[linux-2.6-block.git] / net / ipv4 / tcp_cubic.c

/*
 * TCP CUBIC: Binary Increase Congestion control for TCP v2.3
 * Home page:
 *      http://netsrv.csc.ncsu.edu/twiki/bin/view/Main/BIC
 * This is from the implementation of CUBIC TCP in
 * Sangtae Ha, Injong Rhee and Lisong Xu,
 *  "CUBIC: A New TCP-Friendly High-Speed TCP Variant"
 *  in ACM SIGOPS Operating System Review, July 2008.
 * Available from:
 *  http://netsrv.csc.ncsu.edu/export/cubic_a_new_tcp_2008.pdf
 *
 * CUBIC integrates a new slow start algorithm, called HyStart.
 * The details of HyStart are presented in
 *  Sangtae Ha and Injong Rhee,
 *  "Taming the Elephants: New TCP Slow Start", NCSU TechReport 2008.
 * Available from:
 *  http://netsrv.csc.ncsu.edu/export/hystart_techreport_2008.pdf
 *
 * All testing results are available from:
 * http://netsrv.csc.ncsu.edu/wiki/index.php/TCP_Testing
 *
 * Unless CUBIC is enabled and congestion window is large
 * this behaves the same as the original Reno.
 */

#include <linux/mm.h>
#include <linux/module.h>
#include <linux/math64.h>
#include <net/tcp.h>

#define BICTCP_BETA_SCALE    1024	/* Scale factor beta calculation
					 * max_cwnd = snd_cwnd * beta
					 */
#define	BICTCP_HZ		10	/* BIC HZ 2^10 = 1024 */

/* Two methods of hybrid slow start */
#define HYSTART_ACK_TRAIN	0x1
#define HYSTART_DELAY		0x2

/* Number of delay samples for detecting the increase of delay */
#define HYSTART_MIN_SAMPLES	8
#define HYSTART_DELAY_MIN	(4U<<3)
#define HYSTART_DELAY_MAX	(16U<<3)
#define HYSTART_DELAY_THRESH(x)	clamp(x, HYSTART_DELAY_MIN, HYSTART_DELAY_MAX)

static int fast_convergence __read_mostly = 1;
static int beta __read_mostly = 717;	/* = 717/1024 (BICTCP_BETA_SCALE) */
static int initial_ssthresh __read_mostly;
static int bic_scale __read_mostly = 41;
static int tcp_friendliness __read_mostly = 1;

static int hystart __read_mostly = 1;
static int hystart_detect __read_mostly = HYSTART_ACK_TRAIN | HYSTART_DELAY;
static int hystart_low_window __read_mostly = 16;
static int hystart_ack_delta __read_mostly = 2;

static u32 cube_rtt_scale __read_mostly;
static u32 beta_scale __read_mostly;
static u64 cube_factor __read_mostly;

/* Note parameters that are used for precomputing scale factors are read-only */
module_param(fast_convergence, int, 0644);
MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
module_param(beta, int, 0644);
MODULE_PARM_DESC(beta, "beta for multiplicative increase");
module_param(initial_ssthresh, int, 0644);
MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
module_param(bic_scale, int, 0444);
MODULE_PARM_DESC(bic_scale, "scale (scaled by 1024) value for bic function (bic_scale/1024)");
module_param(tcp_friendliness, int, 0644);
MODULE_PARM_DESC(tcp_friendliness, "turn on/off tcp friendliness");
module_param(hystart, int, 0644);
MODULE_PARM_DESC(hystart, "turn on/off hybrid slow start algorithm");
module_param(hystart_detect, int, 0644);
MODULE_PARM_DESC(hystart_detect, "hyrbrid slow start detection mechanisms"
		 " 1: packet-train 2: delay 3: both packet-train and delay");
module_param(hystart_low_window, int, 0644);
MODULE_PARM_DESC(hystart_low_window, "lower bound cwnd for hybrid slow start");
module_param(hystart_ack_delta, int, 0644);
MODULE_PARM_DESC(hystart_ack_delta, "spacing between ack's indicating train (msecs)");

/* BIC TCP Parameters */
struct bictcp {
	u32	cnt;		/* increase cwnd by 1 after ACKs */
	u32	last_max_cwnd;	/* last maximum snd_cwnd */
	u32	loss_cwnd;	/* congestion window at last loss */
	u32	last_cwnd;	/* the last snd_cwnd */
	u32	last_time;	/* time when updated last_cwnd */
	u32	bic_origin_point;/* origin point of bic function */
	u32	bic_K;		/* time to origin point
				   from the beginning of the current epoch */
	u32	delay_min;	/* min delay (msec << 3) */
	u32	epoch_start;	/* beginning of an epoch */
	u32	ack_cnt;	/* number of acks */
	u32	tcp_cwnd;	/* estimated tcp cwnd */
	u16	unused;
	u8	sample_cnt;	/* number of samples to decide curr_rtt */
	u8	found;		/* the exit point is found? */
	u32	round_start;	/* beginning of each round */
	u32	end_seq;	/* end_seq of the round */
	u32	last_ack;	/* last time when the ACK spacing is close */
	u32	curr_rtt;	/* the minimum rtt of current round */
};

static inline void bictcp_reset(struct bictcp *ca)
{
	ca->cnt = 0;
	ca->last_max_cwnd = 0;
	ca->last_cwnd = 0;
	ca->last_time = 0;
	ca->bic_origin_point = 0;
	ca->bic_K = 0;
	ca->delay_min = 0;
	ca->epoch_start = 0;
	ca->ack_cnt = 0;
	ca->tcp_cwnd = 0;
	ca->found = 0;
}

static inline u32 bictcp_clock(void)
{
#if HZ < 1000
	return ktime_to_ms(ktime_get_real());
#else
	return jiffies_to_msecs(jiffies);
#endif
}

