diff options
Diffstat (limited to 'xlators/cluster/afr/src')
29 files changed, 8025 insertions, 3881 deletions
diff --git a/xlators/cluster/afr/src/Makefile.am b/xlators/cluster/afr/src/Makefile.am index 95db5dd96..35d18a6c0 100644 --- a/xlators/cluster/afr/src/Makefile.am +++ b/xlators/cluster/afr/src/Makefile.am @@ -7,11 +7,11 @@ afr_common_source = afr-dir-read.c afr-dir-write.c afr-inode-read.c \ afr-self-heal-algorithm.c afr-lk-common.c afr-self-heald.c \ $(top_builddir)/xlators/lib/src/libxlator.c -afr_la_LDFLAGS = -module -avoidversion +afr_la_LDFLAGS = -module -avoid-version afr_la_SOURCES = $(afr_common_source) afr.c afr_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la -pump_la_LDFLAGS = -module -avoidversion +pump_la_LDFLAGS = -module -avoid-version pump_la_SOURCES = $(afr_common_source) pump.c pump_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la @@ -21,9 +21,11 @@ noinst_HEADERS = afr.h afr-transaction.h afr-inode-write.h afr-inode-read.h \ afr-self-heald.h $(top_builddir)/xlators/lib/src/libxlator.h \ $(top_builddir)/glusterfsd/src/glusterfsd.h -AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \ +AM_CPPFLAGS = $(GF_CPPFLAGS) \ -I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/xlators/lib/src \ - -I$(top_srcdir)/rpc/rpc-lib/src -shared -nostartfiles $(GF_CFLAGS) + -I$(top_srcdir)/rpc/rpc-lib/src + +AM_CFLAGS = -Wall $(GF_CFLAGS) CLEANFILES = diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c index 9874b2619..af01f2ef2 100644 --- a/xlators/cluster/afr/src/afr-common.c +++ b/xlators/cluster/afr/src/afr-common.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2007-2011 Gluster, Inc. <http://www.gluster.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> @@ -58,10 +49,9 @@ #include "afr-self-heald.h" #include "pump.h" -#define AFR_ICTX_OPENDIR_DONE_MASK 0x0000000200000000ULL -#define AFR_ICTX_SPLIT_BRAIN_MASK 0x0000000100000000ULL +#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); @@ -92,6 +82,11 @@ afr_xattr_req_prepare (xlator_t *this, dict_t *xattr_req, const char *path) 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 @@ -123,6 +118,13 @@ afr_lookup_xattr_req_prepare (afr_local_t *local, xlator_t *this, 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, @@ -200,60 +202,86 @@ out: return ret; } -afr_inode_ctx_t* -afr_inode_ctx_get_from_addr (uint64_t addr, int32_t child_count) +void +afr_inode_ctx_destroy (afr_inode_ctx_t *ctx) { - int ret = -1; - afr_inode_ctx_t *ctx = NULL; - size_t size = 0; + if (!ctx) + return; + GF_FREE (ctx->fresh_children); + GF_FREE (ctx); +} - GF_ASSERT (child_count > 0); +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; - if (!addr) { - ctx = GF_CALLOC (1, sizeof (*ctx), - gf_afr_mt_inode_ctx_t); - if (!ctx) - goto out; - size = sizeof (*ctx->fresh_children); - ctx->fresh_children = GF_CALLOC (child_count, size, - gf_afr_mt_int32_t); - if (!ctx->fresh_children) - goto out; - } else { - ctx = (afr_inode_ctx_t*) (long) addr; + 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; } - ret = 0; + 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: - if (ret && ctx) { - if (ctx->fresh_children) - GF_FREE (ctx->fresh_children); - GF_FREE (ctx); - ctx = NULL; + 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 (xlator_t *this, inode_t *inode, afr_inode_params_t *params) +afr_inode_get_ctx_params (xlator_t *this, inode_t *inode, + afr_inode_params_t *params) { GF_ASSERT (inode); GF_ASSERT (params); - int ret = 0; afr_inode_ctx_t *ctx = NULL; afr_private_t *priv = NULL; int i = 0; - uint64_t ctx_addr = 0; int32_t read_child = -1; int32_t *fresh_children = NULL; priv = this->private; LOCK (&inode->lock); { - ret = __inode_ctx_get (inode, this, &ctx_addr); - if (ret < 0) - goto unlock; - ctx = afr_inode_ctx_get_from_addr (ctx_addr, priv->child_count); + ctx = __afr_inode_ctx_get (inode, this); if (!ctx) goto unlock; switch (params->op) { @@ -272,12 +300,6 @@ afr_inode_get_ctx (xlator_t *this, inode_t *inode, afr_inode_params_t *params) if (ctx->masks & AFR_ICTX_OPENDIR_DONE_MASK) params->u.value = _gf_true; break; - case AFR_INODE_GET_SPLIT_BRAIN: - params->u.value = _gf_false; - if (ctx->masks & AFR_ICTX_SPLIT_BRAIN_MASK) - params->u.value = _gf_true; - ; - break; default: GF_ASSERT (0); break; @@ -290,11 +312,16 @@ unlock: gf_boolean_t afr_is_split_brain (xlator_t *this, inode_t *inode) { - afr_inode_params_t params = {0}; + afr_inode_ctx_t *ctx = NULL; + gf_boolean_t spb = _gf_false; - params.op = AFR_INODE_GET_SPLIT_BRAIN; - afr_inode_get_ctx (this, inode, ¶ms); - return params.u.value; + 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 @@ -303,11 +330,10 @@ 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 (this, inode, ¶ms); + afr_inode_get_ctx_params (this, inode, ¶ms); return params.u.value; } - int32_t afr_inode_get_read_ctx (xlator_t *this, inode_t *inode, int32_t *fresh_children) { @@ -315,7 +341,7 @@ afr_inode_get_read_ctx (xlator_t *this, inode_t *inode, int32_t *fresh_children) params.op = AFR_INODE_GET_READ_CTX; params.u.read_ctx.children = fresh_children; - afr_inode_get_ctx (this, inode, ¶ms); + afr_inode_get_ctx_params (this, inode, ¶ms); return params.u.read_ctx.read_child; } @@ -377,31 +403,14 @@ afr_inode_ctx_set_opendir_done (afr_inode_ctx_t *ctx) } void -afr_inode_ctx_set_splitbrain (afr_inode_ctx_t *ctx, gf_boolean_t set) -{ - uint64_t remaining_mask = 0; - uint64_t mask = 0; - - if (set) { - remaining_mask = (~AFR_ICTX_SPLIT_BRAIN_MASK & ctx->masks); - mask = (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_SPLIT_BRAIN_MASK); - ctx->masks = remaining_mask | mask; - } else { - ctx->masks = (~AFR_ICTX_SPLIT_BRAIN_MASK & ctx->masks); - } -} - -void -afr_inode_set_ctx (xlator_t *this, inode_t *inode, afr_inode_params_t *params) +afr_inode_set_ctx_params (xlator_t *this, inode_t *inode, + afr_inode_params_t *params) { GF_ASSERT (inode); GF_ASSERT (params); - int ret = 0; afr_inode_ctx_t *ctx = NULL; afr_private_t *priv = NULL; - uint64_t ctx_addr = 0; - gf_boolean_t set = _gf_false; int32_t read_child = -1; int32_t *fresh_children = NULL; int32_t *stale_children = NULL; @@ -409,10 +418,7 @@ afr_inode_set_ctx (xlator_t *this, inode_t *inode, afr_inode_params_t *params) priv = this->private; LOCK (&inode->lock); { - ret = __inode_ctx_get (inode, this, &ctx_addr); - if (ret < 0) - ctx_addr = 0; - ctx = afr_inode_ctx_get_from_addr (ctx_addr, priv->child_count); + ctx = __afr_inode_ctx_get (inode, this); if (!ctx) goto unlock; switch (params->op) { @@ -432,33 +438,26 @@ afr_inode_set_ctx (xlator_t *this, inode_t *inode, afr_inode_params_t *params) case AFR_INODE_SET_OPENDIR_DONE: afr_inode_ctx_set_opendir_done (ctx); break; - case AFR_INODE_SET_SPLIT_BRAIN: - set = params->u.value; - afr_inode_ctx_set_splitbrain (ctx, set); - break; default: GF_ASSERT (0); break; } - 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)); - } } unlock: UNLOCK (&inode->lock); } void -afr_set_split_brain (xlator_t *this, inode_t *inode, gf_boolean_t set) +afr_set_split_brain (xlator_t *this, inode_t *inode, afr_spb_state_t mdata_spb, + afr_spb_state_t data_spb) { - afr_inode_params_t params = {0}; + afr_inode_ctx_t *ctx = NULL; - params.op = AFR_INODE_SET_SPLIT_BRAIN; - params.u.value = set; - afr_inode_set_ctx (this, inode, ¶ms); + 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 @@ -467,7 +466,7 @@ 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 (this, inode, ¶ms); + afr_inode_set_ctx_params (this, inode, ¶ms); } void @@ -486,7 +485,7 @@ afr_inode_set_read_ctx (xlator_t *this, inode_t *inode, int32_t 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 (this, inode, ¶ms); + afr_inode_set_ctx_params (this, inode, ¶ms); } void @@ -499,7 +498,7 @@ afr_inode_rm_stale_children (xlator_t *this, inode_t *inode, params.op = AFR_INODE_RM_STALE_CHILDREN; params.u.read_ctx.children = stale_children; - afr_inode_set_ctx (this, inode, ¶ms); + afr_inode_set_ctx_params (this, inode, ¶ms); } gf_boolean_t @@ -543,6 +542,10 @@ afr_is_read_child (int32_t *success_children, int32_t *sources, 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); @@ -559,29 +562,69 @@ 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) +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 = prev_read_child; + read_child = config_read_child; if (afr_is_read_child (success_children, sources, child_count, read_child)) goto out; - read_child = config_read_child; + 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) @@ -601,7 +644,7 @@ out: 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) + int32_t config_read_child, uuid_t gfid) { int read_child = -1; afr_private_t *priv = NULL; @@ -611,7 +654,8 @@ afr_set_read_ctx_from_policy (xlator_t *this, inode_t *inode, priv->child_count, prev_read_child, config_read_child, - NULL); + NULL, + priv->hash_mode, gfid); if (read_child >= 0) afr_inode_set_read_ctx (this, inode, read_child, fresh_children); @@ -667,8 +711,11 @@ afr_get_call_child (xlator_t *this, unsigned char *child_up, int32_t read_child, GF_ASSERT (call_child); GF_ASSERT (last_index); GF_ASSERT (fresh_children); - GF_ASSERT (read_child >= 0); + if (read_child < 0) { + ret = -EIO; + goto out; + } priv = this->private; *call_child = -1; *last_index = -1; @@ -717,6 +764,13 @@ out: } 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; @@ -725,60 +779,51 @@ afr_local_sh_cleanup (afr_local_t *local, xlator_t *this) sh = &local->self_heal; priv = this->private; - if (sh->buf) - GF_FREE (sh->buf); + if (sh->data_sh_info && strcmp (sh->data_sh_info, "")) + GF_FREE (sh->data_sh_info); - if (sh->parentbufs) - GF_FREE (sh->parentbufs); + 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); - if (sh->xattr) { - afr_reset_xattr (sh->xattr, priv->child_count); - GF_FREE (sh->xattr); - } + afr_xattr_array_destroy (sh->xattr, priv->child_count); - if (sh->child_errno) - GF_FREE (sh->child_errno); + GF_FREE (sh->child_errno); afr_matrix_cleanup (sh->pending_matrix, priv->child_count); afr_matrix_cleanup (sh->delta_matrix, priv->child_count); - if (sh->sources) - GF_FREE (sh->sources); + GF_FREE (sh->sources); - if (sh->success) - GF_FREE (sh->success); + GF_FREE (sh->success); - if (sh->locked_nodes) - GF_FREE (sh->locked_nodes); + GF_FREE (sh->locked_nodes); if (sh->healing_fd) { fd_unref (sh->healing_fd); sh->healing_fd = NULL; } - if (sh->linkname) - GF_FREE ((char *)sh->linkname); + GF_FREE ((char *)sh->linkname); - if (sh->success_children) - GF_FREE (sh->success_children); + GF_FREE (sh->success_children); - if (sh->fresh_children) - GF_FREE (sh->fresh_children); + GF_FREE (sh->fresh_children); - if (sh->fresh_parent_dirs) - GF_FREE (sh->fresh_parent_dirs); + GF_FREE (sh->fresh_parent_dirs); loc_wipe (&sh->parent_loc); loc_wipe (&sh->lookup_loc); - if (sh->checksum) - GF_FREE (sh->checksum); + GF_FREE (sh->checksum); - if (sh->write_needed) - GF_FREE (sh->write_needed); + GF_FREE (sh->write_needed); if (sh->healing_fd) fd_unref (sh->healing_fd); } @@ -787,27 +832,26 @@ afr_local_sh_cleanup (afr_local_t *local, xlator_t *this) void afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this) { - afr_private_t * priv = NULL; + 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); - if (local->internal_lock.locked_nodes) - GF_FREE (local->internal_lock.locked_nodes); - - if (local->internal_lock.inode_locked_nodes) - GF_FREE (local->internal_lock.inode_locked_nodes); + GF_FREE (local->internal_lock.locked_nodes); - if (local->internal_lock.entry_locked_nodes) - GF_FREE (local->internal_lock.entry_locked_nodes); + for (i = 0; local->internal_lock.inodelk[i].domain; i++) { + GF_FREE (local->internal_lock.inodelk[i].locked_nodes); + } - if (local->internal_lock.lower_locked_nodes) - GF_FREE (local->internal_lock.lower_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.child_errno); GF_FREE (local->transaction.eager_lock); GF_FREE (local->transaction.basename); @@ -815,6 +859,8 @@ afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this) loc_wipe (&local->transaction.parent_loc); loc_wipe (&local->transaction.new_parent_loc); + + GF_FREE (local->transaction.postop_piggybacked); } @@ -844,17 +890,13 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this) if (local->dict) dict_unref (local->dict); - if (local->child_up) - GF_FREE (local->child_up); + GF_FREE(local->replies); - if (local->child_errno) - GF_FREE (local->child_errno); + GF_FREE (local->child_up); - if (local->fresh_children) - GF_FREE (local->fresh_children); + GF_FREE (local->child_errno); - if (local->fd_open_on) - GF_FREE (local->fd_open_on); + GF_FREE (local->fresh_children); { /* lookup */ if (local->cont.lookup.xattrs) { @@ -872,27 +914,23 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this) inode_unref (local->cont.lookup.inode); } - if (local->cont.lookup.postparents) - GF_FREE (local->cont.lookup.postparents); + GF_FREE (local->cont.lookup.postparents); - if (local->cont.lookup.bufs) - GF_FREE (local->cont.lookup.bufs); + GF_FREE (local->cont.lookup.bufs); - if (local->cont.lookup.success_children) - GF_FREE (local->cont.lookup.success_children); + GF_FREE (local->cont.lookup.success_children); - if (local->cont.lookup.sources) - GF_FREE (local->cont.lookup.sources); + GF_FREE (local->cont.lookup.sources); + afr_matrix_cleanup (local->cont.lookup.pending_matrix, + priv->child_count); } { /* getxattr */ - if (local->cont.getxattr.name) - GF_FREE (local->cont.getxattr.name); + GF_FREE (local->cont.getxattr.name); } { /* lk */ - if (local->cont.lk.locked_nodes) - GF_FREE (local->cont.lk.locked_nodes); + GF_FREE (local->cont.lk.locked_nodes); } { /* create */ @@ -947,8 +985,7 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this) } { /* opendir */ - if (local->cont.opendir.checksum) - GF_FREE (local->cont.opendir.checksum); + GF_FREE (local->cont.opendir.checksum); } { /* readdirp */ @@ -1043,6 +1080,88 @@ afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *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) { @@ -1087,13 +1206,18 @@ afr_lookup_build_response_params (afr_local_t *local, xlator_t *this) 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; @@ -1109,6 +1233,7 @@ afr_lookup_update_lk_counts (afr_local_t *local, xlator_t *this, uint32_t inodelk_count = 0; uint32_t entrylk_count = 0; int ret = -1; + uint32_t parent_entrylk = 0; GF_ASSERT (local); GF_ASSERT (this); @@ -1124,43 +1249,103 @@ afr_lookup_update_lk_counts (afr_local_t *local, xlator_t *this, &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); - if (afr_sh_has_metadata_pending (xattr, this)) { - local->self_heal.do_metadata_self_heal = _gf_true; - gf_log(this->name, GF_LOG_DEBUG, - "metadata self-heal is pending for %s.", - local->loc.path); - } + priv = this->private; - if (afr_sh_has_entry_pending (xattr, this)) { - local->self_heal.do_entry_self_heal = _gf_true; - gf_log(this->name, GF_LOG_DEBUG, - "entry self-heal is pending for %s.", local->loc.path); - } + 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 (afr_sh_has_data_pending (xattr, this)) { - local->self_heal.do_data_self_heal = _gf_true; - gf_log(this->name, GF_LOG_DEBUG, - "data self-heal is pending for %s.", local->loc.path); + 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_INFO, + gf_log (this->name, GF_LOG_DEBUG, "permissions differ for %s ", local->loc.path); local->self_heal.do_metadata_self_heal = _gf_true; } @@ -1168,27 +1353,45 @@ afr_detect_self_heal_by_iatt (afr_local_t *local, xlator_t *this, if (OWNERSHIP_DIFFERS (buf, lookup_buf)) { /* mismatching permissions */ local->self_heal.do_metadata_self_heal = _gf_true; - gf_log (this->name, GF_LOG_INFO, + 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_INFO, + 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_WARNING, + gf_log (this->name, GF_LOG_DEBUG, "%s: gfid different on subvolume", local->loc.path); } } static void -afr_detect_self_heal_by_lookup_status (afr_local_t *local, xlator_t *this, - gf_boolean_t split_brain) +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); @@ -1199,24 +1402,11 @@ afr_detect_self_heal_by_lookup_status (afr_local_t *local, xlator_t *this, 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_INFO, + gf_log(this->name, GF_LOG_DEBUG, "entries are missing in lookup of %s.", local->loc.path); - //If all self-heals are needed no need to check for other rules - goto out; } - if ((local->success_count > 0) && split_brain && - IA_ISREG (local->cont.lookup.inode->ia_type)) { - local->self_heal.do_data_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_WARNING, - "split brain detected during lookup of %s.", - local->loc.path); - } - -out: return; } @@ -1226,6 +1416,8 @@ 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) && @@ -1259,6 +1451,7 @@ afr_lookup_select_read_child (afr_local_t *local, xlator_t *this, 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); @@ -1272,8 +1465,9 @@ afr_lookup_select_read_child (afr_local_t *local, xlator_t *this, 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); + type, *gfid); if (source < 0) { gf_log (this->name, GF_LOG_DEBUG, "failed to select source " "for %s", local->loc.path); @@ -1325,7 +1519,7 @@ afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode, if (background) bg = "background"; - gf_log (this->name, GF_LOG_INFO, + gf_log (this->name, GF_LOG_DEBUG, "%s %s self-heal triggered. path: %s, reason: %s", bg, sh_type_str, local->loc.path, reason); @@ -1396,7 +1590,7 @@ afr_conflicting_iattrs (struct iatt *bufs, int32_t *success_children, child2 = &bufs[success_children[i-1]]; if (FILETYPE_DIFFERS (child1, child2)) { - gf_log (xlator_name, GF_LOG_WARNING, "%s: filetype " + 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; @@ -1405,7 +1599,7 @@ afr_conflicting_iattrs (struct iatt *bufs, int32_t *success_children, if (!gfid || uuid_is_null (child1->ia_gfid)) continue; if (uuid_compare (*gfid, child1->ia_gfid)) { - gf_log (xlator_name, GF_LOG_WARNING, "%s: gfid differs" + gf_log (xlator_name, GF_LOG_DEBUG, "%s: gfid differs" " on subvolume %d", path, success_children[i]); conflicting = _gf_true; goto out; @@ -1488,13 +1682,11 @@ afr_lookup_set_self_heal_params (afr_local_t *local, xlator_t *this) int32_t child1 = -1; int32_t child2 = -1; afr_self_heal_t *sh = NULL; - gf_boolean_t split_brain = _gf_false; priv = this->private; sh = &local->self_heal; - split_brain = afr_is_split_brain (this, local->cont.lookup.inode); - afr_detect_self_heal_by_lookup_status (local, this, split_brain); + 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; @@ -1521,9 +1713,11 @@ afr_lookup_set_self_heal_params (afr_local_t *local, xlator_t *this) afr_lookup_set_self_heal_params_by_xattr (local, this, xattr[child1]); } - if (afr_open_only_data_self_heal (priv->data_self_heal) - && !split_brain) + 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 @@ -1539,8 +1733,8 @@ afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this, if (op_ret == -1) { local->op_ret = -1; - if (afr_error_more_important (local->op_errno, op_errno)) - local->op_errno = op_errno; + local->op_errno = afr_most_important_error(local->op_errno, + op_errno, _gf_true); goto out; } else { @@ -1554,6 +1748,16 @@ afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this, 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, @@ -1628,7 +1832,8 @@ afr_lookup_perform_self_heal (call_frame_t *frame, xlator_t *this, afr_lookup_set_self_heal_params (local, this); if (afr_can_self_heal_proceed (&local->self_heal, priv)) { - if (afr_is_transaction_running (local)) + if (afr_is_transaction_running (local) && + (!local->allow_sh_for_running_transaction)) goto out; reason = "lookup detected pending operations"; @@ -1689,20 +1894,15 @@ afr_lookup_done_success_action (call_frame_t *frame, xlator_t *this, int32_t read_child = -1; int32_t ret = -1; afr_local_t *local = NULL; - afr_private_t *priv = NULL; gf_boolean_t fresh_lookup = _gf_false; local = frame->local; - priv = this->private; fresh_lookup = local->cont.lookup.fresh_lookup; if (local->loc.parent == NULL) fail_conflict = _gf_true; - if (afr_conflicting_iattrs (local->cont.lookup.bufs, - local->cont.lookup.success_children, - priv->child_count, local->loc.path, - this->name)) { + if (afr_lookup_conflicting_entries (local, this)) { if (fail_conflict == _gf_false) ret = 0; goto out; @@ -1734,6 +1934,135 @@ out: 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) { @@ -1750,8 +2079,18 @@ afr_lookup_done (call_frame_t *frame, xlator_t *this) 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); @@ -1803,25 +2142,20 @@ afr_lookup_done (call_frame_t *frame, xlator_t *this) * others in that they must be given higher priority while * returning to the user. * - * The hierarchy is ESTALE > ENOENT > others - * + * The hierarchy is ESTALE > EIO > ENOENT > others */ - -gf_boolean_t -afr_error_more_important (int32_t old_errno, int32_t new_errno) +int32_t +afr_most_important_error(int32_t old_errno, int32_t new_errno, + gf_boolean_t eio) { - gf_boolean_t ret = _gf_true; - - /* Nothing should ever overwrite ESTALE */ - if (old_errno == ESTALE) - ret = _gf_false; - - /* Nothing should overwrite ENOENT, except ESTALE/EIO*/ - else if ((old_errno == ENOENT) && (new_errno != ESTALE) - && (new_errno != EIO)) - ret = _gf_false; - - return ret; + 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 @@ -1840,8 +2174,9 @@ afr_resultant_errno_get (int32_t *children, } else { child = i; } - if (afr_error_more_important (op_errno, child_errno[child])) - op_errno = child_errno[child]; + op_errno = afr_most_important_error(op_errno, + child_errno[child], + _gf_false); } return op_errno; } @@ -1853,8 +2188,8 @@ afr_lookup_handle_error (afr_local_t *local, int32_t op_ret, int32_t op_errno) if (op_errno == ENOENT) local->enoent_count++; - if (afr_error_more_important (local->op_errno, op_errno)) - local->op_errno = op_errno; + local->op_errno = afr_most_important_error(local->op_errno, op_errno, + _gf_false); if (local->op_errno == ESTALE) { local->op_ret = -1; @@ -1901,12 +2236,79 @@ afr_lookup_handle_first_success (afr_local_t *local, xlator_t *this, 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; @@ -1919,6 +2321,11 @@ afr_lookup_handle_success (afr_local_t *local, xlator_t *this, int32_t child_ind 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++; } @@ -1966,6 +2373,7 @@ afr_lookup_cont_init (afr_local_t *local, unsigned int child_count) 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, @@ -1998,6 +2406,11 @@ afr_lookup_cont_init (afr_local_t *local, unsigned int child_count) 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; @@ -2015,7 +2428,7 @@ afr_lookup (call_frame_t *frame, xlator_t *this, 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); @@ -2025,25 +2438,38 @@ afr_lookup (call_frame_t *frame, xlator_t *this, frame->local = local; local->fop = GF_FOP_LOOKUP; - if (!strcmp (loc->path, "/" GF_REPLICATE_TRASH_DIR)) { - op_errno = ENOENT; + loc_copy (&local->loc, loc); + ret = loc_path (&local->loc, NULL); + if (ret < 0) { + op_errno = EINVAL; goto out; } - loc_copy (&local->loc, loc); + if (local->loc.path && + (strcmp (local->loc.path, "/" GF_REPLICATE_TRASH_DIR) == 0)) { + op_errno = EPERM; + ret = -1; + goto out; + } - ret = inode_ctx_get (loc->inode, this, &ctx); + 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, - loc->inode, - NULL); + local->loc.inode, + NULL); } else { LOCK (&priv->read_child_lock); { - local->read_child_index = (++priv->read_child_rr) - % (priv->child_count); + 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; @@ -2074,24 +2500,33 @@ afr_lookup (call_frame_t *frame, xlator_t *this, /* By default assume ENOTCONN. On success it will be set to 0. */ local->op_errno = ENOTCONN; - local->call_count = afr_up_children_count (local->child_up, - priv->child_count); - ret = afr_lookup_xattr_req_prepare (local, this, xattr_req, loc, + 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, - loc); + &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, - loc, local->xattr_req); + &local->loc, local->xattr_req); if (!--call_count) break; } @@ -2185,8 +2620,11 @@ __afr_fd_ctx_set (xlator_t *this, fd_t *fd) goto out; } - INIT_LIST_HEAD (&fd_ctx->paused_calls); + 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) @@ -2214,135 +2652,70 @@ afr_fd_ctx_set (xlator_t *this, fd_t *fd) /* {{{ flush */ int -afr_flush_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 (flush, main_frame, - local->op_ret, local->op_errno, - NULL); - } - - return 0; -} - - -int -afr_flush_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *xdata) +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; - 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 (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->success_count++; - - if (local->success_count == priv->wait_count) { - need_unwind = 1; - } } local->op_errno = op_errno; } UNLOCK (&frame->lock); - if (need_unwind) - afr_flush_unwind (frame, this); - - call_count = afr_frame_return (frame); + call_count = afr_frame_return (frame); - if (call_count == 0) { - local->transaction.resume (frame, this); - } + if (call_count == 0) + AFR_STACK_UNWIND(flush, frame, local->op_ret, + local->op_errno, NULL); return 0; } - -int -afr_flush_wind (call_frame_t *frame, xlator_t *this) +static int +afr_flush_wrapper (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata) { - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - int i = 0; - int call_count = -1; + int i = 0; + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + int call_count = -1; - 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; + 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_wind_cbk, + 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_done (call_frame_t *frame, xlator_t *this) -{ - afr_local_t *local = NULL; - - local = frame->local; - - local->transaction.unwind (frame, this); - - AFR_STACK_DESTROY (frame); - - 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_frame_t *transaction_frame = NULL; + call_stub_t *stub = NULL; int ret = -1; int op_errno = 0; @@ -2352,47 +2725,27 @@ afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata) priv = this->private; - transaction_frame = copy_frame (frame); - if (!transaction_frame) { - op_errno = ENOMEM; - goto out; - } - - 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->op = GF_FOP_FLUSH; - - local->transaction.fop = afr_flush_wind; - local->transaction.done = afr_flush_done; - local->transaction.unwind = afr_flush_unwind; - - local->fd = fd_ref (fd); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - local->transaction.main_frame = frame; - local->transaction.start = 0; - local->transaction.len = 0; + ret = afr_local_init(local, priv, &op_errno); + if (ret < 0) + goto out; - ret = afr_open_fd_fix (transaction_frame, this, _gf_false); - if (ret) { - op_errno = -ret; + 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_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); + afr_delayed_changelog_wake_resume (this, fd, stub); + ret = 0; - ret = 0; out: - if (ret < 0) { - if (transaction_frame) - AFR_STACK_DESTROY (transaction_frame); - - AFR_STACK_UNWIND (flush, frame, -1, op_errno, NULL); - } + if (ret < 0) + AFR_STACK_UNWIND(flush, frame, -1, op_errno, NULL); return 0; } @@ -2406,8 +2759,6 @@ afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd) uint64_t ctx = 0; afr_fd_ctx_t *fd_ctx = NULL; int ret = 0; - afr_fd_paused_call_t *paused_call = NULL; - afr_fd_paused_call_t *tmp = NULL; ret = fd_ctx_get (fd, this, &ctx); if (ret < 0) @@ -2416,28 +2767,18 @@ afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd) fd_ctx = (afr_fd_ctx_t *)(long) ctx; if (fd_ctx) { - if (fd_ctx->pre_op_done) - GF_FREE (fd_ctx->pre_op_done); + GF_FREE (fd_ctx->pre_op_done); - if (fd_ctx->opened_on) - GF_FREE (fd_ctx->opened_on); + GF_FREE (fd_ctx->opened_on); - if (fd_ctx->locked_on) - GF_FREE (fd_ctx->locked_on); + GF_FREE (fd_ctx->locked_on); - if (fd_ctx->pre_op_piggyback) - GF_FREE (fd_ctx->pre_op_piggyback); - list_for_each_entry_safe (paused_call, tmp, &fd_ctx->paused_calls, - call_list) { - list_del_init (&paused_call->call_list); - GF_FREE (paused_call); - } + GF_FREE (fd_ctx->pre_op_piggyback); + GF_FREE (fd_ctx->lock_piggyback); - if (fd_ctx->lock_piggyback) - GF_FREE (fd_ctx->lock_piggyback); + GF_FREE (fd_ctx->lock_acquired); - if (fd_ctx->lock_acquired) - GF_FREE (fd_ctx->lock_acquired); + pthread_mutex_destroy (&fd_ctx->delay_lock); GF_FREE (fd_ctx); } @@ -2475,6 +2816,16 @@ afr_release (xlator_t *this, fd_t *fd) /* {{{ 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) @@ -2483,6 +2834,7 @@ afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int call_count = -1; int child_index = (long) cookie; int read_child = 0; + call_stub_t *stub = NULL; local = frame->local; @@ -2498,13 +2850,13 @@ afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->op_ret = 0; if (local->success_count == 0) { - local->cont.fsync.prebuf = *prebuf; - local->cont.fsync.postbuf = *postbuf; + local->cont.inode_wfop.prebuf = *prebuf; + local->cont.inode_wfop.postbuf = *postbuf; } if (child_index == read_child) { - local->cont.fsync.prebuf = *prebuf; - local->cont.fsync.postbuf = *postbuf; + local->cont.inode_wfop.prebuf = *prebuf; + local->cont.inode_wfop.postbuf = *postbuf; } local->success_count++; @@ -2517,10 +2869,32 @@ afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, call_count = afr_frame_return (frame); if (call_count == 0) { - AFR_STACK_UNWIND (fsync, frame, local->op_ret, local->op_errno, - &local->cont.fsync.prebuf, - &local->cont.fsync.postbuf, - NULL); + /* 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; @@ -2555,6 +2929,10 @@ afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, 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, @@ -3397,8 +3775,7 @@ afr_forget (xlator_t *this, inode_t *inode) goto out; ctx = (afr_inode_ctx_t *)(long)ctx_addr; - if (ctx->fresh_children) - GF_FREE (ctx->fresh_children); + GF_FREE (ctx->fresh_children); GF_FREE (ctx); out: return 0; @@ -3486,6 +3863,14 @@ afr_notify (xlator_t *this, int32_t event, 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]) { @@ -3705,6 +4090,17 @@ afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno) 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; @@ -3716,16 +4112,6 @@ afr_internal_lock_init (afr_internal_lock_t *lk, size_t child_count, { int ret = -ENOMEM; - lk->inode_locked_nodes = GF_CALLOC (sizeof (*lk->inode_locked_nodes), - child_count, gf_afr_mt_char); - if (NULL == lk->inode_locked_nodes) - goto out; - - lk->entry_locked_nodes = GF_CALLOC (sizeof (*lk->entry_locked_nodes), - child_count, gf_afr_mt_char); - if (NULL == lk->entry_locked_nodes) - goto out; - lk->locked_nodes = GF_CALLOC (sizeof (*lk->locked_nodes), child_count, gf_afr_mt_char); if (NULL == lk->locked_nodes) @@ -3784,6 +4170,21 @@ out: } 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; @@ -3796,6 +4197,14 @@ afr_transaction_local_init (afr_local_t *local, xlator_t *this) 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); @@ -3817,14 +4226,6 @@ afr_transaction_local_init (afr_local_t *local, xlator_t *this) if (!local->fresh_children) goto out; - if (local->fd) { - local->fd_open_on = GF_CALLOC (sizeof (*local->fd_open_on), - priv->child_count, - gf_afr_mt_char); - if (!local->fd_open_on) - goto out; - } - local->transaction.pre_op = GF_CALLOC (sizeof (*local->transaction.pre_op), priv->child_count, gf_afr_mt_char); @@ -3836,11 +4237,12 @@ afr_transaction_local_init (afr_local_t *local, xlator_t *this) if (!local->pending) goto out; - local->transaction.child_errno = - GF_CALLOC (sizeof (*local->transaction.child_errno), - priv->child_count, - gf_afr_mt_int32_t); - local->transaction.erase_pending = 1; + 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: @@ -4015,7 +4417,6 @@ afr_priv_destroy (afr_private_t *priv) GF_FREE (priv->shd.pos); GF_FREE (priv->shd.pending); GF_FREE (priv->shd.inprogress); - GF_FREE (priv->shd.sh_times); // 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]); @@ -4030,6 +4431,16 @@ afr_priv_destroy (afr_private_t *priv) 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++) @@ -4056,3 +4467,125 @@ xlator_subvolume_count (xlator_t *this) 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 9cf16b5d2..