[PATCH] dax: fix assertion in dax_iomap_rw()

[PATCH] dax: fix assertion in dax_iomap_rw()

[Jan Kara]

dax_iomap_rw() asserts that inode lock is held when reading from it. The
assert triggers on erofs as it indeed doesn't hold any locks in this
case - naturally because there's nothing to race against when the
filesystem is read-only. Check the locking only if the filesystem is
actually writeable.

Reported-by: syzbot+47680984f2d4969027ea@syzkaller.appspotmail.com
Signed-off-by: Jan Kara <jack@suse.cz>
---
 fs/dax.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/dax.c b/fs/dax.c
index 20ecf652c129..187f8c325744 100644
--- a/fs/dax.c
+++ b/fs/dax.c
@@ -1752,7 +1752,7 @@ dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter,
 	if (iov_iter_rw(iter) == WRITE) {
 		lockdep_assert_held_write(&iomi.inode->i_rwsem);
 		iomi.flags |= IOMAP_WRITE;
-	} else {
+	} else if (!IS_RDONLY(iomi.inode)) {
 		lockdep_assert_held(&iomi.inode->i_rwsem);
 	}
 
-- 
2.51.0

Re: [PATCH] dax: fix assertion in dax_iomap_rw()

[Gao Xiang]

Hi Jan,

On 2025/10/2 16:13, Jan Kara wrote:
> dax_iomap_rw() asserts that inode lock is held when reading from it. The
> assert triggers on erofs as it indeed doesn't hold any locks in this
> case - naturally because there's nothing to race against when the
> filesystem is read-only. Check the locking only if the filesystem is
> actually writeable.
> 
> Reported-by: syzbot+47680984f2d4969027ea@syzkaller.appspotmail.com
> Signed-off-by: Jan Kara <jack@suse.cz>
> ---
>   fs/dax.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/dax.c b/fs/dax.c
> index 20ecf652c129..187f8c325744 100644
> --- a/fs/dax.c
> +++ b/fs/dax.c
> @@ -1752,7 +1752,7 @@ dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter,
>   	if (iov_iter_rw(iter) == WRITE) {
>   		lockdep_assert_held_write(&iomi.inode->i_rwsem);
>   		iomi.flags |= IOMAP_WRITE;
> -	} else {
> +	} else if (!IS_RDONLY(iomi.inode)) {

Thanks, Yuezheng also wrote a similiar patch
days ago (but it seems he didn't cc more related
people),
https://lore.kernel.org/r/20250930054256.2461984-2-Yuezhang.Mo@sony.com

both patches look good to me, thanks for the fix.

Thanks,
Gao Xiang

>   		lockdep_assert_held(&iomi.inode->i_rwsem);
>   	}
>