From: Chi Zhiling <chizhiling@kylinos.cn>

exfat_reconfigure() does not hold s_lock while calling
sync_filesystem(). Therefore, the filesystem can become dirty again
between the sync and clearing the volume dirty flag, leaving a dirty
filesystem with a clean volume flag.

Holding s_lock across sync_filesystem() is not an option because it can
lead to a deadlock: writeback takes s_lock in exfat_write_inode().
Instead, clear the volume dirty flag only when remounting the filesystem
read-only, where no writer can modify the filesystem after the sync.

For a normal read-only remount, reconfigure_super() calls
sb_prepare_remount_readonly() before ->reconfigure(), which returns
-EBUSY if any writer is still active. Afterwards
sb_start_ro_state_change() sets sb->s_readonly_remount, so
mnt_get_write_access() fails with -EROFS and no new writer can start
until the reconfiguration finishes. Therefore no writer can race with
the sync and the clearing.

Forced remounts are different: with SB_FORCE, reconfigure_super() skips
sb_prepare_remount_readonly(), so the writers active at that moment are
not blocked and can still dirty the filesystem after the sync. Do not
clear the volume dirty flag in this case.

Signed-off-by: Chi Zhiling <chizhiling@kylinos.cn>
---
 fs/exfat/super.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/fs/exfat/super.c b/fs/exfat/super.c
index a9ea36ba2693..268732571837 100644
--- a/fs/exfat/super.c
+++ b/fs/exfat/super.c
@@ -775,9 +775,12 @@ static int exfat_reconfigure(struct fs_context *fc)
 	fc->sb_flags |= SB_NODIRATIME;
 
 	sync_filesystem(sb);
-	mutex_lock(&sbi->s_lock);
-	exfat_clear_volume_dirty(sb);
-	mutex_unlock(&sbi->s_lock);
+
+	if ((fc->sb_flags & (SB_FORCE | SB_RDONLY)) == SB_RDONLY) {
+		mutex_lock(&sbi->s_lock);
+		exfat_clear_volume_dirty(sb);
+		mutex_unlock(&sbi->s_lock);
+	}
 
 	if (new_opts->allow_utime == (unsigned short)-1)
 		new_opts->allow_utime = ~new_opts->fs_dmask & 0022;
-- 
2.53.0

> diff --git a/fs/exfat/super.c b/fs/exfat/super.c
> index a9ea36ba2693..268732571837 100644
> --- a/fs/exfat/super.c
> +++ b/fs/exfat/super.c
> @@ -775,9 +775,12 @@ static int exfat_reconfigure(struct fs_context *fc)
>         fc->sb_flags |= SB_NODIRATIME;
>
>         sync_filesystem(sb);
> -       mutex_lock(&sbi->s_lock);
> -       exfat_clear_volume_dirty(sb);
> -       mutex_unlock(&sbi->s_lock);
> +
> +       if ((fc->sb_flags & (SB_FORCE | SB_RDONLY)) == SB_RDONLY) {
Could we also add !sb_rdonly(sb) here?
            if ((fc->sb_flags & (SB_FORCE | SB_RDONLY)) == SB_RDONLY &&
                 !sb_rdonly(sb))
Thanks.


> +               mutex_lock(&sbi->s_lock);
> +               exfat_clear_volume_dirty(sb);
> +               mutex_unlock(&sbi->s_lock);
> +       }
>
>         if (new_opts->allow_utime == (unsigned short)-1)
>                 new_opts->allow_utime = ~new_opts->fs_dmask & 0022;
> --
> 2.53.0
>

On 9/4/26 7:28 AM, Namjae Jeon wrote:
>> diff --git a/fs/exfat/super.c b/fs/exfat/super.c
>> index a9ea36ba2693..268732571837 100644
>> --- a/fs/exfat/super.c
>> +++ b/fs/exfat/super.c
>> @@ -775,9 +775,12 @@ static int exfat_reconfigure(struct fs_context *fc)
>>          fc->sb_flags |= SB_NODIRATIME;
>>
>>          sync_filesystem(sb);
>> -       mutex_lock(&sbi->s_lock);
>> -       exfat_clear_volume_dirty(sb);
>> -       mutex_unlock(&sbi->s_lock);
>> +
>> +       if ((fc->sb_flags & (SB_FORCE | SB_RDONLY)) == SB_RDONLY) {
> Could we also add !sb_rdonly(sb) here?
>              if ((fc->sb_flags & (SB_FORCE | SB_RDONLY)) == SB_RDONLY &&
>                   !sb_rdonly(sb))

Yes, this way we do not call exfat_clear_volume_dirty() when remounting 
from RO to RO. I will update it in v2.

Thanks.