DHT_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
+        return 0;
+}
+
+int
+dht_readv (call_frame_t *frame, xlator_t *this,
+           fd_t *fd, size_t size, off_t off, uint32_t flags, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_READ);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        local->rebalance.offset = off;
+        local->rebalance.size   = size;
+        local->rebalance.flags  = flags;
+        local->call_cnt = 1;
+
+        STACK_WIND (frame, dht_readv_cbk,
+                    subvol, subvol->fops->readv,
+                    fd, size, off, flags, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
+
+        return 0;
+}
+
+int
+dht_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                int op_ret, int op_errno, dict_t *xdata)
+{
+        int          ret = -1;
+        dht_local_t *local = NULL;
+        xlator_t    *subvol = NULL;
+        call_frame_t *prev = NULL;
+
+        local = frame->local;
+        prev = cookie;
+
+        if (!prev || !prev->this)
+                goto out;
+        if (local->call_cnt != 1)
+                goto out;
+        if ((op_ret == -1) && (op_errno == ENOTCONN) &&
+            IA_ISDIR(local->loc.inode->ia_type)) {
+
+                subvol = dht_subvol_next_available (this, prev->this);
+                if (!subvol)
+                        goto out;
+
+                /* check if we are done with visiting every node */
+                if (subvol == local->cached_subvol) {
+                        goto out;
+                }
+
+                STACK_WIND (frame, dht_access_cbk, subvol, subvol->fops->access,
+                            &local->loc, local->rebalance.flags, NULL);
+                return 0;
+        }
+        if ((op_ret == -1) && (op_errno == ENOENT)) {
+                /* File would be migrated to other node */
+                local->op_errno = op_errno;
+                local->rebalance.target_op_fn = dht_access2;
+                ret = dht_rebalance_complete_check (frame->this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STACK_UNWIND (access, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+int
+dht_access2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t *local  = NULL;
+        xlator_t    *subvol = NULL;
+        int          op_errno = EINVAL;
+
+        local = frame->local;
+        if (!local)
+                goto out;
+
+        op_errno = local->op_errno;
+        if (op_ret == -1)
+                goto out;
+
+        local->call_cnt = 2;
+        subvol = local->cached_subvol;
+
+        STACK_WIND (frame, dht_access_cbk, subvol, subvol->fops->access,
+                    &local->loc, local->rebalance.flags, NULL);
+
+        return 0;
+
+out:
+        DHT_STACK_UNWIND (access, frame, -1, op_errno, NULL);
+        return 0;
+}
+
+
+int
+dht_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+            dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_ACCESS);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->rebalance.flags = mask;
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_access_cbk, subvol, subvol->fops->access,
+                    loc, mask, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (access, frame, -1, op_errno, NULL);
+
+        return 0;
+}
+
+
+int
+dht_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int op_ret, int op_errno, dict_t *xdata)
+{
+        dht_local_t  *local  = NULL;
+        inode_t      *inode  = NULL;
+        xlator_t     *subvol = 0;
+
+        local = frame->local;
+
+        local->op_errno = op_errno;
+
+        if (local->call_cnt != 1)
+                goto out;
+
+        /* If context is set, then send flush() it to the destination */
+        dht_inode_ctx_get1 (this, inode, &subvol);
+        if (subvol) {
+                dht_flush2 (this, frame, 0);
+                return 0;
+        }
+
+out:
+        DHT_STACK_UNWIND (flush, frame, op_ret, op_errno, xdata);
+
+        return 0;
+}
+
+int
+dht_flush2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+
+        local = frame->local;
+
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        STACK_WIND (frame, dht_flush_cbk,
+                    subvol, subvol->fops->flush, local->fd, NULL);
+
+        return 0;
+}
+
+
+int
+dht_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FLUSH);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        local->call_cnt = 1;
+
+        STACK_WIND (frame, dht_flush_cbk,
+                    subvol, subvol->fops->flush, fd, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (flush, frame, -1, op_errno, NULL);
+
+        return 0;
+}
+
+
+int
+dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+               dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        inode_t      *inode = NULL;
+        xlator_t     *subvol = 0;
+
+        local = frame->local;
+        prev = cookie;
+
+        local->op_errno = op_errno;
+        if (op_ret == -1 && (op_errno != ENOENT)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+
+        local->op_errno = op_errno;
+        dht_inode_ctx_get1 (this, inode, &subvol);
+        if (!subvol) {
+                local->rebalance.target_op_fn = dht_fsync2;
+
+                /* Check if the rebalance phase1 is true */
+                if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                        dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                        dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+
+                        ret = dht_rebalance_in_progress_check (this, frame);
+                }
+
+                /* Check if the rebalance phase2 is true */
+                if (IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                        ret = dht_rebalance_complete_check (this, frame);
+                }
+                if (!ret)
+                        return 0;
+        } else {
+                dht_fsync2 (this, frame, 0);
+                return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (fsync, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+
+        return 0;
+}
+
+int
+dht_fsync2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+
+        local = frame->local;
+
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        STACK_WIND (frame, dht_fsync_cbk, subvol, subvol->fops->fsync,
+                    local->fd, local->rebalance.flags, NULL);
+
+        return 0;
+}
+
+int
+dht_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync,
+           dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FSYNC);
+        if (!local) {
+                op_errno = ENOMEM;
+
+                goto err;
+        }
+
+        local->call_cnt = 1;
+        local->rebalance.flags = datasync;
+
+        subvol = local->cached_subvol;
+
+        STACK_WIND (frame, dht_fsync_cbk, subvol, subvol->fops->fsync,
+                    fd, datasync, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+/* TODO: for 'lk()' call, we need some other special error, may be ESTALE to
+   indicate that lock migration happened on the fd, so we can consider it as
+   phase 2 of migration */
+int
+dht_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+            int op_ret, int op_errno, struct gf_flock *flock, dict_t *xdata)
+{
+        DHT_STACK_UNWIND (lk, frame, op_ret, op_errno, flock, xdata);
+
+        return 0;
+}
+
+
+int
+dht_lk (call_frame_t *frame, xlator_t *this,
+        fd_t *fd, int cmd, struct gf_flock *flock, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        subvol = dht_subvol_get_cached (this, fd->inode);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        /* TODO: for rebalance, we need to preserve the fop arguments */
+        STACK_WIND (frame, dht_lk_cbk, subvol, subvol->fops->lk, fd,
+                    cmd, flock, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (lk, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+/* Symlinks are currently not migrated, so no need for any check here */
+int
+dht_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int op_ret, int op_errno, const char *path,
+                  struct iatt *stbuf, dict_t *xdata)
