xref: /freebsd/sys/netinet/cc/cc_chd.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
1 /*-
2  * Copyright (c) 2009-2010
3  *	Swinburne University of Technology, Melbourne, Australia
4  * Copyright (c) 2010-2011 The FreeBSD Foundation
5  * All rights reserved.
6  *
7  * This software was developed at the Centre for Advanced Internet
8  * Architectures, Swinburne University of Technology, by David Hayes and
9  * Lawrence Stewart, made possible in part by a grant from the Cisco University
10  * Research Program Fund at Community Foundation Silicon Valley.
11  *
12  * Portions of this software were developed at the Centre for Advanced Internet
13  * Architectures, Swinburne University of Technology, Melbourne, Australia by
14  * David Hayes under sponsorship from the FreeBSD Foundation.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 /*
39  * An implementation of the CAIA-Hamilton delay based congestion control
40  * algorithm, based on "Improved coexistence and loss tolerance for delay based
41  * TCP congestion control" by D. A. Hayes and G. Armitage., in 35th Annual IEEE
42  * Conference on Local Computer Networks (LCN 2010), Denver, Colorado, USA,
43  * 11-14 October 2010.
44  *
45  * Originally released as part of the NewTCP research project at Swinburne
46  * University of Technology's Centre for Advanced Internet Architectures,
47  * Melbourne, Australia, which was made possible in part by a grant from the
48  * Cisco University Research Program Fund at Community Foundation Silicon
49  * Valley. More details are available at:
50  *   http://caia.swin.edu.au/urp/newtcp/
51  */
52 
53 #include <sys/cdefs.h>
54 __FBSDID("$FreeBSD$");
55 
56 #include <sys/param.h>
57 #include <sys/kernel.h>
58 #include <sys/khelp.h>
59 #include <sys/limits.h>
60 #include <sys/malloc.h>
61 #include <sys/module.h>
62 #include <sys/queue.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/sysctl.h>
66 #include <sys/systm.h>
67 
68 #include <net/if.h>
69 #include <net/vnet.h>
70 
71 #include <netinet/cc.h>
72 #include <netinet/tcp_seq.h>
73 #include <netinet/tcp_timer.h>
74 #include <netinet/tcp_var.h>
75 
76 #include <netinet/cc/cc_module.h>
77 
78 #include <netinet/khelp/h_ertt.h>
79 
80 #define	CAST_PTR_INT(X)	(*((int*)(X)))
81 
82 /*
83  * Private signal type for rate based congestion signal.
84  * See <netinet/cc.h> for appropriate bit-range to use for private signals.
85  */
86 #define	CC_CHD_DELAY	0x02000000
87 
88 /* Largest possible number returned by random(). */
89 #define	RANDOM_MAX	INT_MAX
90 
91 static void	chd_ack_received(struct cc_var *ccv, uint16_t ack_type);
92 static void	chd_cb_destroy(struct cc_var *ccv);
93 static int	chd_cb_init(struct cc_var *ccv);
94 static void	chd_cong_signal(struct cc_var *ccv, uint32_t signal_type);
95 static void	chd_conn_init(struct cc_var *ccv);
96 static int	chd_mod_init(void);
97 
98 struct chd {
99 	/*
100 	 * Shadow window - keeps track of what the NewReno congestion window
101 	 * would have been if delay-based cwnd backoffs had not been made. This
102 	 * functionality aids coexistence with loss-based TCP flows which may be
103 	 * sharing links along the path.
104 	 */
105 	unsigned long shadow_w;
106 	/*
107 	 * Loss-based TCP compatibility flag - When set, it turns on the shadow
108 	 * window functionality.
