Re: interrupted resync not restarted properly?

[Neil Brown <neilb@suse.de>]

On Tue, 7 Dec 2010 15:37:00 -0500 "Dailey, Nate" <Nate.Dailey@stratus.com>
wrote:

> It seems to me that resuming an interrupted resync doesn't always work
> right... here's what I'm doing (kernel 2.6.36):
> 
> - start with a 2 disk raid1 with internal bitmap
> - fail/remove one disk and zero the superblock
> - add the disk to the raid1
> - before resync completes, fail/remove the disk again
> - re-add the disk again
> 
> For version 0 superblocks, this works the way I'd expect: on adding the
> disk the second time, the resync continues (or restarts from the
> beginning, not sure).
> 
> But for version 1 superblocks, on adding the disk the second time, the
> resync completes immediately, leaving some part of the array
> out-of-sync.
> 
> Should there be something in the v1 superblock to prevent this?
> 
> If the raid1 is stopped in the middle of the resync (instead of removing
> the target disk) the resync is resumed correctly on re-assembly with
> both devices.
> 
> Nate
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Thanks for the report.
That is pretty bad behaviour.

The following is a patch that I plan to submit to -linus and -stable.  It
doesn't make it work quite as I would like (that would be a lot more code)
but it makes it a lot safer.

Thanks,
NeilBrown

--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -5170,7 +5174,10 @@ static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
 		} else
 			super_types[mddev->major_version].
 				validate_super(mddev, rdev);
-		rdev->saved_raid_disk = rdev->raid_disk;
+		if (test_bit(In_sync, &rdev->flags))
+			rdev->saved_raid_disk = rdev->raid_disk;
+		else
+			rdev->saved_raid_disk = -1;
 
 		clear_bit(In_sync, &rdev->flags); /* just to be sure */
 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))