diff options
Diffstat (limited to 'xlators/cluster/afr/src/afr-common.c')
| -rw-r--r-- | xlators/cluster/afr/src/afr-common.c | 2681 |
1 files changed, 1848 insertions, 833 deletions
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c index d6b358e2b..af01f2ef2 100644 --- a/xlators/cluster/afr/src/afr-common.c +++ b/xlators/cluster/afr/src/afr-common.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2007-2010 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 Affero 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 - Affero General Public License for more details. - - You should have received a copy of the GNU Affero 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> @@ -44,6 +35,7 @@ #include "compat.h" #include "byte-order.h" #include "statedump.h" +#include "inode.h" #include "fd.h" @@ -54,12 +46,12 @@ #include "afr-transaction.h" #include "afr-self-heal.h" #include "afr-self-heal-common.h" +#include "afr-self-heald.h" #include "pump.h" -#define AFR_ICTX_OPENDIR_DONE_MASK 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); @@ -90,6 +82,75 @@ 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 +afr_lookup_xattr_req_prepare (afr_local_t *local, xlator_t *this, + dict_t *xattr_req, loc_t *loc, void **gfid_req) +{ + int ret = -ENOMEM; + + GF_ASSERT (gfid_req); + + *gfid_req = NULL; + local->xattr_req = dict_new (); + if (!local->xattr_req) + goto out; + if (xattr_req) + dict_copy (xattr_req, local->xattr_req); + + afr_xattr_req_prepare (this, local->xattr_req, loc->path); + ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_INODELK_COUNT, 0); + if (ret < 0) { + gf_log (this->name, GF_LOG_WARNING, + "%s: Unable to set dict value for %s", + loc->path, GLUSTERFS_INODELK_COUNT); + } + ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_ENTRYLK_COUNT, 0); + if (ret < 0) { + gf_log (this->name, GF_LOG_WARNING, + "%s: Unable to set dict value for %s", + loc->path, GLUSTERFS_ENTRYLK_COUNT); + } + + ret = dict_set_uint32 (local->xattr_req, GLUSTERFS_PARENT_ENTRYLK, 0); + if (ret < 0) { + gf_log (this->name, GF_LOG_WARNING, + "%s: Unable to set dict value for %s", + loc->path, GLUSTERFS_PARENT_ENTRYLK); + } + + ret = dict_get_ptr (local->xattr_req, "gfid-req", gfid_req); + if (ret) { + gf_log (this->name, GF_LOG_DEBUG, + "%s: failed to get the gfid from dict", loc->path); + *gfid_req = NULL; + } else { + if (loc->parent != NULL) + dict_del (local->xattr_req, "gfid-req"); + } + ret = 0; +out: + return ret; +} + +void +afr_lookup_save_gfid (uuid_t dst, void* new, const loc_t *loc) +{ + inode_t *inode = NULL; + + inode = loc->inode; + if (inode && !uuid_is_null (inode->gfid)) + uuid_copy (dst, inode->gfid); + else if (!uuid_is_null (loc->gfid)) + uuid_copy (dst, loc->gfid); + else if (new && !uuid_is_null (new)) + uuid_copy (dst, new); } int @@ -132,7 +193,7 @@ afr_set_dict_gfid (dict_t *dict, uuid_t gfid) ret = dict_set_dynptr (dict, "gfid-req", pgfid, sizeof (uuid_t)); if (ret) - gf_log (THIS->name, GF_LOG_DEBUG, "gfid set failed"); + gf_log (THIS->name, GF_LOG_ERROR, "gfid set failed"); out: if (ret && pgfid) @@ -141,65 +202,91 @@ 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->mask_type) { - case AFR_ICTX_READ_CHILD_MASK: - fresh_children = params->u.read_ctx.fresh_children; + switch (params->op) { + case AFR_INODE_GET_READ_CTX: + fresh_children = params->u.read_ctx.children; read_child = (int32_t)(ctx->masks & AFR_ICTX_READ_CHILD_MASK); params->u.read_ctx.read_child = read_child; @@ -208,12 +295,13 @@ afr_inode_get_ctx (xlator_t *this, inode_t *inode, afr_inode_params_t *params) for (i = 0; i < priv->child_count; i++) fresh_children[i] = ctx->fresh_children[i]; break; - case AFR_ICTX_OPENDIR_DONE_MASK: - params->u.value = ctx->masks & - AFR_ICTX_OPENDIR_DONE_MASK; + case AFR_INODE_GET_OPENDIR_DONE: + params->u.value = _gf_false; + if (ctx->masks & AFR_ICTX_OPENDIR_DONE_MASK) + params->u.value = _gf_true; break; - case AFR_ICTX_SPLIT_BRAIN_MASK: - params->u.value = ctx->masks & AFR_ICTX_SPLIT_BRAIN_MASK; + default: + GF_ASSERT (0); break; } } @@ -221,14 +309,19 @@ unlock: UNLOCK (&inode->lock); } -uint64_t +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.mask_type = AFR_ICTX_SPLIT_BRAIN_MASK; - 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 @@ -236,37 +329,40 @@ afr_is_opendir_done (xlator_t *this, inode_t *inode) { afr_inode_params_t params = {0}; - params.mask_type = AFR_ICTX_OPENDIR_DONE_MASK; - afr_inode_get_ctx (this, inode, ¶ms); + params.op = AFR_INODE_GET_OPENDIR_DONE; + 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) { afr_inode_params_t params = {0}; - params.mask_type = AFR_ICTX_READ_CHILD_MASK; - params.u.read_ctx.fresh_children = fresh_children; - afr_inode_get_ctx (this, inode, ¶ms); + params.op = AFR_INODE_GET_READ_CTX; + params.u.read_ctx.children = fresh_children; + afr_inode_get_ctx_params (this, inode, ¶ms); return params.u.read_ctx.read_child; } void -afr_inode_ctx_set_read_ctx (afr_inode_ctx_t *ctx, int32_t read_child, - int32_t *fresh_children, int32_t child_count) +afr_inode_ctx_set_read_child (afr_inode_ctx_t *ctx, int32_t read_child) { - uint64_t rest_of_mask = 0; + uint64_t remaining_mask = 0; uint64_t mask = 0; - int i = 0; - rest_of_mask = (~AFR_ICTX_READ_CHILD_MASK & ctx->masks); + remaining_mask = (~AFR_ICTX_READ_CHILD_MASK & ctx->masks); mask = (AFR_ICTX_READ_CHILD_MASK & read_child); - ctx->masks = rest_of_mask | mask; + ctx->masks = remaining_mask | mask; +} - /* avoid memcpy as int, int32_t are used interchangeably - */ +void +afr_inode_ctx_set_read_ctx (afr_inode_ctx_t *ctx, int32_t read_child, + int32_t *fresh_children, int32_t child_count) +{ + int i = 0; + + afr_inode_ctx_set_read_child (ctx, read_child); for (i = 0; i < child_count; i++) { if (fresh_children) ctx->fresh_children[i] = fresh_children[i]; @@ -276,89 +372,92 @@ afr_inode_ctx_set_read_ctx (afr_inode_ctx_t *ctx, int32_t read_child, } void -afr_inode_ctx_set_opendir_done (afr_inode_ctx_t *ctx) +afr_inode_ctx_rm_stale_children (afr_inode_ctx_t *ctx, int32_t *stale_children, + int32_t child_count) { - uint64_t rest_of_mask = 0; - uint64_t mask = 0; + int i = 0; + int32_t read_child = -1; - rest_of_mask = (~AFR_ICTX_OPENDIR_DONE_MASK & ctx->masks); - mask = (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_OPENDIR_DONE_MASK); - ctx->masks = rest_of_mask | mask; + GF_ASSERT (stale_children); + for (i = 0; i < child_count; i++) { + if (stale_children[i] == -1) + break; + afr_children_rm_child (ctx->fresh_children, + stale_children[i], child_count); + } + read_child = (int32_t)(ctx->masks & AFR_ICTX_READ_CHILD_MASK); + if (!afr_is_child_present (ctx->fresh_children, child_count, + read_child)) + afr_inode_ctx_set_read_child (ctx, ctx->fresh_children[0]); } void -afr_inode_ctx_set_splitbrain (afr_inode_ctx_t *ctx, gf_boolean_t set) +afr_inode_ctx_set_opendir_done (afr_inode_ctx_t *ctx) { - uint64_t rest_of_mask = 0; + uint64_t remaining_mask = 0; uint64_t mask = 0; - if (set) { - rest_of_mask = (~AFR_ICTX_SPLIT_BRAIN_MASK & ctx->masks); - mask = (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_SPLIT_BRAIN_MASK); - ctx->masks = rest_of_mask | mask; - } else { - ctx->masks = (~AFR_ICTX_SPLIT_BRAIN_MASK & ctx->masks); - } + remaining_mask = (~AFR_ICTX_OPENDIR_DONE_MASK & ctx->masks); + mask = (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_OPENDIR_DONE_MASK); + ctx->masks = remaining_mask | mask; } void -afr_inode_set_ctx (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; 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->mask_type) { - case AFR_ICTX_READ_CHILD_MASK: + switch (params->op) { + case AFR_INODE_SET_READ_CTX: read_child = params->u.read_ctx.read_child; - fresh_children = params->u.read_ctx.fresh_children; + fresh_children = params->u.read_ctx.children; afr_inode_ctx_set_read_ctx (ctx, read_child, fresh_children, priv->child_count); break; - case AFR_ICTX_OPENDIR_DONE_MASK: + case AFR_INODE_RM_STALE_CHILDREN: + stale_children = params->u.read_ctx.children; + afr_inode_ctx_rm_stale_children (ctx, + stale_children, + priv->child_count); + break; + case AFR_INODE_SET_OPENDIR_DONE: afr_inode_ctx_set_opendir_done (ctx); break; - case AFR_ICTX_SPLIT_BRAIN_MASK: - set = params->u.value; - afr_inode_ctx_set_splitbrain (ctx, set); + 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.mask_type = AFR_ICTX_SPLIT_BRAIN_MASK; - 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 @@ -366,8 +465,8 @@ afr_set_opendir_done (xlator_t *this, inode_t *inode) { afr_inode_params_t params = {0}; - params.mask_type = AFR_ICTX_OPENDIR_DONE_MASK; - afr_inode_set_ctx (this, inode, ¶ms); + params.op = AFR_INODE_SET_OPENDIR_DONE; + afr_inode_set_ctx_params (this, inode, ¶ms); } void @@ -375,14 +474,31 @@ afr_inode_set_read_ctx (xlator_t *this, inode_t *inode, int32_t read_child, int32_t *fresh_children) { afr_inode_params_t params = {0}; + afr_private_t *priv = NULL; + priv = this->private; GF_ASSERT (read_child >= 0); GF_ASSERT (fresh_children); + GF_ASSERT (afr_is_child_present (fresh_children, priv->child_count, + read_child)); - params.mask_type = AFR_ICTX_READ_CHILD_MASK; + params.op = AFR_INODE_SET_READ_CTX; params.u.read_ctx.read_child = read_child; - params.u.read_ctx.fresh_children = fresh_children; - afr_inode_set_ctx (this, inode, ¶ms); + params.u.read_ctx.children = fresh_children; + afr_inode_set_ctx_params (this, inode, ¶ms); +} + +void +afr_inode_rm_stale_children (xlator_t *this, inode_t *inode, + int32_t *stale_children) +{ + afr_inode_params_t params = {0}; + + GF_ASSERT (stale_children); + + params.op = AFR_INODE_RM_STALE_CHILDREN; + params.u.read_ctx.children = stale_children; + afr_inode_set_ctx_params (this, inode, ¶ms); } gf_boolean_t @@ -426,6 +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); @@ -442,29 +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) @@ -484,7 +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; @@ -494,8 +654,11 @@ afr_set_read_ctx_from_policy (xlator_t *this, inode_t *inode, priv->child_count, prev_read_child, config_read_child, - NULL); - afr_inode_set_read_ctx (this, inode, read_child, fresh_children); + NULL, + priv->hash_mode, gfid); + if (read_child >= 0) + afr_inode_set_read_ctx (this, inode, read_child, + fresh_children); } /* afr_next_call_child () @@ -548,8 +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; @@ -598,113 +764,103 @@ 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; afr_private_t *priv = NULL; - int i = 0; - 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->metadata_sh_info && strcmp (sh->metadata_sh_info, "")) + GF_FREE (sh->metadata_sh_info); - if (sh->parentbufs) - GF_FREE (sh->parentbufs); + 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); - if (sh->pending_matrix) { - for (i = 0; i < priv->child_count; i++) { - GF_FREE (sh->pending_matrix[i]); - } - GF_FREE (sh->pending_matrix); - } + afr_matrix_cleanup (sh->pending_matrix, priv->child_count); + afr_matrix_cleanup (sh->delta_matrix, priv->child_count); - if (sh->delta_matrix) { - for (i = 0; i < priv->child_count; i++) { - GF_FREE (sh->delta_matrix[i]); - } - GF_FREE (sh->delta_matrix); - } + GF_FREE (sh->sources); - if (sh->sources) - GF_FREE (sh->sources); + GF_FREE (sh->success); - if (sh->success) - GF_FREE (sh->success); + GF_FREE (sh->locked_nodes); - if (sh->locked_nodes) - GF_FREE (sh->locked_nodes); - - if (sh->healing_fd && !sh->healing_fd_opened) { + 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); + + GF_FREE (sh->checksum); + + GF_FREE (sh->write_needed); + if (sh->healing_fd) + fd_unref (sh->healing_fd); } void afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this) { - int i = 0; - afr_private_t * priv = NULL; + afr_private_t *priv = NULL; + int i = 0; priv = this->private; - for (i = 0; i < priv->child_count; i++) { - if (local->pending && local->pending[i]) - GF_FREE (local->pending[i]); - } - - GF_FREE (local->pending); - - if (local->internal_lock.locked_nodes) - GF_FREE (local->internal_lock.locked_nodes); + afr_matrix_cleanup (local->pending, priv->child_count); + afr_matrix_cleanup (local->transaction.txn_changelog, + priv->child_count); - 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.child_errno); - GF_FREE (local->child_errno); + GF_FREE (local->transaction.pre_op); + GF_FREE (local->transaction.eager_lock); GF_FREE (local->transaction.basename); GF_FREE (local->transaction.new_basename); loc_wipe (&local->transaction.parent_loc); loc_wipe (&local->transaction.new_parent_loc); + + GF_FREE (local->transaction.postop_piggybacked); } @@ -731,11 +887,16 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this) if (local->xattr_req) dict_unref (local->xattr_req); - if (local->child_up) - GF_FREE (local->child_up); + if (local->dict) + dict_unref (local->dict); + + GF_FREE(local->replies); - if (local->fresh_children) - GF_FREE (local->fresh_children); + GF_FREE (local->child_up); + + GF_FREE (local->child_errno); + + GF_FREE (local->fresh_children); { /* lookup */ if (local->cont.lookup.xattrs) { @@ -753,27 +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 */ @@ -807,18 +964,40 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this) dict_unref (local->cont.setxattr.dict); } + { /* fsetxattr */ + if (local->cont.fsetxattr.dict) + dict_unref (local->cont.fsetxattr.dict); + } + { /* removexattr */ GF_FREE (local->cont.removexattr.name); } - + { /* xattrop */ + if (local->cont.xattrop.xattr) + dict_unref (local->cont.xattrop.xattr); + } + { /* fxattrop */ + if (local->cont.fxattrop.xattr) + dict_unref (local->cont.fxattrop.xattr); + } { /* symlink */ GF_FREE (local->cont.symlink.linkpath); } { /* opendir */ - if (local->cont.opendir.checksum) - GF_FREE (local->cont.opendir.checksum); + GF_FREE (local->cont.opendir.checksum); + } + + { /* readdirp */ + if (local->cont.readdir.dict) + dict_unref (local->cont.readdir.dict); } + + if (local->xdata_req) + dict_unref (local->xdata_req); + + if (local->xdata_rsp) + dict_unref (local->xdata_rsp); } @@ -839,23 +1018,41 @@ afr_frame_return (call_frame_t *frame) return call_count; } - -/** - * up_children_count - return the number of children that are up - */ - int -afr_up_children_count (int child_count, unsigned char *child_up) +afr_set_elem_count_get (unsigned char *elems, int child_count) { int i = 0; int ret = 0; for (i = 0; i < child_count; i++) - if (child_up[i]) + if (elems[i]) ret++; return ret; } +/** + * up_children_count - return the number of children that are up + */ + +unsigned int +afr_up_children_count (unsigned char *child_up, unsigned int child_count) +{ + return afr_set_elem_count_get (child_up, child_count); +} + +unsigned int +afr_locked_children_count (unsigned char *children, unsigned int child_count) +{ + return afr_set_elem_count_get (children, child_count); +} + +unsigned int +afr_pre_op_done_children_count (unsigned char *pre_op, + unsigned int child_count) +{ + return afr_set_elem_count_get (pre_op, child_count); +} + gf_boolean_t afr_is_fresh_lookup (loc_t *loc, xlator_t *this) { @@ -883,32 +1080,150 @@ afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *postparent) uuid_copy (loc->pargfid, postparent->ia_gfid); } -void +/* + * 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) { - int32_t read_child = -1; struct iatt *buf = NULL; struct iatt *postparent = NULL; dict_t **xattr = NULL; + int32_t *success_children = NULL; + int32_t *sources = NULL; + afr_private_t *priv = NULL; + int32_t read_child = -1; + int ret = 0; + int i = 0; GF_ASSERT (local); buf = &local->cont.lookup.buf; postparent = &local->cont.lookup.postparent; xattr = &local->cont.lookup.xattr; + priv = this->private; read_child = afr_inode_get_read_ctx (this, local->cont.lookup.inode, - NULL); + local->fresh_children); + if (read_child < 0) { + ret = -1; + goto out; + } + success_children = local->cont.lookup.success_children; + sources = local->cont.lookup.sources; + memset (sources, 0, sizeof (*sources) * priv->child_count); + afr_children_intersection_get (local->fresh_children, success_children, + sources, priv->child_count); + if (!sources[read_child]) { + read_child = -1; + for (i = 0; i < priv->child_count; i++) { + if (sources[i]) { + read_child = i; + break; + } + } + } + if (read_child < 0) { + ret = -1; + goto out; + } + gf_log (this->name, GF_LOG_DEBUG, "Building lookup response from %d", read_child); - *xattr = dict_ref (local->cont.lookup.xattrs[read_child]); + if (!*xattr) + *xattr = dict_ref (local->cont.lookup.xattrs[read_child]); + *buf = local->cont.lookup.bufs[read_child]; *postparent = local->cont.lookup.postparents[read_child]; + if (dict_get (local->xattr_req, QUOTA_SIZE_KEY)) + afr_handle_quota_size (local, this, *xattr); + if (IA_INVAL == local->cont.lookup.inode->ia_type) { /* fix for RT #602 */ local->cont.lookup.inode->ia_type = buf->ia_type; } +out: + return ret; } static void @@ -918,6 +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); @@ -933,99 +1249,164 @@ 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_data_by_xattr (afr_local_t *local, xlator_t *this, - dict_t *xattr) +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.need_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.need_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.need_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.need_metadata_self_heal = _gf_true; + local->self_heal.do_metadata_self_heal = _gf_true; } if (OWNERSHIP_DIFFERS (buf, lookup_buf)) { /* mismatching permissions */ - local->self_heal.need_metadata_self_heal = _gf_true; - gf_log (this->name, GF_LOG_INFO, + local->self_heal.do_metadata_self_heal = _gf_true; + gf_log (this->name, GF_LOG_DEBUG, "ownership differs for %s ", local->loc.path); } if (SIZE_DIFFERS (buf, lookup_buf) && IA_ISREG (buf->ia_type)) { - gf_log (this->name, GF_LOG_INFO, + gf_log (this->name, GF_LOG_DEBUG, "size differs for %s ", local->loc.path); - local->self_heal.need_data_self_heal = _gf_true; + 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_split_brain_status (afr_local_t *local, xlator_t *this) +{ + gf_boolean_t split_brain = _gf_false; + afr_self_heal_t *sh = NULL; + + sh = &local->self_heal; + + split_brain = afr_is_split_brain (this, local->cont.lookup.inode); + split_brain = split_brain || local->cont.lookup.possible_spb; + if ((local->success_count > 0) && split_brain && + IA_ISREG (local->cont.lookup.inode->ia_type)) { + sh->force_confirm_spb = _gf_true; + gf_log (this->name, GF_LOG_DEBUG, + "split brain detected during lookup of %s.", + local->loc.path); + } +} + +static void afr_detect_self_heal_by_lookup_status (afr_local_t *local, xlator_t *this) { GF_ASSERT (local); GF_ASSERT (this); if ((local->success_count > 0) && (local->enoent_count > 0)) { - local->self_heal.need_metadata_self_heal = _gf_true; - local->self_heal.need_data_self_heal = _gf_true; - local->self_heal.need_entry_self_heal = _gf_true; - local->self_heal.need_gfid_self_heal = _gf_true; - local->self_heal.need_missing_entry_self_heal = _gf_true; - gf_log(this->name, GF_LOG_INFO, + local->self_heal.do_metadata_self_heal = _gf_true; + local->self_heal.do_data_self_heal = _gf_true; + local->self_heal.do_entry_self_heal = _gf_true; + local->self_heal.do_gfid_self_heal = _gf_true; + local->self_heal.do_missing_entry_self_heal = _gf_true; + gf_log(this->name, GF_LOG_DEBUG, "entries are missing in lookup of %s.", local->loc.path); - //If all self-heals are needed no need to check for other rules - goto out; } - if ((local->success_count > 0) && - afr_is_split_brain (this, local->cont.lookup.inode) && - IA_ISREG (local->cont.lookup.inode->ia_type)) { - local->self_heal.need_data_self_heal = _gf_true; - local->self_heal.need_gfid_self_heal = _gf_true; - local->self_heal.need_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; } @@ -1035,11 +1416,14 @@ afr_can_self_heal_proceed (afr_self_heal_t *sh, afr_private_t *priv) GF_ASSERT (sh); GF_ASSERT (priv); - return (sh->need_gfid_self_heal - || sh->need_missing_entry_self_heal - || (priv->data_self_heal && sh->need_data_self_heal) - || (priv->metadata_self_heal && sh->need_metadata_self_heal) - || (priv->entry_self_heal && sh->need_entry_self_heal)); + if (sh->force_confirm_spb) + return _gf_true; + return (sh->do_gfid_self_heal + || sh->do_missing_entry_self_heal + || (afr_data_self_heal_enabled (priv->data_self_heal) && + sh->do_data_self_heal) + || (priv->metadata_self_heal && sh->do_metadata_self_heal) + || (priv->entry_self_heal && sh->do_entry_self_heal)); } afr_transaction_type @@ -1066,14 +1450,13 @@ afr_lookup_select_read_child (afr_local_t *local, xlator_t *this, int ret = -1; dict_t **xattrs = NULL; int32_t *success_children = NULL; - struct iatt *bufs = NULL; afr_transaction_type type = AFR_METADATA_TRANSACTION; + uuid_t *gfid = NULL; GF_ASSERT (local); GF_ASSERT (this); GF_ASSERT (local->success_count > 0); - bufs = local->cont.lookup.bufs; success_children = local->cont.lookup.success_children; /*We can take the success_children[0] only because we already *handle the conflicting children other wise, we could select the @@ -1082,8 +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); @@ -1099,29 +1483,32 @@ out: } static inline gf_boolean_t -afr_is_self_heal_running (afr_local_t *local) +afr_is_transaction_running (afr_local_t *local) { - GF_ASSERT (local); + GF_ASSERT (local->fop == GF_FOP_LOOKUP); return ((local->inodelk_count > 0) || (local->entrylk_count > 0)); } -static void +void afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode, - gf_boolean_t is_background, ia_type_t ia_type, + gf_boolean_t background, ia_type_t ia_type, char *reason, void (*gfid_sh_success_cbk) (call_frame_t *sh_frame, xlator_t *this), int (*unwind) (call_frame_t *frame, xlator_t *this, - int32_t op_ret, int32_t op_errno)) + int32_t op_ret, int32_t op_errno, + int32_t sh_failed)) { afr_local_t *local = NULL; char sh_type_str[256] = {0,}; + char *bg = ""; GF_ASSERT (frame); GF_ASSERT (this); GF_ASSERT (inode); + GF_ASSERT (ia_type != IA_INVAL); local = frame->local; - local->self_heal.background = is_background; + local->self_heal.background = background; local->self_heal.type = ia_type; local->self_heal.unwind = unwind; local->self_heal.gfid_sh_success_cbk = gfid_sh_success_cbk; @@ -1130,19 +1517,21 @@ afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode, sh_type_str, sizeof (sh_type_str)); - gf_log (this->name, GF_LOG_INFO, - "background %s self-heal triggered. path: %s", - sh_type_str, local->loc.path); + if (background) + bg = "background"; + gf_log (this->name, GF_LOG_DEBUG, + "%s %s self-heal triggered. path: %s, reason: %s", bg, + sh_type_str, local->loc.path, reason); afr_self_heal (frame, this, inode); } -int +unsigned int afr_gfid_missing_count (const char *xlator_name, int32_t *success_children, struct iatt *bufs, unsigned int child_count, const char *path) { - int gfid_miss_count = 0; + unsigned int gfid_miss_count = 0; int i = 0; struct iatt *child1 = NULL; @@ -1201,7 +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; @@ -1210,7 +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; @@ -1264,8 +1653,27 @@ afr_lookup_conflicting_entries (afr_local_t *local, xlator_t *this) return conflict; } +gf_boolean_t +afr_open_only_data_self_heal (char *data_self_heal) +{ + return !strcmp (data_self_heal, "open"); +} + +gf_boolean_t +afr_data_self_heal_enabled (char *data_self_heal) +{ + gf_boolean_t enabled = _gf_false; + + if (gf_string2boolean (data_self_heal, &enabled) == -1) { + enabled = !strcmp (data_self_heal, "open"); + GF_ASSERT (enabled); + } + + return enabled; +} + static void -afr_lookup_set_self_heal_data (afr_local_t *local, xlator_t *this) +afr_lookup_set_self_heal_params (afr_local_t *local, xlator_t *this) { int i = 0; struct iatt *bufs = NULL; @@ -1273,15 +1681,18 @@ afr_lookup_set_self_heal_data (afr_local_t *local, xlator_t *this) afr_private_t *priv = NULL; int32_t child1 = -1; int32_t child2 = -1; + afr_self_heal_t *sh = NULL; priv = this->private; + sh = &local->self_heal; + afr_detect_self_heal_by_lookup_status (local, this); if (afr_lookup_gfid_missing_count (local, this)) - local->self_heal.need_gfid_self_heal = _gf_true; + local->self_heal.do_gfid_self_heal = _gf_true; if (_gf_true == afr_lookup_conflicting_entries (local, this)) - local->self_heal.need_missing_entry_self_heal = _gf_true; + local->self_heal.do_missing_entry_self_heal = _gf_true; else afr_update_gfid_from_iatts (local->self_heal.sh_gfid_req, local->cont.lookup.bufs, @@ -1299,23 +1710,31 @@ afr_lookup_set_self_heal_data (afr_local_t *local, xlator_t *this) xattr = local->cont.lookup.xattrs; for (i = 0; i < local->success_count; i++) { child1 = local->cont.lookup.success_children[i]; - afr_lookup_set_self_heal_data_by_xattr (local, this, - xattr[child1]); - } + afr_lookup_set_self_heal_params_by_xattr (local, this, + xattr[child1]); + } + if (afr_open_only_data_self_heal (priv->data_self_heal)) + sh->do_data_self_heal = _gf_false; + if (sh->do_metadata_self_heal) + afr_lookup_check_set_metadata_split_brain (local, this); + afr_detect_self_heal_by_split_brain_status (local, this); } int afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this, - int32_t op_ret, int32_t op_errno) + int32_t op_ret, int32_t op_errno, + int32_t sh_failed) { afr_local_t *local = NULL; + int ret = -1; + dict_t *xattr = NULL; local = frame->local; if (op_ret == -1) { local->op_ret = -1; - 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 { @@ -1323,6 +1742,23 @@ afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this, } afr_lookup_done_success_action (frame, this, _gf_true); + xattr = local->cont.lookup.xattr; + if (xattr) { + ret = dict_set_int32 (xattr, "sh-failed", sh_failed); + if (ret) + gf_log (this->name, GF_LOG_ERROR, "%s: Failed to set " + "sh-failed to %d", local->loc.path, sh_failed); + + if (local->self_heal.actual_sh_started == _gf_true && + sh_failed == 0) { + ret = dict_set_int32 (xattr, "actual-sh-done", 1); + if (ret) + gf_log(this->name, GF_LOG_ERROR, "%s: Failed to" + " set actual-sh-done to %d", + local->loc.path, + local->self_heal.actual_sh_started); + } + } out: AFR_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno, local->cont.lookup.inode, &local->cont.lookup.buf, @@ -1374,33 +1810,36 @@ afr_post_gfid_sh_success (call_frame_t *sh_frame, xlator_t *this) } static void -afr_lookup_perform_self_heal_if_needed (call_frame_t *frame, xlator_t *this, - gf_boolean_t *sh_launched) +afr_lookup_perform_self_heal (call_frame_t *frame, xlator_t *this, + gf_boolean_t *sh_launched) { - size_t up_count = 0; + unsigned int up_count = 0; afr_private_t *priv = NULL; afr_local_t *local = NULL; + char *reason = NULL; GF_ASSERT (sh_launched); *sh_launched = _gf_false; priv = this->private; local = frame->local; - up_count = afr_up_children_count (priv->child_count, local->child_up); + up_count = afr_up_children_count (local->child_up, priv->child_count); if (up_count == 1) { gf_log (this->name, GF_LOG_DEBUG, "Only 1 child up - do not attempt to detect self heal"); goto out; } - afr_lookup_set_self_heal_data (local, this); + afr_lookup_set_self_heal_params (local, this); if (afr_can_self_heal_proceed (&local->self_heal, priv)) { - if (afr_is_self_heal_running (local)) + if (afr_is_transaction_running (local) && + (!local->allow_sh_for_running_transaction)) goto out; + reason = "lookup detected pending operations"; afr_launch_self_heal (frame, this, local->cont.lookup.inode, _gf_true, local->cont.lookup.buf.ia_type, - afr_post_gfid_sh_success, + reason, afr_post_gfid_sh_success, afr_self_heal_lookup_unwind); *sh_launched = _gf_true; } @@ -1455,53 +1894,175 @@ 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 (fail_conflict == _gf_false) { + if (afr_lookup_conflicting_entries (local, this)) { + if (fail_conflict == _gf_false) ret = 0; - } else { - local->op_ret = -1; - local->op_errno = EIO; - } goto out; } ret = afr_lookup_select_read_child (local, this, &read_child); - if (ret) { - local->op_ret = -1; - local->op_errno = EIO; - goto out; + if (!afr_is_transaction_running (local) || fresh_lookup) { + if (read_child < 0) + goto out; + + ret = afr_lookup_set_read_ctx (local, this, read_child); + if (ret) + goto out; } - ret = afr_lookup_set_read_ctx (local, this, read_child); + ret = afr_lookup_build_response_params (local, this); + if (ret) + goto out; + afr_update_loc_gfids (&local->loc, + &local->cont.lookup.buf, + &local->cont.lookup.postparent); + + ret = 0; +out: if (ret) { local->op_ret = -1; local->op_errno = EIO; - goto 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; - afr_lookup_build_response_params (local, this); - if (afr_is_fresh_lookup (&local->loc, this)) { - afr_update_loc_gfids (&local->loc, - &local->cont.lookup.buf, - &local->cont.lookup.postparent); + 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; +} - ret = 0; +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 ret; + 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) { @@ -1510,6 +2071,7 @@ afr_lookup_done (call_frame_t *frame, xlator_t *this) afr_local_t *local = NULL; int ret = -1; gf_boolean_t sh_launched = _gf_false; + gf_boolean_t fail_conflict = _gf_false; int gfid_miss_count = 0; int enotconn_count = 0; int up_children_count = 0; @@ -1517,11 +2079,21 @@ 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 (priv->child_count, - local->child_up); + up_children_count = afr_up_children_count (local->child_up, + priv->child_count); enotconn_count = priv->child_count - up_children_count; if ((gfid_miss_count == local->success_count) && (enotconn_count > 0)) { @@ -1533,12 +2105,23 @@ afr_lookup_done (call_frame_t *frame, xlator_t *this) goto unwind; } - ret = afr_lookup_done_success_action (frame, this, _gf_false); + if ((gfid_miss_count == local->success_count) && + uuid_is_null (local->cont.lookup.gfid_req)) { + local->op_ret = -1; + local->op_errno = ENODATA; + gf_log (this->name, GF_LOG_ERROR, "%s: No gfid present", + local->loc.path); + goto unwind; + } + + if (gfid_miss_count && uuid_is_null (local->cont.lookup.gfid_req)) + fail_conflict = _gf_true; + ret = afr_lookup_done_success_action (frame, this, fail_conflict); if (ret) goto unwind; uuid_copy (local->self_heal.sh_gfid_req, local->cont.lookup.gfid_req); - afr_lookup_perform_self_heal_if_needed (frame, this, &sh_launched); + afr_lookup_perform_self_heal (frame, this, &sh_launched); if (sh_launched) { unwind = 0; goto unwind; @@ -1559,24 +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 */ - else if ((old_errno == ENOENT) && (new_errno != ESTALE)) - 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 @@ -1595,8 +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; } @@ -1608,8 +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; @@ -1623,7 +2203,7 @@ afr_set_root_inode_on_first_lookup (afr_local_t *local, xlator_t *this, afr_private_t *priv = NULL; GF_ASSERT (inode); - if (inode->ino != 1) + if (!__is_root_gfid (inode->gfid)) goto out; if (!afr_is_fresh_lookup (&local->loc, this)) goto out; @@ -1656,12 +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; @@ -1674,6 +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++; } @@ -1720,6 +2372,8 @@ afr_lookup_cont_init (afr_local_t *local, unsigned int child_count) int ret = -ENOMEM; struct iatt *iatts = NULL; int32_t *success_children = NULL; + int32_t *sources = NULL; + int32_t **pending_matrix = NULL; GF_ASSERT (local); local->cont.lookup.xattrs = GF_CALLOC (child_count, @@ -1738,15 +2392,25 @@ afr_lookup_cont_init (afr_local_t *local, unsigned int child_count) goto out; local->cont.lookup.bufs = iatts; - success_children = afr_fresh_children_create (child_count); + success_children = afr_children_create (child_count); if (NULL == success_children) goto out; local->cont.lookup.success_children = success_children; - local->fresh_children = afr_fresh_children_create (child_count); + local->fresh_children = afr_children_create (child_count); if (NULL == local->fresh_children) goto out; + sources = GF_CALLOC (sizeof (*sources), child_count, gf_afr_mt_int32_t); + if (NULL == sources) + goto out; + local->cont.lookup.sources = sources; + + pending_matrix = afr_matrix_create (child_count, child_count); + if (NULL == pending_matrix) + goto out; + local->cont.lookup.pending_matrix = pending_matrix; + ret = 0; out: return ret; @@ -1764,42 +2428,55 @@ 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; - ALLOC_OR_GOTO (local, afr_local_t, out); + AFR_LOCAL_ALLOC_OR_GOTO (local, out); local->op_ret = -1; 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; } - if (loc->parent) - local->cont.lookup.parent_ino = loc->parent->ino; - - local->child_up = memdup (priv->child_up, priv->child_count); + local->child_up = memdup (priv->child_up, + sizeof (*local->child_up) * priv->child_count); if (NULL == local->child_up) { op_errno = ENOMEM; goto out; @@ -1811,10 +2488,9 @@ afr_lookup (call_frame_t *frame, xlator_t *this, goto out; } - local->call_count = afr_up_children_count (priv->child_count, - local->child_up); + local->call_count = afr_up_children_count (local->child_up, + priv->child_count); call_count = local->call_count; - if (local->call_count == 0) { ret = -1; op_errno = ENOTCONN; @@ -1824,42 +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; - if (xattr_req == NULL) - local->xattr_req = dict_new (); - else - local->xattr_req = dict_ref (xattr_req); + 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; - afr_xattr_req_prepare (this, local->xattr_req, loc->path); - ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_INODELK_COUNT, 0); - if (ret < 0) { - gf_log (this->name, GF_LOG_WARNING, - "%s: Unable to set dict value for %s", - loc->path, GLUSTERFS_INODELK_COUNT); - } - ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_ENTRYLK_COUNT, 0); - if (ret < 0) { - gf_log (this->name, GF_LOG_WARNING, - "%s: Unable to set dict value for %s", - loc->path, GLUSTERFS_ENTRYLK_COUNT); - } - - ret = dict_get_ptr (xattr_req, "gfid-req", &gfid_req); + ret = afr_lookup_xattr_req_prepare (local, this, xattr_req, &local->loc, + &gfid_req); if (ret) { - gf_log (this->name, GF_LOG_DEBUG, - "failed to get the gfid from dict"); - } else { - uuid_copy (local->cont.lookup.gfid_req, gfid_req); + local->op_errno = -ret; + goto out; + } + afr_lookup_save_gfid (local->cont.lookup.gfid_req, gfid_req, + &local->loc); + local->fop = GF_FOP_LOOKUP; + if (priv->choose_local && !priv->did_discovery) { + if (gfid_req && __is_root_gfid(gfid_req)) { + local->do_discovery = _gf_true; + priv->did_discovery = _gf_true; + } } - if (local->loc.parent != NULL) - dict_del (xattr_req, "gfid-req"); - 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; } @@ -1867,7 +2534,7 @@ afr_lookup (call_frame_t *frame, xlator_t *this, ret = 0; out: - if (ret == -1) + if (ret) AFR_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL); @@ -1878,7 +2545,7 @@ out: /* {{{ open */ int -afr_fd_ctx_set (xlator_t *this, fd_t *fd) +__afr_fd_ctx_set (xlator_t *this, fd_t *fd) { afr_private_t * priv = NULL; int ret = -1; @@ -1890,203 +2557,167 @@ afr_fd_ctx_set (xlator_t *this, fd_t *fd) priv = this->private; - LOCK (&fd->lock); - { - ret = __fd_ctx_get (fd, this, &ctx); + ret = __fd_ctx_get (fd, this, &ctx); - if (ret == 0) - goto unlock; + if (ret == 0) + goto out; - fd_ctx = GF_CALLOC (1, sizeof (afr_fd_ctx_t), - gf_afr_mt_afr_fd_ctx_t); - if (!fd_ctx) { - ret = -ENOMEM; - goto unlock; - } + fd_ctx = GF_CALLOC (1, sizeof (afr_fd_ctx_t), + gf_afr_mt_afr_fd_ctx_t); + if (!fd_ctx) { + ret = -ENOMEM; + goto out; + } - fd_ctx->pre_op_done = GF_CALLOC (sizeof (*fd_ctx->pre_op_done), - priv->child_count, - gf_afr_mt_char); - if (!fd_ctx->pre_op_done) { - ret = -ENOMEM; - goto unlock; - } + fd_ctx->pre_op_done = GF_CALLOC (sizeof (*fd_ctx->pre_op_done), + priv->child_count, + gf_afr_mt_char); + if (!fd_ctx->pre_op_done) { + ret = -ENOMEM; + goto out; + } - fd_ctx->pre_op_piggyback = GF_CALLOC (sizeof (*fd_ctx->pre_op_piggyback), - priv->child_count, - gf_afr_mt_char); - if (!fd_ctx->pre_op_piggyback) { - ret = -ENOMEM; - goto unlock; - } + fd_ctx->pre_op_piggyback = GF_CALLOC (sizeof (*fd_ctx->pre_op_piggyback), + priv->child_count, + gf_afr_mt_char); + if (!fd_ctx->pre_op_piggyback) { + ret = -ENOMEM; + goto out; + } - fd_ctx->opened_on = GF_CALLOC (sizeof (*fd_ctx->opened_on), - priv->child_count, - gf_afr_mt_char); - if (!fd_ctx->opened_on) { - ret = -ENOMEM; - goto unlock; - } + fd_ctx->opened_on = GF_CALLOC (sizeof (*fd_ctx->opened_on), + priv->child_count, + gf_afr_mt_int32_t); + if (!fd_ctx->opened_on) { + ret = -ENOMEM; + goto out; + } - fd_ctx->up_count = priv->up_count; - fd_ctx->down_count = priv->down_count; + fd_ctx->lock_piggyback = GF_CALLOC (sizeof (*fd_ctx->lock_piggyback), + priv->child_count, + gf_afr_mt_char); + if (!fd_ctx->lock_piggyback) { + ret = -ENOMEM; + goto out; + } - fd_ctx->locked_on = GF_CALLOC (sizeof (*fd_ctx->locked_on), - priv->child_count, - gf_afr_mt_char); - if (!fd_ctx->locked_on) { - ret = -ENOMEM; - goto unlock; - } + fd_ctx->lock_acquired = GF_CALLOC (sizeof (*fd_ctx->lock_acquired), + priv->child_count, + gf_afr_mt_char); + if (!fd_ctx->lock_acquired) { + ret = -ENOMEM; + goto out; + } - ret = __fd_ctx_set (fd, this, (uint64_t)(long) fd_ctx); - if (ret) - gf_log (this->name, GF_LOG_DEBUG, - "failed to set fd ctx (%p)", fd); + fd_ctx->up_count = priv->up_count; + fd_ctx->down_count = priv->down_count; - INIT_LIST_HEAD (&fd_ctx->entries); + fd_ctx->locked_on = GF_CALLOC (sizeof (*fd_ctx->locked_on), + priv->child_count, + gf_afr_mt_char); + if (!fd_ctx->locked_on) { + ret = -ENOMEM; + goto out; } -unlock: - UNLOCK (&fd->lock); + + pthread_mutex_init (&fd_ctx->delay_lock, NULL); + INIT_LIST_HEAD (&fd_ctx->entries); + fd_ctx->call_child = -1; + + INIT_LIST_HEAD (&fd_ctx->eager_locked); + + ret = __fd_ctx_set (fd, this, (uint64_t)(long) fd_ctx); + if (ret) + gf_log (this->name, GF_LOG_DEBUG, + "failed to set fd ctx (%p)", fd); out: return ret; } -/* {{{ flush */ int -afr_flush_unwind (call_frame_t *frame, xlator_t *this) +afr_fd_ctx_set (xlator_t *this, fd_t *fd) { - afr_local_t * local = NULL; - call_frame_t *main_frame = NULL; - - local = frame->local; + int ret = -1; - LOCK (&frame->lock); + LOCK (&fd->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); + ret = __afr_fd_ctx_set (this, fd); } + UNLOCK (&fd->lock); - return 0; + return ret; } +/* {{{ flush */ int -afr_flush_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno) +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 (priv->child_count, local->child_up); - - 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); - + 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) +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; - int ret = -1; - int op_ret = -1; - int op_errno = 0; - int call_count = 0; + afr_private_t *priv = NULL; + afr_local_t *local = NULL; + call_stub_t *stub = NULL; + int ret = -1; + int op_errno = 0; VALIDATE_OR_GOTO (frame, out); VALIDATE_OR_GOTO (this, out); @@ -2094,47 +2725,27 @@ afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd) priv = this->private; - ALLOC_OR_GOTO (local, afr_local_t, out); - - ret = AFR_LOCAL_INIT (local, priv); - if (ret < 0) { - op_errno = -ret; - goto out; - } + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - call_count = afr_up_children_count (priv->child_count, local->child_up); + ret = afr_local_init(local, priv, &op_errno); + if (ret < 0) + goto out; - transaction_frame = copy_frame (frame); - if (!transaction_frame) { + local->fd = fd_ref(fd); + stub = fop_flush_stub (frame, afr_flush_wrapper, fd, xdata); + if (!stub) { + ret = -1; op_errno = ENOMEM; goto out; } - transaction_frame->local = local; - - 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); - - local->transaction.main_frame = frame; - local->transaction.start = 0; - local->transaction.len = 0; + afr_delayed_changelog_wake_resume (this, fd, stub); + ret = 0; - afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); - - - op_ret = 0; out: - if (op_ret == -1) { - if (transaction_frame) - AFR_STACK_DESTROY (transaction_frame); - - AFR_STACK_UNWIND (flush, frame, op_ret, op_errno); - } + if (ret < 0) + AFR_STACK_UNWIND(flush, frame, -1, op_errno, NULL); return 0; } @@ -2156,17 +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); + + GF_FREE (fd_ctx->opened_on); - if (fd_ctx->opened_on) - GF_FREE (fd_ctx->opened_on); + GF_FREE (fd_ctx->locked_on); - if (fd_ctx->locked_on) - GF_FREE (fd_ctx->locked_on); + GF_FREE (fd_ctx->pre_op_piggyback); + GF_FREE (fd_ctx->lock_piggyback); - if (fd_ctx->pre_op_piggyback) - GF_FREE (fd_ctx->pre_op_piggyback); + GF_FREE (fd_ctx->lock_acquired); + + pthread_mutex_destroy (&fd_ctx->delay_lock); GF_FREE (fd_ctx); } @@ -2204,14 +2816,25 @@ 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) + struct iatt *postbuf, dict_t *xdata) { afr_local_t *local = NULL; int call_count = -1; int child_index = (long) cookie; int read_child = 0; + call_stub_t *stub = NULL; local = frame->local; @@ -2227,13 +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++; @@ -2246,9 +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); + /* 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; @@ -2257,14 +2903,13 @@ afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, - int32_t datasync) + int32_t datasync, dict_t *xdata) { afr_private_t *priv = NULL; afr_local_t *local = NULL; int ret = -1; int i = 0; int32_t call_count = 0; - int32_t op_ret = -1; int32_t op_errno = 0; VALIDATE_OR_GOTO (frame, out); @@ -2273,19 +2918,20 @@ afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, priv = this->private; - ALLOC_OR_GOTO (local, afr_local_t, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - ret = AFR_LOCAL_INIT (local, priv); - if (ret < 0) { - op_errno = -ret; + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) goto out; - } call_count = local->call_count; - frame->local = local; local->fd = fd_ref (fd); - local->cont.fsync.ino = fd->inode->ino; + + 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]) { @@ -2293,17 +2939,16 @@ afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, (void *) (long) i, priv->children[i], priv->children[i]->fops->fsync, - fd, datasync); + fd, datasync, xdata); if (!--call_count) break; } } - op_ret = 0; + ret = 0; out: - if (op_ret == -1) { - AFR_STACK_UNWIND (fsync, frame, op_ret, op_errno, NULL, NULL); - } + if (ret < 0) + AFR_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL, NULL); return 0; } @@ -2313,7 +2958,8 @@ out: int32_t afr_fsyncdir_cbk (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno) + xlator_t *this, int32_t op_ret, int32_t op_errno, + dict_t *xdata) { afr_local_t *local = NULL; int call_count = -1; @@ -2333,7 +2979,7 @@ afr_fsyncdir_cbk (call_frame_t *frame, void *cookie, if (call_count == 0) AFR_STACK_UNWIND (fsyncdir, frame, local->op_ret, - local->op_errno); + local->op_errno, xdata); return 0; } @@ -2341,14 +2987,13 @@ afr_fsyncdir_cbk (call_frame_t *frame, void *cookie, int32_t afr_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, - int32_t datasync) + int32_t datasync, dict_t *xdata) { afr_private_t *priv = NULL; afr_local_t *local = NULL; int ret = -1; int i = 0; int32_t call_count = 0; - int32_t op_ret = -1; int32_t op_errno = 0; VALIDATE_OR_GOTO (frame, out); @@ -2357,33 +3002,30 @@ afr_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, priv = this->private; - ALLOC_OR_GOTO (local, afr_local_t, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - ret = AFR_LOCAL_INIT (local, priv); - if (ret < 0) { - op_errno = -ret; + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) goto out; - } call_count = local->call_count; - frame->local = local; for (i = 0; i < priv->child_count; i++) { if (local->child_up[i]) { STACK_WIND (frame, afr_fsyncdir_cbk, priv->children[i], priv->children[i]->fops->fsyncdir, - fd, datasync); + fd, datasync, xdata); if (!--call_count) break; } } - op_ret = 0; + ret = 0; out: - if (op_ret == -1) { - AFR_STACK_UNWIND (fsyncdir, frame, op_ret, op_errno); - } + if (ret < 0) + AFR_STACK_UNWIND (fsyncdir, frame, -1, op_errno, NULL); return 0; } @@ -2394,7 +3036,7 @@ out: int32_t afr_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, - dict_t *xattr) + dict_t *xattr, dict_t *xdata) { afr_local_t *local = NULL; int call_count = -1; @@ -2403,8 +3045,11 @@ afr_xattrop_cbk (call_frame_t *frame, void *cookie, LOCK (&frame->lock); { - if (op_ret == 0) + if (op_ret == 0) { + if (!local->cont.xattrop.xattr) + local->cont.xattrop.xattr = dict_ref (xattr); local->op_ret = 0; + } local->op_errno = op_errno; } @@ -2414,7 +3059,7 @@ afr_xattrop_cbk (call_frame_t *frame, void *cookie, if (call_count == 0) AFR_STACK_UNWIND (xattrop, frame, local->op_ret, local->op_errno, - xattr); + local->cont.xattrop.xattr, xdata); return 0; } @@ -2422,14 +3067,13 @@ afr_xattrop_cbk (call_frame_t *frame, void *cookie, int32_t afr_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc, - gf_xattrop_flags_t optype, dict_t *xattr) + gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata) { afr_private_t *priv = NULL; afr_local_t *local = NULL; int ret = -1; int i = 0; int32_t call_count = 0; - int32_t op_ret = -1; int32_t op_errno = 0; VALIDATE_OR_GOTO (frame, out); @@ -2438,33 +3082,30 @@ afr_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc, priv = this->private; - ALLOC_OR_GOTO (local, afr_local_t, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - ret = AFR_LOCAL_INIT (local, priv); - if (ret < 0) { - op_errno = -ret; + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) goto out; - } call_count = local->call_count; - frame->local = local; for (i = 0; i < priv->child_count; i++) { if (local->child_up[i]) { STACK_WIND (frame, afr_xattrop_cbk, priv->children[i], priv->children[i]->fops->xattrop, - loc, optype, xattr); + loc, optype, xattr, xdata); if (!--call_count) break; } } - op_ret = 0; + ret = 0; out: - if (op_ret == -1) { - AFR_STACK_UNWIND (xattrop, frame, op_ret, op_errno, NULL); - } + if (ret < 0) + AFR_STACK_UNWIND (xattrop, frame, -1, op_errno, NULL, NULL); return 0; } @@ -2475,7 +3116,7 @@ out: int32_t afr_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, - dict_t *xattr) + dict_t *xattr, dict_t *xdata) { afr_local_t *local = NULL; @@ -2485,8 +3126,12 @@ afr_fxattrop_cbk (call_frame_t *frame, void *cookie, LOCK (&frame->lock); { - if (op_ret == 0) + if (op_ret == 0) { + if (!local->cont.fxattrop.xattr) + local->cont.fxattrop.xattr = dict_ref (xattr); + local->op_ret = 0; + } local->op_errno = op_errno; } @@ -2496,7 +3141,7 @@ afr_fxattrop_cbk (call_frame_t *frame, void *cookie, if (call_count == 0) AFR_STACK_UNWIND (fxattrop, frame, local->op_ret, local->op_errno, - xattr); + local->cont.fxattrop.xattr, xdata); return 0; } @@ -2504,14 +3149,13 @@ afr_fxattrop_cbk (call_frame_t *frame, void *cookie, int32_t afr_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd, - gf_xattrop_flags_t optype, dict_t *xattr) + gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata) { afr_private_t *priv = NULL; afr_local_t *local = NULL; int ret = -1; int i = 0; int32_t call_count = 0; - int32_t op_ret = -1; int32_t op_errno = 0; VALIDATE_OR_GOTO (frame, out); @@ -2520,33 +3164,30 @@ afr_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd, priv = this->private; - ALLOC_OR_GOTO (local, afr_local_t, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - ret = AFR_LOCAL_INIT (local, priv); - if (ret < 0) { - op_errno = -ret; + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) goto out; - } call_count = local->call_count; - frame->local = local; for (i = 0; i < priv->child_count; i++) { if (local->child_up[i]) { STACK_WIND (frame, afr_fxattrop_cbk, priv->children[i], priv->children[i]->fops->fxattrop, - fd, optype, xattr); + fd, optype, xattr, xdata); if (!--call_count) break; } } - op_ret = 0; + ret = 0; out: - if (op_ret == -1) { - AFR_STACK_UNWIND (fxattrop, frame, op_ret, op_errno, NULL); - } + if (ret < 0) + AFR_STACK_UNWIND (fxattrop, frame, -1, op_errno, NULL, NULL); return 0; } @@ -2555,7 +3196,7 @@ out: int32_t afr_inodelk_cbk (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno) + xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) { afr_local_t *local = NULL; @@ -2576,7 +3217,7 @@ afr_inodelk_cbk (call_frame_t *frame, void *cookie, if (call_count == 0) AFR_STACK_UNWIND (inodelk, frame, local->op_ret, - local->op_errno); + local->op_errno, xdata); return 0; } @@ -2584,14 +3225,14 @@ afr_inodelk_cbk (call_frame_t *frame, void *cookie, int32_t afr_inodelk (call_frame_t *frame, xlator_t *this, - const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *flock) + const char *volume, loc_t *loc, int32_t cmd, + struct gf_flock *flock, dict_t *xdata) { afr_private_t *priv = NULL; afr_local_t *local = NULL; int ret = -1; int i = 0; int32_t call_count = 0; - int32_t op_ret = -1; int32_t op_errno = 0; VALIDATE_OR_GOTO (frame, out); @@ -2600,41 +3241,39 @@ afr_inodelk (call_frame_t *frame, xlator_t *this, priv = this->private; - ALLOC_OR_GOTO (local, afr_local_t, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - ret = AFR_LOCAL_INIT (local, priv); - if (ret < 0) { - op_errno = -ret; + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) goto out; - } call_count = local->call_count; - frame->local = local; for (i = 0; i < priv->child_count; i++) { if (local->child_up[i]) { STACK_WIND (frame, afr_inodelk_cbk, priv->children[i], priv->children[i]->fops->inodelk, - volume, loc, cmd, flock); + volume, loc, cmd, flock, xdata); if (!