diff options
author | Xavier Hernandez <xhernandez@datalab.es> | 2014-07-14 17:34:04 +0200 |
---|---|---|
committer | Vijay Bellur <vbellur@redhat.com> | 2014-09-15 23:12:16 -0700 |
commit | d97863562bb0d2f685df3d2e3aa4bef1299c8307 (patch) | |
tree | b8d9455cdface5425e2452e98751ac75dac358e4 /xlators/cluster | |
parent | 2be54585002cd1c9d02928b89a02047b58dd6aed (diff) |
ec: Optimize read/write performance
This patch significantly improves performance of read/write
operations on a dispersed volume by reusing previous inodelk/
entrylk operations on the same inode/entry. This reduces the
latency of each individual operation considerably.
Inode version and size are also updated when needed instead
of on each request. This gives an additional boost.
Change-Id: I4b98d5508c86b53032e16e295f72a3f83fd8fcac
BUG: 1122586
Signed-off-by: Xavier Hernandez <xhernandez@datalab.es>
Reviewed-on: http://review.gluster.org/8369
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Dan Lambright <dlambrig@redhat.com>
Diffstat (limited to 'xlators/cluster')
-rw-r--r-- | xlators/cluster/ec/src/ec-combine.c | 8 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec-common.c | 545 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec-common.h | 17 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec-data.c | 1 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec-data.h | 31 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec-dir-write.c | 96 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec-generic.c | 96 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec-helpers.c | 3 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec-inode-read.c | 47 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec-inode-write.c | 117 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec-mem-types.h | 3 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec.c | 9 | ||||
-rw-r--r-- | xlators/cluster/ec/src/ec.h | 1 |
13 files changed, 706 insertions, 268 deletions
diff --git a/xlators/cluster/ec/src/ec-combine.c b/xlators/cluster/ec/src/ec-combine.c index 3d088d9be4a..02b7e6024fd 100644 --- a/xlators/cluster/ec/src/ec-combine.c +++ b/xlators/cluster/ec/src/ec-combine.c @@ -735,7 +735,7 @@ void ec_combine(ec_cbk_data_t * cbk, ec_combine_f combine) ec_fop_data_t * fop = cbk->fop; ec_cbk_data_t * ans = NULL, * tmp = NULL; struct list_head * item = NULL; - int32_t needed = 0, report = 0; + int32_t needed = 0, resume = 0; char str[32]; LOCK(&fop->lock); @@ -776,7 +776,7 @@ void ec_combine(ec_cbk_data_t * cbk, ec_combine_f combine) ec_update_bad(fop, cbk->mask); - report = 1; + resume = 1; } ans = list_entry(fop->cbk_list.next, ec_cbk_data_t, list); @@ -788,8 +788,8 @@ void ec_combine(ec_cbk_data_t * cbk, ec_combine_f combine) { ec_dispatch_next(fop, cbk->idx); } - else if (report) + else if (resume) { - ec_report(fop, 0); + ec_resume(fop, 0); } } diff --git a/xlators/cluster/ec/src/ec-common.c b/xlators/cluster/ec/src/ec-common.c index a4423d94aa9..ad04e646d68 100644 --- a/xlators/cluster/ec/src/ec-common.c +++ b/xlators/cluster/ec/src/ec-common.c @@ -316,20 +316,10 @@ void ec_resume_parent(ec_fop_data_t * fop, int32_t error) } } -void ec_report(ec_fop_data_t * fop, int32_t error) -{ - if (!list_empty(&fop->lock_list)) - { - ec_owner_set(fop->frame, fop->frame->root); - } - - ec_resume(fop, error); -} - void ec_complete(ec_fop_data_t * fop) { ec_cbk_data_t * cbk = NULL; - int32_t ready = 0, report = 0; + int32_t resume = 0; LOCK(&fop->lock); @@ -351,21 +341,17 @@ void ec_complete(ec_fop_data_t * fop) } } - report = 1; + resume = 1; } else if ((fop->flags & EC_FLAG_WAITING_WINDS) != 0) { - ready = 1; + resume = 1; } } UNLOCK(&fop->lock); - if (report) - { - ec_report(fop, 0); - } - if (ready) + if (resume) { ec_resume(fop, 0); } @@ -518,7 +504,7 @@ void ec_dispatch_start(ec_fop_data_t * fop) INIT_LIST_HEAD(&fop->cbk_list); - if (!list_empty(&fop->lock_list)) + if (fop->lock_count > 0) { ec_owner_copy(fop->frame, &fop->req_frame->root->lk_owner); } @@ -602,6 +588,7 @@ void ec_dispatch_min(ec_fop_data_t * fop) ec_lock_t * ec_lock_allocate(xlator_t * xl, int32_t kind, loc_t * loc) { + ec_t * ec = xl->private; ec_lock_t * lock; if ((loc->inode == NULL) || @@ -613,15 +600,15 @@ ec_lock_t * ec_lock_allocate(xlator_t * xl, int32_t kind, loc_t * loc) return NULL; } - lock = GF_MALLOC(sizeof(*lock), ec_mt_ec_lock_t); + lock = mem_get0(ec->lock_pool); if (lock != NULL) { - memset(lock, 0, sizeof(*lock)); - lock->kind = kind; + lock->good_mask = -1ULL; + INIT_LIST_HEAD(&lock->waiting); if (!ec_loc_from_loc(xl, &lock->loc, loc)) { - GF_FREE(lock); + mem_put(lock); lock = NULL; } } @@ -634,34 +621,55 @@ void ec_lock_destroy(ec_lock_t * lock) GF_FREE(lock->basename); loc_wipe(&lock->loc); - GF_FREE(lock); + mem_put(lock); } -int32_t ec_locked(call_frame_t * frame, void * cookie, xlator_t * this, - int32_t op_ret, int32_t op_errno, dict_t * xdata) +int32_t ec_lock_compare(ec_lock_t * lock1, ec_lock_t * lock2) { - ec_fop_data_t * fop = cookie; - ec_lock_t * lock = NULL; + int32_t res; - if (op_ret >= 0) + res = uuid_compare(lock1->loc.gfid, lock2->loc.gfid); + if (res != 0) { - lock = fop->data; - lock->mask = fop->good; - fop->parent->mask &= fop->good; - - ec_trace("LOCKED", fop->parent, "lock=%p", lock); + return res; } - else + if (lock1->basename == NULL) { - gf_log(this->name, GF_LOG_WARNING, "Failed to complete preop lock"); + if (lock2->basename == NULL) + { + return 0; + } + return 1; + } + if (lock2->basename == NULL) + { + return -1; } + return strcmp(lock1->basename, lock2->basename); +} - return 0; +void ec_lock_insert(ec_fop_data_t * fop, ec_lock_t * lock) +{ + ec_lock_t * tmp; + + if ((fop->lock_count > 0) && + (ec_lock_compare(fop->locks[0].lock, lock) > 0)) + { + tmp = fop->locks[0].lock; + fop->locks[0].lock = lock; + lock = tmp; + } + fop->locks[fop->lock_count].lock = lock; + fop->locks[fop->lock_count].fop = fop; + fop->lock_count++; + + lock->refs++; } -void ec_lock_entry(ec_fop_data_t * fop, loc_t * loc) +void ec_lock_prepare_entry(ec_fop_data_t * fop, loc_t * loc) { ec_lock_t * lock = NULL; + ec_inode_t * ctx = NULL; char * name = NULL; loc_t tmp; int32_t error; @@ -679,116 +687,106 @@ void ec_lock_entry(ec_fop_data_t * fop, loc_t * loc) return; } - LOCK(&fop->lock); + LOCK(&tmp.inode->lock); - list_for_each_entry(lock, &fop->lock_list, list) + ctx = __ec_inode_get(tmp.inode, fop->xl); + if (ctx == NULL) { - if ((lock->kind == EC_LOCK_ENTRY) && - (lock->loc.inode == tmp.inode) && - (strcmp(lock->basename, name) == 0)) - { - ec_trace("LOCK_ENTRYLK", fop, "lock=%p, parent=%p, path=%s, " - "name=%s. Lock already acquired", - lock, loc->parent, loc->path, name); - - lock = NULL; + __ec_fop_set_error(fop, EIO); - goto unlock; - } + goto unlock; } - lock = ec_lock_allocate(fop->xl, EC_LOCK_ENTRY, &tmp); - if (lock != NULL) + list_for_each_entry(lock, &ctx->entry_locks, list) { - lock->type = ENTRYLK_WRLCK; - lock->basename = name; - - if (list_empty(&fop->lock_list)) + if (strcmp(lock->basename, name) == 0) { - ec_owner_set(fop->frame, fop->frame->root); + ec_trace("LOCK_ENTRYLK", fop, "lock=%p, inode=%p, path=%s, " + "name=%s. Lock already acquired", + lock, tmp.inode, tmp.path, name); + + goto insert; } - list_add_tail(&lock->list, &fop->lock_list); } - else + + lock = ec_lock_allocate(fop->xl, EC_LOCK_ENTRY, &tmp); + if (lock == NULL) { __ec_fop_set_error(fop, EIO); + + goto unlock; } -unlock: - UNLOCK(&fop->lock); + ec_trace("LOCK_CREATE", fop, "lock=%p", lock); - loc_wipe(&tmp); + lock->type = ENTRYLK_WRLCK; + lock->basename = name; + name = NULL; - if (lock != NULL) - { - ec_trace("LOCK_ENTRYLK", fop, "lock=%p, parent=%p, path=%s, " - "basename=%s", lock, lock->loc.inode, - lock->loc.path, lock->basename); + list_add_tail(&lock->list, &ctx->entry_locks); - ec_entrylk(fop->frame, fop->xl, -1, EC_MINIMUM_ALL, ec_locked, lock, - fop->xl->name, &lock->loc, lock->basename, ENTRYLK_LOCK, - lock->type, NULL); - } - else - { - GF_FREE(name); - } +insert: + ec_lock_insert(fop, lock); + +unlock: + UNLOCK(&tmp.inode->lock); + + loc_wipe(&tmp); + GF_FREE(name); } -void ec_lock_inode(ec_fop_data_t * fop, loc_t * loc) +void ec_lock_prepare_inode(ec_fop_data_t * fop, loc_t * loc) { ec_lock_t * lock; + ec_inode_t * ctx; if ((fop->parent != NULL) || (fop->error != 0) || (loc->inode == NULL)) { return; } - LOCK(&fop->lock); + LOCK(&loc->inode->lock); - list_for_each_entry(lock, &fop->lock_list, list) + ctx = __ec_inode_get(loc->inode, fop->xl); + if (ctx == NULL) { - if ((lock->kind == EC_LOCK_INODE) && (lock->loc.inode == loc->inode)) - { - UNLOCK(&fop->lock); - - ec_trace("LOCK_INODELK", fop, "lock=%p, inode=%p. Lock already " - "acquired", lock, loc->inode); + __ec_fop_set_error(fop, EIO); - return; - } + goto unlock; } - lock = ec_lock_allocate(fop->xl, EC_LOCK_INODE, loc); - if (lock != NULL) + if (!list_empty(&ctx->inode_locks)) { - lock->flock.l_type = F_WRLCK; - lock->flock.l_whence = SEEK_SET; + lock = list_entry(ctx->inode_locks.next, ec_lock_t, list); + ec_trace("LOCK_INODELK", fop, "lock=%p, inode=%p. Lock already " + "acquired", lock, loc->inode); - if (list_empty(&fop->lock_list)) - { - ec_owner_set(fop->frame, fop->frame->root); - } - list_add_tail(&lock->list, &fop->lock_list); + goto insert; } - else + + lock = ec_lock_allocate(fop->xl, EC_LOCK_INODE, loc); + if (lock == NULL) { __ec_fop_set_error(fop, EIO); + + goto unlock; } - UNLOCK(&fop->lock); + ec_trace("LOCK_CREATE", fop, "lock=%p", lock); - if (lock != NULL) - { - ec_trace("LOCK_INODELK", fop, "lock=%p, inode=%p, owner=%p", lock, - lock->loc.inode, fop->frame->root); + lock->flock.l_type = F_WRLCK; + lock->flock.l_whence = SEEK_SET; - ec_inodelk(fop->frame, fop->xl, -1, EC_MINIMUM_ALL, ec_locked, lock, - fop->xl->name, &lock->loc, F_SETLKW, &lock->flock, NULL); - } + list_add_tail(&lock->list, &ctx->inode_locks); + +insert: + ec_lock_insert(fop, lock); + +unlock: + UNLOCK(&loc->inode->lock); } -void ec_lock_fd(ec_fop_data_t * fop, fd_t * fd) +void ec_lock_prepare_fd(ec_fop_data_t * fop, fd_t * fd) { loc_t loc; @@ -799,7 +797,7 @@ void ec_lock_fd(ec_fop_data_t * fop, fd_t * fd) if (ec_loc_from_fd(fop->xl, &loc, fd)) { - ec_lock_inode(fop, &loc); + ec_lock_prepare_inode(fop, &loc); loc_wipe(&loc); } @@ -809,6 +807,100 @@ void ec_lock_fd(ec_fop_data_t * fop, fd_t * fd) } } +int32_t ec_locked(call_frame_t * frame, void * cookie, xlator_t * this, + int32_t op_ret, int32_t op_errno, dict_t * xdata) +{ + ec_fop_data_t * fop = cookie; + ec_lock_t * lock = NULL; + + if (op_ret >= 0) + { + lock = fop->data; + lock->mask = fop->good; + lock->acquired = 1; + + fop->parent->mask &= fop->good; + fop->parent->locked++; + + ec_trace("LOCKED", fop->parent, "lock=%p", lock); + + ec_lock(fop->parent); + } + else + { + gf_log(this->name, GF_LOG_WARNING, "Failed to complete preop lock"); + } + + return 0; +} + +void ec_lock(ec_fop_data_t * fop) +{ + ec_lock_t * lock; + + while (fop->locked < fop->lock_count) + { + lock = fop->locks[fop->locked].lock; + + LOCK(&lock->loc.inode->lock); + + if (lock->owner != NULL) + { + ec_trace("LOCK_WAIT", fop, "lock=%p", lock); + + list_add_tail(&fop->locks[fop->locked].wait_list, &lock->waiting); + + fop->jobs++; + fop->refs++; + + UNLOCK(&lock->loc.inode->lock); + + break; + } + lock->owner = fop; + + UNLOCK(&lock->loc.inode->lock); + + if (!lock->acquired) + { + ec_owner_set(fop->frame, lock); + + if (lock->kind == EC_LOCK_ENTRY) + { + ec_trace("LOCK_ACQUIRE", fop, "lock=%p, inode=%p, path=%s, " + "name=%s", lock, lock->loc.inode, lock->loc.path, + lock->basename); + + ec_entrylk(fop->frame, fop->xl, -1, EC_MINIMUM_ALL, ec_locked, + lock, fop->xl->name, &lock->loc, lock->basename, + ENTRYLK_LOCK, lock->type, NULL); + } + else + { + ec_trace("LOCK_ACQUIRE", fop, "lock=%p, inode=%p", lock, + lock->loc.inode); + + ec_inodelk(fop->frame, fop->xl, -1, EC_MINIMUM_ALL, ec_locked, + lock, fop->xl->name, &lock->loc, F_SETLKW, + &lock->flock, NULL); + } + + break; + } + + ec_trace("LOCK_REUSE", fop, "lock=%p", lock); + + if (lock->have_size) + { + fop->pre_size = fop->post_size = lock->size; + fop->have_size = 1; + } + fop->mask &= lock->good_mask; + + fop->locked++; + } +} + int32_t ec_unlocked(call_frame_t * frame, void * cookie, xlator_t * this, int32_t op_ret, int32_t op_errno, dict_t * xdata) { @@ -829,50 +921,68 @@ int32_t ec_unlocked(call_frame_t * frame, void * cookie, xlator_t * this, void ec_unlock(ec_fop_data_t * fop) { - ec_lock_t * lock, * item; - - ec_trace("UNLOCK", fop, ""); + ec_lock_t * lock; + int32_t i, refs; - list_for_each_entry_safe(lock, item, &fop->lock_list, list) + for (i = 0; i < fop->lock_count; i++) { - list_del(&lock->list); + lock = fop->locks[i].lock; - if (lock->mask != 0) - { - switch (lock->kind) - { - case EC_LOCK_ENTRY: - ec_trace("UNLOCK_ENTRYLK", fop, "lock=%p, parent=%p, " - "path=%s, basename=%s", - lock, lock->loc.inode, lock->loc.path, - lock->basename); + LOCK(&lock->loc.inode->lock); - ec_entrylk(fop->frame, fop->xl, lock->mask, EC_MINIMUM_ALL, - ec_unlocked, lock, fop->xl->name, &lock->loc, - lock->basename, ENTRYLK_UNLOCK, lock->type, - NULL); + ec_trace("UNLOCK", fop, "lock=%p", lock); - break; - - case EC_LOCK_INODE: - lock->flock.l_type = F_UNLCK; - ec_trace("UNLOCK_INODELK", fop, "lock=%p, inode=%p", lock, - lock->loc.inode); + refs = --lock->refs; + if (refs == 0) + { + list_del_init(&lock->list); + } - ec_inodelk(fop->frame, fop->xl, lock->mask, EC_MINIMUM_ALL, - ec_unlocked, lock, fop->xl->name, &lock->loc, - F_SETLK, &lock->flock, NULL); + UNLOCK(&lock->loc.inode->lock); - break; + if (refs == 0) + { + if (lock->mask != 0) + { + ec_owner_set(fop->frame, lock); - default: - gf_log(fop->xl->name, GF_LOG_ERROR, "Invalid lock type"); + switch (lock->kind) + { + case EC_LOCK_ENTRY: + ec_trace("UNLOCK_ENTRYLK", fop, "lock=%p, inode=%p, " + "path=%s, basename=%s", + lock, lock->loc.inode, lock->loc.path, + lock->basename); + + ec_entrylk(fop->frame, fop->xl, lock->mask, + EC_MINIMUM_ALL, ec_unlocked, lock, + fop->xl->name, &lock->loc, lock->basename, + ENTRYLK_UNLOCK, lock->type, NULL); + + break; + + case EC_LOCK_INODE: + lock->flock.l_type = F_UNLCK; + ec_trace("UNLOCK_INODELK", fop, "lock=%p, inode=%p", + lock, lock->loc.inode); + + ec_inodelk(fop->frame, fop->xl, lock->mask, + EC_MINIMUM_ALL, ec_unlocked, lock, + fop->xl->name, &lock->loc, F_SETLK, + &lock->flock, NULL); + + break; + + default: + gf_log(fop->xl->name, GF_LOG_ERROR, "Invalid lock " + "type"); + } } - } - loc_wipe(&lock->loc); + ec_trace("LOCK_DESTROY", fop, "lock=%p", lock); - GF_FREE(lock); + ec_lock_destroy(lock); + } } } @@ -883,11 +993,36 @@ int32_t ec_get_size_version_set(call_frame_t * frame, void * cookie, struct iatt * postparent) { ec_fop_data_t * fop = cookie; + ec_inode_t * ctx; + ec_lock_t * lock; if (op_ret >= 0) { - fop->parent->mask &= fop->good; + LOCK(&inode->lock); + + ctx = __ec_inode_get(inode, this); + if ((ctx != NULL) && !list_empty(&ctx->inode_locks)) + { + lock = list_entry(ctx->inode_locks.next, ec_lock_t, list); + + lock->have_size = 1; + lock->size = buf->ia_size; + lock->version = fop->answer->version; + } + + UNLOCK(&inode->lock); + + if (lock != NULL) + { + // Only update parent mask if the lookup has been made with + // inode locked. + fop->parent->mask &= fop->good; + } + fop->parent->pre_size = fop->parent->post_size = buf->ia_size; + + fop->parent->have_size = 1; + } else { @@ -907,11 +1042,18 @@ void ec_get_size_version(ec_fop_data_t * fop) gid_t gid; int32_t error = ENOMEM; - if (fop->parent != NULL) + if (fop->have_size) + { + return; + } + + if ((fop->parent != NULL) && fop->parent->have_size) { fop->pre_size = fop->parent->pre_size; fop->post_size = fop->parent->post_size; + fop->have_size = 1; + return; } @@ -998,10 +1140,10 @@ int32_t ec_update_size_version_done(call_frame_t * frame, void * cookie, return 0; } -void ec_update_size_version(ec_fop_data_t * fop) +void ec_update_size_version(ec_fop_data_t * fop, uint64_t version, + size_t size) { dict_t * dict; - size_t size; uid_t uid; gid_t gid; @@ -1012,20 +1154,20 @@ void ec_update_size_version(ec_fop_data_t * fop) return; } + ec_trace("UPDATE", fop, "version=%ld, size=%ld", version, size); + dict = dict_new(); if (dict == NULL) { goto out; } - if (ec_dict_set_number(dict, EC_XATTR_VERSION, 1) != 0) + if (ec_dict_set_number(dict, EC_XATTR_VERSION, version) != 0) { goto out; } - size = fop->post_size; - if (fop->pre_size != size) + if (size != 0) { - size -= fop->pre_size; if (ec_dict_set_number(dict, EC_XATTR_SIZE, size) != 0) { goto out; @@ -1069,6 +1211,113 @@ out: gf_log(fop->xl->name, GF_LOG_ERROR, "Unable to update version and size"); } +void ec_flush_size_version(ec_fop_data_t * fop) +{ + ec_lock_t * lock; + uint64_t version; + size_t delta; + + GF_ASSERT(fop->lock_count == 1); + + lock = fop->locks[0].lock; + + GF_ASSERT(lock->kind == EC_LOCK_INODE); + + LOCK(&lock->loc.inode->lock); + + GF_ASSERT(lock->owner == fop); + + version = lock->version_delta; + delta = lock->size_delta; + lock->version_delta = 0; + lock->size_delta = 0; + + UNLOCK(&lock->loc.inode->lock); + + if (version > 0) + { + ec_update_size_version(fop, version, delta); + } +} + +void ec_lock_reuse(ec_fop_data_t * fop, int32_t update) +{ + ec_fop_data_t * wait_fop; + ec_lock_t * lock; + ec_lock_link_t * link; + size_t delta = 0; + uint64_t version = 0; + int32_t refs = 0; + int32_t i; + + for (i = 0; i < fop->lock_count; i++) + { + wait_fop = NULL; + + lock = fop->locks[i].lock; + + LOCK(&lock->loc.inode->lock); + + ec_trace("LOCK_DONE", fop, "lock=%p", lock); + + GF_ASSERT(lock->owner == fop); + lock->owner = NULL; + + if (lock->kind == EC_LOCK_INODE) + { + if (update && (fop->error == 0)) + { + lock->version_delta++; + lock->size_delta += fop->post_size - fop->pre_size; + } + version = lock->version_delta; + delta = lock->size_delta; + refs = lock->refs; + if (refs == 1) + { + lock->version_delta = 0; + lock->size_delta = 0; + } + + if (fop->have_size) + { + lock->size = fop->post_size; + lock->have_size = 1; + } + } + lock->good_mask &= fop->mask; + + if (!list_empty(&lock->waiting)) + { + link = list_entry(lock->waiting.next, ec_lock_link_t, wait_list); + list_del_init(&link->wait_list); + + wait_fop = link->fop; + + if (lock->kind == EC_LOCK_INODE) + { + wait_fop->pre_size = wait_fop->post_size = fop->post_size; + wait_fop->have_size = fop->have_size; + } + wait_fop->mask &= fop->mask; + } + + UNLOCK(&lock->loc.inode->lock); + + if (wait_fop != NULL) + { + ec_lock(wait_fop); + + ec_resume(wait_fop, 0); + } + } + + if ((refs == 1) && (version > 0)) + { + ec_update_size_version(fop, version, delta); + } +} + void __ec_manager(ec_fop_data_t * fop, int32_t error) { do diff --git a/xlators/cluster/ec/src/ec-common.h b/xlators/cluster/ec/src/ec-common.h index 83f3ba9637e..4fc89fdde33 100644 --- a/xlators/cluster/ec/src/ec-common.h +++ b/xlators/cluster/ec/src/ec-common.h @@ -47,10 +47,10 @@ #define EC_STATE_DISPATCH 4 #define EC_STATE_PREPARE_ANSWER 5 #define EC_STATE_REPORT 6 -#define EC_STATE_UPDATE_SIZE_AND_VERSION 7 +#define EC_STATE_LOCK_REUSE 7 #define EC_STATE_UNLOCK 8 -#define EC_STATE_WRITE_START 100 +#define EC_STATE_DELAYED_START 100 #define EC_STATE_HEAL_ENTRY_LOOKUP 200 #define EC_STATE_HEAL_ENTRY_PREPARE 201 @@ -81,14 +81,15 @@ void ec_update_bad(ec_fop_data_t * fop, uintptr_t good); void ec_fop_set_error(ec_fop_data_t * fop, int32_t error); -void ec_lock_inode(ec_fop_data_t * fop, loc_t * loc); -void ec_lock_entry(ec_fop_data_t * fop, loc_t * loc); -void ec_lock_fd(ec_fop_data_t * fop, fd_t * fd); - +void ec_lock_prepare_inode(ec_fop_data_t * fop, loc_t * loc); +void ec_lock_prepare_entry(ec_fop_data_t * fop, loc_t * loc); +void ec_lock_prepare_fd(ec_fop_data_t * fop, fd_t * fd); +void ec_lock(ec_fop_data_t * fop); +void ec_lock_reuse(ec_fop_data_t * fop, int32_t update); void ec_unlock(ec_fop_data_t * fop); void ec_get_size_version(ec_fop_data_t * fop); -void ec_update_size_version(ec_fop_data_t * fop); +void ec_flush_size_version(ec_fop_data_t * fop); void ec_dispatch_all(ec_fop_data_t * fop); void ec_dispatch_inc(ec_fop_data_t * fop); @@ -97,8 +98,8 @@ void ec_dispatch_one(ec_fop_data_t * fop); void ec_wait_winds(ec_fop_data_t * fop); +void ec_resume(ec_fop_data_t * fop, int32_t error); void ec_resume_parent(ec_fop_data_t * fop, int32_t error); -void ec_report(ec_fop_data_t * fop, int32_t error); void ec_manager(ec_fop_data_t * fop, int32_t error); diff --git a/xlators/cluster/ec/src/ec-data.c b/xlators/cluster/ec/src/ec-data.c index 0e72fbbd3b6..174586b4051 100644 --- a/xlators/cluster/ec/src/ec-data.c +++ b/xlators/cluster/ec/src/ec-data.c @@ -158,7 +158,6 @@ ec_fop_data_t * ec_fop_data_allocate(call_frame_t * frame, xlator_t * this, fop->minimum = minimum; fop->mask = target; - INIT_LIST_HEAD(&fop->lock_list); INIT_LIST_HEAD(&fop->cbk_list); INIT_LIST_HEAD(&fop->answer_list); diff --git a/xlators/cluster/ec/src/ec-data.h b/xlators/cluster/ec/src/ec-data.h index e83b6ad74eb..cf9ce241adb 100644 --- a/xlators/cluster/ec/src/ec-data.h +++ b/xlators/cluster/ec/src/ec-data.h @@ -37,6 +37,9 @@ typedef union _ec_cbk ec_cbk_t; struct _ec_lock; typedef struct _ec_lock ec_lock_t; +struct _ec_lock_link; +typedef struct _ec_lock_link ec_lock_link_t; + struct _ec_fop_data; typedef struct _ec_fop_data ec_fop_data_t; @@ -60,8 +63,10 @@ struct _ec_fd struct _ec_inode { - uintptr_t bad; - ec_heal_t * heal; + uintptr_t bad; + struct list_head entry_locks; + struct list_head inode_locks; + ec_heal_t * heal; }; typedef int32_t (* fop_heal_cbk_t)(call_frame_t *, void * cookie, xlator_t *, @@ -124,8 +129,18 @@ union _ec_cbk struct _ec_lock { struct list_head list; + struct list_head waiting; uintptr_t mask; + uintptr_t good_mask; int32_t kind; + int32_t refs; + int32_t acquired; + int32_t have_size; + size_t size; + size_t size_delta; + uint64_t version; + uint64_t version_delta; + ec_fop_data_t * owner; loc_t loc; union { @@ -138,6 +153,13 @@ struct _ec_lock }; }; +struct _ec_lock_link +{ + ec_lock_t * lock; + ec_fop_data_t * fop; + struct list_head wait_list; +}; + struct _ec_fop_data { int32_t id; @@ -152,10 +174,13 @@ struct _ec_fop_data xlator_t * xl; call_frame_t * req_frame; // frame of the calling xlator call_frame_t * frame; // frame used by this fop - struct list_head lock_list; // list locks held by this fop struct list_head cbk_list; // sorted list of groups of answers struct list_head answer_list; // list of answers ec_cbk_data_t * answer; // accepted answer + int32_t lock_count; + int32_t locked; + ec_lock_link_t locks[2]; + int32_t have_size; size_t pre_size; size_t post_size; gf_lock_t lock; diff --git a/xlators/cluster/ec/src/ec-dir-write.c b/xlators/cluster/ec/src/ec-dir-write.c index dc1d94e37e8..a28b7ad1937 100644 --- a/xlators/cluster/ec/src/ec-dir-write.c +++ b/xlators/cluster/ec/src/ec-dir-write.c @@ -181,7 +181,8 @@ int32_t ec_manager_create(ec_fop_data_t * fop, int32_t state) /* Fall through */ case EC_STATE_LOCK: - ec_lock_entry(fop, &fop->loc[0]); + ec_lock_prepare_entry(fop, &fop->loc[0]); + ec_lock(fop); return EC_STATE_DISPATCH; @@ -245,11 +246,7 @@ int32_t ec_manager_create(ec_fop_data_t * fop, int32_t state) cbk->xdata); } - if (cbk->op_ret >= 0) - { - return EC_STATE_UPDATE_SIZE_AND_VERSION; - } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -263,14 +260,14 @@ int32_t ec_manager_create(ec_fop_data_t * fop, int32_t state) NULL, NULL, NULL, NULL, NULL, NULL); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; - case EC_STATE_UPDATE_SIZE_AND_VERSION: - ec_update_size_version(fop); + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 1); return EC_STATE_UNLOCK; - case -EC_STATE_UPDATE_SIZE_AND_VERSION: case -EC_STATE_UNLOCK: case EC_STATE_UNLOCK: ec_unlock(fop); @@ -468,7 +465,8 @@ int32_t ec_manager_link(ec_fop_data_t * fop, int32_t state) // Parent entry of fop->loc[0] should be locked, but I don't // receive enough information to do it (fop->loc[0].parent is // NULL). - ec_lock_entry(fop, &fop->loc[1]); + ec_lock_prepare_entry(fop, &fop->loc[1]); + ec_lock(fop); return EC_STATE_GET_SIZE_AND_VERSION; @@ -531,7 +529,7 @@ int32_t ec_manager_link(ec_fop_data_t * fop, int32_t state) &cbk->iatt[1], &cbk->iatt[2], cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_GET_SIZE_AND_VERSION: @@ -546,6 +544,12 @@ int32_t ec_manager_link(ec_fop_data_t * fop, int32_t state) NULL, NULL, NULL, NULL, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -732,7 +736,8 @@ int32_t ec_manager_mkdir(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_entry(fop, &fop->loc[0]); + ec_lock_prepare_entry(fop, &fop->loc[0]); + ec_lock(fop); return EC_STATE_DISPATCH; @@ -785,7 +790,7 @@ int32_t ec_manager_mkdir(ec_fop_data_t * fop, int32_t state) &cbk->iatt[1], &cbk->iatt[2], cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -799,6 +804,12 @@ int32_t ec_manager_mkdir(ec_fop_data_t * fop, int32_t state) NULL, NULL, NULL, NULL, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -982,7 +993,8 @@ int32_t ec_manager_mknod(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_entry(fop, &fop->loc[0]); + ec_lock_prepare_entry(fop, &fop->loc[0]); + ec_lock(fop); return EC_STATE_DISPATCH; @@ -1035,7 +1047,7 @@ int32_t ec_manager_mknod(ec_fop_data_t * fop, int32_t state) &cbk->iatt[1], &cbk->iatt[2], cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -1049,6 +1061,12 @@ int32_t ec_manager_mknod(ec_fop_data_t * fop, int32_t state) NULL, NULL, NULL, NULL, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -1230,8 +1248,9 @@ int32_t ec_manager_rename(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_entry(fop, &fop->loc[0]); - ec_lock_entry(fop, &fop->loc[1]); + ec_lock_prepare_entry(fop, &fop->loc[0]); + ec_lock_prepare_entry(fop, &fop->loc[1]); + ec_lock(fop); return EC_STATE_GET_SIZE_AND_VERSION; @@ -1292,7 +1311,7 @@ int32_t ec_manager_rename(ec_fop_data_t * fop, int32_t state) cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_GET_SIZE_AND_VERSION: @@ -1307,6 +1326,12 @@ int32_t