linux-stable/fs/btrfs
Qu Wenruo ae76d8e3e1 btrfs: scrub: fix grouping of read IO
[REGRESSION]
There are several regression reports about the scrub performance with
v6.4 kernel.

On a PCIe 3.0 device, the old v6.3 kernel can go 3GB/s scrub speed, but
v6.4 can only go 1GB/s, an obvious 66% performance drop.

[CAUSE]
Iostat shows a very different behavior between v6.3 and v6.4 kernel:

  Device         r/s      rkB/s   rrqm/s  %rrqm r_await rareq-sz aqu-sz  %util
  nvme0n1p3  9731.00 3425544.00 17237.00  63.92    2.18   352.02  21.18 100.00
  nvme0n1p3 15578.00  993616.00     5.00   0.03    0.09    63.78   1.32 100.00

The upper one is v6.3 while the lower one is v6.4.

There are several obvious differences:

- Very few read merges
  This turns out to be a behavior change that we no longer do bio
  plug/unplug.

- Very low aqu-sz
  This is due to the submit-and-wait behavior of flush_scrub_stripes(),
  and extra extent/csum tree search.

Both behaviors are not that obvious on SATA SSDs, as SATA SSDs have NCQ
to merge the reads, while SATA SSDs can not handle high queue depth well
either.

[FIX]
For now this patch focuses on the read speed fix. Dev-replace replace
speed needs more work.

For the read part, we go two directions to fix the problems:

- Re-introduce blk plug/unplug to merge read requests
  This is pretty simple, and the behavior is pretty easy to observe.

  This would enlarge the average read request size to 512K.

- Introduce multi-group reads and no longer wait for each group
  Instead of the old behavior, which submits 8 stripes and waits for
  them, here we would enlarge the total number of stripes to 16 * 8.
  Which is 8M per device, the same limit as the old scrub in-flight
  bios size limit.

  Now every time we fill a group (8 stripes), we submit them and
  continue to next stripes.

  Only when the full 16 * 8 stripes are all filled, we submit the
  remaining ones (the last group), and wait for all groups to finish.
  Then submit the repair writes and dev-replace writes.

  This should enlarge the queue depth.

This would greatly improve the merge rate (thus read block size) and
queue depth:

Before (with regression, and cached extent/csum path):

 Device         r/s      rkB/s   rrqm/s  %rrqm r_await rareq-sz aqu-sz  %util
 nvme0n1p3 20666.00 1318240.00    10.00   0.05    0.08    63.79   1.63 100.00

After (with all patches applied):

 nvme0n1p3  5165.00 2278304.00 30557.00  85.54    0.55   441.10   2.81 100.00

i.e. 1287 to 2224 MB/s.

