From c462238d6a6d8ee855bda10f9fff442971540ed2 Mon Sep 17 00:00:00 2001 From: Stephen Hemminger Date: Fri, 20 Apr 2007 17:07:51 -0700 Subject: [TCP]: TCP Illinois congestion control (rev3) This is an implementation of TCP Illinois invented by Shao Liu at University of Illinois. It is a another variant of Reno which adapts the alpha and beta parameters based on RTT. The basic idea is to increase window less rapidly as delay approaches the maximum. See the papers and talks to get a more complete description. Signed-off-by: Stephen Hemminger Signed-off-by: David S. Miller --- net/ipv4/tcp_illinois.c | 284 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 284 insertions(+) create mode 100644 net/ipv4/tcp_illinois.c (limited to 'net/ipv4/tcp_illinois.c') diff --git a/net/ipv4/tcp_illinois.c b/net/ipv4/tcp_illinois.c new file mode 100644 index 000000000000..91a2a34604cd --- /dev/null +++ b/net/ipv4/tcp_illinois.c @@ -0,0 +1,284 @@ +/* + * TCP Illinois congestion control. + * Home page: + * http://www.ews.uiuc.edu/~shaoliu/tcpillinois/index.html + * + * The algorithm is described in: + * "TCP-Illinois: A Loss and Delay-Based Congestion Control Algorithm + * for High-Speed Networks" + * http://www.ews.uiuc.edu/~shaoliu/papersandslides/liubassri06perf.pdf + * + * Implemented from description in paper and ns-2 simulation. + * Copyright (C) 2007 Stephen Hemminger + */ + +#include +#include +#include +#include +#include + +#define ALPHA_SHIFT 7 +#define ALPHA_SCALE (1u<last_alpha = ALPHA_BASE; + ca->min_rtt = 0x7fffffff; +} + +/* + * Keep track of min, max and average RTT + */ +static void tcp_illinois_rtt_calc(struct sock *sk, u32 rtt) +{ + struct tcp_illinois *ca = inet_csk_ca(sk); + + if (rtt < ca->min_rtt) + ca->min_rtt = rtt; + if (rtt > ca->max_rtt) + ca->max_rtt = rtt; + + if (++ca->rtt_cnt == 1) + ca->sum_rtt = rtt; + else + ca->sum_rtt += rtt; +} + +/* max queuing delay */ +static inline u32 max_delay(const struct tcp_illinois *ca) +{ + return ca->max_rtt - ca->min_rtt; +} + +/* average queueing delay */ +static u32 avg_delay(struct tcp_illinois *ca) +{ + u64 avg_rtt = ca->sum_rtt; + + do_div(avg_rtt, ca->rtt_cnt); + + ca->sum_rtt = 0; + ca->rtt_cnt = 0; + + return avg_rtt - ca->min_rtt; +} + +/* + * Compute value of alpha used for additive increase. + * If small window then use 1.0, equivalent to Reno. + * + * For larger windows, adjust based on average delay. + * A. If average delay is at minimum (we are uncongested), + * then use large alpha (10.0) to increase faster. + * B. If average delay is at maximum (getting congested) + * then use small alpha (1.0) + * + * The result is a convex window growth curve. + */ +static u32 alpha(const struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct tcp_illinois *ca = inet_csk_ca(sk); + u32 dm = max_delay(ca); + u32 da = avg_delay(ca); + u32 d1, a; + + if (tp->snd_cwnd < win_thresh) + return ALPHA_BASE; /* same as Reno (1.0) */ + + d1 = dm / 100; + if (da <= d1) { + /* Don't let noise force agressive response */ + if (ca->rtt_low < THETA) { + ++ca->rtt_low; + return ca->last_alpha; + } else + return ALPHA_MAX; + } + + ca->rtt_low = 0; + + /* + * Based on: + * + * (dm - d1) amin amax + * k1 = ------------------- + * amax - amin + * + * (dm - d1) amin + * k2 = ---------------- - d1 + * amax - amin + * + * k1 + * alpha = ---------- + * k2 + da + */ + + dm -= d1; + da -= d1; + + a = (dm * ALPHA_MAX) / (dm - (da * (ALPHA_MAX - ALPHA_MIN)) / ALPHA_MIN); + ca->last_alpha = a; + return a; +} + +/* + * Increase window in response to successful acknowledgment. + */ +static void tcp_illinois_cong_avoid(struct sock *sk, u32 ack, u32 rtt, + u32 in_flight, int flag) +{ + struct tcp_sock *tp = tcp_sk(sk); + + /* RFC2861 only increase cwnd if fully utilized */ + if (!tcp_is_cwnd_limited(sk, in_flight)) + return; + + /* In slow start */ + if (tp->snd_cwnd <= tp->snd_ssthresh) + tcp_slow_start(tp); + + else { + /* additive increase cwnd += alpha / cwnd */ + if ((tp->snd_cwnd_cnt * alpha(sk)) >> ALPHA_SHIFT >= tp->snd_cwnd) { + if (tp->snd_cwnd < tp->snd_cwnd_clamp) + tp->snd_cwnd++; + tp->snd_cwnd_cnt = 0; + } else + tp->snd_cwnd_cnt++; + } +} + +/* + * Beta used for multiplicative decrease. + * For small window sizes returns same value as Reno (0.5) + * + * If delay is small (10% of max) then beta = 1/8 + * If delay is up to 80% of max then beta = 1/2 + * In between is a linear function + */ +static inline u32 