static inline void bictcp_hystart_reset(struct sock *sk)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	ca->round_start = ca->last_ack = bictcp_clock();
	ca->end_seq = tp->snd_nxt;
	ca->curr_rtt = 0;
	ca->sample_cnt = 0;
}

static void bictcp_init(struct sock *sk)
{
	struct bictcp *ca = inet_csk_ca(sk);

	bictcp_reset(ca);
	ca->loss_cwnd = 0;

	if (hystart)
		bictcp_hystart_reset(sk);

	if (!hystart && initial_ssthresh)
		tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
}

static void bictcp_cwnd_event(struct sock *sk, enum tcp_ca_event event)
{
	if (event == CA_EVENT_TX_START) {
		struct bictcp *ca = inet_csk_ca(sk);
		u32 now = tcp_time_stamp;
		s32 delta;

		delta = now - tcp_sk(sk)->lsndtime;

		/* We were application limited (idle) for a while.
		 * Shift epoch_start to keep cwnd growth to cubic curve.
		 */
		if (ca->epoch_start && delta > 0) {
			ca->epoch_start += delta;
			if (after(ca->epoch_start, now))
				ca->epoch_start = now;
		}
		return;
	}
}

/* calculate the cubic root of x using a table lookup followed by one
 * Newton-Raphson iteration.
 * Avg err ~= 0.195%
 */
static u32 cubic_root(u64 a)
{
	u32 x, b, shift;
	/*
	 * cbrt(x) MSB values for x MSB values in [0..63].
	 * Precomputed then refined by hand - Willy Tarreau
	 *
	 * For x in [0..63],
	 *   v = cbrt(x << 18) - 1
	 *   cbrt(x) = (v[x] + 10) >> 6
	 */
	static const u8 v[] = {
		/* 0x00 */    0,   54,   54,   54,  118,  118,  118,  118,
		/* 0x08 */  123,  129,  134,  138,  143,  147,  151,  156,
		/* 0x10 */  157,  161,  164,  168,  170,  173,  176,  179,
		/* 0x18 */  181,  185,  187,  190,  192,  194,  197,  199,
		/* 0x20 */  200,  202,  204,  206,  209,  211,  213,  215,
		/* 0x28 */  217,  219,  221,  222,  224,  225,  227,  229,
		/* 0x30 */  231,  232,  234,  236,  237,  239,  240,  242,
		/* 0x38 */  244,  245,  246,  248,  250,  251,  252,  254,
	};

	b = fls64(a);
	if (b < 7) {
		/* a in [0..63] */
		return ((u32)v[(u32)a] + 35) >> 6;
	}

	b = ((b * 84) >> 8) - 1;
	shift = (a >> (b * 3));

	x = ((u32)(((u32)v[shift] + 10) << b)) >> 6;

	/*
	 * Newton-Raphson iteration
	 *                         2
	 * x    = ( 2 * x  +  a / x  ) / 3
	 *  k+1          k         k
	 */
	x = (2 * x + (u32)div64_u64(a, (u64)x * (u64)(x - 1)));
	x = ((x * 341) >> 10);
	return x;
}

/*
 * Compute congestion window to use.
 */
static inline void bictcp_update(struct bictcp *ca, u32 cwnd, u32 acked)
{
	u32 delta, bic_target, max_cnt;
	u64 offs, t;

	ca->ack_cnt += acked;	/* count the number of ACKed packets */

	if (ca->last_cwnd == cwnd &&
	    (s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
		return;

	/* The CUBIC function can update ca->cnt at most once per jiffy.
	 * On all cwnd reduction events, ca->epoch_start is set to 0,
	 * which will force a recalculation of ca->cnt.
	 */
	if (ca->epoch_start && tcp_time_stamp == ca->last_time)
		goto tcp_friendliness;

	ca->last_cwnd = cwnd;
	ca->last_time = tcp_time_stamp;

	if (ca->epoch_start == 0) {
		ca->epoch_start = tcp_time_stamp;	/* record beginning */
		ca->ack_cnt = acked;			/* start counting */
		ca->tcp_cwnd = cwnd;			/* syn with cubic */

		if (ca->last_max_cwnd <= cwnd) {
			ca->bic_K = 0;
			ca->bic_origin_point = cwnd;
		} else {
			/* Compute new K based on
			 * (wmax-cwnd) * (srtt>>3 / HZ) / c * 2^(3*bictcp_HZ)
			 */
			ca->bic_K = cubic_root(cube_factor
					       * (ca->last_max_cwnd - cwnd));
			ca->bic_origin_point = ca->last_max_cwnd;
		}
	}

	/* cubic function - calc*/
	/* calculate c * time^3 / rtt,
	 *  while considering overflow in calculation of time^3
	 * (so time^3 is done by using 64 bit)
	 * and without the support of division of 64bit numbers
	 * (so all divisions are done by using 32 bit)
	 *  also NOTE the unit of those veriables
	 *	  time  = (t - K) / 2^bictcp_HZ
	 *	  c = bic_scale >> 10
	 * rtt  = (srtt >> 3) / HZ
	 * !!! The following code does not have overflow problems,
	 * if the cwnd < 1 million packets !!!
	 */

	t = (s32)(tcp_time_stamp - ca->epoch_start);
	t += msecs_to_jiffies(ca->delay_min >> 3);
	/* change the unit from HZ to bictcp_HZ */
	t <<= BICTCP_HZ;
	do_div(t, HZ);

	if (t < ca->bic_K)		/* t - K */
		offs = ca->bic_K - t;
	else
		offs = t - ca->bic_K;