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) 2007-2011 Gluster, Inc. <http://www.gluster.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. */ @@ -262,7 +253,7 @@ unlock: goto out; if (!afr_is_opendir_done (this, local->fd->inode) && - up_children_count > 1) { + up_children_count > 1 && priv->entry_self_heal) { /* * This is the first opendir on this inode. We need @@ -367,85 +358,6 @@ struct entry_name { struct list_head list; }; - -static gf_boolean_t -remembered_name (const char *name, struct list_head *entries) -{ - struct entry_name *e = NULL; - gf_boolean_t ret = _gf_false; - - list_for_each_entry (e, entries, list) { - if (!strcmp (name, e->name)) { - ret = _gf_true; - goto out; - } - } - -out: - return ret; -} - - -static void -afr_remember_entries (gf_dirent_t *entries, fd_t *fd) -{ - struct entry_name *n = NULL; - gf_dirent_t *entry = 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 (entry, &entries->list, list) { - n = GF_CALLOC (1, sizeof (*n), gf_afr_mt_entry_name); - n->name = gf_strdup (entry->d_name); - INIT_LIST_HEAD (&n->list); - - list_add (&n->list, &fd_ctx->entries); - } -} - - -static off_t -afr_filter_entries (gf_dirent_t *entries, fd_t *fd) -{ - gf_dirent_t *entry = NULL; - gf_dirent_t *tmp = NULL; - int ret = 0; - uint64_t ctx = 0; - afr_fd_ctx_t *fd_ctx = NULL; - off_t offset = 0; - - 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 -1; - } - - fd_ctx = (afr_fd_ctx_t *)(long) ctx; - - list_for_each_entry_safe (entry, tmp, &entries->list, list) { - offset = entry->d_off; - - if (remembered_name (entry->d_name, &fd_ctx->entries)) { - list_del (&entry->list); - GF_FREE (entry); - } - } - - return offset; -} - - static void afr_forget_entries (fd_t *fd) { @@ -471,32 +383,36 @@ afr_forget_entries (fd_t *fd) } } +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); + } + } +} 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; - gf_dirent_t * entry = NULL; - gf_dirent_t * tmp = NULL; - - local = frame->local; + afr_local_t *local = NULL; if (op_ret == -1) goto out; - list_for_each_entry_safe (entry, tmp, &entries->list, list) { - if ((local->fd->inode == local->fd->inode->table->root) - && !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) { - list_del_init (&entry->list); - GF_FREE (entry); - } - } + local = frame->local; + afr_readdir_filter_trash_dir (entries, local->fd); out: AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries, NULL); - return 0; } @@ -506,125 +422,16 @@ 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; - afr_local_t * local = NULL; - xlator_t ** children = NULL; - int32_t next_call_child = -1; - int ret = 0; - gf_dirent_t * entry = NULL; - gf_dirent_t * tmp = NULL; - int32_t *last_index = NULL; - int32_t read_child = -1; - int32_t *fresh_children = NULL; - uint64_t ctx = 0; - afr_fd_ctx_t *fd_ctx = NULL; - off_t offset = 0; - int32_t call_child = -1; + afr_local_t *local = NULL; - priv = this->private; - children = priv->children; + if (op_ret == -1) + goto out; local = frame->local; - - read_child = (long) cookie; - last_index = &local->cont.readdir.last_index; - fresh_children = local->fresh_children; - - /* the value of the last_index changes if afr_next_call_child is - * called. So to find the call_child of this callback use last_index - * before the next_call_child call. - */ - if (*last_index == -1) - call_child = read_child; - else - call_child = fresh_children[*last_index]; - - if (priv->strict_readdir) { - ret = fd_ctx_get (local->fd, this, &ctx); - if (ret < 0) { - gf_log (this->name, GF_LOG_INFO, - "could not get fd ctx for fd=%p", local->fd); - op_ret = -1; - op_errno = -ret; - goto out; - } - - fd_ctx = (afr_fd_ctx_t *)(long) ctx; - - if (op_ret == -1) { - 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; - gf_log (this->name, GF_LOG_TRACE, - "starting readdir afresh on child %d, offset %"PRId64, - next_call_child, (uint64_t) 0); - - fd_ctx->failed_over = _gf_true; - - STACK_WIND_COOKIE (frame, afr_readdirp_cbk, - (void *) (long) read_child, - children[next_call_child], - children[next_call_child]->fops->readdirp, - local->fd, - local->cont.readdir.size, 0, - local->cont.readdir.dict); - return 0; - } - } - - if (op_ret != -1) { - list_for_each_entry_safe (entry, tmp, &entries->list, list) { - if ((local->fd->inode == local->fd->inode->table->root) - && !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) { - list_del_init (&entry->list); - GF_FREE (entry); - } - } - } - - if (priv->strict_readdir) { - if (fd_ctx->failed_over) { - if (list_empty (&entries->list)) { - gf_log (this->name, GF_LOG_DEBUG, - "no entries found"); - goto out; - } - - offset = afr_filter_entries (entries, local->fd); - - afr_remember_entries (entries, local->fd); - - if (list_empty (&entries->list)) { - /* All the entries we got were duplicate. We - shouldn't send an empty list now, because - that will make the application stop reading. So - try to get more entries */ - - gf_log (this->name, GF_LOG_TRACE, - "trying to fetch non-duplicate entries " - "from offset %"PRId64", child %s", - offset, children[call_child]->name); - - STACK_WIND_COOKIE (frame, afr_readdirp_cbk, - (void *) (long) read_child, - children[call_child], - children[call_child]->fops->readdirp, - local->fd, local->cont.readdir.size, offset, - local->cont.readdir.dict); - return 0; - } - } else { - afr_remember_entries (entries, local->fd); - } - } + afr_readdir_filter_trash_dir (entries, local->fd); out: AFR_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries, NULL); - return 0; } @@ -632,15 +439,14 @@ 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; - uint64_t ctx = 0; - afr_fd_ctx_t *fd_ctx = NULL; - int ret = -1; - int32_t op_errno = 0; - uint64_t read_child = 0; + 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); @@ -665,42 +471,32 @@ afr_do_readdir (call_frame_t *frame, xlator_t *this, 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); + 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 (priv->strict_readdir) { - 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); - op_errno = -ret; - goto out; - } - - fd_ctx = (afr_fd_ctx_t *)(long) ctx; - - if (fd_ctx->last_tried != call_child) { - gf_log (this->name, GF_LOG_TRACE, - "first up child has changed from %d to %d, " - "restarting readdir from offset 0", - fd_ctx->last_tried, call_child); - - fd_ctx->failed_over = _gf_true; - offset = 0; - } - - fd_ctx->last_tried = call_child; - } - if (whichop == GF_FOP_READDIR) STACK_WIND_COOKIE (frame, afr_readdir_cbk, (void *) (long) call_child, @@ -714,10 +510,9 @@ afr_do_readdir (call_frame_t *frame, xlator_t *this, children[call_child]->fops->readdirp, fd, size, offset, dict); - ret = 0; + return 0; out: - if (ret < 0) - AFR_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL); + AFR_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL); return 0; } diff --git a/xlators/cluster/afr/src/afr-dir-read.h b/xlators/cluster/afr/src/afr-dir-read.h index 2ab126adf..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) 2007-2011 Gluster, Inc. <http://www.gluster.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__ diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c index 37d86ee5c..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) 2007-2011 Gluster, Inc. <http://www.gluster.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. */ @@ -65,15 +56,206 @@ afr_build_parent_loc (loc_t *parent, loc_t *child, int32_t *op_errno) *op_errno = ENOMEM; goto out; } - parent->path = dirname (child_path); + 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_mark_new_entry_changelog (call_frame_t *frame, xlator_t *this) +{ + 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; + } + + 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; + } + + 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; +} + +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; + + 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 */ int @@ -81,7 +263,6 @@ afr_create_unwind (call_frame_t *frame, xlator_t *this) { call_frame_t *main_frame = NULL; afr_local_t *local = NULL; - struct iatt *unwind_buf = NULL; local = frame->local; @@ -95,19 +276,14 @@ afr_create_unwind (call_frame_t *frame, xlator_t *this) UNLOCK (&frame->lock); if (main_frame) { - if (local->cont.create.read_child_buf.ia_ino) { - unwind_buf = &local->cont.create.read_child_buf; - } else { - unwind_buf = &local->cont.create.buf; - } - AFR_STACK_UNWIND (create, main_frame, local->op_ret, local->op_errno, local->cont.create.fd, - local->cont.create.inode, - unwind_buf, &local->cont.create.preparent, - &local->cont.create.postparent, - NULL); + 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; @@ -122,29 +298,20 @@ afr_create_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, dict_t *xdata) { afr_local_t *local = NULL; - afr_private_t *priv = NULL; uint64_t ctx = 0; afr_fd_ctx_t *fd_ctx = NULL; int ret = 0; int call_count = -1; int child_index = -1; - int32_t *fresh_children = NULL; local = frame->local; - priv = this->private; child_index = (long) cookie; LOCK (&frame->lock); { - 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 (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); @@ -155,7 +322,6 @@ afr_create_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, } 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); @@ -169,23 +335,14 @@ afr_create_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, fd_ctx->opened_on[child_index] = AFR_FD_OPENED; fd_ctx->flags = local->cont.create.flags; - if (local->success_count == 0) - local->cont.create.buf = *buf; - - if (child_index == local->read_child_index) { - local->cont.create.read_child_buf = *buf; - local->cont.create.preparent = *preparent; - local->cont.create.postparent = *postparent; - } - - local->cont.create.inode = inode; - - fresh_children = local->fresh_children; - fresh_children[local->success_count] = child_index; - local->success_count++; + if (local->success_count == 0) { + if (xdata) + local->xdata_rsp = dict_ref(xdata); + } } - - local->op_errno = op_errno; + __dir_entry_fop_common_cbk (frame, child_index, this, + op_ret, op_errno, inode, buf, + preparent, postparent, NULL, NULL); } unlock: @@ -193,15 +350,8 @@ unlock: call_count = afr_frame_return (frame); - if (call_count == 0) { - afr_set_read_ctx_from_policy (this, inode, - local->fresh_children, - local->read_child_index, - priv->read_child); - local->transaction.unwind (frame, this); - - local->transaction.resume (frame, this); - } + if (call_count == 0) + afr_dir_fop_done (frame, this); return 0; } @@ -269,11 +419,12 @@ afr_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) { - afr_private_t *priv = NULL; - afr_local_t *local = NULL; - call_frame_t *transaction_frame = NULL; - int 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); @@ -305,6 +456,7 @@ afr_create (call_frame_t *frame, xlator_t *this, } UNLOCK (&priv->read_child_lock); + local->op = GF_FOP_CREATE; local->cont.create.flags = flags; local->cont.create.mode = mode; local->cont.create.fd = fd_ref (fd); @@ -323,8 +475,22 @@ afr_create (call_frame_t *frame, xlator_t *this, 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; - afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION); + int_lock->lockee_count++; + ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION); + if (ret < 0) { + op_errno = -ret; + goto out; + } ret = 0; out: @@ -347,7 +513,6 @@ afr_mknod_unwind (call_frame_t *frame, xlator_t *this) { call_frame_t *main_frame = NULL; afr_local_t *local = NULL; - struct iatt *unwind_buf = NULL; local = frame->local; @@ -361,17 +526,12 @@ afr_mknod_unwind (call_frame_t *frame, xlator_t *this) UNLOCK (&frame->lock); if (main_frame) { - if (local->cont.mknod.read_child_buf.ia_ino) { - unwind_buf = &local->cont.mknod.read_child_buf; - } else { - unwind_buf = &local->cont.mknod.buf; - } - AFR_STACK_UNWIND (mknod, main_frame, local->op_ret, local->op_errno, - local->cont.mknod.inode, - unwind_buf, &local->cont.mknod.preparent, - &local->cont.mknod.postparent, + local->cont.dir_fop.inode, + &local->cont.dir_fop.buf, + &local->cont.dir_fop.preparent, + &local->cont.dir_fop.postparent, NULL); } @@ -385,56 +545,23 @@ afr_mknod_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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; - int32_t *fresh_children = NULL; - - local = frame->local; - priv = this->private; child_index = (long) cookie; LOCK (&frame->lock); { - 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) - local->cont.mknod.buf = *buf; - - if (child_index == local->read_child_index) { - local->cont.mknod.read_child_buf = *buf; - local->cont.mknod.preparent = *preparent; - local->cont.mknod.postparent = *postparent; - } - - local->cont.mknod.inode = inode; - - fresh_children = local->fresh_children; - fresh_children[local->success_count] = child_index; - local->success_count++; - } - - local->op_errno = op_errno; + __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); - if (call_count == 0) { - afr_set_read_ctx_from_policy (this, inode, - local->fresh_children, - local->read_child_index, - priv->read_child); - local->transaction.unwind (frame, this); - - local->transaction.resume (frame, this); - } + if (call_count == 0) + afr_dir_fop_done (frame, this); return 0; } @@ -497,11 +624,12 @@ int 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; - int 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); @@ -533,6 +661,7 @@ afr_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode, } UNLOCK (&priv->read_child_lock); + local->op = GF_FOP_MKNOD; local->cont.mknod.mode = mode; local->cont.mknod.dev = dev; local->umask = umask; @@ -550,8 +679,22 @@ afr_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME (loc->path); + int_lock = &local->internal_lock; - afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION); + 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: @@ -575,7 +718,6 @@ afr_mkdir_unwind (call_frame_t *frame, xlator_t *this) { call_frame_t *main_frame = NULL; afr_local_t *local = NULL; - struct iatt *unwind_buf = NULL; local = frame->local; @@ -589,17 +731,12 @@ afr_mkdir_unwind (call_frame_t *frame, xlator_t *this) UNLOCK (&frame->lock); if (main_frame) { - if (local->cont.mkdir.read_child_buf.ia_ino) { - unwind_buf = &local->cont.mkdir.read_child_buf; - } else { - unwind_buf = &local->cont.mkdir.buf; - } - AFR_STACK_UNWIND (mkdir, main_frame, local->op_ret, local->op_errno, - local->cont.mkdir.inode, - unwind_buf, &local->cont.mkdir.preparent, - &local->cont.mkdir.postparent, + local->cont.dir_fop.inode, + &local->cont.dir_fop.buf, + &local->cont.dir_fop.preparent, + &local->cont.dir_fop.postparent, NULL); } @@ -613,56 +750,23 @@ afr_mkdir_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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; - int32_t *fresh_children = NULL; - - local = frame->local; - priv = this->private; child_index = (long) cookie; LOCK (&frame->lock); { - 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) - local->cont.mkdir.buf = *buf; - - if (child_index == local->read_child_index) { - local->cont.mkdir.read_child_buf = *buf; - local->cont.mkdir.preparent = *preparent; - local->cont.mkdir.postparent = *postparent; - } - - local->cont.mkdir.inode = inode; - - fresh_children = local->fresh_children; - fresh_children[local->success_count] = child_index; - local->success_count++; - } - - local->op_errno = op_errno; + __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); - if (call_count == 0) { - afr_set_read_ctx_from_policy (this, inode, - local->fresh_children, - local->read_child_index, - priv->read_child); - local->transaction.unwind (frame, this); - - local->transaction.resume (frame, this); - } + if (call_count == 0) + afr_dir_fop_done (frame, this); return 0; } @@ -721,16 +825,16 @@ afr_mkdir_done (call_frame_t *frame, xlator_t *this) return 0; } - int afr_mkdir (call_frame_t *frame, xlator_t *this, 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; - int 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); @@ -767,6 +871,7 @@ afr_mkdir (call_frame_t *frame, xlator_t *this, 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; @@ -778,8 +883,22 @@ afr_mkdir (call_frame_t *frame, xlator_t *this, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME (loc->path); + int_lock = &local->internal_lock; - afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION); + 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: @@ -804,7 +923,6 @@ afr_link_unwind (call_frame_t *frame, xlator_t *this) { call_frame_t *main_frame = NULL; afr_local_t *local = NULL; - struct iatt *unwind_buf = NULL; local = frame->local; @@ -818,17 +936,12 @@ afr_link_unwind (call_frame_t *frame, xlator_t *this) UNLOCK (&frame->lock); if (main_frame) { - if (local->cont.link.read_child_buf.ia_ino) { - unwind_buf = &local->cont.link.read_child_buf; - } else { - unwind_buf = &local->cont.link.buf; - } - AFR_STACK_UNWIND (link, main_frame, local->op_ret, local->op_errno, - local->cont.link.inode, - unwind_buf, &local->cont.link.preparent, - &local->cont.link.postparent, + local->cont.dir_fop.inode, + &local->cont.dir_fop.buf, + &local->cont.dir_fop.preparent, + &local->cont.dir_fop.postparent, NULL); } @@ -842,57 +955,23 @@ afr_link_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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; - int32_t *fresh_children = NULL; - - local = frame->local; - priv = this->private; child_index = (long) cookie; LOCK (&frame->lock); { - 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) { - local->cont.link.buf = *buf; - } - - if (child_index == local->read_child_index) { - local->cont.link.read_child_buf = *buf; - local->cont.link.preparent = *preparent; - local->cont.link.postparent = *postparent; - } - - local->cont.link.inode = inode; - - fresh_children = local->fresh_children; - fresh_children[local->success_count] = child_index; - local->success_count++; - } - - local->op_errno = op_errno; + __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); - if (call_count == 0) { - afr_set_read_ctx_from_policy (this, inode, - local->fresh_children, - local->read_child_index, - priv->read_child); - local->transaction.unwind (frame, this); - - local->transaction.resume (frame, this); - } + if (call_count == 0) + afr_dir_fop_done (frame, this); return 0; } @@ -921,7 +1000,8 @@ afr_link_wind (call_frame_t *frame, xlator_t *this) 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, + STACK_WIND_COOKIE (frame, afr_link_wind_cbk, + (void *) (long) i, priv->children[i], priv->children[i]->fops->link, &local->loc, @@ -953,11 +1033,12 @@ int afr_link (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; - call_frame_t * transaction_frame = NULL; - int 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); @@ -992,6 +1073,7 @@ afr_link (call_frame_t *frame, xlator_t *this, } UNLOCK (&priv->read_child_lock); + local->op = GF_FOP_LINK; local->transaction.fop = afr_link_wind; local->transaction.done = afr_link_done; local->transaction.unwind = afr_link_unwind; @@ -1003,9 +1085,22 @@ afr_link (call_frame_t *frame, xlator_t *this, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME (newloc->path); + int_lock = &local->internal_lock; - afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION); + 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 (ret < 0) { @@ -1028,7 +1123,6 @@ afr_symlink_unwind (call_frame_t *frame, xlator_t *this) { call_frame_t *main_frame = NULL; afr_local_t *local = NULL; - struct iatt *unwind_buf = NULL; local = frame->local; @@ -1042,17 +1136,12 @@ afr_symlink_unwind (call_frame_t *frame, xlator_t *this) UNLOCK (&frame->lock); if (main_frame) { - if (local->cont.symlink.read_child_buf.ia_ino) { - unwind_buf = &local->cont.symlink.read_child_buf; - } else { - unwind_buf = &local->cont.symlink.buf; - } - AFR_STACK_UNWIND (symlink, main_frame, local->op_ret, local->op_errno, - local->cont.symlink.inode, - unwind_buf, &local->cont.symlink.preparent, - &local->cont.symlink.postparent, + local->cont.dir_fop.inode, + &local->cont.dir_fop.buf, + &local->cont.dir_fop.preparent, + &local->cont.dir_fop.postparent, NULL); } @@ -1066,56 +1155,23 @@ afr_symlink_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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; - int32_t *fresh_children = NULL; - - local = frame->local; - priv = this->private; child_index = (long) cookie; LOCK (&frame->lock); { - 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) - local->cont.symlink.buf = *buf; - - if (child_index == local->read_child_index) { - local->cont.symlink.read_child_buf = *buf; - local->cont.symlink.preparent = *preparent; - local->cont.symlink.postparent = *postparent; - } - - local->cont.symlink.inode = inode; - - fresh_children = local->fresh_children; - fresh_children[local->success_count] = child_index; - local->success_count++; - } - - local->op_errno = op_errno; + __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); - if (call_count == 0) { - afr_set_read_ctx_from_policy (this, inode, - local->fresh_children, - local->read_child_index, - priv->read_child); - local->transaction.unwind (frame, this); - - local->transaction.resume (frame, this); - } + if (call_count == 0) + afr_dir_fop_done (frame, this); return 0; } @@ -1180,11 +1236,12 @@ int afr_symlink (call_frame_t *frame, xlator_t *this, 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; - 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); @@ -1221,6 +1278,7 @@ afr_symlink (call_frame_t *frame, xlator_t *this, 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; @@ -1232,8 +1290,22 @@ afr_symlink (call_frame_t *frame, xlator_t *this, 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; - afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION); + int_lock->lockee_count++; + ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION); + if (ret < 0) { + op_errno = -ret; + goto out; + } ret = 0; out: @@ -1256,7 +1328,6 @@ afr_rename_unwind (call_frame_t *frame, xlator_t *this) { call_frame_t *main_frame = NULL; afr_local_t *local = NULL; - struct iatt *unwind_buf = NULL; local = frame->local; @@ -1270,19 +1341,13 @@ afr_rename_unwind (call_frame_t *frame, xlator_t *this) UNLOCK (&frame->lock); if (main_frame) { - if (local->cont.rename.read_child_buf.ia_ino) { - unwind_buf = &local->cont.rename.read_child_buf; - } else { - unwind_buf = &local->cont.rename.buf; - } - AFR_STACK_UNWIND (rename, main_frame, local->op_ret, local->op_errno, - unwind_buf, - &local->cont.rename.preoldparent, - &local->cont.rename.postoldparent, - &local->cont.rename.prenewparent, - &local->cont.rename.postnewparent, + &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); } @@ -1309,38 +1374,22 @@ afr_rename_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, { 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) { - if (local->success_count == 0) { - local->op_ret = op_ret; - - if (buf) { - local->cont.rename.buf = *buf; - } - - local->success_count++; - } - - if (child_index == local->read_child_index) { - local->cont.rename.read_child_buf = *buf; - - local->cont.rename.preoldparent = *preoldparent; - local->cont.rename.postoldparent = *postoldparent; - local->cont.rename.prenewparent = *prenewparent; - local->cont.rename.postnewparent = *postnewparent; - } - } + if (op_ret > -1) + __dir_entry_fop_common_cbk (frame, child_index, this, + op_ret, op_errno, NULL, buf, + preoldparent, postoldparent, + prenewparent, postnewparent); - local->op_errno = op_errno; } UNLOCK (&frame->lock); call_count = afr_frame_return (frame); - if (call_count == 0) { - local->transaction.unwind (frame, this); - local->transaction.resume (frame, this); - } + if (call_count == 0) + afr_dir_fop_done (frame, this); return 0; } @@ -1401,11 +1450,13 @@ 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; - call_frame_t * transaction_frame = NULL; - int 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; + int nlockee = 0; VALIDATE_OR_GOTO (frame, out); VALIDATE_OR_GOTO (this, out); @@ -1433,6 +1484,7 @@ afr_rename (call_frame_t *frame, xlator_t *this, 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; @@ -1449,8 +1501,44 @@ afr_rename (call_frame_t *frame, xlator_t *this, 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; - afr_transaction (transaction_frame, this, AFR_ENTRY_RENAME_TRANSACTION); + 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; + } ret = 0; out: @@ -1489,8 +1577,8 @@ afr_unlink_unwind (call_frame_t *frame, xlator_t *this) if (main_frame) { AFR_STACK_UNWIND (unlink, main_frame, local->op_ret, local->op_errno, - &local->cont.unlink.preparent, - &local->cont.unlink.postparent, + &local->cont.dir_fop.preparent, + &local->cont.dir_fop.postparent, NULL); } @@ -1514,36 +1602,15 @@ afr_unlink_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, if (child_index == local->read_child_index) { 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.unlink.preparent = *preparent; - local->cont.unlink.postparent = *postparent; - } - - if (child_index == local->read_child_index) { - local->cont.unlink.preparent = *preparent; - local->cont.unlink.postparent = *postparent; - } - - local->success_count++; - } - - local->op_errno = op_errno; + __dir_entry_fop_common_cbk (frame, child_index, this, + op_ret, op_errno, NULL, NULL, + preparent, postparent, NULL, NULL); } UNLOCK (&frame->lock); call_count = afr_frame_return (frame); - - if (call_count == 0) { - local->transaction.unwind (frame, this); - - local->transaction.resume (frame, this); - } + if (call_count == 0) + afr_dir_fop_done (frame, this); return 0; } @@ -1605,11 +1672,12 @@ 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; - call_frame_t * transaction_frame = NULL; - int 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); @@ -1637,6 +1705,7 @@ afr_unlink (call_frame_t *frame, xlator_t *this, 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; @@ -1648,8 +1717,22 @@ afr_unlink (call_frame_t *frame, xlator_t *this, 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; - afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION); + int_lock->lockee_count++; + ret = afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION); + if (ret < 0) { + op_errno = -ret; + goto out; + } ret = 0; out: @@ -1689,8 +1772,8 @@ afr_rmdir_unwind (call_frame_t *frame, xlator_t *this) if (main_frame) { AFR_STACK_UNWIND (rmdir, main_frame, local->op_ret, local->op_errno, - &local->cont.rmdir.preparent, - &local->cont.rmdir.postparent, + &local->cont.dir_fop.preparent, + &local->cont.dir_fop.postparent, NULL); } @@ -1715,36 +1798,22 @@ afr_rmdir_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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); - - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - local->cont.rmdir.preparent = *preparent; - local->cont.rmdir.postparent = *postparent; - - } - - if (child_index == read_child) { - local->cont.rmdir.preparent = *preparent; - local->cont.rmdir.postparent = *postparent; - } - - local->success_count++; - } - 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); + } UNLOCK (&frame->lock); call_count = afr_frame_return (frame); - - if (call_count == 0) { - local->transaction.unwind (frame, this); - local->transaction.resume (frame, this); - } + if (call_count == 0) + afr_dir_fop_done (frame, this); return 0; } @@ -1806,11 +1875,13 @@ 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; - call_frame_t * transaction_frame = NULL; - int 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; + int nlockee = 0; VALIDATE_OR_GOTO (frame, out); VALIDATE_OR_GOTO (this, out); @@ -1836,6 +1907,7 @@ afr_rmdir (call_frame_t *frame, xlator_t *this, 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; @@ -1847,8 +1919,34 @@ afr_rmdir (call_frame_t *frame, xlator_t *this, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME (loc->path); + int_lock = &local->internal_lock; - afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION); + 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; + } ret = 0; out: diff --git a/xlators/cluster/afr/src/afr-dir-write.h b/xlators/cluster/afr/src/afr-dir-write.h index 5a3494ece..