l2tp: fix a race in l2tp_ip_sendmsg()

Commit 081b1b1 (l2tp: fix l2tp_ip_sendmsg() route handling) added
a race, in case IP route cache is disabled.

In this case, we should not do the dst_release(&rt->dst), since it'll
free the dst immediately, instead of waiting a RCU grace period.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: James Chapman <jchapman@katalix.com>
Cc: Denys Fedoryshchenko <denys@visp.net.lb>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/l2tp/l2tp_ip.c

@@ -464,17 +464,20 @@ static int l2tp_ip_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *m
 					   sk->sk_bound_dev_if);
 		if (IS_ERR(rt))
 			goto no_route;
-		if (connected)
+		if (connected) {
 			sk_setup_caps(sk, &rt->dst);
-		else
-			dst_release(&rt->dst); /* safe since we hold rcu_read_lock */
+		} else {
+			skb_dst_set(skb, &rt->dst);
+			goto xmit;
+		}
 	}
 
 	/* We dont need to clone dst here, it is guaranteed to not disappear.
 	 *  __dev_xmit_skb() might force a refcount if needed.
 	 */
 	skb_dst_set_noref(skb, &rt->dst);
 
+xmit:
 	/* Queue the packet to IP for output */
 	rc = ip_queue_xmit(skb, &inet->cork.fl);
 	rcu_read_unlock();