	/* c/rtt * (t-K)^3 */
	delta = (cube_rtt_scale * offs * offs * offs) >> (10+3*BICTCP_HZ);
	if (t < ca->bic_K)                            /* below origin*/
		bic_target = ca->bic_origin_point - delta;
	else                                          /* above origin*/
		bic_target = ca->bic_origin_point + delta;

	/* cubic function - calc bictcp_cnt*/
	if (bic_target > cwnd) {
		ca->cnt = cwnd / (bic_target - cwnd);
	} else {
		ca->cnt = 100 * cwnd;              /* very small increment*/
	}

	/*
	 * The initial growth of cubic function may be too conservative
	 * when the available bandwidth is still unknown.
	 */
	if (ca->last_max_cwnd == 0 && ca->cnt > 20)
		ca->cnt = 20;	/* increase cwnd 5% per RTT */

tcp_friendliness:
	/* TCP Friendly */
	if (tcp_friendliness) {
		u32 scale = beta_scale;

		delta = (cwnd * scale) >> 3;
		while (ca->ack_cnt > delta) {		/* update tcp cwnd */
			ca->ack_cnt -= delta;
			ca->tcp_cwnd++;
		}

		if (ca->tcp_cwnd > cwnd) {	/* if bic is slower than tcp */
			delta = ca->tcp_cwnd - cwnd;
			max_cnt = cwnd / delta;
			if (ca->cnt > max_cnt)
				ca->cnt = max_cnt;
		}
	}

	/* The maximum rate of cwnd increase CUBIC allows is 1 packet per
	 * 2 packets ACKed, meaning cwnd grows at 1.5x per RTT.
	 */
	ca->cnt = max(ca->cnt, 2U);
}

static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	if (!tcp_is_cwnd_limited(sk))
		return;

	if (tcp_in_slow_start(tp)) {
		if (hystart && after(ack, ca->end_seq))
			bictcp_hystart_reset(sk);
		acked = tcp_slow_start(tp, acked);
		if (!acked)
			return;
	}
	bictcp_update(ca, tp->snd_cwnd, acked);
	tcp_cong_avoid_ai(tp, ca->cnt, acked);
}

static u32 bictcp_recalc_ssthresh(struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	ca->epoch_start = 0;	/* end of epoch */

	/* Wmax and fast convergence */
	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
		ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
			/ (2 * BICTCP_BETA_SCALE);
	else
		ca->last_max_cwnd = tp->snd_cwnd;

	ca->loss_cwnd = tp->snd_cwnd;

	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
}

static u32 bictcp_undo_cwnd(struct sock *sk)
{
	struct bictcp *ca = inet_csk_ca(sk);

	return max(tcp_sk(sk)->snd_cwnd, ca->loss_cwnd);
}

static void bictcp_state(struct sock *sk, u8 new_state)
{
	if (new_state == TCP_CA_Loss) {
		bictcp_reset(inet_csk_ca(sk));
		bictcp_hystart_reset(sk);
	}
}

static void hystart_update(struct sock *sk, u32 delay)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	if (ca->found & hystart_detect)
		return;

	if (hystart_detect & HYSTART_ACK_TRAIN) {
		u32 now = bictcp_clock();

		/* first detection parameter - ack-train detection */
		if ((s32)(now - ca->last_ack) <= hystart_ack_delta) {
			ca->last_ack = now;
			if ((s32)(now - ca->round_start) > ca->delay_min >> 4) {
				ca->found |= HYSTART_ACK_TRAIN;
				NET_INC_STATS(sock_net(sk),
					      LINUX_MIB_TCPHYSTARTTRAINDETECT);
				NET_ADD_STATS(sock_net(sk),
					      LINUX_MIB_TCPHYSTARTTRAINCWND,
					      tp->snd_cwnd);
				tp->snd_ssthresh = tp->snd_cwnd;
			}
		}
	}

	if (hystart_detect & HYSTART_DELAY) {
		/* obtain the minimum delay of more than sampling packets */
		if (ca->sample_cnt < HYSTART_MIN_SAMPLES) {
			if (ca->curr_rtt == 0 || ca->curr_rtt > delay)
				ca->curr_rtt = delay;

			ca->sample_cnt++;
		} else {
			if (ca->curr_rtt > ca->delay_min +
			    HYSTART_DELAY_THRESH(ca->delay_min >> 3)) {
				ca->found |= HYSTART_DELAY;
				NET_INC_STATS(sock_net(sk),
					      LINUX_MIB_TCPHYSTARTDELAYDETECT);
				NET_ADD_STATS(sock_net(sk),
					      LINUX_MIB_TCPHYSTARTDELAYCWND,
					      tp->snd_cwnd);
				tp->snd_ssthresh = tp->snd_cwnd;
			}
		}
	}
}

/* Track delayed acknowledgment ratio using sliding window
 * ratio = (15*ratio + sample) / 16
 */
static void bictcp_acked(struct sock *sk, const struct ack_sample *sample)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);
	u32 delay;

	/* Some calls are for duplicates without timetamps */
	if (sample->rtt_us < 0)
		return;

	/* Discard delay samples right after fast recovery */
	if (ca->epoch_start && (s32)(tcp_time_stamp - ca->epoch_start) < HZ)
		return;

	delay = (sample->rtt_us << 3) / USEC_PER_MSEC;
	if (delay == 0)
		delay = 1;

	/* first time call or link delay decreases */
	if (ca->delay_min == 0 || ca->delay_min > delay)
		ca->delay_min = delay;

