io-uring/btrfs: support async buffered writes
Subsystems (4)
v3 cover io-uring/btrfs: support async buffered writes
This patch series adds support for async buffered writes when using both btrfs and io-uring. Currently io-uring only supports buffered writes (for btrfs) in the slow path, by processing them in the io workers. With this patch series it is now possible to support buffered writes in the fast path. To be able to use the fast path, the required pages must be in the page cache, the required locks in btrfs can be granted immediately and no additional blocks need to be read form disk.
This patch series makes use of the changes that have been introduced by a previous patch series: "io-uring/xfs: support async buffered writes"
Performance results:
The new patch improves throughput by over two times (compared to the exiting behavior, where buffered writes are processed by an io-worker process) and also the latency is considerably reduced. Detailled results are part of the changelog of the first commit.
BTRFS changes:
-Add option for NOWAIT IOCB's to tell that searches do not wait on locks. This
adds the nowait option to btrfs_path.
-For NOWAIT buffered writes on PREALLOC or NOCOW extents tell can_nocow_extent()
that we don't want to wait on any locks or metadata IO.
-Support no_flush reservations for nowait buffered writes.
-Add btrfs_try_lock_ordered_range() function.
-Add nowait flag to btrfs_check_nocow_lock() to use it in write code path. -Add nowait parameter to prepare_pages() function.
-Plumb nowait through the write code path.
-Enable nowait buffered writes.
Testing:
This patch has been tested with xfstests, fsx, fio. xfstests shows no new diffs compared to running without the patch series.
Changes:
V3:
- Updated changelog of "btrfs: implement a nowait option for tree searches" to say -EAGAIN.
- Use bool return value in signature of btrfs_try_lock_ordered_range
- Renamed variable tmp to can_nocow in btrfs_buffered_write
- Fixed coding style in get_prepare_fgp_flags
- Set pages[i] to NULL in error code path of lock_and_cleanup_extent_if_need()
- Added const in definition of "bool nowait"
- Removed unlikely from btrfs_buffered_write
- Rephrased changelog for "btrfs: add assert to search functions" and used asserts instead of warnings
- Explained why enocded writes are not supported in the changelog
- Moved performance results to changelog of first commit
V2:
- Replace EWOULDBLOCK with EAGAIN. In Version 1 it was not used consistently
- Export function balance_dirty_pages_ratelimited_flags()
- Add asserts/warnings for search functions when nowait is set, but we don't expect that they are invoked with nowait set.
Josef Bacik (5):
btrfs: implement a nowait option for tree searches
btrfs: make can_nocow_extent nowait compatible
btrfs: add the ability to use NO_FLUSH for data reservations
btrfs: add btrfs_try_lock_ordered_range
btrfs: make btrfs_check_nocow_lock nowait compatible
Stefan Roesch (7):
mm: export balance_dirty_pages_ratelimited_flags()
btrfs: make prepare_pages nowait compatible
btrfs: make lock_and_cleanup_extent_if_need nowait compatible
btrfs: plumb NOWAIT through the write path
btrfs: make balance_dirty_pages nowait compatible
btrfs: assert nowait mode is not used for some btree search functions
btrfs: enable nowait async buffered writes
fs/btrfs/block-group.c | 2 +- fs/btrfs/ctree.c | 43 ++++++++++++- fs/btrfs/ctree.h | 8 ++- fs/btrfs/delalloc-space.c | 13 +++- fs/btrfs/delalloc-space.h | 3 +- fs/btrfs/extent-tree.c | 5 ++ fs/btrfs/file-item.c | 4 +- fs/btrfs/file.c | 124 ++++++++++++++++++++++++++++---------- fs/btrfs/inode.c | 22 ++++--- fs/btrfs/locking.c | 23 +++++++ fs/btrfs/locking.h | 1 + fs/btrfs/ordered-data.c | 28 +++++++++ fs/btrfs/ordered-data.h | 1 + fs/btrfs/relocation.c | 2 +- fs/btrfs/scrub.c | 4 +- fs/btrfs/space-info.c | 3 +- fs/btrfs/tree-log.c | 6 +- mm/page-writeback.c | 1 + 18 files changed, 234 insertions(+), 59 deletions(-)
base-commit: 80e78fcce86de0288793a0ef0f6acf37656ee4cf
Signature · 1 line
2.30.2
PATCH v3 1/12 mm: export balance_dirty_pages_ratelimited_flags()
Export the function balance_dirty_pages_ratelimited_flags(). It is now also called from btrfs.
--- mm/page-writeback.c | 1 + 1 file changed, 1 insertion(+)
mm/page-writeback.c +1 -0
@@ -1933,6 +1933,7 @@ int balance_dirty_pages_ratelimited_flags(struct address_space *mapping,wb_put (wb);return ret; } +EXPORT_SYMBOL_GPL (balance_dirty_pages_ratelimited_flags);/** * balance_dirty_pages_ratelimited - balance dirty memory state.
Signature · 1 line
2.30.2
PATCH v3 2/12 btrfs: implement a nowait option for tree searches
In reply to Stefan RoeschFrom: Josef Bacik <[email protected]>
For NOWAIT IOCB's we'll need a way to tell search to not wait on locks or anything. Accomplish this by adding a path->nowait flag that will use trylocks and skip reading of metadata, returning -EAGAIN in either of these cases. For now we only need this for reads, so only the read side is handled. Add an ASSERT() to catch anybody trying to use this for writes so they know they'll have to implement the write side.