ec_manager_rename(ec_fop_data_t * fop, int32_t state) NULL, NULL, NULL, NULL, NULL, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -1479,7 +1504,8 @@ int32_t ec_manager_rmdir(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_entry(fop, &fop->loc[0]); + ec_lock_prepare_entry(fop, &fop->loc[0]); + ec_lock(fop); return EC_STATE_DISPATCH; @@ -1524,7 +1550,7 @@ int32_t ec_manager_rmdir(ec_fop_data_t * fop, int32_t state) cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -1538,6 +1564,12 @@ int32_t ec_manager_rmdir(ec_fop_data_t * fop, int32_t state) NULL, NULL, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -1719,7 +1751,8 @@ int32_t ec_manager_symlink(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_entry(fop, &fop->loc[0]); + ec_lock_prepare_entry(fop, &fop->loc[0]); + ec_lock(fop); return EC_STATE_DISPATCH; @@ -1772,7 +1805,7 @@ int32_t ec_manager_symlink(ec_fop_data_t * fop, int32_t state) &cbk->iatt[1], &cbk->iatt[2], cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -1786,6 +1819,12 @@ int32_t ec_manager_symlink(ec_fop_data_t * fop, int32_t state) NULL, NULL, NULL, NULL, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -1963,7 +2002,8 @@ int32_t ec_manager_unlink(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_entry(fop, &fop->loc[0]); + ec_lock_prepare_entry(fop, &fop->loc[0]); + ec_lock(fop); return EC_STATE_GET_SIZE_AND_VERSION; @@ -2013,7 +2053,7 @@ int32_t ec_manager_unlink(ec_fop_data_t * fop, int32_t state) cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_GET_SIZE_AND_VERSION: @@ -2028,6 +2068,12 @@ int32_t ec_manager_unlink(ec_fop_data_t * fop, int32_t state) NULL, NULL, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: diff --git a/xlators/cluster/ec/src/ec-generic.c b/xlators/cluster/ec/src/ec-generic.c index 4afec3524c5..4aa02903969 100644 --- a/xlators/cluster/ec/src/ec-generic.c +++ b/xlators/cluster/ec/src/ec-generic.c @@ -91,11 +91,17 @@ int32_t ec_manager_flush(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_fd(fop, fop->fd); + ec_lock_prepare_fd(fop, fop->fd); + ec_lock(fop); return EC_STATE_DISPATCH; case EC_STATE_DISPATCH: + ec_flush_size_version(fop); + + return EC_STATE_DELAYED_START; + + case EC_STATE_DELAYED_START: ec_dispatch_all(fop); return EC_STATE_PREPARE_ANSWER; @@ -135,7 +141,7 @@ int32_t ec_manager_flush(ec_fop_data_t * fop, int32_t state) cbk->op_errno, cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -149,6 +155,12 @@ int32_t ec_manager_flush(ec_fop_data_t * fop, int32_t state) NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -313,7 +325,8 @@ int32_t ec_manager_fsync(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_fd(fop, fop->fd); + ec_lock_prepare_fd(fop, fop->fd); + ec_lock(fop); return EC_STATE_GET_SIZE_AND_VERSION; @@ -323,6 +336,11 @@ int32_t ec_manager_fsync(ec_fop_data_t * fop, int32_t state) return EC_STATE_DISPATCH; case EC_STATE_DISPATCH: + ec_flush_size_version(fop); + + return EC_STATE_DELAYED_START; + + case EC_STATE_DELAYED_START: ec_dispatch_all(fop); return EC_STATE_PREPARE_ANSWER; @@ -371,7 +389,7 @@ int32_t ec_manager_fsync(ec_fop_data_t * fop, int32_t state) cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_GET_SIZE_AND_VERSION: @@ -386,6 +404,12 @@ int32_t ec_manager_fsync(ec_fop_data_t * fop, int32_t state) NULL, NULL, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -526,11 +550,17 @@ int32_t ec_manager_fsyncdir(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_fd(fop, fop->fd); + ec_lock_prepare_fd(fop, fop->fd); + ec_lock(fop); return EC_STATE_DISPATCH; case EC_STATE_DISPATCH: + ec_flush_size_version(fop); + + return EC_STATE_DELAYED_START; + + case EC_STATE_DELAYED_START: ec_dispatch_all(fop); return EC_STATE_PREPARE_ANSWER; @@ -570,7 +600,7 @@ int32_t ec_manager_fsyncdir(ec_fop_data_t * fop, int32_t state) cbk->op_errno, cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -584,6 +614,12 @@ int32_t ec_manager_fsyncdir(ec_fop_data_t * fop, int32_t state) fop->error, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -665,10 +701,12 @@ out: void ec_lookup_rebuild(ec_t * ec, ec_fop_data_t * fop, ec_cbk_data_t * cbk) { ec_cbk_data_t * ans = NULL; + ec_inode_t * ctx = NULL; + ec_lock_t * lock = NULL; data_t * data = NULL; uint8_t * buff = NULL; size_t size = 0; - int32_t i = 0; + int32_t i = 0, have_size = 0; if (cbk->op_ret < 0) { @@ -679,6 +717,22 @@ void ec_lookup_rebuild(ec_t * ec, ec_fop_data_t * fop, ec_cbk_data_t * cbk) ec_loc_prepare(fop->xl, &fop->loc[0], cbk->inode, &cbk->iatt[0]); + LOCK(&cbk->inode->lock); + + ctx = __ec_inode_get(cbk->inode, fop->xl); + if ((ctx != NULL) && !list_empty(&ctx->inode_locks)) + { + lock = list_entry(ctx->inode_locks.next, ec_lock_t, list); + cbk->version = lock->version; + if (lock->have_size) + { + size = lock->size; + have_size = 1; + } + } + + UNLOCK(&cbk->inode->lock); + if (cbk->iatt[0].ia_type == IA_IFREG) { uint8_t * blocks[cbk->count]; @@ -686,6 +740,10 @@ void ec_lookup_rebuild(ec_t * ec, ec_fop_data_t * fop, ec_cbk_data_t * cbk) cbk->size = cbk->iatt[0].ia_size; ec_dict_del_number(cbk->xdata, EC_XATTR_SIZE, &cbk->iatt[0].ia_size); + if (have_size) + { + cbk->iatt[0].ia_size = size; + } size = SIZE_MAX; for (i = 0, ans = cbk; (ans != NULL) && (i < ec->fragments); @@ -1314,7 +1372,15 @@ int32_t ec_manager_xattrop(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_inode(fop, &fop->loc[0]); + if (fop->fd == NULL) + { + ec_lock_prepare_inode(fop, &fop->loc[0]); + } + else + { + ec_lock_prepare_fd(fop, fop->fd); + } + ec_lock(fop); return EC_STATE_DISPATCH; @@ -1373,11 +1439,7 @@ int32_t ec_manager_xattrop(ec_fop_data_t * fop, int32_t state) } } - if (cbk->op_ret >= 0) - { - return EC_STATE_UPDATE_SIZE_AND_VERSION; - } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -1402,14 +1464,14 @@ int32_t ec_manager_xattrop(ec_fop_data_t * fop, int32_t state) } } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; - case EC_STATE_UPDATE_SIZE_AND_VERSION: - ec_update_size_version(fop); + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 1); return EC_STATE_UNLOCK; - case -EC_STATE_UPDATE_SIZE_AND_VERSION: case -EC_STATE_UNLOCK: case EC_STATE_UNLOCK: ec_unlock(fop); diff --git a/xlators/cluster/ec/src/ec-helpers.c b/xlators/cluster/ec/src/ec-helpers.c index 6eb836a1c62..22889337cbc 100644 --- a/xlators/cluster/ec/src/ec-helpers.c +++ b/xlators/cluster/ec/src/ec-helpers.c @@ -503,6 +503,9 @@ ec_inode_t * __ec_inode_get(inode_t * inode, xlator_t * xl) return NULL; } + + INIT_LIST_HEAD(&ctx->entry_locks); + INIT_LIST_HEAD(&ctx->inode_locks); } } else diff --git a/xlators/cluster/ec/src/ec-inode-read.c b/xlators/cluster/ec/src/ec-inode-read.c index a31220ecbc1..0cb5559f62b 100644 --- a/xlators/cluster/ec/src/ec-inode-read.c +++ b/xlators/cluster/ec/src/ec-inode-read.c @@ -252,7 +252,15 @@ int32_t ec_manager_getxattr(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_inode(fop, &fop->loc[0]); + if (fop->fd == NULL) + { + ec_lock_prepare_inode(fop, &fop->loc[0]); + } + else + { + ec_lock_prepare_fd(fop, fop->fd); + } + ec_lock(fop); return EC_STATE_DISPATCH; @@ -311,7 +319,7 @@ int32_t ec_manager_getxattr(ec_fop_data_t * fop, int32_t state) cbk->op_errno, cbk->dict, cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -325,6 +333,12 @@ int32_t ec_manager_getxattr(ec_fop_data_t * fop, int32_t state) fop->error, NULL, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -1216,7 +1230,8 @@ int32_t ec_manager_readv(ec_fop_data_t * fop, int32_t state) /* Fall through */ case EC_STATE_LOCK: - ec_lock_fd(fop, fop->fd); + ec_lock_prepare_fd(fop, fop->fd); + ec_lock(fop); return EC_STATE_GET_SIZE_AND_VERSION; @@ -1276,7 +1291,7 @@ int32_t ec_manager_readv(ec_fop_data_t * fop, int32_t state) &cbk->iatt[0], cbk->buffers, cbk->xdata); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_GET_SIZE_AND_VERSION: @@ -1291,6 +1306,12 @@ int32_t ec_manager_readv(ec_fop_data_t * fop, int32_t state) NULL, 0, NULL, NULL, NULL); } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: @@ -1455,7 +1476,15 @@ int32_t ec_manager_stat(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_inode(fop, &fop->loc[0]); + if (fop->fd == NULL) + { + ec_lock_prepare_inode(fop, &fop->loc[0]); + } + else + { + ec_lock_prepare_fd(fop, fop->fd); + } + ec_lock(fop); return EC_STATE_GET_SIZE_AND_VERSION; @@ -1522,7 +1551,7 @@ int32_t ec_manager_stat(ec_fop_data_t * fop, int32_t state) } } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_GET_SIZE_AND_VERSION: @@ -1548,6 +1577,12 @@ int32_t ec_manager_stat(ec_fop_data_t * fop, int32_t state) } } + return EC_STATE_LOCK_REUSE; + + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 0); + return EC_STATE_UNLOCK; case -EC_STATE_UNLOCK: diff --git a/xlators/cluster/ec/src/ec-inode-write.c b/xlators/cluster/ec/src/ec-inode-write.c index 06a2fef8d17..edc7409854d 100644 --- a/xlators/cluster/ec/src/ec-inode-write.c +++ b/xlators/cluster/ec/src/ec-inode-write.c @@ -92,7 +92,15 @@ int32_t ec_manager_removexattr(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_inode(fop, &fop->loc[0]); + if (fop->fd == NULL) + { + ec_lock_prepare_inode(fop, &fop->loc[0]); + } + else + { + ec_lock_prepare_fd(fop, fop->fd); + } + ec_lock(fop); return EC_STATE_DISPATCH; @@ -149,11 +157,7 @@ int32_t ec_manager_removexattr(ec_fop_data_t * fop, int32_t state) } } - if (cbk->op_ret >= 0) - { - return EC_STATE_UPDATE_SIZE_AND_VERSION; - } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -178,14 +182,14 @@ int32_t ec_manager_removexattr(ec_fop_data_t * fop, int32_t state) } } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; - case EC_STATE_UPDATE_SIZE_AND_VERSION: - ec_update_size_version(fop); + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 1); return EC_STATE_UNLOCK; - case -EC_STATE_UPDATE_SIZE_AND_VERSION: case -EC_STATE_UNLOCK: case EC_STATE_UNLOCK: ec_unlock(fop); @@ -484,7 +488,15 @@ int32_t ec_manager_setattr(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_inode(fop, &fop->loc[0]); + if (fop->fd == NULL) + { + ec_lock_prepare_inode(fop, &fop->loc[0]); + } + else + { + ec_lock_prepare_fd(fop, fop->fd); + } + ec_lock(fop); return EC_STATE_GET_SIZE_AND_VERSION; @@ -556,11 +568,7 @@ int32_t ec_manager_setattr(ec_fop_data_t * fop, int32_t state) } } - if (cbk->op_ret >= 0) - { - return EC_STATE_UPDATE_SIZE_AND_VERSION; - } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_GET_SIZE_AND_VERSION: @@ -586,14 +594,14 @@ int32_t ec_manager_setattr(ec_fop_data_t * fop, int32_t state) } } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; - case EC_STATE_UPDATE_SIZE_AND_VERSION: - ec_update_size_version(fop); + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 1); return EC_STATE_UNLOCK; - case -EC_STATE_UPDATE_SIZE_AND_VERSION: case -EC_STATE_UNLOCK: case EC_STATE_UNLOCK: ec_unlock(fop); @@ -870,7 +878,15 @@ int32_t ec_manager_setxattr(ec_fop_data_t * fop, int32_t state) { case EC_STATE_INIT: case EC_STATE_LOCK: - ec_lock_inode(fop, &fop->loc[0]); + if (fop->fd == NULL) + { + ec_lock_prepare_inode(fop, &fop->loc[0]); + } + else + { + ec_lock_prepare_fd(fop, fop->fd); + } + ec_lock(fop); return EC_STATE_DISPATCH; @@ -926,11 +942,7 @@ int32_t ec_manager_setxattr(ec_fop_data_t * fop, int32_t state) } } - if (cbk->op_ret >= 0) - { - return EC_STATE_UPDATE_SIZE_AND_VERSION; - } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_DISPATCH: @@ -955,14 +967,14 @@ int32_t ec_manager_setxattr(ec_fop_data_t * fop, int32_t state) } } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; - case EC_STATE_UPDATE_SIZE_AND_VERSION: - ec_update_size_version(fop); + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 1); return EC_STATE_UNLOCK; - case -EC_STATE_UPDATE_SIZE_AND_VERSION: case -EC_STATE_UNLOCK: case EC_STATE_UNLOCK: ec_unlock(fop); @@ -1366,7 +1378,15 @@ int32_t ec_manager_truncate(ec_fop_data_t * fop, int32_t state) /* Fall through */ case EC_STATE_LOCK: - ec_lock_inode(fop, &fop->loc[0]); + if (fop->fd == NULL) + { + ec_lock_prepare_inode(fop, &fop->loc[0]); + } + else + { + ec_lock_prepare_fd(fop, fop->fd); + } + ec_lock(fop); return EC_STATE_GET_SIZE_AND_VERSION; @@ -1447,11 +1467,7 @@ int32_t ec_manager_truncate(ec_fop_data_t * fop, int32_t state) } } - if (cbk->op_ret >= 0) - { - return EC_STATE_UPDATE_SIZE_AND_VERSION; - } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_GET_SIZE_AND_VERSION: @@ -1477,14 +1493,14 @@ int32_t ec_manager_truncate(ec_fop_data_t * fop, int32_t state) } } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; - case EC_STATE_UPDATE_SIZE_AND_VERSION: - ec_update_size_version(fop); + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 1); return EC_STATE_UNLOCK; - case -EC_STATE_UPDATE_SIZE_AND_VERSION: case -EC_STATE_UNLOCK: case EC_STATE_UNLOCK: ec_unlock(fop); @@ -2003,7 +2019,8 @@ int32_t ec_manager_writev(ec_fop_data_t * fop, int32_t state) /* Fall through */ case EC_STATE_LOCK: - ec_lock_fd(fop, fop->fd); + ec_lock_prepare_fd(fop, fop->fd); + ec_lock(fop); return EC_STATE_GET_SIZE_AND_VERSION; @@ -2015,9 +2032,9 @@ int32_t ec_manager_writev(ec_fop_data_t * fop, int32_t state) case EC_STATE_DISPATCH: ec_writev_start(fop); - return EC_STATE_WRITE_START; + return EC_STATE_DELAYED_START; - case EC_STATE_WRITE_START: + case EC_STATE_DELAYED_START: ec_dispatch_all(fop); return EC_STATE_PREPARE_ANSWER; @@ -2089,11 +2106,7 @@ int32_t ec_manager_writev(ec_fop_data_t * fop, int32_t state) cbk->xdata); } - if (cbk->op_ret >= 0) - { - return EC_STATE_UPDATE_SIZE_AND_VERSION; - } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; case -EC_STATE_LOCK: case -EC_STATE_GET_SIZE_AND_VERSION: @@ -2108,14 +2121,14 @@ int32_t ec_manager_writev(ec_fop_data_t * fop, int32_t state) NULL, NULL, NULL); } - return EC_STATE_UNLOCK; + return EC_STATE_LOCK_REUSE; - case EC_STATE_UPDATE_SIZE_AND_VERSION: - ec_update_size_version(fop); + case -EC_STATE_LOCK_REUSE: + case EC_STATE_LOCK_REUSE: + ec_lock_reuse(fop, 1); return EC_STATE_UNLOCK; - case -EC_STATE_UPDATE_SIZE_AND_VERSION: case -EC_STATE_UNLOCK: case EC_STATE_UNLOCK: ec_unlock(fop); diff --git a/xlators/cluster/ec/src/ec-mem-types.h b/xlators/cluster/ec/src/ec-mem-types.h index f312d1e333f..a7b0c838c7e 100644 --- a/xlators/cluster/ec/src/ec-mem-types.h +++ b/xlators/cluster/ec/src/ec-mem-types.h @@ -27,11 +27,8 @@ enum gf_ec_mem_types_ { ec_mt_ec_t = gf_common_mt_end + 1, ec_mt_xlator_t, - ec_mt_ec_fop_data_t, - ec_mt_ec_cbk_data_t, ec_mt_ec_inode_t, ec_mt_ec_fd_t, - ec_mt_ec_lock_t, ec_mt_ec_heal_t, ec_mt_end }; diff --git a/xlators/cluster/ec/src/ec.c b/xlators/cluster/ec/src/ec.c index 8554f20df0d..93bee1a4d32 100644 --- a/xlators/cluster/ec/src/ec.c +++ b/xlators/cluster/ec/src/ec.c @@ -151,6 +151,11 @@ void __ec_destroy_private(xlator_t * this) mem_pool_destroy(ec->cbk_pool); } + if (ec->lock_pool != NULL) + { + mem_pool_destroy(ec->lock_pool); + } + LOCK_DESTROY(&ec->lock); GF_FREE(ec); @@ -350,7 +355,9 @@ int32_t init(xlator_t * this) ec->fop_pool = mem_pool_new(ec_fop_data_t, 1024); ec->cbk_pool = mem_pool_new(ec_cbk_data_t, 4096); - if ((ec->fop_pool == NULL) || (ec->cbk_pool == NULL)) + ec->lock_pool = mem_pool_new(ec_lock_t, 1024); + if ((ec->fop_pool == NULL) || (ec->cbk_pool == NULL) || + (ec->lock_pool == NULL)) { gf_log(this->name, GF_LOG_ERROR, "Failed to create memory pools."); diff --git a/xlators/cluster/ec/src/ec.h b/xlators/cluster/ec/src/ec.h index 85c430e2045..0cc8fdb4403 100644 --- a/xlators/cluster/ec/src/ec.h +++ b/xlators/cluster/ec/src/ec.h @@ -49,6 +49,7 @@ struct _ec gf_timer_t * timer; struct mem_pool * fop_pool; struct mem_pool * cbk_pool; + struct mem_pool * lock_pool; }; #endif /* __EC_H__ */ |