[PATCH net-next] Avoid reducing cwnd when ACK+DSACK is received

When the peer has delayed ack enabled, it may reply to a probe with an
ACK+D-SACK, with ack value set to tlp_high_seq. In the current code,
such ACK+DSACK will be missed and only at next, higher ack will the TLP
episode be considered done. Since the DSACK is not present anymore,
this will cost a cwnd reduction.

This patch ensures that this scenario does not cause a cwnd reduction, since
receiving an ACK+DSACK indicates that both the initial segment and the probe
have been received by the peer.

Cc: Gregory Detal <redacted>
Cc: Nandita Dukkipati <redacted>
Cc: Yuchung Cheng <redacted>
Signed-off-by: Sébastien Barré <redacted>

---
 net/ipv4/tcp_input.c | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 075ab4d..fb007cc 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c

@@ -3369,23 +3369,23 @@ static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag)
 
 	/* Mark the end of TLP episode on receiving TLP dupack or when
 	 * ack is after tlp_high_seq.
+	 * With delayed acks, we may also get a regular ACK+DSACK, in which
+	 * case we don't want to reduce the cwnd either.
 	 */
-	if (is_tlp_dupack) {
+	if (is_tlp_dupack ||
+	    !before(ack, tp->tlp_high_seq) && (flag & FLAG_DSACKING_ACK)) {
 		tp->tlp_high_seq = 0;
 		return;
 	}
 
 	if (after(ack, tp->tlp_high_seq)) {
 		tp->tlp_high_seq = 0;
-		/* Don't reduce cwnd if DSACK arrives for TLP retrans. */
-		if (!(flag & FLAG_DSACKING_ACK)) {
-			tcp_init_cwnd_reduction(sk);
-			tcp_set_ca_state(sk, TCP_CA_CWR);
-			tcp_end_cwnd_reduction(sk);
-			tcp_try_keep_open(sk);
-			NET_INC_STATS_BH(sock_net(sk),
-					 LINUX_MIB_TCPLOSSPROBERECOVERY);
-		}
+		tcp_init_cwnd_reduction(sk);
+		tcp_set_ca_state(sk, TCP_CA_CWR);
+		tcp_end_cwnd_reduction(sk);
+		tcp_try_keep_open(sk);
+		NET_INC_STATS_BH(sock_net(sk),
+				 LINUX_MIB_TCPLOSSPROBERECOVERY);
 	}
 }
 

-- 
tg: (52c9b12..) net-next/tlp-dsack-handling (depends on: net-next/master)

The net-next tree is closed, therefore it is not appropriate to submit net-next
changes at this time.

Thanks.

Le 11/12/2014 21:14, David Miller a écrit :

The net-next tree is closed, therefore it is not appropriate to submit net-next
changes at this time.

Sorry for that, I missed your mail announcing that next-next was closed.
Will resubmit when it is open again.

regards,

Sébastien.

Thanks.

On Thu, Dec 11, 2014 at 2:58 PM, Sébastien Barré
[off-list ref] wrote:

When the peer has delayed ack enabled, it may reply to a probe with an
ACK+D-SACK, with ack value set to tlp_high_seq. In the current code,
such ACK+DSACK will be missed and only at next, higher ack will the TLP
episode be considered done. Since the DSACK is not present anymore,
this will cost a cwnd reduction.

This patch ensures that this scenario does not cause a cwnd reduction, since
receiving an ACK+DSACK indicates that both the initial segment and the probe
have been received by the peer.

Cc: Gregory Detal <redacted>
Cc: Nandita Dukkipati <redacted>
Cc: Yuchung Cheng <redacted>
Signed-off-by: Sébastien Barré <redacted>

BTW, I like this idea, and I'll test it.

Two suggestions for the next iteration:

(1) for TCP patch style, I'd suggest a commit first-line like:

  tcp: avoid reducing cwnd when ACK+DSACK is received

(2) Your patch is essentially a reworking of the is_tlp_dupack
computation. So rather than splitting the logic between the
is_tlp_dupack computation and the new expression inside the if
condition, I'd suggest moving and centralizing all the logic in the if
condition. There is no real need for the is_tlp_dupack variable.

neal