diff options
Diffstat (limited to 'xlators/storage/posix/src')
| -rw-r--r-- | xlators/storage/posix/src/Makefile.am | 11 | ||||
| -rw-r--r-- | xlators/storage/posix/src/posix-aio.c | 103 | ||||
| -rw-r--r-- | xlators/storage/posix/src/posix-aio.h | 13 | ||||
| -rw-r--r-- | xlators/storage/posix/src/posix-handle.c | 20 | ||||
| -rw-r--r-- | xlators/storage/posix/src/posix-handle.h | 13 | ||||
| -rw-r--r-- | xlators/storage/posix/src/posix-helpers.c | 521 | ||||
| -rw-r--r-- | xlators/storage/posix/src/posix-mem-types.h | 13 | ||||
| -rw-r--r-- | xlators/storage/posix/src/posix.c | 1237 | ||||
| -rw-r--r-- | xlators/storage/posix/src/posix.h | 76 |
9 files changed, 1586 insertions, 421 deletions
diff --git a/xlators/storage/posix/src/Makefile.am b/xlators/storage/posix/src/Makefile.am index 4eae9798b..88efcc784 100644 --- a/xlators/storage/posix/src/Makefile.am +++ b/xlators/storage/posix/src/Makefile.am @@ -2,17 +2,18 @@ xlator_LTLIBRARIES = posix.la xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/storage -posix_la_LDFLAGS = -module -avoidversion +posix_la_LDFLAGS = -module -avoid-version posix_la_SOURCES = posix.c posix-helpers.c posix-handle.c posix-aio.c posix_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la $(LIBAIO) noinst_HEADERS = posix.h posix-mem-types.h posix-handle.h posix-aio.h -AM_CFLAGS = -fPIC -fno-strict-aliasing -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE \ - -D$(GF_HOST_OS) -Wall -I$(top_srcdir)/libglusterfs/src -shared \ - -nostartfiles -I$(top_srcdir)/rpc/xdr/src \ - -I$(top_srcdir)/rpc/rpc-lib/src $(GF_CFLAGS) +AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \ + -I$(top_srcdir)/rpc/xdr/src \ + -I$(top_srcdir)/rpc/rpc-lib/src + +AM_CFLAGS = -fno-strict-aliasing -Wall $(GF_CFLAGS) CLEANFILES = diff --git a/xlators/storage/posix/src/posix-aio.c b/xlators/storage/posix/src/posix-aio.c index 1cf5c9d26..c3bbddd67 100644 --- a/xlators/storage/posix/src/posix-aio.c +++ b/xlators/storage/posix/src/posix-aio.c @@ -1,13 +1,12 @@ /* - Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> - This file is part of GlusterFS. + Copyright (c) 2006-2012 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 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" @@ -22,6 +21,47 @@ #include <libaio.h> +void +__posix_fd_set_odirect (fd_t *fd, struct posix_fd *pfd, int opflags, + off_t offset, size_t size) +{ + int odirect = 0; + int flags = 0; + int ret = 0; + + odirect = pfd->odirect; + + if ((fd->flags|opflags) & O_DIRECT) { + /* if instructed, use O_DIRECT always */ + odirect = 1; + } else { + /* else use O_DIRECT when feasible */ + if ((offset|size) & 0xfff) + odirect = 0; + else + odirect = 1; + } + + if (!odirect && pfd->odirect) { + flags = fcntl (pfd->fd, F_GETFL); + ret = fcntl (pfd->fd, F_SETFL, (flags & (~O_DIRECT))); + pfd->odirect = 0; + } + + if (odirect && !pfd->odirect) { + flags = fcntl (pfd->fd, F_GETFL); + ret = fcntl (pfd->fd, F_SETFL, (flags | O_DIRECT)); + pfd->odirect = 1; + } + + if (ret) { + gf_log (THIS->name, GF_LOG_WARNING, + "fcntl() failed (%s). fd=%d flags=%d pfd->odirect=%d", + strerror (errno), pfd->fd, flags, pfd->odirect); + } +} + + struct posix_aio_cb { struct iocb iocb; call_frame_t *frame; @@ -96,11 +136,7 @@ posix_aio_readv_complete (struct posix_aio_cb *paiocb, int res, int res2) /* Hack to notify higher layers of EOF. */ - if (postbuf.ia_size == 0) - op_errno = ENOENT; - else if ((offset + iov.iov_len) == postbuf.ia_size) - op_errno = ENOENT; - else if (offset > postbuf.ia_size) + if (!postbuf.ia_size || (offset + iov.iov_len) >= postbuf.ia_size) op_errno = ENOENT; LOCK (&priv->lock); @@ -143,7 +179,7 @@ posix_aio_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, priv = this->private; - ret = posix_fd_ctx_get_off (fd, this, &pfd, offset); + ret = posix_fd_ctx_get (fd, this, &pfd); if (ret < 0) { op_errno = -ret; gf_log (this->name, GF_LOG_WARNING, @@ -187,7 +223,14 @@ posix_aio_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, iocb = &paiocb->iocb; - ret = io_submit (priv->ctxp, 1, &iocb); + LOCK (&fd->lock); + { + __posix_fd_set_odirect (fd, pfd, flags, offset, size); + + ret = io_submit (priv->ctxp, 1, &iocb); + } + UNLOCK (&fd->lock); + if (ret != 1) { gf_log (this->name, GF_LOG_ERROR, "io_submit() returned %d", ret); @@ -293,7 +336,7 @@ posix_aio_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, priv = this->private; - ret = posix_fd_ctx_get_off (fd, this, &pfd, offset); + ret = posix_fd_ctx_get (fd, this, &pfd); if (ret < 0) { op_errno = -ret; gf_log (this->name, GF_LOG_WARNING, @@ -335,7 +378,15 @@ posix_aio_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, } - ret = io_submit (priv->ctxp, 1, &iocb); + LOCK (&fd->lock); + { + __posix_fd_set_odirect (fd, pfd, flags, offset, + iov_length (iov, count)); + + ret = io_submit (priv->ctxp, 1, &iocb); + } + UNLOCK (&fd->lock); + if (ret != 1) { gf_log (this->name, GF_LOG_ERROR, "io_submit() returned %d", ret); @@ -422,7 +473,7 @@ posix_aio_init (xlator_t *this) ret = io_setup (POSIX_AIO_MAX_NR_EVENTS, &priv->ctxp); if ((ret == -1 && errno == ENOSYS) || ret == -ENOSYS) { gf_log (this->name, GF_LOG_WARNING, - "Linux AIO not availble at run-time." + "Linux AIO not available at run-time." " Continuing with synchronous IO"); ret = 0; goto out; @@ -435,8 +486,8 @@ posix_aio_init (xlator_t *this) goto out; } - ret = pthread_create (&priv->aiothread, NULL, - posix_aio_thread, this); + ret = gf_thread_create (&priv->aiothread, NULL, + posix_aio_thread, this); if (ret != 0) { io_destroy (priv->ctxp); goto out; @@ -491,7 +542,7 @@ int posix_aio_on (xlator_t *this) { gf_log (this->name, GF_LOG_INFO, - "Linux AIO not availble at build-time." + "Linux AIO not available at build-time." " Continuing with synchronous IO"); return 0; } @@ -500,9 +551,19 @@ int posix_aio_off (xlator_t *this) { gf_log (this->name, GF_LOG_INFO, - "Linux AIO not availble at build-time." + "Linux AIO not available at build-time." " Continuing with synchronous IO"); return 0; } +void +__posix_fd_set_odirect (fd_t *fd, struct posix_fd *pfd, int opflags, + off_t offset, size_t size) +{ + xlator_t *this = THIS; + gf_log (this->name, GF_LOG_INFO, + "Linux AIO not available at build-time." + " Continuing with synchronous IO"); + return; +} #endif diff --git a/xlators/storage/posix/src/posix-aio.h b/xlators/storage/posix/src/posix-aio.h index dccdd2acc..5bde71601 100644 --- a/xlators/storage/posix/src/posix-aio.h +++ b/xlators/storage/posix/src/posix-aio.h @@ -1,13 +1,12 @@ /* - Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> - This file is part of GlusterFS. + Copyright (c) 2006-2012 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 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 _POSIX_AIO_H #define _POSIX_AIO_H diff --git a/xlators/storage/posix/src/posix-handle.c b/xlators/storage/posix/src/posix-handle.c index 89d64e581..219a582c9 100644 --- a/xlators/storage/posix/src/posix-handle.c +++ b/xlators/storage/posix/src/posix-handle.c @@ -1,13 +1,12 @@ /* - Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> - This file is part of GlusterFS. + Copyright (c) 2011-2012 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 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" @@ -574,13 +573,6 @@ posix_handle_hard (xlator_t *this, const char *oldpath, uuid_t gfid, struct stat } } - ret = lstat (newpath, &newbuf); - if (ret) { - gf_log (this->name, GF_LOG_WARNING, - "lstat on %s failed (%s)", newpath, strerror (errno)); - return -1; - } - if (newbuf.st_ino != oldbuf->st_ino || newbuf.st_dev != oldbuf->st_dev) { gf_log (this->name, GF_LOG_WARNING, diff --git a/xlators/storage/posix/src/posix-handle.h b/xlators/storage/posix/src/posix-handle.h index 99153137f..f1163b727 100644 --- a/xlators/storage/posix/src/posix-handle.h +++ b/xlators/storage/posix/src/posix-handle.h @@ -1,13 +1,12 @@ /* - Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> - This file is part of GlusterFS. + Copyright (c) 2011-2012 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 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 _POSIX_HANDLE_H #define _POSIX_HANDLE_H diff --git a/xlators/storage/posix/src/posix-helpers.c b/xlators/storage/posix/src/posix-helpers.c index ea3ad429a..e295f8850 100644 --- a/xlators/storage/posix/src/posix-helpers.c +++ b/xlators/storage/posix/src/posix-helpers.c @@ -1,13 +1,12 @@ /* - Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> - This file is part of GlusterFS. + Copyright (c) 2006-2012 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 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" @@ -23,6 +22,7 @@ #include <pthread.h> #include <ftw.h> #include <sys/stat.h> +#include <signal.h> #ifndef GF_BSD_HOST_OS #include <alloca.h> @@ -45,16 +45,9 @@ #include "timer.h" #include "glusterfs3-xdr.h" #include "hashfn.h" +#include "glusterfs-acl.h" #include <fnmatch.h> -typedef struct { - xlator_t *this; - const char *real_path; - dict_t *xattr; - struct iatt *stbuf; - loc_t *loc; -} posix_xattr_filler_t; - char *marker_xattrs[] = {"trusted.glusterfs.quota.