+{
+        dht_local_t *local = NULL;
+
+        local = frame->local;
+        if (op_ret == -1)
+                goto err;
+
+        if (!local) {
+                op_ret = -1;
+                op_errno = EINVAL;
+        }
+
+err:
+        DHT_STRIP_PHASE1_FLAGS (stbuf);
+        DHT_STACK_UNWIND (readlink, frame, op_ret, op_errno, path, stbuf, xdata);
+
+        return 0;
+}
+
+
+int
+dht_readlink (call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+              dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_READLINK);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_readlink_cbk,
+                    subvol, subvol->fops->readlink,
+                    loc, size, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (readlink, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+/* Currently no translators on top of 'distribute' will be using
+ * below fops, hence not implementing 'migration' related checks
+ */
+
+int
+dht_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+        DHT_STACK_UNWIND (xattrop, frame, op_ret, op_errno, dict, xdata);
+        return 0;
+}
+
+
+int
+dht_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
+             gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_XATTROP);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        local->call_cnt = 1;
+
+        STACK_WIND (frame,
+                    dht_xattrop_cbk,
+                    subvol, subvol->fops->xattrop,
+                    loc, flags, dict, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (xattrop, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+        DHT_STACK_UNWIND (fxattrop, frame, op_ret, op_errno, dict, xdata);
+        return 0;
+}
+
+
+int
+dht_fxattrop (call_frame_t *frame, xlator_t *this,
+              fd_t *fd, gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        subvol = dht_subvol_get_cached (this, fd->inode);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame,
+                    dht_fxattrop_cbk,
+                    subvol, subvol->fops->fxattrop,
+                    fd, flags, dict, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fxattrop, frame, -1, op_errno, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_inodelk_cbk (call_frame_t *frame, void *cookie,
+                 xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+        DHT_STACK_UNWIND (inodelk, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+
+int32_t
+dht_inodelk (call_frame_t *frame, xlator_t *this, const char *volume,
+             loc_t *loc, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_INODELK);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        local->call_cnt = 1;
+
+        STACK_WIND (frame,
+                    dht_inodelk_cbk,
+                    subvol, subvol->fops->inodelk,
+                    volume, loc, cmd, lock, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (inodelk, frame, -1, op_errno, NULL);
+
+        return 0;
+}
+
+
+int
+dht_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+        DHT_STACK_UNWIND (finodelk, frame, op_ret, op_errno, xdata);
+        return 0;
+}
+
+
+int
+dht_finodelk (call_frame_t *frame, xlator_t *this, const char *volume,
+              fd_t *fd, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        subvol = dht_subvol_get_cached (this, fd->inode);
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+
+        STACK_WIND (frame, dht_finodelk_cbk, subvol, subvol->fops->finodelk,
+                    volume, fd, cmd, lock, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (finodelk, frame, -1, op_errno, NULL);
+
+        return 0;
+}
diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c
new file mode 100644
index 000000000..4b3f3a049
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-inode-write.c
@@ -0,0 +1,1013 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "dht-common.h"
+
+int dht_writev2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_truncate2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_setattr2 (xlator_t *this, call_frame_t *frame, int ret);
+int dht_fallocate2(xlator_t *this, call_frame_t *frame, int op_ret);
+int dht_discard2(xlator_t *this, call_frame_t *frame, int op_ret);
+int dht_zerofill2(xlator_t *this, call_frame_t *frame, int op_ret);
+
+int
+dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                int op_ret, int op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t *local = NULL;
+        int          ret   = -1;
+        xlator_t    *subvol = NULL;
+
+        if (op_ret == -1 && (op_errno != ENOENT)) {
+                goto out;
+        }
+
+        local = frame->local;
+        if (!local) {
+                op_ret = -1;
+                op_errno = EINVAL;
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                /* preserve the modes of source */
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+
+        local->rebalance.target_op_fn = dht_writev2;
+
+        local->op_errno = op_errno;
+        /* Phase 2 of migration */
+        if (IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* Check if the rebalance phase1 is true */
+        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+
+                ret = dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+                if (subvol) {
+                        dht_writev2 (this, frame, 0);
+                        return 0;
+                }
+                ret = dht_rebalance_in_progress_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+
+        DHT_STACK_UNWIND (writev, frame, op_ret, op_errno, prebuf, postbuf,
+                          xdata);
+
+        return 0;
+}
+
+int
+dht_writev2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+
+        local = frame->local;
+
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        STACK_WIND (frame, dht_writev_cbk,
+                    subvol, subvol->fops->writev,
+                    local->fd, local->rebalance.vector, local->rebalance.count,
+                    local->rebalance.offset, local->rebalance.flags,
+                    local->rebalance.iobref, NULL);
+
+        return 0;
+}
+
+int
+dht_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
+            struct iovec *vector, int count, off_t off, uint32_t flags,
+            struct iobref *iobref, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_WRITE);
+        if (!local) {
+
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+
+        local->rebalance.vector = iov_dup (vector, count);
+        local->rebalance.offset = off;
+        local->rebalance.count = count;
+        local->rebalance.flags = flags;
+        local->rebalance.iobref = iobref_ref (iobref);
+        local->call_cnt = 1;
+
+        STACK_WIND (frame, dht_writev_cbk,
+                    subvol, subvol->fops->writev,
+                    fd, vector, count, off, flags, iobref, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+
+int
+dht_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                  int op_ret, int op_errno, struct iatt *prebuf,
+                  struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        xlator_t    *subvol = NULL;
+        inode_t      *inode = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        if ((op_ret == -1) && (op_errno != ENOENT)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+
+        local->rebalance.target_op_fn = dht_truncate2;
+
+        local->op_errno = op_errno;
+        /* Phase 2 of migration */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* Check if the rebalance phase1 is true */
+        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+                inode = (local->fd) ? local->fd->inode : local->loc.inode;
+                dht_inode_ctx_get1 (this, inode, &subvol);
+                if (subvol) {
+                        dht_truncate2 (this, frame, 0);
+                        return 0;
+                }
+                ret = dht_rebalance_in_progress_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (truncate, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+err:
+        return 0;
+}
+
+
+int