109 	 */
110 	int loss_compete;
111 	 /* The maximum round trip time seen within a measured rtt period. */
112 	int maxrtt_in_rtt;
113 	/* The previous qdly that caused cwnd to backoff. */
114 	int prev_backoff_qdly;
115 };
116 
117 static int ertt_id;
118 
119 static VNET_DEFINE(uint32_t, chd_qmin) = 5;
120 static VNET_DEFINE(uint32_t, chd_pmax) = 50;
121 static VNET_DEFINE(uint32_t, chd_loss_fair) = 1;
122 static VNET_DEFINE(uint32_t, chd_use_max) = 1;
123 static VNET_DEFINE(uint32_t, chd_qthresh) = 20;
124 #define	V_chd_qthresh	VNET(chd_qthresh)
125 #define	V_chd_qmin	VNET(chd_qmin)
126 #define	V_chd_pmax	VNET(chd_pmax)
127 #define	V_chd_loss_fair	VNET(chd_loss_fair)
128 #define	V_chd_use_max	VNET(chd_use_max)
129 
130 static MALLOC_DEFINE(M_CHD, "chd data",
131     "Per connection data required for the CHD congestion control algorithm");
132 
133 struct cc_algo chd_cc_algo = {
134 	.name = "chd",
135 	.ack_received = chd_ack_received,
136 	.cb_destroy = chd_cb_destroy,
137 	.cb_init = chd_cb_init,
138 	.cong_signal = chd_cong_signal,
139 	.conn_init = chd_conn_init,
140 	.mod_init = chd_mod_init
141 };
142 
143 static __inline void
144 chd_window_decrease(struct cc_var *ccv)
145 {
146 	unsigned long win;
147 
148 	win = min(CCV(ccv, snd_wnd), CCV(ccv, snd_cwnd)) / CCV(ccv, t_maxseg);
149 	win -= max((win / 2), 1);
150 	CCV(ccv, snd_ssthresh) = max(win, 2) * CCV(ccv, t_maxseg);
151 }
152 
153 /*
154  * Probabilistic backoff function. Returns 1 if we should backoff or 0
155  * otherwise. The calculation of p is similar to the calculation of p in cc_hd.
156  */
157 static __inline int
158 should_backoff(int qdly, int maxqdly, struct chd *chd_data)
159 {
160 	unsigned long p, rand;
161 
162 	rand = random();
163 
164 	if (qdly < V_chd_qthresh) {
165 		chd_data->loss_compete = 0;
166 		p = (((RANDOM_MAX / 100) * V_chd_pmax) /
167 		    (V_chd_qthresh - V_chd_qmin)) *
168 		    (qdly - V_chd_qmin);
169 	} else {
170 		if (qdly > V_chd_qthresh) {
171 			p = (((RANDOM_MAX / 100) * V_chd_pmax) /
172 			    (maxqdly - V_chd_qthresh)) *
173 			    (maxqdly - qdly);
174 			if (V_chd_loss_fair && rand < p)
175 				chd_data->loss_compete = 1;
176 		} else {
177 			p = (RANDOM_MAX / 100) * V_chd_pmax;
178 			chd_data->loss_compete = 0;
179 		}
180 	}
181 
182 	return (rand < p);
183 }
184 
185 static __inline void
186 chd_window_increase(struct cc_var *ccv, int new_measurement)
187 {
188 	struct chd *chd_data;
189 	int incr;
190 
191 	chd_data = ccv->cc_data;
192 	incr = 0;
193 
194 	if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)) {
195 		/* Adapted from NewReno slow start. */
196 		if (V_tcp_do_rfc3465) {
197 			/* In slow-start with ABC enabled. */
198 			if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max)) {
199 				/* Not due to RTO. */
200 				incr = min(ccv->bytes_this_ack,
201 				    V_tcp_abc_l_var * CCV(ccv, t_maxseg));
202 			} else {
203 				/* Due to RTO. */
204 				incr = min(ccv->bytes_this_ack,
205 				    CCV(ccv, t_maxseg));
206 			}
207 		} else
208 			incr = CCV(ccv, t_maxseg);
209 
210 	} else { /* Congestion avoidance. */
211 		if (V_tcp_do_rfc3465) {
212 			if (ccv->flags & CCF_ABC_SENTAWND) {
213 				ccv->flags &= ~CCF_ABC_SENTAWND;
214 				incr = CCV(ccv, t_maxseg);
215 			}
216 		} else if (new_measurement)
217 			incr = CCV(ccv, t_maxseg);
218 	}
219 
220 	if (chd_data->shadow_w > 0) {
221 		/* Track NewReno window. */
222 		chd_data->shadow_w = min(chd_data->shadow_w + incr,
223 		    TCP_MAXWIN << CCV(ccv, snd_scale));
224 	}
225 
226 	CCV(ccv,snd_cwnd) = min(CCV(ccv, snd_cwnd) + incr,
227 	    TCP_MAXWIN << CCV(ccv, snd_scale));
228 }
229 
230 /*
231  * All ACK signals are used for timing measurements to determine delay-based
232  * congestion. However, window increases are only performed when
233  * ack_type == CC_ACK.
234  */
235 static void
236 chd_ack_received(struct cc_var *ccv, uint16_t ack_type)
237 {
238 	struct chd *chd_data;
239 	struct ertt *e_t;
240 	int backoff, new_measurement, qdly, rtt;
241 
242 	e_t = khelp_get_osd(CCV(ccv, osd), ertt_id);
243 	chd_data = ccv->cc_data;
244 	new_measurement = e_t->flags & ERTT_NEW_MEASUREMENT;
245 	backoff = qdly = 0;
246 
247 	chd_data->maxrtt_in_rtt = imax(e_t->rtt, chd_data->maxrtt_in_rtt);
248 
249 	if (new_measurement) {
250 		/*
251 		 * There is a new per RTT measurement, so check to see if there
252 		 * is delay based congestion.