--- fs/btrfs/ctree.c | 39 ++++++++++++++++++++++++++++++++++++--- fs/btrfs/ctree.h | 1 + fs/btrfs/locking.c | 23 +++++++++++++++++++++++ fs/btrfs/locking.h | 1 + 4 files changed, 61 insertions(+), 3 deletions(-)
fs/btrfs/ctree.c +36 -3
@@ -1447,6 +1447,11 @@ read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,return 0 ; } +if (p->nowait) { +free_extent_buffer (tmp); +return -EAGAIN ; + } +if (unlock_up)btrfs_unlock_up_safe (p, level +1 );
@@ -1467,6 +1472,8 @@ read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, ret = -EAGAIN ; goto out; + }else if (p->nowait) { +return -EAGAIN ; }if (unlock_up) {
@@ -1634,7 +1641,13 @@ static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,* We don't know the level of the root node until we actually * have it read locked */ - b =btrfs_read_lock_root_node (root); +if (p->nowait) { + b =btrfs_try_read_lock_root_node (root); +if (IS_ERR (b)) +return b; + }else { + b =btrfs_read_lock_root_node (root); + } level =btrfs_header_level (b);if (level > write_lock_level) goto out;
@@ -1910,6 +1923,13 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,WARN_ON (p->nodes[0 ] !=NULL );BUG_ON (!cow && ins_len); +/* +* For now only allow nowait for read only operations. There's no +* strict reason why we can't, we just only need it for reads so I'm +* only implementing it for reads right now. +*/ +ASSERT (!p->nowait || !cow); +if (ins_len <0 ) { lowest_unlock =2 ;
@@ -1936,7 +1956,12 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,if (p->need_commit_sem) {ASSERT (p->search_commit_root); -down_read (&fs_info->commit_root_sem); +if (p->nowait) { +if (!down_read_trylock (&fs_info->commit_root_sem)) +return -EAGAIN ; + }else { +down_read (&fs_info->commit_root_sem); + } } again:
@@ -2082,7 +2107,15 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,btrfs_tree_lock (b); p->locks[level] =BTRFS_WRITE_LOCK ; }else { -btrfs_tree_read_lock (b); +if (p->nowait) { +if (!btrfs_try_tree_read_lock (b)) { +free_extent_buffer (b); + ret = -EAGAIN ; + goto done; + } + }else { +btrfs_tree_read_lock (b); + } p->locks[level] =BTRFS_READ_LOCK ; } p->nodes[level] = b;
fs/btrfs/ctree.h +1 -0
@@ -443,6 +443,7 @@ struct btrfs_path {* header (ie. sizeof(struct btrfs_item) is not included). */ unsigned int search_for_extension:1; + unsigned int nowait:1; };#define BTRFS_MAX_EXTENT_ITEM_SIZE (r) ((BTRFS_LEAF_DATA_SIZE(r->fs_info) >> 4) - \ sizeof(struct btrfs_item))
fs/btrfs/locking.c +23 -0
@@ -285,6 +285,29 @@ struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)return eb; } +/* +* Loop around taking references on and locking the root node of the tree in +* nowait mode until we end up with a lock on the root node or returning to +* avoid blocking. +* +* Return: root extent buffer with read lock held or -EWOULDBLOCK. +*/ +struct extent_buffer *btrfs_try_read_lock_root_node (struct btrfs_root *root) +{ +struct extent_buffer *eb; + +while (1 ) { + eb =btrfs_root_node (root); +if (!btrfs_try_tree_read_lock (eb)) +return ERR_PTR (-EAGAIN ); +if (eb == root->node) +break ; +btrfs_tree_read_unlock (eb); +free_extent_buffer (eb); + } +return eb; +} +/* * DREW locks * ==========
fs/btrfs/locking.h +1 -0
@@ -94,6 +94,7 @@ int btrfs_try_tree_read_lock(struct extent_buffer *eb);int btrfs_try_tree_write_lock (struct extent_buffer *eb);struct extent_buffer *btrfs_lock_root_node (struct btrfs_root *root);struct extent_buffer *btrfs_read_lock_root_node (struct btrfs_root *root); +struct extent_buffer *btrfs_try_read_lock_root_node (struct btrfs_root *root);#ifdef CONFIG_BTRFS_DEBUG static inline void btrfs_assert_tree_write_locked (struct extent_buffer *eb)
Signature · 1 line
2.30.2
Re: [PATCH v3 02/12] btrfs: implement a nowait option for tree searches
Quoted 128 lines · Stefan Roesch
From: Josef Bacik <[email protected]>
For NOWAIT IOCB's we'll need a way to tell search to not wait on locks or anything. Accomplish this by adding a
path->nowaitflag that will use trylocks and skip reading of metadata, returning-EAGAINin either of these cases. For now we only need this for reads, so only the read side is handled. Add anASSERT()to catch anybody trying to use this for writes so they know they'll have to implement the write side.Signed-off-by Josef Bacik <[email protected]>Signed-off-by Stefan Roesch <[email protected]>--- fs/btrfs/ctree.c | 39 ++++++++++++++++++++++++++++++++++++--- fs/btrfs/ctree.h | 1 + fs/btrfs/locking.c | 23 +++++++++++++++++++++++ fs/btrfs/locking.h | 1 + 4 files changed, 61 insertions(+), 3 deletions(-)fs/btrfs/ctree.c +36
-3@@ -1447,6 +1447,11 @@ read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,return 0 ; } +if (p->nowait) { +free_extent_buffer (tmp); +return -EAGAIN ; + } +if (unlock_up)btrfs_unlock_up_safe (p, level +1 );@@ -1467,6 +1472,8 @@ read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, ret = -EAGAIN ; goto out; + }else if (p->nowait) { +return -EAGAIN ; }if (unlock_up) {@@ -1634,7 +1641,13 @@ static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,* We don't know the level of the root node until we actually * have it read locked */ - b =btrfs_read_lock_root_node (root); +if (p->nowait) { + b =btrfs_try_read_lock_root_node (root); +if (IS_ERR (b)) +return b; + }else { + b =btrfs_read_lock_root_node (root); + } level =btrfs_header_level (b);if (level > write_lock_level) goto out;@@ -1910,6 +1923,13 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,WARN_ON (p->nodes[0 ] !=NULL );BUG_ON (!cow && ins_len); +/* +* For now only allow nowait for read only operations. There's no +* strict reason why we can't, we just only need it for reads so I'm +* only implementing it for reads right now. +*/ +ASSERT (!p->nowait || !cow); +if (ins_len <0 ) { lowest_unlock =2 ;@@ -1936,7 +1956,12 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,if (p->need_commit_sem) {ASSERT (p->search_commit_root); -down_read (&fs_info->commit_root_sem); +if (p->nowait) { +if (!down_read_trylock (&fs_info->commit_root_sem)) +return -EAGAIN ; + }else { +down_read (&fs_info->commit_root_sem); + } } again:@@ -2082,7 +2107,15 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,btrfs_tree_lock (b); p->locks[level] =BTRFS_WRITE_LOCK ; }else { -btrfs_tree_read_lock (b); +if (p->nowait) { +if (!btrfs_try_tree_read_lock (b)) { +free_extent_buffer (b); + ret = -EAGAIN ; + goto done; + } + }else { +btrfs_tree_read_lock (b); + } p->locks[level] =BTRFS_READ_LOCK ; } p->nodes[level] = b;fs/btrfs/ctree.h +1
-0@@ -443,6 +443,7 @@ struct btrfs_path {* header (ie. sizeof(struct btrfs_item) is not included). */ unsigned int search_for_extension:1; + unsigned int nowait:1; };#define BTRFS_MAX_EXTENT_ITEM_SIZE (r) ((BTRFS_LEAF_DATA_SIZE(r->fs_info) >> 4) - \ sizeof(struct btrfs_item))fs/btrfs/locking.c +5
-0@@ -285,6 +285,29 @@ struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)return eb; } +/* +* Loop around taking references on and locking the root node of the tree in +* nowait mode until we end up with a lock on the root node or returning to +* avoid blocking. +*
fs/btrfs/locking.c struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) · line 302
Quoted 10 lines
fs/btrfs/locking.c +10
-0+* Return: root extent buffer with read lock held or -EWOULDBLOCK. +*/ +struct extent_buffer *btrfs_try_read_lock_root_node (struct btrfs_root *root) +{ +struct extent_buffer *eb; + +while (1 ) { + eb =btrfs_root_node (root); +if (!btrfs_try_tree_read_lock (eb)) +return ERR_PTR (-EAGAIN );
There's a leak of the extent buffer here.
