diff options
| -rw-r--r-- | xlators/cluster/afr/src/afr-common.c | 9 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-dir-write.c | 143 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-lk-common.c | 814 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-messages.h | 2 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr-transaction.c | 170 | ||||
| -rw-r--r-- | xlators/cluster/afr/src/afr.h | 26 |
6 files changed, 366 insertions, 798 deletions
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c index 32fa634def1..0f77607fee0 100644 --- a/xlators/cluster/afr/src/afr-common.c +++ b/xlators/cluster/afr/src/afr-common.c @@ -1794,11 +1794,9 @@ afr_local_transaction_cleanup(afr_local_t *local, xlator_t *this) afr_matrix_cleanup(local->pending, priv->child_count); - GF_FREE(local->internal_lock.locked_nodes); - GF_FREE(local->internal_lock.lower_locked_nodes); - afr_entry_lockee_cleanup(&local->internal_lock); + afr_lockees_cleanup(&local->internal_lock); GF_FREE(local->transaction.pre_op); @@ -5708,11 +5706,6 @@ afr_internal_lock_init(afr_internal_lock_t *lk, size_t child_count) { int ret = -ENOMEM; - lk->locked_nodes = GF_CALLOC(sizeof(*lk->locked_nodes), child_count, - gf_afr_mt_char); - if (NULL == lk->locked_nodes) - goto out; - lk->lower_locked_nodes = GF_CALLOC(sizeof(*lk->lower_locked_nodes), child_count, gf_afr_mt_char); if (NULL == lk->lower_locked_nodes) diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c index 486b06b162f..f2c8ef450e1 100644 --- a/xlators/cluster/afr/src/afr-dir-write.c +++ b/xlators/cluster/afr/src/afr-dir-write.c @@ -425,15 +425,11 @@ int afr_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) { - afr_private_t *priv = NULL; afr_local_t *local = NULL; - afr_internal_lock_t *int_lock = NULL; call_frame_t *transaction_frame = NULL; int ret = -1; int op_errno = ENOMEM; - priv = this->private; - transaction_frame = copy_frame(frame); if (!transaction_frame) goto out; @@ -475,16 +471,6 @@ afr_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME(loc->path); - int_lock = &local->internal_lock; - - int_lock->lockee_count = 0; - ret = afr_init_entry_lockee(&int_lock->lockee[0], local, - &local->transaction.parent_loc, - local->transaction.basename, priv->child_count); - if (ret) - goto out; - - int_lock->lockee_count++; ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION); if (ret < 0) { op_errno = -ret; @@ -555,15 +541,11 @@ int afr_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode, dev_t dev, mode_t umask, dict_t *xdata) { - afr_private_t *priv = NULL; afr_local_t *local = NULL; - afr_internal_lock_t *int_lock = NULL; call_frame_t *transaction_frame = NULL; int ret = -1; int op_errno = ENOMEM; - priv = this->private; - transaction_frame = copy_frame(frame); if (!transaction_frame) goto out; @@ -598,16 +580,6 @@ afr_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME(loc->path); - int_lock = &local->internal_lock; - - int_lock->lockee_count = 0; - ret = afr_init_entry_lockee(&int_lock->lockee[0], local, - &local->transaction.parent_loc, - local->transaction.basename, priv->child_count); - if (ret) - goto out; - - int_lock->lockee_count++; ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION); if (ret < 0) { op_errno = -ret; @@ -676,15 +648,11 @@ int afr_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode, mode_t umask, dict_t *xdata) { - afr_private_t *priv = NULL; afr_local_t *local = NULL; - afr_internal_lock_t *int_lock = NULL; call_frame_t *transaction_frame = NULL; int ret = -1; int op_errno = ENOMEM; - priv = this->private; - transaction_frame = copy_frame(frame); if (!transaction_frame) goto out; @@ -726,16 +694,6 @@ afr_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME(loc->path); - int_lock = &local->internal_lock; - - int_lock->lockee_count = 0; - ret = afr_init_entry_lockee(&int_lock->lockee[0], local, - &local->transaction.parent_loc, - local->transaction.basename, priv->child_count); - if (ret) - goto out; - - int_lock->lockee_count++; ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION); if (ret < 0) { op_errno = -ret; @@ -804,15 +762,11 @@ int afr_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc, dict_t *xdata) { - afr_private_t *priv = NULL; afr_local_t *local = NULL; - afr_internal_lock_t *int_lock = NULL; call_frame_t *transaction_frame = NULL; int ret = -1; int op_errno = ENOMEM; - priv = this->private; - transaction_frame = copy_frame(frame); if (!transaction_frame) goto out; @@ -847,16 +801,6 @@ afr_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME(newloc->path); - int_lock = &local->internal_lock; - - int_lock->lockee_count = 0; - ret = afr_init_entry_lockee(&int_lock->lockee[0], local, - &local->transaction.parent_loc, - local->transaction.basename, priv->child_count); - if (ret) - goto out; - - int_lock->lockee_count++; ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION); if (ret < 0) { op_errno = -ret; @@ -926,15 +870,11 @@ int afr_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath, loc_t *loc, mode_t umask, dict_t *xdata) { - afr_private_t *priv = NULL; afr_local_t *local = NULL; - afr_internal_lock_t *int_lock = NULL; call_frame_t *transaction_frame = NULL; int ret = -1; int op_errno = ENOMEM; - priv = this->private; - transaction_frame = copy_frame(frame); if (!transaction_frame) goto out; @@ -968,16 +908,6 @@ afr_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME(loc->path); - int_lock = &local->internal_lock; - - int_lock->lockee_count = 0; - ret = afr_init_entry_lockee(&int_lock->lockee[0], local, - &local->transaction.parent_loc, - local->transaction.basename, priv->child_count); - if (ret) - goto out; - - int_lock->lockee_count++; ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION); if (ret < 0) { op_errno = -ret; @@ -1050,15 +980,10 @@ int afr_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc, dict_t *xdata) { - afr_private_t *priv = NULL; afr_local_t *local = NULL; - afr_internal_lock_t *int_lock = NULL; call_frame_t *transaction_frame = NULL; int ret = -1; int op_errno = ENOMEM; - int nlockee = 0; - - priv = this->private; transaction_frame = copy_frame(frame); if (!transaction_frame) { @@ -1101,35 +1026,6 @@ afr_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME(oldloc->path); local->transaction.new_basename = AFR_BASENAME(newloc->path); - int_lock = &local->internal_lock; - - int_lock->lockee_count = nlockee = 0; - ret = afr_init_entry_lockee( - &int_lock->lockee[nlockee], local, &local->transaction.new_parent_loc, - local->transaction.new_basename, priv->child_count); - if (ret) - goto out; - - nlockee++; - ret = afr_init_entry_lockee(&int_lock->lockee[nlockee], local, - &local->transaction.parent_loc, - local->transaction.basename, priv->child_count); - if (ret) - goto out; - - nlockee++; - if (local->newloc.inode && IA_ISDIR(local->newloc.inode->ia_type)) { - ret = afr_init_entry_lockee(&int_lock->lockee[nlockee], local, - &local->newloc, NULL, priv->child_count); - if (ret) - goto out; - - nlockee++; - } - qsort(int_lock->lockee, nlockee, sizeof(*int_lock->lockee), - afr_entry_lockee_cmp); - int_lock->lockee_count = nlockee; - ret = afr_transaction(transaction_frame, this, AFR_ENTRY_RENAME_TRANSACTION); if (ret < 0) { @@ -1198,15 +1094,11 @@ int afr_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag, dict_t *xdata) { - afr_private_t *priv = NULL; afr_local_t *local = NULL; - afr_internal_lock_t *int_lock = NULL; call_frame_t *transaction_frame = NULL; int ret = -1; int op_errno = ENOMEM; - priv = this->private; - transaction_frame = copy_frame(frame); if (!transaction_frame) goto out; @@ -1239,16 +1131,6 @@ afr_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME(loc->path); - int_lock = &local->internal_lock; - - int_lock->lockee_count = 0; - ret = afr_init_entry_lockee(&int_lock->lockee[0], local, - &local->transaction.parent_loc, - local->transaction.basename, priv->child_count); - if (ret) - goto out; - - int_lock->lockee_count++; ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION); if (ret < 0) { op_errno = -ret; @@ -1315,15 +1197,10 @@ int afr_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags, dict_t *xdata) { - afr_private_t *priv = NULL; afr_local_t *local = NULL; - afr_internal_lock_t *int_lock = NULL; call_frame_t *transaction_frame = NULL; int ret = -1; int op_errno = ENOMEM; - int nlockee = 0; - - priv = this->private; transaction_frame = copy_frame(frame); if (!transaction_frame) @@ -1357,26 +1234,6 @@ afr_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags, local->transaction.main_frame = frame; local->transaction.basename = AFR_BASENAME(loc->path); - int_lock = &local->internal_lock; - - int_lock->lockee_count = nlockee = 0; - ret = afr_init_entry_lockee(&int_lock->lockee[nlockee], local, - &local->transaction.parent_loc, - local->transaction.basename, priv->child_count); - if (ret) - goto out; - - nlockee++; - ret = afr_init_entry_lockee(&int_lock->lockee[nlockee], local, &local->loc, - NULL, priv->child_count); - if (ret) - goto out; - - nlockee++; - qsort(int_lock->lockee, nlockee, sizeof(*int_lock->lockee), - afr_entry_lockee_cmp); - int_lock->lockee_count = nlockee; - ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION); if (ret < 0) { op_errno = -ret; diff --git a/xlators/cluster/afr/src/afr-lk-common.c b/xlators/cluster/afr/src/afr-lk-common.c index a80eab0a349..4091671fa3f 100644 --- a/xlators/cluster/afr/src/afr-lk-common.c +++ b/xlators/cluster/afr/src/afr-lk-common.c @@ -22,11 +22,40 @@ #define LOCKED_YES 0x1 /* for DATA, METADATA, ENTRY and higher_path */ #define LOCKED_LOWER 0x2 /* for lower path */ +void +afr_lockee_cleanup(afr_lockee_t *lockee) +{ + if (lockee->fd) { + fd_unref(lockee->fd); + lockee->fd = NULL; + } else { + loc_wipe(&lockee->loc); + } + + GF_FREE(lockee->basename); + lockee->basename = NULL; + GF_FREE(lockee->locked_nodes); + lockee->locked_nodes = NULL; + + return; +} + +void +afr_lockees_cleanup(afr_internal_lock_t *int_lock) +{ + int i = 0; + + for (i = 0; i < int_lock->lockee_count; i++) { + afr_lockee_cleanup(&int_lock->lockee[i]); + } + + return; +} int afr_entry_lockee_cmp(const void *l1, const void *l2) { - const afr_entry_lockee_t *r1 = l1; - const afr_entry_lockee_t *r2 = l2; + const afr_lockee_t *r1 = l1; + const afr_lockee_t *r2 = l2; int ret = 0; uuid_t gfid1 = {0}; uuid_t gfid2 = {0}; @@ -81,31 +110,14 @@ internal_lock_count(call_frame_t *frame, xlator_t *this) } int -afr_is_inodelk_transaction(afr_transaction_type type) +afr_add_entry_lockee(afr_local_t *local, loc_t *loc, char *basename, + int child_count) { - int ret = 0; - - switch (type) { - case AFR_DATA_TRANSACTION: - case AFR_METADATA_TRANSACTION: - ret = 1; - break; - - case AFR_ENTRY_RENAME_TRANSACTION: - case AFR_ENTRY_TRANSACTION: - ret = 0; - break; - } - - return ret; -} - -int -afr_init_entry_lockee(afr_entry_lockee_t *lockee, afr_local_t *local, - loc_t *loc, char *basename, int child_count) -{ - int ret = -1; + int ret = -ENOMEM; + afr_internal_lock_t *int_lock = &local->internal_lock; + afr_lockee_t *lockee = &int_lock->lockee[int_lock->lockee_count]; + GF_ASSERT(int_lock->lockee_count < AFR_LOCKEE_COUNT_MAX); loc_copy(&lockee->loc, loc); lockee->basename = (basename) ? gf_strdup(basename) : NULL; if (basename && !lockee->basename) @@ -119,28 +131,45 @@ afr_init_entry_lockee(afr_entry_lockee_t *lockee, afr_local_t *local, goto out; ret = 0; + int_lock->lockee_count++; out: + if (ret) { + afr_lockee_cleanup(lockee); + } return ret; } -void -afr_entry_lockee_cleanup(afr_internal_lock_t *int_lock) +int +afr_add_inode_lockee(afr_local_t *local, int child_count) { - int i = 0; + int ret = -ENOMEM; + afr_internal_lock_t *int_lock = &local->internal_lock; + afr_lockee_t *lockee = &int_lock->lockee[int_lock->lockee_count]; - for (i = 0; i < int_lock->lockee_count; i++) { - loc_wipe(&int_lock->lockee[i].loc); - if (int_lock->lockee[i].basename) - GF_FREE(int_lock->lockee[i].basename); - if (int_lock->lockee[i].locked_nodes) - GF_FREE(int_lock->lockee[i].locked_nodes); + if (local->fd) { + lockee->fd = fd_ref(local->fd); + } else { + loc_copy(&lockee->loc, &local->loc); } - return; + lockee->locked_count = 0; + lockee->locked_nodes = GF_CALLOC(child_count, sizeof(*lockee->locked_nodes), + gf_afr_mt_afr_node_character); + + if (!lockee->locked_nodes) + goto out; + + ret = 0; + int_lock->lockee_count++; +out: + if (ret) { + afr_lockee_cleanup(lockee); + } + return ret; } static int -initialize_entrylk_variables(call_frame_t *frame, xlator_t *this) +initialize_internal_lock_variables(call_frame_t *frame, xlator_t *this) { afr_local_t *local = NULL; afr_internal_lock_t *int_lock = NULL; @@ -152,9 +181,10 @@ initialize_entrylk_variables(call_frame_t *frame, xlator_t *this) local = frame->local; int_lock = &local->internal_lock; - int_lock->entrylk_lock_count = 0; + int_lock->lock_count = 0; int_lock->lock_op_ret = -1; int_lock->lock_op_errno = 0; + int_lock->lk_attempted_count = 0; for (i = 0; i < AFR_LOCKEE_COUNT_MAX; i++) { if (!int_lock->lockee[i].locked_nodes) @@ -167,28 +197,6 @@ initialize_entrylk_variables(call_frame_t *frame, xlator_t *this) return 0; } -static int -initialize_inodelk_variables(call_frame_t *frame, xlator_t *this) -{ - afr_local_t *local = NULL; - afr_internal_lock_t *int_lock = NULL; - afr_private_t *priv = NULL; - - priv = this->private; - local = frame->local; - int_lock = &local->internal_lock; - - int_lock->lock_count = 0; - int_lock->lk_attempted_count = 0; - int_lock->lock_op_ret = -1; - int_lock->lock_op_errno = 0; - - memset(int_lock->locked_nodes, 0, - sizeof(*int_lock->locked_nodes) * priv->child_count); - - return 0; -} - int afr_lockee_locked_nodes_count(afr_internal_lock_t *int_lock) { @@ -216,19 +224,74 @@ afr_locked_nodes_count(unsigned char *locked_nodes, int child_count) return call_count; } -/* FIXME: What if UNLOCK fails */ +static void +afr_log_locks_failure(call_frame_t *frame, char *where, char *what, + int op_errno) +{ + xlator_t *this = frame->this; + gf_lkowner_t *lk_owner = &frame->root->lk_owner; + afr_local_t *local = frame->local; + const char *fop = NULL; + char *gfid = NULL; + const char *name = NULL; + + fop = gf_fop_list[local->op]; + + switch (local->transaction.type) { + case AFR_ENTRY_RENAME_TRANSACTION: + case AFR_ENTRY_TRANSACTION: + switch (local->op) { + case GF_FOP_LINK: + gfid = uuid_utoa(local->newloc.pargfid); + name = local->newloc.name; + break; + default: + gfid = uuid_utoa(local->loc.pargfid); + name = local->loc.name; + break; + } + gf_msg(this->name, GF_LOG_WARNING, op_errno, + AFR_MSG_INTERNAL_LKS_FAILED, + "Unable to do entry %s with lk-owner:%s on %s " + "while attempting %s on {pgfid:%s, name:%s}.", + what, lkowner_utoa(lk_owner), where, fop, gfid, name); + break; + case AFR_DATA_TRANSACTION: + case AFR_METADATA_TRANSACTION: + gfid = uuid_utoa(local->inode->gfid); + gf_msg(this->name, GF_LOG_WARNING, op_errno, + AFR_MSG_INTERNAL_LKS_FAILED, + "Unable to do inode %s with lk-owner:%s on %s " + "while attempting %s on gfid:%s.", + what, lkowner_utoa(lk_owner), where, fop, gfid); + break; + } +} + static int32_t afr_unlock_common_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) { afr_local_t *local = NULL; + afr_private_t *priv = NULL; afr_internal_lock_t *int_lock = NULL; + int lockee_num = 0; int call_count = 0; + int child_index = 0; int ret = 0; local = frame->local; int_lock = &local->internal_lock; + priv = this->private; + lockee_num = (int)((long)cookie) / priv->child_count; + child_index = (int)((long)cookie) % priv->child_count; + + if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) { + afr_log_locks_failure(frame, priv->children[child_index]->name, + "unlock", op_errno); + } + int_lock->lockee[lockee_num].locked_nodes[child_index] &= LOCKED_NO; if (local->transaction.type == AFR_DATA_TRANSACTION && op_ret != 1) ret = afr_write_subvol_reset(frame, this); @@ -239,7 +302,6 @@ afr_unlock_common_cbk(call_frame_t *frame, void *cookie, xlator_t *this, UNLOCK(&frame->lock); if (call_count == 0) { - gf_msg_trace(this->name, 0, "All internal locks unlocked"); int_lock->lock_cbk(frame, this); } @@ -247,165 +309,108 @@ afr_unlock_common_cbk(call_frame_t *frame, void *cookie, xlator_t *this, } void -afr_update_uninodelk(afr_local_t *local, afr_internal_lock_t *int_lock, - int32_t child_index) -{ - int_lock->locked_nodes[child_index] &= LOCKED_NO; -} - -static int32_t -afr_unlock_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *xdata) -{ - afr_local_t *local = NULL; - afr_internal_lock_t *int_lock = NULL; - int32_t child_index = (long)cookie; - afr_private_t *priv = NULL; - - local = frame->local; - int_lock = &local->internal_lock; - - priv = this->private; - - if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) { - gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_UNLOCK_FAIL, - "path=%s gfid=%s: unlock failed on subvolume %s " - "with lock owner %s", - local->loc.path, loc_gfid_utoa(&(local->loc)), - priv->children[child_index]->name, - lkowner_utoa(&frame->root->lk_owner)); - } - - afr_update_uninodelk(local, int_lock, child_index); - - afr_unlock_common_cbk(frame, cookie, this, op_ret, op_errno, xdata); - - return 0; -} - -static int -afr_unlock_inodelk(call_frame_t *frame, xlator_t *this) +afr_internal_lock_wind(call_frame_t *frame, + int32_t (*cbk)(call_frame_t *, void *, xlator_t *, + int32_t, int32_t, dict_t *), + void *cookie, int child, int lockee_num, + gf_boolean_t blocking, gf_boolean_t unlock) { - afr_internal_lock_t *int_lock = NULL; - afr_local_t *local = NULL; - afr_private_t *priv = NULL; + afr_local_t *local = frame->local; + xlator_t *this = frame->this; + afr_private_t *priv = this->private; + afr_internal_lock_t *int_lock = &local->internal_lock; + entrylk_cmd cmd = ENTRYLK_LOCK_NB; + int32_t cmd1 = F_SETLK; struct gf_flock flock = { 0, }; - int call_count = 0; - int i = 0; - - local = frame->local; - int_lock = &local->internal_lock; - priv = this->private; - - flock.l_start = int_lock->flock.l_start; - flock.l_len = int_lock->flock.l_len; - flock.l_type = F_UNLCK; - - call_count = afr_locked_nodes_count(int_lock->locked_nodes, - priv->child_count); - - int_lock->lk_call_count = call_count; - - if (!call_count) { - GF_ASSERT(!local->transaction.do_eager_unlock); - gf_msg_trace(this->name, 0, "No internal locks unlocked"); - - int_lock->lock_cbk(frame, this); - goto out; - } - for (i = 0; i < priv->child_count; i++) { - if ((int_lock->locked_nodes[i] & LOCKED_YES) != LOCKED_YES) - continue; - - if (local->fd) { - STACK_WIND_COOKIE( - frame, afr_unlock_inodelk_cbk, (void *)(long)i, - priv->children[i], priv->children[i]->fops->finodelk, - int_lock->domain, local->fd, F_SETLK, &flock, NULL); - } else { - STACK_WIND_COOKIE( - frame, afr_unlock_inodelk_cbk, (void *)(long)i, - priv->children[i], priv->children[i]->fops->inodelk, - int_lock->domain, &local->loc, F_SETLK, &flock, NULL); - } + switch (local->transaction.type) { + case AFR_ENTRY_TRANSACTION: + case AFR_ENTRY_RENAME_TRANSACTION: + if (unlock) { + cmd = ENTRYLK_UNLOCK; + } else if (blocking) { /*Doesn't make sense to have blocking + unlock*/ + cmd = ENTRYLK_LOCK; + } - if (!