^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1) // SPDX-License-Identifier: GPL-2.0-only
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3) * TCP CUBIC: Binary Increase Congestion control for TCP v2.3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 4) * Home page:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 5) * http://netsrv.csc.ncsu.edu/twiki/bin/view/Main/BIC
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 6) * This is from the implementation of CUBIC TCP in
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 7) * Sangtae Ha, Injong Rhee and Lisong Xu,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 8) * "CUBIC: A New TCP-Friendly High-Speed TCP Variant"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 9) * in ACM SIGOPS Operating System Review, July 2008.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 10) * Available from:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 11) * http://netsrv.csc.ncsu.edu/export/cubic_a_new_tcp_2008.pdf
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 12) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 13) * CUBIC integrates a new slow start algorithm, called HyStart.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 14) * The details of HyStart are presented in
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 15) * Sangtae Ha and Injong Rhee,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 16) * "Taming the Elephants: New TCP Slow Start", NCSU TechReport 2008.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 17) * Available from:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 18) * http://netsrv.csc.ncsu.edu/export/hystart_techreport_2008.pdf
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 19) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 20) * All testing results are available from:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 21) * http://netsrv.csc.ncsu.edu/wiki/index.php/TCP_Testing
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 22) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 23) * Unless CUBIC is enabled and congestion window is large
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 24) * this behaves the same as the original Reno.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 25) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 26)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 27) #include <linux/mm.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 28) #include <linux/module.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 29) #include <linux/math64.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 30) #include <net/tcp.h>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 31)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 32) #define BICTCP_BETA_SCALE 1024 /* Scale factor beta calculation
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 33) * max_cwnd = snd_cwnd * beta
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 34) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 35) #define BICTCP_HZ 10 /* BIC HZ 2^10 = 1024 */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 36)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 37) /* Two methods of hybrid slow start */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 38) #define HYSTART_ACK_TRAIN 0x1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 39) #define HYSTART_DELAY 0x2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 40)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 41) /* Number of delay samples for detecting the increase of delay */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 42) #define HYSTART_MIN_SAMPLES 8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 43) #define HYSTART_DELAY_MIN (4000U) /* 4 ms */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 44) #define HYSTART_DELAY_MAX (16000U) /* 16 ms */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 45) #define HYSTART_DELAY_THRESH(x) clamp(x, HYSTART_DELAY_MIN, HYSTART_DELAY_MAX)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 46)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 47) static int fast_convergence __read_mostly = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 48) static int beta __read_mostly = 717; /* = 717/1024 (BICTCP_BETA_SCALE) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 49) static int initial_ssthresh __read_mostly;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 50) static int bic_scale __read_mostly = 41;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 51) static int tcp_friendliness __read_mostly = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 52)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 53) static int hystart __read_mostly = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 54) static int hystart_detect __read_mostly = HYSTART_ACK_TRAIN | HYSTART_DELAY;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 55) static int hystart_low_window __read_mostly = 16;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 56) static int hystart_ack_delta_us __read_mostly = 2000;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 57)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 58) static u32 cube_rtt_scale __read_mostly;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 59) static u32 beta_scale __read_mostly;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 60) static u64 cube_factor __read_mostly;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 61)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 62) /* Note parameters that are used for precomputing scale factors are read-only */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 63) module_param(fast_convergence, int, 0644);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 64) MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 65) module_param(beta, int, 0644);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 66) MODULE_PARM_DESC(beta, "beta for multiplicative increase");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 67) module_param(initial_ssthresh, int, 0644);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 68) MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 69) module_param(bic_scale, int, 0444);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 70) MODULE_PARM_DESC(bic_scale, "scale (scaled by 1024) value for bic function (bic_scale/1024)");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 71) module_param(tcp_friendliness, int, 0644);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 72) MODULE_PARM_DESC(tcp_friendliness, "turn on/off tcp friendliness");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 73) module_param(hystart, int, 0644);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 74) MODULE_PARM_DESC(hystart, "turn on/off hybrid slow start algorithm");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 75) module_param(hystart_detect, int, 0644);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 76) MODULE_PARM_DESC(hystart_detect, "hybrid slow start detection mechanisms"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 77) " 1: packet-train 2: delay 3: both packet-train and delay");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 78) module_param(hystart_low_window, int, 0644);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 79) MODULE_PARM_DESC(hystart_low_window, "lower bound cwnd for hybrid slow start");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 80) module_param(hystart_ack_delta_us, int, 0644);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 81) MODULE_PARM_DESC(hystart_ack_delta_us, "spacing between ack's indicating train (usecs)");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 82)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 83) /* BIC TCP Parameters */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 84) struct bictcp {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 85) u32 cnt; /* increase cwnd by 1 after ACKs */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 86) u32 last_max_cwnd; /* last maximum snd_cwnd */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 87) u32 last_cwnd; /* the last snd_cwnd */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 88) u32 last_time; /* time when updated last_cwnd */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 89) u32 bic_origin_point;/* origin point of bic function */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 90) u32 bic_K; /* time to origin point
