[GitHub/mt8127/android_kernel_alcatel_ttab.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 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_64(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)
{
	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
7e58886b SH	94	/* calculate the cubic root of x using a table lookup followed by one
	95	* Newton-Raphson iteration.
	96	* Avg err ~= 0.195%
df3271f3	97	*/
9eb2d627	98	static u32 cubic_root(u64 a)
df3271f3	99	{
7e58886b SH	100	u32 x, b, shift;
	101	/*
	102	* cbrt(x) MSB values for x MSB values in [0..63].
	103	* Precomputed then refined by hand - Willy Tarreau
	104	*
	105	* For x in [0..63],
	106	* v = cbrt(x << 18) - 1
	107	* cbrt(x) = (v[x] + 10) >> 6
9eb2d627	108	*/
7e58886b SH	109	static const u8 v[] = {
	110	/* 0x00 */ 0, 54, 54, 54, 118, 118, 118, 118,
	111	/* 0x08 */ 123, 129, 134, 138, 143, 147, 151, 156,
	112	/* 0x10 */ 157, 161, 164, 168, 170, 173, 176, 179,
	113	/* 0x18 */ 181, 185, 187, 190, 192, 194, 197, 199,
	114	/* 0x20 */ 200, 202, 204, 206, 209, 211, 213, 215,
	115	/* 0x28 */ 217, 219, 221, 222, 224, 225, 227, 229,
	116	/* 0x30 */ 231, 232, 234, 236, 237, 239, 240, 242,
	117	/* 0x38 */ 244, 245, 246, 248, 250, 251, 252, 254,
	118	};
	119
	120	b = fls64(a);
	121	if (b < 7) {
	122	/* a in [0..63] */
	123	return ((u32)v[(u32)a] + 35) >> 6;
	124	}
	125
	126	b = ((b * 84) >> 8) - 1;
	127	shift = (a >> (b * 3));
	128
	129	x = ((u32)(((u32)v[shift] + 10) << b)) >> 6;
	130
	131	/*
	132	* Newton-Raphson iteration
	133	* 2
	134	* x = ( 2 * x + a / x ) / 3
	135	* k+1 k k
	136	*/
	137	x = (2 * x + (u32)div64_64(a, (u64)x * (u64)(x - 1)));
	138	x = ((x * 341) >> 10);
9eb2d627	139	return x;
df3271f3 SH	140	}
df3271f3 SH	141
df3271f3 SH	142	/*
	143	* Compute congestion window to use.
	144	*/
	145	static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
	146	{
89b3d9aa SH	147	u64 offs;
89b3d9aa SH	148	u32 delta, t, bic_target, min_cnt, max_cnt;
df3271f3 SH	149
	150	ca->ack_cnt++; /* count the number of ACKs */
	151
	152	if (ca->last_cwnd == cwnd &&
	153	(s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
	154	return;
	155
	156	ca->last_cwnd = cwnd;
	157	ca->last_time = tcp_time_stamp;
	158
df3271f3 SH	159	if (ca->epoch_start == 0) {
	160	ca->epoch_start = tcp_time_stamp; /* record the beginning of an epoch */
	161	ca->ack_cnt = 1; /* start counting */
	162	ca->tcp_cwnd = cwnd; /* syn with cubic */
	163
	164	if (ca->last_max_cwnd <= cwnd) {
	165	ca->bic_K = 0;
	166	ca->bic_origin_point = cwnd;
	167	} else {
89b3d9aa SH	168	/* Compute new K based on
	169	* (wmax-cwnd) * (srtt>>3 / HZ) / c * 2^(3*bictcp_HZ)
	170	*/
	171	ca->bic_K = cubic_root(cube_factor
	172	* (ca->last_max_cwnd - cwnd));
df3271f3 SH	173	ca->bic_origin_point = ca->last_max_cwnd;
	174	}
	175	}
	176
e905a9ed YH	177	/* cubic function - calc*/
	178	/* calculate c * time^3 / rtt,
	179	* while considering overflow in calculation of time^3
89b3d9aa	180	* (so time^3 is done by using 64 bit)
df3271f3	181	* and without the support of division of 64bit numbers
89b3d9aa	182	* (so all divisions are done by using 32 bit)
e905a9ed YH	183	* also NOTE the unit of those veriables
	184	* time = (t - K) / 2^bictcp_HZ
	185	* c = bic_scale >> 10
df3271f3 SH	186	* rtt = (srtt >> 3) / HZ
	187	* !!! The following code does not have overflow problems,
	188	* if the cwnd < 1 million packets !!!
e905a9ed	189	*/
df3271f3 SH	190
df3271f3 SH	191	/* change the unit from HZ to bictcp_HZ */
e905a9ed	192	t = ((tcp_time_stamp + (ca->delay_min>>3) - ca->epoch_start)
df3271f3 SH	193	<< BICTCP_HZ) / HZ;
df3271f3 SH	194
e905a9ed	195	if (t < ca->bic_K) /* t - K */
89b3d9aa	196	offs = ca->bic_K - t;
e905a9ed YH	197	else
e905a9ed YH	198	offs = t - ca->bic_K;
df3271f3	199
89b3d9aa SH	200	/* c/rtt * (t-K)^3 */
89b3d9aa SH	201	delta = (cube_rtt_scale * offs * offs * offs) >> (10+3*BICTCP_HZ);
e905a9ed YH	202	if (t < ca->bic_K) /* below origin*/
	203	bic_target = ca->bic_origin_point - delta;
	204	else /* above origin*/
	205	bic_target = ca->bic_origin_point + delta;
df3271f3	206
e905a9ed YH	207	/* cubic function - calc bictcp_cnt*/
e905a9ed YH	208	if (bic_target > cwnd) {
df3271f3	209	ca->cnt = cwnd / (bic_target - cwnd);
e905a9ed YH	210	} else {
	211	ca->cnt = 100 * cwnd; /* very small increment*/
	212	}
df3271f3 SH	213
	214	if (ca->delay_min > 0) {
	215	/* max increment = Smax * rtt / 0.1 */
	216	min_cnt = (cwnd * HZ * 8)/(10 * max_increment * ca->delay_min);
	217	if (ca->cnt < min_cnt)
	218	ca->cnt = min_cnt;
	219	}
	220
e905a9ed	221	/* slow start and low utilization */
df3271f3 SH	222	if (ca->loss_cwnd == 0) /* could be aggressive in slow start */
	223	ca->cnt = 50;
	224
	225	/* TCP Friendly */
	226	if (tcp_friendliness) {
89b3d9aa SH	227	u32 scale = beta_scale;
89b3d9aa SH	228	delta = (cwnd * scale) >> 3;
e905a9ed YH	229	while (ca->ack_cnt > delta) { /* update tcp cwnd */
	230	ca->ack_cnt -= delta;
	231	ca->tcp_cwnd++;
df3271f3 SH	232	}
	233
	234	if (ca->tcp_cwnd > cwnd){ /* if bic is slower than tcp */
89b3d9aa SH	235	delta = ca->tcp_cwnd - cwnd;
89b3d9aa SH	236	max_cnt = cwnd / delta;
df3271f3 SH	237	if (ca->cnt > max_cnt)
	238	ca->cnt = max_cnt;
	239	}
e905a9ed	240	}
df3271f3 SH	241
	242	ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
	243	if (ca->cnt == 0) /* cannot be zero */
	244	ca->cnt = 1;
	245	}
	246
	247
	248	/* Keep track of minimum rtt */
	249	static inline void measure_delay(struct sock *sk)
	250	{
	251	const struct tcp_sock *tp = tcp_sk(sk);
	252	struct bictcp *ca = inet_csk_ca(sk);
	253	u32 delay;
	254
	255	/* No time stamp */
	256	if (!(tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) \|\|
	257	/* Discard delay samples right after fast recovery */
	258	(s32)(tcp_time_stamp - ca->epoch_start) < HZ)
	259	return;
	260
22119240	261	delay = (tcp_time_stamp - tp->rx_opt.rcv_tsecr)<<3;
df3271f3 SH	262	if (delay == 0)
	263	delay = 1;
	264
	265	/* first time call or link delay decreases */
	266	if (ca->delay_min == 0 \|\| ca->delay_min > delay)
	267	ca->delay_min = delay;
	268	}
	269
	270	static void bictcp_cong_avoid(struct sock *sk, u32 ack,
	271	u32 seq_rtt, u32 in_flight, int data_acked)
	272	{
	273	struct tcp_sock *tp = tcp_sk(sk);
	274	struct bictcp *ca = inet_csk_ca(sk);
	275
	276	if (data_acked)
	277	measure_delay(sk);
	278
	279	if (!tcp_is_cwnd_limited(sk, in_flight))
	280	return;
	281
	282	if (tp->snd_cwnd <= tp->snd_ssthresh)
	283	tcp_slow_start(tp);
	284	else {
	285	bictcp_update(ca, tp->snd_cwnd);
	286
	287	/* In dangerous area, increase slowly.
	288	* In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd
	289	*/
	290	if (tp->snd_cwnd_cnt >= ca->cnt) {
	291	if (tp->snd_cwnd < tp->snd_cwnd_clamp)
	292	tp->snd_cwnd++;
	293	tp->snd_cwnd_cnt = 0;
	294	} else
	295	tp->snd_cwnd_cnt++;
	296	}
	297
	298	}
	299
	300	static u32 bictcp_recalc_ssthresh(struct sock *sk)
	301	{
	302	const struct tcp_sock *tp = tcp_sk(sk);
	303	struct bictcp *ca = inet_csk_ca(sk);
	304
	305	ca->epoch_start = 0; /* end of epoch */
	306
	307	/* Wmax and fast convergence */
	308	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
	309	ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
	310	/ (2 * BICTCP_BETA_SCALE);
	311	else
	312	ca->last_max_cwnd = tp->snd_cwnd;
	313
	314	ca->loss_cwnd = tp->snd_cwnd;
	315
	316	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
	317	}
	318
	319	static u32 bictcp_undo_cwnd(struct sock *sk)
	320	{
	321	struct bictcp *ca = inet_csk_ca(sk);
	322
	323	return max(tcp_sk(sk)->snd_cwnd, ca->last_max_cwnd);
	324	}
	325
df3271f3 SH	326	static void bictcp_state(struct sock *sk, u8 new_state)
	327	{
	328	if (new_state == TCP_CA_Loss)
	329	bictcp_reset(inet_csk_ca(sk));
	330	}
	331
	332	/* Track delayed acknowledgment ratio using sliding window
	333	* ratio = (15*ratio + sample) / 16
	334	*/
	335	static void bictcp_acked(struct sock *sk, u32 cnt)
	336	{
	337	const struct inet_connection_sock *icsk = inet_csk(sk);
	338
	339	if (cnt > 0 && icsk->icsk_ca_state == TCP_CA_Open) {
	340	struct bictcp *ca = inet_csk_ca(sk);
	341	cnt -= ca->delayed_ack >> ACK_RATIO_SHIFT;
	342	ca->delayed_ack += cnt;
	343	}
	344	}
	345
	346
	347	static struct tcp_congestion_ops cubictcp = {
	348	.init = bictcp_init,
	349	.ssthresh = bictcp_recalc_ssthresh,
	350	.cong_avoid = bictcp_cong_avoid,
	351	.set_state = bictcp_state,
	352	.undo_cwnd = bictcp_undo_cwnd,
df3271f3 SH	353	.pkts_acked = bictcp_acked,
	354	.owner = THIS_MODULE,
	355	.name = "cubic",
	356	};
	357
	358	static int __init cubictcp_register(void)
	359	{
74975d40	360	BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
89b3d9aa SH	361
	362	/* Precompute a bunch of the scaling factors that are used per-packet
	363	* based on SRTT of 100ms
	364	*/
	365
	366	beta_scale = 8*(BICTCP_BETA_SCALE+beta)/ 3 / (BICTCP_BETA_SCALE - beta);
	367
22119240	368	cube_rtt_scale = (bic_scale * 10); /* 1024c/rtt /
89b3d9aa SH	369
	370	/* calculate the "K" for (wmax-cwnd) = c/rtt * K^3
	371	* so K = cubic_root( (wmax-cwnd)*rtt/c )
	372	* the unit of K is bictcp_HZ=2^10, not HZ
	373	*
	374	* c = bic_scale >> 10
	375	* rtt = 100ms
	376	*
	377	* the following code has been designed and tested for
	378	* cwnd < 1 million packets
	379	* RTT < 100 seconds
	380	* HZ < 1,000,00 (corresponding to 10 nano-second)
	381	*/
	382
	383	/* 1/c * 2^2bictcp_HZ srtt */
	384	cube_factor = 1ull << (10+3BICTCP_HZ); / 2^40 */
	385
	386	/* divide by bic_scale and by constant Srtt (100ms) */
	387	do_div(cube_factor, bic_scale * 10);
	388
df3271f3 SH	389	return tcp_register_congestion_control(&cubictcp);
	390	}
	391
	392	static void __exit cubictcp_unregister(void)
	393	{
	394	tcp_unregister_congestion_control(&cubictcp);
	395	}
	396
	397	module_init(cubictcp_register);
	398	module_exit(cubictcp_unregister);
	399
	400	MODULE_AUTHOR("Sangtae Ha, Stephen Hemminger");
	401	MODULE_LICENSE("GPL");
	402	MODULE_DESCRIPTION("CUBIC TCP");
	403	MODULE_VERSION("2.0");