dm raid: fix restoring of failed devices regression
authorHeinz Mauelshagen <heinzm@redhat.com>
Wed, 10 Aug 2016 00:45:59 +0000 (02:45 +0200)
committerMike Snitzer <snitzer@redhat.com>
Tue, 16 Aug 2016 20:21:31 +0000 (16:21 -0400)
'lvchange --refresh RaidLV' causes a mapped device suspend/resume cycle
aiming at device restore and resync after transient device failures.  This
failed because flag RT_FLAG_RS_RESUMED was always cleared in the suspend path,
thus the device restore wasn't performed in the resume path.

Solve by removing RT_FLAG_RS_RESUMED from the suspend path and resume
unconditionally.  Also, remove superfluous comment from raid_resume().

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
drivers/md/dm-raid.c

index 07d17287fa4a9c4f00faff981408434de2912df1..81ec772b1cc91067d1dc7aa4f9b9774b04c5e380 100644 (file)
@@ -3382,11 +3382,10 @@ static void raid_postsuspend(struct dm_target *ti)
 {
        struct raid_set *rs = ti->private;
 
-       if (test_and_clear_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
-               if (!rs->md.suspended)
-                       mddev_suspend(&rs->md);
-               rs->md.ro = 1;
-       }
+       if (!rs->md.suspended)
+               mddev_suspend(&rs->md);
+
+       rs->md.ro = 1;
 }
 
 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
@@ -3606,25 +3605,15 @@ static void raid_resume(struct dm_target *ti)
                 * devices are reachable again.
                 */
                attempt_restore_of_faulty_devices(rs);
-       } else {
-               mddev->ro = 0;
-               mddev->in_sync = 0;
+       }
 
-               /*
-                * When passing in flags to the ctr, we expect userspace
-                * to reset them because they made it to the superblocks
-                * and reload the mapping anyway.
-                *
-                * -> only unfreeze recovery in case of a table reload or
-                *    we'll have a bogus recovery/reshape position
-                *    retrieved from the superblock by the ctr because
-                *    the ongoing recovery/reshape will change it after read.
-                */
-               clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
+       mddev->ro = 0;
+       mddev->in_sync = 0;
 
-               if (mddev->suspended)
-                       mddev_resume(mddev);
-       }
+       clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
+
+       if (mddev->suspended)
+               mddev_resume(mddev);
 }
 
 static struct target_type raid_target = {