+dht_truncate2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+        inode_t      *inode = NULL;
+
+        local = frame->local;
+
+        inode = local->fd ? local->fd->inode : local->loc.inode;
+
+        dht_inode_ctx_get1 (this, inode, &subvol);
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        if (local->fop == GF_FOP_TRUNCATE) {
+                STACK_WIND (frame, dht_truncate_cbk, subvol,
+                            subvol->fops->truncate, &local->loc,
+                            local->rebalance.offset, NULL);
+        } else {
+                STACK_WIND (frame, dht_truncate_cbk, subvol,
+                            subvol->fops->ftruncate, local->fd,
+                            local->rebalance.offset, NULL);
+        }
+
+        return 0;
+}
+
+int
+dht_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+              dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_TRUNCATE);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->rebalance.offset = offset;
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_truncate_cbk,
+                    subvol, subvol->fops->truncate,
+                    loc, offset, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+int
+dht_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FTRUNCATE);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->rebalance.offset = offset;
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_truncate_cbk,
+                    subvol, subvol->fops->ftruncate,
+                    fd, offset, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int op_ret, int op_errno, struct iatt *prebuf,
+                  struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        xlator_t    *subvol = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        if ((op_ret == -1) && (op_errno != ENOENT)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+        local->rebalance.target_op_fn = dht_fallocate2;
+
+        /* Phase 2 of migration */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* Check if the rebalance phase1 is true */
+        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+                dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+                if (subvol) {
+                        dht_fallocate2 (this, frame, 0);
+                        return 0;
+                }
+                ret = dht_rebalance_in_progress_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (fallocate, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+err:
+        return 0;
+}
+
+int
+dht_fallocate2(xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+
+        local = frame->local;
+
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+	STACK_WIND(frame, dht_fallocate_cbk, subvol, subvol->fops->fallocate,
+		   local->fd, local->rebalance.flags, local->rebalance.offset,
+		   local->rebalance.size, NULL);
+
+        return 0;
+}
+
+int
+dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+	      off_t offset, size_t len, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FALLOCATE);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+	local->rebalance.flags = mode;
+        local->rebalance.offset = offset;
+	local->rebalance.size = len;
+
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_fallocate_cbk,
+                    subvol, subvol->fops->fallocate,
+                    fd, mode, offset, len, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int op_ret, int op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+        xlator_t    *subvol = NULL;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        if ((op_ret == -1) && (op_errno != ENOENT)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+        local->rebalance.target_op_fn = dht_discard2;
+
+        /* Phase 2 of migration */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* Check if the rebalance phase1 is true */
+        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+                dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+                if (subvol) {
+                        dht_discard2 (this, frame, 0);
+                        return 0;
+                }
+                ret = dht_rebalance_in_progress_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (discard, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+err:
+        return 0;
+}
+
+int
+dht_discard2(xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+
+        local = frame->local;
+
+        dht_inode_ctx_get1 (this, local->fd->inode, &subvol);
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+	STACK_WIND(frame, dht_discard_cbk, subvol, subvol->fops->discard,
+		   local->fd, local->rebalance.offset, local->rebalance.size,
+		   NULL);
+
+        return 0;
+}
+
+int
+dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+	    size_t len, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        int           op_errno = -1;
+        dht_local_t  *local = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_DISCARD);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->rebalance.offset = offset;
+	local->rebalance.size = len;
+
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_discard_cbk, subvol, subvol->fops->discard,
+                    fd, offset, len, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+int
+dht_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int op_ret, int op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev  = NULL;
+        int           ret   = -1;
+
+        GF_VALIDATE_OR_GOTO ("dht", frame, err);
+        GF_VALIDATE_OR_GOTO ("dht", this, out);
+        GF_VALIDATE_OR_GOTO ("dht", frame->local, out);
+        GF_VALIDATE_OR_GOTO ("dht", cookie, out);
+
+        local = frame->local;
+        prev = cookie;
+
+        if ((op_ret == -1) && (op_errno != ENOENT)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        if (local->call_cnt != 1) {
+                if (local->stbuf.ia_blocks) {
+                        dht_iatt_merge (this, postbuf, &local->stbuf, NULL);
+                        dht_iatt_merge (this, prebuf, &local->prebuf, NULL);
+                }
+                goto out;
+        }
+        local->rebalance.target_op_fn = dht_zerofill2;
+        /* Phase 2 of migration */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* Check if the rebalance phase1 is true */
+        if (IS_DHT_MIGRATION_PHASE1 (postbuf)) {
+                dht_iatt_merge (this, &local->stbuf, postbuf, NULL);
+                dht_iatt_merge (this, &local->prebuf, prebuf, NULL);
+                ret = fd_ctx_get (local->fd, this, NULL);
+                if (!ret) {
+                        dht_zerofill2 (this, frame, 0);
+                        return 0;
+                }
+                ret = dht_rebalance_in_progress_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (zerofill, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+err:
+        return 0;
+}
+
+int
+dht_zerofill2(xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local          = NULL;
+        xlator_t     *subvol         = NULL;
+        uint64_t      tmp_subvol     = 0;
+        int           ret            = -1;
+
+        local = frame->local;
+
+        if (local->fd)
+                ret = fd_ctx_get (local->fd, this, &tmp_subvol);
+        if (!ret)
+                subvol = (xlator_t *)(long)tmp_subvol;
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        STACK_WIND(frame, dht_zerofill_cbk, subvol, subvol->fops->zerofill,
+                   local->fd, local->rebalance.offset, local->rebalance.size,
+                   NULL);
+
+        return 0;
+}
+
+int
+dht_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata)
+{
+        xlator_t     *subvol       = NULL;
+        int           op_errno     = -1;
+        dht_local_t  *local        = NULL;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_ZEROFILL);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->rebalance.offset = offset;
+        local->rebalance.size = len;
+
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no cached subvolume for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        STACK_WIND (frame, dht_zerofill_cbk, subvol, subvol->fops->zerofill,