CC: stable@vger.kernel.org # 6.4+
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2023-08-21 14:54:49 +02:00
..
tests btrfs: tests: add self tests for extent buffer memory operations 2023-08-21 14:52:16 +02:00
accessors.c btrfs: add eb to btrfs_node_key_ptr_offset 2022-12-05 18:00:58 +01:00
accessors.h btrfs: use helper sizeof_field in struct accessors 2023-08-21 14:52:13 +02:00
acl.c fs: port acl to mnt_idmap 2023-01-19 09:24:28 +01:00
acl.h fs: port ->set_acl() to pass mnt_idmap 2023-01-19 09:24:27 +01:00
async-thread.c btrfs: use alloc_ordered_workqueue() to create ordered workqueues 2023-06-19 13:59:30 +02:00
async-thread.h btrfs: use alloc_ordered_workqueue() to create ordered workqueues 2023-06-19 13:59:30 +02:00
backref.c btrfs: remove v0 extent handling 2023-08-21 14:54:48 +02:00
backref.h btrfs: fix backref walking not returning all inode refs 2023-05-09 22:09:11 +02:00
bio.c btrfs: add an ordered_extent pointer to struct btrfs_bio 2023-06-19 13:59:36 +02:00
bio.h btrfs: add an ordered_extent pointer to struct btrfs_bio 2023-06-19 13:59:36 +02:00
block-group.c btrfs: zoned: don't activate non-DATA BG on allocation 2023-08-21 14:52:19 +02:00
block-group.h btrfs: rename add_new_free_space() to btrfs_add_new_free_space() 2023-08-21 14:52:12 +02:00
block-rsv.c btrfs: account block group tree when calculating global reserve size 2023-07-20 19:22:54 +02:00
block-rsv.h btrfs: move btrfs_check_trunc_cache_free_space into block-rsv.c 2023-06-19 13:59:24 +02:00
btrfs_inode.h btrfs: reduce the number of arguments to btrfs_run_delalloc_range 2023-08-21 14:52:14 +02:00
check-integrity.c btrfs: rename __btrfs_map_block to btrfs_map_block 2023-06-19 13:59:34 +02:00
check-integrity.h
compression.c btrfs: make btrfs_compressed_bioset static 2023-06-19 17:01:44 +02:00
compression.h btrfs: pass an ordered_extent to btrfs_submit_compressed_write 2023-06-19 13:59:36 +02:00
ctree.c btrfs: replace BUG_ON() at split_item() with proper error handling 2023-06-19 13:59:39 +02:00
ctree.h btrfs: fix infinite directory reads 2023-08-14 16:17:37 +02:00
defrag.c btrfs: drop gfp from parameter extent state helpers 2023-06-19 13:59:30 +02:00
defrag.h btrfs: move defrag related prototypes to their own header 2022-12-05 18:00:46 +01:00
delalloc-space.c btrfs: count extents before taking inode's spinlock when reserving metadata 2023-04-17 18:01:19 +02:00
delalloc-space.h btrfs: move delalloc space related prototypes to delalloc-space.h 2022-12-05 18:00:44 +01:00
delayed-inode.c btrfs: remove pointless empty list check when reading delayed dir indexes 2023-08-21 14:54:48 +02:00
delayed-inode.h btrfs: fix infinite directory reads 2023-08-14 16:17:37 +02:00
delayed-ref.c btrfs: use a single switch statement when initializing delayed ref head 2023-06-19 13:59:32 +02:00
delayed-ref.h btrfs: use bool type for delayed ref head fields that are used as booleans 2023-06-19 13:59:32 +02:00
dev-replace.c btrfs: make find_first_extent_bit() return a boolean 2023-08-21 14:52:12 +02:00
dev-replace.h btrfs: move dev-replace prototypes into dev-replace.h 2022-12-05 18:00:47 +01:00
dir-item.c btrfs: move dir-item prototypes into dir-item.h 2022-12-05 18:00:46 +01:00
dir-item.h btrfs: move dir-item prototypes into dir-item.h 2022-12-05 18:00:46 +01:00
discard.c btrfs: unexport btrfs_run_discard_work and make it static 2023-06-19 13:59:25 +02:00
discard.h btrfs: unexport btrfs_run_discard_work and make it static 2023-06-19 13:59:25 +02:00
disk-io.c btrfs: drop redundant check to use fs_devices::metadata_uuid 2023-08-21 14:54:48 +02:00
disk-io.h btrfs: make btrfs_cleanup_fs_roots() static 2023-08-21 14:52:18 +02:00
export.c btrfs: move super_block specific helpers into super.h 2022-12-05 18:00:47 +01:00
export.h btrfs: simplify generation check in btrfs_get_dentry 2022-12-05 18:00:41 +01:00