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) 2007-2011 Gluster, Inc. <http://www.gluster.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__ diff --git a/xlators/cluster/afr/src/afr-inode-read.c b/xlators/cluster/afr/src/afr-inode-read.c index 097bba16e..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) 2007-2011 Gluster, Inc. <http://www.gluster.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. */ @@ -126,6 +117,8 @@ afr_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask, children = priv->children; + AFR_SBRAIN_CHECK_LOC (loc, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -241,6 +234,8 @@ afr_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata) children = priv->children; + AFR_SBRAIN_CHECK_LOC (loc, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -357,6 +352,8 @@ afr_fstat (call_frame_t *frame, xlator_t *this, VALIDATE_OR_GOTO (fd->inode, out); + AFR_SBRAIN_CHECK_FD (fd, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -386,11 +383,8 @@ afr_fstat (call_frame_t *frame, xlator_t *this, local->fd = fd_ref (fd); - ret = afr_open_fd_fix (frame, this, _gf_false); - if (ret) { - op_errno = -ret; - goto out; - } + afr_open_fd_fix (fd, this); + STACK_WIND_COOKIE (frame, afr_fstat_cbk, (void *) (long) call_child, children[call_child], children[call_child]->fops->fstat, @@ -479,6 +473,8 @@ afr_readlink (call_frame_t *frame, xlator_t *this, children = priv->children; + AFR_SBRAIN_CHECK_LOC (loc, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -530,7 +526,7 @@ struct _xattr_key { }; -void +int __gather_xattr_keys (dict_t *dict, char *key, data_t *value, void *data) { @@ -542,13 +538,14 @@ __gather_xattr_keys (dict_t *dict, char *key, data_t *value, xkey = GF_CALLOC (1, sizeof (*xkey), gf_afr_mt_xattr_key); if (!xkey) - return; + return -1; xkey->key = key; INIT_LIST_HEAD (&xkey->list); list_add_tail (&xkey->list, list); } + return 0; } @@ -637,6 +634,96 @@ afr_getxattr_unwind (call_frame_t *frame, int op_ret, int op_errno, } 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, + "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); + + if (xattr) + dict_unref (xattr); + } + + 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) @@ -770,12 +857,364 @@ afr_getxattr_node_uuid_cbk (call_frame_t *frame, void *cookie, unwind: if (unwind) - AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, NULL); + AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict, + NULL); + + return 0; +} + +int32_t +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) +{ + int call_cnt = 0, len = 0; + char *lockinfo_buf = NULL; + dict_t *lockinfo = NULL, *newdict = NULL; + afr_local_t *local = NULL; + + LOCK (&frame->lock); + { + local = frame->local; + + call_cnt = --local->call_count; + + if ((op_ret < 0) || (!dict && !xdata)) { + goto unlock; + } + + 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 (!dict) { + goto unlock; + } + + op_ret = dict_get_ptr_and_len (dict, GF_XATTR_LOCKINFO_KEY, + (void **)&lockinfo_buf, &len); + + if (!lockinfo_buf) { + goto unlock; + } + + 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 (xdata && local->xdata_rsp) { + dict_copy (xdata, local->xdata_rsp); + } + + 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_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) +{ + int call_cnt = 0, len = 0; + char *lockinfo_buf = NULL; + dict_t *lockinfo = NULL, *newdict = NULL; + afr_local_t *local = NULL; + + LOCK (&frame->lock); + { + local = frame->local; + + call_cnt = --local->call_count; + + if ((op_ret < 0) || (!dict && !xdata)) { + goto unlock; + } + + 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 (!dict) { + goto unlock; + } + + op_ret = dict_get_ptr_and_len (dict, GF_XATTR_LOCKINFO_KEY, + (void **)&lockinfo_buf, &len); + + if (!lockinfo_buf) { + goto unlock; + } + + 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 (xdata && local->xdata_rsp) { + dict_copy (xdata, local->xdata_rsp); + } + + 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 (fgetxattr, frame, op_ret, + op_errno, newdict, + local->xdata_rsp); + } + + dict_unref (lockinfo); + + return 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: + 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"); + + unwind: + AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, nxattr, + xdata); + + if (nxattr) + dict_unref (nxattr); + } + + return ret; +} + +int32_t 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) @@ -887,8 +1326,65 @@ afr_getxattr_pathinfo_cbk (call_frame_t *frame, void *cookie, 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_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_local_t *local = NULL; + int32_t callcnt = 0; + + if (!frame || !frame->local || !this) { + gf_log ("", GF_LOG_ERROR, "possible NULL deref"); + goto out; + } + + local = frame->local; + + LOCK (&frame->lock); + { + callcnt = --local->call_count; + + if (!dict || (op_ret < 0)) { + local->op_errno = op_errno; + goto cleanup; + } + + 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; + } + +cleanup: + UNLOCK (&frame->lock); + + if (!callcnt) { + AFR_STACK_UNWIND (getxattr, frame, local->op_ret, + local->op_errno, local->dict, xdata); + } + +out: + return 0; +} + + static gf_boolean_t -afr_is_special_xattr (const char *name, fop_getxattr_cbk_t *cbk) +afr_is_special_xattr (const char *name, fop_getxattr_cbk_t *cbk, + gf_boolean_t is_fgetxattr) { gf_boolean_t is_spl = _gf_true; @@ -898,14 +1394,31 @@ afr_is_special_xattr (const char *name, fop_getxattr_cbk_t *cbk) goto out; } - if (!strcmp (name, GF_XATTR_PATHINFO_KEY)) - *cbk = afr_getxattr_pathinfo_cbk; - - else if (!strncmp (name, GF_XATTR_CLRLK_CMD, - strlen (GF_XATTR_CLRLK_CMD))) - *cbk = afr_getxattr_clrlk_cbk; - else + 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 { is_spl = _gf_false; + } out: return is_spl; @@ -951,7 +1464,7 @@ afr_getxattr (call_frame_t *frame, xlator_t *this, 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); @@ -962,6 +1475,8 @@ afr_getxattr (call_frame_t *frame, xlator_t *this, children = priv->children; + AFR_SBRAIN_CHECK_LOC (loc, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -984,7 +1499,7 @@ afr_getxattr (call_frame_t *frame, xlator_t *this, goto out; } if ((strcmp (GF_XATTR_MARKER_KEY, name) == 0) - && (-1 == frame->root->pid)) { + && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) { local->marker.call_count = priv->child_count; @@ -1000,6 +1515,7 @@ afr_getxattr (call_frame_t *frame, xlator_t *this, sub_volumes, priv->child_count, MARKER_UUID_TYPE, + marker_uuid_default_gauge, priv->vol_uuid)) { gf_log (this->name, GF_LOG_INFO, @@ -1016,7 +1532,7 @@ afr_getxattr (call_frame_t *frame, xlator_t *this, * if we are doing getxattr with pathinfo as the key then we * collect information from all childs */ - if (afr_is_special_xattr (name, &cbk)) { + if (afr_is_special_xattr (name, &cbk, 0)) { afr_getxattr_frm_all_children (this, frame, name, loc, cbk); return 0; @@ -1034,10 +1550,11 @@ afr_getxattr (call_frame_t *frame, xlator_t *this, if (*priv->vol_uuid) { if ((match_uuid_local (name, priv->vol_uuid) == 0) - && (-1 == frame->root->pid)) { + && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) { local->marker.call_count = priv->child_count; - sub_volumes = alloca ( priv->child_count * sizeof (xlator_t *)); + sub_volumes = alloca ( priv->child_count + * sizeof (xlator_t *)); for (i = 0, trav = this->children; trav ; trav = trav->next, i++) { @@ -1045,12 +1562,20 @@ afr_getxattr (call_frame_t *frame, xlator_t *this, } + /* 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)", @@ -1070,7 +1595,8 @@ no_name: goto out; } - read_child = afr_inode_get_read_ctx (this, loc->inode, local->fresh_children); + 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, @@ -1158,18 +1684,44 @@ afr_fgetxattr_unwind (call_frame_t *frame, 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; - + 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); @@ -1180,6 +1732,8 @@ afr_fgetxattr (call_frame_t *frame, xlator_t *this, children = priv->children; + AFR_SBRAIN_CHECK_FD (fd, out); + AFR_LOCAL_ALLOC_OR_GOTO (local, out); frame->local = local; @@ -1193,7 +1747,17 @@ afr_fgetxattr (call_frame_t *frame, xlator_t *this, if (name) local->cont.getxattr.name = gf_strdup (name); - /* pathinfo gets handled only in getxattr() */ + /* 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) { @@ -1201,7 +1765,8 @@ afr_fgetxattr (call_frame_t *frame, xlator_t *this, goto out; } - read_child = afr_inode_get_read_ctx (this, fd->inode, local->fresh_children); + 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, @@ -1221,7 +1786,8 @@ afr_fgetxattr (call_frame_t *frame, xlator_t *this, op_ret = 0; out: if (op_ret == -1) { - AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, NULL, NULL); + AFR_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno, NULL, + NULL); } return 0; } @@ -1323,6 +1889,8 @@ afr_readv (call_frame_t *frame, xlator_t *this, priv = this->private; children = priv->children; + AFR_SBRAIN_CHECK_FD (fd, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -1352,11 +1920,8 @@ afr_readv (call_frame_t *frame, xlator_t *this, local->cont.readv.offset = offset; local->cont.readv.flags = flags; - ret = afr_open_fd_fix (frame, this, _gf_false); - if (ret) { - op_errno = -ret; - goto out; - } + afr_open_fd_fix (fd, this); + STACK_WIND_COOKIE (frame, afr_readv_cbk, (void *) (long) call_child, children[call_child], diff --git a/xlators/cluster/afr/src/afr-inode-read.h b/xlators/cluster/afr/src/afr-inode-read.h index fb04ec418..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) 2007-2011 Gluster, Inc. <http://www.gluster.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__ diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c index 7fa6d096c..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) 2007-2011 Gluster, Inc. <http://www.gluster.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. */ @@ -48,34 +39,133 @@ #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; +} + /* {{{ writev */ -int +void +afr_writev_copy_outvars (call_frame_t *src_frame, call_frame_t *dst_frame) +{ + afr_local_t *src_local = NULL; + afr_local_t *dst_local = NULL; + + src_local = src_frame->local; + dst_local = dst_frame->local; + + 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; +} + +void afr_writev_unwind (call_frame_t *frame, xlator_t *this) { afr_local_t * local = NULL; - call_frame_t *main_frame = 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); { - if (local->transaction.main_frame) - main_frame = local->transaction.main_frame; + fop_frame = local->transaction.main_frame; local->transaction.main_frame = NULL; } UNLOCK (&frame->lock); - if (main_frame) { - AFR_STACK_UNWIND (writev, main_frame, - local->op_ret, local->op_errno, - &local->cont.writev.prebuf, - &local->cont.writev.postbuf, - NULL); + return fop_frame; +} + +int +afr_transaction_writev_unwind (call_frame_t *frame, xlator_t *this) +{ + call_frame_t *fop_frame = NULL; + + fop_frame = afr_transaction_detach_fop_frame (frame); + + if (fop_frame) { + afr_writev_copy_outvars (frame, fop_frame); + afr_writev_unwind (fop_frame, this); } return 0; } +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); + } +} int afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, @@ -83,11 +173,17 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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); @@ -97,32 +193,81 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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.writev.prebuf = *prebuf; - local->cont.writev.postbuf = *postbuf; + __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; + } + + 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; } - if (child_index == read_child) { - local->cont.writev.prebuf = *prebuf; - local->cont.writev.postbuf = *postbuf; - } - } - - local->op_errno = op_errno; + } } UNLOCK (&frame->lock); call_count = afr_frame_return (frame); if (call_count == 0) { - local->transaction.unwind (frame, this); - local->transaction.resume (frame, this); + 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; } @@ -134,6 +279,8 @@ afr_writev_wind (call_frame_t *frame, xlator_t *this) 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; @@ -147,6 +294,28 @@ afr_writev_wind (call_frame_t *frame, xlator_t *this) } 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; + } + + 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]) { @@ -160,13 +329,16 @@ afr_writev_wind (call_frame_t *frame, xlator_t *this) local->cont.writev.offset, local->cont.writev.flags, local->cont.writev.iobref, - NULL); + xdata); if (!--call_count) break; } } + if (xdata) + dict_unref (xdata); + return 0; } @@ -206,7 +378,7 @@ afr_do_writev (call_frame_t *frame, xlator_t *this) } transaction_frame->local = local; - frame->local = NULL; + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local->op = GF_FOP_WRITE; @@ -214,10 +386,17 @@ afr_do_writev (call_frame_t *frame, xlator_t *this) local->transaction.fop = afr_writev_wind; local->transaction.done = afr_writev_done; - local->transaction.unwind = afr_writev_unwind; + 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 { @@ -226,11 +405,15 @@ afr_do_writev (call_frame_t *frame, xlator_t *this) local->cont.writev.count); } - afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); + 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 == -1) { + if (op_ret < 0) { if (transaction_frame) AFR_STACK_DESTROY (transaction_frame); AFR_STACK_UNWIND (writev, frame, op_ret, op_errno, NULL, NULL, NULL); @@ -239,153 +422,74 @@ out: return 0; } -static int -afr_prepare_loc (call_frame_t *frame, fd_t *fd) +static void +afr_trigger_open_fd_self_heal (fd_t *fd, xlator_t *this) { - afr_local_t *local = NULL; - char *name = NULL; - char *path = NULL; - int ret = 0; - - if ((!fd) || (!fd->inode)) - return -1; - - local = frame->local; - ret = inode_path (fd->inode, NULL, (char **)&path); - if (ret <= 0) { - gf_log (frame->this->name, GF_LOG_DEBUG, - "Unable to get path for gfid: %s", - uuid_utoa (fd->inode->gfid)); - return -1; - } - - if (local->loc.path) { - if (strcmp (path, local->loc.path)) - gf_log (frame->this->name, GF_LOG_DEBUG, - "overwriting old loc->path %s with %s", - local->loc.path, path); - GF_FREE ((char *)local->loc.path); - } - local->loc.path = path; - - name = strrchr (local->loc.path, '/'); - if (name) - name++; - local->loc.name = name; - - if (local->loc.inode) { - inode_unref (local->loc.inode); - } - local->loc.inode = inode_ref (fd->inode); - - if (local->loc.parent) { - inode_unref (local->loc.parent); + 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; } - local->loc.parent = inode_parent (local->loc.inode, 0, NULL); - - return 0; -} - -afr_fd_paused_call_t* -afr_paused_call_create (call_frame_t *frame) -{ - afr_local_t *local = NULL; - afr_fd_paused_call_t *paused_call = NULL; + frame = create_frame (this, this->ctx->pool); + if (!frame) + goto out; + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; - GF_ASSERT (local->fop_call_continue); - - paused_call = GF_CALLOC (1, sizeof (*paused_call), - gf_afr_fd_paused_call_t); - if (paused_call) { - INIT_LIST_HEAD (&paused_call->call_list); - paused_call->frame = frame; - } - - return paused_call; -} - -static int -afr_pause_fd_fop (call_frame_t *frame, xlator_t *this, afr_fd_ctx_t *fd_ctx) -{ - afr_fd_paused_call_t *paused_call = NULL; - int ret = 0; - - paused_call = afr_paused_call_create (frame); - if (paused_call) - list_add (&paused_call->call_list, &fd_ctx->paused_calls); - else - ret = -ENOMEM; - - return ret; -} + ret = afr_local_init (local, this->private, &op_errno); + if (ret < 0) + goto out; -static void -afr_trigger_open_fd_self_heal (call_frame_t *frame, xlator_t *this) -{ - afr_local_t *local = NULL; - afr_self_heal_t *sh = NULL; - inode_t *inode = NULL; - char *reason = NULL; + local->loc.inode = inode_ref (fd->inode); + ret = loc_path (&local->loc, NULL); + if (ret < 0) + goto out; - local = frame->local; sh = &local->self_heal; - inode = local->fd->inode; - - sh->do_missing_entry_self_heal = _gf_true; - sh->do_gfid_self_heal = _gf_true; - sh->do_data_self_heal = _gf_true; + 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, inode, _gf_true, inode->ia_type, - reason, NULL, NULL); + afr_launch_self_heal (frame, this, fd->inode, _gf_true, + fd->inode->ia_type, reason, NULL, NULL); + return; +out: + AFR_STACK_DESTROY (frame); } -int -afr_open_fd_fix (call_frame_t *frame, xlator_t *this, gf_boolean_t pause_fop) -{ - int ret = 0; - int i = 0; - afr_fd_ctx_t *fd_ctx = NULL; - inode_t *inode = NULL; - gf_boolean_t need_self_heal = _gf_false; - int *need_open = NULL; - int need_open_count = 0; - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - gf_boolean_t fop_continue = _gf_true; +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; - local = frame->local; priv = this->private; - GF_ASSERT (local->fd); - - inode = local->fd->inode; - //gfid is not set in rebalance, that case needs to be handled. - if (fd_is_anonymous (local->fd) || - !inode || uuid_is_null (inode->gfid)) { - fop_continue = _gf_true; + if (!afr_is_fd_fixable (fd)) goto out; - } - - if (pause_fop) - GF_ASSERT (local->fop_call_continue); - ret = afr_prepare_loc (frame, local->fd); - if (ret < 0) { - //File does not exist we cant open it. - ret = 0; + fd_ctx = afr_fd_ctx_get (fd, this); + if (!fd_ctx) goto out; - } - - fd_ctx = afr_fd_ctx_get (local->fd, this); - if (!fd_ctx) { - ret = -EINVAL; - goto out; - } - LOCK (&local->fd->lock); + LOCK (&fd->lock); { if (fd_ctx->up_count < priv->up_count) { need_self_heal = _gf_true; @@ -393,56 +497,34 @@ afr_open_fd_fix (call_frame_t *frame, xlator_t *this, gf_boolean_t pause_fop) fd_ctx->down_count = priv->down_count; } + need_open = alloca (priv->child_count * sizeof (*need_open)); for (i = 0; i < priv->child_count; i++) { - if ((fd_ctx->opened_on[i] == AFR_FD_NOT_OPENED) && - local->child_up[i]) { - fd_ctx->opened_on[i] = AFR_FD_OPENING; - if (!need_open) - need_open = GF_CALLOC (priv->child_count, - sizeof (*need_open), - gf_afr_mt_int32_t); - need_open[i] = 1; - need_open_count++; - } else if (pause_fop && local->child_up[i] && - (fd_ctx->opened_on[i] == AFR_FD_OPENING)) { - local->fop_paused = _gf_true; - } - } + 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; - if (local->fop_paused) { - GF_ASSERT (pause_fop); - gf_log (this->name, GF_LOG_INFO, "Pause fd %p", - local->fd); - ret = afr_pause_fd_fop (frame, this, fd_ctx); - if (ret) - goto unlock; - fop_continue = _gf_false; + need_open[i] = 1; + need_open_count++; } } -unlock: - UNLOCK (&local->fd->lock); - if (ret) { - gf_log (this->name, GF_LOG_ERROR, "Failed to fix fd for %s", - local->loc.path); - fop_continue = _gf_false; + UNLOCK (&fd->lock); + if (ret) goto out; - } if (need_self_heal) - afr_trigger_open_fd_self_heal (frame, this); + afr_trigger_open_fd_self_heal (fd, this); if (!need_open_count) goto out; - gf_log (this->name, GF_LOG_INFO, "Opening fd %p", local->fd); - afr_fix_open (frame, this, fd_ctx, need_open_count, need_open); - fop_continue = _gf_false; + afr_fix_open (this, fd, need_open_count, need_open); out: - if (need_open) - GF_FREE (need_open); - if (fop_continue && local->fop_call_continue) - local->fop_call_continue (frame, this); - return ret; + return; } int @@ -461,6 +543,11 @@ afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, priv = this->private; + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } + QUORUM_CHECK(writev,out); AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); @@ -477,13 +564,15 @@ afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, local->cont.writev.iobref = iobref_ref (iobref); local->fd = fd_ref (fd); - local->fop_call_continue = afr_do_writev; - ret = afr_open_fd_fix (frame, this, _gf_true); - if (ret) { - op_errno = -ret; - goto out; - } + /* 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)); + + afr_open_fd_fix (fd, this); + + afr_do_writev (frame, this); ret = 0; out: @@ -517,8 +606,8 @@ afr_truncate_unwind (call_frame_t *frame, xlator_t *this) if (main_frame) { AFR_STACK_UNWIND (truncate, main_frame, local->op_ret, local->op_errno, - &local->cont.truncate.prebuf, - &local->cont.truncate.postbuf, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, NULL); } @@ -532,14 +621,11 @@ afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, struct iatt *postbuf, dict_t *xdata) { 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; read_child = afr_inode_get_read_ctx (this, local->loc.inode, NULL); @@ -549,38 +635,22 @@ afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->read_child_returned = _gf_true; } - if (afr_fop_failed (op_ret, op_errno) && op_errno != EFBIG) - afr_transaction_fop_failed (frame, this, child_index); - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - local->cont.truncate.prebuf = *prebuf; - local->cont.truncate.postbuf = *postbuf; - } - - if (child_index == read_child) { - local->cont.truncate.prebuf = *prebuf; - local->cont.truncate.postbuf = *postbuf; - } - - local->success_count++; - - if ((local->success_count >= priv->wait_count) - && local->read_child_returned) { - need_unwind = 1; - } + if (prebuf->ia_size != postbuf->ia_size) + local->stable_write = _gf_false; } - local->op_errno = op_errno; + __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); if (call_count == 0) { + if (local->stable_write && afr_txn_nothing_failed (frame, this)) + local->transaction.unwind (frame, this); + local->transaction.resume (frame, this); } @@ -608,6 +678,7 @@ afr_truncate_wind (call_frame_t *frame, xlator_t *this) } local->call_count = call_count; + local->stable_write = _gf_true; for (i = 0; i < priv->child_count; i++) { if (local->transaction.pre_op[i]) { @@ -686,7 +757,11 @@ afr_truncate (call_frame_t *frame, xlator_t *this, local->transaction.start = offset; local->transaction.len = 0; - afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); + ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); + if (ret < 0) { + op_errno = -ret; + goto out; + } ret = 0; out: @@ -724,8 +799,8 @@ afr_ftruncate_unwind (call_frame_t *frame, xlator_t *this) if (main_frame) { AFR_STACK_UNWIND (ftruncate, main_frame, local->op_ret, local->op_errno, - &local->cont.ftruncate.prebuf, - &local->cont.ftruncate.postbuf, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, NULL); } return 0; @@ -738,14 +813,11 @@ afr_ftruncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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); @@ -755,38 +827,22 @@ afr_ftruncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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.ftruncate.prebuf = *prebuf; - local->cont.ftruncate.postbuf = *postbuf; - } - - if (child_index == read_child) { - local->cont.ftruncate.prebuf = *prebuf; - local->cont.ftruncate.postbuf = *postbuf; - } - - local->success_count++; - - if ((local->success_count >= priv->wait_count) - && local->read_child_returned) { - need_unwind = 1; - } + if (prebuf->ia_size != postbuf->ia_size) + local->stable_write = _gf_false; } - local->op_errno = op_errno; + __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); if (call_count == 0) { + if (local->stable_write && afr_txn_nothing_failed (frame, this)) + local->transaction.unwind (frame, this); + local->transaction.resume (frame, this); } @@ -814,6 +870,7 @@ afr_ftruncate_wind (call_frame_t *frame, xlator_t *this) } local->call_count = call_count; + local->stable_write = _gf_true; for (i = 0; i < priv->child_count; i++) { if (local->transaction.pre_op[i]) { @@ -878,11 +935,15 @@ afr_do_ftruncate (call_frame_t *frame, xlator_t *this) local->transaction.start = local->cont.ftruncate.offset; local->transaction.len = 0; - afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); + 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 == -1) { + if (op_ret < 0) { if (transaction_frame) AFR_STACK_DESTROY (transaction_frame); AFR_STACK_UNWIND (ftruncate, frame, op_ret, op_errno, NULL, @@ -909,6 +970,10 @@ afr_ftruncate (call_frame_t *frame, xlator_t *this, priv = this->private; + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } QUORUM_CHECK(ftruncate,out); AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); @@ -921,13 +986,10 @@ afr_ftruncate (call_frame_t *frame, xlator_t *this, local->cont.ftruncate.offset = offset; local->fd = fd_ref (fd); - local->fop_call_continue = afr_do_ftruncate; - ret = afr_open_fd_fix (frame, this, _gf_true); - if (ret) { - op_errno = -ret; - goto out; - } + afr_open_fd_fix (fd, this); + + afr_do_ftruncate (frame, this); ret = 0; out: @@ -963,8 +1025,8 @@ afr_setattr_unwind (call_frame_t *frame, xlator_t *this) if (main_frame) { AFR_STACK_UNWIND (setattr, main_frame, local->op_ret, local->op_errno, - &local->cont.setattr.preop_buf, - &local->cont.setattr.postop_buf, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, NULL); } @@ -995,29 +1057,14 @@ afr_setattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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.setattr.preop_buf = *preop; - local->cont.setattr.postop_buf = *postop; - } - - if (child_index == read_child) { - local->cont.setattr.preop_buf = *preop; - local->cont.setattr.postop_buf = *postop; - } - - local->success_count++; + __inode_write_fop_cbk (frame, child_index, read_child, this, + &op_ret, &op_errno, preop, postop, + xdata); - if ((local->success_count >= priv->wait_count) - && local->read_child_returned) { - need_unwind = 1; - } + if ((local->success_count >= priv->wait_count) + && local->read_child_returned) { + need_unwind = 1; } - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -1134,7 +1181,11 @@ afr_setattr (call_frame_t *frame, xlator_t *this, 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; + } ret = 0; out: @@ -1168,8 +1219,8 @@ afr_fsetattr_unwind (call_frame_t *frame, xlator_t *this) if (main_frame) { AFR_STACK_UNWIND (fsetattr, main_frame, local->op_ret, local->op_errno, - &local->cont.fsetattr.preop_buf, - &local->cont.fsetattr.postop_buf, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, NULL); } @@ -1200,29 +1251,14 @@ afr_fsetattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->read_child_returned = _gf_true; } - if (afr_fop_failed (op_ret, op_errno)) - afr_transaction_fop_failed (frame, this, child_index); + __inode_write_fop_cbk (frame, child_index, read_child, this, + &op_ret, &op_errno, preop, postop, + xdata); - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - local->cont.fsetattr.preop_buf = *preop; - local->cont.fsetattr.postop_buf = *postop; - } - - if (child_index == read_child) { - local->cont.fsetattr.preop_buf = *preop; - local->cont.fsetattr.postop_buf = *postop; - } - - local->success_count++; - - if ((local->success_count >= priv->wait_count) - && local->read_child_returned) { - need_unwind = 1; - } + if ((local->success_count >= priv->wait_count) + && local->read_child_returned) { + need_unwind = 1; } - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -1310,6 +1346,11 @@ afr_fsetattr (call_frame_t *frame, xlator_t *this, priv = this->private; + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } + QUORUM_CHECK(fsetattr,out); transaction_frame = copy_frame (frame); @@ -1334,17 +1375,17 @@ afr_fsetattr (call_frame_t *frame, xlator_t *this, local->fd = fd_ref (fd); - ret = afr_open_fd_fix (transaction_frame, this, _gf_false); - if (ret) { - op_errno = -ret; - goto out; - } + afr_open_fd_fix (fd, this); 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; + } ret = 0; out: @@ -1390,28 +1431,23 @@ 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; + 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); { - 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; - } + __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; } - - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -1488,17 +1524,16 @@ afr_setxattr (call_frame_t *frame, xlator_t *this, afr_private_t *priv = NULL; afr_local_t *local = NULL; call_frame_t *transaction_frame = NULL; - data_pair_t *trav = NULL; int ret = -1; int op_errno = EINVAL; VALIDATE_OR_GOTO (this, out); GF_IF_INTERNAL_XATTR_GOTO ("trusted.afr.*", dict, - trav, op_errno, out); + op_errno, out); GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.afr.*", dict, - trav, op_errno, out); + op_errno, out); VALIDATE_OR_GOTO (frame, out); VALIDATE_OR_GOTO (this->private, out); @@ -1532,7 +1567,11 @@ afr_setxattr (call_frame_t *frame, xlator_t *this, 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; + } ret = 0; out: @@ -1577,28 +1616,24 @@ 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; + 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); { - 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; - } + __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; } - - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -1677,23 +1712,24 @@ afr_fsetxattr (call_frame_t *frame, xlator_t *this, call_frame_t *transaction_frame = NULL; int ret = -1; int op_errno = EINVAL; - data_pair_t *trav = NULL; VALIDATE_OR_GOTO (frame, out); VALIDATE_OR_GOTO (this, out); VALIDATE_OR_GOTO (this->private, out); GF_IF_INTERNAL_XATTR_GOTO ("trusted.afr.*", dict, - trav, op_errno, out); + op_errno, out); GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.afr.*", dict, - trav, op_errno, out); - - if (ret) - goto out; + op_errno, out); priv = this->private; + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } + QUORUM_CHECK(fsetxattr,out); AFR_LOCAL_ALLOC_OR_GOTO (local, out); @@ -1724,7 +1760,11 @@ afr_fsetxattr (call_frame_t *frame, xlator_t *this, 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; + } ret = 0; out: @@ -1772,28 +1812,23 @@ 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; + 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); { - 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; - } + __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; } - - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -1914,7 +1949,11 @@ afr_removexattr (call_frame_t *frame, xlator_t *this, 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; + } ret = 0; out: @@ -1957,28 +1996,24 @@ int 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; - int call_count = -1; - int need_unwind = 0; + 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); { - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - } - local->success_count++; + __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; - } + if (local->success_count == priv->wait_count) { + need_unwind = 1; } - - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -2071,6 +2106,10 @@ afr_fremovexattr (call_frame_t *frame, xlator_t *this, VALIDATE_OR_GOTO (this->private, out); priv = this->private; + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } QUORUM_CHECK(fremovexattr, out); @@ -2103,11 +2142,15 @@ afr_fremovexattr (call_frame_t *frame, xlator_t *this, local->transaction.start = LLONG_MAX - 1; local->transaction.len = 0; - afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION); + 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 == -1) { + if (op_ret < 0) { if (transaction_frame) AFR_STACK_DESTROY (transaction_frame); AFR_STACK_UNWIND (fremovexattr, frame, op_ret, op_errno, NULL); @@ -2115,3 +2158,704 @@ out: return 0; } + +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; +} + +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 = 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); + + if ((local->success_count >= priv->wait_count) + && local->read_child_returned) { + 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; +} + +static int +afr_fallocate_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_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; +} + +static int +afr_fallocate_done (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + + local = frame->local; + + local->transaction.unwind (frame, this); + + AFR_STACK_DESTROY (frame); + + return 0; +} + +static int +afr_do_fallocate (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; + + local = frame->local; + + transaction_frame = copy_frame (frame); + if (!transaction_frame) { + goto out; + } + + transaction_frame->local = local; + frame->local = NULL; + + local->op = GF_FOP_FALLOCATE; + + local->transaction.fop = afr_fallocate_wind; + local->transaction.done = afr_fallocate_done; + local->transaction.unwind = afr_fallocate_unwind; + + local->transaction.main_frame = frame; + + 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_fallocate (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode, + 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; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + + priv = this->private; + + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } + QUORUM_CHECK(fallocate,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->cont.fallocate.mode = mode; + local->cont.fallocate.offset = offset; + local->cont.fallocate.len = len; + + local->fd = fd_ref (fd); + + afr_open_fd_fix (fd, this); + + afr_do_fallocate (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); + } + + return 0; +} + +/* }}} */ + +/* {{{ discard */ + +static int +afr_discard_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 (discard, main_frame, local->op_ret, + local->op_errno, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, + NULL); + } + return 0; +} + +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; + + 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); + + if ((local->success_count >= priv->wait_count) + && local->read_child_returned) { + 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; +} + +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; +} + +static int +afr_discard_done (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + + local = frame->local; + + local->transaction.unwind (frame, this); + + AFR_STACK_DESTROY (frame); + + 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; + + local = frame->local; + + transaction_frame = copy_frame (frame); + if (!transaction_frame) { + goto out; + } + + transaction_frame->local = local; + frame->local = NULL; + + local->op = GF_FOP_DISCARD; + + local->transaction.fop = afr_discard_wind; + local->transaction.done = afr_discard_done; + local->transaction.unwind = afr_discard_unwind; + + 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_discard (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; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + + priv = this->private; + + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } + QUORUM_CHECK(discard, 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->cont.discard.offset = offset; + local->cont.discard.len = len; + + 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; +} + + +/* {{{ zerofill */ + +static int +afr_zerofill_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 (zerofill, main_frame, local->op_ret, + local->op_errno, + &local->cont.zerofill.prebuf, + &local->cont.zerofill.postbuf, + NULL); + } + return 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; +} + +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; +} + +static int +afr_zerofill_done (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + + local = frame->local; + + local->transaction.unwind (frame, this); + + AFR_STACK_DESTROY (frame); + + return 0; +} + +static int +afr_do_zerofill(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; + + local = frame->local; + + transaction_frame = copy_frame (frame); + if (!transaction_frame) { + goto out; + } + + transaction_frame->local = local; + frame->local = NULL; + + local->op = GF_FOP_ZEROFILL; + + local->transaction.fop = afr_zerofill_wind; + local->transaction.done = afr_zerofill_done; + local->transaction.unwind = afr_zerofill_unwind; + + local->transaction.main_frame = frame; + + local->transaction.start = local->cont.zerofill.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 (zerofill, frame, op_ret, op_errno, NULL, + NULL, NULL); + } + + return 0; +} + +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; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + + priv = this->private; + + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } + QUORUM_CHECK(zerofill, 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->cont.zerofill.offset = offset; + local->cont.zerofill.len = len; + + local->fd = fd_ref (fd); + + afr_open_fd_fix (fd, this); + + afr_do_zerofill(frame, this); + + ret = 0; +out: + 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 437ee687e..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) 2007-2011 Gluster, Inc. <http://www.gluster.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__ @@ -77,4 +68,15 @@ 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 index dfdde2975..060d78f35 100644 --- a/xlators/cluster/afr/src/afr-lk-common.c +++ b/xlators/cluster/afr/src/afr-lk-common.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2007-2011 Gluster, Inc. <http://www.gluster.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" @@ -64,7 +55,36 @@ } while (0); int -afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index); +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; @@ -131,16 +151,9 @@ internal_lock_count (call_frame_t *frame, xlator_t *this) local = frame->local; priv = this->private; - if (local->fd) { - for (i = 0; i < priv->child_count; i++) { - if (local->child_up[i] && local->fd_open_on[i]) - ++call_count; - } - } else { - for (i = 0; i < priv->child_count; i++) { - if (local->child_up[i]) - ++call_count; - } + for (i = 0; i < priv->child_count; i++) { + if (local->child_up[i]) + ++call_count; } return call_count; @@ -296,11 +309,10 @@ afr_trace_inodelk_out (call_frame_t *frame, xlator_t *this, afr_print_verdict (op_ret, op_errno, verdict); gf_log (this->name, GF_LOG_INFO, - "[%s %s] [%s] Lockee={%s} Number={%llu}", + "[%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, - lockee, + verdict, lkowner_utoa (&frame->root->lk_owner), lockee, (unsigned long long) int_lock->lock_number); } @@ -339,10 +351,13 @@ 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 child_index) + 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]; @@ -350,28 +365,40 @@ afr_trace_entrylk_in (call_frame_t *frame, xlator_t *this, 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, &local->loc, local->fd, child_index); + 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}", + "[%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); + (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 child_index) + 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]; @@ -380,20 +407,30 @@ afr_trace_entrylk_out (call_frame_t *frame, xlator_t *this, local = frame->local; int_lock = &local->internal_lock; + priv = this->private; - afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index); + 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}", + "[%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); + (unsigned long long) int_lock->lock_number, + cookie); } @@ -446,6 +483,47 @@ is_afr_lock_transaction (afr_local_t *local) 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) { @@ -463,8 +541,13 @@ initialize_entrylk_variables (call_frame_t *frame, xlator_t *this) int_lock->lock_op_ret = -1; int_lock->lock_op_errno = 0; - for (i = 0; i < priv->child_count; i++) { - int_lock->entry_locked_nodes[i] = 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; @@ -476,19 +559,23 @@ 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; - int i = 0; + afr_inodelk_t *inodelk = NULL; priv = this->private; local = frame->local; int_lock = &local->internal_lock; - int_lock->inodelk_lock_count = 0; - int_lock->lock_op_ret = -1; - int_lock->lock_op_errno = 0; + inodelk = afr_get_inodelk (int_lock, int_lock->domain); - for (i = 0; i < priv->child_count; i++) { - int_lock->inode_locked_nodes[i] = 0; - } + 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; } @@ -510,6 +597,18 @@ lower_path (loc_t *l1, const char *b1, loc_t *l2, const char *b2) } 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) { @@ -557,7 +656,9 @@ afr_unlock_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, { 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; @@ -566,14 +667,18 @@ afr_unlock_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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 %d " - "unlock by %s", local->loc.path, child_index, + 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)); } - int_lock->inode_locked_nodes[child_index] &= LOCKED_NO; + 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; @@ -587,6 +692,7 @@ 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,}; @@ -596,19 +702,20 @@ afr_unlock_inodelk (call_frame_t *frame, xlator_t *this) int i = 0; int piggyback = 0; afr_fd_ctx_t *fd_ctx = NULL; - gf_boolean_t fd_lock_owner = _gf_false; local = frame->local; int_lock = &local->internal_lock; priv = this->private; - flock.l_start = int_lock->lk_flock.l_start; - flock.l_len = int_lock->lk_flock.l_len; + 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 (int_lock->inode_locked_nodes, + call_count = afr_locked_nodes_count (inodelk->locked_nodes, priv->child_count); int_lock->lk_call_count = call_count; @@ -624,18 +731,12 @@ afr_unlock_inodelk (call_frame_t *frame, xlator_t *this) fd_ctx = afr_fd_ctx_get (local->fd, this); for (i = 0; i < priv->child_count; i++) { - if ((int_lock->inode_locked_nodes[i] & LOCKED_YES) - != LOCKED_YES) + if ((inodelk->locked_nodes[i] & LOCKED_YES) != LOCKED_YES) continue; if (local->fd) { flock_use = &flock; if (!local->transaction.eager_lock[i]) { - if (fd_lock_owner) { - afr_set_lk_owner (frame, this, - frame->root); - fd_lock_owner = _gf_false; - } goto wind; } @@ -660,10 +761,6 @@ afr_unlock_inodelk (call_frame_t *frame, xlator_t *this) continue; } - if (!fd_lock_owner) { - afr_set_lk_owner (frame, this, local->fd); - fd_lock_owner = _gf_true; - } flock_use = &full_flock; wind: AFR_TRACE_INODELK_IN (frame, this, @@ -675,7 +772,7 @@ afr_unlock_inodelk (call_frame_t *frame, xlator_t *this) (void *) (long)i, priv->children[i], priv->children[i]->fops->finodelk, - this->name, local->fd, + int_lock->domain, local->fd, F_SETLK, flock_use, NULL); if (!--call_count) @@ -690,7 +787,7 @@ afr_unlock_inodelk (call_frame_t *frame, xlator_t *this) (void *) (long)i, priv->children[i], priv->children[i]->fops->inodelk, - this->name, &local->loc, + int_lock->domain, &local->loc, F_SETLK, &flock, NULL); if (!--call_count) @@ -706,13 +803,22 @@ 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; - int32_t child_index = (long)cookie; + 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, NULL, op_ret, - op_errno, child_index); + 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, @@ -720,6 +826,7 @@ afr_unlock_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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; @@ -728,24 +835,22 @@ afr_unlock_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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; - const char *basename = NULL; - loc_t *loc = NULL; - int call_count = 0; - int i = -1; + 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; - basename = int_lock->lk_basename; - if (int_lock->lk_loc) - loc = int_lock->lk_loc; + call_count = afr_lockee_locked_nodes_count (int_lock); - call_count = afr_locked_nodes_count (int_lock->entry_locked_nodes, - priv->child_count); int_lock->lk_call_count = call_count; if (!call_count){ @@ -755,18 +860,22 @@ afr_unlock_entrylk (call_frame_t *frame, xlator_t *this) goto out; } - for (i = 0; i < priv->child_count; i++) { - if (int_lock->entry_locked_nodes[i] & LOCKED_YES) { - AFR_TRACE_ENTRYLK_IN (frame, this, - AFR_ENTRYLK_NB_TRANSACTION, - AFR_UNLOCK_OP, basename, i); + 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[i], - priv->children[i]->fops->entrylk, - this->name, - loc, basename, + 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) @@ -783,13 +892,20 @@ 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; - int child_index = (long) cookie; + 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) { @@ -805,6 +921,8 @@ afr_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->op_errno = op_errno; int_lock->lock_op_errno = op_errno; } + + int_lock->lk_attempted_count++; } UNLOCK (&frame->lock); @@ -813,10 +931,17 @@ afr_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this, afr_unlock (frame, this); } else { if (op_ret == 0) { - int_lock->locked_nodes[child_index] |= LOCKED_YES; - int_lock->lock_count++; + 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, child_index + 1); + afr_lock_blocking (frame, this, cky + 1); } return 0; @@ -836,78 +961,6 @@ afr_blocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, } static int32_t -afr_lock_lower_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_private_t *priv = NULL; - afr_local_t *local = NULL; - loc_t *lower = NULL; - loc_t *higher = NULL; - const char *higher_name = NULL; - int child_index = (long) cookie; - - priv = this->private; - local = frame->local; - int_lock = &local->internal_lock; - - 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; - } - - local->op_errno = op_errno; - } - } - UNLOCK (&frame->lock); - - if (op_ret != 0) { - afr_unlock (frame, this); - goto out; - } else { - int_lock->lower_locked_nodes[child_index] |= LOCKED_LOWER; - int_lock->lock_count++; - } - - /* The lower path has been locked. Now lock the higher path */ - - lower = lower_path (&local->transaction.parent_loc, - local->transaction.basename, - &local->transaction.new_parent_loc, - 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); - - AFR_TRACE_ENTRYLK_IN (frame, this, AFR_ENTRYLK_TRANSACTION, - AFR_LOCK_OP, higher_name, child_index); - - - 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, NULL); - -out: - 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) { @@ -923,6 +976,7 @@ 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; @@ -933,18 +987,16 @@ afr_copy_locked_nodes (call_frame_t *frame, xlator_t *this) switch (local->transaction.type) { case AFR_DATA_TRANSACTION: case AFR_METADATA_TRANSACTION: - memcpy (int_lock->inode_locked_nodes, - int_lock->locked_nodes, - priv->child_count); - int_lock->inodelk_lock_count = int_lock->lock_count; + 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: - memcpy (int_lock->entry_locked_nodes, - int_lock->locked_nodes, - priv->child_count); - int_lock->entrylk_lock_count = int_lock->lock_count; + /*entrylk_count is being used in both non-blocking and blocking + * modes */ break; } @@ -952,25 +1004,67 @@ afr_copy_locked_nodes (call_frame_t *frame, xlator_t *this) } +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 child_index) +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; - loc_t *lower = NULL; - const char *lower_name = NULL; struct gf_flock flock = {0,}; uint64_t ctx = 0; int ret = 0; - - local = frame->local; - int_lock = &local->internal_lock; - priv = this->private; - - flock.l_start = int_lock->lk_flock.l_start; - flock.l_len = int_lock->lk_flock.l_len; - flock.l_type = int_lock->lk_flock.l_type; + 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); @@ -989,42 +1083,26 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index) return 0; } - - /* skip over children that or down - or don't have the fd open */ - - while ((child_index < priv->child_count) - && (!local->child_up[child_index] || - !local->fd_open_on[child_index])) - - child_index++; - } else { - /* skip over children that are down */ - while ((child_index < priv->child_count) - && !local->child_up[child_index]) - child_index++; } - if ((child_index == priv->child_count) && - 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; + 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"); - afr_copy_locked_nodes (frame, this); + local->op_ret = -1; + int_lock->lock_op_ret = -1; - afr_unlock(frame, this); + afr_copy_locked_nodes (frame, this); - return 0; + afr_unlock(frame, this); + return 0; + } } - if ((child_index == priv->child_count) - || (int_lock->lock_count == int_lock->lk_expected_count)) { - + if (int_lock->lk_expected_count == int_lock->lk_attempted_count) { /* we're done locking */ gf_log (this->name, GF_LOG_DEBUG, @@ -1037,6 +1115,11 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index) 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: @@ -1051,7 +1134,7 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index) (void *) (long) child_index, priv->children[child_index], priv->children[child_index]->fops->finodelk, - this->name, local->fd, + int_lock->domain, local->fd, F_SETLKW, &flock, NULL); } else { @@ -1064,50 +1147,29 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index) (void *) (long) child_index, priv->children[child_index], priv->children[child_index]->fops->inodelk, - this->name, &local->loc, + int_lock->domain, &local->loc, F_SETLKW, &flock, NULL); } break; case AFR_ENTRY_RENAME_TRANSACTION: - { - 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); - - AFR_TRACE_ENTRYLK_IN (frame, this, AFR_ENTRYLK_TRANSACTION, - AFR_LOCK_OP, lower_name, child_index); - - - STACK_WIND_COOKIE (frame, afr_lock_lower_cbk, - (void *) (long) child_index, - priv->children[child_index], - priv->children[child_index]->fops->entrylk, - this->name, lower, lower_name, - ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL); - - break; - } - 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, local->transaction.basename, - child_index); + 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) child_index, + (void *) (long) cookie, priv->children[child_index], priv->children[child_index]->fops->fentrylk, - this->name, local->fd, - local->transaction.basename, + int_lock->domain, local->fd, + int_lock->lockee[lockee_no].basename, ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL); } else { AFR_TRACE_ENTRYLK_IN (frame, this, @@ -1116,12 +1178,12 @@ afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index) child_index); STACK_WIND_COOKIE (frame, afr_blocking_entrylk_cbk, - (void *) (long) child_index, + (void *) (long) cookie, priv->children[child_index], priv->children[child_index]->fops->entrylk, - this->name, - &local->transaction.parent_loc, - local->transaction.basename, + int_lock->domain, + &int_lock->lockee[lockee_no].loc, + int_lock->lockee[lockee_no].basename, ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL); } @@ -1150,11 +1212,12 @@ afr_blocking_lock (call_frame_t *frame, xlator_t *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_expected_count = 2 * up_count; - //fallthrough - case AFR_ENTRY_TRANSACTION: + int_lock->lk_call_count = int_lock->lk_expected_count + = (int_lock->lockee_count * + up_count); initialize_entrylk_variables (frame, this); break; } @@ -1170,38 +1233,51 @@ afr_nonblocking_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, { afr_internal_lock_t *int_lock = NULL; afr_local_t *local = NULL; - int call_count = 0; - int child_index = (long) cookie; + 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, NULL, op_ret, + AFR_LOCK_OP, + int_lock->lockee[lockee_no].basename, op_ret, op_errno, (long) cookie); - LOCK (&frame->lock); - { - call_count = --int_lock->lk_call_count; - } - UNLOCK (&frame->lock); - - if (op_ret < 0 ) { - if (op_errno == ENOSYS) { + 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; + 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++; + } - int_lock->lock_op_errno = op_errno; - local->op_errno = op_errno; - } - } else if (op_ret == 0) { - int_lock->entry_locked_nodes[child_index] |= LOCKED_YES; - 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, @@ -1228,42 +1304,26 @@ afr_nonblocking_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, return 0; } -void -afr_mark_fd_open_on (afr_local_t *local, afr_fd_ctx_t *fd_ctx, - size_t child_count) -{ - int i = 0; - - GF_ASSERT (local->fd_open_on); - - memset (local->fd_open_on, 0, sizeof (*local->fd_open_on)*child_count); - for (i = 0; i < child_count; i++) - if (fd_ctx->opened_on[i] == AFR_FD_OPENED) - local->fd_open_on[i] = 1; -} - 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; - const char *basename = NULL; - loc_t *loc = NULL; - int32_t call_count = 0; + 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); - basename = int_lock->lk_basename; - if (int_lock->lk_loc) - loc = int_lock->lk_loc; - if (local->fd) { fd_ctx = afr_fd_ctx_get (local->fd, this); if (!fd_ctx) { @@ -1276,11 +1336,11 @@ afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this) local->op_errno = EINVAL; int_lock->lock_op_errno = EINVAL; + afr_unlock (frame, this); return -1; } - afr_mark_fd_open_on (local, fd_ctx, priv->child_count); - call_count = internal_lock_count (frame, this); + 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; @@ -1293,46 +1353,52 @@ afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this) /* Send non-blocking entrylk 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] && local->fd_open_on[i]) { - AFR_TRACE_ENTRYLK_IN (frame, this, - AFR_ENTRYLK_NB_TRANSACTION, - AFR_LOCK_OP, basename, i); + 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[i], - priv->children[i]->fops->fentrylk, + priv->children[index], + priv->children[index]->fops->fentrylk, this->name, local->fd, - basename, + int_lock->lockee[lockee_no].basename, ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL); + if (!--call_count) + break; } } } else { - GF_ASSERT (loc); - - call_count = internal_lock_count (frame, this); + 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 < priv->child_count; i++) { - if (local->child_up[i]) { - AFR_TRACE_ENTRYLK_IN (frame, this, - AFR_ENTRYLK_NB_TRANSACTION, - AFR_LOCK_OP, basename, i); + 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[i], - priv->children[i]->fops->entrylk, - this->name, loc, basename, + 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; - } } } @@ -1345,6 +1411,7 @@ 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; @@ -1353,58 +1420,57 @@ afr_nonblocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 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 (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 { - int_lock->inode_locked_nodes[child_index] - |= LOCKED_YES; - int_lock->inodelk_lock_count++; - - if (local->transaction.eager_lock && - local->transaction.eager_lock[child_index] && local->fd) { - fd_ctx = afr_fd_ctx_get (local->fd, this); - /* piggybacked */ - - if (op_ret == 1) { - /* piggybacked */ - } else if (op_ret == 0) { - /* lock acquired from server */ - LOCK (&local->fd->lock); - { - fd_ctx->lock_acquired[child_index]++; - } - UNLOCK (&local->fd->lock); - } - } - } + 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 (int_lock->inodelk_lock_count == - int_lock->lk_expected_count) { + 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; @@ -1428,6 +1494,7 @@ 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; @@ -1438,17 +1505,18 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this) struct gf_flock full_flock = {0,}; struct gf_flock *flock_use = NULL; int piggyback = 0; - gf_boolean_t fd_lock_owner = _gf_false; local = frame->local; int_lock = &local->internal_lock; priv = this->private; - flock.l_start = int_lock->lk_flock.l_start; - flock.l_len = int_lock->lk_flock.l_len; - flock.l_type = int_lock->lk_flock.l_type; + 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 = int_lock->lk_flock.l_type; + full_flock.l_type = inodelk->flock.l_type; initialize_inodelk_variables (frame, this); @@ -1464,11 +1532,11 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this) local->op_errno = EINVAL; int_lock->lock_op_errno = EINVAL; + afr_unlock (frame, this); ret = -1; goto out; } - afr_mark_fd_open_on (local, fd_ctx, priv->child_count); call_count = internal_lock_count (frame, this); int_lock->lk_call_count = call_count; int_lock->lk_expected_count = call_count; @@ -1483,22 +1551,19 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this) /* 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] || !local->fd_open_on[i]) + if (!local->child_up[i]) continue; flock_use = &flock; - if (!priv->eager_lock) { - if (fd_lock_owner) { - afr_set_lk_owner (frame, this, - frame->root); - fd_lock_owner = _gf_false; - } + 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]) { @@ -1517,10 +1582,6 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this) continue; } flock_use = &full_flock; - if (!fd_lock_owner) { - afr_set_lk_owner (frame, this, local->fd); - fd_lock_owner = _gf_true; - } wind: AFR_TRACE_INODELK_IN (frame, this, AFR_INODELK_NB_TRANSACTION, @@ -1530,7 +1591,7 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this) (void *) (long) i, priv->children[i], priv->children[i]->fops->finodelk, - this->name, local->fd, + int_lock->domain, local->fd, F_SETLK, flock_use, NULL); if (!--call_count) @@ -1552,7 +1613,7 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this) (void *) (long) i, priv->children[i], priv->children[i]->fops->inodelk, - this->name, &local->loc, + int_lock->domain, &local->loc, F_SETLK, &flock, NULL); if (!--call_count) @@ -1563,201 +1624,6 @@ out: return ret; } -static int -__is_lower_locked (call_frame_t *frame, xlator_t *this) -{ - afr_internal_lock_t *int_lock = NULL; - afr_private_t *priv = NULL; - afr_local_t *local = NULL; - int count = 0; - int i = 0; - - local = frame->local; - int_lock = &local->internal_lock; - priv = this->private; - - for (i = 0; i < priv->child_count; i++) { - if (int_lock->lower_locked_nodes[i] & LOCKED_LOWER) - count++; - } - - return count; - -} - -static int -__is_higher_locked (call_frame_t *frame, xlator_t *this) -{ - afr_internal_lock_t *int_lock = NULL; - afr_private_t *priv = NULL; - afr_local_t *local = NULL; - int count = 0; - int i = 0; - - local = frame->local; - int_lock = &local->internal_lock; - priv = this->private; - - for (i = 0; i < priv->child_count; i++) { - if (int_lock->locked_nodes[i] & LOCKED_YES) - count++; - } - - return count; - -} - -static int -afr_unlock_lower_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; - const char *basename = NULL; - loc_t *loc = NULL; - int call_count = 0; - int i = -1; - - local = frame->local; - int_lock = &local->internal_lock; - priv = this->private; - - basename = int_lock->lk_basename; - if (int_lock->lk_loc) - loc = int_lock->lk_loc; - - call_count = __is_lower_locked (frame, this); - 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 < priv->child_count; i++) { - if (int_lock->lower_locked_nodes[i] & LOCKED_LOWER) { - AFR_TRACE_ENTRYLK_IN (frame, this, - AFR_ENTRYLK_NB_TRANSACTION, - AFR_UNLOCK_OP, basename, i); - - STACK_WIND_COOKIE (frame, afr_unlock_entrylk_cbk, - (void *) (long) i, - priv->children[i], - priv->children[i]->fops->entrylk, - this->name, - loc, basename, - ENTRYLK_UNLOCK, ENTRYLK_WRLCK, NULL); - - if (!