xref: /freebsd/sys/net/altq/altq_codel.c (revision 19261079) 
1 /*
2  * CoDel - The Controlled-Delay Active Queue Management algorithm
3  *
4  *  Copyright (C) 2013 Ermal Luçi <eri@FreeBSD.org>
5  *  Copyright (C) 2011-2012 Kathleen Nichols <nichols@pollere.com>
6  *  Copyright (C) 2011-2012 Van Jacobson <van@pollere.net>
7  *  Copyright (C) 2012 Michael D. Taht <dave.taht@bufferbloat.net>
8  *  Copyright (C) 2012 Eric Dumazet <edumazet@google.com>
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions, and the following disclaimer,
15  *    without modification.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The names of the authors may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission.
21  *
22  * Alternatively, provided that this notice is retained in full, this
23  * software may be distributed under the terms of the GNU General
24  * Public License ("GPL") version 2, in which case the provisions of the
25  * GPL apply INSTEAD OF those given above.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
28  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
29  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
30  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
31  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
32  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
33  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
34  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
35  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
36  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
37  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
38  * DAMAGE.
39  *
40  * $FreeBSD$
41  */
42 #include "opt_altq.h"
43 #include "opt_inet.h"
44 #include "opt_inet6.h"
45 
46 #ifdef ALTQ_CODEL  /* CoDel is enabled by ALTQ_CODEL option in opt_altq.h */
47 
48 #include <sys/param.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/socket.h>
52 #include <sys/systm.h>
53 
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <netinet/in.h>
57 
58 #include <netpfil/pf/pf.h>
59 #include <netpfil/pf/pf_altq.h>
60 #include <net/altq/if_altq.h>
61 #include <net/altq/altq.h>
62 #include <net/altq/altq_codel.h>
63 
64 static int		 codel_should_drop(struct codel *, class_queue_t *,
65 			    struct mbuf *, u_int64_t);
66 static void		 codel_Newton_step(struct codel_vars *);
67 static u_int64_t	 codel_control_law(u_int64_t t, u_int64_t, u_int32_t);
68 
69 #define	codel_time_after(a, b)		((int64_t)(a) - (int64_t)(b) > 0)
70 #define	codel_time_after_eq(a, b)	((int64_t)(a) - (int64_t)(b) >= 0)
71 #define	codel_time_before(a, b)		((int64_t)(a) - (int64_t)(b) < 0)
72 #define	codel_time_before_eq(a, b)	((int64_t)(a) - (int64_t)(b) <= 0)
73 
74 static int codel_request(struct ifaltq *, int, void *);
75 
76 static int codel_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *);
77 static struct mbuf *codel_dequeue(struct ifaltq *, int);
78 
79 int
80 codel_pfattach(struct pf_altq *a)
81 {
82 	struct ifnet *ifp;
83 
84 	if ((ifp = ifunit(a->ifname)) == NULL || a->altq_disc == NULL)
85 		return (EINVAL);
86 
87 	return (altq_attach(&ifp->if_snd, ALTQT_CODEL, a->altq_disc,
88 	    codel_enqueue, codel_dequeue, codel_request));
89 }
90 
91 int
92 codel_add_altq(struct ifnet *ifp, struct pf_altq *a)
93 {
94 	struct codel_if	*cif;
95 	struct codel_opts	*opts;
96 
97 	if (ifp == NULL)
98 		return (EINVAL);
99 	if (!ALTQ_IS_READY(&ifp->if_snd))
100 		return (ENODEV);
101 
102 	opts = &a->pq_u.codel_opts;
103 
104 	cif = malloc(sizeof(struct codel_if), M_DEVBUF, M_NOWAIT | M_ZERO);
105 	if (cif == NULL)
106 		return (ENOMEM);
107 	cif->cif_bandwidth = a->ifbandwidth;
108 	cif->cif_ifq = &ifp->if_snd;
109 
110 	cif->cl_q = malloc(sizeof(class_queue_t), M_DEVBUF, M_NOWAIT | M_ZERO);
111 	if (cif->cl_q == NULL) {
112 		free(cif, M_DEVBUF);
113 		return (ENOMEM);
114 	}
115 
116 	if (a->qlimit == 0)
117 		a->qlimit = 50;	/* use default. */
118 	qlimit(cif->cl_q) = a->qlimit;
119 	qtype(cif->cl_q) = Q_CODEL;
120 	qlen(cif->cl_q) = 0;
121 	qsize(cif->cl_q) = 0;
122 
123 	if (opts->target == 0)
124 		opts->target = 5;
125 	if (opts->interval == 0)
126 		opts->interval = 100;
127 	cif->codel.params.target = machclk_freq * opts->target / 1000;
128 	cif->codel.params.interval = machclk_freq * opts->interval / 1000;
129 	cif->codel.params.ecn = opts->ecn;
130 	cif->codel.stats.maxpacket = 256;
131 
132 	cif->cl_stats.qlength = qlen(cif->cl_q);
133 	cif->cl_stats.qlimit = qlimit(cif->cl_q);
134 
135 	/* keep the state in pf_altq */
136 	a->altq_disc = cif;
137 
138 	return (0);
139 }
140 
141 int
142 codel_remove_altq(struct pf_altq *a)
143 {
144 	struct codel_if *cif;
145 
146 	if ((cif = a->altq_disc) == NULL)
147 		return (EINVAL);
148 	a->altq_disc = NULL;
149 
150 	if (cif->cl_q)
151 		free(cif->cl_q, M_DEVBUF);
152 	free(cif, M_DEVBUF);
153 
154 	return (0);
155 }
156 
157 int
158 codel_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version)
159 {
160 	struct codel_if *cif;
161 	struct codel_ifstats stats;
162 	int error = 0;
163 
164 	if ((cif = altq_lookup(a->ifname, ALTQT_CODEL)) == NULL)
165 		return (EBADF);
166 
167 	if (*nbytes < sizeof(stats))
168 		return (EINVAL);
169 
170 	stats = cif->cl_stats;
171 	stats.stats = cif->codel.stats;
172 
173 	if ((error = copyout((caddr_t)&stats, ubuf, sizeof(stats))) != 0)
174 		return (error);
175 	*nbytes = sizeof(stats);
176 
177 	return (0);
178 }
179 
180 static int
181 codel_request(struct ifaltq *ifq, int req, void *arg)
182 {
183 	struct codel_if	*cif = (struct codel_if *)ifq->altq_disc;
184 	struct mbuf *m;
185 
186 	IFQ_LOCK_ASSERT(ifq);
187 
188 	switch (req) {
189 	case ALTRQ_PURGE:
190 		if (!ALTQ_IS_ENABLED(cif->cif_ifq))
191 			break;
192 
193 		if (qempty(cif->cl_q))
194 			break;
195 
196 		while ((m = _getq(cif->cl_q)) != NULL) {
197 			PKTCNTR_ADD(&cif->cl_stats.cl_dropcnt, m_pktlen(m));
198 			m_freem(m);
199 			IFQ_DEC_LEN(cif->cif_ifq);
200 		}
201 		cif->cif_ifq->ifq_len = 0;
202 		break;
203 	}
204 
205 	return (0);
206 }
207 
208 static int
209 codel_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr)
210 {
211 
212 	struct codel_if *cif = (struct codel_if *) ifq->altq_disc;
213 
214 	IFQ_LOCK_ASSERT(ifq);
215 
216 	/* grab class set by classifier */
217 	if ((m->m_flags & M_PKTHDR) == 0) {
218 		/* should not happen */
219 		printf("altq: packet for %s does not have pkthdr\n",
220 		   ifq->altq_ifp->if_xname);
221 		m_freem(m);
222 		PKTCNTR_ADD(&cif->cl_stats.cl_dropcnt, m_pktlen(m));
223 		return (ENOBUFS);
224 	}
225 
226 	if (codel_addq(&cif->codel, cif->cl_q, m)) {
227 		PKTCNTR_ADD(&cif->cl_stats.cl_dropcnt, m_pktlen(m));
228 		return (ENOBUFS);
229 	}
230 	IFQ_INC_LEN(ifq);
231 
232 	return (0);
233 }
234 
235 static struct mbuf *
236 codel_dequeue(struct ifaltq *ifq, int op)
