vhost: fix vhost_vq_access_ok() log check
authorStefan Hajnoczi <stefanha@redhat.com>
Wed, 11 Apr 2018 02:35:40 +0000 (10:35 +0800)
committerDavid S. Miller <davem@davemloft.net>
Wed, 11 Apr 2018 14:54:06 +0000 (10:54 -0400)
Commit d65026c6c62e7d9616c8ceb5a53b68bcdc050525 ("vhost: validate log
when IOTLB is enabled") introduced a regression.  The logic was
originally:

  if (vq->iotlb)
      return 1;
  return A && B;

After the patch the short-circuit logic for A was inverted:

  if (A || vq->iotlb)
      return A;
  return B;

This patch fixes the regression by rewriting the checks in the obvious
way, no longer returning A when vq->iotlb is non-NULL (which is hard to
understand).

Reported-by: syzbot+65a84dde0214b0387ccd@syzkaller.appspotmail.com
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/vhost/vhost.c

index f44aead98d604335af5b4c20737568bb3e7b665d..9a18535fafbac32c2cbd127b7486f78de00472c7 100644 (file)
@@ -1244,10 +1244,12 @@ static int vq_log_access_ok(struct vhost_virtqueue *vq,
 /* Caller should have vq mutex and device mutex */
 int vhost_vq_access_ok(struct vhost_virtqueue *vq)
 {
-       int ret = vq_log_access_ok(vq, vq->log_base);
+       if (!vq_log_access_ok(vq, vq->log_base))
+               return 0;
 
-       if (ret || vq->iotlb)
-               return ret;
+       /* Access validation occurs at prefetch time with IOTLB */
+       if (vq->iotlb)
+               return 1;
 
        return vq_access_ok(vq, vq->num, vq->desc, vq->avail, vq->used);
 }