--call_count) - break; - - } - } - -out: - return 0; - -} - - -static int -afr_post_unlock_higher_cbk (call_frame_t *frame, xlator_t *this) -{ - afr_local_t *local = NULL; - - local = frame->local; - - local->transaction.done (frame, this); - return 0; -} - -static int -afr_post_unlock_lower_cbk (call_frame_t *frame, xlator_t *this) -{ - afr_internal_lock_t *int_lock = NULL; - afr_local_t *local = NULL; - loc_t *lower = NULL; - loc_t *higher = NULL; - const char *higher_name = NULL; - - local = frame->local; - int_lock = &local->internal_lock; - - lower = lower_path (&local->transaction.parent_loc, - local->transaction.basename, - &local->transaction.new_parent_loc, - 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); - - if (__is_higher_locked (frame, this)) { - gf_log (this->name, GF_LOG_DEBUG, - "unlocking higher"); - int_lock->lk_basename = higher_name; - int_lock->lk_loc = higher; - int_lock->lock_cbk = afr_post_unlock_higher_cbk; - - afr_unlock_entrylk (frame, this); - } else - local->transaction.done (frame, this); - - return 0; -} - -static int -afr_rename_unlock (call_frame_t *frame, xlator_t *this) -{ - afr_internal_lock_t *int_lock = NULL; - afr_local_t *local = NULL; - loc_t *lower = NULL; - const char *lower_name = NULL; - - local = frame->local; - int_lock = &local->internal_lock; - - 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); - - if (__is_lower_locked (frame, this)) { - gf_log (this->name, GF_LOG_DEBUG, - "unlocking lower"); - int_lock->lk_basename = lower_name; - int_lock->lk_loc = lower; - int_lock->lock_cbk = afr_post_unlock_lower_cbk; - - afr_unlock_lower_entrylk (frame, this); - } else - afr_post_unlock_lower_cbk (frame, this); - - return 0; -} - -static int -afr_rename_transaction (call_frame_t *frame, xlator_t *this) -{ - afr_local_t *local = NULL; - - local = frame->local; - - return (local->transaction.type == - AFR_ENTRY_RENAME_TRANSACTION); - -} - int32_t afr_unlock (call_frame_t *frame, xlator_t *this) { @@ -1769,10 +1635,8 @@ afr_unlock (call_frame_t *frame, xlator_t *this) if (is_afr_lock_transaction (local)) afr_unlock_inodelk (frame, this); else - if (!afr_rename_transaction (frame, this)) - afr_unlock_entrylk (frame, this); - else - afr_rename_unlock (frame, this); + afr_unlock_entrylk (frame, this); + } else { if (is_afr_lock_selfheal (local)) afr_unlock_inodelk (frame, this); @@ -2273,27 +2137,38 @@ out: return ret; } -void -afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src, +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_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; - dst_local = dst->local; - dst_lock = &dst_local->internal_lock; src_local = src->local; src_lock = &src_local->internal_lock; - if (src_lock->inode_locked_nodes) { - memcpy (dst_lock->inode_locked_nodes, - src_lock->inode_locked_nodes, - sizeof (*dst_lock->inode_locked_nodes) * child_count); - memset (src_lock->inode_locked_nodes, 0, - sizeof (*src_lock->inode_locked_nodes) * child_count); + 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->inodelk_lock_count = src_lock->inodelk_lock_count; - src_lock->inodelk_lock_count = 0; + 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 index f5292b3cc..73594f265 100644 --- a/xlators/cluster/afr/src/afr-mem-types.h +++ b/xlators/cluster/afr/src/afr-mem-types.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2008-2011 Gluster, Inc. <http://www.gluster.com> - This file is part of GlusterFS. + 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. */ @@ -50,6 +41,10 @@ enum gf_afr_mem_types_ { 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 index 370d50e7e..643a5d692 100644 --- a/xlators/cluster/afr/src/afr-open.c +++ b/xlators/cluster/afr/src/afr-open.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2007-2011 Gluster, Inc. <http://www.gluster.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> @@ -258,188 +249,126 @@ out: return 0; } -//NOTE: this function should be called with holding the lock on -//fd to which fd_ctx belongs -void -afr_get_resumable_calls (xlator_t *this, afr_fd_ctx_t *fd_ctx, - struct list_head *list) -{ - afr_fd_paused_call_t *paused_call = NULL; - afr_fd_paused_call_t *tmp = NULL; - afr_local_t *call_local = NULL; - afr_private_t *priv = NULL; - int i = 0; - gf_boolean_t call = _gf_false; - - priv = this->private; - list_for_each_entry_safe (paused_call, tmp, &fd_ctx->paused_calls, - call_list) { - call = _gf_true; - call_local = paused_call->frame->local; - for (i = 0; i < priv->child_count; i++) { - if (call_local->child_up[i] && - (fd_ctx->opened_on[i] == AFR_FD_OPENING)) - call = _gf_false; - } - - if (call) { - list_del_init (&paused_call->call_list); - list_add (&paused_call->call_list, list); - } - } -} - -void -afr_resume_calls (xlator_t *this, struct list_head *list) -{ - afr_fd_paused_call_t *paused_call = NULL; - afr_fd_paused_call_t *tmp = NULL; - afr_local_t *call_local = NULL; - - list_for_each_entry_safe (paused_call, tmp, list, call_list) { - list_del_init (&paused_call->call_list); - call_local = paused_call->frame->local; - call_local->fop_call_continue (paused_call->frame, this); - GF_FREE (paused_call); - } -} - 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) + 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; - struct list_head paused_calls = {0}; - gf_boolean_t fop_paused = _gf_false; - fd_t *local_fd = NULL; + 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; - fop_paused = local->fop_paused; if (op_ret >= 0) { - gf_log (this->name, GF_LOG_INFO, "fd for %s opened " + 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); } - local_fd = fd_ref (local->fd); - call_count = afr_frame_return (frame); - //Note: Do not access any thing using the frame outside call_count 0 - - //Note: No frame locking needed for this block of code - fd_ctx = afr_fd_ctx_get (local_fd, this); + 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); + "failed to get fd context, %p", local->fd); goto out; } - LOCK (&local_fd->lock); + 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; } - if (call_count == 0) { - INIT_LIST_HEAD (&paused_calls); - afr_get_resumable_calls (this, fd_ctx, &paused_calls); - } } - UNLOCK (&local_fd->lock); + UNLOCK (&local->fd->lock); out: - if (call_count == 0) { - afr_resume_calls (this, &paused_calls); - //If the fop is paused then resume_calls will continue the fop - if (fop_paused) - goto done; - - if (local->fop_call_continue) - local->fop_call_continue (frame, this); - else - AFR_STACK_DESTROY (frame); - } + call_count = afr_frame_return (frame); + if (call_count == 0) + AFR_STACK_DESTROY (frame); -done: - fd_unref (local_fd); return 0; } -int -afr_fix_open (call_frame_t *frame, xlator_t *this, afr_fd_ctx_t *fd_ctx, - int need_open_count, int *need_open) +void +afr_fix_open (xlator_t *this, fd_t *fd, size_t need_open_count, int *need_open) { - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - int i = 0; - call_frame_t *open_frame = NULL; - afr_local_t *open_local = NULL; - int ret = -1; - ia_type_t ia_type = IA_INVAL; - int32_t op_errno = 0; - - GF_ASSERT (fd_ctx); - GF_ASSERT (need_open_count > 0); - GF_ASSERT (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; - local = frame->local; priv = this->private; - if (!local->fop_call_continue) { - open_frame = copy_frame (frame); - if (!open_frame) { - ret = -ENOMEM; - goto out; - } - AFR_LOCAL_ALLOC_OR_GOTO (open_frame->local, out); - open_local = open_frame->local; - ret = afr_local_init (open_local, priv, &op_errno); - if (ret < 0) - goto out; - loc_copy (&open_local->loc, &local->loc); - open_local->fd = fd_ref (local->fd); - } else { - ret = 0; - open_frame = frame; - open_local = local; + + 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; } - open_local->call_count = need_open_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; + + 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: %d", + gf_log (this->name, GF_LOG_DEBUG, "need open count: %zd", need_open_count); - ia_type = open_local->fd->inode->ia_type; - GF_ASSERT (ia_type != IA_INVAL); for (i = 0; i < priv->child_count; i++) { if (!need_open[i]) continue; - if (IA_IFDIR == ia_type) { + + 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 (open_frame, afr_openfd_fix_open_cbk, + STACK_WIND_COOKIE (frame, afr_openfd_fix_open_cbk, (void*) (long) i, priv->children[i], priv->children[i]->fops->opendir, - &open_local->loc, open_local->fd, + &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 (open_frame, afr_openfd_fix_open_cbk, + STACK_WIND_COOKIE (frame, afr_openfd_fix_open_cbk, (void *)(long) i, priv->children[i], priv->children[i]->fops->open, - &open_local->loc, fd_ctx->flags, - open_local->fd, NULL); + &local->loc, + fd_ctx->flags & (~O_TRUNC), + local->fd, NULL); } } @@ -447,8 +376,7 @@ afr_fix_open (call_frame_t *frame, xlator_t *this, afr_fd_ctx_t *fd_ctx, ret = 0; out: if (op_errno) - ret = -op_errno; - if (ret && open_frame) - AFR_STACK_DESTROY (open_frame); - return ret; + 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 index d340acc75..83846f152 100644 --- a/xlators/cluster/afr/src/afr-self-heal-algorithm.c +++ b/xlators/cluster/afr/src/afr-self-heal-algorithm.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.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. */ @@ -72,8 +63,7 @@ sh_private_cleanup (call_frame_t *frame, xlator_t *this) sh = &local->self_heal; sh_priv = sh->private; - if (sh_priv) - GF_FREE (sh_priv); + GF_FREE (sh_priv); } static int @@ -110,10 +100,10 @@ sh_loop_driver_done (call_frame_t *sh_frame, xlator_t *this, } sh_private_cleanup (sh_frame, this); - if (sh->op_failed) { + 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_INFO, + gf_log (this->name, GF_LOG_DEBUG, "self-heal aborting on %s", local->loc.path); @@ -121,10 +111,10 @@ sh_loop_driver_done (call_frame_t *sh_frame, xlator_t *this, } else { GF_ASSERT (last_loop_frame); if (diff_blocks == total_blocks) { - gf_log (this->name, GF_LOG_INFO, "full self-heal " + gf_log (this->name, GF_LOG_DEBUG, "full self-heal " "completed on %s",local->loc.path); } else { - gf_log (this->name, GF_LOG_INFO, + 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, @@ -153,7 +143,7 @@ sh_loop_finish (call_frame_t *loop_frame, xlator_t *this) } if (loop_sh && loop_sh->data_lock_held) { - afr_sh_data_unlock (loop_frame, this, + afr_sh_data_unlock (loop_frame, this, this->name, sh_destroy_frame); } else { sh_destroy_frame (loop_frame, this); @@ -224,7 +214,7 @@ sh_loop_frame_create (call_frame_t *sh_frame, xlator_t *this, 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_local_copy (local, this); + new_loop_local = afr_self_heal_local_init (local, this); if (!new_loop_local) goto out; new_loop_frame->local = new_loop_local; @@ -283,10 +273,10 @@ sh_loop_start (call_frame_t *sh_frame, xlator_t *this, off_t offset, 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, - sh_loop_lock_success, sh_loop_lock_failure); + _gf_true, this->name, sh_loop_lock_success, sh_loop_lock_failure); return 0; out: - sh->op_failed = 1; + 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); @@ -317,8 +307,9 @@ sh_loop_driver (call_frame_t *sh_frame, xlator_t *this, sh_priv->loops_running--; offset = sh_priv->offset; block_size = sh->block_size; - while ((!sh->eof_reached) && (0 == sh->op_failed) && - (sh_priv->loops_running < priv->data_self_heal_window_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++; @@ -337,7 +328,8 @@ sh_loop_driver (call_frame_t *sh_frame, xlator_t *this, 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 && !sh->op_failed) + 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); @@ -348,7 +340,7 @@ sh_loop_driver (call_frame_t *sh_frame, xlator_t *this, //If we have more loops to form we should finish previous loop after //the next loop lock while (loop--) { - if (sh->op_failed) { + 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); @@ -394,7 +386,7 @@ sh_loop_return (call_frame_t *sh_frame, xlator_t *this, call_frame_t *loop_frame } if (op_ret == -1) { - sh->op_failed = 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); @@ -442,13 +434,22 @@ sh_loop_write_cbk (call_frame_t *loop_frame, void *cookie, xlator_t *this, priv->children[child_index]->name, strerror (op_errno)); - sh->op_failed = 1; + 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); } @@ -515,7 +516,7 @@ sh_loop_read_cbk (call_frame_t *loop_frame, void *cookie, if (op_ret <= 0) { if (op_ret < 0) { - sh->op_failed = 1; + 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)); @@ -537,8 +538,17 @@ sh_loop_read_cbk (call_frame_t *loop_frame, void *cookie, 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; @@ -616,7 +626,7 @@ sh_diff_checksum_cbk (call_frame_t *loop_frame, void *cookie, xlator_t *this, "checksum on %s failed on subvolume %s (%s)", sh_local->loc.path, priv->children[child_index]->name, strerror (op_errno)); - sh->op_failed = 1; + 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); @@ -654,7 +664,8 @@ sh_diff_checksum_cbk (call_frame_t *loop_frame, void *cookie, xlator_t *this, } UNLOCK (&sh_priv->lock); - if (write_needed && !sh->op_failed) { + 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, @@ -743,14 +754,15 @@ out: return sh_priv; } -void -afr_sh_transfer_lock (call_frame_t *dst, call_frame_t *src, +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; @@ -758,9 +770,12 @@ afr_sh_transfer_lock (call_frame_t *dst, call_frame_t *src, src_sh = &src_local->self_heal; GF_ASSERT (src_sh->data_lock_held); GF_ASSERT (!dst_sh->data_lock_held); - afr_lk_transfer_datalock (dst, src, child_count); + 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 @@ -782,7 +797,10 @@ afr_sh_start_loops (call_frame_t *sh_frame, xlator_t *this, ret = sh_loop_frame_create (sh_frame, this, NULL, &first_loop_frame); if (ret) goto out; - afr_sh_transfer_lock (first_loop_frame, sh_frame, priv->child_count); + 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; @@ -792,7 +810,7 @@ afr_sh_start_loops (call_frame_t *sh_frame, xlator_t *this, ret = 0; out: if (ret) { - sh->op_failed = 1; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED); sh_loop_driver_done (sh_frame, this, NULL); } return 0; diff --git a/xlators/cluster/afr/src/afr-self-heal-algorithm.h b/xlators/cluster/afr/src/afr-self-heal-algorithm.h index 27d3d171b..6b20789b1 100644 --- a/xlators/cluster/afr/src/afr-self-heal-algorithm.h +++ b/xlators/cluster/afr/src/afr-self-heal-algorithm.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> - This file is part of GlusterFS. + 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_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 af5aadc3c..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) 2008-2011 Gluster, Inc. <http://www.gluster.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" @@ -27,6 +18,28 @@ #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) { @@ -121,8 +134,8 @@ void afr_sh_set_error (afr_self_heal_t *sh, int32_t op_errno) { sh->op_ret = -1; - if (afr_error_more_important (sh->op_errno, op_errno)) - sh->op_errno = op_errno; + sh->op_errno = afr_most_important_error(sh->op_errno, op_errno, + _gf_false); } void @@ -144,13 +157,85 @@ afr_sh_print_pending_matrix (int32_t *pending_matrix[], xlator_t *this) ptr += sprintf (ptr, "%d ", pending_matrix[i][j]); } sprintf (ptr, "]"); - gf_log (this->name, GF_LOG_DEBUG, - "pending_matrix: %s", buf); + gf_log (this->name, GF_LOG_DEBUG, "pending_matrix: %s", buf); } GF_FREE (buf); } +char* +afr_get_pending_matrix_str (int32_t *pending_matrix[], xlator_t *this) +{ + 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; + + 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); + } + + ptr += sprintf (ptr, "%s", matrix_end); + +out: + return buf; +} + +void +afr_sh_print_split_brain_log (int32_t *pending_matrix[], xlator_t *this, + const char *loc) +{ + char *buf = NULL; + char *free_ptr = NULL; + + buf = afr_get_pending_matrix_str (pending_matrix, this); + if (buf) + free_ptr = buf; + else + buf = ""; + + + 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_init_pending_matrix (int32_t **pending_matrix, size_t child_count) { @@ -416,6 +501,8 @@ afr_find_biggest_witness_among_fools (int32_t *witnesses, { int i = 0; int biggest_witness = -1; + int biggest_witness_idx = -1; + int biggest_witness_cnt = -1; GF_ASSERT (witnesses); GF_ASSERT (characters); @@ -425,10 +512,21 @@ afr_find_biggest_witness_among_fools (int32_t *witnesses, if (characters[i].type != AFR_NODE_FOOL) continue; - if (biggest_witness < witnesses[i]) + 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++; } - return biggest_witness; + + if (biggest_witness_cnt != 1) + return -1; + + return biggest_witness_idx; } int @@ -456,10 +554,84 @@ afr_mark_fool_as_source_by_witness (int32_t *sources, int32_t *witnesses, return nsources; } + +int +afr_mark_fool_as_source_by_idx (int32_t *sources, int child_count, int idx) +{ + if (idx >= 0 && idx < child_count) { + sources[idx] = 1; + return 1; + } + return 0; +} + + +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; + + for (i = 0; i < child_count; i++) { + 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, - int child_count) + int32_t *success_children, + int child_count, struct iatt *bufs) { int32_t biggest_witness = 0; int nsources = 0; @@ -467,6 +639,11 @@ afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix, 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) { @@ -479,12 +656,17 @@ afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix, biggest_witness = afr_find_biggest_witness_among_fools (witnesses, characters, child_count); - nsources = afr_mark_fool_as_source_by_witness (sources, witnesses, - characters, child_count, - biggest_witness); + 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: - if (witnesses) - GF_FREE (witnesses); + GF_FREE (witnesses); return nsources; } @@ -550,6 +732,78 @@ 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) { @@ -715,11 +969,24 @@ afr_mark_sources (xlator_t *this, int32_t *sources, int32_t **pending_matrix, afr_find_character_types (characters, pending_matrix, success_children, child_count); if (afr_sh_all_nodes_innocent (characters, child_count)) { - if (type == AFR_SELF_HEAL_METADATA) + 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; } @@ -741,7 +1008,8 @@ afr_mark_sources (xlator_t *this, int32_t *sources, int32_t **pending_matrix, nsources = afr_mark_biggest_of_fools_as_source (sources, pending_matrix, characters, - child_count); + success_children, + child_count, bufs); } out: @@ -759,50 +1027,108 @@ 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); - for (i = 0; i < priv->child_count; i++) - for (j = 0; j < priv->child_count; j++) - delta_matrix[i][j] = -delta_matrix[i][j]; + + /* + * 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; + } + } + } } int -afr_sh_delta_to_xattr (afr_private_t *priv, +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; + 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) + 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, - 3 * sizeof (int32_t)); - if (ret < 0) - gf_log (THIS->name, GF_LOG_WARNING, + 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; @@ -810,98 +1136,6 @@ afr_sh_delta_to_xattr (afr_private_t *priv, int -afr_sh_has_metadata_pending (dict_t *xattr, xlator_t *this) -{ - /* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */ - int32_t pending[3] = {0,}; - void *pending_raw = NULL; - afr_private_t *priv = NULL; - int ret = -1; - int i = 0; - int j = 0; - - 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) - return 0; - - memcpy (pending, pending_raw, sizeof(pending)); - j = afr_index_for_transaction_type (AFR_METADATA_TRANSACTION); - - if (pending[j]) - return 1; - } - - return 0; -} - - -int -afr_sh_has_data_pending (dict_t *xattr, xlator_t *this) -{ - /* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */ - int32_t pending[3] = {0,}; - void *pending_raw = NULL; - afr_private_t *priv = NULL; - int ret = -1; - int i = 0; - int j = 0; - - 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) - return 0; - - memcpy (pending, pending_raw, sizeof(pending)); - j = afr_index_for_transaction_type (AFR_DATA_TRANSACTION); - - if (pending[j]) - return 1; - } - - return 0; -} - - -int -afr_sh_has_entry_pending (dict_t *xattr, xlator_t *this) -{ - /* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */ - int32_t pending[3] = {0,}; - void *pending_raw = NULL; - afr_private_t *priv = NULL; - int ret = -1; - int i = 0; - int j = 0; - - 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) - return 0; - - memcpy (pending, pending_raw, sizeof(pending)); - j = afr_index_for_transaction_type (AFR_ENTRY_TRANSACTION); - - if (pending[j]) - return 1; - } - - return 0; -} - -int afr_sh_missing_entries_done (call_frame_t *frame, xlator_t *this) { afr_local_t *local = NULL; @@ -912,14 +1146,13 @@ afr_sh_missing_entries_done (call_frame_t *frame, xlator_t *this) afr_sh_reset (frame, this); - if (local->govinda_gOvinda) { - gf_log (this->name, GF_LOG_INFO, + if (local->unhealable) { + gf_log (this->name, GF_LOG_DEBUG, "split brain found, aborting selfheal of %s", local->loc.path); - sh->op_failed = 1; } - if (sh->op_failed) { + if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) { sh->completion_cbk (frame, this); } else { gf_log (this->name, GF_LOG_TRACE, @@ -1005,7 +1238,7 @@ afr_sh_common_lookup_resp_handler (call_frame_t *frame, void *cookie, sh->success_count++; sh->xattr[child_index] = dict_ref (xattr); } else { - gf_log (this->name, GF_LOG_ERROR, "path %s on subvolume" + gf_log (this->name, GF_LOG_DEBUG, "path %s on subvolume" " %s => -1 (%s)", loc->path, priv->children[child_index]->name, strerror (op_errno)); @@ -1151,7 +1384,7 @@ out: if (ret) { gf_log (this->name, GF_LOG_ERROR, "impunge of %s failed, " "reason: %s", local->loc.path, strerror (-ret)); - sh->op_failed = 1; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED); } afr_sh_missing_entries_finish (frame, this); } @@ -1166,7 +1399,7 @@ afr_sh_create_entry_cbk (call_frame_t *frame, xlator_t *this, local = frame->local; sh = &local->self_heal; if (op_ret < 0) - sh->op_failed = 1; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED); afr_sh_missing_entries_finish (frame, this); return 0; } @@ -1190,7 +1423,7 @@ sh_missing_entries_create (call_frame_t *frame, xlator_t *this) if (!afr_valid_ia_type (type)) { gf_log (this->name, GF_LOG_ERROR, "%s: unknown file type: 0%o", local->loc.path, type); - local->govinda_gOvinda = 1; + afr_set_local_for_unhealable (local); afr_sh_missing_entries_finish (frame, this); goto out; } @@ -1212,6 +1445,10 @@ afr_sh_missing_entries_lookup_done (call_frame_t *frame, xlator_t *this, 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; @@ -1219,25 +1456,47 @@ afr_sh_missing_entries_lookup_done (call_frame_t *frame, xlator_t *this, loc = &local->loc; if (op_ret < 0) { - if (op_errno == EIO) - local->govinda_gOvinda = 1; + if (op_errno == EIO) { + afr_set_local_for_unhealable (local); + } // EIO can happen if finding the fresh parent dir failed goto out; } //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, - afr_transaction_type_get (ia_type), - NULL, _gf_false); + 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); - op_errno = EIO; - goto out; + 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; + } } afr_get_fresh_children (sh->success_children, sh->sources, @@ -1254,10 +1513,14 @@ afr_sh_missing_entries_lookup_done (call_frame_t *frame, xlator_t *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); - sh_missing_entries_create (frame, this); + if (split_brain) { + afr_sh_missing_entries_finish (frame, this); + } else { + sh_missing_entries_create (frame, this); + } return; out: - sh->op_failed = 1; + 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; @@ -1341,7 +1604,7 @@ afr_sh_remove_entry_cbk (call_frame_t *frame, xlator_t *this, int child, LOCK (&frame->lock); { afr_sh_set_error (sh, EIO); - sh->op_failed = 1; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED); } UNLOCK (&frame->lock); } @@ -1363,6 +1626,7 @@ afr_sh_call_entry_expunge_remove (call_frame_t *frame, xlator_t *this, 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) { @@ -1377,6 +1641,12 @@ afr_sh_call_entry_expunge_remove (call_frame_t *frame, xlator_t *this, 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); @@ -1416,7 +1686,7 @@ afr_sh_purge_stale_entries_done (call_frame_t *frame, xlator_t *this) sh = &local->self_heal; priv = this->private; - if (sh->op_failed) { + 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, @@ -1509,7 +1779,7 @@ afr_sh_purge_entry_common (call_frame_t *frame, xlator_t *this, if (!purge_condition (local, priv, i)) continue; gf_log (this->name, GF_LOG_INFO, "purging the stale entry %s " - "on %d", local->loc.path, i); + "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], @@ -1629,10 +1899,8 @@ afr_sh_children_lookup_done (call_frame_t *frame, xlator_t *this, sh->child_errno, priv->child_count, ENOENT); if (fresh_child_enoents == fresh_parent_count) { - gf_log (this->name, GF_LOG_INFO, "Deleting stale file %s", - local->loc.path); afr_sh_set_error (sh, ENOENT); - sh->op_failed = 1; + 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, @@ -1646,14 +1914,14 @@ afr_sh_children_lookup_done (call_frame_t *frame, xlator_t *this, afr_sh_purge_stale_entry (frame, this); } else { op_errno = EIO; - local->govinda_gOvinda = 1; + afr_set_local_for_unhealable (local); goto fail; } return; fail: - sh->op_failed = 1; + 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; @@ -1704,6 +1972,7 @@ afr_sh_find_fresh_parents (call_frame_t *frame, xlator_t *this, 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; } @@ -1711,6 +1980,7 @@ afr_sh_find_fresh_parents (call_frame_t *frame, xlator_t *this, 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, @@ -1721,9 +1991,9 @@ afr_sh_find_fresh_parents (call_frame_t *frame, xlator_t *this, return; out: - afr_sh_set_error (sh, EIO); - sh->op_failed = 1; - afr_sh_missing_entries_finish (frame, this); + 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; } @@ -1812,7 +2082,8 @@ afr_sh_common_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, int -afr_sh_post_nb_entrylk_conflicting_sh_cbk (call_frame_t *frame, xlator_t *this) +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; @@ -1825,7 +2096,7 @@ afr_sh_post_nb_entrylk_conflicting_sh_cbk (call_frame_t *frame, xlator_t *this) if (int_lock->lock_op_ret < 0) { gf_log (this->name, GF_LOG_INFO, "Non blocking entrylks failed."); - sh->op_failed = -1; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED); afr_sh_missing_entries_done (frame, this); } else { @@ -1841,40 +2112,14 @@ afr_sh_post_nb_entrylk_conflicting_sh_cbk (call_frame_t *frame, xlator_t *this) } int -afr_sh_post_nb_entrylk_gfid_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; - sh = &local->self_heal; - int_lock = &local->internal_lock; - - if (int_lock->lock_op_ret < 0) { - gf_log (this->name, GF_LOG_INFO, - "Non blocking entrylks 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, &local->loc, - afr_sh_missing_entries_lookup_done, - sh->sh_gfid_req, AFR_LOOKUP_FAIL_CONFLICTS| - AFR_LOOKUP_FAIL_MISSING_GFIDS, - 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; @@ -1886,7 +2131,12 @@ afr_sh_entrylk (call_frame_t *frame, xlator_t *this, loc_t *loc, 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; @@ -1924,27 +2174,31 @@ out: } static int -afr_self_heal_conflicting_entries (call_frame_t *frame, xlator_t *this) +afr_self_heal_missing_entries (call_frame_t *frame, xlator_t *this) { - afr_self_heal_parent_entrylk (frame, this, - afr_sh_post_nb_entrylk_conflicting_sh_cbk); - return 0; -} + afr_local_t *local = NULL; + afr_self_heal_t *sh = NULL; + + local = frame->local; + sh = &local->self_heal; + + sh->sh_type_in_action = AFR_SELF_HEAL_GFID_OR_MISSING_ENTRY; + + afr_set_self_heal_status (sh, AFR_SELF_HEAL_STARTED); -static int -afr_self_heal_gfids (call_frame_t *frame, xlator_t *this) -{ afr_self_heal_parent_entrylk (frame, this, - afr_sh_post_nb_entrylk_gfid_sh_cbk); + afr_sh_post_nb_entrylk_missing_entry_sh_cbk); return 0; } -afr_local_t *afr_local_copy (afr_local_t *l, xlator_t *this) +afr_local_t* +afr_self_heal_local_init (afr_local_t *l, xlator_t *this) { - afr_private_t *priv = NULL; - afr_local_t *lc = NULL; - afr_self_heal_t *sh = NULL; - afr_self_heal_t *shc = NULL; + 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; priv = this->private; @@ -1963,16 +2217,27 @@ afr_local_t *afr_local_copy (afr_local_t *l, xlator_t *this) 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) - loc_copy (&lc->loc, &l->loc); + if (l->loc.path) { + ret = loc_copy (&lc->loc, &l->loc); + if (ret < 0) + goto out; + } lc->child_up = memdup (l->child_up, sizeof (*lc->child_up) * priv->child_count); + if (!lc->child_up) { + ret = -1; + goto out; + } + if (l->xattr_req) lc->xattr_req = dict_ref (l->xattr_req); @@ -1980,40 +2245,25 @@ afr_local_t *afr_local_copy (afr_local_t *l, xlator_t *this) 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); - if (l->internal_lock.inode_locked_nodes) - lc->internal_lock.inode_locked_nodes = - memdup (l->internal_lock.inode_locked_nodes, - sizeof (*lc->internal_lock.inode_locked_nodes) * priv->child_count); - else - lc->internal_lock.inode_locked_nodes = - GF_CALLOC (sizeof (*l->internal_lock.inode_locked_nodes), - priv->child_count, - gf_afr_mt_char); - if (l->internal_lock.entry_locked_nodes) - lc->internal_lock.entry_locked_nodes = - memdup (l->internal_lock.entry_locked_nodes, - sizeof (*lc->internal_lock.entry_locked_nodes) * priv->child_count); - else - lc->internal_lock.entry_locked_nodes = - GF_CALLOC (sizeof (*l->internal_lock.entry_locked_nodes), - priv->child_count, - gf_afr_mt_char); - if (l->internal_lock.locked_nodes) - lc->internal_lock.locked_nodes = - memdup (l->internal_lock.locked_nodes, - sizeof (*lc->internal_lock.locked_nodes) * priv->child_count); - else - lc->internal_lock.locked_nodes = - GF_CALLOC (sizeof (*l->internal_lock.locked_nodes), - priv->child_count, - gf_afr_mt_char); - lc->internal_lock.inodelk_lock_count = - l->internal_lock.inodelk_lock_count; - lc->internal_lock.entrylk_lock_count = - l->internal_lock.entrylk_lock_count; + 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; } @@ -2023,33 +2273,39 @@ 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_boolean_t split_brain = _gf_false; + gf_loglevel_t loglevel = 0; priv = this->private; local = bgsh_frame->local; sh = &local->self_heal; - if (local->govinda_gOvinda) - split_brain = _gf_true; - - afr_set_split_brain (this, sh->inode, split_brain); + if (local->unhealable) { + afr_set_split_brain (this, sh->inode, SPB, SPB); + } afr_self_heal_type_str_get (sh, sh_type_str, sizeof(sh_type_str)); - if (sh->op_failed) { - gf_log (this->name, GF_LOG_ERROR, "background %s self-heal " - "failed on %s", sh_type_str, local->loc.path); + 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 { - gf_log (this->name, GF_LOG_INFO, "background %s self-heal " - "completed on %s", sh_type_str, local->loc.path); + loglevel = GF_LOG_DEBUG; } + afr_log_self_heal_completion_status (local, loglevel); + FRAME_SU_UNDO (bgsh_frame, afr_local_t); 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, - sh->op_failed); + is_self_heal_failed (sh, AFR_CHECK_ALL)); } if (sh->background) { @@ -2098,14 +2354,13 @@ afr_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode) afr_set_lk_owner (sh_frame, this, sh_frame->root); afr_set_low_priority (sh_frame); - sh_local = afr_local_copy (local, this); + sh_local = afr_self_heal_local_init (local, this); if (!sh_local) goto out; sh_frame->local = sh_local; sh = &sh_local->self_heal; sh->inode = inode_ref (inode); - sh->orig_frame = frame; sh->completion_cbk = afr_self_heal_completion_cbk; @@ -2164,12 +2419,11 @@ afr_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode) 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) { - afr_self_heal_conflicting_entries (sh_frame, this); - } else if (sh->do_gfid_self_heal) { - GF_ASSERT (!uuid_is_null (sh->sh_gfid_req)); - afr_self_heal_gfids (sh_frame, this); + 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)) { @@ -2294,3 +2548,265 @@ out: 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 +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; + } + + 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; + + } + +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 339eebe15..473264776 100644 --- a/xlators/cluster/afr/src/afr-self-heal-common.h +++ b/xlators/cluster/afr/src/afr-self-heal-common.