tcp: ensure prior synack rtx behavior with small backlogs

Some applications use a listen() backlog of 1.

Prior kernels were silently enforcing a qlen_log of 4, so that we were
sending up to /proc/sys/net/ipv4/tcp_synack_retries SYNACK messages.

Fixes: ef547f2 ("tcp: remove max_qlen_log")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv4/inet_connection_sock.c

@@ -579,7 +579,7 @@ static void reqsk_timer_handler(unsigned long data)
 	 * ones are about to clog our table.
 	 */
 	qlen = reqsk_queue_len(queue);
-	if ((qlen << 1) > sk_listener->sk_max_ack_backlog) {
+	if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
 		int young = reqsk_queue_len_young(queue) << 1;
 
 		while (thresh > 2) {