beta(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct tcp_illinois *ca = inet_csk_ca(sk); + u32 dm = max_delay(ca); + u32 da = avg_delay(ca); + u32 d2, d3; + + if (tp->snd_cwnd < win_thresh) + return BETA_BASE; + + d2 = dm / 10; + if (da <= d2) + return BETA_MIN; + d3 = (8 * dm) / 10; + if (da >= d3 || d3 <= d2) + return BETA_MAX; + + /* + * Based on: + * + * bmin d3 - bmax d2 + * k3 = ------------------- + * d3 - d2 + * + * bmax - bmin + * k4 = ------------- + * d3 - d2 + * + * b = k3 + k4 da + */ + return (BETA_MIN * d3 - BETA_MAX * d2 + (BETA_MAX - BETA_MIN) * da) + / (d3 - d2); +} + +static u32 tcp_illinois_ssthresh(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + + /* Multiplicative decrease */ + return max((tp->snd_cwnd * beta(sk)) >> BETA_SHIFT, 2U); +} + +/* Extract info for TCP socket info provided via netlink. + * We aren't really doing Vegas, but we can provide RTT info + */ +static void tcp_illinois_get_info(struct sock *sk, u32 ext, + struct sk_buff *skb) +{ + const struct tcp_illinois *ca = inet_csk_ca(sk); + + if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) { + struct tcpvegas_info info = { + .tcpv_enabled = 1, + .tcpv_rttcnt = ca->rtt_cnt, + .tcpv_minrtt = ca->min_rtt, + }; + u64 avg_rtt = ca->sum_rtt; + do_div(avg_rtt, ca->rtt_cnt); + info.tcpv_rtt = avg_rtt; + + nla_put(skb, INET_DIAG_VEGASINFO, sizeof(info), &info); + } +} + +static struct tcp_congestion_ops tcp_illinois = { + .init = tcp_illinois_init, + .ssthresh = tcp_illinois_ssthresh, + .min_cwnd = tcp_reno_min_cwnd, + .cong_avoid = tcp_illinois_cong_avoid, + .rtt_sample = tcp_illinois_rtt_calc, + .get_info = tcp_illinois_get_info, + + .owner = THIS_MODULE, + .name = "illinois", +}; + +static int __init tcp_illinois_register(void) +{ + BUILD_BUG_ON(sizeof(struct tcp_illinois) > ICSK_CA_PRIV_SIZE); + return tcp_register_congestion_control(&tcp_illinois); +} + +static void __exit tcp_illinois_unregister(void) +{ + tcp_unregister_congestion_control(&tcp_illinois); +} + +module_init(tcp_illinois_register); +module_exit(tcp_illinois_unregister); + +MODULE_AUTHOR("Stephen Hemminger, Shao Liu"); +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("TCP Illinois"); +MODULE_VERSION("0.3"); -- cgit v1.2.3 From 65d1b4a7e73fe0e1f5275ad7d2d3547981480886 Mon Sep 17 00:00:00 2001 From: Stephen Hemminger Date: Mon, 23 Apr 2007 22:24:32 -0700 Subject: [TCP]: TCP Illinois update. This version more closely matches the paper, and fixes several math errors. The biggest difference is that it updates alpha/beta once per RTT Signed-off-by: Stephen Hemminger Signed-off-by: David S. Miller --- net/ipv4/tcp_illinois.c | 298 ++++++++++++++++++++++++++++++------------------ 1 file changed, 186 insertions(+), 112 deletions(-) (limited to 'net/ipv4/tcp_illinois.c') diff --git a/net/ipv4/tcp_illinois.c b/net/ipv4/tcp_illinois.c index 91a2a34604cd..ae6298600886 100644 --- a/net/ipv4/tcp_illinois.c +++ b/net/ipv4/tcp_illinois.c @@ -23,74 +23,106 @@ #define ALPHA_MIN ((3*ALPHA_SCALE)/10) /* ~0.3 */ #define ALPHA_MAX (10*ALPHA_SCALE) /* 10.0 */ #define ALPHA_BASE ALPHA_SCALE /* 1.0 */ +#define U32_MAX ((u32)~0U) +#define RTT_MAX (U32_MAX / ALPHA_MAX) /* 3.3 secs */ #define BETA_SHIFT 6 #define BETA_SCALE (1u<end_seq = tp->snd_nxt; + ca->cnt_rtt = 0; + ca->sum_rtt = 0; + + /* TODO: age max_rtt? */ +} + static void tcp_illinois_init(struct sock *sk) { - struct tcp_illinois *ca = inet_csk_ca(sk); + struct illinois *ca = inet_csk_ca(sk); + + ca->alpha = ALPHA_MAX; + ca->beta = BETA_BASE; + ca->base_rtt = 0x7fffffff; + ca->max_rtt = 0; + + ca->acked = 0; + ca->rtt_low = 0; + ca->rtt_above = 0; - ca->last_alpha = ALPHA_BASE; - ca->min_rtt = 0x7fffffff; + rtt_reset(sk); } -/* - * Keep track of min, max and average RTT - */ -static void tcp_illinois_rtt_calc(struct sock *sk, u32 rtt) +/* Measure RTT for each ack. */ +static void tcp_illinois_rtt_sample(struct sock *sk, u32 rtt) { - struct tcp_illinois *ca = inet_csk_ca(sk); + struct illinois *ca = inet_csk_ca(sk); + + /* ignore bogus values, this prevents wraparound in alpha math */ + if (rtt > RTT_MAX) + rtt = RTT_MAX; - if (rtt < ca->min_rtt) - ca->min_rtt = rtt; - if (rtt > ca->max_rtt) + /* keep track of minimum RTT seen so far */ + if (ca->base_rtt > rtt) + ca->base_rtt = rtt; + + /* and max */ + if (ca->max_rtt < rtt) ca->max_rtt = rtt; - if (++ca->rtt_cnt == 1) - ca->sum_rtt = rtt; - else - ca->sum_rtt += rtt; + ++ca->cnt_rtt; + ca->sum_rtt += rtt; } -/* max queuing