From fce5f68bc72d448490a0d41be494ac54a9181b3c Mon Sep 17 00:00:00 2001 From: Ravishankar N Date: Thu, 7 Nov 2019 15:18:30 +0530 Subject: afr: make heal info lockless Changes in locks xlator: Added support for per-domain inodelk count requests. Caller needs to set GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS key in the dict and then set each key with name 'GLUSTERFS_INODELK_DOM_PREFIX:'. In the response dict, the xlator will send the per domain count as values for each of these keys. Changes in AFR: Replaced afr_selfheal_locked_inspect() with afr_lockless_inspect(). Logic has been added to make the latter behave same as the former, thus not breaking the current heal info output behaviour. fixes: bz#1774011 Change-Id: I9ae08ce768b39aeb6ee230207b5b7fa744176952 Signed-off-by: Ravishankar N --- xlators/cluster/afr/src/afr-common.c | 203 +++++++++++++++++-------- xlators/cluster/afr/src/afr-self-heal-common.c | 40 ++++- xlators/cluster/afr/src/afr-self-heal.h | 3 +- xlators/features/locks/src/common.h | 4 + xlators/features/locks/src/locks.h | 8 + xlators/features/locks/src/posix.c | 104 +++++++++++-- 6 files changed, 280 insertions(+), 82 deletions(-) (limited to 'xlators') diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c index bc34a023d2f..7913dfe58bf 100644 --- a/xlators/cluster/afr/src/afr-common.c +++ b/xlators/cluster/afr/src/afr-common.c @@ -6882,68 +6882,6 @@ out: return ret; } -int -afr_selfheal_locked_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid, - inode_t **inode, gf_boolean_t *entry_selfheal, - gf_boolean_t *data_selfheal, - gf_boolean_t *metadata_selfheal, - unsigned char *pending) - -{ - int ret = -1; - fd_t *fd = NULL; - gf_boolean_t dsh = _gf_false; - gf_boolean_t msh = _gf_false; - gf_boolean_t esh = _gf_false; - - ret = afr_selfheal_unlocked_inspect(frame, this, gfid, inode, &dsh, &msh, - &esh); - if (ret) - goto out; - - /* For every heal type hold locks and check if it indeed needs heal */ - - /* Heal-info does an open() on the file being examined so that the - * current eager-lock holding client, if present, at some point sees - * open-fd count being > 1 and releases the eager-lock so that heal-info - * doesn't remain blocked forever until IO completes. - */ - if ((*inode)->ia_type == IA_IFREG) { - ret = afr_selfheal_data_open(this, *inode, &fd); - if (ret < 0) { - gf_msg_debug(this->name, -ret, "%s: Failed to open", - uuid_utoa((*inode)->gfid)); - goto out; - } - } - - if (msh) { - ret = afr_selfheal_locked_metadata_inspect(frame, this, *inode, &msh, - pending); - if (ret == -EIO) - goto out; - } - - if (dsh) { - ret = afr_selfheal_locked_data_inspect(frame, this, fd, &dsh, pending); - if (ret == -EIO || (ret == -EAGAIN)) - goto out; - } - - if (esh) { - ret = afr_selfheal_locked_entry_inspect(frame, this, *inode, &esh, - pending); - } - -out: - *data_selfheal = dsh; - *entry_selfheal = esh; - *metadata_selfheal = msh; - if (fd) - fd_unref(fd); - return ret; -} - static dict_t * afr_set_heal_info(char *status) { @@ -6975,6 +6913,132 @@ out: return dict; } +static int +afr_update_heal_status(xlator_t *this, struct afr_reply *replies, + char *index_vgfid, gf_boolean_t *esh, gf_boolean_t *dsh, + gf_boolean_t *msh) +{ + int ret = -1; + int i = 0; + int io_domain_lk_count = 0; + int shd_domain_lk_count = 0; + afr_private_t *priv = NULL; + char *key1 = NULL; + char *key2 = NULL; + + priv = this->private; + key1 = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 + + strlen(priv->sh_domain)); + key2 = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 + + strlen(priv->sh_domain)); + sprintf(key1, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, this->name); + sprintf(key2, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, priv->sh_domain); + + for (i = 0; i < priv->child_count; i++) { + if ((replies[i].valid != 1) || (replies[i].op_ret != 0)) + continue; + if (!io_domain_lk_count) { + ret = dict_get_int32(replies[i].xdata, key1, &io_domain_lk_count); + } + if (!shd_domain_lk_count) { + ret = dict_get_int32(replies[i].xdata, key2, &shd_domain_lk_count); + } + } + + 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 ((!shd_domain_lk_count && !io_domain_lk_count) || + (!io_domain_lk_count && shd_domain_lk_count)) { + /* Needs heal. */ + ret = 0; + } else { + GF_ASSERT(io_domain_lk_count && !shd_domain_lk_count); + /* No heal needed. */ + *dsh = *esh = *msh = 0; + } + } + return ret; +} + +/*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, + gf_boolean_t *metadata_selfheal, unsigned char *pending) +{ + int ret = -1; + int i = 0; + afr_private_t *priv = NULL; + struct afr_reply *replies = NULL; + gf_boolean_t dsh = _gf_false; + gf_boolean_t msh = _gf_false; + gf_boolean_t esh = _gf_false; + unsigned char *sources = NULL; + unsigned char *sinks = NULL; + unsigned char *valid_on = NULL; + uint64_t *witness = NULL; + + priv = this->private; + replies = alloca0(sizeof(*replies) * priv->child_count); + sources = alloca0(sizeof(*sources) * priv->child_count); + sinks = alloca0(sizeof(*sinks) * priv->child_count); + witness = alloca0(sizeof(*witness) * priv->child_count); + valid_on = alloca0(sizeof(*valid_on) * priv->child_count); + + ret = afr_selfheal_unlocked_inspect(frame, this, gfid, inode, &dsh, &msh, + &esh, replies); + if (ret) + goto out; + for (i = 0; i < priv->child_count; i++) { + if (replies[i].valid && replies[i].op_ret == 0) { + valid_on[i] = 1; + } + } + if (msh) { + ret = afr_selfheal_find_direction(frame, this, replies, + AFR_METADATA_TRANSACTION, valid_on, + sources, sinks, witness, pending); + if (*pending & PFLAG_SBRAIN) + ret = -EIO; + if (ret) + goto out; + } + if (dsh) { + ret = afr_selfheal_find_direction(frame, this, replies, + AFR_DATA_TRANSACTION, valid_on, + sources, sinks, witness, pending); + if (*pending & PFLAG_SBRAIN) + ret = -EIO; + if (ret) + goto out; + } + if (esh) { + ret = afr_selfheal_find_direction(frame, this, replies, + AFR_ENTRY_TRANSACTION, valid_on, + sources, sinks, witness, pending); + if (*pending & PFLAG_SBRAIN) + ret = -EIO; + if (ret) + goto out; + } + + ret = afr_update_heal_status(this, replies, index_vgfid, &esh, &dsh, &msh); +out: + *data_selfheal = dsh; + *entry_selfheal = esh; + *metadata_selfheal = msh; + if (replies) + afr_replies_wipe(replies, priv->child_count); + return ret; +} + int afr_get_heal_info(call_frame_t *frame, xlator_t *this, loc_t *loc) { @@ -6990,6 +7054,14 @@ 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); @@ -6999,9 +7071,10 @@ 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_selfheal_locked_inspect(heal_frame, this, loc->gfid, &inode, - &entry_selfheal, &data_selfheal, - &metadata_selfheal, &pending); + + ret = afr_lockless_inspect(heal_frame, this, loc->gfid, &inode, index_vgfid, + &entry_selfheal, &data_selfheal, + &metadata_selfheal, &pending); if (ret == -ENOMEM) { ret = -1; diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c index f06f7e603a0..781437bf74e 100644 --- a/xlators/cluster/afr/src/afr-self-heal-common.c +++ b/xlators/cluster/afr/src/afr-self-heal-common.c @@ -1831,6 +1831,34 @@ afr_selfheal_unlocked_lookup_on(call_frame_t *frame, inode_t *parent, return inode; } +static int +afr_set_multi_dom_lock_count_request(xlator_t *this, dict_t *dict) +{ + int ret = 0; + afr_private_t *priv = NULL; + char *key = NULL; + + priv = this->private; + key = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 + + strlen(priv->sh_domain)); + + ret = dict_set_uint32(dict, GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS, 1); + if (ret) + return ret; + + sprintf(key, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, this->name); + ret = dict_set_uint32(dict, key, 1); + if (ret) + return ret; + + sprintf(key, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, priv->sh_domain); + ret = dict_set_uint32(dict, key, 