	/* hystart triggers when cwnd is larger than some threshold */
	if (hystart && tcp_in_slow_start(tp) &&
	    tp->snd_cwnd >= hystart_low_window)
		hystart_update(sk, delay);
}

static struct tcp_congestion_ops cubictcp __read_mostly = {
	.init		= bictcp_init,
	.ssthresh	= bictcp_recalc_ssthresh,
	.cong_avoid	= bictcp_cong_avoid,
	.set_state	= bictcp_state,
	.undo_cwnd	= bictcp_undo_cwnd,
	.cwnd_event	= bictcp_cwnd_event,
	.pkts_acked     = bictcp_acked,
	.owner		= THIS_MODULE,
	.name		= "cubic",
};

static int __init cubictcp_register(void)
{
	BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);

	/* Precompute a bunch of the scaling factors that are used per-packet
	 * based on SRTT of 100ms
	 */

	beta_scale = 8*(BICTCP_BETA_SCALE+beta) / 3
		/ (BICTCP_BETA_SCALE - beta);

	cube_rtt_scale = (bic_scale * 10);	/* 1024*c/rtt */

	/* calculate the "K" for (wmax-cwnd) = c/rtt * K^3
	 *  so K = cubic_root( (wmax-cwnd)*rtt/c )
	 * the unit of K is bictcp_HZ=2^10, not HZ
	 *
	 *  c = bic_scale >> 10
	 *  rtt = 100ms
	 *
	 * the following code has been designed and tested for
	 * cwnd < 1 million packets
	 * RTT < 100 seconds
	 * HZ < 1,000,00  (corresponding to 10 nano-second)
	 */

	/* 1/c * 2^2*bictcp_HZ * srtt */
	cube_factor = 1ull << (10+3*BICTCP_HZ); /* 2^40 */

	/* divide by bic_scale and by constant Srtt (100ms) */
	do_div(cube_factor, bic_scale * 10);

	return tcp_register_congestion_control(&cubictcp);
}

static void __exit cubictcp_unregister(void)
{
	tcp_unregister_congestion_control(&cubictcp);
}

module_init(cubictcp_register);
module_exit(cubictcp_unregister);