delay */ -static inline u32 max_delay(const struct tcp_illinois *ca) +/* Capture count of packets covered by ack, to adjust for delayed acks */ +static void tcp_illinois_acked(struct sock *sk, u32 pkts_acked) { - return ca->max_rtt - ca->min_rtt; + struct illinois *ca = inet_csk_ca(sk); + ca->acked = pkts_acked; } -/* average queueing delay */ -static u32 avg_delay(struct tcp_illinois *ca) +/* Maximum queuing delay */ +static inline u32 max_delay(const struct illinois *ca) { - u64 avg_rtt = ca->sum_rtt; - - do_div(avg_rtt, ca->rtt_cnt); + return ca->max_rtt - ca->base_rtt; +} - ca->sum_rtt = 0; - ca->rtt_cnt = 0; +/* Average queuing delay */ +static inline u32 avg_delay(const struct illinois *ca) +{ + u64 t = ca->sum_rtt; - return avg_rtt - ca->min_rtt; + do_div(t, ca->cnt_rtt); + return t - ca->base_rtt; } /* @@ -101,32 +133,31 @@ static u32 avg_delay(struct tcp_illinois *ca) * A. If average delay is at minimum (we are uncongested), * then use large alpha (10.0) to increase faster. * B. If average delay is at maximum (getting congested) - * then use small alpha (1.0) + * then use small alpha (0.3) * * The result is a convex window growth curve. */ -static u32 alpha(const struct sock *sk) +static u32 alpha(struct illinois *ca, u32 da, u32 dm) { - struct tcp_sock *tp = tcp_sk(sk); - struct tcp_illinois *ca = inet_csk_ca(sk); - u32 dm = max_delay(ca); - u32 da = avg_delay(ca); - u32 d1, a; + u32 d1 = dm / 100; /* Low threshold */ - if (tp->snd_cwnd < win_thresh) - return ALPHA_BASE; /* same as Reno (1.0) */ - - d1 = dm / 100; if (da <= d1) { - /* Don't let noise force agressive response */ - if (ca->rtt_low < THETA) { - ++ca->rtt_low; - return ca->last_alpha; - } else + /* If never got out of low delay zone, then use max */ + if (!ca->rtt_above) return ALPHA_MAX; + + /* Wait for 5 good RTT's before allowing alpha to go alpha max. + * This prevents one good RTT from causing sudden window increase. + */ + if (++ca->rtt_low < theta) + return ca->alpha; + + ca->rtt_low = 0; + ca->rtt_above = 0; + return ALPHA_MAX; } - ca->rtt_low = 0; + ca->rtt_above = 1; /* * Based on: @@ -146,37 +177,8 @@ static u32 alpha(const struct sock *sk) dm -= d1; da -= d1; - - a = (dm * ALPHA_MAX) / (dm - (da * (ALPHA_MAX - ALPHA_MIN)) / ALPHA_MIN); - ca->last_alpha = a; - return a; -} - -/* - * Increase window in response to successful acknowledgment. - */ -static void tcp_illinois_cong_avoid(struct sock *sk, u32 ack, u32 rtt, - u32 in_flight, int flag) -{ - struct tcp_sock *tp = tcp_sk(sk); - - /* RFC2861 only increase cwnd if fully utilized */ - if (!tcp_is_cwnd_limited(sk, in_flight)) - return; - - /* In slow start */ - if (tp->snd_cwnd <= tp->snd_ssthresh) - tcp_slow_start(tp); - - else { - /* additive increase cwnd += alpha / cwnd */ - if ((tp->snd_cwnd_cnt * alpha(sk)) >> ALPHA_SHIFT >= tp->snd_cwnd) { - if (tp->snd_cwnd < tp->snd_cwnd_clamp) - tp->snd_cwnd++; - tp->snd_cwnd_cnt = 0; - } else - tp->snd_cwnd_cnt++; - } + return (dm * ALPHA_MAX) / + (dm + (da * (ALPHA_MAX - ALPHA_MIN)) / ALPHA_MIN); } /* @@ -187,20 +189,14 @@ static void tcp_illinois_cong_avoid(struct sock *sk, u32 ack, u32 rtt, * If delay is up to 80% of max then beta = 1/2 * In between is a linear function */ -static inline u32 beta(struct sock *sk) +static u32 beta(u32 da, u32 dm) { - struct tcp_sock *tp = tcp_sk(sk); - struct tcp_illinois *ca = inet_csk_ca(sk); - u32 dm = max_delay(ca); - u32 da = avg_delay(ca); u32 d2, d3; - if (tp->snd_cwnd < win_thresh) - return BETA_BASE; - d2 = dm / 10; if (da <= d2) return BETA_MIN; + d3 = (8 * dm) / 10; if (da >= d3 || d3 <= d2) return BETA_MAX; @@ -222,31 +218,107 @@ static inline u32 beta(struct sock *sk) / (d3 - d2); } +/* Update alpha and beta values once per RTT */ +static void update_params(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct illinois *ca = inet_csk_ca(sk); + + if (tp->snd_cwnd < win_thresh) { + ca->alpha = ALPHA_BASE; + ca->beta = BETA_BASE; + } else if (ca->cnt_rtt > 0) { + u32 dm = max_delay(ca); + u32 da = avg_delay(ca); + + ca->alpha = alpha(ca, da, dm); + ca->beta = beta(da, dm); + } + + rtt_reset(sk); +} + +/* + * In case of loss, reset to default values + */ +static void tcp_illinois_state(struct sock *sk, u8 new_state) +{ + struct illinois *ca = inet_csk_ca(sk); + + if (new_state == TCP_CA_Loss) { + ca->alpha = ALPHA_BASE; + ca->beta = BETA_BASE; + ca->rtt_low = 0; + ca->rtt_above = 0; + rtt_reset(sk); + } +} + +/* + * Increase window in