extent-io-tree.c btrfs: make find_first_extent_bit() return a boolean 2023-08-21 14:52:12 +02:00
extent-io-tree.h btrfs: make find_first_extent_bit() return a boolean 2023-08-21 14:52:12 +02:00
extent-tree.c btrfs: remove v0 extent handling 2023-08-21 14:54:48 +02:00
extent-tree.h btrfs: wait on uncached block groups on every allocation loop 2023-08-21 14:54:47 +02:00
extent_io.c btrfs: zoned: defer advancing meta write pointer 2023-08-21 14:52:19 +02:00
extent_io.h btrfs: zoned: introduce block group context to btrfs_eb_write_context 2023-08-21 14:52:19 +02:00
extent_map.c btrfs: fix incorrect splitting in btrfs_drop_extent_map_range 2023-08-18 14:38:10 +02:00
extent_map.h btrfs: pass the new logical address to split_extent_map 2023-06-19 13:59:33 +02:00
file-item.c btrfs: scrub: avoid unnecessary csum tree search preparing stripes 2023-08-21 14:54:48 +02:00
file-item.h btrfs: scrub: avoid unnecessary csum tree search preparing stripes 2023-08-21 14:54:48 +02:00
file.c btrfs: use LIST_HEAD() to initialize the list_head 2023-08-21 14:54:46 +02:00
file.h btrfs: use cached state when looking for delalloc ranges with fiemap 2022-12-05 18:00:56 +01:00
free-space-cache.c btrfs: zoned: no longer count fresh BG region as zone unusable 2023-08-21 14:52:19 +02:00
free-space-cache.h btrfs: move btrfs_check_trunc_cache_free_space into block-rsv.c 2023-06-19 13:59:24 +02:00
free-space-tree.c btrfs: rename add_new_free_space() to btrfs_add_new_free_space() 2023-08-21 14:52:12 +02:00
free-space-tree.h btrfs: make clear_cache mount option to rebuild FST without disabling it 2023-05-10 14:51:27 +02:00
fs.c btrfs: sysfs: update fs features directory asynchronously 2023-02-13 17:50:35 +01:00
fs.h btrfs: zoned: activate metadata block group on write time 2023-08-21 14:52:19 +02:00
inode-item.c btrfs: remove obsolete delayed ref throttling logic when truncating items 2023-04-17 18:01:19 +02:00
inode-item.h btrfs: move split_flags/combine_flags helpers to inode-item.h 2023-06-19 13:59:25 +02:00
inode.c btrfs: copy dir permission and time when creating a stub subvolume 2023-08-21 14:54:48 +02:00
ioctl.c for-6.5/block-2023-06-23 2023-06-26 12:47:20 -07:00
ioctl.h fs: port ->fileattr_set() to pass mnt_idmap 2023-01-19 09:24:27 +01:00
Kconfig btrfs: deprecate integrity checker feature 2023-08-21 14:52:13 +02:00
locking.c btrfs: add block-group tree to lockdep classes 2023-06-19 13:59:35 +02:00
locking.h btrfs: locking: use atomic for DREW lock writers 2023-04-17 18:01:17 +02:00
lru_cache.c btrfs: send: cache utimes operations for directories if possible 2023-02-15 19:38:50 +01:00
lru_cache.h btrfs: remove btrfs_lru_cache_is_full() inline function 2023-04-17 18:01:18 +02:00
lzo.c btrfs: disable allocation warnings for compression workspaces 2023-06-19 13:59:34 +02:00
Makefile btrfs: send: genericize the backref cache to allow it to be reused 2023-02-13 17:50:35 +01:00
messages.c btrfs: remove v0 extent handling 2023-08-21 14:54:48 +02:00
messages.h btrfs: remove v0 extent handling 2023-08-21 14:54:48 +02:00
misc.h btrfs: export bitmap_test_range_all_{set,zero} 2023-06-19 13:59:22 +02:00
ordered-data.c btrfs: use LIST_HEAD() to initialize the list_head 2023-08-21 14:54:46 +02:00
ordered-data.h btrfs: add a btrfs_finish_ordered_extent helper 2023-06-19 13:59:37 +02:00
orphan.c btrfs: move orphan prototypes into orphan.h 2022-12-05 18:00:47 +01:00
orphan.h btrfs: move orphan prototypes into orphan.h 2022-12-05 18:00:47 +01:00
print-tree.c btrfs: remove v0 extent handling 2023-08-21 14:54:48 +02:00
print-tree.h btrfs: print-tree: pass const extent buffer pointer 2023-06-19 13:59:22 +02:00
props.c btrfs: move super_block specific helpers into super.h 2022-12-05 18:00:47 +01:00
props.h btrfs: make module init/exit match their sequence 2022-12-05 18:00:40 +01:00
qgroup.c btrfs: avoid start and commit empty transaction when flushing qgroups 2023-08-21 14:52:18 +02:00
