tcp: fix tcp_cong_avoid_ai() credit accumulation bug with decreases in w

[ Upstream commit 9949afa4 ]

The recent change to tcp_cong_avoid_ai() to handle stretch ACKs
introduced a bug where snd_cwnd_cnt could accumulate a very large
value while w was large, and then if w was reduced snd_cwnd could be
incremented by a large delta, leading to a large burst and high packet
loss. This was tickled when CUBIC's bictcp_update() sets "ca->cnt =
100 * cwnd".

This bug crept in while preparing the upstream version of
814d488c.

Testing: This patch has been tested in datacenter netperf transfers
and live youtube.com and google.com servers.

Fixes: 814d488c ("tcp: fix the timid additive increase on stretch ACKs")
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>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

net/ipv4/tcp_cong.c

@@ -309,6 +309,12 @@ EXPORT_SYMBOL_GPL(tcp_slow_start);
  */
 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked)
 {
+	/* If credits accumulated at a higher w, apply them gently now. */
+	if (tp->snd_cwnd_cnt >= w) {
+		tp->snd_cwnd_cnt = 0;
+		tp->snd_cwnd++;
+	}
+
 	tp->snd_cwnd_cnt += acked;
 	if (tp->snd_cwnd_cnt >= w) {
 		u32 delta = tp->snd_cwnd_cnt / w;