response to successful acknowledgment. + */ +static void tcp_illinois_cong_avoid(struct sock *sk, u32 ack, u32 rtt, + u32 in_flight, int flag) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct illinois *ca = inet_csk_ca(sk); + + if (after(ack, ca->end_seq)) + update_params(sk); + + /* RFC2861 only increase cwnd if fully utilized */ + if (!tcp_is_cwnd_limited(sk, in_flight)) + return; + + /* In slow start */ + if (tp->snd_cwnd <= tp->snd_ssthresh) + tcp_slow_start(tp); + + else { + u32 delta; + + /* snd_cwnd_cnt is # of packets since last cwnd increment */ + tp->snd_cwnd_cnt += ca->acked; + ca->acked = 1; + + /* This is close approximation of: + * tp->snd_cwnd += alpha/tp->snd_cwnd + */ + delta = (tp->snd_cwnd_cnt * ca->alpha) >> ALPHA_SHIFT; + if (delta >= tp->snd_cwnd) { + tp->snd_cwnd = min(tp->snd_cwnd + delta / tp->snd_cwnd, + (u32) tp->snd_cwnd_clamp); + tp->snd_cwnd_cnt = 0; + } + } +} + static u32 tcp_illinois_ssthresh(struct sock *sk) { struct tcp_sock *tp = tcp_sk(sk); + struct illinois *ca = inet_csk_ca(sk); /* Multiplicative decrease */ - return max((tp->snd_cwnd * beta(sk)) >> BETA_SHIFT, 2U); + return max((tp->snd_cwnd * ca->beta) >> BETA_SHIFT, 2U); } -/* Extract info for TCP socket info provided via netlink. - * We aren't really doing Vegas, but we can provide RTT info - */ -static void tcp_illinois_get_info(struct sock *sk, u32 ext, - struct sk_buff *skb) + +/* Extract info for Tcp socket info provided via netlink. */ +static void tcp_illinois_info(struct sock *sk, u32 ext, + struct sk_buff *skb) { - const struct tcp_illinois *ca = inet_csk_ca(sk); + const struct illinois *ca = inet_csk_ca(sk); if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) { struct tcpvegas_info info = { .tcpv_enabled = 1, - .tcpv_rttcnt = ca->rtt_cnt, - .tcpv_minrtt = ca->min_rtt, + .tcpv_rttcnt = ca->cnt_rtt, + .tcpv_minrtt = ca->base_rtt, }; - u64 avg_rtt = ca->sum_rtt; - do_div(avg_rtt, ca->rtt_cnt); - info.tcpv_rtt = avg_rtt; + u64 t = ca->sum_rtt; + + do_div(t, ca->cnt_rtt); + info.tcpv_rtt = t; nla_put(skb, INET_DIAG_VEGASINFO, sizeof(info), &info); } @@ -257,8 +329,10 @@ static struct tcp_congestion_ops tcp_illinois = { .ssthresh = tcp_illinois_ssthresh, .min_cwnd = tcp_reno_min_cwnd, .cong_avoid = tcp_illinois_cong_avoid, - .rtt_sample = tcp_illinois_rtt_calc, - .get_info = tcp_illinois_get_info, + .set_state = tcp_illinois_state, + .rtt_sample = tcp_illinois_rtt_sample, + .get_info = tcp_illinois_info, + .pkts_acked = tcp_illinois_acked, .owner = THIS_MODULE, .name = "illinois", @@ -266,7 +340,7 @@ static struct tcp_congestion_ops tcp_illinois = { static int __init tcp_illinois_register(void) { - BUILD_BUG_ON(sizeof(struct tcp_illinois) > ICSK_CA_PRIV_SIZE); + BUILD_BUG_ON(sizeof(struct illinois) > ICSK_CA_PRIV_SIZE); return tcp_register_congestion_control(&tcp_illinois); } @@ -281,4 +355,4 @@ module_exit(tcp_illinois_unregister); MODULE_AUTHOR("Stephen Hemminger, Shao Liu"); MODULE_LICENSE("GPL"); MODULE_DESCRIPTION("TCP Illinois"); -MODULE_VERSION("0.3"); +MODULE_VERSION("1.0"); -- cgit v1.2.3 From 164891aadf1721fca4dce473bb0e0998181537c6 Mon Sep 17 00:00:00 2001 From: Stephen Hemminger Date: Mon, 23 Apr 2007 22:26:16 -0700 Subject: [TCP]: Congestion control API update. Do some simple changes to make congestion control API faster/cleaner. * use ktime_t rather than timeval * merge rtt sampling into existing ack callback this means one indirect call versus two per ack. * use flags bits to store options/settings Signed-off-by: Stephen Hemminger Signed-off-by: David S. Miller --- include/linux/skbuff.h | 5 +++++ include/net/tcp.h | 9 +++++---- net/ipv4/tcp_bic.c | 2 +- net/ipv4/tcp_cong.c | 14 +++++++------- net/ipv4/tcp_cubic.c | 2 +- net/ipv4/tcp_htcp.c | 2 +- net/ipv4/tcp_illinois.c | 16 +++++++--------- net/ipv4/tcp_input.c | 25 ++++++++----------------- net/ipv4/tcp_lp.c | 8 +++++--- net/ipv4/tcp_output.c | 2 +- net/ipv4/tcp_vegas.c | 10 +++++++--- net/ipv4/tcp_veno.c | 10 +++++++--- net/ipv4/tcp_westwood.c | 2 +- net/ipv4/tcp_yeah.c | 6 ++++-- net/ipv4/tcp_yeah.h | 7 +++++-- 15 files changed, 65 insertions(+), 55 deletions(-) (limited to 'net/ipv4/tcp_illinois.c') diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h index 50f6f6a094cf..2694cb3ca763 100644 --- a/include/linux/skbuff.h +++ b/include/linux/skbuff.h @@ -1569,6 +1569,11 @@ static inline void __net_timestamp(struct sk_buff *skb) skb->tstamp = ktime_get_real(); } +static inline ktime_t