[PATCH v2 02/12] block: prevent race condition on bi_status in __bio_chain_endio

[Caleb Sander Mateos]

On Fri, Nov 28, 2025 at 12:34 AM zhangshida <starzhangzsd at gmail.com> wrote:
>
> From: Shida Zhang <zhangshida at kylinos.cn>
>
> Andreas point out that multiple completions can race setting
> bi_status.
>
> The check (parent->bi_status) and the subsequent write are not an
> atomic operation. The value of parent->bi_status could have changed
> between the time you read it for the if check and the time you write
> to it. So we use cmpxchg to fix the race, as suggested by Christoph.
>
> Suggested-by: Andreas Gruenbacher <agruenba at redhat.com>
> Suggested-by: Christoph Hellwig <hch at infradead.org>
> Signed-off-by: Shida Zhang <zhangshida at kylinos.cn>
> ---
>  block/bio.c | 7 +++++--
>  1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/block/bio.c b/block/bio.c
> index 55c2c1a0020..aa43435c15f 100644
> --- a/block/bio.c
> +++ b/block/bio.c
> @@ -313,9 +313,12 @@ EXPORT_SYMBOL(bio_reset);
>  static struct bio *__bio_chain_endio(struct bio *bio)
>  {
>         struct bio *parent = bio->bi_private;
> +       blk_status_t *status = &parent->bi_status;

nit: this variable seems unnecessary, just use &parent->bi_status
directly in the one place it's needed?

Best,
Caleb

> +       blk_status_t new_status = bio->bi_status;
> +
> +       if (new_status != BLK_STS_OK)
> +               cmpxchg(status, BLK_STS_OK, new_status);
>
> -       if (bio->bi_status && !parent->bi_status)
> -               parent->bi_status = bio->bi_status;
>         bio_put(bio);
>         return parent;
>  }
> --
> 2.34.1
>
>