--call_count) break; } } - op_ret = 0; + ret = 0; out: - if (op_ret == -1) { - AFR_STACK_UNWIND (inodelk, frame, op_ret, op_errno); - } + if (ret < 0) + AFR_STACK_UNWIND (inodelk, frame, -1, op_errno, NULL); return 0; } int32_t afr_finodelk_cbk (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno) + xlator_t *this, int32_t op_ret, int32_t op_errno, + dict_t *xdata) { afr_local_t *local = NULL; @@ -2655,7 +3294,7 @@ afr_finodelk_cbk (call_frame_t *frame, void *cookie, if (call_count == 0) AFR_STACK_UNWIND (finodelk, frame, local->op_ret, - local->op_errno); + local->op_errno, xdata); return 0; } @@ -2663,14 +3302,14 @@ afr_finodelk_cbk (call_frame_t *frame, void *cookie, int32_t afr_finodelk (call_frame_t *frame, xlator_t *this, - const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock) + const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock, + dict_t *xdata) { afr_private_t *priv = NULL; afr_local_t *local = NULL; int ret = -1; int i = 0; int32_t call_count = 0; - int32_t op_ret = -1; int32_t op_errno = 0; VALIDATE_OR_GOTO (frame, out); @@ -2679,42 +3318,38 @@ afr_finodelk (call_frame_t *frame, xlator_t *this, priv = this->private; - ALLOC_OR_GOTO (local, afr_local_t, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - ret = AFR_LOCAL_INIT (local, priv); - if (ret < 0) { - op_errno = -ret; + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) goto out; - } call_count = local->call_count; - frame->local = local; for (i = 0; i < priv->child_count; i++) { if (local->child_up[i]) { STACK_WIND (frame, afr_finodelk_cbk, priv->children[i], priv->children[i]->fops->finodelk, - volume, fd, cmd, flock); + volume, fd, cmd, flock, xdata); if (!--call_count) break; } } - op_ret = 0; + ret = 0; out: - if (op_ret == -1) { - AFR_STACK_UNWIND (finodelk, frame, op_ret, op_errno); - } + if (ret < 0) + AFR_STACK_UNWIND (finodelk, frame, -1, op_errno, NULL); return 0; } int32_t -afr_entrylk_cbk (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno) - +afr_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, dict_t *xdata) { afr_local_t *local = NULL; int call_count = -1; @@ -2734,7 +3369,7 @@ afr_entrylk_cbk (call_frame_t *frame, void *cookie, if (call_count == 0) AFR_STACK_UNWIND (entrylk, frame, local->op_ret, - local->op_errno); + local->op_errno, xdata); return 0; } @@ -2743,14 +3378,14 @@ afr_entrylk_cbk (call_frame_t *frame, void *cookie, int32_t afr_entrylk (call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc, - const char *basename, entrylk_cmd cmd, entrylk_type type) + const char *basename, entrylk_cmd cmd, entrylk_type type, + dict_t *xdata) { afr_private_t *priv = NULL; afr_local_t *local = NULL; int ret = -1; int i = 0; int32_t call_count = 0; - int32_t op_ret = -1; int32_t op_errno = 0; VALIDATE_OR_GOTO (frame, out); @@ -2759,34 +3394,31 @@ afr_entrylk (call_frame_t *frame, xlator_t *this, priv = this->private; - ALLOC_OR_GOTO (local, afr_local_t, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - ret = AFR_LOCAL_INIT (local, priv); - if (ret < 0) { - op_errno = -ret; + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) goto out; - } call_count = local->call_count; - frame->local = local; for (i = 0; i < priv->child_count; i++) { if (local->child_up[i]) { STACK_WIND (frame, afr_entrylk_cbk, priv->children[i], priv->children[i]->fops->entrylk, - volume, loc, basename, cmd, type); + volume, loc, basename, cmd, type, xdata); if (!--call_count) break; } } - op_ret = 0; + ret = 0; out: - if (op_ret == -1) { - AFR_STACK_UNWIND (entrylk, frame, op_ret, op_errno); - } + if (ret < 0) + AFR_STACK_UNWIND (entrylk, frame, -1, op_errno, NULL); return 0; } @@ -2794,7 +3426,7 @@ out: int32_t afr_fentrylk_cbk (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno) + xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) { afr_local_t *local = NULL; @@ -2815,7 +3447,7 @@ afr_fentrylk_cbk (call_frame_t *frame, void *cookie, if (call_count == 0) AFR_STACK_UNWIND (fentrylk, frame, local->op_ret, - local->op_errno); + local->op_errno, xdata); return 0; } @@ -2824,14 +3456,14 @@ afr_fentrylk_cbk (call_frame_t *frame, void *cookie, int32_t afr_fentrylk (call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd, - const char *basename, entrylk_cmd cmd, entrylk_type type) + const char *basename, entrylk_cmd cmd, + entrylk_type type, dict_t *xdata) { afr_private_t *priv = NULL; afr_local_t *local = NULL; int ret = -1; int i = 0; int32_t call_count = 0; - int32_t op_ret = -1; int32_t op_errno = 0; VALIDATE_OR_GOTO (frame, out); @@ -2840,41 +3472,38 @@ afr_fentrylk (call_frame_t *frame, xlator_t *this, priv = this->private; - ALLOC_OR_GOTO (local, afr_local_t, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - ret = AFR_LOCAL_INIT (local, priv); - if (ret < 0) { - op_errno = -ret; + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) goto out; - } call_count = local->call_count; - frame->local = local; for (i = 0; i < priv->child_count; i++) { if (local->child_up[i]) { STACK_WIND (frame, afr_fentrylk_cbk, priv->children[i], priv->children[i]->fops->fentrylk, - volume, fd, basename, cmd, type); + volume, fd, basename, cmd, type, xdata); if (!--call_count) break; } } - op_ret = 0; + ret = 0; out: - if (op_ret == -1) { - AFR_STACK_UNWIND (fentrylk, frame, op_ret, op_errno); - } + if (ret < 0) + AFR_STACK_UNWIND (fentrylk, frame, -1, op_errno, NULL); return 0; } int32_t afr_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, - struct statvfs *statvfs) + struct statvfs *statvfs, dict_t *xdata) { afr_local_t *local = NULL; int call_count = 0; @@ -2905,7 +3534,7 @@ afr_statfs_cbk (call_frame_t *frame, void *cookie, if (call_count == 0) AFR_STACK_UNWIND (statfs, frame, local->op_ret, local->op_errno, - &local->cont.statfs.buf); + &local->cont.statfs.buf, xdata); return 0; } @@ -2913,7 +3542,7 @@ afr_statfs_cbk (call_frame_t *frame, void *cookie, int32_t afr_statfs (call_frame_t *frame, xlator_t *this, - loc_t *loc) + loc_t *loc, dict_t *xdata) { afr_private_t * priv = NULL; int child_count = 0; @@ -2921,7 +3550,6 @@ afr_statfs (call_frame_t *frame, xlator_t *this, int i = 0; int ret = -1; int call_count = 0; - int32_t op_ret = -1; int32_t op_errno = 0; VALIDATE_OR_GOTO (this, out); @@ -2931,15 +3559,13 @@ afr_statfs (call_frame_t *frame, xlator_t *this, priv = this->private; child_count = priv->child_count; - ALLOC_OR_GOTO (local, afr_local_t, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - ret = AFR_LOCAL_INIT (local, priv); - if (ret < 0) { - op_errno = -ret; + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) goto out; - } - frame->local = local; call_count = local->call_count; for (i = 0; i < child_count; i++) { @@ -2947,24 +3573,24 @@ afr_statfs (call_frame_t *frame, xlator_t *this, STACK_WIND (frame, afr_statfs_cbk, priv->children[i], priv->children[i]->fops->statfs, - loc); + loc, xdata); if (!--call_count) break; } } - op_ret = 0; + ret = 0; out: - if (op_ret == -1) { - AFR_STACK_UNWIND (statfs, frame, op_ret, op_errno, NULL); - } + if (ret < 0) + AFR_STACK_UNWIND (statfs, frame, -1, op_errno, NULL, NULL); return 0; } int32_t afr_lk_unlock_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, struct gf_flock *lock) + int32_t op_ret, int32_t op_errno, struct gf_flock *lock, + dict_t *xdata) { afr_local_t * local = NULL; int call_count = -1; @@ -2974,7 +3600,7 @@ afr_lk_unlock_cbk (call_frame_t *frame, void *cookie, xlator_t *this, if (call_count == 0) AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno, - lock); + lock, xdata); return 0; } @@ -2996,7 +3622,7 @@ afr_lk_unlock (call_frame_t *frame, xlator_t *this) if (call_count == 0) { AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno, - &local->cont.lk.ret_flock); + &local->cont.lk.ret_flock, NULL); return 0; } @@ -3010,7 +3636,7 @@ afr_lk_unlock (call_frame_t *frame, xlator_t *this) priv->children[i], priv->children[i]->fops->lk, local->fd, F_SETLK, - &local->cont.lk.user_flock); + &local->cont.lk.user_flock, NULL); if (!--call_count) break; @@ -3023,7 +3649,7 @@ afr_lk_unlock (call_frame_t *frame, xlator_t *this) int32_t afr_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, struct gf_flock *lock) + int32_t op_ret, int32_t op_errno, struct gf_flock *lock, dict_t *xdata) { afr_local_t *local = NULL; afr_private_t *priv = NULL; @@ -3058,12 +3684,12 @@ afr_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, priv->children[child_index], priv->children[child_index]->fops->lk, local->fd, local->cont.lk.cmd, - &local->cont.lk.user_flock); + &local->cont.lk.user_flock, xdata); } else if (local->op_ret == -1) { /* all nodes have gone down */ AFR_STACK_UNWIND (lk, frame, -1, ENOTCONN, - &local->cont.lk.ret_flock); + &local->cont.lk.ret_flock, NULL); } else { /* locking has succeeded on all nodes that are up */ @@ -3081,7 +3707,7 @@ afr_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, */ AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno, - &local->cont.lk.ret_flock); + &local->cont.lk.ret_flock, NULL); } return 0; @@ -3090,13 +3716,13 @@ afr_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int afr_lk (call_frame_t *frame, xlator_t *this, - fd_t *fd, int32_t cmd, struct gf_flock *flock) + fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata) { afr_private_t *priv = NULL; afr_local_t *local = NULL; int i = 0; - int32_t op_ret = -1; int32_t op_errno = 0; + int ret = -1; VALIDATE_OR_GOTO (frame, out); VALIDATE_OR_GOTO (this, out); @@ -3104,10 +3730,12 @@ afr_lk (call_frame_t *frame, xlator_t *this, priv = this->private; - ALLOC_OR_GOTO (local, afr_local_t, out); - AFR_LOCAL_INIT (local, priv); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; - frame->local = local; + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) + goto