pfctl_altq.c revision 164775
1/*	$OpenBSD: pfctl_altq.c,v 1.86 2005/02/28 14:04:51 henning Exp $	*/
2/* add:	$OpenBSD: pfctl_altq.c,v 1.91 2006/11/28 00:08:50 henning Exp $	*/
3
4/*
5 * Copyright (c) 2002
6 *	Sony Computer Science Laboratories Inc.
7 * Copyright (c) 2002, 2003 Henning Brauer <henning@openbsd.org>
8 *
9 * Permission to use, copy, modify, and distribute this software for any
10 * purpose with or without fee is hereby granted, provided that the above
11 * copyright notice and this permission notice appear in all copies.
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
14 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
15 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
16 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
17 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
18 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
19 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
20 */
21
22#include <sys/cdefs.h>
23__FBSDID("$FreeBSD: head/contrib/pf/pfctl/pfctl_altq.c 164775 2006-11-30 18:55:36Z mlaier $");
24
25#include <sys/param.h>
26#include <sys/ioctl.h>
27#include <sys/socket.h>
28
29#include <net/if.h>
30#include <netinet/in.h>
31#include <net/pfvar.h>
32
33#include <err.h>
34#include <errno.h>
35#include <limits.h>
36#include <math.h>
37#include <stdio.h>
38#include <stdlib.h>
39#include <string.h>
40#include <unistd.h>
41
42#include <altq/altq.h>
43#include <altq/altq_cbq.h>
44#include <altq/altq_priq.h>
45#include <altq/altq_hfsc.h>
46
47#include "pfctl_parser.h"
48#include "pfctl.h"
49
50#define is_sc_null(sc)	(((sc) == NULL) || ((sc)->m1 == 0 && (sc)->m2 == 0))
51
52TAILQ_HEAD(altqs, pf_altq) altqs = TAILQ_HEAD_INITIALIZER(altqs);
53LIST_HEAD(gen_sc, segment) rtsc, lssc;
54
55struct pf_altq	*qname_to_pfaltq(const char *, const char *);
56u_int32_t	 qname_to_qid(const char *);
57
58static int	eval_pfqueue_cbq(struct pfctl *, struct pf_altq *);
59static int	cbq_compute_idletime(struct pfctl *, struct pf_altq *);
60static int	check_commit_cbq(int, int, struct pf_altq *);
61static int	print_cbq_opts(const struct pf_altq *);
62
63static int	eval_pfqueue_priq(struct pfctl *, struct pf_altq *);
64static int	check_commit_priq(int, int, struct pf_altq *);
65static int	print_priq_opts(const struct pf_altq *);
66
67static int	eval_pfqueue_hfsc(struct pfctl *, struct pf_altq *);
68static int	check_commit_hfsc(int, int, struct pf_altq *);
69static int	print_hfsc_opts(const struct pf_altq *,
70		    const struct node_queue_opt *);
71
72static void		 gsc_add_sc(struct gen_sc *, struct service_curve *);
73static int		 is_gsc_under_sc(struct gen_sc *,
74			     struct service_curve *);
75static void		 gsc_destroy(struct gen_sc *);
76static struct segment	*gsc_getentry(struct gen_sc *, double);
77static int		 gsc_add_seg(struct gen_sc *, double, double, double,
78			     double);
79static double		 sc_x2y(struct service_curve *, double);
80
81#ifdef __FreeBSD__
82u_int32_t	 getifspeed(int, char *);
83#else
84u_int32_t	 getifspeed(char *);
85#endif
86u_long		 getifmtu(char *);
87int		 eval_queue_opts(struct pf_altq *, struct node_queue_opt *,
88		     u_int32_t);