This fixes it up:
fs/btrfs/locking.c +3 -1
@@ -298,8 +298,10 @@ struct extent_buffer *btrfs_try_read_lock_root_node(struct btrfs_root *root)while (1 ) { eb =btrfs_root_node (root); -if (!btrfs_try_tree_read_lock (eb)) +if (!btrfs_try_tree_read_lock (eb)) { +free_extent_buffer (eb);return ERR_PTR (-EAGAIN ); + }if (eb == root->node)break ;btrfs_tree_read_unlock (eb);
Thanks.
Quoted 26 lines
fs/btrfs/locking.c +8
-0+if (eb == root->node) +break ; +btrfs_tree_read_unlock (eb); +free_extent_buffer (eb); + } +return eb; +} +/* * DREW locks * ========== fs/btrfs/locking.h +1
-0@@ -94,6 +94,7 @@ int btrfs_try_tree_read_lock(struct extent_buffer *eb);int btrfs_try_tree_write_lock (struct extent_buffer *eb);struct extent_buffer *btrfs_lock_root_node (struct btrfs_root *root);struct extent_buffer *btrfs_read_lock_root_node (struct btrfs_root *root); +struct extent_buffer *btrfs_try_read_lock_root_node (struct btrfs_root *root);#ifdef CONFIG_BTRFS_DEBUG static inline void btrfs_assert_tree_write_locked (struct extent_buffer *eb)Signature · 1 line
2.30.2
Re: [PATCH v3 02/12] btrfs: implement a nowait option for tree searches
Quoted 19 lines · Filipe Manana
Review
Quoted 8 lines · Stefan Roesch
From: Josef Bacik <[email protected]>
(quoted diff) +7
-0+{ + struct extent_buffer *eb; + + while (1) { + eb = btrfs_root_node(root); + if (!btrfs_try_tree_read_lock(eb)) + return ERR_PTR(-EAGAIN);There's a leak of the extent buffer here.
This fixes it up:fs/btrfs/locking.c +0
-0@@ -298,8 +298,10 @@ struct extent_buffer *btrfs_try_read_lock_root_node(struct btrfs_root *root)
fs/btrfs/locking.c struct extent_buffer *btrfs_try_read_lock_root_node(struct btrfs_root *root) · line 307
Quoted 10 lines
fs/btrfs/locking.c +3
-1while (1 ) { eb =btrfs_root_node (root); -if (!btrfs_try_tree_read_lock (eb)) +if (!btrfs_try_tree_read_lock (eb)) { +free_extent_buffer (eb);return ERR_PTR (-EAGAIN ); + }if (eb == root->node)break ;btrfs_tree_read_unlock (eb);
Folded to the commit, thanks.
PATCH v3 3/12 btrfs: make can_nocow_extent nowait compatible
In reply to Stefan RoeschFrom: Josef Bacik <[email protected]>
If we have NOWAIT specified on our IOCB and we're writing into a PREALLOC or NOCOW extent then we need to be able to tell can_nocow_extent that we don't want to wait on any locks or metadata IO. Fix can_nocow_extent to allow for NOWAIT.
--- fs/btrfs/ctree.h | 5 +++-- fs/btrfs/extent-tree.c | 5 +++++ fs/btrfs/file-item.c | 4 +++- fs/btrfs/file.c | 2 +- fs/btrfs/inode.c | 16 ++++++++++------ fs/btrfs/relocation.c | 2 +- fs/btrfs/scrub.c | 4 ++-- fs/btrfs/tree-log.c | 6 +++--- 8 files changed, 28 insertions(+), 16 deletions(-)
fs/btrfs/ctree.h +3 -2
@@ -3274,7 +3274,8 @@ int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,blk_status_t btrfs_csum_one_bio (struct btrfs_inode *inode,struct bio *bio,u64 offset,bool one_ordered);int btrfs_lookup_csums_range (struct btrfs_root *root,u64 start,u64 end, -struct list_head *list,int search_commit); +struct list_head *list,int search_commit, +bool nowait); void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode, const struct btrfs_path *path, struct btrfs_file_extent_item *fi,
@@ -3304,7 +3305,7 @@ struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, u64 start, u64 len); noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,u64 *orig_start,u64 *orig_block_len, -u64 *ram_bytes,bool strict); +u64 *ram_bytes,bool nowait,bool strict);void __btrfs_del_delalloc_inode (struct btrfs_root *root,struct btrfs_inode *inode);
fs/btrfs/extent-tree.c +5 -0
@@ -2220,6 +2220,11 @@ static noinline int check_delayed_ref(struct btrfs_root *root, }if (!mutex_trylock (&head->mutex)) { +if (path->nowait) { +spin_unlock (&delayed_refs->lock); +return -EAGAIN ; + } +refcount_inc (&head->refs);spin_unlock (&delayed_refs->lock);
fs/btrfs/file-item.c +3 -1
@@ -503,7 +503,8 @@ blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst }int btrfs_lookup_csums_range (struct btrfs_root *root,u64 start,u64 end, -struct list_head *list,int search_commit) +struct list_head *list,int search_commit, +bool nowait) {struct btrfs_fs_info *fs_info = root->fs_info;struct btrfs_key key;
@@ -525,6 +526,7 @@ int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,if (!path)return -ENOMEM ; + path->nowait = nowait;if (search_commit) { path->skip_locking =1 ; path->reada =READA_FORWARD ;
fs/btrfs/file.c +1 -1
@@ -1502,7 +1502,7 @@ int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,btrfs_lock_and_flush_ordered_range (inode, lockstart, lockend,NULL ); ret =can_nocow_extent (&inode->vfs_inode, lockstart, &num_bytes, -NULL ,NULL ,NULL , false); +NULL ,NULL ,NULL , false, false);if (ret <=0 ) { ret =0 ;btrfs_drew_write_unlock (&root->snapshot_lock);
fs/btrfs/inode.c +10 -6
@@ -1666,7 +1666,7 @@ static noinline int run_delalloc_zoned(struct btrfs_inode *inode, } static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, -u64 bytenr,u64 num_bytes) +u64 bytenr,u64 num_bytes,bool nowait) {struct btrfs_root *csum_root =btrfs_csum_root (fs_info, bytenr);struct btrfs_ordered_sum *sums;
@@ -1674,7 +1674,8 @@ static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,LIST_HEAD (list); ret =btrfs_lookup_csums_range (csum_root, bytenr, - bytenr + num_bytes -1 , &list,0 ); + bytenr + num_bytes -1 , &list,0 , + nowait);if (ret ==0 &&list_empty (&list))return 0 ;
@@ -1800,6 +1801,7 @@ static int can_nocow_file_extent(struct btrfs_path *path,u8 extent_type;int can_nocow =0 ;int ret =0 ; +bool nowait = path->nowait; fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); extent_type =btrfs_file_extent_type (leaf, fi);
@@ -1876,7 +1878,8 @@ static int can_nocow_file_extent(struct btrfs_path *path,* Force COW if csums exist in the range. This ensures that csums for a * given extent are either valid or do not exist. */ - ret =csum_exist_in_range (root->fs_info, args->disk_bytenr, args->num_bytes); + ret =csum_exist_in_range (root->fs_info, args->disk_bytenr, args->num_bytes, + nowait);WARN_ON_ONCE (ret >0 && is_freespace_inode);if (ret !=0 ) goto out;
