In HMZONED, btrfs use per-Block Group zone_io_lock to serialize the data
write IOs or use per-FS hmzoned_meta_io_lock to serialize the metadata
write IOs.
Even with these serialization, write bios sent from
{btree,btrfs}_write_cache_pages can be reordered by async checksum workers
as these workers are per CPU and not per zone.
To preserve write BIO ordering, we can disable async checksum on HMZONED.
This does not result in lower performance with HDDs as a single CPU core is
fast enough to do checksum for a single zone write stream with the maximum
possible bandwidth of the device. If multiple zones are being written
simultaneously, HDD seek overhead lowers the achievable maximum bandwidth,
resulting again in a per zone checksum serialization not affecting
performance.
Besides, this commit disable async_submit in
btrfs_submit_compressed_write() for the same reason. This part will be
unnecessary once btrfs get the "btrfs: fix cgroup writeback support"
series.
Signed-off-by: Naohiro Aota <naohiro.aota@xxxxxxx>
---
fs/btrfs/compression.c | 5 +++--
fs/btrfs/disk-io.c | 2 ++
fs/btrfs/inode.c | 9 ++++++---
3 files changed, 11 insertions(+), 5 deletions(-)
diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
index 60c47b417a4b..058dea5e432f 100644
--- a/fs/btrfs/compression.c
+++ b/fs/btrfs/compression.c
@@ -322,6 +322,7 @@ blk_status_t btrfs_submit_compressed_write(struct inode *inode, u64 start,
struct block_device *bdev;
blk_status_t ret;
int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
+ int async_submit = !btrfs_fs_incompat(fs_info, HMZONED);
WARN_ON(!PAGE_ALIGNED(start));
cb = kmalloc(compressed_bio_size(fs_info, compressed_len), GFP_NOFS);
@@ -377,7 +378,7 @@ blk_status_t btrfs_submit_compressed_write(struct inode *inode, u64 start,
BUG_ON(ret); /* -ENOMEM */
}
- ret = btrfs_map_bio(fs_info, bio, 0, 1);
+ ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
if (ret) {
bio->bi_status = ret;
bio_endio(bio);
@@ -408,7 +409,7 @@ blk_status_t btrfs_submit_compressed_write(struct inode *inode, u64 start,
BUG_ON(ret); /* -ENOMEM */
}
- ret = btrfs_map_bio(fs_info, bio, 0, 1);
+ ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
if (ret) {
bio->bi_status = ret;
bio_endio(bio);
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index a9632e455eb5..9bae051f9f44 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -873,6 +873,8 @@ static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
static int check_async_write(struct btrfs_fs_info *fs_info,
struct btrfs_inode *bi)
{
+ if (btrfs_fs_incompat(fs_info, HMZONED))
+ return 0;
if (atomic_read(&bi->sync_writers))
return 0;
if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 95f4ce8ac8d0..bb0ae3107e60 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -2075,7 +2075,8 @@ static blk_status_t btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
blk_status_t ret = 0;
int skip_sum;
- int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
+ int async = !atomic_read(&BTRFS_I(inode)->sync_writers) &&
+ !btrfs_fs_incompat(fs_info, HMZONED);
skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
@@ -8383,7 +8384,8 @@ static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
/* Check btrfs_submit_bio_hook() for rules about async submit. */
if (async_submit)
- async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
+ async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers) &&
+ !btrfs_fs_incompat(fs_info, HMZONED);
if (!write) {
ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
@@ -8448,7 +8450,8 @@ static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
}
/* async crcs make it difficult to collect full stripe writes. */
- if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
+ if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK ||
+ btrfs_fs_incompat(fs_info, HMZONED))
async_submit = 0;
else
async_submit = 1;
--
2.23.0
