diff options
author | Brian Foster <bfoster@redhat.com> | 2013-05-08 08:54:11 -0400 |
---|---|---|
committer | Anand Avati <avati@redhat.com> | 2013-06-13 14:37:21 -0700 |
commit | d1ccc4e400728d90f2ef7904661f53deb7199123 (patch) | |
tree | 1c491d1932795d85941861fa683431ee22608668 /xlators/cluster | |
parent | 328ea4b16a276b0e65ca719f60b82ce851dda848 (diff) |
gluster: add fallocate fop support
Implement support for the fallocate file operation. fallocate
allocates blocks for a particular inode such that future writes
to the associated region of the file are guaranteed not to fail
with ENOSPC.
This patch adds fallocate support to the following areas:
- libglusterfs
- mount/fuse
- io-stats
- performance/md-cache,open-behind
- quota
- cluster/afr,dht,stripe
- rpc/xdr
- protocol/client,server
- io-threads
- marker
- storage/posix
- libgfapi
BUG: 949242
Change-Id: Ice8e61351f9d6115c5df68768bc844abbf0ce8bd
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-on: http://review.gluster.org/4969
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
Diffstat (limited to 'xlators/cluster')
-rw-r--r-- | xlators/cluster/afr/src/afr-common.c | 244 | ||||
-rw-r--r-- | xlators/cluster/afr/src/afr.c | 1 | ||||
-rw-r--r-- | xlators/cluster/afr/src/afr.h | 8 | ||||
-rw-r--r-- | xlators/cluster/dht/src/dht-common.h | 2 | ||||
-rw-r--r-- | xlators/cluster/dht/src/dht-inode-write.c | 140 | ||||
-rw-r--r-- | xlators/cluster/dht/src/dht.c | 1 | ||||
-rw-r--r-- | xlators/cluster/stripe/src/stripe.c | 168 |
7 files changed, 564 insertions, 0 deletions
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c index a9acb409490..830ecd99386 100644 --- a/xlators/cluster/afr/src/afr-common.c +++ b/xlators/cluster/afr/src/afr-common.c @@ -2821,6 +2821,250 @@ out: /* }}} */ +/* {{{ fallocate */ + +static int +afr_fallocate_unwind (call_frame_t *frame, xlator_t *this) +{ + afr_local_t * local = NULL; + call_frame_t *main_frame = NULL; + + local = frame->local; + + LOCK (&frame->lock); + { + if (local->transaction.main_frame) + main_frame = local->transaction.main_frame; + local->transaction.main_frame = NULL; + } + UNLOCK (&frame->lock); + + if (main_frame) { + AFR_STACK_UNWIND (fallocate, main_frame, local->op_ret, + local->op_errno, + &local->cont.fallocate.prebuf, + &local->cont.fallocate.postbuf, + NULL); + } + return 0; +} + +static int +afr_fallocate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *prebuf, + struct iatt *postbuf, dict_t *xdata) +{ + afr_local_t * local = NULL; + afr_private_t * priv = NULL; + int child_index = (long) cookie; + int call_count = -1; + int need_unwind = 0; + int read_child = 0; + + local = frame->local; + priv = this->private; + + read_child = afr_inode_get_read_ctx (this, local->fd->inode, NULL); + + LOCK (&frame->lock); + { + if (child_index == read_child) { + local->read_child_returned = _gf_true; + } + + if (afr_fop_failed (op_ret, op_errno)) + afr_transaction_fop_failed (frame, this, child_index); + + if (op_ret != -1) { + if (local->success_count == 0) { + local->op_ret = op_ret; + local->cont.fallocate.prebuf = *prebuf; + local->cont.fallocate.postbuf = *postbuf; + } + + if (child_index == read_child) { + local->cont.fallocate.prebuf = *prebuf; + local->cont.fallocate.postbuf = *postbuf; + } + + local->success_count++; + + if ((local->success_count >= priv->wait_count) + && local->read_child_returned) { + need_unwind = 1; + } + } + local->op_errno = op_errno; + } + UNLOCK (&frame->lock); + + if (need_unwind) + local->transaction.unwind (frame, this); + + call_count = afr_frame_return (frame); + + if (call_count == 0) { + local->transaction.resume (frame, this); + } + + return 0; +} + +static int +afr_fallocate_wind (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + afr_private_t *priv = NULL; + int call_count = -1; + int i = 0; + + local = frame->local; + priv = this->private; + + call_count = afr_pre_op_done_children_count (local->transaction.pre_op, + priv->child_count); + + if (call_count == 0) { + local->transaction.resume (frame, this); + return 0; + } + + local->call_count = call_count; + + for (i = 0; i < priv->child_count; i++) { + if (local->transaction.pre_op[i]) { + STACK_WIND_COOKIE (frame, afr_fallocate_wind_cbk, + (void *) (long) i, + priv->children[i], + priv->children[i]->fops->fallocate, + local->fd, + local->cont.fallocate.mode, + local->cont.fallocate.offset, + local->cont.fallocate.len, + NULL); + + if (!--call_count) + break; + } + } + + return 0; +} + +static int +afr_fallocate_done (call_frame_t *frame, xlator_t *this) +{ + afr_local_t *local = NULL; + + local = frame->local; + + local->transaction.unwind (frame, this); + + AFR_STACK_DESTROY (frame); + + return 0; +} + +static int +afr_do_fallocate(call_frame_t *frame, xlator_t *this) +{ + call_frame_t * transaction_frame = NULL; + afr_local_t * local = NULL; + int op_ret = -1; + int op_errno = 0; + + local = frame->local; + + transaction_frame = copy_frame (frame); + if (!transaction_frame) { + goto out; + } + + transaction_frame->local = local; + frame->local = NULL; + + local->op = GF_FOP_FALLOCATE; + + local->transaction.fop = afr_fallocate_wind; + local->transaction.done = afr_fallocate_done; + local->transaction.unwind = afr_fallocate_unwind; + + local->transaction.main_frame = frame; + + local->transaction.start = local->cont.fallocate.offset; + local->transaction.len = 0; + + /* fallocate can modify the file size */ + op_ret = afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION); + if (op_ret < 0) { + op_errno = -op_ret; + goto out; + } + + op_ret = 0; +out: + if (op_ret < 0) { + if (transaction_frame) + AFR_STACK_DESTROY (transaction_frame); + AFR_STACK_UNWIND (fallocate, frame, op_ret, op_errno, NULL, + NULL, NULL); + } + + return 0; +} + +int +afr_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode, + off_t offset, size_t len, dict_t *xdata) +{ + afr_private_t * priv = NULL; + afr_local_t * local = NULL; + call_frame_t *transaction_frame = NULL; + int ret = -1; + int op_errno = 0; + + VALIDATE_OR_GOTO (frame, out); + VALIDATE_OR_GOTO (this, out); + VALIDATE_OR_GOTO (this->private, out); + + priv = this->private; + + if (afr_is_split_brain (this, fd->inode)) { + op_errno = EIO; + goto out; + } + QUORUM_CHECK(fallocate,out); + + AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out); + local = frame->local; + + ret = afr_local_init (local, priv, &op_errno); + if (ret < 0) + goto out; + + local->cont.fallocate.mode = mode; + local->cont.fallocate.offset = offset; + local->cont.fallocate.len = len; + + local->fd = fd_ref (fd); + + afr_open_fd_fix (fd, this); + + afr_do_fallocate (frame, this); + + ret = 0; +out: + if (ret < 0) { + if (transaction_frame) + AFR_STACK_DESTROY (transaction_frame); + AFR_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL); + } + + return 