qgroup.h btrfs: sink gfp_t parameter to btrfs_qgroup_trace_extent 2022-12-05 18:00:43 +01:00
raid56.c btrfs: scrub: avoid unnecessary csum tree search preparing stripes 2023-08-21 14:54:48 +02:00
raid56.h btrfs: raid56: remove unused BTRFS_RBIO_REBUILD_MISSING 2023-08-21 14:52:12 +02:00
rcu-string.h btrfs: replace strncpy() with strscpy() 2022-12-05 18:00:59 +01:00
ref-verify.c btrfs: move accessor helpers into accessors.h 2022-12-05 18:00:42 +01:00
ref-verify.h
reflink.c btrfs: pass btrfs_inode to btrfs_inode_unlock 2022-12-05 18:00:53 +01:00
reflink.h
relocation.c btrfs: remove v0 extent handling 2023-08-21 14:54:48 +02:00
relocation.h btrfs: pass an ordered_extent to btrfs_reloc_clone_csums 2023-06-19 13:59:36 +02:00
root-tree.c btrfs: move orphan prototypes into orphan.h 2022-12-05 18:00:47 +01:00
root-tree.h btrfs: move root tree prototypes to their own header 2022-12-05 18:00:44 +01:00
scrub.c btrfs: scrub: fix grouping of read IO 2023-08-21 14:54:49 +02:00
scrub.h btrfs: scrub: remove scrub_bio structure 2023-04-17 18:01:24 +02:00
send.c btrfs: use LIST_HEAD() to initialize the list_head 2023-08-21 14:54:46 +02:00
send.h btrfs: send add define for v2 buffer size 2022-12-05 18:00:41 +01:00
space-info.c btrfs: zoned: re-enable metadata over-commit for zoned mode 2023-08-21 14:52:19 +02:00
space-info.h btrfs: update documentation for BTRFS_RESERVE_FLUSH_EVICT flush method 2023-04-17 18:01:18 +02:00
subpage.c btrfs: stop setting PageError in the data I/O path 2023-06-19 13:59:35 +02:00
subpage.h btrfs: stop setting PageError in the data I/O path 2023-06-19 13:59:35 +02:00
super.c btrfs: deprecate integrity checker feature 2023-08-21 14:52:13 +02:00
super.h btrfs: move super_block specific helpers into super.h 2022-12-05 18:00:47 +01:00
sysfs.c btrfs: sysfs: show if ACL support has been compiled in 2023-08-21 14:52:12 +02:00
sysfs.h btrfs: sysfs: update fs features directory asynchronously 2023-02-13 17:50:35 +01:00
transaction.c btrfs: update comment for btrfs_join_transaction_nostart() 2023-08-21 14:52:17 +02:00
transaction.h btrfs: don't hold an extra reference for redirtied buffers 2023-06-19 13:59:26 +02:00
tree-checker.c for-6.5-rc5-tag 2023-08-12 13:28:55 -07:00
tree-checker.h btrfs: move btrfs_verify_level_key into tree-checker.c 2023-06-19 13:59:25 +02:00
tree-log.c btrfs: use LIST_HEAD() to initialize the list_head 2023-08-21 14:54:46 +02:00
tree-log.h btrfs: change for_rename argument of btrfs_record_unlink_dir() to bool 2023-06-19 13:59:26 +02:00
tree-mod-log.c btrfs: avoid tree mod log ENOMEM failures when we don't need to log 2023-06-19 13:59:38 +02:00
tree-mod-log.h btrfs: fix SPDX comment in tree-mod-log.h 2022-12-05 18:00:48 +01:00
ulist.c btrfs: constify ulist parameter of ulist_next() 2022-12-05 18:00:50 +01:00
ulist.h btrfs: constify ulist parameter of ulist_next() 2022-12-05 18:00:50 +01:00
uuid-tree.c btrfs: move uuid tree prototypes to uuid-tree.h 2022-12-05 18:00:46 +01:00
uuid-tree.h btrfs: move uuid tree prototypes to uuid-tree.h 2022-12-05 18:00:46 +01:00
verity.c fsverity: pass pos and size to ->write_merkle_tree_block 2023-01-01 15:46:48 -08:00
verity.h btrfs: move verity prototypes into verity.h 2022-12-05 18:00:47 +01:00
volumes.c btrfs: simplify memcpy either of metadata_uuid or fsid 2023-08-21 14:54:48 +02:00
volumes.h btrfs: add a helper to read the superblock metadata_uuid 2023-08-21 14:54:48 +02:00
xattr.c fs: drop unused posix acl handlers 2023-03-06 09:57:12 +01:00
xattr.h
zlib.c btrfs: disable allocation warnings for compression workspaces 2023-06-19 13:59:34 +02:00
zoned.c btrfs: zoned: do not zone finish data relocation block group 2023-08-21 14:54:47 +02:00
zoned.h btrfs: zoned: reserve zones for an active metadata/system block group 2023-08-21 14:52:19 +02:00
zstd.c btrfs: disable allocation warnings for compression workspaces 2023-06-19 13:59:34 +02:00