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2008-2011 Gluster, Inc. <http://www.gluster.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__ @@ -24,13 +15,6 @@ #define AFR_SH_MIN_PARTICIPANTS 2 typedef enum { - AFR_SELF_HEAL_ENTRY, - AFR_SELF_HEAL_METADATA, - AFR_SELF_HEAL_DATA, - AFR_SELF_HEAL_INVALID = -1, -} afr_self_heal_type; - -typedef enum { AFR_LOOKUP_FAIL_CONFLICTS = 1, AFR_LOOKUP_FAIL_MISSING_GFIDS = 2, } afr_lookup_flags_t; @@ -44,6 +28,10 @@ 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_build_pending_matrix (char **pending_key, int32_t **pending_matrix, unsigned char *ignorant_subvols, @@ -61,7 +49,7 @@ afr_mark_sources (xlator_t *this, int32_t *sources, int32_t **pending_matrix, int32_t *success_children, int32_t *subvol_status); int -afr_sh_delta_to_xattr (afr_private_t *priv, +afr_sh_delta_to_xattr (xlator_t *this, int32_t *delta_matrix[], dict_t *xattr[], int child_count, afr_transaction_type type); @@ -103,13 +91,13 @@ 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, +afr_sh_data_unlock (call_frame_t *frame, xlator_t *this, char *dom, afr_lock_cbk_t lock_cbk); afr_local_t * -afr_local_copy (afr_local_t *l, xlator_t *this); +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, + 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 @@ -131,4 +119,26 @@ 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_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 +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 69494157c..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) 2008-2011 Gluster, Inc. <http://www.gluster.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> @@ -71,15 +62,6 @@ int afr_sh_data_finish (call_frame_t *frame, xlator_t *this); int -afr_sh_data_fxattrop (call_frame_t *frame, xlator_t *this, - afr_fxattrop_cbk_t fxattrop_cbk); - -int -afr_post_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 afr_sh_data_done (call_frame_t *frame, xlator_t *this) { afr_local_t *local = NULL; @@ -109,7 +91,7 @@ afr_sh_data_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this, LOCK (&frame->lock); { if (op_ret == -1) { - gf_log (this->name, GF_LOG_INFO, + 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)); @@ -139,6 +121,11 @@ afr_sh_data_close (call_frame_t *frame, xlator_t *this) 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; @@ -169,6 +156,25 @@ afr_sh_data_close (call_frame_t *frame, xlator_t *this) } 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; +} + +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) @@ -203,29 +209,20 @@ afr_sh_data_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, } int -afr_sh_data_setattr (call_frame_t *frame, xlator_t *this) +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; - int source = 0; int32_t valid = 0; - struct iatt stbuf = {0,}; local = frame->local; sh = &local->self_heal; priv = this->private; - source = sh->source; - - valid |= (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME); - - 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; + valid = (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME); call_count = afr_set_elem_count_get (sh->success, priv->child_count); @@ -245,7 +242,7 @@ afr_sh_data_setattr (call_frame_t *frame, xlator_t *this) (void *) (long) i, priv->children[i], priv->children[i]->fops->setattr, - &local->loc, &stbuf, valid, NULL); + &local->loc, stbuf, valid, NULL); if (!--call_count) break; @@ -269,7 +266,7 @@ afr_sh_data_setattr_fstat_cbk (call_frame_t *frame, void *cookie, GF_ASSERT (sh->source == child_index); if (op_ret != -1) { sh->buf[child_index] = *buf; - afr_sh_data_setattr (frame, this); + 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); @@ -305,31 +302,45 @@ afr_sh_set_timestamps (call_frame_t *frame, xlator_t *this) //Fun fact, lock_cbk is being used for both lock & unlock int -afr_sh_data_unlock (call_frame_t *frame, xlator_t *this, +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; - GF_ASSERT (sh->data_lock_held); - - sh->data_lock_held = _gf_false; + 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); +out: + if (ret) { + int_lock->lock_op_ret = -1; + int_lock->lock_cbk (frame, this); + } return 0; } int afr_sh_data_finish (call_frame_t *frame, xlator_t *this) { - afr_local_t *local = NULL; - afr_self_heal_t *sh = NULL; + afr_local_t *local = NULL; + afr_self_heal_t *sh = NULL; local = frame->local; sh = &local->self_heal; @@ -338,9 +349,9 @@ afr_sh_data_finish (call_frame_t *frame, xlator_t *this) "finishing data selfheal of %s", local->loc.path); if (sh->data_lock_held) - afr_sh_data_unlock (frame, this, afr_sh_data_close); + afr_sh_data_unlock (frame, this, this->name, afr_sh_dom_unlock); else - afr_sh_data_close (frame, this); + afr_sh_dom_unlock (frame, this); return 0; } @@ -357,11 +368,8 @@ afr_sh_data_fail (call_frame_t *frame, xlator_t *this) gf_log (this->name, GF_LOG_DEBUG, "finishing failed data selfheal of %s", local->loc.path); - sh->op_failed = 1; - if (sh->data_lock_held) - afr_sh_data_unlock (frame, this, afr_sh_data_close); - else - afr_sh_data_close (frame, this); + afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED); + afr_sh_data_finish (frame, this); return 0; } @@ -384,13 +392,16 @@ afr_sh_data_erase_pending_cbk (call_frame_t *frame, void *cookie, "log failed on %s for subvol %s, reason: %s", local->loc.path, priv->children[child_index]->name, strerror (op_errno)); - sh->op_failed = 1; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED); } call_count = afr_frame_return (frame); if (call_count == 0) { - if (sh->op_failed) { + 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; } @@ -399,7 +410,7 @@ afr_sh_data_erase_pending_cbk (call_frame_t *frame, void *cookie, goto out; } GF_ASSERT (sh->old_loop_frame); - afr_sh_data_lock (frame, this, 0, 0, + afr_sh_data_lock (frame, this, 0, 0, _gf_true, this->name, afr_post_sh_big_lock_success, afr_post_sh_big_lock_failure); } @@ -410,75 +421,95 @@ out: int afr_sh_data_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; + 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; - sh = &local->self_heal; priv = this->private; + sh = &local->self_heal; - afr_sh_pending_to_delta (priv, sh->xattr, sh->delta_matrix, sh->success, - priv->child_count, AFR_DATA_TRANSACTION); - 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); - - erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count, - gf_afr_mt_dict_t); - - 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]); + 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; } - afr_sh_delta_to_xattr (priv, sh->delta_matrix, erase_xattr, - priv->child_count, AFR_DATA_TRANSACTION); + 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; +} - GF_ASSERT (call_count); - local->call_count = call_count; - for (i = 0; i < priv->child_count; i++) { - if (!erase_xattr[i]) - continue; +/* + * Before erasing xattrs, make sure the data is written to disk + */ +int +afr_sh_data_fsync (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; - gf_log (this->name, GF_LOG_DEBUG, - "erasing pending flags from %s on %s", - local->loc.path, priv->children[i]->name); + local = frame->local; + priv = this->private; + sh = &local->self_heal; - STACK_WIND_COOKIE (frame, afr_sh_data_erase_pending_cbk, - (void *) (long) i, - priv->children[i], - priv->children[i]->fops->fxattrop, - sh->healing_fd, - GF_XATTROP_ADD_ARRAY, erase_xattr[i], - NULL); - if (!--call_count) - break; + 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 (erase_xattr[i]) { - dict_unref (erase_xattr[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); } - GF_FREE (erase_xattr); return 0; } - static struct afr_sh_algorithm * sh_algo_from_name (xlator_t *this, char *name) { int i = 0; + if (name == NULL) + goto out; + while (afr_self_heal_algorithms[i].name) { if (!strcmp (name, afr_self_heal_algorithms[i].name)) { return &afr_self_heal_algorithms[i]; @@ -487,6 +518,7 @@ sh_algo_from_name (xlator_t *this, char *name) i++; } +out: return NULL; } @@ -566,7 +598,7 @@ afr_sh_data_sync_prepare (call_frame_t *frame, xlator_t *this) local = frame->local; sh = &local->self_heal; - sh->algo_completion_cbk = afr_sh_data_erase_pending; + sh->algo_completion_cbk = afr_sh_data_fsync; sh->algo_abort_cbk = afr_sh_data_fail; sh_algo = afr_sh_data_pick_algo (frame, this); @@ -602,7 +634,7 @@ afr_sh_data_trim_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->loc.path, priv->children[child_index]->name, strerror (op_errno)); - sh->op_failed = 1; + 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", @@ -615,7 +647,7 @@ afr_sh_data_trim_cbk (call_frame_t *frame, void *cookie, xlator_t *this, call_count = afr_frame_return (frame); if (call_count == 0) { - if (sh->op_failed) + if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) afr_sh_data_fail (frame, this); else afr_sh_data_sync_prepare (frame, this); @@ -668,6 +700,7 @@ 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); @@ -676,6 +709,15 @@ afr_sh_inode_set_read_ctx (afr_self_heal_t *sh, xlator_t *this) 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); @@ -685,16 +727,254 @@ out: return ret; } -int +char* +afr_get_sizes_str (afr_local_t *local, struct iatt *bufs, xlator_t *this) +{ + 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; +} + +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; + + priv = this->private; + + 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 ++; + + sinks_str = GF_CALLOC (len, sizeof (char), gf_common_mt_char); + + if (!sinks_str) + return NULL; + + off += snprintf (sinks_str + off, len - off, "%s", temp_str); + + 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); + } + } + + return sinks_str; + +} + + +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; + + priv = this->private; + + pending_matrix_str = afr_get_pending_matrix_str (sh->pending_matrix, + this); + if (!pending_matrix_str) + pending_matrix_str = ""; + + sizes_str = afr_get_sizes_str (local, sh->buf, this); + if (!sizes_str) + sizes_str = ""; + + sinks_str = afr_get_sinks_str (this, local, sh); + if (!sinks_str) + sinks_str = ""; + + 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_fxattrop_fstat_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 nsources = 0; - int source = 0; - int i = 0; int ret = 0; + int *old_sources = NULL; + int tstamp_source = 0; + int i = 0; local = frame->local; sh = &local->self_heal; @@ -702,19 +982,17 @@ afr_sh_data_fix (call_frame_t *frame, xlator_t *this) 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_false); - 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; - } - + AFR_DATA_TRANSACTION, NULL, _gf_true); if ((nsources == -1) && (priv->favorite_child != -1) && (sh->child_errno[priv->favorite_child] == 0)) { @@ -732,17 +1010,16 @@ afr_sh_data_fix (call_frame_t *frame, xlator_t *this) } if (nsources == -1) { - 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.", local->loc.path); - - local->govinda_gOvinda = 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, @@ -752,50 +1029,42 @@ afr_sh_data_fix (call_frame_t *frame, xlator_t *this) return 0; } - 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; - - /* 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; - } + 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); - if (sh->background && sh->unwind) { - sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno, - sh->op_failed); - sh->unwound = _gf_true; - } + afr_sh_data_finish (frame, this); + return 0; + } - 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 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); } - - 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); - afr_sh_data_trim_sinks (frame, this); - return 0; } int afr_lookup_select_read_child_by_txn_type (xlator_t *this, afr_local_t *local, dict_t **xattr, - afr_transaction_type txn_type) + afr_transaction_type txn_type, + uuid_t gfid) { afr_private_t *priv = NULL; int read_child = -1; @@ -812,20 +1081,33 @@ afr_lookup_select_read_child_by_txn_type (xlator_t *this, afr_local_t *local, bufs = local->cont.lookup.bufs; success_children = local->cont.lookup.success_children; - pending_matrix = afr_matrix_create (priv->child_count, - priv->child_count); - if (NULL == pending_matrix) - goto out; - + 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_WARNING, "%s: Possible split-brain", + 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; @@ -835,36 +1117,15 @@ afr_lookup_select_read_child_by_txn_type (xlator_t *this, afr_local_t *local, priv->child_count, prev_read_child, config_read_child, - sources); + sources, + priv->hash_mode, gfid); out: - afr_matrix_cleanup (pending_matrix, priv->child_count); gf_log (this->name, GF_LOG_DEBUG, "returning read_child: %d", read_child); return read_child; } int -afr_sh_data_special_file_fix (call_frame_t *frame, xlator_t *this) -{ - afr_private_t *priv = NULL; - afr_self_heal_t *sh = NULL; - afr_local_t *local = NULL; - int i = 0; - - 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_data_erase_pending (frame, this); - - return 0; -} - -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) @@ -914,10 +1175,7 @@ afr_sh_data_fstat_cbk (call_frame_t *frame, void *cookie, afr_sh_data_fail (frame, this); goto out; } - if (IA_ISREG (sh->type)) - afr_sh_data_fix (frame, this); - else - afr_sh_data_special_file_fix (frame, this); + afr_sh_data_fxattrop_fstat_done (frame, this); } out: return 0; @@ -948,6 +1206,7 @@ afr_sh_data_fstat (call_frame_t *frame, xlator_t *this) 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++) { @@ -963,8 +1222,7 @@ afr_sh_data_fstat (call_frame_t *frame, xlator_t *this) } GF_ASSERT (!call_count); out: - if (fstat_children) - GF_FREE (fstat_children); + GF_FREE (fstat_children); return 0; } @@ -1005,45 +1263,6 @@ afr_sh_common_fxattrop_resp_handler (call_frame_t *frame, void *cookie, } int -afr_post_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; - int ret = 0; - afr_local_t *local = NULL; - afr_self_heal_t *sh = NULL; - - afr_sh_common_fxattrop_resp_handler (frame, cookie, this, op_ret, - op_errno, xattr); - - local = frame->local; - sh = &local->self_heal; - call_count = afr_frame_return (frame); - if (call_count) - goto out; - - 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; - } - (void) afr_build_sources (this, sh->xattr, NULL, - sh->pending_matrix, - sh->sources, sh->success_children, - AFR_DATA_TRANSACTION, NULL, _gf_false); - ret = afr_sh_inode_set_read_ctx (sh, this); - if (ret) - afr_sh_data_fail (frame, this); - else - afr_sh_set_timestamps (frame, this); -out: - return 0; -} - -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) @@ -1075,17 +1294,17 @@ out: int -afr_sh_data_fxattrop (call_frame_t *frame, xlator_t *this, - afr_fxattrop_cbk_t fxattrop_cbk) +afr_sh_data_fxattrop (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; + 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; @@ -1096,30 +1315,39 @@ afr_sh_data_fxattrop (call_frame_t *frame, xlator_t *this, local->call_count = call_count; - xattr_req = dict_new(); - if (!xattr_req) { - ret = -1; - goto out; - } - - for (i = 0; i < priv->child_count; i++) { - 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, priv->pending_key[i], - 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; - } - } + 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; + } + } + + 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; + } + } + } afr_reset_xattr (sh->xattr, priv->child_count); afr_reset_children (sh->success_children, priv->child_count); @@ -1128,12 +1356,12 @@ afr_sh_data_fxattrop (call_frame_t *frame, xlator_t *this, sh->success_count = 0; for (i = 0; i < priv->child_count; i++) { if (local->child_up[i]) { - STACK_WIND_COOKIE (frame, fxattrop_cbk, + 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, NULL); + xattr_req[i], NULL); if (!--call_count) break; @@ -1141,12 +1369,15 @@ afr_sh_data_fxattrop (call_frame_t *frame, xlator_t *this, } out: - if (xattr_req) - dict_unref (xattr_req); + if (xattr_req) { + for (i = 0; i < priv->child_count; i++) + if (xattr_req[i]) + dict_unref(xattr_req[i]); + GF_FREE(xattr_req); + } if (ret) { - if (zero_pending) - GF_FREE (zero_pending); + GF_FREE (zero_pending); afr_sh_data_fail (frame, this); } @@ -1163,7 +1394,23 @@ afr_sh_data_big_lock_success (call_frame_t *frame, xlator_t *this) sh = &local->self_heal; sh->data_lock_held = _gf_true; - afr_sh_data_fxattrop (frame, this, afr_sh_data_fxattrop_cbk); + afr_sh_data_fxattrop (frame, this); + return 0; +} + +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; + + 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; } @@ -1212,8 +1459,13 @@ afr_sh_data_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this) "failed for %s. by %s", local->loc.path, lkowner_utoa (&frame->root->lk_owner)); - int_lock->lock_cbk = afr_sh_data_post_blocking_inodelk_cbk; - afr_blocking_lock (frame, this); + 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); + } } else { gf_log (this->name, GF_LOG_DEBUG, "Non Blocking data inodelks " @@ -1226,9 +1478,11 @@ afr_sh_data_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this) } int -afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this, off_t start, off_t len) +afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this, char *dom, + off_t start, off_t len) { afr_internal_lock_t *int_lock = NULL; + afr_inodelk_t *inodelk = NULL; afr_local_t *local = NULL; local = frame->local; @@ -1239,11 +1493,14 @@ afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this, off_t start, off_t le afr_set_lock_number (frame, this); - int_lock->lk_flock.l_start = start; - int_lock->lk_flock.l_len = len; - int_lock->lk_flock.l_type = F_WRLCK; int_lock->lock_cbk = afr_sh_data_post_nonblocking_inodelk_cbk; + 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; + afr_nonblocking_inodelk (frame, this); return 0; @@ -1262,7 +1519,8 @@ afr_post_sh_big_lock_success (call_frame_t *frame, xlator_t *this) sh_loop_finish (sh->old_loop_frame, this); sh->old_loop_frame = NULL; sh->data_lock_held = _gf_true; - afr_sh_data_fxattrop (frame, this, afr_post_sh_data_fxattrop_cbk); + sh->sync_done = _gf_true; + afr_sh_data_fxattrop (frame, this); return 0; } @@ -1285,8 +1543,8 @@ afr_post_sh_big_lock_failure (call_frame_t *frame, xlator_t *this) int afr_sh_data_lock (call_frame_t *frame, xlator_t *this, - off_t start, off_t len, - afr_lock_cbk_t success_handler, + 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; @@ -1297,7 +1555,8 @@ afr_sh_data_lock (call_frame_t *frame, xlator_t *this, sh->data_lock_success_handler = success_handler; sh->data_lock_failure_handler = failure_handler; - return afr_sh_data_lock_rec (frame, this, start, len); + sh->data_lock_block = block; + return afr_sh_data_lock_rec (frame, this, dom, start, len); } int @@ -1328,7 +1587,7 @@ afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->loc.path, priv->children[child_index]->name, strerror (op_errno)); - sh->op_failed = 1; + 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", @@ -1341,7 +1600,7 @@ afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this, call_count = afr_frame_return (frame); if (call_count == 0) { - if (sh->op_failed) { + if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) { afr_sh_data_fail (frame, this); return 0; } @@ -1350,9 +1609,8 @@ afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this, "fd for %s opened, commencing sync", local->loc.path); - afr_sh_data_lock (frame, this, 0, 0, - afr_sh_data_big_lock_success, - afr_sh_data_fail); + afr_sh_data_lock (frame, this, 0, 0, _gf_true, priv->sh_domain, + afr_sh_dom_lock_success, afr_sh_data_fail); } return 0; @@ -1398,20 +1656,93 @@ afr_sh_data_open (call_frame_t *frame, xlator_t *this) return 0; } +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); +} int -afr_self_heal_data (call_frame_t *frame, xlator_t *this) +afr_sh_non_reg_lock_success (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; + 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)) { - afr_sh_data_open (frame, this); + 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; + 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", diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c index b57521b9f..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) 2008-2011 Gluster, Inc. <http://www.gluster.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> @@ -166,73 +157,20 @@ 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; - int need_unwind = 0; local = frame->local; sh = &local->self_heal; - priv = this->private; if (sh->entries_skipped) { - need_unwind = 1; - sh->op_failed = _gf_true; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED); goto out; } - afr_sh_pending_to_delta (priv, sh->xattr, sh->delta_matrix, sh->success, - priv->child_count, AFR_ENTRY_TRANSACTION); - - erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count, - gf_afr_mt_dict_t); - - 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]); - } - } - - if (call_count == 0) - need_unwind = 1; - - afr_sh_delta_to_xattr (priv, sh->delta_matrix, erase_xattr, - priv->child_count, AFR_ENTRY_TRANSACTION); - - local->call_count = call_count; - for (i = 0; i < priv->child_count; i++) { - if (!erase_xattr[i]) - continue; - - gf_log (this->name, GF_LOG_TRACE, - "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], - NULL); - if (!--call_count) - break; - } - - for (i = 0; i < priv->child_count; i++) { - if (erase_xattr[i]) { - dict_unref (erase_xattr[i]); - } - } - GF_FREE (erase_xattr); - + afr_sh_erase_pending (frame, this, AFR_ENTRY_TRANSACTION, + afr_sh_entry_erase_pending_cbk, + afr_sh_entry_finish); + return 0; out: - if (need_unwind) - afr_sh_entry_finish (frame, this); - + afr_sh_entry_finish (frame, this); return 0; } @@ -642,7 +580,8 @@ afr_sh_entry_expunge_entry_cbk (call_frame_t *expunge_frame, void *cookie, if (need_expunge) { gf_log (this->name, GF_LOG_INFO, - "missing entry %s on %s", + "Entry %s is missing on %s and deleting from " + "replica's other bricks", expunge_local->loc.path, priv->children[source]->name); @@ -674,6 +613,19 @@ out: 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, @@ -701,15 +653,7 @@ afr_sh_entry_expunge_entry (call_frame_t *frame, xlator_t *this, sh->expunge_done = afr_sh_entry_expunge_entry_done; name = entry->d_name; - - if ((strcmp (name, ".") == 0) - || (strcmp (name, "..") == 0) - || ((strcmp (local->loc.path, "/") == 0) - && (strcmp (name, GF_REPLICATE_TRASH_DIR) == 0))) { - - gf_log (this->name, GF_LOG_TRACE, - "skipping inspection of %s under %s", - name, local->loc.path); + if (can_skip_entry_self_heal (name, &local->loc)) { op_ret = 0; goto out; } @@ -862,7 +806,7 @@ afr_sh_entry_expunge_all (call_frame_t *frame, xlator_t *this) active_src = next_active_sink (frame, this, sh->active_source); sh->active_source = active_src; - if (sh->op_failed) { + if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) { goto out; } @@ -1081,33 +1025,6 @@ out: return 0; } -void -afr_sh_prepare_new_entry_pending_matrix (int32_t **pending, - int *child_errno, - 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 (child_errno[i]) - continue; - pending[i][midx] = hton32 (1); - if (idx == -1) - continue; - pending[i][idx] = hton32 (1); - } -} - int afr_sh_entry_impunge_perform_xattrop (call_frame_t *impunge_frame, xlator_t *this) @@ -1124,17 +1041,19 @@ afr_sh_entry_impunge_perform_xattrop (call_frame_t *impunge_frame, impunge_sh = &impunge_local->self_heal; active_src = impunge_sh->active_source; - afr_sh_prepare_new_entry_pending_matrix (impunge_local->pending, - impunge_sh->child_errno, - &impunge_sh->entrybuf, - priv->child_count); + 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); + 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, @@ -1205,6 +1124,18 @@ afr_sh_entry_impunge_hardlink_cbk (call_frame_t *impunge_frame, void *cookie, struct iatt *postparent, dict_t *xdata) { 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) @@ -1332,6 +1263,35 @@ afr_sh_entry_impunge_mknod (call_frame_t *impunge_frame, xlator_t *this, 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], @@ -1408,7 +1368,7 @@ afr_sh_entry_impunge_symlink (call_frame_t *impunge_frame, xlator_t *this, priv = this->private; impunge_local = impunge_frame->local; - buf = &impunge_local->cont.symlink.buf; + buf = &impunge_local->cont.dir_fop.buf; dict = dict_new (); if (!dict) { @@ -1668,7 +1628,7 @@ afr_sh_entry_impunge_readlink (call_frame_t *impunge_frame, xlator_t *this, impunge_local = impunge_frame->local; impunge_sh = &impunge_local->self_heal; active_src = impunge_sh->active_source; - impunge_local->cont.symlink.buf = *stbuf; + impunge_local->cont.dir_fop.buf = *stbuf; STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_readlink_cbk, (void *) (long) child_index, @@ -1948,14 +1908,7 @@ afr_sh_entry_impunge_entry (call_frame_t *frame, xlator_t *this, active_src = sh->active_source; sh->impunge_done = afr_sh_entry_impunge_entry_done; - if ((strcmp (entry->d_name, ".") == 0) - || (strcmp (entry->d_name, "..") == 0) - || ((strcmp (local->loc.path, "/") == 0) - && (strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR) == 0))) { - - gf_log (this->name, GF_LOG_TRACE, - "skipping inspection of %s under %s", - entry->d_name, local->loc.path); + if (can_skip_entry_self_heal (entry->d_name, &local->loc)) { op_ret = 0; goto out; } @@ -2024,7 +1977,7 @@ afr_sh_entry_impunge_readdir_cbk (call_frame_t *frame, void *cookie, local->loc.path, priv->children[active_src]->name, strerror (op_errno)); - sh->op_failed = 1; + 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", @@ -2097,7 +2050,7 @@ afr_sh_entry_impunge_all (call_frame_t *frame, xlator_t *this) active_src = next_active_source (frame, this, sh->active_source); sh->active_source = active_src; - if (sh->op_failed) { + if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) { afr_sh_entry_finish (frame, this); return 0; } @@ -2146,7 +2099,7 @@ afr_sh_entry_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->loc.path, priv->children[child_index]->name, strerror (op_errno)); - sh->op_failed = 1; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED); } } UNLOCK (&frame->lock); @@ -2154,7 +2107,7 @@ afr_sh_entry_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this, call_count = afr_frame_return (frame); if (call_count == 0) { - if (sh->op_failed) { + if (is_self_heal_failed (sh, AFR_CHECK_SPECIFIC)) { afr_sh_entry_finish (frame, this); return 0; } @@ -2192,7 +2145,7 @@ afr_sh_entry_open (call_frame_t *frame, xlator_t *this) source = local->self_heal.source; sources = local->self_heal.sources; - sh->block_size = 65536; //131072 + sh->block_size = priv->sh_readdir_size; sh->offset = 0; call_count = sh->active_sinks; @@ -2286,6 +2239,8 @@ afr_sh_entry_sync_prepare (call_frame_t *frame, xlator_t *this) "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; @@ -2308,7 +2263,7 @@ afr_sh_entry_fix (call_frame_t *frame, xlator_t *this, priv = this->private; if (op_ret < 0) { - sh->op_failed = 1; + 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; @@ -2371,7 +2326,7 @@ afr_sh_post_nonblocking_entry_cbk (call_frame_t *frame, xlator_t *this) if (int_lock->lock_op_ret < 0) { gf_log (this->name, GF_LOG_ERROR, "Non Blocking entrylks " "failed for %s.", local->loc.path); - sh->op_failed = 1; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_FAILED); afr_sh_entry_done (frame, this); } else { @@ -2390,14 +2345,18 @@ afr_sh_post_nonblocking_entry_cbk (call_frame_t *frame, xlator_t *this) int afr_self_heal_entry (call_frame_t *frame, xlator_t *this) { - afr_local_t *local = NULL; + 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 { diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c index 9cebd5c07..