lkml.org 
[lkml]   [2003]   [Sep]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: 2.6 md raid5 disk faulty marking bug was: md: bug in file raid5.c, line 1909 in 2.4.22-pre7
    On Tue, Aug 26, 2003 at 09:20:05AM +1000, Neil Brown wrote:
    > On Friday August 22, mfedyk@matchmail.com wrote:
    > > > As far as I can see, the 2.4 code would never set just MD_DISK_REMOVED
    > > > (though it really should cope with it). It is possible that the 2.6
    > > > code does. Has this array had 2.6 running on it? Does it have any
    > > > interesting history?
    > >
    > > Yes, it was running 2.6-test2-mm2 or so (don't remember exactly, I can check
    > > though if needed) previously, but I didn't notice any bug messages from
    > > there, and seeing that it was 2.4 I was surprised to see bug
    > > messages from md.
    >
    > The 2.4 code is very fragile. It can easily bug if the superblock
    > looks wrong in some obscure way.
    >
    > >
    > > Do you have any patches for 2.6 md? Right now this system is still in
    > > testing, and I'd like to help get this code path tested, and fixed.
    >
    > The following should fix it.
    > With this applied to 2.6, you can simply
    > start the array under 2.6
    > stop the array
    > reboot into 2.4
    > and it should be fine again.
    >
    > NeilBrown
    >
    > ==================================================
    > Fix md superblock incompatabilities with 2.4 kernels.
    >
    > 2.4 kernels are very fussy about some values in the superblock, and
    > 2.6 got them wrong. This fixes it.
    >

    Yes, I can confirm this fixes it for me. I applied it to
    2.6.0-test4-mm3-1, without any troubles.

    I tried rebuilding the array with 2.4 (zeroing the superblock in mdadm and
    everything), but that didn't fix it, so I rebuilt with 2.6 and the below
    patch, and do far so good. I'll be beating on this machine some more...

    Mike

    >
    > ----------- Diffstat output ------------
    > ./drivers/md/md.c | 13 ++++++-------
    > 1 files changed, 6 insertions(+), 7 deletions(-)
    >
    > diff ./drivers/md/md.c~current~ ./drivers/md/md.c
    > --- ./drivers/md/md.c~current~ 2003-08-24 08:07:18.000000000 +1000
    > +++ ./drivers/md/md.c 2003-08-26 09:11:39.000000000 +1000
    > @@ -638,14 +638,13 @@ static void super_90_sync(mddev_t *mddev
    > /* make rdev->sb match mddev data..
    > *
    > * 1/ zero out disks
    > - * 2/ Add info for each disk, keeping track of highest desc_nr
    > - * 3/ any empty disks < highest become removed
    > + * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
    > + * 3/ any empty disks < next_spare become removed
    > *
    > * disks[0] gets initialised to REMOVED because
    > * we cannot be sure from other fields if it has
    > * been initialised or not.
    > */
    > - int highest = 0;
    > int i;
    > int active=0, working=0,failed=0,spare=0,nr_disks=0;
    >
    > @@ -716,17 +715,17 @@ static void super_90_sync(mddev_t *mddev
    > spare++;
    > working++;
    > }
    > - if (rdev2->desc_nr > highest)
    > - highest = rdev2->desc_nr;
    > }
    >
    > - /* now set the "removed" bit on any non-trailing holes */
    > - for (i=0; i<highest; i++) {
    > + /* now set the "removed" and "faulty" bits on any missing devices */
    > + for (i=0 ; i < mddev->raid_disks ; i++) {
    > mdp_disk_t *d = &sb->disks[i];
    > if (d->state == 0 && d->number == 0) {
    > d->number = i;
    > d->raid_disk = i;
    > d->state = (1<<MD_DISK_REMOVED);
    > + d->state |= (1<<MD_DISK_FAULTY);
    > + failed++;
    > }
    > }
    > sb->nr_disks = nr_disks;
    >
    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2005-03-22 13:48    [W:0.025 / U:2.268 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site