*", "trusted.glusterfs.*.xtime", NULL}; @@ -108,7 +101,7 @@ out: return ignore; } -static void +static int _posix_xattr_get_set (dict_t *xattr_req, char *key, data_t *data, @@ -182,14 +175,9 @@ _posix_xattr_get_set (dict_t *xattr_req, } } else if (!strcmp (key, GLUSTERFS_OPEN_FD_COUNT)) { loc = filler->loc; - if (loc && !list_empty (&loc->inode->fd_list)) { - ret = dict_set_uint32 (filler->xattr, key, 1); - if (ret < 0) - gf_log (filler->this->name, GF_LOG_WARNING, - "Failed to set dictionary value for %s", - key); - } else { - ret = dict_set_uint32 (filler->xattr, key, 0); + if (loc) { + ret = dict_set_uint32 (filler->xattr, key, + loc->inode->fd_count); if (ret < 0) gf_log (filler->this->name, GF_LOG_WARNING, "Failed to set dictionary value for %s", @@ -202,7 +190,7 @@ _posix_xattr_get_set (dict_t *xattr_req, value = GF_CALLOC (1, xattr_size + 1, gf_posix_mt_char); if (!value) - return; + return -1; xattr_size = sys_lgetxattr (filler->real_path, key, value, xattr_size); @@ -211,7 +199,7 @@ _posix_xattr_get_set (dict_t *xattr_req, "getxattr failed. path: %s, key: %s", filler->real_path, key); GF_FREE (value); - return; + return -1; } value[xattr_size] = '\0'; @@ -226,7 +214,7 @@ _posix_xattr_get_set (dict_t *xattr_req, } } out: - return; + return 0; } @@ -337,11 +325,21 @@ posix_istat (xlator_t *this, uuid_t gfid, const char *basename, ret = lstat (real_path, &lstatbuf); - if (ret == -1) { - if (errno != ENOENT && errno != ELOOP) - gf_log (this->name, GF_LOG_WARNING, - "lstat failed on %s (%s)", - real_path, strerror (errno)); + if (ret != 0) { + if (ret == -1) { + if (errno != ENOENT && errno != ELOOP) + gf_log (this->name, GF_LOG_WARNING, + "lstat failed on %s (%s)", + real_path, strerror (errno)); + } else { + // may be some backend filesystem issue + gf_log (this->name, GF_LOG_ERROR, "lstat failed on " + "%s and return value is %d instead of -1. " + "Please see dmesg output to check whether the " + "failure is due to backend filesystem issue", + real_path, ret); + ret = -1; + } goto out; } @@ -385,11 +383,21 @@ posix_pstat (xlator_t *this, uuid_t gfid, const char *path, ret = lstat (path, &lstatbuf); - if (ret == -1) { - if (errno != ENOENT) - gf_log (this->name, GF_LOG_WARNING, - "lstat failed on %s (%s)", - path, strerror (errno)); + if (ret != 0) { + if (ret == -1) { + if (errno != ENOENT) + gf_log (this->name, GF_LOG_WARNING, + "lstat failed on %s (%s)", + path, strerror (errno)); + } else { + // may be some backend filesytem issue + gf_log (this->name, GF_LOG_ERROR, "lstat failed on " + "%s and return value is %d instead of -1. " + "Please see dmesg output to check whether the " + "failure is due to backend filesystem issue", + path, ret); + ret = -1; + } goto out; } @@ -493,8 +501,8 @@ out: int -posix_set_file_contents (xlator_t *this, const char *path, data_pair_t *trav, - int flags) +posix_set_file_contents (xlator_t *this, const char *path, char *keyp, + data_t *value, int flags) { char * key = NULL; char real_path[PATH_MAX]; @@ -506,7 +514,7 @@ posix_set_file_contents (xlator_t *this, const char *path, data_pair_t *trav, /* XXX: does not handle assigning GFID to created files */ return -1; - key = &(trav->key[15]); + key = &(keyp[15]); sprintf (real_path, "%s/%s", path, key); if (flags & XATTR_REPLACE) { @@ -518,9 +526,8 @@ posix_set_file_contents (xlator_t *this, const char *path, data_pair_t *trav, goto create; } - if (trav->value->len) { - ret = write (file_fd, trav->value->data, - trav->value->len); + if (value->len) { + ret = write (file_fd, value->data, value->len); if (ret == -1) { op_ret = -errno; gf_log (this->name, GF_LOG_ERROR, @@ -552,7 +559,7 @@ posix_set_file_contents (xlator_t *this, const char *path, data_pair_t *trav, goto out; } - ret = write (file_fd, trav->value->data, trav->value->len); + ret = write (file_fd, value->data, value->len); if (ret == -1) { op_ret = -errno; gf_log (this->name, GF_LOG_ERROR, @@ -648,17 +655,17 @@ static int gf_xattr_enotsup_log; int posix_handle_pair (xlator_t *this, const char *real_path, - data_pair_t *trav, int flags) + char *key, data_t *value, int flags) { int sys_ret = -1; int ret = 0; - if (ZR_FILE_CONTENT_REQUEST(trav->key)) { - ret = posix_set_file_contents (this, real_path, trav, flags); + if (ZR_FILE_CONTENT_REQUEST(key)) { + ret = posix_set_file_contents (this, real_path, key, value, + flags); } else { - sys_ret = sys_lsetxattr (real_path, trav->key, - trav->value->data, - trav->value->len, flags); + sys_ret = sys_lsetxattr (real_path, key, value->data, + value->len, flags); if (sys_ret < 0) { if (errno == ENOTSUP) { @@ -670,7 +677,7 @@ posix_handle_pair (xlator_t *this, const char *real_path, "flag)"); } else if (errno == ENOENT) { if (!posix_special_xattr (marker_xattrs, - trav->key)) { + key)) { gf_log (this->name, GF_LOG_ERROR, "setxattr on %s failed: %s", real_path, strerror (errno)); @@ -682,12 +689,12 @@ posix_handle_pair (xlator_t *this, const char *real_path, ((errno == EINVAL) ? GF_LOG_DEBUG : GF_LOG_ERROR), "%s: key:%s error:%s", - real_path, trav->key, + real_path, key, strerror (errno)); #else /* ! DARWIN */ gf_log (this->name, GF_LOG_ERROR, "%s: key:%s error:%s", - real_path, trav->key, + real_path, key, strerror (errno)); #endif /* DARWIN */ } @@ -702,13 +709,13 @@ out: int posix_fhandle_pair (xlator_t *this, int fd, - data_pair_t *trav, int flags) + char *key, data_t *value, int flags) { int sys_ret = -1; int ret = 0; - sys_ret = sys_fsetxattr (fd, trav->key, trav->value->data, - trav->value->len, flags); + sys_ret = sys_fsetxattr (fd, key, value->data, + value->len, flags); if (sys_ret < 0) { if (errno == ENOTSUP) { @@ -729,13 +736,11 @@ posix_fhandle_pair (xlator_t *this, int fd, ((errno == EINVAL) ? GF_LOG_DEBUG : GF_LOG_ERROR), "fd=%d: key:%s error:%s", - fd, trav->key, - strerror (errno)); + fd, key, strerror (errno)); #else /* ! DARWIN */ gf_log (this->name, GF_LOG_ERROR, "fd=%d: key:%s error:%s", - fd, trav->key, - strerror (errno)); + fd, key, strerror (errno)); #endif /* DARWIN */ } @@ -880,8 +885,8 @@ posix_spawn_janitor_thread (xlator_t *this) LOCK (&priv->lock); { if (!priv->janitor_present) { - ret = pthread_create (&priv->janitor, NULL, - posix_janitor_thread_proc, this); + ret = gf_thread_create (&priv->janitor, NULL, + posix_janitor_thread_proc, this); if (ret < 0) { gf_log (this->name, GF_LOG_ERROR, @@ -897,6 +902,74 @@ unlock: UNLOCK (&priv->lock); } +static int +is_fresh_file (struct stat *stat) +{ + struct timeval tv; + + gettimeofday (&tv, NULL); + + if ((stat->st_ctime >= (tv.tv_sec - 1)) + && (stat->st_ctime <= tv.tv_sec)) + return 1; + + return 0; +} + + +int +posix_gfid_heal (xlator_t *this, const char *path, loc_t *loc, dict_t *xattr_req) +{ + /* The purpose of this function is to prevent a race + where an inode creation FOP (like mkdir/mknod/create etc) + races with lookup in the following way: + + {create thread} | {lookup thread} + | + t0 + mkdir ("name") | + t1 + | posix_gfid_set ("name", 2); + t2 + posix_gfid_set ("name", 1); | + t3 + lstat ("name"); | lstat ("name"); + + In the above case mkdir FOP would have resulted with GFID 2 while + it should have been GFID 1. It matters in the case where GFID would + have gotten set to 1 on other subvolumes of replciate/distribute + + The "solution" here is that, if we detect lookup is attempting to + set a GFID on a file which is created very recently, but does not + yet have a GFID (i.e, between t1 and t2), then "fake" it as though + posix_gfid_heal was called at t0 instead. + */ + + uuid_t uuid_curr; + int ret = 0; + struct stat stat = {0, }; + + if (!xattr_req) + goto out; + + if (sys_lstat (path, &stat) != 0) + goto out; + + ret = sys_lgetxattr (path, GFID_XATTR_KEY, uuid_curr, 16); + if (ret != 16) { + if (is_fresh_file (&stat)) { + ret = -1; + errno = ENOENT; + goto out; + } + } + + ret = posix_gfid_set (this, path, loc, xattr_req); +out: + return ret; +} + + int posix_acl_xattr_set (xlator_t *this, const char *path, dict_t *xattr_req) { @@ -910,17 +983,17 @@ posix_acl_xattr_set (xlator_t *this, const char *path, dict_t *xattr_req) if (sys_lstat (path, &stat) != 0) goto out; - data = dict_get (xattr_req, "system.posix_acl_access"); + data = dict_get (xattr_req, POSIX_ACL_ACCESS_XATTR); if (data) { - ret = sys_lsetxattr (path, "system.posix_acl_access", + ret = sys_lsetxattr (path, POSIX_ACL_ACCESS_XATTR, data->data, data->len, 0); if (ret != 0) goto out; } - data = dict_get (xattr_req, "system.posix_acl_default"); + data = dict_get (xattr_req, POSIX_ACL_DEFAULT_XATTR); if (data) { - ret = sys_lsetxattr (path, "system.posix_acl_default", + ret = sys_lsetxattr (path, POSIX_ACL_DEFAULT_XATTR, data->data, data->len, 0); if (ret != 0) goto out; @@ -930,37 +1003,47 @@ out: return ret; } +static int +_handle_entry_create_keyvalue_pair (dict_t *d, char *k, data_t *v, + void *tmp) +{ + int ret = -1; + posix_xattr_filler_t *filler = NULL; + + filler = tmp; + + if (!strcmp (GFID_XATTR_KEY, k) || + !strcmp ("gfid-req", k) || + !strcmp (POSIX_ACL_DEFAULT_XATTR, k) || + !strcmp (POSIX_ACL_ACCESS_XATTR, k) || + ZR_FILE_CONTENT_REQUEST(k)) { + return 0; + } + + ret = posix_handle_pair (filler->this, filler->real_path, k, v, + XATTR_CREATE); + if (ret < 0) { + errno = -ret; + return -1; + } + return 0; +} + int posix_entry_create_xattr_set (xlator_t *this, const char *path, dict_t *dict) { - data_pair_t *trav = NULL; int ret = -1; + posix_xattr_filler_t filler = {0,}; + if (!dict) goto out; - trav = dict->members_list; - while (trav) { - if (!strcmp (GFID_XATTR_KEY, trav->key) || - !strcmp ("gfid-req", trav->key) || - !strcmp ("system.posix_acl_default", trav->key) || - !strcmp ("system.posix_acl_access", trav->key) || - ZR_FILE_CONTENT_REQUEST(trav->key)) { - trav = trav->next; - continue; - } - - ret = posix_handle_pair (this, path, trav, XATTR_CREATE); - if (ret < 0) { - errno = -ret; - ret = -1; - goto out; - } - trav = trav->next; - } + filler.this = this; + filler.real_path = path; - ret = 0; + ret = dict_foreach (dict, _handle_entry_create_keyvalue_pair, &filler); out: return ret; @@ -984,7 +1067,7 @@ __posix_fd_ctx_get (fd_t *fd, xlator_t *this, struct posix_fd **pfd_p) goto out; } - if (fd->pid != -1) + if (!fd_is_anonymous(fd)) /* anonymous fd */ goto out; @@ -1051,34 +1134,258 @@ posix_fd_ctx_get (fd_t *fd, xlator_t *this, struct posix_fd **pfd) return ret; } +static void * +posix_health_check_thread_proc (void *data) +{ + xlator_t *this = NULL; + struct posix_private *priv = NULL; + uint32_t interval = 0; + int ret = -1; + struct stat sb = {0, }; -int -posix_fd_ctx_get_off (fd_t *fd, xlator_t *this, struct posix_fd **pfd, - off_t offset) + this = data; + priv = this->private; + + /* prevent races when the interval is updated */ + interval = priv->health_check_interval; + if (interval == 0) + goto out; + + gf_log (this->name, GF_LOG_DEBUG, "health-check thread started, " + "interval = %d seconds", interval); + + while (1) { + /* aborting sleep() is a request to exit this thread, sleep() + * will normally not return when cancelled */ + ret = sleep (interval); + if (ret > 0) + break; + + /* prevent thread errors while doing the health-check(s) */ + pthread_setcancelstate (PTHREAD_CANCEL_DISABLE, NULL); + + /* Do the health-check, it should be moved to its own function + * in case it gets more complex. */ + ret = stat (priv->base_path, &sb); + if (ret < 0) { + gf_log (this->name, GF_LOG_WARNING, + "stat() on %s returned: %s", priv->base_path, + strerror (errno)); + goto abort; + } + + pthread_setcancelstate (PTHREAD_CANCEL_ENABLE, NULL); + } + +out: + gf_log (this->name, GF_LOG_DEBUG, "health-check thread exiting"); + + LOCK (&priv->lock); + { + priv->health_check_active = _gf_false; + } + UNLOCK (&priv->lock); + + return NULL; + +abort: + /* health-check failed */ + gf_log (this->name, GF_LOG_EMERG, "health-check failed, going down"); + xlator_notify (this->parents->xlator, GF_EVENT_CHILD_DOWN, this); + + ret = sleep (30); + if (ret == 0) { + gf_log (this->name, GF_LOG_EMERG, "still alive! -> SIGTERM"); + kill (getpid(), SIGTERM); + } + + ret = sleep (30); + if (ret == 0) { + gf_log (this->name, GF_LOG_EMERG, "still alive! -> SIGKILL"); + kill (getpid(), SIGKILL); + } + + return NULL; +} + +void +posix_spawn_health_check_thread (xlator_t *xl) { - int ret; - int flags; + struct posix_private *priv = NULL; + int ret = -1; - LOCK (&fd->inode->lock); + priv = xl->private; + + LOCK (&priv->lock); { - ret = __posix_fd_ctx_get (fd, this, pfd); - if (ret) + /* cancel the running thread */ + if (priv->health_check_active == _gf_true) { + pthread_cancel (priv->health_check); + priv->health_check_active = _gf_false; + } + + /* prevent scheduling a check in a tight loop */ + if (priv->health_check_interval == 0) goto unlock; - if ((offset & 0xfff) && (*pfd)->odirect) { - flags = fcntl ((*pfd)->fd, F_GETFL); - ret = fcntl ((*pfd)->fd, F_SETFL, (flags & (~O_DIRECT))); - (*pfd)->odirect = 0; + ret = gf_thread_create (&priv->health_check, NULL, + posix_health_check_thread_proc, xl); + if (ret < 0) { + priv->health_check_interval = 0; + priv->health_check_active = _gf_false; + gf_log (xl->name, GF_LOG_ERROR, + "unable to setup health-check thread: %s", + strerror (errno)); + goto unlock; } - if (((offset & 0xfff) == 0) && (!(*pfd)->odirect)) { - flags = fcntl ((*pfd)->fd, F_GETFL); - ret = fcntl ((*pfd)->fd, F_SETFL, (flags | O_DIRECT)); - (*pfd)->odirect = 1; - } + /* run the thread detached, resources will be freed on exit */ + pthread_detach (priv->health_check); + priv->health_check_active = _gf_true; } unlock: - UNLOCK (&fd->inode->lock); + UNLOCK (&priv->lock); +} - return ret; +int +posix_fsyncer_pick (xlator_t *this, struct list_head *head) +{ + struct posix_private *priv = NULL; + int count = 0; + + priv = this->private; + pthread_mutex_lock (&priv->fsync_mutex); + { + while (list_empty (&priv->fsyncs)) + pthread_cond_wait (&priv->fsync_cond, + &priv->fsync_mutex); + + count = priv->fsync_queue_count; + priv->fsync_queue_count = 0; + list_splice_init (&priv->fsyncs, head); + } + pthread_mutex_unlock (&priv->fsync_mutex); + + return count; +} + + +void +posix_fsyncer_process (xlator_t *this, call_stub_t *stub, gf_boolean_t do_fsync) +{ + struct posix_fd *pfd = NULL; + int ret = -1; + struct posix_private *priv = NULL; + + priv = this->private; + + ret = posix_fd_ctx_get (stub->args.fd, this, &pfd); + if (ret < 0) { + gf_log (this->name, GF_LOG_ERROR, + "could not get fdctx for fd(%s)", + uuid_utoa (stub->args.fd->inode->gfid)); + call_unwind_error (stub, -1, EINVAL); + return; + } + + if (do_fsync) { +#ifdef HAVE_FDATASYNC + if (stub->args.datasync) + ret = fdatasync (pfd->fd); + else +#endif + ret = fsync (pfd->fd); + } else { + ret = 0; + } + + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "could not fstat fd(%s)", + uuid_utoa (stub->args.fd->inode->gfid)); + call_unwind_error (stub, -1, errno); + return; + } + + call_unwind_error (stub, 0, 0); +} + + +static void +posix_fsyncer_syncfs (xlator_t *this, struct list_head *head) +{ + call_stub_t *stub = NULL; + struct posix_fd *pfd = NULL; + int ret = -1; + + stub = list_entry (head->prev, call_stub_t, list); + ret = posix_fd_ctx_get (stub->args.fd, this, &pfd); + if (ret) + return; + +#ifdef GF_LINUX_HOST_OS + /* syncfs() is not "declared" in RHEL's glibc even though + the kernel has support. + */ +#include <sys/syscall.h> +#include <unistd.h> +#ifdef SYS_syncfs + syscall (SYS_syncfs, pfd->fd); +#else + sync(); +#endif +#else + sync(); +#endif +} + + +void * +posix_fsyncer (void *d) +{ + xlator_t *this = d; + struct posix_private *priv = NULL; + call_stub_t *stub = NULL; + call_stub_t *tmp = NULL; + struct list_head list; + int count = 0; + gf_boolean_t do_fsync = _gf_true; + + priv = this->private; + + for (;;) { + INIT_LIST_HEAD (&list); + + count = posix_fsyncer_pick (this, &list); + + usleep (priv->batch_fsync_delay_usec); + + gf_log (this->name, GF_LOG_DEBUG, + "picked %d fsyncs", count); + + switch (priv->batch_fsync_mode) { + case BATCH_NONE: + case BATCH_REVERSE_FSYNC: + break; + case BATCH_SYNCFS: + case BATCH_SYNCFS_SINGLE_FSYNC: + case BATCH_SYNCFS_REVERSE_FSYNC: + posix_fsyncer_syncfs (this, &list); + break; + } + + if (priv->batch_fsync_mode == BATCH_SYNCFS) + do_fsync = _gf_false; + else + do_fsync = _gf_true; + + list_for_each_entry_safe_reverse (stub, tmp, &list, list) { + list_del_init (&stub->list); + + posix_fsyncer_process (this, stub, do_fsync); + + if (priv->batch_fsync_mode == BATCH_SYNCFS_SINGLE_FSYNC) + do_fsync = _gf_false; + } + } } diff --git a/xlators/storage/posix/src/posix-mem-types.h b/xlators/storage/posix/src/posix-mem-types.h index 593bde00c..81752c17e 100644 --- a/xlators/storage/posix/src/posix-mem-types.h +++ b/xlators/storage/posix/src/posix-mem-types.h @@ -1,13 +1,12 @@ /* - Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> - This file is part of GlusterFS. + 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. + 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 __POSIX_MEM_TYPES_H__ #define __POSIX_MEM_TYPES_H__ diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix.c index df0dce727..fb45c7a67 100644 --- a/xlators/storage/posix/src/posix.c +++ b/xlators/storage/posix/src/posix.c @@ -1,13 +1,12 @@ /* - Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> - This file is part of GlusterFS. + Copyright (c) 2006-2012 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 