--call_count) + if (local->fd) { + STACK_WIND_COOKIE(frame, cbk, cookie, priv->children[child], + priv->children[child]->fops->fentrylk, + int_lock->domain, + int_lock->lockee[lockee_num].fd, + int_lock->lockee[lockee_num].basename, cmd, + ENTRYLK_WRLCK, NULL); + } else { + STACK_WIND_COOKIE(frame, cbk, cookie, priv->children[child], + priv->children[child]->fops->entrylk, + int_lock->domain, + &int_lock->lockee[lockee_num].loc, + int_lock->lockee[lockee_num].basename, cmd, + ENTRYLK_WRLCK, NULL); + } break; - } -out: - return 0; -} - -static int32_t -afr_unlock_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *xdata) -{ - afr_local_t *local = NULL; - afr_private_t *priv = NULL; - afr_internal_lock_t *int_lock = NULL; - int32_t child_index = 0; - int lockee_no = 0; - - priv = this->private; - lockee_no = (int)((long)cookie) / priv->child_count; - child_index = (int)((long)cookie) % priv->child_count; - local = frame->local; - int_lock = &local->internal_lock; + case AFR_DATA_TRANSACTION: + case AFR_METADATA_TRANSACTION: + flock = int_lock->lockee[lockee_num].flock; + if (unlock) { + flock.l_type = F_UNLCK; + } else if (blocking) { /*Doesn't make sense to have blocking + unlock*/ + cmd1 = F_SETLKW; + } - if (op_ret < 0) { - gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_ENTRY_UNLOCK_FAIL, - "%s: unlock failed on %s", local->loc.path, - priv->children[child_index]->name); + if (local->fd) { + STACK_WIND_COOKIE( + frame, cbk, cookie, priv->children[child], + priv->children[child]->fops->finodelk, int_lock->domain, + int_lock->lockee[lockee_num].fd, cmd1, &flock, NULL); + } else { + STACK_WIND_COOKIE( + frame, cbk, cookie, priv->children[child], + priv->children[child]->fops->inodelk, int_lock->domain, + &int_lock->lockee[lockee_num].loc, cmd1, &flock, NULL); + } + break; } - - int_lock->lockee[lockee_no].locked_nodes[child_index] &= LOCKED_NO; - afr_unlock_common_cbk(frame, cookie, this, op_ret, op_errno, NULL); - - return 0; } static int -afr_unlock_entrylk(call_frame_t *frame, xlator_t *this) +afr_unlock_now(call_frame_t *frame, xlator_t *this) { afr_internal_lock_t *int_lock = NULL; afr_local_t *local = NULL; afr_private_t *priv = NULL; int call_count = 0; - int index = 0; - int lockee_no = 0; - int copies = 0; + int child_index = 0; + int lockee_num = 0; int i = -1; local = frame->local; int_lock = &local->internal_lock; priv = this->private; - copies = priv->child_count; call_count = afr_lockee_locked_nodes_count(int_lock); int_lock->lk_call_count = call_count; if (!call_count) { + GF_ASSERT(!local->transaction.do_eager_unlock); gf_msg_trace(this->name, 0, "No internal locks unlocked"); int_lock->lock_cbk(frame, this); goto out; } for (i = 0; i < int_lock->lockee_count * priv->child_count; i++) { - lockee_no = i / copies; - index = i % copies; - if (int_lock->lockee[lockee_no].locked_nodes[index] & LOCKED_YES) { - STACK_WIND_COOKIE( - frame, afr_unlock_entrylk_cbk, (void *)(long)i, - priv->children[index], priv->children[index]->fops->entrylk, - int_lock->domain, &int_lock->lockee[lockee_no].loc, - int_lock->lockee[lockee_no].basename, ENTRYLK_UNLOCK, - ENTRYLK_WRLCK, NULL); - + lockee_num = i / priv->child_count; + child_index = i % priv->child_count; + if (int_lock->lockee[lockee_num].locked_nodes[child_index] & + LOCKED_YES) { + afr_internal_lock_wind(frame, afr_unlock_common_cbk, + (void *)(long)i, child_index, lockee_num, + _gf_false, _gf_true); if (!--call_count) break; } @@ -415,18 +420,6 @@ out: return 0; } -int32_t -afr_unlock_now(call_frame_t *frame, xlator_t *this) -{ - afr_local_t *local = frame->local; - - if (afr_is_inodelk_transaction(local->transaction.type)) - afr_unlock_inodelk(frame, this); - else - afr_unlock_entrylk(frame, this); - return 0; -} - static int32_t afr_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, dict_t *xdata) @@ -436,14 +429,14 @@ afr_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, afr_private_t *priv = NULL; int cky = (long)cookie; int child_index = 0; - int lockee_no = 0; + int lockee_num = 0; priv = this->private; local = frame->local; int_lock = &local->internal_lock; child_index = ((int)cky) % priv->child_count; - lockee_no = ((int)cky) / priv->child_count; + lockee_num = ((int)cky) / priv->child_count; LOCK(&frame->lock); { @@ -470,23 +463,16 @@ afr_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, afr_unlock_now(frame, this); } else { if (op_ret == 0) { - if (local->transaction.type == AFR_ENTRY_TRANSACTION || - local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) { - int_lock->lockee[lockee_no] - .locked_nodes[child_index] |= LOCKED_YES; - int_lock->lockee[lockee_no].locked_count++; - int_lock->entrylk_lock_count++; - } else { - int_lock->locked_nodes[child_index] |= LOCKED_YES; - int_lock->lock_count++; - - if (local->transaction.type == AFR_DATA_TRANSACTION) { - LOCK(&local->inode->lock); - { - local->inode_ctx->lock_count++; - } - UNLOCK(&local->inode->lock); + int_lock->lockee[lockee_num] + .locked_nodes[child_index] |= LOCKED_YES; + int_lock->lockee[lockee_num].locked_count++; + int_lock->lock_count++; + if (local->transaction.type == AFR_DATA_TRANSACTION) { + LOCK(&local->inode->lock); + { + local->inode_ctx->lock_count++; } + UNLOCK(&local->inode->lock); } } afr_lock_blocking(frame, this, cky + 1); @@ -495,30 +481,6 @@ afr_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, return 0; } -static int32_t -afr_blocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *xdata) -{ - afr_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata); - return 0; -} - -static int32_t -afr_blocking_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *xdata) -{ - afr_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata); - return 0; -} - -static gf_boolean_t -afr_is_entrylk(afr_transaction_type trans_type) -{ - if (afr_is_inodelk_transaction(trans_type)) - return _gf_false; - return _gf_true; -} - static gf_boolean_t _is_lock_wind_needed(afr_local_t *local, int child_index) { @@ -528,40 +490,12 @@ _is_lock_wind_needed(afr_local_t *local, int child_index) return _gf_true; } -static void -afr_log_entry_locks_failure(xlator_t *this, afr_local_t *local, - afr_internal_lock_t *int_lock) -{ - const char *fop = NULL; - char *pargfid = NULL; - const char *name = NULL; - - fop = gf_fop_list[local->op]; - - switch (local->op) { - case GF_FOP_LINK: - pargfid = uuid_utoa(local->newloc.pargfid); - name = local->newloc.name; - break; - default: - pargfid = uuid_utoa(local->loc.pargfid); - name = local->loc.name; - break; - } - - gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_BLOCKING_LKS_FAILED, - "Unable to obtain sufficient blocking entry locks on at least " - "one child while attempting %s on {pgfid:%s, name:%s}.", - fop, pargfid, name); -} - static gf_boolean_t is_blocking_locks_count_sufficient(call_frame_t *frame, xlator_t *this) { afr_local_t *local = NULL; afr_private_t *priv = NULL; afr_internal_lock_t *int_lock = NULL; - gf_boolean_t is_entrylk = _gf_false; int child = 0; int nlockee = 0; int lockee_count = 0; @@ -571,42 +505,26 @@ is_blocking_locks_count_sufficient(call_frame_t *frame, xlator_t *this) priv = this->private; int_lock = &local->internal_lock; lockee_count = int_lock->lockee_count; - is_entrylk = afr_is_entrylk(local->transaction.type); - - if (!is_entrylk) { - if (int_lock->lock_count == 0) { - gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_BLOCKING_LKS_FAILED, - "Unable to obtain " - "blocking inode lock on even one child for " - "gfid:%s.", - uuid_utoa(local->inode->gfid)); - return _gf_false; - } else { - /*inodelk succeeded on at least one child. */ - return _gf_true; - } - } else { - if (int_lock->entrylk_lock_count == 0) { - afr_log_entry_locks_failure(this, local, int_lock); - return _gf_false; - } - /* For FOPS that take multiple sets of locks (mkdir, rename), - * there must be at least one brick on which the locks from - * all lock sets were successful. */ - for (child = 0; child < priv->child_count; child++) { - ret = _gf_true; - for (nlockee = 0; nlockee < lockee_count; nlockee++) { - if (!(int_lock->lockee[nlockee].locked_nodes[child] & - LOCKED_YES)) - ret = _gf_false; - } - if (ret) - return ret; + if (int_lock->lock_count == 0) { + afr_log_locks_failure(frame, "any subvolume", "lock", + int_lock->lock_op_errno); + return _gf_false; + } + /* For FOPS that take multiple sets of locks (mkdir, rename), + * there must be at least one brick on which the locks from + * all lock sets were successful. */ + for (child = 0; child < priv->child_count; child++) { + ret = _gf_true; + for (nlockee = 0; nlockee < lockee_count; nlockee++) { + if (!(int_lock->lockee[nlockee].locked_nodes[child] & LOCKED_YES)) + ret = _gf_false; } - if (!ret) - afr_log_entry_locks_failure(this, local, int_lock); + if (ret) + return ret; } + if (!ret) + afr_log_locks_failure(frame, "all", "lock", int_lock->lock_op_errno); return ret; } @@ -617,27 +535,16 @@ afr_lock_blocking(call_frame_t *frame, xlator_t *this, int cookie) afr_internal_lock_t *int_lock = NULL; afr_local_t *local = NULL; afr_private_t *priv = NULL; - struct gf_flock flock = { - 0, - }; uint64_t ctx = 0; int ret = 0; int child_index = 0; - int lockee_no = 0; - gf_boolean_t is_entrylk = _gf_false; + int lockee_num = 0; local = frame->local; int_lock = &local->internal_lock; priv = this->private; child_index = cookie % priv->child_count; - lockee_no = cookie / priv->child_count; - is_entrylk = afr_is_entrylk(local->transaction.type); - - if (!is_entrylk) { - flock.l_start = int_lock->flock.l_start; - flock.l_len = int_lock->flock.l_len; - flock.l_type = int_lock->flock.l_type; - } + lockee_num = cookie / priv->child_count; if (local->fd) { ret = fd_ctx_get(local->fd, this, &ctx); @@ -681,52 +588,8 @@ afr_lock_blocking(call_frame_t *frame, xlator_t *this, int cookie) return 0; } - switch (local->transaction.type) { - case AFR_DATA_TRANSACTION: - case AFR_METADATA_TRANSACTION: - - if (local->fd) { - STACK_WIND_COOKIE( - frame, afr_blocking_inodelk_cbk, (void *)(long)child_index, - priv->children[child_index], - priv->children[child_index]->fops->finodelk, - int_lock->domain, local->fd, F_SETLKW, &flock, NULL); - - } else { - STACK_WIND_COOKIE( - frame, afr_blocking_inodelk_cbk, (void *)(long)child_index, - priv->children[child_index], - priv->children[child_index]->fops->inodelk, - int_lock->domain, &local->loc, F_SETLKW, &flock, NULL); - } - - break; - - case AFR_ENTRY_RENAME_TRANSACTION: - case AFR_ENTRY_TRANSACTION: - /*Accounting for child_index increments on 'down' - *and 'fd-less' children */ - - if (local->fd) { - STACK_WIND_COOKIE(frame, afr_blocking_entrylk_cbk, - (void *)(long)cookie, - priv->children[child_index], - priv->children[child_index]->fops->fentrylk, - int_lock->domain, local->fd, - int_lock->lockee[lockee_no].basename, - ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL); - } else { - STACK_WIND_COOKIE( - frame, afr_blocking_entrylk_cbk, (void *)(long)cookie, - priv->children[child_index], - priv->children[child_index]->fops->entrylk, - int_lock->domain, &int_lock->lockee[lockee_no].loc, - int_lock->lockee[lockee_no].basename, ENTRYLK_LOCK, - ENTRYLK_WRLCK, NULL); - } - - break; - } + afr_internal_lock_wind(frame, afr_lock_cbk, (void *)(long)cookie, + child_index, lockee_num, _gf_true, _gf_false); return 0; } @@ -743,20 +606,10 @@ afr_blocking_lock(call_frame_t *frame, xlator_t *this) local = frame->local; int_lock = &local->internal_lock; - switch (local->transaction.type) { - case AFR_DATA_TRANSACTION: - case AFR_METADATA_TRANSACTION: - initialize_inodelk_variables(frame, this); - break; - - case AFR_ENTRY_RENAME_TRANSACTION: - case AFR_ENTRY_TRANSACTION: - up_count = AFR_COUNT(local->child_up, priv->child_count); - int_lock->lk_call_count = int_lock->lk_expected_count = - (int_lock->lockee_count * up_count); - initialize_entrylk_variables(frame, this); - break; - } + up_count = AFR_COUNT(local->child_up, priv->child_count); + int_lock->lk_call_count = int_lock->lk_expected_count = + (int_lock->lockee_count * up_count); + initialize_internal_lock_variables(frame, this); afr_lock_blocking(frame, this, 0); @@ -764,171 +617,20 @@ afr_blocking_lock(call_frame_t *frame, xlator_t *this) } static int32_t -afr_nonblocking_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *xdata) +afr_nb_internal_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, dict_t *xdata) { afr_internal_lock_t *int_lock = NULL; afr_local_t *local = NULL; int call_count = 0; - int child_index = (long)cookie; - int copies = 0; - int index = 0; - int lockee_no = 0; - afr_private_t *priv = NULL; - - priv = this->private; - - copies = priv->child_count; - index = child_index % copies; - lockee_no = child_index / copies; - - local = frame->local; - int_lock = &local->internal_lock; - - LOCK(&frame->lock); - { - if (op_ret < 0) { - if (op_errno == ENOSYS) { - /* return ENOTSUP */ - gf_msg(this->name, GF_LOG_ERROR, ENOSYS, - AFR_MSG_LOCK_XLATOR_NOT_LOADED, - "subvolume does not support " - "locking. please load features/locks" - " xlator on server"); - local->op_ret = op_ret; - int_lock->lock_op_ret = op_ret; - - int_lock->lock_op_errno = op_errno; - local->op_errno = op_errno; - } - } else if (op_ret == 0) { - int_lock->lockee[lockee_no].locked_nodes[index] |= LOCKED_YES; - int_lock->lockee[lockee_no].locked_count++; - int_lock->entrylk_lock_count++; - } - - call_count = --int_lock->lk_call_count; - } - UNLOCK(&frame->lock); - - if (call_count == 0) { - gf_msg_trace(this->name, 0, "Last locking reply received"); - /* all locks successful. Proceed to call FOP */ - if (int_lock->entrylk_lock_count == int_lock->lk_expected_count) { - gf_msg_trace(this->name, 0, "All servers locked. Calling the cbk"); - int_lock->lock_op_ret = 0; - int_lock->lock_cbk(frame, this); - } - /* Not all locks were successful. Unlock and try locking - again, this time with serially blocking locks */ - else { - gf_msg_trace(this->name, 0, - "%d servers locked. Trying again " - "with blocking calls", - int_lock->lock_count); - - afr_unlock_now(frame, this); - } - } - - return 0; -} - -int -afr_nonblocking_entrylk(call_frame_t *frame, xlator_t *this) -{ - afr_internal_lock_t *int_lock = NULL; - afr_local_t *local = NULL; + int child_index = 0; + int lockee_num = 0; afr_private_t *priv = NULL; - afr_fd_ctx_t *fd_ctx = NULL; - int copies = 0; - int index = 0; - int lockee_no = 0; - int32_t call_count = 0; - int i = 0; - local = frame->local; - int_lock = &local->internal_lock; priv = this->private; - copies = priv->child_count; - initialize_entrylk_variables(frame, this); - - if (local->fd) { - fd_ctx = afr_fd_ctx_get(local->fd, this); - if (!fd_ctx) { - gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_FD_CTX_GET_FAILED, - "unable to get fd ctx for fd=%p", local->fd); - - local->op_ret = -1; - int_lock->lock_op_ret = -1; - local->op_errno = EINVAL; - int_lock->lock_op_errno = EINVAL; - - afr_unlock_now(frame, this); - return -1; - } - - call_count = int_lock->lockee_count * internal_lock_count(frame, this); - int_lock->lk_call_count = call_count; - int_lock->lk_expected_count = call_count; - - if (!call_count) { - gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_INFO_COMMON, - "fd not open on any subvolumes. aborting."); - afr_unlock_now(frame, this); - goto out; - } - - /* Send non-blocking entrylk calls only on up children - and where the fd has been opened */ - for (i = 0; i < int_lock->lockee_count * priv->child_count; i++) { - index = i % copies; - lockee_no = i / copies; - if (local->child_up[index]) { - STACK_WIND_COOKIE(frame, afr_nonblocking_entrylk_cbk, - (void *)(long)i, priv->children[index], - priv->children[index]->fops->fentrylk, - this->name, local->fd, - int_lock->lockee[lockee_no].basename, - ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL); - if (!--call_count) - break; - } - } - } else { - call_count = int_lock->lockee_count * internal_lock_count(frame, this); - int_lock->lk_call_count = call_count; - int_lock->lk_expected_count = call_count; - - for (i = 0; i < int_lock->lockee_count * priv->child_count; i++) { - index = i % copies; - lockee_no = i / copies; - if (local->child_up[index]) { - STACK_WIND_COOKIE(frame, afr_nonblocking_entrylk_cbk, - (void *)(long)i, priv->children[index], - priv->children[index]->fops->entrylk, - this->name, &int_lock->lockee[lockee_no].loc, - int_lock->lockee[lockee_no].basename, - ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL); - - if (!--call_count) - break; - } - } - } -out: - return 0; -} - -int32_t -afr_nonblocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *xdata) -{ - afr_internal_lock_t *int_lock = NULL; - afr_local_t *local = NULL; - int call_count = 0; - int child_index = (long)cookie; + child_index = ((long)cookie) % priv->child_count; + lockee_num = ((long)cookie) / priv->child_count; local = frame->local; int_lock = &local->internal_lock; @@ -953,11 +655,14 @@ afr_nonblocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, " xlator on server"); local->op_ret = op_ret; int_lock->lock_op_ret = op_ret; + int_lock->lock_op_errno = op_errno; local->op_errno = op_errno; } - } else { - int_lock->locked_nodes[child_index] |= LOCKED_YES; + } else if (op_ret == 0) { + int_lock->lockee[lockee_num] + .locked_nodes[child_index] |= LOCKED_YES; + int_lock->lockee[lockee_num].locked_count++; int_lock->lock_count++; } @@ -966,7 +671,7 @@ afr_nonblocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, UNLOCK(&frame->lock); if (call_count == 0) { - gf_msg_trace(this->name, 0, "Last inode locking reply received"); + gf_msg_trace(this->name, 0, "Last locking reply received"); /* all locks successful. Proceed to call FOP */ if (int_lock->lock_count == int_lock->lk_expected_count) { gf_msg_trace(this->name, 0, "All servers locked. Calling the cbk"); @@ -977,8 +682,8 @@ afr_nonblocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, again, this time with serially blocking locks */ else { gf_msg_trace(this->name, 0, - "%d servers locked. " - "Trying again with blocking calls", + "%d servers locked. Trying again " + "with blocking calls", int_lock->lock_count); afr_unlock_now(frame, this); @@ -989,12 +694,14 @@ afr_nonblocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, } int -afr_nonblocking_inodelk(call_frame_t *frame, xlator_t *this) +afr_lock_nonblocking(call_frame_t *frame, xlator_t *this) { afr_internal_lock_t *int_lock = NULL; afr_local_t *local = NULL; afr_private_t *priv = NULL; afr_fd_ctx_t *fd_ctx = NULL; + int child = 0; + int lockee_num = 0; int32_t call_count = 0; int i = 0; int ret = 0; @@ -1003,7 +710,7 @@ afr_nonblocking_inodelk(call_frame_t *frame, xlator_t *this) int_lock = &local->internal_lock; priv = this->private; - initialize_inodelk_variables(frame, this); + initialize_internal_lock_variables(frame, this); if (local->fd) { fd_ctx = afr_fd_ctx_get(local->fd, this); @@ -1022,36 +729,29 @@ afr_nonblocking_inodelk(call_frame_t *frame, xlator_t *this) } } - call_count = internal_lock_count(frame, this); + call_count = int_lock->lockee_count * internal_lock_count(frame, this); int_lock->lk_call_count = call_count; int_lock->lk_expected_count = call_count; if (!call_count) { - gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SUBVOLS_DOWN, - "All bricks are down, aborting."); + gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_INFO_COMMON, + "fd