vhost: synchronize IOTLB message with dev cleanup
authorJason Wang <jasowang@redhat.com>
Tue, 22 May 2018 11:58:57 +0000 (19:58 +0800)
committerDavid S. Miller <davem@davemloft.net>
Fri, 25 May 2018 02:09:51 +0000 (22:09 -0400)
DaeRyong Jeong reports a race between vhost_dev_cleanup() and
vhost_process_iotlb_msg():

Thread interleaving:
CPU0 (vhost_process_iotlb_msg) CPU1 (vhost_dev_cleanup)
(In the case of both VHOST_IOTLB_UPDATE and
VHOST_IOTLB_INVALIDATE)

===== =====
vhost_umem_clean(dev->iotlb);
if (!dev->iotlb) {
        ret = -EFAULT;
        break;
}
dev->iotlb = NULL;

The reason is we don't synchronize between them, fixing by protecting
vhost_process_iotlb_msg() with dev mutex.

Reported-by: DaeRyong Jeong <threeearcat@gmail.com>
Fixes: 6b1e6cc7855b0 ("vhost: new device IOTLB API")
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/vhost/vhost.c

index f3bd8e9412245919ba3a73fb9d94677c280d1427..f0be5f35ab28f2a009a0e56d75bbb0c46fbcb4d9 100644 (file)
@@ -981,6 +981,7 @@ static int vhost_process_iotlb_msg(struct vhost_dev *dev,
 {
        int ret = 0;
 
+       mutex_lock(&dev->mutex);
        vhost_dev_lock_vqs(dev);
        switch (msg->type) {
        case VHOST_IOTLB_UPDATE:
@@ -1016,6 +1017,8 @@ static int vhost_process_iotlb_msg(struct vhost_dev *dev,
        }
 
        vhost_dev_unlock_vqs(dev);
+       mutex_unlock(&dev->mutex);
+
        return ret;
 }
 ssize_t vhost_chr_write_iter(struct vhost_dev *dev,