[PATCH] tcp: fix infinite cwnd in tcp_complete_cwr()

[PATCH] tcp: fix infinite cwnd in tcp_complete_cwr()

[Yuchung Cheng]

When the cwnd reduction is done, ssthresh may be infinite
if TCP enters CWR via ECN or F-RTO. If cwnd is not undone, i.e.,
undo_marker is set, tcp_complete_cwr() falsely set cwnd to the
infinite ssthresh value. The correct operation is to keep cwnd
intact because it has been updated in ECN or F-RTO.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
---
 net/ipv4/tcp_input.c |    8 +++++---
 1 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index c93b0cb..ac417de 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -2868,11 +2868,13 @@ static inline void tcp_complete_cwr(struct sock *sk)
 
 	/* Do not moderate cwnd if it's already undone in cwr or recovery. */
 	if (tp->undo_marker) {
-		if (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR)
+		if (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR) {
 			tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_ssthresh);
-		else /* PRR */
+		} else if (tp->snd_ssthresh < TCP_INFINITE_SSTHRESH) {
+			/* PRR algorithm. */
 			tp->snd_cwnd = tp->snd_ssthresh;
-		tp->snd_cwnd_stamp = tcp_time_stamp;
+			tp->snd_cwnd_stamp = tcp_time_stamp;
+		}
 	}
 	tcp_ca_event(sk, CA_EVENT_COMPLETE_CWR);
 }
-- 
1.7.7.3

Re: [PATCH] tcp: fix infinite cwnd in tcp_complete_cwr()

[Neal Cardwell]

On Sat, Apr 28, 2012 at 3:04 PM, Yuchung Cheng <ycheng@google.com> wrote:
> When the cwnd reduction is done, ssthresh may be infinite
> if TCP enters CWR via ECN or F-RTO. If cwnd is not undone, i.e.,
> undo_marker is set, tcp_complete_cwr() falsely set cwnd to the
> infinite ssthresh value. The correct operation is to keep cwnd
> intact because it has been updated in ECN or F-RTO.
>
> Signed-off-by: Yuchung Cheng <ycheng@google.com>
> ---
>  net/ipv4/tcp_input.c |    8 +++++---
>  1 files changed, 5 insertions(+), 3 deletions(-)
>
> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> index c93b0cb..ac417de 100644
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -2868,11 +2868,13 @@ static inline void tcp_complete_cwr(struct sock *sk)
>
>        /* Do not moderate cwnd if it's already undone in cwr or recovery. */
>        if (tp->undo_marker) {
> -               if (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR)
> +               if (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR) {
>                        tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_ssthresh);

In this TCP_CA_CWR case the patch needs to add the assignment:

     tp->snd_cwnd_stamp = tcp_time_stamp;

I know that was in earlier versions of your patch; looks like it just
got accidentally dropped on this latest version.

Otherwise, looks good to me.

neal