@@ -7292,7 +7295,7 @@ static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)*/ noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,u64 *orig_start,u64 *orig_block_len, -u64 *ram_bytes,bool strict) +u64 *ram_bytes,bool nowait,bool strict) {struct btrfs_fs_info *fs_info =btrfs_sb (inode->i_sb);struct can_nocow_file_extent_args nocow_args = {0 };
@@ -7308,6 +7311,7 @@ noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, path =btrfs_alloc_path ();if (!path)return -ENOMEM ; + path->nowait = nowait; ret =btrfs_lookup_file_extent (NULL , root, path,btrfs_ino (BTRFS_I (inode)), offset,0 );
@@ -7577,7 +7581,7 @@ static int btrfs_get_blocks_direct_write(struct extent_map **map, block_start = em->block_start + (start - em->start);if (can_nocow_extent (inode, start, &len, &orig_start, - &orig_block_len, &ram_bytes, false) ==1 ) { + &orig_block_len, &ram_bytes, false, false) ==1 ) { bg =btrfs_inc_nocow_writers (fs_info, block_start);if (bg) can_nocow = true;
@@ -11242,7 +11246,7 @@ static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,free_extent_map (em); em =NULL ; - ret =can_nocow_extent (inode, start, &len,NULL ,NULL ,NULL , true); + ret =can_nocow_extent (inode, start, &len,NULL ,NULL ,NULL , false, true);if (ret <0 ) { goto out; }else if (ret) {
fs/btrfs/relocation.c +1 -1
@@ -4339,7 +4339,7 @@ int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len) disk_bytenr = file_pos + inode->index_cnt; csum_root =btrfs_csum_root (fs_info, disk_bytenr); ret =btrfs_lookup_csums_range (csum_root, disk_bytenr, - disk_bytenr + len -1 , &list,0 ); + disk_bytenr + len -1 , &list,0 , false);if (ret) goto out;
fs/btrfs/scrub.c +2 -2
@@ -3077,7 +3077,7 @@ static int scrub_raid56_data_stripe_for_parity(struct scrub_ctx *sctx, ret =btrfs_lookup_csums_range (csum_root, extent_start, extent_start + extent_size -1 , - &sctx->csum_list,1 ); + &sctx->csum_list,1 , false);if (ret) {scrub_parity_mark_sectors_error (sparity, extent_start, extent_size);
@@ -3303,7 +3303,7 @@ static int scrub_simple_mirror(struct scrub_ctx *sctx,if (extent_flags &BTRFS_EXTENT_FLAG_DATA ) { ret =btrfs_lookup_csums_range (csum_root, cur_logical, cur_logical + scrub_len -1 , - &sctx->csum_list,1 ); + &sctx->csum_list,1 , false);if (ret)break ; }
fs/btrfs/tree-log.c +3 -3
@@ -801,7 +801,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans, ret =btrfs_lookup_csums_range (root->log_root, csum_start, csum_end -1 , - &ordered_sums,0 ); + &ordered_sums,0 , false);if (ret) goto out;/*
@@ -4513,7 +4513,7 @@ static noinline int copy_items(struct btrfs_trans_handle *trans, disk_bytenr += extent_offset; ret =btrfs_lookup_csums_range (csum_root, disk_bytenr, disk_bytenr + extent_num_bytes -1 , - &ordered_sums,0 ); + &ordered_sums,0 , false);if (ret) goto out;
@@ -4709,7 +4709,7 @@ static int log_extent_csums(struct btrfs_trans_handle *trans, ret =btrfs_lookup_csums_range (csum_root, em->block_start + csum_offset, em->block_start + csum_offset + - csum_len -1 , &ordered_sums,0 ); + csum_len -1 , &ordered_sums,0 , false);if (ret)return ret;
Signature · 1 line
2.30.2
PATCH v3 4/12 btrfs: add the ability to use NO_FLUSH for data reservations
In reply to Stefan RoeschFrom: Josef Bacik <[email protected]>
In order to accommodate NOWAIT IOCB's we need to be able to do NO_FLUSH data reservations, so plumb this through the delalloc reservation system.
--- fs/btrfs/block-group.c | 2 +- fs/btrfs/delalloc-space.c | 13 ++++++++++--- fs/btrfs/delalloc-space.h | 3 ++- fs/btrfs/file.c | 2 +- fs/btrfs/inode.c | 4 ++-- fs/btrfs/space-info.c | 3 ++- 6 files changed, 18 insertions(+), 9 deletions(-)
fs/btrfs/block-group.c +1 -1
@@ -2869,7 +2869,7 @@ static int cache_save_setup(struct btrfs_block_group *block_group, cache_size *= fs_info->sectorsize; ret =btrfs_check_data_free_space (BTRFS_I (inode), &data_reserved,0 , - cache_size); + cache_size, false);if (ret) goto out_put;
fs/btrfs/delalloc-space.c +10 -3
@@ -127,9 +127,11 @@ int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) }int btrfs_check_data_free_space (struct btrfs_inode *inode, -struct extent_changeset **reserved,u64 start,u64 len) +struct extent_changeset **reserved,u64 start, +u64 len,bool noflush) {struct btrfs_fs_info *fs_info = inode->root->fs_info; +enum btrfs_reserve_flush_enum flush =BTRFS_RESERVE_FLUSH_DATA ;int ret;/* align the range */
@@ -137,7 +139,12 @@ int btrfs_check_data_free_space(struct btrfs_inode *inode,round_down (start, fs_info->sectorsize); start =round_down (start, fs_info->sectorsize); - ret =btrfs_alloc_data_chunk_ondemand (inode, len); +if (noflush) + flush =BTRFS_RESERVE_NO_FLUSH ; +else if (btrfs_is_free_space_inode (inode)) + flush =BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE ; + + ret =btrfs_reserve_data_bytes (fs_info, len, flush);if (ret <0 )return ret;
@@ -454,7 +461,7 @@ int btrfs_delalloc_reserve_space(struct btrfs_inode *inode, {int ret; - ret =btrfs_check_data_free_space (inode, reserved, start, len); + ret =btrfs_check_data_free_space (inode, reserved, start, len, false);if (ret <0 )return ret; ret =btrfs_delalloc_reserve_metadata (inode, len, len, false);
fs/btrfs/delalloc-space.h +2 -1
@@ -7,7 +7,8 @@ struct extent_changeset;int btrfs_alloc_data_chunk_ondemand (struct btrfs_inode *inode,u64 bytes);int btrfs_check_data_free_space (struct btrfs_inode *inode, -struct extent_changeset **reserved,u64 start,u64 len); +struct extent_changeset **reserved,u64 start,u64 len, +bool noflush);void btrfs_free_reserved_data_space (struct btrfs_inode *inode,struct extent_changeset *reserved,u64 start,u64 len); void btrfs_delalloc_release_space(struct btrfs_inode *inode,
fs/btrfs/file.c +1 -1
@@ -1664,7 +1664,7 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,extent_changeset_release (data_reserved); ret =btrfs_check_data_free_space (BTRFS_I (inode), &data_reserved, pos, - write_bytes); + write_bytes, false);if (ret <0 ) {/* * If we don't have to COW at the offset, reserve
fs/btrfs/inode.c +2 -2
@@ -4881,7 +4881,7 @@ int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len, block_end = block_start + blocksize -1 ; ret =btrfs_check_data_free_space (inode, &data_reserved, block_start, - blocksize); + blocksize, false);if (ret <0 ) {if (btrfs_check_nocow_lock (inode, block_start, &write_bytes) >0 ) {/* For nocow case, no need to reserve data space */
