diff options
Diffstat (limited to 'xlators')
201 files changed, 13236 insertions, 7725 deletions
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c index a2f0b2a..032ab5c 100644 --- a/xlators/cluster/afr/src/afr-common.c +++ b/xlators/cluster/afr/src/afr-common.c @@ -45,7 +45,42 @@ afr_quorum_errno(afr_private_t *priv) return ENOTCONN; } -static void +gf_boolean_t +afr_is_private_directory(afr_private_t *priv, uuid_t pargfid, const char *name, + pid_t pid) +{ + if (!__is_root_gfid(pargfid)) { + return _gf_false; + } + + if (strcmp(name, GF_REPLICATE_TRASH_DIR) == 0) { + /*For backward compatibility /.landfill is private*/ + return _gf_true; + } + + if (pid == GF_CLIENT_PID_GSYNCD) { + /*geo-rep needs to create/sync private directory on slave because + * it appears in changelog*/ + return _gf_false; + } + + if (pid == GF_CLIENT_PID_GLFS_HEAL || pid == GF_CLIENT_PID_SELF_HEALD) { + if (strcmp(name, priv->anon_inode_name) == 0) { + /* anonymous-inode dir is private*/ + return _gf_true; + } + } else { + if (strncmp(name, AFR_ANON_DIR_PREFIX, strlen(AFR_ANON_DIR_PREFIX)) == + 0) { + /* anonymous-inode dir prefix is private for geo-rep to work*/ + return _gf_true; + } + } + + return _gf_false; +} + +void afr_fill_success_replies(afr_local_t *local, afr_private_t *priv, unsigned char *replies) { @@ -885,7 +920,7 @@ __afr_set_in_flight_sb_status(xlator_t *this, afr_local_t *local, metadatamap |= (1 << index); } if (metadatamap_old != metadatamap) { - event = 0; + __afr_inode_need_refresh_set(inode, this); } break; @@ -898,7 +933,7 @@ __afr_set_in_flight_sb_status(xlator_t *this, afr_local_t *local, datamap |= (1 << index); } if (datamap_old != datamap) - event = 0; + __afr_inode_need_refresh_set(inode, this); break; default: @@ -1062,34 +1097,6 @@ out: } int -__afr_inode_event_gen_reset_small(inode_t *inode, xlator_t *this) -{ - int ret = -1; - uint16_t datamap = 0; - uint16_t metadatamap = 0; - uint32_t event = 0; - uint64_t val = 0; - afr_inode_ctx_t *ctx = NULL; - - ret = __afr_inode_ctx_get(this, inode, &ctx); - if (ret) - return ret; - - val = ctx->read_subvol; - - metadatamap = (val & 0x000000000000ffff) >> 0; - datamap = (val & 0x00000000ffff0000) >> 16; - event = 0; - - val = ((uint64_t)metadatamap) | (((uint64_t)datamap) << 16) | - (((uint64_t)event) << 32); - - ctx->read_subvol = val; - - return ret; -} - -int __afr_inode_read_subvol_get(inode_t *inode, xlator_t *this, unsigned char *data, unsigned char *metadata, int *event_p) { @@ -1160,22 +1167,6 @@ out: } int -__afr_inode_event_gen_reset(inode_t *inode, xlator_t *this) -{ - afr_private_t *priv = NULL; - int ret = -1; - - priv = this->private; - - if (priv->child_count <= 16) - ret = __afr_inode_event_gen_reset_small(inode, this); - else - ret = -1; - - return ret; -} - -int afr_inode_read_subvol_get(inode_t *inode, xlator_t *this, unsigned char *data, unsigned char *metadata, int *event_p) { @@ -1241,12 +1232,11 @@ afr_inode_get_readable(call_frame_t *frame, inode_t *inode, xlator_t *this, return 0; } -int +static int afr_inode_split_brain_choice_get(inode_t *inode, xlator_t *this, int *spb_choice) { int ret = -1; - GF_VALIDATE_OR_GOTO(this->name, inode, out); LOCK(&inode->lock); @@ -1258,6 +1248,40 @@ out: return ret; } +/* + * frame is used to get the favourite policy. Since + * afr_inode_split_brain_choice_get was called with afr_open, it is possible to + * have a frame with out local->replies. So in that case, frame is passed as + * null, hence this function will handle the frame NULL case. + */ +int +afr_split_brain_read_subvol_get(inode_t *inode, xlator_t *this, + call_frame_t *frame, int *spb_subvol) +{ + int ret = -1; + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + + GF_VALIDATE_OR_GOTO("afr", this, out); + GF_VALIDATE_OR_GOTO(this->name, this->private, out); + GF_VALIDATE_OR_GOTO(this->name, inode, out); + GF_VALIDATE_OR_GOTO(this->name, spb_subvol, out); + + priv = this->private; + + ret = afr_inode_split_brain_choice_get(inode, this, spb_subvol); + if (*spb_subvol < 0 && priv->fav_child_policy && frame && frame->local) { + local = frame->local; + *spb_subvol = afr_sh_get_fav_by_policy(this, local->replies, inode, + NULL); + if (*spb_subvol >= 0) { + ret = 0; + } + } + +out: + return ret; +} int afr_inode_read_subvol_set(inode_t *inode, xlator_t *this, unsigned char *data, unsigned char *metadata, int event) @@ -1324,30 +1348,22 @@ out: return need_refresh; } -static int -afr_inode_need_refresh_set(inode_t *inode, xlator_t *this) +int +__afr_inode_need_refresh_set(inode_t *inode, xlator_t *this) { int ret = -1; afr_inode_ctx_t *ctx = NULL; - GF_VALIDATE_OR_GOTO(this->name, inode, out); - - LOCK(&inode->lock); - { - ret = __afr_inode_ctx_get(this, inode, &ctx); - if (ret) - goto unlock; - + ret = __afr_inode_ctx_get(this, inode, &ctx); + if (ret == 0) { ctx->need_refresh = _gf_true; } -unlock: - UNLOCK(&inode->lock); -out: + return ret; } int -afr_inode_event_gen_reset(inode_t *inode, xlator_t *this) +afr_inode_need_refresh_set(inode_t *inode, xlator_t *this) { int ret = -1; @@ -1355,7 +1371,7 @@ afr_inode_event_gen_reset(inode_t *inode, xlator_t *this) LOCK(&inode->lock); { - ret = __afr_inode_event_gen_reset(inode, this); + ret = __afr_inode_need_refresh_set(inode, this); } UNLOCK(&inode->lock); out: @@ -1790,7 +1806,7 @@ afr_txn_refresh_done(call_frame_t *frame, xlator_t *this, int err) ret = afr_inode_get_readable(frame, inode, this, local->readable, &event_generation, local->transaction.type); - if (ret == -EIO || (local->is_read_txn && !event_generation)) { + if (ret == -EIO) { /* No readable subvolume even after refresh ==> splitbrain.*/ if (!priv->fav_child_policy) { err = EIO; @@ -2290,8 +2306,9 @@ afr_hash_child(afr_read_subvol_args_t *args, afr_private_t *priv, * need is a low probability that multiple clients * won't converge on the same subvolume. */ + gf_uuid_copy(gfid_copy, args->gfid); pid = getpid(); - memcpy(gfid_copy, &pid, sizeof(pid)); + *(pid_t *)gfid_copy ^= pid; } child = SuperFastHash((char *)gfid_copy, sizeof(gfid_copy)) % priv->child_count; @@ -2875,7 +2892,7 @@ afr_attempt_readsubvol_set(call_frame_t *frame, xlator_t *this, { afr_private_t *priv = NULL; afr_local_t *local = NULL; - int spb_choice = -1; + int spb_subvol = -1; int child_count = -1; if (*read_subvol != -1) @@ -2885,10 +2902,10 @@ afr_attempt_readsubvol_set(call_frame_t *frame, xlator_t *this, local = frame->local; child_count = priv->child_count; - afr_inode_split_brain_choice_get(local->inode, this, &spb_choice); - if ((spb_choice >= 0) && + afr_split_brain_read_subvol_get(local->inode, this, frame, &spb_subvol); + if ((spb_subvol >= 0) && (AFR_COUNT(success_replies, child_count) == child_count)) { - *read_subvol = spb_choice; + *read_subvol = spb_subvol; } else if (!priv->quorum_count || frame->root->pid == GF_CLIENT_PID_GLFS_HEAL) { *read_subvol = afr_first_up_child(frame, this); @@ -2929,6 +2946,7 @@ afr_lookup_done(call_frame_t *frame, xlator_t *this) 0, }; gf_boolean_t locked_entry = _gf_false; + gf_boolean_t in_flight_create = _gf_false; gf_boolean_t can_interpret = _gf_true; inode_t *parent = NULL; ia_type_t ia_type = IA_INVAL; @@ -2972,17 +2990,12 @@ afr_lookup_done(call_frame_t *frame, xlator_t *this) if (!replies[i].valid) continue; - if (locked_entry && replies[i].op_ret == -1 && - replies[i].op_errno == ENOENT) { - /* Second, check entry is still - "underway" in creation */ - local->op_ret = -1; - local->op_errno = ENOENT; - goto error; - } - - if (replies[i].op_ret == -1) + if (replies[i].op_ret == -1) { + if (locked_entry && replies[i].op_errno == ENOENT) { + in_flight_create = _gf_true; + } continue; + } if (read_subvol == -1 || !readable[read_subvol]) { read_subvol = i; @@ -2992,6 +3005,12 @@ afr_lookup_done(call_frame_t *frame, xlator_t *this) } } + if (in_flight_create && !afr_has_quorum(success_replies, this, NULL)) { + local->op_ret = -1; + local->op_errno = ENOENT; + goto error; + } + if (read_subvol == -1) goto error; /* We now have a read_subvol, which is readable[] (if there @@ -3050,7 +3069,7 @@ afr_lookup_done(call_frame_t *frame, xlator_t *this) if (read_subvol == -1) goto cant_interpret; if (ret) { - afr_inode_event_gen_reset(local->inode, this); + afr_inode_need_refresh_set(local->inode, this); dict_del_sizen(local->replies[read_subvol].xdata, GF_CONTENT_KEY); } } else { @@ -3103,7 +3122,7 @@ error: * others in that they must be given higher priority while * returning to the user. * - * The hierarchy is ENODATA > ENOENT > ESTALE > others + * The hierarchy is ENODATA > ENOENT > ESTALE > ENOSPC others */ int @@ -3115,6 +3134,8 @@ afr_higher_errno(int32_t old_errno, int32_t new_errno) return ENOENT; if (old_errno == ESTALE || new_errno == ESTALE) return ESTALE; + if (old_errno == ENOSPC || new_errno == ENOSPC) + return ENOSPC; return new_errno; } @@ -3606,6 +3627,7 @@ afr_discover_unwind(call_frame_t *frame, xlator_t *this) afr_private_t *priv = NULL; afr_local_t *local = NULL; int read_subvol = -1; + int ret = 0; unsigned char *data_readable = NULL; unsigned char *success_replies = NULL; @@ -3627,7 +3649,10 @@ afr_discover_unwind(call_frame_t *frame, xlator_t *this) if (!afr_has_quorum(success_replies, this, frame)) goto unwind; - afr_replies_interpret(frame, this, local->inode, NULL); + ret = afr_replies_interpret(frame, this, local->inode, NULL); + if (ret) { + afr_inode_need_refresh_set(local->inode, this); + } read_subvol = afr_read_subvol_decide(local->inode, this, NULL, data_readable); @@ -3679,7 +3704,7 @@ afr_ta_id_file_check(void *opaque) this = opaque; priv = this->private; - ret = afr_fill_ta_loc(this, &loc); + ret = afr_fill_ta_loc(this, &loc, _gf_false); if (ret) { gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB, "Failed to populate thin-arbiter loc for: %s.", loc.name); @@ -3888,11 +3913,7 @@ afr_discover(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req) afr_read_subvol_get(loc->inode, this, NULL, NULL, &event, AFR_DATA_TRANSACTION, NULL); - if (afr_is_inode_refresh_reqd(loc->inode, this, event, - local->event_generation)) - afr_inode_refresh(frame, this, loc->inode, NULL, afr_discover_do); - else - afr_discover_do(frame, this, 0); + afr_discover_do(frame, this, 0); return 0; out: @@ -3993,11 +4014,10 @@ afr_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req) return 0; } - if (__is_root_gfid(loc->parent->gfid)) { - if (!strcmp(loc->name, GF_REPLICATE_TRASH_DIR)) { - op_errno = EPERM; - goto out; - } + if (afr_is_private_directory(this->private, loc->parent->gfid, loc->name, + frame->root->pid)) { + op_errno = EPERM; + goto out; } local = AFR_FRAME_INIT(frame, op_errno); @@ -4033,11 +4053,7 @@ afr_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req) afr_read_subvol_get(loc->parent, this, NULL, NULL, &event, AFR_DATA_TRANSACTION, NULL); - if (afr_is_inode_refresh_reqd(loc->inode, this, event, - local->event_generation)) - afr_inode_refresh(frame, this, loc->parent, NULL, afr_lookup_do); - else - afr_lookup_do(frame, this, 0); + afr_lookup_do(frame, this, 0); return 0; out: @@ -5665,6 +5681,8 @@ afr_priv_dump(xlator_t *this) GF_ATOMIC_GET(priv->pending_reads[i])); sprintf(key, "child_latency[%d]", i); gf_proc_dump_write(key, "%" PRId64, priv->child_latency[i]); + sprintf(key, "halo_child_up[%d]", i); + gf_proc_dump_write(key, "%d", priv->halo_child_up[i]); } gf_proc_dump_write("data_self_heal", "%d", priv->data_self_heal); gf_proc_dump_write("metadata_self_heal", "%d", priv->metadata_self_heal); @@ -5677,6 +5695,7 @@ afr_priv_dump(xlator_t *this) priv->background_self_heal_count); gf_proc_dump_write("healers", "%d", priv->healers); gf_proc_dump_write("read-hash-mode", "%d", priv->hash_mode); + gf_proc_dump_write("use-anonymous-inode", "%d", priv->use_anon_inode); if (priv->quorum_count == AFR_QUORUM_AUTO) { gf_proc_dump_write("quorum-type", "auto"); } else if (priv->quorum_count == 0) { @@ -5737,13 +5756,31 @@ __afr_get_up_children_count(afr_private_t *priv) return up_children; } +static int +__get_heard_from_all_status(xlator_t *this) +{ + afr_private_t *priv = this->private; + int i; + + for (i = 0; i < priv->child_count; i++) { + if (!priv->last_event[i]) { + return 0; + } + } + if (priv->thin_arbiter_count && !priv->ta_child_up) { + return 0; + } + return 1; +} + glusterfs_event_t -__afr_transform_event_from_state(afr_private_t *priv) +__afr_transform_event_from_state(xlator_t *this) { int i = 0; int up_children = 0; + afr_private_t *priv = this->private; - if (AFR_COUNT(priv->last_event, priv->child_count) == priv->child_count) + if (__get_heard_from_all_status(this)) /* have_heard_from_all. Let afr_notify() do the propagation. */ return GF_EVENT_MAXVAL; @@ -5785,7 +5822,7 @@ afr_notify_cbk(void *data) goto unlock; } priv->timer = NULL; - event = __afr_transform_event_from_state(priv); + event = __afr_transform_event_from_state(this); if (event != GF_EVENT_MAXVAL) propagate = _gf_true; } @@ -5813,20 +5850,6 @@ __afr_launch_notify_timer(xlator_t *this, afr_private_t *priv) } static int -__get_heard_from_all_status(xlator_t *this) -{ - afr_private_t *priv = this->private; - int i; - - for (i = 0; i < priv->child_count; i++) { - if (!priv->last_event[i]) { - return 0; - } - } - return 1; -} - -static int find_best_down_child(xlator_t *this) { afr_private_t *priv = NULL; @@ -5837,7 +5860,7 @@ find_best_down_child(xlator_t *this) priv = this->private; for (i = 0; i < priv->child_count; i++) { - if (priv->child_up[i] && priv->child_latency[i] >= 0 && + if (!priv->child_up[i] && priv->child_latency[i] >= 0 && priv->child_latency[i] < best_latency) { best_child = i; best_latency = priv->child_latency[i]; @@ -5909,7 +5932,9 @@ __afr_handle_ping_event(xlator_t *this, xlator_t *child_xlator, const int idx, "), " "marking child down.", child_latency_msec, halo_max_latency_msec); - *event = GF_EVENT_CHILD_DOWN; + if (priv->halo_child_up[idx]) { + *event = GF_EVENT_CHILD_DOWN; + } } } else if (child_latency_msec < halo_max_latency_msec && priv->child_up[idx] == 0) { @@ -5921,7 +5946,9 @@ __afr_handle_ping_event(xlator_t *this, xlator_t *child_xlator, const int idx, "), " "marking child up.", child_latency_msec, halo_max_latency_msec); - *event = GF_EVENT_CHILD_UP; + if (priv->halo_child_up[idx]) { + *event = GF_EVENT_CHILD_UP; + } } else { gf_log(child_xlator->name, GF_LOG_INFO, "Not marking child %d up, " @@ -5988,7 +6015,10 @@ __afr_handle_child_up_event(xlator_t *this, xlator_t *child_xlator, if (child_latency_msec < 0) { /*set to INT64_MAX-1 so that it is found for best_down_child*/ - priv->child_latency[idx] = AFR_HALO_MAX_LATENCY; + priv->halo_child_up[idx] = 1; + if (priv->child_latency[idx] < 0) { + priv->child_latency[idx] = AFR_HALO_MAX_LATENCY; + } } /* @@ -6077,6 +6107,7 @@ __afr_handle_child_down_event(xlator_t *this, xlator_t *child_xlator, int idx, */ if (child_latency_msec < 0) { priv->child_latency[idx] = child_latency_msec; + priv->halo_child_up[idx] = 0; } priv->child_up[idx] = 0; @@ -6641,6 +6672,8 @@ afr_priv_destroy(afr_private_t *priv) if (!priv) goto out; + + GF_FREE(priv->sh_domain); GF_FREE(priv->last_event); child_count = priv->child_count; @@ -6656,7 +6689,9 @@ afr_priv_destroy(afr_private_t *priv) GF_FREE(priv->local); GF_FREE(priv->pending_key); GF_FREE(priv->children); + GF_FREE(priv->anon_inode); GF_FREE(priv->child_up); + GF_FREE(priv->halo_child_up); GF_FREE(priv->child_latency); LOCK_DESTROY(&priv->lock); @@ -6794,8 +6829,8 @@ afr_is_dirty_count_non_unary(xlator_t *this, struct afr_reply *replies, static int afr_update_heal_status(xlator_t *this, struct afr_reply *replies, - char *index_vgfid, ia_type_t ia_type, gf_boolean_t *esh, - gf_boolean_t *dsh, gf_boolean_t *msh) + ia_type_t ia_type, gf_boolean_t *esh, gf_boolean_t *dsh, + gf_boolean_t *msh, unsigned char pending) { int ret = -1; GF_UNUSED int ret1 = 0; @@ -6825,14 +6860,7 @@ afr_update_heal_status(xlator_t *this, struct afr_reply *replies, } } - if (!strcmp(index_vgfid, GF_XATTROP_INDEX_GFID)) { - if (shd_domain_lk_count) { - ret = -EAGAIN; /*For 'possibly-healing'. */ - } else { - ret = 0; /*needs heal. Just set a non -ve value so that it is - assumed as the source index.*/ - } - } else if (!strcmp(index_vgfid, GF_XATTROP_DIRTY_GFID)) { + if (!pending) { if ((afr_is_dirty_count_non_unary(this, replies, ia_type)) || (!io_domain_lk_count)) { /* Needs heal. */ @@ -6841,6 +6869,13 @@ afr_update_heal_status(xlator_t *this, struct afr_reply *replies, /* No heal needed. */ *dsh = *esh = *msh = 0; } + } else { + if (shd_domain_lk_count) { + ret = -EAGAIN; /*For 'possibly-healing'. */ + } else { + ret = 0; /*needs heal. Just set a non -ve value so that it is + assumed as the source index.*/ + } } return ret; } @@ -6848,8 +6883,8 @@ afr_update_heal_status(xlator_t *this, struct afr_reply *replies, /*return EIO, EAGAIN or pending*/ int afr_lockless_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid, - inode_t **inode, char *index_vgfid, - gf_boolean_t *entry_selfheal, gf_boolean_t *data_selfheal, + inode_t **inode, gf_boolean_t *entry_selfheal, + gf_boolean_t *data_selfheal, gf_boolean_t *metadata_selfheal, unsigned char *pending) { int ret = -1; @@ -6908,8 +6943,8 @@ afr_lockless_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid, goto out; } - ret = afr_update_heal_status(this, replies, index_vgfid, (*inode)->ia_type, - &esh, &dsh, &msh); + ret = afr_update_heal_status(this, replies, (*inode)->ia_type, &esh, &dsh, + &msh, *pending); out: *data_selfheal = dsh; *entry_selfheal = esh; @@ -6934,14 +6969,6 @@ afr_get_heal_info(call_frame_t *frame, xlator_t *this, loc_t *loc) char *status = NULL; call_frame_t *heal_frame = NULL; afr_local_t *heal_local = NULL; - afr_local_t *local = NULL; - char *index_vgfid = NULL; - - local = frame->local; - if (dict_get_str(local->xdata_req, "index-vgfid", &index_vgfid)) { - ret = -1; - goto out; - } /*Use frame with lk-owner set*/ heal_frame = afr_frame_create(frame->this, &op_errno); @@ -6952,7 +6979,7 @@ afr_get_heal_info(call_frame_t *frame, xlator_t *this, loc_t *loc) heal_local = heal_frame->local; heal_frame->local = frame->local; - ret = afr_lockless_inspect(heal_frame, this, loc->gfid, &inode, index_vgfid, + ret = afr_lockless_inspect(heal_frame, this, loc->gfid, &inode, &entry_selfheal, &data_selfheal, &metadata_selfheal, &pending); @@ -7427,7 +7454,7 @@ afr_fav_child_reset_sink_xattrs(void *opaque) ret = afr_selfheal_inodelk(heal_frame, this, inode, this->name, 0, 0, locked_on); { - if (ret < AFR_SH_MIN_PARTICIPANTS) + if (ret < priv->child_count) goto data_unlock; ret = __afr_selfheal_data_prepare( heal_frame, this, inode, locked_on, sources, sinks, @@ -7444,7 +7471,7 @@ afr_fav_child_reset_sink_xattrs(void *opaque) ret = afr_selfheal_inodelk(heal_frame, this, inode, this->name, LLONG_MAX - 1, 0, locked_on); { - if (ret < AFR_SH_MIN_PARTICIPANTS) + if (ret < priv->child_count) goto mdata_unlock; ret = __afr_selfheal_metadata_prepare( heal_frame, this, inode, locked_on, sources, sinks, diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c index 74f71fd..f8bf834 100644 --- a/xlators/cluster/afr/src/afr-dir-read.c +++ b/xlators/cluster/afr/src/afr-dir-read.c @@ -67,7 +67,8 @@ afr_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, } int -afr_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd) +afr_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd, + dict_t *xdata) { afr_private_t *priv = NULL; afr_local_t *local = NULL; @@ -163,8 +164,8 @@ afr_validate_read_subvol(inode_t *inode, xlator_t *this, int par_read_subvol) } static void -afr_readdir_transform_entries(gf_dirent_t *subvol_entries, int subvol, - gf_dirent_t *entries, fd_t *fd) +afr_readdir_transform_entries(call_frame_t *frame, gf_dirent_t *subvol_entries, + int subvol, gf_dirent_t *entries, fd_t *fd) { int ret = -1; gf_dirent_t *entry = NULL; @@ -182,8 +183,8 @@ afr_readdir_transform_entries(gf_dirent_t *subvol_entries, int subvol, list_for_each_entry_safe(entry, tmp, &subvol_entries->list, list) { - if (__is_root_gfid(fd->inode->gfid) && - !strcmp(entry->d_name, GF_REPLICATE_TRASH_DIR)) { + if (afr_is_private_directory(priv, fd->inode->gfid, entry->d_name, + frame->root->pid)) { continue; } @@ -227,8 +228,8 @@ afr_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, } if (op_ret >= 0) - afr_readdir_transform_entries(subvol_entries, (long)cookie, &entries, - local->fd); + afr_readdir_transform_entries(frame, subvol_entries, (long)cookie, + &entries, local->fd); AFR_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, xdata); diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c index e96b7d0..b7cceb7 100644 --- a/xlators/cluster/afr/src/afr-dir-write.c +++ b/xlators/cluster/afr/src/afr-dir-write.c @@ -119,11 +119,11 @@ __afr_dir_write_finalize(call_frame_t *frame, xlator_t *this) continue; if (local->replies[i].op_ret < 0) { if (local->inode) - afr_inode_event_gen_reset(local->inode, this); + afr_inode_need_refresh_set(local->inode, this); if (local->parent) - afr_inode_event_gen_reset(local->parent, this); + afr_inode_need_refresh_set(local->parent, this); if (local->parent2) - afr_inode_event_gen_reset(local->parent2, this); + afr_inode_need_refresh_set(local->parent2, this); continue; } @@ -345,6 +345,7 @@ afr_mark_entry_pending_changelog(call_frame_t *frame, xlator_t *this) afr_private_t *priv = NULL; int pre_op_count = 0; int failed_count = 0; + unsigned char *success_replies = NULL; local = frame->local; priv = this->private; @@ -360,9 +361,16 @@ afr_mark_entry_pending_changelog(call_frame_t *frame, xlator_t *this) failed_count = AFR_COUNT(local->transaction.failed_subvols, priv->child_count); + /* FOP succeeded on all bricks. */ if (pre_op_count == priv->child_count && !failed_count) return; + /* FOP did not suceed on quorum no. of bricks. */ + success_replies = alloca0(priv->child_count); + afr_fill_success_replies(local, priv, success_replies); + if (!afr_has_quorum(success_replies, this, NULL)) + return; + if (priv->thin_arbiter_count) { /*Mark new entry using ta file*/ local->is_new_entry = _gf_true; diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c index c01b413..1d6e4f3 100644 --- a/xlators/cluster/afr/src/afr-inode-write.c +++ b/xlators/cluster/afr/src/afr-inode-write.c @@ -2506,6 +2506,7 @@ afr_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync, call_frame_t *transaction_frame = NULL; int ret = -1; int32_t op_errno = ENOMEM; + int8_t last_fsync = 0; AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out); transaction_frame = copy_frame(frame); @@ -2516,10 +2517,16 @@ afr_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync, if (!local) goto out; - if (xdata) + if (xdata) { local->xdata_req = dict_copy_with_ref(xdata, NULL); - else + if (dict_get_int8(xdata, "last-fsync", &last_fsync) == 0) { + if (last_fsync) { + local->transaction.disable_delayed_post_op = _gf_true; + } + } + } else { local->xdata_req = dict_new(); + } if (!local->xdata_req) goto out; diff --git a/xlators/cluster/afr/src/afr-open.c b/xlators/cluster/afr/src/afr-open.c index a5b004f..6485604 100644 --- a/xlators/cluster/afr/src/afr-open.c +++ b/xlators/cluster/afr/src/afr-open.c @@ -137,7 +137,7 @@ afr_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, { afr_private_t *priv = NULL; afr_local_t *local = NULL; - int spb_choice = 0; + int spb_subvol = 0; int event_generation = 0; int ret = 0; int32_t op_errno = 0; @@ -179,9 +179,9 @@ afr_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, ret = afr_inode_get_readable(frame, local->inode, this, NULL, &event_generation, AFR_DATA_TRANSACTION); if ((ret < 0) && - (afr_inode_split_brain_choice_get(local->inode, this, &spb_choice) == - 0) && - spb_choice < 0) { + (afr_split_brain_read_subvol_get(local->inode, this, NULL, + &spb_subvol) == 0) && + spb_subvol < 0) { afr_inode_refresh(frame, this, local->inode, local->inode->gfid, afr_open_continue); } else { diff --git a/xlators/cluster/afr/src/afr-read-txn.c b/xlators/cluster/afr/src/afr-read-txn.c index 772b59f..6fc2c75 100644 --- a/xlators/cluster/afr/src/afr-read-txn.c +++ b/xlators/cluster/afr/src/afr-read-txn.c @@ -164,7 +164,7 @@ afr_ta_read_txn(void *opaque) xdata_rsp = NULL; /* It doesn't. So query thin-arbiter to see if it blames any data brick. */ - ret = afr_fill_ta_loc(this, &loc); + ret = afr_fill_ta_loc(this, &loc, _gf_true); if (ret) { gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB, "Failed to populate thin-arbiter loc for: %s.", loc.name); @@ -272,7 +272,7 @@ afr_read_txn_refresh_done(call_frame_t *frame, xlator_t *this, int err) int read_subvol = -1; inode_t *inode = NULL; int ret = -1; - int spb_choice = -1; + int spb_subvol = -1; local = frame->local; inode = local->inode; @@ -303,9 +303,9 @@ afr_read_txn_refresh_done(call_frame_t *frame, xlator_t *this, int err) local->read_attempted[read_subvol] = 1; readfn: if (read_subvol == -1) { - ret = afr_inode_split_brain_choice_get(inode, this, &spb_choice); - if ((ret == 0) && spb_choice >= 0) - read_subvol = spb_choice; + ret = afr_split_brain_read_subvol_get(inode, this, frame, &spb_subvol); + if ((ret == 0) && spb_subvol >= 0) + read_subvol = spb_subvol; } if (read_subvol == -1) { diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c index 16f19e7..a580a15 100644 --- a/xlators/cluster/afr/src/afr-self-heal-common.c +++ b/xlators/cluster/afr/src/afr-self-heal-common.c @@ -140,7 +140,7 @@ heal: } } out: - if (gfid_idx && (*gfid_idx == -1) && (ret == 0)) { + if (gfid_idx && (*gfid_idx == -1) && (ret == 0) && local) { ret = -afr_final_errno(local, priv); } loc_wipe(&loc); @@ -1575,7 +1575,6 @@ afr_selfheal_find_direction(call_frame_t *frame, xlator_t *this, char *accused = NULL; /* Accused others without any self-accusal */ char *pending = NULL; /* Have pending operations on others */ char *self_accused = NULL; /* Accused itself */ - int min_participants = -1; priv = this->private; @@ -1599,12 +1598,7 @@ afr_selfheal_find_direction(call_frame_t *frame, xlator_t *this, } } - if (type == AFR_DATA_TRANSACTION || type == AFR_METADATA_TRANSACTION) { - min_participants = priv->child_count; - } else { - min_participants = AFR_SH_MIN_PARTICIPANTS; - } - if (afr_success_count(replies, priv->child_count) < min_participants) { + if (afr_success_count(replies, priv->child_count) < priv->child_count) { /* Treat this just like locks not being acquired */ return -ENOTCONN; } @@ -1911,17 +1905,16 @@ int afr_selfheal_unlocked_discover(call_frame_t *frame, inode_t *inode, uuid_t gfid, struct afr_reply *replies) { - afr_private_t *priv = NULL; afr_local_t *local = NULL; dict_t *dict = NULL; - priv = frame->this->private; local = frame->local; - if (local && local->xattr_req) + + if (local->xattr_req) dict = local->xattr_req; return afr_selfheal_unlocked_discover_on(frame, inode, gfid, replies, - priv->child_up, dict); + local->child_up, dict); } unsigned int @@ -2757,3 +2750,185 @@ afr_choose_source_by_policy(afr_private_t *priv, unsigned char *sources, out: return source; } + +static int +afr_anon_inode_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, inode_t *inode, + struct iatt *buf, struct iatt *preparent, + struct iatt *postparent, dict_t *xdata) +{ + afr_local_t *local = frame->local; + int i = (long)cookie; + + local->replies[i].valid = 1; + local->replies[i].op_ret = op_ret; + local->replies[i].op_errno = op_errno; + if (op_ret == 0) { + local->op_ret = 0; + local->replies[i].poststat = *buf; + local->replies[i].preparent = *preparent; + local->replies[i].postparent = *postparent; + } + if (xdata) { + local->replies[i].xdata = dict_ref(xdata); + } + + syncbarrier_wake(&local->barrier); + return 0; +} + +int +afr_anon_inode_create(xlator_t *this, int child, inode_t **linked_inode) +{ + call_frame_t *frame = NULL; + afr_local_t *local = NULL; + afr_private_t *priv = this->private; + unsigned char *mkdir_on = alloca0(priv->child_count); + unsigned char *lookup_on = alloca0(priv->child_count); + loc_t loc = {0}; + int32_t op_errno = 0; + int32_t child_op_errno = 0; + struct iatt iatt = {0}; + dict_t *xdata = NULL; + uuid_t anon_inode_gfid = {0}; + int mkdir_count = 0; + int i = 0; + + /*Try to mkdir everywhere and return success if the dir exists on 'child' + */ + + if (!priv->use_anon_inode) { + op_errno = EINVAL; + goto out; + } + + frame = afr_frame_create(this, &op_errno); + if (op_errno) { + goto out; + } + local = frame->local; + if (!local->child_up[child]) { + /*Other bricks may need mkdir so don't error out yet*/ + child_op_errno = ENOTCONN; + } + gf_uuid_parse(priv->anon_gfid_str, anon_inode_gfid); + for (i = 0; i < priv->child_count; i++) { + if (!local->child_up[i]) + continue; + + if (priv->anon_inode[i]) { + mkdir_on[i] = 0; + } else { + mkdir_on[i] = 1; + mkdir_count++; + } + } + + if (mkdir_count == 0) { + *linked_inode = inode_find(this->itable, anon_inode_gfid); + if (*linked_inode) { + op_errno = 0; + goto out; + } + } + + loc.parent = inode_ref(this->itable->root); + loc.name = priv->anon_inode_name; + loc.inode = inode_new(this->itable); + if (!loc.inode) { + op_errno = ENOMEM; + goto out; + } + + xdata = dict_new(); + if (!xdata) { + op_errno = ENOMEM; + goto out; + } + + op_errno = -dict_set_gfuuid(xdata, "gfid-req", anon_inode_gfid, _gf_true); + if (op_errno) { + goto out; + } + + if (mkdir_count == 0) { + memcpy(lookup_on, local->child_up, priv->child_count); + goto lookup; + } + + AFR_ONLIST(mkdir_on, frame, afr_anon_inode_mkdir_cbk, mkdir, &loc, 0755, 0, + xdata); + + for (i = 0; i < priv->child_count; i++) { + if (!mkdir_on[i]) { + continue; + } + + if (local->replies[i].op_ret == 0) { + priv->anon_inode[i] = 1; + iatt = local->replies[i].poststat; + } else if (local->replies[i].op_ret < 0 && + local->replies[i].op_errno == EEXIST) { + lookup_on[i] = 1; + } else if (i == child) { + child_op_errno = local->replies[i].op_errno; + } + } + + if (AFR_COUNT(lookup_on, priv->child_count) == 0) { + goto link; + } + +lookup: + AFR_ONLIST(lookup_on, frame, afr_selfheal_discover_cbk, lookup, &loc, + xdata); + for (i = 0; i < priv->child_count; i++) { + if (!lookup_on[i]) { + continue; + } + + if (local->replies[i].op_ret == 0) { + if (gf_uuid_compare(anon_inode_gfid, + local->replies[i].poststat.ia_gfid) == 0) { + priv->anon_inode[i] = 1; + iatt = local->replies[i].poststat; + } else { + if (i == child) + child_op_errno = EINVAL; + gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_DATA, + "%s has gfid: %s", priv->anon_inode_name, + uuid_utoa(local->replies[i].poststat.ia_gfid)); + } + } else if (i == child) { + child_op_errno = local->replies[i].op_errno; + } + } +link: + if (!gf_uuid_is_null(iatt.ia_gfid)) { + *linked_inode = inode_link(loc.inode, loc.parent, loc.name, &iatt); + if (*linked_inode) { + op_errno = 0; + inode_lookup(*linked_inode); + } else { + op_errno = ENOMEM; + } + goto out; + } + +out: + if (xdata) + dict_unref(xdata); + loc_wipe(&loc); + /*child_op_errno takes precedence*/ + if (child_op_errno == 0) { + child_op_errno = op_errno; + } + + if (child_op_errno && *linked_inode) { + inode_unref(*linked_inode); + *linked_inode = NULL; + } + if (frame) + AFR_STACK_DESTROY(frame); + return -child_op_errno; +} diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c index cdff4a5..37bcc2b 100644 --- a/xlators/cluster/afr/src/afr-self-heal-data.c +++ b/xlators/cluster/afr/src/afr-self-heal-data.c @@ -225,24 +225,40 @@ __afr_selfheal_data_read_write(call_frame_t *frame, xlator_t *this, fd_t *fd, return ret; } +static gf_boolean_t +afr_source_sinks_locked(xlator_t *this, unsigned char *locked_on, int source, + unsigned char *healed_sinks) +{ + afr_private_t *priv = this->private; + int i = 0; + + if (!locked_on[source]) + return _gf_false; + + for (i = 0; i < priv->child_count; i++) { + if (healed_sinks[i] && locked_on[i]) + return _gf_true; + } + + return _gf_false; +} + static int afr_selfheal_data_block(call_frame_t *frame, xlator_t *this, fd_t *fd, int source, unsigned char *healed_sinks, off_t offset, size_t size, int type, struct afr_reply *replies) { int ret = -1; - int sink_count = 0; afr_private_t *priv = NULL; unsigned char *data_lock = NULL; priv = this->private; - sink_count = AFR_COUNT(healed_sinks, priv->child_count); data_lock = alloca0(priv->child_count); ret = afr_selfheal_inodelk(frame, this, fd->inode, this->name, offset, size, data_lock); { - if (ret < sink_count) { + if (!afr_source_sinks_locked(this, data_lock, source, healed_sinks)) { ret = -ENOTCONN; goto unlock; } diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c index e706228..64893f4 100644 --- a/xlators/cluster/afr/src/afr-self-heal-entry.c +++ b/xlators/cluster/afr/src/afr-self-heal-entry.c @@ -16,54 +16,170 @@ #include <glusterfs/syncop-utils.h> #include <glusterfs/events.h> -static int -afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name, - inode_t *inode, int child, struct afr_reply *replies) +int +afr_selfheal_entry_anon_inode(xlator_t *this, inode_t *dir, const char *name, + inode_t *inode, int child, + struct afr_reply *replies, + gf_boolean_t *anon_inode) { afr_private_t *priv = NULL; + afr_local_t *local = NULL; xlator_t *subvol = NULL; int ret = 0; + int i = 0; + char g[64] = {0}; + unsigned char *lookup_success = NULL; + call_frame_t *frame = NULL; + loc_t loc2 = { + 0, + }; loc_t loc = { 0, }; - char g[64]; priv = this->private; - subvol = priv->children[child]; + lookup_success = alloca0(priv->child_count); + uuid_utoa_r(replies[child].poststat.ia_gfid, g); + loc.inode = inode_new(inode->table); + if (!loc.inode) { + ret = -ENOMEM; + goto out; + } + + if (replies[child].poststat.ia_type == IA_IFDIR) { + /* This directory may have sub-directory hierarchy which may need to + * be preserved for subsequent heals. So unconditionally move the + * directory to anonymous-inode directory*/ + *anon_inode = _gf_true; + goto anon_inode; + } + + frame = afr_frame_create(this, &ret); + if (!frame) { + ret = -ret; + goto out; + } + local = frame->local; + gf_uuid_copy(loc.gfid, replies[child].poststat.ia_gfid); + AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, &loc, + NULL); + for (i = 0; i < priv->child_count; i++) { + if (local->replies[i].op_ret == 0) { + lookup_success[i] = 1; + } else if (local->replies[i].op_errno != ENOENT && + local->replies[i].op_errno != ESTALE) { + ret = -local->replies[i].op_errno; + } + } + + if (priv->quorum_count) { + if (afr_has_quorum(lookup_success, this, NULL)) { + *anon_inode = _gf_true; + } + } else if (AFR_COUNT(lookup_success, priv->child_count) > 1) { + *anon_inode = _gf_true; + } else if (ret) { + goto out; + } + +anon_inode: + if (!*anon_inode) { + ret = 0; + goto out; + } loc.parent = inode_ref(dir); gf_uuid_copy(loc.pargfid, dir->gfid); loc.name = name; - loc.inode = inode_ref(inode); - if (replies[child].valid && replies[child].op_ret == 0) { - switch (replies[child].poststat.ia_type) { - case IA_IFDIR: - gf_msg(this->name, GF_LOG_WARNING, 0, - AFR_MSG_EXPUNGING_FILE_OR_DIR, - "expunging dir %s/%s (%s) on %s", uuid_utoa(dir->gfid), - name, uuid_utoa_r(replies[child].poststat.ia_gfid, g), - subvol->name); - ret = syncop_rmdir(subvol, &loc, 1, NULL, NULL); - break; - default: - gf_msg(this->name, GF_LOG_WARNING, 0, - AFR_MSG_EXPUNGING_FILE_OR_DIR, - "expunging file %s/%s (%s) on %s", uuid_utoa(dir->gfid), - name, uuid_utoa_r(replies[child].poststat.ia_gfid, g), - subvol->name); - ret = syncop_unlink(subvol, &loc, NULL, NULL); - break; - } + ret = afr_anon_inode_create(this, child, &loc2.parent); + if (ret < 0) + goto out; + + loc2.name = g; + ret = syncop_rename(subvol, &loc, &loc2, NULL, NULL); + if (ret < 0) { + gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_EXPUNGING_FILE_OR_DIR, + "Rename to %s dir %s/%s (%s) on %s failed", + priv->anon_inode_name, uuid_utoa(dir->gfid), name, g, + subvol->name); + } else { + gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR, + "Rename to %s dir %s/%s (%s) on %s successful", + priv->anon_inode_name, uuid_utoa(dir->gfid), name, g, + subvol->name); } +out: loc_wipe(&loc); + loc_wipe(&loc2); + if (frame) { + AFR_STACK_DESTROY(frame); + } return ret; } int +afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name, + inode_t *inode, int child, struct afr_reply *replies) +{ + char g[64] = {0}; + afr_private_t *priv = NULL; + xlator_t *subvol = NULL; + int ret = 0; + loc_t loc = { + 0, + }; + gf_boolean_t anon_inode = _gf_false; + + priv = this->private; + subvol = priv->children[child]; + + if ((!replies[child].valid) || (replies[child].op_ret < 0)) { + /*Nothing to do*/ + ret = 0; + goto out; + } + + if (priv->use_anon_inode) { + ret = afr_selfheal_entry_anon_inode(this, dir, name, inode, child, + replies, &anon_inode); + if (ret < 0 || anon_inode) + goto out; + } + + loc.parent = inode_ref(dir); + loc.inode = inode_new(inode->table); + if (!loc.inode) { + ret = -ENOMEM; + goto out; + } + loc.name = name; + switch (replies[child].poststat.ia_type) { + case IA_IFDIR: + gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR, + "expunging dir %s/%s (%s) on %s", uuid_utoa(dir->gfid), name, + uuid_utoa_r(replies[child].poststat.ia_gfid, g), + subvol->name); + ret = syncop_rmdir(subvol, &loc, 1, NULL, NULL); + break; + default: + gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR, + "expunging file %s/%s (%s) on %s", uuid_utoa(dir->gfid), + name, uuid_utoa_r(replies[child].poststat.ia_gfid, g), + subvol->name); + ret = syncop_unlink(subvol, &loc, NULL, NULL); + break; + } + +out: + loc_wipe(&loc); + return ret; +} + +int afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source, unsigned char *sources, inode_t *dir, const char *name, inode_t *inode, @@ -76,6 +192,9 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source, loc_t srcloc = { 0, }; + loc_t anonloc = { + 0, + }; xlator_t *this = frame->this; afr_private_t *priv = NULL; dict_t *xdata = NULL; @@ -86,15 +205,17 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source, 0, }; unsigned char *newentry = NULL; - char dir_uuid_str[64] = {0}, iatt_uuid_str[64] = {0}; + char iatt_uuid_str[64] = {0}; + char dir_uuid_str[64] = {0}; priv = this->private; iatt = &replies[source].poststat; + uuid_utoa_r(iatt->ia_gfid, iatt_uuid_str); if (iatt->ia_type == IA_INVAL || gf_uuid_is_null(iatt->ia_gfid)) { gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SELF_HEAL_FAILED, "Invalid ia_type (%d) or gfid(%s). source brick=%d, " "pargfid=%s, name=%s", - iatt->ia_type, uuid_utoa_r(iatt->ia_gfid, iatt_uuid_str), source, + iatt->ia_type, iatt_uuid_str, source, uuid_utoa_r(dir->gfid, dir_uuid_str), name); ret = -EINVAL; goto out; @@ -120,14 +241,24 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source, srcloc.inode = inode_ref(inode); gf_uuid_copy(srcloc.gfid, iatt->ia_gfid); - if (iatt->ia_type != IA_IFDIR) - ret = syncop_lookup(priv->children[dst], &srcloc, 0, 0, 0, 0); - if (iatt->ia_type == IA_IFDIR || ret == -ENOENT || ret == -ESTALE) { + ret = syncop_lookup(priv->children[dst], &srcloc, 0, 0, 0, 0); + if (ret == -ENOENT || ret == -ESTALE) { newentry[dst] = 1; ret = afr_selfheal_newentry_mark(frame, this, inode, source, replies, sources, newentry); if (ret) goto out; + } else if (ret == 0 && iatt->ia_type == IA_IFDIR && priv->use_anon_inode) { + // Try rename from hidden directory + ret = afr_anon_inode_create(this, dst, &anonloc.parent); + if (ret < 0) + goto out; + anonloc.inode = inode_ref(inode); + anonloc.name = iatt_uuid_str; + ret = syncop_rename(priv->children[dst], &anonloc, &loc, NULL, NULL); + if (ret == -ENOENT || ret == -ESTALE) + ret = -1; /*This sets 'mismatch' to true*/ + goto out; } mode = st_mode_from_ia(iatt->ia_prot, iatt->ia_type); @@ -166,6 +297,7 @@ out: GF_FREE(linkname); loc_wipe(&loc); loc_wipe(&srcloc); + loc_wipe(&anonloc); return ret; } @@ -578,6 +710,11 @@ afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd, priv = this->private; + if (afr_is_private_directory(priv, fd->inode->gfid, name, + GF_CLIENT_PID_SELF_HEALD)) { + return 0; + } + xattr = dict_new(); if (!xattr) return -ENOMEM; @@ -598,7 +735,7 @@ afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd, ret = afr_selfheal_entrylk(frame, this, fd->inode, this->name, NULL, locked_on); { - if (ret < AFR_SH_MIN_PARTICIPANTS) { + if (ret < priv->child_count) { gf_msg_debug(this->name, 0, "%s: Skipping " "entry self-heal as only %d sub-volumes " @@ -626,7 +763,7 @@ afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd, replies); if ((ret == 0) && (priv->esh_granular) && parent_idx_inode) { - ret = afr_shd_index_purge(subvol, parent_idx_inode, name, + ret = afr_shd_entry_purge(subvol, parent_idx_inode, name, inode->ia_type); /* Why is ret force-set to 0? We do not care about * index purge failing for full heal as it is quite @@ -756,10 +893,6 @@ afr_selfheal_entry_do_subvol(call_frame_t *frame, xlator_t *this, fd_t *fd, if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, "..")) continue; - if (__is_root_gfid(fd->inode->gfid) && - !strcmp(entry->d_name, GF_REPLICATE_TRASH_DIR)) - continue; - ret = afr_selfheal_entry_dirent(iter_frame, this, fd, entry->d_name, loc.inode, subvol, local->need_full_crawl); @@ -822,7 +955,7 @@ afr_selfheal_entry_granular_dirent(xlator_t *subvol, gf_dirent_t *entry, /* The name indices under the pgfid index dir are guaranteed * to be regular files. Hence the hardcoding. */ - afr_shd_index_purge(subvol, parent->inode, entry->d_name, IA_IFREG); + afr_shd_entry_purge(subvol, parent->inode, entry->d_name, IA_IFREG); ret = 0; goto out; } @@ -992,7 +1125,7 @@ __afr_selfheal_entry(call_frame_t *frame, xlator_t *this, fd_t *fd, ret = afr_selfheal_entrylk(frame, this, fd->inode, this->name, NULL, data_lock); { - if (ret < AFR_SH_MIN_PARTICIPANTS) { + if (ret < priv->child_count) { gf_msg_debug(this->name, 0, "%s: Skipping " "entry self-heal as only %d sub-volumes could " @@ -1116,7 +1249,7 @@ afr_selfheal_entry(call_frame_t *frame, xlator_t *this, inode_t *inode) ret = afr_selfheal_tie_breaker_entrylk(frame, this, inode, priv->sh_domain, NULL, locked_on); { - if (ret < AFR_SH_MIN_PARTICIPANTS) { + if (ret < priv->child_count) { gf_msg_debug(this->name, 0, "%s: Skipping " "entry self-heal as only %d sub-volumes could " diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c index f4e31b6..03f43ba 100644 --- a/xlators/cluster/afr/src/afr-self-heal-metadata.c +++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c @@ -190,6 +190,59 @@ out: return ret; } +static int +__afr_selfheal_metadata_mark_pending_xattrs(call_frame_t *frame, xlator_t *this, + inode_t *inode, + struct afr_reply *replies, + unsigned char *sources) +{ + int ret = 0; + int i = 0; + int m_idx = 0; + afr_private_t *priv = NULL; + int raw[AFR_NUM_CHANGE_LOGS] = {0}; + dict_t *xattr = NULL; + + priv = this->private; + m_idx = afr_index_for_transaction_type(AFR_METADATA_TRANSACTION); + raw[m_idx] = 1; + + xattr = dict_new(); + if (!xattr) + return -ENOMEM; + + for (i = 0; i < priv->child_count; i++) { + if (sources[i]) + continue; + ret = dict_set_static_bin(xattr, priv->pending_key[i], raw, + sizeof(int) * AFR_NUM_CHANGE_LOGS); + if (ret) { + ret = -1; + goto out; + } + } + + for (i = 0; i < priv->child_count; i++) { + if (!sources[i]) + continue; + ret = afr_selfheal_post_op(frame, this, inode, i, xattr, NULL); + if (ret < 0) { + gf_msg(this->name, GF_LOG_INFO, -ret, AFR_MSG_SELF_HEAL_INFO, + "Failed to set pending metadata xattr on child %d for %s", i, + uuid_utoa(inode->gfid)); + goto out; + } + } + + afr_replies_wipe(replies, priv->child_count); + ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies); + +out: + if (xattr) + dict_unref(xattr); + return ret; +} + /* * Look for mismatching uid/gid or mode or user xattrs even if * AFR xattrs don't say so, and pick one arbitrarily as winner. */ @@ -210,6 +263,7 @@ __afr_selfheal_metadata_finalize_source(call_frame_t *frame, xlator_t *this, }; int source = -1; int sources_count = 0; + int ret = 0; priv = this->private; @@ -300,7 +354,13 @@ __afr_selfheal_metadata_finalize_source(call_frame_t *frame, xlator_t *this, healed_sinks[i] = 1; } } - + if ((sources_count == priv->child_count) && (source > -1) && + (AFR_COUNT(healed_sinks, priv->child_count) != 0)) { + ret = __afr_selfheal_metadata_mark_pending_xattrs(frame, this, inode, + replies, sources); + if (ret < 0) + return ret; + } out: afr_mark_active_sinks(this, sources, locked_on, healed_sinks); return source; diff --git a/xlators/cluster/afr/src/afr-self-heal-name.c b/xlators/cluster/afr/src/afr-self-heal-name.c index 36640b54..834aac8 100644 --- a/xlators/cluster/afr/src/afr-self-heal-name.c +++ b/xlators/cluster/afr/src/afr-self-heal-name.c @@ -98,21 +98,12 @@ __afr_selfheal_name_expunge(xlator_t *this, inode_t *parent, uuid_t pargfid, const char *bname, inode_t *inode, struct afr_reply *replies) { - loc_t loc = { - 0, - }; int i = 0; afr_private_t *priv = NULL; - char g[64]; int ret = 0; priv = this->private; - loc.parent = inode_ref(parent); - gf_uuid_copy(loc.pargfid, pargfid); - loc.name = bname; - loc.inode = inode_ref(inode); - for (i = 0; i < priv->child_count; i++) { if (!replies[i].valid) continue; @@ -120,30 +111,10 @@ __afr_selfheal_name_expunge(xlator_t *this, inode_t *parent, uuid_t pargfid, if (replies[i].op_ret) continue; - switch (replies[i].poststat.ia_type) { - case IA_IFDIR: - gf_msg(this->name, GF_LOG_WARNING, 0, - AFR_MSG_EXPUNGING_FILE_OR_DIR, - "expunging dir %s/%s (%s) on %s", uuid_utoa(pargfid), - bname, uuid_utoa_r(replies[i].poststat.ia_gfid, g), - priv->children[i]->name); - - ret |= syncop_rmdir(priv->children[i], &loc, 1, NULL, NULL); - break; - default: - gf_msg(this->name, GF_LOG_WARNING, 0, - AFR_MSG_EXPUNGING_FILE_OR_DIR, - "expunging file %s/%s (%s) on %s", uuid_utoa(pargfid), - bname, uuid_utoa_r(replies[i].poststat.ia_gfid, g), - priv->children[i]->name); - - ret |= syncop_unlink(priv->children[i], &loc, NULL, NULL); - break; - } + ret |= afr_selfheal_entry_delete(this, parent, bname, inode, i, + replies); } - loc_wipe(&loc); - return ret; } @@ -381,7 +352,7 @@ __afr_selfheal_name_do(call_frame_t *frame, xlator_t *this, inode_t *parent, ret = __afr_selfheal_assign_gfid(this, parent, pargfid, bname, inode, replies, gfid, locked_on, source, sources, is_gfid_absent, &gfid_idx); - if (ret) + if (ret || (gfid_idx < 0)) return ret; ret = __afr_selfheal_name_impunge(frame, this, parent, pargfid, bname, @@ -514,7 +485,7 @@ afr_selfheal_name_do(call_frame_t *frame, xlator_t *this, inode_t *parent, ret = afr_selfheal_entrylk(frame, this, parent, this->name, bname, locked_on); { - if (ret < AFR_SH_MIN_PARTICIPANTS) { + if (ret < priv->child_count) { ret = -ENOTCONN; goto unlock; } @@ -560,13 +531,15 @@ afr_selfheal_name_unlocked_inspect(call_frame_t *frame, xlator_t *this, struct afr_reply *replies = NULL; inode_t *inode = NULL; int first_idx = -1; + afr_local_t *local = NULL; priv = this->private; + local = frame->local; replies = alloca0(sizeof(*replies) * priv->child_count); inode = afr_selfheal_unlocked_lookup_on(frame, parent, bname, replies, - priv->child_up, NULL); + local->child_up, NULL); if (!inode) return -ENOMEM; diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h index 5e7bde8..48e6dbc 100644 --- a/xlators/cluster/afr/src/afr-self-heal.h +++ b/xlators/cluster/afr/src/afr-self-heal.h @@ -11,8 +11,6 @@ #ifndef _AFR_SELFHEAL_H #define _AFR_SELFHEAL_H -#define AFR_SH_MIN_PARTICIPANTS 2 - /* Perform fop on all UP subvolumes and wait for all callbacks to return */ #define AFR_ONALL(frame, rfn, fop, args...) \ @@ -47,13 +45,16 @@ afr_local_t *__local = frame->local; \ afr_private_t *__priv = frame->this->private; \ int __i = 0; \ - int __count = AFR_COUNT(list, __priv->child_count); \ + int __count = 0; \ + unsigned char *__list = alloca(__priv->child_count); \ \ + memcpy(__list, list, sizeof(*__list) * __priv->child_count); \ + __count = AFR_COUNT(__list, __priv->child_count); \ __local->barrier.waitfor = __count; \ afr_local_replies_wipe(__local, __priv); \ \ for (__i = 0; __i < __priv->child_count; __i++) { \ - if (!list[__i]) \ + if (!__list[__i]) \ continue; \ STACK_WIND_COOKIE(frame, rfn, (void *)(long)__i, \ __priv->children[__i], \ @@ -368,4 +369,9 @@ gf_boolean_t afr_is_file_empty_on_all_children(afr_private_t *priv, struct afr_reply *replies); +int +afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name, + inode_t *inode, int child, struct afr_reply *replies); +int +afr_anon_inode_create(xlator_t *this, int child, inode_t **linked_inode); #endif /* !_AFR_SELFHEAL_H */ diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c index aa774bb..109fd4b 100644 --- a/xlators/cluster/afr/src/afr-self-heald.c +++ b/xlators/cluster/afr/src/afr-self-heald.c @@ -94,7 +94,7 @@ __afr_shd_healer_wait(struct subvol_healer *healer) priv = healer->this->private; disabled_loop: - wait_till.tv_sec = time(NULL) + priv->shd.timeout; + wait_till.tv_sec = gf_time() + priv->shd.timeout; while (!healer->rerun) { ret = pthread_cond_timedwait(&healer->cond, &healer->mutex, &wait_till); @@ -222,7 +222,7 @@ out: } int -afr_shd_index_purge(xlator_t *subvol, inode_t *inode, char *name, +afr_shd_entry_purge(xlator_t *subvol, inode_t *inode, char *name, ia_type_t type) { int ret = 0; @@ -371,7 +371,7 @@ afr_shd_sweep_prepare(struct subvol_healer *healer) event->split_brain_count = 0; event->heal_failed_count = 0; - time(&event->start_time); + event->start_time = gf_time(); event->end_time = 0; _mask_cancellation(); } @@ -386,7 +386,7 @@ afr_shd_sweep_done(struct subvol_healer *healer) event = &healer->crawl_event; shd = &(((afr_private_t *)healer->this->private)->shd); - time(&event->end_time); + event->end_time = gf_time(); history = gf_memdup(event, sizeof(*event)); event->start_time = 0; @@ -424,7 +424,7 @@ afr_shd_index_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent, ret = afr_shd_selfheal(healer, healer->subvol, gfid); if (ret == -ENOENT || ret == -ESTALE) - afr_shd_index_purge(subvol, parent->inode, entry->d_name, val); + afr_shd_entry_purge(subvol, parent->inode, entry->d_name, val); if (ret == 2) /* If bricks crashed in pre-op after creating indices/xattrop @@ -843,6 +843,176 @@ out: return need_heal; } +static int +afr_shd_anon_inode_cleaner(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent, + void *data) +{ + struct subvol_healer *healer = data; + afr_private_t *priv = healer->this->private; + call_frame_t *frame = NULL; + afr_local_t *local = NULL; + int ret = 0; + loc_t loc = {0}; + int count = 0; + int i = 0; + int op_errno = 0; + struct iatt *iatt = NULL; + gf_boolean_t multiple_links = _gf_false; + unsigned char *gfid_present = alloca0(priv->child_count); + unsigned char *entry_present = alloca0(priv->child_count); + char *type = "file"; + + frame = afr_frame_create(healer->this, &ret); + if (!frame) { + ret = -ret; + goto out; + } + local = frame->local; + if (AFR_COUNT(local->child_up, priv->child_count) != priv->child_count) { + gf_msg_debug(healer->this->name, 0, + "Not all bricks are up. Skipping " + "cleanup of %s on %s", + entry->d_name, subvol->name); + ret = 0; + goto out; + } + + loc.inode = inode_new(parent->inode->table); + if (!loc.inode) { + ret = -ENOMEM; + goto out; + } + ret = gf_uuid_parse(entry->d_name, loc.gfid); + if (ret) { + ret = 0; + goto out; + } + AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, &loc, + NULL); + for (i = 0; i < priv->child_count; i++) { + if (local->replies[i].op_ret == 0) { + count++; + gfid_present[i] = 1; + iatt = &local->replies[i].poststat; + if (iatt->ia_type == IA_IFDIR) { + type = "dir"; + } + + if (i == healer->subvol) { + if (local->replies[i].poststat.ia_nlink > 1) { + multiple_links = _gf_true; + } + } + } else if (local->replies[i].op_errno != ENOENT && + local->replies[i].op_errno != ESTALE) { + /*We don't have complete view. Skip the entry*/ + gf_msg_debug(healer->this->name, local->replies[i].op_errno, + "Skipping cleanup of %s on %s", entry->d_name, + subvol->name); + ret = 0; + goto out; + } + } + + /*Inode is deleted from subvol*/ + if (count == 1 || (iatt->ia_type != IA_IFDIR && multiple_links)) { + gf_msg(healer->this->name, GF_LOG_WARNING, 0, + AFR_MSG_EXPUNGING_FILE_OR_DIR, "expunging %s %s/%s on %s", type, + priv->anon_inode_name, entry->d_name, subvol->name); + ret = afr_shd_entry_purge(subvol, parent->inode, entry->d_name, + iatt->ia_type); + if (ret == -ENOENT || ret == -ESTALE) + ret = 0; + } else if (count > 1) { + loc_wipe(&loc); + loc.parent = inode_ref(parent->inode); + loc.name = entry->d_name; + loc.inode = inode_new(parent->inode->table); + if (!loc.inode) { + ret = -ENOMEM; + goto out; + } + AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, + &loc, NULL); + count = 0; + for (i = 0; i < priv->child_count; i++) { + if (local->replies[i].op_ret == 0) { + count++; + entry_present[i] = 1; + iatt = &local->replies[i].poststat; + } else if (local->replies[i].op_errno != ENOENT && + local->replies[i].op_errno != ESTALE) { + /*We don't have complete view. Skip the entry*/ + gf_msg_debug(healer->this->name, local->replies[i].op_errno, + "Skipping cleanup of %s on %s", entry->d_name, + subvol->name); + ret = 0; + goto out; + } + } + for (i = 0; i < priv->child_count; i++) { + if (gfid_present[i] && !entry_present[i]) { + /*Entry is not anonymous on at least one subvol*/ + gf_msg_debug(healer->this->name, 0, + "Valid entry present on %s " + "Skipping cleanup of %s on %s", + priv->children[i]->name, entry->d_name, + subvol->name); + ret = 0; + goto out; + } + } + + gf_msg(healer->this->name, GF_LOG_WARNING, 0, + AFR_MSG_EXPUNGING_FILE_OR_DIR, + "expunging %s %s/%s on all subvols", type, priv->anon_inode_name, + entry->d_name); + ret = 0; + for (i = 0; i < priv->child_count; i++) { + op_errno = -afr_shd_entry_purge(priv->children[i], loc.parent, + entry->d_name, iatt->ia_type); + if (op_errno != ENOENT && op_errno != ESTALE) { + ret |= -op_errno; + } + } + } + +out: + if (frame) + AFR_STACK_DESTROY(frame); + loc_wipe(&loc); + return ret; +} + +static void +afr_cleanup_anon_inode_dir(struct subvol_healer *healer) +{ + int ret = 0; + call_frame_t *frame = NULL; + afr_private_t *priv = healer->this->private; + loc_t loc = {0}; + + ret = afr_anon_inode_create(healer->this, healer->subvol, &loc.inode); + if (ret) + goto out; + + frame = afr_frame_create(healer->this, &ret); + if (!frame) { + ret = -ret; + goto out; + } + + ret = syncop_mt_dir_scan(frame, priv->children[healer->subvol], &loc, + GF_CLIENT_PID_SELF_HEALD, healer, + afr_shd_anon_inode_cleaner, NULL, + priv->shd.max_threads, priv->shd.wait_qlength); +out: + if (frame) + AFR_STACK_DESTROY(frame); + loc_wipe(&loc); + return; +} + void * afr_shd_index_healer(void *data) { @@ -900,6 +1070,10 @@ afr_shd_index_healer(void *data) sleep(1); } while (ret > 0); + if (ret == 0) { + afr_cleanup_anon_inode_dir(healer); + } + if (ret == 0 && pre_crawl_xdata && !healer->crawl_event.heal_failed_count) { afr_shd_ta_check_and_unset_xattrs(this, &loc, healer, @@ -1481,15 +1655,6 @@ afr_xl_op(xlator_t *this, dict_t *input, dict_t *output) case GF_SHD_OP_INDEX_SUMMARY: /* this case has been handled in glfs-heal.c */ break; - case GF_SHD_OP_HEALED_FILES: - case GF_SHD_OP_HEAL_FAILED_FILES: - for (i = 0; i < priv->child_count; i++) { - keylen = snprintf(key, sizeof(key), "%d-%d-status", xl_id, i); - AFR_SET_DICT_AND_LOG(this->name, output, key, keylen, - SOP_NOT_SUPPORTED, - SLEN(SOP_NOT_SUPPORTED)); - } - break; case GF_SHD_OP_SPLIT_BRAIN_FILES: eh_dump(shd->split_brain, output, afr_add_shd_event); break; diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h index 687c28e..18db728 100644 --- a/xlators/cluster/afr/src/afr-self-heald.h +++ b/xlators/cluster/afr/src/afr-self-heald.h @@ -70,6 +70,6 @@ afr_shd_gfid_to_path(xlator_t *this, xlator_t *subvol, uuid_t gfid, char **path_p); int -afr_shd_index_purge(xlator_t *subvol, inode_t *inode, char *name, +afr_shd_entry_purge(xlator_t *subvol, inode_t *inode, char *name, ia_type_t type); #endif /* !_AFR_SELF_HEALD_H */ diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c index 78438f9..a51f79b 100644 --- a/xlators/cluster/afr/src/afr-transaction.c +++ b/xlators/cluster/afr/src/afr-transaction.c @@ -124,9 +124,9 @@ afr_release_notify_lock_for_ta(void *opaque) this = (xlator_t *)opaque; priv = this->private; - ret = afr_fill_ta_loc(this, &loc); + ret = afr_fill_ta_loc(this, &loc, _gf_true); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB, + gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB, "Failed to populate loc for thin-arbiter."); goto out; } @@ -521,42 +521,6 @@ afr_compute_pre_op_sources(call_frame_t *frame, xlator_t *this) local->transaction.pre_op_sources[j] = 0; } -gf_boolean_t -afr_has_arbiter_fop_cbk_quorum(call_frame_t *frame) -{ - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - xlator_t *this = NULL; - gf_boolean_t fop_failed = _gf_false; - unsigned char *pre_op_sources = NULL; - int i = 0; - - local = frame->local; - this = frame->this; - priv = this->private; - pre_op_sources = local->transaction.pre_op_sources; - - /* If the fop failed on the brick, it is not a source. */ - for (i = 0; i < priv->child_count; i++) - if (local->transaction.failed_subvols[i]) - pre_op_sources[i] = 0; - - switch (AFR_COUNT(pre_op_sources, priv->child_count)) { - case 1: - if (pre_op_sources[ARBITER_BRICK_INDEX]) - fop_failed = _gf_true; - break; - case 0: - fop_failed = _gf_true; - break; - } - - if (fop_failed) - return _gf_false; - - return _gf_true; -} - void afr_txn_arbitrate_fop(call_frame_t *frame, xlator_t *this) { @@ -971,12 +935,8 @@ afr_need_dirty_marking(call_frame_t *frame, xlator_t *this) priv->child_count) return _gf_false; - if (priv->arbiter_count) { - if (!afr_has_arbiter_fop_cbk_quorum(frame)) - need_dirty = _gf_true; - } else if (!afr_has_fop_cbk_quorum(frame)) { + if (!afr_has_fop_cbk_quorum(frame)) need_dirty = _gf_true; - } return need_dirty; } @@ -1026,12 +986,8 @@ afr_handle_quorum(call_frame_t *frame, xlator_t *this) * no split-brain with the fix. The problem is eliminated completely. */ - if (priv->arbiter_count) { - if (afr_has_arbiter_fop_cbk_quorum(frame)) - return; - } else if (afr_has_fop_cbk_quorum(frame)) { + if (afr_has_fop_cbk_quorum(frame)) return; - } if (afr_need_dirty_marking(frame, this)) goto set_response; @@ -1073,7 +1029,7 @@ set_response: } int -afr_fill_ta_loc(xlator_t *this, loc_t *loc) +afr_fill_ta_loc(xlator_t *this, loc_t *loc, gf_boolean_t is_gfid_based_fop) { afr_private_t *priv = NULL; @@ -1081,6 +1037,11 @@ afr_fill_ta_loc(xlator_t *this, loc_t *loc) loc->parent = inode_ref(priv->root_inode); gf_uuid_copy(loc->pargfid, loc->parent->gfid); loc->name = priv->pending_key[THIN_ARBITER_BRICK_INDEX]; + if (is_gfid_based_fop && gf_uuid_is_null(priv->ta_gfid)) { + /* Except afr_ta_id_file_check() which is path based, all other gluster + * FOPS need gfid.