md/raid10: fix bug which causes all RAID10 reshapes to move no data.
authorNeilBrown <neilb@suse.de>
Thu, 4 Jul 2013 06:41:53 +0000 (16:41 +1000)
committerNeilBrown <neilb@suse.de>
Thu, 4 Jul 2013 06:42:57 +0000 (16:42 +1000)
The recent comment:
commit 7e83ccbecd608b971f340e951c9e84cd0343002f
    md/raid10: Allow skipping recovery when clean arrays are assembled

Causes raid10 to skip a recovery in certain cases where it is safe to
do so.  Unfortunately it also causes a reshape to be skipped which is
never safe.  The result is that an attempt to reshape a RAID10 will
appear to complete instantly, but no data will have been moves so the
array will now contain garbage.
(If nothing is written, you can recovery by simple performing the
reverse reshape which will also complete instantly).

Bug was introduced in 3.10, so this is suitable for 3.10-stable.

Cc: stable@vger.kernel.org (3.10)
Cc: Martin Wilck <mwilck@arcor.de>
Signed-off-by: NeilBrown <neilb@suse.de>
drivers/md/raid10.c

index 3480bf7c20d41b67a8d26057db8ff890126cf042..cd066b63bdafea61ae1175428f2c261575cfd4d1 100644 (file)
@@ -2931,14 +2931,13 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
         */
        if (mddev->bitmap == NULL &&
            mddev->recovery_cp == MaxSector &&
+           mddev->reshape_position == MaxSector &&
+           !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
            !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
+           !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
            conf->fullsync == 0) {
                *skipped = 1;
-               max_sector = mddev->dev_sectors;
-               if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
-                   test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
-                       max_sector = mddev->resync_max_sectors;
-               return max_sector - sector_nr;
+               return mddev->dev_sectors - sector_nr;
        }
 
  skipped: