Explorar o código

md: avoid races when stopping resync.

There has been a race in raid10 and raid1 for a long time
which has only recently started showing up due to a scheduler changed.

When a sync_read request finishes, as soon as reschedule_retry
is called, another thread can mark the resync request as having
completed, so md_do_sync can finish, ->stop can be called, and
->conf can be freed.  So using conf after reschedule_retry is not
safe.

Similarly, when finishing a sync_write, calling md_done_sync must be
the last thing we do, as it allows a chain of events which will free
conf and other data structures.

The first of these requires action in raid10.c
The second requires action in raid1.c and raid10.c

Cc: stable@kernel.org
Signed-off-by: NeilBrown <neilb@suse.de>
NeilBrown %!s(int64=16) %!d(string=hai) anos
pai
achega
73d5c38a95
Modificáronse 2 ficheiros con 6 adicións e 4 borrados
  1. 2 1
      drivers/md/raid1.c
  2. 4 3
      drivers/md/raid10.c

+ 2 - 1
drivers/md/raid1.c

@@ -1237,8 +1237,9 @@ static void end_sync_write(struct bio *bio, int error)
 	update_head_pos(mirror, r1_bio);
 
 	if (atomic_dec_and_test(&r1_bio->remaining)) {
-		md_done_sync(mddev, r1_bio->sectors, uptodate);
+		sector_t s = r1_bio->sectors;
 		put_buf(r1_bio);
+		md_done_sync(mddev, s, uptodate);
 	}
 }
 

+ 4 - 3
drivers/md/raid10.c

@@ -1236,6 +1236,7 @@ static void end_sync_read(struct bio *bio, int error)
 	/* for reconstruct, we always reschedule after a read.
 	 * for resync, only after all reads
 	 */
+	rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev);
 	if (test_bit(R10BIO_IsRecover, &r10_bio->state) ||
 	    atomic_dec_and_test(&r10_bio->remaining)) {
 		/* we have read all the blocks,
@@ -1243,7 +1244,6 @@ static void end_sync_read(struct bio *bio, int error)
 		 */
 		reschedule_retry(r10_bio);
 	}
-	rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev);
 }
 
 static void end_sync_write(struct bio *bio, int error)
@@ -1264,11 +1264,13 @@ static void end_sync_write(struct bio *bio, int error)
 
 	update_head_pos(i, r10_bio);
 
+	rdev_dec_pending(conf->mirrors[d].rdev, mddev);
 	while (atomic_dec_and_test(&r10_bio->remaining)) {
 		if (r10_bio->master_bio == NULL) {
 			/* the primary of several recovery bios */
-			md_done_sync(mddev, r10_bio->sectors, 1);
+			sector_t s = r10_bio->sectors;
 			put_buf(r10_bio);
+			md_done_sync(mddev, s, 1);
 			break;
 		} else {
 			r10bio_t *r10_bio2 = (r10bio_t *)r10_bio->master_bio;
@@ -1276,7 +1278,6 @@ static void end_sync_write(struct bio *bio, int error)
 			r10_bio = r10_bio2;
 		}
 	}
-	rdev_dec_pending(conf->mirrors[d].rdev, mddev);
 }
 
 /*