MODULE_AUTHOR("Sangtae Ha, Stephen Hemminger");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("CUBIC TCP");
MODULE_VERSION("2.3");
Commit	Line	Data
df3271f3	1	/*
ae27e98a	2	* TCP CUBIC: Binary Increase Congestion control for TCP v2.3
6b3d6263 SH	3	* Home page:
6b3d6263 SH	4	* http://netsrv.csc.ncsu.edu/twiki/bin/view/Main/BIC
df3271f3	5	* This is from the implementation of CUBIC TCP in
ae27e98a SH	6	* Sangtae Ha, Injong Rhee and Lisong Xu,
	7	* "CUBIC: A New TCP-Friendly High-Speed TCP Variant"
	8	* in ACM SIGOPS Operating System Review, July 2008.
df3271f3	9	* Available from:
ae27e98a SH	10	* http://netsrv.csc.ncsu.edu/export/cubic_a_new_tcp_2008.pdf
	11	*
	12	* CUBIC integrates a new slow start algorithm, called HyStart.
	13	* The details of HyStart are presented in
	14	* Sangtae Ha and Injong Rhee,
	15	* "Taming the Elephants: New TCP Slow Start", NCSU TechReport 2008.
	16	* Available from:
	17	* http://netsrv.csc.ncsu.edu/export/hystart_techreport_2008.pdf
	18	*
	19	* All testing results are available from:
	20	* http://netsrv.csc.ncsu.edu/wiki/index.php/TCP_Testing
df3271f3 SH	21	*
	22	* Unless CUBIC is enabled and congestion window is large
	23	* this behaves the same as the original Reno.
	24	*/
	25
df3271f3 SH	26	#include <linux/mm.h>
df3271f3 SH	27	#include <linux/module.h>
6f6d6a1a	28	#include <linux/math64.h>
df3271f3	29	#include <net/tcp.h>
df3271f3 SH	30
	31	#define BICTCP_BETA_SCALE 1024 /* Scale factor beta calculation
	32	* max_cwnd = snd_cwnd * beta
	33	*/
df3271f3 SH	34	#define BICTCP_HZ 10 /* BIC HZ 2^10 = 1024 */
df3271f3 SH	35
ae27e98a SH	36	/* Two methods of hybrid slow start */
	37	#define HYSTART_ACK_TRAIN 0x1
	38	#define HYSTART_DELAY 0x2
	39
	40	/* Number of delay samples for detecting the increase of delay */
	41	#define HYSTART_MIN_SAMPLES 8
2b4636a5	42	#define HYSTART_DELAY_MIN (4U<<3)
ae27e98a SH	43	#define HYSTART_DELAY_MAX (16U<<3)
	44	#define HYSTART_DELAY_THRESH(x) clamp(x, HYSTART_DELAY_MIN, HYSTART_DELAY_MAX)
	45
59758f44	46	static int fast_convergence __read_mostly = 1;
6b3d6263	47	static int beta __read_mostly = 717; /* = 717/1024 (BICTCP_BETA_SCALE) */
66e1e3b2	48	static int initial_ssthresh __read_mostly;
59758f44 SH	49	static int bic_scale __read_mostly = 41;
59758f44 SH	50	static int tcp_friendliness __read_mostly = 1;
df3271f3	51
ae27e98a SH	52	static int hystart __read_mostly = 1;
	53	static int hystart_detect __read_mostly = HYSTART_ACK_TRAIN \| HYSTART_DELAY;
	54	static int hystart_low_window __read_mostly = 16;
aac46324	55	static int hystart_ack_delta __read_mostly = 2;
ae27e98a	56
59758f44 SH	57	static u32 cube_rtt_scale __read_mostly;
	58	static u32 beta_scale __read_mostly;
	59	static u64 cube_factor __read_mostly;
89b3d9aa SH	60
89b3d9aa SH	61	/* Note parameters that are used for precomputing scale factors are read-only */
df3271f3 SH	62	module_param(fast_convergence, int, 0644);
df3271f3 SH	63	MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
6b3d6263	64	module_param(beta, int, 0644);
df3271f3 SH	65	MODULE_PARM_DESC(beta, "beta for multiplicative increase");
	66	module_param(initial_ssthresh, int, 0644);
	67	MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
89b3d9aa	68	module_param(bic_scale, int, 0444);
df3271f3 SH	69	MODULE_PARM_DESC(bic_scale, "scale (scaled by 1024) value for bic function (bic_scale/1024)");
	70	module_param(tcp_friendliness, int, 0644);
	71	MODULE_PARM_DESC(tcp_friendliness, "turn on/off tcp friendliness");
ae27e98a SH	72	module_param(hystart, int, 0644);
	73	MODULE_PARM_DESC(hystart, "turn on/off hybrid slow start algorithm");
	74	module_param(hystart_detect, int, 0644);
	75	MODULE_PARM_DESC(hystart_detect, "hyrbrid slow start detection mechanisms"
	76	" 1: packet-train 2: delay 3: both packet-train and delay");
	77	module_param(hystart_low_window, int, 0644);
	78	MODULE_PARM_DESC(hystart_low_window, "lower bound cwnd for hybrid slow start");
aac46324	79	module_param(hystart_ack_delta, int, 0644);
aac46324	80	MODULE_PARM_DESC(hystart_ack_delta, "spacing between ack's indicating train (msecs)");
df3271f3	81
df3271f3 SH	82	/* BIC TCP Parameters */
	83	struct bictcp {
	84	u32 cnt; /* increase cwnd by 1 after ACKs */
688d1945	85	u32 last_max_cwnd; /* last maximum snd_cwnd */
df3271f3 SH	86	u32 loss_cwnd; /* congestion window at last loss */
	87	u32 last_cwnd; /* the last snd_cwnd */
	88	u32 last_time; /* time when updated last_cwnd */
	89	u32 bic_origin_point;/* origin point of bic function */
688d1945	90	u32 bic_K; /* time to origin point
688d1945	91	from the beginning of the current epoch */
17a6e9f1	92	u32 delay_min; /* min delay (msec << 3) */
df3271f3 SH	93	u32 epoch_start; /* beginning of an epoch */
	94	u32 ack_cnt; /* number of acks */
	95	u32 tcp_cwnd; /* estimated tcp cwnd */