@@ -7766,7 +7766,7 @@ static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,if (write && !(flags &IOMAP_NOWAIT )) { ret =btrfs_check_data_free_space (BTRFS_I (inode), &dio_data->data_reserved, - start, data_alloc_len); + start, data_alloc_len, false);if (!ret) dio_data->data_space_reserved = true; else if (ret && !(BTRFS_I(inode)->flags &
fs/btrfs/space-info.c +2 -1
@@ -1737,7 +1737,8 @@ int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,int ret;ASSERT (flush ==BTRFS_RESERVE_FLUSH_DATA || - flush ==BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE ); + flush ==BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE || + flush ==BTRFS_RESERVE_NO_FLUSH );ASSERT (!current->journal_info || flush !=BTRFS_RESERVE_FLUSH_DATA ); ret =__reserve_bytes (fs_info, data_sinfo, bytes, flush);
Signature · 1 line
2.30.2
PATCH v3 5/12 btrfs: add btrfs_try_lock_ordered_range
In reply to Stefan RoeschFrom: Josef Bacik <[email protected]>
For IOCB_NOWAIT we're going to want to use try lock on the extent lock, and simply bail if there's an ordered extent in the range because the only choice there is to wait for the ordered extent to complete.
--- fs/btrfs/ordered-data.c | 28 ++++++++++++++++++++++++++++ fs/btrfs/ordered-data.h | 1 + 2 files changed, 29 insertions(+)
fs/btrfs/ordered-data.c +28 -0
@@ -1041,6 +1041,34 @@ void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start, } } +/* +* btrfs_try_lock_ordered_range - lock the passed range and ensure all pending +* ordered extents in it are run to completion in nowait mode. +* +* @inode: Inode whose ordered tree is to be searched +* @start: Beginning of range to flush +* @end: Last byte of range to lock +* +* This function returns true if btrfs_lock_ordered_range does not return any +* extents, otherwise false. +*/ +bool btrfs_try_lock_ordered_range (struct btrfs_inode *inode,u64 start,u64 end) +{ +struct btrfs_ordered_extent *ordered; + +if (!try_lock_extent (&inode->io_tree, start, end)) +return false; + + ordered =btrfs_lookup_ordered_range (inode, start, end - start +1 ); +if (!ordered) +return true; + +btrfs_put_ordered_extent (ordered); +unlock_extent (&inode->io_tree, start, end); +return false; +} + +static int clone_ordered_extent (struct btrfs_ordered_extent *ordered,u64 pos,u64 len) {
fs/btrfs/ordered-data.h +1 -0
@@ -218,6 +218,7 @@ void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,void btrfs_lock_and_flush_ordered_range (struct btrfs_inode *inode,u64 start,u64 end,struct extent_state **cached_state); +bool btrfs_try_lock_ordered_range (struct btrfs_inode *inode,u64 start,u64 end);int btrfs_split_ordered_extent (struct btrfs_ordered_extent *ordered,u64 pre,u64 post);int __init ordered_data_init (void );
Signature · 1 line
2.30.2
PATCH v3 6/12 btrfs: make btrfs_check_nocow_lock nowait compatible
In reply to Stefan RoeschFrom: Josef Bacik <[email protected]>
Now all the helpers that btrfs_check_nocow_lock uses handle nowait, add a nowait flag to btrfs_check_nocow_lock so it can be used by the write path.
--- fs/btrfs/ctree.h | 2 +- fs/btrfs/file.c | 33 ++++++++++++++++++++++----------- fs/btrfs/inode.c | 2 +- 3 files changed, 24 insertions(+), 13 deletions(-)
fs/btrfs/ctree.h +1 -1
@@ -3480,7 +3480,7 @@ int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages, struct extent_state **cached, bool noreserve);int btrfs_fdatawrite_range (struct inode *inode,loff_t start,loff_t end);int btrfs_check_nocow_lock (struct btrfs_inode *inode,loff_t pos, -size_t *write_bytes); +size_t *write_bytes,bool nowait);void btrfs_check_nocow_unlock (struct btrfs_inode *inode);/* tree-defrag.c */
fs/btrfs/file.c +22 -11
@@ -1481,7 +1481,7 @@ lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,* NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0. */ int btrfs_check_nocow_lock (struct btrfs_inode *inode,loff_t pos, -size_t *write_bytes) +size_t *write_bytes,bool nowait) {struct btrfs_fs_info *fs_info = inode->root->fs_info;struct btrfs_root *root = inode->root;
@@ -1500,16 +1500,21 @@ int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos, fs_info->sectorsize) - 1; num_bytes = lockend - lockstart +1 ; -btrfs_lock_and_flush_ordered_range (inode, lockstart, lockend,NULL ); +if (nowait) { +if (!btrfs_try_lock_ordered_range (inode, lockstart, lockend)) { +btrfs_drew_write_unlock (&root->snapshot_lock); +return -EAGAIN ; + } + }else { +btrfs_lock_and_flush_ordered_range (inode, lockstart, lockend,NULL ); + } ret =can_nocow_extent (&inode->vfs_inode, lockstart, &num_bytes, -NULL ,NULL ,NULL , false, false); -if (ret <=0 ) { - ret =0 ; +NULL ,NULL ,NULL , nowait, false); +if (ret <=0 )btrfs_drew_write_unlock (&root->snapshot_lock); - }else { +else *write_bytes =min_t (size_t, *write_bytes , num_bytes - pos + lockstart); - }unlock_extent (&inode->io_tree, lockstart, lockend);return ret;
@@ -1666,16 +1671,22 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, &data_reserved, pos, write_bytes, false);if (ret <0 ) { +int can_nocow; +/* * If we don't have to COW at the offset, reserve * metadata only. write_bytes may get smaller than * requested here. */ -if (btrfs_check_nocow_lock (BTRFS_I (inode), pos, - &write_bytes) >0 ) - only_release_metadata = true; -else + can_nocow =btrfs_check_nocow_lock (BTRFS_I (inode), pos, + &write_bytes, false); +if (can_nocow <0 ) + ret = can_nocow; +if (can_nocow >0 ) + ret =0 ; +if (ret)break ; + only_release_metadata = true; } num_pages =DIV_ROUND_UP (write_bytes + offset,PAGE_SIZE );
fs/btrfs/inode.c +1 -1
@@ -4883,7 +4883,7 @@ int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len, ret =btrfs_check_data_free_space (inode, &data_reserved, block_start, blocksize, false);if (ret <0 ) { -if (btrfs_check_nocow_lock (inode, block_start, &write_bytes) >0 ) { +if (btrfs_check_nocow_lock (inode, block_start, &write_bytes, false) >0 ) {/* For nocow case, no need to reserve data space */ only_release_metadata = true; }else {
Signature · 1 line
2.30.2
PATCH v3 7/12 btrfs: make prepare_pages nowait compatible
In reply to Stefan RoeschAdd nowait parameter to the prepare_pages function. In case nowait is specified for an async buffered write request, do a nowait allocation or return -EAGAIN.