net_timedelta(ktime_t t) +{ + return ktime_sub(ktime_get_real(), t); +} + extern __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); extern __sum16 __skb_checksum_complete(struct sk_buff *skb); diff --git a/include/net/tcp.h b/include/net/tcp.h index 43910fe3c448..a385797f160a 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -629,9 +629,12 @@ enum tcp_ca_event { #define TCP_CA_MAX 128 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) +#define TCP_CONG_NON_RESTRICTED 0x1 +#define TCP_CONG_RTT_STAMP 0x2 + struct tcp_congestion_ops { struct list_head list; - int non_restricted; + unsigned long flags; /* initialize private data (optional) */ void (*init)(struct sock *sk); @@ -645,8 +648,6 @@ struct tcp_congestion_ops { /* do new cwnd calculation (required) */ void (*cong_avoid)(struct sock *sk, u32 ack, u32 rtt, u32 in_flight, int good_ack); - /* round trip time sample per acked packet (optional) */ - void (*rtt_sample)(struct sock *sk, u32 usrtt); /* call before changing ca_state (optional) */ void (*set_state)(struct sock *sk, u8 new_state); /* call when cwnd event occurs (optional) */ @@ -654,7 +655,7 @@ struct tcp_congestion_ops { /* new value of cwnd after loss (optional) */ u32 (*undo_cwnd)(struct sock *sk); /* hook for packet ack accounting (optional) */ - void (*pkts_acked)(struct sock *sk, u32 num_acked); + void (*pkts_acked)(struct sock *sk, u32 num_acked, ktime_t last); /* get info for inet_diag (optional) */ void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); diff --git a/net/ipv4/tcp_bic.c b/net/ipv4/tcp_bic.c index 5730333cd0ac..281c9f913257 100644 --- a/net/ipv4/tcp_bic.c +++ b/net/ipv4/tcp_bic.c @@ -206,7 +206,7 @@ static void bictcp_state(struct sock *sk, u8 new_state) /* Track delayed acknowledgment ratio using sliding window * ratio = (15*ratio + sample) / 16 */ -static void bictcp_acked(struct sock *sk, u32 cnt) +static void bictcp_acked(struct sock *sk, u32 cnt, ktime_t last) { const struct inet_connection_sock *icsk = inet_csk(sk); diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c index ccd88407e0cd..86b26539e54b 100644 --- a/net/ipv4/tcp_cong.c +++ b/net/ipv4/tcp_cong.c @@ -126,7 +126,7 @@ int tcp_set_default_congestion_control(const char *name) #endif if (ca) { - ca->non_restricted = 1; /* default is always allowed */ + ca->flags |= TCP_CONG_NON_RESTRICTED; /* default is always allowed */ list_move(&ca->list, &tcp_cong_list); ret = 0; } @@ -181,7 +181,7 @@ void tcp_get_allowed_congestion_control(char *buf, size_t maxlen) *buf = '\0'; rcu_read_lock(); list_for_each_entry_rcu(ca, &tcp_cong_list, list) { - if (!ca->non_restricted) + if (!(ca->flags & TCP_CONG_NON_RESTRICTED)) continue; offs += snprintf(buf + offs, maxlen - offs, "%s%s", @@ -212,16 +212,16 @@ int tcp_set_allowed_congestion_control(char *val) } } - /* pass 2 clear */ + /* pass 2 clear old values */ list_for_each_entry_rcu(ca, &tcp_cong_list, list) - ca->non_restricted = 0; + ca->flags &= ~TCP_CONG_NON_RESTRICTED; /* pass 3 mark as allowed */ while ((name = strsep(&val, " ")) && *name) { ca = tcp_ca_find(name); WARN_ON(!ca); if (ca) - ca->non_restricted = 1; + ca->flags |= TCP_CONG_NON_RESTRICTED; } out: spin_unlock(&tcp_cong_list_lock); @@ -256,7 +256,7 @@ int tcp_set_congestion_control(struct sock *sk, const char *name) if (!ca) err = -ENOENT; - else if (!(ca->non_restricted || capable(CAP_NET_ADMIN))) + else if (!((ca->flags & TCP_CONG_NON_RESTRICTED) || capable(CAP_NET_ADMIN))) err = -EPERM; else if (!try_module_get(ca->owner)) @@ -371,8 +371,8 @@ u32 tcp_reno_min_cwnd(const struct sock *sk) EXPORT_SYMBOL_GPL(tcp_reno_min_cwnd); struct tcp_congestion_ops tcp_reno = { + .flags = TCP_CONG_NON_RESTRICTED, .name = "reno", - .non_restricted = 1, .owner = THIS_MODULE, .ssthresh = tcp_reno_ssthresh, .cong_avoid = tcp_reno_cong_avoid, diff --git a/net/ipv4/tcp_cubic.c b/net/ipv4/tcp_cubic.c index 296845be912b..14224487b16b 100644 --- a/net/ipv4/tcp_cubic.c +++ b/net/ipv4/tcp_cubic.c @@ -334,7 +334,7 @@ static void bictcp_state(struct sock *sk, u8 new_state) /* Track delayed acknowledgment ratio using sliding window * ratio = (15*ratio + sample) / 16 */ -static void bictcp_acked(struct sock *sk, u32 cnt) +static void bictcp_acked(struct sock *sk, u32 cnt, ktime_t last) { const struct inet_connection_sock *icsk = inet_csk(sk); diff --git a/net/ipv4/tcp_htcp.c b/net/ipv4/tcp_htcp.c index 1020eb48d8d1..4ba4a7ae0a85 100644 --- a/net/ipv4/tcp_htcp.c +++ b/net/ipv4/tcp_htcp.c @@ -98,7 +98,7 @@ static inline void measure_rtt(struct sock *sk) } } -static void measure_achieved_throughput(struct sock *sk, u32 pkts_acked) +static void measure_achieved_throughput(struct sock *sk, u32 pkts_acked, ktime_t last) { const struct inet_connection_sock *icsk = inet_csk(sk); const struct tcp_sock *tp = tcp_sk(sk); diff --git a/net/ipv4/tcp_illinois.c b/net/ipv4/tcp_illinois.c index ae6298600886..8e3165917f72 100644 --- a/net/ipv4/tcp_illinois.c +++ b/net/ipv4/tcp_illinois.c @@ -83,9 +83,14 @@ static void tcp_illinois_init(struct sock *sk) } /* Measure RTT for each ack. */ -static void tcp_illinois_rtt_sample(struct sock *sk, u32 rtt) +static void tcp_illinois_acked(struct sock *sk, u32 pkts_acked, ktime_t last) { struct illinois *ca = inet_csk_ca(sk); + u32 rtt; + + ca->acked = pkts_acked; + + rtt = ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC; /* ignore bogus values, this prevents wraparound in alpha math */ if (rtt > RTT_MAX) @@ -103,13 +108,6 @@ static void tcp_illinois_rtt_sample(struct sock *sk, u32 rtt) ca->sum_rtt += rtt; } -/* Capture count of packets covered by ack, to adjust for delayed acks */ -static void tcp_illinois_acked(struct sock *sk, u32 pkts_acked) -{ - struct illinois *ca = inet_csk_ca(sk); - ca->acked = pkts_acked; -} - /* Maximum queuing delay */ static inline u32 max_delay(const struct illinois *ca) { @@ -325,12 +323,12 @@ static void tcp_illinois_info(struct sock *sk, u32 ext, } static struct tcp_congestion_ops tcp_illinois = { + .flags = TCP_CONG_RTT_STAMP, .init = tcp_illinois_init, .ssthresh = tcp_illinois_ssthresh, .min_cwnd = tcp_reno_min_cwnd, .cong_avoid = tcp_illinois_cong_avoid, .set_state = tcp_illinois_state, - .rtt_sample = tcp_illinois_rtt_sample, .get_info = tcp_illinois_info, .pkts_acked = tcp_illinois_acked, diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 633389390788..051f0f815f17 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -2402,14 +2402,6 @@ static int tcp_tso_acked(struct sock *sk, struct sk_buff *skb, return acked; } -static u32 tcp_usrtt(struct timeval *tv) -{ - struct timeval now; - - do_gettimeofday(&now); - return (now.tv_sec - tv->tv_sec) * 1000000 + (now.tv_usec - tv->tv_usec); -} - /* Remove acknowledged frames from the retransmission queue. */ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p) { @@ -2420,9 +2412,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p) int acked = 0; __s32 seq_rtt = -1; u32 pkts_acked = 0; - void (*rtt_sample)(struct sock *sk, u32 usrtt) - = icsk->icsk_ca_ops->rtt_sample; - struct timeval tv = { .tv_sec = 0, .tv_usec = 0 }; + ktime_t last_ackt = ktime_set(0,0); while ((skb = tcp_write_queue_head(sk)) && skb != tcp_send_head(sk)) { @@ -2471,7 +2461,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p) seq_rtt = -1; } else if (seq_rtt < 0) { seq_rtt = now - scb->when; - skb_get_timestamp(skb, &tv); + last_ackt = skb->tstamp; } if (sacked & TCPCB_SACKED_ACKED) tp->sacked_out -= tcp_skb_pcount(skb); @@ -2484,7 +2474,7 @@ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p) } } else if (seq_rtt < 0) { seq_rtt = now - scb->when; - skb_get_timestamp(skb, &tv); + last_ackt = skb->tstamp; } tcp_dec_pcount_approx(&tp->fackets_out, skb); tcp_packets_out_dec(tp, skb); @@ -2494,13 +2484,14 @@ static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p) } if (acked&FLAG_ACKED) { + const struct tcp_congestion_ops *ca_ops + = inet_csk(sk)->icsk_ca_ops; + tcp_ack_update_rtt(sk, acked, seq_rtt); tcp_ack_packets_out(sk); - if (rtt_sample && !(acked & FLAG_RETRANS_DATA_ACKED)) - (*rtt_sample)(sk, tcp_usrtt(&tv)); - if (icsk->icsk_ca_ops->pkts_acked) - icsk->icsk_ca_ops->pkts_acked(sk, pkts_acked); + if (ca_ops->pkts_acked) + ca_ops->pkts_acked(sk, pkts_acked, last_ackt); } #if FASTRETRANS_DEBUG > 0 diff --git a/net/ipv4/tcp_lp.c b/net/ipv4/tcp_lp.c index f0ebaf0e21cb..b4e062ab24a1 100644 --- a/net/ipv4/tcp_lp.c +++ b/net/ipv4/tcp_lp.c @@ -218,7 +218,7 @@ static u32 tcp_lp_owd_calculator(struct sock *sk) * 3. calc smoothed OWD (SOWD). * Most ideas come from the original TCP-LP implementation. */ -static void tcp_lp_rtt_sample(struct sock *sk, u32 usrtt) +static void tcp_lp_rtt_sample(struct sock *sk, u32 rtt) { struct lp *lp = inet_csk_ca(sk); s64 mowd = tcp_lp_owd_calculator(sk); @@ -261,11 +261,13 @@ static void tcp_lp_rtt_sample(struct sock *sk, u32 usrtt) * newReno in increase case. * We work it out by following the idea from