diff options
Diffstat (limited to 'xlators')
-rw-r--r-- | xlators/cluster/dht/src/dht-rename.c | 21 | ||||
-rw-r--r-- | xlators/features/shard/src/shard.c | 343 | ||||
-rw-r--r-- | xlators/storage/posix/src/posix.c | 99 |
3 files changed, 299 insertions, 164 deletions
diff --git a/xlators/cluster/dht/src/dht-rename.c b/xlators/cluster/dht/src/dht-rename.c index f4f834c8342..777c63de685 100644 --- a/xlators/cluster/dht/src/dht-rename.c +++ b/xlators/cluster/dht/src/dht-rename.c @@ -561,7 +561,7 @@ dht_rename_unlock_cbk (call_frame_t *frame, void *cookie, DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno, &local->stbuf, &local->preoldparent, &local->postoldparent, &local->preparent, - &local->postparent, NULL); + &local->postparent, local->xattr); return 0; } @@ -872,6 +872,12 @@ dht_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this, uuid_utoa(local->loc.inode->gfid):""); } } + if (xdata) { + if (!local->xattr) + local->xattr = dict_ref (xdata); + else + local->xattr = dict_copy_with_ref (xdata, local->xattr); + } if ((src_cached == dst_cached) && (dst_hashed != dst_cached)) { link_frame = copy_frame (frame); @@ -1022,7 +1028,6 @@ dht_do_rename (call_frame_t *frame) xlator_t *dst_cached = NULL; xlator_t *this = NULL; xlator_t *rename_subvol = NULL; - dict_t *dict = NULL; local = frame->local; this = frame->this; @@ -1037,11 +1042,12 @@ dht_do_rename (call_frame_t *frame) rename_subvol = dst_hashed; if ((src_cached != dst_hashed) && (rename_subvol == dst_hashed)) { - DHT_MARKER_DONT_ACCOUNT(dict); + DHT_MARKER_DONT_ACCOUNT(local->xattr_req); } if (rename_subvol == src_cached) { - DHT_CHANGELOG_TRACK_AS_RENAME(dict, &local->loc, &local->loc2); + DHT_CHANGELOG_TRACK_AS_RENAME(local->xattr_req, &local->loc, + &local->loc2); } gf_msg_trace (this->name, 0, @@ -1052,10 +1058,7 @@ dht_do_rename (call_frame_t *frame) FRAME_SU_DO (frame, dht_local_t); STACK_WIND (frame, dht_rename_cbk, rename_subvol, rename_subvol->fops->rename, - &local->loc, &local->loc2, dict); - if (dict) - dict_unref (dict); - + &local->loc, &local->loc2, local->xattr_req); return 0; } @@ -1548,6 +1551,8 @@ dht_rename (call_frame_t *frame, xlator_t *this, local->src_cached = src_cached; local->dst_hashed = dst_hashed; local->dst_cached = dst_cached; + if (xdata) + local->xattr_req = dict_ref (xdata); gf_msg (this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_INFO, diff --git a/xlators/features/shard/src/shard.c b/xlators/features/shard/src/shard.c index 154411682f3..dfb0374c13e 100644 --- a/xlators/features/shard/src/shard.c +++ b/xlators/features/shard/src/shard.c @@ -2135,30 +2135,178 @@ err: } int -shard_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, struct iatt *preparent, - struct iatt *postparent, dict_t *xdata) +shard_unlink_shards_do (call_frame_t *frame, xlator_t *this, inode_t *inode); + +int +shard_post_lookup_shards_unlink_handler (call_frame_t *frame, xlator_t *this) { - SHARD_STACK_UNWIND (unlink, frame, op_ret, op_errno, preparent, - postparent, xdata); + shard_local_t *local = NULL; + + local = frame->local; + if ((local->op_ret < 0) && (local->op_errno != ENOENT)) { + if (local->fop == GF_FOP_UNLINK) + SHARD_STACK_UNWIND (unlink, frame, local->op_ret, + local->op_errno, NULL, NULL, NULL); + else + SHARD_STACK_UNWIND (rename, frame, local->op_ret, + local->op_errno, NULL, NULL, NULL, + NULL, NULL, NULL); + return 