9cd981dc	96	u16 unused;
ae27e98a SH	97	u8 sample_cnt; /* number of samples to decide curr_rtt */
	98	u8 found; /* the exit point is found? */
	99	u32 round_start; /* beginning of each round */
	100	u32 end_seq; /* end_seq of the round */
17a6e9f1	101	u32 last_ack; /* last time when the ACK spacing is close */
ae27e98a	102	u32 curr_rtt; /* the minimum rtt of current round */
df3271f3 SH	103	};
	104
	105	static inline void bictcp_reset(struct bictcp *ca)
	106	{
	107	ca->cnt = 0;
	108	ca->last_max_cwnd = 0;
df3271f3 SH	109	ca->last_cwnd = 0;
	110	ca->last_time = 0;
	111	ca->bic_origin_point = 0;
	112	ca->bic_K = 0;
	113	ca->delay_min = 0;
	114	ca->epoch_start = 0;
df3271f3 SH	115	ca->ack_cnt = 0;
df3271f3 SH	116	ca->tcp_cwnd = 0;
ae27e98a SH	117	ca->found = 0;
	118	}
	119
17a6e9f1	120	static inline u32 bictcp_clock(void)
	121	{
	122	#if HZ < 1000
	123	return ktime_to_ms(ktime_get_real());
	124	#else
	125	return jiffies_to_msecs(jiffies);
	126	#endif
	127	}
	128
ae27e98a SH	129	static inline void bictcp_hystart_reset(struct sock *sk)
	130	{
	131	struct tcp_sock *tp = tcp_sk(sk);
	132	struct bictcp *ca = inet_csk_ca(sk);
	133
17a6e9f1	134	ca->round_start = ca->last_ack = bictcp_clock();
ae27e98a SH	135	ca->end_seq = tp->snd_nxt;
	136	ca->curr_rtt = 0;
	137	ca->sample_cnt = 0;
df3271f3 SH	138	}
	139
	140	static void bictcp_init(struct sock *sk)
	141	{
5a45f008 NC	142	struct bictcp *ca = inet_csk_ca(sk);
	143
	144	bictcp_reset(ca);
	145	ca->loss_cwnd = 0;
ae27e98a SH	146
	147	if (hystart)
	148	bictcp_hystart_reset(sk);
	149
	150	if (!hystart && initial_ssthresh)
df3271f3 SH	151	tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
	152	}
	153
30927520 ED	154	static void bictcp_cwnd_event(struct sock *sk, enum tcp_ca_event event)
	155	{
	156	if (event == CA_EVENT_TX_START) {
30927520	157	struct bictcp *ca = inet_csk_ca(sk);
c2e7204d ED	158	u32 now = tcp_time_stamp;
	159	s32 delta;
	160
	161	delta = now - tcp_sk(sk)->lsndtime;
30927520 ED	162
	163	/* We were application limited (idle) for a while.
	164	* Shift epoch_start to keep cwnd growth to cubic curve.
	165	*/
c2e7204d	166	if (ca->epoch_start && delta > 0) {
30927520	167	ca->epoch_start += delta;
c2e7204d ED	168	if (after(ca->epoch_start, now))
	169	ca->epoch_start = now;
	170	}
30927520 ED	171	return;
	172	}
	173	}
	174
7e58886b SH	175	/* calculate the cubic root of x using a table lookup followed by one
	176	* Newton-Raphson iteration.
	177	* Avg err ~= 0.195%
df3271f3	178	*/
9eb2d627	179	static u32 cubic_root(u64 a)
df3271f3	180	{
7e58886b SH	181	u32 x, b, shift;
	182	/*
	183	* cbrt(x) MSB values for x MSB values in [0..63].
	184	* Precomputed then refined by hand - Willy Tarreau
	185	*
	186	* For x in [0..63],
	187	* v = cbrt(x << 18) - 1
	188	* cbrt(x) = (v[x] + 10) >> 6
9eb2d627	189	*/
7e58886b SH	190	static const u8 v[] = {
	191	/* 0x00 */ 0, 54, 54, 54, 118, 118, 118, 118,
	192	/* 0x08 */ 123, 129, 134, 138, 143, 147, 151, 156,
	193	/* 0x10 */ 157, 161, 164, 168, 170, 173, 176, 179,
	194	/* 0x18 */ 181, 185, 187, 190, 192, 194, 197, 199,
	195	/* 0x20 */ 200, 202, 204, 206, 209, 211, 213, 215,
	196	/* 0x28 */ 217, 219, 221, 222, 224, 225, 227, 229,
	197	/* 0x30 */ 231, 232, 234, 236, 237, 239, 240, 242,
	198	/* 0x38 */ 244, 245, 246, 248, 250, 251, 252, 254,
	199	};
	200
	201	b = fls64(a);
	202	if (b < 7) {
	203	/* a in [0..63] */
	204	return ((u32)v[(u32)a] + 35) >> 6;
	205	}
	206
	207	b = ((b * 84) >> 8) - 1;
	208	shift = (a >> (b * 3));
	209
	210	x = ((u32)(((u32)v[shift] + 10) << b)) >> 6;
	211
	212	/*
	213	* Newton-Raphson iteration
	214	* 2
	215	* x = ( 2 * x + a / x ) / 3
	216	* k+1 k k
	217	*/
6f6d6a1a	218	x = (2 * x + (u32)div64_u64(a, (u64)x * (u64)(x - 1)));
7e58886b	219	x = ((x * 341) >> 10);
9eb2d627	220	return x;
df3271f3 SH	221	}
df3271f3 SH	222
df3271f3 SH	223	/*
	224	* Compute congestion window to use.
	225	*/
9cd981dc	226	static inline void bictcp_update(struct bictcp *ca, u32 cwnd, u32 acked)
df3271f3	227	{
2ed0edf9 ED	228	u32 delta, bic_target, max_cnt;
2ed0edf9 ED	229	u64 offs, t;
df3271f3	230
9cd981dc	231	ca->ack_cnt += acked; /* count the number of ACKed packets */
df3271f3 SH	232
	233	if (ca->last_cwnd == cwnd &&
	234	(s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
	235	return;
	236
d6b1a8a9 NC	237	/* The CUBIC function can update ca->cnt at most once per jiffy.
	238	* On all cwnd reduction events, ca->epoch_start is set to 0,
	239	* which will force a recalculation of ca->cnt.
	240	*/
	241	if (ca->epoch_start && tcp_time_stamp == ca->last_time)
	242	goto tcp_friendliness;
	243
df3271f3 SH	244	ca->last_cwnd = cwnd;
	245	ca->last_time = tcp_time_stamp;
	246
df3271f3	247	if (ca->epoch_start == 0) {
688d1945	248	ca->epoch_start = tcp_time_stamp; /* record beginning */