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) 2008-2011 Gluster, Inc. <http://www.gluster.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> @@ -59,20 +50,12 @@ afr_sh_metadata_done (call_frame_t *frame, xlator_t *this) sh = &local->self_heal; afr_sh_reset (frame, this); - if (local->govinda_gOvinda) { - gf_log (this->name, GF_LOG_INFO, - "split-brain detected, aborting selfheal of %s", + if (IA_ISDIR (sh->type)) { + gf_log (this->name, GF_LOG_DEBUG, + "proceeding to entry check on %s", local->loc.path); - sh->op_failed = 1; - sh->completion_cbk (frame, this); + afr_self_heal_entry (frame, this); } else { - 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); - return 0; - } gf_log (this->name, GF_LOG_DEBUG, "proceeding to data check on %s", local->loc.path); @@ -105,6 +88,19 @@ afr_sh_metadata_finish (call_frame_t *frame, xlator_t *this) return 0; } +int +afr_sh_metadata_fail (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; + + 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, @@ -141,80 +137,13 @@ afr_sh_metadata_erase_pending_cbk (call_frame_t *frame, void *cookie, 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 (priv, sh->xattr, sh->delta_matrix, - sh->success, priv->child_count, - AFR_METADATA_TRANSACTION); - - erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count, - gf_afr_mt_dict_t); - if (!erase_xattr) - return -ENOMEM; - - 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 (priv, sh->delta_matrix, erase_xattr, - priv->child_count, AFR_METADATA_TRANSACTION); - - local->call_count = call_count; - - if (call_count == 0) { - gf_log (this->name, GF_LOG_INFO, - "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_TRACE, - "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], - NULL); - if (!--call_count) - break; - } - - for (i = 0; i < priv->child_count; i++) { - if (erase_xattr[i]) { - dict_unref (erase_xattr[i]); - } - } - GF_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; } @@ -251,8 +180,13 @@ afr_sh_metadata_sync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, call_count = afr_frame_return (frame); - if (call_count == 0) + 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; } @@ -278,6 +212,86 @@ afr_sh_metadata_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, return 0; } +int +afr_sh_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, + dict_t *xdata) +{ + int i = 0; + afr_private_t *priv = NULL; + afr_local_t *local = NULL; + + priv = this->private; + local = frame->local; + + if (op_ret < 0) { + afr_sh_metadata_sync_cbk (frame, cookie, + this, -1, op_errno, xdata); + goto out; + } + + 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); + + out: + return 0; +} + +inline void +afr_prune_special_keys (dict_t *xattr_dict) +{ + dict_del (xattr_dict, GF_SELINUX_XATTR_KEY); +} + +inline void +afr_prune_pending_keys (dict_t *xattr_dict, afr_private_t *priv) +{ + int i = 0; + + for (; i < priv->child_count; i++) { + dict_del (xattr_dict, priv->pending_key[i]); + } +} + +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; + + priv = this->private; + local = frame->local; + + if (op_ret < 0) { + afr_sh_metadata_sync_cbk (frame, cookie, + this, -1, op_errno, xdata); + goto out; + } + + afr_prune_pending_keys (xattr, priv); + + afr_prune_special_keys (xattr); + + i = (long) cookie; + + /* 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); + + out: + return 0; +} int afr_sh_metadata_sync (call_frame_t *frame, xlator_t *this, dict_t *xattr) @@ -303,9 +317,10 @@ afr_sh_metadata_sync (call_frame_t *frame, xlator_t *this, dict_t *xattr) /* * 2 calls per sink - setattr, setxattr */ - if (xattr) + if (xattr) { call_count = active_sinks * 2; - else + local->xattr_req = dict_ref (xattr); + } else call_count = active_sinks; local->call_count = call_count; @@ -348,11 +363,11 @@ afr_sh_metadata_sync (call_frame_t *frame, xlator_t *this, dict_t *xattr) if (!xattr) continue; - STACK_WIND_COOKIE (frame, afr_sh_metadata_xattr_cbk, + STACK_WIND_COOKIE (frame, afr_sh_getxattr_cbk, (void *) (long) i, priv->children[i], - priv->children[i]->fops->setxattr, - &local->loc, xattr, 0, NULL); + priv->children[i]->fops->getxattr, + &local->loc, NULL, NULL); call_count--; } @@ -370,8 +385,6 @@ afr_sh_metadata_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, afr_private_t *priv = NULL; int source = 0; - int i; - local = frame->local; sh = &local->self_heal; priv = this->private; @@ -386,16 +399,147 @@ afr_sh_metadata_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, afr_sh_metadata_sync (frame, this, NULL); } else { - for (i = 0; i < priv->child_count; i++) { - dict_del (xattr, priv->pending_key[i]); - } - + 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_sync_prepare (call_frame_t *frame, xlator_t *this) @@ -425,6 +569,9 @@ afr_sh_metadata_sync_prepare (call_frame_t *frame, xlator_t *this) 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, @@ -450,7 +597,7 @@ afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this, priv = this->private; if (op_ret < 0) { - sh->op_failed = 1; + 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; @@ -459,15 +606,6 @@ afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this, sh->pending_matrix, sh->sources, sh->success_children, AFR_METADATA_TRANSACTION, NULL, _gf_false); - 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; - } - if ((nsources == -1) && (priv->favorite_child != -1) && (sh->child_errno[priv->favorite_child] == 0)) { @@ -484,12 +622,18 @@ afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this, } if (nsources == -1) { - gf_log (this->name, GF_LOG_ERROR, - "Unable to self-heal permissions/ownership of '%s' " - "(possible split-brain). Please fix the file on " - "all backend volumes", local->loc.path); + 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; + } - local->govinda_gOvinda = 1; + 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; @@ -528,7 +672,10 @@ afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this, sh->fresh_children); } - afr_sh_metadata_sync_prepare (frame, this); + 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; } @@ -544,9 +691,9 @@ afr_sh_metadata_post_nonblocking_inodelk_cbk (call_frame_t *frame, int_lock = &local->internal_lock; if (int_lock->lock_op_ret < 0) { - gf_log (this->name, GF_LOG_ERROR, "Non Blocking metadata " + gf_log (this->name, GF_LOG_DEBUG, "Non Blocking metadata " "inodelks failed for %s.", local->loc.path); - gf_log (this->name, GF_LOG_ERROR, "Metadata self-heal " + gf_log (this->name, GF_LOG_DEBUG, "Metadata self-heal " "failed for %s.", local->loc.path); afr_sh_metadata_done (frame, this); } else { @@ -568,19 +715,22 @@ int afr_sh_metadata_lock (call_frame_t *frame, xlator_t *this) { 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); - int_lock->lk_flock.l_start = LLONG_MAX - 1; - int_lock->lk_flock.l_len = 0; - int_lock->lk_flock.l_type = F_WRLCK; + 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; afr_nonblocking_inodelk (frame, this); @@ -588,17 +738,29 @@ afr_sh_metadata_lock (call_frame_t *frame, xlator_t *this) 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 (local->self_heal.do_metadata_self_heal && priv->metadata_self_heal) { + 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); diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h index f40c06faa..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) 2008-2011 Gluster, Inc. <http://www.gluster.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__ @@ -30,13 +21,6 @@ #define SIZE_GREATER(buf1,buf2) ((buf1)->ia_size > (buf2)->ia_size) int -afr_sh_has_metadata_pending (dict_t *xattr, xlator_t *this); -int -afr_sh_has_entry_pending (dict_t *xattr, xlator_t *this); -int -afr_sh_has_data_pending (dict_t *xattr, xlator_t *this); - -int afr_self_heal_entry (call_frame_t *frame, xlator_t *this); int @@ -54,5 +38,6 @@ 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); + 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 index 9fc3c964a..1b48a1bca 100644 --- a/xlators/cluster/afr/src/afr-self-heald.c +++ b/xlators/cluster/afr/src/afr-self-heald.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.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 @@ -28,20 +19,18 @@ #include "protocol-common.h" #include "event-history.h" -#define AFR_POLL_TIMEOUT 600 - typedef enum { STOP_CRAWL_ON_SINGLE_SUBVOL = 1 } afr_crawl_flags_t; typedef enum { HEAL = 1, - INFO + INFO, + STATISTICS_TO_BE_HEALED, } shd_crawl_op; typedef struct shd_dump { dict_t *dict; - time_t sh_time; xlator_t *this; int child; } shd_dump_t; @@ -69,9 +58,59 @@ afr_start_crawl (xlator_t *this, int idx, afr_crawl_type_t crawl, 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) { @@ -79,8 +118,7 @@ shd_cleanup_event (void *event) if (!shd_event) goto out; - if (shd_event->path) - GF_FREE (shd_event->path); + GF_FREE (shd_event->path); GF_FREE (shd_event); out: return; @@ -118,8 +156,125 @@ _build_index_loc (xlator_t *this, loc_t *loc, char *name, loc_t *parent) } int -_add_str_to_dict (xlator_t *this, dict_t *output, int child, char *str, - gf_boolean_t dyn) +_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; @@ -138,15 +293,27 @@ _add_str_to_dict (xlator_t *this, dict_t *output, int child, char *str, snprintf (key, sizeof (key), "%d-%d-%"PRIu64, xl_id, child, count); if (dyn) - ret = dict_set_dynstr (output, key, str); + ret = dict_set_dynstr (output, key, path); else - ret = dict_set_str (output, key, str); + ret = dict_set_str (output, key, path); if (ret) { gf_log (this->name, GF_LOG_ERROR, "%s: Could not add to output", - str); + 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) { @@ -160,19 +327,21 @@ out: int _get_path_from_gfid_loc (xlator_t *this, xlator_t *readdir_xl, loc_t *child, - char **fpath) + 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) + 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_DEBUG, "Failed to get path for " + gf_log (this->name, GF_LOG_ERROR, "Failed to get path for " "gfid %s", uuid_utoa (child->gfid)); goto out; } @@ -201,54 +370,57 @@ _add_event_to_dict (circular_buffer_t *cb, void *data) shd_event = cb->data; if (shd_event->child != dump_data->child) goto out; - if (cb->tv.tv_sec >= dump_data->sh_time) - ret = _add_str_to_dict (dump_data->this, dump_data->dict, - dump_data->child, shd_event->path, - _gf_false); + 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_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict, time_t sh_time, - int child) +_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.sh_time = sh_time; dump_data.child = child; eh_dump (eh, &dump_data, _add_event_to_dict); return 0; } + 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) +_add_statistics_to_dict (xlator_t *this, dict_t *dict, int child) { - dict_t *output = NULL; - xlator_t *readdir_xl = NULL; - int ret = -1; - char *path = NULL; - - if (uuid_is_null (childloc->gfid)) - goto out; + shd_dump_t dump_data = {0}; + afr_private_t *priv = NULL; + afr_self_heald_t *shd = NULL; - output = crawl_data->op_data; - readdir_xl = crawl_data->readdir_xl; + priv = this->private; + shd = &priv->shd; - ret = _get_path_from_gfid_loc (this, readdir_xl, childloc, &path); - if (ret) - goto out; + 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; - ret = _add_str_to_dict (this, output, crawl_data->child, path, - _gf_true); -out: - if (ret && path) - GF_FREE (path); - return ret; } void @@ -261,13 +433,13 @@ _remove_stale_index (xlator_t *this, xlator_t *readdir_xl, ret = _build_index_loc (this, &index_loc, fname, parent); if (ret) goto out; - gf_log (this->name, GF_LOG_INFO, "Removing stale index " + 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) { - gf_log (this->name, GF_LOG_ERROR, "%s: Failed to remove" - " index on %s - %s", index_loc.name, - readdir_xl->name, strerror (errno)); + 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); @@ -275,30 +447,108 @@ 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; - shd_event_t *event = NULL; - int32_t sh_failed = 0; - gf_boolean_t split_brain = 0; + 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 (child->gfid)); + parent, uuid_utoa_r (child->gfid, + gfid_str)); goto out; } ret = _get_path_from_gfid_loc (this, crawl_data->readdir_xl, - child, &path); + child, &path, NULL); if (ret) goto out; } else { @@ -309,26 +559,45 @@ _crawl_post_sh_action (xlator_t *this, loc_t *parent, loc_t *child, } } - if (xattr_rsp) + 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) + if ((op_ret < 0 && op_errno == EIO) || split_brain) { eh = shd->split_brain; - else if ((op_ret < 0) || sh_failed) + shd_crawl_event[crawl_data->child]->split_brain_count += 1; + } else if ((op_ret < 0) || sh_failed) { eh = shd->heal_failed; - else + 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; - 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 " - "eh, (%d, %s)", path, op_ret, strerror (op_errno)); - goto out; + + 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: @@ -338,25 +607,56 @@ out: } 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; - if (uuid_is_null (child->gfid)) - gf_log (this->name, GF_LOG_DEBUG, "lookup %s", child->path); - else - gf_log (this->name, GF_LOG_DEBUG, "lookup %s", - uuid_utoa (child->gfid)); + 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); - ret = syncop_lookup (this, child, NULL, + 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; } @@ -371,13 +671,6 @@ afr_crawl_done (int ret, call_frame_t *sync_frame, void *data) void _do_self_heal_on_subvol (xlator_t *this, int child, afr_crawl_type_t crawl) { - afr_private_t *priv = NULL; - afr_self_heald_t *shd = NULL; - - priv = this->private; - shd = &priv->shd; - - time (&shd->sh_times[child]); afr_start_crawl (this, child, crawl, _self_heal_entry, NULL, _gf_true, STOP_CRAWL_ON_SINGLE_SUBVOL, afr_crawl_done); @@ -395,11 +688,11 @@ _crawl_proceed (xlator_t *this, int child, int crawl_flags, char **reason) shd = &priv->shd; if (!shd->enabled) { msg = "Self-heal daemon is not enabled"; - gf_log (this->name, GF_LOG_ERROR, "%s", msg); + gf_log (this->name, GF_LOG_DEBUG, "%s", msg); goto out; } if (!priv->child_up[child]) { - gf_log (this->name, GF_LOG_ERROR, "Stopping crawl for %s , " + 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; @@ -408,7 +701,7 @@ _crawl_proceed (xlator_t *this, int child, int crawl_flags, char **reason) 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_ERROR, "Stopping crawl as " + gf_log (this->name, GF_LOG_DEBUG, "Stopping crawl as " "< 2 children are up"); msg = "< 2 bricks in replica are running"; goto out; @@ -440,20 +733,28 @@ _do_crawl_op_on_local_subvols (xlator_t *this, afr_crawl_type_t crawl, if (op == HEAL) crawl_flags |= STOP_CRAWL_ON_SINGLE_SUBVOL; - 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; + 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; - ret = synctask_new (this->ctx->env, - afr_syncop_find_child_position, - NULL, NULL, &pos_data); + /* + * 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) { @@ -464,22 +765,35 @@ _do_crawl_op_on_local_subvols (xlator_t *this, afr_crawl_type_t crawl, status = "Started self-heal"; _do_self_heal_on_subvol (this, i, crawl); - } else { + } 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); } } - snprintf (key, sizeof (key), "%d-%d-%s", xl_id, - i, subkey); - ret = dict_set_str (output, key, status); + 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; } - snprintf (key, sizeof (key), "%d-%d-%s", xl_id, i, subkey); - ret = dict_set_str (output, key, status); + if (output) { + snprintf (key, sizeof (key), "%d-%d-%s", xl_id, i, + subkey); + ret = dict_set_str (output, key, status); + } } out: return op_ret; @@ -498,8 +812,103 @@ _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_all_subvols_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict) +_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; @@ -511,7 +920,7 @@ _add_all_subvols_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict) for (i = 0; i < priv->child_count; i++) { if (shd->pos[i] != AFR_POS_LOCAL) continue; - _add_eh_to_dict (this, eh, dict, shd->sh_times[i], i); + _add_eh_to_dict (this, eh, dict, i); } return 0; } @@ -549,16 +958,25 @@ afr_xl_op (xlator_t *this, dict_t *input, dict_t *output) ret = 0; break; case GF_AFR_OP_HEALED_FILES: - ret = _add_all_subvols_eh_to_dict (this, shd->healed, output); + ret = _add_local_subvols_eh_to_dict (this, shd->healed, output); break; case GF_AFR_OP_HEAL_FAILED_FILES: - ret = _add_all_subvols_eh_to_dict (this, shd->heal_failed, + ret = _add_local_subvols_eh_to_dict (this, shd->heal_failed, output); break; case GF_AFR_OP_SPLIT_BRAIN_FILES: - ret = _add_all_subvols_eh_to_dict (this, shd->split_brain, + 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; @@ -573,33 +991,48 @@ afr_poll_self_heal (void *data) { afr_private_t *priv = NULL; afr_self_heald_t *shd = NULL; - struct timeval timeout = {0}; + 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; - _do_self_heal_on_subvol (this, child, INDEX); - timeout.tv_sec = AFR_POLL_TIMEOUT; - timeout.tv_usec = 0; + 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) { + 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); @@ -608,7 +1041,7 @@ afr_poll_self_heal (void *data) } static int -afr_local_child_poll_self_heal (int ret, call_frame_t *sync_frame, void *data) +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; @@ -620,8 +1053,9 @@ afr_local_child_poll_self_heal (int ret, call_frame_t *sync_frame, void *data) priv = pos_data->this->private; shd = &priv->shd; shd->pos[pos_data->child] = pos_data->pos; - if (pos_data->pos == AFR_POS_LOCAL) + 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; @@ -645,13 +1079,69 @@ afr_proactive_self_heal (void *data) pos_data->this = this; pos_data->child = child; ret = synctask_new (this->ctx->env, afr_syncop_find_child_position, - afr_local_child_poll_self_heal, NULL, pos_data); + 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) @@ -659,6 +1149,7 @@ afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data, 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}; @@ -668,7 +1159,7 @@ afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data, priv = this->private; if (crawl_data->crawl == FULL) { afr_build_root_loc (this, dirloc); - } else { + } else if (crawl_data->crawl == INDEX) { afr_build_root_loc (this, &rootloc); ret = syncop_getxattr (readdir_xl, &rootloc, &xattr, GF_XATTROP_INDEX_GFID); @@ -692,11 +1183,57 @@ afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data, ret = syncop_lookup (readdir_xl, dirloc, NULL, &iattr, NULL, &parent); if (ret < 0) { - gf_log (this->name, GF_LOG_ERROR, "lookup failed on " - "index dir on %s", readdir_xl->name); + 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; } - inode_link (dirloc->inode, NULL, NULL, &iattr); + ret = _link_inode_update_loc (this, dirloc, &iattr); + if (ret) + goto out; } ret = 0; out: @@ -755,17 +1292,30 @@ 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 = 0; + 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->path = ""; 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; } @@ -779,7 +1329,6 @@ _process_entries (xlator_t *this, loc_t *parentloc, gf_dirent_t *entries, loc_t entry_loc = {0}; fd_t *fd = NULL; struct iatt iattr = {0}; - inode_t *link_inode = NULL; list_for_each_entry_safe (entry, tmp, &entries->list, list) { if (!_crawl_proceed (this, crawl_data->child, @@ -799,8 +1348,6 @@ _process_entries (xlator_t *this, loc_t *parentloc, gf_dirent_t *entries, continue; } - if (crawl_data->crawl == INDEX) - entry_loc.path = NULL;//HACK loc_wipe (&entry_loc); ret = afr_crawl_build_child_loc (this, &entry_loc, parentloc, entry, crawl_data); @@ -810,22 +1357,18 @@ _process_entries (xlator_t *this, loc_t *parentloc, gf_dirent_t *entries, ret = crawl_data->process_entry (this, crawl_data, entry, &entry_loc, parentloc, &iattr); - if (crawl_data->crawl == INDEX) + if (crawl_data->crawl == INDEX_TO_BE_HEALED && ret) { + goto out; + } else if (ret) { continue; + } - if (ret || !IA_ISDIR (iattr.ia_type)) + if ((crawl_data->crawl == INDEX) || + (crawl_data->crawl == INDEX_TO_BE_HEALED)) continue; - link_inode = inode_link (entry_loc.inode, NULL, NULL, &iattr); - if (link_inode == NULL) { - char uuidbuf[64]; - gf_log (this->name, GF_LOG_ERROR, "inode link failed " - "on the inode (%s)", - uuid_utoa_r (entry_loc.gfid, uuidbuf)); - ret = -1; - goto out; - } - + if (!IA_ISDIR (iattr.ia_type)) + continue; fd = NULL; ret = afr_crawl_opendir (this, crawl_data, &fd, &entry_loc); if (ret) @@ -836,10 +1379,11 @@ _process_entries (xlator_t *this, loc_t *parentloc, gf_dirent_t *entries, } ret = 0; out: - if (crawl_data->crawl == INDEX) - entry_loc.path = NULL; - if (entry_loc.path) - loc_wipe (&entry_loc); + 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; } @@ -858,10 +1402,10 @@ _crawl_directory (fd_t *fd, loc_t *loc, afr_crawl_data_t *crawl_data) GF_ASSERT (loc->inode); - if (loc->path) + 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 %s", + gf_log (this->name, GF_LOG_DEBUG, "crawling INDEX %s", uuid_utoa (loc->gfid)); while (1) { @@ -886,6 +1430,9 @@ _crawl_directory (fd_t *fd, loc_t *loc, afr_crawl_data_t *crawl_data) 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; } @@ -927,16 +1474,16 @@ afr_find_child_position (xlator_t *this, int child, afr_child_pos_t *pos) ret = syncop_getxattr (priv->children[child], &loc, &xattr_rsp, GF_XATTR_NODE_UUID_KEY); - if (ret) { - gf_log (this->name, GF_LOG_ERROR, "getxattr failed on %s", - priv->children[child]->name); + 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 %d", child); + "child %s", priv->children[child]->name); goto out; } @@ -945,7 +1492,7 @@ afr_find_child_position (xlator_t *this, int child, afr_child_pos_t *pos) else *pos = AFR_POS_REMOTE; - gf_log (this->name, GF_LOG_INFO, "child %s is %s", + gf_log (this->name, GF_LOG_DEBUG, "child %s is %s", priv->children[child]->name, position_str_get (*pos)); out: if (ret) @@ -991,27 +1538,122 @@ afr_dir_crawl (void *data) goto out; ret = afr_crawl_opendir (this, crawl_data, &fd, &dirloc); - if (ret) + 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_INFO, "Crawl completed " + 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) + 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) { @@ -1047,7 +1689,15 @@ afr_dir_exclusive_crawl (void *data) } 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) { @@ -1094,7 +1744,7 @@ afr_start_crawl (xlator_t *this, int idx, afr_crawl_type_t crawl, crawl_data->crawl = crawl; crawl_data->op_data = op_data; crawl_data->crawl_flags = crawl_flags; - gf_log (this->name, GF_LOG_INFO, "starting crawl %d for %s", + gf_log (this->name, GF_LOG_DEBUG, "starting crawl %d for %s", crawl_data->crawl, priv->children[idx]->name); if (exclusive) @@ -1104,8 +1754,8 @@ afr_start_crawl (xlator_t *this, int idx, afr_crawl_type_t crawl, ret = synctask_new (this->ctx->env, crawler, crawl_done, frame, crawl_data); if (ret) - gf_log (this->name, GF_LOG_ERROR, "Could not create the " - "task for %d ret %d", idx, ret); + gf_log (this->name, GF_LOG_ERROR, "afr crawl failed for child" + " %d with ret %d", idx, ret); out: return; } @@ -1135,4 +1785,3 @@ afr_set_root_gfid (dict_t *dict) return ret; } - diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h index 848738621..e0c083754 100644 --- a/xlators/cluster/afr/src/afr-self-heald.h +++ b/xlators/cluster/afr/src/afr-self-heald.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.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_HEALD_H__ @@ -38,6 +29,19 @@ typedef struct afr_crawl_data_ { 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); @@ -51,4 +55,11 @@ 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 7a95f310a..20306e469 100644 --- a/xlators/cluster/afr/src/afr-transaction.c +++ b/xlators/cluster/afr/src/afr-transaction.c @@ -1,25 +1,17 @@ /* - Copyright (c) 2007-2011 Gluster, Inc. <http://www.gluster.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" @@ -32,7 +24,6 @@ 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) { @@ -82,27 +73,26 @@ afr_fd_ctx_get (fd_t *fd, xlator_t *this) static void -afr_pid_save (call_frame_t *frame) +afr_save_lk_owner (call_frame_t *frame) { afr_local_t * local = NULL; local = frame->local; - local->saved_pid = frame->root->pid; + local->saved_lk_owner = frame->root->lk_owner; } static void -afr_pid_restore (call_frame_t *frame) +afr_restore_lk_owner (call_frame_t *frame) { afr_local_t * local = NULL; local = frame->local; - frame->root->pid = local->saved_pid; + frame->root->lk_owner = local->saved_lk_owner; } - static void __mark_all_pending (int32_t *pending[], int child_count, afr_transaction_type type) @@ -155,51 +145,23 @@ out: return; } - -static void -__mark_pre_op_undone_on_fd (call_frame_t *frame, xlator_t *this, int child_index) -{ - 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 (&local->fd->lock); - { - if (local->transaction.type == AFR_DATA_TRANSACTION) - fd_ctx->pre_op_done[child_index]--; - } - UNLOCK (&local->fd->lock); -out: - return; -} - - static void -__mark_down_children (int32_t *pending[], int child_count, - unsigned char *child_up, afr_transaction_type type) +__mark_non_participant_children (int32_t *pending[], int child_count, + unsigned char *participants, + afr_transaction_type type) { int i = 0; int j = 0; + j = afr_index_for_transaction_type (type); for (i = 0; i < child_count; i++) { - j = afr_index_for_transaction_type (type); - - if (!child_up[i]) + if (!participants[i]) pending[i][j] = 0; } } -static void +void __mark_all_success (int32_t *pending[], int child_count, afr_transaction_type type) { @@ -212,6 +174,54 @@ __mark_all_success (int32_t *pending[], int child_count, } } +void +_set_all_child_errno (int *child_errno, unsigned int child_count) +{ + int i = 0; + + 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_enabled (afr_private_t *priv, afr_transaction_type type) @@ -276,63 +286,41 @@ __fop_changelog_needed (call_frame_t *frame, xlator_t *this) } int -afr_set_pending_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending) +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)); + AFR_NUM_CHANGE_LOGS * sizeof (int32_t)); /* 3 = data+metadata+entry */ if (ret < 0) goto out; } - -out: - return ret; -} - - -static int -afr_set_piggyback_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending, - afr_transaction_type type) -{ - int i = 0; - int ret = 0; - int *arr = NULL; - int index = 0; - size_t pending_xattr_size = 3 * sizeof (int32_t); - /* 3 = data+metadata+entry */ - - index = afr_index_for_transaction_type (type); - - for (i = 0; i < priv->child_count; i++) { - arr = GF_CALLOC (1, pending_xattr_size, - gf_afr_mt_char); - if (!arr) { - ret = -1; - goto out; - } - - memcpy (arr, pending[i], pending_xattr_size); - - arr[index] = hton32 (ntoh32(arr[index]) + 1); - - ret = dict_set_bin (xattr, priv->pending_key[i], - arr, pending_xattr_size); - - if (ret < 0) + 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; } - int afr_lock_server_count (afr_private_t *priv, afr_transaction_type type) { @@ -379,6 +367,11 @@ afr_changelog_post_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this, UNLOCK (&frame->lock); 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 { @@ -445,24 +438,41 @@ afr_transaction_rm_stale_children (call_frame_t *frame, xlator_t *this, afr_inode_rm_stale_children (this, inode, stale_children); out: - if (stale_children) - GF_FREE (stale_children); + 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: - locked_nodes = int_lock->inode_locked_nodes; + inodelk = afr_get_inodelk (int_lock, int_lock->domain); + locked_nodes = inodelk->locked_nodes; break; case AFR_ENTRY_TRANSACTION: case AFR_ENTRY_RENAME_TRANSACTION: - locked_nodes = int_lock->entry_locked_nodes; + /*Because same set of subvols participate in all lockee + * entities*/ + locked_nodes = int_lock->lockee[0].locked_nodes; break; } return locked_nodes; @@ -500,12 +510,141 @@ afr_changelog_post_op_call_count (afr_transaction_type type, 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; + + index = afr_index_for_transaction_type (type); + if (optimized && !pending[child][index]) + op = LOCAL_FIRST; + return op; +} + +void +afr_set_postop_dict (afr_local_t *local, xlator_t *this, dict_t *xattr, + int optimized, int child) +{ + 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"); +} + + +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; + + 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; + + 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 (call_frame_t *frame, xlator_t *this) +afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this) { afr_private_t * priv = this->private; afr_internal_lock_t *int_lock = NULL; - int ret = 0; int i = 0; int call_count = 0; @@ -513,14 +652,17 @@ afr_changelog_post_op (call_frame_t *frame, xlator_t *this) afr_fd_ctx_t *fdctx = NULL; dict_t **xattr = NULL; int piggyback = 0; - int index = 0; int nothing_failed = 1; local = frame->local; int_lock = &local->internal_lock; - __mark_down_children (local->pending, priv->child_count, - local->child_up, local->transaction.type); + __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, @@ -548,38 +690,23 @@ afr_changelog_post_op (call_frame_t *frame, xlator_t *this) goto out; } - /* check if something has failed, to handle piggybacking */ - nothing_failed = 1; - index = afr_index_for_transaction_type (local->transaction.type); - for (i = 0; i < priv->child_count; i++) { - if (local->pending[i][index] == 0) { - nothing_failed = 0; - break; - } - } + nothing_failed = afr_txn_nothing_failed (frame, this); - index = afr_index_for_transaction_type (local->transaction.type); - if (local->optimistic_change_log && - local->transaction.type != AFR_DATA_TRANSACTION) { - /* if nothing_failed, then local->pending[..] == {0 .. 