[deliverable/linux.git] / net / ipv4 / tcp_cubic.c

/*
 * TCP CUBIC: Binary Increase Congestion control for TCP v2.0
 *
 * This is from the implementation of CUBIC TCP in
 * Injong Rhee, Lisong Xu.
 *  "CUBIC: A New TCP-Friendly High-Speed TCP Variant
 *  in PFLDnet 2005
 * Available from:
 *  http://www.csc.ncsu.edu/faculty/rhee/export/bitcp/cubic-paper.pdf
 *
 * 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 <net/tcp.h>
#include <asm/div64.h>

#define BICTCP_BETA_SCALE    1024	/* Scale factor beta calculation
					 * max_cwnd = snd_cwnd * beta
					 */
#define BICTCP_B		4	 /*
					  * In binary search,
					  * go to point (max+min)/N
					  */
#define	BICTCP_HZ		10	/* BIC HZ 2^10 = 1024 */

static int fast_convergence __read_mostly = 1;
static int max_increment __read_mostly = 16;
static int beta __read_mostly = 819;	/* = 819/1024 (BICTCP_BETA_SCALE) */
static int initial_ssthresh __read_mostly = 100;
static int bic_scale __read_mostly = 41;
static int tcp_friendliness __read_mostly = 1;

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(max_increment, int, 0644);
MODULE_PARM_DESC(max_increment, "Limit on increment allowed during binary search");
module_param(beta, int, 0444);
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");

/* 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 */
	u32	epoch_start;	/* beginning of an epoch */
	u32	ack_cnt;	/* number of acks */
	u32	tcp_cwnd;	/* estimated tcp cwnd */
#define ACK_RATIO_SHIFT	4
	u32	delayed_ack;	/* estimate the ratio of Packets/ACKs << 4 */
};

static inline void bictcp_reset(struct bictcp *ca)
{
	ca->cnt = 0;
	ca->last_max_cwnd = 0;
	ca->loss_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->delayed_ack = 2 << ACK_RATIO_SHIFT;
	ca->ack_cnt = 0;
	ca->tcp_cwnd = 0;
}

static void bictcp_init(struct sock *sk)
{
	bictcp_reset(inet_csk_ca(sk));
	if (initial_ssthresh)
		tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
}

/*
 * calculate the cubic root of x using Newton-Raphson
 */
static u32 cubic_root(u64 a)
{
	u32 x;

	/* Initial estimate is based on:
	 * cbrt(x) = exp(log(x) / 3)
	 */
	x = 1u << (fls64(a)/3);

	/* converges to 32 bits in 3 iterations */
	x = (2 * x + (u32)div64_64(a, (u64)x*(u64)x)) / 3;
	x = (2 * x + (u32)div64_64(a, (u64)x*(u64)x)) / 3;
	x = (2 * x + (u32)div64_64(a, (u64)x*(u64)x)) / 3;

	return x;
}

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

	ca->ack_cnt++;	/* count the number of ACKs */

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

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

	if (ca->epoch_start == 0) {
		ca->epoch_start = tcp_time_stamp;	/* record the beginning of an epoch */
		ca->ack_cnt = 1;			/* 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 !!!
	 */

	/* change the unit from HZ to bictcp_HZ */
	t = ((tcp_time_stamp + (ca->delay_min>>3) - ca->epoch_start)
	     << BICTCP_HZ) / 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*/
	}

	if (ca->delay_min > 0) {
		/* max increment = Smax * rtt / 0.1  */
		min_cnt = (cwnd * HZ * 8)/(10 * max_increment * ca->delay_min);
		if (ca->cnt < min_cnt)
			ca->cnt = min_cnt;
	}

	/* slow start and low utilization  */
	if (ca->loss_cwnd == 0)		/* could be aggressive in slow start */
		ca->cnt = 50;

	/* 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;
		}
	}

	ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
	if (ca->cnt == 0)			/* cannot be zero */
		ca->cnt = 1;
}


/* Keep track of minimum rtt */
static inline void measure_delay(struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);
	u32 delay;

	/* No time stamp */
	if (!(tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) ||
	     /* Discard delay samples right after fast recovery */
	    (s32)(tcp_time_stamp - ca->epoch_start) < HZ)
		return;

	delay = (tcp_time_stamp - tp->rx_opt.rcv_tsecr)<<3;
	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;
}

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

	if (data_acked)
		measure_delay(sk);

	if (!tcp_is_cwnd_limited(sk, in_flight))
		return;

	if (tp->snd_cwnd <= tp->snd_ssthresh)
		tcp_slow_start(tp);
	else {
		bictcp_update(ca, tp->snd_cwnd);