+                    fd, offset, len, xdata);
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+
+/* handle cases of migration here for 'setattr()' calls */
+int
+dht_file_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        call_frame_t *prev = NULL;
+        int           ret = -1;
+
+        local = frame->local;
+        prev = cookie;
+
+        local->op_errno = op_errno;
+        if ((op_ret == -1) && (op_errno != ENOENT)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "subvolume %s returned -1 (%s)",
+                        prev->this->name, strerror (op_errno));
+                goto out;
+        }
+
+        if (local->call_cnt != 1)
+                goto out;
+
+        local->rebalance.target_op_fn = dht_setattr2;
+
+        /* Phase 2 of migration */
+        if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) {
+                ret = dht_rebalance_complete_check (this, frame);
+                if (!ret)
+                        return 0;
+        }
+
+        /* At the end of the migration process, whatever 'attr' we
+           have on source file will be migrated to destination file
+           in one shot, hence we don't need to check for in progress
+           state here (ie, PHASE1) */
+out:
+        DHT_STRIP_PHASE1_FLAGS (postbuf);
+        DHT_STRIP_PHASE1_FLAGS (prebuf);
+        DHT_STACK_UNWIND (setattr, frame, op_ret, op_errno,
+                          prebuf, postbuf, xdata);
+
+        return 0;
+}
+
+int
+dht_setattr2 (xlator_t *this, call_frame_t *frame, int op_ret)
+{
+        dht_local_t  *local  = NULL;
+        xlator_t     *subvol = NULL;
+        inode_t      *inode = NULL;
+
+        local = frame->local;
+
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
+
+        dht_inode_ctx_get1 (this, inode, &subvol);
+
+        if (!subvol)
+                subvol = local->cached_subvol;
+
+        local->call_cnt = 2; /* This is the second attempt */
+
+        if (local->fop == GF_FOP_SETATTR) {
+                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
+                            subvol->fops->setattr, &local->loc,
+                            &local->rebalance.stbuf, local->rebalance.flags,
+                            NULL);
+        } else {
+                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
+                            subvol->fops->fsetattr, local->fd,
+                            &local->rebalance.stbuf, local->rebalance.flags,
+                            NULL);
+        }
+
+        return 0;
+}
+
+
+/* Keep the existing code same for all the cases other than regular file */
+int
+dht_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                 int op_ret, int op_errno, struct iatt *statpre,
+                 struct iatt *statpost, dict_t *xdata)
+{
+        dht_local_t  *local = NULL;
+        int           this_call_cnt = 0;
+        call_frame_t *prev = NULL;
+
+
+        local = frame->local;
+        prev = cookie;
+
+        LOCK (&frame->lock);
+        {
+                if (op_ret == -1) {
+                        local->op_errno = op_errno;
+                        gf_log (this->name, GF_LOG_DEBUG,
+                                "subvolume %s returned -1 (%s)",
+                                prev->this->name, strerror (op_errno));
+                        goto unlock;
+                }
+
+                dht_iatt_merge (this, &local->prebuf, statpre, prev->this);
+                dht_iatt_merge (this, &local->stbuf, statpost, prev->this);
+
+                local->op_ret = 0;
+        }
+unlock:
+        UNLOCK (&frame->lock);
+
+        this_call_cnt = dht_frame_return (frame);
+        if (is_last_call (this_call_cnt))
+                DHT_STACK_UNWIND (setattr, frame, local->op_ret, local->op_errno,
+                                  &local->prebuf, &local->stbuf, xdata);
+
+        return 0;
+}
+
+
+int
+dht_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
+             struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        dht_layout_t *layout = NULL;
+        dht_local_t  *local  = NULL;
+        int           op_errno = -1;
+        int           i = -1;
+        int           call_cnt = 0;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (loc, err);
+        VALIDATE_OR_GOTO (loc->inode, err);
+        VALIDATE_OR_GOTO (loc->path, err);
+
+        local = dht_local_init (frame, loc, NULL, GF_FOP_SETATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (!layout_is_sane (layout)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "layout is not sane for path=%s", loc->path);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (IA_ISREG (loc->inode->ia_type)) {
+                /* in the regular file _cbk(), we need to check for
+                   migration possibilities */
+                local->rebalance.stbuf = *stbuf;
+                local->rebalance.flags = valid;
+                local->call_cnt = 1;
+                subvol = local->cached_subvol;
+
+                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
+                            subvol->fops->setattr,
+                            loc, stbuf, valid, xdata);
+
+                return 0;
+        }
+
+        local->call_cnt = call_cnt = layout->cnt;
+
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (frame, dht_setattr_cbk,
+                            layout->list[i].xlator,
+                            layout->list[i].xlator->fops->setattr,
+                            loc, stbuf, valid, xdata);
+        }
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
+
+
+int
+dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+              int32_t valid, dict_t *xdata)
+{
+        xlator_t     *subvol = NULL;
+        dht_layout_t *layout = NULL;
+        dht_local_t  *local  = NULL;
+        int           op_errno = -1;
+        int           i = -1;
+        int           call_cnt = 0;
+
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = dht_local_init (frame, NULL, fd, GF_FOP_FSETATTR);
+        if (!local) {
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        layout = local->layout;
+        if (!layout) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "no layout for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (!layout_is_sane (layout)) {
+                gf_log (this->name, GF_LOG_DEBUG,
+                        "layout is not sane for fd=%p", fd);
+                op_errno = EINVAL;
+                goto err;
+        }
+
+        if (IA_ISREG (fd->inode->ia_type)) {
+                /* in the regular file _cbk(), we need to check for
+                   migration possibilities */
+                local->rebalance.stbuf = *stbuf;
+                local->rebalance.flags = valid;
+                local->call_cnt = 1;
+                subvol = local->cached_subvol;
+
+                STACK_WIND (frame, dht_file_setattr_cbk, subvol,
+                            subvol->fops->fsetattr,
+                            fd, stbuf, valid, xdata);
+
+                return 0;
+        }
+
+        local->call_cnt = call_cnt = layout->cnt;
+
+        for (i = 0; i < call_cnt; i++) {
+                STACK_WIND (frame, dht_setattr_cbk,
+                            layout->list[i].xlator,
+                            layout->list[i].xlator->fops->fsetattr,
+                            fd, stbuf, valid, xdata);
+        }
+
+        return 0;
+
+err:
+        op_errno = (op_errno == -1) ? errno : op_errno;
+        DHT_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+
+        return 0;
+}
diff --git a/xlators/cluster/dht/src/dht-layout.c b/xlators/cluster/dht/src/dht-layout.c
index 45a7df601..38e9970a7 100644
--- a/xlators/cluster/dht/src/dht-layout.c
+++ b/xlators/cluster/dht/src/dht-layout.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2008-2009 Z RESEARCH, Inc. <http://www.zresearch.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef _CONFIG_H
@@ -38,553 +29,764 @@
 dht_layout_t *
 dht_layout_new (xlator_t *this, int cnt)
 {
-	dht_layout_t *layout = NULL;
+        dht_layout_t *layout = NULL;
+        dht_conf_t   *conf = NULL;
 