0} */ - for (i = 0; i < priv->child_count; i++) - local->pending[i][index]++; - } + afr_compute_txn_changelog (local , priv); for (i = 0; i < priv->child_count; i++) { if (!local->transaction.pre_op[i]) continue; - ret = afr_set_pending_dict (priv, xattr[i], local->pending); - - if (ret < 0) - gf_log (this->name, GF_LOG_INFO, - "failed to set pending entry"); - + 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, @@ -589,26 +716,22 @@ afr_changelog_post_op (call_frame_t *frame, xlator_t *this) break; } - LOCK (&local->fd->lock); - { - piggyback = 0; - if (fdctx->pre_op_piggyback[i]) { - fdctx->pre_op_piggyback[i]--; - piggyback = 1; - } - } - UNLOCK (&local->fd->lock); + /* local->transaction.postop_piggybacked[] was + precomputed in is_piggyback_postop() when called from + afr_changelog_post_op_safe() + */ - if (piggyback && !nothing_failed) - ret = afr_set_piggyback_dict (priv, xattr[i], - local->pending, - local->transaction.type); + 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 { - __mark_pre_op_undone_on_fd (frame, this, i); STACK_WIND_COOKIE (frame, afr_changelog_post_op_cbk, (void *) (long) i, @@ -622,7 +745,7 @@ afr_changelog_post_op (call_frame_t *frame, xlator_t *this) break; case AFR_METADATA_TRANSACTION: { - if (nothing_failed) { + if (nothing_failed && local->optimistic_change_log) { afr_changelog_post_op_cbk (frame, (void *)(long)i, this, 1, 0, xattr[i], NULL); @@ -648,7 +771,7 @@ afr_changelog_post_op (call_frame_t *frame, xlator_t *this) case AFR_ENTRY_RENAME_TRANSACTION: { - if (nothing_failed) { + if (nothing_failed && local->optimistic_change_log) { afr_changelog_post_op_cbk (frame, (void *)(long)i, this, 1, 0, xattr[i], NULL); @@ -672,17 +795,14 @@ afr_changelog_post_op (call_frame_t *frame, xlator_t *this) value */ - ret = afr_set_pending_dict (priv, xattr[i], local->pending); - - if (ret < 0) - gf_log (this->name, GF_LOG_INFO, - "failed to set pending entry"); + afr_set_postop_dict (local, this, xattr[i], + local->optimistic_change_log, i); /* fall through */ case AFR_ENTRY_TRANSACTION: { - if (nothing_failed) { + if (nothing_failed && local->optimistic_change_log) { afr_changelog_post_op_cbk (frame, (void *)(long)i, this, 1, 0, xattr[i], NULL); @@ -768,12 +888,7 @@ afr_changelog_pre_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this, (local->op_errno == ENOTSUP)) { local->transaction.resume (frame, this); } else { - __mark_all_success (local->pending, priv->child_count, - local->transaction.type); - - afr_pid_restore (frame); - - local->transaction.fop (frame, this); + afr_transaction_perform_fop (frame, this); } } @@ -826,7 +941,8 @@ afr_changelog_pre_op (call_frame_t *frame, xlator_t *this) for (i = 0; i < priv->child_count; i++) { if (!locked_nodes[i]) continue; - ret = afr_set_pending_dict (priv, xattr[i], local->pending); + ret = afr_set_pending_dict (priv, xattr[i], local->pending, + i, LOCAL_FIRST); if (ret < 0) gf_log (this->name, GF_LOG_INFO, @@ -861,6 +977,8 @@ afr_changelog_pre_op (call_frame_t *frame, xlator_t *this) } 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], @@ -935,7 +1053,8 @@ afr_changelog_pre_op (call_frame_t *frame, xlator_t *this) value */ - ret = afr_set_pending_dict (priv, xattr[i], local->pending); + ret = afr_set_pending_dict (priv, xattr[i], local->pending, + i, LOCAL_FIRST); if (ret < 0) gf_log (this->name, GF_LOG_INFO, @@ -1024,7 +1143,6 @@ afr_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this) gf_log (this->name, GF_LOG_DEBUG, "Non blocking inodelks failed. Proceeding to blocking"); int_lock->lock_cbk = afr_post_blocking_inodelk_cbk; - afr_set_lk_owner (frame, this, frame->root); afr_blocking_lock (frame, this); } else { @@ -1132,12 +1250,14 @@ 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); - int_lock->lk_flock.l_len = local->transaction.len; - int_lock->lk_flock.l_start = local->transaction.start; - int_lock->lk_flock.l_type = F_WRLCK; + inodelk->flock.l_len = local->transaction.len; + inodelk->flock.l_start = local->transaction.start; + inodelk->flock.l_type = F_WRLCK; return 0; } @@ -1152,6 +1272,7 @@ afr_lock_rec (call_frame_t *frame, xlator_t *this) 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: @@ -1165,8 +1286,8 @@ afr_lock_rec (call_frame_t *frame, xlator_t *this) case AFR_ENTRY_RENAME_TRANSACTION: - int_lock->lock_cbk = afr_post_blocking_rename_cbk; - afr_blocking_lock (frame, this); + int_lock->lock_cbk = afr_post_nonblocking_entrylk_cbk; + afr_nonblocking_entrylk (frame, this); break; case AFR_ENTRY_TRANSACTION: @@ -1188,12 +1309,6 @@ afr_lock_rec (call_frame_t *frame, xlator_t *this) int afr_lock (call_frame_t *frame, xlator_t *this) { - afr_pid_save (frame); - - frame->root->pid = (long) frame->root; - - afr_set_lk_owner (frame, this, frame->root); - afr_set_lock_number (frame, this); return afr_lock_rec (frame, this); @@ -1205,28 +1320,463 @@ afr_lock (call_frame_t *frame, xlator_t *this) int afr_internal_lock_finish (call_frame_t *frame, xlator_t *this) { - afr_local_t *local = NULL; - afr_private_t *priv = NULL; + if (__fop_changelog_needed (frame, this)) { + afr_changelog_pre_op (frame, this); + } else { + afr_transaction_perform_fop (frame, this); + } + + return 0; +} + + +void +afr_set_delayed_post_op (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + + /* call this function from any of the related optimizations + which benefit from delaying post op are enabled, namely: + + - changelog piggybacking + - eager locking + */ + + priv = this->private; + if (!priv) + return; + + if (!priv->post_op_delay_secs) + return; + + local = frame->local; + if (!local->transaction.eager_lock_on) + return; + + 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; +} + +gf_boolean_t +is_afr_delayed_changelog_post_op_needed (call_frame_t *frame, xlator_t *this) +{ + 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 (__fop_changelog_needed (frame, this)) { - afr_changelog_pre_op (frame, this); - } else { - __mark_all_success (local->pending, priv->child_count, - local->transaction.type); + if (!local->delayed_post_op) + goto out; + + //Mark pending changelog ASAP + if (afr_any_fops_failed (local, priv)) + goto out; + + if (local->fd && afr_are_multiple_fds_opened (local->fd->inode, this)) + goto out; + + res = _gf_true; +out: + return res; +} + + +void +afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd, + call_stub_t *stub); - afr_pid_restore (frame); +void +afr_delayed_changelog_wake_up_cbk (void *data) +{ + fd_t *fd = NULL; + + fd = data; + + afr_delayed_changelog_wake_up (THIS, fd); +} + + +/* + 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; - local->transaction.fop (frame, this); + 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); + + if (!afr_txn_nothing_failed (frame, frame->this)) { + /* something failed in this transaction, + we will be performing a hard post-op + */ + return _gf_false; + } + + return piggyback; +} + + +/* SET operation */ +int +afr_fd_report_unstable_write (xlator_t *this, fd_t *fd) +{ + afr_fd_ctx_t *fdctx = NULL; + + fdctx = afr_fd_ctx_get (fd, this); + + LOCK(&fd->lock); + { + fdctx->witnessed_unstable_write = _gf_true; + } + UNLOCK(&fd->lock); return 0; } +/* 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; + + fdctx = afr_fd_ctx_get (fd, this); + if (!fdctx) + return _gf_true; + + 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; +} + + +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); + { + 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; + } +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); + } +} + + +void +afr_changelog_post_op (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + + 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); +} + + + +/* 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_internal_lock_t *int_lock = NULL; @@ -1237,6 +1787,19 @@ afr_transaction_resume (call_frame_t *frame, xlator_t *this) 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); + } + + afr_restore_lk_owner (frame); + if (__fop_changelog_needed (frame, this)) { afr_changelog_post_op (frame, this); } else { @@ -1257,7 +1820,8 @@ afr_transaction_resume (call_frame_t *frame, xlator_t *this) */ void -afr_transaction_fop_failed (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; @@ -1266,7 +1830,89 @@ afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this, int child_index priv = this->private; __mark_child_dead (local->pending, priv->child_count, - child_index, local->transaction.type); + child_index, local->transaction.type); +} + + + + 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); } @@ -1275,20 +1921,43 @@ afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type) { afr_local_t * local = NULL; afr_private_t * priv = NULL; + fd_t *fd = NULL; + int ret = -1; local = frame->local; priv = this->private; - afr_transaction_local_init (local, this); - local->transaction.resume = afr_transaction_resume; local->transaction.type = type; + ret = afr_transaction_local_init (local, this); + if (ret < 0) + goto out; + + afr_transaction_eager_lock_init (local, this); + + 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); } - - return 0; + ret = 0; +out: + return ret; } diff --git a/xlators/cluster/afr/src/afr-transaction.h b/xlators/cluster/afr/src/afr-transaction.h index f470f2697..fa626fd0d 100644 --- a/xlators/cluster/afr/src/afr-transaction.h +++ b/xlators/cluster/afr/src/afr-transaction.h @@ -1,25 +1,21 @@ /* - Copyright (c) 2007-2011 Gluster, Inc. <http://www.gluster.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__ +typedef enum { + LOCAL_FIRST = 1, + LOCAL_LAST = 2 +} afr_xattrop_type_t; + void afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this, int child_index); @@ -27,11 +23,29 @@ afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this, 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); +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 f6db7b9e9..c724eb2ae 100644 --- a/xlators/cluster/afr/src/afr.c +++ b/xlators/cluster/afr/src/afr.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2007-2011 Gluster, Inc. <http://www.gluster.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> @@ -116,6 +107,7 @@ reconfigure (xlator_t *this, dict_t *options) { afr_private_t *priv = NULL; xlator_t *read_subvol = NULL; + int read_subvol_index = -1; int ret = -1; int index = -1; char *qtype = NULL; @@ -158,6 +150,9 @@ reconfigure (xlator_t *this, dict_t *options) GF_OPTION_RECONF ("read-subvolume", read_subvol, options, xlator, out); + GF_OPTION_RECONF ("read-hash-mode", priv->hash_mode, + options, uint32, out); + if (read_subvol) { index = xlator_subvolume_index (this, read_subvol); if (index == -1) { @@ -168,11 +163,37 @@ reconfigure (xlator_t *this, dict_t *options) priv->read_child = index; } + GF_OPTION_RECONF ("read-subvolume-index",read_subvol_index, options,int32,out); + + 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; + } + 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: @@ -200,6 +221,7 @@ init (xlator_t *this) 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; @@ -231,7 +253,6 @@ init (xlator_t *this) priv->child_count = child_count; - priv->read_child = -1; GF_OPTION_INIT ("read-subvolume", read_subvol, xlator, out); @@ -243,6 +264,18 @@ init (xlator_t *this) 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); @@ -298,8 +331,15 @@ init (xlator_t *this) 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); + 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; priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count, @@ -340,8 +380,6 @@ init (xlator_t *this) AFR_XATTR_PREFIX, trav->xlator->name); if (-1 == ret) { - gf_log (this->name, GF_LOG_ERROR, - "asprintf failed to set pending key"); ret = -ENOMEM; goto out; } @@ -350,6 +388,13 @@ init (xlator_t *this) i++; } + ret = gf_asprintf (&priv->sh_domain, AFR_SH_DATA_DOMAIN_FMT, + this->name); + if (-1 == ret) { + ret = -ENOMEM; + goto out; + } + priv->last_event = GF_CALLOC (child_count, sizeof (*priv->last_event), gf_afr_mt_int32_t); if (!priv->last_event) { @@ -394,30 +439,30 @@ init (xlator_t *this) if (!priv->shd.timer) goto out; - priv->shd.healed = eh_new (AFR_EH_HEALED_LIMIT, _gf_false); + 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); + 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); + priv->shd.split_brain = eh_new (AFR_EH_SPLIT_BRAIN_LIMIT, _gf_false, + _destroy_shd_event_data); if (!priv->shd.split_brain) goto out; - priv->shd.sh_times = GF_CALLOC (priv->child_count, - sizeof (*priv->shd.sh_times), - gf_afr_mt_time_t); - if (!priv->shd.sh_times) - 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: return ret; @@ -452,6 +497,9 @@ struct xlator_fops fops = { .finodelk = afr_finodelk, .entrylk = afr_entrylk, .fentrylk = afr_fentrylk, + .fallocate = afr_fallocate, + .discard = afr_discard, + .zerofill = afr_zerofill, /* inode read */ .access = afr_access, @@ -504,33 +552,79 @@ struct xlator_cbks cbks = { struct volume_options options[] = { { .key = {"read-subvolume" }, - .type = GF_OPTION_TYPE_XLATOR + .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 + .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, - .default_value = "", .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, - .default_value = "", .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.", - .value = { "diff", "full", "" } + "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, @@ -543,26 +637,43 @@ struct volume_options options[] = { { .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, @@ -571,26 +682,55 @@ struct volume_options options[] = { { .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 = "off", + .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", "" }, + .value = { "none", "auto", "fixed"}, .default_value = "none", .description = "If value is \"fixed\" only allow writes if " "quorum-count bricks are present. If value is " @@ -609,7 +749,45 @@ struct volume_options options[] = { }, { .key = {"node-uuid"}, .type = GF_OPTION_TYPE_STR, - .description = "Local glusterd uuid string", + .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 954e9bb31..21064db58 100644 --- a/xlators/cluster/afr/src/afr.h +++ b/xlators/cluster/afr/src/afr.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2008-2011 Gluster, Inc. <http://www.gluster.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. */ @@ -36,6 +27,11 @@ #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; @@ -66,10 +62,8 @@ typedef enum { AFR_INODE_SET_READ_CTX = 1, AFR_INODE_RM_STALE_CHILDREN, AFR_INODE_SET_OPENDIR_DONE, - AFR_INODE_SET_SPLIT_BRAIN, AFR_INODE_GET_READ_CTX, AFR_INODE_GET_OPENDIR_DONE, - AFR_INODE_GET_SPLIT_BRAIN, } afr_inode_op_t; typedef struct afr_inode_params_ { @@ -83,29 +77,41 @@ typedef struct afr_inode_params_ { } 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; - time_t *sh_times; - gf_timer_t **timer; - eh_t *healed; - eh_t *heal_failed; - eh_t *split_brain; - char *node_uuid; + 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 { @@ -139,6 +145,7 @@ typedef struct _afr_private { gf_boolean_t entry_change_log; /* on/off */ 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 */ @@ -159,14 +166,48 @@ typedef struct _afr_private { 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; 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_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 */ @@ -175,6 +216,8 @@ typedef struct { 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? */ @@ -241,9 +284,10 @@ typedef struct { const char *linkname; gf_boolean_t entries_skipped; - int op_failed; - + 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; @@ -254,17 +298,22 @@ typedef struct { uint8_t *checksum; afr_post_remove_call_t post_remove_call; - loc_t parent_loc; + 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); @@ -272,7 +321,9 @@ typedef struct { void (*gfid_sh_success_cbk) (call_frame_t *sh_frame, xlator_t *this); call_frame_t *sh_frame; -} afr_self_heal_t; +}; + +typedef struct afr_self_heal_ afr_self_heal_t; typedef enum { AFR_DATA_TRANSACTION, /* truncate, write, ... */ @@ -334,11 +385,31 @@ afr_index_for_transaction_type (afr_transaction_type type) return -1; /* make gcc happy */ } +typedef struct { + loc_t loc; + char *basename; + unsigned char *locked_nodes; + int locked_count; + +} afr_entry_lockee_t; + +int +afr_entry_lockee_cmp (const void *l1, const void *l2); + +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; typedef struct { loc_t *lk_loc; - struct gf_flock lk_flock; + int lockee_count; + afr_entry_lockee_t lockee[AFR_LOCKEE_COUNT_MAX]; + + afr_inodelk_t inodelk[AFR_DOM_COUNT_MAX]; const char *lk_basename; const char *lower_basename; const char *higher_basename; @@ -347,23 +418,22 @@ typedef struct { unsigned char *locked_nodes; unsigned char *lower_locked_nodes; - unsigned char *inode_locked_nodes; - unsigned char *entry_locked_nodes; selfheal_lk_type_t selfheal_lk_type; transaction_lk_type_t transaction_lk_type; int32_t lock_count; - int32_t inodelk_lock_count; int32_t entrylk_lock_count; uint64_t lock_number; int32_t lk_call_count; int32_t lk_expected_count; + int32_t lk_attempted_count; 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; typedef struct _afr_locked_fd { @@ -371,21 +441,29 @@ typedef struct _afr_locked_fd { struct list_head list; } afr_locked_fd_t; +struct afr_reply { + int valid; + int32_t op_ret; + int32_t op_errno; +}; + 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; - unsigned int govinda_gOvinda; + unsigned int unhealable; unsigned int read_child_index; unsigned char read_child_returned; unsigned int first_up_child; - pid_t saved_pid; + gf_lkowner_t saved_lk_owner; int32_t op_ret; int32_t op_errno; @@ -396,7 +474,6 @@ typedef struct _afr_local { loc_t newloc; fd_t *fd; - unsigned char *fd_open_on; glusterfs_fop_t fop; @@ -418,13 +495,25 @@ typedef struct _afr_local { dict_t *dict; int optimistic_change_log; + gf_boolean_t delayed_post_op; - gf_boolean_t fop_paused; - int (*fop_call_continue) (call_frame_t *frame, xlator_t *this); - /* - This struct contains the arguments for the "continuation" - (scheme-like) of fops + /* 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; @@ -435,6 +524,7 @@ typedef struct _afr_local { } statfs; struct { + uint32_t parent_entrylk; uuid_t gfid_req; inode_t *inode; struct iatt buf; @@ -446,7 +536,9 @@ typedef struct _afr_local { 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 { @@ -517,7 +609,9 @@ typedef struct _afr_local { struct { struct iatt prebuf; struct iatt postbuf; + } inode_wfop; //common structure for all inode-write-fops + struct { int32_t op_ret; struct iovec *vector; @@ -528,34 +622,21 @@ typedef struct _afr_local { } writev; struct { - struct iatt prebuf; - struct iatt postbuf; - } fsync; - - struct { off_t offset; - struct iatt prebuf; - struct iatt postbuf; } truncate; struct { off_t offset; - struct iatt prebuf; - struct iatt postbuf; } ftruncate; struct { struct iatt in_buf; int32_t valid; - struct iatt preop_buf; - struct iatt postop_buf; } setattr; struct { struct iatt in_buf; int32_t valid; - struct iatt preop_buf; - struct iatt postop_buf; } fsetattr; struct { @@ -583,90 +664,66 @@ typedef struct _afr_local { /* dir write */ struct { - fd_t *fd; - dict_t *params; - int32_t flags; - mode_t mode; inode_t *inode; struct iatt buf; struct iatt preparent; struct iatt postparent; - struct iatt read_child_buf; + 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; - inode_t *inode; - struct iatt buf; - struct iatt preparent; - struct iatt postparent; - struct iatt read_child_buf; } mknod; struct { int32_t mode; dict_t *params; - inode_t *inode; - struct iatt buf; - struct iatt read_child_buf; - struct iatt preparent; - struct iatt postparent; } mkdir; struct { - int32_t op_ret; - int32_t op_errno; - struct iatt preparent; - struct iatt postparent; - } unlink; - - struct { - int flags; - int32_t op_ret; - int32_t op_errno; - struct iatt preparent; - struct iatt postparent; + int flags; } rmdir; struct { - struct iatt buf; - struct iatt read_child_buf; - struct iatt preoldparent; - struct iatt prenewparent; - struct iatt postoldparent; - struct iatt postnewparent; - } rename; - - struct { - inode_t *inode; - struct iatt buf; - struct iatt read_child_buf; - struct iatt preparent; - struct iatt postparent; - } link; - - struct { - inode_t *inode; dict_t *params; - struct iatt buf; - struct iatt read_child_buf; char *linkpath; - struct iatt preparent; - struct iatt postparent; } symlink; + struct { + int32_t mode; + off_t offset; + size_t len; + } fallocate; + + struct { + off_t offset; + size_t len; + } discard; + struct { - int32_t flags; - dir_entry_t *entries; - int32_t count; - } setdents; + off_t offset; + size_t len; + struct iatt prebuf; + struct iatt postbuf; + } zerofill; + + } cont; struct { off_t start, len; + gf_boolean_t eager_lock_on; int *eager_lock; char *basename; @@ -677,12 +734,18 @@ typedef struct _afr_local { afr_transaction_type type; - int success_count; - int erase_pending; - int failure_count; + /* pre-compute the post piggyback status before + entering POST-OP phase + */ + int *postop_piggybacked; + + /* stub to resume on destruction + of the transaction frame */ + call_stub_t *resume_stub; + + struct list_head eager_locked; - int last_tried; - int32_t *child_errno; + int32_t **txn_changelog;//changelog after pre+post ops unsigned char *pre_op; call_frame_t *main_frame; @@ -708,6 +771,8 @@ typedef struct _afr_local { mode_t umask; int xflag; + gf_boolean_t do_discovery; + struct afr_reply *replies; } afr_local_t; typedef enum { @@ -717,11 +782,6 @@ typedef enum { } afr_fd_open_status_t; typedef struct { - struct list_head call_list; - call_frame_t *frame; -} afr_fd_paused_call_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; @@ -740,7 +800,20 @@ typedef struct { struct list_head entries; /* needed for readdir failover */ unsigned char *locked_on; /* which subvolumes locks have been successful */ - struct list_head paused_calls; /* queued calls while fix_open happens */ + + /* 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; @@ -776,6 +849,13 @@ 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_entry_lockee_cleanup (afr_internal_lock_t *int_lock); + +int afr_attempt_lock_recovery (xlator_t *this, int32_t child_index); int @@ -810,8 +890,8 @@ afr_blocking_lock (call_frame_t *frame, xlator_t *this); int afr_internal_lock_finish (call_frame_t *frame, xlator_t *this); -void -afr_lk_transfer_datalock (call_frame_t *dst, call_frame_t *src, +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); @@ -861,7 +941,8 @@ gf_boolean_t afr_is_split_brain (xlator_t *this, inode_t *inode); void -afr_set_split_brain (xlator_t *this, inode_t *inode, gf_boolean_t set); +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, @@ -948,12 +1029,13 @@ afr_first_up_child (unsigned char *child_up, size_t child_count); 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); + 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); + 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, @@ -975,8 +1057,9 @@ 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); -gf_boolean_t -afr_error_more_important (int32_t old_errno, int32_t new_errno); +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); @@ -1019,11 +1102,11 @@ afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode, int (*unwind) (call_frame_t *frame, xlator_t *this, int32_t op_ret, int32_t op_errno, int32_t sh_failed)); -int -afr_fix_open (call_frame_t *frame, xlator_t *this, afr_fd_ctx_t *fd_ctx, - int need_open_count, int *need_open); -int -afr_open_fd_fix (call_frame_t *frame, xlator_t *this, gf_boolean_t pause_fop); +void +afr_fix_open (xlator_t *this, fd_t *fd, size_t need_open_count, int *need_open); + +void +afr_open_fd_fix (fd_t *fd, xlator_t *this); int afr_set_elem_count_get (unsigned char *elems, int child_count); @@ -1050,6 +1133,23 @@ 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). @@ -1069,4 +1169,44 @@ afr_matrix_create (unsigned int m, unsigned int n); } \ } 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 index 8667b4007..a7f72fb30 100644 --- a/xlators/cluster/afr/src/pump.c +++ b/xlators/cluster/afr/src/pump.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2007-2011 Gluster, Inc. <http://www.gluster.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 <unistd.h> @@ -437,7 +428,7 @@ gf_pump_traverse_directory (loc_t *loc) gf_pump_traverse_directory (&entry_loc); } } - } + } } gf_dirent_free (&entries); @@ -448,6 +439,10 @@ gf_pump_traverse_directory (loc_t *loc) } + 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. " @@ -947,6 +942,7 @@ pump_execute_status (call_frame_t *frame, xlator_t *this) uint64_t number_files = 0; char filename[PATH_MAX]; + char summary[PATH_MAX+256]; char *dict_str = NULL; int32_t op_ret = 0; @@ -975,12 +971,15 @@ pump_execute_status (call_frame_t *frame, xlator_t *this) } if (pump_priv->pump_finished) { - snprintf (dict_str, PATH_MAX + 256, "Number of files migrated = %"PRIu64" Migration complete ", - number_files); + snprintf (summary, PATH_MAX+256, + "no_of_files=%"PRIu64, number_files); } else { - snprintf (dict_str, PATH_MAX + 256, "Number of files migrated = %"PRIu64" Current file= %s ", - number_files, filename); + 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 (); @@ -1001,8 +1000,7 @@ out: if (dict) dict_unref (dict); - if (dict_str) - GF_FREE (dict_str); + GF_FREE (dict_str); return 0; } @@ -1282,7 +1280,7 @@ struct _xattr_key { struct list_head list; }; -static void +static int __gather_xattr_keys (dict_t *dict, char *key, data_t *value, void *data) { @@ -1294,13 +1292,14 @@ __gather_xattr_keys (dict_t *dict, char *key, data_t *value, xkey = GF_CALLOC (1, sizeof (*xkey), gf_afr_mt_xattr_key); if (!xkey) - return; + return -1; xkey->key = key; INIT_LIST_HEAD (&xkey->list); list_add_tail (&xkey->list, list); } + return 0; } static void @@ -1607,8 +1606,6 @@ pump_command_reply (call_frame_t *frame, xlator_t *this) gf_log (this->name, GF_LOG_INFO, "Command succeeded"); - dict_unref (local->dict); - AFR_STACK_UNWIND (setxattr, frame, local->op_ret, @@ -1654,7 +1651,6 @@ pump_setxattr (call_frame_t *frame, xlator_t *this, afr_private_t * priv = NULL; afr_local_t * local = NULL; call_frame_t *transaction_frame = NULL; - data_pair_t * trav = NULL; int ret = -1; int op_errno = 0; @@ -1663,7 +1659,7 @@ pump_setxattr (call_frame_t *frame, xlator_t *this, VALIDATE_OR_GOTO (this->private, out); GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.pump*", dict, - trav, op_errno, out); + op_errno, out); priv = this->private; if (!priv->use_afr_in_pump) { @@ -2260,6 +2256,17 @@ pump_release (xlator_t *this, } +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, @@ -2430,14 +2437,13 @@ init (xlator_t *this) priv->metadata_self_heal = 1; priv->entry_self_heal = 1; - priv->data_self_heal_algorithm = ""; - 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 */ @@ -2495,6 +2501,12 @@ init (xlator_t *this) 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; @@ -2526,7 +2538,7 @@ init (xlator_t *this) goto out; } - pump_priv->env = syncenv_new (0); + pump_priv->env = this->ctx->env; if (!pump_priv->env) { gf_log (this->name, GF_LOG_ERROR, "Could not create new sync-environment"); @@ -2567,9 +2579,6 @@ fini (xlator_t *this) if (!pump_priv) goto afr_priv; - if (pump_priv->env) - syncenv_destroy (pump_priv->env); - GF_FREE (pump_priv->resume_path); LOCK_DESTROY (&pump_priv->resume_path_lock); LOCK_DESTROY (&pump_priv->pump_state_lock); @@ -2624,6 +2633,7 @@ struct xlator_dumpops dumpops = { struct xlator_cbks cbks = { .release = pump_release, .releasedir = pump_releasedir, + .forget = pump_forget, }; struct volume_options options[] = { diff --git a/xlators/cluster/afr/src/pump.h b/xlators/cluster/afr/src/pump.h index 4213a635e..bc4c31a78 100644 --- a/xlators/cluster/afr/src/pump.h +++ b/xlators/cluster/afr/src/pump.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2007-2011 Gluster, Inc. <http://www.gluster.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 __PUMP_H__ |