Update recovery_offset even when external metadata is used.

The update of ->recovery_offset in sync_sbs is appropriate even then external
metadata is in use.  However sync_sbs is only called when native
metadata is used.

So move that update in to the top of md_update_sb (which is the only
caller of sync_sbs) before the test on ->external.

This moves the update out of ->write_lock protection, but those fields
only need ->reconfig_mutex protection which they still have.

Also move the test on ->persistent up to where ->external is set as
for metadata update purposes they are the same.

Clear MD_CHANGE_DEVS and MD_CHANGE_CLEAN as they can only be confusing
if ->external is set or ->persistent isn't.

Finally move the update of ->utime down as it is only relevent (like
the ->events update) for native metadata.

Signed-off-by: NeilBrown <neilb@suse.de>
Reported-by: "Kwolek, Adam" <adam.kwolek@intel.com>

diff --git a/drivers/md/md.c b/drivers/md/md.c
index 11567c7..c148b63 100644 (file)
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -2136,16 +2136,6 @@ static void sync_sbs(mddev_t * mddev, int nospares)
         * with the rest of the array)
         */
        mdk_rdev_t *rdev;
-
-       /* First make sure individual recovery_offsets are correct */
-       list_for_each_entry(rdev, &mddev->disks, same_set) {
-               if (rdev->raid_disk >= 0 &&
-                   mddev->delta_disks >= 0 &&
-                   !test_bit(In_sync, &rdev->flags) &&
-                   mddev->curr_resync_completed > rdev->recovery_offset)
-                               rdev->recovery_offset = mddev->curr_resync_completed;
-
-       }       
        list_for_each_entry(rdev, &mddev->disks, same_set) {
                if (rdev->sb_events == mddev->events ||
                    (nospares &&
@@ -2167,12 +2157,27 @@ static void md_update_sb(mddev_t * mddev, int force_change)
        int sync_req;
        int nospares = 0;
 
-       mddev->utime = get_seconds();
-       if (mddev->external)
-               return;
 repeat:
+       /* First make sure individual recovery_offsets are correct */
+       list_for_each_entry(rdev, &mddev->disks, same_set) {
+               if (rdev->raid_disk >= 0 &&
+                   mddev->delta_disks >= 0 &&
+                   !test_bit(In_sync, &rdev->flags) &&
+                   mddev->curr_resync_completed > rdev->recovery_offset)
+                               rdev->recovery_offset = mddev->curr_resync_completed;
+
+       }       
+       if (mddev->external || !mddev->persistent) {
+               clear_bit(MD_CHANGE_DEVS, &mddev->flags);
+               clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
+               wake_up(&mddev->sb_wait);
+               return;
+       }
+
        spin_lock_irq(&mddev->write_lock);
 
+       mddev->utime = get_seconds();
+
        set_bit(MD_CHANGE_PENDING, &mddev->flags);
        if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
                force_change = 1;
@@ -2221,19 +2226,6 @@ repeat:
                MD_BUG();
                mddev->events --;
        }
-
-       /*
-        * do not write anything to disk if using
-        * nonpersistent superblocks
-        */
-       if (!mddev->persistent) {
-               if (!mddev->external)
-                       clear_bit(MD_CHANGE_PENDING, &mddev->flags);
-
-               spin_unlock_irq(&mddev->write_lock);
-               wake_up(&mddev->sb_wait);
-               return;
-       }
        sync_sbs(mddev, nospares);
        spin_unlock_irq(&mddev->write_lock);