253 		 */
254 		rtt = V_chd_use_max ? chd_data->maxrtt_in_rtt : e_t->rtt;
255 		chd_data->maxrtt_in_rtt = 0;
256 
257 		if (rtt && e_t->minrtt && !IN_RECOVERY(CCV(ccv, t_flags))) {
258 			qdly = rtt - e_t->minrtt;
259 			if (qdly > V_chd_qmin) {
260 				/*
261 				 * Probabilistic delay based congestion
262 				 * indication.
263 				 */
264 				backoff = should_backoff(qdly,
265 				    e_t->maxrtt - e_t->minrtt, chd_data);
266 			} else
267 				chd_data->loss_compete = 0;
268 		}
269 		/* Reset per RTT measurement flag to start a new measurement. */
270 		e_t->flags &= ~ERTT_NEW_MEASUREMENT;
271 	}
272 
273 	if (backoff) {
274 		/*
275 		 * Update shadow_w before delay based backoff.
276 		 */
277 		if (chd_data->loss_compete ||
278 		    qdly > chd_data->prev_backoff_qdly) {
279 			/*
280 			 * Delay is higher than when we backed off previously,
281 			 * so it is possible that this flow is competing with
282 			 * loss based flows.
283 			 */
284 			chd_data->shadow_w = max(CCV(ccv, snd_cwnd),
285 			    chd_data->shadow_w);
286 		} else {
287 			/*
288 			 * Reset shadow_w, as it is probable that this flow is
289 			 * not competing with loss based flows at the moment.
290 			 */
291 			chd_data->shadow_w = 0;
292 		}
293 
294 		chd_data->prev_backoff_qdly = qdly;
295 		/*
296 		 * Send delay-based congestion signal to the congestion signal
297 		 * handler.
298 		 */
299 		chd_cong_signal(ccv, CC_CHD_DELAY);
300 
301 	} else if (ack_type == CC_ACK)
302 		chd_window_increase(ccv, new_measurement);
303 }
304 
305 static void
306 chd_cb_destroy(struct cc_var *ccv)
307 {
308 
309 	if (ccv->cc_data != NULL)
310 		free(ccv->cc_data, M_CHD);
311 }
312 
313 static int
314 chd_cb_init(struct cc_var *ccv)
315 {
316 	struct chd *chd_data;
317 
318 	chd_data = malloc(sizeof(struct chd), M_CHD, M_NOWAIT);
319 	if (chd_data == NULL)
320 		return (ENOMEM);
321 
322 	chd_data->shadow_w = 0;
323 	ccv->cc_data = chd_data;
324 
325 	return (0);
326 }
327 
328 static void
329 chd_cong_signal(struct cc_var *ccv, uint32_t signal_type)
330 {
331 	struct ertt *e_t;
332 	struct chd *chd_data;
333 	int qdly;
334 
335 	e_t = khelp_get_osd(CCV(ccv, osd), ertt_id);
336 	chd_data = ccv->cc_data;
337 	qdly = imax(e_t->rtt, chd_data->maxrtt_in_rtt) - e_t->minrtt;
338 
339 	switch(signal_type) {
340 	case CC_CHD_DELAY:
341 		chd_window_decrease(ccv); /* Set new ssthresh. */
342 		CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
343 		CCV(ccv, snd_recover) = CCV(ccv, snd_max);
344 		ENTER_CONGRECOVERY(CCV(ccv, t_flags));
345 		break;
346 
347 	case CC_NDUPACK: /* Packet loss. */
348 		/*
349 		 * Only react to loss as a congestion signal if qdly >
350 		 * V_chd_qthresh.  If qdly is less than qthresh, presume that
351 		 * this is a non congestion related loss. If qdly is greater
352 		 * than qthresh, assume that we are competing with loss based
353 		 * tcp flows and restore window from any unnecessary backoffs,
354 		 * before the decrease.
355 		 */
356 		if (!IN_RECOVERY(CCV(ccv, t_flags)) && qdly > V_chd_qthresh) {
357 			if (chd_data->loss_compete) {
358 				CCV(ccv, snd_cwnd) = max(CCV(ccv, snd_cwnd),
359 				    chd_data->shadow_w);
360 			}
361 			chd_window_decrease(ccv);
362 		} else {
363 			 /*
364 			  * This loss isn't congestion related, or already
365 			  * recovering from congestion.