*/ + return -EINVAL; + } gf_uuid_copy(loc->gfid, priv->ta_gfid); loc->inode = inode_new(loc->parent->table); if (!loc->inode) { @@ -1090,86 +1051,6 @@ afr_fill_ta_loc(xlator_t *this, loc_t *loc) return 0; } -int -afr_changelog_thin_arbiter_post_op(xlator_t *this, afr_local_t *local) -{ - int ret = 0; - afr_private_t *priv = NULL; - dict_t *xattr = NULL; - int failed_count = 0; - struct gf_flock flock = { - 0, - }; - loc_t loc = { - 0, - }; - int i = 0; - - priv = this->private; - if (!priv->thin_arbiter_count) - return 0; - - failed_count = AFR_COUNT(local->transaction.failed_subvols, - priv->child_count); - if (!failed_count) - return 0; - - GF_ASSERT(failed_count == 1); - ret = afr_fill_ta_loc(this, &loc); - if (ret) { - gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB, - "Failed to populate thin-arbiter loc for: %s.", loc.name); - goto out; - } - - xattr = dict_new(); - if (!xattr) { - ret = -ENOMEM; - goto out; - } - for (i = 0; i < priv->child_count; i++) { - ret = dict_set_static_bin(xattr, priv->pending_key[i], - local->pending[i], - AFR_NUM_CHANGE_LOGS * sizeof(int)); - if (ret) - goto out; - } - - flock.l_type = F_WRLCK; - flock.l_start = 0; - flock.l_len = 0; - - /*TODO: Convert to two domain locking. */ - ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX], - AFR_TA_DOM_NOTIFY, &loc, F_SETLKW, &flock, NULL, NULL); - if (ret) - goto out; - - ret = syncop_xattrop(priv->children[THIN_ARBITER_BRICK_INDEX], &loc, - GF_XATTROP_ADD_ARRAY, xattr, NULL, NULL, NULL); - - if (ret == -EINVAL) { - gf_msg(this->name, GF_LOG_INFO, -ret, AFR_MSG_THIN_ARB, - "Thin-arbiter has denied post-op on %s for gfid %s.", - priv->pending_key[THIN_ARBITER_BRICK_INDEX], - uuid_utoa(local->inode->gfid)); - - } else if (ret) { - gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB, - "Post-op on thin-arbiter id file %s failed for gfid %s.", - priv->pending_key[THIN_ARBITER_BRICK_INDEX], - uuid_utoa(local->inode->gfid)); - } - flock.l_type = F_UNLCK; - syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX], AFR_TA_DOM_NOTIFY, - &loc, F_SETLK, &flock, NULL, NULL); -out: - if (xattr) - dict_unref(xattr); - - return ret; -} - static int afr_ta_post_op_done(int ret, call_frame_t *frame, void *opaque) { @@ -1264,9 +1145,9 @@ afr_ta_post_op_do(void *opaque) this = local->transaction.frame->this; priv = this->private; - ret = afr_fill_ta_loc(this, &loc); + ret = afr_fill_ta_loc(this, &loc, _gf_true); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB, + gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB, "Failed to populate loc for thin-arbiter."); goto out; } @@ -2466,8 +2347,13 @@ afr_is_delayed_changelog_post_op_needed(call_frame_t *frame, xlator_t *this, goto out; } - if ((local->op != GF_FOP_WRITE) && (local->op != GF_FOP_FXATTROP)) { - /*Only allow writes but shard does [f]xattrops on writes, so + if (local->transaction.disable_delayed_post_op) { + goto out; + } + + if ((local->op != GF_FOP_WRITE) && (local->op != GF_FOP_FXATTROP) && + (local->op != GF_FOP_FSYNC)) { + /*Only allow writes/fsyncs but shard does [f]xattrops on writes, so * they are fine too*/ goto out; } diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c index ec7aa22..df7366f 100644 --- a/xlators/cluster/afr/src/afr.c +++ b/xlators/cluster/afr/src/afr.c @@ -135,6 +135,27 @@ set_data_self_heal_algorithm(afr_private_t *priv, char *algo) } } +void +afr_handle_anon_inode_options(afr_private_t *priv, dict_t *options) +{ + char *volfile_id_str = NULL; + uuid_t anon_inode_gfid = {0}; + + /*If volume id is not present don't enable anything*/ + if (dict_get_str(options, "volume-id", &volfile_id_str)) + return; + GF_ASSERT(strlen(AFR_ANON_DIR_PREFIX) + strlen(volfile_id_str) <= NAME_MAX); + /*anon_inode_name is not supposed to change once assigned*/ + if (!priv->anon_inode_name[0]) { + snprintf(priv->anon_inode_name, sizeof(priv->anon_inode_name), "%s-%s", + AFR_ANON_DIR_PREFIX, volfile_id_str); + gf_uuid_parse(volfile_id_str, anon_inode_gfid); + /*Flip a bit to make sure volfile-id and anon-gfid are not same*/ + anon_inode_gfid[0] ^= 1; + uuid_utoa_r(anon_inode_gfid, priv->anon_gfid_str); + } +} + int reconfigure(xlator_t *this, dict_t *options) { @@ -168,7 +189,8 @@ reconfigure(xlator_t *this, dict_t *options) bool, out); GF_OPTION_RECONF("data-self-heal", data_self_heal, options, str, out); - gf_string2boolean(data_self_heal, &priv->data_self_heal); + if (gf_string2boolean(data_self_heal, &priv->data_self_heal) == -1) + goto out; GF_OPTION_RECONF("entry-self-heal", priv->entry_self_heal, options, bool, out); @@ -289,6 +311,10 @@ reconfigure(xlator_t *this, dict_t *options) consistent_io = _gf_false; priv->consistent_io = consistent_io; + afr_handle_anon_inode_options(priv, options); + + GF_OPTION_RECONF("use-anonymous-inode", priv->use_anon_inode, options, bool, + out); if (priv->shd.enabled) { if ((priv->shd.enabled != enabled_old) || (timeout_old != priv->shd.timeout)) @@ -485,7 +511,8 @@ init(xlator_t *this) GF_OPTION_INIT("heal-wait-queue-length", priv->heal_wait_qlen, uint32, out); GF_OPTION_INIT("data-self-heal", data_self_heal, str, out); - gf_string2boolean(data_self_heal, &priv->data_self_heal); + if (gf_string2boolean(data_self_heal, &priv->data_self_heal) == -1) + goto out; GF_OPTION_INIT("data-self-heal-algorithm", data_self_heal_algorithm, str, out); @@ -539,7 +566,9 @@ init(xlator_t *this) GF_OPTION_INIT("consistent-metadata", priv->consistent_metadata, bool, out); GF_OPTION_INIT("consistent-io", priv->consistent_io, bool, out); + afr_handle_anon_inode_options(priv, this->options); + GF_OPTION_INIT("use-anonymous-inode", priv->use_anon_inode, bool, out); if (priv->quorum_count != 0) priv->consistent_io = _gf_false; @@ -551,13 +580,19 @@ init(xlator_t *this) goto out; } + priv->anon_inode = GF_CALLOC(sizeof(unsigned char), child_count, + gf_afr_mt_char); + priv->child_up = GF_CALLOC(sizeof(unsigned char), child_count, gf_afr_mt_char); priv->child_latency = GF_MALLOC(sizeof(*priv->child_latency) * child_count, gf_afr_mt_child_latency_t); + priv->halo_child_up = GF_CALLOC(sizeof(unsigned char), child_count, + gf_afr_mt_char); - if (!priv->child_up || !priv->child_latency) { + if (!priv->child_up || !priv->child_latency || !priv->halo_child_up || + !priv->anon_inode) { ret = -ENOMEM; goto out; } @@ -1282,6 +1317,14 @@ struct volume_options options[] = { .tags = {"replicate"}, .description = "This option exists only for backward compatibility " "and configuring it doesn't have any effect"}, + {.key = {"use-anonymous-inode"}, + .type = GF_OPTION_TYPE_BOOL, + .default_value = "no", + .op_version = {GD_OP_VERSION_8_0}, + .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE, + .tags = {"replicate"}, + .description = "Setting this option heals directory renames efficiently"}, + {.key = {NULL}}, }; diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h index 8845656..d62f9a9 100644 --- a/xlators/cluster/afr/src/afr.h +++ b/xlators/cluster/afr/src/afr.h @@ -42,6 +42,7 @@ #define AFR_LK_HEAL_DOM "afr.lock-heal.domain" #define AFR_HALO_MAX_LATENCY 99999 +#define AFR_ANON_DIR_PREFIX ".glusterfs-anonymous-inode" #define PFLAG_PENDING (1 << 0) #define PFLAG_SBRAIN (1 << 1) @@ -190,7 +191,9 @@ typedef struct _afr_private { struct list_head ta_waitq; struct list_head ta_onwireq; + unsigned char *anon_inode; unsigned char *child_up; + unsigned char *halo_child_up; int64_t *child_latency; unsigned char *local; @@ -274,10 +277,15 @@ typedef struct _afr_private { gf_boolean_t esh_granular; gf_boolean_t consistent_io; gf_boolean_t data_self_heal; /* on/off */ + gf_boolean_t use_anon_inode; /*For lock healing.*/ struct list_head saved_locks; struct list_head lk_healq; + + /*For anon-inode handling */ + char anon_inode_name[NAME_MAX + 1]; + char anon_gfid_str[UUID_SIZE + 1]; } afr_private_t; typedef enum { @@ -900,7 +908,7 @@ typedef struct _afr_local { gf_boolean_t uninherit_done; gf_boolean_t uninherit_value; - /* post-op hook */ + gf_boolean_t disable_delayed_post_op; } transaction; syncbarrier_t barrier; @@ -996,7 +1004,10 @@ afr_inode_read_subvol_set(inode_t *inode, xlator_t *this, int event_generation); int -afr_inode_event_gen_reset(inode_t *inode, xlator_t *this); +__afr_inode_need_refresh_set(inode_t *inode, xlator_t *this); + +int +afr_inode_need_refresh_set(inode_t *inode, xlator_t *this); int afr_read_subvol_select_by_policy(inode_t *inode, xlator_t *this, @@ -1267,8 +1278,8 @@ int afr_inode_split_brain_choice_set(inode_t *inode, xlator_t *this, int spb_choice); int -afr_inode_split_brain_choice_get(inode_t *inode, xlator_t *this, - int *spb_choice); +afr_split_brain_read_subvol_get(inode_t *inode, xlator_t *this, + call_frame_t *frame, int *spb_subvol); int afr_get_child_index_from_name(xlator_t *this, char *name); @@ -1353,7 +1364,7 @@ int afr_set_inode_local(xlator_t *this, afr_local_t *local, inode_t *inode); int -afr_fill_ta_loc(xlator_t *this, loc_t *loc); +afr_fill_ta_loc(xlator_t *this, loc_t *loc, gf_boolean_t is_gfid_based_fop); int afr_ta_post_op_lock(xlator_t *this, loc_t *loc); @@ -1401,4 +1412,12 @@ afr_is_lock_mode_mandatory(dict_t *xdata); void afr_dom_lock_release(call_frame_t *frame); + +void +afr_fill_success_replies(afr_local_t *local, afr_private_t *priv, + unsigned char *replies); + +gf_boolean_t +afr_is_private_directory(afr_private_t *priv, uuid_t pargfid, const char *name, + pid_t pid); #endif /* __AFR_H__ */ diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c index b762f1a..8ba0cc4 100644 --- a/xlators/cluster/dht/src/dht-common.c +++ b/xlators/cluster/dht/src/dht-common.c @@ -17,6 +17,7 @@ #include <glusterfs/quota-common-utils.h> #include <glusterfs/upcall-utils.h> #include "glusterfs/compat-errno.h" // for ENODATA on BSD +#include <glusterfs/common-utils.h> #include <sys/time.h> #include <libgen.h> @@ -43,15 +44,6 @@ dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, static int dht_rmdir_unlock(call_frame_t *frame, xlator_t *this); -char *xattrs_to_heal[] = {"user.", - POSIX_ACL_ACCESS_XATTR, - POSIX_ACL_DEFAULT_XATTR, - QUOTA_LIMIT_KEY, - QUOTA_LIMIT_OBJECTS_KEY, - GF_SELINUX_XATTR_KEY, - GF_XATTR_MDATA_KEY, - NULL}; - static const char *dht_dbg_vxattrs[] = {DHT_DBG_HASHED_SUBVOL_PATTERN, NULL}; /* Check the xdata to make sure EBADF has been set by client xlator */ @@ -84,6 +76,8 @@ dht_set_fixed_dir_stat(struct iatt *stat) static gf_boolean_t dht_match_xattr(const char *key) { + char **xattrs_to_heal = get_xattrs_to_heal(); + return gf_get_index_by_elem(xattrs_to_heal, (char *)key) >= 0; } @@ -388,7 +382,7 @@ out: /* Code to save hashed subvol on inode ctx as a mds subvol */ -static int +int dht_inode_ctx_mdsvol_set(inode_t *inode, xlator_t *this, xlator_t *mds_subvol) { dht_inode_ctx_t *ctx = NULL; @@ -619,13 +613,14 @@ dht_discover_complete(xlator_t *this, call_frame_t *discover_frame) if (local->need_xattr_heal && !heal_path) { local->need_xattr_heal = 0; - ret = dht_dir_xattr_heal(this, local); - if (ret) - gf_msg(this->name, GF_LOG_ERROR, ret, + ret = dht_dir_xattr_heal(this, local, &op_errno); + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_DIR_XATTR_HEAL_FAILED, "xattr heal failed for " "directory gfid is %s ", gfid_local); + } } } @@ -695,6 +690,7 @@ dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int ret = -1; dht_conf_t *conf = 0; dht_layout_t *layout = NULL; + int32_t mds_heal_fresh_lookup = 0; GF_VALIDATE_OR_GOTO(this->name, frame, out); GF_VALIDATE_OR_GOTO(this->name, frame->local, out); @@ -702,6 +698,7 @@ dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, local = frame->local; conf = this->private; layout = local->selfheal.layout; + mds_heal_fresh_lookup = local->mds_heal_fresh_lookup; if (op_ret) { gf_msg_debug(this->name, op_ret, @@ -722,7 +719,7 @@ dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, layout); } out: - if (local && local->mds_heal_fresh_lookup) + if (mds_heal_fresh_lookup) DHT_STACK_DESTROY(frame); return 0; } @@ -1256,7 +1253,7 @@ err: to non hashed subvol */ int -dht_dir_xattr_heal(xlator_t *this, dht_local_t *local) +dht_dir_xattr_heal(xlator_t *this, dht_local_t *local, int *op_errno) { dht_local_t *copy_local = NULL; call_frame_t *copy = NULL; @@ -1268,6 +1265,7 @@ dht_dir_xattr_heal(xlator_t *this, dht_local_t *local) "No gfid exists for path %s " "so healing xattr is not possible", local->loc.path); + *op_errno = EIO; goto out; } @@ -1281,6 +1279,7 @@ dht_dir_xattr_heal(xlator_t *this, dht_local_t *local) "Memory allocation failed " "for path %s gfid %s ", local->loc.path, gfid_local); + *op_errno = ENOMEM; DHT_STACK_DESTROY(copy); } else { copy_local->stbuf = local->stbuf; @@ -1295,6 +1294,7 @@ dht_dir_xattr_heal(xlator_t *this, dht_local_t *local) "Synctask creation failed to heal xattr " "for path %s gfid %s ", local->loc.path, gfid_local); + *op_errno = ENOMEM; DHT_STACK_DESTROY(copy); } } @@ -1435,15 +1435,31 @@ dht_lookup_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, dht_aggregate_xattr(local->xattr, xattr); } + if (__is_root_gfid(stbuf->ia_gfid)) { + ret = dht_dir_has_layout(xattr, conf->xattr_name); + if (ret >= 0) { + if (is_greater_time(local->prebuf.ia_ctime, + local->prebuf.ia_ctime_nsec, + stbuf->ia_ctime, stbuf->ia_ctime_nsec)) { + /* Choose source */ + local->prebuf.ia_gid = stbuf->ia_gid; + local->prebuf.ia_uid = stbuf->ia_uid; + + local->prebuf.ia_ctime = stbuf->ia_ctime; + local->prebuf.ia_ctime_nsec = stbuf->ia_ctime_nsec; + local->prebuf.ia_prot = stbuf->ia_prot; + } + } + } + if (local->stbuf.ia_type != IA_INVAL) { /* This is not the first subvol to respond * Compare values to see if attrs need to be healed */ - if (!__is_root_gfid(stbuf->ia_gfid) && - ((local->stbuf.ia_gid != stbuf->ia_gid) || - (local->stbuf.ia_uid != stbuf->ia_uid) || - (is_permission_different(&local->stbuf.ia_prot, - &stbuf->ia_prot)))) { + if ((local->stbuf.ia_gid != stbuf->ia_gid) || + (local->stbuf.ia_uid != stbuf->ia_uid) || + (is_permission_different(&local->stbuf.ia_prot, + &stbuf->ia_prot))) { local->need_attrheal = 1; } } @@ -1635,7 +1651,7 @@ dht_revalidate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, uint32_t vol_commit_hash = 0; xlator_t *subvol = NULL; int32_t check_mds = 0; - int errst = 0; + int errst = 0, i = 0; int32_t mds_xattr_val[1] = {0}; GF_VALIDATE_OR_GOTO("dht", frame, err); @@ -1702,6 +1718,14 @@ dht_revalidate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, local->need_lookup_everywhere = 1; } else if (IA_ISDIR(local->loc.inode->ia_type)) { + layout = local->layout; + for (i = 0; i < layout->cnt; i++) { + if (layout->list[i].xlator == prev) { + layout->list[i].err = op_errno; + break; + } + } + local->need_selfheal = 1; } } @@ -2137,31 +2161,18 @@ static int dht_fill_dict_to_avoid_unlink_of_migrating_file(dict_t *dict) { int ret = 0; - xlator_t *this = NULL; - char *linktoskip_key = NULL; - this = THIS; - GF_VALIDATE_OR_GOTO("dht", this, err); - - if (dht_is_tier_xlator(this)) - linktoskip_key = TIER_SKIP_NON_LINKTO_UNLINK; - else - linktoskip_key = DHT_SKIP_NON_LINKTO_UNLINK; - - ret = dict_set_int32(dict, linktoskip_key, 1); + ret = dict_set_int32_sizen(dict, DHT_SKIP_NON_LINKTO_UNLINK, 1); if (ret) - goto err; + return -1; - ret = dict_set_int32(dict, DHT_SKIP_OPEN_FD_UNLINK, 1); + ret = dict_set_int32_sizen(dict, DHT_SKIP_OPEN_FD_UNLINK, 1); if (ret) - goto err; + return -1; return 0; - -err: - return -1; } static int32_t @@ -4290,6 +4301,8 @@ dht_find_local_subvol_cbk(call_frame_t *frame, void *cookie, xlator_t *this, index = conf->local_subvols_cnt; uuid_list_copy = gf_strdup(uuid_list); + if (!uuid_list_copy) + goto unlock; for (uuid_str = strtok_r(uuid_list, " ", &saveptr); uuid_str; uuid_str = next_uuid_str) { @@ -4580,18 +4593,8 @@ dht_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, dict_del(xattr, conf->xattr_name); dict_del(xattr, conf->mds_xattr_key); - /* filter out following two xattrs that need not - * be visible on the mount point for geo-rep - - * trusted.tier.fix.layout.complete and - * trusted.tier.tier-dht.commithash - */ - dict_del(xattr, conf->commithash_xattr_name); - if (frame->root->pid >= 0 && dht_is_tier_xlator(this)) { - dict_del(xattr, GF_XATTR_TIER_LAYOUT_FIXED_KEY); - } - if (frame->root->pid >= 0) { GF_REMOVE_INTERNAL_XATTR("trusted.glusterfs.quota*", xattr); GF_REMOVE_INTERNAL_XATTR("trusted.pgfid*", xattr); @@ -4865,6 +4868,60 @@ out: return 0; } +/* Virtual Xattr which returns 1 if all subvols are up, + else returns 0. Geo-rep then uses this virtual xattr + after a fresh mount and starts the I/O. +*/ + +enum dht_vxattr_subvol { + DHT_VXATTR_SUBVOLS_UP = 1, + DHT_VXATTR_SUBVOLS_DOWN = 0, +}; + +int +dht_vgetxattr_subvol_status(call_frame_t *frame, xlator_t *this, + const char *key) +{ + dht_local_t *local = NULL; + int ret = -1; + int op_errno = ENODATA; + int value = DHT_VXATTR_SUBVOLS_UP; + int i = 0; + dht_conf_t *conf = NULL; + + conf = this->private; + local = frame->local; + + if (!key) { + op_errno = EINVAL; + goto out; + } + local->xattr = dict_new(); + if (!local->xattr) { + op_errno = ENOMEM; + goto out; + } + for (i = 0; i < conf->subvolume_cnt; i++) { + if (!conf->subvolume_status[i]) { + value = DHT_VXATTR_SUBVOLS_DOWN; + gf_msg_debug(this->name, 0, "subvol %s is down ", + conf->subvolumes[i]->name); + break; + } + } + ret = dict_set_int8(local->xattr, (char *)key, value); + if (ret < 0) { + op_errno = -ret; + ret = -1; + goto out; + } + ret = 0; + +out: + DHT_STACK_UNWIND(getxattr, frame, ret, op_errno, local->xattr, NULL); + return 0; +} + int dht_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *key, dict_t *xdata) @@ -4922,6 +4979,11 @@ dht_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *key, goto err; } + if (strncmp(key, DHT_SUBVOL_STATUS_KEY, SLEN(DHT_SUBVOL_STATUS_KEY)) == 0) { + dht_vgetxattr_subvol_status(frame, this, key); + return 0; + } + /* skip over code which is irrelevant if !DHT_IS_DIR(layout) */ if (!DHT_IS_DIR(layout)) goto no_dht_is_dir; @@ -5371,11 +5433,13 @@ dht_dir_common_set_remove_xattr(call_frame_t *frame, xlator_t *this, loc_t *loc, int call_cnt = 0; dht_local_t *local = NULL; char gfid_local[GF_UUID_BUF_SIZE] = {0}; + char **xattrs_to_heal; conf = this->private; local = frame->local; call_cnt = conf->subvolume_cnt; local->flags = flags; + xattrs_to_heal = get_xattrs_to_heal(); if (!gf_uuid_is_null(local->gfid)) { gf_uuid_unparse(local->gfid, gfid_local); @@ -5808,22 +5872,7 @@ dht_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr, if (local->rebalance.target_node) { local->flags = forced_rebalance; - /* Flag to suggest its a tiering migration - * The reason for this dic key-value is that - * promotions and demotions are multithreaded - * so the original frame from gf_defrag_start() - * is not carried. A new frame will be created when - * we do syncop_setxattr(). This does not have the - * frame->root->pid of the original frame. So we pass - * this dic key-value when we do syncop_setxattr() to do - * data migration and set the frame->root->pid to - * GF_CLIENT_PID_TIER_DEFRAG in dht_setxattr() just before - * calling dht_start_rebalance_task() */ - tmp = dict_get(xattr, TIERING_MIGRATION_KEY); - if (tmp) - frame->root->pid = GF_CLIENT_PID_TIER_DEFRAG; - else - frame->root->pid = GF_CLIENT_PID_DEFRAG; + frame->root->pid = GF_CLIENT_PID_DEFRAG; ret = dht_start_rebalance_task(this, frame); if (!ret) @@ -6635,10 +6684,9 @@ dht_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, layout = local->layout; - /* We have seen crashes in while running "rm -rf" on tier volumes - when the layout was NULL on the hot tier. This will skip the - entries on the subvol without a layout, hence preventing the crash - but rmdir might fail with "directory not empty" errors*/ + /* This will skip the entries on the subvol without a layout, + * hence preventing the crash but rmdir might fail with + * "directory not empty" errors*/ if (layout == NULL) goto done; @@ -8282,6 +8330,11 @@ dht_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, xlator_t *prev = NULL; int ret = -1; dht_local_t *local = NULL; + gf_boolean_t parent_layout_changed = _gf_false; + char pgfid[GF_UUID_BUF_SIZE] = {0}; + xlator_t *subvol = NULL; + + local = frame->local; local = frame->local; if (!local) { @@ -8290,8 +8343,69 @@ dht_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, goto out; } - if (op_ret == -1) + if (op_ret == -1) { + local->op_errno = op_errno; + parent_layout_changed = (xdata && + dict_get(xdata, GF_PREOP_CHECK_FAILED)) + ? _gf_true + : _gf_false; + + if (parent_layout_changed) { + if (local && local->lock[0].layout.parent_layout.locks) { + /* Returning failure as the layout could not be fixed even under + * the lock */ + goto out; + } + + gf_uuid_unparse(local->loc.parent->gfid, pgfid); + gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_PARENT_LAYOUT_CHANGED, + "create (%s/%s) (path: %s): parent layout " + "changed. Attempting a layout refresh and then a " + "retry", + pgfid, local->loc.name, local->loc.path); + + /* + dht_refresh_layout needs directory info in local->loc.Hence, + storing the parent_loc in local->loc and storing the create + context in local->loc2. We will restore this information in + dht_creation_do. + */ + + loc_wipe(&local->loc2); + + ret = loc_copy(&local->loc2, &local->loc); + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, + "loc_copy failed %s", local->loc.path); + + goto out; + } + + loc_wipe(&local->loc); + + ret = dht_build_parent_loc(this, &local->loc, &local->loc2, + &op_errno); + + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_LOC_FAILED, + "parent loc build failed"); + goto out; + } + + subvol = dht_subvol_get_hashed(this, &local->loc2); + + ret = dht_create_lock(frame, subvol); + if (ret < 0) { + gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR, + "locking parent failed"); + goto out; + } + + return 0; + } + goto out; + } prev = cookie; @@ -8412,6 +8526,8 @@ dht_create_wind_to_avail_subvol(call_frame_t *frame, xlator_t *this, gf_msg_debug(this->name, 0, "creating %s on %s", loc->path, subvol->name); + dht_set_parent_layout_in_dict(loc, this, local); + STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol, subvol->fops->create, loc, flags, mode, umask, fd, params); @@ -8420,10 +8536,6 @@ dht_create_wind_to_avail_subvol(call_frame_t *frame, xlator_t *this, avail_subvol = dht_free_disk_available_subvol(this, subvol, local); if (avail_subvol != subvol) { - local->params = dict_ref(params); - local->flags = flags; - local->mode = mode; - local->umask = umask; local->cached_subvol = avail_subvol; local->hashed_subvol = subvol; @@ -8439,6 +8551,8 @@ dht_create_wind_to_avail_subvol(call_frame_t *frame, xlator_t *this, gf_msg_debug(this->name, 0, "creating %s on %s", loc->path, subvol->name); + dht_set_parent_layout_in_dict(loc, this, local); + STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol, subvol->fops->create, loc, flags, mode, umask, fd, params); @@ -8654,7 +8768,7 @@ err: return 0; } -static int32_t +int32_t dht_create_lock(call_frame_t *frame, xlator_t *subvol) { dht_local_t *local = NULL; @@ -8700,6 +8814,60 @@ err: } int +dht_set_parent_layout_in_dict(loc_t *loc, xlator_t *this, dht_local_t *local) +{ + dht_conf_t *conf = this->private; + dht_layout_t *parent_layout = NULL; + int *parent_disk_layout = NULL; + xlator_t *hashed_subvol = NULL; + char pgfid[GF_UUID_BUF_SIZE] = {0}; + int ret = 0; + + gf_uuid_unparse(loc->parent->gfid, pgfid); + + parent_layout = dht_layout_get(this, loc->parent); + hashed_subvol = dht_subvol_get_hashed(this, loc); + + ret = dht_disk_layout_extract_for_subvol(this, parent_layout, hashed_subvol, + &parent_disk_layout); + if (ret == -1) { + gf_msg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_PARENT_LAYOUT_CHANGED, + "%s (%s/%s) (path: %s): " + "extracting in-memory layout of parent failed. ", + gf_fop_list[local->fop], pgfid, loc->name, loc->path); + goto err; + } + + ret = dict_set_str_sizen(local->params, GF_PREOP_PARENT_KEY, + conf->xattr_name); + if (ret < 0) { + gf_msg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_PARENT_LAYOUT_CHANGED, + "%s (%s/%s) (path: %s): " + "setting %s key in params dictionary failed. ", + gf_fop_list[local->fop], pgfid, loc->name, loc->path, + GF_PREOP_PARENT_KEY); + goto err; + } + + ret = dict_set_bin(local->params, conf->xattr_name, parent_disk_layout, + 4 * 4); + if (ret < 0) { + gf_msg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_PARENT_LAYOUT_CHANGED, + "%s (%s/%s) (path: %s): " + "setting parent-layout in params dictionary failed. ", + gf_fop_list[local->fop], pgfid, loc->name, loc->path); + goto err; + } + +err: + dht_layout_unref(this, parent_layout); + return ret; +} + +int dht_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, mode_t mode, mode_t umask, fd_t *fd, dict_t *params) { @@ -8725,6 +8893,11 @@ dht_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, goto err; } + local->params = dict_ref(params); + local->flags = flags; + local->mode = mode; + local->umask = umask; + if (dht_filter_loc_subvol_key(this, loc, &local->loc, &subvol)) { gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO, "creating %s on %s (got create on %s)", local->loc.path, @@ -8740,10 +8913,6 @@ dht_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, if (hashed_subvol && (hashed_subvol != subvol)) { /* Create the linkto file and then the data file */ - local->params = dict_ref(params); - local->flags = flags; - local->mode = mode; - local->umask = umask; local->cached_subvol = subvol; local->hashed_subvol = hashed_subvol; @@ -8756,6 +8925,9 @@ dht_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, * file as we expect a lookup everywhere if there are problems * with the parent layout */ + + dht_set_parent_layout_in_dict(loc, this, local); + STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol, subvol->fops->create, &local->loc, flags, mode, umask, fd, params); @@ -8807,11 +8979,6 @@ dht_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, goto err; } - local->params = dict_ref(params); - local->flags = flags; - local->mode = mode; - local->umask = umask; - loc_wipe(&local->loc); ret = dht_build_parent_loc(this, &local->loc, loc, &op_errno); @@ -10646,23 +10813,17 @@ dht_notify(xlator_t *this, int event, void *data, ...) int had_heard_from_all = 0; int have_heard_from_all = 0; - struct timeval time = { - 0, - }; gf_defrag_info_t *defrag = NULL; dict_t *dict = NULL; gf_defrag_type cmd = 0; dict_t *output = NULL; va_list ap; - dht_methods_t *methods = NULL; struct gf_upcall *up_data = NULL; struct gf_upcall_cache_invalidation *up_ci = NULL; conf = this->private; GF_VALIDATE_OR_GOTO(this->name, conf, out); - methods = &(conf->methods); - /* had all subvolumes reported status once till now? */ had_heard_from_all = 1; for (i = 0; i < conf->subvolume_cnt; i++) { @@ -10692,12 +10853,11 @@ dht_notify(xlator_t *this, int event, void *data, ...) break; } - gettimeofday(&time, NULL); LOCK(&conf->subvolume_lock); { conf->subvolume_status[cnt] = 1; conf->last_event[cnt] = event; - conf->subvol_up_time[cnt] = time.tv_sec; + conf->subvol_up_time[cnt] = gf_time(); } UNLOCK(&conf->subvolume_lock); @@ -10805,21 +10965,13 @@ dht_notify(xlator_t *this, int event, void *data, ...) if (defrag->is_exiting) goto unlock; if ((cmd == GF_DEFRAG_CMD_STATUS) || - (cmd == GF_DEFRAG_CMD_STATUS_TIER) || (cmd == GF_DEFRAG_CMD_DETACH_STATUS)) gf_defrag_status_get(conf, output); - else if (cmd == GF_DEFRAG_CMD_START_DETACH_TIER) - gf_defrag_start_detach_tier(defrag); else if (cmd == GF_DEFRAG_CMD_DETACH_START) defrag->cmd = GF_DEFRAG_CMD_DETACH_START; else if (cmd == GF_DEFRAG_CMD_STOP || - cmd == GF_DEFRAG_CMD_STOP_DETACH_TIER || cmd == GF_DEFRAG_CMD_DETACH_STOP) gf_defrag_stop(conf, GF_DEFRAG_STATUS_STOPPED, output); - else if (cmd == GF_DEFRAG_CMD_PAUSE_TIER) - ret = gf_defrag_pause_tier(this, defrag); - else if (cmd == GF_DEFRAG_CMD_RESUME_TIER) - ret = gf_defrag_resume_tier(this, defrag); } unlock: UNLOCK(&defrag->lock); @@ -10894,15 +11046,13 @@ dht_notify(xlator_t *this, int event, void *data, ...) * thread has already started. */ if (conf->defrag && !run_defrag) { - if (methods->migration_needed(this)) { - run_defrag = 1; - ret = gf_thread_create(&conf->defrag->th, NULL, gf_defrag_start, - this, "dhtdg"); - if (ret) { - GF_FREE(conf->defrag); - conf->defrag = NULL; - kill(getpid(), SIGTERM); - } + run_defrag = 1; + ret = gf_thread_create(&conf->defrag->th, NULL, gf_defrag_start, + this, "dhtdg"); + if (ret) { + GF_FREE(conf->defrag); + conf->defrag = NULL; + kill(getpid(), SIGTERM); } } } @@ -11047,28 +11197,6 @@ out: return ret; } -int32_t -dht_migration_needed(xlator_t *this) -{ - gf_defrag_info_t *defrag = NULL; - dht_conf_t *conf = NULL; - int ret = 0; - - conf = this->private; - - GF_VALIDATE_OR_GOTO("dht", conf, out); - GF_VALIDATE_OR_GOTO("dht", conf->defrag, out); - - defrag = conf->defrag; - - if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) && - (defrag->cmd != GF_DEFRAG_CMD_START_DETACH_TIER)) - ret = 1; - -out: - return ret; -} - /* This function should not be called more then once during a FOP handling path. It is valid only for for ops on files @@ -11103,14 +11231,6 @@ dht_set_local_rebalance(xlator_t *this, dht_local_t *local, struct iatt *stbuf, return 0; } -gf_boolean_t -dht_is_tier_xlator(xlator_t *this) -{ - if (strcmp(this->type, "cluster/tier") == 0) - return _gf_true; - return _gf_false; -} - int32_t dht_release(xlator_t *this, fd_t *fd) { @@ -11250,3 +11370,22 @@ dht_pt_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *key, FIRST_CHILD(this)->fops->fgetxattr, fd, key, xdata); return 0; } + +/* The job of this function is to check if all the xlators have updated + * error in the layout. */ +int +dht_dir_layout_error_check(xlator_t *this, inode_t *inode) +{ + dht_layout_t *layout = NULL; + int i = 0; + + layout = dht_layout_get(this, inode); + for (i = 0; i < layout->cnt; i++) { + if (layout->list[i].err == 0) { + return 0; + } + } + + /* Returning the first xlator error as all xlators have errors */ + return layout->list[0].err; +} diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h index 288bdf7..fe0dc3d 100644 --- a/xlators/cluster/dht/src/dht-common.h +++ b/xlators/cluster/dht/src/dht-common.h @@ -24,7 +24,6 @@ #define _DHT_H #define GF_XATTR_FIX_LAYOUT_KEY "distribute.fix.layout" -#define GF_XATTR_TIER_LAYOUT_FIXED_KEY "trusted.tier.fix.layout.complete" #define GF_XATTR_FILE_MIGRATE_KEY "trusted.distribute.migrate-data" #define DHT_MDS_STR "mds" #define GF_DHT_LOOKUP_UNHASHED_OFF 0 @@ -36,22 +35,21 @@ #define DHT_LAYOUT_HEAL_DOMAIN "dht.layout.heal" /* Namespace synchronization */ #define DHT_ENTRY_SYNC_DOMAIN "dht.entry.sync" -#define TIERING_MIGRATION_KEY "tiering.migration" #define DHT_LAYOUT_HASH_INVALID 1 #define MAX_REBAL_THREADS sysconf(_SC_NPROCESSORS_ONLN) #define DHT_DIR_STAT_BLOCKS 8 #define DHT_DIR_STAT_SIZE 4096 +/* Virtual xattr for subvols status */ + +#define DHT_SUBVOL_STATUS_KEY "dht.subvol.status" + /* Virtual xattrs for debugging */ #define DHT_DBG_HASHED_SUBVOL_PATTERN "dht.file.hashed-subvol.*" #define DHT_DBG_HASHED_SUBVOL_KEY "dht.file.hashed-subvol." -/* Array to hold custom xattr keys - */ -extern char *xattrs_to_heal[]; - /* Rebalance nodeuuid flags */ #define REBAL_NODEUUID_MINE 0x01 @@ -242,19 +240,6 @@ typedef gf_boolean_t (*dht_need_heal_t)(call_frame_t *frame, dht_layout_t **inmem, dht_layout_t **ondisk); -typedef struct { - uint64_t blocks_used; - uint64_t pblocks_used; - uint64_t files_used; - uint64_t pfiles_used; - uint64_t unhashed_blocks_used; - uint64_t unhashed_pblocks_used; - uint64_t unhashed_files_used; - uint64_t unhashed_pfiles_used; - uint64_t unhashed_fsid; - uint64_t hashed_fsid; -} tier_statvfs_t; - struct dht_local { loc_t loc; loc_t loc2; @@ -272,7 +257,6 @@ struct dht_local { struct iatt preparent; struct iatt postparent; struct statvfs statvfs; - tier_statvfs_t tier_statvfs; fd_t *fd; inode_t *inode; dict_t *params; @@ -405,14 +389,7 @@ enum gf_defrag_type { GF_DEFRAG_CMD_STATUS = 1 + 2, GF_DEFRAG_CMD_START_LAYOUT_FIX = 1 + 3, GF_DEFRAG_CMD_START_FORCE = 1 + 4, - GF_DEFRAG_CMD_START_TIER = 1 + 5, - GF_DEFRAG_CMD_STATUS_TIER = 1 + 6, - GF_DEFRAG_CMD_START_DETACH_TIER = 1 + 7, - GF_DEFRAG_CMD_STOP_DETACH_TIER = 1 + 8, - GF_DEFRAG_CMD_PAUSE_TIER = 1 + 9, - GF_DEFRAG_CMD_RESUME_TIER = 1 + 10, GF_DEFRAG_CMD_DETACH_STATUS = 1 + 11, - GF_DEFRAG_CMD_STOP_TIER = 1 + 12, GF_DEFRAG_CMD_DETACH_START = 1 + 13, GF_DEFRAG_CMD_DETACH_COMMIT = 1 + 14, GF_DEFRAG_CMD_DETACH_COMMIT_FORCE = 1 + 15, @@ -463,75 +440,6 @@ struct dht_container { int local_subvol_index; }; -typedef enum tier_mode_ { - TIER_MODE_NONE = 0, - TIER_MODE_TEST, - TIER_MODE_WM -} tier_mode_t; - -typedef enum tier_pause_state_ { - TIER_RUNNING = 0, - TIER_REQUEST_PAUSE, - TIER_PAUSED -} tier_pause_state_t; - -/* This Structure is only used in tiering fixlayout */ -typedef struct gf_tier_fix_layout_arg { - xlator_t *this; - dict_t *fix_layout; - pthread_t thread_id; -} gf_tier_fix_layout_arg_t; - -typedef struct gf_tier_conf { - int is_tier; - int watermark_hi; - int watermark_low; - int watermark_last; - unsigned long block_size; - fsblkcnt_t blocks_total; - fsblkcnt_t blocks_used; - uint64_t max_migrate_bytes; - int max_migrate_files; - int query_limit; - tier_mode_t mode; - int percent_full; - /* These flags are only used for tier-compact */ - gf_boolean_t compact_active; - /* These 3 flags are set to true when the client changes the */ - /* compaction mode on the command line. */ - /* When they are set, the daemon will trigger compaction as */ - /* soon as possible to activate or deactivate compaction. */ - /* If in the middle of a compaction, then the switches take */ - /* effect on the next compaction, not the current one. */ - /* If the user switches it off, we want to avoid needless */ - /* compactions. */ - /* If the user switches it on, they want to compact as soon */ - /* as possible. */ - gf_boolean_t compact_mode_switched; - gf_boolean_t compact_mode_switched_hot; - gf_boolean_t compact_mode_switched_cold; - int tier_max_promote_size; - int tier_promote_frequency; - int tier_demote_frequency; - int tier_compact_hot_frequency; - int tier_compact_cold_frequency; - uint64_t st_last_promoted_size; - uint64_t st_last_demoted_size; - struct synctask *pause_synctask; - gf_timer_t *pause_timer; - pthread_mutex_t pause_mutex; - int promote_in_progress; - int demote_in_progress; - /* This Structure is only used in tiering fixlayout */ - gf_tier_fix_layout_arg_t tier_fix_layout_arg; - /* Indicates the index of the first queryfile picked - * in the last cycle of promote or demote */ - int32_t last_promote_qfile_index; - int32_t last_demote_qfile_index; - tier_pause_state_t pause_state; - char volname[GD_VOLUME_NAME_MAX + 1]; -} gf_tier_conf_t; - typedef struct nodeuuid_info { char info; /* Set to 1 is this is my node's uuid*/ uuid_t uuid; /* Store the nodeuuid as well for debugging*/ @@ -559,17 +467,10 @@ struct gf_defrag_info_ { int cmd; inode_t *root_inode; uuid_t node_uuid; - struct timeval start_time; + time_t start_time; uint32_t new_commit_hash; gf_defrag_status_t defrag_status; gf_defrag_pattern_list_t *defrag_pattern; - gf_tier_conf_t tier_conf; - - /*Data Tiering params for scanner*/ - uint64_t total_files_promoted; - uint64_t total_files_demoted; - int write_freq_threshold; - int read_freq_threshold; pthread_cond_t parallel_migration_cond; pthread_mutex_t dfq_mutex; @@ -605,7 +506,6 @@ typedef struct gf_defrag_info_ gf_defrag_info_t; struct dht_methods_s { int32_t (*migration_get_dst_subvol)(xlator_t *this, dht_local_t *local); int32_t (*migration_other)(xlator_t *this, gf_defrag_info_t *defrag); - int32_t (*migration_needed)(xlator_t *this); xlator_t *(*layout_search)(xlator_t *this, dht_layout_t *layout, const char *name); }; @@ -626,7 +526,7 @@ struct dht_conf { int subvolume_cnt; int32_t refresh_interval; gf_lock_t subvolume_lock; - struct timeval last_stat_fetch; + time_t last_stat_fetch; gf_lock_t layout_lock; dict_t *leaf_to_subvol; void *private; /* Can be used by wrapper xlators over @@ -748,6 +648,8 @@ struct dir_dfmeta { struct list_head **head; struct list_head **iterator; int *fetch_entries; + /* fds corresponding to local subvols only */ + fd_t **lfd; }; typedef struct dht_migrate_info { @@ -879,7 +781,6 @@ dht_layout_anomalies(xlator_t *this, loc_t *loc, dht_layout_t *layout, 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); @@ -897,9 +798,6 @@ int dht_disk_layout_extract(xlator_t *this, dht_layout_t *layout, int pos, int32_t **disk_layout_p); int -dht_disk_layout_merge(xlator_t *this, dht_layout_t *layout, int pos, - void *disk_layout_raw, int disk_layout_len); -int dht_disk_layout_extract_for_subvol(xlator_t *this, dht_layout_t *layout, xlator_t *subvol, int32_t **disk_layout_p); @@ -945,7 +843,7 @@ dht_selfheal_new_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t cbk, int dht_selfheal_restore(call_frame_t *frame, dht_selfheal_dir_cbk_t cbk, loc_t *loc, dht_layout_t *layout); -int +void dht_layout_sort_volname(dht_layout_t *layout); int @@ -962,14 +860,14 @@ dht_get_du_info_for_subvol(xlator_t *this, int subvol_idx); int dht_layout_preset(xlator_t *this, xlator_t *subvol, inode_t *inode); int -dht_layout_index_for_subvol(dht_layout_t *layout, xlator_t *subvol); -int dht_layout_set(xlator_t *this, inode_t *inode, dht_layout_t *layout); ; void dht_layout_unref(xlator_t *this, dht_layout_t *layout); dht_layout_t * dht_layout_ref(xlator_t *this, dht_layout_t *layout); +int +dht_layout_index_for_subvol(dht_layout_t *layout, xlator_t *subvol); xlator_t * dht_first_up_subvol(xlator_t *this); xlator_t * @@ -1236,30 +1134,8 @@ dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata); int -dht_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata); - -int gf_defrag_status_get(dht_conf_t *conf, dict_t *dict); -void -gf_defrag_set_pause_state(gf_tier_conf_t *tier_conf, tier_pause_state_t state); - -tier_pause_state_t -gf_defrag_get_pause_state(gf_tier_conf_t *tier_conf); - -int -gf_defrag_pause_tier(xlator_t *this, gf_defrag_info_t *defrag); - -tier_pause_state_t -gf_defrag_check_pause_tier(gf_tier_conf_t *defrag); - -int -gf_defrag_resume_tier(xlator_t *this, gf_defrag_info_t *defrag); - -int -gf_defrag_start_detach_tier(gf_defrag_info_t *defrag); - int gf_defrag_stop(dht_conf_t *conf, gf_defrag_status_t status, dict_t *output); @@ -1291,10 +1167,6 @@ int dht_dir_attr_heal(void *data); int dht_dir_attr_heal_done(int ret, call_frame_t *sync_frame, void *data); -int -dht_dir_has_layout(dict_t *xattr, char *name); -gf_boolean_t -dht_is_subvol_in_layout(dht_layout_t *layout, xlator_t *xlator); xlator_t * dht_subvol_with_free_space_inodes(xlator_t *this, xlator_t *subvol, xlator_t *ignore, dht_layout_t *layout, @@ -1303,15 +1175,18 @@ xlator_t * dht_subvol_maxspace_nonzeroinode(xlator_t *this, xlator_t *subvol, dht_layout_t *layout); int +dht_dir_has_layout(dict_t *xattr, char *name); +int dht_linkfile_attr_heal(call_frame_t *frame, xlator_t *this); -void -dht_layout_dump(dht_layout_t *layout, const char *prefix); int32_t dht_priv_dump(xlator_t *this); int32_t dht_inodectx_dump(xlator_t *this, inode_t *inode); +gf_boolean_t +dht_is_subvol_in_layout(dht_layout_t *layout, xlator_t *xlator); + int dht_inode_ctx_get_mig_info(xlator_t *this, inode_t *inode, xlator_t **src_subvol, xlator_t **dst_subvol); @@ -1341,9 +1216,6 @@ dht_layout_missing_dirs(dht_layout_t *layout); int dht_refresh_layout(call_frame_t *frame); -gf_boolean_t -dht_is_tier_xlator(xlator_t *this); - int dht_build_parent_loc(xlator_t *this, loc_t *parent, loc_t *child, int32_t *op_errno); @@ -1456,7 +1328,7 @@ dht_dir_set_heal_xattr(xlator_t *this, dht_local_t *local, dict_t *dst, dict_t *src, int *uret, int *uflag); int -dht_dir_xattr_heal(xlator_t *this, dht_local_t *local); +dht_dir_xattr_heal(xlator_t *this, dht_local_t *local, int *op_errno); int dht_common_mark_mdsxattr(call_frame_t *frame, int *errst, int flag); @@ -1498,4 +1370,15 @@ dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata); +int32_t +dht_create_lock(call_frame_t *frame, xlator_t *subvol); + +int +dht_set_parent_layout_in_dict(loc_t *loc, xlator_t *this, dht_local_t *local); + +int +dht_dir_layout_error_check(xlator_t *this, inode_t *inode); + +int +dht_inode_ctx_mdsvol_set(inode_t *inode, xlator_t *this, xlator_t *mds_subvol); #endif /* _DHT_H */ diff --git a/xlators/cluster/dht/src/dht-diskusage.c b/xlators/cluster/dht/src/dht-diskusage.c index 27097ca..c058882 100644 --- a/xlators/cluster/dht/src/dht-diskusage.c +++ b/xlators/cluster/dht/src/dht-diskusage.c @@ -151,22 +151,18 @@ dht_get_du_info(call_frame_t *frame, xlator_t *this, loc_t *loc) dht_conf_t *conf = NULL; call_frame_t *statfs_frame = NULL; dht_local_t *statfs_local = NULL; - struct timeval tv = { - 0, - }; loc_t tmp_loc = { 0, }; + time_t now; conf = this->private; - - gettimeofday(&tv, NULL); - + now = gf_time(); /* make it root gfid, should be enough to get the proper info back */ tmp_loc.gfid[15] = 1; - if (tv.tv_sec > (conf->refresh_interval + conf->last_stat_fetch.tv_sec)) { + if (now > (conf->refresh_interval + conf->last_stat_fetch)) { statfs_frame = copy_frame(frame); if (!statfs_frame) { goto err; @@ -198,7 +194,7 @@ dht_get_du_info(call_frame_t *frame, xlator_t *this, loc_t *loc) statfs_local->params); } - conf->last_stat_fetch.tv_sec = tv.tv_sec; + conf->last_stat_fetch = now; } return 0; err: diff --git a/xlators/cluster/dht/src/dht-hashfn.c b/xlators/cluster/dht/src/dht-hashfn.c index 1dfe2a7..acda67c 100644 --- a/xlators/cluster/dht/src/dht-hashfn.c +++ b/xlators/cluster/dht/src/dht-hashfn.c @@ -78,6 +78,9 @@ dht_hash_compute(xlator_t *this, int type, const char *name, uint32_t *hash_p) priv = this->private; + if (name == NULL) + return -1; + len = strlen(name) + 1; rsync_friendly_name = alloca(len); diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c index 023d437..3f2fe43 100644 --- a/xlators/cluster/dht/src/dht-helper.c +++ b/xlators/cluster/dht/src/dht-helper.c @@ -64,8 +64,8 @@ __dht_fd_ctx_set(xlator_t *this, fd_t *fd, xlator_t *dst) ret = __fd_ctx_set(fd, this, value); if (ret < 0) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_FD_CTX_SET_FAILED, - "Failed to set fd ctx in fd=0x%p", fd); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_FD_CTX_SET_FAILED, + "fd=0x%p", fd, NULL); GF_REF_PUT(fd_ctx); } out: @@ -96,8 +96,8 @@ dht_fd_ctx_set(xlator_t *this, fd_t *fd, xlator_t *dst) /* Overwrite and hope for the best*/ fd_ctx->opened_on_dst = (uint64_t)(uintptr_t)dst; UNLOCK(&fd->lock); - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_INVALID_VALUE, - "Different dst found in the fd ctx"); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_INVALID_VALUE, + NULL); goto out; } @@ -383,9 +383,9 @@ dht_check_and_open_fd_on_subvol_complete(int ret, call_frame_t *frame, &local->rebalance.flock, local->xattr_req); break; default: - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UNKNOWN_FOP, - "Unknown FOP on fd (%p) on file %s @ %s", fd, - uuid_utoa(fd->inode->gfid), subvol->name); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UNKNOWN_FOP, "fd=%p", + fd, "gfid=%s", uuid_utoa(fd->inode->gfid), "name=%s", + subvol->name, NULL); break; } @@ -458,9 +458,9 @@ handle_err: break; default: - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UNKNOWN_FOP, - "Unknown FOP on fd (%p) on file %s @ %s", fd, - uuid_utoa(fd->inode->gfid), subvol->name); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UNKNOWN_FOP, "fd=%p", + fd, "gfid=%s", uuid_utoa(fd->inode->gfid), "name=%s", + subvol->name, NULL); break; } @@ -513,10 +513,9 @@ dht_check_and_open_fd_on_subvol_task(void *data) fd, NULL, NULL); if (ret < 0) { - gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_OPEN_FD_ON_DST_FAILED, - "Failed to open the fd" - " (%p, flags=0%o) on file %s @ %s", - fd, fd->flags, uuid_utoa(fd->inode->gfid), subvol->name); + gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_OPEN_FD_ON_DST_FAILED, + "fd=%p", fd, "flags=0%o", fd->flags, "gfid=%s", + uuid_utoa(fd->inode->gfid), "name=%s", subvol->name, NULL); /* This can happen if the cached subvol was updated in the * inode_ctx and the fd was opened on the new cached suvol * after this fop was wound on the old cached subvol. @@ -562,10 +561,8 @@ dht_check_and_open_fd_on_subvol(xlator_t *this, call_frame_t *frame) dht_check_and_open_fd_on_subvol_complete, frame, frame); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, 0, - "Failed to create synctask" - " to check and open fd=%p", - local->fd); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SYNCTASK_CREATE_FAILED, + "to-check-and-open fd=%p", local->fd, NULL); } return ret; @@ -674,9 +671,7 @@ dht_get_subvol_from_id(xlator_t *this, int client_id) ret = gf_asprintf(&sid, "%d", client_id); if (ret == -1) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_ASPRINTF_FAILED, - "asprintf failed while " - "fetching subvol from the id"); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_ASPRINTF_FAILED, NULL); goto out; } @@ -1336,9 +1331,9 @@ dht_migration_complete_check_task(void *data) * migrated by two different layers. Raise * a warning here. */ - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_HAS_MIGINFO, - "%s: Found miginfo in the inode ctx", - tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid)); + gf_smsg( + this->name, GF_LOG_WARNING, 0, DHT_MSG_HAS_MIGINFO, "tmp=%s", + tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), NULL); miginfo = (void *)(uintptr_t)tmp_miginfo; GF_REF_PUT(miginfo); @@ -1359,10 +1354,9 @@ dht_migration_complete_check_task(void *data) ret = syncop_lookup(this, &tmp_loc, &stbuf, 0, 0, 0); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_FILE_LOOKUP_FAILED, - "%s: failed to lookup the file on %s", - tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), - this->name); + gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_FILE_LOOKUP_FAILED, + "tmp=%s", tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), + "name=%s", this->name, NULL); local->op_errno = -ret; ret = -1; goto out; @@ -1370,18 +1364,15 @@ dht_migration_complete_check_task(void *data) dst_node = dht_subvol_get_cached(this, tmp_loc.inode); if (linkto_target && dst_node != linkto_target) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_INVALID_LINKFILE, - "linkto target (%s) is " - "different from cached-subvol (%s). Treating %s as " - "destination subvol", - linkto_target->name, dst_node->name, dst_node->name); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_INVALID_LINKFILE, + "linkto_target_name=%s", linkto_target->name, "dst_name=%s", + dst_node->name, NULL); } if (gf_uuid_compare(stbuf.ia_gfid, tmp_loc.inode->gfid)) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH, - "%s: gfid different on the target file on %s", - tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), - dst_node->name); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH, "tmp=%s", + tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), + "dst_name=%s", dst_node->name, NULL); ret = -1; local->op_errno = EIO; goto out; @@ -1463,12 +1454,10 @@ dht_migration_complete_check_task(void *data) (iter_fd->flags & ~(O_CREAT | O_EXCL | O_TRUNC)), iter_fd, NULL, NULL); if (ret < 0) { - gf_msg(this->name, GF_LOG_ERROR, -ret, - DHT_MSG_OPEN_FD_ON_DST_FAILED, - "failed" - " to open the fd" - " (%p, flags=0%o) on file %s @ %s", - iter_fd, iter_fd->flags, path, dst_node->name); + gf_smsg(this->name, GF_LOG_ERROR, -ret, + DHT_MSG_OPEN_FD_ON_DST_FAILED, "id=%p", iter_fd, + "flags=0%o", iter_fd->flags, "path=%s", path, "name=%s", + dst_node->name, NULL); open_failed = 1; local->op_errno = -ret; @@ -1622,9 +1611,9 @@ dht_rebalance_inprogress_task(void *data) * migrated by two different layers. Raise * a warning here. */ - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_HAS_MIGINFO, - "%s: Found miginfo in the inode ctx", - tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid)); + gf_smsg( + this->name, GF_LOG_WARNING, 0, DHT_MSG_HAS_MIGINFO, "tmp=%s", + tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), NULL); miginfo = (void *)(uintptr_t)tmp_miginfo; GF_REF_PUT(miginfo); } @@ -1633,17 +1622,16 @@ dht_rebalance_inprogress_task(void *data) } if (ret < 0) { - gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_GET_XATTR_FAILED, - "%s: failed to get the 'linkto' xattr", local->loc.path); + gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_GET_XATTR_FAILED, + "path=%s", local->loc.path, NULL); ret = -1; goto out; } dst_node = dht_linkfile_subvol(this, NULL, NULL, dict); if (!dst_node) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SUBVOL_NOT_FOUND, - "%s: failed to get the 'linkto' xattr from dict", - local->loc.path); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GET_XATTR_FAILED, + "path=%s", local->loc.path, NULL); ret = -1; goto out; } @@ -1660,20 +1648,17 @@ dht_rebalance_inprogress_task(void *data) /* lookup on dst */ ret = syncop_lookup(dst_node, &tmp_loc, &stbuf, NULL, NULL, NULL); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, -ret, - DHT_MSG_FILE_LOOKUP_ON_DST_FAILED, - "%s: failed to lookup the file on %s", - tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), - dst_node->name); + gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_FILE_LOOKUP_FAILED, + "tmp=%s", tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), + "name=%s", dst_node->name, NULL); ret = -1; goto out; } if (gf_uuid_compare(stbuf.ia_gfid, tmp_loc.inode->gfid)) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH, - "%s: gfid different on the target file on %s", - tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), - dst_node->name); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH, "tmp=%s", + tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), + "name=%s", dst_node->name, NULL); ret = -1; goto out; } @@ -1741,11 +1726,10 @@ dht_rebalance_inprogress_task(void *data) (iter_fd->flags & ~(O_CREAT | O_EXCL | O_TRUNC)), iter_fd, NULL, NULL); if (ret < 0) { - gf_msg(this->name, GF_LOG_ERROR, -ret, - DHT_MSG_OPEN_FD_ON_DST_FAILED, - "failed to send open " - "the fd (%p, flags=0%o) on file %s @ %s", - iter_fd, iter_fd->flags, path, dst_node->name); + gf_smsg(this->name, GF_LOG_ERROR, -ret, + DHT_MSG_OPEN_FD_ON_DST_FAILED, "fd=%p", iter_fd, + "flags=0%o", iter_fd->flags, "path=%s", path, "name=%s", + dst_node->name, NULL); ret = -1; open_failed = 1; } else { @@ -1777,9 +1761,8 @@ unlock: ret = dht_inode_ctx_set_mig_info(this, inode, src_node, dst_node); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED, - "%s: failed to set inode-ctx target file at %s", local->loc.path, - dst_node->name); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED, + "path=%s", local->loc.path, "name=%s", dst_node->name, NULL); goto out; } @@ -2001,10 +1984,9 @@ dht_heal_path(xlator_t *this, char *path, inode_table_t *itable) ret = syncop_lookup(this, &loc, &iatt, NULL, NULL, NULL); if (ret) { - gf_msg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_SELFHEAL_FAILED, - "Healing of path %s failed on subvolume %s for " - "directory %s", - path, this->name, bname); + gf_smsg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_SELFHEAL_FAILED, + "path=%s", path, "subvolume=%s", this->name, "bname=%s", + bname, NULL); goto out; } @@ -2062,10 +2044,8 @@ dht_heal_full_path(void *data) ret = syncop_getxattr(source, &loc, &dict, GET_ANCESTRY_PATH_KEY, NULL, NULL); if (ret) { - gf_msg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_SELFHEAL_FAILED, - "Failed to get path from subvol %s. Aborting " - "directory healing.", - source->name); + gf_smsg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_HEAL_ABORT, + "subvol=%s", source->name, NULL); goto out; } @@ -2103,6 +2083,7 @@ dht_heal_full_path_done(int op_ret, call_frame_t *heal_frame, void *data) dht_local_t *local = NULL; xlator_t *this = NULL; int ret = -1; + int op_errno = 0; local = heal_frame->local; main_frame = local->main_frame; @@ -2112,10 +2093,12 @@ dht_heal_full_path_done(int op_ret, call_frame_t *heal_frame, void *data) dht_set_fixed_dir_stat(&local->postparent); if (local->need_xattr_heal) { local->need_xattr_heal = 0; - ret = dht_dir_xattr_heal(this, local); - if (ret) - gf_msg(this->name, GF_LOG_ERROR, ret, DHT_MSG_DIR_XATTR_HEAL_FAILED, - "xattr heal failed for directory %s ", local->loc.path); + ret = dht_dir_xattr_heal(this, local, &op_errno); + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, op_errno, + DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s", local->loc.path, + NULL); + } } DHT_STACK_UNWIND(lookup, main_frame, 0, 0, local->inode, &local->stbuf, @@ -2203,8 +2186,8 @@ dht_get_lock_subvolume(xlator_t *this, struct gf_flock *lock, if (ret) { gf_uuid_unparse(inode->gfid, gfid); UNLOCK(&inode->lock); - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SET_INODE_CTX_FAILED, - "Failed to set lock_subvol in inode ctx for gfid %s", gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SET_INODE_CTX_FAILED, + "lock_subvol gfid=%s", gfid, NULL); goto post_unlock; } subvol = cached_subvol; @@ -2234,8 +2217,8 @@ dht_lk_inode_unref(call_frame_t *frame, int32_t op_ret) inode = local->loc.inode ? local->loc.inode : local->fd->inode; } if (!inode) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LOCK_INODE_UNREF_FAILED, - "Found a NULL inode. Failed to unref the inode"); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LOCK_INODE_UNREF_FAILED, + NULL); goto out; } @@ -2261,11 +2244,8 @@ dht_lk_inode_unref(call_frame_t *frame, int32_t op_ret) inode_unref(inode); } else { gf_uuid_unparse(inode->gfid, gfid); - gf_msg(this->name, GF_LOG_WARNING, 0, - DHT_MSG_LOCK_INODE_UNREF_FAILED, - "Unlock request failed for gfid %s." - "Failed to unref the inode", - gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, + DHT_MSG_LOCK_INODE_UNREF_FAILED, "gfid=%s", gfid, NULL); goto out; } default: @@ -2287,12 +2267,11 @@ dht_dir_set_heal_xattr(xlator_t *this, dht_local_t *local, dict_t *dst, int luret = -1; int luflag = -1; int i = 0; + char **xattrs_to_heal; if (!src || !dst) { - gf_msg(this->name, GF_LOG_WARNING, EINVAL, DHT_MSG_DICT_SET_FAILED, - "src or dst is NULL. Failed to set " - " dictionary value for path %s", - local->loc.path); + gf_smsg(this->name, GF_LOG_WARNING, EINVAL, DHT_MSG_DST_NULL_SET_FAILED, + "path=%s", local->loc.path, NULL); return; } /* Check if any user xattr present in src dict and set @@ -2303,17 +2282,18 @@ dht_dir_set_heal_xattr(xlator_t *this, dht_local_t *local, dict_t *dst, and set it to dst dict, here index start from 1 because user xattr already checked in previous statement */ + + xattrs_to_heal = get_xattrs_to_heal(); + for (i = 1; xattrs_to_heal[i]; i++) { keyval = dict_get(src, xattrs_to_heal[i]); if (keyval) { luflag = 1; ret = dict_set(dst, xattrs_to_heal[i], keyval); if (ret) - gf_msg(this->name, GF_LOG_WARNING, ENOMEM, - DHT_MSG_DICT_SET_FAILED, - "Failed to set dictionary value:key = %s for " - "path %s", - xattrs_to_heal[i], local->loc.path); + gf_smsg(this->name, GF_LOG_WARNING, ENOMEM, + DHT_MSG_DICT_SET_FAILED, "key=%s", xattrs_to_heal[i], + "path=%s", local->loc.path, NULL); keyval = NULL; } } diff --git a/xlators/cluster/dht/src/dht-inode-read.c b/xlators/cluster/dht/src/dht-inode-read.c index 8007fc7..dbb8070 100644 --- a/xlators/cluster/dht/src/dht-inode-read.c +++ b/xlators/cluster/dht/src/dht-inode-read.c @@ -10,25 +10,25 @@ #include "dht-common.h" -int +static int dht_access2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret); -int +static int dht_readv2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret); -int +static int dht_attr2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret); -int +static int dht_open2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret); -int +static int dht_flush2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret); -int +static int dht_lk2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret); -int +static int dht_fsync2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret); -int +static int dht_common_xattrop2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret); -int +static int dht_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, fd_t *fd, dict_t *xdata) { @@ -67,7 +67,7 @@ out: return 0; } -int +static int dht_open2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -216,7 +216,7 @@ err: return 0; } -int +static int dht_attr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -258,7 +258,7 @@ out: return 0; } -int +static int dht_attr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, struct iatt *stbuf, dict_t *xdata) { @@ -473,7 +473,7 @@ out: return 0; } -int +static int dht_readv2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -558,7 +558,7 @@ err: return 0; } -int +static int dht_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, dict_t *xdata) { @@ -606,7 +606,7 @@ out: return 0; } -int +static int dht_access2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -735,7 +735,7 @@ out: return 0; } -int +static int dht_flush2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -881,7 +881,7 @@ out: return 0; } -int +static int dht_fsync2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -959,7 +959,7 @@ err: /* TODO: for 'lk()' call, we need some other special error, may be ESTALE to indicate that lock migration happened on the fd, so we can consider it as phase 2 of migration */ -int +static int dht_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, struct gf_flock *flock, dict_t *xdata) { @@ -1006,7 +1006,7 @@ out: return 0; } -int +static int dht_lk2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -1087,7 +1087,7 @@ err: return 0; } -int +static int dht_lease_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, struct gf_lease *lease, dict_t *xdata) { @@ -1129,7 +1129,7 @@ err: } /* Symlinks are currently not migrated, so no need for any check here */ -int +static int dht_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, const char *path, struct iatt *stbuf, dict_t *xdata) @@ -1305,7 +1305,7 @@ out: return 0; } -int +static int dht_common_xattrop2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { @@ -1364,7 +1364,7 @@ out: return 0; } -int +static int dht_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata) { @@ -1452,7 +1452,7 @@ err: return 0; } -int +static int dht_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata) { @@ -1525,7 +1525,7 @@ err: * below fops, hence not implementing 'migration' related checks */ -int +static int dht_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c index b6b349d..2f23ce9 100644 --- a/xlators/cluster/dht/src/dht-inode-write.c +++ b/xlators/cluster/dht/src/dht-inode-write.c @@ -10,17 +10,17 @@ #include "dht-common.h" -int +static int dht_writev2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret); -int +static int dht_truncate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret); -int +static int dht_setattr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret); -int +static int dht_fallocate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret); -int +static int dht_discard2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret); -int +static int dht_zerofill2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret); int @@ -93,30 +93,28 @@ dht_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, /* Check if the rebalance phase1 is true */ if (IS_DHT_MIGRATION_PHASE1(postbuf)) { - if (!dht_is_tier_xlator(this)) { + if (!local->xattr_req) { + local->xattr_req = dict_new(); if (!local->xattr_req) { - local->xattr_req = dict_new(); - if (!local->xattr_req) { - gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM, - "insufficient memory"); - local->op_errno = ENOMEM; - local->op_ret = -1; - goto out; - } - } - - ret = dict_set_uint32(local->xattr_req, - GF_PROTECT_FROM_EXTERNAL_WRITES, 1); - if (ret) { - gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_DICT_SET_FAILED, 0, - "Failed to set key %s in dictionary", - GF_PROTECT_FROM_EXTERNAL_WRITES); + gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM, + "insufficient memory"); local->op_errno = ENOMEM; local->op_ret = -1; goto out; } } + ret = dict_set_uint32(local->xattr_req, GF_PROTECT_FROM_EXTERNAL_WRITES, + 1); + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_DICT_SET_FAILED, 0, + "Failed to set key %s in dictionary", + GF_PROTECT_FROM_EXTERNAL_WRITES); + local->op_errno = ENOMEM; + local->op_ret = -1; + goto out; + } + dht_iatt_merge(this, &local->stbuf, postbuf); dht_iatt_merge(this, &local->prebuf, prebuf); @@ -142,7 +140,7 @@ out: return 0; } -int +static int dht_writev2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -336,7 +334,7 @@ err: return 0; } -int +static int dht_truncate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -555,7 +553,7 @@ err: return 0; } -int +static int dht_fallocate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -731,7 +729,7 @@ err: return 0; } -int +static int dht_discard2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -902,7 +900,7 @@ err: return 0; } -int +static int dht_zerofill2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; @@ -1049,7 +1047,7 @@ out: return 0; } -int +static int dht_setattr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret) { dht_local_t *local = NULL; diff --git a/xlators/cluster/dht/src/dht-layout.c b/xlators/cluster/dht/src/dht-layout.c index dc14d72..fda904c 100644 --- a/xlators/cluster/dht/src/dht-layout.c +++ b/xlators/cluster/dht/src/dht-layout.c @@ -131,9 +131,8 @@ dht_layout_search(xlator_t *this, dht_layout_t *layout, const char *name) ret = dht_hash_compute(this, layout->type, name, &hash); if (ret != 0) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_COMPUTE_HASH_FAILED, - "hash computation failed for type=%d name=%s", layout->type, - name); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_COMPUTE_HASH_FAILED, + "type=%d", layout->type, "name=%s", name, NULL); goto out; } @@ -145,8 +144,8 @@ dht_layout_search(xlator_t *this, dht_layout_t *layout, const char *name) } if (!subvol) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED, - "no subvolume for hash (value) = 0x%x", hash); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED, + "hash-value=0x%x", hash, NULL); } out: @@ -255,7 +254,7 @@ dht_disk_layout_extract_for_subvol(xlator_t *this, dht_layout_t *layout, return dht_disk_layout_extract(this, layout, i, disk_layout_p); } -int +static int dht_disk_layout_merge(xlator_t *this, dht_layout_t *layout, int pos, void *disk_layout_raw, int disk_layout_len) { @@ -266,8 +265,8 @@ dht_disk_layout_merge(xlator_t *this, dht_layout_t *layout, int pos, int disk_layout[4]; if (!disk_layout_raw) { - gf_msg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_LAYOUT_MERGE_FAILED, - "error no layout on disk for merge"); + gf_smsg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_LAYOUT_MERGE_FAILED, + NULL); return -1; } @@ -284,10 +283,8 @@ dht_disk_layout_merge(xlator_t *this, dht_layout_t *layout, int pos, case DHT_HASH_TYPE_DM: break; default: - gf_msg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_INVALID_DISK_LAYOUT, - "Invalid disk layout: " - "Catastrophic error layout with unknown type found %d", - disk_layout[1]); + gf_smsg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_INVALID_DISK_LAYOUT, + "layout=%d", disk_layout[1], NULL); return -1; } @@ -355,8 +352,8 @@ dht_layout_merge(xlator_t *this, dht_layout_t *layout, xlator_t *subvol, ret = dht_disk_layout_merge(this, layout, i, disk_layout_raw, disk_layout_len); if (ret != 0) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_MERGE_FAILED, - "layout merge from subvolume %s failed", subvol->name); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_MERGE_FAILED, + "subvolume=%s", subvol->name, NULL); goto out; } @@ -415,8 +412,7 @@ dht_layout_range_swap(dht_layout_t *layout, int i, int j) layout->list[j].start = start_swap; layout->list[j].stop = stop_swap; } - -int64_t +static int64_t dht_layout_entry_cmp_volname(dht_layout_t *layout, int i, int j) { return (strcmp(layout->list[i].xlator->name, layout->list[j].xlator->name)); @@ -439,7 +435,7 @@ dht_is_subvol_in_layout(dht_layout_t *layout, xlator_t *xlator) return _gf_false; } -int64_t +static int64_t dht_layout_entry_cmp(dht_layout_t *layout, int i, int j) { int64_t diff = 0; @@ -475,7 +471,7 @@ dht_layout_sort(dht_layout_t *layout) return 0; } -int +void dht_layout_sort_volname(dht_layout_t *layout) { int i = 0; @@ -491,8 +487,6 @@ dht_layout_sort_volname(dht_layout_t *layout) dht_layout_entry_swap(layout, i, j); } } - - return 0; } void @@ -625,8 +619,8 @@ dht_layout_normalize(xlator_t *this, loc_t *loc, dht_layout_t *layout) ret = dht_layout_sort(layout); if (ret == -1) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_SORT_FAILED, - "sort failed?! how the ...."); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_SORT_FAILED, + NULL); goto out; } @@ -642,10 +636,9 @@ dht_layout_normalize(xlator_t *this, loc_t *loc, dht_layout_t *layout) " gfid = %s", loc->path, gfid); } else { - gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_ANOMALIES_INFO, - "Found anomalies in %s (gfid = %s). " - "Holes=%d overlaps=%d", - loc->path, gfid, holes, overlaps); + gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_ANOMALIES_INFO, + "path=%s", loc->path, "gfid=%s", gfid, "holes=%d", holes, + "overlaps=%d", overlaps, NULL); } ret = -1; } @@ -712,12 +705,11 @@ dht_layout_dir_mismatch(xlator_t *this, dht_layout_t *layout, xlator_t *subvol, if (!xattr) { if (err == 0) { if (loc) { - gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_DICT_GET_FAILED, - "%s: xattr dictionary is NULL", loc->path); + gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_XATTR_DICT_NULL, + "path=%s", loc->path, NULL); } else { - gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_DICT_GET_FAILED, - "path not found: " - "xattr dictionary is NULL"); + gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_XATTR_DICT_NULL, + "path not found", NULL); } ret = -1; } @@ -729,13 +721,13 @@ dht_layout_dir_mismatch(xlator_t *this, dht_layout_t *layout, xlator_t *subvol, if (dict_ret < 0) { if (err == 0 && layout->list[pos].stop) { if (loc) { - gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_DISK_LAYOUT_MISSING, - "%s: Disk layout missing, gfid = %s", loc->path, gfid); + gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_DISK_LAYOUT_MISSING, + "path=%s", loc->path, "gfid=%s", gfid, NULL); } else { - gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_DISK_LAYOUT_MISSING, - "path not found: " - "Disk layout missing, gfid = %s", - gfid); + gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_DISK_LAYOUT_MISSING, + "path not found" + "gfid=%s", + gfid, NULL); } ret = -1; } @@ -751,12 +743,13 @@ dht_layout_dir_mismatch(xlator_t *this, dht_layout_t *layout, xlator_t *subvol, if ((layout->list[pos].start != start_off) || (layout->list[pos].stop != stop_off) || (layout->list[pos].commit_hash != commit_hash)) { - gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_INFO, - "subvol: %s; inode layout: 0x%x - 0x%x, 0x%x; " - "disk layout: 0x%x - 0x%x, 0x%x", - layout->list[pos].xlator->name, layout->list[pos].start, - layout->list[pos].stop, layout->list[pos].commit_hash, start_off, - stop_off, commit_hash); + gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_INFO, "subvol=%s", + layout->list[pos].xlator->name, "inode-layout:start=0x%x", + layout->list[pos].start, "inode-layout:stop=0x%x", + layout->list[pos].stop, "layout-commit-hash=0x%x; ", + layout->list[pos].commit_hash, "disk-layout:start-off=0x%x", + start_off, "disk-layout:top-off=0x%x", stop_off, + "commit-hash=0x%x", commit_hash, NULL); ret = 1; } else { ret = 0; @@ -778,9 +771,8 @@ dht_layout_preset(xlator_t *this, xlator_t *subvol, inode_t *inode) layout = dht_layout_for_subvol(this, subvol); if (!layout) { - gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_NO_LAYOUT_INFO, - "no pre-set layout for subvolume %s", - subvol ? subvol->name : "<nil>"); + gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_NO_LAYOUT_INFO, + "subvolume=%s", subvol ? subvol->name : "<nil>", NULL); ret = -1; goto out; } diff --git a/xlators/cluster/dht/src/dht-linkfile.c b/xlators/cluster/dht/src/dht-linkfile.c index 2f4e181..89ec6cc 100644 --- a/xlators/cluster/dht/src/dht-linkfile.c +++ b/xlators/cluster/dht/src/dht-linkfile.c @@ -11,7 +11,7 @@ #include <glusterfs/compat.h> #include "dht-common.h" -int +static int dht_linkfile_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf, dict_t *xattr, @@ -34,17 +34,16 @@ dht_linkfile_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, is_linkfile = check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name); if (!is_linkfile) - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NOT_LINK_FILE_ERROR, - "got non-linkfile %s:%s, gfid = %s", prev->name, local->loc.path, - gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NOT_LINK_FILE_ERROR, + "name=%s", prev->name, "path=%s", local->loc.path, "gfid=%s", + gfid, NULL); out: local->linkfile.linkfile_cbk(frame, cookie, this, op_ret, op_errno, inode, stbuf, postparent, postparent, xattr); return 0; } -#define is_equal(a, b) ((a) == (b)) -int +static int dht_linkfile_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf, struct iatt *preparent, @@ -73,9 +72,8 @@ dht_linkfile_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, goto out; ret = dict_set_uint32(xattrs, conf->link_xattr_name, 256); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, - "Failed to set dictionary value. key : %s", - conf->link_xattr_name); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, + "mame=%s", conf->link_xattr_name, NULL); goto out; } @@ -125,27 +123,23 @@ dht_linkfile_create(call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk, ret = dict_set_gfuuid(dict, "gfid-req", local->gfid, true); if (ret) - gf_msg("dht-linkfile", GF_LOG_INFO, 0, DHT_MSG_DICT_SET_FAILED, - "%s: Failed to set dictionary value: " - "key = gfid-req, gfid = %s ", - loc->path, gfid); + gf_smsg("dht-linkfile", GF_LOG_INFO, 0, DHT_MSG_DICT_SET_FAILED, + "path=%s", loc->path, "gfid=%s", gfid, NULL); } else { gf_uuid_unparse(loc->gfid, gfid); } ret = dict_set_str(dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes"); if (ret) - gf_msg("dht-linkfile", GF_LOG_INFO, 0, DHT_MSG_DICT_SET_FAILED, - "%s: Failed to set dictionary value: key = %s," - " gfid = %s", - loc->path, GLUSTERFS_INTERNAL_FOP_KEY, gfid); + gf_smsg("dht-linkfile", GF_LOG_INFO, 0, DHT_MSG_DICT_SET_FAILED, + "path=%s", loc->path, "key=%s", GLUSTERFS_INTERNAL_FOP_KEY, + "gfid=%s", gfid, NULL); ret = dict_set_str(dict, conf->link_xattr_name, tovol->name); if (ret < 0) { - gf_msg(frame->this->name, GF_LOG_INFO, 0, DHT_MSG_CREATE_LINK_FAILED, - "%s: failed to initialize linkfile data, gfid = %s", loc->path, - gfid); + gf_smsg(frame->this->name, GF_LOG_INFO, 0, DHT_MSG_CREATE_LINK_FAILED, + "path=%s", loc->path, "gfid=%s", gfid, NULL); goto out; } @@ -186,10 +180,9 @@ dht_linkfile_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, if (op_ret == -1) { gf_uuid_unparse(local->loc.gfid, gfid); - gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_UNLINK_FAILED, - "Unlinking linkfile %s (gfid = %s)on " - "subvolume %s failed ", - local->loc.path, gfid, subvol->name); + gf_smsg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_UNLINK_FAILED, + "path=%s", local->loc.path, "gfid=%s", gfid, "subvolume=%s", + subvol->name, NULL); } DHT_STACK_DESTROY(frame); @@ -257,7 +250,7 @@ out: return subvol; } -int +static int dht_linkfile_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, struct iatt *statpre, struct iatt *statpost, dict_t *xdata) @@ -269,10 +262,9 @@ dht_linkfile_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, loc = &local->loc; if (op_ret) - gf_msg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_SETATTR_FAILED, - "Failed to set attr uid/gid on %s" - " :<gfid:%s> ", - (loc->path ? loc->path : "NULL"), uuid_utoa(local->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_SETATTR_FAILED, + "path=%s", (loc->path ? loc->path : "NULL"), "gfid=%s", + uuid_utoa(local->gfid), NULL); DHT_STACK_DESTROY(frame); diff --git a/xlators/cluster/dht/src/dht-lock.c b/xlators/cluster/dht/src/dht-lock.c index f9bac4f..638821c 100644 --- a/xlators/cluster/dht/src/dht-lock.c +++ b/xlators/cluster/dht/src/dht-lock.c @@ -44,7 +44,8 @@ dht_log_lk_array(char *name, gf_loglevel_t log_level, dht_lock_t **lk_array, if (!lk_buf) goto out; - gf_msg(name, log_level, 0, DHT_MSG_LK_ARRAY_INFO, "%d. %s", i, lk_buf); + gf_smsg(name, log_level, 0, DHT_MSG_LK_ARRAY_INFO, "index=%d", i, + "lk_buf=%s", lk_buf, NULL); GF_FREE(lk_buf); } @@ -313,11 +314,9 @@ dht_unlock_entrylk_done(call_frame_t *frame, void *cookie, xlator_t *this, gfid); if (op_ret < 0) { - gf_msg(this->name, GF_LOG_WARNING, op_errno, - DHT_MSG_PARENT_LAYOUT_CHANGED, - "unlock failed on gfid: %s, stale lock might be left " - "in DHT_LAYOUT_HEAL_DOMAIN", - gfid); + gf_smsg(this->name, GF_LOG_WARNING, op_errno, + DHT_MSG_UNLOCK_GFID_FAILED, "gfid=%s", gfid, + "DHT_LAYOUT_HEAL_DOMAIN", NULL); } DHT_STACK_DESTROY(frame); @@ -339,9 +338,10 @@ dht_unlock_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, uuid_utoa_r(local->lock[0].ns.directory_ns.locks[lk_index]->loc.gfid, gfid); if (op_ret < 0) { - gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_UNLOCKING_FAILED, - "unlocking failed on %s:%s", - local->lock[0].ns.directory_ns.locks[lk_index]->xl->name, gfid); + gf_smsg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_UNLOCKING_FAILED, + "name=%s", + local->lock[0].ns.directory_ns.locks[lk_index]->xl->name, + "gfid=%s", gfid, NULL); } else { local->lock[0].ns.directory_ns.locks[lk_index]->locked = 0; } @@ -375,9 +375,9 @@ dht_unlock_entrylk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count, lock_frame = dht_lock_frame(frame); if (lock_frame == NULL) { - gf_msg(frame->this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED, - "cannot allocate a frame, not unlocking following " - "entrylks:"); + gf_smsg(frame->this->name, GF_LOG_WARNING, 0, + DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS, + NULL); dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count); goto done; @@ -385,9 +385,9 @@ dht_unlock_entrylk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count, ret = dht_local_entrylk_init(lock_frame, lk_array, lk_count, entrylk_cbk); if (ret < 0) { - gf_msg(frame->this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED, - "storing locks in local failed, not unlocking " - "following entrylks:"); + gf_smsg(frame->this->name, GF_LOG_WARNING, 0, + DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK, + NULL); dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count); @@ -446,21 +446,17 @@ dht_unlock_entrylk_wrapper(call_frame_t *frame, dht_elock_wrap_t *entrylk) lock_frame = copy_frame(frame); if (lock_frame == NULL) { - gf_msg(frame->this->name, GF_LOG_WARNING, ENOMEM, - DHT_MSG_PARENT_LAYOUT_CHANGED, - "mkdir (%s/%s) (path: %s): " - "copy frame failed", - pgfid, local->loc.name, local->loc.path); + gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM, + DHT_MSG_COPY_FRAME_FAILED, "pgfid=%s", pgfid, "name=%s", + local->loc.name, "path=%s", local->loc.path, NULL); goto done; } lock_local = dht_local_init(lock_frame, NULL, NULL, 0); if (lock_local == NULL) { - gf_msg(frame->this->name, GF_LOG_WARNING, ENOMEM, - DHT_MSG_PARENT_LAYOUT_CHANGED, - "mkdir (%s/%s) (path: %s): " - "local creation failed", - pgfid, local->loc.name, local->loc.path); + gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM, + DHT_MSG_CREATE_FAILED, "local", "pgfid=%s", pgfid, "name=%s", + local->loc.name, "path=%s", local->loc.path, NULL); goto done; } @@ -700,9 +696,10 @@ dht_unlock_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, uuid_utoa_r(local->lock[0].layout.my_layout.locks[lk_index]->loc.gfid, gfid); - gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_UNLOCKING_FAILED, - "unlocking failed on %s:%s", - local->lock[0].layout.my_layout.locks[lk_index]->xl->name, gfid); + gf_smsg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_UNLOCKING_FAILED, + "name=%s", + local->lock[0].layout.my_layout.locks[lk_index]->xl->name, + "gfid=%s", gfid, NULL); } else { local->lock[0].layout.my_layout.locks[lk_index]->locked = 0; } @@ -727,11 +724,9 @@ dht_unlock_inodelk_done(call_frame_t *frame, void *cookie, xlator_t *this, gfid); if (op_ret < 0) { - gf_msg(this->name, GF_LOG_WARNING, op_errno, - DHT_MSG_PARENT_LAYOUT_CHANGED, - "unlock failed on gfid: %s, stale lock might be left " - "in DHT_LAYOUT_HEAL_DOMAIN", - gfid); + gf_smsg(this->name, GF_LOG_WARNING, op_errno, + DHT_MSG_UNLOCK_GFID_FAILED, "DHT_LAYOUT_HEAL_DOMAIN gfid=%s", + gfid, NULL); } DHT_STACK_DESTROY(frame); @@ -762,9 +757,9 @@ dht_unlock_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count, lock_frame = dht_lock_frame(frame); if (lock_frame == NULL) { - gf_msg(frame->this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED, - "cannot allocate a frame, not unlocking following " - "locks:"); + gf_smsg(frame->this->name, GF_LOG_WARNING, 0, + DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS, + NULL); dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count); goto done; @@ -772,9 +767,9 @@ dht_unlock_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count, ret = dht_local_inodelk_init(lock_frame, lk_array, lk_count, inodelk_cbk); if (ret < 0) { - gf_msg(frame->this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED, - "storing locks in local failed, not unlocking " - "following locks:"); + gf_smsg(frame->this->name, GF_LOG_WARNING, 0, + DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK, + NULL); dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count); @@ -834,21 +829,17 @@ dht_unlock_inodelk_wrapper(call_frame_t *frame, dht_ilock_wrap_t *inodelk) lock_frame = copy_frame(frame); if (lock_frame == NULL) { - gf_msg(frame->this->name, GF_LOG_WARNING, ENOMEM, - DHT_MSG_PARENT_LAYOUT_CHANGED, - "mkdir (%s/%s) (path: %s): " - "copy frame failed", - pgfid, local->loc.name, local->loc.path); + gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM, + DHT_MSG_COPY_FRAME_FAILED, "pgfid=%s", pgfid, "name=%s", + local->loc.name, "path=%s", local->loc.path, NULL); goto done; } lock_local = dht_local_init(lock_frame, NULL, NULL, 0); if (lock_local == NULL) { - gf_msg(frame->this->name, GF_LOG_WARNING, ENOMEM, - DHT_MSG_PARENT_LAYOUT_CHANGED, - "mkdir (%s/%s) (path: %s): " - "local creation failed", - pgfid, local->loc.name, local->loc.path); + gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM, + DHT_MSG_CREATE_FAILED, "local", "gfid=%s", pgfid, "name=%s", + local->loc.name, "path=%s", local->loc.path, NULL); goto done; } @@ -1039,13 +1030,12 @@ dht_blocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, gfid); local->lock[0].layout.my_layout.op_ret = -1; local->lock[0].layout.my_layout.op_errno = op_errno; - gf_msg(this->name, GF_LOG_ERROR, op_errno, - DHT_MSG_INODELK_FAILED, - "inodelk failed on subvol %s. gfid:%s", - local->lock[0] - .layout.my_layout.locks[lk_index] - ->xl->name, - gfid); + gf_smsg(this->name, GF_LOG_ERROR, op_errno, + DHT_MSG_INODELK_FAILED, "subvol=%s", + local->lock[0] + .layout.my_layout.locks[lk_index] + ->xl->name, + "gfid=%s", gfid, NULL); goto cleanup; } break; @@ -1060,13 +1050,12 @@ dht_blocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, gfid); local->lock[0].layout.my_layout.op_ret = -1; local->lock[0].layout.my_layout.op_errno = op_errno; - gf_msg(this->name, GF_LOG_ERROR, op_errno, - DHT_MSG_INODELK_FAILED, - "inodelk failed on subvol %s. gfid:%s", - local->lock[0] - .layout.my_layout.locks[lk_index] - ->xl->name, - gfid); + gf_smsg(this->name, GF_LOG_ERROR, op_errno, + DHT_MSG_INODELK_FAILED, "subvol=%s", + local->lock[0] + .layout.my_layout.locks[lk_index] + ->xl->name, + "gfid=%s", gfid, NULL); goto cleanup; } break; @@ -1077,11 +1066,11 @@ dht_blocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, gfid); local->lock[0].layout.my_layout.op_ret = -1; local->lock[0].layout.my_layout.op_errno = op_errno; - gf_msg( + gf_smsg( this->name, GF_LOG_ERROR, op_errno, DHT_MSG_INODELK_FAILED, - "inodelk failed on subvol %s, gfid:%s", + "subvol=%s", local->lock[0].layout.my_layout.locks[lk_index]->xl->name, - gfid); + "gfid=%s", gfid, NULL); goto cleanup; } } @@ -1153,19 +1142,16 @@ dht_blocking_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count, lock_frame = dht_lock_frame(frame); if (lock_frame == NULL) { gf_uuid_unparse(tmp_local->loc.gfid, gfid); - gf_msg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_LOCK_FRAME_FAILED, - "memory allocation failed for lock_frame. gfid:%s" - " path:%s", - gfid, tmp_local->loc.path); + gf_smsg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_LOCK_FRAME_FAILED, + "gfid=%s", gfid, "path=%s", tmp_local->loc.path, NULL); goto out; } ret = dht_local_inodelk_init(lock_frame, lk_array, lk_count, inodelk_cbk); if (ret < 0) { gf_uuid_unparse(tmp_local->loc.gfid, gfid); - gf_msg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_LOCAL_LOCK_INIT_FAILED, - "dht_local_lock_init failed, gfid: %s path:%s", gfid, - tmp_local->loc.path); + gf_smsg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_LOCAL_LOCK_INIT_FAILED, + "gfid=%s", gfid, "path=%s", tmp_local->loc.path, NULL); goto out; } @@ -1246,11 +1232,10 @@ dht_blocking_entrylk_after_inodelk(call_frame_t *frame, void *cookie, if (ret < 0) { local->op_ret = -1; local->op_errno = EIO; - gf_msg(this->name, GF_LOG_WARNING, local->op_errno, - DHT_MSG_ENTRYLK_ERROR, - "%s (%s/%s): " - "dht_blocking_entrylk failed after taking inodelk", - gf_fop_list[local->fop], pgfid, entrylk->locks[0]->basename); + gf_smsg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_ENTRYLK_FAILED_AFT_INODELK, "fop=%s", + gf_fop_list[local->fop], "pgfid=%s", pgfid, "basename=%s", + entrylk->locks[0]->basename, NULL); goto err; } @@ -1310,10 +1295,9 @@ dht_protect_namespace(call_frame_t *frame, loc_t *loc, xlator_t *subvol, ret = dht_build_parent_loc(this, &parent, loc, &op_errno); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_LOC_FAILED, - "gfid:%s (name:%s) (path: %s): " - "parent loc build failed", - loc->gfid, loc->name, loc->path); + gf_smsg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_LOC_FAILED, + "gfid=%s", loc->gfid, "name=%s", loc->name, "path=%s", + loc->path, NULL); goto out; } gf_uuid_unparse(parent.gfid, pgfid); @@ -1322,10 +1306,10 @@ dht_protect_namespace(call_frame_t *frame, loc_t *loc, xlator_t *subvol, inodelk->locks = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer); if (inodelk->locks == NULL) { local->op_errno = ENOMEM; - gf_msg(this->name, GF_LOG_WARNING, local->op_errno, DHT_MSG_NO_MEMORY, - "%s (%s/%s) (path: %s): " - "calloc failure", - gf_fop_list[local->fop], pgfid, loc->name, loc->path); + gf_smsg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_CALLOC_FAILED, "fop=%s", gf_fop_list[local->fop], + "pgfid=%s", pgfid, "name=%s", loc->name, "path=%s", loc->path, + NULL); goto out; } @@ -1334,10 +1318,10 @@ dht_protect_namespace(call_frame_t *frame, loc_t *loc, xlator_t *subvol, FAIL_ON_ANY_ERROR); if (inodelk->locks[0] == NULL) { local->op_errno = ENOMEM; - gf_msg(this->name, GF_LOG_WARNING, local->op_errno, DHT_MSG_NO_MEMORY, - "%s (%s/%s) (path: %s): " - "inodelk: lock allocation failed", - gf_fop_list[local->fop], pgfid, loc->name, loc->path); + gf_smsg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_LOCK_ALLOC_FAILED, "inodelk-fop=%s", + gf_fop_list[local->fop], "pgfid=%s", pgfid, "name=%s", + loc->name, "path=%s", loc->path, NULL); goto err; } inodelk->lk_count = count; @@ -1346,10 +1330,10 @@ dht_protect_namespace(call_frame_t *frame, loc_t *loc, xlator_t *subvol, entrylk->locks = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer); if (entrylk->locks == NULL) { local->op_errno = ENOMEM; - gf_msg(this->name, GF_LOG_WARNING, local->op_errno, DHT_MSG_NO_MEMORY, - "%s (%s/%s) (path: %s): " - "entrylk: calloc failure", - gf_fop_list[local->fop], pgfid, loc->name, loc->path); + gf_smsg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_CALLOC_FAILED, "entrylk-fop=%s", + gf_fop_list[local->fop], "pgfid=%s", pgfid, "name=%s", + loc->name, "path=%s", loc->path, NULL); goto err; } @@ -1359,10 +1343,10 @@ dht_protect_namespace(call_frame_t *frame, loc_t *loc, xlator_t *subvol, FAIL_ON_ANY_ERROR); if (entrylk->locks[0] == NULL) { local->op_errno = ENOMEM; - gf_msg(this->name, GF_LOG_WARNING, local->op_errno, DHT_MSG_NO_MEMORY, - "%s (%s/%s) (path: %s): " - "entrylk: lock allocation failed", - gf_fop_list[local->fop], pgfid, loc->name, loc->path); + gf_smsg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_LOCK_ALLOC_FAILED, "entrylk-fop=%s", + gf_fop_list[local->fop], "pgfid=%s", pgfid, "name=%s", + loc->name, "path=%s", loc->path, NULL); goto err; } @@ -1376,11 +1360,11 @@ dht_protect_namespace(call_frame_t *frame, loc_t *loc, xlator_t *subvol, dht_blocking_entrylk_after_inodelk); if (ret < 0) { local->op_errno = EIO; - gf_msg(this->name, GF_LOG_WARNING, local->op_errno, - DHT_MSG_INODELK_ERROR, - "%s (%s/%s) (path: %s): " - "dht_blocking_inodelk failed", - gf_fop_list[local->fop], pgfid, loc->name, loc->path); + gf_smsg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_BLOCK_INODELK_FAILED, "fop=%s", gf_fop_list[local->fop], + "pgfid=%s", pgfid, "name=%s", loc->name, "path=%s", loc->path, + NULL); + goto err; } diff --git a/xlators/cluster/dht/src/dht-mem-types.h b/xlators/cluster/dht/src/dht-mem-types.h index 33f9832..e3c4471 100644 --- a/xlators/cluster/dht/src/dht-mem-types.h +++ b/xlators/cluster/dht/src/dht-mem-types.h @@ -30,10 +30,7 @@ enum gf_dht_mem_types_ { gf_dht_mt_container_t, gf_dht_mt_octx_t, gf_dht_mt_miginfo_t, - gf_tier_mt_bricklist_t, - gf_tier_mt_ipc_ctr_params_t, gf_dht_mt_fd_ctx_t, - gf_tier_mt_qfile_array_t, gf_dht_ret_cache_t, gf_dht_nodeuuids_t, gf_dht_mt_end diff --git a/xlators/cluster/dht/src/dht-messages.h b/xlators/cluster/dht/src/dht-messages.h index afc7460..601f8da 100644 --- a/xlators/cluster/dht/src/dht-messages.h +++ b/xlators/cluster/dht/src/dht-messages.h @@ -38,12 +38,11 @@ GLFS_MSGID( DHT_MSG_REBALANCE_STATUS, DHT_MSG_REBALANCE_STOPPED, DHT_MSG_RENAME_FAILED, DHT_MSG_SETATTR_FAILED, DHT_MSG_SUBVOL_INSUFF_INODES, DHT_MSG_SUBVOL_INSUFF_SPACE, DHT_MSG_UNLINK_FAILED, - DHT_MSG_LAYOUT_SET_FAILED, DHT_MSG_LOG_FIXED_LAYOUT, DHT_MSG_LOG_TIER_ERROR, - DHT_MSG_LOG_TIER_STATUS, DHT_MSG_GET_XATTR_FAILED, - DHT_MSG_FILE_LOOKUP_FAILED, DHT_MSG_OPEN_FD_FAILED, - DHT_MSG_SET_INODE_CTX_FAILED, DHT_MSG_UNLOCKING_FAILED, - DHT_MSG_DISK_LAYOUT_NULL, DHT_MSG_SUBVOL_INFO, DHT_MSG_CHUNK_SIZE_INFO, - DHT_MSG_LAYOUT_FORM_FAILED, DHT_MSG_SUBVOL_ERROR, + DHT_MSG_LAYOUT_SET_FAILED, DHT_MSG_LOG_FIXED_LAYOUT, + DHT_MSG_GET_XATTR_FAILED, DHT_MSG_FILE_LOOKUP_FAILED, + DHT_MSG_OPEN_FD_FAILED, DHT_MSG_SET_INODE_CTX_FAILED, + DHT_MSG_UNLOCKING_FAILED, DHT_MSG_DISK_LAYOUT_NULL, DHT_MSG_SUBVOL_INFO, + DHT_MSG_CHUNK_SIZE_INFO, DHT_MSG_LAYOUT_FORM_FAILED, DHT_MSG_SUBVOL_ERROR, DHT_MSG_LAYOUT_SORT_FAILED, DHT_MSG_REGEX_INFO, DHT_MSG_FOPEN_FAILED, DHT_MSG_SET_HOSTNAME_FAILED, DHT_MSG_BRICK_ERROR, DHT_MSG_SYNCOP_FAILED, DHT_MSG_MIGRATE_INFO, DHT_MSG_SOCKET_ERROR, DHT_MSG_CREATE_FD_FAILED, @@ -69,8 +68,7 @@ GLFS_MSGID( DHT_MSG_INIT_LOCAL_SUBVOL_FAILED, DHT_MSG_SYS_CALL_GET_TIME_FAILED, DHT_MSG_NO_DISK_USAGE_STATUS, DHT_MSG_SUBVOL_DOWN_ERROR, DHT_MSG_REBAL_THROTTLE_INFO, DHT_MSG_COMMIT_HASH_INFO, - DHT_MSG_REBAL_STRUCT_SET, DHT_MSG_HAS_MIGINFO, DHT_MSG_LOG_IPC_TIER_ERROR, - DHT_MSG_TIER_PAUSED, DHT_MSG_TIER_RESUME, DHT_MSG_SETTLE_HASH_FAILED, + DHT_MSG_REBAL_STRUCT_SET, DHT_MSG_HAS_MIGINFO, DHT_MSG_SETTLE_HASH_FAILED, DHT_MSG_DEFRAG_PROCESS_DIR_FAILED, DHT_MSG_FD_CTX_SET_FAILED, DHT_MSG_STALE_LOOKUP, DHT_MSG_PARENT_LAYOUT_CHANGED, DHT_MSG_LOCK_MIGRATION_FAILED, DHT_MSG_LOCK_INODE_UNREF_FAILED, @@ -79,6 +77,310 @@ GLFS_MSGID( DHT_MSG_ENTRYLK_ERROR, DHT_MSG_INODELK_ERROR, DHT_MSG_LOC_FAILED, DHT_MSG_UNKNOWN_FOP, DHT_MSG_MIGRATE_FILE_SKIPPED, DHT_MSG_DIR_XATTR_HEAL_FAILED, DHT_MSG_HASHED_SUBVOL_DOWN, - DHT_MSG_NON_HASHED_SUBVOL_DOWN); + DHT_MSG_NON_HASHED_SUBVOL_DOWN, DHT_MSG_SYNCTASK_CREATE_FAILED, + DHT_MSG_DIR_HEAL_ABORT, DHT_MSG_MIGRATE_SKIP, DHT_MSG_FD_CREATE_FAILED, + DHT_MSG_DICT_NEW_FAILED, DHT_MSG_FAILED_TO_OPEN, DHT_MSG_CREATE_FAILED, + DHT_MSG_FILE_NOT_EXIST, DHT_MSG_CHOWN_FAILED, DHT_MSG_FALLOCATE_FAILED, + DHT_MSG_FTRUNCATE_FAILED, DHT_MSG_STATFS_FAILED, DHT_MSG_WRITE_CROSS, + DHT_MSG_NEW_TARGET_FOUND, DHT_MSG_INSUFF_MEMORY, DHT_MSG_SET_XATTR_FAILED, + DHT_MSG_SET_MODE_FAILED, DHT_MSG_FILE_EXISTS_IN_DEST, + DHT_MSG_SYMLINK_FAILED, DHT_MSG_LINKFILE_DEL_FAILED, DHT_MSG_MKNOD_FAILED, + DHT_MSG_MIGRATE_CLEANUP_FAILED, DHT_MSG_LOCK_MIGRATE, + DHT_MSG_PARENT_BUILD_FAILED, DHT_MSG_HASHED_SUBVOL_NOT_FOUND, + DHT_MSG_ACQUIRE_ENTRYLK_FAILED, DHT_MSG_CREATE_DST_FAILED, + DHT_MSG_MIGRATION_EXIT, DHT_MSG_CHANGED_DST, DHT_MSG_TRACE_FAILED, + DHT_MSG_WRITE_LOCK_FAILED, DHT_MSG_GETACTIVELK_FAILED, DHT_MSG_STAT_FAILED, + DHT_MSG_UNLINK_PERFORM_FAILED, DHT_MSG_CLANUP_SOURCE_FILE_FAILED, + DHT_MSG_UNLOCK_FILE_FAILED, DHT_MSG_REMOVE_XATTR_FAILED, + DHT_MSG_DATA_MIGRATE_ABORT, DHT_MSG_DEFRAG_NULL, DHT_MSG_PARENT_NULL, + DHT_MSG_GFID_NOT_PRESENT, DHT_MSG_CHILD_LOC_FAILED, + DHT_MSG_SET_LOOKUP_FAILED, DHT_MSG_DIR_REMOVED, DHT_MSG_FIX_NOT_COMP, + DHT_MSG_SUBVOL_DETER_FAILED, DHT_MSG_LOCAL_SUBVOL, DHT_MSG_NODE_UUID, + DHT_MSG_SIZE_FILE, DHT_MSG_GET_DATA_SIZE_FAILED, + DHT_MSG_PTHREAD_JOIN_FAILED, DHT_MSG_COUNTER_THREAD_CREATE_FAILED, + DHT_MSG_MIGRATION_INIT_QUEUE_FAILED, DHT_MSG_PAUSED_TIMEOUT, DHT_MSG_WOKE, + DHT_MSG_ABORT_REBALANCE, DHT_MSG_CREATE_TASK_REBAL_FAILED, + DHT_MSG_REBAL_ESTIMATE_NOT_AVAIL, DHT_MSG_ADD_CHOICES_ERROR, + DHT_MSG_GET_CHOICES_ERROR, DHT_MSG_PREPARE_STATUS_ERROR, + DHT_MSG_SET_CHOICE_FAILED, DHT_MSG_SET_HASHED_SUBVOL_FAILED, + DHT_MSG_XATTR_HEAL_NOT_POSS, DHT_MSG_LINKTO_FILE_FAILED, + DHT_MSG_STALE_LINKFILE_DELETE, DHT_MSG_NO_SUBVOL_FOR_LINKTO, + DHT_MSG_SUBVOL_RETURNED, DHT_MSG_UNKNOWN_LOCAL_XSEL, DHT_MSG_GET_XATTR_ERR, + DHT_MSG_ALLOC_OR_FILL_FAILED, DHT_MSG_GET_REAL_NAME_FAILED, + DHT_MSG_COPY_UUID_FAILED, DHT_MSG_MDS_DETER_FAILED, + DHT_MSG_CREATE_REBAL_FAILED, DHT_MSG_LINK_LAYOUT_FAILED, + DHT_MSG_NO_SUBVOL_IN_LAYOUT, DHT_MSG_MEM_ALLOC_FAILED, + DHT_MSG_SET_IN_PARAMS_DICT_FAILED, DHT_MSG_LOC_COPY_FAILED, + DHT_MSG_PARENT_LOC_FAILED, DHT_MSG_CREATE_LOCK_FAILED, + DHT_MSG_PREV_ATTEMPT_FAILED, DHT_MSG_REFRESH_ATTEMPT, + DHT_MSG_ACQUIRE_LOCK_FAILED, DHT_MSG_CREATE_STUB_FAILED, + DHT_MSG_WIND_LOCK_REQ_FAILED, DHT_MSG_REFRESH_FAILED, + DHT_MSG_CACHED_SUBVOL_ERROR, DHT_MSG_NO_LINK_SUBVOL, DHT_MSG_SET_KEY_FAILED, + DHT_MSG_REMOVE_LINKTO_FAILED, DHT_MSG_LAYOUT_DICT_SET_FAILED, + DHT_MSG_XATTR_DICT_NULL, DHT_MSG_DUMMY_ALLOC_FAILED, DHT_MSG_DICT_IS_NULL, + DHT_MSG_LINK_INODE_FAILED, DHT_MSG_SELFHEAL_FAILED, DHT_MSG_NO_MDS_SUBVOL, + DHT_MSG_LIST_XATTRS_FAILED, DHT_MSG_RESET_INTER_XATTR_FAILED, + DHT_MSG_MDS_DOWN_UNABLE_TO_SET, DHT_MSG_WIND_UNLOCK_FAILED, + DHT_MSG_COMMIT_HASH_FAILED, DHT_MSG_UNLOCK_GFID_FAILED, + DHT_MSG_UNLOCK_FOLLOW_ENTRYLK, DHT_MSG_COPY_FRAME_FAILED, + DHT_MSG_UNLOCK_FOLLOW_LOCKS, DHT_MSG_ENTRYLK_FAILED_AFT_INODELK, + DHT_MSG_CALLOC_FAILED, DHT_MSG_LOCK_ALLOC_FAILED, + DHT_MSG_BLOCK_INODELK_FAILED, + DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK, + DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS, + DHT_MSG_DST_NULL_SET_FAILED); + +#define DHT_MSG_FD_CTX_SET_FAILED_STR "Failed to set fd ctx" +#define DHT_MSG_INVALID_VALUE_STR "Different dst found in the fd ctx" +#define DHT_MSG_UNKNOWN_FOP_STR "Unknown FOP on file" +#define DHT_MSG_OPEN_FD_ON_DST_FAILED_STR "Failed to open the fd on file" +#define DHT_MSG_SYNCTASK_CREATE_FAILED_STR "Failed to create synctask" +#define DHT_MSG_ASPRINTF_FAILED_STR \ + "asprintf failed while fetching subvol from the id" +#define DHT_MSG_HAS_MIGINFO_STR "Found miginfo in the inode ctx" +#define DHT_MSG_FILE_LOOKUP_FAILED_STR "failed to lookup the file" +#define DHT_MSG_INVALID_LINKFILE_STR \ + "linkto target is different from cached-subvol. treating as destination " \ + "subvol" +#define DHT_MSG_GFID_MISMATCH_STR "gfid different on the target file" +#define DHT_MSG_GET_XATTR_FAILED_STR "failed to get 'linkto' xattr" +#define DHT_MSG_SET_INODE_CTX_FAILED_STR "failed to set inode-ctx target file" +#define DHT_MSG_DIR_SELFHEAL_FAILED_STR "Healing of path failed" +#define DHT_MSG_DIR_HEAL_ABORT_STR \ + "Failed to get path from subvol. Aborting directory healing" +#define DHT_MSG_DIR_XATTR_HEAL_FAILED_STR "xattr heal failed for directory" +#define DHT_MSG_LOCK_INODE_UNREF_FAILED_STR \ + "Found a NULL inode. Failed to unref the inode" +#define DHT_MSG_DICT_SET_FAILED_STR "Failed to set dictionary value" +#define DHT_MSG_NOT_LINK_FILE_ERROR_STR "got non-linkfile" +#define DHT_MSG_CREATE_LINK_FAILED_STR "failed to initialize linkfile data" +#define DHT_MSG_UNLINK_FAILED_STR "Unlinking linkfile on subvolume failed" +#define DHT_MSG_MIGRATE_FILE_FAILED_STR "Migrate file failed" +#define DHT_MSG_NO_MEMORY_STR "could not allocate memory for dict" +#define DHT_MSG_SUBVOL_ERROR_STR "Failed to get linkto subvol" +#define DHT_MSG_MIGRATE_HARDLINK_FILE_FAILED_STR "link failed on subvol" +#define DHT_MSG_MIGRATE_FILE_SKIPPED_STR "Migration skipped" +#define DHT_MSG_FD_CREATE_FAILED_STR "fd create failed" +#define DHT_MSG_DICT_NEW_FAILED_STR "dict_new failed" +#define DHT_MSG_FAILED_TO_OPEN_STR "failed to open" +#define DHT_MSG_CREATE_FAILED_STR "failed to create" +#define DHT_MSG_FILE_NOT_EXIST_STR "file does not exist" +#define DHT_MSG_CHOWN_FAILED_STR "chown failed" +#define DHT_MSG_FALLOCATE_FAILED_STR "fallocate failed" +#define DHT_MSG_FTRUNCATE_FAILED_STR "ftruncate failed" +#define DHT_MSG_STATFS_FAILED_STR "failed to get statfs" +#define DHT_MSG_WRITE_CROSS_STR \ + "write will cross min-fre-disk for file on subvol. looking for new subvol" +#define DHT_MSG_SUBVOL_INSUFF_SPACE_STR \ + "Could not find any subvol with space accommodating the file. Cosider " \ + "adding bricks" +#define DHT_MSG_NEW_TARGET_FOUND_STR "New target found for file" +#define DHT_MSG_INSUFF_MEMORY_STR "insufficient memory" +#define DHT_MSG_SET_XATTR_FAILED_STR "failed to set xattr" +#define DHT_MSG_SET_MODE_FAILED_STR "failed to set mode" +#define DHT_MSG_FILE_EXISTS_IN_DEST_STR "file exists in destination" +#define DHT_MSG_LINKFILE_DEL_FAILED_STR "failed to delete the linkfile" +#define DHT_MSG_SYMLINK_FAILED_STR "symlink failed" +#define DHT_MSG_MKNOD_FAILED_STR "mknod failed" +#define DHT_MSG_SETATTR_FAILED_STR "failed to perform setattr" +#define DHT_MSG_MIGRATE_CLEANUP_FAILED_STR \ + "Migrate file cleanup failed: failed to fstat file" +#define DHT_MSG_LOCK_MIGRATE_STR "locks will be migrated for file" +#define DHT_MSG_PARENT_BUILD_FAILED_STR \ + "failed to build parent loc, which is needed to acquire entrylk to " \ + "synchronize with renames on this path. Skipping migration" +#define DHT_MSG_HASHED_SUBVOL_NOT_FOUND_STR \ + "cannot find hashed subvol which is needed to synchronize with renames " \ + "on this path. Skipping migration" +#define DHT_MSG_ACQUIRE_ENTRYLK_FAILED_STR "failed to acquire entrylk on subvol" +#define DHT_MSG_CREATE_DST_FAILED_STR "create dst failed for file" +#define DHT_MSG_MIGRATION_EXIT_STR "Exiting migration" +#define DHT_MSG_CHANGED_DST_STR "destination changed fo file" +#define DHT_MSG_TRACE_FAILED_STR "Trace failed" +#define DHT_MSG_WRITE_LOCK_FAILED_STR "write lock failed" +#define DHT_MSG_GETACTIVELK_FAILED_STR "getactivelk failed for file" +#define DHT_MSG_STAT_FAILED_STR "failed to do a stat" +#define DHT_MSG_UNLINK_PERFORM_FAILED_STR "failed to perform unlink" +#define DHT_MSG_MIGRATE_FILE_COMPLETE_STR "completed migration" +#define DHT_MSG_CLANUP_SOURCE_FILE_FAILED_STR "failed to cleanup source file" +#define DHT_MSG_UNLOCK_FILE_FAILED_STR "failed to unlock file" +#define DHT_MSG_REMOVE_XATTR_FAILED_STR "remove xattr failed" +#define DHT_MSG_SOCKET_ERROR_STR "Failed to unlink listener socket" +#define DHT_MSG_HASHED_SUBVOL_GET_FAILED_STR "Failed to get hashed subvolume" +#define DHT_MSG_CACHED_SUBVOL_GET_FAILED_STR "Failed to get cached subvolume" +#define DHT_MSG_MIGRATE_DATA_FAILED_STR "migrate-data failed" +#define DHT_MSG_DEFRAG_NULL_STR "defrag is NULL" +#define DHT_MSG_DATA_MIGRATE_ABORT_STR \ + "Readdirp failed. Aborting data migration for dict" +#define DHT_MSG_LAYOUT_FIX_FAILED_STR "fix layout failed" +#define DHT_MSG_PARENT_NULL_STR "parent is NULL" +#define DHT_MSG_GFID_NOT_PRESENT_STR "gfid not present" +#define DHT_MSG_CHILD_LOC_FAILED_STR "Child loc build failed" +#define DHT_MSG_SET_LOOKUP_FAILED_STR "Failed to set lookup" +#define DHT_MSG_DIR_LOOKUP_FAILED_STR "lookup failed" +#define DHT_MSG_DIR_REMOVED_STR "Dir renamed or removed. Skipping" +#define DHT_MSG_READDIR_ERROR_STR "readdir failed, Aborting fix-layout" +#define DHT_MSG_SETTLE_HASH_FAILED_STR "Settle hash failed" +#define DHT_MSG_DEFRAG_PROCESS_DIR_FAILED_STR "gf_defrag_process_dir failed" +#define DHT_MSG_FIX_NOT_COMP_STR \ + "Unable to retrieve fixlayout xattr. Assume background fix layout not " \ + "complete" +#define DHT_MSG_SUBVOL_DETER_FAILED_STR \ + "local subvolume determination failed with error" +#define DHT_MSG_LOCAL_SUBVOL_STR "local subvol" +#define DHT_MSG_NODE_UUID_STR "node uuid" +#define DHT_MSG_SIZE_FILE_STR "Total size files" +#define DHT_MSG_GET_DATA_SIZE_FAILED_STR \ + "Failed to get the total data size. Unable to estimate time to complete " \ + "rebalance" +#define DHT_MSG_PTHREAD_JOIN_FAILED_STR \ + "file_counter_thread: pthread_join failed" +#define DHT_MSG_COUNTER_THREAD_CREATE_FAILED_STR \ + "Failed to create the file counter thread" +#define DHT_MSG_MIGRATION_INIT_QUEUE_FAILED_STR \ + "Failed to initialise migration queue" +#define DHT_MSG_REBALANCE_STOPPED_STR "Received stop command on rebalance" +#define DHT_MSG_PAUSED_TIMEOUT_STR "Request pause timer timeout" +#define DHT_MSG_WOKE_STR "woken" +#define DHT_MSG_ABORT_REBALANCE_STR "Aborting rebalance" +#define DHT_MSG_REBALANCE_START_FAILED_STR \ + "Failed to start rebalance: look up on / failed" +#define DHT_MSG_CREATE_TASK_REBAL_FAILED_STR \ + "Could not create task for rebalance" +#define DHT_MSG_REBAL_ESTIMATE_NOT_AVAIL_STR \ + "Rebalance estimates will not be available" +#define DHT_MSG_REBALANCE_STATUS_STR "Rebalance status" +#define DHT_MSG_DATA_NULL_STR "data value is NULL" +#define DHT_MSG_ADD_CHOICES_ERROR_STR "Error to add choices in buffer" +#define DHT_MSG_GET_CHOICES_ERROR_STR "Error to get choices" +#define DHT_MSG_PREPARE_STATUS_ERROR_STR "Error to prepare status" +#define DHT_MSG_SET_CHOICE_FAILED_STR "Failed to set full choice" +#define DHT_MSG_AGGREGATE_QUOTA_XATTR_FAILED_STR \ + "Failed to aggregate quota xattr" +#define DHT_MSG_FILE_TYPE_MISMATCH_STR \ + "path exists as a file on one subvolume and directory on another. Please " \ + "fix it manually" +#define DHT_MSG_LAYOUT_SET_FAILED_STR "failed to set layout for subvolume" +#define DHT_MSG_LAYOUT_MERGE_FAILED_STR "failed to merge layouts for subvolume" +#define DHT_MSG_SET_HASHED_SUBVOL_FAILED_STR "Failed to set hashed subvolume" +#define DHT_MSG_XATTR_HEAL_NOT_POSS_STR \ + "No gfid exists for path. so healing xattr is not possible" +#define DHT_MSG_REVALIDATE_CBK_INFO_STR "Revalidate: subvolume returned -1" +#define DHT_MSG_LAYOUT_MISMATCH_STR "Mismatching layouts" +#define DHT_MSG_UNLINK_LOOKUP_INFO_STR "lookup_unlink retuened" +#define DHT_MSG_LINKTO_FILE_FAILED_STR \ + "Could not unlink the linkto file as either fd is open and/or linkto " \ + "xattr is set" +#define DHT_MSG_LAYOUT_PRESET_FAILED_STR \ + "Could not set pre-set layout for subvolume" +#define DHT_MSG_FILE_ON_MULT_SUBVOL_STR \ + "multiple subvolumes have file (preferably rename the file in the " \ + "backend, and do a fresh lookup" +#define DHT_MSG_STALE_LINKFILE_DELETE_STR \ + "attempting deletion of stale linkfile" +#define DHT_MSG_LINK_FILE_LOOKUP_INFO_STR "Lookup on following linkfile" +#define DHT_MSG_NO_SUBVOL_FOR_LINKTO_STR "No link subvolume for linkto" +#define DHT_MSG_SUBVOL_RETURNED_STR "Subvolume returned -1" +#define DHT_MSG_UNKNOWN_LOCAL_XSEL_STR "Unknown local->xsel" +#define DHT_MSG_DICT_GET_FAILED_STR "Failed to get" +#define DHT_MSG_UUID_PARSE_ERROR_STR "Failed to parse uuid" +#define DHT_MSG_GET_XATTR_ERR_STR "getxattr err for dir" +#define DHT_MSG_ALLOC_OR_FILL_FAILED_STR "alloc or fill failed" +#define DHT_MSG_UPGRADE_BRICKS_STR \ + "At least one of the bricks does not support this operation. Please " \ + "upgrade all bricks" +#define DHT_MSG_GET_REAL_NAME_FAILED_STR "Failed to get real filename" +#define DHT_MSG_LAYOUT_NULL_STR "Layout is NULL" +#define DHT_MSG_COPY_UUID_FAILED_STR "Failed to copy node uuid key" +#define DHT_MSG_MDS_DETER_FAILED_STR \ + "Cannot determine MDS, fetching xattr randomly from a subvol" +#define DHT_MSG_HASHED_SUBVOL_DOWN_STR \ + "MDS is down for path, so fetching xattr randomly from subvol" +#define DHT_MSG_CREATE_REBAL_FAILED_STR \ + "failed to create a new rebalance synctask" +#define DHT_MSG_FIX_LAYOUT_INFO_STR "fixing the layout" +#define DHT_MSG_OPERATION_NOT_SUP_STR "wrong directory-spread-count value" +#define DHT_MSG_LINK_LAYOUT_FAILED_STR "failed to link the layout in inode" +#define DHT_MSG_NO_SUBVOL_IN_LAYOUT_STR "no subvolume in layout for path" +#define DHT_MSG_INODE_LK_ERROR_STR "mknod lock failed for file" +#define DHT_MSG_MEM_ALLOC_FAILED_STR "mem allocation failed" +#define DHT_MSG_PARENT_LAYOUT_CHANGED_STR \ + "extracting in-memory layout of parent failed" +#define DHT_MSG_SET_IN_PARAMS_DICT_FAILED_STR \ + "setting in params dictionary failed" +#define DHT_MSG_LOC_COPY_FAILED_STR "loc_copy failed" +#define DHT_MSG_LOC_FAILED_STR "parent loc build failed" +#define DHT_MSG_PARENT_LOC_FAILED_STR "locking parent failed" +#define DHT_MSG_CREATE_LOCK_FAILED_STR "Create lock failed" +#define DHT_MSG_PREV_ATTEMPT_FAILED_STR \ + "mkdir loop detected. parent layout didn't change even though previous " \ + "attempt of mkdir failed because of in-memory layout not matching with " \ + "that on disk." +#define DHT_MSG_REFRESH_ATTEMPT_STR \ + "mkdir parent layout changed. Attempting a refresh and then a retry" +#define DHT_MSG_ACQUIRE_LOCK_FAILED_STR \ + "Acquiring lock on parent to guard against layout-change failed" +#define DHT_MSG_CREATE_STUB_FAILED_STR "creating stub failed" +#define DHT_MSG_WIND_LOCK_REQ_FAILED_STR \ + "cannot wind lock request to guard parent layout" +#define DHT_MSG_REFRESH_FAILED_STR "refreshing parent layout failed." +#define DHT_MSG_CACHED_SUBVOL_ERROR_STR "On cached subvol" +#define DHT_MSG_NO_LINK_SUBVOL_STR "Linkfile does not have link subvolume" +#define DHT_MSG_SET_KEY_FAILED_STR "failed to set key" +#define DHT_MSG_CHILD_DOWN_STR "Received CHILD_DOWN. Exiting" +#define DHT_MSG_LOG_FIXED_LAYOUT_STR "log layout fixed" +#define DHT_MSG_REBAL_STRUCT_SET_STR "local->rebalance already set" +#define DHT_MSG_REMOVE_LINKTO_FAILED_STR "Removal of linkto failed at subvol" +#define DHT_MSG_LAYOUT_DICT_SET_FAILED_STR "dht layout dict set failed" +#define DHT_MSG_SUBVOL_INFO_STR "creating subvolume" +#define DHT_MSG_COMPUTE_HASH_FAILED_STR "hash computation failed" +#define DHT_MSG_INVALID_DISK_LAYOUT_STR \ + "Invalid disk layout: Catastrophic error layout with unknown type found" +#define DHT_MSG_LAYOUT_SORT_FAILED_STR "layout sort failed" +#define DHT_MSG_ANOMALIES_INFO_STR "Found anomalies" +#define DHT_MSG_XATTR_DICT_NULL_STR "xattr dictionary is NULL" +#define DHT_MSG_DISK_LAYOUT_MISSING_STR "Disk layout missing" +#define DHT_MSG_LAYOUT_INFO_STR "layout info" +#define DHT_MSG_SUBVOL_NO_LAYOUT_INFO_STR "no pre-set layout for subvol" +#define DHT_MSG_SELFHEAL_XATTR_FAILED_STR "layout setxattr failed" +#define DHT_MSG_DIR_SELFHEAL_XATTR_FAILED_STR "Directory self heal xattr failed" +#define DHT_MSG_DUMMY_ALLOC_FAILED_STR "failed to allocate dummy layout" +#define DHT_MSG_DICT_IS_NULL_STR \ + "dict is NULL, need to make sure gfids are same" +#define DHT_MSG_ENTRYLK_ERROR_STR "acquiring entrylk after inodelk failed" +#define DHT_MSG_NO_DISK_USAGE_STATUS_STR "no du stats" +#define DHT_MSG_LINK_INODE_FAILED_STR "linking inode failed" +#define DHT_MSG_SELFHEAL_FAILED_STR "Directory selfheal failed" +#define DHT_MSG_NO_MDS_SUBVOL_STR "No mds subvol" +#define DHT_MSG_LIST_XATTRS_FAILED_STR "failed to list xattrs" +#define DHT_MSG_RESET_INTER_XATTR_FAILED_STR "Failed to reset internal xattr" +#define DHT_MSG_MDS_DOWN_UNABLE_TO_SET_STR \ + "mds subvol is down, unable to set xattr" +#define DHT_MSG_DIR_ATTR_HEAL_FAILED_STR \ + "Directory attr heal failed. Failed to set uid/gid" +#define DHT_MSG_WIND_UNLOCK_FAILED_STR \ + "Winding unlock failed: stale locks left on brick" +#define DHT_MSG_COMMIT_HASH_FAILED_STR "Directory commit hash updaten failed" +#define DHT_MSG_LK_ARRAY_INFO_STR "lk info" +#define DHT_MSG_UNLOCK_GFID_FAILED_STR \ + "unlock failed on gfid: stale lock might be left" +#define DHT_MSG_UNLOCKING_FAILED_STR "unlocking failed" +#define DHT_MSG_UNLOCK_FOLLOW_ENTRYLK_STR "not unlocking following entrylks" +#define DHT_MSG_COPY_FRAME_FAILED_STR "copy frame failed" +#define DHT_MSG_UNLOCK_FOLLOW_LOCKS_STR "not unlocking following locks" +#define DHT_MSG_INODELK_FAILED_STR "inodelk failed on subvol" +#define DHT_MSG_LOCK_FRAME_FAILED_STR "memory allocation failed for lock_frame" +#define DHT_MSG_LOCAL_LOCK_INIT_FAILED_STR "dht_local_lock_init failed" +#define DHT_MSG_ENTRYLK_FAILED_AFT_INODELK_STR \ + "dht_blocking_entrylk failed after taking inodelk" +#define DHT_MSG_BLOCK_INODELK_FAILED_STR "dht_blocking_inodelk failed" +#define DHT_MSG_CALLOC_FAILED_STR "calloc failed" +#define DHT_MSG_LOCK_ALLOC_FAILED_STR "lock allocation failed" +#define DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS_STR \ + "cannot allocate a frame, not unlocking following entrylks" +#define DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK_STR \ + "storing locks in local failed, not unlocking following entrylks" +#define DHT_MSG_DST_NULL_SET_FAILED_STR \ + "src or dst is NULL, Failed to set dictionary value" #endif /* _DHT_MESSAGES_H_ */ diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c index 0d68ed5..8ba8082 100644 --- a/xlators/cluster/dht/src/dht-rebalance.c +++ b/xlators/cluster/dht/src/dht-rebalance.c @@ -16,8 +16,8 @@ #include "glusterfs/compat-errno.h" // for ENODATA on BSD #define GF_DISK_SECTOR_SIZE 512 -#define DHT_REBALANCE_PID 4242 /* Change it if required */ -#define DHT_REBALANCE_BLKSIZE (1024 * 1024) /* 1 MB */ +#define DHT_REBALANCE_PID 4242 /* Change it if required */ +#define DHT_REBALANCE_BLKSIZE 1048576 /* 1 MB */ #define MAX_MIGRATE_QUEUE_COUNT 500 #define MIN_MIGRATE_QUEUE_COUNT 200 #define MAX_REBAL_TYPE_SIZE 16 @@ -45,7 +45,10 @@ gf_defrag_free_dir_dfmeta(struct dir_dfmeta *meta, int local_subvols_cnt) if (meta) { for (i = 0; i < local_subvols_cnt; i++) { - gf_dirent_free(&meta->equeue[i]); + if (meta->equeue) + gf_dirent_free(&meta->equeue[i]); + if (meta->lfd && meta->lfd[i]) + fd_unref(meta->lfd[i]); } GF_FREE(meta->equeue); @@ -53,6 +56,7 @@ gf_defrag_free_dir_dfmeta(struct dir_dfmeta *meta, int local_subvols_cnt) GF_FREE(meta->iterator); GF_FREE(meta->offset_var); GF_FREE(meta->fetch_entries); + GF_FREE(meta->lfd); GF_FREE(meta); } } @@ -84,26 +88,6 @@ dht_set_global_defrag_error(gf_defrag_info_t *defrag, int ret) return; } -static gf_boolean_t -dht_is_tier_command(int cmd) -{ - gf_boolean_t is_tier = _gf_false; - - switch (cmd) { - case GF_DEFRAG_CMD_START_TIER: - case GF_DEFRAG_CMD_STATUS_TIER: - case GF_DEFRAG_CMD_START_DETACH_TIER: - case GF_DEFRAG_CMD_STOP_DETACH_TIER: - case GF_DEFRAG_CMD_PAUSE_TIER: - case GF_DEFRAG_CMD_RESUME_TIER: - is_tier = _gf_true; - break; - default: - break; - } - return is_tier; -} - static int dht_send_rebalance_event(xlator_t *this, int cmd, gf_defrag_status_t status) { @@ -112,8 +96,6 @@ dht_send_rebalance_event(xlator_t *this, int cmd, gf_defrag_status_t status) char *tmpstr = NULL; char *ptr = NULL; char *suffix = "-dht"; - dht_conf_t *conf = NULL; - gf_defrag_info_t *defrag = NULL; int len = 0; eventtypes_t event = EVENT_LAST; @@ -132,21 +114,14 @@ dht_send_rebalance_event(xlator_t *this, int cmd, gf_defrag_status_t status) break; } - if (dht_is_tier_command(cmd)) { - /* We should have the tier volume name*/ - conf = this->private; - defrag = conf->defrag; - volname = defrag->tier_conf.volname; - } else { - /* DHT volume */ - len = strlen(this->name) - strlen(suffix); - tmpstr = gf_strdup(this->name); - if (tmpstr) { - ptr = tmpstr + len; - if (!strcmp(ptr, suffix)) { - tmpstr[len] = '\0'; - volname = tmpstr; - } + /* DHT volume */ + len = strlen(this->name) - strlen(suffix); + tmpstr = gf_strdup(this->name); + if (tmpstr) { + ptr = tmpstr + len; + if (!strcmp(ptr, suffix)) { + tmpstr[len] = '\0'; + volname = tmpstr; } } @@ -172,75 +147,6 @@ dht_strip_out_acls(dict_t *dict) } } -static int -dht_write_with_holes(xlator_t *to, fd_t *fd, struct iovec *vec, int count, - int32_t size, off_t offset, struct iobref *iobref, - int *fop_errno) -{ - int i = 0; - int ret = -1; - int start_idx = 0; - int tmp_offset = 0; - int write_needed = 0; - int buf_len = 0; - int size_pending = 0; - char *buf = NULL; - - /* loop through each vector */ - for (i = 0; i < count; i++) { - buf = vec[i].iov_base; - buf_len = vec[i].iov_len; - - for (start_idx = 0; (start_idx + GF_DISK_SECTOR_SIZE) <= buf_len; - start_idx += GF_DISK_SECTOR_SIZE) { - if (mem_0filled(buf + start_idx, GF_DISK_SECTOR_SIZE) != 0) { - write_needed = 1; - continue; - } - - if (write_needed) { - ret = syncop_write( - to, fd, (buf + tmp_offset), (start_idx - tmp_offset), - (offset + tmp_offset), iobref, 0, NULL, NULL); - /* 'path' will be logged in calling function */ - if (ret < 0) { - gf_log(THIS->name, GF_LOG_WARNING, "failed to write (%s)", - strerror(-ret)); - *fop_errno = -ret; - ret = -1; - goto out; - } - - write_needed = 0; - } - tmp_offset = start_idx + GF_DISK_SECTOR_SIZE; - } - - if ((start_idx < buf_len) || write_needed) { - /* This means, last chunk is not yet written.. write it */ - ret = syncop_write(to, fd, (buf + tmp_offset), - (buf_len - tmp_offset), (offset + tmp_offset), - iobref, 0, NULL, NULL); - if (ret < 0) { - /* 'path' will be logged in calling function */ - gf_log(THIS->name, GF_LOG_WARNING, "failed to write (%s)", - strerror(-ret)); - *fop_errno = -ret; - ret = -1; - goto out; - } - } - - size_pending = (size - buf_len); - if (!size_pending) - break; - } - - ret = size; -out: - return ret; -} - /* return values: -1 : failure @@ -648,7 +554,7 @@ out: static int __dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from, loc_t *loc, struct iatt *stbuf, fd_t **dst_fd, - int *fop_errno) + int *fop_errno, int file_has_holes) { int ret = -1; int ret2 = -1; @@ -703,26 +609,23 @@ __dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from, goto out; } - if (!!dht_is_tier_xlator(this)) { - xdata = dict_new(); - if (!xdata) { - *fop_errno = ENOMEM; - ret = -1; - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, - DHT_MSG_MIGRATE_FILE_FAILED, "%s: dict_new failed)", - loc->path); - goto out; - } + xdata = dict_new(); + if (!xdata) { + *fop_errno = ENOMEM; + ret = -1; + gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MIGRATE_FILE_FAILED, + "%s: dict_new failed)", loc->path); + goto out; + } - ret = dict_set_int32(xdata, GF_CLEAN_WRITE_PROTECTION, 1); - if (ret) { - *fop_errno = ENOMEM; - ret = -1; - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, - "%s: failed to set dictionary value: key = %s ", loc->path, - GF_CLEAN_WRITE_PROTECTION); - goto out; - } + ret = dict_set_int32_sizen(xdata, GF_CLEAN_WRITE_PROTECTION, 1); + if (ret) { + *fop_errno = ENOMEM; + ret = -1; + gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, + "%s: failed to set dictionary value: key = %s ", loc->path, + GF_CLEAN_WRITE_PROTECTION); + goto out; } ret = syncop_lookup(to, loc, &new_stbuf, NULL, xdata, NULL); @@ -817,7 +720,7 @@ __dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from, /* No need to bother about 0 byte size files */ if (stbuf->ia_size > 0) { - if (conf->use_fallocate) { + if (conf->use_fallocate && !file_has_holes) { ret = syncop_fallocate(to, fd, 0, 0, stbuf->ia_size, NULL, NULL); if (ret < 0) { if (ret == -EOPNOTSUPP || ret == -EINVAL || ret == -ENOSYS) { @@ -844,9 +747,7 @@ __dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from, goto out; } } - } - - if (!conf->use_fallocate) { + } else { ret = syncop_ftruncate(to, fd, stbuf->ia_size, NULL, NULL, NULL, NULL); if (ret < 0) { @@ -1097,22 +998,90 @@ __dht_rebalance_migrate_data(xlator_t *this, gf_defrag_info_t *defrag, int ret = 0; int count = 0; off_t offset = 0; + off_t data_offset = 0; + off_t hole_offset = 0; struct iovec *vector = NULL; struct iobref *iobref = NULL; uint64_t total = 0; size_t read_size = 0; + size_t data_block_size = 0; dict_t *xdata = NULL; dht_conf_t *conf = NULL; conf = this->private; + /* if file size is '0', no need to enter this loop */ while (total < ia_size) { - read_size = (((ia_size - total) > DHT_REBALANCE_BLKSIZE) - ? DHT_REBALANCE_BLKSIZE - : (ia_size - total)); + /* This is a regular file - read it sequentially */ + if (!hole_exists) { + read_size = (((ia_size - total) > DHT_REBALANCE_BLKSIZE) + ? DHT_REBALANCE_BLKSIZE + : (ia_size - total)); + } else { + /* This is a sparse file - read only the data segments in the file + */ + + /* If the previous data block is fully copied, find the next data + * segment + * starting at the offset of the last read and written byte, */ + if (data_block_size <= 0) { + ret = syncop_seek(from, src, offset, GF_SEEK_DATA, NULL, + &data_offset); + if (ret) { + if (ret == -ENXIO) + ret = 0; /* No more data segments */ + else + *fop_errno = -ret; /* Error occurred */ + + break; + } + + /* If the position of the current data segment is greater than + * the position of the next hole, find the next hole in order to + * calculate the length of the new data segment */ + if (data_offset > hole_offset) { + /* Starting at the offset of the last data segment, find the + * next hole */ + ret = syncop_seek(from, src, data_offset, GF_SEEK_HOLE, + NULL, &hole_offset); + if (ret) { + /* If an error occurred here it's a real error because + * if the seek for a data segment was successful then + * necessarily another hole must exist (EOF is a hole) + */ + *fop_errno = -ret; + break; + } + + /* Calculate the total size of the current data block */ + data_block_size = hole_offset - data_offset; + } + } else { + /* There is still data in the current segment, move the + * data_offset to the position of the last written byte */ + data_offset = offset; + } + + /* Calculate how much data needs to be read and written. If the data + * segment's length is bigger than DHT_REBALANCE_BLKSIZE, read and + * write DHT_REBALANCE_BLKSIZE data length and the rest in the + * next iteration(s) */ + read_size = ((data_block_size > DHT_REBALANCE_BLKSIZE) + ? DHT_REBALANCE_BLKSIZE + : data_block_size); + + /* Calculate the remaining size of the data block - maybe there's no + * need to seek for data in the next iteration */ + data_block_size -= read_size; + + /* Set offset to the offset of the data segment so read and write + * will have the correct position */ + offset = data_offset; + } ret = syncop_readv(from, src, read_size, offset, 0, &vector, &count, &iobref, NULL, NULL, NULL); + if (!ret || (ret < 0)) { if (!ret) { /* File was probably truncated*/ @@ -1124,57 +1093,42 @@ __dht_rebalance_migrate_data(xlator_t *this, gf_defrag_info_t *defrag, break; } - if (hole_exists) { - ret = dht_write_with_holes(to, dst, vector, count, ret, offset, - iobref, fop_errno); - } else { - if (!conf->force_migration && !dht_is_tier_xlator(this)) { + if (!conf->force_migration) { + if (!xdata) { + xdata = dict_new(); if (!xdata) { - xdata = dict_new(); - if (!xdata) { - gf_msg("dht", GF_LOG_ERROR, 0, - DHT_MSG_MIGRATE_FILE_FAILED, - "insufficient memory"); - ret = -1; - *fop_errno = ENOMEM; - break; - } + gf_msg("dht", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED, + "insufficient memory"); + ret = -1; + *fop_errno = ENOMEM; + break; + } - /* Fail this write and abort rebalance if we - * detect a write from client since migration of - * this file started. This is done to avoid - * potential data corruption due to out of order - * writes from rebalance and client to the same - * region (as compared between src and dst - * files). See - * https://github.com/gluster/glusterfs/issues/308 - * for more details. - */ - ret = dict_set_int32(xdata, GF_AVOID_OVERWRITE, 1); - if (ret) { - gf_msg("dht", GF_LOG_ERROR, 0, ENOMEM, - "failed to set dict"); - ret = -1; - *fop_errno = ENOMEM; - break; - } + /* Fail this write and abort rebalance if we + * detect a write from client since migration of + * this file started. This is done to avoid + * potential data corruption due to out of order + * writes from rebalance and client to the same + * region (as compared between src and dst + * files). See + * https://github.com/gluster/glusterfs/issues/308 + * for more details. + */ + ret = dict_set_int32_sizen(xdata, GF_AVOID_OVERWRITE, 1); + if (ret) { + gf_msg("dht", GF_LOG_ERROR, 0, ENOMEM, + "failed to set dict"); + ret = -1; + *fop_errno = ENOMEM; + break; } } - ret = syncop_writev(to, dst, vector, count, offset, iobref, 0, NULL, - NULL, xdata, NULL); - if (ret < 0) { - *fop_errno = -ret; - } - } - - if ((defrag && defrag->cmd == GF_DEFRAG_CMD_START_TIER) && - (gf_defrag_get_pause_state(&defrag->tier_conf) != TIER_RUNNING)) { - gf_msg("tier", GF_LOG_INFO, 0, DHT_MSG_TIER_PAUSED, - "Migrate file paused"); - ret = -1; } + ret = syncop_writev(to, dst, vector, count, offset, iobref, 0, NULL, + NULL, xdata, NULL); if (ret < 0) { + *fop_errno = -ret; break; } @@ -1568,6 +1522,7 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, xlator_t *old_target = NULL; xlator_t *hashed_subvol = NULL; fd_t *linkto_fd = NULL; + dict_t *xdata = NULL; if (from == to) { gf_msg_debug(this->name, 0, @@ -1578,21 +1533,6 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, goto out; } - /* If defrag is NULL, it should be assumed that migration is triggered - * from client using the trusted.distribute.migrate-data virtual xattr - */ - defrag = conf->defrag; - - /* migration of files from clients is restricted to non-tiered clients - * for now */ - if (!defrag && dht_is_tier_xlator(this)) { - ret = ENOTSUP; - goto out; - } - - if (defrag && defrag->tier_conf.is_tier) - log_level = GF_LOG_TRACE; - gf_log(this->name, log_level, "%s: attempting to move from %s to %s", loc->path, from->name, to->name); @@ -1739,9 +1679,13 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, goto out; } + /* Try to preserve 'holes' while migrating data */ + if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE)) + file_has_holes = 1; + /* create the destination, with required modes/xattr */ ret = __dht_rebalance_create_dst_file(this, to, from, loc, &stbuf, &dst_fd, - fop_errno); + fop_errno, file_has_holes); if (ret) { gf_msg(this->name, GF_LOG_ERROR, 0, 0, "Create dst failed" @@ -1785,8 +1729,8 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, * destination. We need to do update this only post migration * as in case of failure the linkto needs to point to the source * subvol */ - ret = __dht_rebalance_create_dst_file(this, to, from, loc, &stbuf, - &dst_fd, fop_errno); + ret = __dht_rebalance_create_dst_file( + this, to, from, loc, &stbuf, &dst_fd, fop_errno, file_has_holes); if (ret) { gf_log(this->name, GF_LOG_ERROR, "Create dst failed" @@ -1873,9 +1817,6 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, ret = 0; goto out; } - /* Try to preserve 'holes' while migrating data */ - if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE)) - file_has_holes = 1; ret = __dht_rebalance_migrate_data(this, defrag, from, to, src_fd, dst_fd, stbuf.ia_size, file_has_holes, @@ -1890,7 +1831,15 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to, /* TODO: Sync the locks */ - ret = syncop_fsync(to, dst_fd, 0, NULL, NULL, NULL, NULL); + xdata = dict_new(); + if (!xdata || dict_set_int8(xdata, "last-fsync", 1)) { + gf_log(this->name, GF_LOG_ERROR, + "%s: failed to set last-fsync flag on " + "%s (%s)", + loc->path, to->name, strerror(ENOMEM)); + } + + ret = syncop_fsync(to, dst_fd, 0, NULL, NULL, xdata, NULL); if (ret) { gf_log(this->name, GF_LOG_WARNING, "%s: failed to fsync on %s (%s)", loc->path, to->name, strerror(-ret)); @@ -2333,14 +2282,12 @@ out: } } - if (!dht_is_tier_xlator(this)) { - lk_ret = syncop_removexattr(to, loc, GF_PROTECT_FROM_EXTERNAL_WRITES, - NULL, NULL); - if (lk_ret && (lk_ret != -ENODATA) && (lk_ret != -ENOATTR)) { - gf_msg(this->name, GF_LOG_WARNING, -lk_ret, 0, - "%s: removexattr failed key %s", loc->path, - GF_PROTECT_FROM_EXTERNAL_WRITES); - } + lk_ret = syncop_removexattr(to, loc, GF_PROTECT_FROM_EXTERNAL_WRITES, NULL, + NULL); + if (lk_ret && (lk_ret != -ENODATA) && (lk_ret != -ENOATTR)) { + gf_msg(this->name, GF_LOG_WARNING, -lk_ret, 0, + "%s: removexattr failed key %s", loc->path, + GF_PROTECT_FROM_EXTERNAL_WRITES); } if (dict) @@ -2353,11 +2300,15 @@ out: if (dst_fd) syncop_close(dst_fd); + if (src_fd) syncop_close(src_fd); if (linkto_fd) syncop_close(linkto_fd); + if (xdata) + dict_unref(xdata); + loc_wipe(&tmp_loc); loc_wipe(&parent_loc); @@ -2587,10 +2538,10 @@ out: * all hardlinks. */ -int +gf_boolean_t gf_defrag_should_i_migrate(xlator_t *this, int local_subvol_index, uuid_t gfid) { - int ret = 0; + gf_boolean_t ret = _gf_false; int i = local_subvol_index; char *str = NULL; uint32_t hashval = 0; @@ -2612,12 +2563,11 @@ gf_defrag_should_i_migrate(xlator_t *this, int local_subvol_index, uuid_t gfid) } str = uuid_utoa_r(gfid, buf); - ret = dht_hash_compute(this, 0, str, &hashval); - if (ret == 0) { + if (dht_hash_compute(this, 0, str, &hashval) == 0) { index = (hashval % entry->count); if (entry->elements[index].info == REBAL_NODEUUID_MINE) { /* Index matches this node's nodeuuid.