tcp: add reordering timer in RACK loss detection

This patch makes RACK install a reordering timer when it suspects
some packets might be lost, but wants to delay the decision
a little bit to accomodate reordering.

It does not create a new timer but instead repurposes the existing
RTO timer, because both are meant to retransmit packets.
Specifically it arms a timer ICSK_TIME_REO_TIMEOUT when
the RACK timing check fails. The wait time is set to

  RACK.RTT + RACK.reo_wnd - (NOW - Packet.xmit_time) + fudge

This translates to expecting a packet (Packet) should take
(RACK.RTT + RACK.reo_wnd + fudge) to deliver after it was sent.

When there are multiple packets that need a timer, we use one timer
with the maximum timeout. Therefore the timer conservatively uses
the maximum window to expire N packets by one timeout, instead of
N timeouts to expire N packets sent at different times.

The fudge factor is 2 jiffies to ensure when the timer fires, all
the suspected packets would exceed the deadline and be marked lost
by tcp_rack_detect_loss(). It has to be at least 1 jiffy because the
clock may tick between calling icsk_reset_xmit_timer(timeout) and
actually hang the timer. The next jiffy is to lower-bound the timeout
to 2 jiffies when reo_wnd is < 1ms.

When the reordering timer fires (tcp_rack_reo_timeout): If we aren't
in Recovery we'll enter fast recovery and force fast retransmit.
This is very similar to the early retransmit (RFC5827) except RACK
is not constrained to only enter recovery for small outstanding
flights.

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

7 files changed, 61 insertions, 11 deletions

diff --git a/net/ipv4/inet_diag.c b/net/ipv4/inet_diag.c
index 4dea33e5f295..d216e40623d3 100644
--- a/net/ipv4/inet_diag.c
+++ b/net/ipv4/inet_diag.c
@@ -216,6 +216,7 @@ int inet_sk_diag_fill(struct sock *sk, struct inet_connection_sock *icsk,
 
 	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
 	    icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
+	    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
 	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
 		r->idiag_timer = 1;
 		r->idiag_retrans = icsk->icsk_retransmits;
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 8ccd171999bf..be1191829963 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -2522,8 +2522,7 @@ static void tcp_init_cwnd_reduction(struct sock *sk)
 	tcp_ecn_queue_cwr(tp);
 }
 
-static void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked,
-			       int flag)
+void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	int sndcnt = 0;
@@ -2691,7 +2690,7 @@ void tcp_simple_retransmit(struct sock *sk)
 }
 EXPORT_SYMBOL(tcp_simple_retransmit);
 
-static void tcp_enter_recovery(struct sock *sk, bool ece_ack)
+void tcp_enter_recovery(struct sock *sk, bool ece_ack)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	int mib_idx;
@@ -3031,6 +3030,7 @@ void tcp_rearm_rto(struct sock *sk)
 		u32 rto = inet_csk(sk)->icsk_rto;
 		/* Offset the time elapsed after installing regular RTO */
 		if (icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
+		    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
 		    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
 			struct sk_buff *skb = tcp_write_queue_head(sk);
 			const u32 rto_time_stamp =
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 56d756ecfb59..ebf3e0c4967a 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -2230,6 +2230,7 @@ static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
 
 	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
 	    icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
+	    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
 	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
 		timer_active	= 1;
 		timer_expires	= icsk->icsk_timeout;
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 1d5331a1b1dc..0ba9026cb70d 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -2960,7 +2960,8 @@ begin_fwd:
 		if (tcp_in_cwnd_reduction(sk))
 			tp->prr_out += tcp_skb_pcount(skb);
 
-		if (skb == tcp_write_queue_head(sk))
+		if (skb == tcp_write_queue_head(sk) &&
+		    icsk->icsk_pending != ICSK_TIME_REO_TIMEOUT)
 			inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 						  inet_csk(sk)->icsk_rto,
 						  TCP_RTO_MAX);
diff --git a/net/ipv4/tcp_recovery.c b/net/ipv4/tcp_recovery.c
index 557363cde58a..eb39b1b6d1dc 100644
--- a/net/ipv4/tcp_recovery.c
+++ b/net/ipv4/tcp_recovery.c
@@ -32,19 +32,18 @@ static void tcp_rack_mark_skb_lost(struct sock *sk, struct sk_buff *skb)
  * The current version is only used after recovery starts but can be
  * easily extended to detect the first loss.
  */
-static void tcp_rack_detect_loss(struct sock *sk, const struct skb_mstamp *now)
+static void tcp_rack_detect_loss(struct sock *sk, const struct skb_mstamp *now,
+				 u32 *reo_timeout)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct sk_buff *skb;
 	u32 reo_wnd;
 
