vhost/vsock: fix reset orphans race with close timeout
authorStefan Hajnoczi <stefanha@redhat.com>
Thu, 6 Dec 2018 19:14:34 +0000 (19:14 +0000)
committerMichael S. Tsirkin <mst@redhat.com>
Thu, 6 Dec 2018 19:22:34 +0000 (14:22 -0500)
If a local process has closed a connected socket and hasn't received a
RST packet yet, then the socket remains in the table until a timeout
expires.

When a vhost_vsock instance is released with the timeout still pending,
the socket is never freed because vhost_vsock has already set the
SOCK_DONE flag.

Check if the close timer is pending and let it close the socket.  This
prevents the race which can leak sockets.

Reported-by: Maximilian Riemensberger <riemensberger@cadami.net>
Cc: Graham Whaley <graham.whaley@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
drivers/vhost/vsock.c

index 34bc3ab40c6da8d5637e6e5bd19e6fb27ec27fba..731e2ea2aeca6e1deba84566c15680e544809e9f 100644 (file)
@@ -563,13 +563,21 @@ static void vhost_vsock_reset_orphans(struct sock *sk)
         * executing.
         */
 
-       if (!vhost_vsock_get(vsk->remote_addr.svm_cid)) {
-               sock_set_flag(sk, SOCK_DONE);
-               vsk->peer_shutdown = SHUTDOWN_MASK;
-               sk->sk_state = SS_UNCONNECTED;
-               sk->sk_err = ECONNRESET;
-               sk->sk_error_report(sk);
-       }
+       /* If the peer is still valid, no need to reset connection */
+       if (vhost_vsock_get(vsk->remote_addr.svm_cid))
+               return;
+
+       /* If the close timeout is pending, let it expire.  This avoids races
+        * with the timeout callback.
+        */
+       if (vsk->close_work_scheduled)
+               return;
+
+       sock_set_flag(sk, SOCK_DONE);
+       vsk->peer_shutdown = SHUTDOWN_MASK;
+       sk->sk_state = SS_UNCONNECTED;
+       sk->sk_err = ECONNRESET;
+       sk->sk_error_report(sk);
 }
 
 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)