tcp: better retrans tracking for defer-accept

For passive TCP connections using TCP_DEFER_ACCEPT facility,
we incorrectly increment req->retrans each time timeout triggers
while no SYNACK is sent.

SYNACK are not sent for TCP_DEFER_ACCEPT that were established (for
which we received the ACK from client). Only the last SYNACK is sent
so that we can receive again an ACK from client, to move the req into
accept queue. We plan to change this later to avoid the useless
retransmit (and potential problem as this SYNACK could be lost)

TCP_INFO later gives wrong information to user, claiming imaginary
retransmits.

Decouple req->retrans field into two independent fields :

num_retrans : number of retransmit
num_timeout : number of timeouts

num_timeout is the counter that is incremented at each timeout,
regardless of actual SYNACK being sent or not, and used to
compute the exponential timeout.

Introduce inet_rtx_syn_ack() helper to increment num_retrans
only if ->rtx_syn_ack() succeeded.

Use inet_rtx_syn_ack() from tcp_check_req() to increment num_retrans
when we re-send a SYNACK in answer to a (retransmitted) SYN.
Prior to this patch, we were not counting these retransmits.

Change tcp_v[46]_rtx_synack() to increment TCP_MIB_RETRANSSEGS
only if a synack packet was successfully queued.

Reported-by: Yuchung Cheng <[e-mail]>
Signed-off-by: Eric Dumazet <[e-mail]>
Cc: Julian Anastasov <[e-mail]>
Cc: Vijay Subramanian <[e-mail]>
Cc: Elliott Hughes <[e-mail]>
Cc: Neal Cardwell <[e-mail]>
Signed-off-by: David S. Miller <[e-mail]>

include/net/request_sock.h

···
 					   struct request_sock *req);
 };
 
+extern int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req);
+
 /* struct request_sock - mini sock to represent a connection request
  */
 struct request_sock {
 	struct request_sock		*dl_next; /* Must be first member! */
 	u16				mss;
-	u8				retrans;
-	u8				cookie_ts; /* syncookie: encode tcpopts in timestamp */
+	u8				num_retrans; /* number of retransmits */
+	u8				cookie_ts:1; /* syncookie: encode tcpopts in timestamp */
+	u8				num_timeout:7; /* number of timeouts */
 	/* The following two fields can be easily recomputed I think -AK */
 	u32				window_clamp; /* window clamp at creation time */
 	u32				rcv_wnd;	  /* rcv_wnd offered first time */
···
 {
 	struct listen_sock *lopt = queue->listen_opt;
 
-	if (req->retrans == 0)
+	if (req->num_timeout == 0)
 		--lopt->qlen_young;
 
 	return --lopt->qlen;
···
 	struct listen_sock *lopt = queue->listen_opt;
 
 	req->expires = jiffies + timeout;
-	req->retrans = 0;
+	req->num_retrans = 0;
+	req->num_timeout = 0;
 	req->sk = NULL;
 	req->dl_next = lopt->syn_table[hash];
 

net/dccp/minisocks.c

···
 			 * To protect against Request floods, increment retrans
 			 * counter (backoff, monitored by dccp_response_timer).
 			 */
-			req->retrans++;
-			req->rsk_ops->rtx_syn_ack(sk, req, NULL);
+			inet_rtx_syn_ack(sk, req);
 		}
 		/* Network Duplicate, discard packet */
 		return NULL;

net/ipv4/inet_connection_sock.c

···
 				  int *expire, int *resend)
 {
 	if (!rskq_defer_accept) {
-		*expire = req->retrans >= thresh;
+		*expire = req->num_timeout >= thresh;
 		*resend = 1;
 		return;
 	}
-	*expire = req->retrans >= thresh &&
-		  (!inet_rsk(req)->acked || req->retrans >= max_retries);
+	*expire = req->num_timeout >= thresh &&
+		  (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
 	/*
 	 * Do not resend while waiting for data after ACK,
 	 * start to resend on end of deferring period to give
 	 * last chance for data or ACK to create established socket.
 	 */
 	*resend = !inet_rsk(req)->acked ||
-		  req->retrans >= rskq_defer_accept - 1;
+		  req->num_timeout >= rskq_defer_accept - 1;
 }
+
+int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req)
+{
+	int err = req->rsk_ops->rtx_syn_ack(parent, req, NULL);
+
+	if (!err)
+		req->num_retrans++;
+	return err;
+}
+EXPORT_SYMBOL(inet_rtx_syn_ack);
 
 void inet_csk_reqsk_queue_prune(struct sock *parent,
 				const unsigned long interval,
···
 				req->rsk_ops->syn_ack_timeout(parent, req);
 				if (!expire &&
 				    (!resend ||
-				     !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
+				     !inet_rtx_syn_ack(parent, req) ||
 				     inet_rsk(req)->acked)) {
 					unsigned long timeo;
 
-					if (req->retrans++ == 0)
+					if (req->num_timeout++ == 0)
 						lopt->qlen_young--;
-					timeo = min((timeout << req->retrans), max_rto);
+					timeo = min(timeout << req->num_timeout,
+						    max_rto);
 					req->expires = now + timeo;
 					reqp = &req->dl_next;
 					continue;

net/ipv4/inet_diag.c

···
 	r->idiag_family = sk->sk_family;
 	r->idiag_state = TCP_SYN_RECV;
 	r->idiag_timer = 1;
-	r->idiag_retrans = req->retrans;
+	r->idiag_retrans = req->num_retrans;
 
 	r->id.idiag_if = sk->sk_bound_dev_if;
 	sock_diag_save_cookie(req, r->id.idiag_cookie);

net/ipv4/syncookies.c

···
 	}
 
 	req->expires	= 0UL;
-	req->retrans	= 0;
+	req->num_retrans = 0;
 
 	/*
 	 * We need to lookup the route here to get at the correct

net/ipv4/tcp_input.c

···
 				 */
 				if (req) {
 					tcp_synack_rtt_meas(sk, req);
-					tp->total_retrans = req->retrans;
+					tp->total_retrans = req->num_retrans;
 
 					reqsk_fastopen_remove(sk, req, false);
 				} else {

net/ipv4/tcp_ipv4.c

···
 }
 
 static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
-			      struct request_values *rvp)
+			     struct request_values *rvp)
 {
-	TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
-	return tcp_v4_send_synack(sk, NULL, req, rvp, 0, false);
+	int res = tcp_v4_send_synack(sk, NULL, req, rvp, 0, false);
+
+	if (!res)
+		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
+	return res;
 }
 
 /*
···
 	struct sock *child;
 	int err;
 
-	req->retrans = 0;
+	req->num_retrans = 0;
+	req->num_timeout = 0;
 	req->sk = NULL;
 
 	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
···
 
 	tcp_initialize_rcv_mss(newsk);
 	tcp_synack_rtt_meas(newsk, req);
-	newtp->total_retrans = req->retrans;
+	newtp->total_retrans = req->num_retrans;
 
 #ifdef CONFIG_TCP_MD5SIG
 	/* Copy over the MD5 key from the original socket */
···
 		0, 0, /* could print option size, but that is af dependent. */
 		1,    /* timers active (only the expire timer) */
 		jiffies_delta_to_clock_t(delta),
-		req->retrans,
+		req->num_timeout,
 		from_kuid_munged(seq_user_ns(f), uid),
 		0,  /* non standard timer */
 		0, /* open_requests have no inode */

net/ipv4/tcp_minisocks.c

···
 			 * it can be estimated (approximately)
 			 * from another data.
 			 */
-			tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans);
+			tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
 			paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
 		}
 	}
···
 		 * Note that even if there is new data in the SYN packet
 		 * they will be thrown away too.
 		 */
-		req->rsk_ops->rtx_syn_ack(sk, req, NULL);
+		inet_rtx_syn_ack(sk, req);
 		return NULL;
 	}
 
···
 	/* Got ACK for our SYNACK, so update baseline for SYNACK RTT sample. */
 	if (tmp_opt.saw_tstamp && tmp_opt.rcv_tsecr)
 		tcp_rsk(req)->snt_synack = tmp_opt.rcv_tsecr;
-	else if (req->retrans) /* don't take RTT sample if retrans && ~TS */
+	else if (req->num_retrans) /* don't take RTT sample if retrans && ~TS */
 		tcp_rsk(req)->snt_synack = 0;
 
 	/* For Fast Open no more processing is needed (sk is the
···
 		return sk;
 
 	/* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
-	if (req->retrans < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
+	if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
 	    TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
 		inet_rsk(req)->acked = 1;
 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);

net/ipv4/tcp_timer.c

···
 	req = tcp_sk(sk)->fastopen_rsk;
 	req->rsk_ops->syn_ack_timeout(sk, req);
 
-	if (req->retrans >= max_retries) {
+	if (req->num_timeout >= max_retries) {
 		tcp_write_err(sk);
 		return;
 	}
···
 	 * regular retransmit because if the child socket has been accepted
 	 * it's not good to give up too easily.
 	 */
-	req->rsk_ops->rtx_syn_ack(sk, req, NULL);
-	req->retrans++;
+	inet_rtx_syn_ack(sk, req);
+	req->num_timeout++;
 	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
-			  TCP_TIMEOUT_INIT << req->retrans, TCP_RTO_MAX);
+			  TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
 }
 
 /*

net/ipv6/syncookies.c

···
 		ireq6->iif = inet6_iif(skb);
 
 	req->expires = 0UL;
-	req->retrans = 0;
+	req->num_retrans = 0;
 	ireq->ecn_ok		= ecn_ok;
 	ireq->snd_wscale	= tcp_opt.snd_wscale;
 	ireq->sack_ok		= tcp_opt.sack_ok;

net/ipv6/tcp_ipv6.c

···
 			     struct request_values *rvp)
 {
 	struct flowi6 fl6;
+	int res;
 
-	TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
-	return tcp_v6_send_synack(sk, NULL, &fl6, req, rvp, 0);
+	res = tcp_v6_send_synack(sk, NULL, &fl6, req, rvp, 0);
+	if (!res)
+		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
+	return res;
 }
 
 static void tcp_v6_reqsk_destructor(struct request_sock *req)
···
 
 	tcp_initialize_rcv_mss(newsk);
 	tcp_synack_rtt_meas(newsk, req);
-	newtp->total_retrans = req->retrans;
+	newtp->total_retrans = req->num_retrans;
 
 	newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
 	newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
···
 		   0,0, /* could print option size, but that is af dependent. */
 		   1,   /* timers active (only the expire timer) */
 		   jiffies_to_clock_t(ttd),
-		   req->retrans,
+		   req->num_timeout,
 		   from_kuid_munged(seq_user_ns(seq), uid),
 		   0,  /* non standard timer */
 		   0, /* open_requests have no inode */