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 91) from the beginning of the current epoch */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 92) u32 delay_min; /* min delay (usec) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 93) u32 epoch_start; /* beginning of an epoch */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 94) u32 ack_cnt; /* number of acks */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 95) u32 tcp_cwnd; /* estimated tcp cwnd */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 96) u16 unused;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 97) u8 sample_cnt; /* number of samples to decide curr_rtt */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 98) u8 found; /* the exit point is found? */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 99) u32 round_start; /* beginning of each round */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 100) u32 end_seq; /* end_seq of the round */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 101) u32 last_ack; /* last time when the ACK spacing is close */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 102) u32 curr_rtt; /* the minimum rtt of current round */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 103) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 104)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 105) static inline void bictcp_reset(struct bictcp *ca)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 106) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 107) ca->cnt = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 108) ca->last_max_cwnd = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 109) ca->last_cwnd = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 110) ca->last_time = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 111) ca->bic_origin_point = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 112) ca->bic_K = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 113) ca->delay_min = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 114) ca->epoch_start = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 115) ca->ack_cnt = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 116) ca->tcp_cwnd = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 117) ca->found = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 118) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 119)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 120) static inline u32 bictcp_clock_us(const struct sock *sk)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 121) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 122) return tcp_sk(sk)->tcp_mstamp;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 123) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 124)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 125) static inline void bictcp_hystart_reset(struct sock *sk)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 126) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 127) struct tcp_sock *tp = tcp_sk(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 128) struct bictcp *ca = inet_csk_ca(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 129)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 130) ca->round_start = ca->last_ack = bictcp_clock_us(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 131) ca->end_seq = tp->snd_nxt;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 132) ca->curr_rtt = ~0U;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 133) ca->sample_cnt = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 134) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 135)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 136) static void bictcp_init(struct sock *sk)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 137) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 138) struct bictcp *ca = inet_csk_ca(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 139)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 140) bictcp_reset(ca);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 141)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 142) if (hystart)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 143) bictcp_hystart_reset(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 144)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 145) if (!hystart && initial_ssthresh)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 146) tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 147) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 148)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 149) static void bictcp_cwnd_event(struct sock *sk, enum tcp_ca_event event)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 150) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 151) if (event == CA_EVENT_TX_START) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 152) struct bictcp *ca = inet_csk_ca(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 153) u32 now = tcp_jiffies32;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 154) s32 delta;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 155)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 156) delta = now - tcp_sk(sk)->lsndtime;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 157)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 158) /* We were application limited (idle) for a while.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 159) * Shift epoch_start to keep cwnd growth to cubic curve.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 160) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 161) if (ca->epoch_start && delta > 0) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 162) ca->epoch_start += delta;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 163) if (after(ca->epoch_start, now))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 164) ca->epoch_start = now;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 165) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 166) return;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 167) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 168) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 169)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 170) /* calculate the cubic root of x using a table lookup followed by one
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 171) * Newton-Raphson iteration.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 172) * Avg err ~= 0.195%
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 173) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 174) static u32 cubic_root(u64 a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 175) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 176) u32 x, b, shift;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 177) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 178) * cbrt(x) MSB values for x MSB values in [0..63].
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 179) * Precomputed then refined by hand - Willy Tarreau
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 180) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 181) * For x in [0..63],
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 182) * v = cbrt(x << 18) - 1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 183) * cbrt(x) = (v[x] + 10) >> 6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 184) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 185) static const u8 v[] = {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 186) /* 0x00 */ 0, 54, 54, 54, 118, 118, 118, 118,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 187) /* 0x08 */ 123, 129, 134, 138, 143, 147, 151, 156,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 188) /* 0x10 */ 157, 161, 164, 168, 170, 173, 176, 179,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 189) /* 0x18 */ 181, 185, 187, 190, 192, 194, 197, 199,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 190) /* 0x20 */ 200, 202, 204, 206, 209, 211, 213, 215,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 191) /* 0x28 */ 217, 219, 221, 222, 224, 225, 227, 229,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 192) /* 0x30 */ 231, 232, 234, 236, 237, 239, 240, 242,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 193) /* 0x38 */ 244, 245, 246, 248, 250, 251, 252, 254,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 194) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 195)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 196) b = fls64(a);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 197) if (b < 7) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 198) /* a in [0..63] */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 199) return ((u32)v[(u32)a] + 35) >> 6;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 200) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 201)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 202) b = ((b * 84) >> 8) - 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 203) shift = (a >> (b * 3));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 204)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 205) x = ((u32)(((u32)v[shift] + 10) << b)) >> 6;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 206)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 207) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 208) * Newton-Raphson iteration
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 209) * 2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 210) * x = ( 2 * x + a / x ) / 3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 211) * k+1 k k
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 212) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 213) x = (2 * x + (u32)div64_u64(a, (u64)x * (u64)(x - 1)));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 214) x = ((x * 341) >> 10);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 215) return x;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 216) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 217)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 218) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 219) * Compute congestion window to use.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 220) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 221) static inline void bictcp_update(struct bictcp *ca, u32 cwnd, u32 acked)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 222) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 223) u32 delta, bic_target, max_cnt;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 224) u64 offs, t;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 225)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 226) ca->ack_cnt += acked; /* count the number of ACKed packets */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 227)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 228) if (ca->last_cwnd == cwnd &&
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 229) (s32)(tcp_jiffies32 - ca->last_time) <= HZ / 32)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 230) return;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 231)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 232) /* The CUBIC function can update ca->cnt at most once per jiffy.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 233) * On all cwnd reduction events, ca->epoch_start is set to 0,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 234) * which will force a recalculation of ca->cnt.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 235) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 236) if (ca->epoch_start && tcp_jiffies32 == ca->last_time)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 237) goto tcp_friendliness;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 238)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 239) ca->last_cwnd = cwnd;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 240) ca->last_time = tcp_jiffies32;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 241)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 242) if (ca->epoch_start == 0) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 243) ca->epoch_start = tcp_jiffies32; /* record beginning */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 244) ca->ack_cnt = acked; /* start counting */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 245) ca->tcp_cwnd = cwnd; /* syn with cubic */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 246)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 247) if (ca->last_max_cwnd <= cwnd) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 248) ca->bic_K = 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 249) ca->bic_origin_point = cwnd;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 250) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 251) /* Compute new K based on
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 252) * (wmax-cwnd) * (srtt>>3 / HZ) / c * 2^(3*bictcp_HZ)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 253) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 254) ca->bic_K = cubic_root(cube_factor
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 255) * (ca->last_max_cwnd - cwnd));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 256) ca->bic_origin_point = ca->last_max_cwnd;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 257) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 258) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 259)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 260) /* cubic function - calc*/
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 261) /* calculate c * time^3 / rtt,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 262) * while considering overflow in calculation of time^3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 263) * (so time^3 is done by using 64 bit)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 264) * and without the support of division of 64bit numbers
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 265) * (so all divisions are done by using 32 bit)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 266) * also NOTE the unit of those veriables
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 267) * time = (t - K) / 2^bictcp_HZ
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 268) * c = bic_scale >> 10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 269) * rtt = (srtt >> 3) / HZ
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 270) * !!! The following code does not have overflow problems,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 271) * if the cwnd < 1 million packets !!!