		/* In dangerous area, increase slowly.
		 * In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd
		 */
		if (tp->snd_cwnd_cnt >= ca->cnt) {
			if (tp->snd_cwnd < tp->snd_cwnd_clamp)
				tp->snd_cwnd++;
			tp->snd_cwnd_cnt = 0;
		} else
			tp->snd_cwnd_cnt++;
	}

}

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->last_max_cwnd);
}

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

/* Track delayed acknowledgment ratio using sliding window
 * ratio = (15*ratio + sample) / 16
 */
static void bictcp_acked(struct sock *sk, u32 cnt)
{
	const struct inet_connection_sock *icsk = inet_csk(sk);

	if (cnt > 0 && icsk->icsk_ca_state == TCP_CA_Open) {
		struct bictcp *ca = inet_csk_ca(sk);
		cnt -= ca->delayed_ack >> ACK_RATIO_SHIFT;
		ca->delayed_ack += cnt;
	}
}


static struct tcp_congestion_ops cubictcp = {
	.init		= bictcp_init,
	.ssthresh	= bictcp_recalc_ssthresh,
	.cong_avoid	= bictcp_cong_avoid,
	.set_state	= bictcp_state,
	.undo_cwnd	= bictcp_undo_cwnd,
	.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("2.0");
Commit	Line	Data
df3271f3 SH	1	/*
	2	* TCP CUBIC: Binary Increase Congestion control for TCP v2.0
	3	*
	4	* This is from the implementation of CUBIC TCP in
	5	* Injong Rhee, Lisong Xu.
	6	* "CUBIC: A New TCP-Friendly High-Speed TCP Variant
	7	* in PFLDnet 2005
	8	* Available from:
	9	* http://www.csc.ncsu.edu/faculty/rhee/export/bitcp/cubic-paper.pdf
	10	*
	11	* Unless CUBIC is enabled and congestion window is large
	12	* this behaves the same as the original Reno.
	13	*/
	14
df3271f3 SH	15	#include <linux/mm.h>
	16	#include <linux/module.h>
	17	#include <net/tcp.h>
89b3d9aa	18	#include <asm/div64.h>
df3271f3 SH	19
	20	#define BICTCP_BETA_SCALE 1024 /* Scale factor beta calculation
	21	* max_cwnd = snd_cwnd * beta
	22	*/
	23	#define BICTCP_B 4 /*
	24	* In binary search,
	25	* go to point (max+min)/N
	26	*/
	27	#define BICTCP_HZ 10 /* BIC HZ 2^10 = 1024 */
	28
59758f44 SH	29	static int fast_convergence __read_mostly = 1;
	30	static int max_increment __read_mostly = 16;
	31	static int beta __read_mostly = 819; /* = 819/1024 (BICTCP_BETA_SCALE) */
	32	static int initial_ssthresh __read_mostly = 100;
	33	static int bic_scale __read_mostly = 41;
	34	static int tcp_friendliness __read_mostly = 1;
df3271f3	35
59758f44 SH	36	static u32 cube_rtt_scale __read_mostly;
	37	static u32 beta_scale __read_mostly;
	38	static u64 cube_factor __read_mostly;
89b3d9aa SH	39
89b3d9aa SH	40	/* Note parameters that are used for precomputing scale factors are read-only */
df3271f3 SH	41	module_param(fast_convergence, int, 0644);
	42	MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
	43	module_param(max_increment, int, 0644);
	44	MODULE_PARM_DESC(max_increment, "Limit on increment allowed during binary search");
89b3d9aa	45	module_param(beta, int, 0444);
df3271f3 SH	46	MODULE_PARM_DESC(beta, "beta for multiplicative increase");
	47	module_param(initial_ssthresh, int, 0644);
	48	MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
89b3d9aa	49	module_param(bic_scale, int, 0444);
df3271f3 SH	50	MODULE_PARM_DESC(bic_scale, "scale (scaled by 1024) value for bic function (bic_scale/1024)");
	51	module_param(tcp_friendliness, int, 0644);
	52	MODULE_PARM_DESC(tcp_friendliness, "turn on/off tcp friendliness");
	53
df3271f3 SH	54	/* BIC TCP Parameters */
	55	struct bictcp {
	56	u32 cnt; /* increase cwnd by 1 after ACKs */
	57	u32 last_max_cwnd; /* last maximum snd_cwnd */
	58	u32 loss_cwnd; /* congestion window at last loss */
	59	u32 last_cwnd; /* the last snd_cwnd */
	60	u32 last_time; /* time when updated last_cwnd */
	61	u32 bic_origin_point;/* origin point of bic function */
	62	u32 bic_K; /* time to origin point from the beginning of the current epoch */
	63	u32 delay_min; /* min delay */
	64	u32 epoch_start; /* beginning of an epoch */
	65	u32 ack_cnt; /* number of acks */
	66	u32 tcp_cwnd; /* estimated tcp cwnd */
	67	#define ACK_RATIO_SHIFT 4
	68	u32 delayed_ack; /* estimate the ratio of Packets/ACKs << 4 */
	69	};
	70
	71	static inline void bictcp_reset(struct bictcp *ca)
	72	{
	73	ca->cnt = 0;
	74	ca->last_max_cwnd = 0;
	75	ca->loss_cwnd = 0;
	76	ca->last_cwnd = 0;
	77	ca->last_time = 0;
	78	ca->bic_origin_point = 0;
	79	ca->bic_K = 0;
	80	ca->delay_min = 0;
	81	ca->epoch_start = 0;
	82	ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
	83	ca->ack_cnt = 0;
	84	ca->tcp_cwnd = 0;
	85	}
	86
	87	static void bictcp_init(struct sock *sk)
	88	{
	89	bictcp_reset(inet_csk_ca(sk));
	90	if (initial_ssthresh)
	91	tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
	92	}
	93
df3271f3	94	/*
9eb2d627	95	* calculate the cubic root of x using Newton-Raphson
df3271f3	96	*/
9eb2d627	97	static u32 cubic_root(u64 a)
df3271f3	98	{
c5f5877c	99	u32 x;
9eb2d627 SH	100
	101	/* Initial estimate is based on:
	102	* cbrt(x) = exp(log(x) / 3)
	103	*/
	104	x = 1u << (fls64(a)/3);
	105
c5f5877c SH	106	/* converges to 32 bits in 3 iterations */
	107	x = (2 * x + (u32)div64_64(a, (u64)x*(u64)x)) / 3;
	108	x = (2 * x + (u32)div64_64(a, (u64)x*(u64)x)) / 3;
	109	x = (2 * x + (u32)div64_64(a, (u64)x*(u64)x)) / 3;
9eb2d627 SH	110
9eb2d627 SH	111	return x;
df3271f3 SH	112	}
df3271f3 SH	113
df3271f3 SH	114	/*
	115	* Compute congestion window to use.
	116	*/
	117	static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
	118	{
89b3d9aa SH	119	u64 offs;
89b3d9aa SH	120	u32 delta, t, bic_target, min_cnt, max_cnt;
df3271f3 SH	121
	122	ca->ack_cnt++; /* count the number of ACKs */
	123
	124	if (ca->last_cwnd == cwnd &&
	125	(s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
	126	return;
	127
	128	ca->last_cwnd = cwnd;
	129	ca->last_time = tcp_time_stamp;
	130
df3271f3 SH	131	if (ca->epoch_start == 0) {
	132	ca->epoch_start = tcp_time_stamp; /* record the beginning of an epoch */
	133	ca->ack_cnt = 1; /* start counting */
	134	ca->tcp_cwnd = cwnd; /* syn with cubic */
	135
	136	if (ca->last_max_cwnd <= cwnd) {
	137	ca->bic_K = 0;
	138	ca->bic_origin_point = cwnd;
	139	} else {
89b3d9aa SH	140	/* Compute new K based on
	141	* (wmax-cwnd) * (srtt>>3 / HZ) / c * 2^(3*bictcp_HZ)
	142	*/
	143	ca->bic_K = cubic_root(cube_factor
	144	* (ca->last_max_cwnd - cwnd));
df3271f3 SH	145	ca->bic_origin_point = ca->last_max_cwnd;
	146	}
	147	}
	148
e905a9ed YH	149	/* cubic function - calc*/
	150	/* calculate c * time^3 / rtt,
	151	* while considering overflow in calculation of time^3
89b3d9aa	152	* (so time^3 is done by using 64 bit)
df3271f3	153	* and without the support of division of 64bit numbers
89b3d9aa	154	* (so all divisions are done by using 32 bit)
e905a9ed YH	155	* also NOTE the unit of those veriables
	156	* time = (t - K) / 2^bictcp_HZ
	157	* c = bic_scale >> 10