1); + if (ret) + return ret; + + return 0; +} + int afr_selfheal_unlocked_discover_on(call_frame_t *frame, inode_t *inode, uuid_t gfid, struct afr_reply *replies, @@ -1857,6 +1885,11 @@ afr_selfheal_unlocked_discover_on(call_frame_t *frame, inode_t *inode, return -ENOMEM; } + if (afr_set_multi_dom_lock_count_request(frame->this, xattr_req)) { + dict_unref(xattr_req); + return -1; + } + loc.inode = inode_ref(inode); gf_uuid_copy(loc.gfid, gfid); @@ -2252,7 +2285,8 @@ int afr_selfheal_unlocked_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid, inode_t **link_inode, gf_boolean_t *data_selfheal, gf_boolean_t *metadata_selfheal, - gf_boolean_t *entry_selfheal) + gf_boolean_t *entry_selfheal, + struct afr_reply *replies_dst) { afr_private_t *priv = NULL; inode_t *inode = NULL; @@ -2388,6 +2422,8 @@ afr_selfheal_unlocked_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid, ret = 0; out: + if (replies && replies_dst) + afr_replies_copy(replies_dst, replies, priv->child_count); if (inode) inode_unref(inode); if (replies) @@ -2507,7 +2543,7 @@ afr_selfheal_do(call_frame_t *frame, xlator_t *this, uuid_t gfid) ret = afr_selfheal_unlocked_inspect(frame, this, gfid, &inode, &data_selfheal, &metadata_selfheal, - &entry_selfheal); + &entry_selfheal, NULL); if (ret) goto out; diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h index 55d94e20cea..5e7bde8689d 100644 --- a/xlators/cluster/afr/src/afr-self-heal.h +++ b/xlators/cluster/afr/src/afr-self-heal.h @@ -325,7 +325,8 @@ int afr_selfheal_unlocked_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid, inode_t **link_inode, gf_boolean_t *data_selfheal, gf_boolean_t *metadata_selfheal, - gf_boolean_t *entry_selfheal); + gf_boolean_t *entry_selfheal, + struct afr_reply *replies); int afr_selfheal_do(call_frame_t *frame, xlator_t *this, uuid_t gfid); diff --git a/xlators/features/locks/src/common.h b/xlators/features/locks/src/common.h index 4006062606e..0916c299e84 100644 --- a/xlators/features/locks/src/common.h +++ b/xlators/features/locks/src/common.h @@ -44,6 +44,10 @@ fd_unref(__local->fd); \ if (__local->inode) \ inode_unref(__local->inode); \ + if (__local->xdata) { \ + dict_unref(__local->xdata); \ + __local->xdata = NULL; \ + } \ mem_put(__local); \ } \ } while (0) diff --git a/xlators/features/locks/src/locks.h b/xlators/features/locks/src/locks.h index 0ab2aa6cbae..3305350afb1 100644 --- a/xlators/features/locks/src/locks.h +++ b/xlators/features/locks/src/locks.h @@ -240,6 +240,7 @@ typedef struct { gf_boolean_t inodelk_count_req; gf_boolean_t posixlk_count_req; gf_boolean_t parent_entrylk_req; + gf_boolean_t multiple_dom_lk_requests; int update_mlock_enforced_flag; } pl_local_t; @@ -261,6 +262,13 @@ typedef struct _locks_ctx { struct list_head metalk_list; } pl_ctx_t; +typedef struct _multi_dom_lk_data { + xlator_t *this; + inode_t *inode; + dict_t *xdata_rsp; + gf_boolean_t keep_max; +} multi_dom_lk_data; + typedef enum { DECREMENT, INCREMENT } pl_count_op_t; pl_ctx_t * diff --git a/xlators/features/locks/src/posix.c b/xlators/features/locks/src/posix.c index d2f5db5116b..e03a6063128 100644 --- a/xlators/features/locks/src/posix.c +++ b/xlators/features/locks/src/posix.c @@ -151,13 +151,20 @@ fetch_pathinfo(xlator_t *, inode_t *, int32_t *, char **); gf_boolean_t pl_has_xdata_requests(dict_t *xdata) { - static char *reqs[] = {GLUSTERFS_ENTRYLK_COUNT, GLUSTERFS_INODELK_COUNT, - GLUSTERFS_INODELK_DOM_COUNT, GLUSTERFS_POSIXLK_COUNT, - GLUSTERFS_PARENT_ENTRYLK, NULL}; - static int reqs_size[] = { - SLEN(GLUSTERFS_ENTRYLK_COUNT), SLEN(GLUSTERFS_INODELK_COUNT), - SLEN(GLUSTERFS_INODELK_DOM_COUNT), SLEN(GLUSTERFS_POSIXLK_COUNT), - SLEN(GLUSTERFS_PARENT_ENTRYLK), 0}; + static char *reqs[] = {GLUSTERFS_ENTRYLK_COUNT, + GLUSTERFS_INODELK_COUNT, + GLUSTERFS_INODELK_DOM_COUNT, + GLUSTERFS_POSIXLK_COUNT, + GLUSTERFS_PARENT_ENTRYLK, + GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS, + NULL}; + static