Re: rxrpc: Fix deadlock between call creation and sendmsg/recvmsg

[David Howells <dhowells@redhat.com>]

Colin Ian King <colin.king@canonical.com> wrote:

> I think the following part of the patch is problematic:
> 
>         call = rxrpc_find_call_by_user_ID(rx, user_call_ID);
>         if (!call) {
> +               ret = -EBADSLT;
>                 if (cmd != RXRPC_CMD_SEND_DATA)
> -                       return -EBADSLT;
> +                       goto error_release_sock;
> +               ret = -EBUSY;
> 
> At this point call is null, so the following code is performing a null
> pointer dereference on call when accessing call->state.
> 
> Detected by CoverityScan CID#1414316 ("Dereference after null check")
> 
> +               if (call->state == RXRPC_CALL_UNINITIALISED ||
> +                   call->state == RXRPC_CALL_CLIENT_AWAIT_CONN ||
> +                   call->state == RXRPC_CALL_SERVER_PREALLOC ||
> +                   call->state == RXRPC_CALL_SERVER_SECURING ||
> +                   call->state == RXRPC_CALL_SERVER_ACCEPTING)
> +                       goto error_release_sock;
>                 call = rxrpc_new_client_call_for_sendmsg(rx, msg,
> user_call_ID,

Good catch, thanks,  I put the state check in the wrong half of the
if-statement.

David