2023-11-01 15:47:50

by Ojaswin Mujoo

Subject: [PATCH] ext4: Enable dioread_nolock as default for bs < ps case

dioread_nolock was originally disabled as a default option for bs < ps
scenarios due to a data corruption issue. Since then, we've had some
fixes in this area which address such issues. Enable dioread_nolock by
default and remove the experimental warning message for bs < ps path.

dioread for bs < ps has been tested on a 64k pagesize machine using:

kvm-xfstest -C 3 -g auto

with the following configs:

64k adv bigalloc_4k bigalloc_64k data_journal encrypt
dioread_nolock dioread_nolock_4k ext3 ext3conv nojournal

And no new regressions were seen compared to baseline kernel.

Suggested-by: Ritesh Harjani (IBM) <[email protected]>
Signed-off-by: Ojaswin Mujoo <[email protected]>
Reviewed-by: Ritesh Harjani (IBM) <[email protected]>
---
fs/ext4/super.c | 11 +----------
1 file changed, 1 insertion(+), 10 deletions(-)

diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index d062383ea50e..99c8ae97112c 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -2785,15 +2785,6 @@ static int ext4_check_opt_consistency(struct fs_context *fc,
return -EINVAL;
}

- if (ctx_test_mount_opt(ctx, EXT4_MOUNT_DIOREAD_NOLOCK)) {
- int blocksize =
- BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
- if (blocksize < PAGE_SIZE)
- ext4_msg(NULL, KERN_WARNING, "Warning: mounting with an "
- "experimental mount option 'dioread_nolock' "
- "for blocksize < PAGE_SIZE");
- }
-
err = ext4_check_test_dummy_encryption(fc, sb);
if (err)
return err;
@@ -4402,7 +4393,7 @@ static void ext4_set_def_opts(struct super_block *sb,
((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
set_opt(sb, DELALLOC);

- if (sb->s_blocksize == PAGE_SIZE)
+ if (sb->s_blocksize <= PAGE_SIZE)
set_opt(sb, DIOREAD_NOLOCK);
}

--
2.39.3

2024-01-09 02:53:50

by Theodore Ts'o

Subject: Re: [PATCH] ext4: Enable dioread_nolock as default for bs < ps case

On Wed, 01 Nov 2023 21:17:17 +0530, Ojaswin Mujoo wrote:
> dioread_nolock was originally disabled as a default option for bs < ps
> scenarios due to a data corruption issue. Since then, we've had some
> fixes in this area which address such issues. Enable dioread_nolock by
> default and remove the experimental warning message for bs < ps path.
>
> dioread for bs < ps has been tested on a 64k pagesize machine using:
>
> [...]

Applied, thanks!

[1/1] ext4: Enable dioread_nolock as default for bs < ps case
commit: e89fdcc425b6feea4dfb33877e9256757905d763

Best regards,
--
Theodore Ts'o <[email protected]>