TCP-LP's paper directly */ -static void tcp_lp_pkts_acked(struct sock *sk, u32 num_acked) +static void tcp_lp_pkts_acked(struct sock *sk, u32 num_acked, ktime_t last) { struct tcp_sock *tp = tcp_sk(sk); struct lp *lp = inet_csk_ca(sk); + tcp_lp_rtt_sample(sk, ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC); + /* calc inference */ if (tcp_time_stamp > tp->rx_opt.rcv_tsecr) lp->inference = 3 * (tcp_time_stamp - tp->rx_opt.rcv_tsecr); @@ -312,11 +314,11 @@ static void tcp_lp_pkts_acked(struct sock *sk, u32 num_acked) } static struct tcp_congestion_ops tcp_lp = { + .flags = TCP_CONG_RTT_STAMP, .init = tcp_lp_init, .ssthresh = tcp_reno_ssthresh, .cong_avoid = tcp_lp_cong_avoid, .min_cwnd = tcp_reno_min_cwnd, - .rtt_sample = tcp_lp_rtt_sample, .pkts_acked = tcp_lp_pkts_acked, .owner = THIS_MODULE, diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index 3a60aea744ae..e70a6840cb64 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -409,7 +409,7 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it, /* If congestion control is doing timestamping, we must * take such a timestamp before we potentially clone/copy. */ - if (icsk->icsk_ca_ops->rtt_sample) + if (icsk->icsk_ca_ops->flags & TCP_CONG_RTT_STAMP) __net_timestamp(skb); if (likely(clone_it)) { diff --git a/net/ipv4/tcp_vegas.c b/net/ipv4/tcp_vegas.c index 87e72bef6d08..f4104eeb5f26 100644 --- a/net/ipv4/tcp_vegas.c +++ b/net/ipv4/tcp_vegas.c @@ -120,10 +120,13 @@ static void tcp_vegas_init(struct sock *sk) * o min-filter RTT samples from a much longer window (forever for now) * to find the propagation delay (baseRTT) */ -static void tcp_vegas_rtt_calc(struct sock *sk, u32 usrtt) +static void tcp_vegas_pkts_acked(struct sock *sk, u32 cnt, ktime_t last) { struct vegas *vegas = inet_csk_ca(sk); - u32 vrtt = usrtt + 1; /* Never allow zero rtt or baseRTT */ + u32 vrtt; + + /* Never allow zero rtt or baseRTT */ + vrtt = (ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC) + 1; /* Filter to find propagation delay: */ if (vrtt < vegas->baseRTT) @@ -353,11 +356,12 @@ static void tcp_vegas_get_info(struct sock *sk, u32 ext, } static struct tcp_congestion_ops tcp_vegas = { + .flags = TCP_CONG_RTT_STAMP, .init = tcp_vegas_init, .ssthresh = tcp_reno_ssthresh, .cong_avoid = tcp_vegas_cong_avoid, .min_cwnd = tcp_reno_min_cwnd, - .rtt_sample = tcp_vegas_rtt_calc, + .pkts_acked = tcp_vegas_pkts_acked, .set_state = tcp_vegas_state, .cwnd_event = tcp_vegas_cwnd_event, .get_info = tcp_vegas_get_info, diff --git a/net/ipv4/tcp_veno.c b/net/ipv4/tcp_veno.c index ce57bf302f6c..0b50d0607a0e 100644 --- a/net/ipv4/tcp_veno.c +++ b/net/ipv4/tcp_veno.c @@ -69,10 +69,13 @@ static void tcp_veno_init(struct sock *sk) } /* Do rtt sampling needed for Veno. */ -static void tcp_veno_rtt_calc(struct sock *sk, u32 usrtt) +static void tcp_veno_pkts_acked(struct sock *sk, u32 cnt, ktime_t last) { struct veno *veno = inet_csk_ca(sk); - u32 vrtt = usrtt + 1; /* Never allow zero rtt or basertt */ + u32 vrtt; + + /* Never allow zero rtt or baseRTT */ + vrtt = (ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC) + 1; /* Filter to find propagation delay: */ if (vrtt < veno->basertt) @@ -199,10 +202,11 @@ static u32 tcp_veno_ssthresh(struct sock *sk) } static struct tcp_congestion_ops tcp_veno = { + .flags = TCP_CONG_RTT_STAMP, .init = tcp_veno_init, .ssthresh = tcp_veno_ssthresh, .cong_avoid = tcp_veno_cong_avoid, - .rtt_sample = tcp_veno_rtt_calc, + .pkts_acked = tcp_veno_pkts_acked, .set_state = tcp_veno_state, .cwnd_event = tcp_veno_cwnd_event, diff --git a/net/ipv4/tcp_westwood.c b/net/ipv4/tcp_westwood.c index ae1026a67720..e61e09dd513e 100644 --- a/net/ipv4/tcp_westwood.c +++ b/net/ipv4/tcp_westwood.c @@ -100,7 +100,7 @@ static void westwood_filter(struct westwood *w, u32 delta) * Called after processing group of packets. * but all westwood needs is the last sample of srtt. */ -static void tcp_westwood_pkts_acked(struct sock *sk, u32 cnt) +static void tcp_westwood_pkts_acked(struct sock *sk, u32 cnt, ktime_t last) { struct westwood *w = inet_csk_ca(sk); if (cnt > 0) diff --git a/net/ipv4/tcp_yeah.c b/net/ipv4/tcp_yeah.c index 46dd1bee583a..81ef02c1649a 100644 --- a/net/ipv4/tcp_yeah.c +++ b/net/ipv4/tcp_yeah.c @@ -64,13 +64,15 @@ static void tcp_yeah_init(struct sock *sk) } -static void tcp_yeah_pkts_acked(struct sock *sk, u32 pkts_acked) +static void tcp_yeah_pkts_acked(struct sock *sk, u32 pkts_acked, ktime_t last) { const