dm io: fix duplicate bio completion due to missing ref count
authorMike Snitzer <snitzer@redhat.com>
Tue, 20 Jun 2017 23:14:30 +0000 (19:14 -0400)
committerSasha Levin <alexander.levin@microsoft.com>
Wed, 21 Mar 2018 03:49:49 +0000 (23:49 -0400)
[ Upstream commit feb7695fe9fb83084aa29de0094774f4c9d4c9fc ]

If only a subset of the devices associated with multiple regions support
a given special operation (eg. DISCARD) then the dec_count() that is
used to set error for the region must increment the io->count.

Otherwise, when the dec_count() is called it can cause the dm-io
caller's bio to be completed multiple times.  As was reported against
the dm-mirror target that had mirror legs with a mix of discard
capabilities.

Bug: https://bugzilla.kernel.org/show_bug.cgi?id=196077
Reported-by: Zhang Yi <yizhan@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
drivers/md/dm-io.c

index 74adcd2c967ec8680d0cfd5c9d876c93074dbe10..01d7b5785b8e51cc8f8efa0c78a9306be9be0e2f 100644 (file)
@@ -299,6 +299,7 @@ static void do_region(int rw, unsigned region, struct dm_io_region *where,
        else if (rw & REQ_WRITE_SAME)
                special_cmd_max_sectors = q->limits.max_write_same_sectors;
        if ((rw & (REQ_DISCARD | REQ_WRITE_SAME)) && special_cmd_max_sectors == 0) {
+               atomic_inc(&io->count);
                dec_count(io, region, -EOPNOTSUPP);
                return;
        }