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 272) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 273)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 274) t = (s32)(tcp_jiffies32 - ca->epoch_start);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 275) t += usecs_to_jiffies(ca->delay_min);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 276) /* change the unit from HZ to bictcp_HZ */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 277) t <<= BICTCP_HZ;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 278) do_div(t, HZ);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 279)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 280) if (t < ca->bic_K) /* t - K */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 281) offs = ca->bic_K - t;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 282) else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 283) offs = t - ca->bic_K;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 284)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 285) /* c/rtt * (t-K)^3 */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 286) delta = (cube_rtt_scale * offs * offs * offs) >> (10+3*BICTCP_HZ);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 287) if (t < ca->bic_K) /* below origin*/
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 288) bic_target = ca->bic_origin_point - delta;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 289) else /* above origin*/
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 290) bic_target = ca->bic_origin_point + delta;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 291)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 292) /* cubic function - calc bictcp_cnt*/
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 293) if (bic_target > cwnd) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 294) ca->cnt = cwnd / (bic_target - cwnd);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 295) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 296) ca->cnt = 100 * cwnd; /* very small increment*/
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 297) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 298)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 299) /*
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 300) * The initial growth of cubic function may be too conservative
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 301) * when the available bandwidth is still unknown.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 302) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 303) if (ca->last_max_cwnd == 0 && ca->cnt > 20)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 304) ca->cnt = 20; /* increase cwnd 5% per RTT */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 305)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 306) tcp_friendliness:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 307) /* TCP Friendly */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 308) if (tcp_friendliness) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 309) u32 scale = beta_scale;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 310)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 311) delta = (cwnd * scale) >> 3;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 312) while (ca->ack_cnt > delta) { /* update tcp cwnd */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 313) ca->ack_cnt -= delta;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 314) ca->tcp_cwnd++;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 315) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 316)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 317) if (ca->tcp_cwnd > cwnd) { /* if bic is slower than tcp */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 318) delta = ca->tcp_cwnd - cwnd;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 319) max_cnt = cwnd / delta;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 320) if (ca->cnt > max_cnt)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 321) ca->cnt = max_cnt;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 322) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 323) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 324)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 325) /* The maximum rate of cwnd increase CUBIC allows is 1 packet per
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 326) * 2 packets ACKed, meaning cwnd grows at 1.5x per RTT.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 327) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 328) ca->cnt = max(ca->cnt, 2U);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 329) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 330)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 331) static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 332) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 333) struct tcp_sock *tp = tcp_sk(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 334) struct bictcp *ca = inet_csk_ca(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 335)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 336) if (!tcp_is_cwnd_limited(sk))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 337) return;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 338)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 339) if (tcp_in_slow_start(tp)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 340) acked = tcp_slow_start(tp, acked);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 341) if (!acked)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 342) return;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 343) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 344) bictcp_update(ca, tp->snd_cwnd, acked);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 345) tcp_cong_avoid_ai(tp, ca->cnt, acked);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 346) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 347)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 348) static u32 bictcp_recalc_ssthresh(struct sock *sk)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 349) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 350) const struct tcp_sock *tp = tcp_sk(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 351) struct bictcp *ca = inet_csk_ca(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 352)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 353) ca->epoch_start = 0; /* end of epoch */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 354)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 355) /* Wmax and fast convergence */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 356) if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 357) ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 358) / (2 * BICTCP_BETA_SCALE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 359) else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 360) ca->last_max_cwnd = tp->snd_cwnd;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 361)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 362) return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 363) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 364)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 365) static void bictcp_state(struct sock *sk, u8 new_state)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 366) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 367) if (new_state == TCP_CA_Loss) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 368) bictcp_reset(inet_csk_ca(sk));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 369) bictcp_hystart_reset(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 370) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 371) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 372)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 373) /* Account for TSO/GRO delays.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 374) * Otherwise short RTT flows could get too small ssthresh, since during