 
-	layout = CALLOC (1, layout_size (cnt));
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto out;
-	}
+        conf = this->private;
+
+        layout = GF_CALLOC (1, layout_size (cnt),
+                            gf_dht_mt_dht_layout_t);
+        if (!layout) {
+                goto out;
+        }
+
+        layout->type = DHT_HASH_TYPE_DM;
+        layout->cnt = cnt;
 
-	layout->cnt = cnt;
+        if (conf) {
+                layout->spread_cnt = conf->dir_spread_cnt;
+                layout->gen = conf->gen;
+        }
 
+        layout->ref = 1;
 out:
-	return layout;
+        return layout;
 }
 
 
 dht_layout_t *
 dht_layout_get (xlator_t *this, inode_t *inode)
 {
-        uint64_t layout = 0;
-        int      ret    = -1;
-
-        ret = inode_ctx_get (inode, this, &layout);
+        dht_conf_t   *conf = NULL;
+        dht_layout_t *layout = NULL;
+
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        LOCK (&conf->layout_lock);
+        {
+                dht_inode_ctx_layout_get (inode, this, &layout);
+                if (layout) {
+                        layout->ref++;
+                }
+        }
+        UNLOCK (&conf->layout_lock);
 
-        return (dht_layout_t *)(long)layout;
+out:
+        return layout;
 }
 
 
-xlator_t *
-dht_layout_search (xlator_t *this, dht_layout_t *layout, const char *name)
+int
+dht_layout_set (xlator_t *this, inode_t *inode, dht_layout_t *layout)
 {
-	uint32_t   hash = 0;
-        xlator_t  *subvol = NULL;
-	int        i = 0;
-	int        ret = 0;
+        dht_conf_t   *conf = NULL;
+        int           oldret = -1;
+        int           ret = 0;
+        dht_layout_t *old_layout;
+
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        LOCK (&conf->layout_lock);
+        {
+                oldret = dht_inode_ctx_layout_get (inode, this, &old_layout);
+                layout->ref++;
+                dht_inode_ctx_layout_set (inode, this, layout);
+        }
+        UNLOCK (&conf->layout_lock);
+
+        if (!oldret) {
+                dht_layout_unref (this, old_layout);
+        }
 
+out:
+        return ret;
+}
 
-	ret = dht_hash_compute (layout->type, name, &hash);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"hash computation failed for type=%d name=%s",
-			layout->type, name);
-		goto out;
-	}
 
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].start <= hash
-		    && layout->list[i].stop >= hash) {
-			subvol = layout->list[i].xlator;
-			break;
-		}
-	}
+void
+dht_layout_unref (xlator_t *this, dht_layout_t *layout)
+{
+        dht_conf_t  *conf = NULL;
+        int          ref = 0;
 
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume for hash (value) = %u", hash);
-	}
+        if (!layout || layout->preset || !this->private)
+                return;
 
-out:
-	return subvol;
+        conf = this->private;
+
+        LOCK (&conf->layout_lock);
+        {
+                ref = --layout->ref;
+        }
+        UNLOCK (&conf->layout_lock);
+
+        if (!ref)
+                GF_FREE (layout);
 }
 
 
 dht_layout_t *
-dht_layout_for_subvol (xlator_t *this, xlator_t *subvol)
+dht_layout_ref (xlator_t *this, dht_layout_t *layout)
 {
-	dht_conf_t   *conf = NULL;
-	dht_layout_t *layout = NULL;
-	int           i = 0;
+        dht_conf_t  *conf = NULL;
 
+        if (layout->preset || !this->private)
+                return layout;
 