+	*reo_timeout = 0;
 	/* To be more reordering resilient, allow min_rtt/4 settling delay
 	 * (lower-bounded to 1000uS). We use min_rtt instead of the smoothed
 	 * RTT because reordering is often a path property and less related
 	 * to queuing or delayed ACKs.
-	 *
-	 * TODO: measure and adapt to the observed reordering delay, and
-	 * use a timer to retransmit like the delayed early retransmit.
 	 */
 	reo_wnd = 1000;
 	if (tp->rack.reord && tcp_min_rtt(tp) != ~0U)
@@ -66,10 +65,23 @@ static void tcp_rack_detect_loss(struct sock *sk, const struct skb_mstamp *now)
 			 * A packet is lost if its elapsed time is beyond
 			 * the recent RTT plus the reordering window.
 			 */
-			if (skb_mstamp_us_delta(now, &skb->skb_mstamp) >
-			    tp->rack.rtt_us + reo_wnd) {
+			u32 elapsed = skb_mstamp_us_delta(now,
+							  &skb->skb_mstamp);
+			s32 remaining = tp->rack.rtt_us + reo_wnd - elapsed;
+
+			if (remaining < 0) {
 				tcp_rack_mark_skb_lost(sk, skb);
+				continue;
 			}
+
+			/* Skip ones marked lost but not yet retransmitted */
+			if ((scb->sacked & TCPCB_LOST) &&
+			    !(scb->sacked & TCPCB_SACKED_RETRANS))
+				continue;
+
+			/* Record maximum wait time (+1 to avoid 0) */
+			*reo_timeout = max_t(u32, *reo_timeout, 1 + remaining);
+
 		} else if (!(scb->sacked & TCPCB_RETRANS)) {
 			/* Original data are sent sequentially so stop early
 			 * b/c the rest are all sent after rack_sent
@@ -82,12 +94,19 @@ static void tcp_rack_detect_loss(struct sock *sk, const struct skb_mstamp *now)
 void tcp_rack_mark_lost(struct sock *sk, const struct skb_mstamp *now)
 {
 	struct tcp_sock *tp = tcp_sk(sk);
+	u32 timeout;
 
 	if (inet_csk(sk)->icsk_ca_state < TCP_CA_Recovery || !tp->rack.advanced)
 		return;
+
 	/* Reset the advanced flag to avoid unnecessary queue scanning */
 	tp->rack.advanced = 0;
-	tcp_rack_detect_loss(sk, now);
+	tcp_rack_detect_loss(sk, now, &timeout);
+	if (timeout) {
+		timeout = usecs_to_jiffies(timeout + TCP_REO_TIMEOUT_MIN);
+		inet_csk_reset_xmit_timer(sk, ICSK_TIME_REO_TIMEOUT,
+					  timeout, inet_csk(sk)->icsk_rto);
+	}
 }
 
 /* Record the most recently (re)sent time among the (s)acked packets
@@ -123,3 +142,27 @@ void tcp_rack_advance(struct tcp_sock *tp, u8 sacked,
 	tp->rack.mstamp = *xmit_time;
 	tp->rack.advanced = 1;
 }
+
+/* We have waited long enough to accommodate reordering. Mark the expired
+ * packets lost and retransmit them.
+ */
+void tcp_rack_reo_timeout(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct skb_mstamp now;
+	u32 timeout, prior_inflight;
+
+	skb_mstamp_get(&now);
+	prior_inflight = tcp_packets_in_flight(tp);
+	tcp_rack_detect_loss(sk, &now, &timeout);
+	if (prior_inflight != tcp_packets_in_flight(tp)) {
+		if (inet_csk(sk)->icsk_ca_state != TCP_CA_Recovery) {
+			tcp_enter_recovery(sk, false);
+			if (!inet_csk(sk)->icsk_ca_ops->cong_control)
+				tcp_cwnd_reduction(sk, 1, 0);
+		}
+		tcp_xmit_retransmit_queue(sk);
+	}
+	if (inet_csk(sk)->icsk_pending != ICSK_TIME_RETRANS)
+		tcp_rearm_rto(sk);
+}
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index 29a9bd5f1225..953c02a8566e 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -563,6 +563,9 @@ void tcp_write_timer_handler(struct sock *sk)
 	event = icsk->icsk_pending;
 
 	switch (event) {
+	case ICSK_TIME_REO_TIMEOUT:
+		tcp_rack_reo_timeout(sk);
+		break;
 	case ICSK_TIME_EARLY_RETRANS:
 		tcp_resume_early_retransmit(sk);
 		break;
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 228965dca3c5..f52c3742b404 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -1746,6 +1746,7 @@ static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
 
 	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
 	    icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
+	    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
 	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
 		timer_active	= 1;
 		timer_expires	= icsk->icsk_timeout;