--- fs/btrfs/file.c | 42 +++++++++++++++++++++++++++++++++++------- 1 file changed, 35 insertions(+), 7 deletions(-)
fs/btrfs/file.c +35 -7
@@ -1339,26 +1339,54 @@ static int prepare_uptodate_page(struct inode *inode,return 0 ; } +static int get_prepare_fgp_flags (bool nowait) +{ +int fgp_flags =FGP_LOCK |FGP_ACCESSED |FGP_CREAT ; + +if (nowait) + fgp_flags |=FGP_NOWAIT ; + +return fgp_flags; +} + +static gfp_t get_prepare_gfp_flags (struct inode *inode,bool nowait) +{ +gfp_t gfp; + + gfp =btrfs_alloc_write_mask (inode->i_mapping); +if (nowait) { + gfp &= ~__GFP_DIRECT_RECLAIM; + gfp |=GFP_NOWAIT ; + } + +return gfp; +} +/* * this just gets pages into the page cache and locks them down. */ static noinline int prepare_pages(struct inode *inode, struct page **pages,size_t num_pages,loff_t pos, -size_t write_bytes,bool force_uptodate) +size_t write_bytes,bool force_uptodate, +bool nowait) {int i;unsigned long index = pos >>PAGE_SHIFT ; -gfp_t mask =btrfs_alloc_write_mask (inode->i_mapping); +gfp_t mask =get_prepare_gfp_flags (inode, nowait); +int fgp_flags =get_prepare_fgp_flags (nowait);int err =0 ;int faili;for (i =0 ; i < num_pages; i++) { again: - pages[i] =find_or_create_page (inode->i_mapping, index + i, - mask | __GFP_WRITE); + pages[i] =pagecache_get_page (inode->i_mapping, index + i, + fgp_flags, mask | __GFP_WRITE);if (!pages[i]) { faili = i -1 ; - err = -ENOMEM ; +if (nowait) + err = -EAGAIN ; +else + err = -ENOMEM ; goto fail; }
@@ -1376,7 +1404,7 @@ static noinline int prepare_pages(struct inode *inode, struct page **pages, pos + write_bytes, false);if (err) {put_page (pages[i]); -if (err == -EAGAIN ) { +if (!nowait && err == -EAGAIN ) { err =0 ; goto again; }
@@ -1716,7 +1744,7 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,*/ ret =prepare_pages (inode, pages, num_pages, pos, write_bytes, - force_page_uptodate); + force_page_uptodate, false);if (ret) {btrfs_delalloc_release_extents (BTRFS_I (inode), reserve_bytes);
Signature · 1 line
2.30.2
PATCH v3 8/12 btrfs: make lock_and_cleanup_extent_if_need nowait compatible
In reply to Stefan RoeschThis adds the nowait parameter to lock_and_cleanup_extent_if_need(). If the nowait parameter is specified we try to lock the extent in nowait mode.
--- fs/btrfs/file.c | 19 ++++++++++++++++--- 1 file changed, 16 insertions(+), 3 deletions(-)
fs/btrfs/file.c +16 -3
@@ -1439,7 +1439,7 @@ static noinline int lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages, size_t num_pages, loff_t pos, size_t write_bytes, - u64 *lockstart, u64 *lockend, + u64 *lockstart, u64 *lockend, bool nowait, struct extent_state **cached_state) { struct btrfs_fs_info *fs_info = inode->root->fs_info;
@@ -1454,8 +1454,21 @@ lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,if (start_pos < inode->vfs_inode.i_size) {struct btrfs_ordered_extent *ordered; -lock_extent_bits (&inode->io_tree, start_pos, last_pos, +if (nowait) { +if (!try_lock_extent (&inode->io_tree, start_pos, last_pos)) { +for (i =0 ; i < num_pages; i++) { +unlock_page (pages[i]); +put_page (pages[i]); + pages[i] =NULL ; + } + +return -EAGAIN ; + } + }else { +lock_extent_bits (&inode->io_tree, start_pos, last_pos, cached_state); + } + ordered =btrfs_lookup_ordered_range (inode, start_pos, last_pos - start_pos +1 ); if (ordered &&
@@ -1754,7 +1767,7 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, extents_locked =lock_and_cleanup_extent_if_need (BTRFS_I (inode), pages, num_pages, pos, write_bytes, &lockstart, - &lockend, &cached_state); + &lockend, false, &cached_state);if (extents_locked <0 ) {if (extents_locked == -EAGAIN ) goto again;
Signature · 1 line
2.30.2
PATCH v3 9/12 btrfs: plumb NOWAIT through the write path
In reply to Stefan RoeschWe have everywhere setup for nowait, plumb NOWAIT through the write path.
--- fs/btrfs/file.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)
fs/btrfs/file.c +13 -6
@@ -1653,8 +1653,9 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,bool force_page_uptodate = false;loff_t old_isize =i_size_read (inode);unsigned int ilock_flags =0 ; +const bool nowait = iocb->ki_flags &IOCB_NOWAIT ; -if (iocb->ki_flags &IOCB_NOWAIT ) +if (nowait) ilock_flags |=BTRFS_ILOCK_TRY ; ret =btrfs_inode_lock (inode, ilock_flags);
@@ -1710,17 +1711,22 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,extent_changeset_release (data_reserved); ret =btrfs_check_data_free_space (BTRFS_I (inode), &data_reserved, pos, - write_bytes, false); + write_bytes, nowait);if (ret <0 ) {int can_nocow; +if (nowait && (ret == -ENOSPC || ret == -EAGAIN )) { + ret = -EAGAIN ; +break ; + } +/* * If we don't have to COW at the offset, reserve * metadata only. write_bytes may get smaller than * requested here. */ can_nocow =btrfs_check_nocow_lock (BTRFS_I (inode), pos, - &write_bytes, false); + &write_bytes, nowait);if (can_nocow <0 ) ret = can_nocow;if (can_nocow >0 )
@@ -1737,7 +1743,7 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,WARN_ON (reserve_bytes ==0 ); ret =btrfs_delalloc_reserve_metadata (BTRFS_I (inode), reserve_bytes, - reserve_bytes, false); + reserve_bytes, nowait);if (ret) {if (!only_release_metadata) btrfs_free_reserved_data_space(BTRFS_I(inode),
@@ -1767,10 +1773,11 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, extents_locked =lock_and_cleanup_extent_if_need (BTRFS_I (inode), pages, num_pages, pos, write_bytes, &lockstart, - &lockend, false, &cached_state); + &lockend, nowait, &cached_state);if (extents_locked <0 ) { -if (extents_locked == -EAGAIN ) +if (!nowait && extents_locked == -EAGAIN ) goto again; +btrfs_delalloc_release_extents (BTRFS_I (inode), reserve_bytes); ret = extents_locked;
Signature · 1 line
2.30.2
PATCH v3 10/12 btrfs: make balance_dirty_pages nowait compatible
In reply to Stefan RoeschThis replaces the call to function balance_dirty_pages_ratelimited() in the function btrfs_buffered_write() with a call to balance_dirty_pages_ratelimited_flags().
