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

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 814d488. Testing: This patch has been tested in datacenter netperf transfers and live youtube.com and google.com servers. Fixes: 814d488 ("tcp: fix the timid additive increase on stretch ACKs") Signed-off-by: Neal Cardwell <[email protected]> Signed-off-by: Yuchung Cheng <[email protected]> Signed-off-by: Eric Dumazet <[email protected]> Signed-off-by: David S. Miller <[email protected]>
fastly · Mar 11, 2015 · 9949afa · 9949afa
1 parent 366c1bd
commit 9949afa
Showing 1 changed file with 6 additions and 0 deletions.
diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c
@@ -378,6 +378,12 @@ 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;