diff options
Diffstat (limited to 'xlators/cluster')
30 files changed, 2891 insertions, 1095 deletions
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c index 79644b740..af01f2ef2 100644 --- a/xlators/cluster/afr/src/afr-common.c +++ b/xlators/cluster/afr/src/afr-common.c @@ -51,7 +51,7 @@ #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); @@ -782,6 +782,9 @@ afr_local_sh_cleanup (afr_local_t *local, xlator_t *this) if (sh->data_sh_info && strcmp (sh->data_sh_info, "")) GF_FREE (sh->data_sh_info); + if (sh->metadata_sh_info && strcmp (sh->metadata_sh_info, "")) + GF_FREE (sh->metadata_sh_info); + GF_FREE (sh->buf); GF_FREE (sh->parentbufs); @@ -1077,6 +1080,88 @@ afr_update_loc_gfids (loc_t *loc, struct iatt *buf, struct iatt *postparent) uuid_copy (loc->pargfid, postparent->ia_gfid); } +/* + * Quota size xattrs are not maintained by afr. There is a + * possibility that they differ even when both the directory changelog xattrs + * suggest everything is fine. So if there is at least one 'source' check among + * the sources which has the maximum quota size. Otherwise check among all the + * available ones for maximum quota size. This way if there is a source and + * stale copies it always votes for the 'source'. + * */ + +static void +afr_handle_quota_size (afr_local_t *local, xlator_t *this, + dict_t *rsp_dict) +{ + int32_t *sources = NULL; + dict_t *xattr = NULL; + data_t *max_data = NULL; + int64_t max_quota_size = -1; + data_t *data = NULL; + int64_t *size = NULL; + int64_t quota_size = -1; + afr_private_t *priv = NULL; + int i = 0; + int ret = -1; + gf_boolean_t source_present = _gf_false; + + priv = this->private; + sources = local->cont.lookup.sources; + + if (rsp_dict == NULL) { + gf_log_callingfn (this->name, GF_LOG_ERROR, "%s: Invalid " + "response dictionary", local->loc.path); + return; + } + + for (i = 0; i < priv->child_count; i++) { + if (sources[i]) { + source_present = _gf_true; + break; + } + } + + for (i = 0; i < priv->child_count; i++) { + /* + * If there is at least one source lets check + * for maximum quota sizes among sources, otherwise take the + * maximum of the ones present to be on the safer side. + */ + if (source_present && !sources[i]) + continue; + + xattr = local->cont.lookup.xattrs[i]; + if (!xattr) + continue; + + data = dict_get (xattr, QUOTA_SIZE_KEY); + if (!data) + continue; + + size = (int64_t*)data->data; + quota_size = ntoh64(*size); + gf_log (this->name, GF_LOG_DEBUG, "%s: %d, size: %"PRId64, + local->loc.path, i, quota_size); + if (quota_size > max_quota_size) { + if (max_data) + data_unref (max_data); + + max_quota_size = quota_size; + max_data = data_ref (data); + } + } + + if (max_data) { + ret = dict_set (rsp_dict, QUOTA_SIZE_KEY, max_data); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "%s: Failed to set " + "quota size", local->loc.path); + } + + data_unref (max_data); + } +} + int afr_lookup_build_response_params (afr_local_t *local, xlator_t *this) { @@ -1121,13 +1206,18 @@ afr_lookup_build_response_params (afr_local_t *local, xlator_t *this) ret = -1; goto out; } + gf_log (this->name, GF_LOG_DEBUG, "Building lookup response from %d", read_child); if (!*xattr) *xattr = dict_ref (local->cont.lookup.xattrs[read_child]); + *buf = local->cont.lookup.bufs[read_child]; *postparent = local->cont.lookup.postparents[read_child]; + if (dict_get (local->xattr_req, QUOTA_SIZE_KEY)) + afr_handle_quota_size (local, this, *xattr); + if (IA_INVAL == local->cont.lookup.inode->ia_type) { /* fix for RT #602 */ local->cont.lookup.inode->ia_type = buf->ia_type; @@ -1742,7 +1832,8 @@ afr_lookup_perform_self_heal (call_frame_t *frame, xlator_t *this, afr_lookup_set_self_heal_params (local, this); if (afr_can_self_heal_proceed (&local->self_heal, priv)) { - if (afr_is_transaction_running (local)) + if (afr_is_transaction_running (local) && + (!local->allow_sh_for_running_transaction)) goto out; reason = "lookup detected pending operations"; @@ -2337,7 +2428,7 @@ afr_lookup (call_frame_t *frame, xlator_t *this, int call_count = 0; uint64_t ctx = 0; int32_t op_errno = 0; - + int allow_sh = 0; priv = this->private; AFR_LOCAL_ALLOC_OR_GOTO (local, out); @@ -2409,6 +2500,11 @@ 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; + ret = dict_get_int32 (xattr_req, "allow-sh-for-running-transaction", + &allow_sh); + dict_del (xattr_req, "allow-sh-for-running-transaction"); + local->allow_sh_for_running_transaction = allow_sh; + ret = afr_lookup_xattr_req_prepare (local, this, xattr_req, &local->loc, &gfid_req); if (ret) { @@ -2592,9 +2688,11 @@ afr_flush_wrapper (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata) int i = 0; afr_local_t *local = NULL; afr_private_t *priv = NULL; + int call_count = -1; priv = this->private; local = frame->local; + call_count = local->call_count; for (i = 0; i < priv->child_count; i++) { if (local->child_up[i]) { @@ -2603,6 +2701,9 @@ afr_flush_wrapper (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata) priv->children[i], priv->children[i]->fops->flush, local->fd, NULL); + if (!--call_count) + break; + } } @@ -2749,13 +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++; @@ -2779,8 +2880,9 @@ afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, */ stub = fop_fsync_cbk_stub (frame, afr_fsync_unwind_cbk, local->op_ret, local->op_errno, - &local->cont.fsync.prebuf, - &local->cont.fsync.postbuf, xdata); + &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; @@ -2852,491 +2954,6 @@ out: /* }}} */ -/* {{{ fallocate */ - -static int -afr_fallocate_unwind (call_frame_t *frame, xlator_t *this) -{ - afr_local_t * local = NULL; - call_frame_t *main_frame = NULL; - - local = frame->local; - - LOCK (&frame->lock); - { - if (local->transaction.main_frame) - main_frame = local->transaction.main_frame; - local->transaction.main_frame = NULL; - } - UNLOCK (&frame->lock); - - if (main_frame) { - AFR_STACK_UNWIND (fallocate, main_frame, local->op_ret, - local->op_errno, - &local->cont.fallocate.prebuf, - &local->cont.fallocate.postbuf, - NULL); - } - return 0; -} - -static int -afr_fallocate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, struct iatt *prebuf, - struct iatt *postbuf, dict_t *xdata) -{ - afr_local_t * local = NULL; - afr_private_t * priv = NULL; - int child_index = (long) cookie; - int call_count = -1; - int need_unwind = 0; - int read_child = 0; - - local = frame->local; - priv = this->private; - - read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL); - - LOCK (&frame->lock); - { - if (child_index == read_child) { - local->read_child_returned = _gf_true; - } - - if (afr_fop_failed (op_ret, op_errno)) - afr_transaction_fop_failed (frame, this, child_index); - - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - local->cont.fallocate.prebuf = *prebuf; - local->cont.fallocate.postbuf = *postbuf; - } - - if (child_index == read_child) { - local->cont.fallocate.prebuf = *prebuf; - local->cont.fallocate.postbuf = *postbuf; - } - - local->success_count++; - - if ((local->success_count >= priv->wait_count) - && local->read_child_returned) { - need_unwind = 1; - } - } - local->op_errno = op_errno; - } - UNLOCK (&frame->lock); - - if (need_unwind) - local->transaction.unwind (frame, this); - - call_count = afr_frame_return (frame); - - if (call_count == 0) { - local->transaction.resume (frame, this); - } - - return 0; -} - -static int -afr_fallocate_wind (call_frame_t *frame, xlator_t *this) -{ - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - int call_count = -1; - int i = 0; - - local = frame->local; - priv = this->private; - - call_count = afr_pre_op_done_children_count (local->transaction.pre_op, - priv->child_count); - - if (call_count == 0) { - local->transaction.resume (frame, this); - return 0; - } - - local->call_count = call_count; - - for (i = 0; i < priv->child_count; i++) { - if (local->transaction.pre_op[i]) { - STACK_WIND_COOKIE (frame, afr_fallocate_wind_cbk, - (void *) (long) i, - priv->children[i], - priv->children[i]->fops->fallocate, - local->fd, - local->cont.fallocate.mode, - local->cont.fallocate.offset, - local->cont.fallocate.len, - NULL); - - if (!--call_count) - break; - } - } - - return 0; -} - -static int -afr_fallocate_done (call_frame_t *frame, xlator_t *this) -{ - afr_local_t *local = NULL; - - local = frame->local; - - local->transaction.unwind (frame, this); - - AFR_STACK_DESTROY (frame); - - return 0; -} - -static int -afr_do_fallocate(call_frame_t *frame, xlator_t *this) -{ - call_frame_t * transaction_frame = NULL; - afr_local_t * local = NULL; - int op_ret = -1; - int op_errno = 0; - - local = frame->local; - - transaction_frame = copy_frame (frame); - if (!transaction_frame) { - goto out; - } - - transaction_frame->local = local; - frame->local = NULL; - - local->op = GF_FOP_FALLOCATE; - - local->transaction.fop = afr_fallocate_wind; - local->transaction.done = afr_fallocate_done; - local->transaction.unwind = afr_fallocate_unwind; - - local->transaction.main_frame = frame; - - local->transaction.start = local->cont.fallocate.offset; - local->transaction.len = 0; - - /* fallocate can modify the file size */ - op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); - if (op_ret < 0) { - op_errno = -op_ret; - goto out; - } - - op_ret = 0; -out: - if (op_ret < 0) { - if (transaction_frame) - AFR_STACK_DESTROY (transaction_frame); - AFR_STACK_UNWIND (fallocate, frame, op_ret, op_errno, NULL, - NULL, NULL); - } - - return 0; -} - -int -afr_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode, - off_t offset, size_t len, dict_t *xdata) -{ - afr_private_t * priv = NULL; - afr_local_t * local = NULL; - call_frame_t *transaction_frame = NULL; - int ret = -1; - int op_errno = 0; - - VALIDATE_OR_GOTO (frame, out); - VALIDATE_OR_GOTO (this, out); - VALIDATE_OR_GOTO (this->private, out); - - priv = this->private; - - if (afr_is_split_brain (this, fd->inode)) { - op_errno = EIO; - goto out; - } - QUORUM_CHECK(fallocate,out); - - AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); - local = frame->local; - - ret = afr_local_init (local, priv, &op_errno); - if (ret < 0) - goto out; - - local->cont.fallocate.mode = mode; - local->cont.fallocate.offset = offset; - local->cont.fallocate.len = len; - - local->fd = fd_ref (fd); - - afr_open_fd_fix (fd, this); - - afr_do_fallocate (frame, this); - - ret = 0; -out: - if (ret < 0) { - if (transaction_frame) - AFR_STACK_DESTROY (transaction_frame); - AFR_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL); - } - - return 0; -} - -/* }}} */ - -/* {{{ discard */ - -static int -afr_discard_unwind (call_frame_t *frame, xlator_t *this) -{ - afr_local_t * local = NULL; - call_frame_t *main_frame = NULL; - - local = frame->local; - - LOCK (&frame->lock); - { - if (local->transaction.main_frame) - main_frame = local->transaction.main_frame; - local->transaction.main_frame = NULL; - } - UNLOCK (&frame->lock); - - if (main_frame) { - AFR_STACK_UNWIND (discard, main_frame, local->op_ret, - local->op_errno, - &local->cont.discard.prebuf, - &local->cont.discard.postbuf, - NULL); - } - return 0; -} - -static int -afr_discard_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, struct iatt *prebuf, - struct iatt *postbuf, dict_t *xdata) -{ - afr_local_t * local = NULL; - afr_private_t * priv = NULL; - int child_index = (long) cookie; - int call_count = -1; - int need_unwind = 0; - int read_child = 0; - - local = frame->local; - priv = this->private; - - read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL); - - LOCK (&frame->lock); - { - if (child_index == read_child) { - local->read_child_returned = _gf_true; - } - - if (afr_fop_failed (op_ret, op_errno)) - afr_transaction_fop_failed (frame, this, child_index); - - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - local->cont.discard.prebuf = *prebuf; - local->cont.discard.postbuf = *postbuf; - } - - if (child_index == read_child) { - local->cont.discard.prebuf = *prebuf; - local->cont.discard.postbuf = *postbuf; - } - - local->success_count++; - - if ((local->success_count >= priv->wait_count) - && local->read_child_returned) { - need_unwind = 1; - } - } - local->op_errno = op_errno; - } - UNLOCK (&frame->lock); - - if (need_unwind) - local->transaction.unwind (frame, this); - - call_count = afr_frame_return (frame); - - if (call_count == 0) { - local->transaction.resume (frame, this); - } - - return 0; -} - -static int -afr_discard_wind (call_frame_t *frame, xlator_t *this) -{ - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - int call_count = -1; - int i = 0; - - local = frame->local; - priv = this->private; - - call_count = afr_pre_op_done_children_count (local->transaction.pre_op, - priv->child_count); - - if (call_count == 0) { - local->transaction.resume (frame, this); - return 0; - } - - local->call_count = call_count; - - for (i = 0; i < priv->child_count; i++) { - if (local->transaction.pre_op[i]) { - STACK_WIND_COOKIE (frame, afr_discard_wind_cbk, - (void *) (long) i, - priv->children[i], - priv->children[i]->fops->discard, - local->fd, - local->cont.discard.offset, - local->cont.discard.len, - NULL); - - if (!--call_count) - break; - } - } - - return 0; -} - -static int -afr_discard_done (call_frame_t *frame, xlator_t *this) -{ - afr_local_t *local = NULL; - - local = frame->local; - - local->transaction.unwind (frame, this); - - AFR_STACK_DESTROY (frame); - - return 0; -} - -static int -afr_do_discard(call_frame_t *frame, xlator_t *this) -{ - call_frame_t * transaction_frame = NULL; - afr_local_t * local = NULL; - int op_ret = -1; - int op_errno = 0; - - local = frame->local; - - transaction_frame = copy_frame (frame); - if (!transaction_frame) { - goto out; - } - - transaction_frame->local = local; - frame->local = NULL; - - local->op = GF_FOP_DISCARD; - - local->transaction.fop = afr_discard_wind; - local->transaction.done = afr_discard_done; - local->transaction.unwind = afr_discard_unwind; - - local->transaction.main_frame = frame; - - local->transaction.start = local->cont.discard.offset; - local->transaction.len = 0; - - op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); - if (op_ret < 0) { - op_errno = -op_ret; - goto out; - } - - op_ret = 0; -out: - if (op_ret < 0) { - if (transaction_frame) - AFR_STACK_DESTROY (transaction_frame); - AFR_STACK_UNWIND (discard, frame, op_ret, op_errno, NULL, - NULL, NULL); - } - - return 0; -} - -int -afr_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, - size_t len, dict_t *xdata) -{ - afr_private_t * priv = NULL; - afr_local_t * local = NULL; - call_frame_t *transaction_frame = NULL; - int ret = -1; - int op_errno = 0; - - VALIDATE_OR_GOTO (frame, out); - VALIDATE_OR_GOTO (this, out); - VALIDATE_OR_GOTO (this->private, out); - - priv = this->private; - - if (afr_is_split_brain (this, fd->inode)) { - op_errno = EIO; - goto out; - } - QUORUM_CHECK(discard, out); - - AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); - local = frame->local; - - ret = afr_local_init (local, priv, &op_errno); - if (ret < 0) - goto out; - - local->cont.discard.offset = offset; - local->cont.discard.len = len; - - local->fd = fd_ref (fd); - - afr_open_fd_fix (fd, this); - - afr_do_discard(frame, this); - - ret = 0; -out: - if (ret < 0) { - if (transaction_frame) - AFR_STACK_DESTROY (transaction_frame); - AFR_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL); - } - - return 0; -} - -/* }}} */ - /* {{{ fsync */ int32_t @@ -4482,6 +4099,8 @@ afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno) goto out; } + local->append_write = _gf_false; + ret = 0; out: return ret; @@ -4812,6 +4431,16 @@ afr_priv_destroy (afr_private_t *priv) if (priv->shd.split_brain) eh_destroy (priv->shd.split_brain); + for (i = 0; i < priv->child_count; i++) + { + if (priv->shd.statistics[i]) + eh_destroy (priv->shd.statistics[i]); + } + + GF_FREE (priv->shd.statistics); + + GF_FREE (priv->shd.crawl_events); + GF_FREE (priv->last_event); if (priv->pending_key) { for (i = 0; i < priv->child_count; i++) @@ -4915,3 +4544,48 @@ afr_handle_open_fd_count (call_frame_t *frame, xlator_t *this) } UNLOCK (&inode->lock); } + +int +afr_initialise_statistics (xlator_t *this) +{ + afr_private_t *priv = NULL; + int ret = -1; + int i = 0; + int child_count = 0; + eh_t *stats_per_brick = NULL; + shd_crawl_event_t ***shd_crawl_events = NULL; + priv = this->private; + + priv->shd.statistics = GF_CALLOC (sizeof(eh_t *), priv->child_count, + gf_common_mt_eh_t); + if (!priv->shd.statistics) { + ret = -1; + goto out; + } + child_count = priv->child_count; + for (i=0; i < child_count ; i++) { + stats_per_brick = eh_new (AFR_STATISTICS_HISTORY_SIZE, + _gf_false, + _destroy_crawl_event_data); + if (!stats_per_brick) { + ret = -1; + goto out; + } + priv->shd.statistics[i] = stats_per_brick; + + } + + shd_crawl_events = (shd_crawl_event_t***)(&priv->shd.crawl_events); + *shd_crawl_events = GF_CALLOC (sizeof(shd_crawl_event_t*), + priv->child_count, + gf_afr_mt_shd_crawl_event_t); + + if (!priv->shd.crawl_events) { + ret = -1; + goto out; + } + ret = 0; +out: + return ret; + +} diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c index 2bb707853..689dd84e6 100644 --- a/xlators/cluster/afr/src/afr-dir-read.c +++ b/xlators/cluster/afr/src/afr-dir-read.c @@ -253,7 +253,7 @@ unlock: goto out; if (!afr_is_opendir_done (this, local->fd->inode) && - up_children_count > 1) { + up_children_count > 1 && priv->entry_self_heal) { /* * This is the first opendir on this inode. We need @@ -410,9 +410,9 @@ afr_readdir_cbk (call_frame_t *frame, void *cookie, local = frame->local; afr_readdir_filter_trash_dir (entries, local->fd); - AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries, NULL); out: + AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries, NULL); return 0; } @@ -429,9 +429,9 @@ afr_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local = frame->local; afr_readdir_filter_trash_dir (entries, local->fd); - AFR_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries, NULL); out: + AFR_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries, NULL); return 0; } diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c index 5a22696ce..1943b719b 100644 --- a/xlators/cluster/afr/src/afr-dir-write.c +++ b/xlators/cluster/afr/src/afr-dir-write.c @@ -235,26 +235,6 @@ out: } void -afr_dir_fop_handle_all_fop_failures (call_frame_t *frame) -{ - xlator_t *this = NULL; - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - - this = frame->this; - local = frame->local; - priv = this->private; - - if (local->op_ret >= 0) - goto out; - - __mark_all_success (local->pending, priv->child_count, - local->transaction.type); -out: - return; -} - -void afr_dir_fop_done (call_frame_t *frame, xlator_t *this) { afr_local_t *local = NULL; @@ -273,7 +253,6 @@ afr_dir_fop_done (call_frame_t *frame, xlator_t *this) done: local->transaction.unwind (frame, this); afr_dir_fop_mark_entry_pending_changelog (frame, this); - afr_dir_fop_handle_all_fop_failures (frame); local->transaction.resume (frame, this); } diff --git a/xlators/cluster/afr/src/afr-inode-read.c b/xlators/cluster/afr/src/afr-inode-read.c index 00dd22214..e06e3b2f2 100644 --- a/xlators/cluster/afr/src/afr-inode-read.c +++ b/xlators/cluster/afr/src/afr-inode-read.c @@ -117,6 +117,8 @@ afr_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask, children = priv->children; + AFR_SBRAIN_CHECK_LOC (loc, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -232,6 +234,8 @@ afr_stat (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata) children = priv->children; + AFR_SBRAIN_CHECK_LOC (loc, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -348,10 +352,7 @@ afr_fstat (call_frame_t *frame, xlator_t *this, VALIDATE_OR_GOTO (fd->inode, out); - if (afr_is_split_brain (this, fd->inode)) { - op_errno = EIO; - goto out; - } + AFR_SBRAIN_CHECK_FD (fd, out); AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -472,6 +473,8 @@ afr_readlink (call_frame_t *frame, xlator_t *this, children = priv->children; + AFR_SBRAIN_CHECK_LOC (loc, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -1472,6 +1475,8 @@ afr_getxattr (call_frame_t *frame, xlator_t *this, children = priv->children; + AFR_SBRAIN_CHECK_LOC (loc, out); + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; @@ -1727,10 +1732,8 @@ afr_fgetxattr (call_frame_t *frame, xlator_t *this, children = priv->children; - if (afr_is_split_brain (this, fd->inode)) { - op_errno = EIO; - goto out; - } + AFR_SBRAIN_CHECK_FD (fd, out); + AFR_LOCAL_ALLOC_OR_GOTO (local, out); frame->local = local; @@ -1886,10 +1889,7 @@ afr_readv (call_frame_t *frame, xlator_t *this, priv = this->private; children = priv->children; - if (afr_is_split_brain (this, fd->inode)) { - op_errno = EIO; - goto out; - } + AFR_SBRAIN_CHECK_FD (fd, out); AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); local = frame->local; diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c index 68570f15a..c1ec69a55 100644 --- a/xlators/cluster/afr/src/afr-inode-write.c +++ b/xlators/cluster/afr/src/afr-inode-write.c @@ -39,6 +39,46 @@ #include "afr-transaction.h" #include "afr-self-heal-common.h" +void +__inode_write_fop_cbk (call_frame_t *frame, int child_index, int read_child, + xlator_t *this, int32_t *op_ret, int32_t *op_errno, + struct iatt *prebuf, struct iatt *postbuf, dict_t *xdata) +{ + afr_local_t *local = NULL; + + local = frame->local; + + if (afr_fop_failed (*op_ret, *op_errno)) { + local->child_errno[child_index] = *op_errno; + + switch (local->op) { + case GF_FOP_TRUNCATE: + case GF_FOP_FTRUNCATE: + if (*op_errno != EFBIG) + afr_transaction_fop_failed (frame, this, + child_index); + break; + default: + afr_transaction_fop_failed (frame, this, child_index); + break; + } + local->op_errno = *op_errno; + goto out; + } + + if ((local->success_count == 0) || (read_child == child_index)) { + local->op_ret = *op_ret; + if (prebuf) + local->cont.inode_wfop.prebuf = *prebuf; + if (postbuf) + local->cont.inode_wfop.postbuf = *postbuf; + } + + local->success_count++; +out: + return; +} + /* {{{ writev */ void @@ -52,8 +92,8 @@ afr_writev_copy_outvars (call_frame_t *src_frame, call_frame_t *dst_frame) dst_local->op_ret = src_local->op_ret; dst_local->op_errno = src_local->op_errno; - dst_local->cont.writev.prebuf = src_local->cont.writev.prebuf; - dst_local->cont.writev.postbuf = src_local->cont.writev.postbuf; + dst_local->cont.inode_wfop.prebuf = src_local->cont.inode_wfop.prebuf; + dst_local->cont.inode_wfop.postbuf = src_local->cont.inode_wfop.postbuf; } void @@ -64,8 +104,8 @@ afr_writev_unwind (call_frame_t *frame, xlator_t *this) AFR_STACK_UNWIND (writev, frame, local->op_ret, local->op_errno, - &local->cont.writev.prebuf, - &local->cont.writev.postbuf, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, NULL); } @@ -133,14 +173,17 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, struct iatt *postbuf, dict_t *xdata) { afr_local_t * local = NULL; + afr_private_t *priv = NULL; call_frame_t *fop_frame = NULL; int child_index = (long) cookie; int call_count = -1; int read_child = 0; int ret = 0; uint32_t open_fd_count = 0; + uint32_t write_is_append = 0; local = frame->local; + priv = this->private; read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL); @@ -150,12 +193,14 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->read_child_returned = _gf_true; } + __inode_write_fop_cbk (frame, child_index, read_child, this, + &op_ret, &op_errno, prebuf, postbuf, + xdata); + local->replies[child_index].valid = 1; local->replies[child_index].op_ret = op_ret; local->replies[child_index].op_errno = op_errno; - if (afr_fop_failed (op_ret, op_errno)) - afr_transaction_fop_failed (frame, this, child_index); /* stage the best case return value for unwind */ if ((local->success_count == 0) || (op_ret > local->op_ret)) { @@ -173,14 +218,15 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->open_fd_count = open_fd_count; local->update_open_fd_count = _gf_true; } + + write_is_append = 0; + ret = dict_get_uint32 (xdata, + GLUSTERFS_WRITE_IS_APPEND, + &write_is_append); + if (ret || !write_is_append) + local->append_write = _gf_false; } - if ((local->success_count == 0) || - (child_index == read_child)) { - local->cont.writev.prebuf = *prebuf; - local->cont.writev.postbuf = *postbuf; - } - local->success_count++; } } UNLOCK (&frame->lock); @@ -192,10 +238,20 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, if (local->update_open_fd_count) afr_handle_open_fd_count (frame, this); - if (!local->stable_write) + if (!local->stable_write && !local->append_write) + /* An appended write removes the necessity to + fsync() the file. This is because self-heal + has the logic to check for larger file when + the xattrs are not reliably pointing at + a stale file. + */ afr_fd_report_unstable_write (this, local->fd); afr_writev_handle_short_writes (frame, this); + if (afr_any_fops_failed (local, priv)) { + //Don't unwind until post-op is complete + local->transaction.resume (frame, this); + } else { /* * Generally inode-write fops do transaction.unwind then * transaction.resume, but writev needs to make sure that @@ -207,10 +263,11 @@ afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, * completed. */ - fop_frame = afr_transaction_detach_fop_frame (frame); - afr_writev_copy_outvars (frame, fop_frame); - local->transaction.resume (frame, this); - afr_writev_unwind (fop_frame, this); + fop_frame = afr_transaction_detach_fop_frame (frame); + afr_writev_copy_outvars (frame, fop_frame); + local->transaction.resume (frame, this); + afr_writev_unwind (fop_frame, this); + } } return 0; } @@ -251,6 +308,13 @@ afr_writev_wind (call_frame_t *frame, xlator_t *this) if (xdata) { ret = dict_set_uint32 (xdata, GLUSTERFS_OPEN_FD_COUNT, sizeof (uint32_t)); + ret = dict_set_uint32 (xdata, GLUSTERFS_WRITE_IS_APPEND, + 0); + /* Set append_write to be true speculatively. If on any + server it turns not be true, we unset it in the + callback. + */ + local->append_write = _gf_true; } for (i = 0; i < priv->child_count; i++) { @@ -542,8 +606,8 @@ afr_truncate_unwind (call_frame_t *frame, xlator_t *this) if (main_frame) { AFR_STACK_UNWIND (truncate, main_frame, local->op_ret, local->op_errno, - &local->cont.truncate.prebuf, - &local->cont.truncate.postbuf, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, NULL); } @@ -557,14 +621,11 @@ afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, struct iatt *postbuf, dict_t *xdata) { afr_local_t * local = NULL; - afr_private_t * priv = NULL; int child_index = (long) cookie; int read_child = 0; int call_count = -1; - int need_unwind = 0; local = frame->local; - priv = this->private; read_child = afr_inode_get_read_ctx (this, local->loc.inode, NULL); @@ -574,38 +635,22 @@ afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->read_child_returned = _gf_true; } - if (afr_fop_failed (op_ret, op_errno) && op_errno != EFBIG) - afr_transaction_fop_failed (frame, this, child_index); - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - local->cont.truncate.prebuf = *prebuf; - local->cont.truncate.postbuf = *postbuf; - } - - if (child_index == read_child) { - local->cont.truncate.prebuf = *prebuf; - local->cont.truncate.postbuf = *postbuf; - } - - local->success_count++; - - if ((local->success_count >= priv->wait_count) - && local->read_child_returned) { - need_unwind = 1; - } + if (prebuf->ia_size != postbuf->ia_size) + local->stable_write = _gf_false; } - local->op_errno = op_errno; + __inode_write_fop_cbk (frame, child_index, read_child, this, + &op_ret, &op_errno, prebuf, postbuf, + xdata); } UNLOCK (&frame->lock); - if (need_unwind) - local->transaction.unwind (frame, this); - call_count = afr_frame_return (frame); if (call_count == 0) { + if (local->stable_write && afr_txn_nothing_failed (frame, this)) + local->transaction.unwind (frame, this); + local->transaction.resume (frame, this); } @@ -633,6 +678,7 @@ afr_truncate_wind (call_frame_t *frame, xlator_t *this) } local->call_count = call_count; + local->stable_write = _gf_true; for (i = 0; i < priv->child_count; i++) { if (local->transaction.pre_op[i]) { @@ -753,8 +799,8 @@ afr_ftruncate_unwind (call_frame_t *frame, xlator_t *this) if (main_frame) { AFR_STACK_UNWIND (ftruncate, main_frame, local->op_ret, local->op_errno, - &local->cont.ftruncate.prebuf, - &local->cont.ftruncate.postbuf, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, NULL); } return 0; @@ -767,14 +813,11 @@ afr_ftruncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, struct iatt *postbuf, dict_t *xdata) { afr_local_t * local = NULL; - afr_private_t * priv = NULL; int child_index = (long) cookie; int call_count = -1; - int need_unwind = 0; int read_child = 0; local = frame->local; - priv = this->private; read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL); @@ -784,38 +827,22 @@ afr_ftruncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->read_child_returned = _gf_true; } - if (afr_fop_failed (op_ret, op_errno)) - afr_transaction_fop_failed (frame, this, child_index); - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - local->cont.ftruncate.prebuf = *prebuf; - local->cont.ftruncate.postbuf = *postbuf; - } - - if (child_index == read_child) { - local->cont.ftruncate.prebuf = *prebuf; - local->cont.ftruncate.postbuf = *postbuf; - } - - local->success_count++; - - if ((local->success_count >= priv->wait_count) - && local->read_child_returned) { - need_unwind = 1; - } + if (prebuf->ia_size != postbuf->ia_size) + local->stable_write = _gf_false; } - local->op_errno = op_errno; + __inode_write_fop_cbk (frame, child_index, read_child, this, + &op_ret, &op_errno, prebuf, postbuf, + xdata); } UNLOCK (&frame->lock); - if (need_unwind) - local->transaction.unwind (frame, this); - call_count = afr_frame_return (frame); if (call_count == 0) { + if (local->stable_write && afr_txn_nothing_failed (frame, this)) + local->transaction.unwind (frame, this); + local->transaction.resume (frame, this); } @@ -843,6 +870,7 @@ afr_ftruncate_wind (call_frame_t *frame, xlator_t *this) } local->call_count = call_count; + local->stable_write = _gf_true; for (i = 0; i < priv->child_count; i++) { if (local->transaction.pre_op[i]) { @@ -997,8 +1025,8 @@ afr_setattr_unwind (call_frame_t *frame, xlator_t *this) if (main_frame) { AFR_STACK_UNWIND (setattr, main_frame, local->op_ret, local->op_errno, - &local->cont.setattr.preop_buf, - &local->cont.setattr.postop_buf, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, NULL); } @@ -1029,29 +1057,14 @@ afr_setattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->read_child_returned = _gf_true; } - if (afr_fop_failed (op_ret, op_errno)) - afr_transaction_fop_failed (frame, this, child_index); - - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - local->cont.setattr.preop_buf = *preop; - local->cont.setattr.postop_buf = *postop; - } - - if (child_index == read_child) { - local->cont.setattr.preop_buf = *preop; - local->cont.setattr.postop_buf = *postop; - } - - local->success_count++; + __inode_write_fop_cbk (frame, child_index, read_child, this, + &op_ret, &op_errno, preop, postop, + xdata); - if ((local->success_count >= priv->wait_count) - && local->read_child_returned) { - need_unwind = 1; - } + if ((local->success_count >= priv->wait_count) + && local->read_child_returned) { + need_unwind = 1; } - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -1206,8 +1219,8 @@ afr_fsetattr_unwind (call_frame_t *frame, xlator_t *this) if (main_frame) { AFR_STACK_UNWIND (fsetattr, main_frame, local->op_ret, local->op_errno, - &local->cont.fsetattr.preop_buf, - &local->cont.fsetattr.postop_buf, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, NULL); } @@ -1238,29 +1251,14 @@ afr_fsetattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->read_child_returned = _gf_true; } - if (afr_fop_failed (op_ret, op_errno)) - afr_transaction_fop_failed (frame, this, child_index); - - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - local->cont.fsetattr.preop_buf = *preop; - local->cont.fsetattr.postop_buf = *postop; - } - - if (child_index == read_child) { - local->cont.fsetattr.preop_buf = *preop; - local->cont.fsetattr.postop_buf = *postop; - } - - local->success_count++; + __inode_write_fop_cbk (frame, child_index, read_child, this, + &op_ret, &op_errno, preop, postop, + xdata); - if ((local->success_count >= priv->wait_count) - && local->read_child_returned) { - need_unwind = 1; - } + if ((local->success_count >= priv->wait_count) + && local->read_child_returned) { + need_unwind = 1; } - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -1433,28 +1431,23 @@ int afr_setxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) { - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - int call_count = -1; - int need_unwind = 0; + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + int call_count = -1; + int need_unwind = 0; + int child_index = (long) cookie; local = frame->local; priv = this->private; LOCK (&frame->lock); { - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - } - local->success_count++; - - if (local->success_count == priv->child_count) { - need_unwind = 1; - } + __inode_write_fop_cbk (frame, child_index, -1, this, + &op_ret, &op_errno, NULL, NULL, + xdata); + if (local->success_count == priv->child_count) { + need_unwind = 1; } - - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -1623,28 +1616,24 @@ int afr_fsetxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) { - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - int call_count = -1; - int need_unwind = 0; + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + int call_count = -1; + int need_unwind = 0; + int child_index = (long) cookie; local = frame->local; priv = this->private; LOCK (&frame->lock); { - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - } - local->success_count++; - if (local->success_count == priv->child_count) { - need_unwind = 1; - } + __inode_write_fop_cbk (frame, child_index, -1, this, + &op_ret, &op_errno, NULL, NULL, + xdata); + if (local->success_count == priv->child_count) { + need_unwind = 1; } - - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -1823,28 +1812,23 @@ int afr_removexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) { - afr_local_t * local = NULL; - afr_private_t * priv = NULL; - int call_count = -1; - int need_unwind = 0; + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + int call_count = -1; + int need_unwind = 0; + int child_index = (long) cookie; local = frame->local; priv = this->private; LOCK (&frame->lock); { - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - } - local->success_count++; - - if (local->success_count == priv->wait_count) { - need_unwind = 1; - } + __inode_write_fop_cbk (frame, child_index, -1, this, + &op_ret, &op_errno, NULL, NULL, + xdata); + if (local->success_count == priv->wait_count) { + need_unwind = 1; } - - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -2012,28 +1996,24 @@ int afr_fremovexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) { - afr_local_t * local = NULL; - afr_private_t * priv = NULL; - int call_count = -1; - int need_unwind = 0; + afr_local_t * local = NULL; + afr_private_t * priv = NULL; + int call_count = -1; + int need_unwind = 0; + int child_index = (long) cookie; local = frame->local; priv = this->private; LOCK (&frame->lock); { - if (op_ret != -1) { - if (local->success_count == 0) { - local->op_ret = op_ret; - } - local->success_count++; + __inode_write_fop_cbk (frame, child_index, -1, this, + &op_ret, &op_errno, NULL, NULL, + xdata); - if (local->success_count == priv->wait_count) { - need_unwind = 1; - } + if (local->success_count == priv->wait_count) { + need_unwind = 1; } - - local->op_errno = op_errno; } UNLOCK (&frame->lock); @@ -2178,3 +2158,704 @@ out: return 0; } + +static int +afr_fallocate_unwind (call_frame_t *frame, xlator_t *this) +{ + afr_local_t * local = NULL; + call_frame_t *main_frame = NULL; + + local = frame->local; + + LOCK (&frame->lock); + { + if (local->transaction.main_frame) + main_frame = local->transaction.main_frame; + local->transaction.main_frame = NULL; + } + UNLOCK (&frame->lock); + + if (main_frame) { + AFR_STACK_UNWIND (fallocate, main_frame, local->op_ret, + local->op_errno, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, + NULL); + } + return 0; +} + +static int +afr_fallocate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *prebuf, + struct iatt *postbuf, dict_t *xdata) +{ + afr_local_t * local = NULL; + afr_private_t * priv = NULL; + int child_index = (long) cookie; + int call_count = -1; + int need_unwind = 0; + int read_child = 0; + + local = frame->local; + priv = this->private; + + read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL); + + LOCK (&frame->lock); + { + if (child_index == read_child) { + local->read_child_returned = _gf_true; + } + + __inode_write_fop_cbk (frame, child_index, read_child, this, + &op_ret, &op_errno, prebuf, postbuf, + xdata); + + if ((local->success_count >= priv->wait_count) + && local->read_child_returned) { + need_unwind = 1; + } + } + UNLOCK (&frame->lock); + + if (need_unwind) + local->transaction.unwind (frame, this); + + call_count = afr_frame_return (frame); + + if (call_count == 0) { + local->transaction.resume (frame, this); + } + + return 0; +} + +static int +afr_fallocate_wind (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + int call_count = -1; + int i = 0; + + local = frame->local; + priv = this->private; + + call_count = afr_pre_op_done_children_count (local->transaction.pre_op, + priv->child_count); + + if (call_count == 0) { + local->transaction.resume (frame, this); + return 0; + } + + local->call_count = call_count; + + for (i = 0; i < priv->child_count; i++) { + if (local->transaction.pre_op[i]) { + STACK_WIND_COOKIE (frame, afr_fallocate_wind_cbk, + (void *) (long) i, + priv->children[i], + priv->children[i]->fops->fallocate, + local->fd, + local->cont.fallocate.mode, + local->cont.fallocate.offset, + local->cont.fallocate.len, + NULL); + + if (!