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 375) * slow start we begin with small TSO packets and ca->delay_min would
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 376) * not account for long aggregation delay when TSO packets get bigger.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 377) * Ideally even with a very small RTT we would like to have at least one
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 378) * TSO packet being sent and received by GRO, and another one in qdisc layer.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 379) * We apply another 100% factor because @rate is doubled at this point.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 380) * We cap the cushion to 1ms.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 381) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 382) static u32 hystart_ack_delay(struct sock *sk)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 383) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 384) unsigned long rate;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 385)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 386) rate = READ_ONCE(sk->sk_pacing_rate);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 387) if (!rate)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 388) return 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 389) return min_t(u64, USEC_PER_MSEC,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 390) div64_ul((u64)GSO_MAX_SIZE * 4 * USEC_PER_SEC, rate));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 391) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 392)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 393) static void hystart_update(struct sock *sk, u32 delay)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 394) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 395) struct tcp_sock *tp = tcp_sk(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 396) struct bictcp *ca = inet_csk_ca(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 397) u32 threshold;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 398)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 399) if (after(tp->snd_una, ca->end_seq))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 400) bictcp_hystart_reset(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 401)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 402) if (hystart_detect & HYSTART_ACK_TRAIN) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 403) u32 now = bictcp_clock_us(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 404)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 405) /* first detection parameter - ack-train detection */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 406) if ((s32)(now - ca->last_ack) <= hystart_ack_delta_us) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 407) ca->last_ack = now;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 408)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 409) threshold = ca->delay_min + hystart_ack_delay(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 410)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 411) /* Hystart ack train triggers if we get ack past
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 412) * ca->delay_min/2.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 413) * Pacing might have delayed packets up to RTT/2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 414) * during slow start.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 415) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 416) if (sk->sk_pacing_status == SK_PACING_NONE)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 417) threshold >>= 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 418)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 419) if ((s32)(now - ca->round_start) > threshold) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 420) ca->found = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 421) pr_debug("hystart_ack_train (%u > %u) delay_min %u (+ ack_delay %u) cwnd %u\n",
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 422) now - ca->round_start, threshold,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 423) ca->delay_min, hystart_ack_delay(sk), tp->snd_cwnd);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 424) NET_INC_STATS(sock_net(sk),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 425) LINUX_MIB_TCPHYSTARTTRAINDETECT);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 426) NET_ADD_STATS(sock_net(sk),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 427) LINUX_MIB_TCPHYSTARTTRAINCWND,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 428) tp->snd_cwnd);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 429) tp->snd_ssthresh = tp->snd_cwnd;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 430) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 431) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 432) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 433)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 434) if (hystart_detect & HYSTART_DELAY) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 435) /* obtain the minimum delay of more than sampling packets */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 436) if (ca->curr_rtt > delay)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 437) ca->curr_rtt = delay;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 438) if (ca->sample_cnt < HYSTART_MIN_SAMPLES) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 439) ca->sample_cnt++;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 440) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 441) if (ca->curr_rtt > ca->delay_min +
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 442) HYSTART_DELAY_THRESH(ca->delay_min >> 3)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 443) ca->found = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 444) NET_INC_STATS(sock_net(sk),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 445) LINUX_MIB_TCPHYSTARTDELAYDETECT);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 446) NET_ADD_STATS(sock_net(sk),