not open on any subvolumes. aborting."); afr_unlock_now(frame, this); goto out; } - /* Send non-blocking inodelk calls only on up children + /* Send non-blocking lock calls only on up children and where the fd has been opened */ - for (i = 0; i < priv->child_count; i++) { - if (!local->child_up[i]) - continue; - - if (local->fd) { - STACK_WIND_COOKIE( - frame, afr_nonblocking_inodelk_cbk, (void *)(long)i, - priv->children[i], priv->children[i]->fops->finodelk, - int_lock->domain, local->fd, F_SETLK, &int_lock->flock, NULL); - } else { - STACK_WIND_COOKIE( - frame, afr_nonblocking_inodelk_cbk, (void *)(long)i, - priv->children[i], priv->children[i]->fops->inodelk, - int_lock->domain, &local->loc, F_SETLK, &int_lock->flock, NULL); + for (i = 0; i < int_lock->lockee_count * priv->child_count; i++) { + child = i % priv->child_count; + lockee_num = i / priv->child_count; + if (local->child_up[child]) { + afr_internal_lock_wind(frame, afr_nb_internal_lock_cbk, + (void *)(long)i, child, lockee_num, + _gf_false, _gf_false); + if (!--call_count) + break; } - if (!--call_count) - break; } out: return ret; diff --git a/xlators/cluster/afr/src/afr-messages.h b/xlators/cluster/afr/src/afr-messages.h index baf1a6c291f..c9c99270e98 100644 --- a/xlators/cluster/afr/src/afr-messages.h +++ b/xlators/cluster/afr/src/afr-messages.h @@ -30,7 +30,7 @@ GLFS_MSGID(AFR, AFR_MSG_QUORUM_FAIL, AFR_MSG_QUORUM_MET, AFR_MSG_GFID_NULL, AFR_MSG_FD_CREATE_FAILED, AFR_MSG_DICT_SET_FAILED, AFR_MSG_EXPUNGING_FILE_OR_DIR, AFR_MSG_MIGRATION_IN_PROGRESS, AFR_MSG_CHILD_MISCONFIGURED, AFR_MSG_VOL_MISCONFIGURED, - AFR_MSG_BLOCKING_LKS_FAILED, AFR_MSG_INVALID_FD, AFR_MSG_LOCK_INFO, + AFR_MSG_INTERNAL_LKS_FAILED, AFR_MSG_INVALID_FD, AFR_MSG_LOCK_INFO, AFR_MSG_LOCK_XLATOR_NOT_LOADED, AFR_MSG_FD_CTX_GET_FAILED, AFR_MSG_INVALID_SUBVOL, AFR_MSG_PUMP_XLATOR_ERROR, AFR_MSG_SELF_HEAL_INFO, AFR_MSG_READ_SUBVOL_ERROR, diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c index 8d97ff53253..f0cb7ab34f3 100644 --- a/xlators/cluster/afr/src/afr-transaction.c +++ b/xlators/cluster/afr/src/afr-transaction.c @@ -797,21 +797,9 @@ afr_changelog_post_op_fail(call_frame_t *frame, xlator_t *this, int op_errno) unsigned char * afr_locked_nodes_get(afr_transaction_type type, afr_internal_lock_t *int_lock) { - unsigned char *locked_nodes = NULL; - switch (type) { - case AFR_DATA_TRANSACTION: - case AFR_METADATA_TRANSACTION: - locked_nodes = int_lock->locked_nodes; - break; - - case AFR_ENTRY_TRANSACTION: - case AFR_ENTRY_RENAME_TRANSACTION: - /*Because same set of subvols participate in all lockee - * entities*/ - locked_nodes = int_lock->lockee[0].locked_nodes; - break; - } - return locked_nodes; + /*Because same set of subvols participate in all lockee + * entities*/ + return int_lock->lockee[0].locked_nodes; } int @@ -2057,31 +2045,7 @@ err: } int -afr_post_nonblocking_inodelk_cbk(call_frame_t *frame, xlator_t *this) -{ - afr_internal_lock_t *int_lock = NULL; - afr_local_t *local = NULL; - - local = frame->local; - int_lock = &local->internal_lock; - - /* Initiate blocking locks if non-blocking has failed */ - if (int_lock->lock_op_ret < 0) { - gf_msg_debug(this->name, 0, - "Non blocking inodelks failed. Proceeding to blocking"); - int_lock->lock_cbk = afr_internal_lock_finish; - afr_blocking_lock(frame, this); - } else { - gf_msg_debug(this->name, 0, - "Non blocking inodelks done. Proceeding to FOP"); - afr_internal_lock_finish(frame, this); - } - - return 0; -} - -int -afr_post_nonblocking_entrylk_cbk(call_frame_t *frame, xlator_t *this) +afr_post_nonblocking_lock_cbk(call_frame_t *frame, xlator_t *this) { afr_internal_lock_t *int_lock = NULL; afr_local_t *local = NULL; @@ -2092,12 +2056,12 @@ afr_post_nonblocking_entrylk_cbk(call_frame_t *frame, xlator_t *this) /* Initiate blocking locks if non-blocking has failed */ if (int_lock->lock_op_ret < 0) { gf_msg_debug(this->name, 0, - "Non blocking entrylks failed. Proceeding to blocking"); + "Non blocking locks failed. Proceeding to blocking"); int_lock->lock_cbk = afr_internal_lock_finish; afr_blocking_lock(frame, this); } else { gf_msg_debug(this->name, 0, - "Non blocking entrylks done. Proceeding to FOP"); + "Non blocking locks done. Proceeding to FOP"); afr_internal_lock_finish(frame, this); } @@ -2115,7 +2079,7 @@ afr_post_blocking_rename_cbk(call_frame_t *frame, xlator_t *this) int_lock = &local->internal_lock; if (int_lock->lock_op_ret < 0) { - gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_BLOCKING_LKS_FAILED, + gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_INTERNAL_LKS_FAILED, "Blocking entrylks failed."); afr_transaction_done(frame, this); @@ -2146,25 +2110,26 @@ afr_post_lower_unlock_cbk(call_frame_t *frame, xlator_t *this) } int -afr_set_transaction_flock(xlator_t *this, afr_local_t *local) +afr_set_transaction_flock(xlator_t *this, afr_local_t *local, + afr_lockee_t *lockee) { - afr_internal_lock_t *int_lock = NULL; afr_private_t *priv = NULL; + struct gf_flock *flock = NULL; - int_lock = &local->internal_lock; priv = this->private; + flock = &lockee->flock; if ((priv->arbiter_count || local->transaction.eager_lock_on || priv->full_lock) && local->transaction.type == AFR_DATA_TRANSACTION) { /*Lock entire file to avoid network split brains.*/ - int_lock->flock.l_len = 0; - int_lock->flock.l_start = 0; + flock->l_len = 0; + flock->l_start = 0; } else { - int_lock->flock.l_len = local->transaction.len; - int_lock->flock.l_start = local->transaction.start; + flock->l_len = local->transaction.len; + flock->l_start = local->transaction.start; } - int_lock->flock.l_type = F_WRLCK; + flock->l_type = F_WRLCK; return 0; } @@ -2174,26 +2139,21 @@ afr_lock(call_frame_t *frame, xlator_t *this) { afr_internal_lock_t *int_lock = NULL; afr_local_t *local = NULL; + int i = 0; local = frame->local; int_lock = &local->internal_lock; + int_lock->lock_cbk = afr_post_nonblocking_lock_cbk; int_lock->domain = this->name; switch (local->transaction.type) { case AFR_DATA_TRANSACTION: case AFR_METADATA_TRANSACTION: - afr_set_transaction_flock(this, local); - - int_lock->lock_cbk = afr_post_nonblocking_inodelk_cbk; - - afr_nonblocking_inodelk(frame, this); - break; - - case AFR_ENTRY_RENAME_TRANSACTION: + for (i = 0; i < int_lock->lockee_count; i++) { + afr_set_transaction_flock(this, local, &int_lock->lockee[i]); + } - int_lock->lock_cbk = afr_post_nonblocking_entrylk_cbk; - afr_nonblocking_entrylk(frame, this); break; case AFR_ENTRY_TRANSACTION: @@ -2202,11 +2162,11 @@ afr_lock(call_frame_t *frame, xlator_t *this) int_lock->lk_loc = &local->transaction.parent_loc; else GF_ASSERT(local->fd); - - int_lock->lock_cbk = afr_post_nonblocking_entrylk_cbk; - afr_nonblocking_entrylk(frame, this); + break; + case AFR_ENTRY_RENAME_TRANSACTION: break; } + afr_lock_nonblocking(frame, this); return 0; } @@ -2277,17 +2237,19 @@ afr_has_lock_conflict(afr_local_t *local, gf_boolean_t waitlist_check) /* }}} */ static void afr_copy_inodelk_vars(afr_internal_lock_t *dst, afr_internal_lock_t *src, - xlator_t *this) + xlator_t *this, int lockee_num) { afr_private_t *priv = this->private; + afr_lockee_t *sl = &src->lockee[lockee_num]; + afr_lockee_t *dl = &dst->lockee[lockee_num]; dst->domain = src->domain; - dst->flock.l_len = src->flock.l_len; - dst->flock.l_start = src->flock.l_start; - dst->flock.l_type = src->flock.l_type; - dst->lock_count = src->lock_count; - memcpy(dst->locked_nodes, src->locked_nodes, - priv->child_count * sizeof(*dst->locked_nodes)); + dl->flock.l_len = sl->flock.l_len; + dl->flock.l_start = sl->flock.l_start; + dl->flock.l_type = sl->flock.l_type; + dl->locked_count = sl->locked_count; + memcpy(dl->locked_nodes, sl->locked_nodes, + priv->child_count * sizeof(*dl->locked_nodes)); } void @@ -2308,7 +2270,7 @@ __afr_transaction_wake_shared(afr_local_t *local, struct list_head *shared) if (conflict && !list_empty(&lock->owners)) return; afr_copy_inodelk_vars(&each->internal_lock, &local->internal_lock, - each->transaction.frame->this); + each->transaction.frame->this, 0); list_move_tail(&each->transaction.wait_list, shared); list_add_tail(&each->transaction.owner_list, &lock->owners); } @@ -2785,7 +2747,7 @@ __afr_eager_lock_handle(afr_local_t *local, gf_boolean_t *take_lock, *timer_local = list_entry(lock->post_op.next, afr_local_t, transaction.owner_list); afr_copy_inodelk_vars(&local->internal_lock, - &(*timer_local)->internal_lock, this); + &(*timer_local)->internal_lock, this, 0); lock->delay_timer = NULL; *do_pre_op = _gf_true; list_add_tail(&local->transaction.owner_list, &lock->owners); @@ -2802,7 +2764,7 @@ __afr_eager_lock_handle(afr_local_t *local, gf_boolean_t *take_lock, owner_local = list_entry(lock->owners.next, afr_local_t, transaction.owner_list); afr_copy_inodelk_vars(&local->internal_lock, - &owner_local->internal_lock, this); + &owner_local->internal_lock, this, 0); *take_lock = _gf_false; *do_pre_op = _gf_true; } @@ -2872,6 +2834,62 @@ fail: } int +afr_transaction_lockee_init(call_frame_t *frame) +{ + afr_local_t *local = frame->local; + afr_internal_lock_t *int_lock = &local->internal_lock; + afr_private_t *priv = frame->this->private; + int ret = 0; + + switch (local->transaction.type) { + case AFR_DATA_TRANSACTION: + case AFR_METADATA_TRANSACTION: + ret = afr_add_inode_lockee(local, priv->child_count); + break; + + case AFR_ENTRY_TRANSACTION: + case AFR_ENTRY_RENAME_TRANSACTION: + ret = afr_add_entry_lockee(local, &local->transaction.parent_loc, + local->transaction.basename, + priv->child_count); + if (ret) { + goto out; + } + if (local->op == GF_FOP_RENAME) { + ret = afr_add_entry_lockee( + local, &local->transaction.new_parent_loc, + local->transaction.new_basename, priv->child_count); + if (ret) { + goto out; + } + + if (local->newloc.inode && + IA_ISDIR(local->newloc.inode->ia_type)) { + ret = afr_add_entry_lockee(local, &local->newloc, NULL, + priv->child_count); + if (ret) { + goto out; + } + } + } else if (local->op == GF_FOP_RMDIR) { + ret = afr_add_entry_lockee(local, &local->loc, NULL, + priv->child_count); + if (ret) { + goto out; + } + } + + if (int_lock->lockee_count > 1) { + qsort(int_lock->lockee, int_lock->lockee_count, + sizeof(*int_lock->lockee), afr_entry_lockee_cmp); + } + break; + } +out: + return ret; +} + +int afr_transaction(call_frame_t *frame, xlator_t *this, afr_transaction_type type) { afr_local_t *local = NULL; @@ -2904,6 +2922,10 @@ afr_transaction(call_frame_t *frame, xlator_t *this, afr_transaction_type type) if (ret < 0) goto out; + ret = afr_transaction_lockee_init(frame); + if (ret) + goto out; + if (type != AFR_METADATA_TRANSACTION) { goto txn_start; } diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h index 9c12d2d9ac0..cdcc6027fdb 100644 --- a/xlators/cluster/afr/src/afr.h +++ b/xlators/cluster/afr/src/afr.h @@ -288,12 +288,14 @@ afr_index_from_ia_type(ia_type_t type) } typedef struct { + struct gf_flock flock; loc_t loc; + fd_t *fd; char *basename; unsigned char *locked_nodes; int locked_count; -} afr_entry_lockee_t; +} afr_lockee_t; int afr_entry_lockee_cmp(const void *l1, const void *l2); @@ -302,20 +304,17 @@ typedef struct { loc_t *lk_loc; int lockee_count; - afr_entry_lockee_t lockee[AFR_LOCKEE_COUNT_MAX]; + afr_lockee_t lockee[AFR_LOCKEE_COUNT_MAX]; - struct gf_flock flock; const char *lk_basename; const char *lower_basename; const char *higher_basename; char lower_locked; char higher_locked; - unsigned char *locked_nodes; unsigned char *lower_locked_nodes; int32_t lock_count; - int32_t entrylk_lock_count; int32_t lk_call_count; int32_t lk_expected_count; @@ -980,11 +979,14 @@ int xattr_is_equal(dict_t *this, char *key1, data_t *value1, void *data); int -afr_init_entry_lockee(afr_entry_lockee_t *lockee, afr_local_t *local, - loc_t *loc, char *basename, int child_count); +afr_add_entry_lockee(afr_local_t *local, loc_t *loc, char *basename, + int child_count); + +int +afr_add_inode_lockee(afr_local_t *local, int child_count); void -afr_entry_lockee_cleanup(afr_internal_lock_t *int_lock); +afr_lockees_cleanup(afr_internal_lock_t *int_lock); int afr_attempt_lock_recovery(xlator_t *this, int32_t child_index); @@ -1002,10 +1004,7 @@ int32_t afr_unlock(call_frame_t *frame, xlator_t *this); int -afr_nonblocking_entrylk(call_frame_t *frame, xlator_t *this); - -int -afr_nonblocking_inodelk(call_frame_t *frame, xlator_t *this); +afr_lock_nonblocking(call_frame_t *frame, xlator_t *this); int afr_blocking_lock(call_frame_t *frame, xlator_t *this); @@ -1272,9 +1271,6 @@ afr_writev_copy_outvars(call_frame_t *src_frame, call_frame_t *dst_frame); void afr_update_uninodelk(afr_local_t *local, afr_internal_lock_t *int_lock, int32_t child_index); -int -afr_is_inodelk_transaction(afr_transaction_type type); - afr_fd_ctx_t * __afr_fd_ctx_get(fd_t *fd, xlator_t *this); |