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" @@ -24,6 +23,8 @@ #include <pthread.h> #include <ftw.h> #include <sys/stat.h> +#include <signal.h> +#include <sys/uio.h> #ifndef GF_BSD_HOST_OS #include <alloca.h> @@ -51,8 +52,10 @@ #include "glusterfs3-xdr.h" #include "hashfn.h" #include "posix-aio.h" +#include "glusterfs-acl.h" extern char *marker_xattrs[]; +#define ALIGN_SIZE 4096 #undef HAVE_SET_FSID #ifdef HAVE_SET_FSID @@ -106,7 +109,6 @@ posix_lookup (call_frame_t *frame, xlator_t *this, VALIDATE_OR_GOTO (frame, out); VALIDATE_OR_GOTO (this, out); VALIDATE_OR_GOTO (loc, out); - VALIDATE_OR_GOTO (loc->path, out); /* The Hidden directory should be for housekeeping purpose and it should not get any gfid on it */ @@ -129,7 +131,7 @@ posix_lookup (call_frame_t *frame, xlator_t *this, MAKE_ENTRY_HANDLE (real_path, par_path, this, loc, &buf); if (uuid_is_null (loc->inode->gfid)) { - posix_gfid_set (this, real_path, loc, xdata); + posix_gfid_heal (this, real_path, loc, xdata); MAKE_ENTRY_HANDLE (real_path, par_path, this, loc, &buf); } @@ -210,7 +212,8 @@ posix_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata) if (op_ret == -1) { op_errno = errno; - gf_log (this->name, GF_LOG_ERROR, + gf_log (this->name, (op_errno == ENOENT)? + GF_LOG_DEBUG:GF_LOG_ERROR, "lstat on %s failed: %s", real_path, strerror (op_errno)); goto out; @@ -561,6 +564,289 @@ out: return 0; } +static int32_t +posix_do_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags, + off_t offset, size_t len, struct iatt *statpre, + struct iatt *statpost) +{ + struct posix_fd *pfd = NULL; + int32_t ret = -1; + + DECLARE_OLD_FS_ID_VAR; + + SET_FS_ID (frame->root->uid, frame->root->gid); + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + + ret = posix_fd_ctx_get (fd, this, &pfd); + if (ret < 0) { + gf_log (this->name, GF_LOG_DEBUG, + "pfd is NULL from fd=%p", fd); + goto out; + } + + ret = posix_fdstat (this, pfd->fd, statpre); + if (ret == -1) { + ret = -errno; + gf_log (this->name, GF_LOG_ERROR, + "fallocate (fstat) failed on fd=%p: %s", fd, + strerror (errno)); + goto out; + } + + ret = sys_fallocate(pfd->fd, flags, offset, len); + if (ret == -1) { + ret = -errno; + goto out; + } + + ret = posix_fdstat (this, pfd->fd, statpost); + if (ret == -1) { + ret = -errno; + gf_log (this->name, GF_LOG_ERROR, + "fallocate (fstat) failed on fd=%p: %s", fd, + strerror (errno)); + goto out; + } + +out: + SET_TO_OLD_FS_ID (); + + return ret; +} + +char* +_page_aligned_alloc (size_t size, char **aligned_buf) +{ + char *alloc_buf = NULL; + char *buf = NULL; + + alloc_buf = GF_CALLOC (1, (size + ALIGN_SIZE), gf_posix_mt_char); + if (!alloc_buf) + goto out; + /* page aligned buffer */ + buf = GF_ALIGN_BUF (alloc_buf, ALIGN_SIZE); + *aligned_buf = buf; +out: + return alloc_buf; +} + +static int32_t +_posix_do_zerofill(int fd, off_t offset, size_t len, int o_direct) +{ + size_t num_vect = 0; + int32_t num_loop = 1; + int32_t idx = 0; + int32_t op_ret = -1; + int32_t vect_size = VECTOR_SIZE; + size_t remain = 0; + size_t extra = 0; + struct iovec *vector = NULL; + char *iov_base = NULL; + char *alloc_buf = NULL; + + if (len == 0) + return 0; + if (len < VECTOR_SIZE) + vect_size = len; + + num_vect = len / (vect_size); + remain = len % vect_size ; + if (num_vect > MAX_NO_VECT) { + extra = num_vect % MAX_NO_VECT; + num_loop = num_vect / MAX_NO_VECT; + num_vect = MAX_NO_VECT; + } + + vector = GF_CALLOC (num_vect, sizeof(struct iovec), + gf_common_mt_iovec); + if (!vector) + return -1; + if (o_direct) { + alloc_buf = _page_aligned_alloc(vect_size, &iov_base); + if (!alloc_buf) { + gf_log ("_posix_do_zerofill", GF_LOG_DEBUG, + "memory alloc failed, vect_size %d: %s", + vect_size, strerror(errno)); + GF_FREE(vector); + return -1; + } + } else { + iov_base = GF_CALLOC (vect_size, sizeof(char), + gf_common_mt_char); + if (!iov_base) { + GF_FREE(vector); + return -1; + } + } + + for (idx = 0; idx < num_vect; idx++) { + vector[idx].iov_base = iov_base; + vector[idx].iov_len = vect_size; + } + lseek(fd, offset, SEEK_SET); + for (idx = 0; idx < num_loop; idx++) { + op_ret = writev(fd, vector, num_vect); + if (op_ret < 0) + goto err; + } + if (extra) { + op_ret = writev(fd, vector, extra); + if (op_ret < 0) + goto err; + } + if (remain) { + vector[0].iov_len = remain; + op_ret = writev(fd, vector , 1); + if (op_ret < 0) + goto err; + } +err: + if (o_direct) + GF_FREE(alloc_buf); + else + GF_FREE(iov_base); + GF_FREE(vector); + return op_ret; +} + +static int32_t +posix_do_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, + off_t offset, size_t len, struct iatt *statpre, + struct iatt *statpost) +{ + struct posix_fd *pfd = NULL; + int32_t ret = -1; + + DECLARE_OLD_FS_ID_VAR; + + SET_FS_ID (frame->root->uid, frame->root->gid); + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (fd, out); + + ret = posix_fd_ctx_get (fd, this, &pfd); + if (ret < 0) { + gf_log (this->name, GF_LOG_DEBUG, + "pfd is NULL from fd=%p", fd); + goto out; + } + + ret = posix_fdstat (this, pfd->fd, statpre); + if (ret == -1) { + ret = -errno; + gf_log (this->name, GF_LOG_ERROR, + "pre-operation fstat failed on fd = %p: %s", fd, + strerror (errno)); + goto out; + } + ret = _posix_do_zerofill(pfd->fd, offset, len, pfd->flags & O_DIRECT); + if (ret < 0) { + ret = -errno; + gf_log(this->name, GF_LOG_ERROR, + "zerofill failed on fd %d length %ld %s", + pfd->fd, len, strerror(errno)); + goto out; + } + if (pfd->flags & (O_SYNC|O_DSYNC)) { + ret = fsync (pfd->fd); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "fsync() in writev on fd %d failed: %s", + pfd->fd, strerror (errno)); + ret = -errno; + goto out; + } + } + + ret = posix_fdstat (this, pfd->fd, statpost); + if (ret == -1) { + ret = -errno; + gf_log (this->name, GF_LOG_ERROR, + "post operation fstat failed on fd=%p: %s", fd, + strerror (errno)); + goto out; + } + +out: + SET_TO_OLD_FS_ID (); + + return ret; +} + +static int32_t +_posix_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t keep_size, + off_t offset, size_t len, dict_t *xdata) +{ + int32_t ret; + int32_t flags = 0; + struct iatt statpre = {0,}; + struct iatt statpost = {0,}; + + if (keep_size) + flags = FALLOC_FL_KEEP_SIZE; + + ret = posix_do_fallocate(frame, this, fd, flags, offset, len, + &statpre, &statpost); + if (ret < 0) + goto err; + + STACK_UNWIND_STRICT(fallocate, frame, 0, 0, &statpre, &statpost, NULL); + return 0; + +err: + STACK_UNWIND_STRICT(fallocate, frame, -1, -ret, NULL, NULL, NULL); + return 0; +} + +static int32_t +posix_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + size_t len, dict_t *xdata) +{ + int32_t ret; + int32_t flags = FALLOC_FL_KEEP_SIZE|FALLOC_FL_PUNCH_HOLE; + struct iatt statpre = {0,}; + struct iatt statpost = {0,}; + + ret = posix_do_fallocate(frame, this, fd, flags, offset, len, + &statpre, &statpost); + if (ret < 0) + goto err; + + STACK_UNWIND_STRICT(discard, frame, 0, 0, &statpre, &statpost, NULL); + return 0; + +err: + STACK_UNWIND_STRICT(discard, frame, -1, -ret, NULL, NULL, NULL); + return 0; + +} + +static int32_t +posix_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + size_t len, dict_t *xdata) +{ + int32_t ret = 0; + struct iatt statpre = {0,}; + struct iatt statpost = {0,}; + + ret = posix_do_zerofill(frame, this, fd, offset, len, + &statpre, &statpost); + if (ret < 0) + goto err; + + STACK_UNWIND_STRICT(zerofill, frame, 0, 0, &statpre, &statpost, NULL); + return 0; + +err: + STACK_UNWIND_STRICT(zerofill, frame, -1, -ret, NULL, NULL, NULL); + return 0; + +} + int32_t posix_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd, dict_t *xdata) @@ -1532,7 +1818,7 @@ posix_link (call_frame_t *frame, xlator_t *this, /* * On most systems (Linux being the notable exception), link(2) * first resolves symlinks. If the target is a directory or - * is nonexistent, it will fail. linkat(2) operates on the + * is nonexistent, it will fail. linkat(2) operates on the * symlink instead of its target when the AT_SYMLINK_FOLLOW * flag is not supplied. */ @@ -1718,6 +2004,9 @@ posix_create (call_frame_t *frame, xlator_t *this, goto out; } + if (was_present) + goto fill_stat; + op_ret = posix_gfid_set (this, real_path, loc, xdata); if (op_ret) { gf_log (this->name, GF_LOG_ERROR, @@ -1748,6 +2037,7 @@ posix_create (call_frame_t *frame, xlator_t *this, strerror (errno)); } +fill_stat: op_ret = posix_fdstat (this, _fd, &stbuf); if (op_ret == -1) { op_errno = errno; @@ -1965,11 +2255,7 @@ posix_readv (call_frame_t *frame, xlator_t *this, } /* Hack to notify higher layers of EOF. */ - if (stbuf.ia_size == 0) - op_errno = ENOENT; - else if ((offset + vec.iov_len) == stbuf.ia_size) - op_errno = ENOENT; - else if (offset > stbuf.ia_size) + if (!stbuf.ia_size || (offset + vec.iov_len) >= stbuf.ia_size) op_errno = ENOENT; op_ret = vec.iov_len; @@ -2014,14 +2300,12 @@ err: return op_ret; } - int32_t __posix_writev (int fd, struct iovec *vector, int count, off_t startoff, int odirect) { int32_t op_ret = 0; int idx = 0; - int align = 4096; int max_buf_size = 0; int retval = 0; char *buf = NULL; @@ -2037,7 +2321,7 @@ __posix_writev (int fd, struct iovec *vector, int count, off_t startoff, max_buf_size = vector[idx].iov_len; } - alloc_buf = GF_MALLOC (1 * (max_buf_size + align), gf_posix_mt_char); + alloc_buf = _page_aligned_alloc (max_buf_size, &buf); if (!alloc_buf) { op_ret = -errno; goto err; @@ -2045,9 +2329,6 @@ __posix_writev (int fd, struct iovec *vector, int count, off_t startoff, internal_off = startoff; for (idx = 0; idx < count; idx++) { - /* page aligned buffer */ - buf = GF_ALIGN_BUF (alloc_buf, align); - memcpy (buf, vector[idx].iov_base, vector[idx].iov_len); /* not sure whether writev works on O_DIRECT'd fd */ @@ -2067,6 +2348,48 @@ err: return op_ret; } +dict_t* +_fill_writev_xdata (fd_t *fd, dict_t *xdata, xlator_t *this, int is_append) +{ + dict_t *rsp_xdata = NULL; + int32_t ret = 0; + inode_t *inode = NULL; + + if (fd) + inode = fd->inode; + + if (!fd || !fd->inode || uuid_is_null (fd->inode->gfid)) { + gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid Args: " + "fd: %p inode: %p gfid:%s", fd, inode?inode:0, + inode?uuid_utoa(inode->gfid):"N/A"); + goto out; + } + + if (!xdata || !dict_get (xdata, GLUSTERFS_OPEN_FD_COUNT)) + goto out; + + rsp_xdata = dict_new(); + if (!rsp_xdata) + goto out; + + ret = dict_set_uint32 (rsp_xdata, GLUSTERFS_OPEN_FD_COUNT, + fd->inode->fd_count); + if (ret < 0) { + gf_log (this->name, GF_LOG_WARNING, "%s: Failed to set " + "dictionary value for %s", uuid_utoa (fd->inode->gfid), + GLUSTERFS_OPEN_FD_COUNT); + } + + ret = dict_set_uint32 (rsp_xdata, GLUSTERFS_WRITE_IS_APPEND, + is_append); + if (ret < 0) { + gf_log (this->name, GF_LOG_WARNING, "%s: Failed to set " + "dictionary value for %s", uuid_utoa (fd->inode->gfid), + GLUSTERFS_WRITE_IS_APPEND); + } +out: + return rsp_xdata; +} int32_t posix_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, @@ -2081,6 +2404,9 @@ posix_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt preop = {0,}; struct iatt postop = {0,}; int ret = -1; + dict_t *rsp_xdata = NULL; + int is_append = 0; + gf_boolean_t locked = _gf_false; VALIDATE_OR_GOTO (frame, out); VALIDATE_OR_GOTO (this, out); @@ -2102,6 +2428,17 @@ posix_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, _fd = pfd->fd; + if (xdata && dict_get (xdata, GLUSTERFS_WRITE_IS_APPEND)) { + /* The write_is_append check and write must happen + atomically. Else another write can overtake this + write after the check and get written earlier. + + So lock before preop-stat and unlock after write. + */ + locked = _gf_true; + LOCK(&fd->inode->lock); + } + op_ret = posix_fdstat (this, _fd, &preop); if (op_ret == -1) { op_errno = errno; @@ -2111,8 +2448,19 @@ posix_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, goto out; } + if (locked) { + if (preop.ia_size == offset || (fd->flags & O_APPEND)) + is_append = 1; + } + op_ret = __posix_writev (_fd, vector, count, offset, (pfd->flags & O_DIRECT)); + + if (locked) { + UNLOCK (&fd->inode->lock); + locked = _gf_false; + } + if (op_ret < 0) { op_errno = -op_ret; op_ret = -1; @@ -2128,14 +2476,21 @@ posix_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, UNLOCK (&priv->lock); if (op_ret >= 0) { + rsp_xdata = _fill_writev_xdata (fd, xdata, this, is_append); /* wiretv successful, we also need to get the stat of * the file we wrote to */ - if (pfd->flushwrites) { - /* NOTE: ignore the error, if one occurs at this - * point */ - fsync (_fd); + if (flags & (O_SYNC|O_DSYNC)) { + ret = fsync (_fd); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "fsync() in writev on fd %d failed: %s", + _fd, strerror (errno)); + op_ret = -1; + op_errno = errno; + goto out; + } } ret = posix_fdstat (this, _fd, &postop); @@ -2151,9 +2506,16 @@ posix_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, out: + if (locked) { + UNLOCK (&fd->inode->lock); + locked = _gf_false; + } + STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, &preop, &postop, - NULL); + rsp_xdata); + if (rsp_xdata) + dict_unref (rsp_xdata); return 0; } @@ -2280,6 +2642,33 @@ out: } +int +posix_batch_fsync (call_frame_t *frame, xlator_t *this, + fd_t *fd, int datasync, dict_t *xdata) +{ + call_stub_t *stub = NULL; + struct posix_private *priv = NULL; + + priv = this->private; + + stub = fop_fsync_stub (frame, default_fsync, fd, datasync, xdata); + if (!stub) { + STACK_UNWIND_STRICT (fsync, frame, -1, ENOMEM, 0, 0, 0); + return 0; + } + + pthread_mutex_lock (&priv->fsync_mutex); + { + list_add_tail (&stub->list, &priv->fsyncs); + priv->fsync_queue_count++; + pthread_cond_signal (&priv->fsync_cond); + } + pthread_mutex_unlock (&priv->fsync_mutex); + + return 0; +} + + int32_t posix_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync, dict_t *xdata) @@ -2291,6 +2680,7 @@ posix_fsync (call_frame_t *frame, xlator_t *this, int ret = -1; struct iatt preop = {0,}; struct iatt postop = {0,}; + struct posix_private *priv = NULL; DECLARE_OLD_FS_ID_VAR; @@ -2306,6 +2696,12 @@ posix_fsync (call_frame_t *frame, xlator_t *this, goto out; #endif + priv = this->private; + if (priv->batch_fsync_mode && xdata && dict_get (xdata, "batch-fsync")) { + posix_batch_fsync (frame, this, fd, datasync, xdata); + return 0; + } + ret = posix_fd_ctx_get (fd, this, &pfd); if (ret < 0) { op_errno = -ret; @@ -2367,6 +2763,17 @@ out: } static int gf_posix_xattr_enotsup_log; +static int +_handle_setxattr_keyvalue_pair (dict_t *d, char *k, data_t *v, + void *tmp) +{ + posix_xattr_filler_t *filler = NULL; + + filler = tmp; + + return posix_handle_pair (filler->this, filler->real_path, k, v, + filler->flags); +} int32_t posix_setxattr (call_frame_t *frame, xlator_t *this, @@ -2375,8 +2782,8 @@ posix_setxattr (call_frame_t *frame, xlator_t *this, int32_t op_ret = -1; int32_t op_errno = 0; char * real_path = NULL; - data_pair_t * trav = NULL; - int ret = -1; + + posix_xattr_filler_t filler = {0,}; DECLARE_OLD_FS_ID_VAR; SET_FS_ID (frame->root->uid, frame->root->gid); @@ -2391,18 +2798,13 @@ posix_setxattr (call_frame_t *frame, xlator_t *this, op_ret = -1; dict_del (dict, GFID_XATTR_KEY); - trav = dict->members_list; - - while (trav) { - ret = posix_handle_pair (this, real_path, trav, flags); - if (ret < 0) { - op_errno = -ret; - goto out; - } - trav = trav->next; - } - - op_ret = 0; + filler.real_path = real_path; + filler.this = this; + filler.flags = flags; + op_ret = dict_foreach (dict, _handle_setxattr_keyvalue_pair, + &filler); + if (op_ret < 0) + op_errno = -op_ret; out: SET_TO_OLD_FS_ID (); @@ -2412,6 +2814,53 @@ out: return 0; } + +int +posix_xattr_get_real_filename (call_frame_t *frame, xlator_t *this, loc_t *loc, + const char *key, dict_t *dict, dict_t *xdata) +{ + char *real_path = NULL; + struct dirent *dirent = NULL; + DIR *fd = NULL; + const char *fname = NULL; + char *found = NULL; + int ret = -1; + int op_ret = -1; + + MAKE_INODE_HANDLE (real_path, this, loc, NULL); + + fd = opendir (real_path); + if (!fd) + return -errno; + + fname = key + strlen (GF_XATTR_GET_REAL_FILENAME_KEY); + + while ((dirent = readdir (fd))) { + if (strcasecmp (dirent->d_name, fname) == 0) { + found = gf_strdup (dirent->d_name); + if (!found) { + closedir (fd); + return -ENOMEM; + } + break; + } + } + + closedir (fd); + + if (!found) + return -ENOENT; + + ret = dict_set_dynstr (dict, (char *)key, found); + if (ret) { + GF_FREE (found); + return -ENOMEM; + } + ret = strlen (found) + 1; + + return ret; +} + /** * posix_getxattr - this function returns a dictionary with all the * key:value pair present as xattr. used for @@ -2466,9 +2915,29 @@ posix_getxattr (call_frame_t *frame, xlator_t *this, dict = dict_new (); if (!dict) { + op_errno = ENOMEM; goto out; } + if (loc->inode && name && + (strncmp (name, GF_XATTR_GET_REAL_FILENAME_KEY, + strlen (GF_XATTR_GET_REAL_FILENAME_KEY)) == 0)) { + ret = posix_xattr_get_real_filename (frame, this, loc, + name, dict, xdata); + if (ret < 0) { + op_ret = -1; + op_errno = -ret; + gf_log (this->name, (op_errno == ENOENT) ? + GF_LOG_DEBUG : GF_LOG_WARNING, + "Failed to get real filename (%s, %s): %s", + loc->path, name, strerror (op_errno)); + goto out; + } + + size = ret; + goto done; + } + if (loc->inode && name && !strcmp (name, GLUSTERFS_OPEN_FD_COUNT)) { if (!list_empty (&loc->inode->fd_list)) { ret = dict_set_uint32 (dict, (char *)name, 1); @@ -2492,8 +2961,13 @@ posix_getxattr (call_frame_t *frame, xlator_t *this, else rpath = real_path; - (void) snprintf (host_buf, 1024, "<POSIX(%s):%s:%s>", - priv->base_path, priv->hostname, rpath); + (void) snprintf (host_buf, 1024, + "<POSIX(%s):%s:%s>", priv->base_path, + ((priv->node_uuid_pathinfo + && !uuid_is_null(priv->glusterd_uuid)) + ? uuid_utoa (priv->glusterd_uuid) + : priv->hostname), + rpath); dyn_rpath = gf_strdup (host_buf); if (!dyn_rpath) { @@ -2569,6 +3043,11 @@ posix_getxattr (call_frame_t *frame, xlator_t *this, "supported (try remounting" " brick with 'user_xattr' " "flag)"); + } else if (op_errno == ENOATTR || + op_errno == ENODATA) { + gf_log (this->name, GF_LOG_DEBUG, + "No such attribute:%s for file %s", + key, real_path); } else { gf_log (this->name, GF_LOG_ERROR, "getxattr failed on %s: %s (%s)", @@ -2886,6 +3365,17 @@ out: return 0; } +static int +_handle_fsetxattr_keyvalue_pair (dict_t *d, char *k, data_t *v, + void *tmp) +{ + posix_xattr_filler_t *filler = NULL; + + filler = tmp; + + return posix_fhandle_pair (filler->this, filler->fd, k, v, + filler->flags); +} int32_t posix_fsetxattr (call_frame_t *frame, xlator_t *this, @@ -2895,8 +3385,9 @@ posix_fsetxattr (call_frame_t *frame, xlator_t *this, int32_t op_errno = 0; struct posix_fd * pfd = NULL; int _fd = -1; - data_pair_t * trav = NULL; - int ret = -1; + int ret = -1; + + posix_xattr_filler_t filler = {0,}; DECLARE_OLD_FS_ID_VAR; SET_FS_ID (frame->root->uid, frame->root->gid); @@ -2917,18 +3408,13 @@ posix_fsetxattr (call_frame_t *frame, xlator_t *this, dict_del (dict, GFID_XATTR_KEY); - trav = dict->members_list; - - while (trav) { - ret = posix_fhandle_pair (this, _fd, trav, flags); - if (ret < 0) { - op_errno = -ret; - goto out; - } - trav = trav->next; - } - - op_ret = 0; + filler.fd = _fd; + filler.this = this; + filler.flags = flags; + op_ret = dict_foreach (dict, _handle_fsetxattr_keyvalue_pair, + &filler); + if (op_ret < 0) + op_errno = -op_ret; out: SET_TO_OLD_FS_ID (); @@ -2938,6 +3424,28 @@ out: return 0; } +int +_posix_remove_xattr (dict_t *dict, char *key, data_t *value, void *data) +{ + int32_t op_ret = 0; + xlator_t *this = NULL; + posix_xattr_filler_t *filler = NULL; + + filler = (posix_xattr_filler_t *) data; + this = filler->this; + + op_ret = sys_lremovexattr (filler->real_path, key); + if (op_ret == -1) { + filler->op_errno = errno; + if (errno != ENOATTR && errno != EPERM) + gf_log (this->name, GF_LOG_ERROR, + "removexattr failed on %s (for %s): %s", + filler->real_path, key, strerror (errno)); + } + + return op_ret; +} + int32_t posix_removexattr (call_frame_t *frame, xlator_t *this, @@ -2946,6 +3454,7 @@ posix_removexattr (call_frame_t *frame, xlator_t *this, int32_t op_ret = -1; int32_t op_errno = 0; char * real_path = NULL; + posix_xattr_filler_t filler = {0,}; DECLARE_OLD_FS_ID_VAR; @@ -2961,6 +3470,22 @@ posix_removexattr (call_frame_t *frame, xlator_t *this, SET_FS_ID (frame->root->uid, frame->root->gid); + /** + * sending an empty key name with xdata containing the + * list of key(s) to be removed implies "bulk remove request" + * for removexattr. + */ + if (name && (strcmp (name, "") == 0) && xdata) { + filler.real_path = real_path; + filler.this = this; + op_ret = dict_foreach (xdata, _posix_remove_xattr, &filler); + if (op_ret) { + op_errno = filler.op_errno; + } + + goto out; + } + op_ret = sys_lremovexattr (real_path, name); if (op_ret == -1) { op_errno = errno; @@ -2988,7 +3513,6 @@ posix_fremovexattr (call_frame_t *frame, xlator_t *this, int32_t op_errno = 0; struct posix_fd * pfd = NULL; int _fd = -1; - uint64_t tmp_pfd = 0; int ret = -1; DECLARE_OLD_FS_ID_VAR; @@ -2999,15 +3523,13 @@ posix_fremovexattr (call_frame_t *frame, xlator_t *this, goto out; } - ret = fd_ctx_get (fd, this, &tmp_pfd); + ret = posix_fd_ctx_get (fd, this, &pfd); if (ret < 0) { op_errno = -ret; gf_log (this->name, GF_LOG_WARNING, "pfd is NULL from fd=%p", fd); goto out; } - pfd = (struct posix_fd *)(long)tmp_pfd; - _fd = pfd->fd; @@ -3085,9 +3607,31 @@ posix_print_xattr (dict_t *this, static void __add_array (int32_t *dest, int32_t *src, int count) { + int i = 0; + int32_t destval = 0; + for (i = 0; i < count; i++) { + destval = ntoh32 (dest[i]); + if (destval == 0xffffffff) + continue; + dest[i] = hton32 (destval + ntoh32 (src[i])); + } +} + +static void +__or_array (int32_t *dest, int32_t *src, int count) +{ + int i = 0; + for (i = 0; i < count; i++) { + dest[i] = hton32 (ntoh32 (dest[i]) | ntoh32 (src[i])); + } +} + +static void +__and_array (int32_t *dest, int32_t *src, int count) +{ int i = 0; for (i = 0; i < count; i++) { - dest[i] = hton32 (ntoh32 (dest[i]) + ntoh32 (src[i])); + dest[i] = hton32 (ntoh32 (dest[i]) & ntoh32 (src[i])); } } @@ -3100,6 +3644,159 @@ __add_long_array (int64_t *dest, int64_t *src, int count) } } +static int +_posix_handle_xattr_keyvalue_pair (dict_t *d, char *k, data_t *v, + void *tmp) +{ + int size = 0; + int count = 0; + int op_ret = 0; + int op_errno = 0; + gf_xattrop_flags_t optype = 0; + char *array = NULL; + inode_t *inode = NULL; + xlator_t *this = NULL; + posix_xattr_filler_t *filler = NULL; + + filler = tmp; + + optype = (gf_xattrop_flags_t)(filler->flags); + this = filler->this; + inode = filler->inode; + + count = v->len; + array = GF_CALLOC (count, sizeof (char), gf_posix_mt_char); + + LOCK (&inode->lock); + { + if (filler->real_path) { + size = sys_lgetxattr (filler->real_path, k, + (char *)array, v->len); + } else { + size = sys_fgetxattr (filler->fd, k, (char *)array, + v->len); + } + + op_errno = errno; + if ((size == -1) && (op_errno != ENODATA) && + (op_errno != ENOATTR)) { + if (op_errno == ENOTSUP) { + GF_LOG_OCCASIONALLY(gf_posix_xattr_enotsup_log, + this->name, GF_LOG_WARNING, + "Extended attributes not " + "supported by filesystem"); + } else if (op_errno != ENOENT || + !posix_special_xattr (marker_xattrs, + k)) { + if (filler->real_path) + gf_log (this->name, GF_LOG_ERROR, + "getxattr failed on %s while doing " + "xattrop: Key:%s (%s)", + filler->real_path, + k, strerror (op_errno)); + else + gf_log (this->name, GF_LOG_ERROR, + "fgetxattr failed on fd=%d while doing " + "xattrop: Key:%s (%s)", + filler->fd, + k, strerror (op_errno)); + } + + op_ret = -1; + goto unlock; + } + + switch (optype) { + + case GF_XATTROP_ADD_ARRAY: + __add_array ((int32_t *) array, (int32_t *) v->data, + v->len / 4); + break; + + case GF_XATTROP_ADD_ARRAY64: + __add_long_array ((int64_t *) array, (int64_t *) v->data, + v->len / 8); + break; + + case GF_XATTROP_OR_ARRAY: + __or_array ((int32_t *) array, + (int32_t *) v->data, + v->len / 4); + break; + + case GF_XATTROP_AND_ARRAY: + __and_array ((int32_t *) array, + (int32_t *) v->data, + v->len / 4); + break; + + default: + gf_log (this->name, GF_LOG_ERROR, + "Unknown xattrop type (%d) on %s. Please send " + "a bug report to gluster-devel@nongnu.org", + optype, filler->real_path); + op_ret = -1; + op_errno = EINVAL; + goto unlock; + } + + if (filler->real_path) { + size = sys_lsetxattr (filler->real_path, k, array, + v->len, 0); + } else { + size = sys_fsetxattr (filler->fd, k, (char *)array, + v->len, 0); + } + } +unlock: + UNLOCK (&inode->lock); + + if (op_ret == -1) + goto out; + + op_errno = errno; + if (size == -1) { + if (filler->real_path) + gf_log (this->name, GF_LOG_ERROR, + "setxattr failed on %s while doing xattrop: " + "key=%s (%s)", filler->real_path, + k, strerror (op_errno)); + else + gf_log (this->name, GF_LOG_ERROR, + "fsetxattr failed on fd=%d while doing xattrop: " + "key=%s (%s)", filler->fd, + k, strerror (op_errno)); + + op_ret = -1; + goto out; + } else { + size = dict_set_bin (d, k, array, v->len); + + if (size != 0) { + if (filler->real_path) + gf_log (this->name, GF_LOG_DEBUG, + "dict_set_bin failed (path=%s): " + "key=%s (%s)", filler->real_path, + k, strerror (-size)); + else + gf_log (this->name, GF_LOG_DEBUG, + "dict_set_bin failed (fd=%d): " + "key=%s (%s)", filler->fd, + k, strerror (-size)); + + op_ret = -1; + op_errno = EINVAL; + goto out; + } + array = NULL; + } + + array = NULL; + +out: + return op_ret; +} + /** * xattrop - xattr operations - for internal use by GlusterFS * @optype: ADD_ARRAY: @@ -3111,36 +3808,24 @@ int do_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd, gf_xattrop_flags_t optype, dict_t *xattr) { - char *real_path = NULL; - char *array = NULL; - int size = 0; - int count = 0; - - int op_ret = 0; - int op_errno = 0; - - int ret = 0; - int _fd = -1; - struct posix_fd *pfd = NULL; - - data_pair_t *trav = NULL; - - char * path = NULL; - inode_t * inode = NULL; + int op_ret = 0; + int op_errno = 0; + int _fd = -1; + char *real_path = NULL; + struct posix_fd *pfd = NULL; + inode_t *inode = NULL; + posix_xattr_filler_t filler = {0,}; VALIDATE_OR_GOTO (frame, out); VALIDATE_OR_GOTO (xattr, out); VALIDATE_OR_GOTO (this, out); - trav = xattr->members_list; - if (fd) { - ret = posix_fd_ctx_get (fd, this, &pfd); - if (ret < 0) { + op_ret = posix_fd_ctx_get (fd, this, &pfd); + if (op_ret < 0) { gf_log (this->name, GF_LOG_WARNING, "failed to get pfd from fd=%p", fd); - op_ret = -1; op_errno = EBADFD; goto out; } @@ -3151,136 +3836,21 @@ do_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd, MAKE_INODE_HANDLE (real_path, this, loc, NULL); if (real_path) { - path = gf_strdup (real_path); inode = loc->inode; } else if (fd) { inode = fd->inode; } - while (trav && inode) { - count = trav->value->len; - array = GF_CALLOC (count, sizeof (char), - gf_posix_mt_char); - - LOCK (&inode->lock); - { - if (loc) { - size = sys_lgetxattr (real_path, trav->key, (char *)array, - trav->value->len); - } else { - size = sys_fgetxattr (_fd, trav->key, (char *)array, - trav->value->len); - } - - op_errno = errno; - if ((size == -1) && (op_errno != ENODATA) && - (op_errno != ENOATTR)) { - if (op_errno == ENOTSUP) { - GF_LOG_OCCASIONALLY(gf_posix_xattr_enotsup_log, - this->name,GF_LOG_WARNING, - "Extended attributes not " - "supported by filesystem"); - } else if (op_errno != ENOENT || - !posix_special_xattr (marker_xattrs, - trav->key)) { - if (loc) - gf_log (this->name, GF_LOG_ERROR, - "getxattr failed on %s while doing " - "xattrop: Key:%s (%s)", path, - trav->key, strerror (op_errno)); - else - gf_log (this->name, GF_LOG_ERROR, - "fgetxattr failed on fd=%d while doing " - "xattrop: Key:%s (%s)", _fd, - trav->key, strerror (op_errno)); - } - - op_ret = -1; - goto unlock; - } - - switch (optype) { + filler.this = this; + filler.fd = _fd; + filler.real_path = real_path; + filler.flags = (int)optype; + filler.inode = inode; - case GF_XATTROP_ADD_ARRAY: - __add_array ((int32_t *) array, (int32_t *) trav->value->data, - trav->value->len / 4); - break; - - case GF_XATTROP_ADD_ARRAY64: - __add_long_array ((int64_t *) array, (int64_t *) trav->value->data, - trav->value->len / 8); - break; - - default: - gf_log (this->name, GF_LOG_ERROR, - "Unknown xattrop type (%d) on %s. Please send " - "a bug report to gluster-devel@nongnu.org", - optype, path); - op_ret = -1; - op_errno = EINVAL; - goto unlock; - } - - if (loc) { - size = sys_lsetxattr (real_path, trav->key, array, - trav->value->len, 0); - } else { - size = sys_fsetxattr (_fd, trav->key, (char *)array, - trav->value->len, 0); - } - } - unlock: - UNLOCK (&inode->lock); - - if (op_ret == -1) - goto out; - - op_errno = errno; - if (size == -1) { - if (loc) - gf_log (this->name, GF_LOG_ERROR, - "setxattr failed on %s while doing xattrop: " - "key=%s (%s)", path, - trav->key, strerror (op_errno)); - else - gf_log (this->name, GF_LOG_ERROR, - "fsetxattr failed on fd=%d while doing xattrop: " - "key=%s (%s)", _fd, - trav->key, strerror (op_errno)); - - op_ret = -1; - goto out; - } else { - size = dict_set_bin (xattr, trav->key, array, - trav->value->len); - - if (size != 0) { - if (loc) - gf_log (this->name, GF_LOG_DEBUG, - "dict_set_bin failed (path=%s): " - "key=%s (%s)", path, - trav->key, strerror (-size)); - else - gf_log (this->name, GF_LOG_DEBUG, - "dict_set_bin failed (fd=%d): " - "key=%s (%s)", _fd, - trav->key, strerror (-size)); - - op_ret = -1; - op_errno = EINVAL; - goto out; - } - array = NULL; - } - - array = NULL; - trav = trav->next; - } + op_ret = dict_foreach (xattr, _posix_handle_xattr_keyvalue_pair, + &filler); out: - GF_FREE (array); - - GF_FREE (path); STACK_UNWIND_STRICT (xattrop, frame, op_ret, op_errno, xattr, NULL); return 0; @@ -3685,6 +4255,8 @@ posix_readdirp_fill (xlator_t *this, fd_t *fd, gf_dirent_t *entries, dict_t *dic struct iatt stbuf = {0, }; uuid_t gfid; + if (list_empty(&entries->list)) + return 0; itable = fd->inode->table; @@ -3773,8 +4345,23 @@ posix_do_readdir (call_frame_t *frame, xlator_t *this, */ ret = dict_get_int32 (dict, GF_READDIR_SKIP_DIRS, &skip_dirs); - count = posix_fill_readdir (fd, dir, off, size, &entries, this, - skip_dirs); + LOCK (&fd->lock); + { + /* posix_fill_readdir performs multiple separate individual + readdir() calls to fill up the buffer. + + In case of NFS where the same anonymous FD is shared between + different applications, reading a common directory can + result in the anonymous fd getting re-used unsafely between + the two readdir requests (in two different io-threads). + + It would also help, in the future, to replace the loop + around readdir() with a single large getdents() call. + */ + count = posix_fill_readdir (fd, dir, off, size, &entries, this, + skip_dirs); + } + UNLOCK (&fd->lock); /* pick ENOENT to indicate EOF */ op_errno = errno; @@ -3849,23 +4436,26 @@ int32_t posix_rchecksum (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, int32_t len, dict_t *xdata) { - char *buf = NULL; - int _fd = -1; - struct posix_fd *pfd = NULL; - int op_ret = -1; - int op_errno = 0; - int ret = 0; - int32_t weak_checksum = 0; - unsigned char strong_checksum[MD5_DIGEST_LENGTH]; + char *alloc_buf = NULL; + char *buf = NULL; + int _fd = -1; + struct posix_fd *pfd = NULL; + int op_ret = -1; + int op_errno = 0; + int ret = 0; + int32_t weak_checksum = 0; + unsigned char strong_checksum[MD5_DIGEST_LENGTH] = {0}; + struct posix_private *priv = NULL; VALIDATE_OR_GOTO (frame, out); VALIDATE_OR_GOTO (this, out); VALIDATE_OR_GOTO (fd, out); + priv = this->private; memset (strong_checksum, 0, MD5_DIGEST_LENGTH); - buf = GF_CALLOC (1, len, gf_posix_mt_char); - if (!buf) { + alloc_buf = _page_aligned_alloc (len, &buf); + if (!alloc_buf) { op_errno = ENOMEM; goto out; } @@ -3880,15 +4470,25 @@ posix_rchecksum (call_frame_t *frame, xlator_t *this, _fd = pfd->fd; - ret = pread (_fd, buf, len, offset); - if (ret < 0) { - gf_log (this->name, GF_LOG_WARNING, - "pread of %d bytes returned %d (%s)", - len, ret, strerror (errno)); + LOCK (&fd->lock); + { + if (priv->aio_capable && priv->aio_init_done) + __posix_fd_set_odirect (fd, pfd, 0, offset, len); + + ret = pread (_fd, buf, len, offset); + if (ret < 0) { + gf_log (this->name, GF_LOG_WARNING, + "pread of %d bytes returned %d (%s)", + len, ret, strerror (errno)); + + op_errno = errno; + } - op_errno = errno; - goto out; } + UNLOCK (&fd->lock); + + if (ret < 0) + goto out; weak_checksum = gf_rsync_weak_checksum ((unsigned char *) buf, (size_t) len); gf_rsync_strong_checksum ((unsigned char *) buf, (size_t) len, (unsigned char *) strong_checksum); @@ -3898,7 +4498,7 @@ out: STACK_UNWIND_STRICT (rchecksum, frame, op_ret, op_errno, weak_checksum, strong_checksum, NULL); - GF_FREE (buf); + GF_FREE (alloc_buf); return 0; } @@ -3947,15 +4547,71 @@ mem_acct_init (xlator_t *this) return ret; } +static int +posix_set_owner (xlator_t *this, uid_t uid, gid_t gid) +{ + struct posix_private *priv = NULL; + int ret = -1; + + priv = this->private; + + ret = sys_chown (priv->base_path, uid, gid); + if (ret) + gf_log (this->name, GF_LOG_ERROR, "Failed to set " + "uid/gid for brick path %s, %s", + priv->base_path, strerror (errno)); + + return ret; +} + + +static int +set_batch_fsync_mode (struct posix_private *priv, const char *str) +{ + if (strcmp (str, "none") == 0) + priv->batch_fsync_mode = BATCH_NONE; + else if (strcmp (str, "syncfs") == 0) + priv->batch_fsync_mode = BATCH_SYNCFS; + else if (strcmp (str, "syncfs-single-fsync") == 0) + priv->batch_fsync_mode = BATCH_SYNCFS_SINGLE_FSYNC; + else if (strcmp (str, "syncfs-reverse-fsync") == 0) + priv->batch_fsync_mode = BATCH_SYNCFS_REVERSE_FSYNC; + else if (strcmp (str, "reverse-fsync") == 0) + priv->batch_fsync_mode = BATCH_REVERSE_FSYNC; + else + return -1; + + return 0; +} + int reconfigure (xlator_t *this, dict_t *options) { int ret = -1; struct posix_private *priv = NULL; + uid_t uid = -1; + gid_t gid = -1; + char *batch_fsync_mode_str = NULL; priv = this->private; + GF_OPTION_RECONF ("brick-uid", uid, options, uint32, out); + GF_OPTION_RECONF ("brick-gid", gid, options, uint32, out); + posix_set_owner (this, uid, gid); + + GF_OPTION_RECONF ("batch-fsync-delay-usec", priv->batch_fsync_delay_usec, + options, uint32, out); + + GF_OPTION_RECONF ("batch-fsync-mode", batch_fsync_mode_str, + options, str, out); + + if (set_batch_fsync_mode (priv, batch_fsync_mode_str) != 0) { + gf_log (this->name, GF_LOG_ERROR, "Unknown mode string: %s", + batch_fsync_mode_str); + goto out; + } + GF_OPTION_RECONF ("linux-aio", priv->aio_configured, options, bool, out); @@ -3964,6 +4620,20 @@ reconfigure (xlator_t *this, dict_t *options) else posix_aio_off (this); + GF_OPTION_RECONF ("node-uuid-pathinfo", priv->node_uuid_pathinfo, + options, bool, out); + + if (priv->node_uuid_pathinfo && + (uuid_is_null (priv->glusterd_uuid))) { + gf_log (this->name, GF_LOG_INFO, + "glusterd uuid is NULL, pathinfo xattr would" + " fallback to <hostname>:<export>"); + } + + GF_OPTION_RECONF ("health-check-interval", priv->health_check_interval, + options, uint32, out); + posix_spawn_health_check_thread (this); + ret = 0; out: return ret; @@ -3991,6 +4661,9 @@ init (xlator_t *this) uuid_t gfid = {0,}; uuid_t rootgfid = {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1}; char *guuid = NULL; + uid_t uid = -1; + gid_t gid = -1; + char *batch_fsync_mode_str; dir_data = dict_get (this->options, "directory"); @@ -4083,16 +4756,13 @@ init (xlator_t *this) goto out; } } else if ((size == -1) && (errno == ENODATA)) { - /* Using the export for first time */ - size = sys_lsetxattr (dir_data->data, - "trusted.glusterfs.volume-id", - dict_uuid, 16, 0); - if (size == -1) { + gf_log (this->name, GF_LOG_ERROR, - "failed to set volume id on export"); + "Extended attribute trusted.glusterfs." + "volume-id is absent"); ret = -1; goto out; - } + } else if ((size == -1) && (errno != ENODATA)) { /* Wrong 'volume-id' is set, it should be error */ gf_log (this->name, GF_LOG_WARNING, @@ -4146,7 +4816,7 @@ init (xlator_t *this) } } - size = sys_lgetxattr (dir_data->data, "system.posix_acl_access", + size = sys_lgetxattr (dir_data->data, POSIX_ACL_ACCESS_XATTR, NULL, 0); if ((size < 0) && (errno == ENOTSUP)) gf_log (this->name, GF_LOG_WARNING, @@ -4309,6 +4979,10 @@ init (xlator_t *this) _private->aio_init_done = _gf_false; _private->aio_capable = _gf_false; + GF_OPTION_INIT ("brick-uid", uid, uint32, out); + GF_OPTION_INIT ("brick-gid", gid, uint32, out); + posix_set_owner (this, uid, gid); + GF_OPTION_INIT ("linux-aio", _private->aio_configured, bool, out); if (_private->aio_configured) { @@ -4322,11 +4996,48 @@ init (xlator_t *this) } } + GF_OPTION_INIT ("node-uuid-pathinfo", + _private->node_uuid_pathinfo, bool, out); + if (_private->node_uuid_pathinfo && + (uuid_is_null (_private->glusterd_uuid))) { + gf_log (this->name, GF_LOG_INFO, + "glusterd uuid is NULL, pathinfo xattr would" + " fallback to <hostname>:<export>"); + } + + _private->health_check_active = _gf_false; + GF_OPTION_INIT ("health-check-interval", + _private->health_check_interval, uint32, out); + if (_private->health_check_interval) + posix_spawn_health_check_thread (this); + pthread_mutex_init (&_private->janitor_lock, NULL); pthread_cond_init (&_private->janitor_cond, NULL); INIT_LIST_HEAD (&_private->janitor_fds); posix_spawn_janitor_thread (this); + + pthread_mutex_init (&_private->fsync_mutex, NULL); + pthread_cond_init (&_private->fsync_cond, NULL); + INIT_LIST_HEAD (&_private->fsyncs); + + ret = gf_thread_create (&_private->fsyncer, NULL, posix_fsyncer, this); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "fsyncer thread" + " creation failed (%s)", strerror (errno)); + goto out; + } + + GF_OPTION_INIT ("batch-fsync-mode", batch_fsync_mode_str, str, out); + + if (set_batch_fsync_mode (_private, batch_fsync_mode_str) != 0) { + gf_log (this->name, GF_LOG_ERROR, "Unknown mode string: %s", + batch_fsync_mode_str); + goto out; + } + + GF_OPTION_INIT ("batch-fsync-delay-usec", _private->batch_fsync_delay_usec, + uint32, out); out: return ret; } @@ -4392,6 +5103,9 @@ struct xlator_fops fops = { .fxattrop = posix_fxattrop, .setattr = posix_setattr, .fsetattr = posix_fsetattr, + .fallocate = _posix_fallocate, + .discard = posix_discard, + .zerofill = posix_zerofill, }; struct xlator_cbks cbks = { @@ -4420,10 +5134,59 @@ struct volume_options options[] = { { .key = {"glusterd-uuid"}, .type = GF_OPTION_TYPE_STR }, { - .key = {"linux-aio"}, - .type = GF_OPTION_TYPE_BOOL, - .default_value = "off", - .description = "Support for native Linux AIO" + .key = {"linux-aio"}, + .type = GF_OPTION_TYPE_BOOL, + .default_value = "off", + .description = "Support for native Linux AIO" + }, + { + .key = {"brick-uid"}, + .type = GF_OPTION_TYPE_INT, + .min = 0, + .validate = GF_OPT_VALIDATE_MIN, + .description = "Support for setting uid of brick's owner" + }, + { + .key = {"brick-gid"}, + .type = GF_OPTION_TYPE_INT, + .min = 0, + .validate = GF_OPT_VALIDATE_MIN, + .description = "Support for setting gid of brick's owner" + }, + { .key = {"node-uuid-pathinfo"}, + .type = GF_OPTION_TYPE_BOOL, + .default_value = "off", + .description = "return glusterd's node-uuid in pathinfo xattr" + " string instead of hostname" + }, + { + .key = {"health-check-interval"}, + .type = GF_OPTION_TYPE_INT, + .min = 0, + .default_value = "30", + .validate = GF_OPT_VALIDATE_MIN, + .description = "Interval in seconds for a filesystem health check, " + "set to 0 to disable" + }, + { .key = {"batch-fsync-mode"}, + .type = GF_OPTION_TYPE_STR, + .default_value = "reverse-fsync", + .description = "Possible values:\n" + "\t- syncfs: Perform one syncfs() on behalf oa batch" + "of fsyncs.\n" + "\t- syncfs-single-fsync: Perform one syncfs() on behalf of a batch" + " of fsyncs and one fsync() per batch.\n" + "\t- syncfs-reverse-fsync: Preform one syncfs() on behalf of a batch" + " of fsyncs and fsync() each file in the batch in reverse order.\n" + " in reverse order.\n" + "\t- reverse-fsync: Perform fsync() of each file in the batch in" + " reverse order." + }, + { .key = {"batch-fsync-delay-usec"}, + .type = GF_OPTION_TYPE_INT, + .default_value = "0", + .description = "Num of usecs to wait for aggregating fsync" + " requests", }, { .key = {NULL} } }; diff --git a/xlators/storage/posix/src/posix.h b/xlators/storage/posix/src/posix.h index 669bd5cbb..3121db271 100644 --- a/xlators/storage/posix/src/posix.h +++ b/xlators/storage/posix/src/posix.h @@ -1,13 +1,12 @@ /* - Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> - This file is part of GlusterFS. + Copyright (c) 2006-2012 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 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 _POSIX_H #define _POSIX_H @@ -44,12 +43,15 @@ #include "timer.h" #include "posix-mem-types.h" #include "posix-handle.h" +#include "call-stub.h" #ifdef HAVE_LIBAIO #include <libaio.h> #include "posix-aio.h" #endif +#define VECTOR_SIZE 64 * 1024 /* vector size 64KB*/ +#define MAX_NO_VECT 1024 /** * posix_fd - internal structure common to file and directory fd's */ @@ -58,7 +60,6 @@ struct posix_fd { int fd; /* fd returned by the kernel */ int32_t flags; /* flags for open/creat */ DIR * dir; /* handle returned by the kernel */ - int flushwrites; int odirect; struct list_head list; /* to add to the janitor list */ }; @@ -126,8 +127,45 @@ struct posix_private { io_context_t ctxp; pthread_t aiothread; #endif + + /* node-uuid in pathinfo xattr */ + gf_boolean_t node_uuid_pathinfo; + + pthread_t fsyncer; + struct list_head fsyncs; + pthread_mutex_t fsync_mutex; + pthread_cond_t fsync_cond; + int fsync_queue_count; + + enum { + BATCH_NONE = 0, + BATCH_SYNCFS, + BATCH_SYNCFS_SINGLE_FSYNC, + BATCH_REVERSE_FSYNC, + BATCH_SYNCFS_REVERSE_FSYNC + } batch_fsync_mode; + + uint32_t batch_fsync_delay_usec; + + /* seconds to sleep between health checks */ + uint32_t health_check_interval; + pthread_t health_check; + gf_boolean_t health_check_active; }; +typedef struct { + xlator_t *this; + const char *real_path; + dict_t *xattr; + struct iatt *stbuf; + loc_t *loc; + inode_t *inode; /* for all do_xattrop() key handling */ + int fd; + int flags; + int32_t op_errno; +} posix_xattr_filler_t; + + #define POSIX_BASE_PATH(this) (((struct posix_private *)this->private)->base_path) #define POSIX_BASE_PATH_LEN(this) (((struct posix_private *)this->private)->base_path_length) @@ -142,23 +180,29 @@ int posix_pstat (xlator_t *this, uuid_t gfid, const char *real_path, struct iatt *iatt); dict_t *posix_lookup_xattr_fill (xlator_t *this, const char *path, loc_t *loc, dict_t *xattr, struct iatt *buf); -int posix_handle_pair (xlator_t *this, const char *real_path, - data_pair_t *trav, int flags); -int posix_fhandle_pair (xlator_t *this, int fd, data_pair_t *trav, int flags); +int posix_handle_pair (xlator_t *this, const char *real_path, char *key, + data_t *value, int flags); +int posix_fhandle_pair (xlator_t *this, int fd, char *key, data_t *value, + int flags); void posix_spawn_janitor_thread (xlator_t *this); int posix_get_file_contents (xlator_t *this, uuid_t pargfid, const char *name, char **contents); -int posix_set_file_contents (xlator_t *this, const char *path, - data_pair_t *trav, int flags); +int posix_set_file_contents (xlator_t *this, const char *path, char *key, + data_t *value, int flags); int posix_acl_xattr_set (xlator_t *this, const char *path, dict_t *xattr_req); -int posix_gfid_heal (xlator_t *this, const char *path, dict_t *xattr_req); +int posix_gfid_heal (xlator_t *this, const char *path, loc_t *loc, dict_t *xattr_req); int posix_entry_create_xattr_set (xlator_t *this, const char *path, dict_t *dict); int posix_fd_ctx_get (fd_t *fd, xlator_t *this, struct posix_fd **pfd); -int posix_fd_ctx_get_off (fd_t *fd, xlator_t *this, struct posix_fd **pfd, - off_t off); void posix_fill_ino_from_gfid (xlator_t *this, struct iatt *buf); gf_boolean_t posix_special_xattr (char **pattern, char *key); + +void +__posix_fd_set_odirect (fd_t *fd, struct posix_fd *pfd, int opflags, + off_t offset, size_t size); +void posix_spawn_health_check_thread (xlator_t *this); + +void *posix_fsyncer (void *); #endif /* _POSIX_H */ |