Hi Dan,
On Tue, Nov 26, 2019 at 03:22:19PM +0300, Dan Carpenter wrote:
> Hello Stefano Garzarella,
>
> This is a semi-automatic email about new static checker warnings.
>
> The patch c0cfa2d8a788: "vsock: add multi-transports support" from
> Nov 14, 2019, leads to the following Smatch complaint:
>
> net/vmw_vsock/af_vsock.c:1014 vsock_poll()
> error: we previously assumed 'transport' could be null (see line 987)
>
> net/vmw_vsock/af_vsock.c
> 986 /* If there is something in the queue then we can read. */
> 987 if (transport && transport->stream_is_active(vsk) &&
> ^^^^^^^^^
> The patch adds a check for transport.
>
> 988 !(sk->sk_shutdown & RCV_SHUTDOWN)) {
> 989 bool data_ready_now = false;
> 990 int ret = transport->notify_poll_in(
> 991 vsk, 1, &data_ready_now);
> 992 if (ret < 0) {
> 993 mask |= EPOLLERR;
> 994 } else {
> 995 if (data_ready_now)
> 996 mask |= EPOLLIN | EPOLLRDNORM;
> 997
> 998 }
> 999 }
> 1000
> 1001 /* Sockets whose connections have been closed, reset, or
> 1002 * terminated should also be considered read, and we check the
> 1003 * shutdown flag for that.
> 1004 */
> 1005 if (sk->sk_shutdown & RCV_SHUTDOWN ||
> 1006 vsk->peer_shutdown & SEND_SHUTDOWN) {
> 1007 mask |= EPOLLIN | EPOLLRDNORM;
> 1008 }
> 1009
> 1010 /* Connected sockets that can produce data can be written. */
> 1011 if (sk->sk_state == TCP_ESTABLISHED) {
> 1012 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
> 1013 bool space_avail_now = false;
> 1014 int ret = transport->notify_poll_out(
> ^^^^^^^^^^^^^^^^^^^^^^^^^^
> But it's not checked here. I guess probably TCP_ESTABLISHED means that
> transport is non-NULL? If so then just ignore the warning.
Exactly, TCP_ESTABLISHED is set (by the transport) only when a socket is
assigned to a transport, so transport cannot be null here.
Thanks for warning me,
Stefano