It also moves the function after the again label. This can cause the function to be called a bit later, but this should have no impact in the real world.
--- fs/btrfs/file.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
fs/btrfs/file.c +5 -2
@@ -1654,6 +1654,7 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,loff_t old_isize =i_size_read (inode);unsigned int ilock_flags =0 ;const bool nowait = iocb->ki_flags &IOCB_NOWAIT ; +unsigned int bdp_flags = nowait ?BDP_ASYNC :0 ;if (nowait) ilock_flags |=BTRFS_ILOCK_TRY ;
@@ -1756,6 +1757,10 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb, release_bytes = reserve_bytes; again: + ret =balance_dirty_pages_ratelimited_flags (inode->i_mapping, bdp_flags); +if (ret) +break ; +/* * This is going to setup the pages array with the number of * pages we want, so we don't really need to worry about the
@@ -1860,8 +1865,6 @@ static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,cond_resched (); -balance_dirty_pages_ratelimited (inode->i_mapping); - pos += copied; num_written += copied; }
Signature · 1 line
2.30.2
PATCH v3 11/12 btrfs: assert nowait mode is not used for some btree search functions
In reply to Stefan RoeschThis adds nowait asserts to btree search functions which are not used by buffered IO and direct IO paths.
--- fs/btrfs/ctree.c | 4 ++++ 1 file changed, 4 insertions(+)
fs/btrfs/ctree.c +4 -0
@@ -2164,6 +2164,7 @@ int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, lowest_level = p->lowest_level;WARN_ON (p->nodes[0 ] !=NULL ); +ASSERT (!p->nowait);if (p->search_commit_root) {BUG_ON (time_seq);
@@ -4465,6 +4466,7 @@ int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,int ret =1 ;int keep_locks = path->keep_locks; +ASSERT (!path->nowait); path->keep_locks =1 ; again: cur =btrfs_read_lock_root_node (root);
@@ -4645,6 +4647,8 @@ int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,int ret;int i; +ASSERT (!path->nowait); + nritems =btrfs_header_nritems (path->nodes[0 ]);if (nritems ==0 )return 1 ;
Signature · 1 line
2.30.2
PATCH v3 12/12 btrfs: enable nowait async buffered writes
In reply to Stefan RoeschEnable nowait async buffered writes in btrfs_do_write_iter() and btrfs_file_open().
In this version encoded buffered writes have the optimization not enabled. Encoded writes are enabled by using an ioctl. io-uring currently do not support ioctl's. This might be enabled in the future.
Performance results:
For fio the following results have been obtained with a queue depth of 1 and 4k block size (runtime 600 secs):
sequential writes:
without patch with patch libaio psync
iops: 55k 134k 117K 148K bw: 221MB/s 538MB/s 469MB/s 592MB/s clat: 15286ns 82ns 994ns 6340ns
For an io depth of 1, the new patch improves throughput by over two times (compared to the exiting behavior, where buffered writes are processed by an io-worker process) and also the latency is considerably reduced. To achieve the same or better performance with the exisiting code an io depth of 4 is required. Increasing the iodepth further does not lead to improvements.
The tests have been run like this:
./fio --name=seq-writers --ioengine=psync --iodepth=1 --rw=write \
--bs=4k --direct=0 --size=100000m --time_based --runtime=600 \
--numjobs=1 --filename=...
./fio --name=seq-writers --ioengine=io_uring --iodepth=1 --rw=write \
--bs=4k --direct=0 --size=100000m --time_based --runtime=600 \
--numjobs=1 --filename=...
./fio --name=seq-writers --ioengine=libaio --iodepth=1 --rw=write \
--bs=4k --direct=0 --size=100000m --time_based --runtime=600 \
--numjobs=1 --filename=...
Testing:
This patch has been tested with xfstests, fsx, fio. xfstests shows no new diffs compared to running without the patch series.
--- fs/btrfs/file.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
fs/btrfs/file.c +4 -4
@@ -2107,13 +2107,13 @@ ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from,if (BTRFS_FS_ERROR (inode->root->fs_info))return -EROFS ; -if ((iocb->ki_flags &IOCB_NOWAIT ) && !(iocb->ki_flags &IOCB_DIRECT )) -return -EOPNOTSUPP ; -if (sync)atomic_inc (&inode->sync_writers);if (encoded) { +if (iocb->ki_flags &IOCB_NOWAIT ) +return -EOPNOTSUPP ; + num_written =btrfs_encoded_write (iocb, from, encoded); num_sync = encoded->len; }else if (iocb->ki_flags &IOCB_DIRECT ) {
@@ -3755,7 +3755,7 @@ static int btrfs_file_open(struct inode *inode, struct file *filp) {int ret; - filp->f_mode |=FMODE_NOWAIT |FMODE_BUF_RASYNC ; + filp->f_mode |=FMODE_NOWAIT |FMODE_BUF_RASYNC |FMODE_BUF_WASYNC ; ret =fsverity_file_open (inode, filp);if (ret)
Signature · 1 line
2.30.2
Quoted 12 lines · Stefan Roesch
This patch series adds support for async buffered writes when using both btrfs and io-uring. Currently io-uring only supports buffered writes (for btrfs) in the slow path, by processing them in the io workers. With this patch series it is now possible to support buffered writes in the fast path. To be able to use the fast path, the required pages must be in the page cache, the required locks in btrfs can be granted immediately and no additional blocks need to be read form disk.
This patch series makes use of the changes that have been introduced by a previous patch series: "io-uring/xfs: support async buffered writes"
Quoted 6 lines
Performance results:
The new patch improves throughput by over two times (compared to the exiting behavior, where buffered writes are processed by an io-worker process) and also the latency is considerably reduced. Detailled results are part of the changelog of the first commit.
Not in the first, but in the last commit instead.
Looks fine to me now, thanks.
Quoted 85 lines
BTRFS changes:
-Add option for NOWAIT IOCB's to tell that searches do not wait on locks. This
adds the nowait option tobtrfs_path.
