md: revert incorrect fix for read error handling in raid1.
NeilBrown [Tue, 1 Dec 2009 06:30:59 +0000 (17:30 +1100)]
commit 4706b349f was a forward port of a fix that was needed
for SLES10.  But in fact it is not needed in mainline because
the earlier commit dd00a99e7a fixes the same problem in a
better way.
Further, this commit introduces a bug in the way it interacts with
the automatic read-error-correction.  If, after a read error is
successfully corrected, the same disk is chosen to re-read - the
re-read won't be attempted but an error will be returned instead.

After reverting that commit, there is the possibility that a
read error on a read-only array (where read errors cannot
be corrected as that requires a write) will repeatedly read the same
device and continue to get an error.
So in the "Array is readonly" case, fail the drive immediately on
a read error.

Signed-off-by: NeilBrown <neilb@suse.de>
Cc: stable@kernel.org

drivers/md/raid1.c

index a053423..e07ce2e 100644 (file)
@@ -1650,11 +1650,12 @@ static void raid1d(mddev_t *mddev)
                                               r1_bio->sector,
                                               r1_bio->sectors);
                                unfreeze_array(conf);
-                       }
+                       } else
+                               md_error(mddev,
+                                        conf->mirrors[r1_bio->read_disk].rdev);
 
                        bio = r1_bio->bios[r1_bio->read_disk];
-                       if ((disk=read_balance(conf, r1_bio)) == -1 ||
-                           disk == r1_bio->read_disk) {
+                       if ((disk=read_balance(conf, r1_bio)) == -1) {
                                printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
                                       " read error for block %llu\n",
                                       bdevname(bio->bi_bdev,b),