0; +} + +/* }}} */ + /* {{{ fsync */ int32_t diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c index bee10fd01dc..4d6c714801f 100644 --- a/xlators/cluster/afr/src/afr.c +++ b/xlators/cluster/afr/src/afr.c @@ -483,6 +483,7 @@ struct xlator_fops fops = { .finodelk = afr_finodelk, .entrylk = afr_entrylk, .fentrylk = afr_fentrylk, + .fallocate = afr_fallocate, /* inode read */ .access = afr_access, diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h index 387ed12ec2f..185f4f7134a 100644 --- a/xlators/cluster/afr/src/afr.h +++ b/xlators/cluster/afr/src/afr.h @@ -652,6 +652,14 @@ typedef struct _afr_local { dict_t *params; char *linkpath; } symlink; + + struct { + int32_t mode; + off_t offset; + size_t len; + struct iatt prebuf; + struct iatt postbuf; + } fallocate; } cont; struct { diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h index 9de861360c5..de35e843f47 100644 --- a/xlators/cluster/dht/src/dht-common.h +++ b/xlators/cluster/dht/src/dht-common.h @@ -691,6 +691,8 @@ int32_t dht_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf, int32_t valid, dict_t *xdata); int32_t dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf, int32_t valid, dict_t *xdata); +int32_t dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, + int32_t mode, off_t offset, size_t len, dict_t *xdata); int32_t dht_init (xlator_t *this); void dht_fini (xlator_t *this); diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c index b87d2f73284..56ed36ef0a9 100644 --- a/xlators/cluster/dht/src/dht-inode-write.c +++ b/xlators/cluster/dht/src/dht-inode-write.c @@ -19,6 +19,7 @@ int dht_writev2 (xlator_t *this, call_frame_t *frame, int ret); int dht_truncate2 (xlator_t *this, call_frame_t *frame, int ret); int dht_setattr2 (xlator_t *this, call_frame_t *frame, int ret); +int dht_fallocate2(xlator_t *this, call_frame_t *frame, int op_ret); int dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this, @@ -348,6 +349,145 @@ err: return 0; } + +int +dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int op_ret, int op_errno, struct iatt *prebuf, + struct iatt *postbuf, dict_t *xdata) +{ + dht_local_t *local = NULL; + call_frame_t *prev = NULL; + int ret = -1; + + GF_VALIDATE_OR_GOTO ("dht", frame, err); + GF_VALIDATE_OR_GOTO ("dht", this, out); + GF_VALIDATE_OR_GOTO ("dht", frame->local, out); + GF_VALIDATE_OR_GOTO ("dht", cookie, out); + + local = frame->local; + prev = cookie; + + if ((op_ret == -1) && (op_errno != ENOENT)) { + local->op_errno = op_errno; + local->op_ret = -1; + gf_log (this->name, GF_LOG_DEBUG, + "subvolume %s returned -1 (%s)", + prev->this->name, strerror (op_errno)); + + goto out; + } + + if (local->call_cnt != 1) { + if (local->stbuf.ia_blocks) { + dht_iatt_merge (this, postbuf, &local->stbuf, NULL); + dht_iatt_merge (this, prebuf, &local->prebuf, NULL); + } + goto out; + } + local->rebalance.target_op_fn = dht_fallocate2; + + /* Phase 2 of migration */ + if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2 (postbuf)) { + ret = dht_rebalance_complete_check (this, frame); + if (!ret) + return 0; + } + + /* Check if the rebalance phase1 is true */ + if (IS_DHT_MIGRATION_PHASE1 (postbuf)) { + dht_iatt_merge (this, &local->stbuf, postbuf, NULL); + dht_iatt_merge (this, &local->prebuf, prebuf, NULL); + ret = fd_ctx_get (local->fd, this, NULL); + if (!ret) { + dht_fallocate2 (this, frame, 0); + return 0; + } + ret = dht_rebalance_in_progress_check (this, frame); + if (!ret) + return 0; + } + +out: + DHT_STRIP_PHASE1_FLAGS (postbuf); + DHT_STRIP_PHASE1_FLAGS (prebuf); + DHT_STACK_UNWIND (fallocate, frame, op_ret, op_errno, + prebuf, postbuf, xdata); +err: + return 0; +} + +int +dht_fallocate2(xlator_t *this, call_frame_t *frame, int op_ret) +{ + dht_local_t *local = NULL; + xlator_t *subvol = NULL; + uint64_t tmp_subvol = 0; + int ret = -1; + + local = frame->local; + + if (local->fd) + ret = fd_ctx_get (local->fd, this, &tmp_subvol); + if (!ret) + subvol = (xlator_t *)(long)tmp_subvol; + + if (!subvol) + subvol = local->cached_subvol; + + local->call_cnt = 2; /* This is the second attempt */ + + STACK_WIND(frame, dht_fallocate_cbk, subvol, subvol->fops->fallocate, + local->fd, local->rebalance.flags, local->rebalance.offset, + local->rebalance.size, NULL); + + return 0; +} + +int +dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode, + off_t offset, size_t len, dict_t *xdata) +{ + xlator_t *subvol = NULL; + int op_errno = -1; + dht_local_t *local = NULL; + + VALIDATE_OR_GOTO (frame, err); + VALIDATE_OR_GOTO (this, err); + VALIDATE_OR_GOTO (fd, err); + + local = dht_local_init (frame, NULL, fd, GF_FOP_FALLOCATE); + if (!local) { + op_errno = ENOMEM; + goto err; + } + + local->rebalance.flags = mode; + local->rebalance.offset = offset; + local->rebalance.size = len; + + local->call_cnt = 1; + subvol = local->cached_subvol; + if (!subvol) { + gf_log (this->name, GF_LOG_DEBUG, + "no cached subvolume for fd=%p", fd); + op_errno = EINVAL; + goto err; + } + + STACK_WIND (frame, dht_fallocate_cbk, + subvol, subvol->fops->fallocate, + fd, mode, offset, len, xdata); + + return 0; + +err: + op_errno = (op_errno == -1) ? errno : op_errno; + DHT_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL); + + return 0; +} + + /* handle cases of migration here for 'setattr()' calls */ int dht_file_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c index 814f0a8eb6f..d3031e2032b 100644 --- a/xlators/cluster/dht/src/dht.c +++ b/xlators/cluster/dht/src/dht.c @@ -70,6 +70,7 @@ struct xlator_fops fops = { .fxattrop = dht_fxattrop, .setattr = dht_setattr, .fsetattr = dht_fsetattr, + .fallocate = dht_fallocate, }; struct xlator_dumpops dumpops = { diff --git a/xlators/cluster/stripe/src/stripe.c b/xlators/cluster/stripe/src/stripe.c index dadd3fec539..3c56464b8c1 100644 --- a/xlators/cluster/stripe/src/stripe.c +++ b/xlators/cluster/stripe/src/stripe.c @@ -3774,6 +3774,173 @@ err: int32_t +stripe_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, struct iatt *prebuf, + struct iatt *postbuf, dict_t *xdata) +{ + int32_t callcnt = 0; + stripe_local_t *local = NULL; + stripe_local_t *mlocal = NULL; + call_frame_t *prev = NULL; + call_frame_t *mframe = NULL; + + if (!this || !frame || !frame->local || !cookie) { + gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref"); + goto out; + } + + prev = cookie; + local = frame->local; + mframe = local->orig_frame; + mlocal = mframe->local; + + LOCK(&frame->lock); + { + callcnt = ++mlocal->call_count; + + if (op_ret == 0) { + mlocal->post_buf = *postbuf; + mlocal->pre_buf = *prebuf; + + mlocal->prebuf_blocks += prebuf->ia_blocks; + mlocal->postbuf_blocks += postbuf->ia_blocks; + + correct_file_size(prebuf, mlocal->fctx, prev); + correct_file_size(postbuf, mlocal->fctx, prev); + + if (mlocal->prebuf_size < prebuf->ia_size) + mlocal->prebuf_size = prebuf->ia_size; + if (mlocal->postbuf_size < postbuf->ia_size) + mlocal->postbuf_size = postbuf->ia_size; + } + + /* return the first failure */ + if (mlocal->op_ret == 0) { + mlocal->op_ret = op_ret; + mlocal->op_errno = op_errno; + } + } + UNLOCK (&frame->lock); + + if ((callcnt == mlocal->wind_count) && mlocal->unwind) { + mlocal->pre_buf.ia_size = mlocal->prebuf_size; + mlocal->pre_buf.ia_blocks = mlocal->prebuf_blocks; + mlocal->post_buf.ia_size = mlocal->postbuf_size; + mlocal->post_buf.ia_blocks = mlocal->postbuf_blocks; + + STRIPE_STACK_UNWIND (fallocate, mframe, mlocal->op_ret, + mlocal->op_errno, &mlocal->pre_buf, + &mlocal->post_buf, NULL); + } +out: + STRIPE_STACK_DESTROY(frame); + return 0; +} + +int32_t +stripe_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode, + off_t offset, size_t len, dict_t *xdata) +{ + stripe_local_t *local = NULL; + stripe_fd_ctx_t *fctx = NULL; + int32_t op_errno = 1; + int32_t idx = 0; + int32_t offset_offset = 0; + int32_t remaining_size = 0; + off_t fill_size = 0; + uint64_t stripe_size = 0; + uint64_t tmp_fctx = 0; + off_t dest_offset = 0; + call_frame_t *fframe = NULL; + stripe_local_t *flocal = NULL; + + VALIDATE_OR_GOTO (frame, err); + VALIDATE_OR_GOTO (this, err); + VALIDATE_OR_GOTO (fd, err); + VALIDATE_OR_GOTO (fd->inode, err); + + inode_ctx_get (fd->inode, this, &tmp_fctx); + if (!tmp_fctx) { + op_errno = EINVAL; + goto err; + } + fctx = (stripe_fd_ctx_t *)(long)tmp_fctx; + stripe_size = fctx->stripe_size; + + STRIPE_VALIDATE_FCTX (fctx, err); + + remaining_size = len; + + local = mem_get0 (this->local_pool); + if (!local) { + op_errno = ENOMEM; + goto err; + } + frame->local = local; + local->stripe_size = stripe_size; + local->fctx = fctx; + + if (!stripe_size) { + gf_log (this->name, GF_LOG_DEBUG, + "Wrong stripe size for the file"); + op_errno = EINVAL; + goto err; + } + + while (1) { + fframe = copy_frame(frame); + flocal = mem_get0(this->local_pool); + if (!flocal) { + op_errno = ENOMEM; + goto err; + } + flocal->orig_frame = frame; + fframe->local = flocal; + + /* send fallocate request to the associated child node */ + idx = (((offset + offset_offset) / + local->stripe_size) % fctx->stripe_count); + + fill_size = (local->stripe_size - + ((offset + offset_offset) % local->stripe_size)); + if (fill_size > remaining_size) + fill_size = remaining_size; + + remaining_size -= fill_size; + + local->wind_count++; + if (remaining_size == 0) + local->unwind = 1; + + dest_offset = offset + offset_offset; + if (fctx->stripe_coalesce) + dest_offset = coalesced_offset(dest_offset, + local->stripe_size, fctx->stripe_count); + + /* + * TODO: Create a separate handler for coalesce mode that sends a + * single fallocate per-child (since the ranges are linear). + */ + STACK_WIND(fframe, stripe_fallocate_cbk, fctx->xl_array[idx], + fctx->xl_array[idx]->fops->fallocate, fd, mode, + dest_offset, fill_size, xdata); + + offset_offset += fill_size; + if (remaining_size == 0) + break; + } + + return 0; +err: + if (fframe) + STRIPE_STACK_DESTROY(fframe); + + STRIPE_STACK_UNWIND (fallocate, frame, -1, op_errno, NULL, NULL, NULL); + return 0; +} + + +int32_t stripe_release (xlator_t *this, fd_t *fd) { return 0; @@ -5221,6 +5388,7 @@ struct xlator_fops fops = { .removexattr = stripe_removexattr, .fremovexattr = stripe_fremovexattr, .readdirp = stripe_readdirp, + .fallocate = stripe_fallocate, }; struct xlator_cbks cbks = { |