struct inet_connection_sock *icsk = inet_csk(sk); struct yeah *yeah = inet_csk_ca(sk); if (icsk->icsk_ca_state == TCP_CA_Open) yeah->pkts_acked = pkts_acked; + + tcp_vegas_pkts_acked(sk, pkts_acked, last); } static void tcp_yeah_cong_avoid(struct sock *sk, u32 ack, @@ -237,11 +239,11 @@ static u32 tcp_yeah_ssthresh(struct sock *sk) { } static struct tcp_congestion_ops tcp_yeah = { + .flags = TCP_CONG_RTT_STAMP, .init = tcp_yeah_init, .ssthresh = tcp_yeah_ssthresh, .cong_avoid = tcp_yeah_cong_avoid, .min_cwnd = tcp_reno_min_cwnd, - .rtt_sample = tcp_vegas_rtt_calc, .set_state = tcp_vegas_state, .cwnd_event = tcp_vegas_cwnd_event, .get_info = tcp_vegas_get_info, diff --git a/net/ipv4/tcp_yeah.h b/net/ipv4/tcp_yeah.h index a62d82038fd0..33ad5385c188 100644 --- a/net/ipv4/tcp_yeah.h +++ b/net/ipv4/tcp_yeah.h @@ -81,10 +81,13 @@ static void tcp_vegas_state(struct sock *sk, u8 ca_state) * o min-filter RTT samples from a much longer window (forever for now) * to find the propagation delay (baseRTT) */ -static void tcp_vegas_rtt_calc(struct sock *sk, u32 usrtt) +static void tcp_vegas_pkts_acked(struct sock *sk, u32 cnt, ktime_t last) { struct vegas *vegas = inet_csk_ca(sk); - u32 vrtt = usrtt + 1; /* Never allow zero rtt or baseRTT */ + u32 vrtt; + + /* Never allow zero rtt or baseRTT */ + vrtt = (ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC) + 1; /* Filter to find propagation delay: */ if (vrtt < vegas->baseRTT) -- cgit v1.2.3 From 84299b3bc4eaedc0734fcc9052b01291e44445fc Mon Sep 17 00:00:00 2001 From: YOSHIFUJI Hideaki Date: Tue, 24 Apr 2007 16:21:38 -0700 Subject: [TCP]: Fix linkage errors on i386. To avoid raw division, use ktime_to_timeval() to get usec. Signed-off-by: YOSHIFUJI Hideaki Signed-off-by: David S. Miller --- include/linux/ktime.h | 6 ++++++ net/ipv4/tcp_illinois.c | 2 +- net/ipv4/tcp_lp.c | 2 +- net/ipv4/tcp_vegas.c | 2 +- net/ipv4/tcp_veno.c | 2 +- 5 files changed, 10 insertions(+), 4 deletions(-) (limited to 'net/ipv4/tcp_illinois.c') diff --git a/include/linux/ktime.h b/include/linux/ktime.h index 248305bb9a18..81bb9c7a4eb3 100644 --- a/include/linux/ktime.h +++ b/include/linux/ktime.h @@ -259,6 +259,12 @@ static inline s64 ktime_to_ns(const ktime_t kt) #endif +static inline s64 ktime_to_us(const ktime_t kt) +{ + struct timeval tv = ktime_to_timeval(kt); + return (s64) tv.tv_sec * USEC_PER_SEC + tv.tv_usec; +} + /* * The resolution of the clocks. The resolution value is returned in * the clock_getres() system call to give application programmers an diff --git a/net/ipv4/tcp_illinois.c b/net/ipv4/tcp_illinois.c index 8e3165917f72..4adc47c55351 100644 --- a/net/ipv4/tcp_illinois.c +++ b/net/ipv4/tcp_illinois.c @@ -90,7 +90,7 @@ static void tcp_illinois_acked(struct sock *sk, u32 pkts_acked, ktime_t last) ca->acked = pkts_acked; - rtt = ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC; + rtt = ktime_to_us(net_timedelta(last)); /* ignore bogus values, this prevents wraparound in alpha math */ if (rtt > RTT_MAX) diff --git a/net/ipv4/tcp_lp.c b/net/ipv4/tcp_lp.c index b4e062ab24a1..43294ad9f63e 100644 --- a/net/ipv4/tcp_lp.c +++ b/net/ipv4/tcp_lp.c @@ -266,7 +266,7 @@ static void tcp_lp_pkts_acked(struct sock *sk, u32 num_acked, ktime_t last) struct tcp_sock *tp = tcp_sk(sk); struct lp *lp = inet_csk_ca(sk); - tcp_lp_rtt_sample(sk, ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC); + tcp_lp_rtt_sample(sk, ktime_to_us(net_timedelta(last))); /* calc inference */ if (tcp_time_stamp > tp->rx_opt.rcv_tsecr) diff --git a/net/ipv4/tcp_vegas.c b/net/ipv4/tcp_vegas.c index 0f0ee7f732c3..73e19cf7df21 100644 --- a/net/ipv4/tcp_vegas.c +++ b/net/ipv4/tcp_vegas.c @@ -118,7 +118,7 @@ void tcp_vegas_pkts_acked(struct sock *sk, u32 cnt, ktime_t last) u32 vrtt; /* Never allow zero rtt or baseRTT */ - vrtt = (ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC) + 1; + vrtt = ktime_to_us(net_timedelta(last)) + 1; /* Filter to find propagation delay: */ if (vrtt < vegas->baseRTT) diff --git a/net/ipv4/tcp_veno.c b/net/ipv4/tcp_veno.c index 0b50d0607a0e..9edb340f2f95 100644 --- a/net/ipv4/tcp_veno.c +++ b/net/ipv4/tcp_veno.c @@ -75,7 +75,7 @@ static void tcp_veno_pkts_acked(struct sock *sk, u32 cnt, ktime_t last) u32 vrtt; /* Never allow zero rtt or baseRTT */ - vrtt = (ktime_to_ns(net_timedelta(last)) / NSEC_PER_USEC) + 1; + vrtt = ktime_to_us(net_timedelta(last)) + 1; /* Filter to find propagation delay: */ if (vrtt < veno->basertt) -- cgit v1.2.3