rxrpc: Fix client call queueing, waiting for channel

rxrpc_get_client_conn() adds a new call to the front of the waiting_calls
queue if the connection it's going to use already exists.  This is bad as
it allows calls to get starved out.

Fix this by adding to the tail instead.

Also change the other enqueue point in the same function to put it on the
front (ie. when we have a new connection).  This makes the point that in
the case of a new connection the new call goes at the front (though it
doesn't actually matter since the queue should be unoccupied).

Fixes: 45025bce ("rxrpc: Improve management and caching of client connection objects")
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/rxrpc/conn_client.c

@@ -353,7 +353,7 @@ static int rxrpc_get_client_conn(struct rxrpc_sock *rx,
 	 * normally have to take channel_lock but we do this before anyone else
 	 * can see the connection.
 	 */
-	list_add_tail(&call->chan_wait_link, &candidate->waiting_calls);
+	list_add(&call->chan_wait_link, &candidate->waiting_calls);
 
 	if (cp->exclusive) {
 		call->conn = candidate;
@@ -432,7 +432,7 @@ static int rxrpc_get_client_conn(struct rxrpc_sock *rx,
 	call->conn = conn;
 	call->security_ix = conn->security_ix;
 	call->service_id = conn->service_id;
-	list_add(&call->chan_wait_link, &conn->waiting_calls);
+	list_add_tail(&call->chan_wait_link, &conn->waiting_calls);
 	spin_unlock(&conn->channel_lock);
 	_leave(" = 0 [extant %d]", conn->debug_id);
 	return 0;