Merge branch 'tcp-xmit-timer-rearming'

Neal Cardwell says:

====================
tcp: fix xmit timer rearming to avoid stalls

This patch series is a bug fix for a TCP loss recovery performance bug
reported independently in recent netdev threads:

 (i)  July 26, 2017: netdev thread "TCP fast retransmit issues"
 (ii) July 26, 2017: netdev thread:
       "[PATCH V2 net-next] TLP: Don't reschedule PTO when there's one
       outstanding TLP retransmission"

Many thanks to Klavs Klavsen and Mao Wenan for the detailed reports,
traces, and packetdrill test cases, which enabled us to root-cause
this issue and verify the fix.

- v1 -> v2:
 - In patch 2/3, changed an unclear comment in the pre-existing code
   in tcp_schedule_loss_probe() to be more clear (thanks to Eric Dumazet
   for suggesting we improve this).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 5 insertions, 18 deletions

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 2f1588bf73da..276406a83a37 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -2377,24 +2377,15 @@ bool tcp_schedule_loss_probe(struct sock *sk)
 {
 	struct inet_connection_sock *icsk = inet_csk(sk);
 	struct tcp_sock *tp = tcp_sk(sk);
-	u32 timeout, tlp_time_stamp, rto_time_stamp;
 	u32 rtt = usecs_to_jiffies(tp->srtt_us >> 3);
+	u32 timeout, rto_delta_us;
 
-	/* No consecutive loss probes. */
-	if (WARN_ON(icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)) {
-		tcp_rearm_rto(sk);
-		return false;
-	}
 	/* Don't do any loss probe on a Fast Open connection before 3WHS
 	 * finishes.
 	 */
 	if (tp->fastopen_rsk)
 		return false;
 
-	/* TLP is only scheduled when next timer event is RTO. */
-	if (icsk->icsk_pending != ICSK_TIME_RETRANS)
-		return false;
-
 	/* Schedule a loss probe in 2*RTT for SACK capable connections
 	 * in Open state, that are either limited by cwnd or application.
 	 */
@@ -2417,14 +2408,10 @@ bool tcp_schedule_loss_probe(struct sock *sk)
 				(rtt + (rtt >> 1) + TCP_DELACK_MAX));
 	timeout = max_t(u32, timeout, msecs_to_jiffies(10));
 
-	/* If RTO is shorter, just schedule TLP in its place. */
-	tlp_time_stamp = tcp_jiffies32 + timeout;
-	rto_time_stamp = (u32)inet_csk(sk)->icsk_timeout;
-	if ((s32)(tlp_time_stamp - rto_time_stamp) > 0) {
-		s32 delta = rto_time_stamp - tcp_jiffies32;
-		if (delta > 0)
-			timeout = delta;
-	}
+	/* If the RTO formula yields an earlier time, then use that time. */
+	rto_delta_us = tcp_rto_delta_us(sk);  /* How far in future is RTO? */
+	if (rto_delta_us > 0)
+		timeout = min_t(u32, timeout, usecs_to_jiffies(rto_delta_us));
 
 	inet_csk_reset_xmit_timer(sk, ICSK_TIME_LOSS_PROBE, timeout,
 				  TCP_RTO_MAX);