-For NOWAIT buffered writes on PREALLOC or NOCOW extents tellcan_nocow_extent()
that we don't want to wait on any locks or metadata IO.
-Supportno_flushreservations for nowait buffered writes.
-Addbtrfs_try_lock_ordered_range()function.
-Add nowait flag tobtrfs_check_nocow_lock()to use it in write code path. -Add nowait parameter toprepare_pages()function.
-Plumb nowait through the write code path.
-Enable nowait buffered writes.Testing:
This patch has been tested with xfstests, fsx, fio. xfstests shows no new diffs compared to running without the patch series.Changes:
V3:
- Updated changelog of "btrfs: implement a nowait option for tree searches" to say-EAGAIN.
- Use bool return value in signature ofbtrfs_try_lock_ordered_range
- Renamed variable tmp tocan_nocowinbtrfs_buffered_write
- Fixed coding style inget_prepare_fgp_flags
- Set pages[i] to NULL in error code path oflock_and_cleanup_extent_if_need()
- Added const in definition of "bool nowait"
- Removed unlikely frombtrfs_buffered_write
- Rephrased changelog for "btrfs: add assert to search functions" and used asserts instead of warnings
- Explained why enocded writes are not supported in the changelog
- Moved performance results to changelog of first commitV2:
- Replace EWOULDBLOCK with EAGAIN. In Version 1 it was not used consistently
- Export functionbalance_dirty_pages_ratelimited_flags()
- Add asserts/warnings for search functions when nowait is set, but we don't expect that they are invoked with nowait set.Josef Bacik (5):
btrfs: implement a nowait option for tree searches
btrfs: makecan_nocow_extentnowait compatible
btrfs: add the ability to useNO_FLUSHfor data reservations
btrfs: addbtrfs_try_lock_ordered_range
btrfs: makebtrfs_check_nocow_locknowait compatibleStefan Roesch (7):
mm: exportbalance_dirty_pages_ratelimited_flags()
btrfs: makeprepare_pagesnowait compatible
btrfs: makelock_and_cleanup_extent_if_neednowait compatible
btrfs: plumb NOWAIT through the write path
btrfs: makebalance_dirty_pagesnowait compatible
btrfs: assert nowait mode is not used for some btree search functions
btrfs: enable nowait async buffered writesfs/btrfs/block-group.c | 2 +- fs/btrfs/ctree.c | 43 ++++++++++++- fs/btrfs/ctree.h | 8 ++- fs/btrfs/delalloc-space.c | 13 +++- fs/btrfs/delalloc-space.h | 3 +- fs/btrfs/extent-tree.c | 5 ++ fs/btrfs/file-item.c | 4 +- fs/btrfs/file.c | 124 ++++++++++++++++++++++++++++---------- fs/btrfs/inode.c | 22 ++++--- fs/btrfs/locking.c | 23 +++++++ fs/btrfs/locking.h | 1 + fs/btrfs/ordered-data.c | 28 +++++++++ fs/btrfs/ordered-data.h | 1 + fs/btrfs/relocation.c | 2 +- fs/btrfs/scrub.c | 4 +- fs/btrfs/space-info.c | 3 +- fs/btrfs/tree-log.c | 6 +- mm/page-writeback.c | 1 + 18 files changed, 234 insertions(+), 59 deletions(-)base-commit:
80e78fcce86de0288793a0ef0f6acf37656ee4cf--
2.30.2
Quoted 11 lines · Stefan Roesch
This patch series adds support for async buffered writes when using both btrfs and io-uring. Currently io-uring only supports buffered writes (for btrfs) in the slow path, by processing them in the io workers. With this patch series it is now possible to support buffered writes in the fast path. To be able to use the fast path, the required pages must be in the page cache, the required locks in btrfs can be granted immediately and no additional blocks need to be read form disk.
This patch series makes use of the changes that have been introduced by a previous patch series: "io-uring/xfs: support async buffered writes"
Quoted 6 lines
Performance results:
The new patch improves throughput by over two times (compared to the exiting behavior, where buffered writes are processed by an io-worker process) and also the latency is considerably reduced. Detailled results are part of the changelog of the first commit.
Thanks. It's late for including this patches to 6.1 queue but it's now in for-next and will be added to misc-next after rc1, targeting merge to 6.2. I did some minor fixups, so please don't resend full series unless there's a significant change. Incremental changes are fine if needed.
Quoted 19 lines · David Sterba
Quoted 11 lines · Stefan Roesch
This patch series adds support for async buffered writes when using both btrfs and io-uring. Currently io-uring only supports buffered writes (for btrfs) in the slow path, by processing them in the io workers. With this patch series it is now possible to support buffered writes in the fast path. To be able to use the fast path, the required pages must be in the page cache, the required locks in btrfs can be granted immediately and no additional blocks need to be read form disk.
This patch series makes use of the changes that have been introduced by a previous patch series: "io-uring/xfs: support async buffered writes"
Quoted 6 lines
Performance results:
The new patch improves throughput by over two times (compared to the exiting behavior, where buffered writes are processed by an io-worker process) and also the latency is considerably reduced. Detailled results are part of the changelog of the first commit.
Quoted 4 lines
Thanks. It's late for including this patches to 6.1 queue but it's now in for-next and will be added to misc-next after rc1, targeting merge to 6.2. I did some minor fixups, so please don't resend full series unless there's a significant change. Incremental changes are fine if needed.
I'm revisiting the merge target, the potential risk seems to be low here, straightforward changes for a separate feature, so it's now in the 6.1 queue. We still don't have ack for the function export so that would be good to have.
Jens Axboe
maintainerQuoted 25 lines · David Sterba
Quoted 19 lines · David Sterba
Quoted 11 lines · Stefan Roesch
This patch series adds support for async buffered writes when using both btrfs and io-uring. Currently io-uring only supports buffered writes (for btrfs) in the slow path, by processing them in the io workers. With this patch series it is now possible to support buffered writes in the fast path. To be able to use the fast path, the required pages must be in the page cache, the required locks in btrfs can be granted immediately and no additional blocks need to be read form disk.
This patch series makes use of the changes that have been introduced by a previous patch series: "io-uring/xfs: support async buffered writes"
Quoted 6 lines
Performance results:
The new patch improves throughput by over two times (compared to the exiting behavior, where buffered writes are processed by an io-worker process) and also the latency is considerably reduced. Detailled results are part of the changelog of the first commit.
Quoted 4 lines
Thanks. It's late for including this patches to 6.1 queue but it's now in for-next and will be added to misc-next after rc1, targeting merge to 6.2. I did some minor fixups, so please don't resend full series unless there's a significant change. Incremental changes are fine if needed.
Quoted 4 lines
I'm revisiting the merge target, the potential risk seems to be low here, straightforward changes for a separate feature, so it's now in the 6.1 queue. We still don't have ack for the function export so that would be good to have.
I don't think the ack there is a big deal, iomap already uses it but can't be modular. So it's just adding a modular user of the same thing, really.
6.1 sounds good to me!
Signature · 1 line
Jens Axboe