237 {
238 	struct codel_if *cif = (struct codel_if *)ifq->altq_disc;
239 	struct mbuf *m;
240 
241 	IFQ_LOCK_ASSERT(ifq);
242 
243 	if (IFQ_IS_EMPTY(ifq))
244 		return (NULL);
245 
246 	if (op == ALTDQ_POLL)
247 		return (qhead(cif->cl_q));
248 
249 	m = codel_getq(&cif->codel, cif->cl_q);
250 	if (m != NULL) {
251 		IFQ_DEC_LEN(ifq);
252 		PKTCNTR_ADD(&cif->cl_stats.cl_xmitcnt, m_pktlen(m));
253 		return (m);
254 	}
255 
256 	return (NULL);
257 }
258 
259 struct codel *
260 codel_alloc(int target, int interval, int ecn)
261 {
262 	struct codel *c;
263 
264 	c = malloc(sizeof(*c), M_DEVBUF, M_NOWAIT | M_ZERO);
265 	if (c != NULL) {
266 		c->params.target = machclk_freq * target / 1000;
267 		c->params.interval = machclk_freq * interval / 1000;
268 		c->params.ecn = ecn;
269 		c->stats.maxpacket = 256;
270 	}
271 
272 	return (c);
273 }
274 
275 void
276 codel_destroy(struct codel *c)
277 {
278 
279 	free(c, M_DEVBUF);
280 }
281 
282 #define	MTAG_CODEL	1438031249
283 int
284 codel_addq(struct codel *c, class_queue_t *q, struct mbuf *m)
285 {
286 	struct m_tag *mtag;
287 	uint64_t *enqueue_time;
288 
289 	if (qlen(q) < qlimit(q)) {
290 		mtag = m_tag_locate(m, MTAG_CODEL, 0, NULL);
291 		if (mtag == NULL)
292 			mtag = m_tag_alloc(MTAG_CODEL, 0, sizeof(uint64_t),
293 			    M_NOWAIT);
294 		if (mtag == NULL) {
295 			m_freem(m);
296 			return (-1);
297 		}
298 		enqueue_time = (uint64_t *)(mtag + 1);
299 		*enqueue_time = read_machclk();
300 		m_tag_prepend(m, mtag);
301 		_addq(q, m);
302 		return (0);
303 	}
304 	c->drop_overlimit++;
305 	m_freem(m);
306 
307 	return (-1);
308 }
309 
310 static int
311 codel_should_drop(struct codel *c, class_queue_t *q, struct mbuf *m,
312     u_int64_t now)
313 {
314 	struct m_tag *mtag;
315 	uint64_t *enqueue_time;
316 
317 	if (m == NULL) {
318 		c->vars.first_above_time = 0;
319 		return (0);
320 	}
321 
322 	mtag = m_tag_locate(m, MTAG_CODEL, 0, NULL);
323 	if (mtag == NULL) {
324 		/* Only one warning per second. */
325 		if (ppsratecheck(&c->last_log, &c->last_pps, 1))
326 			printf("%s: could not found the packet mtag!\n",
327 			    __func__);
328 		c->vars.first_above_time = 0;
329 		return (0);
330 	}
331 	enqueue_time = (uint64_t *)(mtag + 1);
332 	c->vars.ldelay = now - *enqueue_time;
333 	c->stats.maxpacket = MAX(c->stats.maxpacket, m_pktlen(m));
334 
335 	if (codel_time_before(c->vars.ldelay, c->params.target) ||
336 	    qsize(q) <= c->stats.maxpacket) {
337 		/* went below - stay below for at least interval */
338 		c->vars.first_above_time = 0;
339 		return (0);
340 	}
341 	if (c->vars.first_above_time == 0) {
342 		/* just went above from below. If we stay above
343 		 * for at least interval we'll say it's ok to drop
344 		 */
345 		c->vars.first_above_time = now + c->params.interval;
346 		return (0);
347 	}
348 	if (codel_time_after(now, c->vars.first_above_time))
349 		return (1);
350 
351 	return (0);
352 }
353 
354 /*
355  * Run a Newton method step:
356  * new_invsqrt = (invsqrt / 2) * (3 - count * invsqrt^2)
357  *
358  * Here, invsqrt is a fixed point number (< 1.0), 32bit mantissa, aka Q0.32
359  */
360 static void
361 codel_Newton_step(struct codel_vars *vars)
362 {
363 	uint32_t invsqrt, invsqrt2;
364 	uint64_t val;
365 
366 /* sizeof_in_bits(rec_inv_sqrt) */
367 #define	REC_INV_SQRT_BITS (8 * sizeof(u_int16_t))
368 /* needed shift to get a Q0.32 number from rec_inv_sqrt */