*/ - ret = 1; + ret = _gf_true; goto out; } @@ -2630,12 +2580,12 @@ gf_defrag_should_i_migrate(xlator_t *this, int local_subvol_index, uuid_t gfid) /* None of the bricks in the subvol are up. * CHILD_DOWN will kill the process soon */ - return 0; + return _gf_false; } if (entry->elements[index].info == REBAL_NODEUUID_MINE) { /* Index matches this node's nodeuuid.*/ - ret = 1; + ret = _gf_true; goto out; } } @@ -2684,6 +2634,7 @@ gf_defrag_migrate_single_file(void *opaque) struct iatt *iatt_ptr = NULL; gf_boolean_t update_skippedcount = _gf_true; int i = 0; + gf_boolean_t should_i_migrate = 0; rebal_entry = (struct dht_container *)opaque; if (!rebal_entry) { @@ -2738,11 +2689,29 @@ gf_defrag_migrate_single_file(void *opaque) goto out; } + should_i_migrate = gf_defrag_should_i_migrate( + this, rebal_entry->local_subvol_index, entry->d_stat.ia_gfid); + gf_uuid_copy(entry_loc.gfid, entry->d_stat.ia_gfid); gf_uuid_copy(entry_loc.pargfid, loc->gfid); ret = syncop_lookup(this, &entry_loc, &iatt, NULL, NULL, NULL); + + if (!should_i_migrate) { + /* this node isn't supposed to migrate the file. suppressing any + * potential error from lookup as this file is under migration by + * another node */ + if (ret) { + gf_msg_debug(this->name, -ret, + "Ignoring lookup failure: node isn't migrating %s", + entry_loc.path); + ret = 0; + } + gf_msg_debug(this->name, 0, "Don't migrate %s ", entry_loc.path); + goto out; + } + if (ret) { gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED, "Migrate file failed: %s lookup failed", entry_loc.path); @@ -2763,12 +2732,6 @@ gf_defrag_migrate_single_file(void *opaque) goto out; } - if (!gf_defrag_should_i_migrate(this, rebal_entry->local_subvol_index, - entry->d_stat.ia_gfid)) { - gf_msg_debug(this->name, 0, "Don't migrate %s ", entry_loc.path); - goto out; - } - iatt_ptr = &iatt; hashed_subvol = dht_subvol_get_hashed(this, &entry_loc); @@ -2911,8 +2874,7 @@ gf_defrag_migrate_single_file(void *opaque) if (defrag->stats == _gf_true) { gettimeofday(&end, NULL); - elapsed = (end.tv_sec - start.tv_sec) * 1e6 + - (end.tv_usec - start.tv_usec); + elapsed = gf_tvdiff(&start, &end); gf_log(this->name, GF_LOG_INFO, "Migration of " "file:%s size:%" PRIu64 @@ -3091,9 +3053,9 @@ int static gf_defrag_get_entry(xlator_t *this, int i, 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 *perrno) + int *perrno) { - int ret = -1; + int ret = 0; char is_linkfile = 0; gf_dirent_t *df_entry = NULL; struct dht_container *tmp_container = NULL; @@ -3109,6 +3071,13 @@ int static gf_defrag_get_entry(xlator_t *this, int i, } if (dir_dfmeta->fetch_entries[i] == 1) { + if (!fd) { + dir_dfmeta->fetch_entries[i] = 0; + dir_dfmeta->offset_var[i].readdir_done = 1; + ret = 0; + goto out; + } + ret = syncop_readdirp(conf->local_subvols[i], fd, 131072, dir_dfmeta->offset_var[i].offset, &(dir_dfmeta->equeue[i]), xattr_req, NULL); @@ -3268,7 +3237,6 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, dict_t *migrate_data, int *perrno) { int ret = -1; - fd_t *fd = NULL; dht_conf_t *conf = NULL; gf_dirent_t entries; dict_t *xattr_req = NULL; @@ -3289,7 +3257,7 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, int dfc_index = 0; int throttle_up = 0; struct dir_dfmeta *dir_dfmeta = NULL; - int should_commit_hash = 1; + xlator_t *old_THIS = NULL; gf_log(this->name, GF_LOG_INFO, "migrate data called on %s", loc->path); gettimeofday(&dir_start, NULL); @@ -3302,28 +3270,53 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, goto out; } - fd = fd_create(loc->inode, defrag->pid); - if (!fd) { - gf_log(this->name, GF_LOG_ERROR, "Failed to create fd"); + old_THIS = THIS; + THIS = this; + + dir_dfmeta = GF_CALLOC(1, sizeof(*dir_dfmeta), gf_common_mt_pointer); + if (!dir_dfmeta) { + gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta is NULL"); ret = -1; goto out; } - ret = syncop_opendir(this, loc, fd, NULL, NULL); - if (ret) { - gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_DATA_FAILED, - "Migrate data failed: Failed to open dir %s", loc->path); - *perrno = -ret; + dir_dfmeta->lfd = GF_CALLOC(local_subvols_cnt, sizeof(fd_t *), + gf_common_mt_pointer); + if (!dir_dfmeta->lfd) { + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_INSUFF_MEMORY, + "for dir_dfmeta", NULL); ret = -1; + *perrno = ENOMEM; goto out; } - fd_bind(fd); - dir_dfmeta = GF_CALLOC(1, sizeof(*dir_dfmeta), gf_common_mt_pointer); - if (!dir_dfmeta) { - gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta is NULL"); - ret = -1; - goto out; + for (i = 0; i < local_subvols_cnt; i++) { + dir_dfmeta->lfd[i] = fd_create(loc->inode, defrag->pid); + if (!dir_dfmeta->lfd[i]) { + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_FD_CREATE_FAILED, + NULL); + *perrno = ENOMEM; + ret = -1; + goto out; + } + + ret = syncop_opendir(conf->local_subvols[i], loc, dir_dfmeta->lfd[i], + NULL, NULL); + if (ret) { + fd_unref(dir_dfmeta->lfd[i]); + dir_dfmeta->lfd[i] = NULL; + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_FAILED_TO_OPEN, + "dir: %s", loc->path, "subvol: %s", + conf->local_subvols[i]->name, NULL); + + if (conf->decommission_in_progress) { + *perrno = -ret; + ret = -1; + goto out; + } + } else { + fd_bind(dir_dfmeta->lfd[i]); + } } dir_dfmeta->head = GF_CALLOC(local_subvols_cnt, sizeof(*(dir_dfmeta->head)), @@ -3358,6 +3351,7 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, ret = -1; goto out; } + ret = gf_defrag_ctx_subvols_init(dir_dfmeta->offset_var, this); if (ret) { gf_log(this->name, GF_LOG_ERROR, @@ -3370,7 +3364,8 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, dir_dfmeta->fetch_entries = GF_CALLOC(local_subvols_cnt, sizeof(int), gf_common_mt_int); if (!dir_dfmeta->fetch_entries) { - gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->fetch_entries is NULL"); + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_INSUFF_MEMORY, + "for dir_dfmeta->fetch_entries", NULL); ret = -1; goto out; } @@ -3440,8 +3435,13 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, ldfq_count <= MAX_MIGRATE_QUEUE_COUNT && !dht_dfreaddirp_done(dir_dfmeta->offset_var, local_subvols_cnt)) { ret = gf_defrag_get_entry(this, dfc_index, &container, loc, conf, - defrag, fd, migrate_data, dir_dfmeta, - xattr_req, &should_commit_hash, perrno); + defrag, dir_dfmeta->lfd[dfc_index], + migrate_data, dir_dfmeta, xattr_req, + perrno); + + if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED) { + goto out; + } if (ret) { gf_log(this->name, GF_LOG_WARNING, @@ -3481,27 +3481,19 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, } gettimeofday(&end, NULL); - elapsed = (end.tv_sec - dir_start.tv_sec) * 1e6 + - (end.tv_usec - dir_start.tv_usec); + elapsed = gf_tvdiff(&dir_start, &end); gf_log(this->name, GF_LOG_INFO, "Migration operation on dir %s took " "%.2f secs", loc->path, elapsed / 1e6); ret = 0; out: - + THIS = old_THIS; gf_defrag_free_dir_dfmeta(dir_dfmeta, local_subvols_cnt); if (xattr_req) dict_unref(xattr_req); - if (fd) - fd_unref(fd); - - if (ret == 0 && should_commit_hash == 0) { - ret = 2; - } - /* It does not matter if it errored out - this number is * used to calculate rebalance estimated time to complete. * No locking required as dirs are processed by a single thread. @@ -3509,6 +3501,7 @@ out: defrag->num_dirs_processed++; return ret; } + int gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, dict_t *fix_layout) @@ -3523,7 +3516,6 @@ gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, * rebalance is complete. */ if (defrag->cmd == GF_DEFRAG_CMD_START_LAYOUT_FIX || - defrag->cmd == GF_DEFRAG_CMD_START_DETACH_TIER || defrag->cmd == GF_DEFRAG_CMD_DETACH_START) { return 0; } @@ -3569,114 +3561,6 @@ gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, return 0; } -/* Function for doing a named lookup on file inodes during an attach tier - * So that a hardlink lookup heal i.e gfid to parent gfid lookup heal - * happens on pre-existing data. This is required so that the ctr database has - * hardlinks of all the exisitng file in the volume. CTR xlator on the - * brick/server side does db update/insert of the hardlink on a namelookup. - * Currently the namedlookup is done synchronous to the fixlayout that is - * triggered by attach tier. This is not performant, adding more time to - * fixlayout. The performant approach is record the hardlinks on a compressed - * datastore and then do the namelookup asynchronously later, giving the ctr db - * eventual consistency - * */ -int -gf_fix_layout_tier_attach_lookup(xlator_t *this, loc_t *parent_loc, - gf_dirent_t *file_dentry) -{ - int ret = -1; - dict_t *lookup_xdata = NULL; - dht_conf_t *conf = NULL; - loc_t file_loc = { - 0, - }; - struct iatt iatt = { - 0, - }; - - GF_VALIDATE_OR_GOTO("tier", this, out); - - GF_VALIDATE_OR_GOTO(this->name, parent_loc, out); - - GF_VALIDATE_OR_GOTO(this->name, file_dentry, out); - - GF_VALIDATE_OR_GOTO(this->name, this->private, out); - - if (!parent_loc->inode) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "%s/%s parent is NULL", parent_loc->path, file_dentry->d_name); - goto out; - } - - conf = this->private; - - loc_wipe(&file_loc); - - if (gf_uuid_is_null(file_dentry->d_stat.ia_gfid)) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "%s/%s gfid not present", parent_loc->path, file_dentry->d_name); - goto out; - } - - gf_uuid_copy(file_loc.gfid, file_dentry->d_stat.ia_gfid); - - if (gf_uuid_is_null(parent_loc->gfid)) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "%s/%s" - " gfid not present", - parent_loc->path, file_dentry->d_name); - goto out; - } - - gf_uuid_copy(file_loc.pargfid, parent_loc->gfid); - - ret = dht_build_child_loc(this, &file_loc, parent_loc, file_dentry->d_name); - if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "Child loc build failed"); - ret = -1; - goto out; - } - - lookup_xdata = dict_new(); - if (!lookup_xdata) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "Failed creating lookup dict for %s", file_dentry->d_name); - goto out; - } - - ret = dict_set_int32(lookup_xdata, CTR_ATTACH_TIER_LOOKUP, 1); - if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR, - "Failed to set lookup flag"); - goto out; - } - - gf_uuid_copy(file_loc.parent->gfid, parent_loc->gfid); - - /* Sending lookup to cold tier only */ - ret = syncop_lookup(conf->subvolumes[0], &file_loc, &iatt, NULL, - lookup_xdata, NULL); - if (ret) { - /* If the file does not exist on the cold tier than it must */ - /* have been discovered on the hot tier. This is not an error. */ - gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LOG_TIER_STATUS, - "%s lookup to cold tier on attach heal failed", file_loc.path); - goto out; - } - - ret = 0; - -out: - - loc_wipe(&file_loc); - - if (lookup_xdata) - dict_unref(lookup_xdata); - - return ret; -} - int gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, dict_t *fix_layout, dict_t *migrate_data) @@ -3696,7 +3580,6 @@ 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; @@ -3799,16 +3682,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, "..")) continue; if (!IA_ISDIR(entry->d_stat.ia_type)) { - /* If its a fix layout during the attach - * tier operation do lookups on files - * on cold subvolume so that there is a - * CTR DB Lookup Heal triggered on existing - * data. - * */ - if (defrag->cmd == GF_DEFRAG_CMD_START_TIER) { - gf_fix_layout_tier_attach_lookup(this, loc, entry); - } - continue; } loc_wipe(&entry_loc); @@ -3825,8 +3698,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, goto out; } else { - should_commit_hash = 0; - continue; } } @@ -3889,7 +3760,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, ret = -1; goto out; } else { - should_commit_hash = 0; continue; } } @@ -3902,7 +3772,12 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, ret = gf_defrag_fix_layout(this, defrag, &entry_loc, fix_layout, migrate_data); - if (ret && ret != 2) { + if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED || + defrag->defrag_status == GF_DEFRAG_STATUS_FAILED) { + goto out; + } + + if (ret) { gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_FIX_FAILED, "Fix layout failed for %s", entry_loc.path); @@ -3933,6 +3808,17 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, */ ret = syncop_setxattr(this, loc, fix_layout, 0, NULL, NULL); + + /* In case of a race where the directory is deleted just before + * layout setxattr, the errors are updated in the layout structure. + * We can use this information to make a decision whether the directory + * is deleted entirely. + */ + if (ret == 0) { + ret = dht_dir_layout_error_check(this, loc->inode); + ret = -ret; + } + if (ret) { if (-ret == ENOENT || -ret == ESTALE) { gf_msg(this->name, GF_LOG_INFO, -ret, DHT_MSG_LAYOUT_FIX_FAILED, @@ -3943,6 +3829,7 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, defrag->total_failures++; } ret = 0; + goto out; } else { gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_LAYOUT_FIX_FAILED, "Setxattr failed for %s", loc->path); @@ -3957,11 +3844,10 @@ 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)) { + if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) { ret = gf_defrag_process_dir(this, defrag, loc, migrate_data, &perrno); - if (ret && (ret != 2)) { + if (ret) { if (perrno == ENOENT || perrno == ESTALE) { ret = 0; goto out; @@ -3977,18 +3863,13 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc, if (conf->decommission_in_progress) { goto out; } - - should_commit_hash = 0; } - } else if (ret == 2) { - should_commit_hash = 0; } } gf_msg_trace(this->name, 0, "fix layout called on %s", loc->path); - if (should_commit_hash && - gf_defrag_settle_hash(this, defrag, loc, fix_layout) != 0) { + if (gf_defrag_settle_hash(this, defrag, loc, fix_layout) != 0) { defrag->total_failures++; gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SETTLE_HASH_FAILED, @@ -4012,245 +3893,34 @@ out: if (fd) fd_unref(fd); - if (ret == 0 && should_commit_hash == 0) { - ret = 2; - } - - return ret; -} - -/****************************************************************************** - * Tier background Fix layout functions - ******************************************************************************/ -/* This is the background tier fixlayout thread */ -void * -gf_tier_do_fix_layout(void *args) -{ - gf_tier_fix_layout_arg_t *tier_fix_layout_arg = args; - int ret = -1; - xlator_t *this = NULL; - dht_conf_t *conf = NULL; - gf_defrag_info_t *defrag = NULL; - dict_t *dict = NULL; - loc_t loc = { - 0, - }; - struct iatt iatt = { - 0, - }; - struct iatt parent = { - 0, - }; - - GF_VALIDATE_OR_GOTO("tier", tier_fix_layout_arg, out); - GF_VALIDATE_OR_GOTO("tier", tier_fix_layout_arg->this, out); - this = tier_fix_layout_arg->this; - - conf = this->private; - GF_VALIDATE_OR_GOTO(this->name, conf, out); - - defrag = conf->defrag; - GF_VALIDATE_OR_GOTO(this->name, defrag, out); - GF_VALIDATE_OR_GOTO(this->name, defrag->root_inode, out); - - GF_VALIDATE_OR_GOTO(this->name, tier_fix_layout_arg->fix_layout, out); - - /* Get Root loc_t */ - dht_build_root_loc(defrag->root_inode, &loc); - ret = syncop_lookup(this, &loc, &iatt, &parent, NULL, NULL); - if (ret) { - gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_REBALANCE_START_FAILED, - "Lookup on root failed."); - ret = -1; - goto out; - } - - /* Start the crawl */ - gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LOG_TIER_STATUS, - "Tiering Fixlayout started"); - - ret = gf_defrag_fix_layout(this, defrag, &loc, - tier_fix_layout_arg->fix_layout, NULL); - if (ret && ret != 2) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_FAILED, - "Tiering fixlayout failed."); - ret = -1; - goto out; - } - - if (ret != 2 && - gf_defrag_settle_hash(this, defrag, &loc, - tier_fix_layout_arg->fix_layout) != 0) { - defrag->total_failures++; - ret = -1; - goto out; - } - - dict = dict_new(); - if (!dict) { - ret = -1; - goto out; - } - - ret = dict_set_str(dict, GF_XATTR_TIER_LAYOUT_FIXED_KEY, "yes"); - if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_FAILED, - "Failed to set dictionary value: key = %s", - GF_XATTR_TIER_LAYOUT_FIXED_KEY); - ret = -1; - goto out; - } - - /* Marking the completion of tiering fix layout via a xattr on root */ - ret = syncop_setxattr(this, &loc, dict, 0, NULL, NULL); - if (ret) { - gf_log(this->name, GF_LOG_ERROR, - "Failed to set tiering fix " - "layout completed xattr on %s", - loc.path); - ret = -1; - goto out; - } - - ret = 0; -out: - if (ret && defrag) - defrag->total_failures++; - - if (dict) - dict_unref(dict); - - return NULL; -} - -int -gf_tier_start_fix_layout(xlator_t *this, loc_t *loc, gf_defrag_info_t *defrag, - dict_t *fix_layout) -{ - int ret = -1; - dict_t *tier_dict = NULL; - gf_tier_fix_layout_arg_t *tier_fix_layout_arg = NULL; - - tier_dict = dict_new(); - if (!tier_dict) { - gf_log("tier", GF_LOG_ERROR, - "Tier fix layout failed :" - "Creation of tier_dict failed"); - ret = -1; - goto out; - } - - /* Check if layout is fixed already */ - ret = syncop_getxattr(this, loc, &tier_dict, GF_XATTR_TIER_LAYOUT_FIXED_KEY, - NULL, NULL); - if (ret != 0) { - tier_fix_layout_arg = &defrag->tier_conf.tier_fix_layout_arg; - - /*Fill crawl arguments */ - tier_fix_layout_arg->this = this; - tier_fix_layout_arg->fix_layout = fix_layout; - - /* Spawn the fix layout thread so that its done in the - * background */ - ret = gf_thread_create(&tier_fix_layout_arg->thread_id, NULL, - gf_tier_do_fix_layout, tier_fix_layout_arg, - "tierfixl"); - if (ret) { - gf_log("tier", GF_LOG_ERROR, - "Thread creation failed. " - "Background fix layout for tiering will not " - "work."); - defrag->total_failures++; - goto out; - } - } - ret = 0; -out: - if (tier_dict) - dict_unref(tier_dict); - return ret; } -void -gf_tier_clear_fix_layout(xlator_t *this, loc_t *loc, gf_defrag_info_t *defrag) -{ - int ret = -1; - dict_t *dict = NULL; - - GF_VALIDATE_OR_GOTO("tier", this, out); - GF_VALIDATE_OR_GOTO(this->name, loc, out); - GF_VALIDATE_OR_GOTO(this->name, defrag, out); - - /* Check if background fixlayout is completed. This is not - * multi-process safe i.e there is a possibility that by the time - * we move to remove the xattr there it might have been cleared by some - * other detach process from other node. We ignore the error if such - * a thing happens */ - ret = syncop_getxattr(this, loc, &dict, GF_XATTR_TIER_LAYOUT_FIXED_KEY, - NULL, NULL); - if (ret) { - /* Background fixlayout not complete - nothing to clear*/ - gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_LOG_TIER_STATUS, - "Unable to retrieve fixlayout xattr." - "Assume background fix layout not complete"); - goto out; - } - - ret = syncop_removexattr(this, loc, GF_XATTR_TIER_LAYOUT_FIXED_KEY, NULL, - NULL); - if (ret) { - gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_LOG_TIER_STATUS, - "Failed removing tier fix layout " - "xattr from %s", - loc->path); - goto out; - } - ret = 0; -out: - if (dict) - dict_unref(dict); -} - -void -gf_tier_wait_fix_lookup(gf_defrag_info_t *defrag) -{ - if (defrag->tier_conf.tier_fix_layout_arg.thread_id) { - pthread_join(defrag->tier_conf.tier_fix_layout_arg.thread_id, NULL); - } -} -/******************Tier background Fix layout functions END********************/ - int dht_init_local_subvols_and_nodeuuids(xlator_t *this, dht_conf_t *conf, loc_t *loc) { dict_t *dict = NULL; - gf_defrag_info_t *defrag = NULL; uuid_t *uuid_ptr = NULL; int ret = -1; int i = 0; int j = 0; - defrag = conf->defrag; - - if (defrag->cmd != GF_DEFRAG_CMD_START_TIER) { - /* Find local subvolumes */ - ret = syncop_getxattr(this, loc, &dict, GF_REBAL_FIND_LOCAL_SUBVOL, - NULL, NULL); - if (ret && (ret != -ENODATA)) { - gf_msg(this->name, GF_LOG_ERROR, -ret, 0, - "local " - "subvolume determination failed with error: %d", - -ret); - ret = -1; - goto out; - } - - if (!ret) - goto out; + /* Find local subvolumes */ + ret = syncop_getxattr(this, loc, &dict, GF_REBAL_FIND_LOCAL_SUBVOL, NULL, + NULL); + if (ret && (ret != -ENODATA)) { + gf_msg(this->name, GF_LOG_ERROR, -ret, 0, + "local " + "subvolume determination failed with error: %d", + -ret); + ret = -1; + goto out; } + if (!ret) + goto out; + ret = syncop_getxattr(this, loc, &dict, GF_REBAL_OLD_FIND_LOCAL_SUBVOL, NULL, NULL); if (ret) { @@ -4341,9 +4011,6 @@ dht_file_counter_thread(void *args) struct timespec time_to_wait = { 0, }; - struct timeval now = { - 0, - }; uint64_t tmp_size = 0; if (!args) @@ -4353,9 +4020,8 @@ dht_file_counter_thread(void *args) dht_build_root_loc(defrag->root_inode, &root_loc); while (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) { - gettimeofday(&now, NULL); - time_to_wait.tv_sec = now.tv_sec + 600; - time_to_wait.tv_nsec = 0; + timespec_now(&time_to_wait); + time_to_wait.tv_sec += 600; pthread_mutex_lock(&defrag->fc_mutex); pthread_cond_timedwait(&defrag->fc_wakeup_cond, &defrag->fc_mutex, @@ -4428,7 +4094,7 @@ gf_defrag_estimates_init(xlator_t *this, loc_t *loc, pthread_t *filecnt_thread) goto out; } - ret = gf_thread_create(filecnt_thread, NULL, &dht_file_counter_thread, + ret = gf_thread_create(filecnt_thread, NULL, dht_file_counter_thread, (void *)defrag, "dhtfcnt"); if (ret) { @@ -4485,7 +4151,7 @@ gf_defrag_parallel_migration_init(xlator_t *this, gf_defrag_info_t *defrag, /*Spawn Threads Here*/ while (index < thread_spawn_count) { - ret = gf_thread_create(&(tid[index]), NULL, &gf_defrag_task, + ret = gf_thread_create(&(tid[index]), NULL, gf_defrag_task, (void *)defrag, "dhtmig%d", (index + 1) & 0x3ff); if (ret != 0) { gf_msg("DHT", GF_LOG_ERROR, ret, 0, "Thread[%d] creation failed. ", @@ -4559,7 +4225,6 @@ gf_defrag_start_crawl(void *data) dict_t *migrate_data = NULL; dict_t *status = NULL; glusterfs_ctx_t *ctx = NULL; - dht_methods_t *methods = NULL; call_frame_t *statfs_frame = NULL; xlator_t *old_THIS = NULL; int ret = -1; @@ -4575,7 +4240,6 @@ gf_defrag_start_crawl(void *data) int thread_index = 0; pthread_t *tid = NULL; pthread_t filecnt_thread; - gf_boolean_t is_tier_detach = _gf_false; gf_boolean_t fc_thread_started = _gf_false; this = data; @@ -4594,7 +4258,8 @@ gf_defrag_start_crawl(void *data) if (!defrag) goto exit; - gettimeofday(&defrag->start_time, NULL); + defrag->start_time = gf_time(); + dht_build_root_inode(this, &defrag->root_inode); if (!defrag->root_inode) goto out; @@ -4728,43 +4393,17 @@ gf_defrag_start_crawl(void *data) } } - if (defrag->cmd == GF_DEFRAG_CMD_START_TIER) { - /* Fix layout for attach tier */ - ret = gf_tier_start_fix_layout(this, &loc, defrag, fix_layout); - if (ret) { - goto out; - } - - methods = &(conf->methods); - - /* Calling tier_start of tier.c */ - methods->migration_other(this, defrag); - if (defrag->cmd == GF_DEFRAG_CMD_START_DETACH_TIER || - defrag->cmd == GF_DEFRAG_CMD_DETACH_START) { - ret = dict_set_str(migrate_data, GF_XATTR_FILE_MIGRATE_KEY, - "force"); - if (ret) - goto out; - } - } else { - ret = gf_defrag_fix_layout(this, defrag, &loc, fix_layout, - migrate_data); - if (ret && ret != 2) { - defrag->total_failures++; - ret = -1; - goto out; - } - - if (ret != 2 && - gf_defrag_settle_hash(this, defrag, &loc, fix_layout) != 0) { - defrag->total_failures++; - ret = -1; - goto out; - } + ret = gf_defrag_fix_layout(this, defrag, &loc, fix_layout, migrate_data); + if (ret) { + defrag->total_failures++; + ret = -1; + goto out; + } - if (defrag->cmd == GF_DEFRAG_CMD_START_DETACH_TIER || - defrag->cmd == GF_DEFRAG_CMD_DETACH_START) - is_tier_detach = _gf_true; + if (gf_defrag_settle_hash(this, defrag, &loc, fix_layout) != 0) { + defrag->total_failures++; + ret = -1; + goto out; } gf_log("DHT", GF_LOG_INFO, "crawling file-system completed"); @@ -4778,19 +4417,6 @@ out: defrag->defrag_status = GF_DEFRAG_STATUS_FAILED; } - if (defrag->cmd == GF_DEFRAG_CMD_START_TIER) { - /* Wait for the tier fixlayout to - * complete if its was started.*/ - gf_tier_wait_fix_lookup(defrag); - } - - if (is_tier_detach && ret == 0) { - /* If it was a detach remove the tier fix-layout - * xattr on root. Ignoring the failure, as nothing has to be - * done, logging is done in gf_tier_clear_fix_layout */ - gf_tier_clear_fix_layout(this, &loc, defrag); - } - gf_defrag_parallel_migration_cleanup(defrag, tid, thread_index); if ((defrag->defrag_status != GF_DEFRAG_STATUS_STOPPED) && @@ -4889,9 +4515,6 @@ gf_defrag_get_estimates_based_on_size(dht_conf_t *conf) uint64_t total_processed = 0; uint64_t tmp_count = 0; uint64_t time_to_complete = 0; - struct timeval now = { - 0, - }; double elapsed = 0; defrag = conf->defrag; @@ -4899,8 +4522,7 @@ gf_defrag_get_estimates_based_on_size(dht_conf_t *conf) if (!g_totalsize) goto out; - gettimeofday(&now, NULL); - elapsed = now.tv_sec - defrag->start_time.tv_sec; + elapsed = gf_time() - defrag->start_time; /* Don't calculate the estimates for the first 10 minutes. * It is unlikely to be accurate and estimates are not required @@ -4950,13 +4572,8 @@ gf_defrag_status_get(dht_conf_t *conf, dict_t *dict) uint64_t lookup = 0; uint64_t failures = 0; uint64_t skipped = 0; - uint64_t promoted = 0; - uint64_t demoted = 0; char *status = ""; double elapsed = 0; - struct timeval end = { - 0, - }; uint64_t time_to_complete = 0; uint64_t time_left = 0; gf_defrag_info_t *defrag = conf->defrag; @@ -4973,17 +4590,12 @@ gf_defrag_status_get(dht_conf_t *conf, dict_t *dict) lookup = defrag->num_files_lookedup; failures = defrag->total_failures; skipped = defrag->skipped; - promoted = defrag->total_files_promoted; - demoted = defrag->total_files_demoted; - gettimeofday(&end, NULL); - - elapsed = end.tv_sec - defrag->start_time.tv_sec; + elapsed = gf_time() - defrag->start_time; /* The rebalance is still in progress */ - if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) && - (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED)) { + if (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) { time_to_complete = gf_defrag_get_estimates_based_on_size(conf); if (time_to_complete && (time_to_complete > elapsed)) @@ -4998,14 +4610,6 @@ gf_defrag_status_get(dht_conf_t *conf, dict_t *dict) if (!dict) goto log; - ret = dict_set_uint64(dict, "promoted", promoted); - if (ret) - gf_log(THIS->name, GF_LOG_WARNING, "failed to set promoted count"); - - ret = dict_set_uint64(dict, "demoted", demoted); - if (ret) - gf_log(THIS->name, GF_LOG_WARNING, "failed to set demoted count"); - ret = dict_set_uint64(dict, "files", files); if (ret) gf_log(THIS->name, GF_LOG_WARNING, "failed to set file count"); @@ -5071,159 +4675,6 @@ out: return 0; } -void -gf_defrag_set_pause_state(gf_tier_conf_t *tier_conf, tier_pause_state_t state) -{ - pthread_mutex_lock(&tier_conf->pause_mutex); - tier_conf->pause_state = state; - pthread_mutex_unlock(&tier_conf->pause_mutex); -} - -tier_pause_state_t -gf_defrag_get_pause_state(gf_tier_conf_t *tier_conf) -{ - int state; - - pthread_mutex_lock(&tier_conf->pause_mutex); - state = tier_conf->pause_state; - pthread_mutex_unlock(&tier_conf->pause_mutex); - - return state; -} - -tier_pause_state_t -gf_defrag_check_pause_tier(gf_tier_conf_t *tier_conf) -{ - int woke = 0; - int state = -1; - - pthread_mutex_lock(&tier_conf->pause_mutex); - - if (tier_conf->pause_state == TIER_RUNNING) - goto out; - - if (tier_conf->pause_state == TIER_PAUSED) - goto out; - - if (tier_conf->promote_in_progress || tier_conf->demote_in_progress) - goto out; - - tier_conf->pause_state = TIER_PAUSED; - - if (tier_conf->pause_synctask) { - synctask_wake(tier_conf->pause_synctask); - tier_conf->pause_synctask = 0; - woke = 1; - } - - gf_msg("tier", GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED, "woken %d", woke); - - gf_event(EVENT_TIER_PAUSE, "vol=%s", tier_conf->volname); -out: - state = tier_conf->pause_state; - - pthread_mutex_unlock(&tier_conf->pause_mutex); - - return state; -} - -void -gf_defrag_pause_tier_timeout(void *data) -{ - xlator_t *this = NULL; - dht_conf_t *conf = NULL; - gf_defrag_info_t *defrag = NULL; - - this = (xlator_t *)data; - GF_VALIDATE_OR_GOTO("tier", this, out); - - conf = this->private; - GF_VALIDATE_OR_GOTO(this->name, conf, out); - - defrag = conf->defrag; - GF_VALIDATE_OR_GOTO(this->name, defrag, out); - - gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED, - "Request pause timer timeout"); - - gf_defrag_check_pause_tier(&defrag->tier_conf); - -out: - return; -} - -int -gf_defrag_pause_tier(xlator_t *this, gf_defrag_info_t *defrag) -{ - int ret = 0; - struct timespec delta = { - 0, - }; - int delay = 2; - - if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) - goto out; - - /* - * Set flag requesting to pause tiering. Wait 'delay' seconds for - * tiering to actually stop as indicated by the pause state - * before returning success or failure. - */ - gf_defrag_set_pause_state(&defrag->tier_conf, TIER_REQUEST_PAUSE); - - /* - * If migration is not underway, can pause immediately. - */ - gf_defrag_check_pause_tier(&defrag->tier_conf); - if (gf_defrag_get_pause_state(&defrag->tier_conf) == TIER_PAUSED) - goto out; - - gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED, - "Request pause tier"); - - defrag->tier_conf.pause_synctask = synctask_get(); - delta.tv_sec = delay; - delta.tv_nsec = 0; - defrag->tier_conf.pause_timer = gf_timer_call_after( - this->ctx, delta, gf_defrag_pause_tier_timeout, this); - - synctask_yield(defrag->tier_conf.pause_synctask); - - if (gf_defrag_get_pause_state(&defrag->tier_conf) == TIER_PAUSED) - goto out; - - gf_defrag_set_pause_state(&defrag->tier_conf, TIER_RUNNING); - - ret = -1; -out: - - gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED, - "Pause tiering ret=%d", ret); - - return ret; -} - -int -gf_defrag_resume_tier(xlator_t *this, gf_defrag_info_t *defrag) -{ - gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_RESUME, - "Pause end. Resume tiering"); - - gf_defrag_set_pause_state(&defrag->tier_conf, TIER_RUNNING); - - gf_event(EVENT_TIER_RESUME, "vol=%s", defrag->tier_conf.volname); - - return 0; -} - -int -gf_defrag_start_detach_tier(gf_defrag_info_t *defrag) -{ - defrag->cmd = GF_DEFRAG_CMD_START_DETACH_TIER; - - return 0; -} - int gf_defrag_stop(dht_conf_t *conf, gf_defrag_status_t status, dict_t *output) { diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c index e7d8028..3e24065 100644 --- a/xlators/cluster/dht/src/dht-selfheal.c +++ b/xlators/cluster/dht/src/dht-selfheal.c @@ -8,9 +8,7 @@ cases as published by the Free Software Foundation. */ -#include "dht-common.h" #include "dht-lock.h" -#include <glusterfs/glusterfs-acl.h> #define DHT_SET_LAYOUT_RANGE(layout, i, srt, chunk, path) \ do { \ @@ -35,7 +33,7 @@ } \ } while (0) -int +static int dht_selfheal_layout_lock(call_frame_t *frame, dht_layout_t *layout, gf_boolean_t newdir, dht_selfheal_layout_t healer, dht_need_heal_t should_heal); @@ -146,8 +144,8 @@ dht_refresh_layout_done(call_frame_t *frame) ret = dht_layout_sort(refreshed); if (ret == -1) { - gf_msg(frame->this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_SORT_FAILED, - "sorting the layout failed"); + gf_smsg(frame->this->name, GF_LOG_WARNING, 0, + DHT_MSG_LAYOUT_SORT_FAILED, NULL); goto err; } @@ -203,10 +201,9 @@ dht_refresh_layout_cbk(call_frame_t *frame, void *cookie, xlator_t *this, if (op_ret == -1) { gf_uuid_unparse(local->loc.gfid, gfid); local->op_errno = op_errno; - gf_msg(this->name, GF_LOG_ERROR, op_errno, - DHT_MSG_FILE_LOOKUP_FAILED, - "lookup of %s on %s returned error, gfid: %s", - local->loc.path, prev->name, gfid); + gf_smsg(this->name, GF_LOG_ERROR, op_errno, + DHT_MSG_FILE_LOOKUP_FAILED, "path=%s", local->loc.path, + "name=%s", prev->name, "gfid=%s", gfid, NULL); goto unlock; } @@ -267,9 +264,8 @@ dht_refresh_layout(call_frame_t *frame) conf->subvolume_cnt); if (!local->selfheal.refreshed_layout) { gf_uuid_unparse(local->loc.gfid, gfid); - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, - "mem allocation for layout failed, path:%s gfid:%s", - local->loc.path, gfid); + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED, + "path=%s", local->loc.path, "gfid=%s", gfid, NULL); goto out; } @@ -281,9 +277,8 @@ dht_refresh_layout(call_frame_t *frame) gf_uuid_unparse(local->loc.gfid, gfid); local->xattr_req = dict_new(); if (local->xattr_req == NULL) { - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, - "dict mem allocation failed, path:%s gfid:%s", - local->loc.path, gfid); + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, + "path=%s", local->loc.path, "gfid=%s", gfid, NULL); goto out; } } @@ -291,9 +286,9 @@ dht_refresh_layout(call_frame_t *frame) if (dict_get(local->xattr_req, conf->xattr_name) == 0) { ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4); if (ret) - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, - "%s: Failed to set dictionary value:key = %s", - local->loc.path, conf->xattr_name); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, + "path=%s", local->loc.path, "key=%s", conf->xattr_name, + NULL); } for (i = 0; i < call_cnt; i++) { @@ -526,7 +521,7 @@ out: return fixit; } -int +static int dht_selfheal_layout_lock(call_frame_t *frame, dht_layout_t *layout, gf_boolean_t newdir, dht_selfheal_layout_t healer, dht_need_heal_t should_heal) @@ -558,10 +553,8 @@ dht_selfheal_layout_lock(call_frame_t *frame, dht_layout_t *layout, lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_char); if (lk_array == NULL) { gf_uuid_unparse(local->stbuf.ia_gfid, gfid); - gf_msg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, - "mem allocation failed for " - "lk_array, gfid:%s path: %s", - gfid, local->loc.path); + gf_smsg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED, + "lk_array-gfid=%s", gfid, "path=%s", local->loc.path, NULL); goto err; } @@ -571,10 +564,9 @@ dht_selfheal_layout_lock(call_frame_t *frame, dht_layout_t *layout, DHT_LAYOUT_HEAL_DOMAIN, NULL, FAIL_ON_ANY_ERROR); if (lk_array[i] == NULL) { gf_uuid_unparse(local->stbuf.ia_gfid, gfid); - gf_msg(THIS->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, - "mem allocation " - "failed for lk_array, gfid:%s path:%s", - gfid, local->loc.path); + gf_smsg(THIS->name, GF_LOG_ERROR, ENOMEM, + DHT_MSG_MEM_ALLOC_FAILED, "lk_array-gfid=%s", gfid, + "path=%s", local->loc.path, NULL); goto err; } } @@ -583,10 +575,8 @@ dht_selfheal_layout_lock(call_frame_t *frame, dht_layout_t *layout, lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_char); if (lk_array == NULL) { gf_uuid_unparse(local->stbuf.ia_gfid, gfid); - gf_msg(THIS->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, - "mem allocation failed for " - "lk_array, gfid:%s path:%s", - gfid, local->loc.path); + gf_smsg(THIS->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED, + "lk_array-gfid=%s", gfid, "path=%s", local->loc.path, NULL); goto err; } @@ -595,10 +585,8 @@ dht_selfheal_layout_lock(call_frame_t *frame, dht_layout_t *layout, NULL, FAIL_ON_ANY_ERROR); if (lk_array[0] == NULL) { gf_uuid_unparse(local->stbuf.ia_gfid, gfid); - gf_msg(THIS->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, - "mem allocation failed for " - "lk_array, gfid:%s path:%s", - gfid, local->loc.path); + gf_smsg(THIS->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED, + "lk_array-gfid=%s", gfid, "path=%s", local->loc.path, NULL); goto err; } } @@ -624,7 +612,7 @@ err: return -1; } -int +static int dht_selfheal_dir_xattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, dict_t *xdata) { @@ -646,10 +634,9 @@ dht_selfheal_dir_xattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, err = 0; } else { gf_uuid_unparse(local->loc.gfid, gfid); - gf_msg(this->name, GF_LOG_ERROR, op_errno, - DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, - "layout setxattr failed on %s, path:%s gfid:%s", subvol->name, - local->loc.path, gfid); + gf_smsg(this->name, GF_LOG_ERROR, op_errno, + DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, "name=%s", subvol->name, + "path=%s", local->loc.path, "gfid=%s", gfid, NULL); err = op_errno; } @@ -696,7 +683,7 @@ dht_set_user_xattr(dict_t *dict, char *k, data_t *v, void *data) return ret; } -int +static int dht_selfheal_dir_xattr_persubvol(call_frame_t *frame, loc_t *loc, dht_layout_t *layout, int i, xlator_t *req_subvol) @@ -738,19 +725,17 @@ dht_selfheal_dir_xattr_persubvol(call_frame_t *frame, loc_t *loc, ret = dict_set_str(xdata, GLUSTERFS_INTERNAL_FOP_KEY, "yes"); if (ret < 0) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, - "%s: Failed to set dictionary value: key = %s," - " gfid = %s", - loc->path, GLUSTERFS_INTERNAL_FOP_KEY, gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, + "path=%s", loc->path, "key=%s", GLUSTERFS_INTERNAL_FOP_KEY, + "gfid=%s", gfid, NULL); goto err; } ret = dict_set_int8(xdata, DHT_IATT_IN_XDATA_KEY, 1); if (ret < 0) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, - "%s: Failed to set dictionary value: key = %s," - " gfid = %s", - loc->path, DHT_IATT_IN_XDATA_KEY, gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, + "path=%s", loc->path, "key=%s", DHT_IATT_IN_XDATA_KEY, + "gfid=%s", gfid, NULL); goto err; } @@ -758,21 +743,21 @@ dht_selfheal_dir_xattr_persubvol(call_frame_t *frame, loc_t *loc, ret = dht_disk_layout_extract(this, layout, i, &disk_layout); if (ret == -1) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, - "Directory self heal xattr failed:" - " %s: (subvol %s) Failed to extract disk layout," - " gfid = %s", - loc->path, subvol->name, gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, + DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, + "extract-disk-layout-failed, path=%s", loc->path, "subvol=%s", + subvol->name, "gfid=%s", gfid, NULL); goto err; } ret = dict_set_bin(xattr, conf->xattr_name, disk_layout, 4 * 4); if (ret == -1) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, - "Directory self heal xattr failed:" - "%s: (subvol %s) Failed to set xattr dictionary," - " gfid = %s", - loc->path, subvol->name, gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, + DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, "path=%s", loc->path, + "subvol=%s", subvol->name, + "set-xattr-dictionary-failed" + "gfid=%s", + gfid, NULL); goto err; } disk_layout = NULL; @@ -788,20 +773,17 @@ dht_selfheal_dir_xattr_persubvol(call_frame_t *frame, loc_t *loc, if (data) { ret = dict_add(xattr, QUOTA_LIMIT_KEY, data); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, - "%s: Failed to set dictionary value:" - " key = %s", - loc->path, QUOTA_LIMIT_KEY); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, + "path=%s", loc->path, "key=%s", QUOTA_LIMIT_KEY, NULL); } } data = dict_get(local->xattr, QUOTA_LIMIT_OBJECTS_KEY); if (data) { ret = dict_add(xattr, QUOTA_LIMIT_OBJECTS_KEY, data); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, - "%s: Failed to set dictionary value:" - " key = %s", - loc->path, QUOTA_LIMIT_OBJECTS_KEY); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, + "path=%s", loc->path, "key=%s", QUOTA_LIMIT_OBJECTS_KEY, + NULL); } } } @@ -830,7 +812,7 @@ err: return 0; } -int +static int dht_fix_dir_xattr(call_frame_t *frame, loc_t *loc, dht_layout_t *layout) { dht_local_t *local = NULL; @@ -879,7 +861,7 @@ out: return 0; } -int +static int dht_selfheal_dir_xattr(call_frame_t *frame, loc_t *loc, dht_layout_t *layout) { dht_local_t *local = NULL; @@ -939,9 +921,8 @@ dht_selfheal_dir_xattr(call_frame_t *frame, loc_t *loc, dht_layout_t *layout) dummy = dht_layout_new(this, 1); if (!dummy) { gf_uuid_unparse(loc->gfid, gfid); - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, - "failed to allocate dummy layout, path:%s gfid:%s", loc->path, - gfid); + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_DUMMY_ALLOC_FAILED, + "path=%s", loc->path, "gfid=%s", gfid, NULL); goto out; } for (i = 0; i < conf->subvolume_cnt && missing_xattr; i++) { @@ -957,38 +938,6 @@ out: return 0; } -gf_boolean_t -dht_is_subvol_part_of_layout(dht_layout_t *layout, xlator_t *xlator) -{ - int i = 0; - gf_boolean_t ret = _gf_false; - - for (i = 0; i < layout->cnt; i++) { - if (!strcmp(layout->list[i].xlator->name, xlator->name)) { - ret = _gf_true; - break; - } - } - - return ret; -} - -int -dht_layout_index_from_conf(dht_layout_t *layout, xlator_t *xlator) -{ - int i = -1; - int j = 0; - - for (j = 0; j < layout->cnt; j++) { - if (!strcmp(layout->list[j].xlator->name, xlator->name)) { - i = j; - break; - } - } - - return i; -} - int dht_selfheal_dir_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, struct iatt *statpre, @@ -1079,7 +1028,7 @@ dht_selfheal_dir_setattr(call_frame_t *frame, loc_t *loc, struct iatt *stbuf, return 0; } -int +static int dht_selfheal_dir_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf, struct iatt *preparent, @@ -1109,11 +1058,10 @@ dht_selfheal_dir_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, if (op_ret) { gf_uuid_unparse(local->loc.gfid, gfid); - gf_msg(this->name, - ((op_errno == EEXIST) ? GF_LOG_DEBUG : GF_LOG_WARNING), op_errno, - DHT_MSG_DIR_SELFHEAL_FAILED, - "Directory selfheal failed: path = %s, gfid = %s", - local->loc.path, gfid); + gf_smsg(this->name, + ((op_errno == EEXIST) ? GF_LOG_DEBUG : GF_LOG_WARNING), + op_errno, DHT_MSG_DIR_SELFHEAL_FAILED, "path=%s", + local->loc.path, "gfid=%s", gfid, NULL); goto out; } dht_iatt_merge(this, &local->preparent, preparent); @@ -1132,89 +1080,7 @@ out: return 0; } -void -dht_selfheal_dir_mkdir_setacl(dict_t *xattr, dict_t *dict) -{ - data_t *acl_default = NULL; - data_t *acl_access = NULL; - xlator_t *this = NULL; - int ret = -1; - - GF_ASSERT(xattr); - GF_ASSERT(dict); - - this = THIS; - GF_ASSERT(this); - - acl_default = dict_get(xattr, POSIX_ACL_DEFAULT_XATTR); - - if (!acl_default) { - gf_msg_debug(this->name, 0, "ACL_DEFAULT xattr not present"); - goto cont; - } - ret = dict_set(dict, POSIX_ACL_DEFAULT_XATTR, acl_default); - if (ret) - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, - "Failed to set dictionary value.key = %s", - POSIX_ACL_DEFAULT_XATTR); -cont: - acl_access = dict_get(xattr, POSIX_ACL_ACCESS_XATTR); - if (!acl_access) { - gf_msg_debug(this->name, 0, "ACL_ACCESS xattr not present"); - goto out; - } - ret = dict_set(dict, POSIX_ACL_ACCESS_XATTR, acl_access); - if (ret) - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, - "Failed to set dictionary value.key = %s", - POSIX_ACL_ACCESS_XATTR); - -out: - return; -} - -void -dht_selfheal_dir_mkdir_setquota(dict_t *src, dict_t *dst) -{ - data_t *quota_limit_key = NULL; - data_t *quota_limit_obj_key = NULL; - xlator_t *this = NULL; - int ret = -1; - - GF_ASSERT(src); - GF_ASSERT(dst); - - this = THIS; - GF_ASSERT(this); - - quota_limit_key = dict_get(src, QUOTA_LIMIT_KEY); - if (!quota_limit_key) { - gf_msg_debug(this->name, 0, "QUOTA_LIMIT_KEY xattr not present"); - goto cont; - } - ret = dict_set(dst, QUOTA_LIMIT_KEY, quota_limit_key); - if (ret) - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, - "Failed to set dictionary value.key = %s", QUOTA_LIMIT_KEY); - -cont: - quota_limit_obj_key = dict_get(src, QUOTA_LIMIT_OBJECTS_KEY); - if (!quota_limit_obj_key) { - gf_msg_debug(this->name, 0, - "QUOTA_LIMIT_OBJECTS_KEY xattr not present"); - goto out; - } - ret = dict_set(dst, QUOTA_LIMIT_OBJECTS_KEY, quota_limit_obj_key); - if (ret) - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, - "Failed to set dictionary value.key = %s", - QUOTA_LIMIT_OBJECTS_KEY); - -out: - return; -} - -int +static int dht_selfheal_dir_mkdir_lookup_done(call_frame_t *frame, xlator_t *this) { dht_local_t *local = NULL; @@ -1238,10 +1104,8 @@ dht_selfheal_dir_mkdir_lookup_done(call_frame_t *frame, xlator_t *this) ret = dict_set_gfuuid(dict, "gfid-req", local->gfid, true); if (ret) - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, - "%s: Failed to set dictionary value:" - " key = gfid-req", - loc->path); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, + "path=%s", loc->path, "key=gfid-req", NULL); } else if (local->params) { /* Send the dictionary from higher layers directly */ @@ -1253,18 +1117,15 @@ dht_selfheal_dir_mkdir_lookup_done(call_frame_t *frame, xlator_t *this) dht_dir_set_heal_xattr(this, local, dict, local->xattr, NULL, NULL); if (!dict) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, - "dict is NULL, need to make sure gfids are same"); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_IS_NULL, NULL); dict = dict_new(); if (!dict) return -1; } ret = dict_set_flag(dict, GF_INTERNAL_CTX_KEY, GF_DHT_HEAL_DIR); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, - "Failed to set dictionary value for" - " key = %s at path: %s", - GF_INTERNAL_CTX_KEY, loc->path); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, "key=%s", + GF_INTERNAL_CTX_KEY, "path=%s", loc->path, NULL); /* We can still continue. As heal can still happen * unless quota limits have reached for the dir. */ @@ -1296,7 +1157,7 @@ err: return 0; } -int +static int dht_selfheal_dir_mkdir_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf, @@ -1390,7 +1251,7 @@ err: return 0; } -int +static int dht_selfheal_dir_mkdir_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) @@ -1410,19 +1271,14 @@ dht_selfheal_dir_mkdir_lock_cbk(call_frame_t *frame, void *cookie, local->call_cnt = conf->subvolume_cnt; if (op_ret < 0) { - /* We get this error when the directory entry was not created - * on a newky attached tier subvol. Hence proceed and do mkdir - * on the tier subvol. - */ if (op_errno == EINVAL) { local->call_cnt = 1; dht_selfheal_dir_mkdir_lookup_done(frame, this); return 0; } - gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_ENTRYLK_ERROR, - "acquiring entrylk after inodelk failed for %s", - local->loc.path); + gf_smsg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_ENTRYLK_ERROR, + "path=%s", local->loc.path, NULL); local->op_errno = op_errno; goto err; @@ -1436,10 +1292,8 @@ dht_selfheal_dir_mkdir_lock_cbk(call_frame_t *frame, void *cookie, ret = dict_set_int32(local->xattr_req, "list-xattr", 1); if (ret) - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, - "Failed to set dictionary key list-xattr value " - " for path %s ", - local->loc.path); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, "path=%s", + local->loc.path, NULL); for (i = 0; i < conf->subvolume_cnt; i++) { if (mds_subvol && conf->subvolumes[i] == mds_subvol) { @@ -1462,18 +1316,21 @@ err: return 0; } -int +static int dht_selfheal_dir_mkdir(call_frame_t *frame, loc_t *loc, dht_layout_t *layout, int force) { int missing_dirs = 0; int i = 0; + int op_errno = 0; int ret = -1; dht_local_t *local = NULL; xlator_t *this = NULL; + dht_conf_t *conf = NULL; local = frame->local; this = frame->this; + conf = this->private; local->selfheal.force_mkdir = force; local->selfheal.hole_cnt = 0; @@ -1490,13 +1347,12 @@ dht_selfheal_dir_mkdir(call_frame_t *frame, loc_t *loc, dht_layout_t *layout, if (!__is_root_gfid(local->stbuf.ia_gfid)) { if (local->need_xattr_heal) { local->need_xattr_heal = 0; - ret = dht_dir_xattr_heal(this, local); - if (ret) - gf_msg(this->name, GF_LOG_ERROR, ret, - DHT_MSG_DIR_XATTR_HEAL_FAILED, - "%s:xattr heal failed for " - "directory (gfid = %s)", - local->loc.path, local->gfid); + ret = dht_dir_xattr_heal(this, local, &op_errno); + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, op_errno, + DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s", + local->loc.path, "gfid=%s", local->gfid, NULL); + } } else { if (!gf_uuid_is_null(local->gfid)) gf_uuid_copy(loc->gfid, local->gfid); @@ -1505,28 +1361,53 @@ dht_selfheal_dir_mkdir(call_frame_t *frame, loc_t *loc, dht_layout_t *layout, if (!ret) return 0; - gf_msg(this->name, GF_LOG_INFO, 0, - DHT_MSG_DIR_XATTR_HEAL_FAILED, - "%s: Failed to set mds xattr " - "for directory (gfid = %s)", - local->loc.path, local->gfid); + gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_SET_XATTR_FAILED, + "path=%s", local->loc.path, "gfid=%s", local->gfid, + NULL); } } dht_selfheal_dir_setattr(frame, loc, &local->stbuf, 0xffffffff, layout); return 0; } - if (local->hashed_subvol == NULL) - local->hashed_subvol = dht_subvol_get_hashed(this, loc); + /* MDS xattr is populated only while DHT is having more than one + subvol.In case of graph switch while adding more dht subvols need to + consider hash subvol as a MDS to avoid MDS check failure at the time + of running fop on directory + */ + if (!dict_get(local->xattr, conf->mds_xattr_key) && + (conf->subvolume_cnt > 1)) { + if (local->hashed_subvol == NULL) { + local->hashed_subvol = dht_subvol_get_hashed(this, loc); + if (local->hashed_subvol == NULL) { + local->op_errno = EINVAL; + gf_smsg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_HASHED_SUBVOL_GET_FAILED, "gfid=%s", + loc->pargfid, "name=%s", loc->name, "path=%s", + loc->path, NULL); + goto err; + } + } + ret = dht_inode_ctx_mdsvol_set(local->inode, this, + local->hashed_subvol); + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED, + "Failed to set hashed subvol for %s on inode vol is %s", + local->loc.path, + local->hashed_subvol ? local->hashed_subvol->name : "NULL"); + goto err; + } + } if (local->hashed_subvol == NULL) { - local->op_errno = EINVAL; - gf_msg(this->name, GF_LOG_WARNING, local->op_errno, - DHT_MSG_HASHED_SUBVOL_GET_FAILED, - "(%s/%s) (path: %s): " - "hashed subvolume not found", - loc->pargfid, loc->name, loc->path); - goto err; + local->hashed_subvol = dht_subvol_get_hashed(this, loc); + if (local->hashed_subvol == NULL) { + local->op_errno = EINVAL; + gf_smsg(this->name, GF_LOG_WARNING, local->op_errno, + DHT_MSG_HASHED_SUBVOL_GET_FAILED, "gfid=%s", loc->pargfid, + "name=%s", loc->name, "path=%s", loc->path, NULL); + goto err; + } } local->current = &local->lock[0]; @@ -1542,7 +1423,7 @@ err: return -1; } -int +static int dht_selfheal_layout_alloc_start(xlator_t *this, loc_t *loc, dht_layout_t *layout) { @@ -1651,8 +1532,6 @@ dht_selfheal_layout_new_directory(call_frame_t *frame, loc_t *loc, dht_layout_t *new_layout); void -dht_layout_entry_swap(dht_layout_t *layout, int i, int j); -void dht_layout_range_swap(dht_layout_t *layout, int i, int j); /* @@ -1661,7 +1540,7 @@ dht_layout_range_swap(dht_layout_t *layout, int i, int j); */ #define OV_ENTRY(x, y) table[x * new->cnt + y] -void +static void dht_selfheal_layout_maximize_overlap(call_frame_t *frame, loc_t *loc, dht_layout_t *new, dht_layout_t *old) { @@ -1738,7 +1617,7 @@ dht_selfheal_layout_maximize_overlap(call_frame_t *frame, loc_t *loc, } } -dht_layout_t * +static dht_layout_t * dht_fix_layout_of_directory(call_frame_t *frame, loc_t *loc, dht_layout_t *layout) { @@ -1763,9 +1642,8 @@ dht_fix_layout_of_directory(call_frame_t *frame, loc_t *loc, new_layout = dht_layout_new(this, priv->subvolume_cnt); if (!new_layout) { gf_uuid_unparse(loc->gfid, gfid); - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY, - "mem allocation failed for new_layout, path:%s gfid:%s", - loc->path, gfid); + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED, + "new_layout, path=%s", loc->path, "gfid=%s", gfid, NULL); goto done; } @@ -1775,10 +1653,9 @@ dht_fix_layout_of_directory(call_frame_t *frame, loc_t *loc, if (subvol_down) { gf_uuid_unparse(loc->gfid, gfid); - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_FIX_FAILED, - "Layout fix failed: %u subvolume(s) are down" - ". Skipping fix layout. path:%s gfid:%s", - subvol_down, loc->path, gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_FIX_FAILED, + "subvol-down=%u", subvol_down, "Skipping-fix-layout", "path=%s", + loc->path, "gfid=%s", gfid, NULL); GF_FREE(new_layout); return NULL; } @@ -1796,10 +1673,10 @@ dht_fix_layout_of_directory(call_frame_t *frame, loc_t *loc, if (priv->du_stats) { for (i = 0; i < priv->subvolume_cnt; ++i) { - gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_SUBVOL_INFO, - "subvolume %d (%s): %u chunks, path:%s", i, - priv->subvolumes[i]->name, priv->du_stats[i].chunks, - loc->path); + gf_smsg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_SUBVOL_INFO, + "index=%d", i, "name=%s", priv->subvolumes[i]->name, + "chunks=%u", priv->du_stats[i].chunks, "path=%s", loc->path, + NULL); /* Maximize overlap if the bricks are all the same * size. @@ -1811,8 +1688,8 @@ dht_fix_layout_of_directory(call_frame_t *frame, loc_t *loc, } } } else { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NO_DISK_USAGE_STATUS, - "no du stats ?!?"); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NO_DISK_USAGE_STATUS, + NULL); } /* First give it a layout as though it is a new directory. This @@ -1843,7 +1720,7 @@ done: * Having to call this 2x for each entry in the layout is pretty horrible, but * that's what all of this layout-sorting nonsense gets us. */ -uint32_t +static uint32_t dht_get_chunks_from_xl(xlator_t *parent, xlator_t *child) { dht_conf_t *priv = parent->private; @@ -1961,7 +1838,7 @@ done: return; } -int +static int dht_selfheal_dir_getafix(call_frame_t *frame, loc_t *loc, dht_layout_t *layout) { dht_local_t *local = NULL; @@ -2020,9 +1897,8 @@ dht_selfheal_new_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk, linked_inode = inode_link(loc->inode, loc->parent, loc->name, &local->stbuf); if (!linked_inode) { - gf_msg(frame->this->name, GF_LOG_WARNING, 0, - DHT_MSG_DIR_SELFHEAL_FAILED, - "linking inode failed (%s/%s) => %s", pgfid, loc->name, gfid); + gf_smsg(frame->this->name, GF_LOG_WARNING, 0, DHT_MSG_LINK_INODE_FAILED, + "pgfid=%s", pgfid, "name=%s", loc->name, "gfid=%s", gfid, NULL); ret = -1; goto out; } @@ -2094,9 +1970,18 @@ dht_selfheal_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk, local->selfheal.dir_cbk = dir_cbk; local->selfheal.layout = dht_layout_ref(this, layout); - if (local->need_attrheal && !IA_ISINVAL(local->mds_stbuf.ia_type)) { - /*Use the one in the mds_stbuf*/ - local->stbuf = local->mds_stbuf; + if (local->need_attrheal) { + if (__is_root_gfid(local->stbuf.ia_gfid)) { + local->stbuf.ia_gid = local->prebuf.ia_gid; + local->stbuf.ia_uid = local->prebuf.ia_uid; + + local->stbuf.ia_ctime = local->prebuf.ia_ctime; + local->stbuf.ia_ctime_nsec = local->prebuf.ia_ctime_nsec; + local->stbuf.ia_prot = local->prebuf.ia_prot; + + } else if (!IA_ISINVAL(local->mds_stbuf.ia_type)) { + local->stbuf = local->mds_stbuf; + } } if (!