Re: fail_last_dev and FailFast/LastDev flag incompatibility

[Guoqing Jiang <guoqing.jiang@linux.dev>]

On 2/9/22 5:40 PM, Mariusz Tkaczyk wrote:
> Hi All,
> During my work under failed arrays handling[1]*improvements*,

Sorry, I disagree, will comment your new version later.

> I discovered potential issue with "failfast" and metadata writes. In
> commit message[2] Neil mentioned that:
> "If we get a failure writing metadata but the device doesn't
> fail, it must be the last device so we re-write without
> FAILFAST".
>
> Obviously, this is not true for RAID456 (again)[1] but it is also not
> true for RAID1 and RAID10 with "fail_las_dev"[3] functionality enabled.
>
> I did a quick check and can see that setter for "LastDev" flag is
> called if "Faulty" on device is not set. I proposed some changes in the
> area in my patchset[4] but after discussion we decided to drop changes
> here. Current approach is not correct for all branches, so my proposal
> is to change:
>
> diff --git a/drivers/md/md.c b/drivers/md/md.c
> index 7b024912f1eb..3daec14ef6b2 100644
> --- a/drivers/md/md.c
> +++ b/drivers/md/md.c
> @@ -931,7 +931,7 @@ static void super_written(struct bio *bio)
>                  pr_err("md: %s gets error=%d\n", __func__,
>                         blk_status_to_errno(bio->bi_status));
>                  md_error(mddev, rdev);
> -               if (!test_bit(Faulty, &rdev->flags)
> +               if (test_bit(MD_BROKEN, mddev->flag)
>                      && (bio->bi_opf & MD_FAILFAST)) {
>                          set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
>                          set_bit(LastDev, &rdev->flags);

IIUC, there is no problem with checking Faulty since super_written is
against rdev while MD_BROKEN is supposed to mean array is broken.

Thanks,
Guoqing