raid5: don't increment read_errors on EILSEQ return

[ Upstream commit b76b4715eba0d0ed574f58918b29c1b2f0fa37a8 ]

While MD continues to count read errors returned by the lower layer.
If those errors are -EILSEQ, instead of -EIO, it should NOT increase
the read_errors count.

When RAID6 is set up on dm-integrity target that detects massive
corruption, the leg will be ejected from the array.  Even if the
issue is correctable with a sector re-write and the array has
necessary redundancy to correct it.

The leg is ejected because it runs up the rdev->read_errors beyond
conf->max_nr_stripes.  The return status in dm-drypt when there is
a data integrity error is -EILSEQ (BLK_STS_PROTECTION).

Signed-off-by: Nigel Croxon <ncroxon@redhat.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index cc0bd528136db6b974aabe4a7b278896fd182c9e..9f2059e185f7fe8488e1b040dcc340d43c1e3bdc 100644 (file)
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -2538,7 +2538,8 @@ static void raid5_end_read_request(struct bio * bi)
                int set_bad = 0;
 
                clear_bit(R5_UPTODATE, &sh->dev[i].flags);
-               atomic_inc(&rdev->read_errors);
+               if (!(bi->bi_status == BLK_STS_PROTECTION))
+                       atomic_inc(&rdev->read_errors);
                if (test_bit(R5_ReadRepl, &sh->dev[i].flags))
                        pr_warn_ratelimited(
                                "md/raid:%s: read error on replacement device (sector %llu on %s).\n",