0; + } + local->op_ret = 0; + local->op_errno = 0; + + shard_unlink_shards_do (frame, this, + (local->fop == GF_FOP_RENAME) + ? local->loc2.inode + : local->loc.inode); return 0; } int -shard_unlink_base_file (call_frame_t *frame, xlator_t *this) +shard_rename_cbk (call_frame_t *frame, xlator_t *this); + +int32_t +shard_unlink_cbk (call_frame_t *frame, xlator_t *this); + +int +shard_post_resolve_unlink_handler (call_frame_t *frame, xlator_t *this) { shard_local_t *local = NULL; local = frame->local; if (local->op_ret < 0) { - shard_unlink_cbk (frame, 0, this, local->op_ret, - local->op_errno, NULL, NULL, NULL); + if (local->op_errno == ENOENT) { + /* If lookup on /.shard fails with ENOENT, it probably + * means that the file is being unlinked before it + * could grow beyond its first block. In this case, + * unlink boils down to unlinking the base file and + * unwinding the call. + */ + local->op_ret = 0; + local->first_block = local->last_block = 0; + local->num_blocks = 1; + if (local->fop == GF_FOP_UNLINK) + shard_unlink_cbk (frame, this); + else + shard_rename_cbk (frame, this); + return 0; + } else { + if (local->fop == GF_FOP_UNLINK) + SHARD_STACK_UNWIND (unlink, frame, + local->op_ret, + local->op_errno, NULL, NULL, + NULL); + else + shard_rename_cbk (frame, this); + return 0; + } + } + + if (!local->call_count) + shard_unlink_shards_do (frame, this, + (local->fop == GF_FOP_RENAME) + ? local->loc2.inode + : local->loc.inode); + else + shard_common_lookup_shards (frame, this, + (local->fop == GF_FOP_RENAME) + ? local->loc2.inode + : local->loc.inode, + shard_post_lookup_shards_unlink_handler); + return 0; +} + +int +shard_unlink_base_file_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, + struct iatt *preparent, struct iatt *postparent, + dict_t *xdata) +{ + int ret = 0; + uint32_t link_count = 0; + shard_local_t *local = NULL; + shard_priv_t *priv = NULL; + + local = frame->local; + priv = this->private; + + if (op_ret < 0) { + SHARD_STACK_UNWIND (unlink, frame, op_ret, op_errno, NULL, NULL, + NULL); return 0; } - STACK_WIND (frame, shard_unlink_cbk, FIRST_CHILD(this), + /* Because link() does not create links for all but the + * base shard, unlink() must delete these shards only when the + * link count is 1. We can return safely now. + */ + if ((xdata) && (!dict_get_uint32 (xdata, GET_LINK_COUNT, &link_count)) + && (link_count > 1)) + goto unwind; + + local->first_block = get_lowest_block (0, local->block_size); + local->last_block = get_highest_block (0, local->prebuf.ia_size, + local->block_size); + local->num_blocks = local->last_block - local->first_block + 1; + + /* num_blocks = 1 implies that the file has not crossed its + * shard block size. So unlink boils down to unlinking just the + * base file. We can safely return now. + */ + if (local->num_blocks == 1) + goto unwind; + + local->inode_list = GF_CALLOC (local->num_blocks, sizeof (inode_t *), + gf_shard_mt_inode_list); + if (!local->inode_list) + goto unwind; + + /* Save the xdata and preparent and postparent iatts now. This will be + * used at the time of unwinding the call to the parent xl. + */ + local->preoldparent = *preparent; + local->postoldparent = *postparent; + if (xdata) + local->xattr_rsp = dict_ref (xdata); + + local->dot_shard_loc.inode = inode_find (this->itable, + priv->dot_shard_gfid); + if (!local->dot_shard_loc.inode) { + ret = shard_init_dot_shard_loc (this, local); + if (ret) + goto unwind; + shard_lookup_dot_shard (frame, this, + shard_post_resolve_unlink_handler); + } else { + shard_common_resolve_shards (frame, this, local->loc.inode, + shard_post_resolve_unlink_handler); + } + + return 0; + +unwind: + SHARD_STACK_UNWIND (unlink, frame, op_ret, op_errno, preparent, + postparent, xdata); + return 0; +} + +int +shard_unlink_base_file (call_frame_t *frame, xlator_t *this) +{ + shard_local_t *local = NULL; + + local = frame->local; + + if (dict_set_uint32 (local->xattr_req, GET_LINK_COUNT, 0)) + gf_msg (this->name, GF_LOG_WARNING, 0, + SHARD_MSG_DICT_SET_FAILED, "Failed to set " + GET_LINK_COUNT" in dict"); + + /* To-Do: Request open-fd count on base file */ + STACK_WIND (frame, shard_unlink_base_file_cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->unlink, &local->loc, local->xflag, local->xattr_req); return 0; @@ -2199,6 +2347,17 @@ shard_unlink_block_inode (shard_local_t *local, int shard_block_num) int shard_rename_cbk (call_frame_t *frame, xlator_t *this); +int32_t +shard_unlink_cbk (call_frame_t *frame, xlator_t *this) +{ + shard_local_t *local = frame->local; + + SHARD_STACK_UNWIND (unlink, frame, local->op_ret, local->op_errno, + &local->preoldparent, &local->postoldparent, + local->xattr_rsp); + return 0; +} + int shard_unlink_shards_do_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, @@ -2225,7 +2384,7 @@ done: SHARD_UNSET_ROOT_FS_ID (frame, local); if (local->fop == GF_FOP_UNLINK) - shard_unlink_base_file (frame, this); + shard_unlink_cbk (frame, this); else if (local->fop == GF_FOP_RENAME) shard_rename_cbk (frame, this); else @@ -2254,9 +2413,16 @@ shard_unlink_shards_do (call_frame_t *frame, xlator_t *this, inode_t *inode) priv = this->private; local = frame->local; + + /* local->num_blocks includes the base file block. This function only + * deletes the shards under /.shard. So subtract num_blocks by 1. + */ local->call_count = call_count = local->num_blocks - 1; last_block = local->last_block; + /* Ignore the inode associated with the base file and start counting + * from 1. + */ for (i = 1; i < local->num_blocks; i++) { if (!local->inode_list[i]) continue; @@ -2266,20 +2432,15 @@ shard_unlink_shards_do (call_frame_t *frame, xlator_t *this, inode_t *inode) if (!count) { /* callcount = 0 implies that all of the shards that need to be * unlinked are non-existent (in other words the file is full of - * holes). So shard xlator would now proceed to do the final - * unlink on the base file. + * holes). So shard xlator can simply return the fop to its + * parent now. */ gf_msg_debug (this->name, 0, "All shards that need to be " "unlinked are non-existent: %s", uuid_utoa (inode->gfid)); local->num_blocks = 1; if (local->fop == GF_FOP_UNLINK) { - gf_msg_debug (this->name, 0, "Proceeding to unlink the" - " base file"); - STACK_WIND (frame, shard_unlink_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->unlink, - &local->loc, local->flags, - local->xattr_req); + shard_unlink_cbk (frame, this); } else if (local->fop == GF_FOP_RENAME) { gf_msg_debug (this->name, 0, "Resuming rename()"); shard_rename_cbk (frame, this); @@ -2291,6 +2452,8 @@ shard_unlink_shards_do (call_frame_t *frame, xlator_t *this, inode_t *inode) cur_block = 1; SHARD_SET_ROOT_FS_ID (frame, local); + /* Ignore the base file and start iterating from the first block shard. + */ while (cur_block <= last_block) { if (!local->inode_list[cur_block]) { cur_block++; @@ -2347,86 +2510,6 @@ next: } int -shard_post_lookup_shards_unlink_handler (call_frame_t *frame, xlator_t *this) -{ - shard_local_t *local = NULL; - - local = frame->local; - - if ((local->op_ret < 0) && (local->op_errno != ENOENT)) { - if (local->fop == GF_FOP_UNLINK) - SHARD_STACK_UNWIND (unlink, frame, local->op_ret, - local->op_errno, NULL, NULL, NULL); - else - SHARD_STACK_UNWIND (rename, frame, local->op_ret, - local->op_errno, NULL, NULL, NULL, - NULL, NULL, NULL); - return 0; - } - local->op_ret = 0; - local->op_errno = 0; - - shard_unlink_shards_do (frame, this, - (local->fop == GF_FOP_RENAME) - ? local->loc2.inode - : local->loc.inode); - return 0; -} - -int -shard_post_resolve_unlink_handler (call_frame_t *frame, xlator_t *this) -{ - shard_local_t *local = NULL; - - local = frame->local; - - if (local->op_ret < 0) { - if (local->op_errno == ENOENT) { - /* If lookup on /.shard fails with ENOENT, it probably - * means that the file is being unlinked before it - * could grow beyond its first block. In this case, - * unlink boils down to unlinking the base file and - * unwinding the call. - */ - local->op_ret = 0; - local->first_block = local->last_block = 0; - local->num_blocks = 1; - if (local->fop == GF_FOP_UNLINK) - STACK_WIND (frame, shard_unlink_cbk, - FIRST_CHILD(this), - FIRST_CHILD (this)->fops->unlink, - &local->loc, local->xflag, - local->xattr_req); - else - shard_rename_cbk (frame, this); - return 0; - } else { - if (local->fop == GF_FOP_UNLINK) - SHARD_STACK_UNWIND (unlink, frame, - local->op_ret, - local->op_errno, NULL, NULL, - NULL); - else - shard_rename_cbk (frame, this); - return 0; - } - } - - if (!local->call_count) - shard_unlink_shards_do (frame, this, - (local->fop == GF_FOP_RENAME) - ? local->loc2.inode - : local->loc.inode); - else - shard_common_lookup_shards (frame, this, - (local->fop == GF_FOP_RENAME) - ? local->loc2.inode - : local->loc.inode, - shard_post_lookup_shards_unlink_handler); - return 0; -} - -int shard_post_lookup_unlink_handler (call_frame_t *frame, xlator_t *this) { int ret = -1; @@ -2442,46 +2525,7 @@ shard_post_lookup_unlink_handler (call_frame_t *frame, xlator_t *this) return 0; } - local->first_block = get_lowest_block (0, local->block_size); - local->last_block = get_highest_block (0, local->prebuf.ia_size, - local->block_size); - local->num_blocks = local->last_block - local->first_block + 1; - - if ((local->num_blocks == 1) || (local->prebuf.ia_nlink > 1)) { - /* num_blocks = 1 implies that the file has not crossed its - * shard block size. So unlink boils down to unlinking just the - * base file. - * Because link() does not create links for all but the - * base shard, unlink() must delete these shards only when the - * link count is 1. - */ - STACK_WIND (frame, shard_unlink_cbk, FIRST_CHILD (this), - FIRST_CHILD (this)->fops->unlink, &local->loc, - local->xflag, local->xattr_req); - return 0; - } - - local->inode_list = GF_CALLOC (local->num_blocks, sizeof (inode_t *), - gf_shard_mt_inode_list); - if (!local->inode_list) - goto out; - - local->dot_shard_loc.inode = inode_find (this->itable, - priv->dot_shard_gfid); - if (!local->dot_shard_loc.inode) { - ret = shard_init_dot_shard_loc (this, local); - if (ret) - goto out; - shard_lookup_dot_shard (frame, this, - shard_post_resolve_unlink_handler); - } else { - shard_common_resolve_shards (frame, this, local->loc.inode, - shard_post_resolve_unlink_handler); - } - return 0; - -out: - SHARD_STACK_UNWIND (unlink, frame, -1, ENOMEM, NULL, NULL, NULL); + shard_unlink_base_file (frame, this); return 0; } @@ -2524,7 +2568,6 @@ shard_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag, shard_lookup_base_file (frame, this, &local->loc, shard_post_lookup_unlink_handler); - return 0; err: SHARD_STACK_UNWIND (unlink, frame, -1, ENOMEM, NULL, NULL, NULL); @@ -2549,9 +2592,10 @@ shard_rename_cbk (call_frame_t *frame, xlator_t *this) int shard_rename_unlink_dst_shards_do (call_frame_t *frame, xlator_t *this) { - int ret = -1; - shard_local_t *local = NULL; - shard_priv_t *priv = NULL; + int ret = -1; + uint32_t link_count = 0; + shard_local_t *local = NULL; + shard_priv_t *priv = NULL; local = frame->local; priv = this->private; @@ -2561,7 +2605,14 @@ shard_rename_unlink_dst_shards_do (call_frame_t *frame, xlator_t *this) local->dst_block_size); local->num_blocks = local->last_block - local->first_block + 1; - if ((local->num_blocks == 1) || (local->postbuf.ia_nlink > 1)) { + if ((local->xattr_rsp) && + (!dict_get_uint32 (local->xattr_rsp, GET_LINK_COUNT, &link_count)) + && (link_count > 1)) { + shard_rename_cbk (frame, this); + return 0; + } + + if (local->num_blocks == 1) { shard_rename_cbk (frame, this); return 0; } @@ -2664,6 +2715,12 @@ shard_rename_src_base_file (call_frame_t *frame, xlator_t *this) local = frame->local; + if (dict_set_uint32 (local->xattr_req, GET_LINK_COUNT, 0)) + gf_msg (this->name, GF_LOG_WARNING, 0, + SHARD_MSG_DICT_SET_FAILED, "Failed to set " + GET_LINK_COUNT" in dict"); + + /* To-Do: Request open-fd count on dst base file */ STACK_WIND (frame, shard_rename_src_cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->rename, &local->loc, &local->loc2, local->xattr_req); diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix.c index 6cd7df54909..b621580d3e6 100644 --- a/xlators/storage/posix/src/posix.c +++ b/xlators/storage/posix/src/posix.c @@ -1717,13 +1717,13 @@ out: int32_t posix_unlink_gfid_handle_and_entry (xlator_t *this, const char *real_path, struct iatt *stbuf, int32_t *op_errno, - loc_t *loc) + loc_t *loc, gf_boolean_t get_link_count, + dict_t *rsp_dict) { - int32_t ret = 0; - struct posix_private *priv = NULL; - int fd_count = 0; - - priv = this->private; + int fd_count = 0; + int32_t ret = 0; + struct iatt prebuf = {0,}; + gf_boolean_t locked = _gf_false; /* Unlink the gfid_handle_first */ if (stbuf && stbuf->ia_nlink == 1) { @@ -1746,6 +1746,18 @@ posix_unlink_gfid_handle_and_entry (xlator_t *this, const char *real_path, } } + if (get_link_count) { + LOCK (&loc->inode->lock); + locked = _gf_true; + ret = posix_pstat (this, loc->gfid, real_path, &prebuf); + if (ret) { + gf_msg (this->name, GF_LOG_ERROR, errno, + P_MSG_LSTAT_FAILED, "lstat on %s failed", + real_path); + goto err; + } + } + /* Unlink the actual file */ ret = sys_unlink (real_path); if (ret == -1) { @@ -1756,9 +1768,23 @@ posix_unlink_gfid_handle_and_entry (xlator_t *this, const char *real_path, goto