df3271f3 SH	158	* rtt = (srtt >> 3) / HZ
	159	* !!! The following code does not have overflow problems,
	160	* if the cwnd < 1 million packets !!!
e905a9ed	161	*/
df3271f3 SH	162
df3271f3 SH	163	/* change the unit from HZ to bictcp_HZ */
e905a9ed	164	t = ((tcp_time_stamp + (ca->delay_min>>3) - ca->epoch_start)
df3271f3 SH	165	<< BICTCP_HZ) / HZ;
df3271f3 SH	166
e905a9ed	167	if (t < ca->bic_K) /* t - K */
89b3d9aa	168	offs = ca->bic_K - t;
e905a9ed YH	169	else
e905a9ed YH	170	offs = t - ca->bic_K;
df3271f3	171
89b3d9aa SH	172	/* c/rtt * (t-K)^3 */
89b3d9aa SH	173	delta = (cube_rtt_scale * offs * offs * offs) >> (10+3*BICTCP_HZ);
e905a9ed YH	174	if (t < ca->bic_K) /* below origin*/
	175	bic_target = ca->bic_origin_point - delta;
	176	else /* above origin*/
	177	bic_target = ca->bic_origin_point + delta;
df3271f3	178
e905a9ed YH	179	/* cubic function - calc bictcp_cnt*/
e905a9ed YH	180	if (bic_target > cwnd) {
df3271f3	181	ca->cnt = cwnd / (bic_target - cwnd);
e905a9ed YH	182	} else {
	183	ca->cnt = 100 * cwnd; /* very small increment*/
	184	}
df3271f3 SH	185
	186	if (ca->delay_min > 0) {
	187	/* max increment = Smax * rtt / 0.1 */
	188	min_cnt = (cwnd * HZ * 8)/(10 * max_increment * ca->delay_min);
	189	if (ca->cnt < min_cnt)
	190	ca->cnt = min_cnt;
	191	}
	192
e905a9ed	193	/* slow start and low utilization */
df3271f3 SH	194	if (ca->loss_cwnd == 0) /* could be aggressive in slow start */
	195	ca->cnt = 50;
	196
	197	/* TCP Friendly */
	198	if (tcp_friendliness) {
89b3d9aa SH	199	u32 scale = beta_scale;
89b3d9aa SH	200	delta = (cwnd * scale) >> 3;
e905a9ed YH	201	while (ca->ack_cnt > delta) { /* update tcp cwnd */
	202	ca->ack_cnt -= delta;
	203	ca->tcp_cwnd++;
df3271f3 SH	204	}
	205
	206	if (ca->tcp_cwnd > cwnd){ /* if bic is slower than tcp */
89b3d9aa SH	207	delta = ca->tcp_cwnd - cwnd;
89b3d9aa SH	208	max_cnt = cwnd / delta;
df3271f3 SH	209	if (ca->cnt > max_cnt)
	210	ca->cnt = max_cnt;
	211	}
e905a9ed	212	}
df3271f3 SH	213
	214	ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
	215	if (ca->cnt == 0) /* cannot be zero */
	216	ca->cnt = 1;
	217	}
	218
	219
	220	/* Keep track of minimum rtt */
	221	static inline void measure_delay(struct sock *sk)
	222	{
	223	const struct tcp_sock *tp = tcp_sk(sk);
	224	struct bictcp *ca = inet_csk_ca(sk);
	225	u32 delay;
	226
	227	/* No time stamp */
	228	if (!(tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) \|\|
	229	/* Discard delay samples right after fast recovery */
	230	(s32)(tcp_time_stamp - ca->epoch_start) < HZ)
	231	return;
	232
22119240	233	delay = (tcp_time_stamp - tp->rx_opt.rcv_tsecr)<<3;
df3271f3 SH	234	if (delay == 0)
	235	delay = 1;
	236
	237	/* first time call or link delay decreases */
	238	if (ca->delay_min == 0 \|\| ca->delay_min > delay)
	239	ca->delay_min = delay;
	240	}
	241
	242	static void bictcp_cong_avoid(struct sock *sk, u32 ack,
	243	u32 seq_rtt, u32 in_flight, int data_acked)
	244	{
	245	struct tcp_sock *tp = tcp_sk(sk);
	246	struct bictcp *ca = inet_csk_ca(sk);
	247
	248	if (data_acked)
	249	measure_delay(sk);
	250
	251	if (!tcp_is_cwnd_limited(sk, in_flight))
	252	return;
	253
	254	if (tp->snd_cwnd <= tp->snd_ssthresh)
	255	tcp_slow_start(tp);
	256	else {
	257	bictcp_update(ca, tp->snd_cwnd);
	258
	259	/* In dangerous area, increase slowly.
	260	* In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd
	261	*/
	262	if (tp->snd_cwnd_cnt >= ca->cnt) {
	263	if (tp->snd_cwnd < tp->snd_cwnd_clamp)
	264	tp->snd_cwnd++;
	265	tp->snd_cwnd_cnt = 0;
	266	} else
	267	tp->snd_cwnd_cnt++;
	268	}
	269
	270	}
	271
	272	static u32 bictcp_recalc_ssthresh(struct sock *sk)
	273	{
	274	const struct tcp_sock *tp = tcp_sk(sk);
	275	struct bictcp *ca = inet_csk_ca(sk);
	276
	277	ca->epoch_start = 0; /* end of epoch */
	278
	279	/* Wmax and fast convergence */
	280	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
	281	ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
	282	/ (2 * BICTCP_BETA_SCALE);
	283	else
	284	ca->last_max_cwnd = tp->snd_cwnd;
	285
	286	ca->loss_cwnd = tp->snd_cwnd;
	287
	288	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
	289	}
	290
	291	static u32 bictcp_undo_cwnd(struct sock *sk)
	292	{
	293	struct bictcp *ca = inet_csk_ca(sk);
	294
	295	return max(tcp_sk(sk)->snd_cwnd, ca->last_max_cwnd);
	296	}
	297
df3271f3 SH	298	static void bictcp_state(struct sock *sk, u8 new_state)
	299	{
	300	if (new_state == TCP_CA_Loss)
	301	bictcp_reset(inet_csk_ca(sk));
	302	}
	303
	304	/* Track delayed acknowledgment ratio using sliding window
	305	* ratio = (15*ratio + sample) / 16
	306	*/
	307	static void bictcp_acked(struct sock *sk, u32 cnt)
	308	{
	309	const struct inet_connection_sock *icsk = inet_csk(sk);
	310
	311	if (cnt > 0 && icsk->icsk_ca_state == TCP_CA_Open) {
	312	struct bictcp *ca = inet_csk_ca(sk);
	313	cnt -= ca->delayed_ack >> ACK_RATIO_SHIFT;
	314	ca->delayed_ack += cnt;
	315	}
	316	}
	317
	318
	319	static struct tcp_congestion_ops cubictcp = {
	320	.init = bictcp_init,
	321	.ssthresh = bictcp_recalc_ssthresh,
	322	.cong_avoid = bictcp_cong_avoid,
	323	.set_state = bictcp_state,
	324	.undo_cwnd = bictcp_undo_cwnd,
df3271f3 SH	325	.pkts_acked = bictcp_acked,
	326	.owner = THIS_MODULE,
	327	.name = "cubic",
	328	};
	329
	330	static int __init cubictcp_register(void)
	331	{
74975d40	332	BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
89b3d9aa SH	333
	334	/* Precompute a bunch of the scaling factors that are used per-packet
	335	* based on SRTT of 100ms
	336	*/
	337
	338	beta_scale = 8*(BICTCP_BETA_SCALE+beta)/ 3 / (BICTCP_BETA_SCALE - beta);
	339
22119240	340	cube_rtt_scale = (bic_scale * 10); /* 1024c/rtt /
89b3d9aa SH	341
	342	/* calculate the "K" for (wmax-cwnd) = c/rtt * K^3
	343	* so K = cubic_root( (wmax-cwnd)*rtt/c )
	344	* the unit of K is bictcp_HZ=2^10, not HZ
	345	*
	346	* c = bic_scale >> 10
	347	* rtt = 100ms
	348	*
	349	* the following code has been designed and tested for
	350	* cwnd < 1 million packets
	351	* RTT < 100 seconds
	352	* HZ < 1,000,00 (corresponding to 10 nano-second)
	353	*/
	354
	355	/* 1/c * 2^2bictcp_HZ srtt */
	356	cube_factor = 1ull << (10+3BICTCP_HZ); / 2^40 */
	357
	358	/* divide by bic_scale and by constant Srtt (100ms) */
	359	do_div(cube_factor, bic_scale * 10);
	360
df3271f3 SH	361	return tcp_register_congestion_control(&cubictcp);
	362	}
	363
	364	static void __exit cubictcp_unregister(void)
	365	{
	366	tcp_unregister_congestion_control(&cubictcp);
	367	}
	368
	369	module_init(cubictcp_register);
	370	module_exit(cubictcp_unregister);
	371
	372	MODULE_AUTHOR("Sangtae Ha, Stephen Hemminger");
	373	MODULE_LICENSE("GPL");
	374	MODULE_DESCRIPTION("CUBIC TCP");
	375	MODULE_VERSION("2.0");