md/raid1: count resync requests in nr_pending.
authorNeilBrown <neilb@suse.de>
Tue, 16 Sep 2014 02:14:14 +0000 (12:14 +1000)
committerNeilBrown <neilb@suse.de>
Mon, 22 Sep 2014 01:26:01 +0000 (11:26 +1000)
Both normal IO and resync IO can be retried with reschedule_retry()
and so be counted into ->nr_queued, but only normal IO gets counted in
->nr_pending.

Before the recent improvement to RAID1 resync there could only
possibly have been one or the other on the queue.  When handling a
read failure it could only be normal IO.  So when handle_read_error()
called freeze_array() the fact that freeze_array only compares
->nr_queued against ->nr_pending was safe.

But now that these two types can interleave, we can have both normal
and resync IO requests queued, so we need to count them both in
nr_pending.

This error can lead to freeze_array() hanging if there is a read
error, so it is suitable for -stable.

Fixes: 79ef3a8aa1cb1523cc231c9a90a278333c21f761
cc: stable@vger.kernel.org (v3.13+)
Reported-by: Brassow Jonathan <jbrassow@redhat.com>
Signed-off-by: NeilBrown <neilb@suse.de>
drivers/md/raid1.c

index acfccec..35649dd 100644 (file)
@@ -856,6 +856,7 @@ static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
                             conf->next_resync + RESYNC_SECTORS),
                            conf->resync_lock);
 
+       conf->nr_pending++;
        spin_unlock_irq(&conf->resync_lock);
 }
 
@@ -865,6 +866,7 @@ static void lower_barrier(struct r1conf *conf)
        BUG_ON(conf->barrier <= 0);
        spin_lock_irqsave(&conf->resync_lock, flags);
        conf->barrier--;
+       conf->nr_pending--;
        spin_unlock_irqrestore(&conf->resync_lock, flags);
        wake_up(&conf->wait_barrier);
 }