diff options
Diffstat (limited to 'xlators')
| -rw-r--r-- | xlators/cluster/afr/src/afr-common.c | 205 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-read-txn.c | 14 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-self-heal-common.c | 159 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-self-heal-data.c | 21 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-self-heal-entry.c | 3 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-self-heal-metadata.c | 15 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-self-heal.h | 24 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-transaction.c | 9 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr.h | 1 | ||||
| -rw-r--r-- | xlators/cluster/dht/src/dht-helper.c | 7 | ||||
| -rw-r--r-- | xlators/cluster/dht/src/dht-rebalance.c | 29 | ||||
| -rw-r--r-- | xlators/mgmt/glusterd/src/glusterd-geo-rep.c | 32 | ||||
| -rw-r--r-- | xlators/storage/posix/src/posix.c | 26 |
13 files changed, 431 insertions, 114 deletions
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c index 747577c9380..0f878a9be86 100644 --- a/xlators/cluster/afr/src/afr-common.c +++ b/xlators/cluster/afr/src/afr-common.c @@ -979,6 +979,13 @@ ret: return -err; } +int +afr_fav_child_reset_sink_xattrs (void *opaque); + +int +afr_fav_child_reset_sink_xattrs_cbk (int ret, call_frame_t *frame, + void *opaque); + gf_boolean_t afr_selfheal_enabled (xlator_t *this) { @@ -994,6 +1001,82 @@ afr_selfheal_enabled (xlator_t *this) return data || priv->metadata_self_heal || priv->entry_self_heal; } + +int +afr_txn_refresh_done (call_frame_t *frame, xlator_t *this, int err) +{ + + call_frame_t *heal_frame = NULL; + afr_local_t *heal_local = NULL; + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + inode_t *inode = NULL; + int event_generation = 0; + int read_subvol = -1; + int op_errno = ENOMEM; + int ret = 0; + + local = frame->local; + inode = local->inode; + priv = this->private; + + if (err) + goto refresh_done; + + if (local->op == GF_FOP_LOOKUP) + goto refresh_done; + + ret = afr_inode_get_readable (frame, inode, this, local->readable, + &event_generation, + local->transaction.type); + + if (ret == -EIO || (local->is_read_txn && !event_generation)) { + /* No readable subvolume even after refresh ==> splitbrain.*/ + if (!priv->fav_child_policy) { + err = -EIO; + goto refresh_done; + } + read_subvol = afr_sh_get_fav_by_policy (this, local->replies, + inode, NULL); + if (read_subvol == -1) { + err = -EIO; + goto refresh_done; + } + + heal_frame = copy_frame (frame); + if (!heal_frame) { + err = -EIO; + goto refresh_done; + } + heal_frame->root->pid = GF_CLIENT_PID_SELF_HEALD; + heal_local = AFR_FRAME_INIT (heal_frame, op_errno); + if (!heal_local) { + err = -EIO; + AFR_STACK_DESTROY (heal_frame); + goto refresh_done; + } + heal_local->xdata_req = dict_new(); + if (!heal_local->xdata_req) { + err = -EIO; + AFR_STACK_DESTROY (heal_frame); + goto refresh_done; + } + heal_local->heal_frame = frame; + ret = synctask_new (this->ctx->env, + afr_fav_child_reset_sink_xattrs, + afr_fav_child_reset_sink_xattrs_cbk, + heal_frame, + heal_frame); + return 0; + } + +refresh_done: + afr_local_replies_wipe (local, this->private); + local->refreshfn (frame, this, err); + + return 0; +} + int afr_inode_refresh_done (call_frame_t *frame, xlator_t *this) { @@ -1012,8 +1095,6 @@ afr_inode_refresh_done (call_frame_t *frame, xlator_t *this) err = afr_inode_refresh_err (frame, this); - afr_local_replies_wipe (local, this->private); - if (ret && afr_selfheal_enabled (this) && start_heal) { heal_frame = copy_frame (frame); if (!heal_frame) @@ -1033,7 +1114,7 @@ afr_inode_refresh_done (call_frame_t *frame, xlator_t *this) } refresh_done: - local->refreshfn (frame, this, err); + afr_txn_refresh_done (frame, this, err); return 0; } @@ -5391,6 +5472,7 @@ afr_selfheal_locked_metadata_inspect (call_frame_t *frame, xlator_t *this, unsigned char *sources = NULL; unsigned char *sinks = NULL; unsigned char *healed_sinks = NULL; + unsigned char *undid_pending = NULL; struct afr_reply *locked_replies = NULL; afr_private_t *priv = this->private; @@ -5399,6 +5481,7 @@ afr_selfheal_locked_metadata_inspect (call_frame_t *frame, xlator_t *this, sources = alloca0 (priv->child_count); sinks = alloca0 (priv->child_count); healed_sinks = alloca0 (priv->child_count); + undid_pending = alloca0 (priv->child_count); locked_replies = alloca0 (sizeof (*locked_replies) * priv->child_count); @@ -5415,6 +5498,7 @@ afr_selfheal_locked_metadata_inspect (call_frame_t *frame, xlator_t *this, ret = __afr_selfheal_metadata_prepare (frame, this, inode, locked_on, sources, sinks, healed_sinks, + undid_pending, locked_replies, pending); *msh = afr_decide_heal_info (priv, sources, ret); @@ -5438,6 +5522,7 @@ afr_selfheal_locked_data_inspect (call_frame_t *frame, xlator_t *this, unsigned char *sources = NULL; unsigned char *sinks = NULL; unsigned char *healed_sinks = NULL; + unsigned char *undid_pending = NULL; afr_private_t *priv = NULL; fd_t *fd = NULL; struct afr_reply *locked_replies = NULL; @@ -5451,6 +5536,7 @@ afr_selfheal_locked_data_inspect (call_frame_t *frame, xlator_t *this, sources = alloca0 (priv->child_count); sinks = alloca0 (priv->child_count); healed_sinks = alloca0 (priv->child_count); + undid_pending = alloca0 (priv->child_count); /* Heal-info does an open() on the file being examined so that the * current eager-lock holding client, if present, at some point sees @@ -5490,6 +5576,7 @@ afr_selfheal_locked_data_inspect (call_frame_t *frame, xlator_t *this, ret = __afr_selfheal_data_prepare (frame, this, inode, data_lock, sources, sinks, healed_sinks, + undid_pending, locked_replies, pflag); *dsh = afr_decide_heal_info (priv, sources, ret); @@ -6025,3 +6112,115 @@ afr_get_msg_id (char *op_type) return AFR_MSG_ADD_BRICK_STATUS; return -1; } + +int +afr_fav_child_reset_sink_xattrs_cbk (int ret, call_frame_t *heal_frame, + void *opaque) +{ + + call_frame_t *txn_frame = NULL; + afr_local_t *local = NULL; + afr_local_t *heal_local = NULL; + xlator_t *this = NULL; + + heal_local = heal_frame->local; + txn_frame = heal_local->heal_frame; + local = txn_frame->local; + this = txn_frame->this; + + /* Refresh the inode agan and proceed with the transaction.*/ + afr_inode_refresh (txn_frame, this, local->inode, NULL, + local->refreshfn); + + if (heal_frame) + AFR_STACK_DESTROY (heal_frame); + + return 0; +} + +int +afr_fav_child_reset_sink_xattrs (void *opaque) +{ + call_frame_t *heal_frame = NULL; + call_frame_t *txn_frame = NULL; + xlator_t *this = NULL; + gf_boolean_t d_spb = _gf_false; + gf_boolean_t m_spb = _gf_false; + afr_local_t *heal_local = NULL; + afr_local_t *txn_local = NULL; + afr_private_t *priv = NULL; + inode_t *inode = NULL; + unsigned char *locked_on = NULL; + unsigned char *sources = NULL; + unsigned char *sinks = NULL; + unsigned char *healed_sinks = NULL; + unsigned char *undid_pending = NULL; + struct afr_reply *locked_replies = NULL; + int ret = 0; + + heal_frame = (call_frame_t *) opaque; + heal_local = heal_frame->local; + txn_frame = heal_local->heal_frame; + txn_local = txn_frame->local; + this = txn_frame->this; + inode = txn_local->inode; + priv = this->private; + locked_on = alloca0 (priv->child_count); + sources = alloca0 (priv->child_count); + sinks = alloca0 (priv->child_count); + healed_sinks = alloca0 (priv->child_count); + undid_pending = alloca0 (priv->child_count); + locked_replies = alloca0 (sizeof (*locked_replies) * priv->child_count); + + ret = _afr_is_split_brain (txn_frame, this, txn_local->replies, + AFR_DATA_TRANSACTION, &d_spb); + + ret = _afr_is_split_brain (txn_frame, this, txn_local->replies, + AFR_METADATA_TRANSACTION, &m_spb); + + /* Take appropriate locks and reset sink xattrs. */ + if (d_spb) { + ret = afr_selfheal_inodelk (heal_frame, this, inode, this->name, + 0, 0, locked_on); + { + if (ret < AFR_SH_MIN_PARTICIPANTS) + goto data_unlock; + ret = __afr_selfheal_data_prepare (heal_frame, this, + inode, locked_on, + sources, sinks, + healed_sinks, + undid_pending, + locked_replies, + NULL); + } +data_unlock: + afr_selfheal_uninodelk (heal_frame, this, inode, this->name, + 0, 0, locked_on); + } + + if (m_spb) { + memset (locked_on, 0, sizeof (*locked_on) * priv->child_count); + memset (undid_pending, 0, + sizeof (*undid_pending) * priv->child_count); + ret = afr_selfheal_inodelk (heal_frame, this, inode, this->name, + LLONG_MAX-1, 0, locked_on); + { + if (ret < AFR_SH_MIN_PARTICIPANTS) + goto mdata_unlock; + ret = __afr_selfheal_metadata_prepare (heal_frame, this, + inode, locked_on, + sources, sinks, + healed_sinks, + undid_pending, + locked_replies, + NULL); + + } +mdata_unlock: + afr_selfheal_uninodelk (heal_frame, this, inode, this->name, + LLONG_MAX-1, 0, locked_on); + } + + return ret; + +} diff --git a/xlators/cluster/afr/src/afr-read-txn.c b/xlators/cluster/afr/src/afr-read-txn.c index 74749f029c8..926f7c4dc47 100644 --- a/xlators/cluster/afr/src/afr-read-txn.c +++ b/xlators/cluster/afr/src/afr-read-txn.c @@ -64,7 +64,6 @@ afr_read_txn_refresh_done (call_frame_t *frame, xlator_t *this, int err) { afr_local_t *local = NULL; int read_subvol = 0; - int event_generation = 0; inode_t *inode = NULL; int ret = -1; int spb_choice = -1; @@ -76,18 +75,12 @@ afr_read_txn_refresh_done (call_frame_t *frame, xlator_t *this, int err) local->op_errno = -err; local->op_ret = -1; read_subvol = -1; + gf_msg (this->name, GF_LOG_ERROR, EIO, AFR_MSG_SPLIT_BRAIN, + "Failing %s on gfid %s: split-brain observed.", + gf_fop_list[local->op], uuid_utoa (inode->gfid)); goto readfn; } - ret = afr_inode_get_readable (frame, inode, this, local->readable, - &event_generation, - local->transaction.type); - - if (ret == -EIO || !event_generation) - /* Even after refresh, we don't have a good - read subvolume. Time to bail */ - AFR_READ_TXN_SET_ERROR_AND_GOTO (-1, EIO, -1, readfn); - read_subvol = afr_read_subvol_select_by_policy (inode, this, local->readable, NULL); if (read_subvol == -1) @@ -208,6 +201,7 @@ afr_read_txn (call_frame_t *frame, xlator_t *this, inode_t *inode, local->readfn = readfn; local->inode = inode_ref (inode); + local->is_read_txn = _gf_true; if (priv->quorum_reads && priv->quorum_count && !afr_has_quorum (priv->child_up, this)) { diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c index faee8dbb89b..40cfd01f4e9 100644 --- a/xlators/cluster/afr/src/afr-self-heal-common.c +++ b/xlators/cluster/afr/src/afr-self-heal-common.c @@ -144,8 +144,10 @@ err: int afr_selfheal_undo_pending (call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *sources, unsigned char *sinks, - unsigned char *healed_sinks, afr_transaction_type type, - struct afr_reply *replies, unsigned char *locked_on) + unsigned char *healed_sinks, + unsigned char *undid_pending, + afr_transaction_type type, struct afr_reply *replies, + unsigned char *locked_on) { afr_private_t *priv = NULL; afr_local_t *local = NULL; @@ -213,6 +215,10 @@ afr_selfheal_undo_pending (call_frame_t *frame, xlator_t *this, inode_t *inode, and inspected on. */ continue; + if (undid_pending[i]) + /* We already unset the pending xattrs in + * _afr_fav_child_reset_sink_xattrs(). */ + continue; xattr = afr_selfheal_output_xattr (this, local->need_full_crawl, type, output_dirty, @@ -734,6 +740,42 @@ afr_sh_fav_by_size (xlator_t *this, struct afr_reply *replies, inode_t *inode) return fav_child; } +int +afr_sh_get_fav_by_policy (xlator_t *this, struct afr_reply *replies, + inode_t *inode, char **policy_str) +{ + afr_private_t *priv = NULL; + int fav_child = -1; + + priv = this->private; + switch (priv->fav_child_policy) { + case AFR_FAV_CHILD_BY_SIZE: + fav_child = afr_sh_fav_by_size (this, replies, inode); + if (policy_str && fav_child >= 0) + *policy_str = "SIZE"; + break; + case AFR_FAV_CHILD_BY_CTIME: + fav_child = afr_sh_fav_by_ctime (this, replies, inode); + if (policy_str && fav_child >= 0) + *policy_str = "CTIME"; + break; + case AFR_FAV_CHILD_BY_MTIME: + fav_child = afr_sh_fav_by_mtime (this, replies, inode); + if (policy_str && fav_child >= 0) + *policy_str = "MTIME"; + break; + case AFR_FAV_CHILD_BY_MAJORITY: + fav_child = afr_sh_fav_by_majority (this, replies, inode); + if (policy_str && fav_child >= 0) + *policy_str = "MAJORITY"; + break; + case AFR_FAV_CHILD_NONE: + default: + break; + } + + return fav_child; +} int afr_mark_split_brain_source_sinks_by_policy (call_frame_t *frame, @@ -755,24 +797,9 @@ afr_mark_split_brain_source_sinks_by_policy (call_frame_t *frame, time_t time; priv = this->private; - if (priv->fav_child_policy == AFR_FAV_CHILD_BY_MAJORITY) { - fav_child = afr_sh_fav_by_majority (this, replies, inode); - if (fav_child >= 0) - policy_str = "MAJORITY"; - } else if (priv->fav_child_policy == AFR_FAV_CHILD_BY_MTIME) { - fav_child = afr_sh_fav_by_mtime (this, replies, inode); - if (fav_child >= 0) - policy_str = "MTIME"; - } else if (priv->fav_child_policy == AFR_FAV_CHILD_BY_CTIME) { - fav_child = afr_sh_fav_by_ctime (this, replies, inode); - if (fav_child >= 0) - policy_str = "CTIME"; - } else if (priv->fav_child_policy == AFR_FAV_CHILD_BY_SIZE) { - fav_child = afr_sh_fav_by_size (this, replies, inode); - if (fav_child >= 0) - policy_str = "SIZE"; - } + fav_child = afr_sh_get_fav_by_policy (this, replies, inode, + &policy_str); if (fav_child > priv->child_count - 1) { gf_msg (this->name, GF_LOG_ERROR, 0, AFR_MSG_SBRAIN_FAV_CHILD_POLICY, "Invalid child (%d) " @@ -828,6 +855,7 @@ afr_mark_split_brain_source_sinks (call_frame_t *frame, xlator_t *this, dict_t *xdata_req = NULL; int heal_op = -1; int ret = -1; + int source = -1; local = frame->local; priv = this->private; @@ -837,27 +865,96 @@ afr_mark_split_brain_source_sinks (call_frame_t *frame, xlator_t *this, if (ret) goto autoheal; - ret = afr_mark_split_brain_source_sinks_by_heal_op (frame, this, + source = afr_mark_split_brain_source_sinks_by_heal_op (frame, this, sources, sinks, healed_sinks, locked_on, replies, type, heal_op); - return ret; + return source; autoheal: /* Automatically heal if fav_child_policy is set. */ if (priv->fav_child_policy != AFR_FAV_CHILD_NONE) { - ret = afr_mark_split_brain_source_sinks_by_policy (frame, this, - inode, - sources, - sinks, + source = afr_mark_split_brain_source_sinks_by_policy (frame, + this, + inode, + sources, + sinks, healed_sinks, - locked_on, - replies, - type); + locked_on, + replies, + type); + if (source != -1) { + ret = dict_set_int32 (xdata_req, "fav-child-policy", 1); + if (ret) + return -1; + } } - return ret; + return source; +} + +int +_afr_fav_child_reset_sink_xattrs (call_frame_t *frame, xlator_t *this, + inode_t *inode, int source, + unsigned char *healed_sinks, + unsigned char *undid_pending, + afr_transaction_type type, + unsigned char *locked_on, + struct afr_reply *replies) +{ + afr_private_t *priv = NULL; + afr_local_t *local = NULL; + int *input_dirty = NULL; + int **input_matrix = NULL; + int *output_dirty = NULL; + int **output_matrix = NULL; + dict_t *xattr = NULL; + dict_t *xdata = NULL; + int i = 0; + + priv = this->private; + local = frame->local; + + if (!dict_get (local->xdata_req, "fav-child-policy")) + return 0; + + xdata = dict_new(); + if (!xdata) + return -1; + + input_dirty = alloca0 (priv->child_count * sizeof (int)); + input_matrix = ALLOC_MATRIX (priv->child_count, int); + output_dirty = alloca0 (priv->child_count * sizeof (int)); + output_matrix = ALLOC_MATRIX (priv->child_count, int); + + afr_selfheal_extract_xattr (this, replies, type, input_dirty, + input_matrix); + + for (i = 0; i < priv->child_count; i++) { + if (i == source || !healed_sinks[i]) + continue; + output_dirty[i] = -input_dirty[i]; + output_matrix[i][source] = -input_matrix[i][source]; + } + + for (i = 0; i < priv->child_count; i++) { + if (!healed_sinks[i] || !locked_on[i]) + continue; + xattr = afr_selfheal_output_xattr (this, _gf_false, type, + output_dirty, output_matrix, + i, NULL); + + afr_selfheal_post_op (frame, this, inode, i, xattr, xdata); + + undid_pending[i] = 1; + dict_unref (xattr); + } + + if (xdata) + dict_unref (xdata); + + return 0; } gf_boolean_t @@ -1895,11 +1992,15 @@ afr_selfheal (xlator_t *this, uuid_t gfid) { int ret = -1; call_frame_t *frame = NULL; + afr_local_t *local = NULL; frame = afr_frame_create (this); if (!frame) return ret; + local = frame->local; + local->xdata_req = dict_new(); + ret = afr_selfheal_do (frame, this, gfid); if (frame) diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c index 87542799a5b..9c12e433097 100644 --- a/xlators/cluster/afr/src/afr-self-heal-data.c +++ b/xlators/cluster/afr/src/afr-self-heal-data.c @@ -581,6 +581,7 @@ __afr_selfheal_data_finalize_source (call_frame_t *frame, xlator_t *this, unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on, + unsigned char *undid_pending, struct afr_reply *replies, uint64_t *witness) { @@ -601,6 +602,12 @@ __afr_selfheal_data_finalize_source (call_frame_t *frame, xlator_t *this, AFR_DATA_TRANSACTION); if (source < 0) return -EIO; + + _afr_fav_child_reset_sink_xattrs (frame, this, inode, source, + healed_sinks, undid_pending, + AFR_DATA_TRANSACTION, + locked_on, replies); + return source; } @@ -640,6 +647,7 @@ __afr_selfheal_data_prepare (call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *locked_on, unsigned char *sources, unsigned char *sinks, unsigned char *healed_sinks, + unsigned char *undid_pending, struct afr_reply *replies, gf_boolean_t *pflag) { int ret = -1; @@ -675,8 +683,8 @@ __afr_selfheal_data_prepare (call_frame_t *frame, xlator_t *this, source = __afr_selfheal_data_finalize_source (frame, this, inode, sources, sinks, healed_sinks, - locked_on, replies, - witness); + locked_on, undid_pending, + replies, witness); if (source < 0) return -EIO; @@ -694,6 +702,7 @@ __afr_selfheal_data (call_frame_t *frame, xlator_t *this, fd_t *fd, unsigned char *sinks = NULL; unsigned char *data_lock = NULL; unsigned char *healed_sinks = NULL; + unsigned char *undid_pending = NULL; struct afr_reply *locked_replies = NULL; int source = -1; gf_boolean_t did_sh = _gf_true; @@ -705,6 +714,7 @@ __afr_selfheal_data (call_frame_t *frame, xlator_t *this, fd_t *fd, sinks = alloca0 (priv->child_count); healed_sinks = alloca0 (priv->child_count); data_lock = alloca0 (priv->child_count); + undid_pending = alloca0 (priv->child_count); locked_replies = alloca0 (sizeof (*locked_replies) * priv->child_count); @@ -724,9 +734,8 @@ __afr_selfheal_data (call_frame_t *frame, xlator_t *this, fd_t *fd, ret = __afr_selfheal_data_prepare (frame, this, fd->inode, data_lock, sources, sinks, - healed_sinks, - locked_replies, - NULL); + healed_sinks, undid_pending, + locked_replies, NULL); if (ret < 0) goto unlock; @@ -785,7 +794,7 @@ restore_time: } ret = afr_selfheal_undo_pending (frame, this, fd->inode, sources, sinks, healed_sinks, - AFR_DATA_TRANSACTION, + undid_pending, AFR_DATA_TRANSACTION, locked_replies, data_lock); skip_undo_pending: afr_selfheal_uninodelk (frame, this, fd->inode, this->name, 0, 0, diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c index 573b5435eae..9ab566b8242 100644 --- a/xlators/cluster/afr/src/afr-self-heal-entry.c +++ b/xlators/cluster/afr/src/afr-self-heal-entry.c @@ -897,6 +897,7 @@ __afr_selfheal_entry (call_frame_t *frame, xlator_t *this, fd_t *fd, unsigned char *data_lock = NULL; unsigned char *postop_lock = NULL; unsigned char *healed_sinks = NULL; + unsigned char *undid_pending = NULL; struct afr_reply *locked_replies = NULL; afr_local_t *local = NULL; afr_private_t *priv = NULL; @@ -908,6 +909,7 @@ __afr_selfheal_entry (call_frame_t *frame, xlator_t *this, fd_t *fd, sources = alloca0 (priv->child_count); sinks = alloca0 (priv->child_count); healed_sinks = alloca0 (priv->child_count); + undid_pending = alloca0 (priv->child_count); data_lock = alloca0 (priv->child_count); postop_lock = alloca0 (priv->child_count); @@ -980,6 +982,7 @@ unlock: ret = afr_selfheal_undo_pending (frame, this, fd->inode, sources, sinks, healed_sinks, + undid_pending, AFR_ENTRY_TRANSACTION, locked_replies, postop_lock); } diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c index 85aaca7cefd..a67717bf9d4 100644 --- a/xlators/cluster/afr/src/afr-self-heal-metadata.c +++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c @@ -202,6 +202,7 @@ __afr_selfheal_metadata_finalize_source (call_frame_t *frame, xlator_t *this, unsigned char *sources, unsigned char *sinks, unsigned char *healed_sinks, + unsigned char *undid_pending, unsigned char *locked_on, struct afr_reply *replies) { @@ -223,8 +224,14 @@ __afr_selfheal_metadata_finalize_source (call_frame_t *frame, xlator_t *this, healed_sinks, locked_on, replies, AFR_METADATA_TRANSACTION); - if (source >= 0) + if (source >= 0) { + _afr_fav_child_reset_sink_xattrs (frame, this, inode, + source, healed_sinks, + undid_pending, + AFR_METADATA_TRANSACTION, + locked_on, replies); return source; + } /* If this is a directory mtime/ctime only split brain use the most recent */ @@ -304,6 +311,7 @@ int __afr_selfheal_metadata_prepare (call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *locked_on, unsigned char *sources, unsigned char *sinks, unsigned char *healed_sinks, + unsigned char *undid_pending, struct afr_reply *replies, gf_boolean_t *pflag) { int ret = -1; @@ -358,6 +366,7 @@ __afr_selfheal_metadata_prepare (call_frame_t *frame, xlator_t *this, inode_t *i source = __afr_selfheal_metadata_finalize_source (frame, this, inode, sources, sinks, healed_sinks, + undid_pending, locked_on, replies); if (source < 0) @@ -375,6 +384,7 @@ afr_selfheal_metadata (call_frame_t *frame, xlator_t *this, inode_t *inode) unsigned char *sinks = NULL; unsigned char *data_lock = NULL; unsigned char *healed_sinks = NULL; + unsigned char *undid_pending = NULL; struct afr_reply *locked_replies = NULL; gf_boolean_t did_sh = _gf_true; int source = -1; @@ -384,6 +394,7 @@ afr_selfheal_metadata (call_frame_t *frame, xlator_t *this, inode_t *inode) sources = alloca0 (priv->child_count); sinks = alloca0 (priv->child_count); healed_sinks = alloca0 (priv->child_count); + undid_pending = alloca0 (priv->child_count); data_lock = alloca0 (priv->child_count); locked_replies = alloca0 (sizeof (*locked_replies) * priv->child_count); @@ -399,6 +410,7 @@ afr_selfheal_metadata (call_frame_t *frame, xlator_t *this, inode_t *inode) ret = __afr_selfheal_metadata_prepare (frame, this, inode, data_lock, sources, sinks, healed_sinks, + undid_pending, locked_replies, NULL); if (ret < 0) goto unlock; @@ -417,6 +429,7 @@ afr_selfheal_metadata (call_frame_t *frame, xlator_t *this, inode_t *inode) ret = afr_selfheal_undo_pending (frame, this, inode, sources, sinks, healed_sinks, + undid_pending, AFR_METADATA_TRANSACTION, locked_replies, data_lock); } diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h