369 #define	REC_INV_SQRT_SHIFT (32 - REC_INV_SQRT_BITS)
370 
371 	invsqrt = ((u_int32_t)vars->rec_inv_sqrt) << REC_INV_SQRT_SHIFT;
372 	invsqrt2 = ((u_int64_t)invsqrt * invsqrt) >> 32;
373 	val = (3LL << 32) - ((u_int64_t)vars->count * invsqrt2);
374 	val >>= 2; /* avoid overflow in following multiply */
375 	val = (val * invsqrt) >> (32 - 2 + 1);
376 
377 	vars->rec_inv_sqrt = val >> REC_INV_SQRT_SHIFT;
378 }
379 
380 static u_int64_t
381 codel_control_law(u_int64_t t, u_int64_t interval, u_int32_t rec_inv_sqrt)
382 {
383 
384 	return (t + (u_int32_t)(((u_int64_t)interval *
385 	    (rec_inv_sqrt << REC_INV_SQRT_SHIFT)) >> 32));
386 }
387 
388 struct mbuf *
389 codel_getq(struct codel *c, class_queue_t *q)
390 {
391 	struct mbuf	*m;
392 	u_int64_t	 now;
393 	int		 drop;
394 
395 	if ((m = _getq(q)) == NULL) {
396 		c->vars.dropping = 0;
397 		return (m);
398 	}
399 
400 	now = read_machclk();
401 	drop = codel_should_drop(c, q, m, now);
402 	if (c->vars.dropping) {
403 		if (!drop) {
404 			/* sojourn time below target - leave dropping state */
405 			c->vars.dropping = 0;
406 		} else if (codel_time_after_eq(now, c->vars.drop_next)) {
407 			/* It's time for the next drop. Drop the current
408 			 * packet and dequeue the next. The dequeue might
409 			 * take us out of dropping state.
410 			 * If not, schedule the next drop.
411 			 * A large backlog might result in drop rates so high
412 			 * that the next drop should happen now,
413 			 * hence the while loop.
414 			 */
415 			while (c->vars.dropping &&
416 			    codel_time_after_eq(now, c->vars.drop_next)) {
417 				c->vars.count++; /* don't care of possible wrap
418 						  * since there is no more
419 						  * divide */
420 				codel_Newton_step(&c->vars);
421 				/* TODO ECN */
422 				PKTCNTR_ADD(&c->stats.drop_cnt, m_pktlen(m));
423 				m_freem(m);
424 				m = _getq(q);
425 				if (!codel_should_drop(c, q, m, now))
426 					/* leave dropping state */
427 					c->vars.dropping = 0;
428 				else
429 					/* and schedule the next drop */
430 					c->vars.drop_next =
431 					    codel_control_law(c->vars.drop_next,
432 						c->params.interval,
433 						c->vars.rec_inv_sqrt);
434 			}
435 		}
436 	} else if (drop) {
437 		/* TODO ECN */
438 		PKTCNTR_ADD(&c->stats.drop_cnt, m_pktlen(m));
439 		m_freem(m);
440 
441 		m = _getq(q);
442 		drop = codel_should_drop(c, q, m, now);
443 
444 		c->vars.dropping = 1;
445 		/* if min went above target close to when we last went below it
446 		 * assume that the drop rate that controlled the queue on the
447 		 * last cycle is a good starting point to control it now.
448 		 */
449 		if (codel_time_before(now - c->vars.drop_next,
450 		    16 * c->params.interval)) {
451 			c->vars.count = (c->vars.count - c->vars.lastcount) | 1;
452 			/* we dont care if rec_inv_sqrt approximation
453 			 * is not very precise :
454 			 * Next Newton steps will correct it quadratically.
455 			 */
456 			codel_Newton_step(&c->vars);
457 		} else {
458 			c->vars.count = 1;
459 			c->vars.rec_inv_sqrt = ~0U >> REC_INV_SQRT_SHIFT;
460 		}
461 		c->vars.lastcount = c->vars.count;
462 		c->vars.drop_next = codel_control_law(now, c->params.interval,
463 		    c->vars.rec_inv_sqrt);
464 	}
465 
466 	return (m);
467 }
468 
469 void
470 codel_getstats(struct codel *c, struct codel_stats *s)
471 {
472 	*s = c->stats;
473 }
474 
475 #endif /* ALTQ_CODEL */
476