raid5: Retry R5_ReadNoMerge flag when hit a read error.

Because of block layer merge, one bio fails will cause other bios
which belongs to the same request fails, so raid5_end_read_request
will record all these bios as badblocks.
If retry request with R5_ReadNoMerge flag to avoid bios merge,
badblocks can only record sector which is bad exactly.

test:
hdparm --yes-i-know-what-i-am-doing --make-bad-sector 300000 /dev/sdb
mdadm -C /dev/md0 -l5 -n3 /dev/sd[bcd] --assume-clean
mdadm /dev/md0 -f /dev/sdd
mdadm /dev/md0 -r /dev/sdd
mdadm --zero-superblock /dev/sdd
mdadm /dev/md0 -a /dev/sdd

1. Without this patch:
cat /sys/block/md0/md/rd*/bad_blocks
299776 256
299776 256

2. With this patch:
cat /sys/block/md0/md/rd*/bad_blocks
300000 8
300000 8

Signed-off-by: Bian Yu <bianyu@kedacom.com>
Signed-off-by: NeilBrown <neilb@suse.de>

diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 1fb28c5..2c89077 100644 (file)
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -2056,6 +2056,9 @@ static void raid5_end_read_request(struct bio * bi, int error)
                               mdname(conf->mddev), bdn);
                else
                        retry = 1;
+               if (set_bad && test_bit(In_sync, &rdev->flags)
+                   && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
+                       retry = 1;
                if (retry)
                        if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
                                set_bit(R5_ReadError, &sh->dev[i].flags);