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 447) LINUX_MIB_TCPHYSTARTDELAYCWND,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 448) tp->snd_cwnd);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 449) tp->snd_ssthresh = tp->snd_cwnd;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 450) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 451) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 452) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 453) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 454)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 455) static void bictcp_acked(struct sock *sk, const struct ack_sample *sample)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 456) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 457) const struct tcp_sock *tp = tcp_sk(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 458) struct bictcp *ca = inet_csk_ca(sk);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 459) u32 delay;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 460)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 461) /* Some calls are for duplicates without timetamps */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 462) if (sample->rtt_us < 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 463) return;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 464)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 465) /* Discard delay samples right after fast recovery */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 466) if (ca->epoch_start && (s32)(tcp_jiffies32 - ca->epoch_start) < HZ)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 467) return;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 468)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 469) delay = sample->rtt_us;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 470) if (delay == 0)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 471) delay = 1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 472)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 473) /* first time call or link delay decreases */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 474) if (ca->delay_min == 0 || ca->delay_min > delay)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 475) ca->delay_min = delay;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 476)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 477) /* hystart triggers when cwnd is larger than some threshold */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 478) if (!ca->found && tcp_in_slow_start(tp) && hystart &&
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 479) tp->snd_cwnd >= hystart_low_window)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 480) hystart_update(sk, delay);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 481) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 482)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 483) static struct tcp_congestion_ops cubictcp __read_mostly = {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 484) .init = bictcp_init,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 485) .ssthresh = bictcp_recalc_ssthresh,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 486) .cong_avoid = bictcp_cong_avoid,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 487) .set_state = bictcp_state,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 488) .undo_cwnd = tcp_reno_undo_cwnd,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 489) .cwnd_event = bictcp_cwnd_event,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 490) .pkts_acked = bictcp_acked,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 491) .owner = THIS_MODULE,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 492) .name = "cubic",
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 493) };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 494)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 495) static int __init cubictcp_register(void)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 496) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 497) BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 498)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 499) /* Precompute a bunch of the scaling factors that are used per-packet
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 500) * based on SRTT of 100ms
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 501) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 502)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 503) beta_scale = 8*(BICTCP_BETA_SCALE+beta) / 3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 504) / (BICTCP_BETA_SCALE - beta);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 505)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 506) cube_rtt_scale = (bic_scale * 10); /* 1024*c/rtt */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 507)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 508) /* calculate the "K" for (wmax-cwnd) = c/rtt * K^3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 509) * so K = cubic_root( (wmax-cwnd)*rtt/c )
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 510) * the unit of K is bictcp_HZ=2^10, not HZ
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 511) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 512) * c = bic_scale >> 10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 513) * rtt = 100ms
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 514) *
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 515) * the following code has been designed and tested for
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 516) * cwnd < 1 million packets
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 517) * RTT < 100 seconds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 518) * HZ < 1,000,00 (corresponding to 10 nano-second)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 519) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 520)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 521) /* 1/c * 2^2*bictcp_HZ * srtt */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 522) cube_factor = 1ull << (10+3*BICTCP_HZ); /* 2^40 */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 523)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 524) /* divide by bic_scale and by constant Srtt (100ms) */
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 525) do_div(cube_factor, bic_scale * 10);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 526)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 527) return tcp_register_congestion_control(&cubictcp);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 528) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 529)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 530) static void __exit cubictcp_unregister(void)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 531) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 532) tcp_unregister_congestion_control(&cubictcp);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 533) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 534)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 535) module_init(cubictcp_register);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 536) module_exit(cubictcp_unregister);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 537)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 538) MODULE_AUTHOR("Sangtae Ha, Stephen Hemminger");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 539) MODULE_LICENSE("GPL");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 540) MODULE_DESCRIPTION("CUBIC TCP");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 541) MODULE_VERSION("2.3");