index 1705c967f60..500227abe24 100644 --- a/xlators/cluster/afr/src/afr-self-heal.h +++ b/xlators/cluster/afr/src/afr-self-heal.h @@ -167,8 +167,10 @@ afr_selfheal_extract_xattr (xlator_t *this, struct afr_reply *replies, int afr_selfheal_undo_pending (call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *sources, unsigned char *sinks, - unsigned char *healed_sinks, afr_transaction_type type, - struct afr_reply *replies, unsigned char *locked_on); + unsigned char *healed_sinks, + unsigned char *undid_pending, + afr_transaction_type type, struct afr_reply *replies, + unsigned char *locked_on); int afr_selfheal_recreate_entry (xlator_t *this, int dst, int source, inode_t *dir, @@ -229,6 +231,19 @@ afr_mark_split_brain_source_sinks (call_frame_t *frame, xlator_t *this, afr_transaction_type type); int +afr_sh_get_fav_by_policy (xlator_t *this, struct afr_reply *replies, + inode_t *inode, char **policy_str); + +int +_afr_fav_child_reset_sink_xattrs (call_frame_t *frame, xlator_t *this, + inode_t *inode, int source, + unsigned char *healed_sinks, + unsigned char *undid_pending, + afr_transaction_type type, + unsigned char *locked_on, + struct afr_reply *replies); + +int afr_get_child_index_from_name (xlator_t *this, char *name); gf_boolean_t @@ -239,8 +254,8 @@ __afr_selfheal_data_prepare (call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *locked_on, unsigned char *sources, unsigned char *sinks, unsigned char *healed_sinks, - struct afr_reply *replies, - gf_boolean_t *flag); + unsigned char *undid_pending, + struct afr_reply *replies, gf_boolean_t *flag); int __afr_selfheal_metadata_prepare (call_frame_t *frame, xlator_t *this, @@ -248,6 +263,7 @@ __afr_selfheal_metadata_prepare (call_frame_t *frame, xlator_t *this, unsigned char *sources, unsigned char *sinks, unsigned char *healed_sinks, + unsigned char *undid_pending, struct afr_reply *replies, gf_boolean_t *flag); int diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c index 4d0ed943a81..9cb735ea7fa 100644 --- a/xlators/cluster/afr/src/afr-transaction.c +++ b/xlators/cluster/afr/src/afr-transaction.c @@ -2261,19 +2261,12 @@ afr_write_txn_refresh_done (call_frame_t *frame, xlator_t *this, int err) if (err) { local->op_errno = -err; local->op_ret = -1; - goto fail; - } - ret = afr_inode_get_readable (frame, local->inode, this, - local->readable, NULL, - local->transaction.type); - if (ret < 0) { gf_msg (this->name, GF_LOG_ERROR, -ret, AFR_MSG_SPLIT_BRAIN, "Failing %s on gfid %s: split-brain observed.", gf_fop_list[local->op], uuid_utoa (local->inode->gfid)); - local->op_ret = -1; - local->op_errno = -ret; goto fail; } + afr_transaction_start (frame, this); return 0; fail: diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h index d09aa6852c8..bbfa309b868 100644 --- a/xlators/cluster/afr/src/afr.h +++ b/xlators/cluster/afr/src/afr.h @@ -808,6 +808,7 @@ typedef struct _afr_local { call_frame_t *heal_frame; gf_boolean_t need_full_crawl; + gf_boolean_t is_read_txn; } afr_local_t; diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c index 255c0823aac..5e7f5fa4eb1 100644 --- a/xlators/cluster/dht/src/dht-helper.c +++ b/xlators/cluster/dht/src/dht-helper.c @@ -1320,6 +1320,9 @@ unlock: UNLOCK(&inode->lock); out: + if (dict) { + dict_unref (dict); + } loc_wipe (&tmp_loc); @@ -1588,6 +1591,10 @@ unlock: ret = 0; out: + if (dict) { + dict_unref (dict); + } + loc_wipe (&tmp_loc); return ret; } diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c index 10fd878041e..ebc8a9c2492 100644 --- a/xlators/cluster/dht/src/dht-rebalance.c +++ b/xlators/cluster/dht/src/dht-rebalance.c @@ -2351,7 +2351,7 @@ gf_defrag_get_entry (xlator_t *this, int i, struct dht_container **container, loc_t *loc, dht_conf_t *conf, gf_defrag_info_t *defrag, fd_t *fd, dict_t *migrate_data, struct dir_dfmeta *dir_dfmeta, dict_t *xattr_req, - int *should_commit_hash) + int *should_commit_hash, int *perrno) { int ret = -1; char is_linkfile = 0; @@ -2362,7 +2362,6 @@ gf_defrag_get_entry (xlator_t *this, int i, struct dht_container **container, struct dht_container *tmp_container = NULL; xlator_t *hashed_subvol = NULL; xlator_t *cached_subvol = NULL; - int fop_errno = 0; if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) { ret = -1; @@ -2390,7 +2389,7 @@ gf_defrag_get_entry (xlator_t *this, int i, struct dht_container **container, DHT_MSG_MIGRATE_DATA_FAILED, "Readdirp failed. Aborting data migration for " "directory: %s", loc->path); - fop_errno = -ret; + *perrno = -ret; ret = -1; goto out; } @@ -2647,13 +2646,12 @@ out: dict_unref (xattr_rsp); - errno = fop_errno; return ret; } int gf_defrag_process_dir (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, - dict_t *migrate_data) + dict_t *migrate_data, int *perrno) { int ret = -1; fd_t *fd = NULL; @@ -2673,7 +2671,6 @@ gf_defrag_process_dir (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, int throttle_up = 0; struct dir_dfmeta *dir_dfmeta = NULL; int should_commit_hash = 1; - int fop_errno = 0; gf_log (this->name, GF_LOG_INFO, "migrate data called on %s", loc->path); @@ -2700,7 +2697,7 @@ gf_defrag_process_dir (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, DHT_MSG_MIGRATE_DATA_FAILED, "Migrate data failed: Failed to open dir %s", loc->path); - fop_errno = -ret; + *perrno = -ret; ret = -1; goto out; } @@ -2846,11 +2843,10 @@ gf_defrag_process_dir (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, loc, conf, defrag, fd, migrate_data, dir_dfmeta, xattr_req, - &should_commit_hash); + &should_commit_hash, perrno); if (ret) { - fop_errno = errno; - gf_log ("this->name", GF_LOG_WARNING, "Found " + gf_log (this->name, GF_LOG_WARNING, "Found " "error from gf_defrag_get_entry"); ret = -1; @@ -2910,7 +2906,6 @@ out: ret = 2; } - errno = fop_errno; return ret; } int @@ -3096,6 +3091,7 @@ gf_defrag_fix_layout (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, inode_t *linked_inode = NULL, *inode = NULL; dht_conf_t *conf = NULL; int should_commit_hash = 1; + int perrno = 0; conf = this->private; if (!conf) { @@ -3116,7 +3112,7 @@ gf_defrag_fix_layout (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, } if (-ret == ENOENT || -ret == ESTALE) { - gf_msg (this->name, GF_LOG_INFO, errno, + gf_msg (this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_LOOKUP_FAILED, "Dir:%s renamed or removed. Skipping", loc->path); @@ -3134,10 +3130,11 @@ gf_defrag_fix_layout (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) && (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX)) { - ret = gf_defrag_process_dir (this, defrag, loc, migrate_data); + ret = gf_defrag_process_dir (this, defrag, loc, migrate_data, + &perrno); - if (ret && ret != 2) { - if (errno == ENOENT || errno == ESTALE) { + if (ret && (ret != 2)) { + if (perrno == ENOENT || perrno == ESTALE) { ret = 0; goto out; } else { @@ -3292,7 +3289,7 @@ gf_defrag_fix_layout (xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, NULL, NULL); if (ret) { if (-ret == ENOENT || -ret == ESTALE) { - gf_msg (this->name, GF_LOG_INFO, errno, + gf_msg (this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_LOOKUP_FAILED, "Dir:%s renamed or removed. " "Skipping", loc->path); diff --git a/xlators/mgmt/glusterd/src/glusterd-geo-rep.c b/xlators/mgmt/glusterd/src/glusterd-geo-rep.c index 9a702b70313..66f26fc5256 100644 --- a/xlators/mgmt/glusterd/src/glusterd-geo-rep.c +++ b/xlators/mgmt/glusterd/src/glusterd-geo-rep.c @@ -1346,7 +1346,6 @@ static int _get_slave_idx_slave_voluuid (dict_t *dict, char *key, data_t *value, void *data) { - char *slave_voluuid = NULL; char *slave_info = NULL; xlator_t *this = NULL; struct slave_vol_config *slave_cfg = NULL; @@ -2863,10 +2862,8 @@ out: static int get_slavehost_from_voluuid (dict_t *dict, char *key, data_t *value, void *data) { - char *slave_voluuid = NULL; char *slave_info = NULL; char *tmp = NULL; - char tmp_char = 0; char *slave_host = NULL; xlator_t *this = NULL; struct slave_vol_config *slave_vol = NULL; @@ -3004,14 +3001,10 @@ glusterd_op_stage_gsync_create (dict_t *dict, char **op_errstr) glusterd_volinfo_t *volinfo = NULL; struct stat stbuf = {0,}; xlator_t *this = NULL; - char *georep_session_wrkng_dir = NULL; struct slave_vol_config slave1 = {{0},}; - int type = 0; - char monitor_status[NAME_MAX] = {0,}; char old_slave_url[SLAVE_URL_INFO_MAX] = {0}; char old_confpath[PATH_MAX] = {0}; gf_boolean_t is_running = _gf_false; - int ret_status = 0; char *statedir = NULL; char statefiledir[PATH_MAX] = {0,}; gf_boolean_t is_different_slavehost = _gf_false; @@ -3429,7 +3422,6 @@ glusterd_op_stage_gsync_set (dict_t *dict, char **op_errstr) { int ret = 0; int type = 0; - int pfd = -1; char *volname = NULL; char *slave = NULL; char *slave_url = NULL; @@ -3444,7 +3436,6 @@ glusterd_op_stage_gsync_set (dict_t *dict, char **op_errstr) gf_boolean_t exists = _gf_false; glusterd_volinfo_t *volinfo = NULL; char errmsg[PATH_MAX] = {0,}; - char pidfile[PATH_MAX] = {0,}; dict_t *ctx = NULL; gf_boolean_t is_force = 0; gf_boolean_t is_running = _gf_false; @@ -3691,16 +3682,6 @@ glusterd_op_stage_gsync_set (dict_t *dict, char **op_errstr) goto out; } - pfd = gsyncd_getpidfile (volname, slave, pidfile, - conf_path, &is_template_in_use); - if (is_template_in_use) { - snprintf (errmsg, sizeof(errmsg), "pid-file entry " - "missing in the config file(%s).", - conf_path); - ret = -1; - goto out; - } - ret = gsync_verify_config_options (dict, op_errstr, volname); goto out; break; @@ -3748,6 +3729,7 @@ glusterd_op_stage_gsync_set (dict_t *dict, char **op_errstr) } out: + if (path_list) GF_FREE (path_list); @@ -3772,7 +3754,6 @@ gd_pause_or_resume_gsync (dict_t *dict, char *master, char *slave, char pidfile[PATH_MAX] = {0,}; char errmsg[PATH_MAX] = ""; char buf [1024] = {0,}; - int i = 0; gf_boolean_t is_template_in_use = _gf_false; char monitor_status[NAME_MAX] = {0,}; char *statefile = NULL; @@ -4408,13 +4389,10 @@ int glusterd_read_status_file (glusterd_volinfo_t *volinfo, char *slave, char *conf_path, dict_t *dict, char *node) { - char brick_state_file[PATH_MAX] = ""; - char brick_path[PATH_MAX] = ""; char temp_conf_path[PATH_MAX] = ""; char *working_conf_path = NULL; char *georep_session_wrkng_dir = NULL; char *master = NULL; - char tmp[1024] = ""; char sts_val_name[1024] = ""; char monitor_status[NAME_MAX] = ""; char *statefile = NULL; @@ -4429,14 +4407,12 @@ glusterd_read_status_file (glusterd_volinfo_t *volinfo, char *slave, char *brick_host_uuid = NULL; int brick_host_uuid_length = 0; int gsync_count = 0; - int i = 0; int ret = 0; glusterd_brickinfo_t *brickinfo = NULL; gf_gsync_status_t *sts_val = NULL; gf_boolean_t is_template_in_use = _gf_false; glusterd_conf_t *priv = NULL; struct stat stbuf = {0,}; - dict_t *statusd = NULL; xlator_t *this = NULL; this = THIS; @@ -4845,7 +4821,6 @@ glusterd_get_gsync_status_mst_slv (glusterd_volinfo_t *volinfo, { char *statefile = NULL; uuid_t uuid = {0, }; - glusterd_conf_t *priv = NULL; int ret = 0; gf_boolean_t is_template_in_use = _gf_false; struct stat stbuf = {0, }; @@ -4858,8 +4833,6 @@ glusterd_get_gsync_status_mst_slv (glusterd_volinfo_t *volinfo, GF_ASSERT (slave); GF_ASSERT (this->private); - priv = this->private; - ret = glusterd_gsync_get_uuid (slave, volinfo, uuid); if (ret) { gf_msg (this->name, GF_LOG_INFO, 0, GD_MSG_SESSION_INACTIVE, @@ -5489,7 +5462,6 @@ glusterd_op_gsync_set (dict_t *dict, char **op_errstr, dict_t *rsp_dict) char *status_msg = NULL; gf_boolean_t is_running = _gf_false; char *conf_path = NULL; - char errmsg[PATH_MAX] = ""; char *key = NULL; xlator_t *this = NULL; @@ -6277,8 +6249,6 @@ glusterd_op_gsync_create (dict_t *dict, char **op_errstr, dict_t *rsp_dict) xlator_t *this = NULL; char old_working_dir[PATH_MAX] = {0}; char new_working_dir[PATH_MAX] = {0}; - char *slave_info = NULL; - char slave_url_info[SLAVE_URL_INFO_MAX] = {0}; char *slave_voluuid = NULL; char *old_slavehost = NULL; gf_boolean_t is_existing_session = _gf_false; diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix.c index c40a087ec46..8f85f8c8ba1 100644 --- a/xlators/storage/posix/src/posix.c +++ b/xlators/storage/posix/src/posix.c @@ -374,7 +374,8 @@ posix_do_chown (xlator_t *this, static int posix_do_utimes (xlator_t *this, const char *path, - struct iatt *stbuf) + struct iatt *stbuf, + int valid) { int32_t ret = -1; struct timeval tv[2] = {{0,},{0,}}; @@ -391,10 +392,23 @@ posix_do_utimes (xlator_t *this, if (S_ISLNK (stat.st_mode)) is_symlink = 1; - tv[0].tv_sec = stbuf->ia_atime; - tv[0].tv_usec = stbuf->ia_atime_nsec / 1000; - tv[1].tv_sec = stbuf->ia_mtime; - tv[1].tv_usec = stbuf->ia_mtime_nsec / 1000; + if ((valid & GF_SET_ATTR_ATIME) == GF_SET_ATTR_ATIME) { + tv[0].tv_sec = stbuf->ia_atime; + tv[0].tv_usec = stbuf->ia_atime_nsec / 1000; + } else { + /* atime is not given, use current values */ + tv[0].tv_sec = ST_ATIM_SEC (&stat); + tv[0].tv_usec = ST_ATIM_NSEC (&stat) / 1000; + } + + if ((valid & GF_SET_ATTR_MTIME) == GF_SET_ATTR_MTIME) { + tv[1].tv_sec = stbuf->ia_mtime; + tv[1].tv_usec = stbuf->ia_mtime_nsec / 1000; + } else { + /* mtime is not given, use current values */ + tv[1].tv_sec = ST_MTIM_SEC (&stat); + tv[1].tv_usec = ST_MTIM_NSEC (&stat) / 1000; + } ret = lutimes (path, tv); if ((ret == -1) && (errno == ENOSYS)) { @@ -463,7 +477,7 @@ posix_setattr (call_frame_t *frame, xlator_t *this, } if (valid & (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME)) { - op_ret = posix_do_utimes (this, real_path, stbuf); + op_ret = posix_do_utimes (this, real_path, stbuf, valid); if (op_ret == -1) { op_errno = errno; gf_msg (this->name, GF_LOG_ERROR, errno, |