out; local->cont.lk.locked_nodes = GF_CALLOC (priv->child_count, sizeof (*local->cont.lk.locked_nodes), @@ -3126,13 +3754,12 @@ afr_lk (call_frame_t *frame, xlator_t *this, STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) 0, priv->children[i], priv->children[i]->fops->lk, - fd, cmd, flock); + fd, cmd, flock, xdata); - op_ret = 0; + ret = 0; out: - if (op_ret == -1) { - AFR_STACK_UNWIND (lk, frame, op_ret, op_errno, NULL); - } + if (ret < 0) + AFR_STACK_UNWIND (lk, frame, -1, op_errno, NULL, NULL); return 0; } @@ -3148,8 +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; @@ -3170,41 +3796,23 @@ afr_priv_dump (xlator_t *this) GF_ASSERT (priv); snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s", this->type, this->name); gf_proc_dump_add_section(key_prefix); - gf_proc_dump_build_key(key, key_prefix, "child_count"); - gf_proc_dump_write(key, "%u", priv->child_count); - gf_proc_dump_build_key(key, key_prefix, "read_child_rr"); - gf_proc_dump_write(key, "%u", priv->read_child_rr); + gf_proc_dump_write("child_count", "%u", priv->child_count); + gf_proc_dump_write("read_child_rr", "%u", priv->read_child_rr); for (i = 0; i < priv->child_count; i++) { - gf_proc_dump_build_key(key, key_prefix, "child_up[%d]", i); + sprintf (key, "child_up[%d]", i); gf_proc_dump_write(key, "%d", priv->child_up[i]); - gf_proc_dump_build_key(key, key_prefix, - "pending_key[%d]", i); + sprintf (key, "pending_key[%d]", i); gf_proc_dump_write(key, "%s", priv->pending_key[i]); } - gf_proc_dump_build_key(key, key_prefix, "data_self_heal"); - gf_proc_dump_write(key, "%d", priv->data_self_heal); - gf_proc_dump_build_key(key, key_prefix, "metadata_self_heal"); - gf_proc_dump_write(key, "%d", priv->metadata_self_heal); - gf_proc_dump_build_key(key, key_prefix, "entry_self_heal"); - gf_proc_dump_write(key, "%d", priv->entry_self_heal); - gf_proc_dump_build_key(key, key_prefix, "data_change_log"); - gf_proc_dump_write(key, "%d", priv->data_change_log); - gf_proc_dump_build_key(key, key_prefix, "metadata_change_log"); - gf_proc_dump_write(key, "%d", priv->metadata_change_log); - gf_proc_dump_build_key(key, key_prefix, "entry_change_log"); - gf_proc_dump_write(key, "%d", priv->entry_change_log); - gf_proc_dump_build_key(key, key_prefix, "read_child"); - gf_proc_dump_write(key, "%d", priv->read_child); - gf_proc_dump_build_key(key, key_prefix, "favorite_child"); - gf_proc_dump_write(key, "%d", priv->favorite_child); - gf_proc_dump_build_key(key, key_prefix, "data_lock_server_count"); - gf_proc_dump_write(key, "%u", priv->data_lock_server_count); - gf_proc_dump_build_key(key, key_prefix, "metadata_lock_server_count"); - gf_proc_dump_write(key, "%u", priv->metadata_lock_server_count); - gf_proc_dump_build_key(key, key_prefix, "entry_lock_server_count"); - gf_proc_dump_write(key, "%u", priv->entry_lock_server_count); - gf_proc_dump_build_key(key, key_prefix, "wait_count"); - gf_proc_dump_write(key, "%u", priv->wait_count); + gf_proc_dump_write("data_self_heal", "%s", priv->data_self_heal); + gf_proc_dump_write("metadata_self_heal", "%d", priv->metadata_self_heal); + gf_proc_dump_write("entry_self_heal", "%d", priv->entry_self_heal); + gf_proc_dump_write("data_change_log", "%d", priv->data_change_log); + gf_proc_dump_write("metadata_change_log", "%d", priv->metadata_change_log); + gf_proc_dump_write("entry-change_log", "%d", priv->entry_change_log); + gf_proc_dump_write("read_child", "%d", priv->read_child); + gf_proc_dump_write("favorite_child", "%d", priv->favorite_child); + gf_proc_dump_write("wait_count", "%u", priv->wait_count); return 0; } @@ -3234,24 +3842,35 @@ find_child_index (xlator_t *this, xlator_t *child) int32_t afr_notify (xlator_t *this, int32_t event, - void *data, ...) + void *data, void *data2) { afr_private_t *priv = NULL; int i = -1; int up_children = 0; int down_children = 0; int propagate = 0; - int had_heard_from_all = 0; int have_heard_from_all = 0; int idx = -1; int ret = -1; + int call_psh = 0; + int up_child = AFR_ALL_CHILDREN; + dict_t *input = NULL; + dict_t *output = NULL; priv = this->private; if (!priv) return 0; + /* + * We need to reset this in case children come up in "staggered" + * fashion, so that we discover a late-arriving local subvolume. Note + * that we could end up issuing N lookups to the first subvolume, and + * O(N^2) overall, but N is small for AFR so it shouldn't be an issue. + */ + priv->did_discovery = _gf_false; + had_heard_from_all = 1; for (i = 0; i < priv->child_count; i++) { if (!priv->last_event[i]) { @@ -3262,7 +3881,7 @@ afr_notify (xlator_t *this, int32_t event, /* parent xlators dont need to know about every child_up, child_down * because of afr ha. If all subvolumes go down, child_down has * to be triggered. In that state when 1 subvolume comes up child_up - * needs to be triggered. dht optimises revalidate lookup by sending + * needs to be triggered. dht optimizes revalidate lookup by sending * it only to one of its subvolumes. When child up/down happens * for afr's subvolumes dht should be notified by child_modified. The * subsequent revalidate lookup happens on all the dht's subvolumes @@ -3279,9 +3898,19 @@ afr_notify (xlator_t *this, int32_t event, case GF_EVENT_CHILD_UP: LOCK (&priv->lock); { + /* + * This only really counts if the child was never up + * (value = -1) or had been down (value = 0). See + * comment at GF_EVENT_CHILD_DOWN for a more detailed + * explanation. + */ + if (priv->child_up[idx] != 1) { + priv->up_count++; + } priv->child_up[idx] = 1; - priv->up_count++; + call_psh = 1; + up_child = idx; for (i = 0; i < priv->child_count; i++) if (priv->child_up[i] == 1) up_children++; @@ -3302,8 +3931,22 @@ afr_notify (xlator_t *this, int32_t event, case GF_EVENT_CHILD_DOWN: LOCK (&priv->lock); { + /* + * If a brick is down when we start, we'll get a + * CHILD_DOWN to indicate its initial state. There + * was never a CHILD_UP in this case, so if we + * increment "down_count" the difference between than + * and "up_count" will no longer be the number of + * children that are currently up. This has serious + * implications e.g. for quorum enforcement, so we + * don't increment these values unless the event + * represents an actual state transition between "up" + * (value = 1) and anything else. + */ + if (priv->child_up[idx] == 1) { + priv->down_count++; + } priv->child_up[idx] = 0; - priv->down_count++; for (i = 0; i < priv->child_count; i++) if (priv->child_up[i] == 0) @@ -3328,7 +3971,16 @@ afr_notify (xlator_t *this, int32_t event, priv->last_event[idx] = event; } UNLOCK (&priv->lock); + break; + + case GF_EVENT_TRANSLATOR_OP: + input = data; + output = data2; + ret = afr_xl_op (this, input, output); + goto out; + break; + default: propagate = 1; break; @@ -3355,6 +4007,8 @@ afr_notify (xlator_t *this, int32_t event, LOCK (&priv->lock); { + up_children = afr_up_children_count (priv->child_up, + priv->child_count); for (i = 0; i < priv->child_count; i++) { if (priv->last_event[i] == GF_EVENT_CHILD_UP) { event = GF_EVENT_CHILD_UP; @@ -3374,6 +4028,8 @@ afr_notify (xlator_t *this, int32_t event, ret = 0; if (propagate) ret = default_notify (this, event, data); + if (call_psh && priv->shd.iamshd) + afr_proactive_self_heal ((void*) (long) up_child); out: return ret; @@ -3398,29 +4054,56 @@ afr_first_up_child (unsigned char *child_up, size_t child_count) } int -AFR_LOCAL_INIT (afr_local_t *local, afr_private_t *priv) +afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno) { + int ret = -1; + local->op_ret = -1; local->op_errno = EUCLEAN; - local->call_count = afr_up_children_count (priv->child_count, - priv->child_up); - if (local->call_count == 0) { - gf_log (THIS->name, GF_LOG_INFO, "no subvolumes up"); - return -ENOTCONN; - } - - local->child_up = GF_CALLOC (sizeof (*local->child_up), - priv->child_count, + local->child_up = GF_CALLOC (priv->child_count, + sizeof (*local->child_up), gf_afr_mt_char); if (!local->child_up) { - return -ENOMEM; + if (op_errno) + *op_errno = ENOMEM; + goto out; } memcpy (local->child_up, priv->child_up, sizeof (*local->child_up) * priv->child_count); + local->call_count = afr_up_children_count (local->child_up, + priv->child_count); + if (local->call_count == 0) { + gf_log (THIS->name, GF_LOG_INFO, "no subvolumes up"); + if (op_errno) + *op_errno = ENOTCONN; + goto out; + } - return 0; + local->child_errno = GF_CALLOC (priv->child_count, + sizeof (*local->child_errno), + gf_afr_mt_int32_t); + if (!local->child_errno) { + if (op_errno) + *op_errno = ENOMEM; + goto out; + } + + local->transaction.postop_piggybacked = GF_CALLOC (priv->child_count, + sizeof (int), + gf_afr_mt_int32_t); + if (!local->transaction.postop_piggybacked) { + if (op_errno) + *op_errno = ENOMEM; + goto out; + } + + local->append_write = _gf_false; + + ret = 0; +out: + return ret; } int @@ -3429,16 +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) @@ -3458,56 +4131,118 @@ out: return ret; } +void +afr_matrix_cleanup (int32_t **matrix, unsigned int m) +{ + int i = 0; + + if (!matrix) + goto out; + for (i = 0; i < m; i++) { + GF_FREE (matrix[i]); + } + + GF_FREE (matrix); +out: + return; +} + +int32_t** +afr_matrix_create (unsigned int m, unsigned int n) +{ + int32_t **matrix = NULL; + int i = 0; + + matrix = GF_CALLOC (sizeof (*matrix), m, gf_afr_mt_int32_t); + if (!matrix) + goto out; + + for (i = 0; i < m; i++) { + matrix[i] = GF_CALLOC (sizeof (*matrix[i]), n, + gf_afr_mt_int32_t); + if (!matrix[i]) + goto out; + } + return matrix; +out: + afr_matrix_cleanup (matrix, m); + return NULL; +} + +int +afr_inodelk_init (afr_inodelk_t *lk, char *dom, size_t child_count) +{ + int ret = -ENOMEM; + + lk->domain = dom; + lk->locked_nodes = GF_CALLOC (sizeof (*lk->locked_nodes), + child_count, gf_afr_mt_char); + if (NULL == lk->locked_nodes) + goto out; + ret = 0; +out: + return ret; +} + int -afr_transaction_local_init (afr_local_t *local, afr_private_t *priv) +afr_transaction_local_init (afr_local_t *local, xlator_t *this) { - int i; - int child_up_count = 0; - int ret = -ENOMEM; + int child_up_count = 0; + int ret = -ENOMEM; + afr_private_t *priv = NULL; + priv = this->private; ret = afr_internal_lock_init (&local->internal_lock, priv->child_count, AFR_TRANSACTION_LK); if (ret < 0) goto out; + if ((local->transaction.type == AFR_DATA_TRANSACTION) || + (local->transaction.type == AFR_METADATA_TRANSACTION)) { + ret = afr_inodelk_init (&local->internal_lock.inodelk[0], + this->name, priv->child_count); + if (ret < 0) + goto out; + } + ret = -ENOMEM; - child_up_count = afr_up_children_count (priv->child_count, local->child_up); + child_up_count = afr_up_children_count (local->child_up, + priv->child_count); if (priv->optimistic_change_log && child_up_count == priv->child_count) local->optimistic_change_log = 1; local->first_up_child = afr_first_up_child (local->child_up, priv->child_count); - local->child_errno = GF_CALLOC (sizeof (*local->child_errno), - priv->child_count, - gf_afr_mt_int32_t); - if (!local->child_errno) + local->transaction.eager_lock = + GF_CALLOC (sizeof (*local->transaction.eager_lock), + priv->child_count, + gf_afr_mt_int32_t); + + if (!local->transaction.eager_lock) goto out; - local->pending = GF_CALLOC (sizeof (*local->pending), - priv->child_count, - gf_afr_mt_int32_t); + local->fresh_children = afr_children_create (priv->child_count); + if (!local->fresh_children) + goto out; - if (!local->pending) + local->transaction.pre_op = GF_CALLOC (sizeof (*local->transaction.pre_op), + priv->child_count, + gf_afr_mt_char); + if (!local->transaction.pre_op) goto out; - local->fresh_children = afr_fresh_children_create (priv->child_count); - if (!local->fresh_children) + local->pending = afr_matrix_create (priv->child_count, + AFR_NUM_CHANGE_LOGS); + if (!local->pending) goto out; - for (i = 0; i < priv->child_count; i++) { - local->pending[i] = GF_CALLOC (sizeof (*local->pending[i]), - 3, /* data + metadata + entry */ - gf_afr_mt_int32_t); - if (!local->pending[i]) - goto out; - } + local->transaction.txn_changelog = afr_matrix_create (priv->child_count, + AFR_NUM_CHANGE_LOGS); + if (!local->transaction.txn_changelog) + 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; + INIT_LIST_HEAD (&local->transaction.eager_locked); ret = 0; out: @@ -3523,54 +4258,334 @@ afr_reset_children (int32_t *fresh_children, int32_t child_count) } int32_t* -afr_fresh_children_create (int32_t child_count) +afr_children_create (int32_t child_count) { - int32_t *fresh_children = NULL; + int32_t *children = NULL; int i = 0; GF_ASSERT (child_count > 0); - fresh_children = GF_CALLOC (child_count, sizeof (*fresh_children), - gf_afr_mt_int32_t); - if (NULL == fresh_children) + children = GF_CALLOC (child_count, sizeof (*children), + gf_afr_mt_int32_t); + if (NULL == children) goto out; for (i = 0; i < child_count; i++) - fresh_children[i] = -1; + children[i] = -1; out: - return fresh_children; + return children; } void -afr_fresh_children_add_child (int32_t *fresh_children, int32_t child, - int32_t child_count) +afr_children_add_child (int32_t *children, int32_t child, + int32_t child_count) { gf_boolean_t child_found = _gf_false; int i = 0; for (i = 0; i < child_count; i++) { - if (fresh_children[i] == -1) + if (children[i] == -1) break; - if (fresh_children[i] == child) { + if (children[i] == child) { child_found = _gf_true; break; } } + if (!child_found) { GF_ASSERT (i < child_count); - fresh_children[i] = child; + children[i] = child; + } +} + +void +afr_children_rm_child (int32_t *children, int32_t child, int32_t child_count) +{ + int i = 0; + + GF_ASSERT ((child >= 0) && (child < child_count)); + for (i = 0; i < child_count; i++) { + if (children[i] == -1) + break; + if (children[i] == child) { + if (i != (child_count - 1)) + memmove (children + i, children + i + 1, + sizeof (*children)*(child_count - i - 1)); + children[child_count - 1] = -1; + break; + } } } int -afr_get_children_count (int32_t *fresh_children, unsigned int child_count) +afr_get_children_count (int32_t *children, unsigned int child_count) { int count = 0; int i = 0; for (i = 0; i < child_count; i++) { - if (fresh_children[i] == -1) + if (children[i] == -1) break; count++; } return count; } + +void +afr_set_low_priority (call_frame_t *frame) +{ + frame->root->pid = LOW_PRIO_PROC_PID; +} + +int +afr_child_fd_ctx_set (xlator_t *this, fd_t *fd, int32_t child, + int flags) +{ + int ret = 0; + uint64_t ctx = 0; + afr_fd_ctx_t *fd_ctx = NULL; + + GF_ASSERT (fd && fd->inode); + ret = afr_fd_ctx_set (this, fd); + if (ret < 0) { + gf_log (this->name, GF_LOG_ERROR, + "could not set fd ctx for fd=%p", fd); + goto out; + } + + ret = fd_ctx_get (fd, this, &ctx); + if (ret < 0) { + gf_log (this->name, GF_LOG_ERROR, + "could not get fd ctx for fd=%p", fd); + goto out; + } + + fd_ctx = (afr_fd_ctx_t *)(long) ctx; + fd_ctx->opened_on[child] = AFR_FD_OPENED; + if (!IA_ISDIR (fd->inode->ia_type)) { + fd_ctx->flags = flags; + } + ret = 0; +out: + return ret; +} + +gf_boolean_t +afr_have_quorum (char *logname, afr_private_t *priv) +{ + unsigned int quorum = 0; + + GF_VALIDATE_OR_GOTO(logname,priv,out); + + quorum = priv->quorum_count; + if (quorum != AFR_QUORUM_AUTO) { + return (priv->up_count >= (priv->down_count + quorum)); + } + + quorum = priv->child_count / 2 + 1; + if (priv->up_count >= (priv->down_count + quorum)) { + return _gf_true; + } + + /* + * Special case for even numbers of nodes: if we have exactly half + * and that includes the first ("senior-most") node, then that counts + * as quorum even if it wouldn't otherwise. This supports e.g. N=2 + * while preserving the critical property that there can only be one + * such group. + */ + if ((priv->child_count % 2) == 0) { + quorum = priv->child_count / 2; + if (priv->up_count >= (priv->down_count + quorum)) { + if (priv->child_up[0]) { + return _gf_true; + } + } + } + +out: + return _gf_false; +} + +void +afr_priv_destroy (afr_private_t *priv) +{ + int i = 0; + + if (!priv) + goto out; + inode_unref (priv->root_inode); + GF_FREE (priv->shd.pos); + GF_FREE (priv->shd.pending); + GF_FREE (priv->shd.inprogress); +// for (i = 0; i < priv->child_count; i++) +// if (priv->shd.timer && priv->shd.timer[i]) +// gf_timer_call_cancel (this->ctx, priv->shd.timer[i]); + GF_FREE (priv->shd.timer); + + if (priv->shd.healed) + eh_destroy (priv->shd.healed); + + if (priv->shd.heal_failed) + eh_destroy (priv->shd.heal_failed); + + if (priv->shd.split_brain) + eh_destroy (priv->shd.split_brain); + + for (i = 0; i < priv->child_count; i++) + { + if (priv->shd.statistics[i]) + eh_destroy (priv->shd.statistics[i]); + } + + GF_FREE (priv->shd.statistics); + + GF_FREE (priv->shd.crawl_events); + + GF_FREE (priv->last_event); + if (priv->pending_key) { + for (i = 0; i < priv->child_count; i++) + GF_FREE (priv->pending_key[i]); + } + GF_FREE (priv->pending_key); + GF_FREE (priv->children); + GF_FREE (priv->child_up); + LOCK_DESTROY (&priv->lock); + LOCK_DESTROY (&priv->read_child_lock); + pthread_mutex_destroy (&priv->mutex); + GF_FREE (priv); +out: + return; +} + +int +xlator_subvolume_count (xlator_t *this) +{ + int i = 0; + xlator_list_t *list = NULL; + + for (list = this->children; list; list = list->next) + i++; + return i; +} + +inline gf_boolean_t +afr_is_errno_set (int *child_errno, int child) +{ + return child_errno[child]; +} + +inline gf_boolean_t +afr_is_errno_unset (int *child_errno, int child) +{ + return !afr_is_errno_set (child_errno, child); +} + +void +afr_prepare_new_entry_pending_matrix (int32_t **pending, + gf_boolean_t (*is_pending) (int *, int), + int *ctx, struct iatt *buf, + unsigned int child_count) +{ + int midx = 0; + int idx = 0; + int i = 0; + + midx = afr_index_for_transaction_type (AFR_METADATA_TRANSACTION); + if (IA_ISDIR (buf->ia_type)) + idx = afr_index_for_transaction_type (AFR_ENTRY_TRANSACTION); + else if (IA_ISREG (buf->ia_type)) + idx = afr_index_for_transaction_type (AFR_DATA_TRANSACTION); + else + idx = -1; + for (i = 0; i < child_count; i++) { + if (is_pending (ctx, i)) { + pending[i][midx] = hton32 (1); + if (idx == -1) + continue; + pending[i][idx] = hton32 (1); + } + } +} + +gf_boolean_t +afr_is_fd_fixable (fd_t *fd) +{ + if (!fd || !fd->inode) + return _gf_false; + else if (fd_is_anonymous (fd)) + return _gf_false; + else if (uuid_is_null (fd->inode->gfid)) + return _gf_false; + + return _gf_true; +} + +void +afr_handle_open_fd_count (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + inode_t *inode = NULL; + afr_inode_ctx_t *ctx = NULL; + + local = frame->local; + + if (local->fd) + inode = local->fd->inode; + else + inode = local->loc.inode; + + if (!inode) + return; + + LOCK (&inode->lock); + { + ctx = __afr_inode_ctx_get (inode, this); + ctx->open_fd_count = local->open_fd_count; + } + UNLOCK (&inode->lock); +} + +int +afr_initialise_statistics (xlator_t *this) +{ + afr_private_t *priv = NULL; + int ret = -1; + int i = 0; + int child_count = 0; + eh_t *stats_per_brick = NULL; + shd_crawl_event_t ***shd_crawl_events = NULL; + priv = this->private; + + priv->shd.statistics = GF_CALLOC (sizeof(eh_t *), priv->child_count, + gf_common_mt_eh_t); + if (!priv->shd.statistics) { + ret = -1; + goto out; + } + child_count = priv->child_count; + for (i=0; i < child_count ; i++) { + stats_per_brick = eh_new (AFR_STATISTICS_HISTORY_SIZE, + _gf_false, + _destroy_crawl_event_data); + if (!stats_per_brick) { + ret = -1; + goto out; + } + priv->shd.statistics[i] = stats_per_brick; + + } + + shd_crawl_events = (shd_crawl_event_t***)(&priv->shd.crawl_events); + *shd_crawl_events = GF_CALLOC (sizeof(shd_crawl_event_t*), + priv->child_count, + gf_afr_mt_shd_crawl_event_t); + + if (!priv->shd.crawl_events) { + ret = -1; + goto out; + } + ret = 0; +out: + return ret; + +} |