swapfile: avoid NULL pointer dereference in swapon when s_bdev is NULL
authorSuresh Jayaraman <sjayaraman@suse.de>
Wed, 30 Sep 2009 08:53:48 +0000 (10:53 +0200)
committerJens Axboe <jens.axboe@oracle.com>
Thu, 1 Oct 2009 19:15:46 +0000 (21:15 +0200)
While testing Swap over NFS patchset, I noticed an oops that was triggered
during swapon. Investigating further, the NULL pointer deference is due to the
SSD device check/optimization in the swapon code that assumes s_bdev could never
be NULL.

inode->i_sb->s_bdev could be NULL in a few cases. For e.g. one such case is
loopback NFS mount, there could be others as well. Fix this by ensuring s_bdev
is not NULL before we try to deference s_bdev.

Signed-off-by: Suresh Jayaraman <sjayaraman@suse.de>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
mm/swapfile.c

index 4de7f02f820b03bfcf36b5fc8d6827b5eecd38cb..a1bc6b9af9a23634f4185e85775d94a6500bbb0e 100644 (file)
@@ -1974,12 +1974,14 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
                goto bad_swap;
        }
 
-       if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
-               p->flags |= SWP_SOLIDSTATE;
-               p->cluster_next = 1 + (random32() % p->highest_bit);
+       if (p->bdev) {
+               if (blk_queue_nonrot(bdev_get_queue(p->bdev))) {
+                       p->flags |= SWP_SOLIDSTATE;
+                       p->cluster_next = 1 + (random32() % p->highest_bit);
+               }
+               if (discard_swap(p) == 0)
+                       p->flags |= SWP_DISCARDABLE;
        }
-       if (discard_swap(p) == 0)
-               p->flags |= SWP_DISCARDABLE;
 
        mutex_lock(&swapon_mutex);
        spin_lock(&swap_lock);