__is_root_gfid(local->stbuf.ia_gfid)) { @@ -2106,9 +1991,9 @@ dht_selfheal_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk, linked_inode = inode_link(loc->inode, loc->parent, loc->name, &local->stbuf); if (!linked_inode) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DIR_SELFHEAL_FAILED, - "linking inode failed (%s/%s) => %s", pgfid, loc->name, - gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LINK_INODE_FAILED, + "pgfid=%s", pgfid, "name=%s", loc->name, "gfid=%s", gfid, + NULL); ret = 0; goto sorry_no_fix; } @@ -2134,19 +2019,17 @@ dht_selfheal_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk, misc = local->selfheal.misc; if (down) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DIR_SELFHEAL_FAILED, - "%s: Directory selfheal failed: %d subvolumes down." - "Not fixing. gfid = %s", - loc->path, down, gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SELFHEAL_FAILED, + "path=%s", loc->path, "subvol-down=%d", down, "Not-fixing", + "gfid=%s", gfid, NULL); ret = 0; goto sorry_no_fix; } if (misc) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DIR_SELFHEAL_FAILED, - "%s: Directory selfheal failed : %d subvolumes " - "have unrecoverable errors. gfid = %s", - loc->path, misc, gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SELFHEAL_FAILED, + "path=%s", loc->path, "misc=%d", misc, "unrecoverable-errors", + "gfid=%s", gfid, NULL); ret = 0; goto sorry_no_fix; @@ -2232,29 +2115,28 @@ dht_dir_heal_xattrs(void *data) gf_uuid_unparse(local->loc.gfid, gfid); if (!mds_subvol) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DIR_XATTR_HEAL_FAILED, - "No mds subvol for %s gfid = %s", local->loc.path, gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NO_MDS_SUBVOL, "path=%s", + local->loc.path, "gfid=%s", gfid, NULL); goto out; } if ((local->loc.inode && gf_uuid_is_null(local->loc.inode->gfid)) || gf_uuid_is_null(local->loc.gfid)) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DIR_XATTR_HEAL_FAILED, - "No gfid present so skip heal for path %s gfid = %s", - local->loc.path, gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_NOT_PRESENT, + "skip-heal path=%s", local->loc.path, "gfid=%s", gfid, NULL); goto out; } internal_xattr = dict_new(); if (!internal_xattr) { - gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, 0, - "dictionary creation failed"); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CREATE_FAILED, + "dictionary", NULL); goto out; } xdata = dict_new(); if (!xdata) { - gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, 0, - "dictionary creation failed"); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CREATE_FAILED, + "dictionary", NULL); goto out; } @@ -2262,18 +2144,17 @@ dht_dir_heal_xattrs(void *data) user_xattr = dict_new(); if (!user_xattr) { - gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, 0, - "dictionary creation failed"); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CREATE_FAILED, + "dictionary", NULL); goto out; } ret = syncop_listxattr(local->mds_subvol, &local->loc, &mds_xattr, NULL, NULL); if (ret < 0) { - gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_DIR_XATTR_HEAL_FAILED, - "failed to list xattrs for " - "%s: on %s ", - local->loc.path, local->mds_subvol->name); + gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_LIST_XATTRS_FAILED, + "path=%s", local->loc.path, "name=%s", local->mds_subvol->name, + NULL); } if (!mds_xattr) @@ -2288,10 +2169,9 @@ dht_dir_heal_xattrs(void *data) dict_get(user_xattr, QUOTA_LIMIT_OBJECTS_KEY)) { ret = dict_set_int32(xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, - "Failed to set dictionary value: key = %s," - " path = %s", - GLUSTERFS_INTERNAL_FOP_KEY, local->loc.path); + gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, + "key=%s", GLUSTERFS_INTERNAL_FOP_KEY, "path=%s", + local->loc.path, NULL); goto out; } } @@ -2303,16 +2183,25 @@ dht_dir_heal_xattrs(void *data) if (subvol == mds_subvol) continue; if (uret || uflag) { + /* Custom xattr heal is required - let posix handle it */ + ret = dict_set_int8(xdata, "sync_backend_xattrs", _gf_true); + if (ret) { + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED, + "path=%s", local->loc.path, "key=%s", + "sync_backend_xattrs", NULL); + goto out; + } + ret = syncop_setxattr(subvol, &local->loc, user_xattr, 0, xdata, NULL); if (ret) { xattr_hashed = 1; - gf_msg(this->name, GF_LOG_ERROR, -ret, - DHT_MSG_DIR_XATTR_HEAL_FAILED, - "Directory xattr heal failed. Failed to set" - "user xattr on path %s on " - "subvol %s, gfid = %s ", - local->loc.path, subvol->name, gfid); + gf_smsg(this->name, GF_LOG_ERROR, -ret, + DHT_MSG_DIR_XATTR_HEAL_FAILED, + "set-user-xattr-failed path=%s", local->loc.path, + "subvol=%s", subvol->name, "gfid=%s", gfid, NULL); + } else { + dict_del(xdata, "sync_backend_xattrs"); } } } @@ -2321,21 +2210,17 @@ dht_dir_heal_xattrs(void *data) ret = dht_dict_set_array(internal_xattr, conf->mds_xattr_key, allzero, 1); if (ret) { - gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED, - "Failed to set dictionary value:key = %s for " - "path %s", - conf->mds_xattr_key, local->loc.path); + gf_smsg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED, + "key=%s", conf->mds_xattr_key, "path=%s", local->loc.path, + NULL); goto out; } ret = syncop_setxattr(mds_subvol, &local->loc, internal_xattr, 0, NULL, NULL); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, -ret, - DHT_MSG_DIR_XATTR_HEAL_FAILED, - "Failed to reset internal xattr " - "on path %s on subvol %s" - "gfid = %s ", - local->loc.path, mds_subvol->name, gfid); + gf_smsg(this->name, GF_LOG_ERROR, -ret, + DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s", local->loc.path, + "subvol=%s", mds_subvol->name, "gfid=%s", gfid, NULL); } } @@ -2386,8 +2271,8 @@ dht_dir_attr_heal(void *data) call_cnt = conf->subvolume_cnt; if (!__is_root_gfid(local->stbuf.ia_gfid) && (!mds_subvol)) { - gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DIR_ATTR_HEAL_FAILED, - "No mds subvol for %s gfid = %s", local->loc.path, gfid); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NO_MDS_SUBVOL, "path=%s", + local->loc.path, "gfid=%s", gfid, NULL); goto out; } @@ -2395,11 +2280,9 @@ dht_dir_attr_heal(void *data) for (i = 0; i < conf->subvolume_cnt; i++) { if (conf->subvolumes[i] == mds_subvol) { if (!conf->subvolume_status[i]) { - gf_msg(this->name, GF_LOG_ERROR, 0, - DHT_MSG_HASHED_SUBVOL_DOWN, - "mds subvol is down for path " - " %s gfid is %s Unable to set xattr ", - local->loc.path, gfid); + gf_smsg(this->name, GF_LOG_ERROR, 0, + DHT_MSG_MDS_DOWN_UNABLE_TO_SET, "path=%s", + local->loc.path, "gfid=%s", gfid, NULL); goto out; } } @@ -2425,10 +2308,9 @@ dht_dir_attr_heal(void *data) if (ret) { gf_uuid_unparse(local->loc.gfid, gfid); - gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_DIR_ATTR_HEAL_FAILED, - "Directory attr heal failed. Failed to set" - " uid/gid on path %s on subvol %s, gfid = %s ", - local->loc.path, subvol->name, gfid); + gf_smsg(this->name, GF_LOG_ERROR, -ret, + DHT_MSG_DIR_ATTR_HEAL_FAILED, "path=%s", local->loc.path, + "subvol=%s", subvol->name, "gfid=%s", gfid, NULL); } } out: @@ -2443,7 +2325,7 @@ dht_dir_attr_heal_done(int ret, call_frame_t *sync_frame, void *data) } /* EXIT: dht_update_commit_hash_for_layout */ -int +static int dht_update_commit_hash_for_layout_done(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) @@ -2463,7 +2345,7 @@ dht_update_commit_hash_for_layout_done(call_frame_t *frame, void *cookie, return 0; } -int +static int dht_update_commit_hash_for_layout_unlock(call_frame_t *frame, xlator_t *this) { dht_local_t *local = NULL; @@ -2481,11 +2363,8 @@ dht_update_commit_hash_for_layout_unlock(call_frame_t *frame, xlator_t *this) local->op_ret = -1; } - gf_msg(this->name, GF_LOG_WARNING, errno, - DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, - "Winding unlock failed: stale locks left on brick" - " %s", - local->loc.path); + gf_smsg(this->name, GF_LOG_WARNING, errno, DHT_MSG_WIND_UNLOCK_FAILED, + "path=%s", local->loc.path, NULL); dht_update_commit_hash_for_layout_done(frame, NULL, this, 0, 0, NULL); } @@ -2493,7 +2372,7 @@ dht_update_commit_hash_for_layout_unlock(call_frame_t *frame, xlator_t *this) return 0; } -int +static int dht_update_commit_hash_for_layout_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret, int op_errno, dict_t *xdata) @@ -2520,7 +2399,7 @@ dht_update_commit_hash_for_layout_cbk(call_frame_t *frame, void *cookie, return 0; } -int +static int dht_update_commit_hash_for_layout_resume(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) @@ -2548,11 +2427,8 @@ dht_update_commit_hash_for_layout_resume(call_frame_t *frame, void *cookie, if (!xattr) { local->op_errno = errno; - gf_msg(this->name, GF_LOG_WARNING, errno, - DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, - "Directory commit hash update failed:" - " %s: Allocation failed", - local->loc.path); + gf_smsg(this->name, GF_LOG_WARNING, errno, DHT_MSG_COMMIT_HASH_FAILED, + "allocation-failed path=%s", local->loc.path, NULL); goto err; } @@ -2563,11 +2439,10 @@ dht_update_commit_hash_for_layout_resume(call_frame_t *frame, void *cookie, if (ret < 0) { local->op_errno = ENOENT; - gf_msg(this->name, GF_LOG_WARNING, 0, - DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, - "Directory commit hash update failed:" - " %s: (subvol %s) Failed to find disk layout", - local->loc.path, conf->local_subvols[i]->name); + gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_COMMIT_HASH_FAILED, + "path=%s", local->loc.path, "subvol=%s", + conf->local_subvols[i]->name, "find-disk-layout-failed", + NULL); goto err; } @@ -2581,12 +2456,10 @@ dht_update_commit_hash_for_layout_resume(call_frame_t *frame, void *cookie, if (ret == -1) { local->op_errno = errno; - gf_msg(this->name, GF_LOG_WARNING, errno, - DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, - "Directory commit hash update failed:" - " %s: (subvol %s) Failed to extract disk" - " layout", - local->loc.path, conf->local_subvols[i]->name); + gf_smsg(this->name, GF_LOG_WARNING, errno, + DHT_MSG_COMMIT_HASH_FAILED, "path=%s", local->loc.path, + "subvol=%s", conf->local_subvols[i]->name, + "extract-disk-layout-failed", NULL); goto err; } @@ -2595,11 +2468,9 @@ dht_update_commit_hash_for_layout_resume(call_frame_t *frame, void *cookie, if (!xattr[i]) { local->op_errno = errno; - gf_msg(this->name, GF_LOG_WARNING, errno, - DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, - "Directory commit hash update failed:" - " %s: Allocation failed", - local->loc.path); + gf_smsg(this->name, GF_LOG_WARNING, errno, + DHT_MSG_COMMIT_HASH_FAILED, "path=%s Allocation-failed", + local->loc.path, NULL); goto err; } @@ -2608,12 +2479,10 @@ dht_update_commit_hash_for_layout_resume(call_frame_t *frame, void *cookie, if (ret != 0) { local->op_errno = ENOMEM; - gf_msg(this->name, GF_LOG_WARNING, 0, - DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, - "Directory self heal xattr failed:" - "%s: (subvol %s) Failed to set xattr" - " dictionary,", - local->loc.path, conf->local_subvols[i]->name); + gf_smsg(this->name, GF_LOG_WARNING, 0, + DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, "path=%s", + local->loc.path, "subvol=%s", conf->local_subvols[i]->name, + "set-xattr-failed", NULL); goto err; } diff --git a/xlators/cluster/dht/src/dht-shared.c b/xlators/cluster/dht/src/dht-shared.c index 0b73121..bb72b0f 100644 --- a/xlators/cluster/dht/src/dht-shared.c +++ b/xlators/cluster/dht/src/dht-shared.c @@ -17,23 +17,6 @@ #define MAX(a, b) (((a) > (b)) ? (a) : (b)) #endif -#define GF_DECIDE_DEFRAG_THROTTLE_COUNT(throttle_count, conf) \ - { \ - throttle_count = MAX((sysconf(_SC_NPROCESSORS_ONLN) - 4), 4); \ - pthread_mutex_lock(&conf->defrag->dfq_mutex); \ - \ - if (!strcasecmp(conf->dthrottle, "lazy")) \ - conf->defrag->recon_thread_count = 1; \ - \ - else if (!strcasecmp(conf->dthrottle, "normal")) \ - conf->defrag->recon_thread_count = (throttle_count / 2); \ - \ - else if (!strcasecmp(conf->dthrottle, "aggressive")) \ - conf->defrag->recon_thread_count = throttle_count; \ - \ - pthread_mutex_unlock(&conf->defrag->dfq_mutex); \ - } - /* TODO: - use volumename in xattr instead of "dht" - use NS locks @@ -41,9 +24,7 @@ - complete linkfile selfheal */ -extern dht_methods_t dht_methods; - -void +static void dht_layout_dump(dht_layout_t *layout, const char *prefix) { char key[GF_DUMP_MAX_BUF_LEN]; @@ -51,8 +32,6 @@ dht_layout_dump(dht_layout_t *layout, const char *prefix) if (!layout) goto out; - if (!prefix) - goto out; gf_proc_dump_build_key(key, prefix, "cnt"); gf_proc_dump_write(key, "%d", layout->cnt); @@ -161,9 +140,9 @@ dht_priv_dump(xlator_t *this) } } - if (conf->last_stat_fetch.tv_sec) + if (conf->last_stat_fetch) gf_proc_dump_write("last_stat_fetch", "%s", - ctime(&conf->last_stat_fetch.tv_sec)); + ctime(&conf->last_stat_fetch)); UNLOCK(&conf->subvolume_lock); @@ -263,7 +242,7 @@ out: return ret; } -int +static int dht_parse_decommissioned_bricks(xlator_t *this, dht_conf_t *conf, const char *bricks) { @@ -309,14 +288,10 @@ out: return ret; } -int +static void dht_decommissioned_remove(xlator_t *this, dht_conf_t *conf) { int i = 0; - int ret = -1; - - if (!conf) - goto out; for (i = 0; i < conf->subvolume_cnt; i++) { if (conf->decommissioned_bricks[i]) { @@ -324,13 +299,9 @@ dht_decommissioned_remove(xlator_t *this, dht_conf_t *conf) conf->decommission_subvols_cnt--; } } - - ret = 0; -out: - - return ret; } -void + +static void dht_init_regex(xlator_t *this, dict_t *odict, char *name, regex_t *re, gf_boolean_t *re_valid, dht_conf_t *conf) { @@ -387,7 +358,7 @@ out: return ret; } -int +static int dht_configure_throttle(xlator_t *this, dht_conf_t *conf, char *temp_str) { int rebal_thread_count = 0; @@ -526,9 +497,7 @@ dht_reconfigure(xlator_t *this, dict_t *options) if (ret == -1) goto out; } else { - ret = dht_decommissioned_remove(this, conf); - if (ret == -1) - goto out; + dht_decommissioned_remove(this, conf); } dht_init_regex(this, options, "rsync-hash-regex", &conf->rsync_regex, @@ -568,6 +537,8 @@ gf_defrag_pattern_list_fill(xlator_t *this, gf_defrag_info_t *defrag, pattern_str = strtok_r(data, ",", &tmp_str); while (pattern_str) { dup_str = gf_strdup(pattern_str); + if (!dup_str) + goto out; pattern_list = GF_CALLOC(1, sizeof(gf_defrag_pattern_list_t), 1); if (!pattern_list) { goto out; @@ -614,7 +585,7 @@ out: return ret; } -int +static int dht_init_methods(xlator_t *this) { int ret = -1; @@ -627,7 +598,6 @@ dht_init_methods(xlator_t *this) methods = &(conf->methods); methods->migration_get_dst_subvol = dht_migration_get_dst_subvol; - methods->migration_needed = dht_migration_needed; methods->migration_other = NULL; methods->layout_search = dht_layout_search; @@ -1076,84 +1046,6 @@ struct volume_options dht_options[] = { /* NUFA option */ {.key = {"local-volume-name"}, .type = GF_OPTION_TYPE_XLATOR}, - /* tier options */ - { - .key = {"tier-pause"}, - .type = GF_OPTION_TYPE_BOOL, - .default_value = "off", - }, - - { - .key = {"tier-promote-frequency"}, - .type = GF_OPTION_TYPE_INT, - .default_value = "120", - }, - - { - .key = {"tier-demote-frequency"}, - .type = GF_OPTION_TYPE_INT, - .default_value = "3600", - }, - - { - .key = {"write-freq-threshold"}, - .type = GF_OPTION_TYPE_INT, - .default_value = "0", - }, - - { - .key = {"read-freq-threshold"}, - .type = GF_OPTION_TYPE_INT, - .default_value = "0", - }, - { - .key = {"watermark-hi"}, - .type = GF_OPTION_TYPE_PERCENT, - .default_value = "90", - }, - { - .key = {"watermark-low"}, - .type = GF_OPTION_TYPE_PERCENT, - .default_value = "75", - }, - { - .key = {"tier-mode"}, - .type = GF_OPTION_TYPE_STR, - .default_value = "test", - }, - { - .key = {"tier-compact"}, - .type = GF_OPTION_TYPE_BOOL, - .default_value = "off", - }, - {.key = {"tier-hot-compact-frequency"}, - .type = GF_OPTION_TYPE_INT, - .default_value = "604800", - .description = "Frequency to compact DBs on hot tier in system"}, - {.key = {"tier-cold-compact-frequency"}, - .type = GF_OPTION_TYPE_INT, - .default_value = "604800", - .description = "Frequency to compact DBs on cold tier in system"}, - { - .key = {"tier-max-mb"}, - .type = GF_OPTION_TYPE_INT, - .default_value = "4000", - }, - { - .key = {"tier-max-promote-file-size"}, - .type = GF_OPTION_TYPE_INT, - .default_value = "0", - }, - { - .key = {"tier-max-files"}, - .type = GF_OPTION_TYPE_INT, - .default_value = "10000", - }, - { - .key = {"tier-query-limit"}, - .type = GF_OPTION_TYPE_INT, - .default_value = "100", - }, /* switch option */ {.key = {"pattern.switch.case"}, .type = GF_OPTION_TYPE_ANY}, diff --git a/xlators/cluster/dht/src/nufa.c b/xlators/cluster/dht/src/nufa.c index 5931363..3648a56 100644 --- a/xlators/cluster/dht/src/nufa.c +++ b/xlators/cluster/dht/src/nufa.c @@ -595,7 +595,6 @@ nufa_init(xlator_t *this) dht_methods_t dht_methods = { .migration_get_dst_subvol = dht_migration_get_dst_subvol, - .migration_needed = dht_migration_needed, .layout_search = dht_layout_search, }; diff --git a/xlators/cluster/ec/src/ec-combine.c b/xlators/cluster/ec/src/ec-combine.c index dedcfb0..703a30e 100644 --- a/xlators/cluster/ec/src/ec-combine.c +++ b/xlators/cluster/ec/src/ec-combine.c @@ -179,13 +179,14 @@ ec_iatt_combine(ec_fop_data_t *fop, struct iatt *dst, struct iatt *src, "links: %u-%u, uid: %u-%u, gid: %u-%u, " "rdev: %" PRIu64 "-%" PRIu64 ", size: %" PRIu64 "-%" PRIu64 ", " - "mode: %o-%o)", + "mode: %o-%o), %s", dst[i].ia_ino, src[i].ia_ino, dst[i].ia_nlink, src[i].ia_nlink, dst[i].ia_uid, src[i].ia_uid, dst[i].ia_gid, src[i].ia_gid, dst[i].ia_rdev, src[i].ia_rdev, dst[i].ia_size, src[i].ia_size, st_mode_from_ia(dst[i].ia_prot, dst[i].ia_type), - st_mode_from_ia(src[i].ia_prot, dst[i].ia_type)); + st_mode_from_ia(src[i].ia_prot, dst[i].ia_type), + ec_msg_str(fop)); return 0; } @@ -342,9 +343,8 @@ out: } static int32_t -ec_dict_data_concat(const char *fmt, ec_cbk_data_t *cbk, int32_t which, - char *key, char *new_key, const char *def, - gf_boolean_t global, ...) +ec_dict_data_concat(ec_cbk_data_t *cbk, int32_t which, char *key, char *new_key, + const char *def, gf_boolean_t global, const char *fmt, ...) { ec_t *ec = cbk->fop->xl->private; data_t *data[ec->nodes]; @@ -356,7 +356,7 @@ ec_dict_data_concat(const char *fmt, ec_cbk_data_t *cbk, int32_t which, ec_dict_list(data, cbk, which, key, global); - va_start(args, global); + va_start(args, fmt); err = ec_concat_prepare(cbk->fop->xl, &pre, &sep, &post, fmt, args); va_end(args); @@ -729,14 +729,14 @@ ec_dict_data_combine(dict_t *dict, char *key, data_t *value, void *arg) if ((strcmp(key, GF_XATTR_PATHINFO_KEY) == 0) || (strcmp(key, GF_XATTR_USER_PATHINFO_KEY) == 0)) { - return ec_dict_data_concat("(<EC:%s> { })", data->cbk, data->which, key, - NULL, NULL, _gf_false, + return ec_dict_data_concat(data->cbk, data->which, key, NULL, NULL, + _gf_false, _gf_false, "(<EC:%s> { })", data->cbk->fop->xl->name); } if (strncmp(key, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD)) == 0) { - return ec_dict_data_concat("{\n}", data->cbk, data->which, key, NULL, - NULL, _gf_false); + return ec_dict_data_concat(data->cbk, data->which, key, NULL, NULL, + _gf_false, "{\n}"); } if (strncmp(key, GF_XATTR_LOCKINFO_KEY, SLEN(GF_XATTR_LOCKINFO_KEY)) == 0) { @@ -766,9 +766,9 @@ ec_dict_data_combine(dict_t *dict, char *key, data_t *value, void *arg) if (XATTR_IS_NODE_UUID(key)) { if (data->cbk->fop->int32) { /* List of node uuid is requested */ - return ec_dict_data_concat("{ }", data->cbk, data->which, key, + return ec_dict_data_concat(data->cbk, data->which, key, GF_XATTR_LIST_NODE_UUIDS_KEY, UUID0_STR, - _gf_true); + _gf_true, "{ }"); } else { return ec_dict_data_uuid(data->cbk, data->which, key); } diff --git a/xlators/cluster/ec/src/ec-common.c b/xlators/cluster/ec/src/ec-common.c index 9045a33..b955efd 100644 --- a/xlators/cluster/ec/src/ec-common.c +++ b/xlators/cluster/ec/src/ec-common.c @@ -230,7 +230,7 @@ ec_child_next(ec_t *ec, ec_fop_data_t *fop, uint32_t idx) int32_t ec_heal_report(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, uintptr_t mask, uintptr_t good, - uintptr_t bad, dict_t *xdata) + uintptr_t bad, uint32_t pending, dict_t *xdata) { if (op_ret < 0) { gf_msg(this->name, GF_LOG_DEBUG, op_errno, EC_MSG_HEAL_FAIL, @@ -316,17 +316,19 @@ ec_check_status(ec_fop_data_t *fop) } } - gf_msg(fop->xl->name, GF_LOG_WARNING, 0, EC_MSG_OP_FAIL_ON_SUBVOLS, - "Operation failed on %d of %d subvolumes.(up=%s, mask=%s, " - "remaining=%s, good=%s, bad=%s, %s)", - gf_bits_count(ec->xl_up & ~(fop->remaining | fop->good)), ec->nodes, - ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes), - ec_bin(str2, sizeof(str2), fop->mask, ec->nodes), - ec_bin(str3, sizeof(str3), fop->remaining, ec->nodes), - ec_bin(str4, sizeof(str4), fop->good, ec->nodes), - ec_bin(str5, sizeof(str5), ec->xl_up & ~(fop->remaining | fop->good), - ec->nodes), - ec_msg_str(fop)); + gf_msg( + fop->xl->name, GF_LOG_WARNING, 0, EC_MSG_OP_FAIL_ON_SUBVOLS, + "Operation failed on %d of %d subvolumes.(up=%s, mask=%s, " + "remaining=%s, good=%s, bad=%s," + "(Least significant bit represents first client/brick of subvol), %s)", + gf_bits_count(ec->xl_up & ~(fop->remaining | fop->good)), ec->nodes, + ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes), + ec_bin(str2, sizeof(str2), fop->mask, ec->nodes), + ec_bin(str3, sizeof(str3), fop->remaining, ec->nodes), + ec_bin(str4, sizeof(str4), fop->good, ec->nodes), + ec_bin(str5, sizeof(str5), ec->xl_up & ~(fop->remaining | fop->good), + ec->nodes), + ec_msg_str(fop)); if (fop->use_fd) { if (fop->fd != NULL) { ec_fheal(NULL, fop->xl, -1, EC_MINIMUM_ONE, ec_heal_report, NULL, @@ -614,10 +616,10 @@ ec_msg_str(ec_fop_data_t *fop) loc_t *loc2 = NULL; char gfid1[64] = {0}; char gfid2[64] = {0}; + ec_fop_data_t *parent = fop->parent; if (fop->errstr) return fop->errstr; - if (!fop->use_fd) { loc1 = &fop->loc[0]; loc2 = &fop->loc[1]; @@ -625,23 +627,45 @@ ec_msg_str(ec_fop_data_t *fop) if (fop->id == GF_FOP_RENAME) { gf_asprintf(&fop->errstr, "FOP : '%s' failed on '%s' and '%s' with gfids " - "%s and %s respectively", + "%s and %s respectively. Parent FOP: %s", ec_fop_name(fop->id), loc1->path, loc2->path, uuid_utoa_r(loc1->gfid, gfid1), - uuid_utoa_r(loc2->gfid, gfid2)); + uuid_utoa_r(loc2->gfid, gfid2), + parent ? ec_fop_name(parent->id) : "No Parent"); } else { - gf_asprintf(&fop->errstr, "FOP : '%s' failed on '%s' with gfid %s", - ec_fop_name(fop->id), loc1->path, - uuid_utoa_r(loc1->gfid, gfid1)); + gf_asprintf( + &fop->errstr, + "FOP : '%s' failed on '%s' with gfid %s. Parent FOP: %s", + ec_fop_name(fop->id), loc1->path, + uuid_utoa_r(loc1->gfid, gfid1), + parent ? ec_fop_name(parent->id) : "No Parent"); } } else { - gf_asprintf(&fop->errstr, "FOP : '%s' failed on gfid %s", - ec_fop_name(fop->id), - uuid_utoa_r(fop->fd->inode->gfid, gfid1)); + gf_asprintf( + &fop->errstr, "FOP : '%s' failed on gfid %s. Parent FOP: %s", + ec_fop_name(fop->id), uuid_utoa_r(fop->fd->inode->gfid, gfid1), + parent ? ec_fop_name(parent->id) : "No Parent"); } return fop->errstr; } +static void +ec_log_insufficient_vol(ec_fop_data_t *fop, int32_t have, uint32_t need, + int32_t loglevel) +{ + ec_t *ec = fop->xl->private; + char str1[32], str2[32], str3[32]; + + gf_msg(ec->xl->name, loglevel, 0, EC_MSG_CHILDS_INSUFFICIENT, + "Insufficient available children for this request: " + "Have : %d, Need : %u : Child UP : %s " + "Mask: %s, Healing : %s : %s ", + have, need, ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes), + ec_bin(str2, sizeof(str2), fop->mask, ec->nodes), + ec_bin(str3, sizeof(str3), fop->healing, ec->nodes), + ec_msg_str(fop)); +} + static int32_t ec_child_select(ec_fop_data_t *fop) { @@ -699,11 +723,7 @@ ec_child_select(ec_fop_data_t *fop) ec_trace("SELECT", fop, ""); if ((num < fop->minimum) && (num < ec->fragments)) { - gf_msg(ec->xl->name, GF_LOG_ERROR, 0, EC_MSG_CHILDS_INSUFFICIENT, - "Insufficient available children " - "for this request (have %d, need " - "%d). %s", - num, fop->minimum, ec_msg_str(fop)); + ec_log_insufficient_vol(fop, num, fop->minimum, GF_LOG_ERROR); return 0; } @@ -711,11 +731,7 @@ ec_child_select(ec_fop_data_t *fop) (fop->locks[0].update[EC_DATA_TXN] || fop->locks[0].update[EC_METADATA_TXN])) { if (ec->quorum_count && (num < ec->quorum_count)) { - gf_msg(ec->xl->name, GF_LOG_ERROR, 0, EC_MSG_CHILDS_INSUFFICIENT, - "Insufficient available children " - "for this request (have %d, need " - "%d). %s", - num, ec->quorum_count, ec_msg_str(fop)); + ec_log_insufficient_vol(fop, num, ec->quorum_count, GF_LOG_ERROR); return 0; } } @@ -2240,7 +2256,7 @@ ec_unlocked(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, if (op_ret < 0) { gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_UNLOCK_FAILED, - "entry/inode unlocking failed (%s)", ec_fop_name(link->fop->id)); + "entry/inode unlocking failed :(%s)", ec_msg_str(link->fop)); } else { ec_trace("UNLOCKED", link->fop, "lock=%p", link->lock); } diff --git a/xlators/cluster/ec/src/ec-dir-read.c b/xlators/cluster/ec/src/ec-dir-read.c index ef6b06f..f71dcfa 100644 --- a/xlators/cluster/ec/src/ec-dir-read.c +++ b/xlators/cluster/ec/src/ec-dir-read.c @@ -386,9 +386,16 @@ ec_manager_readdir(ec_fop_data_t *fop, int32_t state) /* Return error if opendir has not been successfully called on * any subvolume. */ ctx = ec_fd_get(fop->fd, fop->xl); - if ((ctx == NULL) || (ctx->open == 0)) { - fop->error = EINVAL; + if (ctx == NULL) { + fop->error = ENOMEM; + } else if (ctx->open == 0) { + fop->error = EBADFD; + } + if (fop->error) { + gf_msg(fop->xl->name, GF_LOG_ERROR, fop->error, + EC_MSG_INVALID_REQUEST, "EC is not winding readdir: %s", + ec_msg_str(fop)); return EC_STATE_REPORT; } diff --git a/xlators/cluster/ec/src/ec-heal.c b/xlators/cluster/ec/src/ec-heal.c index 81f6add..7d991f0 100644 --- a/xlators/cluster/ec/src/ec-heal.c +++ b/xlators/cluster/ec/src/ec-heal.c @@ -70,6 +70,7 @@ struct ec_name_data { char *name; inode_t *parent; default_args_cbk_t *replies; + uint32_t heal_pending; }; static char *ec_ignore_xattrs[] = {GF_SELINUX_XATTR_KEY, QUOTA_SIZE_KEY, NULL}; @@ -994,6 +995,7 @@ ec_set_new_entry_dirty(ec_t *ec, loc_t *loc, struct iatt *ia, ret = -ENOTCONN; goto out; } + out: if (xattr) dict_unref(xattr); @@ -1172,6 +1174,7 @@ ec_create_name(call_frame_t *frame, ec_t *ec, inode_t *parent, char *name, dict_t *xdata = NULL; char *linkname = NULL; ec_config_t config; + /* There should be just one gfid key */ EC_REPLIES_ALLOC(replies, ec->nodes); if (gfid_db->count != 1) { @@ -1416,6 +1419,11 @@ __ec_heal_name(call_frame_t *frame, ec_t *ec, inode_t *parent, char *name, ret = ec_create_name(frame, ec, parent, name, replies, gfid_db, enoent, participants); + if (ret >= 0) { + /* If ec_create_name() succeeded we return 1 to indicate that a new + * file has been created and it will need to be healed. */ + ret = 1; + } out: cluster_replies_wipe(replies, ec->nodes); loc_wipe(&loc); @@ -1493,18 +1501,22 @@ ec_name_heal_handler(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent, ret = ec_heal_name(name_data->frame, ec, parent->inode, entry->d_name, name_on); - if (ret < 0) + if (ret < 0) { memset(name_on, 0, ec->nodes); + } else { + name_data->heal_pending += ret; + } for (i = 0; i < ec->nodes; i++) if (name_data->participants[i] && !name_on[i]) name_data->failed_on[i] = 1; + return 0; } int ec_heal_names(call_frame_t *frame, ec_t *ec, inode_t *inode, - unsigned char *participants) + unsigned char *participants, uint32_t *pending) { int i = 0; int j = 0; @@ -1517,7 +1529,7 @@ ec_heal_names(call_frame_t *frame, ec_t *ec, inode_t *inode, name_data.frame = frame; name_data.participants = participants; name_data.failed_on = alloca0(ec->nodes); - ; + name_data.heal_pending = 0; for (i = 0; i < ec->nodes; i++) { if (!participants[i]) @@ -1536,6 +1548,8 @@ ec_heal_names(call_frame_t *frame, ec_t *ec, inode_t *inode, break; } } + *pending += name_data.heal_pending; + loc_wipe(&loc); return ret; } @@ -1543,7 +1557,7 @@ ec_heal_names(call_frame_t *frame, ec_t *ec, inode_t *inode, int __ec_heal_entry(call_frame_t *frame, ec_t *ec, inode_t *inode, unsigned char *heal_on, unsigned char *sources, - unsigned char *healed_sinks) + unsigned char *healed_sinks, uint32_t *pending) { unsigned char *locked_on = NULL; unsigned char *output = NULL; @@ -1588,7 +1602,7 @@ unlock: if (sources[i] || healed_sinks[i]) participants[i] = 1; } - ret = ec_heal_names(frame, ec, inode, participants); + ret = ec_heal_names(frame, ec, inode, participants, pending); if (EC_COUNT(participants, ec->nodes) <= ec->fragments) goto out; @@ -1609,7 +1623,8 @@ out: int ec_heal_entry(call_frame_t *frame, ec_t *ec, inode_t *inode, - unsigned char *sources, unsigned char *healed_sinks) + unsigned char *sources, unsigned char *healed_sinks, + uint32_t *pending) { unsigned char *locked_on = NULL; unsigned char *up_subvols = NULL; @@ -1640,7 +1655,7 @@ ec_heal_entry(call_frame_t *frame, ec_t *ec, inode_t *inode, goto unlock; } ret = __ec_heal_entry(frame, ec, inode, locked_on, sources, - healed_sinks); + healed_sinks, pending); } unlock: cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame, @@ -1961,14 +1976,14 @@ ec_manager_heal_block(ec_fop_data_t *fop, int32_t state) if (fop->cbks.heal) { fop->cbks.heal(fop->req_frame, fop->data, fop->xl, 0, 0, (heal->good | heal->bad), heal->good, heal->bad, - NULL); + 0, NULL); } return EC_STATE_END; case -EC_STATE_REPORT: if (fop->cbks.heal) { fop->cbks.heal(fop->req_frame, fop->data, fop->xl, -1, - fop->error, 0, 0, 0, NULL); + fop->error, 0, 0, 0, 0, NULL); } return EC_STATE_END; @@ -2005,14 +2020,15 @@ out: if (fop != NULL) { ec_manager(fop, error); } else { - func(frame, heal, this, -1, error, 0, 0, 0, NULL); + func(frame, heal, this, -1, error, 0, 0, 0, 0, NULL); } } int32_t ec_heal_block_done(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, uintptr_t mask, - uintptr_t good, uintptr_t bad, dict_t *xdata) + uintptr_t good, uintptr_t bad, uint32_t pending, + dict_t *xdata) { ec_heal_t *heal = cookie; @@ -2481,6 +2497,58 @@ out: return ret; } +int +ec_heal_purge_stale_index(call_frame_t *frame, ec_t *ec, inode_t *inode) +{ + int i = 0; + int ret = 0; + dict_t **xattr = NULL; + loc_t loc = {0}; + uint64_t dirty_xattr[EC_VERSION_SIZE] = {0}; + unsigned char *on = NULL; + default_args_cbk_t *replies = NULL; + dict_t *dict = NULL; + + /* Allocate the required memory */ + loc.inode = inode_ref(inode); + gf_uuid_copy(loc.gfid, inode->gfid); + on = alloca0(ec->nodes); + EC_REPLIES_ALLOC(replies, ec->nodes); + xattr = GF_CALLOC(ec->nodes, sizeof(*xattr), gf_common_mt_pointer); + if (!xattr) { + ret = -ENOMEM; + goto out; + } + dict = dict_new(); + if (!dict) { + ret = -ENOMEM; + goto out; + } + for (i = 0; i < ec->nodes; i++) { + xattr[i] = dict; + on[i] = 1; + } + ret = dict_set_static_bin(dict, EC_XATTR_DIRTY, dirty_xattr, + (sizeof(*dirty_xattr) * EC_VERSION_SIZE)); + if (ret < 0) { + ret = -ENOMEM; + goto out; + } + PARALLEL_FOP_ONLIST(ec->xl_list, on, ec->nodes, replies, frame, + ec_wind_xattrop_parallel, &loc, GF_XATTROP_ADD_ARRAY64, + xattr, NULL); +out: + if (dict) { + dict_unref(dict); + } + if (xattr) { + GF_FREE(xattr); + } + cluster_replies_wipe(replies, ec->nodes); + loc_wipe(&loc); + return ret; +} + void ec_heal_do(xlator_t *this, void *data, loc_t *loc, int32_t partial) { @@ -2498,6 +2566,7 @@ ec_heal_do(xlator_t *this, void *data, loc_t *loc, int32_t partial) intptr_t mbad = 0; intptr_t good = 0; intptr_t bad = 0; + uint32_t pending = 0; ec_fop_data_t *fop = data; gf_boolean_t blocking = _gf_false; ec_heal_need_t need_heal = EC_HEAL_NONEED; @@ -2533,7 +2602,7 @@ ec_heal_do(xlator_t *this, void *data, loc_t *loc, int32_t partial) if (loc->name && strlen(loc->name)) { ret = ec_heal_name(frame, ec, loc->parent, (char *)loc->name, participants); - if (ret == 0) { + if (ret >= 0) { gf_msg_debug(this->name, 0, "%s: name heal " "successful on %" PRIXPTR, @@ -2551,23 +2620,34 @@ ec_heal_do(xlator_t *this, void *data, loc_t *loc, int32_t partial) /* Mount triggers heal only when it detects that it must need heal, shd * triggers heals periodically which need not be thorough*/ - if (ec->shd.iamshd) { + if (ec->shd.iamshd && (ret <= 0)) { ec_heal_inspect(frame, ec, loc->inode, up_subvols, _gf_false, _gf_false, &need_heal); - if (need_heal == EC_HEAL_NONEED) { + if (need_heal == EC_HEAL_PURGE_INDEX) { + gf_msg(ec->xl->name, GF_LOG_INFO, 0, EC_MSG_HEAL_FAIL, + "Index entry needs to be purged for: %s ", + uuid_utoa(loc->gfid)); + /* We need to send zero-xattrop so that stale index entry could be + * removed. We need not take lock on this entry to do so as + * xattrop on a brick is atomic. */ + ec_heal_purge_stale_index(frame, ec, loc->inode); + goto out; + } else if (need_heal == EC_HEAL_NONEED) { gf_msg(ec->xl->name, GF_LOG_DEBUG, 0, EC_MSG_HEAL_FAIL, "Heal is not required for : %s ", uuid_utoa(loc->gfid)); goto out; } } + sources = alloca0(ec->nodes); healed_sinks = alloca0(ec->nodes); if (IA_ISREG(loc->inode->ia_type)) { ret = ec_heal_data(frame, ec, blocking, loc->inode, sources, healed_sinks); } else if (IA_ISDIR(loc->inode->ia_type) && !partial) { - ret = ec_heal_entry(frame, ec, loc->inode, sources, healed_sinks); + ret = ec_heal_entry(frame, ec, loc->inode, sources, healed_sinks, + &pending); } else { ret = 0; memcpy(sources, participants, ec->nodes); @@ -2597,10 +2677,11 @@ out: if (fop->cbks.heal) { fop->cbks.heal(fop->req_frame, fop->data, fop->xl, op_ret, op_errno, ec_char_array_to_mask(participants, ec->nodes), - mgood & good, mbad & bad, NULL); + mgood & good, mbad & bad, pending, NULL); } if (frame) STACK_DESTROY(frame->root); + return; } @@ -2648,7 +2729,7 @@ ec_heal_fail(ec_t *ec, ec_fop_data_t *fop) { if (fop->cbks.heal) { fop->cbks.heal(fop->req_frame, fop->data, ec->xl, -1, fop->error, 0, 0, - 0, NULL); + 0, 0, NULL); } ec_fop_data_release(fop); } @@ -2835,7 +2916,7 @@ fail: if (fop) ec_fop_data_release(fop); if (func) - func(frame, data, this, -1, err, 0, 0, 0, NULL); + func(frame, data, this, -1, err, 0, 0, 0, 0, NULL); } int @@ -2964,6 +3045,13 @@ _need_heal_calculate(ec_t *ec, uint64_t *dirty, unsigned char *sources, goto out; } } + /* If lock count is 0, all dirty flags are 0 and all the + * versions are macthing then why are we here. It looks + * like something went wrong while removing the index entries + * after completing a successful heal or fop. In this case + * we need to remove this index entry to avoid triggering heal + * in a loop and causing lookups again and again*/ + *need_heal = EC_HEAL_PURGE_INDEX; } else { for (i = 0; i < ec->nodes; i++) { /* Since each lock can only increment the dirty diff --git a/xlators/cluster/ec/src/ec-heald.c b/xlators/cluster/ec/src/ec-heald.c index 956e73c..5c1586b 100644 --- a/xlators/cluster/ec/src/ec-heald.c +++ b/xlators/cluster/ec/src/ec-heald.c @@ -62,7 +62,7 @@ __ec_shd_healer_wait(struct subvol_healer *healer) ec = healer->this->private; disabled_loop: - wait_till.tv_sec = time(NULL) + ec->shd.timeout; + wait_till.tv_sec = gf_time() + ec->shd.timeout; while (!healer->rerun) { ret = pthread_cond_timedwait(&healer->cond, &healer->mutex, &wait_till); @@ -156,19 +156,78 @@ ec_shd_index_purge(xlator_t *subvol, inode_t *inode, char *name) return ret; } +static gf_boolean_t +ec_is_heal_completed(char *status) +{ + char *bad_pos = NULL; + char *zero_pos = NULL; + + if (!status) { + return _gf_false; + } + + /*Logic: + * Status will be of the form Good: <binary>, Bad: <binary> + * If heal completes, if we do strchr for '0' it should be present after + * 'Bad:' i.e. strRchr for ':' + * */ + + zero_pos = strchr(status, '0'); + bad_pos = strrchr(status, ':'); + if (!zero_pos || !bad_pos) { + /*malformed status*/ + return _gf_false; + } + + if (zero_pos > bad_pos) { + return _gf_true; + } + + return _gf_false; +} + int ec_shd_selfheal(struct subvol_healer *healer, int child, loc_t *loc, gf_boolean_t full) { + dict_t *xdata = NULL; + dict_t *dict = NULL; + uint32_t count; int32_t ret; + char *heal_status = NULL; + ec_t *ec = healer->this->private; + + GF_ATOMIC_INC(ec->stats.shd.attempted); + ret = syncop_getxattr(healer->this, loc, &dict, EC_XATTR_HEAL, NULL, + &xdata); + if (ret == 0) { + if (dict && (dict_get_str(dict, EC_XATTR_HEAL, &heal_status) == 0)) { + if (ec_is_heal_completed(heal_status)) { + GF_ATOMIC_INC(ec->stats.shd.completed); + } + } + } - ret = syncop_getxattr(healer->this, loc, NULL, EC_XATTR_HEAL, NULL, NULL); - if (!full && (ret >= 0) && (loc->inode->ia_type == IA_IFDIR)) { + if (!full && (loc->inode->ia_type == IA_IFDIR)) { /* If we have just healed a directory, it's possible that - * other index entries have appeared to be healed. We put a - * mark so that we can check it later and restart a scan - * without delay. */ - healer->rerun = _gf_true; + * other index entries have appeared to be healed. */ + if ((xdata != NULL) && + (dict_get_uint32(xdata, EC_XATTR_HEAL_NEW, &count) == 0) && + (count > 0)) { + /* Force a rerun of the index healer. */ + gf_msg_debug(healer->this->name, 0, "%d more entries to heal", + count); + + healer->rerun = _gf_true; + } + } + + if (xdata != NULL) { + dict_unref(xdata); + } + + if (dict) { + dict_unref(dict); } return ret; diff --git a/xlators/cluster/ec/src/ec-inode-read.c b/xlators/cluster/ec/src/ec-inode-read.c index a891ccd..dad5f4d 100644 --- a/xlators/cluster/ec/src/ec-inode-read.c +++ b/xlators/cluster/ec/src/ec-inode-read.c @@ -390,7 +390,8 @@ ec_manager_getxattr(ec_fop_data_t *fop, int32_t state) int32_t ec_getxattr_heal_cbk(call_frame_t *frame, void *cookie, xlator_t *xl, int32_t op_ret, int32_t op_errno, uintptr_t mask, - uintptr_t good, uintptr_t bad, dict_t *xdata) + uintptr_t good, uintptr_t bad, uint32_t pending, + dict_t *xdata) { fop_getxattr_cbk_t func = cookie; ec_t *ec = xl->private; @@ -398,6 +399,25 @@ ec_getxattr_heal_cbk(call_frame_t *frame, void *cookie, xlator_t *xl, char *str; char bin1[65], bin2[65]; + /* We try to return the 'pending' information in xdata, but if this cannot + * be set, we will ignore it silently. We prefer to report the success or + * failure of the heal itself. */ + if (xdata == NULL) { + xdata = dict_new(); + } else { + dict_ref(xdata); + } + if (xdata != NULL) { + if (dict_set_uint32(xdata, EC_XATTR_HEAL_NEW, pending) != 0) { + /* dict_set_uint32() is marked as 'warn_unused_result' and gcc + * enforces to check the result in this case. However we don't + * really care if it succeeded or not. We'll just do the same. + * + * This empty 'if' avoids the warning, and it will be removed by + * the optimizer. */ + } + } + if (op_ret >= 0) { dict = dict_new(); if (dict == NULL) { @@ -431,11 +451,14 @@ ec_getxattr_heal_cbk(call_frame_t *frame, void *cookie, xlator_t *xl, } out: - func(frame, NULL, xl, op_ret, op_errno, dict, NULL); + func(frame, NULL, xl, op_ret, op_errno, dict, xdata); if (dict != NULL) { dict_unref(dict); } + if (xdata != NULL) { + dict_unref(xdata); + } return 0; } diff --git a/xlators/cluster/ec/src/ec-locks.c b/xlators/cluster/ec/src/ec-locks.c index 8e84977..601960d 100644 --- a/xlators/cluster/ec/src/ec-locks.c +++ b/xlators/cluster/ec/src/ec-locks.c @@ -24,9 +24,36 @@ ec_lock_check(ec_fop_data_t *fop, uintptr_t *mask) ec_t *ec = fop->xl->private; ec_cbk_data_t *ans = NULL; ec_cbk_data_t *cbk = NULL; - uintptr_t locked = 0, notlocked = 0; + uintptr_t locked = 0; + int32_t good = 0; + int32_t eagain = 0; + int32_t estale = 0; int32_t error = -1; + /* There are some errors that we'll handle in an special way while trying + * to acquire a lock. + * + * EAGAIN: If it's found during a parallel non-blocking lock request, we + * consider that there's contention on the inode, so we consider + * the acquisition a failure and try again with a sequential + * blocking lock request. This will ensure that we get a lock on + * as many bricks as possible (ignoring EAGAIN here would cause + * unnecessary triggers of self-healing). + * + * If it's found during a sequential blocking lock request, it's + * considered an error. Lock will only succeed if there are + * enough other bricks locked. + * + * ESTALE: This can appear during parallel or sequential lock request if + * the inode has just been unlinked. We consider this error is + * not recoverable, but we also don't consider it as fatal. So, + * if it happens during parallel lock, we won't attempt a + * sequential one unless there are EAGAIN errors on other + * bricks (and are enough to form a quorum), but if we reach + * quorum counting the ESTALE bricks, we consider the whole + * result of the operation is ESTALE instead of EIO. + */ + list_for_each_entry(ans, &fop->cbk_list, list) { if (ans->op_ret >= 0) { @@ -34,24 +61,23 @@ ec_lock_check(ec_fop_data_t *fop, uintptr_t *mask) error = EIO; } locked |= ans->mask; + good = ans->count; cbk = ans; - } else { - if (ans->op_errno == EAGAIN) { - switch (fop->uint32) { - case EC_LOCK_MODE_NONE: - case EC_LOCK_MODE_ALL: - /* Goal is to treat non-blocking lock as failure - * even if there is a single EAGAIN*/ - notlocked |= ans->mask; - break; - } - } + } else if (ans->op_errno == ESTALE) { + estale += ans->count; + } else if ((ans->op_errno == EAGAIN) && + (fop->uint32 != EC_LOCK_MODE_INC)) { + eagain += ans->count; } } if (error == -1) { - if (gf_bits_count(locked | notlocked) >= ec->fragments) { - if (notlocked == 0) { + /* If we have enough quorum with succeeded and EAGAIN answers, we + * ignore for now any ESTALE answer. If there are EAGAIN answers, + * we retry with a sequential blocking lock request if needed. + * Otherwise we succeed. */ + if ((good + eagain) >= ec->fragments) { + if (eagain == 0) { if (fop->answer == NULL) { fop->answer = cbk; } @@ -64,21 +90,28 @@ ec_lock_check(ec_fop_data_t *fop, uintptr_t *mask) case EC_LOCK_MODE_NONE: error = EAGAIN; break; - case EC_LOCK_MODE_ALL: fop->uint32 = EC_LOCK_MODE_INC; break; - default: + /* This shouldn't happen because eagain cannot be > 0 + * when fop->uint32 is EC_LOCK_MODE_INC. */ error = EIO; break; } } } else { - if (fop->answer && fop->answer->op_ret < 0) + /* We have been unable to find enough candidates that will be able + * to take the lock. If we have quorum on some answer, we return + * it. Otherwise we check if ESTALE answers allow us to reach + * quorum. If so, we return ESTALE. */ + if (fop->answer && fop->answer->op_ret < 0) { error = fop->answer->op_errno; - else + } else if ((good + eagain + estale) >= ec->fragments) { + error = ESTALE; + } else { error = EIO; + } } } diff --git a/xlators/cluster/ec/src/ec-types.h b/xlators/cluster/ec/src/ec-types.h index 7829b8c..de9b89b 100644 --- a/xlators/cluster/ec/src/ec-types.h +++ b/xlators/cluster/ec/src/ec-types.h @@ -130,7 +130,12 @@ typedef void (*ec_resume_f)(ec_fop_data_t *, int32_t); enum _ec_read_policy { EC_ROUND_ROBIN, EC_GFID_HASH, EC_READ_POLICY_MAX }; -enum _ec_heal_need { EC_HEAL_NONEED, EC_HEAL_MAYBE, EC_HEAL_MUST }; +enum _ec_heal_need { + EC_HEAL_NONEED, + EC_HEAL_MAYBE, + EC_HEAL_MUST, + EC_HEAL_PURGE_INDEX +}; enum _ec_stripe_part { EC_STRIPE_HEAD, EC_STRIPE_TAIL }; @@ -186,10 +191,10 @@ struct _ec_inode { typedef int32_t (*fop_heal_cbk_t)(call_frame_t *, void *, xlator_t *, int32_t, int32_t, uintptr_t, uintptr_t, uintptr_t, - dict_t *); + uint32_t, dict_t *); typedef int32_t (*fop_fheal_cbk_t)(call_frame_t *, void *, xlator_t *, int32_t, int32_t, uintptr_t, uintptr_t, uintptr_t, - dict_t *); + uint32_t, dict_t *); union _ec_cbk { fop_access_cbk_t access; @@ -621,6 +626,11 @@ struct _ec_statistics { requests. (Basically memory allocation errors). */ } stripe_cache; + struct { + gf_atomic_t attempted; /*Number of heals attempted on + files/directories*/ + gf_atomic_t completed; /*Number of heals complted on files/directories*/ + } shd; }; struct _ec { diff --git a/xlators/cluster/ec/src/ec.c b/xlators/cluster/ec/src/ec.c index 66b4e63..7344be4 100644 --- a/xlators/cluster/ec/src/ec.c +++ b/xlators/cluster/ec/src/ec.c @@ -325,13 +325,18 @@ ec_get_event_from_state(ec_t *ec) void ec_up(xlator_t *this, ec_t *ec) { + char str1[32], str2[32]; + if (ec->timer != NULL) { gf_timer_call_cancel(this->ctx, ec->timer); ec->timer = NULL; } ec->up = 1; - gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_EC_UP, "Going UP"); + gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_EC_UP, + "Going UP : Child UP = %s Child Notify = %s", + ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes), + ec_bin(str2, sizeof(str2), ec->xl_notify, ec->nodes)); gf_event(EVENT_EC_MIN_BRICKS_UP, "subvol=%s", this->name); } @@ -339,13 +344,18 @@ ec_up(xlator_t *this, ec_t *ec) void ec_down(xlator_t *this, ec_t *ec) { + char str1[32], str2[32]; + if (ec->timer != NULL) { gf_timer_call_cancel(this->ctx, ec->timer); ec->timer = NULL; } ec->up = 0; - gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_EC_DOWN, "Going DOWN"); + gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_EC_DOWN, + "Going DOWN : Child UP = %s Child Notify = %s", + ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes), + ec_bin(str2, sizeof(str2), ec->xl_notify, ec->nodes)); gf_event(EVENT_EC_MIN_BRICKS_NOT_UP, "subvol=%s", this->name); } @@ -700,6 +710,8 @@ ec_statistics_init(ec_t *ec) GF_ATOMIC_INIT(ec->stats.stripe_cache.evicts, 0); GF_ATOMIC_INIT(ec->stats.stripe_cache.allocs, 0); GF_ATOMIC_INIT(ec->stats.stripe_cache.errors, 0); + GF_ATOMIC_INIT(ec->stats.shd.attempted, 0); + GF_ATOMIC_INIT(ec->stats.shd.completed, 0); } static int @@ -1569,6 +1581,10 @@ ec_dump_private(xlator_t *this) GF_ATOMIC_GET(ec->stats.stripe_cache.allocs)); gf_proc_dump_write("errors", "%" GF_PRI_ATOMIC, GF_ATOMIC_GET(ec->stats.stripe_cache.errors)); + gf_proc_dump_write("heals-attempted", "%" GF_PRI_ATOMIC, + GF_ATOMIC_GET(ec->stats.shd.attempted)); + gf_proc_dump_write("heals-completed", "%" GF_PRI_ATOMIC, + GF_ATOMIC_GET(ec->stats.shd.completed)); return 0; } diff --git a/xlators/cluster/ec/src/ec.h b/xlators/cluster/ec/src/ec.h index 1b210d9..6f6de6d 100644 --- a/xlators/cluster/ec/src/ec.h +++ b/xlators/cluster/ec/src/ec.h @@ -18,6 +18,7 @@ #define EC_XATTR_SIZE EC_XATTR_PREFIX "size" #define EC_XATTR_VERSION EC_XATTR_PREFIX "version" #define EC_XATTR_HEAL EC_XATTR_PREFIX "heal" +#define EC_XATTR_HEAL_NEW EC_XATTR_PREFIX "heal-new" #define EC_XATTR_DIRTY EC_XATTR_PREFIX "dirty" #define EC_STRIPE_CACHE_MAX_SIZE 10 #define EC_VERSION_SIZE 2 diff --git a/xlators/debug/error-gen/src/error-gen.c b/xlators/debug/error-gen/src/error-gen.c index ff993f7..d45655e 100644 --- a/xlators/debug/error-gen/src/error-gen.c +++ b/xlators/debug/error-gen/src/error-gen.c @@ -31,9 +31,9 @@ sys_error_t error_no_list[] = { [GF_FOP_LOOKUP] = {.error_no_count = 4, .error_no = {ENOENT, ENOTDIR, ENAMETOOLONG, EAGAIN}}, - [GF_FOP_STAT] = {.error_no_count = 7, - .error_no = {EACCES, EBADF, EFAULT, ENAMETOOLONG, ENOENT, - ENOMEM, ENOTDIR}}, + [GF_FOP_STAT] = {.error_no_count = 6, + .error_no = {EACCES, EFAULT, ENAMETOOLONG, ENOENT, ENOMEM, + ENOTDIR}}, [GF_FOP_READLINK] = {.error_no_count = 8, .error_no = {EACCES, EFAULT, EINVAL, EIO, ENAMETOOLONG, ENOENT, ENOMEM, ENOTDIR}}, @@ -79,21 +79,20 @@ sys_error_t error_no_list[] = { [GF_FOP_WRITE] = {.error_no_count = 7, .error_no = {EINVAL, EBADF, EFAULT, EISDIR, ENAMETOOLONG, ENOSPC, GF_ERROR_SHORT_WRITE}}, - [GF_FOP_STATFS] = {.error_no_count = 10, - .error_no = {EACCES, EBADF, EFAULT, EINTR, EIO, - ENAMETOOLONG, ENOENT, ENOMEM, ENOSYS, - ENOTDIR}}, + [GF_FOP_STATFS] = {.error_no_count = 9, + .error_no = {EACCES, EFAULT, EINTR, EIO, ENAMETOOLONG, + ENOENT, ENOMEM, ENOSYS, ENOTDIR}}, [GF_FOP_FLUSH] = {.error_no_count = 5, .error_no = {EACCES, EFAULT, ENAMETOOLONG, ENOSYS, ENOENT}}, [GF_FOP_FSYNC] = {.error_no_count = 4, .error_no = {EBADF, EIO, EROFS, EINVAL}}, - [GF_FOP_SETXATTR] = {.error_no_count = 4, - .error_no = {EACCES, EBADF, EINTR, ENAMETOOLONG}}, - [GF_FOP_GETXATTR] = {.error_no_count = 4, - .error_no = {EACCES, EBADF, ENAMETOOLONG, EINTR}}, - [GF_FOP_REMOVEXATTR] = {.error_no_count = 4, - .error_no = {EACCES, EBADF, ENAMETOOLONG, EINTR}}, + [GF_FOP_SETXATTR] = {.error_no_count = 3, + .error_no = {EACCES, EINTR, ENAMETOOLONG}}, + [GF_FOP_GETXATTR] = {.error_no_count = 3, + .error_no = {EACCES, ENAMETOOLONG, EINTR}}, + [GF_FOP_REMOVEXATTR] = {.error_no_count = 3, + .error_no = {EACCES, ENAMETOOLONG, EINTR}}, [GF_FOP_FSETXATTR] = {.error_no_count = 4, .error_no = {EACCES, EBADF, EINTR, ENAMETOOLONG}}, [GF_FOP_FGETXATTR] = {.error_no_count = 4, @@ -125,26 +124,25 @@ sys_error_t error_no_list[] = { ENOENT}}, [GF_FOP_FXATTROP] = {.error_no_count = 4, .error_no = {EBADF, EIO, EROFS, EINVAL}}, - [GF_FOP_INODELK] = {.error_no_count = 4, - .error_no = {EACCES, EBADF, EINTR, ENAMETOOLONG}}, + [GF_FOP_INODELK] = {.error_no_count = 3, + .error_no = {EACCES, EINTR, ENAMETOOLONG}}, [GF_FOP_FINODELK] = {.error_no_count = 4, .error_no = {EACCES, EBADF, EINTR, ENAMETOOLONG}}, - [GF_FOP_ENTRYLK] = {.error_no_count = 4, - .error_no = {EACCES, EBADF, ENAMETOOLONG, EINTR}}, + [GF_FOP_ENTRYLK] = {.error_no_count = 3, + .error_no = {EACCES, ENAMETOOLONG, EINTR}}, [GF_FOP_FENTRYLK] = {.error_no_count = 10, .error_no = {EACCES, EEXIST, EFAULT, EISDIR, EMFILE, ENAMETOOLONG, ENFILE, ENODEV, ENOENT, ENOMEM}}, - [GF_FOP_SETATTR] = {.error_no_count = 11, + [GF_FOP_SETATTR] = {.error_no_count = 10, .error_no = {EACCES, EFAULT, EIO, ENAMETOOLONG, ENOENT, - ENOMEM, ENOTDIR, EPERM, EROFS, EBADF, - EIO}}, + ENOMEM, ENOTDIR, EPERM, EROFS, EIO}}, [GF_FOP_FSETATTR] = {.error_no_count = 11, .error_no = {EACCES, EFAULT, EIO, ENAMETOOLONG, ENOENT, ENOMEM, ENOTDIR, EPERM, EROFS, EBADF, EIO}}, - [GF_FOP_GETSPEC] = {.error_no_count = 4, - .error_no = {EACCES, EBADF, ENAMETOOLONG, EINTR}}}; + [GF_FOP_GETSPEC] = {.error_no_count = 3, + .error_no = {EACCES, ENAMETOOLONG, EINTR}}}; int generate_rand_no(int op_no) @@ -1509,8 +1507,8 @@ init(xlator_t *this) this->private = pvt; - /* Give some seed value here */ - srand(time(NULL)); + /* Give some seed value here. */ + srand(gf_time()); ret = 0; out: diff --git a/xlators/debug/io-stats/src/io-stats.c b/xlators/debug/io-stats/src/io-stats.c index 4e1f6e5..aa00c44 100644 --- a/xlators/debug/io-stats/src/io-stats.c +++ b/xlators/debug/io-stats/src/io-stats.c @@ -135,7 +135,7 @@ struct ios_global_stats { gf_atomic_t block_count_read[IOS_BLOCK_COUNT_SIZE]; gf_atomic_t fop_hits[GF_FOP_MAXVALUE]; gf_atomic_t upcall_hits[GF_UPCALL_FLAGS_MAXVALUE]; - struct timeval started_at; + time_t started_at; struct ios_lat latency[GF_FOP_MAXVALUE]; uint64_t nr_opens; uint64_t max_nr_opens; @@ -292,9 +292,7 @@ is_fop_latency_started(call_frame_t *frame) begin = &frame->begin; \ end = &frame->end; \ \ - elapsed = ((end->tv_sec - begin->tv_sec) * 1e9 + \ - (end->tv_nsec - begin->tv_nsec)) / \ - 1000; \ + elapsed = gf_tsdiff(begin, end) / 1000.0; \ throughput = op_ret / elapsed; \ \ conf = this->private; \ @@ -678,10 +676,7 @@ ios_dump_throughput_stats(struct ios_stat_head *list_head, xlator_t *this, FILE *logfp, ios_stats_thru_t type) { struct ios_stat_list *entry = NULL; - struct timeval time = { - 0, - }; - char timestr[256] = { + char timestr[GF_TIMESTR_SIZE] = { 0, }; @@ -689,12 +684,9 @@ ios_dump_throughput_stats(struct ios_stat_head *list_head, xlator_t *this, { list_for_each_entry(entry, &list_head->iosstats->list, list) { - gf_time_fmt(timestr, sizeof timestr, - entry->iosstat->thru_counters[type].time.tv_sec, - gf_timefmt_FT); - snprintf(timestr + strlen(timestr), - sizeof timestr - strlen(timestr), ".%" GF_PRI_SUSECONDS, - time.tv_usec); + gf_time_fmt_tv(timestr, sizeof timestr, + &entry->iosstat->thru_counters[type].time, + gf_timefmt_FT); ios_log(this, logfp, "%s \t %-10.2f \t %s", timestr, entry->value, entry->iosstat->filename); @@ -773,9 +765,8 @@ err: int io_stats_dump_global_to_json_logfp(xlator_t *this, - struct ios_global_stats *stats, - struct timeval *now, int interval, - FILE *logfp) + struct ios_global_stats *stats, time_t now, + int interval, FILE *logfp) { int i = 0; int j = 0; @@ -801,10 +792,7 @@ io_stats_dump_global_to_json_logfp(xlator_t *this, }; dict_t *xattr = NULL; - interval_sec = ((now->tv_sec * 1000000.0 + now->tv_usec) - - (stats->started_at.tv_sec * 1000000.0 + - stats->started_at.tv_usec)) / - 1000000.0; + interval_sec = (double)(now - stats->started_at); conf = this->private; @@ -956,8 +944,8 @@ io_stats_dump_global_to_json_logfp(xlator_t *this, } if (interval == -1) { - ios_log(this, logfp, "\"%s.%s.uptime\": %" PRId64 ",", key_prefix, - str_prefix, (uint64_t)(now->tv_sec - stats->started_at.tv_sec)); + ios_log(this, logfp, "\"%s.%s.uptime\": %" PRIu64 ",", key_prefix, + str_prefix, (uint64_t)(now - stats->started_at)); ios_log(this, logfp, "\"%s.%s.bytes_read\": " "%" GF_PRI_ATOMIC ",", @@ -1209,14 +1197,14 @@ out: int io_stats_dump_global_to_logfp(xlator_t *this, struct ios_global_stats *stats, - struct timeval *now, int interval, FILE *logfp) + time_t now, int interval, FILE *logfp) { int i = 0; int per_line = 0; int index = 0; struct ios_stat_head *list_head = NULL; struct ios_conf *conf = NULL; - char timestr[256] = { + char timestr[GF_TIMESTR_SIZE] = { 0, }; char str_header[128] = {0}; @@ -1232,8 +1220,8 @@ io_stats_dump_global_to_logfp(xlator_t *this, struct ios_global_stats *stats, ios_log(this, logfp, "\n=== Cumulative stats ==="); else ios_log(this, logfp, "\n=== Interval %d stats ===", interval); - ios_log(this, logfp, " Duration : %" PRId64 " secs", - (uint64_t)(now->tv_sec - stats->started_at.tv_sec)); + ios_log(this, logfp, " Duration : %" PRIu64 " secs", + (uint64_t)(now - stats->started_at)); ios_log(this, logfp, " BytesRead : %" GF_PRI_ATOMIC, GF_ATOMIC_GET(stats->data_read)); ios_log(this, logfp, " BytesWritten : %" GF_PRI_ATOMIC "\n", @@ -1325,11 +1313,8 @@ io_stats_dump_global_to_logfp(xlator_t *this, struct ios_global_stats *stats, if (interval == -1) { LOCK(&conf->lock); { - gf_time_fmt(timestr, sizeof timestr, - conf->cumulative.max_openfd_time.tv_sec, gf_timefmt_FT); - snprintf(timestr + strlen(timestr), - sizeof timestr - strlen(timestr), ".