tcp: do not pass timestamp to tcp_rack_detect_loss()
authorEric Dumazet <edumazet@google.com>
Tue, 25 Apr 2017 17:15:33 +0000 (10:15 -0700)
committerDavid S. Miller <davem@davemloft.net>
Wed, 26 Apr 2017 18:44:37 +0000 (14:44 -0400)
We can use tp->tcp_mstamp as it contains a recent timestamp.

This removes a call to skb_mstamp_get() from tcp_rack_reo_timeout()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_recovery.c

index d8acbd9f477a2ac6b0f8eee1bf59f3ab43abff07..fdac262e277b2f25492f155bbb295d6d87e31d02 100644 (file)
@@ -45,8 +45,7 @@ static bool tcp_rack_sent_after(const struct skb_mstamp *t1,
  * or tcp_time_to_recover()'s "Trick#1: the loss is proven" code path will
  * make us enter the CA_Recovery state.
  */
-static void tcp_rack_detect_loss(struct sock *sk, const struct skb_mstamp *now,
-                                u32 *reo_timeout)
+static void tcp_rack_detect_loss(struct sock *sk, u32 *reo_timeout)
 {
        struct tcp_sock *tp = tcp_sk(sk);
        struct sk_buff *skb;
@@ -79,7 +78,7 @@ static void tcp_rack_detect_loss(struct sock *sk, const struct skb_mstamp *now,
                         * A packet is lost if its elapsed time is beyond
                         * the recent RTT plus the reordering window.
                         */
-                       u32 elapsed = skb_mstamp_us_delta(now,
+                       u32 elapsed = skb_mstamp_us_delta(&tp->tcp_mstamp,
                                                          &skb->skb_mstamp);
                        s32 remaining = tp->rack.rtt_us + reo_wnd - elapsed;
 
@@ -115,7 +114,7 @@ void tcp_rack_mark_lost(struct sock *sk, const struct skb_mstamp *now)
 
        /* Reset the advanced flag to avoid unnecessary queue scanning */
        tp->rack.advanced = 0;
-       tcp_rack_detect_loss(sk, now, &timeout);
+       tcp_rack_detect_loss(sk, &timeout);
        if (timeout) {
                timeout = usecs_to_jiffies(timeout + TCP_REO_TIMEOUT_MIN);
                inet_csk_reset_xmit_timer(sk, ICSK_TIME_REO_TIMEOUT,
@@ -165,12 +164,10 @@ void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
 void tcp_rack_reo_timeout(struct sock *sk)
 {
        struct tcp_sock *tp = tcp_sk(sk);
-       struct skb_mstamp now;
        u32 timeout, prior_inflight;
 
-       skb_mstamp_get(&now);
        prior_inflight = tcp_packets_in_flight(tp);
-       tcp_rack_detect_loss(sk, &now, &timeout);
+       tcp_rack_detect_loss(sk, &timeout);
        if (prior_inflight != tcp_packets_in_flight(tp)) {
                if (inet_csk(sk)->icsk_ca_state != TCP_CA_Recovery) {
                        tcp_enter_recovery(sk, false);