tcp: fix stretch ACK bugs in Reno
authorNeal Cardwell <ncardwell@google.com>
Thu, 29 Jan 2015 01:01:37 +0000 (20:01 -0500)
committerDavid S. Miller <davem@davemloft.net>
Thu, 29 Jan 2015 06:18:38 +0000 (22:18 -0800)
Change Reno to properly handle stretch ACKs in additive increase mode
by passing in the count of ACKed packets to tcp_cong_avoid_ai().

In addition, if snd_cwnd crosses snd_ssthresh during slow start
processing, and we then exit slow start mode, we need to carry over
any remaining "credit" for packets ACKed and apply that to additive
increase by passing this remaining "acked" count to
tcp_cong_avoid_ai().

Reported-by: Eyal Perry <eyalpe@mellanox.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_cong.c

index faaee53..8670e68 100644 (file)
@@ -335,11 +335,13 @@ void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked)
                return;
 
        /* In "safe" area, increase. */
-       if (tp->snd_cwnd <= tp->snd_ssthresh)
-               tcp_slow_start(tp, acked);
+       if (tp->snd_cwnd <= tp->snd_ssthresh) {
+               acked = tcp_slow_start(tp, acked);
+               if (!acked)
+                       return;
+       }
        /* In dangerous area, increase slowly. */
-       else
-               tcp_cong_avoid_ai(tp, tp->snd_cwnd, 1);
+       tcp_cong_avoid_ai(tp, tp->snd_cwnd, acked);
 }
 EXPORT_SYMBOL_GPL(tcp_reno_cong_avoid);