%" GF_PRI_SUSECONDS, - conf->cumulative.max_openfd_time.tv_usec); + gf_time_fmt_tv(timestr, sizeof timestr, + &conf->cumulative.max_openfd_time, gf_timefmt_FT); ios_log(this, logfp, "Current open fd's: %" PRId64 " Max open fd's: %" PRId64 " time %s", @@ -1381,7 +1366,7 @@ io_stats_dump_global_to_logfp(xlator_t *this, struct ios_global_stats *stats, int io_stats_dump_global_to_dict(xlator_t *this, struct ios_global_stats *stats, - struct timeval *now, int interval, dict_t *dict) + time_t now, int interval, dict_t *dict) { int ret = 0; char key[64] = {0}; @@ -1407,7 +1392,7 @@ io_stats_dump_global_to_dict(xlator_t *this, struct ios_global_stats *stats, interval); snprintf(key, sizeof(key), "%d-duration", interval); - sec = (uint64_t)(now->tv_sec - stats->started_at.tv_sec); + sec = now - stats->started_at; ret = dict_set_uint64(dict, key, sec); if (ret) { gf_log(this->name, GF_LOG_ERROR, @@ -1530,9 +1515,8 @@ out: } int -io_stats_dump_global(xlator_t *this, struct ios_global_stats *stats, - struct timeval *now, int interval, - struct ios_dump_args *args) +io_stats_dump_global(xlator_t *this, struct ios_global_stats *stats, time_t now, + int interval, struct ios_dump_args *args) { int ret = -1; @@ -1590,13 +1574,13 @@ ios_dump_args_init(struct ios_dump_args *args, ios_dump_type_t type, } static void -ios_global_stats_clear(struct ios_global_stats *stats, struct timeval *now) +ios_global_stats_clear(struct ios_global_stats *stats, time_t now) { GF_ASSERT(stats); GF_ASSERT(now); memset(stats, 0, sizeof(*stats)); - stats->started_at = *now; + stats->started_at = now; } int @@ -1607,7 +1591,7 @@ io_stats_dump(xlator_t *this, struct ios_dump_args *args, ios_info_op_t op, struct ios_global_stats cumulative = {}; struct ios_global_stats incremental = {}; int increment = 0; - struct timeval now; + time_t now = 0; GF_ASSERT(this); GF_ASSERT(args); @@ -1615,8 +1599,8 @@ io_stats_dump(xlator_t *this, struct ios_dump_args *args, ios_info_op_t op, GF_ASSERT(args->type < IOS_DUMP_TYPE_MAX); conf = this->private; + now = gf_time(); - gettimeofday(&now, NULL); LOCK(&conf->lock); { if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_CUMULATIVE) @@ -1629,17 +1613,17 @@ io_stats_dump(xlator_t *this, struct ios_dump_args *args, ios_info_op_t op, if (!is_peek) { increment = conf->increment++; - ios_global_stats_clear(&conf->incremental, &now); + ios_global_stats_clear(&conf->incremental, now); } } } UNLOCK(&conf->lock); if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_CUMULATIVE) - io_stats_dump_global(this, &cumulative, &now, -1, args); + io_stats_dump_global(this, &cumulative, now, -1, args); if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_INCREMENTAL) - io_stats_dump_global(this, &incremental, &now, increment, args); + io_stats_dump_global(this, &incremental, now, increment, args); return 0; } @@ -1649,9 +1633,8 @@ io_stats_dump_fd(xlator_t *this, struct ios_fd *iosfd) { struct ios_conf *conf = NULL; struct timeval now; - uint64_t sec = 0; - uint64_t usec = 0; int i = 0; + double usecs = 0; uint64_t data_read = 0; uint64_t data_written = 0; uint64_t block_count_read = 0; @@ -1666,23 +1649,15 @@ io_stats_dump_fd(xlator_t *this, struct ios_fd *iosfd) return 0; gettimeofday(&now, NULL); - - if (iosfd->opened_at.tv_usec > now.tv_usec) { - now.tv_usec += 1000000; - now.tv_usec--; - } - - sec = now.tv_sec - iosfd->opened_at.tv_sec; - usec = now.tv_usec - iosfd->opened_at.tv_usec; + usecs = gf_tvdiff(&iosfd->opened_at, &now); gf_log(this->name, GF_LOG_INFO, "--- fd stats ---"); if (iosfd->filename) gf_log(this->name, GF_LOG_INFO, " Filename : %s", iosfd->filename); - if (sec) - gf_log(this->name, GF_LOG_INFO, - " Lifetime : %" PRId64 "secs, %" PRId64 "usecs", sec, usec); + if (usecs) + gf_log(this->name, GF_LOG_INFO, " Lifetime : %lf secs", usecs); data_read = GF_ATOMIC_GET(iosfd->data_read); if (data_read) @@ -1785,9 +1760,7 @@ update_ios_latency(struct ios_conf *conf, call_frame_t *frame, begin = &frame->begin; end = &frame->end; - elapsed = ((end->tv_sec - begin->tv_sec) * 1e9 + - (end->tv_nsec - begin->tv_nsec)) / - 1000; + elapsed = gf_tsdiff(begin, end) / 1000.0; update_ios_latency_stats(&conf->cumulative, elapsed, op); update_ios_latency_stats(&conf->incremental, elapsed, op); @@ -1808,7 +1781,7 @@ io_stats_dump_stats_to_dict(xlator_t *this, dict_t *resp, struct ios_stat_list *entry = NULL; int ret = -1; ios_stats_thru_t index = IOS_STATS_THRU_MAX; - char timestr[256] = { + char timestr[GF_TIMESTR_SIZE] = { 0, }; char *dict_timestr = NULL; @@ -1827,14 +1800,9 @@ io_stats_dump_stats_to_dict(xlator_t *this, dict_t *resp, ret = dict_set_uint64(resp, "max-open", conf->cumulative.max_nr_opens); - gf_time_fmt(timestr, sizeof timestr, - conf->cumulative.max_openfd_time.tv_sec, - gf_timefmt_FT); - if (conf->cumulative.max_openfd_time.tv_sec) - snprintf(timestr + strlen(timestr), - sizeof timestr - strlen(timestr), - ".%" GF_PRI_SUSECONDS, - conf->cumulative.max_openfd_time.tv_usec); + gf_time_fmt_tv(timestr, sizeof timestr, + &conf->cumulative.max_openfd_time, + gf_timefmt_FT); dict_timestr = gf_strdup(timestr); if (!dict_timestr) @@ -3606,26 +3574,21 @@ ios_destroy_top_stats(struct ios_conf *conf) return; } -static int +static void io_stats_clear(struct ios_conf *conf) { - struct timeval now; - int ret = -1; + time_t now = 0; GF_ASSERT(conf); + now = gf_time(); - if (!gettimeofday(&now, NULL)) { - LOCK(&conf->lock); - { - ios_global_stats_clear(&conf->cumulative, &now); - ios_global_stats_clear(&conf->incremental, &now); - conf->increment = 0; - } - UNLOCK(&conf->lock); - ret = 0; + LOCK(&conf->lock); + { + ios_global_stats_clear(&conf->cumulative, now); + ios_global_stats_clear(&conf->incremental, now); + conf->increment = 0; } - - return ret; + UNLOCK(&conf->lock); } int32_t @@ -3866,7 +3829,7 @@ ios_conf_destroy(struct ios_conf *conf) _ios_destroy_dump_thread(conf); ios_destroy_sample_buf(conf->ios_sample_buf); LOCK_DESTROY(&conf->lock); - GF_FREE(conf->dnscache); + gf_dnscache_deinit(conf->dnscache); GF_FREE(conf); } @@ -3889,7 +3852,7 @@ ios_init_stats(struct ios_global_stats *stats) for (i = 0; i < GF_UPCALL_FLAGS_MAXVALUE; i++) GF_ATOMIC_INIT(stats->upcall_hits[i], 0); - gettimeofday(&stats->started_at, NULL); + stats->started_at = gf_time(); } int @@ -3978,11 +3941,14 @@ init(xlator_t *this) gf_log(this->name, GF_LOG_ERROR, "Out of memory."); goto out; } - ret = -1; GF_OPTION_INIT("ios-dnscache-ttl-sec", conf->ios_dnscache_ttl_sec, int32, out); conf->dnscache = gf_dnscache_init(conf->ios_dnscache_ttl_sec); + if (!conf->dnscache) { + ret = -1; + goto out; + } GF_OPTION_INIT("sys-log-level", sys_log_str, str, out); if (sys_log_str) { @@ -4133,12 +4099,9 @@ notify(xlator_t *this, int32_t event, void *data, ...) } if (GF_IOS_INFO_CLEAR == op) { - ret = io_stats_clear(this->private); - if (ret) - gf_log(this->name, GF_LOG_ERROR, - "Failed to clear info stats"); + io_stats_clear(this->private); - ret = dict_set_int32(output, "stats-cleared", ret ? 0 : 1); + ret = dict_set_int32(output, "stats-cleared", 1); if (ret) gf_log(this->name, GF_LOG_ERROR, "Failed to set stats-cleared" diff --git a/xlators/debug/trace/src/trace.c b/xlators/debug/trace/src/trace.c index 3db2e26..6ed0ca0 100644 --- a/xlators/debug/trace/src/trace.c +++ b/xlators/debug/trace/src/trace.c @@ -22,13 +22,13 @@ static void trace_stat_to_str(struct iatt *buf, char *str, size_t len) { - char atime_buf[200] = { + char atime_buf[GF_TIMESTR_SIZE] = { 0, }; - char mtime_buf[200] = { + char mtime_buf[GF_TIMESTR_SIZE] = { 0, }; - char ctime_buf[200] = { + char ctime_buf[GF_TIMESTR_SIZE] = { 0, }; @@ -64,7 +64,7 @@ trace_stat_to_str(struct iatt *buf, char *str, size_t len) int dump_history_trace(circular_buffer_t *cb, void *data) { - char timestr[256] = { + char timestr[GF_TIMESTR_SIZE] = { 0, }; @@ -72,9 +72,7 @@ dump_history_trace(circular_buffer_t *cb, void *data) gettimeofday () fails, it's safe to check tm and then dump the time at which the entry was added to the buffer */ - gf_time_fmt(timestr, sizeof timestr, cb->tv.tv_sec, gf_timefmt_Ymd_T); - snprintf(timestr + strlen(timestr), 256 - strlen(timestr), - ".%" GF_PRI_SUSECONDS, cb->tv.tv_usec); + gf_time_fmt_tv(timestr, sizeof timestr, &cb->tv, gf_timefmt_Ymd_T); gf_proc_dump_write("TIME", "%s", timestr); gf_proc_dump_write("FOP", "%s\n", (char *)cb->data); @@ -2209,10 +2207,10 @@ int trace_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf, int32_t valid, dict_t *xdata) { - char actime_str[256] = { + char actime_str[GF_TIMESTR_SIZE] = { 0, }; - char modtime_str[256] = { + char modtime_str[GF_TIMESTR_SIZE] = { 0, }; trace_conf_t *conf = NULL; @@ -2278,10 +2276,10 @@ int trace_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf, int32_t valid, dict_t *xdata) { - char actime_str[256] = { + char actime_str[GF_TIMESTR_SIZE] = { 0, }; - char modtime_str[256] = { + char modtime_str[GF_TIMESTR_SIZE] = { 0, }; trace_conf_t *conf = NULL; diff --git a/xlators/features/Makefile.am b/xlators/features/Makefile.am index 194634b..c57897f 100644 --- a/xlators/features/Makefile.am +++ b/xlators/features/Makefile.am @@ -2,9 +2,13 @@ if BUILD_CLOUDSYNC CLOUDSYNC_DIR = cloudsync endif +if BUILD_METADISP + METADISP_DIR = metadisp +endif + SUBDIRS = locks quota read-only quiesce marker index barrier arbiter upcall \ compress changelog gfid-access snapview-client snapview-server trash \ shard bit-rot leases selinux sdfs namespace $(CLOUDSYNC_DIR) thin-arbiter \ - utime + utime $(METADISP_DIR) CLEANFILES = diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h b/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h index 3d40089..5bc5103 100644 --- a/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h +++ b/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h @@ -47,6 +47,55 @@ GLFS_MSGID(BITROT_BITD, BRB_MSG_FD_CREATE_FAILED, BRB_MSG_READV_FAILED, BRB_MSG_SCRUB_THREAD_CLEANUP, BRB_MSG_SCRUBBER_CLEANED, BRB_MSG_GENERIC_SSM_INFO, BRB_MSG_ZERO_TIMEOUT_BUG, BRB_MSG_BAD_OBJ_READDIR_FAIL, BRB_MSG_SSM_FAILED, - BRB_MSG_SCRUB_WAIT_FAILED); + BRB_MSG_SCRUB_WAIT_FAILED, BRB_MSG_TRIGGER_SIGN_FAILED, + BRB_MSG_EVENT_UNHANDLED, BRB_MSG_COULD_NOT_SCHEDULE_SCRUB, + BRB_MSG_THREAD_CREATION_FAILED, BRB_MSG_MEM_POOL_ALLOC, + BRB_MSG_SAVING_HASH_FAILED); +#define BRB_MSG_FD_CREATE_FAILED_STR "failed to create fd for the inode" +#define BRB_MSG_READV_FAILED_STR "readv failed" +#define BRB_MSG_BLOCK_READ_FAILED_STR "reading block failed" +#define BRB_MSG_NO_MEMORY_STR "failed to allocate memory" +#define BRB_MSG_CALC_CHECKSUM_FAILED_STR "calculating checksum failed" +#define BRB_MSG_GET_SIGN_FAILED_STR "failed to get the signature" +#define BRB_MSG_SET_SIGN_FAILED_STR "signing failed" +#define BRB_MSG_OP_FAILED_STR "failed on object" +#define BRB_MSG_TRIGGER_SIGN_FAILED_STR "Could not trigger signing" +#define BRB_MSG_READ_AND_SIGN_FAILED_STR "reading and signing of object failed" +#define BRB_MSG_SET_TIMER_FAILED_STR "Failed to allocate object expiry timer" +#define BRB_MSG_GET_SUBVOL_FAILED_STR \ + "failed to get the subvolume for the brick" +#define BRB_MSG_PATH_FAILED_STR "path failed" +#define BRB_MSG_SKIP_OBJECT_STR "Entry is marked corrupted. skipping" +#define BRB_MSG_PARTIAL_VERSION_PRESENCE_STR \ + "PArtial version xattr presence detected, ignoring" +#define BRB_MSG_TRIGGER_SIGN_STR "Triggering signing" +#define BRB_MSG_CRAWLING_START_STR \ + "Crawling brick, scanning for unsigned objects" +#define BRB_MSG_CRAWLING_FINISH_STR "Completed crawling brick" +#define BRB_MSG_REGISTER_FAILED_STR "Register to changelog failed" +#define BRB_MSG_SPAWN_FAILED_STR "failed to spawn" +#define BRB_MSG_CONNECTED_TO_BRICK_STR "Connected to brick" +#define BRB_MSG_LOOKUP_FAILED_STR "lookup on root failed" +#define BRB_MSG_GET_INFO_FAILED_STR "failed to get stub info" +#define BRB_MSG_SCRUB_THREAD_CLEANUP_STR "Error cleaning up scanner thread" +#define BRB_MSG_SCRUBBER_CLEANED_STR "clened up scrubber for brick" +#define BRB_MSG_SUBVOL_CONNECT_FAILED_STR \ + "callback handler for subvolume failed" +#define BRB_MSG_MEM_ACNT_FAILED_STR "Memory accounting init failed" +#define BRB_MSG_EVENT_UNHANDLED_STR "Event unhandled for child" +#define BRB_MSG_INVALID_SUBVOL_STR "Got event from invalid subvolume" +#define BRB_MSG_RESCHEDULE_SCRUBBER_FAILED_STR \ + "on demand scrub schedule failed. Scrubber is not in pending state." +#define BRB_MSG_COULD_NOT_SCHEDULE_SCRUB_STR \ + "Could not schedule ondemand scrubbing. Scrubbing will continue " \ + "according to old frequency." +#define BRB_MSG_THREAD_CREATION_FAILED_STR "thread creation failed" +#define BRB_MSG_RATE_LIMIT_INFO_STR "Rate Limit Info" +#define BRB_MSG_MEM_POOL_ALLOC_STR "failed to allocate mem-pool for timer" +#define BRB_MSG_NO_CHILD_STR "FATAL: no children" +#define BRB_MSG_TIMER_WHEEL_UNAVAILABLE_STR "global timer wheel unavailable" +#define BRB_MSG_BITROT_LOADED_STR "bit-rot xlator loaded" +#define BRB_MSG_SAVING_HASH_FAILED_STR \ + "failed to allocate memory for saving hash of the object" #endif /* !_BITROT_BITD_MESSAGES_H_ */ diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c index 34e20f9..5cef2ff 100644 --- a/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c +++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c @@ -40,21 +40,21 @@ br_inc_scrubbed_file(br_scrub_stats_t *scrub_stat) } void -br_update_scrub_start_time(br_scrub_stats_t *scrub_stat, struct timeval *tv) +br_update_scrub_start_time(br_scrub_stats_t *scrub_stat, time_t time) { if (!scrub_stat) return; pthread_mutex_lock(&scrub_stat->lock); { - scrub_stat->scrub_start_tv.tv_sec = tv->tv_sec; + scrub_stat->scrub_start_time = time; } pthread_mutex_unlock(&scrub_stat->lock); } void br_update_scrub_finish_time(br_scrub_stats_t *scrub_stat, char *timestr, - struct timeval *tv) + time_t time) { int lst_size = 0; @@ -67,10 +67,10 @@ br_update_scrub_finish_time(br_scrub_stats_t *scrub_stat, char *timestr, pthread_mutex_lock(&scrub_stat->lock); { - scrub_stat->scrub_end_tv.tv_sec = tv->tv_sec; + scrub_stat->scrub_end_time = time; - scrub_stat->scrub_duration = scrub_stat->scrub_end_tv.tv_sec - - scrub_stat->scrub_start_tv.tv_sec; + scrub_stat->scrub_duration = scrub_stat->scrub_end_time - + scrub_stat->scrub_start_time; snprintf(scrub_stat->last_scrub_time, lst_size, "%s", timestr); } diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h index 24128b9..f022aa8 100644 --- a/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h +++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h @@ -15,20 +15,22 @@ #include <sys/time.h> #include <pthread.h> +#include <glusterfs/common-utils.h> + struct br_scrub_stats { - uint64_t scrubbed_files; /* Total number of scrubbed file */ + uint64_t scrubbed_files; /* Total number of scrubbed files. */ - uint64_t unsigned_files; /* Total number of unsigned file */ + uint64_t unsigned_files; /* Total number of unsigned files. */ - uint64_t scrub_duration; /* Duration of last scrub */ + uint64_t scrub_duration; /* Duration of last scrub. */ - char last_scrub_time[1024]; /*last scrub completion time */ + char last_scrub_time[GF_TIMESTR_SIZE]; /* Last scrub completion time. */ - struct timeval scrub_start_tv; /* Scrubbing starting time*/ + time_t scrub_start_time; /* Scrubbing starting time. */ - struct timeval scrub_end_tv; /* Scrubbing finishing time */ + time_t scrub_end_time; /* Scrubbing finishing time. */ - int8_t scrub_running; /* Scrub running or not */ + int8_t scrub_running; /* Whether scrub running or not. */ pthread_mutex_t lock; }; @@ -40,9 +42,9 @@ br_inc_unsigned_file_count(br_scrub_stats_t *scrub_stat); void br_inc_scrubbed_file(br_scrub_stats_t *scrub_stat); void -br_update_scrub_start_time(br_scrub_stats_t *scrub_stat, struct timeval *tv); +br_update_scrub_start_time(br_scrub_stats_t *scrub_stat, time_t time); void br_update_scrub_finish_time(br_scrub_stats_t *scrub_stat, char *timestr, - struct timeval *tv); + time_t time); #endif /* __BIT_ROT_SCRUB_STATUS_H__ */ diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c index d20ecc7..289dd53 100644 --- a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c +++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c @@ -601,25 +601,23 @@ br_fsscan_deactivate(xlator_t *this) static void br_scrubber_log_time(xlator_t *this, const char *sfx) { - char timestr[1024] = { - 0, - }; - struct timeval tv = { + char timestr[GF_TIMESTR_SIZE] = { 0, }; br_private_t *priv = NULL; + time_t now = 0; + now = gf_time(); priv = this->private; - gettimeofday(&tv, NULL); - gf_time_fmt(timestr, sizeof(timestr), tv.tv_sec, gf_timefmt_FT); + gf_time_fmt(timestr, sizeof(timestr), now, gf_timefmt_FT); if (strcasecmp(sfx, "started") == 0) { - br_update_scrub_start_time(&priv->scrub_stat, &tv); + br_update_scrub_start_time(&priv->scrub_stat, now); gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_START, "Scrubbing %s at %s", sfx, timestr); } else { - br_update_scrub_finish_time(&priv->scrub_stat, timestr, &tv); + br_update_scrub_finish_time(&priv->scrub_stat, timestr, now); gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_FINISH, "Scrubbing %s at %s", sfx, timestr); } @@ -628,15 +626,13 @@ br_scrubber_log_time(xlator_t *this, const char *sfx) static void br_fsscanner_log_time(xlator_t *this, br_child_t *child, const char *sfx) { - char timestr[1024] = { - 0, - }; - struct timeval tv = { + char timestr[GF_TIMESTR_SIZE] = { 0, }; + time_t now = 0; - gettimeofday(&tv, NULL); - gf_time_fmt(timestr, sizeof(timestr), tv.tv_sec, gf_timefmt_FT); + now = gf_time(); + gf_time_fmt(timestr, sizeof(timestr), now, gf_timefmt_FT); if (strcasecmp(sfx, "started") == 0) { gf_msg_debug(this->name, 0, "Scrubbing \"%s\" %s at %s", @@ -919,10 +915,7 @@ br_fsscan_schedule(xlator_t *this) { uint32_t timo = 0; br_private_t *priv = NULL; - struct timeval tv = { - 0, - }; - char timestr[1024] = { + char timestr[GF_TIMESTR_SIZE] = { 0, }; struct br_scrubber *fsscrub = NULL; @@ -933,8 +926,7 @@ br_fsscan_schedule(xlator_t *this) fsscrub = &priv->fsscrub; scrub_monitor = &priv->scrub_monitor; - (void)gettimeofday(&tv, NULL); - scrub_monitor->boot = tv.tv_sec; + scrub_monitor->boot = gf_time(); timo = br_fsscan_calculate_timeout(fsscrub->frequency); if (timo == 0) { @@ -975,12 +967,10 @@ int32_t br_fsscan_activate(xlator_t *this) { uint32_t timo = 0; - char timestr[1024] = { - 0, - }; - struct timeval now = { + char timestr[GF_TIMESTR_SIZE] = { 0, }; + time_t now = 0; br_private_t *priv = NULL; struct br_scrubber *fsscrub = NULL; struct br_monitor *scrub_monitor = NULL; @@ -989,7 +979,7 @@ br_fsscan_activate(xlator_t *this) fsscrub = &priv->fsscrub; scrub_monitor = &priv->scrub_monitor; - (void)gettimeofday(&now, NULL); + now = gf_time(); timo = br_fsscan_calculate_timeout(fsscrub->frequency); if (timo == 0) { gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG, @@ -1003,7 +993,7 @@ br_fsscan_activate(xlator_t *this) } pthread_mutex_unlock(&scrub_monitor->donelock); - gf_time_fmt(timestr, sizeof(timestr), (now.tv_sec + timo), gf_timefmt_FT); + gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT); (void)gf_tw_mod_timer(priv->timer_wheel, scrub_monitor->timer, timo); _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING); @@ -1020,12 +1010,10 @@ br_fsscan_reschedule(xlator_t *this) { int32_t ret = 0; uint32_t timo = 0; - char timestr[1024] = { - 0, - }; - struct timeval now = { + char timestr[GF_TIMESTR_SIZE] = { 0, }; + time_t now = 0; br_private_t *priv = NULL; struct br_scrubber *fsscrub = NULL; struct br_monitor *scrub_monitor = NULL; @@ -1037,7 +1025,7 @@ br_fsscan_reschedule(xlator_t *this) if (!fsscrub->frequency_reconf) return 0; - (void)gettimeofday(&now, NULL); + now = gf_time(); timo = br_fsscan_calculate_timeout(fsscrub->frequency); if (timo == 0) { gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG, @@ -1045,7 +1033,7 @@ br_fsscan_reschedule(xlator_t *this) return -1; } - gf_time_fmt(timestr, sizeof(timestr), (now.tv_sec + timo), gf_timefmt_FT); + gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT); pthread_mutex_lock(&scrub_monitor->donelock); { @@ -1073,23 +1061,19 @@ br_fsscan_ondemand(xlator_t *this) { int32_t ret = 0; uint32_t timo = 0; - char timestr[1024] = { - 0, - }; - struct timeval now = { + char timestr[GF_TIMESTR_SIZE] = { 0, }; + time_t now = 0; br_private_t *priv = NULL; struct br_monitor *scrub_monitor = NULL; priv = this->private; scrub_monitor = &priv->scrub_monitor; - (void)gettimeofday(&now, NULL); - + now = gf_time(); timo = BR_SCRUB_ONDEMAND; - - gf_time_fmt(timestr, sizeof(timestr), (now.tv_sec + timo), gf_timefmt_FT); + gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT); pthread_mutex_lock(&scrub_monitor->donelock); { @@ -1799,7 +1783,7 @@ br_collect_bad_objects_of_child(xlator_t *this, br_child_t *child, dict_t *dict, tmp_count = total_count; for (j = 0; j < count; j++) { - len = snprintf(key, PATH_MAX, "quarantine-%d", j); + len = snprintf(key, sizeof(key), "quarantine-%d", j); ret = dict_get_strn(child_dict, key, len, &entry); if (ret) continue; @@ -1810,7 +1794,7 @@ br_collect_bad_objects_of_child(xlator_t *this, br_child_t *child, dict_t *dict, if ((len < 0) || (len >= PATH_MAX)) { continue; } - snprintf(main_key, PATH_MAX, "quarantine-%d", tmp_count); + snprintf(main_key, sizeof(main_key), "quarantine-%d", tmp_count); ret = dict_set_dynstr_with_alloc(dict, main_key, tmp); if (!ret) diff --git a/xlators/features/bit-rot/src/bitd/bit-rot.c b/xlators/features/bit-rot/src/bitd/bit-rot.c index 6d0c0b5..a2f1c34 100644 --- a/xlators/features/bit-rot/src/bitd/bit-rot.c +++ b/xlators/features/bit-rot/src/bitd/bit-rot.c @@ -241,8 +241,8 @@ br_object_open(xlator_t *this, br_object_t *object, inode_t *inode, ret = -EINVAL; fd = fd_create(inode, 0); if (!fd) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED, - "failed to create fd for the inode %s", uuid_utoa(inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED, + "gfid=%s", uuid_utoa(inode->gfid), NULL); goto out; } @@ -296,8 +296,8 @@ br_object_read_block_and_sign(xlator_t *this, fd_t *fd, br_child_t *child, NULL, NULL, NULL); if (ret < 0) { - gf_msg(this->name, GF_LOG_ERROR, errno, BRB_MSG_READV_FAILED, - "readv on %s failed", uuid_utoa(fd->inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, errno, BRB_MSG_READV_FAILED, + "gfid=%s", uuid_utoa(fd->inode->gfid), NULL); ret = -1; goto out; } @@ -347,9 +347,9 @@ br_calculate_obj_checksum(unsigned char *md, br_child_t *child, fd_t *fd, ret = br_object_read_block_and_sign(this, fd, child, offset, block, &sha256); if (ret < 0) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_BLOCK_READ_FAILED, - "reading block with offset %" PRIu64 " of object %s failed", - offset, uuid_utoa(fd->inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_BLOCK_READ_FAILED, + "offset=%" PRIu64, offset, "object-gfid=%s", + uuid_utoa(fd->inode->gfid), NULL); break; } @@ -391,28 +391,23 @@ br_object_read_sign(inode_t *linked_inode, fd_t *fd, br_object_t *object, md = GF_MALLOC(SHA256_DIGEST_LENGTH, gf_common_mt_char); if (!md) { - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, - "failed to allocate memory for saving hash of the " - "object %s", - uuid_utoa(fd->inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_SAVING_HASH_FAILED, + "object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL); goto out; } ret = br_object_checksum(md, object, fd, iatt); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_CALC_CHECKSUM_FAILED, - "calculating checksum " - "for the object %s failed", - uuid_utoa(linked_inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_CALC_CHECKSUM_FAILED, + "object-gfid=%s", uuid_utoa(linked_inode->gfid), NULL); goto free_signature; } sign = br_prepare_signature(md, SHA256_DIGEST_LENGTH, BR_SIGNATURE_TYPE_SHA256, object); if (!sign) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED, - "failed to get the signature for the object %s", - uuid_utoa(fd->inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED, + "object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL); goto free_signature; } @@ -420,17 +415,16 @@ br_object_read_sign(inode_t *linked_inode, fd_t *fd, br_object_t *object, signature_size(SHA256_DIGEST_LENGTH), _gf_true); if (!xattr) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED, - "dict allocation for signing failed for the object %s", - uuid_utoa(fd->inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED, + "dict-allocation object-gfid=%s", uuid_utoa(fd->inode->gfid), + NULL); goto free_isign; } ret = syncop_fsetxattr(object->child->xl, fd, xattr, 0, NULL, NULL); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED, - "fsetxattr of signature to the object %s failed", - uuid_utoa(fd->inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED, + "fsetxattr object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL); goto unref_dict; } @@ -463,8 +457,8 @@ br_log_object(xlator_t *this, char *op, uuid_t gfid, int32_t op_errno) "[reason: %s]", op, uuid_utoa(gfid), strerror(op_errno)); } else { - gf_msg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_OP_FAILED, - "%s() failed on object %s", op, uuid_utoa(gfid)); + gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_OP_FAILED, "op=%s", + op, "gfid=%s", uuid_utoa(gfid), NULL); } } @@ -478,8 +472,8 @@ br_log_object_path(xlator_t *this, char *op, const char *path, int32_t op_errno) "[reason: %s]", op, path, strerror(op_errno)); } else { - gf_msg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_OP_FAILED, - "%s() failed on object %s", op, path); + gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_OP_FAILED, "op=%s", + op, "path=%s", path, NULL); } } @@ -508,8 +502,8 @@ br_trigger_sign(xlator_t *this, br_child_t *child, inode_t *linked_inode, ret = -1; fd = fd_create(linked_inode, 0); if (!fd) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED, - "Failed to create fd [GFID %s]", uuid_utoa(linked_inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED, + "gfid=%s", uuid_utoa(linked_inode->gfid), NULL); goto cleanup_dict; } @@ -533,9 +527,9 @@ cleanup_dict: dict_unref(dict); out: if (ret) { - gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_TRIGGER_SIGN, - "Could not trigger signingd for %s (reopen hint: %d)", - uuid_utoa(linked_inode->gfid), val); + gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_TRIGGER_SIGN_FAILED, + "gfid=%s", uuid_utoa(linked_inode->gfid), "reopen-hint-val=%d", + val, NULL); } } @@ -615,10 +609,8 @@ br_sign_object(br_object_t *object) ret = br_object_read_sign(linked_inode, fd, object, &iatt); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_READ_AND_SIGN_FAILED, - "reading and signing of " - "the object %s failed", - uuid_utoa(linked_inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_READ_AND_SIGN_FAILED, + "gfid=%s", uuid_utoa(linked_inode->gfid), NULL); goto unref_fd; } @@ -672,8 +664,8 @@ br_process_object(void *arg) ret = br_sign_object(object); if (ret && !br_object_sign_softerror(-ret)) - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SIGN_FAILED, - "SIGNING FAILURE [%s]", uuid_utoa(object->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED, + "gfid=%s", uuid_utoa(object->gfid), NULL); GF_FREE(object); } @@ -775,9 +767,8 @@ br_schedule_object_reopen(xlator_t *this, br_object_t *object, timer = br_initialize_timer(this, object, child, ev); if (!timer) - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_TIMER_FAILED, - "Failed to allocate object expiry timer [GFID: %s]", - uuid_utoa(object->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_TIMER_FAILED, + "gfid=%s", uuid_utoa(object->gfid), NULL); return timer ? 0 : -1; } @@ -824,15 +815,15 @@ br_brick_callback(void *xl, char *brick, void *data, changelog_event_t *ev) child = br_get_child_from_brick_path(this, brick); if (!child) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SUBVOL_FAILED, - "failed to get the subvolume for the brick %s", brick); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SUBVOL_FAILED, + "brick=%s", brick, NULL); goto out; } object = br_initialize_object(this, child, ev); if (!object) { - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, - "failed to allocate object memory [GFID: %s]", uuid_utoa(gfid)); + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, + "object-gfid=%s", uuid_utoa(gfid), NULL); goto out; } @@ -884,8 +875,8 @@ br_check_object_need_sign(xlator_t *this, dict_t *xattr, br_child_t *child) ret = dict_get_ptr(xattr, GLUSTERFS_GET_OBJECT_SIGNATURE, (void **)&sign); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED, - "failed to get object signature info"); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED, + "object-info", NULL); goto out; } @@ -924,9 +915,9 @@ br_prepare_loc(xlator_t *this, br_child_t *child, loc_t *parent, ret = inode_path(parent->inode, entry->d_name, (char **)&loc->path); if (ret < 0 || !loc->path) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_PATH_FAILED, - "inode_path on %s (parent: %s) failed", entry->d_name, - uuid_utoa(parent->inode->gfid)); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_PATH_FAILED, + "inode_path=%s", entry->d_name, "parent-gfid=%s", + uuid_utoa(parent->inode->gfid), NULL); goto out; } @@ -1018,8 +1009,8 @@ bitd_oneshot_crawl(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent, */ if (bitd_is_bad_file(this, child, &loc, NULL)) { - gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SKIP_OBJECT, - "Entry [%s] is marked corrupted.. skipping.", loc.path); + gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SKIP_OBJECT, "path=%s", + loc.path, NULL); goto unref_inode; } @@ -1036,12 +1027,9 @@ bitd_oneshot_crawl(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent, if (op_errno == ENODATA && (iatt.ia_size != 0)) need_signing = _gf_true; if (op_errno == EINVAL) - gf_msg(this->name, GF_LOG_WARNING, 0, - BRB_MSG_PARTIAL_VERSION_PRESENCE, - "Partial " - "version xattr presence detected, ignoring " - "[GFID: %s]", - uuid_utoa(linked_inode->gfid)); + gf_smsg(this->name, GF_LOG_WARNING, 0, + BRB_MSG_PARTIAL_VERSION_PRESENCE, "gfid=%s", + uuid_utoa(linked_inode->gfid), NULL); } else { need_signing = br_check_object_need_sign(this, xattr, child); @@ -1061,9 +1049,9 @@ bitd_oneshot_crawl(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent, if (!need_signing) goto unref_dict; - gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_TRIGGER_SIGN, - "Triggering signing for %s [GFID: %s | Brick: %s]", loc.path, - uuid_utoa(linked_inode->gfid), child->brick_path); + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_TRIGGER_SIGN, "path=%s", + loc.path, "gfid=%s", uuid_utoa(linked_inode->gfid), "Brick-path=%s", + child->brick_path, NULL); br_trigger_sign(this, child, linked_inode, &loc, need_reopen); ret = 0; @@ -1096,17 +1084,16 @@ br_oneshot_signer(void *arg) THIS = this; - gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_CRAWLING_START, - "Crawling brick [%s], scanning for unsigned objects", - child->brick_path); + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_CRAWLING_START, "brick-path=%s", + child->brick_path, NULL); loc.inode = child->table->root; (void)syncop_ftw_throttle(child->xl, &loc, GF_CLIENT_PID_BITD, child, bitd_oneshot_crawl, BR_CRAWL_THROTTLE_COUNT, BR_CRAWL_THROTTLE_ZZZ); - gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_CRAWLING_FINISH, - "Completed crawling brick [%s]", child->brick_path); + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_CRAWLING_FINISH, + "brick-path=%s", child->brick_path, NULL); return NULL; } @@ -1150,9 +1137,7 @@ br_enact_signer(xlator_t *this, br_child_t *child, br_stub_init_t *stub) ret = gf_changelog_register_generic(brick, 1, 1, this->ctx->cmd_args.log_file, -1, this); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, errno, BRB_MSG_REGISTER_FAILED, - "Register to changelog " - "failed"); + gf_smsg(this->name, GF_LOG_ERROR, errno, BRB_MSG_REGISTER_FAILED, NULL); goto dealloc; } @@ -1160,8 +1145,8 @@ br_enact_signer(xlator_t *this, br_child_t *child, br_stub_init_t *stub) ret = gf_thread_create(&child->thread, NULL, br_oneshot_signer, child, "brosign"); if (ret) - gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SPAWN_FAILED, - "failed to spawn FS crawler thread"); + gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SPAWN_FAILED, + "FS-crawler-thread", NULL); else child->threadrunning = 1; @@ -1189,9 +1174,9 @@ br_launch_scrubber(xlator_t *this, br_child_t *child, struct br_scanfs *fsscan, ret = gf_thread_create(&child->thread, NULL, br_fsscanner, child, "brfsscan"); if (ret != 0) { - gf_msg(this->name, GF_LOG_ALERT, 0, BRB_MSG_SPAWN_FAILED, - "failed to spawn bitrot scrubber daemon [Brick: %s]", - child->brick_path); + gf_smsg(this->name, GF_LOG_ALERT, 0, BRB_MSG_SPAWN_FAILED, + "bitrot-scrubber-daemon Brick-path=%s", child->brick_path, + NULL); goto error_return; } @@ -1279,8 +1264,8 @@ br_child_enaction(xlator_t *this, br_child_t *child, br_stub_init_t *stub) if (!ret) { child->witnessed = 1; _br_set_child_state(child, BR_CHILD_STATE_CONNECTED); - gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_CONNECTED_TO_BRICK, - "Connected to brick %s..", child->brick_path); + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_CONNECTED_TO_BRICK, + "brick-path=%s", child->brick_path, NULL); } } pthread_mutex_unlock(&child->lock); @@ -1327,8 +1312,8 @@ br_brick_connect(xlator_t *this, br_child_t *child) if (ret) { op_errno = -ret; ret = -1; - gf_msg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_LOOKUP_FAILED, - "lookup on root failed"); + gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_LOOKUP_FAILED, + NULL); goto wipeloc; } @@ -1337,15 +1322,14 @@ br_brick_connect(xlator_t *this, br_child_t *child) if (ret) { op_errno = -ret; ret = -1; - gf_msg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_GET_INFO_FAILED, - "failed to get stub info"); + gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_GET_INFO_FAILED, + NULL); goto wipeloc; } ret = dict_get_ptr(xattr, GLUSTERFS_GET_BR_STUB_INIT_TIME, (void **)&stub); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_INFO_FAILED, - "failed to extract stub information"); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_INFO_FAILED, NULL); goto free_dict; } @@ -1415,11 +1399,10 @@ br_cleanup_scrubber(xlator_t *this, br_child_t *child) */ ret = gf_thread_cleanup_xint(child->thread); if (ret) - gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_THREAD_CLEANUP, - "Error cleaning up scanner thread"); + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_THREAD_CLEANUP, NULL); - gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUBBER_CLEANED, - "Cleaned up scrubber for brick [%s]", child->brick_path); + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUBBER_CLEANED, + "brick-path=%s", child->brick_path, NULL); return 0; } @@ -1504,9 +1487,8 @@ br_handle_events(void *arg) child = childev->child; ret = childev->call(this, child); if (ret) - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SUBVOL_CONNECT_FAILED, - "callback handler for subvolume [%s] failed", - child->xl->name); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SUBVOL_CONNECT_FAILED, + "name=%s", child->xl->name, NULL); GF_FREE(childev); } @@ -1524,8 +1506,7 @@ mem_acct_init(xlator_t *this) ret = xlator_mem_acct_init(this, gf_br_stub_mt_end + 1); if (ret != 0) { - gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_MEM_ACNT_FAILED, - "Memory accounting init failed"); + gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_MEM_ACNT_FAILED, NULL); return ret; } @@ -1542,8 +1523,8 @@ _br_qchild_event(xlator_t *this, br_child_t *child, br_child_handler *call) childev = GF_CALLOC(1, sizeof(*childev), gf_br_mt_br_child_event_t); if (!childev) { - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, - "Event unhandled for child.. [Brick: %s]", child->xl->name); + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_EVENT_UNHANDLED, + "Brick-name=%s", child->xl->name, NULL); return; } @@ -1638,10 +1619,8 @@ notify(xlator_t *this, int32_t event, void *data, ...) switch (event) { case GF_EVENT_CHILD_UP: if (idx < 0) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_INVALID_SUBVOL, - "Got event %d from " - "invalid subvolume", - event); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_INVALID_SUBVOL, + "event=%d", event, NULL); goto out; } @@ -1669,9 +1648,8 @@ notify(xlator_t *this, int32_t event, void *data, ...) case GF_EVENT_CHILD_DOWN: if (idx < 0) { - gf_msg(this->name, GF_LOG_ERROR, 0, - BRB_MSG_INVALID_SUBVOL_CHILD, - "Got event %d from invalid subvolume", event); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_INVALID_SUBVOL, + "event=%d", event, NULL); goto out; } @@ -1712,11 +1690,9 @@ notify(xlator_t *this, int32_t event, void *data, ...) "called"); if (scrub_monitor->state != BR_SCRUB_STATE_PENDING) { - gf_msg(this->name, GF_LOG_ERROR, 0, - BRB_MSG_RESCHEDULE_SCRUBBER_FAILED, - "on demand scrub schedule failed. Scrubber is " - "not in pending state. Current state is %d", - scrub_monitor->state); + gf_smsg(this->name, GF_LOG_ERROR, 0, + BRB_MSG_RESCHEDULE_SCRUBBER_FAILED, "Current-state=%d", + scrub_monitor->state, NULL); return -2; } @@ -1728,11 +1704,8 @@ notify(xlator_t *this, int32_t event, void *data, ...) pthread_mutex_unlock(&priv->lock); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, - BRB_MSG_RESCHEDULE_SCRUBBER_FAILED, - "Could not schedule ondemand scrubbing. " - "Scrubbing will continue according to " - "old frequency."); + gf_smsg(this->name, GF_LOG_ERROR, 0, + BRB_MSG_COULD_NOT_SCHEDULE_SCRUB, NULL); } gf_msg_debug(this->name, 0, "returning %d", ret); break; @@ -1744,22 +1717,26 @@ out: return 0; } -/** - * Initialize signer specific structures, spawn worker threads. - */ - static void br_fini_signer(xlator_t *this, br_private_t *priv) { int i = 0; - for (; i < BR_WORKERS; i++) { + if (priv == NULL) + return; + + for (; i < priv->signer_th_count; i++) { (void)gf_thread_cleanup_xint(priv->obj_queue->workers[i]); } + GF_FREE(priv->obj_queue->workers); pthread_cond_destroy(&priv->object_cond); } +/** + * Initialize signer specific structures, spawn worker threads. + */ + static int32_t br_init_signer(xlator_t *this, br_private_t *priv) { @@ -1779,13 +1756,17 @@ br_init_signer(xlator_t *this, br_private_t *priv) goto cleanup_cond; INIT_LIST_HEAD(&priv->obj_queue->objects); - for (i = 0; i < BR_WORKERS; i++) { + priv->obj_queue->workers = GF_CALLOC( + priv->signer_th_count, sizeof(pthread_t), gf_br_mt_br_worker_t); + if (!priv->obj_queue->workers) + goto cleanup_obj_queue; + + for (i = 0; i < priv->signer_th_count; i++) { ret = gf_thread_create(&priv->obj_queue->workers[i], NULL, br_process_object, this, "brpobj"); if (ret != 0) { - gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SPAWN_FAILED, - "thread creation" - " failed"); + gf_smsg(this->name, GF_LOG_ERROR, -ret, + BRB_MSG_THREAD_CREATION_FAILED, NULL); ret = -1; goto cleanup_threads; } @@ -1797,7 +1778,9 @@ cleanup_threads: for (i--; i >= 0; i--) { (void)gf_thread_cleanup_xint(priv->obj_queue->workers[i]); } + GF_FREE(priv->obj_queue->workers); +cleanup_obj_queue: GF_FREE(priv->obj_queue); cleanup_cond: @@ -1850,18 +1833,17 @@ br_rate_limit_signer(xlator_t *this, int child_count, int numbricks) if (contribution == 0) contribution = 1; spec.rate = BR_HASH_CALC_READ_SIZE * contribution; - spec.maxlimit = BR_WORKERS * BR_HASH_CALC_READ_SIZE; + spec.maxlimit = priv->signer_th_count * BR_HASH_CALC_READ_SIZE; #endif if (!spec.rate) - gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_RATE_LIMIT_INFO, - "[Rate Limit Info] \"FULL THROTTLE\""); + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_RATE_LIMIT_INFO, + "FULL THROTTLE", NULL); else - gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_RATE_LIMIT_INFO, - "[Rate Limit Info] \"tokens/sec (rate): %lu, " - "maxlimit: %lu\"", - spec.rate, spec.maxlimit); + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_RATE_LIMIT_INFO, + "tokens/sec-rate=%lu", spec.rate, "maxlimit=%lu", spec.maxlimit, + NULL); priv->tbf = tbf_init(&spec, 1); return priv->tbf ? 0 : -1; @@ -1870,11 +1852,16 @@ br_rate_limit_signer(xlator_t *this, int child_count, int numbricks) static int32_t br_signer_handle_options(xlator_t *this, br_private_t *priv, dict_t *options) { - if (options) + if (options) { GF_OPTION_RECONF("expiry-time", priv->expiry_time, options, uint32, error_return); - else + GF_OPTION_RECONF("signer-threads", priv->signer_th_count, options, + uint32, error_return); + } else { GF_OPTION_INIT("expiry-time", priv->expiry_time, uint32, error_return); + GF_OPTION_INIT("signer-threads", priv->signer_th_count, uint32, + error_return); + } return 0; @@ -1890,6 +1877,8 @@ br_signer_init(xlator_t *this, br_private_t *priv) GF_OPTION_INIT("expiry-time", priv->expiry_time, uint32, error_return); GF_OPTION_INIT("brick-count", numbricks, int32, error_return); + GF_OPTION_INIT("signer-threads", priv->signer_th_count, uint32, + error_return); ret = br_rate_limit_signer(this, priv->child_count, numbricks); if (ret) @@ -1976,8 +1965,8 @@ br_init_children(xlator_t *this, br_private_t *priv) child->timer_pool = mem_pool_new(struct gf_tw_timer_list, 4096); if (!child->timer_pool) { - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, - "failed to allocate mem-pool for timer"); + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_MEM_POOL_ALLOC, + NULL); errno = ENOMEM; goto freechild; } @@ -2003,15 +1992,13 @@ init(xlator_t *this) br_private_t *priv = NULL; if (!this->children) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_NO_CHILD, - "FATAL: no children"); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_NO_CHILD, NULL); goto out; } priv = GF_CALLOC(1, sizeof(*priv), gf_br_mt_br_private_t); if (!priv) { - gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, - "failed to allocate memory (->priv)"); + gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, NULL); goto out; } @@ -2029,8 +2016,8 @@ init(xlator_t *this) priv->timer_wheel = glusterfs_ctx_tw_get(this->ctx); if (!priv->timer_wheel) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_TIMER_WHEEL_UNAVAILABLE, - "global timer wheel unavailable"); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_TIMER_WHEEL_UNAVAILABLE, + NULL); goto cleanup; } @@ -2052,15 +2039,14 @@ init(xlator_t *this) ret = gf_thread_create(&priv->thread, NULL, br_handle_events, this, "brhevent"); if (ret != 0) { - gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SPAWN_FAILED, - "thread creation failed"); + gf_smsg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_THREAD_CREATION_FAILED, + NULL); ret = -1; } if (!ret) { - gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_BITROT_LOADED, - "bit-rot xlator loaded in \"%s\" mode", - (priv->iamscrubber) ? "SCRUBBER" : "SIGNER"); + gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_BITROT_LOADED, "mode=%s", + (priv->iamscrubber) ? "SCRUBBER" : "SIGNER", NULL); return 0; } @@ -2107,9 +2093,8 @@ br_reconfigure_monitor(xlator_t *this) ret = br_scrub_state_machine(this, _gf_false); if (ret) { - gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_RESCHEDULE_SCRUBBER_FAILED, - "Could not reschedule scrubber for the volume. Scrubbing " - "will continue according to old frequency."); + gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_COULD_NOT_SCHEDULE_SCRUB, + NULL); } } @@ -2220,6 +2205,15 @@ struct volume_options options[] = { .description = "Pause/Resume scrub. Upon resume, scrubber " "continues from where it left off.", }, + { + .key = {"signer-threads"}, + .type = GF_OPTION_TYPE_INT, + .default_value = BR_WORKERS, + .op_version = {GD_OP_VERSION_8_0}, + .flags = OPT_FLAG_SETTABLE, + .description = "Number of signing process threads. As a best " + "practice, set this to the number of processor cores", + }, {.key = {NULL}}, }; diff --git a/xlators/features/bit-rot/src/bitd/bit-rot.h b/xlators/features/bit-rot/src/bitd/bit-rot.h index a4d4fd7..8ac7dcd 100644 --- a/xlators/features/bit-rot/src/bitd/bit-rot.h +++ b/xlators/features/bit-rot/src/bitd/bit-rot.h @@ -30,12 +30,6 @@ #include <openssl/sha.h> -/** - * TODO: make this configurable. As a best practice, set this to the - * number of processor cores. - */ -#define BR_WORKERS 4 - typedef enum scrub_throttle { BR_SCRUB_THROTTLE_VOID = -1, BR_SCRUB_THROTTLE_LAZY = 0, @@ -108,12 +102,12 @@ struct br_child { typedef struct br_child br_child_t; struct br_obj_n_workers { - struct list_head objects; /* queue of objects expired from the - timer wheel and ready to be picked - up for signing */ - pthread_t workers[BR_WORKERS]; /* Threads which pick up the objects - from the above queue and start - signing each object */ + struct list_head objects; /* queue of objects expired from the + timer wheel and ready to be picked + up for signing */ + pthread_t *workers; /* Threads which pick up the objects + from the above queue and start + signing each object */ }; struct br_scrubber { @@ -209,6 +203,8 @@ struct br_private { uint32_t expiry_time; /* objects "wait" time */ + uint32_t signer_th_count; /* Number of signing process threads */ + tbf_t *tbf; /* token bucket filter */ gf_boolean_t iamscrubber; /* function as a fs scrubber */ diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h b/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h index 40bcda1..9d93caf 100644 --- a/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h +++ b/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h @@ -29,6 +29,7 @@ enum br_mem_types { gf_br_stub_mt_sigstub_t, gf_br_mt_br_child_event_t, gf_br_stub_mt_misc, + gf_br_mt_br_worker_t, gf_br_stub_mt_end, }; diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub-messages.h b/xlators/features/bit-rot/src/stub/bit-rot-stub-messages.h index 8d2b7f0..6c15a16 100644 --- a/xlators/features/bit-rot/src/stub/bit-rot-stub-messages.h +++ b/xlators/features/bit-rot/src/stub/bit-rot-stub-messages.h @@ -44,7 +44,8 @@ GLFS_MSGID(BITROT_STUB, BRS_MSG_NO_MEMORY, BRS_MSG_SET_EVENT_FAILED, BRS_MSG_NON_BITD_PID, BRS_MSG_SIGN_PREPARE_FAIL, BRS_MSG_USING_DEFAULT_THREAD_SIZE, BRS_MSG_ALLOC_MEM_FAILED, BRS_MSG_DICT_ALLOC_FAILED, BRS_MSG_CREATE_GF_DIRENT_FAILED, - BRS_MSG_ALLOC_FAILED, BRS_MSG_PATH_XATTR_GET_FAILED); + BRS_MSG_ALLOC_FAILED, BRS_MSG_PATH_XATTR_GET_FAILED, + BRS_MSG_VERSION_PREPARE_FAIL); #define BRS_MSG_MEM_ACNT_FAILED_STR "Memory accounting init failed" #define BRS_MSG_BAD_OBJ_THREAD_FAIL_STR "pthread_init failed" @@ -68,6 +69,8 @@ GLFS_MSGID(BITROT_STUB, BRS_MSG_NO_MEMORY, BRS_MSG_SET_EVENT_FAILED, "daemon. Unwinding the fop" #define BRS_MSG_SIGN_PREPARE_FAIL_STR \ "failed to prepare the signature. Unwinding the fop" +#define BRS_MSG_VERSION_PREPARE_FAIL_STR \ + "failed to prepare the version. Unwinding the fop" #define BRS_MSG_STUB_ALLOC_FAILED_STR "failed to allocate stub fop, Unwinding" #define BRS_MSG_BAD_OBJ_MARK_FAIL_STR "failed to mark object as bad" #define BRS_MSG_NON_SCRUB_BAD_OBJ_MARK_STR \ diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub.c b/xlators/features/bit-rot/src/stub/bit-rot-stub.c index 605a5e4..447dd47 100644 --- a/xlators/features/bit-rot/src/stub/bit-rot-stub.c +++ b/xlators/features/bit-rot/src/stub/bit-rot-stub.c @@ -424,8 +424,8 @@ br_stub_prepare_version_request(xlator_t *this, dict_t *dict, priv = this->private; br_set_ongoingversion(obuf, oversion, priv->boot); - return dict_set_static_bin(dict, BITROT_CURRENT_VERSION_KEY, (void *)obuf, - sizeof(br_version_t)); + return dict_set_bin(dict, BITROT_CURRENT_VERSION_KEY, (void *)obuf, + sizeof(br_version_t)); } static int @@ -436,8 +436,7 @@ br_stub_prepare_signing_request(dict_t *dict, br_signature_t *sbuf, br_set_signature(sbuf, sign, signaturelen, &size); - return dict_set_static_bin(dict, BITROT_SIGNING_VERSION_KEY, (void *)sbuf, - size); + return dict_set_bin(dict, BITROT_SIGNING_VERSION_KEY, (void *)sbuf, size); } /** @@ -854,23 +853,27 @@ br_stub_perform_incversioning(xlator_t *this, call_frame_t *frame, op_errno = ENOMEM; dict = dict_new(); if (!dict) - goto done; + goto out; ret = br_stub_alloc_versions(&obuf, NULL, 0); - if (ret) - goto dealloc_dict; + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_ALLOC_MEM_FAILED, + "gfid=%s", uuid_utoa(fd->inode->gfid), NULL); + goto out; + } ret = br_stub_prepare_version_request(this, dict, obuf, writeback_version); - if (ret) - goto dealloc_versions; + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_VERSION_PREPARE_FAIL, + "gfid=%s", uuid_utoa(fd->inode->gfid), NULL); + br_stub_dealloc_versions(obuf); + goto out; + } ret = br_stub_fd_versioning( this, frame, stub, dict, fd, br_stub_fd_incversioning_cbk, writeback_version, BR_STUB_INCREMENTAL_VERSIONING, !WRITEBACK_DURABLE); - -dealloc_versions: - br_stub_dealloc_versions(obuf); -dealloc_dict: - dict_unref(dict); -done: +out: + if (dict) + dict_unref(dict); if (ret) { if (local) frame->local = NULL; @@ -1025,31 +1028,36 @@ static int br_stub_prepare_signature(xlator_t *this, dict_t *dict, inode_t *inode, br_isignature_t *sign, int *fakesuccess) { - int32_t ret = 0; + int32_t ret = -1; size_t signaturelen = 0; br_signature_t *sbuf = NULL; if (!br_is_signature_type_valid(sign->signaturetype)) - goto error_return; + goto out; signaturelen = sign->signaturelen; ret = br_stub_alloc_versions(NULL, &sbuf, signaturelen); - if (ret) - goto error_return; + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_ALLOC_MEM_FAILED, + "gfid=%s", uuid_utoa(inode->gfid), NULL); + ret = -1; + goto out; + } ret = br_stub_prepare_signing_request(dict, sbuf, sign, signaturelen); - if (ret) - goto dealloc_versions; + if (ret) { + gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_SIGN_PREPARE_FAIL, + "gfid=%s", uuid_utoa(inode->gfid), NULL); + ret = -1; + br_stub_dealloc_versions(sbuf); + goto out; + } + /* At this point sbuf has been added to dict, so the memory will be freed + * when the data from the dict is destroyed + */ ret = br_stub_compare_sign_version(this, inode, sbuf, dict, fakesuccess); - if (ret) - goto dealloc_versions; - - return 0; - -dealloc_versions: - br_stub_dealloc_versions(sbuf); -error_return: - return -1; +out: + return ret; } static void diff --git a/xlators/features/changelog/src/changelog-helpers.c b/xlators/features/changelog/src/changelog-helpers.c index 71fe1f0..e561997 100644 --- a/xlators/features/changelog/src/changelog-helpers.c +++ b/xlators/features/changelog/src/changelog-helpers.c @@ -242,8 +242,7 @@ changelog_write(int fd, char *buffer, size_t len) } int -htime_update(xlator_t *this, changelog_priv_t *priv, unsigned long ts, - char *buffer) +htime_update(xlator_t *this, changelog_priv_t *priv, time_t ts, char *buffer) { char changelog_path[PATH_MAX + 1] = { 0, @@ -273,7 +272,7 @@ htime_update(xlator_t *this, changelog_priv_t *priv, unsigned long ts, goto out; } - len = snprintf(x_value, sizeof(x_value), "%lu:%d", ts, + len = snprintf(x_value, sizeof(x_value), "%ld:%d", ts, priv->rollover_count); if (len >= sizeof(x_value)) { ret = -1; @@ -382,8 +381,7 @@ out: } static int -changelog_rollover_changelog(xlator_t *this, changelog_priv_t *priv, - unsigned long ts) +changelog_rollover_changelog(xlator_t *this, changelog_priv_t *priv, time_t ts) { int ret = -1; int notify = 0; @@ -421,16 +419,14 @@ changelog_rollover_changelog(xlator_t *this, changelog_priv_t *priv, priv->changelog_fd = -1; } - time_t time = (time_t)ts; - - /* Get GMT time */ - gmt = gmtime(&time); + /* Get GMT time. */ + gmt = gmtime(&ts); strftime(yyyymmdd, sizeof(yyyymmdd), "%Y/%m/%d", gmt); (void)snprintf(ofile, PATH_MAX, "%s/" CHANGELOG_FILE_NAME, priv->changelog_dir); - (void)snprintf(nfile, PATH_MAX, "%s/%s/" CHANGELOG_FILE_NAME ".%lu", + (void)snprintf(nfile, PATH_MAX, "%s/%s/" CHANGELOG_FILE_NAME ".%ld", priv->changelog_dir, yyyymmdd, ts); (void)snprintf(nfile_dir, PATH_MAX, "%s/%s", priv->changelog_dir, yyyymmdd); @@ -593,7 +589,7 @@ out: * returns -1 on failure or error */ int -htime_open(xlator_t *this, changelog_priv_t *priv, unsigned long ts) +htime_open(xlator_t *this, changelog_priv_t *priv, time_t ts) { int ht_file_fd = -1; int ht_dir_fd = -1; @@ -723,7 +719,7 @@ out: * returns -1 on failure or error */ int -htime_create(xlator_t *this, changelog_priv_t *priv, unsigned long ts) +htime_create(xlator_t *this, changelog_priv_t *priv, time_t ts) { int ht_file_fd = -1; int ht_dir_fd = -1; @@ -741,12 +737,12 @@ htime_create(xlator_t *this, changelog_priv_t *priv, unsigned long ts) int32_t len = 0; gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_NEW_HTIME_FILE, - "name=%lu", ts, NULL); + "name=%ld", ts, NULL); CHANGELOG_FILL_HTIME_DIR(priv->changelog_dir, ht_dir_path); /* get the htime file name in ht_file_path */ - len = snprintf(ht_file_path, PATH_MAX, "%s/%s.%lu", ht_dir_path, + len = snprintf(ht_file_path, PATH_MAX, "%s/%s.%ld", ht_dir_path, HTIME_FILE_NAME, ts); if ((len < 0) || (len >= PATH_MAX)) { ret = -1; @@ -792,7 +788,7 @@ htime_create(xlator_t *this, changelog_priv_t *priv, unsigned long ts) goto out; } - (void)snprintf(ht_file_bname, sizeof(ht_file_bname), "%s.%lu", + (void)snprintf(ht_file_bname, sizeof(ht_file_bname), "%s.