xnu-3248.20.55.tar.gz

[apple/xnu.git] / bsd / netinet / tcp_ledbat.c

diff --git a/bsd/netinet/tcp_ledbat.c b/bsd/netinet/tcp_ledbat.c
index 1d1d5e5e7ab28f8d870efed24058b7600a58cf68..aa2d32dbc711dad9cf37be7f59df4d4c09367823 100644 (file)
--- a/bsd/netinet/tcp_ledbat.c
+++ b/bsd/netinet/tcp_ledbat.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2010-2011 Apple Inc. All rights reserved.
+ * Copyright (c) 2010-2014 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  *
@@ -58,7 +58,7 @@
 int tcp_ledbat_init(struct tcpcb *tp);
 int tcp_ledbat_cleanup(struct tcpcb *tp);
 void tcp_ledbat_cwnd_init(struct tcpcb *tp);
-void tcp_ledbat_inseq_ack_rcvd(struct tcpcb *tp, struct tcphdr *th);
+void tcp_ledbat_congestion_avd(struct tcpcb *tp, struct tcphdr *th);
 void tcp_ledbat_ack_rcvd(struct tcpcb *tp, struct tcphdr *th);
 void tcp_ledbat_pre_fr(struct tcpcb *tp);
 void tcp_ledbat_post_fr(struct tcpcb *tp, struct tcphdr *th);
@@ -72,7 +72,7 @@ struct tcp_cc_algo tcp_cc_ledbat = {
        .init = tcp_ledbat_init,
        .cleanup = tcp_ledbat_cleanup,
        .cwnd_init = tcp_ledbat_cwnd_init,
-       .inseq_ack_rcvd = tcp_ledbat_inseq_ack_rcvd,
+       .congestion_avd = tcp_ledbat_congestion_avd,
        .ack_rcvd = tcp_ledbat_ack_rcvd,
        .pre_fr = tcp_ledbat_pre_fr,
        .post_fr = tcp_ledbat_post_fr,
@@ -82,10 +82,6 @@ struct tcp_cc_algo tcp_cc_ledbat = {
        .switch_to = tcp_ledbat_switch_cc
 };
 
-extern int tcp_do_rfc3465;
-extern int tcp_do_rfc3465_lim2;
-extern uint32_t get_base_rtt(struct tcpcb *tp);
-
 /* Target queuing delay in milliseconds. This includes the processing 
  * and scheduling delay on both of the end-hosts. A LEDBAT sender tries 
  * to keep queuing delay below this limit. When the queuing delay
@@ -108,15 +104,15 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, bg_target_qdelay, CTLFLAG_RW | CTLFLAG_LOCKE
  * max_allowed_cwnd = allowed_increase + (tether * flight_size)
  * cwnd = min(cwnd, max_allowed_cwnd)
  *
- * 'Allowed_increase' parameter is set to 2. If the flight size is zero, then
- * we want the congestion window to be at least 2 packets to reduce the
- * delay induced by delayed ack. This helps when the receiver is acking every
- * other packet.
+ * 'Allowed_increase' parameter is set to 8. If the flight size is zero, then
+ * we want the congestion window to be at least 8 packets to reduce the
+ * delay induced by delayed ack. This helps when the receiver is acking 
+ * more than 2 packets at a time (stretching acks for better performance).
  * 
  * 'Tether' is also set to 2. We do not want this to limit the growth of cwnd
  * during slow-start.
  */ 
-int allowed_increase = 2;
+int allowed_increase = 8;
 SYSCTL_INT(_net_inet_tcp, OID_AUTO, bg_allowed_increase, CTLFLAG_RW | CTLFLAG_LOCKED, 
        &allowed_increase, 1, "Additive constant used to calculate max allowed congestion window");
 
@@ -224,11 +220,11 @@ tcp_ledbat_cwnd_init(struct tcpcb *tp) {
  * This gets called only during congestion avoidance phase.
  */
 void
-tcp_ledbat_inseq_ack_rcvd(struct tcpcb *tp, struct tcphdr *th) {
+tcp_ledbat_congestion_avd(struct tcpcb *tp, struct tcphdr *th) {
        int acked = 0;
        u_int32_t incr = 0;
 
-       acked = th->th_ack - tp->snd_una;
+       acked = BYTES_ACKED(th, tp);
        tp->t_bytes_acked += acked;
        if (tp->t_bytes_acked > tp->snd_cwnd) {
                tp->t_bytes_acked -= tp->snd_cwnd;
@@ -260,7 +256,7 @@ tcp_ledbat_ack_rcvd(struct tcpcb *tp, struct tcphdr *th) {
        register u_int incr = tp->t_maxseg;
        int acked = 0;
 
-       acked = th->th_ack - tp->snd_una;
+       acked = BYTES_ACKED(th, tp);
        tp->t_bytes_acked += acked;
        if (cw >= tp->bg_ssthresh) {
                /* congestion-avoidance */
@@ -318,9 +314,13 @@ tcp_ledbat_post_fr(struct tcpcb *tp, struct tcphdr *th) {
         * snd_ssthresh outstanding data.  But in case we
         * would be inclined to send a burst, better to do
         * it via the slow start mechanism.
+        *
+        * If the flight size is zero, then make congestion 
+        * window to be worth at least 2 segments to avoid 
+        * delayed acknowledgement (draft-ietf-tcpm-rfc3782-bis-05).
         */
        if (ss < (int32_t)tp->snd_ssthresh)
-               tp->snd_cwnd = ss + tp->t_maxseg;
+               tp->snd_cwnd = max(ss, tp->t_maxseg) + tp->t_maxseg;
        else
                tp->snd_cwnd = tp->snd_ssthresh;
        tp->t_bytes_acked = 0;
@@ -373,14 +373,12 @@ tcp_ledbat_after_timeout(struct tcpcb *tp) {
                u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
                if (win < 2)
                        win = 2;
-               tp->snd_cwnd = tp->t_maxseg;
                tp->snd_ssthresh = win * tp->t_maxseg;
-               tp->t_bytes_acked = 0;
-               tp->t_dupacks = 0;
 
                if (tp->bg_ssthresh > tp->snd_ssthresh)
                        tp->bg_ssthresh = tp->snd_ssthresh;
 
+               tp->snd_cwnd = tp->t_maxseg;
                tcp_cc_resize_sndbuf(tp);
        }
 }
@@ -401,6 +399,12 @@ tcp_ledbat_after_timeout(struct tcpcb *tp) {
 
 int
 tcp_ledbat_delay_ack(struct tcpcb *tp, struct tcphdr *th) {
+       /* If any flag other than TH_ACK is set, set "end-of-write" bit */
+       if (th->th_flags & ~TH_ACK)
+               tp->t_flagsext |= TF_STREAMEOW;
+       else
+               tp->t_flagsext &= ~(TF_STREAMEOW);
+
        if ((tp->t_flags & TF_RXWIN0SENT) == 0 &&
                (th->th_flags & TH_PUSH) == 0 &&
                (tp->t_unacksegs == 1))