-	conf = this->private;
+        conf = this->private;
+        LOCK (&conf->layout_lock);
+        {
+                layout->ref++;
+        }
+        UNLOCK (&conf->layout_lock);
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		if (conf->subvolumes[i] == subvol) {
-			layout = conf->file_layouts[i];
-			break;
-		}
-	}
+        return layout;
+}
+
+
+xlator_t *
+dht_layout_search (xlator_t *this, dht_layout_t *layout, const char *name)
+{
+        uint32_t   hash = 0;
+        xlator_t  *subvol = NULL;
+        int        i = 0;
+        int        ret = 0;
+
+
+        ret = dht_hash_compute (this, layout->type, name, &hash);
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "hash computation failed for type=%d name=%s",
+                        layout->type, name);
+                goto out;
+        }
+
+        for (i = 0; i < layout->cnt; i++) {
+                if (layout->list[i].start <= hash
+                    && layout->list[i].stop >= hash) {
+                        subvol = layout->list[i].xlator;
+                        break;
+                }
+        }
 
-	return layout;
+        if (!subvol) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "no subvolume for hash (value) = %u", hash);
+        }
+
+out:
+        return subvol;
+}
+
+
+dht_layout_t *
+dht_layout_for_subvol (xlator_t *this, xlator_t *subvol)
+{
+        dht_conf_t   *conf = NULL;
+        dht_layout_t *layout = NULL;
+        int           i = 0;
+
+        conf = this->private;
+        if (!conf)
+                goto out;
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (conf->subvolumes[i] == subvol) {
+                        layout = conf->file_layouts[i];
+                        break;
+                }
+        }
+
+out:
+        return layout;
 }
 
 
 int
 dht_layouts_init (xlator_t *this, dht_conf_t *conf)
 {
-	dht_layout_t *layout = NULL;
-	int           i = 0;
-	int           ret = -1;
-	
-
-	conf->file_layouts = CALLOC (conf->subvolume_cnt,
-				     sizeof (dht_layout_t *));
-	if (!conf->file_layouts) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto out;
-	}
+        dht_layout_t *layout = NULL;
+        int           i = 0;
+        int           ret = -1;
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		layout = dht_layout_new (this, 1);
+        if (!conf)
+                goto out;
 
-		if (!layout) {
-			goto out;
-		}
+        conf->file_layouts = GF_CALLOC (conf->subvolume_cnt,
+                                        sizeof (dht_layout_t *),
+                                        gf_dht_mt_dht_layout_t);
+        if (!conf->file_layouts) {
+                goto out;
+        }
 
-		layout->preset = 1;
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+                layout = dht_layout_new (this, 1);
 
-		layout->list[0].xlator = conf->subvolumes[i];
+                if (!layout) {
+                        goto out;
+                }
 
-		conf->file_layouts[i] = layout;
-	}
+                layout->preset = 1;
+
+                layout->list[0].xlator = conf->subvolumes[i];
 
-	ret = 0;
+                conf->file_layouts[i] = layout;
+        }
+
+        ret = 0;
 out:
-	return ret;
+        return ret;
 }
 
 
 int
 dht_disk_layout_extract (xlator_t *this, dht_layout_t *layout,
-			 int pos, int32_t **disk_layout_p)
+                         int pos, int32_t **disk_layout_p)
 {
-	int      ret = -1;
-	int32_t *disk_layout = NULL;
-
-	disk_layout = CALLOC (5, sizeof (int));
-	if (!disk_layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		goto out;
-	}
+        int      ret = -1;
+        int32_t *disk_layout = NULL;
+
+        disk_layout = GF_CALLOC (5, sizeof (int),
+                                 gf_dht_mt_int32_t);
+        if (!disk_layout) {
+                goto out;
+        }
+
+        disk_layout[0] = hton32 (1);
+        disk_layout[1] = hton32 (layout->type);
+        disk_layout[2] = hton32 (layout->list[pos].start);
+        disk_layout[3] = hton32 (layout->list[pos].stop);
 
-	disk_layout[0] = hton32 (1);
-	disk_layout[1] = hton32 (layout->type);
-	disk_layout[2] = hton32 (layout->list[pos].start);
-	disk_layout[3] = hton32 (layout->list[pos].stop);
+        if (disk_layout_p)
+                *disk_layout_p = disk_layout;
+        else
+                GF_FREE (disk_layout);
 
-	if (disk_layout_p)
-		*disk_layout_p = disk_layout;
-	ret = 0;
+        ret = 0;
 
 out:
-	return ret;
+        return ret;
 }
 
 
 int
 dht_disk_layout_merge (xlator_t *this, dht_layout_t *layout,
-		       int pos, int32_t *disk_layout)
+		       int pos, void *disk_layout_raw, int disk_layout_len)
 {
-	int      cnt = 0;
-	int      type = 0;
-	int      start_off = 0;
-	int      stop_off = 0;
-
-	/* TODO: assert disk_layout_ptr is of required length */
+        int      cnt = 0;
+        int      type = 0;
+        int      start_off = 0;
+        int      stop_off = 0;
+        int      disk_layout[4];
+
+	if (!disk_layout_raw) {
+		gf_log (this->name, GF_LOG_CRITICAL,
+                        "error no layout on disk for merge");
+		return -1;
+	}
 
-	cnt  = ntoh32 (disk_layout[0]);
-	if (cnt != 1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"disk layout has invalid count %d", cnt);
+	GF_ASSERT (disk_layout_len == sizeof (disk_layout));
+
+        memcpy (disk_layout, disk_layout_raw, disk_layout_len);
+
+        cnt  = ntoh32 (disk_layout[0]);
+        if (cnt != 1) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "disk layout has invalid count %d", cnt);
+                return -1;
+        }
+
+        type = ntoh32 (disk_layout[1]);
+	switch (type) {
+        case DHT_HASH_TYPE_DM_USER:
+                gf_log (this->name, GF_LOG_DEBUG, "found user-set layout");
+                layout->type = type;
+                /* Fall through. */
+	case DHT_HASH_TYPE_DM:
+		break;
+        default:
+		gf_log (this->name, GF_LOG_CRITICAL,
+			"Catastrophic error layout with unknown type found %d",
+			disk_layout[1]);
 		return -1;
 	}
 