%ld", HTIME_FILE_NAME, ts); if (sys_fsetxattr(ht_dir_fd, HTIME_CURRENT, ht_file_bname, strlen(ht_file_bname), 0)) { @@ -963,8 +959,8 @@ out: } int -changelog_start_next_change(xlator_t *this, changelog_priv_t *priv, - unsigned long ts, gf_boolean_t finale) +changelog_start_next_change(xlator_t *this, changelog_priv_t *priv, time_t ts, + gf_boolean_t finale) { int ret = -1; @@ -985,21 +981,12 @@ changelog_entry_length() return sizeof(changelog_log_data_t); } -int +void changelog_fill_rollover_data(changelog_log_data_t *cld, gf_boolean_t is_last) { - struct timeval tv = { - 0, - }; - cld->cld_type = CHANGELOG_TYPE_ROLLOVER; - - if (gettimeofday(&tv, NULL)) - return -1; - - cld->cld_roll_time = (unsigned long)tv.tv_sec; + cld->cld_roll_time = gf_time(); cld->cld_finale = is_last; - return 0; } int @@ -1274,7 +1261,7 @@ changelog_rollover(void *data) while (1) { (void)pthread_testcancel(); - tv.tv_sec = time(NULL) + priv->rollover_time; + tv.tv_sec = gf_time() + priv->rollover_time; tv.tv_nsec = 0; ret = 0; /* Reset ret to zero */ @@ -1355,12 +1342,7 @@ changelog_rollover(void *data) if (priv->explicit_rollover == _gf_true) sleep(1); - ret = changelog_fill_rollover_data(&cld, _gf_false); - if (ret) { - gf_smsg(this->name, GF_LOG_ERROR, 0, - CHANGELOG_MSG_ROLLOVER_DATA_FILL_FAILED, NULL); - continue; - } + changelog_fill_rollover_data(&cld, _gf_false); _mask_cancellation(); diff --git a/xlators/features/changelog/src/changelog-helpers.h b/xlators/features/changelog/src/changelog-helpers.h index 0906b16..38fa759 100644 --- a/xlators/features/changelog/src/changelog-helpers.h +++ b/xlators/features/changelog/src/changelog-helpers.h @@ -31,7 +31,7 @@ */ typedef struct changelog_log_data { /* rollover related */ - unsigned long cld_roll_time; + time_t cld_roll_time; /* reopen changelog? */ gf_boolean_t cld_finale; @@ -97,12 +97,6 @@ struct changelog_encoder { typedef struct changelog_time_slice { /** - * just in case we need nanosecond granularity some day. - * field is unused as of now (maybe we'd need it later). - */ - struct timeval tv_start; - - /** * version of changelog file, incremented each time changes * rollover. */ @@ -190,8 +184,12 @@ typedef struct changelog_ev_selector { /* changelog's private structure */ struct changelog_priv { + /* changelog journalling */ gf_boolean_t active; + /* changelog live notifications */ + gf_boolean_t rpc_active; + /* to generate unique socket file per brick */ char *changelog_brick; @@ -419,11 +417,11 @@ changelog_local_t * changelog_local_init(xlator_t *this, inode_t *inode, uuid_t gfid, int xtra_records, gf_boolean_t update_flag); int -changelog_start_next_change(xlator_t *this, changelog_priv_t *priv, - unsigned long ts, gf_boolean_t finale); +changelog_start_next_change(xlator_t *this, changelog_priv_t *priv, time_t ts, + gf_boolean_t finale); int changelog_open_journal(xlator_t *this, changelog_priv_t *priv); -int +void changelog_fill_rollover_data(changelog_log_data_t *cld, gf_boolean_t is_last); int changelog_inject_single_event(xlator_t *this, changelog_priv_t *priv, @@ -447,12 +445,11 @@ changelog_fsync_thread(void *data); int changelog_forget(xlator_t *this, inode_t *inode); int -htime_update(xlator_t *this, changelog_priv_t *priv, unsigned long ts, - char *buffer); +htime_update(xlator_t *this, changelog_priv_t *priv, time_t ts, char *buffer); int -htime_open(xlator_t *this, changelog_priv_t *priv, unsigned long ts); +htime_open(xlator_t *this, changelog_priv_t *priv, time_t ts); int -htime_create(xlator_t *this, changelog_priv_t *priv, unsigned long ts); +htime_create(xlator_t *this, changelog_priv_t *priv, time_t ts); /* Geo-Rep snapshot dependency changes */ void diff --git a/xlators/features/changelog/src/changelog-messages.h b/xlators/features/changelog/src/changelog-messages.h index 4dd56b8..cb0e16c 100644 --- a/xlators/features/changelog/src/changelog-messages.h +++ b/xlators/features/changelog/src/changelog-messages.h @@ -59,12 +59,12 @@ GLFS_MSGID( CHANGELOG_MSG_NO_HTIME_CURRENT, CHANGELOG_MSG_HTIME_CURRENT, CHANGELOG_MSG_NEW_HTIME_FILE, CHANGELOG_MSG_MKDIR_ERROR, CHANGELOG_MSG_PATH_NOT_FOUND, CHANGELOG_MSG_XATTR_INIT_FAILED, - CHANGELOG_MSG_WROTE_TO_CSNAP, CHANGELOG_MSG_ROLLOVER_DATA_FILL_FAILED, + CHANGELOG_MSG_WROTE_TO_CSNAP, CHANGELOG_MSG_UNUSED_0, CHANGELOG_MSG_GET_BUFFER_FAILED, CHANGELOG_MSG_BARRIER_STATE_NOTIFY, CHANGELOG_MSG_BARRIER_DISABLED, CHANGELOG_MSG_BARRIER_ALREADY_DISABLED, CHANGELOG_MSG_BARRIER_ON_ERROR, CHANGELOG_MSG_BARRIER_ENABLE, CHANGELOG_MSG_BARRIER_KEY_NOT_FOUND, CHANGELOG_MSG_ERROR_IN_DICT_GET, - CHANGELOG_MSG_GET_TIME_FAILURE, CHANGELOG_MSG_HTIME_FETCH_FAILED, + CHANGELOG_MSG_UNUSED_1, CHANGELOG_MSG_UNUSED_2, CHANGELOG_MSG_DEQUEUING_BARRIER_FOPS, CHANGELOG_MSG_DEQUEUING_BARRIER_FOPS_FINISHED, CHANGELOG_MSG_BARRIER_TIMEOUT, CHANGELOG_MSG_TIMEOUT_ADD_FAILED, @@ -123,8 +123,6 @@ GLFS_MSGID( #define CHANGELOG_MSG_GET_TIME_OP_FAILED_STR "Problem rolling over changelog(s)" #define CHANGELOG_MSG_BARRIER_INFO_STR "Explicit wakeup on barrier notify" #define CHANGELOG_MSG_SELECT_FAILED_STR "pthread_cond_timedwait failed" -#define CHANGELOG_MSG_ROLLOVER_DATA_FILL_FAILED_STR \ - "failed to fill rollover data" #define CHANGELOG_MSG_INJECT_FSYNC_FAILED_STR "failed to inject fsync event" #define CHANGELOG_MSG_LOCAL_INIT_FAILED_STR \ "changelog local initialization failed" @@ -144,9 +142,7 @@ GLFS_MSGID( #define CHANGELOG_MSG_BARRIER_KEY_NOT_FOUND_STR "barrier key not found" #define CHANGELOG_MSG_ERROR_IN_DICT_GET_STR \ "Something went wrong in dict_get_str_boolean" -#define CHANGELOG_MSG_GET_TIME_FAILURE_STR "gettimeofday() failure" #define CHANGELOG_MSG_DIR_OPTIONS_NOT_SET_STR "changelog-dir option is not set" -#define CHANGELOG_MSG_HTIME_FETCH_FAILED_STR "unable to fetch htime" #define CHANGELOG_MSG_FREEUP_FAILED_STR "could not cleanup bootstrapper" #define CHANGELOG_MSG_CHILD_MISCONFIGURED_STR \ "translator needs a single subvolume" diff --git a/xlators/features/changelog/src/changelog-rpc-common.c b/xlators/features/changelog/src/changelog-rpc-common.c index afcc3a8..125246a 100644 --- a/xlators/features/changelog/src/changelog-rpc-common.c +++ b/xlators/features/changelog/src/changelog-rpc-common.c @@ -262,6 +262,9 @@ changelog_rpc_server_destroy(xlator_t *this, rpcsvc_t *rpc, char *sockfile, struct rpcsvc_program *prog = NULL; rpc_transport_t *trans = NULL; + if (!rpc) + return; + while (*progs) { prog = *progs; (void)rpcsvc_program_unregister(rpc, prog); diff --git a/xlators/features/changelog/src/changelog.c b/xlators/features/changelog/src/changelog.c index 0a03cfa..6a6e5af 100644 --- a/xlators/features/changelog/src/changelog.c +++ b/xlators/features/changelog/src/changelog.c @@ -34,6 +34,12 @@ static struct changelog_bootstrap cb_bootstrap[] = { }, }; +static int +changelog_init_rpc(xlator_t *this, changelog_priv_t *priv); + +static int +changelog_init(xlator_t *this, changelog_priv_t *priv); + /* Entry operations - TYPE III */ /** @@ -1997,6 +2003,11 @@ notify(xlator_t *this, int event, void *data, ...) uint64_t clntcnt = 0; changelog_clnt_t *conn = NULL; gf_boolean_t cleanup_notify = _gf_false; + char sockfile[UNIX_PATH_MAX] = { + 0, + }; + rpcsvc_listener_t *listener = NULL; + rpcsvc_listener_t *next = NULL; INIT_LIST_HEAD(&queue); @@ -2010,23 +2021,40 @@ notify(xlator_t *this, int event, void *data, ...) "cleanup changelog rpc connection of brick %s", priv->victim->name); - this->cleanup_starting = 1; - changelog_destroy_rpc_listner(this, priv); - conn = &priv->connections; - if (conn) - changelog_ev_cleanup_connections(this, conn); - xprtcnt = GF_ATOMIC_GET(priv->xprtcnt); - clntcnt = GF_ATOMIC_GET(priv->clntcnt); - - if (!xprtcnt && !clntcnt) { - LOCK(&priv->lock); - { - cleanup_notify = priv->notify_down; - priv->notify_down = _gf_true; + if (priv->rpc_active) { + this->cleanup_starting = 1; + changelog_destroy_rpc_listner(this, priv); + conn = &priv->connections; + if (conn) + changelog_ev_cleanup_connections(this, conn); + xprtcnt = GF_ATOMIC_GET(priv->xprtcnt); + clntcnt = GF_ATOMIC_GET(priv->clntcnt); + if (!xprtcnt && !clntcnt) { + LOCK(&priv->lock); + { + cleanup_notify = priv->notify_down; + priv->notify_down = _gf_true; + } + UNLOCK(&priv->lock); + if (priv->rpc) { + list_for_each_entry_safe(listener, next, + &priv->rpc->listeners, list) + { + if (listener->trans) { + rpc_transport_unref(listener->trans); + } + } + rpcsvc_destroy(priv->rpc); + priv->rpc = NULL; + } + CHANGELOG_MAKE_SOCKET_PATH(priv->changelog_brick, sockfile, + UNIX_PATH_MAX); + sys_unlink(sockfile); + if (!cleanup_notify) + default_notify(this, GF_EVENT_PARENT_DOWN, data); } - UNLOCK(&priv->lock); - if (!cleanup_notify) - default_notify(this, GF_EVENT_PARENT_DOWN, data); + } else { + default_notify(this, GF_EVENT_PARENT_DOWN, data); } goto out; } @@ -2224,23 +2252,11 @@ static int changelog_init(xlator_t *this, changelog_priv_t *priv) { int i = 0; - int ret = -1; - struct timeval tv = { - 0, - }; + int ret = 0; changelog_log_data_t cld = { 0, }; - ret = gettimeofday(&tv, NULL); - if (ret) { - gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_GET_TIME_FAILURE, - NULL); - goto out; - } - - priv->slice.tv_start = tv; - priv->maps[CHANGELOG_TYPE_DATA] = "D "; priv->maps[CHANGELOG_TYPE_METADATA] = "M "; priv->maps[CHANGELOG_TYPE_METADATA_XATTR] = "M "; @@ -2259,9 +2275,7 @@ changelog_init(xlator_t *this, changelog_priv_t *priv) * in case there was an encoding change. so... things are kept * simple here. */ - ret = changelog_fill_rollover_data(&cld, _gf_false); - if (ret) - goto out; + changelog_fill_rollover_data(&cld, _gf_false); ret = htime_open(this, priv, cld.cld_roll_time); /* call htime open with cld's rollover_time */ @@ -2405,6 +2419,22 @@ changelog_barrier_pthread_destroy(changelog_priv_t *priv) LOCK_DESTROY(&priv->bflags.lock); } +static void +changelog_cleanup_rpc(xlator_t *this, changelog_priv_t *priv) +{ + /* terminate rpc server */ + if (!this->cleanup_starting) + changelog_destroy_rpc_listner(this, priv); + + (void)changelog_cleanup_rpc_threads(this, priv); + /* cleanup rot buffs */ + rbuf_dtor(priv->rbuf); + + /* cleanup poller thread */ + if (priv->poller) + (void)changelog_thread_cleanup(this, priv->poller); +} + int reconfigure(xlator_t *this, dict_t *options) { @@ -2413,6 +2443,9 @@ reconfigure(xlator_t *this, dict_t *options) changelog_priv_t *priv = NULL; gf_boolean_t active_earlier = _gf_true; gf_boolean_t active_now = _gf_true; + gf_boolean_t rpc_active_earlier = _gf_true; + gf_boolean_t rpc_active_now = _gf_true; + gf_boolean_t iniate_rpc = _gf_false; changelog_time_slice_t *slice = NULL; changelog_log_data_t cld = { 0, @@ -2423,9 +2456,6 @@ reconfigure(xlator_t *this, dict_t *options) char csnap_dir[PATH_MAX] = { 0, }; - struct timeval tv = { - 0, - }; uint32_t timeout = 0; priv = this->private; @@ -2434,6 +2464,7 @@ reconfigure(xlator_t *this, dict_t *options) ret = -1; active_earlier = priv->active; + rpc_active_earlier = priv->rpc_active; /* first stop the rollover and the fsync thread */ changelog_cleanup_helper_threads(this, priv); @@ -2467,6 +2498,29 @@ reconfigure(xlator_t *this, dict_t *options) goto out; GF_OPTION_RECONF("changelog", active_now, options, bool, out); + GF_OPTION_RECONF("changelog-notification", rpc_active_now, options, bool, + out); + + /* If journalling is enabled, enable rpc notifications */ + if (active_now && !active_earlier) { + if (!rpc_active_earlier) + iniate_rpc = _gf_true; + } + + if (rpc_active_now && !rpc_active_earlier) { + iniate_rpc = _gf_true; + } + + /* TODO: Disable of changelog-notifications is not supported for now + * as there is no clean way of cleaning up of rpc resources + */ + + if (iniate_rpc) { + ret = changelog_init_rpc(this, priv); + if (ret) + goto out; + priv->rpc_active = _gf_true; + } /** * changelog_handle_change() handles changes that could possibly @@ -2493,9 +2547,7 @@ reconfigure(xlator_t *this, dict_t *options) out); if (active_now || active_earlier) { - ret = changelog_fill_rollover_data(&cld, !active_now); - if (ret) - goto out; + changelog_fill_rollover_data(&cld, !active_now); slice = &priv->slice; @@ -2514,13 +2566,7 @@ reconfigure(xlator_t *this, dict_t *options) if (!active_earlier) { gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_RECONFIGURE, NULL); - if (gettimeofday(&tv, NULL)) { - gf_smsg(this->name, GF_LOG_ERROR, 0, - CHANGELOG_MSG_HTIME_FETCH_FAILED, NULL); - ret = -1; - goto out; - } - htime_create(this, priv, tv.tv_sec); + htime_create(this, priv, gf_time()); } ret = changelog_spawn_helper_threads(this, priv); } @@ -2597,6 +2643,7 @@ changelog_init_options(xlator_t *this, changelog_priv_t *priv) goto dealloc_2; GF_OPTION_INIT("changelog", priv->active, bool, dealloc_2); + GF_OPTION_INIT("changelog-notification", priv->rpc_active, bool, dealloc_2); GF_OPTION_INIT("capture-del-path", priv->capture_del_path, bool, dealloc_2); GF_OPTION_INIT("op-mode", tmp, str, dealloc_2); @@ -2635,22 +2682,6 @@ error_return: return -1; } -static void -changelog_cleanup_rpc(xlator_t *this, changelog_priv_t *priv) -{ - /* terminate rpc server */ - if (!this->cleanup_starting) - changelog_destroy_rpc_listner(this, priv); - - (void)changelog_cleanup_rpc_threads(this, priv); - /* cleanup rot buffs */ - rbuf_dtor(priv->rbuf); - - /* cleanup poller thread */ - if (priv->poller) - (void)changelog_thread_cleanup(this, priv->poller); -} - static int changelog_init_rpc(xlator_t *this, changelog_priv_t *priv) { @@ -2747,10 +2778,13 @@ init(xlator_t *this) INIT_LIST_HEAD(&priv->queue); priv->barrier_enabled = _gf_false; - /* RPC ball rolling.. */ - ret = changelog_init_rpc(this, priv); - if (ret) - goto cleanup_barrier; + if (priv->rpc_active || priv->active) { + /* RPC ball rolling.. */ + ret = changelog_init_rpc(this, priv); + if (ret) + goto cleanup_barrier; + priv->rpc_active = _gf_true; + } ret = changelog_init(this, priv); if (ret) @@ -2762,7 +2796,9 @@ init(xlator_t *this) return 0; cleanup_rpc: - changelog_cleanup_rpc(this, priv); + if (priv->rpc_active) { + changelog_cleanup_rpc(this, priv); + } cleanup_barrier: changelog_barrier_pthread_destroy(priv); cleanup_options: @@ -2788,9 +2824,11 @@ fini(xlator_t *this) priv = this->private; if (priv) { - /* terminate RPC server/threads */ - changelog_cleanup_rpc(this, priv); - + if (priv->active || priv->rpc_active) { + /* terminate RPC server/threads */ + changelog_cleanup_rpc(this, priv); + GF_FREE(priv->ev_dispatcher); + } /* call barrier_disable to cancel timer */ if (priv->barrier_enabled) __chlog_barrier_disable(this, &queue); @@ -2859,6 +2897,13 @@ struct volume_options options[] = { .flags = OPT_FLAG_SETTABLE, .level = OPT_STATUS_BASIC, .tags = {"journal", "georep", "glusterfind"}}, + {.key = {"changelog-notification"}, + .type = GF_OPTION_TYPE_BOOL, + .default_value = "off", + .description = "enable/disable changelog live notification", + .op_version = {3}, + .level = OPT_STATUS_BASIC, + .tags = {"bitrot", "georep"}}, {.key = {"changelog-brick"}, .type = GF_OPTION_TYPE_PATH, .description = "brick path to generate unique socket file name." diff --git a/xlators/features/cloudsync/src/Makefile.am b/xlators/features/cloudsync/src/Makefile.am index 0c3966c..e2a277e 100644 --- a/xlators/features/cloudsync/src/Makefile.am +++ b/xlators/features/cloudsync/src/Makefile.am @@ -21,9 +21,9 @@ cloudsync_la_SOURCES = $(cloudsync_sources) $(cloudsynccommon_sources) nodist_cloudsync_la_SOURCES = cloudsync-autogen-fops.c cloudsync-autogen-fops.h BUILT_SOURCES = cloudsync-autogen-fops.h -cloudsync_la_LDFLAGS = $(LIB_DL) -module $(GF_XLATOR_DEFAULT_LDFLAGS) +cloudsync_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS) -cloudsync_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la +cloudsync_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la $(LIB_DL) AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \ -DCS_PLUGINDIR=\"$(libdir)/glusterfs/$(PACKAGE_VERSION)/cloudsync-plugins\" diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.c b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.c index 7680260..23c3599 100644 --- a/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.c +++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.c @@ -237,7 +237,7 @@ aws_form_request(char *resource, char **date, char *reqtype, char *bucketid, int date_len = -1; int res_len = -1; - ctime = time(NULL); + ctime = gf_time(); gtime = gmtime(&ctime); date_len = strftime(httpdate, sizeof(httpdate), diff --git a/xlators/features/index/src/index.c b/xlators/features/index/src/index.c index 4ece7ff..4abb2c7 100644 --- a/xlators/features/index/src/index.c +++ b/xlators/features/index/src/index.c @@ -2104,7 +2104,7 @@ index_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req) worker_enqueue(this, stub); return 0; normal: - ret = dict_get_str(xattr_req, "link-count", &flag); + ret = dict_get_str_sizen(xattr_req, "link-count", &flag); if ((ret == 0) && (strcmp(flag, GF_XATTROP_INDEX_COUNT) == 0)) { STACK_WIND(frame, index_lookup_cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->lookup, loc, xattr_req); @@ -2592,7 +2592,7 @@ notify(xlator_t *this, int event, void *data, ...) if ((event == GF_EVENT_PARENT_DOWN) && victim->cleanup_starting) { stub_cnt = GF_ATOMIC_GET(priv->stub_cnt); - clock_gettime(CLOCK_REALTIME, &sleep_till); + timespec_now_realtime(&sleep_till); sleep_till.tv_sec += 1; /* Wait for draining stub from queue before notify PARENT_DOWN */ diff --git a/xlators/features/leases/src/leases-internal.c b/xlators/features/leases/src/leases-internal.c index 67fdd53..56dee24 100644 --- a/xlators/features/leases/src/leases-internal.c +++ b/xlators/features/leases/src/leases-internal.c @@ -897,7 +897,7 @@ __recall_lease(xlator_t *this, lease_inode_ctx_t *lease_ctx) } priv = this->private; - recall_time = time(NULL); + recall_time = gf_time(); list_for_each_entry_safe(lease_entry, tmp, &lease_ctx->lease_id_list, lease_id_list) { @@ -1367,7 +1367,7 @@ expired_recall_cleanup(void *data) gf_msg_debug(this->name, 0, "Started the expired_recall_cleanup thread"); while (1) { - time_now = time(NULL); + time_now = gf_time(); pthread_mutex_lock(&priv->mutex); { if (priv->fini) { diff --git a/xlators/features/locks/src/clear.c b/xlators/features/locks/src/clear.c index 116aed6..ab1eac6 100644 --- a/xlators/features/locks/src/clear.c +++ b/xlators/features/locks/src/clear.c @@ -181,9 +181,9 @@ clrlk_clear_posixlk(xlator_t *this, pl_inode_t *pl_inode, clrlk_args *args, if (plock->blocked) { bcount++; pl_trace_out(this, plock->frame, NULL, NULL, F_SETLKW, - &plock->user_flock, -1, EAGAIN, NULL); + &plock->user_flock, -1, EINTR, NULL); - STACK_UNWIND_STRICT(lk, plock->frame, -1, EAGAIN, + STACK_UNWIND_STRICT(lk, plock->frame, -1, EINTR, &plock->user_flock, NULL); } else { diff --git a/xlators/features/locks/src/common.c b/xlators/features/locks/src/common.c index c78d537..a2c6be9 100644 --- a/xlators/features/locks/src/common.c +++ b/xlators/features/locks/src/common.c @@ -460,11 +460,16 @@ pl_inode_get(xlator_t *this, inode_t *inode, pl_local_t *local) INIT_LIST_HEAD(&pl_inode->blocked_calls); INIT_LIST_HEAD(&pl_inode->metalk_list); INIT_LIST_HEAD(&pl_inode->queued_locks); + INIT_LIST_HEAD(&pl_inode->waiting); gf_uuid_copy(pl_inode->gfid, inode->gfid); pl_inode->check_mlock_info = _gf_true; pl_inode->mlock_enforced = _gf_false; + /* -2 means never looked up. -1 means something went wrong and link + * tracking is disabled. */ + pl_inode->links = -2; + ret = __inode_ctx_put(inode, this, (uint64_t)(long)(pl_inode)); if (ret) { pthread_mutex_destroy(&pl_inode->mutex); @@ -600,13 +605,11 @@ static void __insert_lock(pl_inode_t *pl_inode, posix_lock_t *lock) { if (lock->blocked) - gettimeofday(&lock->blkd_time, NULL); + lock->blkd_time = gf_time(); else - gettimeofday(&lock->granted_time, NULL); + lock->granted_time = gf_time(); list_add_tail(&lock->list, &pl_inode->ext_list); - - return; } /* Return true if the locks overlap, false otherwise */ @@ -962,7 +965,7 @@ grant_blocked_locks(xlator_t *this, pl_inode_t *pl_inode) struct list_head granted_list; posix_lock_t *tmp = NULL; posix_lock_t *lock = NULL; - + pl_local_t *local = NULL; INIT_LIST_HEAD(&granted_list); pthread_mutex_lock(&pl_inode->mutex); @@ -977,9 +980,9 @@ grant_blocked_locks(xlator_t *this, pl_inode_t *pl_inode) pl_trace_out(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock, 0, 0, NULL); - - STACK_UNWIND_STRICT(lk, lock->frame, 0, 0, &lock->user_flock, NULL); - + local = lock->frame->local; + PL_STACK_UNWIND_AND_FREE(local, lk, lock->frame, 0, 0, + &lock->user_flock, NULL); __destroy_lock(lock); } @@ -999,6 +1002,7 @@ pl_send_prelock_unlock(xlator_t *this, pl_inode_t *pl_inode, struct list_head granted_list; posix_lock_t *tmp = NULL; posix_lock_t *lock = NULL; + pl_local_t *local = NULL; int ret = -1; @@ -1026,9 +1030,9 @@ pl_send_prelock_unlock(xlator_t *this, pl_inode_t *pl_inode, pl_trace_out(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock, 0, 0, NULL); - - STACK_UNWIND_STRICT(lk, lock->frame, 0, 0, &lock->user_flock, NULL); - + local = lock->frame->local; + PL_STACK_UNWIND_AND_FREE(local, lk, lock->frame, 0, 0, + &lock->user_flock, NULL); __destroy_lock(lock); } @@ -1289,3 +1293,299 @@ pl_is_lk_owner_valid(gf_lkowner_t *owner, client_t *client) } return _gf_true; } + +static int32_t +pl_inode_from_loc(loc_t *loc, inode_t **pinode) +{ + inode_t *inode = NULL; + int32_t error = 0; + + if (loc->inode != NULL) { + inode = inode_ref(loc->inode); + goto done; + } + + if (loc->parent == NULL) { + error = EINVAL; + goto done; + } + + if (!gf_uuid_is_null(loc->gfid)) { + inode = inode_find(loc->parent->table, loc->gfid); + if (inode != NULL) { + goto done; + } + } + + if (loc->name == NULL) { + error = EINVAL; + goto done; + } + + inode = inode_grep(loc->parent->table, loc->parent, loc->name); + if (inode == NULL) { + /* We haven't found any inode. This means that the file doesn't exist + * or that even if it exists, we don't have any knowledge about it, so + * we don't have locks on it either, which is fine for our purposes. */ + goto done; + } + +done: + *pinode = inode; + + return error; +} + +static gf_boolean_t +pl_inode_has_owners(xlator_t *xl, client_t *client, pl_inode_t *pl_inode, + struct timespec *now, struct list_head *contend) +{ + pl_dom_list_t *dom; + pl_inode_lock_t *lock; + gf_boolean_t has_owners = _gf_false; + + list_for_each_entry(dom, &pl_inode->dom_list, inode_list) + { + list_for_each_entry(lock, &dom->inodelk_list, list) + { + /* If the lock belongs to the same client, we assume it's related + * to the same operation, so we allow the removal to continue. */ + if (lock->client == client) { + continue; + } + /* If the lock belongs to an internal process, we don't block the + * removal. */ + if (lock->client_pid < 0) { + continue; + } + if (contend == NULL) { + return _gf_true; + } + has_owners = _gf_true; + inodelk_contention_notify_check(xl, lock, now, contend); + } + } + + return has_owners; +} + +int32_t +pl_inode_remove_prepare(xlator_t *xl, call_frame_t *frame, loc_t *loc, + pl_inode_t **ppl_inode, struct list_head *contend) +{ + struct timespec now; + inode_t *inode; + pl_inode_t *pl_inode; + int32_t error; + + pl_inode = NULL; + + error = pl_inode_from_loc(loc, &inode); + if ((error != 0) || (inode == NULL)) { + goto done; + } + + pl_inode = pl_inode_get(xl, inode, NULL); + if (pl_inode == NULL) { + inode_unref(inode); + error = ENOMEM; + goto done; + } + + /* pl_inode_from_loc() already increments ref count for inode, so + * we only assign here our reference. */ + pl_inode->inode = inode; + + timespec_now(&now); + + pthread_mutex_lock(&pl_inode->mutex); + + if (pl_inode->removed) { + error = ESTALE; + goto unlock; + } + + if (pl_inode_has_owners(xl, frame->root->client, pl_inode, &now, contend)) { + error = -1; + /* We skip the unlock here because the caller must create a stub when + * we return -1 and do a call to pl_inode_remove_complete(), which + * assumes the lock is still acquired and will release it once + * everything else is prepared. */ + goto done; + } + + pl_inode->is_locked = _gf_true; + pl_inode->remove_running++; + +unlock: + pthread_mutex_unlock(&pl_inode->mutex); + +done: + *ppl_inode = pl_inode; + + return error; +} + +int32_t +pl_inode_remove_complete(xlator_t *xl, pl_inode_t *pl_inode, call_stub_t *stub, + struct list_head *contend) +{ + pl_inode_lock_t *lock; + int32_t error = -1; + + if (stub != NULL) { + list_add_tail(&stub->list, &pl_inode->waiting); + pl_inode->is_locked = _gf_true; + } else { + error = ENOMEM; + + while (!list_empty(contend)) { + lock = list_first_entry(contend, pl_inode_lock_t, list); + list_del_init(&lock->list); + __pl_inodelk_unref(lock); + } + } + + pthread_mutex_unlock(&pl_inode->mutex); + + if (error < 0) { + inodelk_contention_notify(xl, contend); + } + + inode_unref(pl_inode->inode); + + return error; +} + +void +pl_inode_remove_wake(struct list_head *list) +{ + call_stub_t *stub; + + while (!list_empty(list)) { + stub = list_first_entry(list, call_stub_t, list); + list_del_init(&stub->list); + + call_resume(stub); + } +} + +void +pl_inode_remove_cbk(xlator_t *xl, pl_inode_t *pl_inode, int32_t error) +{ + struct list_head contend, granted; + struct timespec now; + pl_dom_list_t *dom; + + if (pl_inode == NULL) { + return; + } + + INIT_LIST_HEAD(&contend); + INIT_LIST_HEAD(&granted); + timespec_now(&now); + + pthread_mutex_lock(&pl_inode->mutex); + + if (error == 0) { + if (pl_inode->links >= 0) { + pl_inode->links--; + } + if (pl_inode->links == 0) { + pl_inode->removed = _gf_true; + } + } + + pl_inode->remove_running--; + + if ((pl_inode->remove_running == 0) && list_empty(&pl_inode->waiting)) { + pl_inode->is_locked = _gf_false; + + list_for_each_entry(dom, &pl_inode->dom_list, inode_list) + { + __grant_blocked_inode_locks(xl, pl_inode, &granted, dom, &now, + &contend); + } + } + + pthread_mutex_unlock(&pl_inode->mutex); + + unwind_granted_inodes(xl, pl_inode, &granted); + + inodelk_contention_notify(xl, &contend); + + inode_unref(pl_inode->inode); +} + +void +pl_inode_remove_unlocked(xlator_t *xl, pl_inode_t *pl_inode, + struct list_head *list) +{ + call_stub_t *stub, *tmp; + + if (!pl_inode->is_locked) { + return; + } + + list_for_each_entry_safe(stub, tmp, &pl_inode->waiting, list) + { + if (!pl_inode_has_owners(xl, stub->frame->root->client, pl_inode, NULL, + NULL)) { + list_move_tail(&stub->list, list); + } + } +} + +/* This function determines if an inodelk attempt can be done now or it needs + * to wait. + * + * Possible return values: + * < 0: An error occurred. Currently only -ESTALE can be returned if the + * inode has been deleted previously by unlink/rmdir/rename + * = 0: The lock can be attempted. + * > 0: The lock needs to wait because a conflicting remove operation is + * ongoing. + */ +int32_t +pl_inode_remove_inodelk(pl_inode_t *pl_inode, pl_inode_lock_t *lock) +{ + pl_dom_list_t *dom; + pl_inode_lock_t *ilock; + + /* If the inode has been deleted, we won't allow any lock. */ + if (pl_inode->removed) { + return -ESTALE; + } + + /* We only synchronize with locks made for regular operations coming from + * the user. Locks done for internal purposes are hard to control and could + * lead to long delays or deadlocks quite easily. */ + if (lock->client_pid < 0) { + return 0; + } + if (!pl_inode->is_locked) { + return 0; + } + if (pl_inode->remove_running > 0) { + return 1; + } + + list_for_each_entry(dom, &pl_inode->dom_list, inode_list) + { + list_for_each_entry(ilock, &dom->inodelk_list, list) + { + /* If a lock from the same client is already granted, we allow this + * one to continue. This is necessary to prevent deadlocks when + * multiple locks are taken for the same operation. + * + * On the other side it's unlikely that the same client sends + * completely unrelated locks for the same inode. + */ + if (ilock->client == lock->client) { + return 0; + } + } + } + + return 1; +} diff --git a/xlators/features/locks/src/common.h b/xlators/features/locks/src/common.h index 0916c29..281223b 100644 --- a/xlators/features/locks/src/common.h +++ b/xlators/features/locks/src/common.h @@ -105,6 +105,15 @@ void __pl_inodelk_unref(pl_inode_lock_t *lock); void +__grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode, + struct list_head *granted, pl_dom_list_t *dom, + struct timespec *now, struct list_head *contend); + +void +unwind_granted_inodes(xlator_t *this, pl_inode_t *pl_inode, + struct list_head *granted); + +void grant_blocked_entry_locks(xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom, struct timespec *now, struct list_head *contend); @@ -204,6 +213,16 @@ pl_metalock_is_active(pl_inode_t *pl_inode); void __pl_queue_lock(pl_inode_t *pl_inode, posix_lock_t *reqlock); +void +inodelk_contention_notify_check(xlator_t *xl, pl_inode_lock_t *lock, + struct timespec *now, + struct list_head *contend); + +void +entrylk_contention_notify_check(xlator_t *xl, pl_entry_lock_t *lock, + struct timespec *now, + struct list_head *contend); + gf_boolean_t pl_does_monkey_want_stuck_lock(); @@ -218,4 +237,26 @@ pl_local_init(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd); gf_boolean_t pl_is_lk_owner_valid(gf_lkowner_t *owner, client_t *client); + +int32_t +pl_inode_remove_prepare(xlator_t *xl, call_frame_t *frame, loc_t *loc, + pl_inode_t **ppl_inode, struct list_head *contend); + +int32_t +pl_inode_remove_complete(xlator_t *xl, pl_inode_t *pl_inode, call_stub_t *stub, + struct list_head *contend); + +void +pl_inode_remove_wake(struct list_head *list); + +void +pl_inode_remove_cbk(xlator_t *xl, pl_inode_t *pl_inode, int32_t error); + +void +pl_inode_remove_unlocked(xlator_t *xl, pl_inode_t *pl_inode, + struct list_head *list); + +int32_t +pl_inode_remove_inodelk(pl_inode_t *pl_inode, pl_inode_lock_t *lock); + #endif /* __COMMON_H__ */ diff --git a/xlators/features/locks/src/entrylk.c b/xlators/features/locks/src/entrylk.c index 0911659..fd772c8 100644 --- a/xlators/features/locks/src/entrylk.c +++ b/xlators/features/locks/src/entrylk.c @@ -121,7 +121,6 @@ __stale_entrylk(xlator_t *this, pl_entry_lock_t *candidate_lock, pl_entry_lock_t *requested_lock, time_t *lock_age_sec) { posix_locks_private_t *priv = NULL; - struct timeval curr; priv = this->private; @@ -129,8 +128,7 @@ __stale_entrylk(xlator_t *this, pl_entry_lock_t *candidate_lock, * chance? Or just the locks we are attempting to acquire? */ if (names_conflict(candidate_lock->basename, requested_lock->basename)) { - gettimeofday(&curr, NULL); - *lock_age_sec = curr.tv_sec - candidate_lock->granted_time.tv_sec; + *lock_age_sec = gf_time() - candidate_lock->granted_time; if (*lock_age_sec > priv->revocation_secs) return _gf_true; } @@ -204,9 +202,9 @@ out: return revoke_lock; } -static gf_boolean_t -__entrylk_needs_contention_notify(xlator_t *this, pl_entry_lock_t *lock, - struct timespec *now) +void +entrylk_contention_notify_check(xlator_t *this, pl_entry_lock_t *lock, + struct timespec *now, struct list_head *contend) { posix_locks_private_t *priv; int64_t elapsed; @@ -216,7 +214,7 @@ __entrylk_needs_contention_notify(xlator_t *this, pl_entry_lock_t *lock, /* If this lock is in a list, it means that we are about to send a * notification for it, so no need to do anything else. */ if (!list_empty(&lock->contend)) { - return _gf_false; + return; } elapsed = now->tv_sec; @@ -225,7 +223,7 @@ __entrylk_needs_contention_notify(xlator_t *this, pl_entry_lock_t *lock, elapsed--; } if (elapsed < priv->notify_contention_delay) { - return _gf_false; + return; } /* All contention notifications will be sent outside of the locked @@ -238,7 +236,7 @@ __entrylk_needs_contention_notify(xlator_t *this, pl_entry_lock_t *lock, lock->contention_time = *now; - return _gf_true; + list_add_tail(&lock->contend, contend); } void @@ -332,9 +330,7 @@ __entrylk_grantable(xlator_t *this, pl_dom_list_t *dom, pl_entry_lock_t *lock, break; } } - if (__entrylk_needs_contention_notify(this, tmp, now)) { - list_add_tail(&tmp->contend, contend); - } + entrylk_contention_notify_check(this, tmp, now, contend); } } @@ -546,14 +542,10 @@ static int __lock_blocked_add(xlator_t *this, pl_inode_t *pinode, pl_dom_list_t *dom, pl_entry_lock_t *lock, int nonblock) { - struct timeval now; - if (nonblock) goto out; - gettimeofday(&now, NULL); - - lock->blkd_time = now; + lock->blkd_time = gf_time(); list_add_tail(&lock->blocked_locks, &dom->blocked_entrylks); gf_msg_trace(this->name, 0, "Blocking lock: {pinode=%p, basename=%s}", @@ -614,7 +606,7 @@ __lock_entrylk(xlator_t *this, pl_inode_t *pinode, pl_entry_lock_t *lock, } __pl_entrylk_ref(lock); - gettimeofday(&lock->granted_time, NULL); + lock->granted_time = gf_time(); list_add(&lock->domain_list, &dom->entrylk_list); ret = 0; @@ -697,10 +689,9 @@ __grant_blocked_entry_locks(xlator_t *this, pl_inode_t *pl_inode, bl_ret = __lock_entrylk(bl->this, pl_inode, bl, 0, dom, now, contend); if (bl_ret == 0) { - list_add(&bl->blocked_locks, granted); + list_add_tail(&bl->blocked_locks, granted); } } - return; } /* Grants locks if possible which are blocked on a lock */ diff --git a/xlators/features/locks/src/inodelk.c b/xlators/features/locks/src/inodelk.c index 6022e5a..d4e51d6 100644 --- a/xlators/features/locks/src/inodelk.c +++ b/xlators/features/locks/src/inodelk.c @@ -140,15 +140,13 @@ __stale_inodelk(xlator_t *this, pl_inode_lock_t *candidate_lock, pl_inode_lock_t *requested_lock, time_t *lock_age_sec) { posix_locks_private_t *priv = NULL; - struct timeval curr; priv = this->private; /* Question: Should we just prune them all given the * chance? Or just the locks we are attempting to acquire? */ if (inodelk_conflict(candidate_lock, requested_lock)) { - gettimeofday(&curr, NULL); - *lock_age_sec = curr.tv_sec - candidate_lock->granted_time.tv_sec; + *lock_age_sec = gf_time() - candidate_lock->granted_time; if (*lock_age_sec > priv->revocation_secs) return _gf_true; } @@ -229,9 +227,9 @@ out: return revoke_lock; } -static gf_boolean_t -__inodelk_needs_contention_notify(xlator_t *this, pl_inode_lock_t *lock, - struct timespec *now) +void +inodelk_contention_notify_check(xlator_t *this, pl_inode_lock_t *lock, + struct timespec *now, struct list_head *contend) { posix_locks_private_t *priv; int64_t elapsed; @@ -241,7 +239,7 @@ __inodelk_needs_contention_notify(xlator_t *this, pl_inode_lock_t *lock, /* If this lock is in a list, it means that we are about to send a * notification for it, so no need to do anything else. */ if (!list_empty(&lock->contend)) { - return _gf_false; + return; } elapsed = now->tv_sec; @@ -250,7 +248,7 @@ __inodelk_needs_contention_notify(xlator_t *this, pl_inode_lock_t *lock, elapsed--; } if (elapsed < priv->notify_contention_delay) { - return _gf_false; + return; } /* All contention notifications will be sent outside of the locked @@ -263,7 +261,7 @@ __inodelk_needs_contention_notify(xlator_t *this, pl_inode_lock_t *lock, lock->contention_time = *now; - return _gf_true; + list_add_tail(&lock->contend, contend); } void @@ -351,9 +349,7 @@ __inodelk_grantable(xlator_t *this, pl_dom_list_t *dom, pl_inode_lock_t *lock, break; } } - if (__inodelk_needs_contention_notify(this, l, now)) { - list_add_tail(&l->contend, contend); - } + inodelk_contention_notify_check(this, l, now, contend); } } @@ -399,15 +395,11 @@ static int __lock_blocked_add(xlator_t *this, pl_dom_list_t *dom, pl_inode_lock_t *lock, int can_block) { - struct timeval now; - if (can_block == 0) { goto out; } - gettimeofday(&now, NULL); - - lock->blkd_time = now; + lock->blkd_time = gf_time(); list_add_tail(&lock->blocked_locks, &dom->blocked_inodelks); gf_msg_trace(this->name, 0, @@ -433,12 +425,17 @@ __lock_inodelk(xlator_t *this, pl_inode_t *pl_inode, pl_inode_lock_t *lock, struct list_head *contend) { pl_inode_lock_t *conf = NULL; - int ret = -EINVAL; + int ret; - conf = __inodelk_grantable(this, dom, lock, now, contend); - if (conf) { - ret = __lock_blocked_add(this, dom, lock, can_block); - goto out; + ret = pl_inode_remove_inodelk(pl_inode, lock); + if (ret < 0) { + return ret; + } + if (ret == 0) { + conf = __inodelk_grantable(this, dom, lock, now, contend); + } + if ((ret > 0) || (conf != NULL)) { + return __lock_blocked_add(this, dom, lock, can_block); } /* To prevent blocked locks starvation, check if there are any blocked @@ -460,17 +457,13 @@ __lock_inodelk(xlator_t *this, pl_inode_t *pl_inode, pl_inode_lock_t *lock, "starvation"); } - ret = __lock_blocked_add(this, dom, lock, can_block); - goto out; + return __lock_blocked_add(this, dom, lock, can_block); } __pl_inodelk_ref(lock); - gettimeofday(&lock->granted_time, NULL); + lock->granted_time = gf_time(); list_add(&lock->list, &dom->inodelk_list); - ret = 0; - -out: - return ret; + return 0; } /* Return true if the two inodelks have exactly same lock boundaries */ @@ -502,33 +495,36 @@ static pl_inode_lock_t * __inode_unlock_lock(xlator_t *this, pl_inode_lock_t *lock, pl_dom_list_t *dom) { pl_inode_lock_t *conf = NULL; + inode_t *inode = NULL; + + inode = lock->pl_inode->inode; conf = find_matching_inodelk(lock, dom); if (!conf) { gf_log(this->name, GF_LOG_ERROR, " Matching lock not found for unlock %llu-%llu, by %s " - "on %p", + "on %p for gfid:%s", (unsigned long long)lock->fl_start, (unsigned long long)lock->fl_end, lkowner_utoa(&lock->owner), - lock->client); + lock->client, inode ? uuid_utoa(inode->gfid) : "UNKNOWN"); goto out; } __delete_inode_lock(conf); gf_log(this->name, GF_LOG_DEBUG, - " Matching lock found for unlock %llu-%llu, by %s on %p", + " Matching lock found for unlock %llu-%llu, by %s on %p for gfid:%s", (unsigned long long)lock->fl_start, (unsigned long long)lock->fl_end, - lkowner_utoa(&lock->owner), lock->client); + lkowner_utoa(&lock->owner), lock->client, + inode ? uuid_utoa(inode->gfid) : "UNKNOWN"); out: return conf; } -static void +void __grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode, struct list_head *granted, pl_dom_list_t *dom, struct timespec *now, struct list_head *contend) { - int bl_ret = 0; pl_inode_lock_t *bl = NULL; pl_inode_lock_t *tmp = NULL; @@ -541,52 +537,48 @@ __grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode, { list_del_init(&bl->blocked_locks); - bl_ret = __lock_inodelk(this, pl_inode, bl, 1, dom, now, contend); + bl->status = __lock_inodelk(this, pl_inode, bl, 1, dom, now, contend); - if (bl_ret == 0) { - list_add(&bl->blocked_locks, granted); + if (bl->status != -EAGAIN) { + list_add_tail(&bl->blocked_locks, granted); } } - return; } -/* Grant all inodelks blocked on a lock */ void -grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode, - pl_dom_list_t *dom, struct timespec *now, - struct list_head *contend) +unwind_granted_inodes(xlator_t *this, pl_inode_t *pl_inode, + struct list_head *granted) { - struct list_head granted; pl_inode_lock_t *lock; pl_inode_lock_t *tmp; + int32_t op_ret; + int32_t op_errno; - INIT_LIST_HEAD(&granted); - - pthread_mutex_lock(&pl_inode->mutex); - { - __grant_blocked_inode_locks(this, pl_inode, &granted, dom, now, - contend); - } - pthread_mutex_unlock(&pl_inode->mutex); - - list_for_each_entry_safe(lock, tmp, &granted, blocked_locks) + list_for_each_entry_safe(lock, tmp, granted, blocked_locks) { - gf_log(this->name, GF_LOG_TRACE, - "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64 " => Granted", - lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid, - lkowner_utoa(&lock->owner), lock->user_flock.l_start, - lock->user_flock.l_len); - + if (lock->status == 0) { + op_ret = 0; + op_errno = 0; + gf_log(this->name, GF_LOG_TRACE, + "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64 + " => Granted", + lock->fl_type == F_UNLCK ? "Unlock" : "Lock", + lock->client_pid, lkowner_utoa(&lock->owner), + lock->user_flock.l_start, lock->user_flock.l_len); + } else { + op_ret = -1; + op_errno = -lock->status; + } pl_trace_out(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock, - 0, 0, lock->volume); + op_ret, op_errno, lock->volume); - STACK_UNWIND_STRICT(inodelk, lock->frame, 0, 0, NULL); + STACK_UNWIND_STRICT(inodelk, lock->frame, op_ret, op_errno, NULL); lock->frame = NULL; } pthread_mutex_lock(&pl_inode->mutex); { - list_for_each_entry_safe(lock, tmp, &granted, blocked_locks) + list_for_each_entry_safe(lock, tmp, granted, blocked_locks) { list_del_init(&lock->blocked_locks); __pl_inodelk_unref(lock); @@ -595,6 +587,26 @@ grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode, pthread_mutex_unlock(&pl_inode->mutex); } +/* Grant all inodelks blocked on a lock */ +void +grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode, + pl_dom_list_t *dom, struct timespec *now, + struct list_head *contend) +{ + struct list_head granted; + + INIT_LIST_HEAD(&granted); + + pthread_mutex_lock(&pl_inode->mutex); + { + __grant_blocked_inode_locks(this, pl_inode, &granted, dom, now, + contend); + } + pthread_mutex_unlock(&pl_inode->mutex); + + unwind_granted_inodes(this, pl_inode, &granted); +} + static void pl_inodelk_log_cleanup(pl_inode_lock_t *lock) { @@ -656,7 +668,7 @@ pl_inodelk_client_cleanup(xlator_t *this, pl_ctx_t *ctx) * and blocked lists, then this means that a parallel * unlock on another inodelk (L2 say) may have 'granted' * L1 and added it to 'granted' list in - * __grant_blocked_node_locks() (although using the + * __grant_blocked_inode_locks() (although using the * 'blocked_locks' member). In that case, the cleanup * codepath must try and grant other overlapping * blocked inodelks from other clients, now that L1 is @@ -741,6 +753,7 @@ pl_inode_setlk(xlator_t *this, pl_ctx_t *ctx, pl_inode_t *pl_inode, gf_boolean_t need_inode_unref = _gf_false; struct list_head *pcontend = NULL; struct list_head contend; + struct list_head wake; struct timespec now = {}; short fl_type; @@ -792,6 +805,8 @@ pl_inode_setlk(xlator_t *this, pl_ctx_t *ctx, pl_inode_t *pl_inode, timespec_now(&now); } + INIT_LIST_HEAD(&wake); + if (ctx) pthread_mutex_lock(&ctx->lock); pthread_mutex_lock(&pl_inode->mutex); @@ -814,18 +829,17 @@ pl_inode_setlk(xlator_t *this, pl_ctx_t *ctx, pl_inode_t *pl_inode, lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid, lkowner_utoa(&lock->owner), lock->user_flock.l_start, lock->user_flock.l_len); - if (can_block) + if (can_block) { unref = _gf_false; - /* For all but the case where a non-blocking - * lock attempt fails, the extra ref taken at - * the start of this function must be negated. - */ - else - need_inode_unref = _gf_true; + } } - - if (ctx && (!ret || can_block)) + /* For all but the case where a non-blocking lock attempt fails + * with -EAGAIN, the extra ref taken at the start of this function + * must be negated. */ + need_inode_unref = (ret != 0) && ((ret != -EAGAIN) || !can_block); + if (ctx && !need_inode_unref) { list_add_tail(&lock->client_list, &ctx->inodelk_lockers); + } } else { /* Irrespective of whether unlock succeeds or not, * the extra inode ref that was done at the start of @@ -843,6 +857,8 @@ pl_inode_setlk(xlator_t *this, pl_ctx_t *ctx, pl_inode_t *pl_inode, list_del_init(&retlock->client_list); __pl_inodelk_unref(retlock); + pl_inode_remove_unlocked(this, pl_inode, &wake); + ret = 0; } out: @@ -853,6 +869,8 @@ pl_inode_setlk(xlator_t *this, pl_ctx_t *ctx, pl_inode_t *pl_inode, if (ctx) pthread_mutex_unlock(&ctx->lock); + pl_inode_remove_wake(&wake); + /* The following (extra) unref corresponds to the ref that * was done at the time the lock was granted. */ @@ -1033,10 +1051,14 @@ pl_common_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, inode); if (ret < 0) { - if ((can_block) && (F_UNLCK != lock_type)) { - goto out; + if (ret == -EAGAIN) { + if (can_block && (F_UNLCK != lock_type)) { + goto out; + } + gf_log(this->name, GF_LOG_TRACE, "returning EAGAIN"); + } else { + gf_log(this->name, GF_LOG_TRACE, "returning %d", ret); } - gf_log(this->name, GF_LOG_TRACE, "returning EAGAIN"); op_errno = -ret; goto unwind; } diff --git a/xlators/features/locks/src/locks.h b/xlators/features/locks/src/locks.h index 3305350..c868eb4 100644 --- a/xlators/features/locks/src/locks.h +++ b/xlators/features/locks/src/locks.h @@ -43,9 +43,8 @@ struct __posix_lock { fd_t *fd; call_frame_t *frame; - struct timeval blkd_time; /*time at which lock was queued into blkd list*/ - struct timeval - granted_time; /*time at which lock was queued into active list*/ + time_t blkd_time; /* time at which lock was queued into blkd list */ + time_t granted_time; /* time at which lock was queued into active list */ /* These two together serve to uniquely identify each process across nodes */ @@ -85,9 +84,9 @@ struct __pl_inode_lock { call_frame_t *frame; - struct timeval blkd_time; /*time at which lock was queued into blkd list*/ - struct timeval - granted_time; /*time at which lock was queued into active list*/ + time_t blkd_time; /* time at which lock was queued into blkd list */ + time_t granted_time; /* time at which lock was queued into active list */ + /*last time at which lock contention was detected and notified*/ struct timespec contention_time; @@ -102,6 +101,9 @@ struct __pl_inode_lock { struct list_head client_list; /* list of all locks from a client */ short fl_type; + + int32_t status; /* Error code when we try to grant a lock in blocked + state */ }; typedef struct __pl_inode_lock pl_inode_lock_t; @@ -136,9 +138,9 @@ struct __entry_lock { const char *basename; - struct timeval blkd_time; /*time at which lock was queued into blkd list*/ - struct timeval - granted_time; /*time at which lock was queued into active list*/ + time_t blkd_time; /* time at which lock was queued into blkd list */ + time_t granted_time; /* time at which lock was queued into active list */ + /*last time at which lock contention was detected and notified*/ struct timespec contention_time; @@ -164,13 +166,14 @@ struct __pl_inode { struct list_head rw_list; /* list of waiting r/w requests */ struct list_head reservelk_list; /* list of reservelks */ struct list_head blocked_reservelks; /* list of blocked reservelks */ - struct list_head - blocked_calls; /* List of blocked lock calls while a reserve is held*/ - struct list_head metalk_list; /* Meta lock list */ - /* This is to store the incoming lock - requests while meta lock is enabled */ - struct list_head queued_locks; - int mandatory; /* if mandatory locking is enabled */ + struct list_head blocked_calls; /* List of blocked lock calls while a + reserve is held*/ + struct list_head metalk_list; /* Meta lock list */ + struct list_head queued_locks; /* This is to store the incoming lock + requests while meta lock is enabled */ + struct list_head waiting; /* List of pending fops waiting to unlink/rmdir + the inode. */ + int mandatory; /* if mandatory locking is enabled */ inode_t *refkeeper; /* hold refs on an inode while locks are held to prevent pruning */ @@ -197,7 +200,13 @@ struct __pl_inode { */ int fop_wind_count; pthread_cond_t check_fop_wind_count; + gf_boolean_t track_fop_wind_count; + + int32_t links; /* Number of hard links the inode has. */ + uint32_t remove_running; /* Number of remove operations running. */ + gf_boolean_t is_locked; /* Regular locks will be blocked. */ + gf_boolean_t removed; /* The inode has been deleted. */ }; typedef struct __pl_inode pl_inode_t; diff --git a/xlators/features/locks/src/posix.c b/xlators/features/locks/src/posix.c index af530aa..cf0ae4c 100644 --- a/xlators/features/locks/src/posix.c +++ b/xlators/features/locks/src/posix.c @@ -148,6 +148,29 @@ fetch_pathinfo(xlator_t *, inode_t *, int32_t *, char **); } \ } while (0) +#define PL_INODE_REMOVE(_fop, _frame, _xl, _loc1, _loc2, _cont, _cbk, \ + _args...) \ + ({ \ + struct list_head contend; \ + pl_inode_t *__pl_inode; \ + call_stub_t *__stub; \ + int32_t __error; \ + INIT_LIST_HEAD(&contend); \ + __error = pl_inode_remove_prepare(_xl, _frame, _loc2 ? _loc2 : _loc1, \ + &__pl_inode, &contend); \ + if (__error < 0) { \ + __stub = fop_##_fop##_stub(_frame, _cont, ##_args); \ + __error = pl_inode_remove_complete(_xl, __pl_inode, __stub, \ + &contend); \ + } else if (__error == 0) { \ + PL_LOCAL_GET_REQUESTS(_frame, _xl, xdata, ((fd_t *)NULL), _loc1, \ + _loc2); \ + STACK_WIND_COOKIE(_frame, _cbk, __pl_inode, FIRST_CHILD(_xl), \ + FIRST_CHILD(_xl)->fops->_fop, ##_args); \ + } \ + __error; \ + }) + gf_boolean_t pl_has_xdata_requests(dict_t *xdata) { @@ -471,6 +494,9 @@ pl_inodelk_xattr_fill_multiple(dict_t *this, char *key, data_t *value, char *save_ptr = NULL; tmp_key = gf_strdup(key); + if (!tmp_key) + return -1; + strtok_r(tmp_key, ":", &save_ptr); if (!*save_ptr) { if (tmp_key) @@ -2962,11 +2988,85 @@ out: return ret; } +static int32_t +pl_request_link_count(dict_t **pxdata) +{ + dict_t *xdata; + + xdata = *pxdata; + if (xdata == NULL) { + xdata = dict_new(); + if (xdata == NULL) { + return ENOMEM; + } + } else { + dict_ref(xdata); + } + + if (dict_set_uint32(xdata, GET_LINK_COUNT, 0) != 0) { + dict_unref(xdata); + return ENOMEM; + } + + *pxdata = xdata; + + return 0; +} + +static int32_t +pl_check_link_count(dict_t *xdata) +{ + int32_t count; + + /* In case we are unable to read the link count from xdata, we take a + * conservative approach and return -2, which will prevent the inode from + * being considered deleted. In fact it will cause link tracking for this + * inode to be disabled completely to avoid races. */ + + if (xdata == NULL) { + return -2; + } + + if (dict_get_int32(xdata, GET_LINK_COUNT, &count) != 0) { + return -2; + } + + return count; +} + int32_t pl_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata, struct iatt *postparent) { + pl_inode_t *pl_inode; + + if (op_ret >= 0) { + pl_inode = pl_inode_get(this, inode, NULL); + if (pl_inode == NULL) { + PL_STACK_UNWIND(lookup, xdata, frame, -1, ENOMEM, NULL, NULL, NULL, + NULL); + return 0; + } + + pthread_mutex_lock(&pl_inode->mutex); + + /* We only update the link count if we previously didn't know it. + * Doing it always can lead to races since lookup is not executed + * atomically most of the times. */ + if (pl_inode->links == -2) { + pl_inode->links = pl_check_link_count(xdata); + if (buf->ia_type == IA_IFDIR) { + /* Directories have at least 2 links. To avoid special handling + * for directories, we simply decrement the value here to make + * them equivalent to regular files. */ + pl_inode->links--; + } + } + + pthread_mutex_unlock(&pl_inode->mutex); + } + PL_STACK_UNWIND(lookup, xdata, frame, op_ret, op_errno, inode, buf, xdata, postparent); return 0; @@ -2975,9 +3075,17 @@ pl_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t pl_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata) { - PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL); - STACK_WIND(frame, pl_lookup_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->lookup, loc, xdata); + int32_t error; + + error = pl_request_link_count(&xdata); + if (error == 0) { + PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL); + STACK_WIND(frame, pl_lookup_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->lookup, loc, xdata); + dict_unref(xdata); + } else { + STACK_UNWIND_STRICT(lookup, frame, -1, error, NULL, NULL, NULL, NULL); + } return 0; } @@ -3502,10 +3610,10 @@ pl_dump_lock(char *str, int size, struct gf_flock *flock, gf_lkowner_t *owner, time_t *blkd_time, gf_boolean_t active) { char *type_str = NULL; - char granted[256] = { + char granted[GF_TIMESTR_SIZE] = { 0, }; - char blocked[256] = { + char blocked[GF_TIMESTR_SIZE] = { 0, }; @@ -3556,10 +3664,10 @@ __dump_entrylks(pl_inode_t *pl_inode) { pl_dom_list_t *dom = NULL; pl_entry_lock_t *lock = NULL; - char blocked[256] = { + char blocked[GF_TIMESTR_SIZE] = { 0, }; - char granted[256] = { + char granted[GF_TIMESTR_SIZE] = { 0, }; int count = 0; @@ -3579,10 +3687,10 @@ __dump_entrylks(pl_inode_t *pl_inode) list_for_each_entry(lock, &dom->entrylk_list, domain_list) { - gf_time_fmt(granted, sizeof(granted), lock->granted_time.tv_sec, + gf_time_fmt(granted, sizeof(granted), lock->granted_time, gf_timefmt_FT); gf_proc_dump_build_key(key, k, "entrylk[%d](ACTIVE)", count); - if (lock->blkd_time.tv_sec == 0) { + if (lock->blkd_time == 0) { snprintf(tmp, sizeof(tmp), ENTRY_GRNTD_FMT, lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" : "ENTRYLK_WRLCK", @@ -3590,7 +3698,7 @@ __dump_entrylks(pl_inode_t *pl_inode) lkowner_utoa(&lock->owner), lock->client, lock->connection_id, granted); } else { - gf_time_fmt(blocked, sizeof(blocked), lock->blkd_time.tv_sec, + gf_time_fmt(blocked, sizeof(blocked), lock->blkd_time, gf_timefmt_FT); snprintf(tmp, sizeof(tmp), ENTRY_BLKD_GRNTD_FMT, lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" @@ -3607,7 +3715,7 @@ __dump_entrylks(pl_inode_t *pl_inode) list_for_each_entry(lock, &dom->blocked_entrylks, blocked_locks) { - gf_time_fmt(blocked, sizeof(blocked), lock->blkd_time.tv_sec, + gf_time_fmt(blocked, sizeof(blocked), lock->blkd_time, gf_timefmt_FT); gf_proc_dump_build_key(key, k, "entrylk[%d](BLOCKED)", count); @@ -3659,9 +3767,8 @@ __dump_inodelks(pl_inode_t *pl_inode) SET_FLOCK_PID(&lock->user_flock, lock); pl_dump_lock(tmp, sizeof(tmp), &lock->user_flock, &lock->owner, - lock->client, lock->connection_id, - &lock->granted_time.tv_sec, &lock->blkd_time.tv_sec, - _gf_true); + lock->client, lock->connection_id, &lock->granted_time, + &lock->blkd_time, _gf_true); gf_proc_dump_write(key, "%s", tmp); count++; @@ -3673,8 +3780,8 @@ __dump_inodelks(pl_inode_t *pl_inode) count); SET_FLOCK_PID(&lock->user_flock, lock); pl_dump_lock(tmp, sizeof(tmp), &lock->user_flock, &lock->owner, - lock->client, lock->connection_id, 0, - &lock->blkd_time.tv_sec, _gf_false); + lock->client, lock->connection_id, 0, &lock->blkd_time, + _gf_false); gf_proc_dump_write(key, "%s", tmp); count++; @@ -3707,9 +3814,8 @@ __dump_posixlks(pl_inode_t *pl_inode) gf_proc_dump_build_key(key, "posixlk", "posixlk[%d](%s)", count, lock->blocked ? "BLOCKED" : "ACTIVE"); pl_dump_lock(tmp, sizeof(tmp), &lock->user_flock, &lock->owner, - lock->client, lock->client_uid, &lock->granted_time.tv_sec, - &lock->blkd_time.tv_sec, - (lock->blocked) ? _gf_false : _gf_true); + lock->client, lock->client_uid, &lock->granted_time, + &lock->blkd_time, (lock->blocked) ? _gf_false : _gf_true); gf_proc_dump_write(key, "%s", tmp); count++; @@ -3793,6 +3899,10 @@ unlock: gf_proc_dump_write("posixlk-count", "%d", count); __dump_posixlks(pl_inode); } + + gf_proc_dump_write("links", "%d", pl_inode->links); + gf_proc_dump_write("removes_pending", "%u", pl_inode->remove_running); + gf_proc_dump_write("removed", "%u", pl_inode->removed); } pthread_mutex_unlock(&pl_inode->mutex); @@ -4104,6 +4214,10 @@ fini(xlator_t *this) if (!priv) return; this->private = NULL; + if (this->local_pool) { + mem_pool_destroy(this->local_pool); + this->local_pool = NULL; + } GF_FREE(priv->brickname); GF_FREE(priv); @@ -4134,8 +4248,11 @@ pl_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, struct iatt *postoldparent, struct iatt *prenewparent, struct iatt *postnewparent, dict_t *xdata) { + pl_inode_remove_cbk(this, cookie, op_ret < 0 ? op_errno : 0); + PL_STACK_UNWIND(rename, xdata, frame, op_ret, op_errno, buf, preoldparent, postoldparent, prenewparent, postnewparent, xdata); + return 0; } @@ -4143,10 +4260,15 @@ int32_t pl_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc, dict_t *xdata) { - PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), oldloc, newloc); + int32_t error; + + error = PL_INODE_REMOVE(rename, frame, this, oldloc, newloc, pl_rename, + pl_rename_cbk, oldloc, newloc, xdata); + if (error > 0) { + STACK_UNWIND_STRICT(rename, frame, -1, error, NULL, NULL, NULL, NULL, + NULL, NULL); + } - STACK_WIND(frame, pl_rename_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata); return 0; } @@ -4270,8 +4392,11 @@ pl_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *preparent, struct iatt *postparent, dict_t *xdata) { + pl_inode_remove_cbk(this, cookie, op_ret < 0 ? op_errno : 0); + PL_STACK_UNWIND(unlink, xdata, frame, op_ret, op_errno, preparent, postparent, xdata); + return 0; } @@ -4279,9 +4404,14 @@ int32_t pl_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag, dict_t *xdata) { - PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL); - STACK_WIND(frame, pl_unlink_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata); + int32_t error; + + error = PL_INODE_REMOVE(unlink, frame, this, loc, NULL, pl_unlink, + pl_unlink_cbk, loc, xflag, xdata); + if (error > 0) { + STACK_UNWIND_STRICT(unlink, frame, -1, error, NULL, NULL, NULL); + } + return 0; } @@ -4348,8 +4478,11 @@ pl_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *preparent, struct iatt *postparent, dict_t *xdata) { + pl_inode_remove_cbk(this, cookie, op_ret < 0 ? op_errno : 0); + PL_STACK_UNWIND_FOR_CLIENT(rmdir, xdata, frame, op_ret, op_errno, preparent, postparent, xdata); + return 0; } @@ -4357,9 +4490,14 @@ int pl_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags, dict_t *xdata) { - PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL); - STACK_WIND(frame, pl_rmdir_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->rmdir, loc, xflags, xdata); + int32_t error; + + error = PL_INODE_REMOVE(rmdir, frame, this, loc, NULL, pl_rmdir, + pl_rmdir_cbk, loc, xflags, xdata); + if (error > 0) { + STACK_UNWIND_STRICT(rmdir, frame, -1, error, NULL, NULL, NULL); + } + return 0; } @@ -4389,6 +4527,19 @@ pl_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, inode_t *inode, struct iatt *buf, struct iatt *preparent, struct iatt *postparent, dict_t *xdata) { + pl_inode_t *pl_inode = (pl_inode_t *)cookie; + + if (op_ret >= 0) { + pthread_mutex_lock(&pl_inode->mutex); + + /* TODO: can happen pl_inode->links == 0 ? */ + if (pl_inode->links >= 0) { + pl_inode->links++; + } + + pthread_mutex_unlock(&pl_inode->mutex); + } + PL_STACK_UNWIND_FOR_CLIENT(link, xdata, frame, op_ret, op_errno, inode, buf, preparent, postparent, xdata); return 0; @@ -4398,9 +4549,18 @@ int pl_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc, dict_t *xdata) { + pl_inode_t *pl_inode; + + pl_inode = pl_inode_get(this, oldloc->inode, NULL); + if (pl_inode == NULL) { + STACK_UNWIND_STRICT(link, frame, -1, ENOMEM, NULL, NULL, NULL, NULL, + NULL); + return 0; + } + PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), oldloc, newloc); - STACK_WIND(frame, pl_link_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata); + STACK_WIND_COOKIE(frame, pl_link_cbk, pl_inode, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata); return 0; } diff --git a/xlators/features/marker/src/marker-quota.c b/xlators/features/marker/src/marker-quota.c index 4176ab6..3de2ea1 100644 --- a/xlators/features/marker/src/marker-quota.c +++ b/xlators/features/marker/src/marker-quota.c @@ -1713,21 +1713,17 @@ mq_initiate_quota_task(void *opaque) } out: - if (dirty) { - if (ret < 0) { - /* On failure clear dirty status flag. - * In the next lookup inspect_directory_xattr - * can set the status flag and fix the - * dirty directory. - * Do the same if the dir was dirty before - * txn - */ - ret = mq_inode_ctx_get(parent_loc.inode, this, &parent_ctx); - if (ret == 0) - mq_set_ctx_dirty_status(parent_ctx, _gf_false); - } else { - ret = mq_mark_dirty(this, &parent_loc, 0); - } + if ((dirty) && (ret < 0)) { + /* On failure clear dirty status flag. + * In the next lookup inspect_directory_xattr + * can set the status flag and fix the + * dirty directory. + * Do the same if the dir was dirty before + * txn + */ + ret = mq_inode_ctx_get(parent_loc.inode, this, &parent_ctx); + if (ret == 0) + mq_set_ctx_dirty_status(parent_ctx, _gf_false); } if (locked) diff --git a/xlators/features/metadisp/Makefile.am b/xlators/features/metadisp/Makefile.am new file mode 100644 index 0000000..a985f42 --- /dev/null +++ b/xlators/features/metadisp/Makefile.am @@ -0,0 +1,3 @@ +SUBDIRS = src + +CLEANFILES = diff --git a/xlators/features/metadisp/src/Makefile.am b/xlators/features/metadisp/src/Makefile.am new file mode 100644 index 0000000..1520ad8 --- /dev/null +++ b/xlators/features/metadisp/src/Makefile.am @@ -0,0 +1,38 @@ +noinst_PYTHON = gen-fops.py + +EXTRA_DIST = fops-tmpl.c + +xlator_LTLIBRARIES = metadisp.la +xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features + +nodist_metadisp_la_SOURCES = fops.c + +BUILT_SOURCES = fops.c + +metadisp_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS) + +metadisp_la_SOURCES = metadisp.c \ + metadisp-unlink.c \ + metadisp-stat.c \ + metadisp-lookup.c \ + metadisp-readdir.c \ + metadisp-create.c \ + metadisp-open.c \ + metadisp-fsync.c \ + metadisp-setattr.c \ + backend.c + +metadisp_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la + +noinst_HEADERS = metadisp.h metadisp-fops.h + +AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \ + -I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src + +AM_CFLAGS = -Wall $(GF_CFLAGS) + +fops.c: fops-tmpl.c $(top_srcdir)/libglusterfs/src/generator.py gen-fops.py + PYTHONPATH=$(top_srcdir)/libglusterfs/src \ + $(PYTHON) $(srcdir)/gen-fops.py $(srcdir)/fops-tmpl.c > $@ + +CLEANFILES = $(nodist_metadisp_la_SOURCES) diff --git a/xlators/features/metadisp/src/backend.c b/xlators/features/metadisp/src/backend.c new file mode 100644 index 0000000..ee2c25b --- /dev/null +++ b/xlators/features/metadisp/src/backend.c @@ -0,0 +1,45 @@ +#define GFID_STR_LEN 37 + +#include "metadisp.h" + +/* + * backend.c + * + * functions responsible for converting user-facing paths to backend-style + * "/$GFID" paths. + */ + +int32_t +build_backend_loc(uuid_t gfid, loc_t *src_loc, loc_t *dst_loc) +{ + static uuid_t root = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1}; + char gfid_buf[GFID_STR_LEN + 1] = { + 0, + }; + char *path = NULL; + + GF_VALIDATE_OR_GOTO("metadisp", src_loc, out); + GF_VALIDATE_OR_GOTO("metadisp", dst_loc, out); + + loc_copy(dst_loc, src_loc); + memcpy(dst_loc->pargfid, root, sizeof(root)); + GF_FREE((char *)dst_loc->path); // we are overwriting path so nuke + // whatever loc_copy gave us + + uuid_utoa_r(gfid, gfid_buf); + + path = GF_CALLOC(GFID_STR_LEN + 1, sizeof(char), + gf_common_mt_char); // freed via loc_wipe + + path[0] = '/'; + strncpy(path + 1, gfid_buf, GFID_STR_LEN); + path[GFID_STR_LEN] = 0; + dst_loc->path = path; + if (src_loc->name) + dst_loc->name = strrchr(dst_loc->path, '/'); + if (dst_loc->name) + dst_loc->name++; + return 0; +out: + return -1; +} diff --git a/xlators/features/metadisp/src/fops-tmpl.c b/xlators/features/metadisp/src/fops-tmpl.c new file mode 100644 index 0000000..4385b7d --- /dev/null +++ b/xlators/features/metadisp/src/fops-tmpl.c @@ -0,0 +1,10 @@ +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include <glusterfs/xlator.h> +#include "metadisp.h" +#include "metadisp-fops.h" + +#pragma generate diff --git a/xlators/features/metadisp/src/gen-fops.py b/xlators/features/metadisp/src/gen-fops.py new file mode 100644 index 0000000..8b5e120 --- /dev/null +++ b/xlators/features/metadisp/src/gen-fops.py @@ -0,0 +1,160 @@ +#!/usr/bin/python + +import sys +from generator import fop_subs, generate + +FN_METADATA_CHILD_GENERIC = """ +int32_t +metadisp_@NAME@ (call_frame_t *frame, xlator_t *this, + @LONG_ARGS@) +{ + METADISP_TRACE("@NAME@ metadata"); + STACK_WIND (frame, default_@NAME@_cbk, + METADATA_CHILD(this), METADATA_CHILD(this)->fops->@NAME@, + @SHORT_ARGS@); + return 0; +} +""" + +FN_GENERIC_TEMPLATE = """ +int32_t +metadisp_@NAME@ (call_frame_t *frame, xlator_t *this, + @LONG_ARGS@) +{ + METADISP_TRACE("@NAME@ generic"); + STACK_WIND (frame, default_@NAME@_cbk, + DATA_CHILD(this), DATA_CHILD(this)->fops->@NAME@, + @SHORT_ARGS@); + return 0; +} +""" + +FN_DATAFD_TEMPLATE = """ +int32_t +metadisp_@NAME@ (call_frame_t *frame, xlator_t *this, + @LONG_ARGS@) +{ + METADISP_TRACE("@NAME@ datafd"); + xlator_t *child = NULL; + child = DATA_CHILD(this); + STACK_WIND (frame, default_@NAME@_cbk, + child, child->fops->@NAME@, + @SHORT_ARGS@); + return 0; +} +""" + +FN_DATALOC_TEMPLATE = """ +int32_t +metadisp_@NAME@ (call_frame_t *frame, xlator_t *this, + @LONG_ARGS@) +{ + METADISP_TRACE("@NAME@ dataloc"); + loc_t backend_loc = { + 0, + }; + if (build_backend_loc(loc->gfid, loc, &backend_loc)) { + goto unwind; + } + xlator_t *child = NULL; + child = DATA_CHILD(this); + STACK_WIND (frame, default_@NAME@_cbk, + child, child->fops->@NAME@, + @SHORT_ARGS@); + return 0; + +unwind: + STACK_UNWIND_STRICT(lookup, frame, -1, EINVAL, NULL, NULL, NULL, NULL); + return 0; +} +""" + +FOPS_LINE_TEMPLATE = "\t.