Re: [PATCH 2/2] md/raid10: handle replacement devices in fix_recovery_read_error

[Yu Kuai]

Hi,

在 2023/07/07 16:33, Song Liu 写道:
> On Tue, Jun 27, 2023 at 11:42 AM <linan666@xxxxxxxxxxxxxxx> wrote:
> > From: Li Nan <linan122@xxxxxxxxxx>
> >
> > In fix_recovery_read_error(), the handling of replacement devices is
> > missing. Add it. If io error is from replacement, error this device
> > directly. If io error is from other device, just set badblocks for
> > replacement.
> >
> > Signed-off-by: Li Nan <linan122@xxxxxxxxxx>
> > ---
> >   drivers/md/raid10.c | 10 ++++++++--
> >   1 file changed, 8 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
> > index 5105273f60e9..6d9025089455 100644
> > --- a/drivers/md/raid10.c
> > +++ b/drivers/md/raid10.c
> > @@ -2551,7 +2551,7 @@ static void fix_recovery_read_error(struct r10bio *r10_bio)
> >
> >          while (sectors) {
> >                  int s = sectors;
> > -               struct md_rdev *rdev;
> > +               struct md_rdev *rdev, *repl;
> >                  sector_t addr;
> >                  int ok;
> >
> > @@ -2559,6 +2559,7 @@ static void fix_recovery_read_error(struct r10bio *r10_bio)
> >                          s = PAGE_SIZE >> 9;
> >
> >                  rdev = conf->mirrors[dr].rdev;
> > +               repl = conf->mirrors[dw].replacement;
> >                  addr = r10_bio->devs[0].addr + sect,
> >                  ok = sync_page_io(rdev,
> >                                    addr,
> > @@ -2580,6 +2581,9 @@ static void fix_recovery_read_error(struct r10bio *r10_bio)
> >                                          set_bit(MD_RECOVERY_NEEDED,
> >                                                  &rdev->mddev->recovery);
> >                          }
> > +                       if (repl && !sync_page_io(repl, addr, s << 9,
> > +                           pages[idx], REQ_OP_WRITE, false))
> > +                               md_error(mddev, repl);
> >                  }
> >                  if (!ok) {
> >                          /* We don't worry if we cannot set a bad block -
> > @@ -2592,7 +2596,9 @@ static void fix_recovery_read_error(struct r10bio *r10_bio)
> >                                  /* need bad block on destination too */
> >                                  rdev = conf->mirrors[dw].rdev;
> >                                  addr = r10_bio->devs[1].addr + sect;
> > -                               if (!rdev_set_badblocks(rdev, addr, s, 0)) {
> > +                               if (!rdev_set_badblocks(rdev, addr, s, 0) ||
> > +                                   (repl &&
> > +                                   !rdev_set_badblocks(repl, addr, s, 0))) {
>
> Do we really want this in the if () statement? Shall we always set
> badblock on both rdev and repl?

I think this is wrong to set repl badblocks inside this, because if
setting badblocks for rdev failed, repl is still not handled.

By the way, I think it's better to at least try to read from all
possible copies before setting badblocks for repl.

Thanks,
Kuai
> Thanks,
> Song
>
> >                                          /* just abort the recovery */
> >                                          pr_notice("md/raid10:%s: recovery aborted due to read error\n",
> >                                                    mdname(mddev));
> > --
> > 2.39.2
> .