-	/* TODO: assert type is compatible */
-	type      = ntoh32 (disk_layout[1]);
-	start_off = ntoh32 (disk_layout[2]);
-	stop_off  = ntoh32 (disk_layout[3]);
+        start_off = ntoh32 (disk_layout[2]);
+        stop_off  = ntoh32 (disk_layout[3]);
 
-	layout->list[pos].start = start_off;
-	layout->list[pos].stop  = stop_off;
+        layout->list[pos].start = start_off;
+        layout->list[pos].stop  = stop_off;
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"merged to layout: %u - %u (type %d) from %s",
-		start_off, stop_off, type,
-		layout->list[pos].xlator->name);
+        gf_log (this->name, GF_LOG_TRACE,
+                "merged to layout: %u - %u (type %d) from %s",
+                start_off, stop_off, type,
+                layout->list[pos].xlator->name);
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_layout_merge (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
-		  int op_ret, int op_errno, dict_t *xattr)
+                  int op_ret, int op_errno, dict_t *xattr)
 {
-	int      i     = 0;
-	int      ret   = -1;
-	int      err   = -1;
-	int32_t *disk_layout = NULL;
-
-
-	if (op_ret != 0) {
-		err = op_errno;
-	}
-
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].xlator == NULL) {
-			layout->list[i].err    = err;
-			layout->list[i].xlator = subvol;
-			break;
-		}
-	}
-
-	if (op_ret != 0) {
-		ret = 0;
-		goto out;
-	}
-
-	if (xattr) {
-		/* during lookup and not mkdir */
-		ret = dict_get_ptr (xattr, "trusted.glusterfs.dht",
-				    VOID(&disk_layout));
-	}
-
-	if (ret != 0) {
-		layout->list[i].err = -1;
-		gf_log (this->name, GF_LOG_DEBUG,
-			"missing disk layout on %s. err = %d",
-			subvol->name, err);
-		ret = 0;
-		goto out;
-	}
-
-	ret = dht_disk_layout_merge (this, layout, i, disk_layout);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"layout merge from subvolume %s failed",
-			subvol->name);
-		goto out;
-	}
-	layout->list[i].err = 0;
+        int         i     = 0;
+        int         ret   = -1;
+        int         err   = -1;
+        void       *disk_layout_raw = NULL;
+        int         disk_layout_len = 0;
+        dht_conf_t *conf = this->private;
+
+        if (op_ret != 0) {
+                err = op_errno;
+        }
+
+        for (i = 0; i < layout->cnt; i++) {
+                if (layout->list[i].xlator == NULL) {
+                        layout->list[i].err    = err;
+                        layout->list[i].xlator = subvol;
+                        break;
+                }
+        }
+
+        if (op_ret != 0) {
+                ret = 0;
+                goto out;
+        }
+
+        if (xattr) {
+                /* during lookup and not mkdir */
+                ret = dict_get_ptr_and_len (xattr, conf->xattr_name,
+					    &disk_layout_raw, &disk_layout_len);
+        }
+
+        if (ret != 0) {
+                layout->list[i].err = 0;
+                gf_log (this->name, GF_LOG_TRACE,
+                        "missing disk layout on %s. err = %d",
+                        subvol->name, err);
+                ret = 0;
+                goto out;
+        }
+
+        ret = dht_disk_layout_merge (this, layout, i, disk_layout_raw,
+				     disk_layout_len);
+        if (ret != 0) {
+                gf_log (this->name, GF_LOG_WARNING,
+                        "layout merge from subvolume %s failed",
+                        subvol->name);
+                goto out;
+        }
+        layout->list[i].err = 0;
 
 out:
-	return ret;
+        return ret;
 }
 
 
 void
 dht_layout_entry_swap (dht_layout_t *layout, int i, int j)
 {
-	uint32_t  start_swap = 0;
-	uint32_t  stop_swap = 0;
-	xlator_t *xlator_swap = 0;
-	int       err_swap = 0;
-
-
-	start_swap  = layout->list[i].start;
-	stop_swap   = layout->list[i].stop;
-	xlator_swap = layout->list[i].xlator;
-	err_swap    = layout->list[i].err;
-
-	layout->list[i].start  = layout->list[j].start;
-	layout->list[i].stop   = layout->list[j].stop;
-	layout->list[i].xlator = layout->list[j].xlator;
-	layout->list[i].err    = layout->list[j].err;
-
-	layout->list[j].start  = start_swap;
-	layout->list[j].stop   = stop_swap;
-	layout->list[j].xlator = xlator_swap;
-	layout->list[j].err    = err_swap;
+        uint32_t  start_swap = 0;
+        uint32_t  stop_swap = 0;
+        xlator_t *xlator_swap = 0;
+        int       err_swap = 0;
+
+        start_swap  = layout->list[i].start;
+        stop_swap   = layout->list[i].stop;
+        xlator_swap = layout->list[i].xlator;
+        err_swap    = layout->list[i].err;
+
+        layout->list[i].start  = layout->list[j].start;
+        layout->list[i].stop   = layout->list[j].stop;
+        layout->list[i].xlator = layout->list[j].xlator;
+        layout->list[i].err    = layout->list[j].err;
+
+        layout->list[j].start  = start_swap;
+        layout->list[j].stop   = stop_swap;
+        layout->list[j].xlator = xlator_swap;
+        layout->list[j].err    = err_swap;
+}
+
+void
+dht_layout_range_swap (dht_layout_t *layout, int i, int j)
+{
+        uint32_t  start_swap = 0;
+        uint32_t  stop_swap = 0;
+
+        start_swap  = layout->list[i].start;
+        stop_swap   = layout->list[i].stop;
+
+        layout->list[i].start  = layout->list[j].start;
+        layout->list[i].stop   = layout->list[j].stop;
+
+        layout->list[j].start  = start_swap;
+        layout->list[j].stop   = stop_swap;
 }
 