@NAME@ = metadisp_@NAME@," + +skipped = [ + "readdir", + "readdirp", + "lookup", + "fsync", + "stat", + "open", + "create", + "unlink", + "setattr", + # TODO: implement "inodelk", +] + + +def gen_fops(): + done = skipped + + # + # these are fops that wind to the DATA_CHILD + # + # NOTE: re-written in order from google doc: + # https://docs.google.com/document/d/1KEwVtSNvDhs4qb63gWx2ulCp5GJjge77NGJk4p_Ms4Q + for name in [ + "writev", + "readv", + "ftruncate", + "zerofill", + "discard", + "seek", + "fstat", + ]: + done = done + [name] + print(generate(FN_DATAFD_TEMPLATE, name, fop_subs)) + + for name in ["truncate"]: + done = done + [name] + print(generate(FN_DATALOC_TEMPLATE, name, fop_subs)) + + # these are fops that operate solely on dentries, folders, + # or extended attributes. Therefore, they must always + # wind to METADATA_CHILD and should never perform + # any path rewriting + # + # NOTE: re-written in order from google doc: + # https://docs.google.com/document/d/1KEwVtSNvDhs4qb63gWx2ulCp5GJjge77NGJk4p_Ms4Q + for name in [ + "mkdir", + "symlink", + "link", + "rename", + "mknod", + "opendir", + # "readdir, # special-cased + # "readdirp, # special-cased + "fsyncdir", + # "setattr", # special-cased + "readlink", + "fentrylk", + "access", + # TODO: these wind to both, + # data for backend-attributes and metadata for the rest + "xattrop", + "setxattr", + "getxattr", + "removexattr", + "fgetxattr", + "fsetxattr", + "fremovexattr", + ]: + + done = done + [name] + print(generate(FN_METADATA_CHILD_GENERIC, name, fop_subs)) + + print("struct xlator_fops fops = {") + for name in done: + print(generate(FOPS_LINE_TEMPLATE, name, fop_subs)) + + print("};") + + +for l in open(sys.argv[1], "r").readlines(): + if l.find("#pragma generate") != -1: + print("/* BEGIN GENERATED CODE - DO NOT MODIFY */") + gen_fops() + print("/* END GENERATED CODE */") + else: + print(l[:-1]) diff --git a/xlators/features/metadisp/src/metadisp-create.c b/xlators/features/metadisp/src/metadisp-create.c new file mode 100644 index 0000000..f8c9798 --- /dev/null +++ b/xlators/features/metadisp/src/metadisp-create.c @@ -0,0 +1,101 @@ +#include "metadisp.h" +#include <glusterfs/call-stub.h> + +/** + * Create, like stat, is a two-step process. We send a create + * to the METADATA_CHILD, then send another create to the DATA_CHILD. + * + * We do the metadata child first to ensure that the ACLs are enforced. + */ + +int32_t +metadisp_create_dentry_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, fd_t *fd, + inode_t *inode, struct iatt *buf, + struct iatt *preparent, struct iatt *postparent, + dict_t *xdata) +{ + STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, buf, + preparent, postparent, xdata); + return 0; +} + +int32_t +metadisp_create_resume(call_frame_t *frame, xlator_t *this, loc_t *loc, + int32_t flags, mode_t mode, mode_t umask, fd_t *fd, + dict_t *xdata) +{ + // create the backend data inode + STACK_WIND(frame, metadisp_create_dentry_cbk, DATA_CHILD(this), + DATA_CHILD(this)->fops->create, loc, flags, mode, umask, fd, + xdata); + return 0; +} + +int32_t +metadisp_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode, + struct iatt *buf, struct iatt *preparent, + struct iatt *postparent, dict_t *xdata) +{ + METADISP_TRACE("%d %d", op_ret, op_errno); + call_stub_t *stub = cookie; + if (op_ret != 0) { + STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, buf, + preparent, postparent, xdata); + return 0; + } + + if (stub == NULL) { + goto unwind; + } + + if (stub->poison) { + call_stub_destroy(stub); + return 0; + } + + call_resume(stub); + return 0; + +unwind: + STACK_UNWIND_STRICT(create, frame, -1, EINVAL, NULL, NULL, NULL, NULL, NULL, + NULL); + return 0; +} + +int32_t +metadisp_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, + mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata) +{ + METADISP_TRACE("."); + + loc_t backend_loc = { + 0, + }; + call_stub_t *stub = NULL; + uuid_t *gfid_req = NULL; + + RESOLVE_GFID_REQ(xdata, gfid_req, out); + + if (build_backend_loc(*gfid_req, loc, &backend_loc)) { + goto unwind; + } + + frame->local = loc; + + stub = fop_create_stub(frame, metadisp_create_resume, &backend_loc, flags, + mode, umask, fd, xdata); + + STACK_WIND_COOKIE(frame, metadisp_create_cbk, stub, METADATA_CHILD(this), + METADATA_CHILD(this)->fops->create, loc, flags, mode, + umask, fd, xdata); + return 0; + +unwind: + STACK_UNWIND_STRICT(create, frame, -1, EINVAL, NULL, NULL, NULL, NULL, NULL, + NULL); + return 0; +out: + return -1; +} diff --git a/xlators/features/metadisp/src/metadisp-fops.h b/xlators/features/metadisp/src/metadisp-fops.h new file mode 100644 index 0000000..56dd427 --- /dev/null +++ b/xlators/features/metadisp/src/metadisp-fops.h @@ -0,0 +1,51 @@ +#ifndef GF_METADISP_FOPS_H_ +#define GF_METADISP_FOPS_H_ + +#include <glusterfs/xlator.h> +#include <glusterfs/dict.h> +#include <glusterfs/glusterfs.h> + +#include <sys/types.h> + +/* fops in here are defined in their own file. Every other fop is just defined + * inline of fops.c */ + +int +metadisp_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, + off_t off, dict_t *xdata); + +int +metadisp_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, + off_t off, dict_t *dict); + +int +metadisp_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata); + +int +metadisp_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, + mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata); + +int +metadisp_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, + fd_t *fd, dict_t *xdata); + +int +metadisp_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata); + +int +metadisp_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, + loc_t *loc, int32_t cmd, struct gf_flock *lock, dict_t *xdata); + +int +metadisp_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags, + dict_t *xdata); + +int +metadisp_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag, + dict_t *xdata); + +int +metadisp_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, + struct iatt *stbuf, int32_t valid, dict_t *xdata); + +#endif diff --git a/xlators/features/metadisp/src/metadisp-fsync.c b/xlators/features/metadisp/src/metadisp-fsync.c new file mode 100644 index 0000000..2e46fa8 --- /dev/null +++ b/xlators/features/metadisp/src/metadisp-fsync.c @@ -0,0 +1,54 @@ + +#include "metadisp.h" +#include <glusterfs/call-stub.h> + +int32_t +metadisp_fsync_resume(call_frame_t *frame, xlator_t *this, fd_t *fd, + int32_t flags, dict_t *xdata) +{ + STACK_WIND(frame, default_fsync_cbk, DATA_CHILD(this), + DATA_CHILD(this)->fops->fsync, fd, flags, xdata); + return 0; +} + +int32_t +metadisp_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *prebuf, + struct iatt *postbuf, dict_t *xdata) +{ + call_stub_t *stub = NULL; + if (cookie) { + stub = cookie; + } + + if (op_ret != 0) { + goto unwind; + } + + if (stub->poison) { + call_stub_destroy(stub); + stub = NULL; + return 0; + } + + call_resume(stub); + return 0; + +unwind: + if (stub) { + call_stub_destroy(stub); + } + STACK_UNWIND_STRICT(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata); + return 0; +} + +int32_t +metadisp_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags, + dict_t *xdata) +{ + call_stub_t *stub = NULL; + stub = fop_fsync_stub(frame, metadisp_fsync_resume, fd, flags, xdata); + STACK_WIND_COOKIE(frame, metadisp_fsync_cbk, stub, METADATA_CHILD(this), + METADATA_CHILD(this)->fops->fsync, fd, flags, xdata); + return 0; +} diff --git a/xlators/features/metadisp/src/metadisp-lookup.c b/xlators/features/metadisp/src/metadisp-lookup.c new file mode 100644 index 0000000..27d90c9 --- /dev/null +++ b/xlators/features/metadisp/src/metadisp-lookup.c @@ -0,0 +1,90 @@ +#include "metadisp.h" +#include <glusterfs/call-stub.h> + +/** + * Lookup, like stat, is a two-step process for grabbing the metadata details + * as well as the data details. + */ + +int32_t +metadisp_backend_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, inode_t *inode, + struct iatt *buf, dict_t *xdata, + struct iatt *postparent) +{ + METADISP_TRACE("backend_lookup_cbk"); + if (op_errno == ENOENT) { + op_errno = ENODATA; + op_ret = -1; + } + STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xdata, + postparent); + return 0; +} + +int32_t +metadisp_backend_lookup_resume(call_frame_t *frame, xlator_t *this, loc_t *loc, + dict_t *xdata) +{ + METADISP_TRACE("backend_lookup_resume"); + loc_t backend_loc = { + 0, + }; + if (build_backend_loc(loc->gfid, loc, &backend_loc)) { + goto unwind; + } + + STACK_WIND(frame, metadisp_backend_lookup_cbk, DATA_CHILD(this), + DATA_CHILD(this)->fops->lookup, &backend_loc, xdata); + return 0; + +unwind: + STACK_UNWIND_STRICT(lookup, frame, -1, EINVAL, NULL, NULL, NULL, NULL); + return 0; +} + +int32_t +metadisp_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, inode_t *inode, + struct iatt *buf, dict_t *xdata, struct iatt *postparent) +{ + METADISP_TRACE("%d %d", op_ret, op_errno); + call_stub_t *stub = NULL; + stub = cookie; + + if (op_ret != 0) { + goto unwind; + } + + if (!IA_ISREG(buf->ia_type)) { + goto unwind; + } else if (!stub) { + op_errno = EINVAL; + goto unwind; + } + + METADISP_TRACE("resuming stub"); + + // memcpy(stub->args.loc.gfid, buf->ia_gfid, sizeof(uuid_t)); + call_resume(stub); + return 0; +unwind: + METADISP_TRACE("unwinding %d %d", op_ret, op_errno); + STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xdata, + postparent); + if (stub) { + call_stub_destroy(stub); + } + return 0; +} + +int32_t +metadisp_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata) +{ + METADISP_TRACE("lookup"); + call_stub_t *stub = NULL; + stub = fop_lookup_stub(frame, metadisp_backend_lookup_resume, loc, xdata); + STACK_WIND_COOKIE(frame, metadisp_lookup_cbk, stub, METADATA_CHILD(this), + METADATA_CHILD(this)->fops->lookup, loc, xdata); + return 0; +} diff --git a/xlators/features/metadisp/src/metadisp-open.c b/xlators/features/metadisp/src/metadisp-open.c new file mode 100644 index 0000000..64814af --- /dev/null +++ b/xlators/features/metadisp/src/metadisp-open.c @@ -0,0 +1,70 @@ +#include <glusterfs/call-stub.h> +#include "metadisp.h" + +int32_t +metadisp_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata) +{ + METADISP_TRACE("got open results %d %d", op_ret, op_errno); + + call_stub_t *stub = NULL; + if (cookie) { + stub = cookie; + } + + if (op_ret != 0) { + goto unwind; + } + + if (!stub) { + goto unwind; + } + + if (stub->poison) { + call_stub_destroy(stub); + stub = NULL; + return 0; + } + + call_resume(stub); + return 0; + +unwind: + if (stub) { + call_stub_destroy(stub); + } + STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, fd, xdata); + return 0; +} + +int32_t +metadisp_open_resume(call_frame_t *frame, xlator_t *this, loc_t *loc, + int32_t flags, fd_t *fd, dict_t *xdata) +{ + STACK_WIND_COOKIE(frame, metadisp_open_cbk, NULL, DATA_CHILD(this), + DATA_CHILD(this)->fops->open, loc, flags, fd, xdata); + return 0; +} + +int32_t +metadisp_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, + fd_t *fd, dict_t *xdata) +{ + call_stub_t *stub = NULL; + loc_t backend_loc = { + 0, + }; + + if (build_backend_loc(loc->gfid, loc, &backend_loc)) { + goto unwind; + } + + stub = fop_open_stub(frame, metadisp_open_resume, &backend_loc, flags, fd, + xdata); + STACK_WIND_COOKIE(frame, metadisp_open_cbk, stub, METADATA_CHILD(this), + METADATA_CHILD(this)->fops->open, loc, flags, fd, xdata); + return 0; +unwind: + STACK_UNWIND_STRICT(open, frame, -1, EINVAL, NULL, NULL); + return 0; +} diff --git a/xlators/features/metadisp/src/metadisp-readdir.c b/xlators/features/metadisp/src/metadisp-readdir.c new file mode 100644 index 0000000..5f840b1 --- /dev/null +++ b/xlators/features/metadisp/src/metadisp-readdir.c @@ -0,0 +1,65 @@ +#include "metadisp.h" + +/** + * With a change to the posix xlator, readdir and readdirp are shockingly + * simple. + * + * The issue with separating the backend data of the files + * with the metadata is that readdirs must now read from multiple sources + * to coalesce the directory entries. + * + * The way we do this is to tell the METADATA_CHILD that when it's + * running readdirp, each file entry should have a stat wound to + * 'stat-source-of-truth'. + * + * see metadisp_stat for how it handles winds _from_posix. + */ + +int32_t +metadisp_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, + off_t off, dict_t *xdata) +{ + METADISP_TRACE("."); + /* + * Always use readdirp, even if the original was readdir. Why? Because NFS. + * There are multiple translations between Gluster, UNIX, and NFS stat + * structures in that path. One of them uses the type etc. from the stat + * structure, which is only filled in by readdirp. If we use readdir, the + * entries do actually go all the way back to the client and are visible in + * getdents, but then the readdir throws them away because of the + * uninitialized type. + */ + GF_UNUSED int32_t ret; + if (!xdata) { + xdata = dict_new(); + } + + // ret = dict_set_int32 (xdata, "list-xattr", 1); + + // I'm my own source of truth! + ret = dict_set_static_ptr(xdata, "stat-source-of-truth", (void *)this); + + STACK_WIND(frame, default_readdirp_cbk, METADATA_CHILD(this), + METADATA_CHILD(this)->fops->readdirp, fd, size, off, xdata); + + return 0; +} + +int32_t +metadisp_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, + off_t off, dict_t *xdata) +{ + METADISP_TRACE("."); + if (!xdata) { + xdata = dict_new(); + } + GF_UNUSED int32_t ret; + // ret = dict_set_int32 (xdata, "list-xattr", 1); + + // I'm my own source of truth! + ret = dict_set_static_ptr(xdata, "stat-source-of-truth", (void *)this); + + STACK_WIND(frame, default_readdirp_cbk, METADATA_CHILD(this), + METADATA_CHILD(this)->fops->readdirp, fd, size, off, xdata); + return 0; +} diff --git a/xlators/features/metadisp/src/metadisp-setattr.c b/xlators/features/metadisp/src/metadisp-setattr.c new file mode 100644 index 0000000..6991cf6 --- /dev/null +++ b/xlators/features/metadisp/src/metadisp-setattr.c @@ -0,0 +1,90 @@ +#include "metadisp.h" +#include <glusterfs/call-stub.h> + +int32_t +metadisp_backend_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, + struct iatt *statpre, struct iatt *statpost, + dict_t *xdata) + +{ + METADISP_TRACE("backend_setattr_cbk"); + if (op_errno == ENOENT) { + op_errno = ENODATA; + op_ret = -1; + } + STACK_UNWIND_STRICT(setattr, frame, op_ret, op_errno, statpre, statpost, + xdata); + return 0; +} + +int32_t +metadisp_backend_setattr_resume(call_frame_t *frame, xlator_t *this, loc_t *loc, + struct iatt *stbuf, int32_t valid, + dict_t *xdata) + +{ + METADISP_TRACE("backend_setattr_resume"); + loc_t backend_loc = { + 0, + }; + if (build_backend_loc(loc->gfid, loc, &backend_loc)) { + goto unwind; + } + + STACK_WIND(frame, metadisp_backend_setattr_cbk, DATA_CHILD(this), + DATA_CHILD(this)->fops->setattr, &backend_loc, stbuf, valid, + xdata); + return 0; + +unwind: + STACK_UNWIND_STRICT(setattr, frame, -1, EINVAL, NULL, NULL, NULL); + return 0; +} + +int32_t +metadisp_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *statpre, + struct iatt *statpost, dict_t *xdata) +{ + METADISP_TRACE("%d %d", op_ret, op_errno); + call_stub_t *stub = NULL; + stub = cookie; + + if (op_ret != 0) { + goto unwind; + } + + if (!IA_ISREG(statpost->ia_type)) { + goto unwind; + } else if (!stub) { + op_errno = EINVAL; + goto unwind; + } + + METADISP_TRACE("resuming stub"); + call_resume(stub); + return 0; +unwind: + METADISP_TRACE("unwinding %d %d", op_ret, op_errno); + STACK_UNWIND_STRICT(setattr, frame, op_ret, op_errno, statpre, statpost, + xdata); + if (stub) { + call_stub_destroy(stub); + } + return 0; +} + +int32_t +metadisp_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, + struct iatt *stbuf, int32_t valid, dict_t *xdata) +{ + METADISP_TRACE("setattr"); + call_stub_t *stub = NULL; + stub = fop_setattr_stub(frame, metadisp_backend_setattr_resume, loc, stbuf, + valid, xdata); + STACK_WIND_COOKIE(frame, metadisp_setattr_cbk, stub, METADATA_CHILD(this), + METADATA_CHILD(this)->fops->setattr, loc, stbuf, valid, + xdata); + return 0; +} diff --git a/xlators/features/metadisp/src/metadisp-stat.c b/xlators/features/metadisp/src/metadisp-stat.c new file mode 100644 index 0000000..b06d0db --- /dev/null +++ b/xlators/features/metadisp/src/metadisp-stat.c @@ -0,0 +1,124 @@ +#include "metadisp.h" +#include <glusterfs/call-stub.h> + +/** + * The stat flow in METADISP is complicated because we must + * do ensure a few things: + * 1. stat, on the path within the metadata layer, + * MUST get the backend FD of the data layer. + * --- we wind to the metadata layer, then the data layer. + * + * 2. the metadata layer MUST be able to ask the data + * layer for stat information. + * --- this is 'syncop-internal-from-posix' + * + * 3. when the metadata exists BUT the data is missing, + * we MUST mark the backend file as bad and heal it. + */ + +int32_t +metadisp_stat_backend_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *buf, + dict_t *xdata) +{ + METADISP_TRACE("got backend stat results %d %d", op_ret, op_errno); + if (op_errno == ENOENT) { + STACK_UNWIND_STRICT(open, frame, -1, ENODATA, NULL, NULL); + return 0; + } + STACK_UNWIND_STRICT(stat, frame, op_ret, op_errno, buf, xdata); + return 0; +} + +int32_t +metadisp_stat_resume(call_frame_t *frame, xlator_t *this, loc_t *loc, + dict_t *xdata) +{ + METADISP_TRACE("winding stat to path %s", loc->path); + if (gf_uuid_is_null(loc->gfid)) { + METADISP_TRACE("bad object, sending EUCLEAN"); + STACK_UNWIND_STRICT(open, frame, -1, EUCLEAN, NULL, NULL); + return 0; + } + + STACK_WIND(frame, metadisp_stat_backend_cbk, SECOND_CHILD(this), + SECOND_CHILD(this)->fops->stat, loc, xdata); + return 0; +} + +int32_t +metadisp_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *buf, + dict_t *xdata) +{ + call_stub_t *stub = NULL; + + METADISP_TRACE("got stat results %d %d", op_ret, op_errno); + + if (cookie) { + stub = cookie; + } + + if (op_ret != 0) { + goto unwind; + } + + // only use the stub for the files + if (!IA_ISREG(buf->ia_type)) { + goto unwind; + } + + if (stub->poison) { + call_stub_destroy(stub); + stub = NULL; + return 0; + } + + call_resume(stub); + return 0; + +unwind: + if (stub) { + call_stub_destroy(stub); + } + STACK_UNWIND_STRICT(stat, frame, op_ret, op_errno, buf, xdata); + return 0; +} + +int32_t +metadisp_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata) +{ + call_stub_t *stub = NULL; + int32_t ret = 0; + loc_t backend_loc = { + 0, + }; + METADISP_FILTER_ROOT(stat, loc, xdata); + + if (build_backend_loc(loc->gfid, loc, &backend_loc)) { + goto unwind; + } + + if (dict_get_int32(xdata, "syncop-internal-from-posix", &ret) == 0) { + // if we've just been sent a stat from posix, then we know + // that we must send down a stat for a file to the second child. + // + // that means we can skip the stat for the first child and just + // send to the data disk. + METADISP_TRACE("got syncop-internal-from-posix"); + STACK_WIND(frame, default_stat_cbk, DATA_CHILD(this), + DATA_CHILD(this)->fops->stat, &backend_loc, xdata); + return 0; + } + + // we do not know if the request is for a file, folder, etc. wind + // to first child to find out. + stub = fop_stat_stub(frame, metadisp_stat_resume, &backend_loc, xdata); + METADISP_TRACE("winding stat to first child %s", loc->path); + STACK_WIND_COOKIE(frame, metadisp_stat_cbk, stub, METADATA_CHILD(this), + METADATA_CHILD(this)->fops->stat, loc, xdata); + return 0; +unwind: + STACK_UNWIND_STRICT(stat, frame, -1, EINVAL, NULL, NULL); + return 0; +} diff --git a/xlators/features/metadisp/src/metadisp-unlink.c b/xlators/features/metadisp/src/metadisp-unlink.c new file mode 100644 index 0000000..1f6a8eb --- /dev/null +++ b/xlators/features/metadisp/src/metadisp-unlink.c @@ -0,0 +1,160 @@ + +#include "metadisp.h" +#include <glusterfs/call-stub.h> + +/** + * The unlink flow in metadisp is complicated because we must + * do ensure that UNLINK causes both the metadata objects + * to get removed and the data objects to get removed. + */ + +int32_t +metadisp_unlink_resume(call_frame_t *frame, xlator_t *this, loc_t *loc, + int xflag, dict_t *xdata) +{ + METADISP_TRACE("winding backend unlink to path %s", loc->path); + STACK_WIND(frame, default_unlink_cbk, DATA_CHILD(this), + DATA_CHILD(this)->fops->unlink, loc, xflag, xdata); + return 0; +} + +int32_t +metadisp_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *preparent, + struct iatt *postparent, dict_t *xdata) +{ + METADISP_TRACE(". %d %d", op_ret, op_errno); + + int ret = 0; + call_stub_t *stub = NULL; + int nlink = 0; + + if (cookie) { + stub = cookie; + } + + if (op_ret != 0) { + goto unwind; + } + + if (stub->poison) { + call_stub_destroy(stub); + stub = NULL; + return 0; + } + + ret = dict_get_uint32(xdata, GF_RESPONSE_LINK_COUNT_XDATA, &nlink); + if (ret != 0) { + op_errno = EINVAL; + op_ret = -1; + goto unwind; + } + METADISP_TRACE("frontend hardlink count %d %d", ret, nlink); + if (nlink > 1) { + goto unwind; + } + + call_resume(stub); + return 0; + +unwind: + if (stub) { + call_stub_destroy(stub); + } + STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, preparent, postparent, + xdata); + return 0; +} + +int32_t +metadisp_unlink_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, inode_t *inode, + struct iatt *buf, dict_t *xdata, + struct iatt *postparent) +{ + call_stub_t *stub = NULL; + + if (cookie) { + stub = cookie; + } + + if (op_ret != 0) { + goto unwind; + } + + // fail fast on empty gfid so we don't loop forever + if (gf_uuid_is_null(buf->ia_gfid)) { + op_ret = -1; + op_errno = ENODATA; + goto unwind; + } + + // fill gfid since the stub is incomplete + memcpy(stub->args.loc.gfid, buf->ia_gfid, sizeof(uuid_t)); + memcpy(stub->args.loc.pargfid, postparent->ia_gfid, sizeof(uuid_t)); + + if (stub->poison) { + call_stub_destroy(stub); + stub = NULL; + return 0; + } + + call_resume(stub); + return 0; + +unwind: + if (stub) { + call_stub_destroy(stub); + } + STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, NULL, NULL, NULL); + return 0; +} + +int32_t +metadisp_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag, + dict_t *xdata) +{ + call_stub_t *stub = NULL; + loc_t backend_loc = { + 0, + }; + + if (gf_uuid_is_null(loc->gfid)) { + METADISP_TRACE("winding lookup for unlink to path %s", loc->path); + + // loop back to ourselves after a lookup + stub = fop_unlink_stub(frame, metadisp_unlink, loc, xflag, xdata); + STACK_WIND_COOKIE(frame, metadisp_unlink_lookup_cbk, stub, + METADATA_CHILD(this), + METADATA_CHILD(this)->fops->lookup, loc, xdata); + return 0; + } + + if (build_backend_loc(loc->gfid, loc, &backend_loc)) { + goto unwind; + } + + // + // ensure we get the link count on the unlink response, so we can + // account for hardlinks before winding to the backend. + // NOTE: + // multiple xlators use GF_REQUEST_LINK_COUNT_XDATA. confirmation + // is needed to ensure that multiple requests will work in the same + // xlator stack. + // + if (!xdata) { + xdata = dict_new(); + } + dict_set_int32(xdata, GF_REQUEST_LINK_COUNT_XDATA, 1); + + METADISP_TRACE("winding frontend unlink to path %s", loc->path); + stub = fop_unlink_stub(frame, metadisp_unlink_resume, &backend_loc, xflag, + xdata); + + STACK_WIND_COOKIE(frame, metadisp_unlink_cbk, stub, METADATA_CHILD(this), + METADATA_CHILD(this)->fops->unlink, loc, xflag, xdata); + return 0; +unwind: + STACK_UNWIND_STRICT(unlink, frame, -1, EINVAL, NULL, NULL, NULL); + return 0; +} diff --git a/xlators/features/metadisp/src/metadisp.c b/xlators/features/metadisp/src/metadisp.c new file mode 100644 index 0000000..3c8f150 --- /dev/null +++ b/xlators/features/metadisp/src/metadisp.c @@ -0,0 +1,46 @@ +#include <glusterfs/call-stub.h> + +#include "metadisp.h" +#include "metadisp-fops.h" + +int32_t +init(xlator_t *this) +{ + if (!this->children) { + gf_log(this->name, GF_LOG_ERROR, + "not configured with children. exiting"); + return -1; + } + + if (!this->parents) { + gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile "); + } + + return 0; +} + +void +fini(xlator_t *this) +{ + return; +} + +/* defined in fops.c */ +struct xlator_fops fops; + +struct xlator_cbks cbks = {}; + +struct volume_options options[] = { + {.key = {NULL}}, +}; + +xlator_api_t xlator_api = { + .init = init, + .fini = fini, + .fops = &fops, + .cbks = &cbks, + .options = options, + .op_version = {1}, + .identifier = "metadisp", + .category = GF_EXPERIMENTAL, +}; diff --git a/xlators/features/metadisp/src/metadisp.h b/xlators/features/metadisp/src/metadisp.h new file mode 100644 index 0000000..c8fd7a1 --- /dev/null +++ b/xlators/features/metadisp/src/metadisp.h @@ -0,0 +1,45 @@ +/* + Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ +#ifndef GF_METADISP_H_ +#define GF_METADISP_H_ + +#include <glusterfs/glusterfs.h> +#include <glusterfs/logging.h> +#include <glusterfs/dict.h> +#include <glusterfs/xlator.h> +#include <glusterfs/defaults.h> + +#define METADATA_CHILD(_this) FIRST_CHILD(_this) +#define DATA_CHILD(_this) SECOND_CHILD(_this) + +int32_t +build_backend_loc(uuid_t gfid, loc_t *src_loc, loc_t *dst_loc); + +#define METADISP_TRACE(_args...) gf_log("metadisp", GF_LOG_INFO, _args) + +#define METADISP_FILTER_ROOT(_op, _args...) \ + if (strcmp(loc->path, "/") == 0) { \ + STACK_WIND(frame, default_##_op##_cbk, METADATA_CHILD(this), \ + METADATA_CHILD(this)->fops->_op, _args); \ + return 0; \ + } + +#define METADISP_FILTER_ROOT_BY_GFID(_op, _gfid, _args...) \ + if (__is_root_gfid(_gfid)) { \ + STACK_WIND(frame, default_##_op##_cbk, METADATA_CHILD(this), \ + METADATA_CHILD(this)->fops->_op, _args); \ + return 0; \ + } + +#define RESOLVE_GFID_REQ(_dict, _dest, _lbl) \ + VALIDATE_OR_GOTO(dict_get_ptr(_dict, "gfid-req", (void **)&_dest) == 0, \ + _lbl) + +#endif /* __TEMPLATE_H__ */ diff --git a/xlators/features/quota/src/quota-enforcer-client.c b/xlators/features/quota/src/quota-enforcer-client.c index 1a4c2e3..480d64a 100644 --- a/xlators/features/quota/src/quota-enforcer-client.c +++ b/xlators/features/quota/src/quota-enforcer-client.c @@ -32,12 +32,6 @@ #include <malloc.h> #endif -#ifdef HAVE_MALLOC_STATS -#ifdef DEBUG -#include <mcheck.h> -#endif -#endif - #include "quota.h" #include "quota-messages.h" @@ -362,16 +356,28 @@ quota_enforcer_notify(struct rpc_clnt *rpc, void *mydata, { xlator_t *this = NULL; int ret = 0; + quota_priv_t *priv = NULL; this = mydata; - + priv = this->private; switch (event) { case RPC_CLNT_CONNECT: { + pthread_mutex_lock(&priv->conn_mutex); + { + priv->conn_status = _gf_true; + } + pthread_mutex_unlock(&priv->conn_mutex); gf_msg_trace(this->name, 0, "got RPC_CLNT_CONNECT"); break; } case RPC_CLNT_DISCONNECT: { + pthread_mutex_lock(&priv->conn_mutex); + { + priv->conn_status = _gf_false; + pthread_cond_signal(&priv->conn_cond); + } + pthread_mutex_unlock(&priv->conn_mutex); gf_msg_trace(this->name, 0, "got RPC_CLNT_DISCONNECT"); break; } diff --git a/xlators/features/quota/src/quota.c b/xlators/features/quota/src/quota.c index 01b01b5..18df9ae 100644 --- a/xlators/features/quota/src/quota.c +++ b/xlators/features/quota/src/quota.c @@ -586,9 +586,6 @@ quota_validate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, quota_meta_t size = { 0, }; - struct timeval tv = { - 0, - }; local = frame->local; @@ -626,13 +623,12 @@ quota_validate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, * loop of validation and checking * limit when timeout is zero. */ - gettimeofday(&tv, NULL); LOCK(&ctx->lock); { ctx->size = size.size; + ctx->validate_time = gf_time(); ctx->file_count = size.file_count; ctx->dir_count = size.dir_count; - memcpy(&ctx->tv, &tv, sizeof(struct timeval)); } UNLOCK(&ctx->lock); @@ -644,27 +640,10 @@ unwind: return 0; } -static uint64_t -quota_time_elapsed(struct timeval *now, struct timeval *then) +static inline gf_boolean_t +quota_timeout(time_t t, uint32_t timeout) { - return (now->tv_sec - then->tv_sec); -} - -int32_t -quota_timeout(struct timeval *tv, int32_t timeout) -{ - struct timeval now = { - 0, - }; - int32_t timed_out = 0; - - gettimeofday(&now, NULL); - - if (quota_time_elapsed(&now, tv) >= timeout) { - timed_out = 1; - } - - return timed_out; + return (gf_time() - t) >= timeout; } /* Return: 1 if new entry added @@ -1128,7 +1107,7 @@ quota_check_object_limit(call_frame_t *frame, quota_inode_ctx_t *ctx, timeout = priv->hard_timeout; } - if (!just_validated && quota_timeout(&ctx->tv, timeout)) { + if (!just_validated && quota_timeout(ctx->validate_time, timeout)) { need_validate = 1; } else if ((object_aggr_count) > ctx->object_hard_lim) { hard_limit_exceeded = 1; @@ -1195,7 +1174,7 @@ quota_check_size_limit(call_frame_t *frame, quota_inode_ctx_t *ctx, timeout = priv->hard_timeout; } - if (!just_validated && quota_timeout(&ctx->tv, timeout)) { + if (!just_validated && quota_timeout(ctx->validate_time, timeout)) { need_validate = 1; } else if (wouldbe_size >= ctx->hard_lim) { hard_limit_exceeded = 1; @@ -4314,9 +4293,6 @@ quota_statfs_validate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, quota_meta_t size = { 0, }; - struct timeval tv = { - 0, - }; local = frame->local; @@ -4348,13 +4324,12 @@ quota_statfs_validate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, op_errno = EINVAL; } - gettimeofday(&tv, NULL); LOCK(&ctx->lock); { ctx->size = size.size; + ctx->validate_time = gf_time(); ctx->file_count = size.file_count; ctx->dir_count = size.dir_count; - memcpy(&ctx->tv, &tv, sizeof(struct timeval)); } UNLOCK(&ctx->lock); @@ -4873,7 +4848,7 @@ off: void quota_log_helper(char **usage_str, int64_t cur_size, inode_t *inode, - char **path, struct timeval *cur_time) + char **path, time_t *cur_time) { xlator_t *this = THIS; @@ -4892,7 +4867,7 @@ quota_log_helper(char **usage_str, int64_t cur_size, inode_t *inode, if (!(*path)) *path = uuid_utoa(inode->gfid); - gettimeofday(cur_time, NULL); + *cur_time = gf_time(); } /* Logs if @@ -4903,9 +4878,7 @@ void quota_log_usage(xlator_t *this, quota_inode_ctx_t *ctx, inode_t *inode, int64_t delta) { - struct timeval cur_time = { - 0, - }; + time_t cur_time = 0; char *usage_str = NULL; char *path = NULL; int64_t cur_size = 0; @@ -4931,12 +4904,12 @@ quota_log_usage(xlator_t *this, quota_inode_ctx_t *ctx, inode_t *inode, "path=%s", usage_str, priv->volume_uuid, path); - ctx->prev_log = cur_time; + ctx->prev_log_time = cur_time; } /* Usage is above soft limit */ else if (cur_size > ctx->soft_lim && - quota_timeout(&ctx->prev_log, priv->log_timeout)) { + quota_timeout(ctx->prev_log_time, priv->log_timeout)) { quota_log_helper(&usage_str, cur_size, inode, &path, &cur_time); gf_msg(this->name, GF_LOG_ALERT, 0, Q_MSG_CROSSED_SOFT_LIMIT, @@ -4947,7 +4920,7 @@ quota_log_usage(xlator_t *this, quota_inode_ctx_t *ctx, inode_t *inode, "path=%s", usage_str, priv->volume_uuid, path); - ctx->prev_log = cur_time; + ctx->prev_log_time = cur_time; } if (path) @@ -5008,6 +4981,43 @@ quota_forget(xlator_t *this, inode_t *inode) return 0; } +int +notify(xlator_t *this, int event, void *data, ...) +{ + quota_priv_t *priv = NULL; + int ret = 0; + rpc_clnt_t *rpc = NULL; + gf_boolean_t conn_status = _gf_true; + xlator_t *victim = data; + + priv = this->private; + if (!priv || !priv->is_quota_on) + goto out; + + if (event == GF_EVENT_PARENT_DOWN) { + rpc = priv->rpc_clnt; + if (rpc) { + rpc_clnt_disable(rpc); + pthread_mutex_lock(&priv->conn_mutex); + { + conn_status = priv->conn_status; + while (conn_status) { + (void)pthread_cond_wait(&priv->conn_cond, + &priv->conn_mutex); + conn_status = priv->conn_status; + } + } + pthread_mutex_unlock(&priv->conn_mutex); + gf_log(this->name, GF_LOG_INFO, + "Notify GF_EVENT_PARENT_DOWN for brick %s", victim->name); + } + } + +out: + ret = default_notify(this, event, data); + return ret; +} + int32_t init(xlator_t *this) { @@ -5050,6 +5060,10 @@ init(xlator_t *this) goto err; } + pthread_mutex_init(&priv->conn_mutex, NULL); + pthread_cond_init(&priv->conn_cond, NULL); + priv->conn_status = _gf_false; + if (priv->is_quota_on) { rpc = quota_enforcer_init(this, this->options); if (rpc == NULL) { @@ -5143,9 +5157,9 @@ quota_priv_dump(xlator_t *this) if (ret) goto out; else { - gf_proc_dump_write("soft-timeout", "%d", priv->soft_timeout); - gf_proc_dump_write("hard-timeout", "%d", priv->hard_timeout); - gf_proc_dump_write("alert-time", "%d", priv->log_timeout); + gf_proc_dump_write("soft-timeout", "%u", priv->soft_timeout); + gf_proc_dump_write("hard-timeout", "%u", priv->hard_timeout); + gf_proc_dump_write("alert-time", "%u", priv->log_timeout); gf_proc_dump_write("quota-on", "%d", priv->is_quota_on); gf_proc_dump_write("statfs", "%d", priv->consider_statfs); gf_proc_dump_write("volume-uuid", "%s", priv->volume_uuid); @@ -5163,20 +5177,22 @@ fini(xlator_t *this) { quota_priv_t *priv = NULL; rpc_clnt_t *rpc = NULL; - int i = 0, cnt = 0; priv = this->private; if (!priv) return; rpc = priv->rpc_clnt; priv->rpc_clnt = NULL; - this->private = NULL; if (rpc) { - cnt = GF_ATOMIC_GET(rpc->refcount); - for (i = 0; i < cnt; i++) - rpc_clnt_unref(rpc); + rpc_clnt_connection_cleanup(&rpc->conn); + rpc_clnt_unref(rpc); } + + this->private = NULL; LOCK_DESTROY(&priv->lock); + pthread_mutex_destroy(&priv->conn_mutex); + pthread_cond_destroy(&priv->conn_cond); + GF_FREE(priv); if (this->local_pool) { mem_pool_destroy(this->local_pool); @@ -5308,6 +5324,7 @@ struct volume_options options[] = { xlator_api_t xlator_api = { .init = init, .fini = fini, + .notify = notify, .reconfigure = reconfigure, .mem_acct_init = mem_acct_init, .op_version = {1}, /* Present from the initial version */ diff --git a/xlators/features/quota/src/quota.h b/xlators/features/quota/src/quota.h index 00012c2..0395d78 100644 --- a/xlators/features/quota/src/quota.h +++ b/xlators/features/quota/src/quota.h @@ -153,8 +153,8 @@ struct quota_inode_ctx { int64_t object_soft_lim; struct iatt buf; struct list_head parents; - struct timeval tv; - struct timeval prev_log; + time_t validate_time; + time_t prev_log_time; gf_boolean_t ancestry_built; gf_lock_t lock; }; @@ -199,6 +199,7 @@ struct quota_local { typedef struct quota_local quota_local_t; struct quota_priv { + /* FIXME: consider time_t for timeouts. */ uint32_t soft_timeout; uint32_t hard_timeout; uint32_t log_timeout; @@ -214,6 +215,9 @@ struct quota_priv { char *volume_uuid; uint64_t validation_count; int32_t quotad_conn_status; + pthread_mutex_t conn_mutex; + pthread_cond_t conn_cond; + gf_boolean_t conn_status; }; typedef struct quota_priv quota_priv_t; diff --git a/xlators/features/read-only/src/worm-helper.c b/xlators/features/read-only/src/worm-helper.c index 25fbd4a..df45f2a 100644 --- a/xlators/features/read-only/src/worm-helper.c +++ b/xlators/features/read-only/src/worm-helper.c @@ -41,7 +41,7 @@ worm_init_state(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr) GF_VALIDATE_OR_GOTO("worm", this, out); GF_VALIDATE_OR_GOTO(this->name, file_ptr, out); - start_time = time(NULL); + start_time = gf_time(); dict = dict_new(); if (!dict) { gf_log(this->name, GF_LOG_ERROR, "Error creating the dict"); @@ -94,7 +94,7 @@ worm_set_state(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr, if (ret) goto out; stbuf->ia_mtime = stpre.ia_mtime; - stbuf->ia_atime = time(NULL) + retention_state->ret_period; + stbuf->ia_atime = gf_time() + retention_state->ret_period; if (fop_with_fd) ret = syncop_fsetattr(this, (fd_t *)file_ptr, stbuf, GF_SET_ATTR_ATIME, @@ -286,6 +286,7 @@ gf_worm_state_transition(xlator_t *this, gf_boolean_t fop_with_fd, { int op_errno = EROFS; int ret = -1; + time_t now = 0; uint64_t com_period = 0; uint64_t start_time = 0; dict_t *dict = NULL; @@ -337,8 +338,10 @@ gf_worm_state_transition(xlator_t *this, gf_boolean_t fop_with_fd, goto out; } - if (ret == -1 && (time(NULL) - start_time) >= com_period) { - if ((time(NULL) - stbuf.ia_mtime) >= com_period) { + now = gf_time(); + + if (ret == -1 && (now - start_time) >= com_period) { + if ((now - stbuf.ia_mtime) >= com_period) { ret = worm_set_state(this, fop_with_fd, file_ptr, &reten_state, &stbuf); if (ret) { @@ -352,10 +355,10 @@ gf_worm_state_transition(xlator_t *this, gf_boolean_t fop_with_fd, op_errno = 0; goto out; } - } else if (ret == -1 && (time(NULL) - start_time) < com_period) { + } else if (ret == -1 && (now - start_time) < com_period) { op_errno = 0; goto out; - } else if (reten_state.retain && ((time(NULL) >= stbuf.ia_atime))) { + } else if (reten_state.retain && ((now >= stbuf.ia_atime))) { gf_worm_state_lookup(this, fop_with_fd, file_ptr, &reten_state, &stbuf); } if (reten_state.worm && !reten_state.retain && priv->worm_files_deletable && diff --git a/xlators/features/read-only/src/worm.c b/xlators/features/read-only/src/worm.c index 46078c1..1cc5526 100644 --- a/xlators/features/read-only/src/worm.c +++ b/xlators/features/read-only/src/worm.c @@ -440,8 +440,6 @@ worm_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, { int ret = 0; read_only_priv_t *priv = NULL; - dict_t *dict = NULL; - // In case of an error exit because fd can be NULL and this would // cause an segfault when performing fsetxattr . We explicitly // unwind to avoid future problems @@ -452,24 +450,12 @@ worm_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, priv = this->private; GF_ASSERT(priv); if (priv->worm_file) { - dict = dict_new(); - if (!dict) { - gf_log(this->name, GF_LOG_ERROR, - "Error creating the " - "dict"); - goto out; - } - ret = dict_set_int8(dict, "trusted.worm_file", 1); + ret = fd_ctx_set(fd, this, 1); if (ret) { gf_log(this->name, GF_LOG_ERROR, - "Error in setting " - "the dict"); - goto out; - } - ret = syncop_fsetxattr(this, fd, dict, 0, NULL, NULL); - if (ret) { - gf_log(this->name, GF_LOG_ERROR, "Error setting xattr"); - goto out; + "Failed to set the fd ctx " + "for gfid:%s . Worm feature may not work for the gfid", + uuid_utoa(inode->gfid)); } ret = worm_init_state(this, _gf_true, fd); if (ret) { @@ -480,8 +466,6 @@ worm_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, out: STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, buf, preparent, postparent, xdata); - if (dict) - dict_unref(dict); return ret; } @@ -617,7 +601,62 @@ struct xlator_fops fops = { .lk = ro_lk, }; -struct xlator_cbks cbks; +int32_t +worm_release(xlator_t *this, fd_t *fd) +{ + dict_t *dict = NULL; + int ret = -1; + dict = dict_new(); + uint64_t value = 0; + loc_t loc = { + 0, + }; + read_only_priv_t *priv = NULL; + priv = this->private; + + if (priv->worm_file) { + if (!dict) { + gf_log(this->name, GF_LOG_ERROR, "Error creating the dict"); + goto out; + } + + ret = fd_ctx_get(fd, this, &value); + if (ret) { + gf_log(this->name, GF_LOG_DEBUG, "Failed to get the fd ctx"); + } + if (!value) { + goto out; + } + + ret = dict_set_int8(dict, "trusted.worm_file", 1); + if (ret) { + gf_log(this->name, GF_LOG_ERROR, + "Error in setting " + "the dict"); + goto out; + } + + loc.inode = inode_ref(fd->inode); + gf_uuid_copy(loc.gfid, fd->inode->gfid); + ret = syncop_setxattr(this, &loc, dict, 0, NULL, NULL); + if (ret) { + gf_log(this->name, GF_LOG_ERROR, "Error setting xattr"); + goto out; + } + + gf_worm_state_transition(this, _gf_false, &loc, GF_FOP_WRITE); + } + +out: + loc_wipe(&loc); + if (dict) + dict_unref(dict); + return 0; +} + +struct xlator_cbks cbks = { + .release = worm_release, +}; struct volume_options options[] = { {.key = {"worm"}, diff --git a/xlators/features/shard/src/shard.c b/xlators/features/shard/src/shard.c index c59e244..e5f9306 100644 --- a/xlators/features/shard/src/shard.c +++ b/xlators/features/shard/src/shard.c @@ -513,6 +513,9 @@ shard_local_wipe(shard_local_t *local) loc_wipe(&local->int_entrylk.loc); loc_wipe(&local->newloc); + if (local->name) + GF_FREE(local->name); + if (local->int_entrylk.basename) GF_FREE(local->int_entrylk.basename); if (local->fd) @@ -1001,6 +1004,10 @@ shard_initiate_evicted_inode_fsync(xlator_t *this, inode_t *inode) } int +shard_common_inode_write_post_lookup_shards_handler(call_frame_t *frame, + xlator_t *this); + +int shard_common_resolve_shards(call_frame_t *frame, xlator_t *this, shard_post_resolve_fop_handler_t post_res_handler) { @@ -1017,21 +1024,47 @@ shard_common_resolve_shards(call_frame_t *frame, xlator_t *this, inode_t *fsync_inode = NULL; shard_priv_t *priv = NULL; shard_local_t *local = NULL; + uint64_t resolve_count = 0; priv = this->private; local = frame->local; local->call_count = 0; shard_idx_iter = local->first_block; res_inode = local->resolver_base_inode; + + if ((local->op_ret < 0) || (local->resolve_not)) + goto out; + + /* If this prealloc FOP is for fresh file creation, then the size of the + * file will be 0. Then there will be no shards associated with this file. + * So we can skip the lookup process for the shards which do not exists + * and directly issue mknod to crete shards. + * + * In case the prealloc fop is to extend the preallocated file to bigger + * size then just lookup and populate inodes of existing shards and + * update the create count + */ + if (local->fop == GF_FOP_FALLOCATE) { + if (!local->prebuf.ia_size) { + local->inode_list[0] = inode_ref(res_inode); + local->create_count = local->last_block; + shard_common_inode_write_post_lookup_shards_handler(frame, this); + return 0; + } + if (local->prebuf.ia_size < local->total_size) + local->create_count = local->last_block - + ((local->prebuf.ia_size - 1) / + local->block_size); + } + + resolve_count = local->last_block - local->create_count; + if (res_inode) gf_uuid_copy(gfid, res_inode->gfid); else gf_uuid_copy(gfid, local->base_gfid); - if ((local->op_ret < 0) || (local->resolve_not)) - goto out; - - while (shard_idx_iter <= local->last_block) { + while (shard_idx_iter <= resolve_count) { i++; if (shard_idx_iter == 0) { local->inode_list[i] = inode_ref(res_inode); @@ -1659,26 +1692,24 @@ err: } int -shard_lookup_base_file_cbk(call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, inode_t *inode, - struct iatt *buf, dict_t *xdata, - struct iatt *postparent) +shard_set_iattr_invoke_post_handler(call_frame_t *frame, xlator_t *this, + inode_t *inode, int32_t op_ret, + int32_t op_errno, struct iatt *buf, + dict_t *xdata) { int ret = -1; int32_t mask = SHARD_INODE_WRITE_MASK; - shard_local_t *local = NULL; + shard_local_t *local = frame->local; shard_inode_ctx_t ctx = { 0, }; - local = frame->local; - if (op_ret < 0) { gf_msg(this->name, GF_LOG_ERROR, op_errno, SHARD_MSG_BASE_FILE_LOOKUP_FAILED, "Lookup on base file" " failed : %s", - loc_gfid_utoa(&(local->loc))); + uuid_utoa(inode->gfid)); local->op_ret = op_ret; local->op_errno = op_errno; goto unwind; @@ -1712,18 +1743,57 @@ unwind: } int -shard_lookup_base_file(call_frame_t *frame, xlator_t *this, loc_t *loc, - shard_post_fop_handler_t handler) +shard_fstat_base_file_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *buf, + dict_t *xdata) +{ + shard_local_t *local = frame->local; + + shard_set_iattr_invoke_post_handler(frame, this, local->fd->inode, op_ret, + op_errno, buf, xdata); + return 0; +} + +int +shard_lookup_base_file_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, inode_t *inode, + struct iatt *buf, dict_t *xdata, + struct iatt *postparent) +{ + /* In case of op_ret < 0, inode passed to this function will be NULL + ex: in case of op_errno = ENOENT. So refer prefilled inode data + which is part of local. + Note: Reassigning/overriding the inode passed to this cbk with inode + which is part of *struct shard_local_t* won't cause any issue as + both inodes have same reference/address as of the inode passed */ + inode = ((shard_local_t *)frame->local)->loc.inode; + + shard_set_iattr_invoke_post_handler(frame, this, inode, op_ret, op_errno, + buf, xdata); + return 0; +} + +/* This function decides whether to make file based lookup or + * fd based lookup (fstat) depending on the 3rd and 4th arg. + * If fd != NULL and loc == NULL then call is for fstat + * If fd == NULL and loc != NULL then call is for file based + * lookup. Please pass args based on the requirement. + */ +int +shard_refresh_base_file(call_frame_t *frame, xlator_t *this, loc_t *loc, + fd_t *fd, shard_post_fop_handler_t handler) { int ret = -1; + inode_t *inode = NULL; shard_local_t *local = NULL; dict_t *xattr_req = NULL; gf_boolean_t need_refresh = _gf_false; local = frame->local; local->handler = handler; + inode = fd ? fd->inode : loc->inode; - ret = shard_inode_ctx_fill_iatt_from_cache(loc->inode, this, &local->prebuf, + ret = shard_inode_ctx_fill_iatt_from_cache(inode, this, &local->prebuf, &need_refresh); /* By this time, inode ctx should have been created either in create, * mknod, readdirp or lookup. If not it is a bug! @@ -1732,7 +1802,7 @@ shard_lookup_base_file(call_frame_t *frame, xlator_t *this, loc_t *loc, gf_msg_debug(this->name, 0, "Skipping lookup on base file: %s" "Serving prebuf off the inode ctx cache", - uuid_utoa(loc->gfid)); + uuid_utoa(inode->gfid)); goto out; } @@ -1743,10 +1813,14 @@ shard_lookup_base_file(call_frame_t *frame, xlator_t *this, loc_t *loc, goto out; } - SHARD_MD_READ_FOP_INIT_REQ_DICT(this, xattr_req, loc->gfid, local, out); + SHARD_MD_READ_FOP_INIT_REQ_DICT(this, xattr_req, inode->gfid, local, out); - STACK_WIND(frame, shard_lookup_base_file_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->lookup, loc, xattr_req); + if (fd) + STACK_WIND(frame, shard_fstat_base_file_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fstat, fd, xattr_req); + else + STACK_WIND(frame, shard_lookup_base_file_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->lookup, loc, xattr_req); dict_unref(xattr_req); return 0; @@ -2015,8 +2089,8 @@ shard_truncate_last_shard(call_frame_t *frame, xlator_t *this, inode_t *inode) */ if (!inode) { gf_msg_debug(this->name, 0, - "Last shard to be truncated absent in backend: %d of " - "gfid %s. Directly proceeding to update file size", + "Last shard to be truncated absent in backend: %" PRIu64 + " of gfid %s. Directly proceeding to update file size", local->first_block, uuid_utoa(local->loc.inode->gfid)); shard_update_file_size(frame, this, NULL, &local->loc, shard_post_update_size_truncate_handler); @@ -2399,7 +2473,7 @@ shard_common_lookup_shards(call_frame_t *frame, xlator_t *this, inode_t *inode, int count = 0; int call_count = 0; int32_t shard_idx_iter = 0; - int last_block = 0; + int lookup_count = 0; char path[PATH_MAX] = { 0, }; @@ -2419,7 +2493,7 @@ shard_common_lookup_shards(call_frame_t *frame, xlator_t *this, inode_t *inode, local = frame->local; count = call_count = local->call_count; shard_idx_iter = local->first_block; - last_block = local->last_block; + lookup_count = local->last_block - local->create_count; local->pls_fop_handler = handler; if (local->lookup_shards_barriered) local->barrier.waitfor = local->call_count; @@ -2429,7 +2503,7 @@ shard_common_lookup_shards(call_frame_t *frame, xlator_t *this, inode_t *inode, else gf_uuid_copy(gfid, local->base_gfid); - while (shard_idx_iter <= last_block) { + while (shard_idx_iter <= lookup_count) { if (local->inode_list[i]) { i++; shard_idx_iter++; @@ -2574,6 +2648,7 @@ shard_truncate_begin(call_frame_t *frame, xlator_t *this) local->block_size); local->num_blocks = local->last_block - local->first_block + 1; + GF_ASSERT(local->num_blocks > 0); local->resolver_base_inode = (local->fop == GF_FOP_TRUNCATE) ? local->loc.inode : local->fd->inode; @@ -2723,8 +2798,8 @@ shard_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset, local->resolver_base_inode = loc->inode; GF_ATOMIC_INIT(local->delta_blocks, 0); - shard_lookup_base_file(frame, this, &local->loc, - shard_post_lookup_truncate_handler); + shard_refresh_base_file(frame, this, &local->loc, NULL, + shard_post_lookup_truncate_handler); return 0; err: @@ -2779,8 +2854,8 @@ shard_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset, local->resolver_base_inode = fd->inode; GF_ATOMIC_INIT(local->delta_blocks, 0); - shard_lookup_base_file(frame, this, &local->loc, - shard_post_lookup_truncate_handler); + shard_refresh_base_file(frame, this, NULL, fd, + shard_post_lookup_truncate_handler); return 0; err: shard_common_failure_unwind(GF_FOP_FTRUNCATE, frame, -1, ENOMEM); @@ -2924,8 +2999,8 @@ shard_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc, if (!local->xattr_req) goto err; - shard_lookup_base_file(frame, this, &local->loc, - shard_post_lookup_link_handler); + shard_refresh_base_file(frame, this, &local->loc, NULL, + shard_post_lookup_link_handler); return 0; err: shard_common_failure_unwind(GF_FOP_LINK, frame, -1, ENOMEM); @@ -2939,13 +3014,20 @@ int shard_post_lookup_shards_unlink_handler(call_frame_t *frame, xlator_t *this) { shard_local_t *local = NULL; + uuid_t gfid = { + 0, + }; local = frame->local; + if (local->resolver_base_inode) + gf_uuid_copy(gfid, local->resolver_base_inode->gfid); + else + gf_uuid_copy(gfid, local->base_gfid); + if ((local->op_ret < 0) && (local->op_errno != ENOENT)) { gf_msg(this->name, GF_LOG_ERROR, local->op_errno, SHARD_MSG_FOP_FAILED, - "failed to delete shards of %s", - uuid_utoa(local->resolver_base_inode->gfid)); + "failed to delete shards of %s", uuid_utoa(gfid)); return 0; } local->op_ret = 0; @@ -4247,8 +4329,8 @@ shard_post_inodelk_fop_handler(call_frame_t *frame, xlator_t *this) switch (local->fop) { case GF_FOP_UNLINK: case GF_FOP_RENAME: - shard_lookup_base_file(frame, this, &local->int_inodelk.loc, - shard_post_lookup_base_shard_rm_handler); + shard_refresh_base_file(frame, this, &local->int_inodelk.loc, NULL, + shard_post_lookup_base_shard_rm_handler); break; default: gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_INVALID_FOP, @@ -4503,8 +4585,8 @@ shard_rename_src_cbk(call_frame_t *frame, void *cookie, xlator_t *this, if (local->block_size) { local->tmp_loc.inode = inode_new(this->itable); gf_uuid_copy(local->tmp_loc.gfid, (local->loc.inode)->gfid); - shard_lookup_base_file(frame, this, &local->tmp_loc, - shard_post_rename_lookup_handler); + shard_refresh_base_file(frame, this, &local->tmp_loc, NULL, + shard_post_rename_lookup_handler); } else { shard_rename_cbk(frame, this); } @@ -5143,6 +5225,7 @@ shard_post_lookup_readv_handler(call_frame_t *frame, xlator_t *this) local->block_size); local->num_blocks = local->last_block - local->first_block + 1; + GF_ASSERT(local->num_blocks > 0); local->resolver_base_inode = local->loc.inode; local->inode_list = GF_CALLOC(local->num_blocks, sizeof(inode_t *), @@ -5239,8 +5322,8 @@ shard_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, local->loc.inode = inode_ref(fd->inode); gf_uuid_copy(local->loc.gfid, fd->inode->gfid); - shard_lookup_base_file(frame, this, &local->loc, - shard_post_lookup_readv_handler); + shard_refresh_base_file(frame, this, NULL, fd, + shard_post_lookup_readv_handler); return 0; err: shard_common_failure_unwind(GF_FOP_READ, frame, -1, ENOMEM); @@ -5603,6 +5686,8 @@ shard_common_inode_write_post_resolve_handler(call_frame_t *frame, shard_common_lookup_shards( frame, this, local->resolver_base_inode, shard_common_inode_write_post_lookup_shards_handler); + } else if (local->create_count) { + shard_common_inode_write_post_lookup_shards_handler(frame, this); } else { shard_common_inode_write_do(frame, this); } @@ -5633,6 +5718,7 @@ shard_common_inode_write_post_lookup_handler(call_frame_t *frame, local->last_block = get_highest_block(local->offset, local->total_size, local->block_size); local->num_blocks = local->last_block - local->first_block + 1; + GF_ASSERT(local->num_blocks > 0); local->inode_list = GF_CALLOC(local->num_blocks, sizeof(inode_t *), gf_shard_mt_inode_list); if (!local->inode_list) { @@ -5641,9 +5727,9 @@ shard_common_inode_write_post_lookup_handler(call_frame_t *frame, } gf_msg_trace(this->name, 0, - "%s: gfid=%s first_block=%" PRIu32 + "%s: gfid=%s first_block=%" PRIu64 " " - "last_block=%" PRIu32 " num_blocks=%" PRIu32 " offset=%" PRId64 + "last_block=%" PRIu64 " num_blocks=%" PRIu64 " offset=%" PRId64 " total_size=%zu flags=%" PRId32 "", gf_fop_list[local->fop], uuid_utoa(local->resolver_base_inode->gfid), @@ -6038,8 +6124,8 @@ shard_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync, local->loc.inode = inode_ref(fd->inode); gf_uuid_copy(local->loc.gfid, fd->inode->gfid); - shard_lookup_base_file(frame, this, &local->loc, - shard_post_lookup_fsync_handler); + shard_refresh_base_file(frame, this, NULL, fd, + shard_post_lookup_fsync_handler); return 0; err: shard_common_failure_unwind(GF_FOP_FSYNC, frame, -1, ENOMEM); @@ -6231,48 +6317,210 @@ shard_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, } int32_t -shard_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc, - const char *name, dict_t *xdata) +shard_modify_and_set_iatt_in_dict(dict_t *xdata, shard_local_t *local, + char *key) { - int op_errno = EINVAL; + int ret = 0; + struct iatt *tmpbuf = NULL; + struct iatt *stbuf = NULL; + data_t *data = NULL; - if (frame->root->pid != GF_CLIENT_PID_GSYNCD) { - GF_IF_NATIVE_XATTR_GOTO(SHARD_XATTR_PREFIX "*", name, op_errno, out); + if (!xdata) + return 0; + + data = dict_get(xdata, key); + if (!data) + return 0; + + tmpbuf = data_to_iatt(data, key); + stbuf = GF_MALLOC(sizeof(struct iatt), gf_common_mt_char); + if (stbuf == NULL) { + local->op_ret = -1; + local->op_errno = ENOMEM; + goto err; } + *stbuf = *tmpbuf; + stbuf->ia_size = local->prebuf.ia_size; + stbuf->ia_blocks = local->prebuf.ia_blocks; + ret = dict_set_iatt(xdata, key, stbuf, false); + if (ret < 0) { + local->op_ret = -1; + local->op_errno = ENOMEM; + goto err; + } + return 0; - if (xdata && (frame->root->pid != GF_CLIENT_PID_GSYNCD)) { - dict_del(xdata, GF_XATTR_SHARD_BLOCK_SIZE); - dict_del(xdata, GF_XATTR_SHARD_FILE_SIZE); +err: + GF_FREE(stbuf); + return -1; +} + +int32_t +shard_common_remove_xattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, dict_t *xdata) +{ + int ret = -1; + shard_local_t *local = NULL; + + local = frame->local; + + if (op_ret < 0) { + local->op_ret = op_ret; + local->op_errno = op_errno; + goto err; } - STACK_WIND_TAIL(frame, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->removexattr, loc, name, xdata); + ret = shard_modify_and_set_iatt_in_dict(xdata, local, GF_PRESTAT); + if (ret < 0) + goto err; + + ret = shard_modify_and_set_iatt_in_dict(xdata, local, GF_POSTSTAT); + if (ret < 0) + goto err; + + if (local->fd) + SHARD_STACK_UNWIND(fremovexattr, frame, local->op_ret, local->op_errno, + xdata); + else + SHARD_STACK_UNWIND(removexattr, frame, local->op_ret, local->op_errno, + xdata); return 0; -out: - shard_common_failure_unwind(GF_FOP_REMOVEXATTR, frame, -1, op_errno); + +err: + shard_common_failure_unwind(local->fop, frame, local->op_ret, + local->op_errno); return 0; } int32_t -shard_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, - const char *name, dict_t *xdata) +shard_post_lookup_remove_xattr_handler(call_frame_t *frame, xlator_t *this) { - int op_errno = EINVAL; + shard_local_t *local = NULL; + + local = frame->local; + if (local->op_ret < 0) { + shard_common_failure_unwind(local->fop, frame, local->op_ret, + local->op_errno); + return 0; + } + + if (local->fd) + STACK_WIND(frame, shard_common_remove_xattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fremovexattr, local->fd, + local->name, local->xattr_req); + else + STACK_WIND(frame, shard_common_remove_xattr_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->removexattr, &local->loc, + local->name, local->xattr_req); + return 0; +} + +int32_t +shard_common_remove_xattr(call_frame_t *frame, xlator_t *this, + glusterfs_fop_t fop, loc_t *loc, fd_t *fd, + const char *name, dict_t *xdata) +{ + int ret = -1; + int op_errno = ENOMEM; + uint64_t block_size = 0; + shard_local_t *local = NULL; + inode_t *inode = loc ? loc->inode : fd->inode; + + if ((IA_ISDIR(inode->ia_type)) || (IA_ISLNK(inode->ia_type))) { + if (loc) + STACK_WIND_TAIL(frame, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->removexattr, loc, name, + xdata); + else + STACK_WIND_TAIL(frame, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fremovexattr, fd, name, + xdata); + return 0; + } + + /* If shard's special xattrs are attempted to be removed, + * fail the fop with EPERM (except if the client is gsyncd). + */ if (frame->root->pid != GF_CLIENT_PID_GSYNCD) { - GF_IF_NATIVE_XATTR_GOTO(SHARD_XATTR_PREFIX "*", name, op_errno, out); + GF_IF_NATIVE_XATTR_GOTO(SHARD_XATTR_PREFIX "*", name, op_errno, err); } + /* Repeat the same check for bulk-removexattr */ if (xdata && (frame->root->pid != GF_CLIENT_PID_GSYNCD)) { dict_del(xdata, GF_XATTR_SHARD_BLOCK_SIZE); dict_del(xdata, GF_XATTR_SHARD_FILE_SIZE); } - STACK_WIND_TAIL(frame, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata); + ret = shard_inode_ctx_get_block_size(inode, this, &block_size); + if (ret) { + gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED, + "Failed to get block size from inode ctx of %s", + uuid_utoa(inode->gfid)); + goto err; + } + + if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) { + if (loc) + STACK_WIND_TAIL(frame, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->removexattr, loc, name, + xdata); + else + STACK_WIND_TAIL(frame, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->fremovexattr, fd, name, + xdata); + return 0; + } + + local = mem_get0(this->local_pool); + if (!local) + goto err; + + frame->local = local; + local->fop = fop; + if (loc) { + if (loc_copy(&local->loc, loc) != 0) + goto err; + } + + if (fd) { + local->fd = fd_ref(fd); + local->loc.inode = inode_ref(fd->inode); + gf_uuid_copy(local->loc.gfid, fd->inode->gfid); + } + + if (name) { + local->name = gf_strdup(name); + if (!local->name) + goto err; + } + + if (xdata) + local->xattr_req = dict_ref(xdata); + + shard_refresh_base_file(frame, this, loc, fd, + shard_post_lookup_remove_xattr_handler); return 0; -out: - shard_common_failure_unwind(GF_FOP_FREMOVEXATTR, frame, -1, op_errno); +err: + shard_common_failure_unwind(fop, frame, -1, op_errno); + return 0; +} + +int32_t +shard_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc, + const char *name, dict_t *xdata) +{ + shard_common_remove_xattr(frame, this, GF_FOP_REMOVEXATTR, loc, NULL, name, + xdata); + return 0; +} + +int32_t +shard_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, + const char *name, dict_t *xdata) +{ + shard_common_remove_xattr(frame, this, GF_FOP_FREMOVEXATTR, NULL, fd, name, + xdata); return 0; } @@ -6353,38 +6601,164 @@ out: } int32_t -shard_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict, - int32_t flags, dict_t *xdata) +shard_common_set_xattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, dict_t *xdata) { - int op_errno = EINVAL; + int ret = -1; + shard_local_t *local = NULL; - if (frame->root->pid != GF_CLIENT_PID_GSYNCD) { - GF_IF_INTERNAL_XATTR_GOTO(SHARD_XATTR_PREFIX "*", dict, op_errno, out); + local = frame->local; +</ |