err; } + if (locked) { + UNLOCK (&loc->inode->lock); + locked = _gf_false; + } + + ret = dict_set_uint32 (rsp_dict, GET_LINK_COUNT, prebuf.ia_nlink); + if (ret) + gf_msg (this->name, GF_LOG_WARNING, 0, P_MSG_SET_XDATA_FAIL, + "failed to set "GET_LINK_COUNT" for %s", real_path); + return 0; err: + if (locked) { + UNLOCK (&loc->inode->lock); + locked = _gf_false; + } return -1; } @@ -1849,6 +1875,7 @@ posix_unlink (call_frame_t *frame, xlator_t *this, void *uuid = NULL; char uuid_str[GF_UUID_BUF_SIZE] = {0}; char gfid_str[GF_UUID_BUF_SIZE] = {0}; + gf_boolean_t get_link_count = _gf_false; DECLARE_OLD_FS_ID_VAR; @@ -1972,18 +1999,23 @@ posix_unlink (call_frame_t *frame, xlator_t *this, } } - op_ret = posix_unlink_gfid_handle_and_entry (this, real_path, &stbuf, - &op_errno, loc); - if (op_ret == -1) { - goto out; - } - unwind_dict = dict_new (); if (!unwind_dict) { op_errno = -ENOMEM; op_ret = -1; goto out; } + + if (xdata && dict_get (xdata, GET_LINK_COUNT)) + get_link_count = _gf_true; + op_ret = posix_unlink_gfid_handle_and_entry (this, real_path, &stbuf, + &op_errno, loc, + get_link_count, + unwind_dict); + if (op_ret == -1) { + goto out; + } + if (fdstat_requested) { op_ret = posix_fdstat (this, fd, &postbuf); if (op_ret == -1) { @@ -2307,6 +2339,8 @@ posix_rename (call_frame_t *frame, xlator_t *this, char *pgfid_xattr_key = NULL; int32_t nlink_samepgfid = 0; dict_t *unwind_dict = NULL; + gf_boolean_t locked = _gf_false; + gf_boolean_t get_link_count = _gf_false; DECLARE_OLD_FS_ID_VAR; @@ -2333,6 +2367,13 @@ posix_rename (call_frame_t *frame, xlator_t *this, goto out; } + unwind_dict = dict_new (); + if (!unwind_dict) { + op_ret = -1; + op_errno = ENOMEM; + goto out; + } + op_ret = posix_pstat (this, oldloc->pargfid, par_oldpath, &preoldparent); if (op_ret == -1) { op_errno = errno; @@ -2399,6 +2440,22 @@ posix_rename (call_frame_t *frame, xlator_t *this, this, unlock); } + if ((xdata) && (dict_get (xdata, GET_LINK_COUNT)) + && (real_newpath) && (was_present)) { + LOCK (&newloc->inode->lock); + locked = _gf_true; + get_link_count = _gf_true; + op_ret = posix_pstat (this, newloc->gfid, real_newpath, + &stbuf); + if ((op_ret == -1) && (errno != ENOENT)) { + op_errno = errno; + gf_msg (this->name, GF_LOG_ERROR, errno, + P_MSG_LSTAT_FAILED, + "lstat on %s failed", real_newpath); + goto unlock; + } + } + op_ret = sys_rename (real_oldpath, real_newpath); if (op_ret == -1) { op_errno = errno; @@ -2426,6 +2483,18 @@ posix_rename (call_frame_t *frame, xlator_t *this, goto unlock; } + if (locked) { + UNLOCK (&newloc->inode->lock); + locked = _gf_false; + } + + if ((get_link_count) && + (dict_set_uint32 (unwind_dict, GET_LINK_COUNT, + stbuf.ia_nlink))) + gf_msg (this->name, GF_LOG_WARNING, 0, + P_MSG_SET_XDATA_FAIL, "failed to set " + GET_LINK_COUNT" for %s", real_newpath); + if (!IA_ISDIR (oldloc->inode->ia_type) && priv->update_pgfid_nlinks) { MAKE_PGFID_XATTR_KEY (pgfid_xattr_key, @@ -2439,6 +2508,10 @@ posix_rename (call_frame_t *frame, xlator_t *this, } } unlock: + if (locked) { + UNLOCK (&newloc->inode->lock); + locked = _gf_false; + } UNLOCK (&oldloc->inode->lock); if (op_ret < 0) { @@ -2487,7 +2560,7 @@ unlock: } if (was_present) - unwind_dict = posix_dict_set_nlink (xdata, NULL, nlink); + unwind_dict = posix_dict_set_nlink (xdata, unwind_dict, nlink); op_ret = 0; out: |