9cd981dc	249	ca->ack_cnt = acked; /* start counting */
df3271f3 SH	250	ca->tcp_cwnd = cwnd; /* syn with cubic */
	251
	252	if (ca->last_max_cwnd <= cwnd) {
	253	ca->bic_K = 0;
	254	ca->bic_origin_point = cwnd;
	255	} else {
89b3d9aa SH	256	/* Compute new K based on
	257	* (wmax-cwnd) * (srtt>>3 / HZ) / c * 2^(3*bictcp_HZ)
	258	*/
	259	ca->bic_K = cubic_root(cube_factor
	260	* (ca->last_max_cwnd - cwnd));
df3271f3 SH	261	ca->bic_origin_point = ca->last_max_cwnd;
	262	}
	263	}
	264
e905a9ed YH	265	/* cubic function - calc*/
	266	/* calculate c * time^3 / rtt,
	267	* while considering overflow in calculation of time^3
89b3d9aa	268	* (so time^3 is done by using 64 bit)
df3271f3	269	* and without the support of division of 64bit numbers
89b3d9aa	270	* (so all divisions are done by using 32 bit)
e905a9ed YH	271	* also NOTE the unit of those veriables
	272	* time = (t - K) / 2^bictcp_HZ
	273	* c = bic_scale >> 10
df3271f3 SH	274	* rtt = (srtt >> 3) / HZ
	275	* !!! The following code does not have overflow problems,
	276	* if the cwnd < 1 million packets !!!
e905a9ed	277	*/
df3271f3	278
2ed0edf9 ED	279	t = (s32)(tcp_time_stamp - ca->epoch_start);
2ed0edf9 ED	280	t += msecs_to_jiffies(ca->delay_min >> 3);
df3271f3	281	/* change the unit from HZ to bictcp_HZ */
2ed0edf9 ED	282	t <<= BICTCP_HZ;
2ed0edf9 ED	283	do_div(t, HZ);
df3271f3	284
e905a9ed	285	if (t < ca->bic_K) /* t - K */
89b3d9aa	286	offs = ca->bic_K - t;
e905a9ed YH	287	else
e905a9ed YH	288	offs = t - ca->bic_K;
df3271f3	289
89b3d9aa SH	290	/* c/rtt * (t-K)^3 */
89b3d9aa SH	291	delta = (cube_rtt_scale * offs * offs * offs) >> (10+3*BICTCP_HZ);
688d1945	292	if (t < ca->bic_K) /* below origin*/
e905a9ed	293	bic_target = ca->bic_origin_point - delta;
688d1945	294	else /* above origin*/
e905a9ed	295	bic_target = ca->bic_origin_point + delta;
df3271f3	296
e905a9ed YH	297	/* cubic function - calc bictcp_cnt*/
e905a9ed YH	298	if (bic_target > cwnd) {
df3271f3	299	ca->cnt = cwnd / (bic_target - cwnd);
e905a9ed YH	300	} else {
	301	ca->cnt = 100 * cwnd; /* very small increment*/
	302	}
df3271f3	303
b5ccd073 SH	304	/*
	305	* The initial growth of cubic function may be too conservative
	306	* when the available bandwidth is still unknown.
	307	*/
5a45f008	308	if (ca->last_max_cwnd == 0 && ca->cnt > 20)
b5ccd073 SH	309	ca->cnt = 20; /* increase cwnd 5% per RTT */
b5ccd073 SH	310
d6b1a8a9	311	tcp_friendliness:
df3271f3 SH	312	/* TCP Friendly */
df3271f3 SH	313	if (tcp_friendliness) {
89b3d9aa	314	u32 scale = beta_scale;
688d1945	315
89b3d9aa	316	delta = (cwnd * scale) >> 3;
e905a9ed YH	317	while (ca->ack_cnt > delta) { /* update tcp cwnd */
	318	ca->ack_cnt -= delta;
	319	ca->tcp_cwnd++;
df3271f3 SH	320	}
df3271f3 SH	321
688d1945	322	if (ca->tcp_cwnd > cwnd) { /* if bic is slower than tcp */
89b3d9aa SH	323	delta = ca->tcp_cwnd - cwnd;
89b3d9aa SH	324	max_cnt = cwnd / delta;
df3271f3 SH	325	if (ca->cnt > max_cnt)
	326	ca->cnt = max_cnt;
	327	}
e905a9ed	328	}
df3271f3	329
d578e18c NC	330	/* The maximum rate of cwnd increase CUBIC allows is 1 packet per
	331	* 2 packets ACKed, meaning cwnd grows at 1.5x per RTT.
	332	*/
	333	ca->cnt = max(ca->cnt, 2U);
df3271f3 SH	334	}
df3271f3 SH	335
24901551	336	static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
df3271f3 SH	337	{
	338	struct tcp_sock *tp = tcp_sk(sk);
	339	struct bictcp *ca = inet_csk_ca(sk);
	340
24901551	341	if (!tcp_is_cwnd_limited(sk))
df3271f3 SH	342	return;
df3271f3 SH	343
071d5080	344	if (tcp_in_slow_start(tp)) {
ae27e98a SH	345	if (hystart && after(ack, ca->end_seq))
ae27e98a SH	346	bictcp_hystart_reset(sk);
9cd981dc NC	347	acked = tcp_slow_start(tp, acked);
	348	if (!acked)
	349	return;
df3271f3	350	}
9cd981dc NC	351	bictcp_update(ca, tp->snd_cwnd, acked);
9cd981dc NC	352	tcp_cong_avoid_ai(tp, ca->cnt, acked);
df3271f3 SH	353	}
	354
	355	static u32 bictcp_recalc_ssthresh(struct sock *sk)
	356	{
	357	const struct tcp_sock *tp = tcp_sk(sk);
	358	struct bictcp *ca = inet_csk_ca(sk);
	359
	360	ca->epoch_start = 0; /* end of epoch */
	361
	362	/* Wmax and fast convergence */
	363	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
	364	ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
	365	/ (2 * BICTCP_BETA_SCALE);
	366	else
	367	ca->last_max_cwnd = tp->snd_cwnd;
	368
	369	ca->loss_cwnd = tp->snd_cwnd;
	370
	371	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
	372	}
	373
	374	static u32 bictcp_undo_cwnd(struct sock *sk)
	375	{
	376	struct bictcp *ca = inet_csk_ca(sk);
	377
5a45f008	378	return max(tcp_sk(sk)->snd_cwnd, ca->loss_cwnd);
df3271f3 SH	379	}
df3271f3 SH	380
df3271f3 SH	381	static void bictcp_state(struct sock *sk, u8 new_state)
df3271f3 SH	382	{
ae27e98a	383	if (new_state == TCP_CA_Loss) {
df3271f3	384	bictcp_reset(inet_csk_ca(sk));
