tcp: properly track retry time on passive Fast Open

This patch addresses a corner issue on timeout behavior of a
passive Fast Open socket.  A passive Fast Open server may write
and close the socket when it is re-trying SYN-ACK to complete
the handshake. After the handshake is completely, the server does
not properly stamp the recovery start time (tp->retrans_stamp is
0), and the socket may abort immediately on the very first FIN
timeout, instead of retying until it passes the system or user
specified limit.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv4/tcp_timer.c

@@ -378,6 +378,7 @@ static void tcp_fastopen_synack_timer(struct sock *sk)
 	struct inet_connection_sock *icsk = inet_csk(sk);
 	int max_retries = icsk->icsk_syn_retries ? :
 	    sock_net(sk)->ipv4.sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
+	struct tcp_sock *tp = tcp_sk(sk);
 	struct request_sock *req;
 
 	req = tcp_sk(sk)->fastopen_rsk;
@@ -395,6 +396,8 @@ static void tcp_fastopen_synack_timer(struct sock *sk)
 	inet_rtx_syn_ack(sk, req);
 	req->num_timeout++;
 	icsk->icsk_retransmits++;
+	if (!tp->retrans_stamp)
+		tp->retrans_stamp = tcp_time_stamp(tp);
 	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 			  TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
 }