[PATCH net] tcp: when rearming RTO, if RTO time is in past then fire RTO ASAP

In some situations tcp_send_loss_probe() can realize that it's unable
to send a loss probe (TLP), and falls back to calling tcp_rearm_rto()
to schedule an RTO timer. In such cases, sometimes tcp_rearm_rto()
realizes that the RTO was eligible to fire immediately or at some
point in the past (delta_us <= 0). Previously in such cases
tcp_rearm_rto() was scheduling such "overdue" RTOs to happen at now +
icsk_rto, which caused needless delays of hundreds of milliseconds
(and non-linear behavior that made reproducible testing
difficult). This commit changes the logic to schedule "overdue" RTOs
ASAP, rather than at now + icsk_rto.

Suggested-by: Yuchung Cheng <redacted>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <redacted>
Signed-off-by: Eric Dumazet <redacted>
---
 net/ipv4/tcp_input.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 53de1424c13c..bab7f0493098 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c

@@ -3009,8 +3009,7 @@ void tcp_rearm_rto(struct sock *sk)
 			/* delta_us may not be positive if the socket is locked
 			 * when the retrans timer fires and is rescheduled.
 			 */
-			if (delta_us > 0)
-				rto = usecs_to_jiffies(delta_us);
+			rto = usecs_to_jiffies(max_t(int, delta_us, 1));
 		}
 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, rto,
 					  TCP_RTO_MAX);

-- 
2.14.1.480.gb18f417b89-goog

On Wed, Aug 16, 2017 at 5:53 PM, Neal Cardwell [off-list ref] wrote:

In some situations tcp_send_loss_probe() can realize that it's unable
to send a loss probe (TLP), and falls back to calling tcp_rearm_rto()
to schedule an RTO timer. In such cases, sometimes tcp_rearm_rto()
realizes that the RTO was eligible to fire immediately or at some
point in the past (delta_us <= 0). Previously in such cases
tcp_rearm_rto() was scheduling such "overdue" RTOs to happen at now +
icsk_rto, which caused needless delays of hundreds of milliseconds
(and non-linear behavior that made reproducible testing
difficult). This commit changes the logic to schedule "overdue" RTOs
ASAP, rather than at now + icsk_rto.

Suggested-by: Yuchung Cheng <redacted>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <redacted>
Signed-off-by: Eric Dumazet <redacted>
---

Forgot to mention:

Fixes: 6ba8a3b19e76 ("tcp: Tail loss probe (TLP)")

This is a -stable candidate.

thanks,
neal

From: Neal Cardwell <ncardwell@google.com>
Date: Wed, 16 Aug 2017 17:53:36 -0400

In some situations tcp_send_loss_probe() can realize that it's unable
to send a loss probe (TLP), and falls back to calling tcp_rearm_rto()
to schedule an RTO timer. In such cases, sometimes tcp_rearm_rto()
realizes that the RTO was eligible to fire immediately or at some
point in the past (delta_us <= 0). Previously in such cases
tcp_rearm_rto() was scheduling such "overdue" RTOs to happen at now +
icsk_rto, which caused needless delays of hundreds of milliseconds
(and non-linear behavior that made reproducible testing
difficult). This commit changes the logic to schedule "overdue" RTOs
ASAP, rather than at now + icsk_rto.

Suggested-by: Yuchung Cheng <redacted>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <redacted>
Signed-off-by: Eric Dumazet <redacted>

Applied with Fixes: tag added and queued up for -stable, thanks Neal.