366 			  */
367 			CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd);
368 			CCV(ccv, snd_recover) = CCV(ccv, snd_max);
369 		}
370 
371 		if (chd_data->shadow_w > 0) {
372 			chd_data->shadow_w = max(chd_data->shadow_w /
373 			    CCV(ccv, t_maxseg) / 2, 2) * CCV(ccv, t_maxseg);
374 		}
375 		ENTER_FASTRECOVERY(CCV(ccv, t_flags));
376 		break;
377 
378 	default:
379 		newreno_cc_algo.cong_signal(ccv, signal_type);
380 	}
381 }
382 
383 static void
384 chd_conn_init(struct cc_var *ccv)
385 {
386 	struct chd *chd_data;
387 
388 	chd_data = ccv->cc_data;
389 	chd_data->prev_backoff_qdly = 0;
390 	chd_data->maxrtt_in_rtt = 0;
391 	chd_data->loss_compete = 0;
392 	/*
393 	 * Initialise the shadow_cwnd to be equal to snd_cwnd in case we are
394 	 * competing with loss based flows from the start.
395 	 */
396 	chd_data->shadow_w = CCV(ccv, snd_cwnd);
397 }
398 
399 static int
400 chd_mod_init(void)
401 {
402 
403 	ertt_id = khelp_get_id("ertt");
404 	if (ertt_id <= 0) {
405 		printf("%s: h_ertt module not found\n", __func__);
406 		return (ENOENT);
407 	}
408 
409 	chd_cc_algo.after_idle = newreno_cc_algo.after_idle;
410 	chd_cc_algo.post_recovery = newreno_cc_algo.post_recovery;
411 
412 	return (0);
413 }
414 
415 static int
416 chd_loss_fair_handler(SYSCTL_HANDLER_ARGS)
417 {
418 	int error;
419 	uint32_t new;
420 
421 	new = V_chd_loss_fair;
422 	error = sysctl_handle_int(oidp, &new, 0, req);
423 	if (error == 0 && req->newptr != NULL) {
424 		if (CAST_PTR_INT(req->newptr) > 1)
425 			error = EINVAL;
426 		else
427 			V_chd_loss_fair = new;
428 	}
429 
430 	return (error);
431 }
432 
433 static int
434 chd_pmax_handler(SYSCTL_HANDLER_ARGS)
435 {
436 	int error;
437 	uint32_t new;
438 
439 	new = V_chd_pmax;
440 	error = sysctl_handle_int(oidp, &new, 0, req);
441 	if (error == 0 && req->newptr != NULL) {
442 		if (CAST_PTR_INT(req->newptr) == 0 ||
443 		    CAST_PTR_INT(req->newptr) > 100)
444 			error = EINVAL;
445 		else
446 			V_chd_pmax = new;
447 	}
448 
449 	return (error);
450 }
451 
452 static int
453 chd_qthresh_handler(SYSCTL_HANDLER_ARGS)
454 {
455 	int error;
456 	uint32_t new;
457 
458 	new = V_chd_qthresh;
459 	error = sysctl_handle_int(oidp, &new, 0, req);
460 	if (error == 0 && req->newptr != NULL) {
461 		if (CAST_PTR_INT(req->newptr) <= V_chd_qmin)
462 			error = EINVAL;
463 		else
464 			V_chd_qthresh = new;
465 	}
466 
467 	return (error);
468 }
469 
470 SYSCTL_DECL(_net_inet_tcp_cc_chd);
471 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, chd, CTLFLAG_RW, NULL,
472     "CAIA Hamilton delay-based congestion control related settings");
473 
474 SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, loss_fair,
475     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW,
476     &VNET_NAME(chd_loss_fair), 1, &chd_loss_fair_handler,
477     "IU", "Flag to enable shadow window functionality.");
478 
479 SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, pmax,
480     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW,
481     &VNET_NAME(chd_pmax), 5, &chd_pmax_handler,
482     "IU", "Per RTT maximum backoff probability as a percentage");
483 
484 SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, queue_threshold,
485     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW,
486     &VNET_NAME(chd_qthresh), 20, &chd_qthresh_handler,
487     "IU", "Queueing congestion threshold in ticks");
488 
489 SYSCTL_UINT(_net_inet_tcp_cc_chd, OID_AUTO, queue_min,
490     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(chd_qmin), 5,
491     "Minimum queueing delay threshold in ticks");
492 
493 SYSCTL_UINT(_net_inet_tcp_cc_chd,  OID_AUTO, use_max,
494     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(chd_use_max), 1,
495     "Use the maximum RTT seen within the measurement period (RTT) "
496     "as the basic delay measurement for the algorithm.");
497 
498 DECLARE_CC_MODULE(chd, &chd_cc_algo);
499 MODULE_DEPEND(chd, ertt, 1, 1, 1);
500