int reqs_size[] = {SLEN(GLUSTERFS_ENTRYLK_COUNT), + SLEN(GLUSTERFS_INODELK_COUNT), + SLEN(GLUSTERFS_INODELK_DOM_COUNT), + SLEN(GLUSTERFS_POSIXLK_COUNT), + SLEN(GLUSTERFS_PARENT_ENTRYLK), + SLEN(GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS), + 0}; int i = 0; if (!xdata) @@ -176,6 +183,9 @@ pl_get_xdata_requests(pl_local_t *local, dict_t *xdata) if (!local || !xdata) return; + GF_ASSERT(local->xdata == NULL); + local->xdata = dict_ref(xdata); + if (dict_get_sizen(xdata, GLUSTERFS_ENTRYLK_COUNT)) { local->entrylk_count_req = 1; dict_del_sizen(xdata, GLUSTERFS_ENTRYLK_COUNT); @@ -184,6 +194,10 @@ pl_get_xdata_requests(pl_local_t *local, dict_t *xdata) local->inodelk_count_req = 1; dict_del_sizen(xdata, GLUSTERFS_INODELK_COUNT); } + if (dict_get_sizen(xdata, GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS)) { + local->multiple_dom_lk_requests = 1; + dict_del_sizen(xdata, GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS); + } local->inodelk_dom_count_req = dict_get_sizen(xdata, GLUSTERFS_INODELK_DOM_COUNT); @@ -211,7 +225,7 @@ pl_needs_xdata_response(pl_local_t *local) if (local->parent_entrylk_req || local->entrylk_count_req || local->inodelk_dom_count_req || local->inodelk_count_req || - local->posixlk_count_req) + local->posixlk_count_req || local->multiple_dom_lk_requests) return _gf_true; return _gf_false; @@ -410,6 +424,71 @@ pl_posixlk_xattr_fill(xlator_t *this, inode_t *inode, dict_t *dict, } } +void +pl_inodelk_xattr_fill_each(xlator_t *this, inode_t *inode, dict_t *dict, + char *domname, gf_boolean_t keep_max, char *key) +{ + int32_t count = 0; + int32_t maxcount = -1; + int ret = -1; + + if (keep_max) { + ret = dict_get_int32(dict, key, &maxcount); + if (ret < 0) + gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s", + GLUSTERFS_INODELK_COUNT); + } + count = get_inodelk_count(this, inode, domname); + if (maxcount >= count) + return; + + ret = dict_set_int32(dict, key, count); + if (ret < 0) { + gf_msg_debug(this->name, 0, + "Failed to set count for " + "key %s", + key); + } + + return; +} + +static int +pl_inodelk_xattr_fill_multiple(dict_t *this, char *key, data_t *value, + void *data) +{ + multi_dom_lk_data *d = data; + char *tmp_key = NULL; + char *save_ptr = NULL; + + tmp_key = gf_strdup(key); + strtok_r(tmp_key, ":", &save_ptr); + GF_ASSERT(*save_ptr); // Should contain the domain string. + + pl_inodelk_xattr_fill_each(d->this, d->inode, d->xdata_rsp, save_ptr, + d->keep_max, key); + if (tmp_key) + GF_FREE(tmp_key); + + return 0; +} + +void +pl_fill_multiple_dom_lk_requests(xlator_t *this, pl_local_t *local, + inode_t *inode, dict_t *dict, + gf_boolean_t keep_max) +{ + multi_dom_lk_data data; + + data.this = this; + data.inode = inode; + data.xdata_rsp = dict; + data.keep_max = keep_max; + + dict_foreach_fnmatch(local->xdata, GLUSTERFS_INODELK_DOM_PREFIX "*", + pl_inodelk_xattr_fill_multiple, &data); +} + void pl_set_xdata_response(xlator_t *this, pl_local_t *local, inode_t *parent, inode_t *inode, char *name, dict_t *xdata, @@ -437,6 +516,9 @@ pl_set_xdata_response(xlator_t *this, pl_local_t *local, inode_t *parent, if (local->posixlk_count_req) pl_posixlk_xattr_fill(this, inode, xdata, max_lock); + + if (local->multiple_dom_lk_requests) + pl_fill_multiple_dom_lk_requests(this, local, inode, xdata, max_lock); } /* Checks whether the region where fop is acting upon conflicts @@ -775,9 +857,6 @@ pl_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, { pl_local_t *local = frame->local; - if (local->xdata) - dict_unref(local->xdata); - pl_track_io_fop_count(local, this, DECREMENT); if (local->op == GF_FOP_TRUNCATE) @@ -935,9 +1014,6 @@ unwind: "ret: %d, error: %s", op_ret, strerror(op_errno)); - if (local->xdata) - dict_unref(local->xdata); - switch (local->op) { case GF_FOP_TRUNCATE: PL_STACK_UNWIND(truncate, xdata, frame, op_ret, op_errno, buf, -- cgit