ae27e98a SH	385	bictcp_hystart_reset(sk);
	386	}
	387	}
	388
	389	static void hystart_update(struct sock *sk, u32 delay)
	390	{
	391	struct tcp_sock *tp = tcp_sk(sk);
	392	struct bictcp *ca = inet_csk_ca(sk);
	393
6e3a8a93 ED	394	if (ca->found & hystart_detect)
	395	return;
	396
	397	if (hystart_detect & HYSTART_ACK_TRAIN) {
17a6e9f1	398	u32 now = bictcp_clock();
ae27e98a SH	399
ae27e98a SH	400	/* first detection parameter - ack-train detection */
17a6e9f1	401	if ((s32)(now - ca->last_ack) <= hystart_ack_delta) {
17a6e9f1	402	ca->last_ack = now;
6e3a8a93	403	if ((s32)(now - ca->round_start) > ca->delay_min >> 4) {
ae27e98a	404	ca->found \|= HYSTART_ACK_TRAIN;
c10d9310 ED	405	NET_INC_STATS(sock_net(sk),
	406	LINUX_MIB_TCPHYSTARTTRAINDETECT);
	407	NET_ADD_STATS(sock_net(sk),
	408	LINUX_MIB_TCPHYSTARTTRAINCWND,
	409	tp->snd_cwnd);
6e3a8a93 ED	410	tp->snd_ssthresh = tp->snd_cwnd;
6e3a8a93 ED	411	}
ae27e98a	412	}
6e3a8a93	413	}
ae27e98a	414
6e3a8a93	415	if (hystart_detect & HYSTART_DELAY) {
ae27e98a SH	416	/* obtain the minimum delay of more than sampling packets */
	417	if (ca->sample_cnt < HYSTART_MIN_SAMPLES) {
	418	if (ca->curr_rtt == 0 \|\| ca->curr_rtt > delay)
	419	ca->curr_rtt = delay;
	420
	421	ca->sample_cnt++;
	422	} else {
	423	if (ca->curr_rtt > ca->delay_min +
42eef7a0	424	HYSTART_DELAY_THRESH(ca->delay_min >> 3)) {
ae27e98a	425	ca->found \|= HYSTART_DELAY;
c10d9310 ED	426	NET_INC_STATS(sock_net(sk),
	427	LINUX_MIB_TCPHYSTARTDELAYDETECT);
	428	NET_ADD_STATS(sock_net(sk),
	429	LINUX_MIB_TCPHYSTARTDELAYCWND,
	430	tp->snd_cwnd);
6e3a8a93 ED	431	tp->snd_ssthresh = tp->snd_cwnd;
6e3a8a93 ED	432	}
ae27e98a	433	}
ae27e98a	434	}
df3271f3 SH	435	}
	436
	437	/* Track delayed acknowledgment ratio using sliding window
	438	* ratio = (15*ratio + sample) / 16
	439	*/
756ee172	440	static void bictcp_acked(struct sock sk, const struct ack_sample sample)
df3271f3	441	{
ae27e98a	442	const struct tcp_sock *tp = tcp_sk(sk);
e7d0c885 SH	443	struct bictcp *ca = inet_csk_ca(sk);
e7d0c885 SH	444	u32 delay;
df3271f3	445
e7d0c885	446	/* Some calls are for duplicates without timetamps */
756ee172	447	if (sample->rtt_us < 0)
e7d0c885 SH	448	return;
	449
	450	/* Discard delay samples right after fast recovery */
cd6b423a	451	if (ca->epoch_start && (s32)(tcp_time_stamp - ca->epoch_start) < HZ)
e7d0c885 SH	452	return;
e7d0c885 SH	453
756ee172	454	delay = (sample->rtt_us << 3) / USEC_PER_MSEC;
e7d0c885 SH	455	if (delay == 0)
	456	delay = 1;
	457
	458	/* first time call or link delay decreases */
	459	if (ca->delay_min == 0 \|\| ca->delay_min > delay)
	460	ca->delay_min = delay;
ae27e98a SH	461
ae27e98a SH	462	/* hystart triggers when cwnd is larger than some threshold */
071d5080	463	if (hystart && tcp_in_slow_start(tp) &&
ae27e98a SH	464	tp->snd_cwnd >= hystart_low_window)
ae27e98a SH	465	hystart_update(sk, delay);
e7d0c885	466	}
df3271f3	467
a252bebe	468	static struct tcp_congestion_ops cubictcp __read_mostly = {
df3271f3 SH	469	.init = bictcp_init,
	470	.ssthresh = bictcp_recalc_ssthresh,
	471	.cong_avoid = bictcp_cong_avoid,
	472	.set_state = bictcp_state,
	473	.undo_cwnd = bictcp_undo_cwnd,
30927520	474	.cwnd_event = bictcp_cwnd_event,
df3271f3 SH	475	.pkts_acked = bictcp_acked,
	476	.owner = THIS_MODULE,
	477	.name = "cubic",
	478	};
	479
	480	static int __init cubictcp_register(void)
	481	{
74975d40	482	BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
89b3d9aa SH	483
	484	/* Precompute a bunch of the scaling factors that are used per-packet
	485	* based on SRTT of 100ms
	486	*/
	487
688d1945	488	beta_scale = 8*(BICTCP_BETA_SCALE+beta) / 3
688d1945	489	/ (BICTCP_BETA_SCALE - beta);
89b3d9aa	490
22119240	491	cube_rtt_scale = (bic_scale * 10); /* 1024c/rtt /
89b3d9aa SH	492
	493	/* calculate the "K" for (wmax-cwnd) = c/rtt * K^3
	494	* so K = cubic_root( (wmax-cwnd)*rtt/c )
	495	* the unit of K is bictcp_HZ=2^10, not HZ
	496	*
	497	* c = bic_scale >> 10
	498	* rtt = 100ms
	499	*
	500	* the following code has been designed and tested for
	501	* cwnd < 1 million packets
	502	* RTT < 100 seconds
	503	* HZ < 1,000,00 (corresponding to 10 nano-second)
	504	*/
	505
	506	/* 1/c * 2^2bictcp_HZ srtt */
	507	cube_factor = 1ull << (10+3BICTCP_HZ); / 2^40 */
	508
	509	/* divide by bic_scale and by constant Srtt (100ms) */
	510	do_div(cube_factor, bic_scale * 10);
	511
df3271f3 SH	512	return tcp_register_congestion_control(&cubictcp);
	513	}
	514
	515	static void __exit cubictcp_unregister(void)
	516	{
	517	tcp_unregister_congestion_control(&cubictcp);
	518	}
	519
	520	module_init(cubictcp_register);
	521	module_exit(cubictcp_unregister);
	522
	523	MODULE_AUTHOR("Sangtae Ha, Stephen Hemminger");
	524	MODULE_LICENSE("GPL");
	525	MODULE_DESCRIPTION("CUBIC TCP");
ae27e98a	526	MODULE_VERSION("2.3");