--call_count) + break; + } + } + + return 0; +} + +static int +afr_fallocate_done (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + + local = frame->local; + + local->transaction.unwind (frame, this); + + AFR_STACK_DESTROY (frame); + + return 0; +} + +static int +afr_do_fallocate (call_frame_t *frame, xlator_t *this) +{ + call_frame_t * transaction_frame = NULL; + afr_local_t * local = NULL; + int op_ret = -1; + int op_errno = 0; + + local = frame->local; + + transaction_frame = copy_frame (frame); + if (!transaction_frame) { + goto out; + } + + transaction_frame->local = local; + frame->local = NULL; + + local->op = GF_FOP_FALLOCATE; + + local->transaction.fop = afr_fallocate_wind; + local->transaction.done = afr_fallocate_done; + local->transaction.unwind = afr_fallocate_unwind; + + local->transaction.main_frame = frame; + + local->transaction.start = local->cont.fallocate.offset; + local->transaction.len = 0; + + /* fallocate can modify the file size */ + op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); + if (op_ret < 0) { + op_errno = -op_ret; + goto out; + } + + op_ret = 0; +out: + if (op_ret < 0) { + if (transaction_frame) + AFR_STACK_DESTROY (transaction_frame); + AFR_STACK_UNWIND (fallocate, frame, op_ret, op_errno, NULL, + NULL, NULL); + } + + return 0; +} + +int +afr_fallocate (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode, + off_t offset, size_t len, dict_t *xdata) +{ + afr_private_t * priv = NULL; + afr_local_t * local = NULL; + call_frame_t *transaction_frame = NULL; + int ret = -1; + int op_errno = 0; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + + priv = this->private; + + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } + QUORUM_CHECK(fallocate,out); + + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; + + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) + goto out; + + local->cont.fallocate.mode = mode; + local->cont.fallocate.offset = offset; + local->cont.fallocate.len = len; + + local->fd = fd_ref (fd); + + afr_open_fd_fix (fd, this); + + afr_do_fallocate (frame, this); + + ret = 0; +out: + if (ret < 0) { + if (transaction_frame) + AFR_STACK_DESTROY (transaction_frame); + AFR_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL); + } + + return 0; +} + +/* }}} */ + +/* {{{ discard */ + +static int +afr_discard_unwind (call_frame_t *frame, xlator_t *this) +{ + afr_local_t * local = NULL; + call_frame_t *main_frame = NULL; + + local = frame->local; + + LOCK (&frame->lock); + { + if (local->transaction.main_frame) + main_frame = local->transaction.main_frame; + local->transaction.main_frame = NULL; + } + UNLOCK (&frame->lock); + + if (main_frame) { + AFR_STACK_UNWIND (discard, main_frame, local->op_ret, + local->op_errno, + &local->cont.inode_wfop.prebuf, + &local->cont.inode_wfop.postbuf, + NULL); + } + return 0; +} + +static int +afr_discard_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *prebuf, + struct iatt *postbuf, dict_t *xdata) +{ + afr_local_t * local = NULL; + afr_private_t * priv = NULL; + int child_index = (long) cookie; + int call_count = -1; + int need_unwind = 0; + int read_child = 0; + + local = frame->local; + priv = this->private; + + read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL); + + LOCK (&frame->lock); + { + if (child_index == read_child) { + local->read_child_returned = _gf_true; + } + + __inode_write_fop_cbk (frame, child_index, read_child, this, + &op_ret, &op_errno, prebuf, postbuf, + xdata); + + if ((local->success_count >= priv->wait_count) + && local->read_child_returned) { + need_unwind = 1; + } + } + UNLOCK (&frame->lock); + + if (need_unwind) + local->transaction.unwind (frame, this); + + call_count = afr_frame_return (frame); + + if (call_count == 0) { + local->transaction.resume (frame, this); + } + + return 0; +} + +static int +afr_discard_wind (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + int call_count = -1; + int i = 0; + + local = frame->local; + priv = this->private; + + call_count = afr_pre_op_done_children_count (local->transaction.pre_op, + priv->child_count); + + if (call_count == 0) { + local->transaction.resume (frame, this); + return 0; + } + + local->call_count = call_count; + + for (i = 0; i < priv->child_count; i++) { + if (local->transaction.pre_op[i]) { + STACK_WIND_COOKIE (frame, afr_discard_wind_cbk, + (void *) (long) i, + priv->children[i], + priv->children[i]->fops->discard, + local->fd, + local->cont.discard.offset, + local->cont.discard.len, + NULL); + + if (!--call_count) + break; + } + } + + return 0; +} + +static int +afr_discard_done (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + + local = frame->local; + + local->transaction.unwind (frame, this); + + AFR_STACK_DESTROY (frame); + + return 0; +} + +static int +afr_do_discard (call_frame_t *frame, xlator_t *this) +{ + call_frame_t * transaction_frame = NULL; + afr_local_t * local = NULL; + int op_ret = -1; + int op_errno = 0; + + local = frame->local; + + transaction_frame = copy_frame (frame); + if (!transaction_frame) { + goto out; + } + + transaction_frame->local = local; + frame->local = NULL; + + local->op = GF_FOP_DISCARD; + + local->transaction.fop = afr_discard_wind; + local->transaction.done = afr_discard_done; + local->transaction.unwind = afr_discard_unwind; + + local->transaction.main_frame = frame; + + local->transaction.start = local->cont.discard.offset; + local->transaction.len = 0; + + op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); + if (op_ret < 0) { + op_errno = -op_ret; + goto out; + } + + op_ret = 0; +out: + if (op_ret < 0) { + if (transaction_frame) + AFR_STACK_DESTROY (transaction_frame); + AFR_STACK_UNWIND (discard, frame, op_ret, op_errno, NULL, + NULL, NULL); + } + + return 0; +} + +int +afr_discard (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + size_t len, dict_t *xdata) +{ + afr_private_t * priv = NULL; + afr_local_t * local = NULL; + call_frame_t *transaction_frame = NULL; + int ret = -1; + int op_errno = 0; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + + priv = this->private; + + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } + QUORUM_CHECK(discard, out); + + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; + + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) + goto out; + + local->cont.discard.offset = offset; + local->cont.discard.len = len; + + local->fd = fd_ref (fd); + + afr_open_fd_fix (fd, this); + + afr_do_discard(frame, this); + + ret = 0; +out: + if (ret < 0) { + if (transaction_frame) + AFR_STACK_DESTROY (transaction_frame); + AFR_STACK_UNWIND (discard, frame, -1, op_errno, NULL, NULL, NULL); + } + + return 0; +} + + +/* {{{ zerofill */ + +static int +afr_zerofill_unwind (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + call_frame_t *main_frame = NULL; + + local = frame->local; + + LOCK (&frame->lock); + { + if (local->transaction.main_frame) { + main_frame = local->transaction.main_frame; + } + local->transaction.main_frame = NULL; + } + UNLOCK (&frame->lock); + + if (main_frame) { + AFR_STACK_UNWIND (zerofill, main_frame, local->op_ret, + local->op_errno, + &local->cont.zerofill.prebuf, + &local->cont.zerofill.postbuf, + NULL); + } + return 0; +} + +static int +afr_zerofill_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *prebuf, + struct iatt *postbuf, dict_t *xdata) +{ + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + int child_index = (long) cookie; + int call_count = -1; + int need_unwind = 0; + int read_child = 0; + + local = frame->local; + priv = this->private; + + read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL); + + LOCK (&frame->lock); + { + if (child_index == read_child) { + local->read_child_returned = _gf_true; + } + + if (afr_fop_failed (op_ret, op_errno)) { + afr_transaction_fop_failed (frame, this, child_index); + } + + if (op_ret != -1) { + if (local->success_count == 0) { + local->op_ret = op_ret; + local->cont.zerofill.prebuf = *prebuf; + local->cont.zerofill.postbuf = *postbuf; + } + + if (child_index == read_child) { + local->cont.zerofill.prebuf = *prebuf; + local->cont.zerofill.postbuf = *postbuf; + } + + local->success_count++; + + if ((local->success_count >= priv->wait_count) + && local->read_child_returned) { + need_unwind = 1; + } + } + local->op_errno = op_errno; + } + UNLOCK (&frame->lock); + + if (need_unwind) { + local->transaction.unwind (frame, this); + } + call_count = afr_frame_return (frame); + + if (call_count == 0) { + local->transaction.resume (frame, this); + } + + return 0; +} + +static int +afr_zerofill_wind (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + int call_count = -1; + int i = 0; + + local = frame->local; + priv = this->private; + + call_count = afr_pre_op_done_children_count (local->transaction.pre_op, + priv->child_count); + + if (call_count == 0) { + local->transaction.resume (frame, this); + return 0; + } + + local->call_count = call_count; + + for (i = 0; i < priv->child_count; i++) { + if (local->transaction.pre_op[i]) { + STACK_WIND_COOKIE (frame, afr_zerofill_wind_cbk, + (void *) (long) i, + priv->children[i], + priv->children[i]->fops->zerofill, + local->fd, + local->cont.zerofill.offset, + local->cont.zerofill.len, + NULL); + + if (!--call_count) + break; + } + } + + return 0; +} + +static int +afr_zerofill_done (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + + local = frame->local; + + local->transaction.unwind (frame, this); + + AFR_STACK_DESTROY (frame); + + return 0; +} + +static int +afr_do_zerofill(call_frame_t *frame, xlator_t *this) +{ + call_frame_t *transaction_frame = NULL; + afr_local_t *local = NULL; + int op_ret = -1; + int op_errno = 0; + + local = frame->local; + + transaction_frame = copy_frame (frame); + if (!transaction_frame) { + goto out; + } + + transaction_frame->local = local; + frame->local = NULL; + + local->op = GF_FOP_ZEROFILL; + + local->transaction.fop = afr_zerofill_wind; + local->transaction.done = afr_zerofill_done; + local->transaction.unwind = afr_zerofill_unwind; + + local->transaction.main_frame = frame; + + local->transaction.start = local->cont.zerofill.offset; + local->transaction.len = 0; + + op_ret = afr_transaction (transaction_frame, this, + AFR_DATA_TRANSACTION); + if (op_ret < 0) { + op_errno = -op_ret; + goto out; + } + + op_ret = 0; +out: + if (op_ret < 0) { + if (transaction_frame) { + AFR_STACK_DESTROY (transaction_frame); + } + AFR_STACK_UNWIND (zerofill, frame, op_ret, op_errno, NULL, + NULL, NULL); + } + + return 0; +} + +int +afr_zerofill (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + size_t len, dict_t *xdata) +{ + afr_private_t *priv = NULL; + afr_local_t *local = NULL; + call_frame_t *transaction_frame = NULL; + int ret = -1; + int op_errno = 0; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + + priv = this->private; + + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } + QUORUM_CHECK(zerofill, out); + + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; + + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) { + goto out; + } + local->cont.zerofill.offset = offset; + local->cont.zerofill.len = len; + + local->fd = fd_ref (fd); + + afr_open_fd_fix (fd, this); + + afr_do_zerofill(frame, this); + + ret = 0; +out: + if (ret < 0) { + if (transaction_frame) { + AFR_STACK_DESTROY (transaction_frame); + } + AFR_STACK_UNWIND (zerofill, frame, -1, op_errno, NULL, + NULL, NULL); + } + + return 0; +} + +/* }}} */ + + diff --git a/xlators/cluster/afr/src/afr-inode-write.h b/xlators/cluster/afr/src/afr-inode-write.h index ed11079fd..8e93ca44a 100644 --- a/xlators/cluster/afr/src/afr-inode-write.h +++ b/xlators/cluster/afr/src/afr-inode-write.h @@ -68,4 +68,15 @@ int32_t afr_fremovexattr (call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name, dict_t *xdata); +int +afr_discard (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + size_t len, dict_t *xdata); + +int +afr_fallocate (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode, + off_t offset, size_t len, dict_t *xdata); + +int +afr_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + size_t len, dict_t *xdata); #endif /* __INODE_WRITE_H__ */ diff --git a/xlators/cluster/afr/src/afr-lk-common.c b/xlators/cluster/afr/src/afr-lk-common.c index 116e1ef69..060d78f35 100644 --- a/xlators/cluster/afr/src/afr-lk-common.c +++ b/xlators/cluster/afr/src/afr-lk-common.c @@ -568,11 +568,14 @@ initialize_inodelk_variables (call_frame_t *frame, xlator_t *this) inodelk = afr_get_inodelk (int_lock, int_lock->domain); inodelk->lock_count = 0; + int_lock->lk_attempted_count = 0; int_lock->lock_op_ret = -1; int_lock->lock_op_errno = 0; memset (inodelk->locked_nodes, 0, sizeof (*inodelk->locked_nodes) * priv->child_count); + memset (int_lock->locked_nodes, 0, + sizeof (*int_lock->locked_nodes) * priv->child_count); return 0; } @@ -655,6 +658,7 @@ afr_unlock_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, afr_internal_lock_t *int_lock = NULL; afr_inodelk_t *inodelk = NULL; int32_t child_index = (long)cookie; + afr_private_t *priv = NULL; local = frame->local; int_lock = &local->internal_lock; @@ -663,9 +667,12 @@ afr_unlock_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, AFR_UNLOCK_OP, NULL, op_ret, op_errno, child_index); + priv = this->private; + if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) { - gf_log (this->name, GF_LOG_INFO, "%s: unlock failed on %d " - "unlock by %s", local->loc.path, child_index, + gf_log (this->name, GF_LOG_INFO, "%s: unlock failed on subvolume %s " + "with lock owner %s", local->loc.path, + priv->children[child_index]->name, lkowner_utoa (&frame->root->lk_owner)); } @@ -1329,6 +1336,7 @@ afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this) local->op_errno = EINVAL; int_lock->lock_op_errno = EINVAL; + afr_unlock (frame, this); return -1; } @@ -1524,6 +1532,7 @@ afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this) local->op_errno = EINVAL; int_lock->lock_op_errno = EINVAL; + afr_unlock (frame, this); ret = -1; goto out; } diff --git a/xlators/cluster/afr/src/afr-mem-types.h b/xlators/cluster/afr/src/afr-mem-types.h index e01ab366f..73594f265 100644 --- a/xlators/cluster/afr/src/afr-mem-types.h +++ b/xlators/cluster/afr/src/afr-mem-types.h @@ -41,7 +41,10 @@ enum gf_afr_mem_types_ { gf_afr_mt_shd_event_t, gf_afr_mt_time_t, gf_afr_mt_pos_data_t, - gf_afr_mt_reply_t, + gf_afr_mt_reply_t, + gf_afr_mt_stats_t, + gf_afr_mt_shd_crawl_event_t, + gf_afr_mt_uint64_t, gf_afr_mt_end }; #endif diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c index 929def242..ef92b4205 100644 --- a/xlators/cluster/afr/src/afr-self-heal-common.c +++ b/xlators/cluster/afr/src/afr-self-heal-common.c @@ -18,12 +18,24 @@ #include "afr-self-heal.h" #include "pump.h" -#define ADD_FMT_STRING(msg, off, sh_str, status) \ +#define ADD_FMT_STRING(msg, off, sh_str, status, print_log) \ do { \ if (AFR_SELF_HEAL_NOT_ATTEMPTED != status) { \ off += snprintf (msg + off, sizeof (msg) - off, \ " "sh_str" self heal %s,", \ get_sh_completion_status (status));\ + print_log = 1; \ + } \ + } while (0) + +#define ADD_FMT_STRING_SYNC(msg, off, sh_str, status, print_log) \ + do { \ + if (AFR_SELF_HEAL_SYNC_BEGIN == status || \ + AFR_SELF_HEAL_FAILED == status) { \ + off += snprintf (msg + off, sizeof (msg) - off, \ + " "sh_str" self heal %s,", \ + get_sh_completion_status (status));\ + print_log = 1; \ } \ } while (0) @@ -489,6 +501,8 @@ afr_find_biggest_witness_among_fools (int32_t *witnesses, { int i = 0; int biggest_witness = -1; + int biggest_witness_idx = -1; + int biggest_witness_cnt = -1; GF_ASSERT (witnesses); GF_ASSERT (characters); @@ -498,10 +512,21 @@ afr_find_biggest_witness_among_fools (int32_t *witnesses, if (characters[i].type != AFR_NODE_FOOL) continue; - if (biggest_witness < witnesses[i]) + if (biggest_witness < witnesses[i]) { biggest_witness = witnesses[i]; + biggest_witness_idx = i; + biggest_witness_cnt = 1; + continue; + } + + if (biggest_witness == witnesses[i]) + biggest_witness_cnt++; } - return biggest_witness; + + if (biggest_witness_cnt != 1) + return -1; + + return biggest_witness_idx; } int @@ -529,10 +554,84 @@ afr_mark_fool_as_source_by_witness (int32_t *sources, int32_t *witnesses, return nsources; } + +int +afr_mark_fool_as_source_by_idx (int32_t *sources, int child_count, int idx) +{ + if (idx >= 0 && idx < child_count) { + sources[idx] = 1; + return 1; + } + return 0; +} + + +static int +afr_find_largest_file_size (struct iatt *bufs, int32_t *success_children, + int child_count) +{ + int idx = -1; + int i = -1; + int child = -1; + uint64_t max_size = 0; + uint64_t min_size = 0; + int num_children = 0; + + for (i = 0; i < child_count; i++) { + if (success_children[i] == -1) + break; + + child = success_children[i]; + if (bufs[child].ia_size > max_size) { + max_size = bufs[child].ia_size; + idx = child; + } + + if ((num_children == 0) || (bufs[child].ia_size < min_size)) { + min_size = bufs[child].ia_size; + } + + num_children++; + } + + /* If sizes are same for all of them, finding sources will have to + * happen with pending changelog. So return -1 + */ + if ((num_children > 1) && (min_size == max_size)) + return -1; + return idx; +} + + +static int +afr_find_newest_file (struct iatt *bufs, int32_t *success_children, + int child_count) +{ + int idx = -1; + int i = -1; + int child = -1; + uint64_t max_ctime = 0; + + for (i = 0; i < child_count; i++) { + if (success_children[i] == -1) + break; + + child = success_children[i]; + if (bufs[child].ia_ctime > max_ctime) { + max_ctime = bufs[child].ia_ctime; + idx = child; + } + } + + return idx; +} + + static int afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix, afr_node_character *characters, - int child_count) + int32_t *success_children, + int child_count, struct iatt *bufs) { int32_t biggest_witness = 0; int nsources = 0; @@ -540,6 +639,11 @@ afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix, GF_ASSERT (child_count > 0); + biggest_witness = afr_find_largest_file_size (bufs, success_children, + child_count); + if (biggest_witness != -1) + goto found; + witnesses = GF_CALLOC (child_count, sizeof (*witnesses), gf_afr_mt_int32_t); if (NULL == witnesses) { @@ -552,9 +656,15 @@ afr_mark_biggest_of_fools_as_source (int32_t *sources, int32_t **pending_matrix, biggest_witness = afr_find_biggest_witness_among_fools (witnesses, characters, child_count); - nsources = afr_mark_fool_as_source_by_witness (sources, witnesses, - characters, child_count, - biggest_witness); + if (biggest_witness != -1) + goto found; + + biggest_witness = afr_find_newest_file (bufs, success_children, + child_count); + +found: + nsources = afr_mark_fool_as_source_by_idx (sources, child_count, + biggest_witness); out: GF_FREE (witnesses); return nsources; @@ -898,7 +1008,8 @@ afr_mark_sources (xlator_t *this, int32_t *sources, int32_t **pending_matrix, nsources = afr_mark_biggest_of_fools_as_source (sources, pending_matrix, characters, - child_count); + success_children, + child_count, bufs); } out: @@ -2111,6 +2222,7 @@ afr_self_heal_local_init (afr_local_t *l, xlator_t *this) shc->background = sh->background; shc->type = sh->type; shc->data_sh_info = ""; + shc->metadata_sh_info = ""; uuid_copy (shc->sh_gfid_req, sh->sh_gfid_req); if (l->loc.path) { @@ -2630,7 +2742,8 @@ get_sh_completion_status (afr_self_heal_status status) char *not_attempted = " is not attempted"; char *failed = " failed"; - char *successfull_complt = " is successfully completed"; + char *started = " is started"; + char *sync_begin = " is successfully completed"; char *result = " has unknown status"; switch (status) @@ -2642,7 +2755,10 @@ get_sh_completion_status (afr_self_heal_status status) result = failed; break; case AFR_SELF_HEAL_STARTED: - result = successfull_complt; + result = started; + break; + case AFR_SELF_HEAL_SYNC_BEGIN: + result = sync_begin; break; } @@ -2659,26 +2775,38 @@ afr_log_self_heal_completion_status (afr_local_t *local, gf_loglevel_t loglvl) afr_sh_status_for_all_type all_status = sh->afr_all_sh_status; xlator_t *this = NULL; size_t off = 0; + int data_sh = 0; + int metadata_sh = 0; + int print_log = 0; this = THIS; ADD_FMT_STRING (sh_log, off, "gfid or missing entry", - all_status.gfid_or_missing_entry_self_heal); - ADD_FMT_STRING (sh_log, off, "metadata", all_status.metadata_self_heal); + all_status.gfid_or_missing_entry_self_heal, print_log); + ADD_FMT_STRING_SYNC (sh_log, off, "metadata", + all_status.metadata_self_heal, print_log); if (sh->background) { - ADD_FMT_STRING (sh_log, off, "backgroung data", - all_status.data_self_heal); + ADD_FMT_STRING_SYNC (sh_log, off, "backgroung data", + all_status.data_self_heal, print_log); } else { - ADD_FMT_STRING (sh_log, off, "foreground data", - all_status.data_self_heal); + ADD_FMT_STRING_SYNC (sh_log, off, "foreground data", + all_status.data_self_heal, print_log); } - ADD_FMT_STRING (sh_log, off, "entry", all_status.entry_self_heal); + ADD_FMT_STRING_SYNC (sh_log, off, "entry", all_status.entry_self_heal, + print_log); - if (AFR_SELF_HEAL_STARTED == all_status.data_self_heal) { - gf_log (this->name, loglvl, "%s %s on %s", sh_log, - sh->data_sh_info, local->loc.path); - } else { - gf_log (this->name, loglvl, "%s on %s", sh_log, - local->loc.path); - } + if (AFR_SELF_HEAL_SYNC_BEGIN == all_status.data_self_heal && + strcmp (sh->data_sh_info, "") && sh->data_sh_info ) + data_sh = 1; + if (AFR_SELF_HEAL_SYNC_BEGIN == all_status.metadata_self_heal && + strcmp (sh->metadata_sh_info, "") && sh->metadata_sh_info) + metadata_sh = 1; + + if (!print_log) + return; + + gf_log (this->name, loglvl, "%s %s %s on %s", sh_log, + ((data_sh == 1) ? sh->data_sh_info : ""), + ((metadata_sh == 1) ? sh->metadata_sh_info : ""), + local->loc.path); } diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c index cab848832..9de26ee56 100644 --- a/xlators/cluster/afr/src/afr-self-heal-data.c +++ b/xlators/cluster/afr/src/afr-self-heal-data.c @@ -728,39 +728,170 @@ out: } char* -afr_get_sizes_str (struct iatt *bufs, xlator_t *this) +afr_get_sizes_str (afr_local_t *local, struct iatt *bufs, xlator_t *this) { afr_private_t *priv = NULL; int i = 0; - char num[128] = {0}; + char num[1024] = {0}; size_t len = 0; char *sizes_str = NULL; size_t off = 0; - char *fmt_str = "%llu "; + char *fmt_str = "%llu bytes on %s, "; + char *child_down = " %s,"; + char *child_unknown = " %s,"; + int down_child_present = 0; + int down_count = 0; + int unknown_count = 0; + int unknown_child_present = 0; + char *down_subvol_1 = " down subvolume is "; + char *unknown_subvol_1 = " unknown subvolume is "; + char *down_subvol_2 = " down subvolumes are "; + char *unknown_subvol_2 = " unknown subvolumes are "; priv = this->private; + for (i = 0; i < priv->child_count; i++) { - len += snprintf (num, sizeof (num), fmt_str, - (unsigned long long) bufs[i].ia_size); + if (local->child_up[i] == 1) { + len += snprintf (num, sizeof (num), fmt_str, + (unsigned long long) bufs[i].ia_size, + priv->children[i]->name); + } else if (local->child_up[i] == 0) { + len += snprintf (num, sizeof (num), child_down, + priv->children[i]->name); + if (!down_child_present) + down_child_present = 1; + down_count ++; + } else if (local->child_up[i] == -1) { + len += snprintf (num, sizeof (num), child_unknown, + priv->children[i]->name); + if (!unknown_child_present) + unknown_child_present = 1; + unknown_count++; + } + + } + + if (down_child_present) { + if (down_count > 1) + len += snprintf (num, sizeof (num), "%s", + down_subvol_2); + else + len += snprintf (num, sizeof (num), "%s", + down_subvol_1); + } + if (unknown_child_present) { + if (unknown_count > 1) + len += snprintf (num, sizeof (num), "%s", + unknown_subvol_2); + else + len += snprintf (num, sizeof (num), "%s", + unknown_subvol_1); } len++;//for '\0' + sizes_str = GF_CALLOC (len, sizeof (char), gf_common_mt_char); + if (!sizes_str) return NULL; for (i = 0; i < priv->child_count; i++) { - off += snprintf (sizes_str + off, len - off, fmt_str, - (unsigned long long) bufs[i].ia_size); + if (local->child_up[i] == 1) { + off += snprintf (sizes_str + off, len - off, fmt_str, + (unsigned long long) bufs[i].ia_size, + priv->children[i]->name); + } + } + + if (down_child_present) { + if (down_count > 1) { + off += snprintf (sizes_str + off, len - off, "%s", + down_subvol_2); + } else { + off += snprintf (sizes_str + off, len - off, "%s", + down_subvol_1); + } + } + + for (i = 0; i < priv->child_count; i++) { + if (local->child_up[i] == 0) { + off += snprintf (sizes_str + off, len - off, child_down, + priv->children[i]->name); + } + } + + if (unknown_child_present) { + if (unknown_count > 1) { + off += snprintf (sizes_str + off, len - off, "%s", + unknown_subvol_2); + } else { + off += snprintf (sizes_str + off, len - off, "%s", + unknown_subvol_1); + } + } + + for (i = 0; i < priv->child_count; i++) { + if (local->child_up[i] == -1) { + off += snprintf (sizes_str + off, len - off, + child_unknown, + priv->children[i]->name); + + } } + return sizes_str; } +char* +afr_get_sinks_str (xlator_t *this, afr_local_t *local, afr_self_heal_t *sh) +{ + afr_private_t *priv = NULL; + int i = 0; + char num[1024] = {0}; + size_t len = 0; + char *sinks_str = NULL; + char *temp_str = " to sinks "; + char *str_format = " %s,"; + char off = 0; + + priv = this->private; + + len += snprintf (num, sizeof (num), "%s", temp_str); + for (i = 0; i < priv->child_count; i++) { + if ((sh->sources[i] == 0) && (local->child_up[i] == 1)) { + len += snprintf (num, sizeof (num), str_format, + priv->children[i]->name); + } + } + + len ++; + + sinks_str = GF_CALLOC (len, sizeof (char), gf_common_mt_char); + + if (!sinks_str) + return NULL; + + off += snprintf (sinks_str + off, len - off, "%s", temp_str); + + for (i = 0; i < priv->child_count; i++) { + if ((sh->sources[i] == 0) && (local->child_up[i] == 1)) { + off += snprintf (sinks_str + off, len - off, + str_format, + priv->children[i]->name); + } + } + + return sinks_str; + +} + + void -afr_set_data_sh_info_str (afr_self_heal_t *sh, xlator_t *this) +afr_set_data_sh_info_str (afr_local_t *local, afr_self_heal_t *sh, xlator_t *this) { char *pending_matrix_str = NULL; char *sizes_str = NULL; + char *sinks_str = NULL; afr_private_t *priv = NULL; priv = this->private; @@ -770,13 +901,17 @@ afr_set_data_sh_info_str (afr_self_heal_t *sh, xlator_t *this) if (!pending_matrix_str) pending_matrix_str = ""; - sizes_str = afr_get_sizes_str (sh->buf, this); + sizes_str = afr_get_sizes_str (local, sh->buf, this); if (!sizes_str) sizes_str = ""; - gf_asprintf (&sh->data_sh_info, " from %s with %s sizes %s", - priv->children[sh->source]->name, sizes_str, - pending_matrix_str); + sinks_str = afr_get_sinks_str (this, local, sh); + if (!sinks_str) + sinks_str = ""; + + gf_asprintf (&sh->data_sh_info, " data self heal from %s %s with " + "%s data %s", priv->children[sh->source]->name, sinks_str, + sizes_str, pending_matrix_str); if (pending_matrix_str && strcmp (pending_matrix_str, "")) GF_FREE (pending_matrix_str); @@ -825,6 +960,7 @@ afr_sh_data_fix (call_frame_t *frame, xlator_t *this) sh->active_sinks); sh->actual_sh_started = _gf_true; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN); afr_sh_data_trim_sinks (frame, this); } @@ -905,7 +1041,7 @@ afr_sh_data_fxattrop_fstat_done (call_frame_t *frame, xlator_t *this) } afr_sh_data_setattr (frame, this, &sh->buf[tstamp_source]); } else { - afr_set_data_sh_info_str (sh, this); + afr_set_data_sh_info_str (local, sh, this); if (nsources == 0) { gf_log (this->name, GF_LOG_DEBUG, "No self-heal needed for %s", diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c index 60d1a5f47..53491a1d7 100644 --- a/xlators/cluster/afr/src/afr-self-heal-entry.c +++ b/xlators/cluster/afr/src/afr-self-heal-entry.c @@ -2240,6 +2240,7 @@ afr_sh_entry_sync_prepare (call_frame_t *frame, xlator_t *this) local->loc.path); sh->actual_sh_started = _gf_true; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN); afr_sh_entry_open (frame, this); return 0; diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c index 6d629ce98..fd5da6cfd 100644 --- a/xlators/cluster/afr/src/afr-self-heal-metadata.c +++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c @@ -406,6 +406,140 @@ afr_sh_metadata_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, return 0; } +static void +afr_set_metadata_sh_info_str (afr_local_t *local, afr_self_heal_t *sh, + xlator_t *this) +{ + afr_private_t *priv = NULL; + int i = 0; + char num[1024] = {0}; + size_t len = 0; + char *string = NULL; + size_t off = 0; + char *source_child = " from source %s to"; + char *format = " %s, "; + char *string_msg = " metadata self heal"; + char *pending_matrix_str = NULL; + int down_child_present = 0; + int unknown_child_present = 0; + char *down_subvol_1 = " down subvolume is "; + char *unknown_subvol_1 = " unknown subvolume is"; + char *down_subvol_2 = " down subvolumes are "; + char *unknown_subvol_2 = " unknown subvolumes are "; + int down_count = 0; + int unknown_count = 0; + + priv = this->private; + + pending_matrix_str = afr_get_pending_matrix_str (sh->pending_matrix, + this); + + if (!pending_matrix_str) + pending_matrix_str = ""; + + len += snprintf (num, sizeof (num), "%s", string_msg); + + for (i = 0; i < priv->child_count; i++) { + if ((sh->source == i) && (local->child_up[i] == 1)) { + len += snprintf (num, sizeof (num), source_child, + priv->children[i]->name); + } else if ((local->child_up[i] == 1) && (sh->sources[i] == 0)) { + len += snprintf (num, sizeof (num), format, + priv->children[i]->name); + } else if (local->child_up[i] == 0) { + len += snprintf (num, sizeof (num), format, + priv->children[i]->name); + if (!down_child_present) + down_child_present = 1; + down_count++; + } else if (local->child_up[i] == -1) { + len += snprintf (num, sizeof (num), format, + priv->children[i]->name); + if (!unknown_child_present) + unknown_child_present = 1; + unknown_count++; + } + } + + if (down_child_present) { + if (down_count > 1) { + len += snprintf (num, sizeof (num), "%s", + down_subvol_2); + } else { + len += snprintf (num, sizeof (num), "%s", + down_subvol_1); + } + } + if (unknown_child_present) { + if (unknown_count > 1) { + len += snprintf (num, sizeof (num), "%s", + unknown_subvol_2); + } else { + len += snprintf (num, sizeof (num), "%s", + unknown_subvol_1); + } + } + + len ++; + + string = GF_CALLOC (len, sizeof (char), gf_common_mt_char); + if (!string) + return; + + off += snprintf (string + off, len - off, "%s", string_msg); + for (i=0; i < priv->child_count; i++) { + if ((sh->source == i) && (local->child_up[i] == 1)) + off += snprintf (string + off, len - off, source_child, + priv->children[i]->name); + } + + for (i = 0; i < priv->child_count; i++) { + if ((local->child_up[i] == 1)&& (sh->sources[i] == 0)) + off += snprintf (string + off, len - off, format, + priv->children[i]->name); + } + + if (down_child_present) { + if (down_count > 1) { + off += snprintf (string + off, len - off, "%s", + down_subvol_2); + } else { + off += snprintf (string + off, len - off, "%s", + down_subvol_1); + } + } + + for (i = 0; i < priv->child_count; i++) { + if (local->child_up[i] == 0) + off += snprintf (string + off, len - off, format, + priv->children[i]->name); + } + + if (unknown_child_present) { + if (unknown_count > 1) { + off += snprintf (string + off, len - off, "%s", + unknown_subvol_2); + } else { + off += snprintf (string + off, len - off, "%s", + unknown_subvol_1); + } + } + + for (i = 0; i < priv->child_count; i++) { + if (local->child_up[i] == -1) + off += snprintf (string + off, len - off, format, + priv->children[i]->name); + } + + gf_asprintf (&sh->metadata_sh_info, "%s metadata %s,", string, + pending_matrix_str); + + if (pending_matrix_str && strcmp (pending_matrix_str, "")) + GF_FREE (pending_matrix_str); + + if (string && strcmp (string, "")) + GF_FREE (string); +} int afr_sh_metadata_sync_prepare (call_frame_t *frame, xlator_t *this) @@ -436,6 +570,8 @@ afr_sh_metadata_sync_prepare (call_frame_t *frame, xlator_t *this) sh->active_sinks); sh->actual_sh_started = _gf_true; + afr_set_self_heal_status (sh, AFR_SELF_HEAL_SYNC_BEGIN); + afr_set_metadata_sh_info_str (local, sh, this); STACK_WIND (frame, afr_sh_metadata_getxattr_cbk, priv->children[source], priv->children[source]->fops->getxattr, diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c index f33b04eed..1b48a1bca 100644 --- a/xlators/cluster/afr/src/afr-self-heald.c +++ b/xlators/cluster/afr/src/afr-self-heald.c @@ -25,7 +25,8 @@ typedef enum { typedef enum { HEAL = 1, - INFO + INFO, + STATISTICS_TO_BE_HEALED, } shd_crawl_op; typedef struct shd_dump { @@ -84,6 +85,33 @@ _loc_assign_gfid_path (loc_t *loc) } void +_destroy_crawl_event_data (void *data) +{ + shd_crawl_event_t *crawl_event = NULL; + + if (!data) + goto out; + + crawl_event = (shd_crawl_event_t *)data; + GF_FREE (crawl_event->start_time_str); + GF_FREE (crawl_event->end_time_str); + +out: + return; +} + +void +_destroy_shd_event_data (void *data) +{ + shd_event_t *event = NULL; + if (!data) + goto out; + event = (shd_event_t*)data; + GF_FREE (event->path); +out: + return; +} +void shd_cleanup_event (void *event) { shd_event_t *shd_event = event; @@ -128,6 +156,123 @@ _build_index_loc (xlator_t *this, loc_t *loc, char *name, loc_t *parent) } int +_add_crawl_stats_to_dict (xlator_t *this, dict_t *output, int child, + shd_crawl_event_t *shd_event, struct timeval *tv) +{ + int ret = 0; + uint64_t count = 0; + char key[256] = {0}; + int xl_id = 0; + uint64_t healed_count = 0; + uint64_t split_brain_count = 0; + uint64_t heal_failed_count = 0; + char *start_time_str = NULL; + char *end_time_str = NULL; + char *crawl_type = NULL; + int progress = -1; + + healed_count = shd_event->healed_count; + split_brain_count = shd_event->split_brain_count; + heal_failed_count = shd_event->heal_failed_count; + start_time_str = shd_event->start_time_str; + end_time_str = shd_event->end_time_str; + crawl_type = shd_event->crawl_type; + + if (!start_time_str) { + ret = -1; + goto out; + } + + + ret = dict_get_int32 (output, this->name, &xl_id); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "xl does not have id"); + goto out; + } + + snprintf (key, sizeof (key), "statistics-%d-%d-count", xl_id, child); + ret = dict_get_uint64 (output, key, &count); + + snprintf (key, sizeof (key), "statistics_healed_cnt-%d-%d-%"PRIu64, + xl_id, child, count); + ret = dict_set_uint64(output, key, healed_count); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_" + "healed_count to outout"); + goto out; + } + snprintf (key, sizeof (key), "statistics_sb_cnt-%d-%d-%"PRIu64, + xl_id, child, count); + ret = dict_set_uint64 (output, key, split_brain_count); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_" + "split_brain_count to outout"); + goto out; + } + snprintf (key, sizeof (key), "statistics_crawl_type-%d-%d-%"PRIu64, + xl_id, child, count); + ret = dict_set_dynstr (output, key, gf_strdup (crawl_type)); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_" + "crawl_type to output"); + goto out; + } + snprintf (key, sizeof (key), "statistics_heal_failed_cnt-%d-%d-%"PRIu64, + xl_id, child, count); + ret = dict_set_uint64 (output, key, heal_failed_count); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_" + "healed_failed_count to outout"); + goto out; + } + snprintf (key, sizeof (key), "statistics_strt_time-%d-%d-%"PRIu64, + xl_id, child, count); + ret = dict_set_dynstr (output, key, gf_strdup(start_time_str)); + + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_" + "crawl_start_time to outout"); + goto out; + } + + snprintf (key, sizeof (key), "statistics_end_time-%d-%d-%"PRIu64, + xl_id, child, count); + + if (!end_time_str) + end_time_str = "Could not determine the end time"; + ret = dict_set_dynstr (output, key, gf_strdup(end_time_str)); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_" + "crawl_end_time to outout"); + goto out; + } + snprintf (key, sizeof (key), "statistics_inprogress-%d-%d-%"PRIu64, + xl_id, child, count); + + if (shd_event->crawl_inprogress == _gf_true) + progress = 1; + else + progress = 0; + + ret = dict_set_int32 (output, key, progress); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Could not add statistics_" + "inprogress to outout"); + goto out; + } + + snprintf (key, sizeof (key), "statistics-%d-%d-count",xl_id, child); + ret = dict_set_uint64 (output, key, count + 1); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, "Could not increment the " + "counter."); + goto out; + } +out: + return ret; +} + +int _add_path_to_dict (xlator_t *this, dict_t *output, int child, char *path, struct timeval *tv, gf_boolean_t dyn) { @@ -233,6 +378,20 @@ out: } int +_add_crawl_event_statistics_to_dict (circular_buffer_t *cb, void *data) +{ + int ret = 0; + shd_dump_t *dump_data = NULL; + shd_crawl_event_t *shd_event = NULL; + + dump_data = data; + shd_event = cb->data; + ret = _add_crawl_stats_to_dict (dump_data->this, dump_data->dict, + dump_data->child, shd_event, &cb->tv); + return ret; +} + +int _add_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict, int child) { shd_dump_t dump_data = {0}; @@ -244,6 +403,26 @@ _add_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict, int child) return 0; } + +int +_add_statistics_to_dict (xlator_t *this, dict_t *dict, int child) +{ + shd_dump_t dump_data = {0}; + afr_private_t *priv = NULL; + afr_self_heald_t *shd = NULL; + + priv = this->private; + shd = &priv->shd; + + dump_data.this = this; + dump_data.dict = dict; + dump_data.child = child; + eh_dump (shd->statistics[child], &dump_data, + _add_crawl_event_statistics_to_dict); + return 0; + +} + void _remove_stale_index (xlator_t *this, xlator_t *readdir_xl, loc_t *parent, char *fname) @@ -269,6 +448,46 @@ out: } int +_count_hard_links_under_base_indices_dir (xlator_t *this, + afr_crawl_data_t *crawl_data, + gf_dirent_t *entry, loc_t *childloc, + loc_t *parentloc, struct iatt *iattr) +{ + xlator_t *readdir_xl = crawl_data->readdir_xl; + struct iatt parent = {0}; + int ret = 0; + dict_t *output = NULL; + int xl_id = 0; + char key[256] = {0}; + int child = -1; + uint64_t hardlinks = 0; + + output = crawl_data->op_data; + child = crawl_data->child; + + ret = syncop_lookup (readdir_xl, childloc, NULL, iattr, NULL, &parent); + if (ret) + goto out; + + ret = dict_get_int32 (output, this->name, &xl_id); + if (ret) + goto out; + + snprintf (key, sizeof (key), "%d-%d-hardlinks", xl_id, child); + ret = dict_get_uint64 (output, key, &hardlinks); + + /*Removing the count of base_entry under indices/base_indicies and + * entry under indices/xattrop */ + hardlinks = hardlinks + iattr->ia_nlink - 2; + ret = dict_set_uint64 (output, key, hardlinks); + if (ret) + goto out; + +out: + return ret; +} + +int _add_summary_to_dict (xlator_t *this, afr_crawl_data_t *crawl_data, gf_dirent_t *entry, loc_t *childloc, loc_t *parentloc, struct iatt *iattr) @@ -307,16 +526,18 @@ _crawl_post_sh_action (xlator_t *this, loc_t *parent, loc_t *child, int32_t op_ret, int32_t op_errno, dict_t *xattr_rsp, afr_crawl_data_t *crawl_data) { - int ret = 0; - afr_private_t *priv = NULL; - afr_self_heald_t *shd = NULL; - eh_t *eh = NULL; - char *path = NULL; - char gfid_str[64] = {0}; - shd_event_t *event = NULL; - int32_t sh_failed = 0; - gf_boolean_t split_brain = 0; - int32_t actual_sh_done = 0; + int ret = 0; + afr_private_t *priv = NULL; + afr_self_heald_t *shd = NULL; + eh_t *eh = NULL; + char *path = NULL; + char gfid_str[64] = {0}; + shd_event_t *event = NULL; + int32_t sh_failed = 0; + gf_boolean_t split_brain = 0; + int32_t actual_sh_done = 0; + shd_crawl_event_t **shd_crawl_event = NULL; + priv = this->private; shd = &priv->shd; if (crawl_data->crawl == INDEX) { @@ -343,16 +564,19 @@ _crawl_post_sh_action (xlator_t *this, loc_t *parent, loc_t *child, ret = dict_get_int32 (xattr_rsp, "actual-sh-done", &actual_sh_done); } - split_brain = afr_is_split_brain (this, child->inode); + shd_crawl_event = (shd_crawl_event_t**)(shd->crawl_events); + split_brain = afr_is_split_brain (this, child->inode); if ((op_ret < 0 && op_errno == EIO) || split_brain) { eh = shd->split_brain; + shd_crawl_event[crawl_data->child]->split_brain_count += 1; } else if ((op_ret < 0) || sh_failed) { eh = shd->heal_failed; + shd_crawl_event[crawl_data->child]->heal_failed_count += 1; } else if (actual_sh_done == 1) { - eh = shd->healed; + eh = shd->healed; + shd_crawl_event[crawl_data->child]->healed_count += 1; } - ret = -1; if (eh != NULL) { @@ -408,10 +632,20 @@ _self_heal_entry (xlator_t *this, afr_crawl_data_t *crawl_data, gf_dirent_t *ent struct iatt parentbuf = {0}; int ret = 0; dict_t *xattr_rsp = NULL; + dict_t *xattr_req = NULL; + + xattr_req = dict_new (); + if (!xattr_req) { + errno = ENOMEM; + ret = -1; + goto out; + } + + ret = dict_set_int32 (xattr_req, "allow-sh-for-running-transaction", 1); gf_log (this->name, GF_LOG_DEBUG, "lookup %s", child->path); - ret = syncop_lookup (this, child, NULL, + ret = syncop_lookup (this, child, xattr_req, iattr, &xattr_rsp, &parentbuf); _crawl_post_sh_action (this, parent, child, ret, errno, xattr_rsp, crawl_data); @@ -420,6 +654,9 @@ _self_heal_entry (xlator_t *this, afr_crawl_data_t *crawl_data, gf_dirent_t *ent if (ret == 0) ret = _link_inode_update_loc (this, child, iattr); +out: + if (xattr_req) + dict_unref(xattr_req); return ret; } @@ -528,12 +765,20 @@ _do_crawl_op_on_local_subvols (xlator_t *this, afr_crawl_type_t crawl, status = "Started self-heal"; _do_self_heal_on_subvol (this, i, crawl); - } else if (output) { + } else if (output && (op == INFO)) { status = ""; afr_start_crawl (this, i, INDEX, _add_summary_to_dict, output, _gf_false, 0, NULL); + } else if (output && + (op == STATISTICS_TO_BE_HEALED)) { + status = ""; + afr_start_crawl (this, i, + INDEX_TO_BE_HEALED, + _count_hard_links_under_base_indices_dir, + output, _gf_false, + 0, NULL); } } if (output) { @@ -567,8 +812,103 @@ _get_index_summary_on_local_subvols (xlator_t *this, dict_t *output) return _do_crawl_op_on_local_subvols (this, INDEX, INFO, output); } +void +afr_fill_completed_crawl_statistics_to_dict (xlator_t *this, dict_t *dict) +{ + afr_private_t *priv = NULL; + afr_self_heald_t *shd = NULL; + int i = 0; + priv = this->private; + shd= &priv->shd; + for (i = 0; i < priv->child_count; i++) { + if (shd->pos[i] != AFR_POS_LOCAL) + continue; + _add_statistics_to_dict (this, dict, i); + } + + return ; +} + +static void +reset_crawl_event (shd_crawl_event_t *crawl_event) +{ + crawl_event->healed_count = 0; + crawl_event->split_brain_count = 0; + crawl_event->heal_failed_count = 0; + GF_FREE (crawl_event->start_time_str); + crawl_event->start_time_str = NULL; + crawl_event->end_time_str = NULL; + crawl_event->crawl_type = NULL; + crawl_event->crawl_inprogress = _gf_false; + return; +} + +static void +afr_copy_crawl_event_struct (shd_crawl_event_t *src, shd_crawl_event_t *dst) +{ + dst->healed_count = src->healed_count; + dst->split_brain_count = src->split_brain_count; + dst->heal_failed_count = src->heal_failed_count; + dst->start_time_str = gf_strdup (src->start_time_str); + dst->end_time_str = "Crawl is already in progress"; + dst->crawl_type = src->crawl_type; + dst->crawl_inprogress = _gf_true; + return; +} + +static int +afr_fill_crawl_statistics_of_running_crawl(xlator_t *this, dict_t *dict) +{ + shd_crawl_event_t *evnt = NULL; + int ret = 0; + afr_private_t *priv = NULL; + afr_self_heald_t *shd = NULL; + int i = 0; + priv = this->private; + shd = &priv->shd; + + evnt = GF_CALLOC (1, sizeof (shd_crawl_event_t), + gf_afr_mt_shd_crawl_event_t); + if (!evnt) { + ret = -1; + goto out; + } + LOCK (&priv->lock); + { + for (i = 0; i < priv->child_count; i++) { + if (shd->pos[i] != AFR_POS_LOCAL) + continue; + + reset_crawl_event (evnt); + + if (!shd->crawl_events[i]) { + continue; + } + + afr_copy_crawl_event_struct (shd->crawl_events[i], + evnt); + _add_crawl_stats_to_dict (this, dict, i, evnt, NULL); + + } + } + UNLOCK (&priv->lock); + reset_crawl_event (evnt); + GF_FREE (evnt); + +out: + return ret; +} + +static int +_add_local_subvols_crawl_statistics_to_dict (xlator_t *this, dict_t *dict) +{ + int ret = 0; + afr_fill_completed_crawl_statistics_to_dict (this, dict); + ret = afr_fill_crawl_statistics_of_running_crawl (this, dict); + return ret; +} int -_add_all_subvols_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict) +_add_local_subvols_eh_to_dict (xlator_t *this, eh_t *eh, dict_t *dict) { afr_private_t *priv = NULL; afr_self_heald_t *shd = NULL; @@ -618,16 +958,25 @@ afr_xl_op (xlator_t *this, dict_t *input, dict_t *output) ret = 0; break; case GF_AFR_OP_HEALED_FILES: - ret = _add_all_subvols_eh_to_dict (this, shd->healed, output); + ret = _add_local_subvols_eh_to_dict (this, shd->healed, output); break; case GF_AFR_OP_HEAL_FAILED_FILES: - ret = _add_all_subvols_eh_to_dict (this, shd->heal_failed, + ret = _add_local_subvols_eh_to_dict (this, shd->heal_failed, output); break; case GF_AFR_OP_SPLIT_BRAIN_FILES: - ret = _add_all_subvols_eh_to_dict (this, shd->split_brain, + ret = _add_local_subvols_eh_to_dict (this, shd->split_brain, output); break; + case GF_AFR_OP_STATISTICS: + ret = _add_local_subvols_crawl_statistics_to_dict (this, output); + break; + case GF_AFR_OP_STATISTICS_HEAL_COUNT: + case GF_AFR_OP_STATISTICS_HEAL_COUNT_PER_REPLICA: + ret = _do_crawl_op_on_local_subvols (this, INDEX_TO_BE_HEALED, + STATISTICS_TO_BE_HEALED, + output); + break; default: gf_log (this->name, GF_LOG_ERROR, "Unknown set op %d", op); break; @@ -642,7 +991,7 @@ afr_poll_self_heal (void *data) { afr_private_t *priv = NULL; afr_self_heald_t *shd = NULL; - struct timeval timeout = {0}; + struct timespec timeout = {0}; xlator_t *this = NULL; long child = (long)data; gf_timer_t *old_timer = NULL; @@ -666,7 +1015,7 @@ afr_poll_self_heal (void *data) if (shd->enabled && (shd->pos[child] == AFR_POS_LOCAL)) _do_self_heal_on_subvol (this, child, INDEX); timeout.tv_sec = shd->timeout; - timeout.tv_usec = 0; + timeout.tv_nsec = 0; //notify and previous timer should be synchronized. LOCK (&priv->lock); { @@ -800,6 +1149,7 @@ afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data, afr_private_t *priv = NULL; dict_t *xattr = NULL; void *index_gfid = NULL; + void *base_indices_holder_vgfid = NULL; loc_t rootloc = {0}; struct iatt iattr = {0}; struct iatt parent = {0}; @@ -809,7 +1159,7 @@ afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data, priv = this->private; if (crawl_data->crawl == FULL) { afr_build_root_loc (this, dirloc); - } else { + } else if (crawl_data->crawl == INDEX) { afr_build_root_loc (this, &rootloc); ret = syncop_getxattr (readdir_xl, &rootloc, &xattr, GF_XATTROP_INDEX_GFID); @@ -843,6 +1193,47 @@ afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data, ret = _link_inode_update_loc (this, dirloc, &iattr); if (ret) goto out; + } else if (crawl_data->crawl == INDEX_TO_BE_HEALED) { + afr_build_root_loc (this, &rootloc); + ret = syncop_getxattr (readdir_xl, &rootloc, &xattr, + GF_BASE_INDICES_HOLDER_GFID); + if (ret < 0) + goto out; + ret = dict_get_ptr (xattr, GF_BASE_INDICES_HOLDER_GFID, + &base_indices_holder_vgfid); + if (ret < 0) { + gf_log (this->name, GF_LOG_ERROR, "index gfid empty " + "on %s", readdir_xl->name); + ret = -1; + goto out; + } + if (!base_indices_holder_vgfid) { + gf_log (this->name, GF_LOG_ERROR, "Base indices holder" + "virtual gfid is null on %s", readdir_xl->name); + ret = -1; + goto out; + } + uuid_copy (dirloc->gfid, base_indices_holder_vgfid); + dirloc->path = ""; + dirloc->inode = inode_new (priv->root_inode->table); + ret = syncop_lookup (readdir_xl, dirloc, NULL, &iattr, NULL, + &parent); + if (ret < 0) { + if (errno != ENOENT) { + gf_log (this->name, GF_LOG_ERROR, "lookup " + "failed for base_indices_holder dir" + " on %s - (%s)", readdir_xl->name, + strerror (errno)); + + } else { + gf_log (this->name, GF_LOG_ERROR, "base_indices" + "_holder is not yet created."); + } + goto out; + } + ret = _link_inode_update_loc (this, dirloc, &iattr); + if (ret) + goto out; } ret = 0; out: @@ -907,6 +1298,16 @@ afr_crawl_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, priv = this->private; if (crawl_data->crawl == FULL) { ret = afr_build_child_loc (this, child, parent, entry->d_name); + } else if (crawl_data->crawl == INDEX_TO_BE_HEALED) { + ret = _build_index_loc (this, child, entry->d_name, parent); + if (ret) + goto out; + child->inode = inode_new (priv->root_inode->table); + if (!child->inode) { + ret = -1; + goto out; + } + child->path = NULL; } else { child->inode = inode_new (priv->root_inode->table); if (!child->inode) @@ -956,10 +1357,14 @@ _process_entries (xlator_t *this, loc_t *parentloc, gf_dirent_t *entries, ret = crawl_data->process_entry (this, crawl_data, entry, &entry_loc, parentloc, &iattr); - if (ret) + if (crawl_data->crawl == INDEX_TO_BE_HEALED && ret) { + goto out; + } else if (ret) { continue; + } - if (crawl_data->crawl == INDEX) + if ((crawl_data->crawl == INDEX) || + (crawl_data->crawl == INDEX_TO_BE_HEALED)) continue; if (!IA_ISDIR (iattr.ia_type)) @@ -974,6 +1379,10 @@ _process_entries (xlator_t *this, loc_t *parentloc, gf_dirent_t *entries, } ret = 0; out: + if ((crawl_data->crawl == INDEX_TO_BE_HEALED) && ret) { + gf_log (this->name, GF_LOG_ERROR,"Failed to get the hardlink " + "count"); + } loc_wipe (&entry_loc); return ret; } @@ -1021,6 +1430,9 @@ _crawl_directory (fd_t *fd, loc_t *loc, afr_crawl_data_t *crawl_data) ret = _process_entries (this, loc, &entries, &offset, crawl_data); + if ((ret < 0) && (crawl_data->crawl == INDEX_TO_BE_HEALED)) { + goto out; + } gf_dirent_free (&entries); free_entries = _gf_false; } @@ -1126,8 +1538,13 @@ afr_dir_crawl (void *data) goto out; ret = afr_crawl_opendir (this, crawl_data, &fd, &dirloc); - if (ret) + if (ret) { + if (crawl_data->crawl == INDEX_TO_BE_HEALED) { + gf_log (this->name, GF_LOG_ERROR, "Failed to open base_" + "indices_holder"); + } goto out; + } ret = _crawl_directory (fd, &dirloc, crawl_data); if (ret) @@ -1141,12 +1558,102 @@ afr_dir_crawl (void *data) out: if (fd) fd_unref (fd); - if (crawl_data->crawl == INDEX) + if ((crawl_data->crawl == INDEX) || + (crawl_data->crawl == INDEX_TO_BE_HEALED )) dirloc.path = NULL; loc_wipe (&dirloc); return ret; } +char * +get_crawl_type_in_string (afr_crawl_type_t crawl) +{ + char *index = "INDEX"; + char *full = "FULL"; + char *crawl_type = NULL; + + if (crawl == INDEX){ + crawl_type = index; + } else if (crawl == FULL) { + crawl_type = full; + } + + return crawl_type; +} + +static int +afr_allocate_crawl_event (xlator_t *this, int child, afr_crawl_type_t crawl) +{ + afr_private_t *priv = NULL; + afr_self_heald_t *shd = NULL; + int ret = 0; + shd_crawl_event_t *crawl_event = NULL; + time_t get_time = 0; + + priv = this->private; + shd = &priv->shd; + + crawl_event = GF_CALLOC (sizeof (shd_crawl_event_t), 1, + gf_afr_mt_shd_crawl_event_t); + if (!crawl_event) { + ret = -1; + goto out; + } + + get_time = time(NULL); + if (get_time == ((time_t)-1)) { + ret = -1; + goto out; + } + + crawl_event->start_time_str = gf_strdup (ctime(&get_time)); + + crawl_event->crawl_type = get_crawl_type_in_string (crawl); + if (!crawl_event->crawl_type) { + ret = -1; + goto out; + } + LOCK (&priv->lock); + { + shd->crawl_events[child] = crawl_event; + } + UNLOCK (&priv->lock); + ret = 0; +out: + return ret; + +} + +static int +afr_put_crawl_event_in_eh (xlator_t *this, int child) +{ + afr_private_t *priv = NULL; + afr_self_heald_t *shd = NULL; + int ret = 0; + time_t get_time = 0; + shd_crawl_event_t **crawl_event = NULL; + + priv = this->private; + shd = &priv->shd; + + get_time = time(NULL); + if (get_time == ((time_t)-1)) { + ret = -1; + goto out; + } + crawl_event = (shd_crawl_event_t**)shd->crawl_events; + LOCK (&priv->lock); + { + crawl_event[child]->end_time_str = gf_strdup (ctime(&get_time)); + ret = eh_save_history (shd->statistics[child], + crawl_event[child]); + crawl_event[child] = NULL; + } + UNLOCK (&priv->lock); +out: + return ret; +} + static int afr_dir_exclusive_crawl (void *data) { @@ -1182,7 +1689,15 @@ afr_dir_exclusive_crawl (void *data) } do { + ret = afr_allocate_crawl_event (this, child, crawl_data->crawl); + if (ret) + goto out; afr_dir_crawl (data); + + ret = afr_put_crawl_event_in_eh (this, child); + if (ret < 0) + goto out; + LOCK (&priv->lock); { if (shd->pending[child] != NONE) { @@ -1270,4 +1785,3 @@ afr_set_root_gfid (dict_t *dict) return ret; } - diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h index 32a8aaca5..e0c083754 100644 --- a/xlators/cluster/afr/src/afr-self-heald.h +++ b/xlators/cluster/afr/src/afr-self-heald.h @@ -29,6 +29,19 @@ typedef struct afr_crawl_data_ { struct iatt *iattr); } afr_crawl_data_t; +typedef struct crawl_event_stats_ { + uint64_t healed_count; + uint64_t split_brain_count; + uint64_t heal_failed_count; + char *start_time_str; + char *end_time_str; + char *crawl_type; + gf_boolean_t crawl_inprogress; +} shd_crawl_event_t; + +void _destroy_crawl_event_data (void *data); +void _destroy_shd_event_data (void *data); + typedef int (*process_entry_cbk_t) (xlator_t *this, afr_crawl_data_t *crawl_data, gf_dirent_t *entry, loc_t *child, loc_t *parent, struct iatt *iattr); diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c index 817645505..20306e469 100644 --- a/xlators/cluster/afr/src/afr-transaction.c +++ b/xlators/cluster/afr/src/afr-transaction.c @@ -586,6 +586,59 @@ afr_txn_nothing_failed (call_frame_t *frame, xlator_t *this) return _gf_true; } +static void +afr_dir_fop_handle_all_fop_failures (call_frame_t *frame) +{ + xlator_t *this = NULL; + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + + this = frame->this; + local = frame->local; + priv = this->private; + + if ((local->transaction.type != AFR_ENTRY_TRANSACTION) && + (local->transaction.type != AFR_ENTRY_RENAME_TRANSACTION)) + return; + + if (local->op_ret >= 0) + goto out; + + __mark_all_success (local->pending, priv->child_count, + local->transaction.type); +out: + return; +} + +static void +afr_data_handle_quota_errors (call_frame_t *frame, xlator_t *this) +{ + int i = 0; + afr_private_t *priv = NULL; + afr_local_t *local = NULL; + gf_boolean_t all_quota_failures = _gf_false; + + local = frame->local; + priv = this->private; + if (local->transaction.type != AFR_DATA_TRANSACTION) + return; + /* + * Idea is to not leave the file in FOOL-FOOL scenario in case on + * all the bricks data transaction failed with EDQUOT to avoid + * increasing un-necessary load of self-heals in the system. + */ + all_quota_failures = _gf_true; + for (i = 0; i < priv->child_count; i++) { + if (local->transaction.pre_op[i] && + (local->child_errno[i] != EDQUOT)) { + all_quota_failures = _gf_false; + break; + } + } + if (all_quota_failures) + __mark_all_success (local->pending, priv->child_count, + local->transaction.type); +} int afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this) @@ -608,6 +661,9 @@ afr_changelog_post_op_now (call_frame_t *frame, xlator_t *this) local->transaction.pre_op, local->transaction.type); + afr_data_handle_quota_errors (frame, this); + afr_dir_fop_handle_all_fop_failures (frame); + if (local->fd) afr_transaction_rm_stale_children (frame, this, local->fd->inode, @@ -1338,10 +1394,21 @@ afr_are_multiple_fds_opened (inode_t *inode, xlator_t *this) } gf_boolean_t +afr_any_fops_failed (afr_local_t *local, afr_private_t *priv) +{ + if (local->success_count != priv->child_count) + return _gf_true; + return _gf_false; +} + +gf_boolean_t is_afr_delayed_changelog_post_op_needed (call_frame_t *frame, xlator_t *this) { afr_local_t *local = NULL; gf_boolean_t res = _gf_false; + afr_private_t *priv = NULL; + + priv = this->private; local = frame->local; if (!local) @@ -1350,6 +1417,10 @@ is_afr_delayed_changelog_post_op_needed (call_frame_t *frame, xlator_t *this) if (!local->delayed_post_op) goto out; + //Mark pending changelog ASAP + if (afr_any_fops_failed (local, priv)) + goto out; + if (local->fd && afr_are_multiple_fds_opened (local->fd->inode, this)) goto out; @@ -1622,13 +1693,13 @@ afr_changelog_post_op_safe (call_frame_t *frame, xlator_t *this) } - void +void afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd, - call_stub_t *stub) + call_stub_t *stub) { afr_fd_ctx_t *fd_ctx = NULL; call_frame_t *prev_frame = NULL; - struct timeval delta = {0, }; + struct timespec delta = {0, }; afr_private_t *priv = NULL; afr_local_t *local = NULL; @@ -1639,7 +1710,7 @@ afr_delayed_changelog_post_op (xlator_t *this, call_frame_t *frame, fd_t *fd, goto out; delta.tv_sec = priv->post_op_delay_secs; - delta.tv_usec = 0; + delta.tv_nsec = 0; pthread_mutex_lock (&fd_ctx->delay_lock); { @@ -1669,7 +1740,7 @@ out: } - void +void afr_changelog_post_op (call_frame_t *frame, xlator_t *this) { afr_local_t *local = NULL; @@ -1691,14 +1762,14 @@ afr_changelog_post_op (call_frame_t *frame, xlator_t *this) The @stub gets saved in @local and gets resumed in afr_local_cleanup() */ - void +void afr_delayed_changelog_wake_resume (xlator_t *this, fd_t *fd, call_stub_t *stub) { afr_delayed_changelog_post_op (this, NULL, fd, stub); } - void +void afr_delayed_changelog_wake_up (xlator_t *this, fd_t *fd) { afr_delayed_changelog_post_op (this, NULL, fd, NULL); @@ -1748,8 +1819,9 @@ afr_transaction_resume (call_frame_t *frame, xlator_t *this) * afr_transaction_fop_failed - inform that an fop failed */ - void -afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this, int child_index) +void +afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this, + int child_index) { afr_local_t * local = NULL; afr_private_t * priv = NULL; diff --git a/xlators/cluster/afr/src/afr-transaction.h b/xlators/cluster/afr/src/afr-transaction.h index 108131276..fa626fd0d 100644 --- a/xlators/cluster/afr/src/afr-transaction.h +++ b/xlators/cluster/afr/src/afr-transaction.h @@ -43,4 +43,9 @@ afr_delayed_changelog_wake_up (xlator_t *this, fd_t *fd); void __mark_all_success (int32_t *pending[], int child_count, afr_transaction_type type); +gf_boolean_t +afr_any_fops_failed (afr_local_t *local, afr_private_t *priv); + +gf_boolean_t +afr_txn_nothing_failed (call_frame_t *frame, xlator_t *this); #endif /* __TRANSACTION_H__ */ diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c index 4338b799a..c724eb2ae 100644 --- a/xlators/cluster/afr/src/afr.c +++ b/xlators/cluster/afr/src/afr.c @@ -388,7 +388,8 @@ init (xlator_t *this) i++; } - ret = gf_asprintf (&priv->sh_domain, "%s-self-heal", this->name); + ret = gf_asprintf (&priv->sh_domain, AFR_SH_DATA_DOMAIN_FMT, + this->name); if (-1 == ret) { ret = -ENOMEM; goto out; @@ -438,15 +439,18 @@ init (xlator_t *this) if (!priv->shd.timer) goto out; - priv->shd.healed = eh_new (AFR_EH_HEALED_LIMIT, _gf_false); + priv->shd.healed = eh_new (AFR_EH_HEALED_LIMIT, _gf_false, + _destroy_shd_event_data); if (!priv->shd.healed) goto out; - priv->shd.heal_failed = eh_new (AFR_EH_HEAL_FAIL_LIMIT, _gf_false); + priv->shd.heal_failed = eh_new (AFR_EH_HEAL_FAIL_LIMIT, _gf_false, + _destroy_shd_event_data); if (!priv->shd.heal_failed) goto out; - priv->shd.split_brain = eh_new (AFR_EH_SPLIT_BRAIN_LIMIT, _gf_false); + priv->shd.split_brain = eh_new (AFR_EH_SPLIT_BRAIN_LIMIT, _gf_false, + _destroy_shd_event_data); if (!priv->shd.split_brain) goto out; @@ -456,7 +460,9 @@ init (xlator_t *this) priv->root_inode = inode_ref (this->itable->root); GF_OPTION_INIT ("node-uuid", priv->shd.node_uuid, str, out); GF_OPTION_INIT ("heal-timeout", priv->shd.timeout, int32, out); - + ret = afr_initialise_statistics (this); + if (ret) + goto out; ret = 0; out: return ret; @@ -493,6 +499,7 @@ struct xlator_fops fops = { .fentrylk = afr_fentrylk, .fallocate = afr_fallocate, .discard = afr_discard, + .zerofill = afr_zerofill, /* inode read */ .access = afr_access, diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h index 49d281aca..21064db58 100644 --- a/xlators/cluster/afr/src/afr.h +++ b/xlators/cluster/afr/src/afr.h @@ -28,6 +28,7 @@ #define AFR_XATTR_PREFIX "trusted.afr" #define AFR_PATHINFO_HEADER "REPLICATE:" #define AFR_SH_READDIR_SIZE_KEY "self-heal-readdir-size" +#define AFR_SH_DATA_DOMAIN_FMT "%s:self-heal" #define AFR_LOCKEE_COUNT_MAX 3 #define AFR_DOM_COUNT_MAX 3 @@ -93,21 +94,24 @@ typedef struct afr_inode_ctx_ { typedef enum { NONE, INDEX, + INDEX_TO_BE_HEALED, FULL, } afr_crawl_type_t; typedef struct afr_self_heald_ { - gf_boolean_t enabled; - gf_boolean_t iamshd; - afr_crawl_type_t *pending; - gf_boolean_t *inprogress; - afr_child_pos_t *pos; - gf_timer_t **timer; - eh_t *healed; - eh_t *heal_failed; - eh_t *split_brain; - char *node_uuid; - int timeout; + gf_boolean_t enabled; + gf_boolean_t iamshd; + afr_crawl_type_t *pending; + gf_boolean_t *inprogress; + afr_child_pos_t *pos; + gf_timer_t **timer; + eh_t *healed; + eh_t *heal_failed; + eh_t *split_brain; + eh_t **statistics; + void **crawl_events; + char *node_uuid; + int timeout; } afr_self_heald_t; typedef struct _afr_private { @@ -180,6 +184,7 @@ typedef enum { AFR_SELF_HEAL_NOT_ATTEMPTED, AFR_SELF_HEAL_STARTED, AFR_SELF_HEAL_FAILED, + AFR_SELF_HEAL_SYNC_BEGIN, } afr_self_heal_status; typedef struct { @@ -294,6 +299,8 @@ struct afr_self_heal_ { afr_post_remove_call_t post_remove_call; char *data_sh_info; + char *metadata_sh_info; + loc_t parent_loc; call_frame_t *orig_frame; call_frame_t *old_loop_frame; @@ -490,15 +497,23 @@ typedef struct _afr_local { int optimistic_change_log; gf_boolean_t delayed_post_op; + /* Is the current writev() going to perform a stable write? i.e, is fd->flags or @flags writev param have O_SYNC or O_DSYNC? */ - gf_boolean_t stable_write; + gf_boolean_t stable_write; + + /* This write appended to the file. Nnot necessarily O_APPEND, + just means the offset of write was at the end of file. + */ + gf_boolean_t append_write; + + int allow_sh_for_running_transaction; + - /* - This struct contains the arguments for the "continuation" - (scheme-like) of fops + /* This struct contains the arguments for the "continuation" + (scheme-like) of fops */ int op; @@ -594,7 +609,9 @@ typedef struct _afr_local { struct { struct iatt prebuf; struct iatt postbuf; + } inode_wfop; //common structure for all inode-write-fops + struct { int32_t op_ret; struct iovec *vector; @@ -605,34 +622,21 @@ typedef struct _afr_local { } writev; struct { - struct iatt prebuf; - struct iatt postbuf; - } fsync; - - struct { off_t offset; - struct iatt prebuf; - struct iatt postbuf; } truncate; struct { off_t offset; - struct iatt prebuf; - struct iatt postbuf; } ftruncate; struct { struct iatt in_buf; int32_t valid; - struct iatt preop_buf; - struct iatt postop_buf; } setattr; struct { struct iatt in_buf; int32_t valid; - struct iatt preop_buf; - struct iatt postop_buf; } fsetattr; struct { @@ -699,17 +703,21 @@ typedef struct _afr_local { int32_t mode; off_t offset; size_t len; - struct iatt prebuf; - struct iatt postbuf; } fallocate; struct { off_t offset; size_t len; - struct iatt prebuf; - struct iatt postbuf; } discard; + struct { + off_t offset; + size_t len; + struct iatt prebuf; + struct iatt postbuf; + } zerofill; + + } cont; struct { @@ -1162,6 +1170,27 @@ afr_xattr_array_destroy (dict_t **xattr, unsigned int child_count); } while (0); +#define AFR_SBRAIN_MSG "Failed on %s as split-brain is seen. Returning EIO." + +#define AFR_SBRAIN_CHECK_FD(fd, label) do { \ + if (fd->inode && afr_is_split_brain (this, fd->inode)) { \ + op_errno = EIO; \ + gf_log (this->name, GF_LOG_WARNING, \ + AFR_SBRAIN_MSG ,uuid_utoa (fd->inode->gfid)); \ + goto label; \ + } \ +} while (0) + +#define AFR_SBRAIN_CHECK_LOC(loc, label) do { \ + if (loc->inode && afr_is_split_brain (this, loc->inode)) { \ + op_errno = EIO; \ + loc_path (loc, NULL); \ + gf_log (this->name, GF_LOG_WARNING, \ + AFR_SBRAIN_MSG , loc->path); \ + goto label; \ + } \ +} while (0) + int afr_fd_report_unstable_write (xlator_t *this, fd_t *fd); diff --git a/xlators/cluster/afr/src/pump.c b/xlators/cluster/afr/src/pump.c index c7b858691..a7f72fb30 100644 --- a/xlators/cluster/afr/src/pump.c +++ b/xlators/cluster/afr/src/pump.c @@ -2538,7 +2538,7 @@ init (xlator_t *this) goto out; } - pump_priv->env = syncenv_new (0); + pump_priv->env = this->ctx->env; if (!pump_priv->env) { gf_log (this->name, GF_LOG_ERROR, "Could not create new sync-environment"); @@ -2579,9 +2579,6 @@ fini (xlator_t *this) if (!pump_priv) goto afr_priv; - if (pump_priv->env) - syncenv_destroy (pump_priv->env); - GF_FREE (pump_priv->resume_path); LOCK_DESTROY (&pump_priv->resume_path_lock); LOCK_DESTROY (&pump_priv->pump_state_lock); diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c index 8b34d1a73..8f61339e6 100644 --- a/xlators/cluster/dht/src/dht-common.c +++ b/xlators/cluster/dht/src/dht-common.c @@ -22,6 +22,7 @@ #include "dht-common.h" #include "defaults.h" #include "byte-order.h" +#include "glusterfs-acl.h" #include <sys/time.h> #include <libgen.h> @@ -154,7 +155,6 @@ dht_discover_complete (xlator_t *this, call_frame_t *discover_frame) int ret = -1; dht_layout_t *layout = NULL; dht_conf_t *conf = NULL; - uint32_t missing = 0; local = discover_frame->local; layout = local->layout; @@ -191,33 +191,20 @@ dht_discover_complete (xlator_t *this, call_frame_t *discover_frame) goto out; } } else { - ret = dht_layout_normalize (this, &local->loc, layout, - &missing); - if (ret < 0) { - gf_log (this->name, GF_LOG_DEBUG, - "normalizing failed on %s (internal error)", - local->loc.path); - op_errno = EIO; - goto out; - } - if (missing == conf->subvolume_cnt) { + ret = dht_layout_normalize (this, &local->loc, layout); + if ((ret < 0) || ((ret > 0) && (local->op_ret != 0))) { + /* either the layout is incorrect or the directory is + * not found even in one subvolume. + */ gf_log (this->name, GF_LOG_DEBUG, - "normalizing failed on %s, ENOENT errors: %u)", - local->loc.path, missing); - op_errno = ENOENT; - goto out; - } - if (ret != 0) { - gf_log (this->name, GF_LOG_WARNING, "normalizing failed on %s " - "(overlaps/holes present)", local->loc.path); - /* We may need to do the lookup again */ - /* in discover call, parent is not know, and basename - * of entry is also not available. Without which we - * cannot build a layout correctly to heal it. Hence - * returning ESTALE */ - op_errno = ESTALE; - goto out; + "(overlaps/holes present: %s, " + "ENOENT errors: %d)", local->loc.path, + (ret < 0) ? "yes" : "no", (ret > 0) ? ret : 0); + if ((ret > 0) && (ret == conf->subvolume_cnt)) { + op_errno = ESTALE; + goto out; + } } if (local->inode) @@ -421,7 +408,6 @@ dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this, dht_layout_t *layout = NULL; int ret = -1; int is_dir = 0; - uint32_t missing = 0; GF_VALIDATE_OR_GOTO ("dht", frame, out); GF_VALIDATE_OR_GOTO ("dht", this, out); @@ -455,7 +441,7 @@ dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this, op_ret, op_errno, xattr); if (op_ret == -1) { - local->op_errno = ENOENT; + local->op_errno = op_errno; gf_log (this->name, GF_LOG_DEBUG, "lookup of %s on %s returned error (%s)", local->loc.path, prev->this->name, @@ -502,16 +488,9 @@ unlock: } if (local->op_ret == 0) { - ret = dht_layout_normalize (this, &local->loc, layout, - &missing); - - /* - * Arguably, we shouldn't do self-heal just because - * bricks are missing as long as there are no other - * anomalies. For now, though, just preserve the - * existing behavior. - */ - if ((ret != 0) || (missing != 0)) { + ret = dht_layout_normalize (this, &local->loc, layout); + + if (ret != 0) { gf_log (this->name, GF_LOG_DEBUG, "fixing assignment on %s", local->loc.path); @@ -3065,7 +3044,7 @@ dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, list_for_each_entry (orig_entry, (&orig_entries->list), list) { next_offset = orig_entry->d_off; if (check_is_dir (NULL, (&orig_entry->d_stat), NULL) && - (prev->this != dht_first_up_subvol (this))) { + (prev->this != local->first_up_subvol)) { continue; } if (check_is_linkfile (NULL, (&orig_entry->d_stat), @@ -3147,13 +3126,16 @@ done: } if (conf->readdir_optimize == _gf_true) { - if (next_subvol != dht_first_up_subvol (this)) { + if (next_subvol != local->first_up_subvol) { ret = dict_set_int32 (local->xattr, GF_READDIR_SKIP_DIRS, 1); if (ret) gf_log (this->name, GF_LOG_ERROR, "dict set failed"); - } + } else { + dict_del (local->xattr, + GF_READDIR_SKIP_DIRS); + } } STACK_WIND (frame, dht_readdirp_cbk, @@ -3299,6 +3281,7 @@ dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, local->fd = fd_ref (fd); local->size = size; local->xattr_req = (dict)? dict_ref (dict) : NULL; + local->first_up_subvol = dht_first_up_subvol (this); dht_deitransform (this, yoff, &xvol, (uint64_t *)&xoff); @@ -3317,13 +3300,16 @@ dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, "failed to set '%s' key", conf->link_xattr_name); if (conf->readdir_optimize == _gf_true) { - if (xvol != dht_first_up_subvol (this)) { + if (xvol != local->first_up_subvol) { ret = dict_set_int32 (local->xattr, GF_READDIR_SKIP_DIRS, 1); if (ret) gf_log (this->name, GF_LOG_ERROR, "Dict set failed"); + } else { + dict_del (local->xattr, + GF_READDIR_SKIP_DIRS); } } } @@ -5187,8 +5173,8 @@ unlock: * not need to handle CHILD_DOWN event here. */ if (conf->defrag) { - ret = pthread_create (&conf->defrag->th, NULL, - gf_defrag_start, this); + ret = gf_thread_create (&conf->defrag->th, NULL, + gf_defrag_start, this); if (ret) { conf->defrag = NULL; GF_FREE (conf->defrag); diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h index 8a62b9eec..5ccd66799 100644 --- a/xlators/cluster/dht/src/dht-common.h +++ b/xlators/cluster/dht/src/dht-common.h @@ -183,6 +183,7 @@ struct dht_local { xlator_t *link_subvol; struct dht_rebalance_ rebalance; + xlator_t *first_up_subvol; }; typedef struct dht_local dht_local_t; @@ -211,6 +212,10 @@ enum gf_defrag_status_t { GF_DEFRAG_STATUS_STOPPED, GF_DEFRAG_STATUS_COMPLETE, GF_DEFRAG_STATUS_FAILED, + GF_DEFRAG_STATUS_LAYOUT_FIX_STARTED, + GF_DEFRAG_STATUS_LAYOUT_FIX_STOPPED, + GF_DEFRAG_STATUS_LAYOUT_FIX_COMPLETE, + GF_DEFRAG_STATUS_LAYOUT_FIX_FAILED, }; typedef enum gf_defrag_status_t gf_defrag_status_t; @@ -395,26 +400,18 @@ typedef enum { } while (0) #define is_greater_time(a, an, b, bn) (((a) < (b)) || (((a) == (b)) && ((an) < (bn)))) - -dht_layout_t *dht_layout_new (xlator_t *this, int cnt); -dht_layout_t *dht_layout_get (xlator_t *this, inode_t *inode); -dht_layout_t *dht_layout_for_subvol (xlator_t *this, xlator_t *subvol); -xlator_t *dht_layout_search (xlator_t *this, dht_layout_t *layout, - const char *name); -int dht_layout_normalize (xlator_t *this, loc_t *loc, - dht_layout_t *layout, - uint32_t *missing_p); -int dht_layout_anomalies (xlator_t *this, loc_t *loc, - dht_layout_t *layout, - uint32_t *holes_p, - uint32_t *overlaps_p, - uint32_t *missing_p, - uint32_t *down_p, - uint32_t *misc_p, - uint32_t *no_space_p); -int dht_layout_dir_mismatch (xlator_t *this, dht_layout_t *layout, - xlator_t *subvol, loc_t *loc, - dict_t *xattr); +dht_layout_t *dht_layout_new (xlator_t *this, int cnt); +dht_layout_t *dht_layout_get (xlator_t *this, inode_t *inode); +dht_layout_t *dht_layout_for_subvol (xlator_t *this, xlator_t *subvol); +xlator_t *dht_layout_search (xlator_t *this, dht_layout_t *layout, + const char *name); +int dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout); +int dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout, + uint32_t *holes_p, uint32_t *overlaps_p, + uint32_t *missing_p, uint32_t *down_p, + uint32_t *misc_p, uint32_t *no_space_p); +int dht_layout_dir_mismatch (xlator_t *this, dht_layout_t *layout, + xlator_t *subvol, loc_t *loc, dict_t *xattr); xlator_t *dht_linkfile_subvol (xlator_t *this, inode_t *inode, struct iatt *buf, dict_t *xattr); @@ -698,6 +695,8 @@ int32_t dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode, off_t offset, size_t len, dict_t *xdata); int32_t dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, size_t len, dict_t *xdata); +int32_t dht_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, + off_t offset, size_t len, dict_t *xdata); int32_t dht_init (xlator_t *this); void dht_fini (xlator_t *this); @@ -782,4 +781,7 @@ dht_priv_dump (xlator_t *this); int32_t dht_inodectx_dump (xlator_t *this, inode_t *inode); +int +dht_inode_ctx_get1 (xlator_t *this, inode_t *inode, xlator_t **subvol); + #endif/* _DHT_H */ diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c index 1cf343d24..311a48112 100644 --- a/xlators/cluster/dht/src/dht-helper.c +++ b/xlators/cluster/dht/src/dht-helper.c @@ -18,6 +18,28 @@ #include "xlator.h" #include "dht-common.h" +static inline int +dht_inode_ctx_set1 (xlator_t *this, inode_t *inode, xlator_t *subvol) +{ + uint64_t tmp_subvol = 0; + + tmp_subvol = (long)subvol; + return inode_ctx_set1 (inode, this, &tmp_subvol); +} + +int +dht_inode_ctx_get1 (xlator_t *this, inode_t *inode, xlator_t **subvol) +{ + int ret = -1; + uint64_t tmp_subvol = 0; + + ret = inode_ctx_get1 (inode, this, &tmp_subvol); + if (tmp_subvol && subvol) + *subvol = (xlator_t *)tmp_subvol; + + return ret; +} + int dht_frame_return (call_frame_t *frame) @@ -340,20 +362,6 @@ out: return local; } - -char * -basestr (const char *str) -{ - char *basestr = NULL; - - basestr = strrchr (str, '/'); - if (basestr) - basestr ++; - - return basestr; -} - - xlator_t * dht_first_up_subvol (xlator_t *this) { @@ -727,6 +735,10 @@ dht_migration_complete_check_task (void *data) loc_t tmp_loc = {0,}; char *path = NULL; dht_conf_t *conf = NULL; + inode_t *inode = NULL; + fd_t *iter_fd = NULL; + uint64_t tmp_subvol = 0; + int open_failed = 0; this = THIS; frame = data; @@ -738,6 +750,8 @@ dht_migration_complete_check_task (void *data) if (!local->loc.inode && !local->fd) goto out; + inode = (!local->fd) ? local->loc.inode : local->fd->inode; + /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr * as root:root. If a fd is already open, access check wont be done*/ @@ -800,10 +814,7 @@ dht_migration_complete_check_task (void *data) /* update inode ctx (the layout) */ dht_layout_unref (this, local->layout); - if (!local->loc.inode) - ret = dht_layout_preset (this, dst_node, local->fd->inode); - else - ret = dht_layout_preset (this, dst_node, local->loc.inode); + ret = dht_layout_preset (this, dst_node, inode); if (ret != 0) { gf_log (this->name, GF_LOG_DEBUG, "%s: could not set preset layout for subvol %s", @@ -821,10 +832,7 @@ dht_migration_complete_check_task (void *data) goto out; } - if (!local->loc.inode) - ret = dht_layout_set (this, local->fd->inode, layout); - else - ret = dht_layout_set (this, local->loc.inode, layout); + ret = dht_layout_set (this, inode, layout); if (ret) { gf_log (this->name, GF_LOG_ERROR, "%s: failed to set the new layout", @@ -835,41 +843,46 @@ dht_migration_complete_check_task (void *data) local->cached_subvol = dst_node; ret = 0; - if (!local->fd) + /* once we detect the migration complete, the inode-ctx2 is no more + required.. delete the ctx and also, it means, open() already + done on all the fd of inode */ + ret = inode_ctx_reset1 (inode, this, &tmp_subvol); + if (tmp_subvol) goto out; - /* once we detect the migration complete, the fd-ctx is no more - required.. delete the ctx */ - ret = fd_ctx_del (local->fd, this, NULL); - if (!ret) + + if (list_empty (&inode->fd_list)) goto out; /* perform open as root:root. There is window between linkfile * creation(root:root) and setattr with the correct uid/gid */ SYNCTASK_SETID(0, 0); - /* if 'local->fd' (ie, fd based operation), send a 'open()' on - destination if not already done */ - if (local->loc.inode) { - ret = syncop_open (dst_node, &local->loc, - local->fd->flags, local->fd); - } else { - tmp_loc.inode = local->fd->inode; - inode_path (local->fd->inode, NULL, &path); - if (path) - tmp_loc.path = path; - ret = syncop_open (dst_node, &tmp_loc, - local->fd->flags, local->fd); - GF_FREE (path); + /* perform 'open()' on all the fd's present on the inode */ + tmp_loc.inode = inode; + inode_path (inode, NULL, &path); + if (path) + tmp_loc.path = path; + list_for_each_entry (iter_fd, &inode->fd_list, inode_list) { + if (fd_is_anonymous (iter_fd)) + continue; + + ret = syncop_open (dst_node, &tmp_loc, + iter_fd->flags, iter_fd); + if (ret == -1) { + gf_log (this->name, GF_LOG_ERROR, "failed to open " + "the fd (%p, flags=0%o) on file %s @ %s", + iter_fd, iter_fd->flags, path, dst_node->name); + open_failed = 1; + } } + GF_FREE (path); + SYNCTASK_SETID (frame->root->uid, frame->root->gid); - if (ret == -1) { - gf_log (this->name, GF_LOG_ERROR, - "%s: failed to send open() on target file at %s", - local->loc.path, dst_node->name); + if (open_failed) { + ret = -1; goto out; } - ret = 0; out: @@ -914,6 +927,9 @@ dht_rebalance_inprogress_task (void *data) struct iatt stbuf = {0,}; loc_t tmp_loc = {0,}; dht_conf_t *conf = NULL; + inode_t *inode = NULL; + fd_t *iter_fd = NULL; + int open_failed = 0; this = THIS; frame = data; @@ -925,6 +941,8 @@ dht_rebalance_inprogress_task (void *data) if (!local->loc.inode && !local->fd) goto out; + inode = (!local->fd) ? local->loc.inode : local->fd->inode; + /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr * as root:root. If a fd is already open, access check wont be done*/ if (local->loc.inode) { @@ -976,35 +994,47 @@ dht_rebalance_inprogress_task (void *data) } ret = 0; + + if (list_empty (&inode->fd_list)) + goto done; + /* perform open as root:root. There is window between linkfile * creation(root:root) and setattr with the correct uid/gid */ SYNCTASK_SETID (0, 0); - if (local->loc.inode) { - ret = syncop_open (dst_node, &local->loc, - local->fd->flags, local->fd); - } else { - tmp_loc.inode = local->fd->inode; - inode_path (local->fd->inode, NULL, &path); - if (path) - tmp_loc.path = path; + + tmp_loc.inode = inode; + inode_path (inode, NULL, &path); + if (path) + tmp_loc.path = path; + + list_for_each_entry (iter_fd, &inode->fd_list, inode_list) { + if (fd_is_anonymous (iter_fd)) + continue; + ret = syncop_open (dst_node, &tmp_loc, - local->fd->flags, local->fd); - GF_FREE (path); + iter_fd->flags, iter_fd); + if (ret == -1) { + gf_log (this->name, GF_LOG_ERROR, "failed to send open " + "the fd (%p, flags=0%o) on file %s @ %s", + iter_fd, iter_fd->flags, path, dst_node->name); + open_failed = 1; + } } + GF_FREE (path); - if (ret == -1) { - gf_log (this->name, GF_LOG_ERROR, - "%s: failed to send open() on target file at %s", - local->loc.path, dst_node->name); + SYNCTASK_SETID (frame->root->uid, frame->root->gid); + + if (open_failed) { + ret = -1; goto out; } - SYNCTASK_SETID (frame->root->uid, frame->root->gid); - ret = fd_ctx_set (local->fd, this, (uint64_t)(long)dst_node); +done: + ret = dht_inode_ctx_set1 (this, inode, dst_node); if (ret) { gf_log (this->name, GF_LOG_ERROR, - "%s: failed to set fd-ctx target file at %s", + "%s: failed to set inode-ctx target file at %s", local->loc.path, dst_node->name); goto out; } diff --git a/xlators/cluster/dht/src/dht-inode-read.c b/xlators/cluster/dht/src/dht-inode-read.c index 10f59ab26..ece84151a 100644 --- a/xlators/cluster/dht/src/dht-inode-read.c +++ b/xlators/cluster/dht/src/dht-inode-read.c @@ -130,10 +130,11 @@ int dht_file_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, struct iatt *stbuf, dict_t *xdata) { - uint64_t tmp_subvol = 0; + xlator_t *subvol = 0; dht_local_t *local = NULL; call_frame_t *prev = NULL; int ret = -1; + inode_t *inode = NULL; GF_VALIDATE_OR_GOTO ("dht", frame, err); GF_VALIDATE_OR_GOTO ("dht", this, out); @@ -157,19 +158,20 @@ dht_file_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->op_errno = op_errno; /* Check if the rebalance phase2 is true */ if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (stbuf)) { - if (local->fd) - ret = fd_ctx_get (local->fd, this, &tmp_subvol); - if (ret) { + inode = (local->fd) ? local->fd->inode : local->loc.inode; + ret = dht_inode_ctx_get1 (this, inode, &subvol); + if (!subvol) { /* Phase 2 of migration */ local->rebalance.target_op_fn = dht_attr2; ret = dht_rebalance_complete_check (this, frame); + if (!ret) + return 0; } else { /* value is already set in fd_ctx, that means no need to check for whether its complete or not. */ dht_attr2 (this, frame, 0); - } - if (!ret) return 0; + } } out: @@ -382,6 +384,8 @@ dht_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this, { dht_local_t *local = NULL; int ret = 0; + inode_t *inode = NULL; + xlator_t *subvol = 0; local = frame->local; if (!local) { @@ -400,17 +404,18 @@ dht_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this, local->op_errno = op_errno; if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (stbuf)) { /* File would be migrated to other node */ - ret = fd_ctx_get (local->fd, this, NULL); - if (ret) { + ret = dht_inode_ctx_get1 (this, inode, &subvol); + if (!subvol) { local->rebalance.target_op_fn = dht_readv2; ret = dht_rebalance_complete_check (this, frame); + if (!ret) + return 0; } else { /* value is already set in fd_ctx, that means no need to check for whether its complete or not. */ dht_readv2 (this, frame, 0); - } - if (!ret) return 0; + } } out: @@ -616,8 +621,9 @@ int dht_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, dict_t *xdata) { - dht_local_t *local = NULL; - int ret = -1; + dht_local_t *local = NULL; + inode_t *inode = NULL; + xlator_t *subvol = 0; local = frame->local; @@ -627,8 +633,8 @@ dht_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this, goto out; /* If context is set, then send flush() it to the destination */ - ret = fd_ctx_get (local->fd, this, NULL); - if (!ret) { + dht_inode_ctx_get1 (this, inode, &subvol); + if (subvol) { dht_flush2 (this, frame, 0); return 0; } @@ -644,14 +650,10 @@ dht_flush2 (xlator_t *this, call_frame_t *frame, int op_ret) { dht_local_t *local = NULL; xlator_t *subvol = NULL; - uint64_t tmp_subvol = 0; - int ret = -1; local = frame->local; - ret = fd_ctx_get (local->fd, this, &tmp_subvol); - if (!ret) - subvol = (xlator_t *)(long)tmp_subvol; + dht_inode_ctx_get1 (this, local->fd->inode, &subvol); if (!subvol) subvol = local->cached_subvol; @@ -713,6 +715,8 @@ dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, dht_local_t *local = NULL; call_frame_t *prev = NULL; int ret = -1; + inode_t *inode = NULL; + xlator_t *subvol = 0; local = frame->local; prev = cookie; @@ -734,8 +738,8 @@ dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, } local->op_errno = op_errno; - ret = fd_ctx_get (local->fd, this, NULL); - if (ret) { + dht_inode_ctx_get1 (this, inode, &subvol); + if (!subvol) { local->rebalance.target_op_fn = dht_fsync2; /* Check if the rebalance phase1 is true */ @@ -750,11 +754,12 @@ dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, if (IS_DHT_MIGRATION_PHASE2 (postbuf)) { ret = dht_rebalance_complete_check (this, frame); } + if (!ret) + return 0; } else { dht_fsync2 (this, frame, 0); - } - if (!ret) return 0; + } out: DHT_STRIP_PHASE1_FLAGS (postbuf); @@ -770,15 +775,10 @@ dht_fsync2 (xlator_t *this, call_frame_t *frame, int op_ret) { dht_local_t *local = NULL; xlator_t *subvol = NULL; - uint64_t tmp_subvol = 0; - int ret = -1; local = frame->local; - ret = fd_ctx_get (local->fd, this, &tmp_subvol); - if (!ret) - subvol = (xlator_t *)(long)tmp_subvol; - + dht_inode_ctx_get1 (this, local->fd->inode, &subvol); if (!subvol) subvol = local->cached_subvol; diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c index 9bcd84ae1..4b3f3a049 100644 --- a/xlators/cluster/dht/src/dht-inode-write.c +++ b/xlators/cluster/dht/src/dht-inode-write.c @@ -21,6 +21,7 @@ int dht_truncate2 (xlator_t *this, call_frame_t *frame, int ret); int dht_setattr2 (xlator_t *this, call_frame_t *frame, int ret); int dht_fallocate2(xlator_t *this, call_frame_t *frame, int op_ret); int dht_discard2(xlator_t *this, call_frame_t *frame, int op_ret); +int dht_zerofill2(xlator_t *this, call_frame_t *frame, int op_ret); int dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this, @@ -29,6 +30,7 @@ dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this, { dht_local_t *local = NULL; int ret = -1; + xlator_t *subvol = NULL; if (op_ret == -1 && (op_errno != ENOENT)) { goto out; @@ -65,8 +67,8 @@ dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this, dht_iatt_merge (this, &local->stbuf, postbuf, NULL); dht_iatt_merge (this, &local->prebuf, prebuf, NULL); - ret = fd_ctx_get (local->fd, this, NULL); - if (!ret) { + ret = dht_inode_ctx_get1 (this, local->fd->inode, &subvol); + if (subvol) { dht_writev2 (this, frame, 0); return 0; } @@ -90,14 +92,10 @@ dht_writev2 (xlator_t *this, call_frame_t *frame, int op_ret) { dht_local_t *local = NULL; xlator_t *subvol = NULL; - uint64_t tmp_subvol = 0; - int ret = -1; local = frame->local; - ret = fd_ctx_get (local->fd, this, &tmp_subvol); - if (!ret) - subvol = (xlator_t *)(long)tmp_subvol; + dht_inode_ctx_get1 (this, local->fd->inode, &subvol); if (!subvol) subvol = local->cached_subvol; @@ -172,6 +170,8 @@ dht_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this, dht_local_t *local = NULL; call_frame_t *prev = NULL; int ret = -1; + xlator_t *subvol = NULL; + inode_t *inode = NULL; GF_VALIDATE_OR_GOTO ("dht", frame, err); GF_VALIDATE_OR_GOTO ("dht", this, out); @@ -213,8 +213,9 @@ dht_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this, if (IS_DHT_MIGRATION_PHASE1 (postbuf)) { dht_iatt_merge (this, &local->stbuf, postbuf, NULL); dht_iatt_merge (this, &local->prebuf, prebuf, NULL); - ret = fd_ctx_get (local->fd, this, NULL); - if (!ret || !local->fd) { + inode = (local->fd) ? local->fd->inode : local->loc.inode; + dht_inode_ctx_get1 (this, inode, &subvol); + if (subvol) { dht_truncate2 (this, frame, 0); return 0; } @@ -238,16 +239,13 @@ dht_truncate2 (xlator_t *this, call_frame_t *frame, int op_ret) { dht_local_t *local = NULL; xlator_t *subvol = NULL; - uint64_t tmp_subvol = 0; - int ret = -1; + inode_t *inode = NULL; local = frame->local; - if (local->fd) - ret = fd_ctx_get (local->fd, this, &tmp_subvol); - if (!ret) - subvol = (xlator_t *)(long)tmp_subvol; + inode = local->fd ? local->fd->inode : local->loc.inode; + dht_inode_ctx_get1 (this, inode, &subvol); if (!subvol) subvol = local->cached_subvol; @@ -359,6 +357,7 @@ dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, dht_local_t *local = NULL; call_frame_t *prev = NULL; int ret = -1; + xlator_t *subvol = NULL; GF_VALIDATE_OR_GOTO ("dht", frame, err); GF_VALIDATE_OR_GOTO ("dht", this, out); @@ -398,8 +397,8 @@ dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, if (IS_DHT_MIGRATION_PHASE1 (postbuf)) { dht_iatt_merge (this, &local->stbuf, postbuf, NULL); dht_iatt_merge (this, &local->prebuf, prebuf, NULL); - ret = fd_ctx_get (local->fd, this, NULL); - if (!ret) { + dht_inode_ctx_get1 (this, local->fd->inode, &subvol); + if (subvol) { dht_fallocate2 (this, frame, 0); return 0; } @@ -422,15 +421,10 @@ dht_fallocate2(xlator_t *this, call_frame_t *frame, int op_ret) { dht_local_t *local = NULL; xlator_t *subvol = NULL; - uint64_t tmp_subvol = 0; - int ret = -1; local = frame->local; - if (local->fd) - ret = fd_ctx_get (local->fd, this, &tmp_subvol); - if (!ret) - subvol = (xlator_t *)(long)tmp_subvol; + dht_inode_ctx_get1 (this, local->fd->inode, &subvol); if (!subvol) subvol = local->cached_subvol; @@ -497,6 +491,7 @@ dht_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this, dht_local_t *local = NULL; call_frame_t *prev = NULL; int ret = -1; + xlator_t *subvol = NULL; GF_VALIDATE_OR_GOTO ("dht", frame, err); GF_VALIDATE_OR_GOTO ("dht", this, out); @@ -536,8 +531,8 @@ dht_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this, if (IS_DHT_MIGRATION_PHASE1 (postbuf)) { dht_iatt_merge (this, &local->stbuf, postbuf, NULL); dht_iatt_merge (this, &local->prebuf, prebuf, NULL); - ret = fd_ctx_get (local->fd, this, NULL); - if (!ret) { + dht_inode_ctx_get1 (this, local->fd->inode, &subvol); + if (subvol) { dht_discard2 (this, frame, 0); return 0; } @@ -560,15 +555,10 @@ dht_discard2(xlator_t *this, call_frame_t *frame, int op_ret) { dht_local_t *local = NULL; xlator_t *subvol = NULL; - uint64_t tmp_subvol = 0; - int ret = -1; local = frame->local; - if (local->fd) - ret = fd_ctx_get (local->fd, this, &tmp_subvol); - if (!ret) - subvol = (xlator_t *)(long)tmp_subvol; + dht_inode_ctx_get1 (this, local->fd->inode, &subvol); if (!subvol) subvol = local->cached_subvol; @@ -624,6 +614,141 @@ err: return 0; } +int +dht_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, struct iatt *prebuf, + struct iatt *postbuf, dict_t *xdata) +{ + dht_local_t *local = NULL; + call_frame_t *prev = NULL; + int ret = -1; + + GF_VALIDATE_OR_GOTO ("dht", frame, err); + GF_VALIDATE_OR_GOTO ("dht", this, out); + GF_VALIDATE_OR_GOTO ("dht", frame->local, out); + GF_VALIDATE_OR_GOTO ("dht", cookie, out); + + local = frame->local; + prev = cookie; + + if ((op_ret == -1) && (op_errno != ENOENT)) { + local->op_errno = op_errno; + local->op_ret = -1; + gf_log (this->name, GF_LOG_DEBUG, + "subvolume %s returned -1 (%s)", + prev->this->name, strerror (op_errno)); + goto out; + } + + if (local->call_cnt != 1) { + if (local->stbuf.ia_blocks) { + dht_iatt_merge (this, postbuf, &local->stbuf, NULL); + dht_iatt_merge (this, prebuf, &local->prebuf, NULL); + } + goto out; + } + local->rebalance.target_op_fn = dht_zerofill2; + /* Phase 2 of migration */ + if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) { + ret = dht_rebalance_complete_check (this, frame); + if (!ret) + return 0; + } + + /* Check if the rebalance phase1 is true */ + if (IS_DHT_MIGRATION_PHASE1 (postbuf)) { + dht_iatt_merge (this, &local->stbuf, postbuf, NULL); + dht_iatt_merge (this, &local->prebuf, prebuf, NULL); + ret = fd_ctx_get (local->fd, this, NULL); + if (!ret) { + dht_zerofill2 (this, frame, 0); + return 0; + } + ret = dht_rebalance_in_progress_check (this, frame); + if (!ret) + return 0; + } + +out: + DHT_STRIP_PHASE1_FLAGS (postbuf); + DHT_STRIP_PHASE1_FLAGS (prebuf); + DHT_STACK_UNWIND (zerofill, frame, op_ret, op_errno, + prebuf, postbuf, xdata); +err: + return 0; +} + +int +dht_zerofill2(xlator_t *this, call_frame_t *frame, int op_ret) +{ + dht_local_t *local = NULL; + xlator_t *subvol = NULL; + uint64_t tmp_subvol = 0; + int ret = -1; + + local = frame->local; + + if (local->fd) + ret = fd_ctx_get (local->fd, this, &tmp_subvol); + if (!ret) + subvol = (xlator_t *)(long)tmp_subvol; + + if (!subvol) + subvol = local->cached_subvol; + + local->call_cnt = 2; /* This is the second attempt */ + + STACK_WIND(frame, dht_zerofill_cbk, subvol, subvol->fops->zerofill, + local->fd, local->rebalance.offset, local->rebalance.size, + NULL); + + return 0; +} + +int +dht_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + size_t len, dict_t *xdata) +{ + xlator_t *subvol = NULL; + int op_errno = -1; + dht_local_t *local = NULL; + + VALIDATE_OR_GOTO (frame, err); + VALIDATE_OR_GOTO (this, err); + VALIDATE_OR_GOTO (fd, err); + + local = dht_local_init (frame, NULL, fd, GF_FOP_ZEROFILL); + if (!local) { + op_errno = ENOMEM; + goto err; + } + + local->rebalance.offset = offset; + local->rebalance.size = len; + + local->call_cnt = 1; + subvol = local->cached_subvol; + if (!subvol) { + gf_log (this->name, GF_LOG_DEBUG, + "no cached subvolume for fd=%p", fd); + op_errno = EINVAL; + goto err; + } + + STACK_WIND (frame, dht_zerofill_cbk, subvol, subvol->fops->zerofill, + fd, offset, len, xdata); + + return 0; + +err: + op_errno = (op_errno == -1) ? errno : op_errno; + DHT_STACK_UNWIND (zerofill, frame, -1, op_errno, NULL, NULL, NULL); + + return 0; +} + + + /* handle cases of migration here for 'setattr()' calls */ int dht_file_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, @@ -675,15 +800,13 @@ dht_setattr2 (xlator_t *this, call_frame_t *frame, int op_ret) { dht_local_t *local = NULL; xlator_t *subvol = NULL; - uint64_t tmp_subvol = 0; - int ret = -1; + inode_t *inode = NULL; local = frame->local; - if (local->fd) - ret = fd_ctx_get (local->fd, this, &tmp_subvol); - if (!ret) - subvol = (xlator_t *)(long)tmp_subvol; + inode = (local->fd) ? local->fd->inode : local->loc.inode; + + dht_inode_ctx_get1 (this, inode, &subvol); if (!subvol) subvol = local->cached_subvol; diff --git a/xlators/cluster/dht/src/dht-layout.c b/xlators/cluster/dht/src/dht-layout.c index 07e8cbae4..38e9970a7 100644 --- a/xlators/cluster/dht/src/dht-layout.c +++ b/xlators/cluster/dht/src/dht-layout.c @@ -454,12 +454,19 @@ dht_layout_entry_cmp (dht_layout_t *layout, int i, int j) { int64_t diff = 0; + /* swap zero'ed out layouts to front, if needed */ + if (!layout->list[j].start && !layout->list[j].stop) { + diff = (int64_t) layout->list[i].stop + - (int64_t) layout->list[j].stop; + goto out; + } if (layout->list[i].err || layout->list[j].err) diff = layout->list[i].err - layout->list[j].err; else diff = (int64_t) layout->list[i].start - (int64_t) layout->list[j].start; +out: return diff; } @@ -534,13 +541,13 @@ dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout, case -1: case ENOENT: missing++; - break; + continue; case ENOTCONN: down++; - break; + continue; case ENOSPC: no_space++; - break; + continue; case 0: /* if err == 0 and start == stop, then it is a non misc++; * participating subvolume(spread-cnt). Then, do not @@ -552,6 +559,7 @@ dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout, break; default: misc++; + continue; } is_virgin = 0; @@ -593,8 +601,7 @@ dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout, int -dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout, - uint32_t *missing_p) +dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout) { int ret = 0; int i = 0; @@ -606,7 +613,6 @@ dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout, ret = dht_layout_sort (layout); if (ret == -1) { - /* defensive coding; this can't happen currently */ gf_log (this->name, GF_LOG_WARNING, "sort failed?! how the ...."); goto out; @@ -616,26 +622,23 @@ dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout, &holes, &overlaps, &missing, &down, &misc, NULL); if (ret == -1) { - /* defensive coding; this can't happen currently */ gf_log (this->name, GF_LOG_WARNING, "error while finding anomalies in %s -- not good news", loc->path); goto out; } - ret = holes + overlaps; - if (ret) { + if (holes || overlaps) { if (missing == layout->cnt) { gf_log (this->name, GF_LOG_DEBUG, "directory %s looked up first time", loc->path); } else { gf_log (this->name, GF_LOG_INFO, - "found anomalies in %s. holes=%d overlaps=%d" - " missing=%d down=%d misc=%d", - loc->path, holes, overlaps, missing, down, - misc); + "found anomalies in %s. holes=%d overlaps=%d", + loc->path, holes, overlaps); } + ret = -1; } for (i = 0; i < layout->cnt; i++) { @@ -650,14 +653,14 @@ dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout, (layout->list[i].xlator ? layout->list[i].xlator->name : "<>")); + if ((layout->list[i].err == ENOENT) && (ret >= 0)) { + ret++; + } } } out: - if (missing_p) { - *missing_p = missing; - } return ret; } diff --git a/xlators/cluster/dht/src/dht-linkfile.c b/xlators/cluster/dht/src/dht-linkfile.c index 57cc202da..dbc9d0b3c 100644 --- a/xlators/cluster/dht/src/dht-linkfile.c +++ b/xlators/cluster/dht/src/dht-linkfile.c @@ -148,15 +148,6 @@ dht_linkfile_create (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk, goto out; } - ret = dict_set_uint32 (dict, GLUSTERFS_CREATE_MODE_KEY, - DHT_LINKFILE_MODE); - if (ret < 0) { - gf_log (frame->this->name, GF_LOG_WARNING, - "%s: failed to set mode in xdata", - loc->path); - goto out; - } - local->link_subvol = fromvol; /* Always create as root:root. dht_linkfile_attr_heal fixes the * ownsership */ diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c index 5300ff2fe..bcb19f23e 100644 --- a/xlators/cluster/dht/src/dht-rebalance.c +++ b/xlators/cluster/dht/src/dht-rebalance.c @@ -243,7 +243,7 @@ out: static inline int __dht_rebalance_create_dst_file (xlator_t *to, xlator_t *from, loc_t *loc, struct iatt *stbuf, - dict_t *dict, fd_t **dst_fd) + dict_t *dict, fd_t **dst_fd, dict_t *xattr) { xlator_t *this = NULL; int ret = -1; @@ -307,6 +307,12 @@ __dht_rebalance_create_dst_file (xlator_t *to, xlator_t *from, loc_t *loc, struc goto out; } + ret = syncop_fsetxattr (to, fd, xattr, 0); + if (ret == -1) + gf_log (this->name, GF_LOG_WARNING, + "%s: failed to set xattr on %s (%s)", + loc->path, to->name, strerror (errno)); + ret = syncop_ftruncate (to, fd, stbuf->ia_size); if (ret < 0) gf_log (this->name, GF_LOG_ERROR, @@ -340,6 +346,9 @@ __dht_check_free_space (xlator_t *to, xlator_t *from, loc_t *loc, int ret = -1; xlator_t *this = NULL; + uint64_t src_statfs_blocks = 1; + uint64_t dst_statfs_blocks = 1; + this = THIS; ret = syncop_statfs (from, loc, &src_statfs); @@ -363,22 +372,34 @@ __dht_check_free_space (xlator_t *to, xlator_t *from, loc_t *loc, if (flag != GF_DHT_MIGRATE_DATA) goto check_avail_space; - if (((dst_statfs.f_bavail * - dst_statfs.f_bsize) / GF_DISK_SECTOR_SIZE) < - (((src_statfs.f_bavail * src_statfs.f_bsize) / - GF_DISK_SECTOR_SIZE) - stbuf->ia_blocks)) { - gf_log (this->name, GF_LOG_WARNING, - "data movement attempted from node (%s) with" - " higher disk space to a node (%s) with " - "lesser disk space (%s)", from->name, - to->name, loc->path); - - /* this is not a 'failure', but we don't want to - consider this as 'success' too :-/ */ - ret = 1; - goto out; + /* Check: + During rebalance `migrate-data` - Destination subvol experiences + a `reduction` in 'blocks' of free space, at the same time source + subvol gains certain 'blocks' of free space. A valid check is + necessary here to avoid errorneous move to destination where + the space could be scantily available. + */ + if (stbuf) { + dst_statfs_blocks = ((dst_statfs.f_bavail * + dst_statfs.f_bsize) / + GF_DISK_SECTOR_SIZE); + src_statfs_blocks = ((src_statfs.f_bavail * + src_statfs.f_bsize) / + GF_DISK_SECTOR_SIZE); + if ((dst_statfs_blocks - stbuf->ia_blocks) < + (src_statfs_blocks + stbuf->ia_blocks)) { + gf_log (this->name, GF_LOG_WARNING, + "data movement attempted from node (%s) with" + " higher disk space to a node (%s) with " + "lesser disk space (%s)", from->name, + to->name, loc->path); + + /* this is not a 'failure', but we don't want to + consider this as 'success' too :-/ */ + ret = 1; + goto out; + } } - check_avail_space: if (((dst_statfs.f_bavail * dst_statfs.f_bsize) / GF_DISK_SECTOR_SIZE) < stbuf->ia_blocks) { @@ -708,9 +729,16 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, goto out; } + /* TODO: move all xattr related operations to fd based operations */ + ret = syncop_listxattr (from, loc, &xattr); + if (ret == -1) + gf_log (this->name, GF_LOG_WARNING, + "%s: failed to get xattr from %s (%s)", + loc->path, from->name, strerror (errno)); + /* create the destination, with required modes/xattr */ ret = __dht_rebalance_create_dst_file (to, from, loc, &stbuf, - dict, &dst_fd); + dict, &dst_fd, xattr); if (ret) goto out; @@ -727,6 +755,7 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, goto out; } + ret = syncop_fstat (from, src_fd, &stbuf); if (ret) { gf_log (this->name, GF_LOG_ERROR, "failed to lookup %s on %s (%s)", @@ -756,19 +785,6 @@ dht_migrate_file (xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, goto out; } - /* TODO: move all xattr related operations to fd based operations */ - ret = syncop_listxattr (from, loc, &xattr); - if (ret == -1) - gf_log (this->name, GF_LOG_WARNING, - "%s: failed to get xattr from %s (%s)", - loc->path, from->name, strerror (errno)); - - ret = syncop_setxattr (to, loc, xattr, 0); - if (ret == -1) - gf_log (this->name, GF_LOG_WARNING, - "%s: failed to set xattr on %s (%s)", - loc->path, to->name, strerror (errno)); - /* TODO: Sync the locks */ ret = syncop_fsync (to, dst_fd, 0); @@ -1760,6 +1776,8 @@ log: case GF_DEFRAG_STATUS_FAILED: status = "failed"; break; + default: + break; } gf_log (THIS->name, GF_LOG_INFO, "Rebalance is %s. Time taken is %.2f " diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c index b220a0e25..3fe96b1c7 100644 --- a/xlators/cluster/dht/src/dht-selfheal.c +++ b/xlators/cluster/dht/src/dht-selfheal.c @@ -17,6 +17,7 @@ #include "glusterfs.h" #include "xlator.h" #include "dht-common.h" +#include "glusterfs-acl.h" #define DHT_SET_LAYOUT_RANGE(layout,i,srt,chunk,cnt,path) do { \ layout->list[i].start = srt; \ @@ -564,9 +565,33 @@ dht_get_layout_count (xlator_t *this, dht_layout_t *layout, int new_layout) for (i = 0; i < layout->cnt; i++) { err = layout->list[i].err; - if (err == -1 || err == 0) { - layout->list[i].err = -1; + if (err == -1 || err == 0 || err == ENOENT) { + /* Setting list[i].err = -1 is an indication for + dht_selfheal_layout_new_directory() to assign + a range. We set it to -1 based on any one of + the three criteria: + + - err == -1 already, which means directory + existed but layout was not set on it. + + - err == 0, which means directory exists and + has an old layout piece which will be + overwritten now. + + - err == ENOENT, which means directory does + not exist (possibly racing with mkdir or + finishing half done mkdir). The missing + directory will be attempted to be recreated. + + It is important to note that it is safe + to race with mkdir() as self-heal and + mkdir are idempotent operations. Both will + strive to set the directory and layouts to + the same final state. + */ count++; + if (!err) + layout->list[i].err = -1; } } @@ -776,7 +801,7 @@ dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc, DHT_RESET_LAYOUT_RANGE (layout); for (i = start_subvol; i < layout->cnt; i++) { err = layout->list[i].err; - if (err == -1) { + if (err == -1 || err == ENOENT) { DHT_SET_LAYOUT_RANGE(layout, i, start, chunk, cnt, loc->path); if (--cnt == 0) { @@ -789,7 +814,7 @@ dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc, for (i = 0; i < start_subvol; i++) { err = layout->list[i].err; - if (err == -1) { + if (err == -1 || err == ENOENT) { DHT_SET_LAYOUT_RANGE(layout, i, start, chunk, cnt, loc->path); if (--cnt == 0) { diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c index 0349b63a9..fc0ca2f77 100644 --- a/xlators/cluster/dht/src/dht.c +++ b/xlators/cluster/dht/src/dht.c @@ -72,6 +72,7 @@ struct xlator_fops fops = { .fsetattr = dht_fsetattr, .fallocate = dht_fallocate, .discard = dht_discard, + .zerofill = dht_zerofill, }; struct xlator_dumpops dumpops = { diff --git a/xlators/cluster/stripe/src/stripe.c b/xlators/cluster/stripe/src/stripe.c index 056dea2eb..69b510e23 100644 --- a/xlators/cluster/stripe/src/stripe.c +++ b/xlators/cluster/stripe/src/stripe.c @@ -4107,6 +4107,191 @@ err: } int32_t +stripe_zerofill_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) +{ + int32_t callcnt = 0; + stripe_local_t *local = NULL; + stripe_local_t *mlocal = NULL; + call_frame_t *prev = NULL; + call_frame_t *mframe = NULL; + + if (!this || !frame || !frame->local || !cookie) { + gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref"); + goto out; + } + + prev = cookie; + local = frame->local; + mframe = local->orig_frame; + mlocal = mframe->local; + + LOCK(&frame->lock); + { + callcnt = ++mlocal->call_count; + + if (op_ret == 0) { + mlocal->post_buf = *postbuf; + mlocal->pre_buf = *prebuf; + + mlocal->prebuf_blocks += prebuf->ia_blocks; + mlocal->postbuf_blocks += postbuf->ia_blocks; + + correct_file_size(prebuf, mlocal->fctx, prev); + correct_file_size(postbuf, mlocal->fctx, prev); + + if (mlocal->prebuf_size < prebuf->ia_size) + mlocal->prebuf_size = prebuf->ia_size; + if (mlocal->postbuf_size < postbuf->ia_size) + mlocal->postbuf_size = postbuf->ia_size; + } + + /* return the first failure */ + if (mlocal->op_ret == 0) { + mlocal->op_ret = op_ret; + mlocal->op_errno = op_errno; + } + } + UNLOCK (&frame->lock); + + if ((callcnt == mlocal->wind_count) && mlocal->unwind) { + mlocal->pre_buf.ia_size = mlocal->prebuf_size; + mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks; + mlocal->post_buf.ia_size = mlocal->postbuf_size; + mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks; + + STRIPE_STACK_UNWIND (zerofill, mframe, mlocal->op_ret, + mlocal->op_errno, &mlocal->pre_buf, + &mlocal->post_buf, NULL); + } +out: + STRIPE_STACK_DESTROY(frame); + return 0; +} + +int32_t +stripe_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, + size_t len, dict_t *xdata) +{ + stripe_local_t *local = NULL; + stripe_fd_ctx_t *fctx = NULL; + int32_t op_errno = 1; + int32_t idx = 0; + int32_t offset_offset = 0; + int32_t remaining_size = 0; + off_t fill_size = 0; + uint64_t stripe_size = 0; + uint64_t tmp_fctx = 0; + off_t dest_offset = 0; + call_frame_t *fframe = NULL; + stripe_local_t *flocal = NULL; + + VALIDATE_OR_GOTO (frame, err); + VALIDATE_OR_GOTO (this, err); + VALIDATE_OR_GOTO (fd, err); + VALIDATE_OR_GOTO (fd->inode, err); + + inode_ctx_get (fd->inode, this, &tmp_fctx); + if (!tmp_fctx) { + op_errno = EINVAL; + goto err; + } + fctx = (stripe_fd_ctx_t *)(long)tmp_fctx; + stripe_size = fctx->stripe_size; + + STRIPE_VALIDATE_FCTX (fctx, err); + + remaining_size = len; + + local = mem_get0 (this->local_pool); + if (!local) { + op_errno = ENOMEM; + goto err; + } + fctx = (stripe_fd_ctx_t *)(long)tmp_fctx; + stripe_size = fctx->stripe_size; + + STRIPE_VALIDATE_FCTX (fctx, err); + + remaining_size = len; + + local = mem_get0 (this->local_pool); + if (!local) { + op_errno = ENOMEM; + goto err; + } + fctx = (stripe_fd_ctx_t *)(long)tmp_fctx; + stripe_size = fctx->stripe_size; + + STRIPE_VALIDATE_FCTX (fctx, err); + + remaining_size = len; + + local = mem_get0 (this->local_pool); + if (!local) { + op_errno = ENOMEM; + goto err; + } + frame->local = local; + local->stripe_size = stripe_size; + local->fctx = fctx; + + if (!stripe_size) { + gf_log (this->name, GF_LOG_DEBUG, + "Wrong stripe size for the file"); + op_errno = EINVAL; + goto err; + } + + while (1) { + fframe = copy_frame(frame); + flocal = mem_get0(this->local_pool); + if (!flocal) { + op_errno = ENOMEM; + goto err; + } + flocal->orig_frame = frame; + fframe->local = flocal; + + idx = (((offset + offset_offset) / + local->stripe_size) % fctx->stripe_count); + + fill_size = (local->stripe_size - + ((offset + offset_offset) % local->stripe_size)); + if (fill_size > remaining_size) + fill_size = remaining_size; + + remaining_size -= fill_size; + + local->wind_count++; + if (remaining_size == 0) + local->unwind = 1; + + dest_offset = offset + offset_offset; + if (fctx->stripe_coalesce) + dest_offset = coalesced_offset(dest_offset, + local->stripe_size, + fctx->stripe_count); + + STACK_WIND(fframe, stripe_zerofill_cbk, fctx->xl_array[idx], + fctx->xl_array[idx]->fops->zerofill, fd, + dest_offset, fill_size, xdata); + offset_offset += fill_size; + if (remaining_size == 0) + break; + } + + return 0; +err: + if (fframe) + STRIPE_STACK_DESTROY(fframe); + + STRIPE_STACK_UNWIND (zerofill, frame, -1, op_errno, NULL, NULL, NULL); + return 0; +} + +int32_t stripe_release (xlator_t *this, fd_t *fd) { return 0; @@ -4280,6 +4465,37 @@ stripe_setxattr_cbk (call_frame_t *frame, void *cookie, return 0; } +#ifdef HAVE_BD_XLATOR +int +stripe_is_bd (dict_t *this, char *key, data_t *value, void *data) +{ + gf_boolean_t *is_bd = data; + + if (data == NULL) + return 0; + + if (XATTR_IS_BD (key)) + *is_bd = _gf_true; + + return 0; +} + +inline gf_boolean_t +stripe_setxattr_is_bd (dict_t *dict) +{ + gf_boolean_t is_bd = _gf_false; + + if (dict == NULL) + goto out; + + dict_foreach (dict, stripe_is_bd, &is_bd); +out: + return is_bd; +} +#else +#define stripe_setxattr_is_bd(dict) _gf_false +#endif + int stripe_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict, int flags, dict_t *xdata) @@ -4289,6 +4505,7 @@ stripe_setxattr (call_frame_t *frame, xlator_t *this, stripe_private_t *priv = NULL; stripe_local_t *local = NULL; int i = 0; + gf_boolean_t is_bd = _gf_false; VALIDATE_OR_GOTO (frame, err); VALIDATE_OR_GOTO (this, err); @@ -4311,11 +4528,15 @@ stripe_setxattr (call_frame_t *frame, xlator_t *this, local->wind_count = priv->child_count; local->op_ret = local->op_errno = 0; + is_bd = stripe_setxattr_is_bd (dict); + /** * Set xattrs for directories on all subvolumes. Additionally - * this power is only given to a special client. + * this power is only given to a special client. Bd xlator + * also needs xattrs for regular files (ie LVs) */ - if ((frame->root->pid == GF_CLIENT_PID_GSYNCD) && IA_ISDIR (loc->inode->ia_type)) { + if (((frame->root->pid == GF_CLIENT_PID_GSYNCD) && + IA_ISDIR (loc->inode->ia_type)) || is_bd) { for (i = 0; i < priv->child_count; i++, trav = trav->next) { STACK_WIND (frame, stripe_setxattr_cbk, trav->xlator, trav->xlator->fops->setxattr, @@ -4346,21 +4567,21 @@ stripe_fsetxattr_cbk (call_frame_t *frame, void *cookie, int -stripe_is_lockinfo (dict_t *this, - char *key, - data_t *value, - void *data) +stripe_is_special_key (dict_t *this, + char *key, + data_t *value, + void *data) { - gf_boolean_t *is_lockinfo = NULL; + gf_boolean_t *is_special = NULL; if (data == NULL) { goto out; } - is_lockinfo = data; + is_special = data; - if (XATTR_IS_LOCKINFO (key)) - *is_lockinfo = _gf_true; + if (XATTR_IS_LOCKINFO (key) || XATTR_IS_BD (key)) + *is_special = _gf_true; out: return 0; @@ -4437,7 +4658,7 @@ stripe_fsetxattr_is_special (dict_t *dict) goto out; } - dict_foreach (dict, stripe_is_lockinfo, &is_spl); + dict_foreach (dict, stripe_is_special_key, &is_spl); out: return is_spl; @@ -5199,10 +5420,10 @@ stripe_vgetxattr_cbk (call_frame_t *frame, void *cookie, xattr->pos = cky; xattr->xattr_value = gf_memdup (xattr_val, - xattr->xattr_value, xattr->xattr_len); - local->xattr_total_len += xattr->xattr_len + 1; + if (xattr->xattr_value != NULL) + local->xattr_total_len += xattr->xattr_len + 1; } } out: @@ -5558,6 +5779,7 @@ struct xlator_fops fops = { .readdirp = stripe_readdirp, .fallocate = stripe_fallocate, .discard = stripe_discard, + .zerofill = stripe_zerofill, }; struct xlator_cbks cbks = { @@ -5583,10 +5805,10 @@ struct volume_options options[] = { }, { .key = {"coalesce"}, .type = GF_OPTION_TYPE_BOOL, - .default_value = "false", - .description = "Enable coalesce mode to flatten striped files as " - "stored on the server (i.e., eliminate holes caused " - "by the traditional format)." + .default_value = "true", + .description = "Enable/Disable coalesce mode to flatten striped " + "files as stored on the server (i.e., eliminate holes " + "caused by the traditional format)." }, { .key = {NULL} }, }; |