 int64_t
 dht_layout_entry_cmp_volname (dht_layout_t *layout, int i, int j)
 {
-	return (strcmp (layout->list[i].xlator->name, 
-			layout->list[j].xlator->name));
+        return (strcmp (layout->list[i].xlator->name,
+                        layout->list[j].xlator->name));
+}
+
+
+gf_boolean_t
+dht_is_subvol_in_layout (dht_layout_t *layout, xlator_t *xlator)
+{
+        int i = 0;
+
+        for (i = 0; i < layout->cnt; i++) {
+                if (!strcmp (layout->list[i].xlator->name, xlator->name))
+                        return _gf_true;
+        }
+        return _gf_false;
 }
 
 int64_t
 dht_layout_entry_cmp (dht_layout_t *layout, int i, int j)
 {
-	int64_t diff = 0;
+        int64_t diff = 0;
+
+        /* swap zero'ed out layouts to front, if needed */
+        if (!layout->list[j].start && !layout->list[j].stop) {
+                diff = (int64_t) layout->list[i].stop
+                       - (int64_t) layout->list[j].stop;
+                       goto out;
+        }
+        if (layout->list[i].err || layout->list[j].err)
+                diff = layout->list[i].err - layout->list[j].err;
+        else
+                diff = (int64_t) layout->list[i].start
+                        - (int64_t) layout->list[j].start;
 
-	if (layout->list[i].err || layout->list[j].err)
-		diff = layout->list[i].err - layout->list[j].err;
-	else
-		diff = (int64_t) layout->list[i].start
-			- (int64_t) layout->list[j].start;
-
-	return diff;
+out:
+        return diff;
 }
 
 
 int
 dht_layout_sort (dht_layout_t *layout)
 {
-	int       i = 0;
-	int       j = 0;
-	int64_t   ret = 0;
-
-	/* TODO: O(n^2) -- bad bad */
-
-	for (i = 0; i < layout->cnt - 1; i++) {
-		for (j = i + 1; j < layout->cnt; j++) {
-			ret = dht_layout_entry_cmp (layout, i, j);
-			if (ret > 0)
-				dht_layout_entry_swap (layout, i, j);
-		}
-	}
+        int       i = 0;
+        int       j = 0;
+        int64_t   ret = 0;
+
+        /* TODO: O(n^2) -- bad bad */
+
+        for (i = 0; i < layout->cnt - 1; i++) {
+                for (j = i + 1; j < layout->cnt; j++) {
+                        ret = dht_layout_entry_cmp (layout, i, j);
+                        if (ret > 0)
+                                dht_layout_entry_swap (layout, i, j);
+                }
+        }
 
-	return 0;
+        return 0;
 }
 
 int
 dht_layout_sort_volname (dht_layout_t *layout)
 {
-	int       i = 0;
-	int       j = 0;
-	int64_t   ret = 0;
-
-	/* TODO: O(n^2) -- bad bad */
-
-	for (i = 0; i < layout->cnt - 1; i++) {
-		for (j = i + 1; j < layout->cnt; j++) {
-			ret = dht_layout_entry_cmp_volname (layout, i, j);
-			if (ret > 0)
-				dht_layout_entry_swap (layout, i, j);
-		}
-	}
+        int       i = 0;
+        int       j = 0;
+        int64_t   ret = 0;
+
+        /* TODO: O(n^2) -- bad bad */
+
+        for (i = 0; i < layout->cnt - 1; i++) {
+                for (j = i + 1; j < layout->cnt; j++) {
+                        ret = dht_layout_entry_cmp_volname (layout, i, j);
+                        if (ret > 0)
+                                dht_layout_entry_swap (layout, i, j);
+                }
+        }
 
-	return 0;
+        return 0;
 }
 
 
 int
 dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
-		      uint32_t *holes_p, uint32_t *overlaps_p,
-		      uint32_t *missing_p, uint32_t *down_p, uint32_t *misc_p)
+                      uint32_t *holes_p, uint32_t *overlaps_p,
+                      uint32_t *missing_p, uint32_t *down_p, uint32_t *misc_p,
+                      uint32_t *no_space_p)
 {
-	dht_conf_t *conf = NULL;
-	uint32_t    holes    = 0;
-	uint32_t    overlaps = 0;
-	uint32_t    missing  = 0;
-	uint32_t    down     = 0;
-	uint32_t    misc     = 0;
-	uint32_t    hole_cnt = 0;
-	uint32_t    overlap_cnt = 0;
-	int         i = 0;
-	int         ret = 0;
-	uint32_t    prev_stop = 0;
-	uint32_t    last_stop = 0;
-	char        is_virgin = 1;
-
-
-	conf = this->private;
-
-	/* TODO: explain WTF is happening */
-
-	last_stop = layout->list[0].start - 1;
-	prev_stop = last_stop;
-
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].err) {
-			switch (layout->list[i].err) {
-			case -1:
-			case ENOENT:
-				missing++;
-				break;
-			case ENOTCONN:
-				down++;
-				break;
-			default:
-				misc++;
-			}
-			continue;
-		}
-
-		is_virgin = 0;
-
-		if ((prev_stop + 1) < layout->list[i].start) {
-			hole_cnt++;
-			holes += (layout->list[i].start - (prev_stop + 1));
-		}
-
-		if ((prev_stop + 1) > layout->list[i].start) {
-			overlap_cnt++;
-			overlaps += ((prev_stop + 1) - layout->list[i].start);
author	Anand Avati <avati@redhat.com>	2013-09-15 15:00:06 -0700
committer	Anand Avati <avati@redhat.com>	2013-09-25 14:09:15 -0700
commit	56769c4db9e25f10c35b7f9b6ccf24dc27b35f07 (patch)
tree	2746758353de959cfabb11221e2325c07bca4a0e /contrib/timer-wheel/timer-wheel.c
parent	22e982e31bc232f3d9597a41f49e715428e010ce (diff)