Bug #376

———- Forwarded message ———-
From: Neal Cardwell ncardwell@google.com
Date: Sun, Apr 29, 2012 at 8:49 PM
Subject: Re: [PATCH] tcp: fix infinite cwnd in tcp_complete_cwr()
To: Yuchung Cheng ycheng@google.com
Cc: davem@davemloft.net, ilpo.jarvinen@helsinki.fi,
nanditad@google.com, netdev@vger.kernel.org

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.

final patch

———- Forwarded message ———-
From: Yuchung Cheng ycheng@google.com
Date: Mon, Apr 30, 2012 at 9:00 AM
Subject: [PATCH v2] tcp: fix infinite cwnd in tcp_complete_cwr()
To: davem@davemloft.net, ilpo.jarvinen@helsinki.fi, ncardwell@google.com
Cc: nanditad@google.com, netdev@vger.kernel.org, Yuchung Cheng
ycheng@google.com

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
—
ChangeLog since v1:
 - Add snd_cwnd_stamp timestamping in CWR mode

 net/ipv4/tcp_input.c |    9 **—
 1 files changed, 6 insertions(+), 3 deletions(-)

diff –git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index c93b0cb..e0a9b89 100644
— a/net/ipv4/tcp_input.c
**+ b/net/ipv4/tcp_input.c
@ -2868,11 +2868,14@ 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 */
+                       tp->snd_cwnd_stamp = tcp_time_stamp;
+               } 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);
 }