vsock/virtio: avoid potential deadlock when vsock device remove

[ Upstream commit 49b0b6f ]

There's a potential deadlock case when remove the vsock device or
process the RESET event:

  vsock_for_each_connected_socket:
      spin_lock_bh(&vsock_table_lock) ----------- (1)
      ...
          virtio_vsock_reset_sock:
              lock_sock(sk) --------------------- (2)
      ...
      spin_unlock_bh(&vsock_table_lock)

lock_sock() may do initiative schedule when the 'sk' is owned by
other thread at the same time, we would receivce a warning message
that "scheduling while atomic".

Even worse, if the next task (selected by the scheduler) try to
release a 'sk', it need to request vsock_table_lock and the deadlock
occur, cause the system into softlockup state.
  Call trace:
   queued_spin_lock_slowpath
   vsock_remove_bound
   vsock_remove_sock
   virtio_transport_release
   __vsock_release
   vsock_release
   __sock_release
   sock_close
   __fput
   ____fput

So we should not require sk_lock in this case, just like the behavior
in vhost_vsock or vmci.

Fixes: 0ea9e1d ("VSOCK: Introduce virtio_transport.ko")
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Link: https://lore.kernel.org/r/20210812053056.1699-1-longpeng2@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>

net/vmw_vsock/virtio_transport.c

@@ -356,11 +356,14 @@ static void virtio_vsock_event_fill(struct virtio_vsock *vsock)
 
 static void virtio_vsock_reset_sock(struct sock *sk)
 {
-	lock_sock(sk);
+	/* vmci_transport.c doesn't take sk_lock here either.  At least we're
+	 * under vsock_table_lock so the sock cannot disappear while we're
+	 * executing.
+	 */
+
 	sk->sk_state = TCP_CLOSE;
 	sk->sk_err = ECONNRESET;
 	sk->sk_error_report(sk);
-	release_sock(sk);
 }
 
 static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock)