xref: /openbmc/linux/net/ipv4/tcp_cdg.c (revision c059ee9d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * CAIA Delay-Gradient (CDG) congestion control
4  *
5  * This implementation is based on the paper:
6  *   D.A. Hayes and G. Armitage. "Revisiting TCP congestion control using
7  *   delay gradients." In IFIP Networking, pages 328-341. Springer, 2011.
8  *
9  * Scavenger traffic (Less-than-Best-Effort) should disable coexistence
10  * heuristics using parameters use_shadow=0 and use_ineff=0.
11  *
12  * Parameters window, backoff_beta, and backoff_factor are crucial for
13  * throughput and delay. Future work is needed to determine better defaults,
14  * and to provide guidelines for use in different environments/contexts.
15  *
16  * Except for window, knobs are configured via /sys/module/tcp_cdg/parameters/.
17  * Parameter window is only configurable when loading tcp_cdg as a module.
18  *
19  * Notable differences from paper/FreeBSD:
20  *   o Using Hybrid Slow start and Proportional Rate Reduction.
21  *   o Add toggle for shadow window mechanism. Suggested by David Hayes.
22  *   o Add toggle for non-congestion loss tolerance.
23  *   o Scaling parameter G is changed to a backoff factor;
24  *     conversion is given by: backoff_factor = 1000/(G * window).
25  *   o Limit shadow window to 2 * cwnd, or to cwnd when application limited.
26  *   o More accurate e^-x.
27  */
28 #include <linux/kernel.h>
29 #include <linux/random.h>
30 #include <linux/module.h>
31 #include <linux/sched/clock.h>
32 
33 #include <net/tcp.h>
34 
35 #define HYSTART_ACK_TRAIN	1
36 #define HYSTART_DELAY		2
37 
38 static int window __read_mostly = 8;
39 static unsigned int backoff_beta __read_mostly = 0.7071 * 1024; /* sqrt 0.5 */
40 static unsigned int backoff_factor __read_mostly = 42;
41 static unsigned int hystart_detect __read_mostly = 3;
42 static unsigned int use_ineff __read_mostly = 5;
43 static bool use_shadow __read_mostly = true;
44 static bool use_tolerance __read_mostly;
45 
46 module_param(window, int, 0444);
47 MODULE_PARM_DESC(window, "gradient window size (power of two <= 256)");
48 module_param(backoff_beta, uint, 0644);
49 MODULE_PARM_DESC(backoff_beta, "backoff beta (0-1024)");
50 module_param(backoff_factor, uint, 0644);
51 MODULE_PARM_DESC(backoff_factor, "backoff probability scale factor");
52 module_param(hystart_detect, uint, 0644);
53 MODULE_PARM_DESC(hystart_detect, "use Hybrid Slow start "
54 		 "(0: disabled, 1: ACK train, 2: delay threshold, 3: both)");
55 module_param(use_ineff, uint, 0644);
56 MODULE_PARM_DESC(use_ineff, "use ineffectual backoff detection (threshold)");
57 module_param(use_shadow, bool, 0644);
58 MODULE_PARM_DESC(use_shadow, "use shadow window heuristic");
59 module_param(use_tolerance, bool, 0644);
60 MODULE_PARM_DESC(use_tolerance, "use loss tolerance heuristic");
61 
62 struct cdg_minmax {
63 	union {
64 		struct {
65 			s32 min;
66 			s32 max;
67 		};
68 		u64 v64;
69 	};
70 };
71 
72 enum cdg_state {
73 	CDG_UNKNOWN = 0,
74 	CDG_NONFULL = 1,
75 	CDG_FULL    = 2,
76 	CDG_BACKOFF = 3,
77 };
78 
79 struct cdg {
80 	struct cdg_minmax rtt;
81 	struct cdg_minmax rtt_prev;
82 	struct cdg_minmax *gradients;
83 	struct cdg_minmax gsum;
84 	bool gfilled;
85 	u8  tail;
86 	u8  state;
87 	u8  delack;
88 	u32 rtt_seq;
89 	u32 shadow_wnd;
90 	u16 backoff_cnt;
91 	u16 sample_cnt;
92 	s32 delay_min;
93 	u32 last_ack;
94 	u32 round_start;
95 };
96 
97 /**
98  * nexp_u32 - negative base-e exponential
99  * @ux: x in units of micro
100  *
101  * Returns exp(ux * -1e-6) * U32_MAX.
102  */
103 static u32 __pure nexp_u32(u32 ux)
104 {
105 	static const u16 v[] = {
106 		/* exp(-x)*65536-1 for x = 0, 0.000256, 0.000512, ... */
107 		65535,
108 		65518, 65501, 65468, 65401, 65267, 65001, 64470, 63422,
109 		61378, 57484, 50423, 38795, 22965, 8047,  987,   14,
110 	};
111 	u32 msb = ux >> 8;
112 	u32 res;
113 	int i;
114 
115 	/* Cut off when ux >= 2^24 (actual result is <= 222/U32_MAX). */
116 	if (msb > U16_MAX)
117 		return 0;
118 
119 	/* Scale first eight bits linearly: */
120 	res = U32_MAX - (ux & 0xff) * (U32_MAX / 1000000);
121 
122 	/* Obtain e^(x + y + ...) by computing e^x * e^y * ...: */
123 	for (i = 1; msb; i++, msb >>= 1) {
124 		u32 y = v[i & -(msb & 1)] + U32_C(1);
125 
126 		res = ((u64)res * y) >> 16;
127 	}
128 
129 	return res;
130 }
131 
132 /* Based on the HyStart algorithm (by Ha et al.) that is implemented in
133  * tcp_cubic. Differences/experimental changes:
134  *   o Using Hayes' delayed ACK filter.
135  *   o Using a usec clock for the ACK train.
136  *   o Reset ACK train when application limited.
137  *   o Invoked at any cwnd (i.e. also when cwnd < 16).
138  *   o Invoked only when cwnd < ssthresh (i.e. not when cwnd == ssthresh).
139  */
140 static void tcp_cdg_hystart_update(struct sock *sk)
141 {
142 	struct cdg *ca = inet_csk_ca(sk);
143 	struct tcp_sock *tp = tcp_sk(sk);
144 
145 	ca->delay_min = min_not_zero(ca->delay_min, ca->rtt.min);
146 	if (ca->delay_min == 0)
147 		return;
148 
149 	if (hystart_detect & HYSTART_ACK_TRAIN) {
150 		u32 now_us = tp->tcp_mstamp;
151 
152 		if (ca->last_ack == 0 || !tcp_is_cwnd_limited(sk)) {
153 			ca->last_ack = now_us;
154 			ca->round_start = now_us;
155 		} else if (before(now_us, ca->last_ack + 3000)) {
156 			u32 base_owd = max(ca->delay_min / 2U, 125U);
157 
158 			ca->last_ack = now_us;
159 			if (after(now_us, ca->round_start + base_owd)) {
160 				NET_INC_STATS(sock_net(sk),
161 					      LINUX_MIB_TCPHYSTARTTRAINDETECT);
162 				NET_ADD_STATS(sock_net(sk),
163 					      LINUX_MIB_TCPHYSTARTTRAINCWND,
164 					      tcp_snd_cwnd(tp));
165 				tp->snd_ssthresh = tcp_snd_cwnd(tp);
166 				return;
167 			}
168 		}
169 	}
170 
171 	if (hystart_detect & HYSTART_DELAY) {
172 		if (ca->sample_cnt < 8) {
173 			ca->sample_cnt++;
174 		} else {
175 			s32 thresh = max(ca->delay_min + ca->delay_min / 8U,
176 					 125U);
177 
178 			if (ca->rtt.min > thresh) {
179 				NET_INC_STATS(sock_net(sk),
180 					      LINUX_MIB_TCPHYSTARTDELAYDETECT);
181 				NET_ADD_STATS(sock_net(sk),
182 					      LINUX_MIB_TCPHYSTARTDELAYCWND,
183 					      tcp_snd_cwnd(tp));
184 				tp->snd_ssthresh = tcp_snd_cwnd(tp);
185 			}
186 		}
187 	}
188 }
189 
190 static s32 tcp_cdg_grad(struct cdg *ca)
191 {
192 	s32 gmin = ca->rtt.min - ca->rtt_prev.min;
193 	s32 gmax = ca->rtt.max - ca->rtt_prev.max;
194 	s32 grad;
195 
196 	if (ca->gradients) {
197 		ca->gsum.min += gmin - ca->gradients[ca->tail].min;
198 		ca->gsum.max += gmax - ca->gradients[ca->tail].max;
199 		ca->gradients[ca->tail].min = gmin;
200 		ca->gradients[ca->tail].max = gmax;
201 		ca->tail = (ca->tail + 1) & (window - 1);
202 		gmin = ca->gsum.min;
203 		gmax = ca->gsum.max;
204 	}
205 
206 	/* We keep sums to ignore gradients during cwnd reductions;
207 	 * the paper's smoothed gradients otherwise simplify to:
208 	 * (rtt_latest - rtt_oldest) / window.
209 	 *
210 	 * We also drop division by window here.
211 	 */
212 	grad = gmin > 0 ? gmin : gmax;
213 
214 	/* Extrapolate missing values in gradient window: */
215 	if (!ca->gfilled) {
216 		if (!ca->gradients && window > 1)
217 			grad *= window; /* Memory allocation failed. */
218 		else if (ca->tail == 0)
219 			ca->gfilled = true;
220 		else
221 			grad = (grad * window) / (int)ca->tail;
222 	}
223 
224 	/* Backoff was effectual: */
225 	if (gmin <= -32 || gmax <= -32)
226 		ca->backoff_cnt = 0;
227 
228 	if (use_tolerance) {
229 		/* Reduce small variations to zero: */
230 		gmin = DIV_ROUND_CLOSEST(gmin, 64);
231 		gmax = DIV_ROUND_CLOSEST(gmax, 64);
232 
233 		if (gmin > 0 && gmax <= 0)
234 			ca->state = CDG_FULL;
235 		else if ((gmin > 0 && gmax > 0) || gmax < 0)
236 			ca->state = CDG_NONFULL;
237 	}
238 	return grad;
239 }
240 
241 static bool tcp_cdg_backoff(struct sock *sk, u32 grad)
242 {
243 	struct cdg *ca = inet_csk_ca(sk);
244 	struct tcp_sock *tp = tcp_sk(sk);
245 
246 	if (prandom_u32() <= nexp_u32(grad * backoff_factor))
247 		return false;
248 
249 	if (use_ineff) {
250 		ca->backoff_cnt++;
251 		if (ca->backoff_cnt > use_ineff)
252 			return false;
253 	}
254 
255 	ca->shadow_wnd = max(ca->shadow_wnd, tcp_snd_cwnd(tp));
256 	ca->state = CDG_BACKOFF;
257 	tcp_enter_cwr(sk);
258 	return true;
259 }
260 
261 /* Not called in CWR or Recovery state. */
262 static void tcp_cdg_cong_avoid(struct sock *sk, u32 ack, u32 acked)
263 {
264 	struct cdg *ca = inet_csk_ca(sk);
265 	struct tcp_sock *tp = tcp_sk(sk);
266 	u32 prior_snd_cwnd;
267 	u32 incr;
268 
269 	if (tcp_in_slow_start(tp) && hystart_detect)
270 		tcp_cdg_hystart_update(sk);
271 
272 	if (after(ack, ca->rtt_seq) && ca->rtt.v64) {
273 		s32 grad = 0;
274 
275 		if (ca->rtt_prev.v64)
276 			grad = tcp_cdg_grad(ca);
277 		ca->rtt_seq = tp->snd_nxt;
278 		ca->rtt_prev = ca->rtt;
279 		ca->rtt.v64 = 0;
280 		ca->last_ack = 0;
281 		ca->sample_cnt = 0;
282 
283 		if (grad > 0 && tcp_cdg_backoff(sk, grad))
284 			return;
285 	}
286 
287 	if (!tcp_is_cwnd_limited(sk)) {
288 		ca->shadow_wnd = min(ca->shadow_wnd, tcp_snd_cwnd(tp));
289 		return;
290 	}
291 
292 	prior_snd_cwnd = tcp_snd_cwnd(tp);
293 	tcp_reno_cong_avoid(sk, ack, acked);
294 
295 	incr = tcp_snd_cwnd(tp) - prior_snd_cwnd;
296 	ca->shadow_wnd = max(ca->shadow_wnd, ca->shadow_wnd + incr);
297 }
298 
299 static void tcp_cdg_acked(struct sock *sk, const struct ack_sample *sample)
300 {
301 	struct cdg *ca = inet_csk_ca(sk);
302 	struct tcp_sock *tp = tcp_sk(sk);
303 
304 	if (sample->rtt_us <= 0)
305 		return;
306 
307 	/* A heuristic for filtering delayed ACKs, adapted from:
308 	 * D.A. Hayes. "Timing enhancements to the FreeBSD kernel to support
309 	 * delay and rate based TCP mechanisms." TR 100219A. CAIA, 2010.
310 	 */
311 	if (tp->sacked_out == 0) {
312 		if (sample->pkts_acked == 1 && ca->delack) {
313 			/* A delayed ACK is only used for the minimum if it is
314 			 * provenly lower than an existing non-zero minimum.
315 			 */
316 			ca->rtt.min = min(ca->rtt.min, sample->rtt_us);
317 			ca->delack--;
318 			return;
319 		} else if (sample->pkts_acked > 1 && ca->delack < 5) {
320 			ca->delack++;
321 		}
322 	}
323 
324 	ca->rtt.min = min_not_zero(ca->rtt.min, sample->rtt_us);
325 	ca->rtt.max = max(ca->rtt.max, sample->rtt_us);
326 }
327 
328 static u32 tcp_cdg_ssthresh(struct sock *sk)
329 {
330 	struct cdg *ca = inet_csk_ca(sk);
331 	struct tcp_sock *tp = tcp_sk(sk);
332 
333 	if (ca->state == CDG_BACKOFF)
334 		return max(2U, (tcp_snd_cwnd(tp) * min(1024U, backoff_beta)) >> 10);
335 
336 	if (ca->state == CDG_NONFULL && use_tolerance)
337 		return tcp_snd_cwnd(tp);
338 
339 	ca->shadow_wnd = min(ca->shadow_wnd >> 1, tcp_snd_cwnd(tp));
340 	if (use_shadow)
341 		return max3(2U, ca->shadow_wnd, tcp_snd_cwnd(tp) >> 1);
342 	return max(2U, tcp_snd_cwnd(tp) >> 1);
343 }
344 
345 static void tcp_cdg_cwnd_event(struct sock *sk, const enum tcp_ca_event ev)
346 {
347 	struct cdg *ca = inet_csk_ca(sk);
348 	struct tcp_sock *tp = tcp_sk(sk);
349 	struct cdg_minmax *gradients;
350 
351 	switch (ev) {
352 	case CA_EVENT_CWND_RESTART:
353 		gradients = ca->gradients;
354 		if (gradients)
355 			memset(gradients, 0, window * sizeof(gradients[0]));
356 		memset(ca, 0, sizeof(*ca));
357 
358 		ca->gradients = gradients;
359 		ca->rtt_seq = tp->snd_nxt;
360 		ca->shadow_wnd = tcp_snd_cwnd(tp);
361 		break;
362 	case CA_EVENT_COMPLETE_CWR:
363 		ca->state = CDG_UNKNOWN;
364 		ca->rtt_seq = tp->snd_nxt;
365 		ca->rtt_prev = ca->rtt;
366 		ca->rtt.v64 = 0;
367 		break;
368 	default:
369 		break;
370 	}
371 }
372 
373 static void tcp_cdg_init(struct sock *sk)
374 {
375 	struct cdg *ca = inet_csk_ca(sk);
376 	struct tcp_sock *tp = tcp_sk(sk);
377 
378 	/* We silently fall back to window = 1 if allocation fails. */
379 	if (window > 1)
380 		ca->gradients = kcalloc(window, sizeof(ca->gradients[0]),
381 					GFP_NOWAIT | __GFP_NOWARN);
382 	ca->rtt_seq = tp->snd_nxt;
383 	ca->shadow_wnd = tcp_snd_cwnd(tp);
384 }
385 
386 static void tcp_cdg_release(struct sock *sk)
387 {
388 	struct cdg *ca = inet_csk_ca(sk);
389 
390 	kfree(ca->gradients);
391 }
392 
393 static struct tcp_congestion_ops tcp_cdg __read_mostly = {
394 	.cong_avoid = tcp_cdg_cong_avoid,
395 	.cwnd_event = tcp_cdg_cwnd_event,
396 	.pkts_acked = tcp_cdg_acked,
397 	.undo_cwnd = tcp_reno_undo_cwnd,
398 	.ssthresh = tcp_cdg_ssthresh,
399 	.release = tcp_cdg_release,
400 	.init = tcp_cdg_init,
401 	.owner = THIS_MODULE,
402 	.name = "cdg",
403 };
404 
405 static int __init tcp_cdg_register(void)
406 {
407 	if (backoff_beta > 1024 || window < 1 || window > 256)
408 		return -ERANGE;
409 	if (!is_power_of_2(window))
410 		return -EINVAL;
411 
412 	BUILD_BUG_ON(sizeof(struct cdg) > ICSK_CA_PRIV_SIZE);
413 	tcp_register_congestion_control(&tcp_cdg);
414 	return 0;
415 }
416 
417 static void __exit tcp_cdg_unregister(void)
418 {
419 	tcp_unregister_congestion_control(&tcp_cdg);
420 }
421 
422 module_init(tcp_cdg_register);
423 module_exit(tcp_cdg_unregister);
424 MODULE_AUTHOR("Kenneth Klette Jonassen");
425 MODULE_LICENSE("GPL");
426 MODULE_DESCRIPTION("TCP CDG");
427