89u_int32_t	 eval_bwspec(struct node_queue_bw *, u_int32_t);
90void		 print_hfsc_sc(const char *, u_int, u_int, u_int,
91		     const struct node_hfsc_sc *);
92
93void
94pfaltq_store(struct pf_altq *a)
95{
96	struct pf_altq	*altq;
97
98	if ((altq = malloc(sizeof(*altq))) == NULL)
99		err(1, "malloc");
100	memcpy(altq, a, sizeof(struct pf_altq));
101	TAILQ_INSERT_TAIL(&altqs, altq, entries);
102}
103
104void
105pfaltq_free(struct pf_altq *a)
106{
107	struct pf_altq	*altq;
108
109	TAILQ_FOREACH(altq, &altqs, entries) {
110		if (strncmp(a->ifname, altq->ifname, IFNAMSIZ) == 0 &&
111		    strncmp(a->qname, altq->qname, PF_QNAME_SIZE) == 0) {
112			TAILQ_REMOVE(&altqs, altq, entries);
113			free(altq);
114			return;
115		}
116	}
117}
118
119struct pf_altq *
120pfaltq_lookup(const char *ifname)
121{
122	struct pf_altq	*altq;
123
124	TAILQ_FOREACH(altq, &altqs, entries) {
125		if (strncmp(ifname, altq->ifname, IFNAMSIZ) == 0 &&
126		    altq->qname[0] == 0)
127			return (altq);
128	}
129	return (NULL);
130}
131
132struct pf_altq *
133qname_to_pfaltq(const char *qname, const char *ifname)
134{
135	struct pf_altq	*altq;
136
137	TAILQ_FOREACH(altq, &altqs, entries) {
138		if (strncmp(ifname, altq->ifname, IFNAMSIZ) == 0 &&
139		    strncmp(qname, altq->qname, PF_QNAME_SIZE) == 0)
140			return (altq);
141	}
142	return (NULL);
143}
144
145u_int32_t
146qname_to_qid(const char *qname)
147{
148	struct pf_altq	*altq;
149
150	/*
151	 * We guarantee that same named queues on different interfaces
152	 * have the same qid, so we do NOT need to limit matching on
153	 * one interface!
154	 */
155
156	TAILQ_FOREACH(altq, &altqs, entries) {
157		if (strncmp(qname, altq->qname, PF_QNAME_SIZE) == 0)
158			return (altq->qid);
159	}
160	return (0);
161}
162
163void
164print_altq(const struct pf_altq *a, unsigned level, struct node_queue_bw *bw,
165	struct node_queue_opt *qopts)
166{
167	if (a->qname[0] != 0) {
168		print_queue(a, level, bw, 0, qopts);
169		return;
170	}
171
172	printf("altq on %s ", a->ifname);
173
174	switch (a->scheduler) {
175	case ALTQT_CBQ:
176		if (!print_cbq_opts(a))
177			printf("cbq ");
178		break;
179	case ALTQT_PRIQ:
180		if (!print_priq_opts(a))
181			printf("priq ");
182		break;
183	case ALTQT_HFSC:
184		if (!print_hfsc_opts(a, qopts))
185			printf("hfsc ");
186		break;
187	}
188
189	if (bw != NULL && bw->bw_percent > 0) {
190		if (bw->bw_percent < 100)
191			printf("bandwidth %u%% ", bw->bw_percent);
192	} else
193		printf("bandwidth %s ", rate2str((double)a->ifbandwidth));
194
195	if (a->qlimit != DEFAULT_QLIMIT)
196		printf("qlimit %u ", a->qlimit);
197	printf("tbrsize %u ", a->tbrsize);
198}
199
200void
201print_queue(const struct pf_altq *a, unsigned level, struct node_queue_bw *bw,
202    int print_interface, struct node_queue_opt *qopts)
203{
204	unsigned	i;
205
206	printf("queue ");
207	for (i = 0; i < level; ++i)
208		printf(" ");
209	printf("%s ", a->qname);
210	if (print_interface)
211		printf("on %s ", a->ifname);
212	if (a->scheduler == ALTQT_CBQ || a->scheduler == ALTQT_HFSC) {
213		if (bw != NULL && bw->bw_percent > 0) {
214			if (bw->bw_percent < 100)
215				printf("bandwidth %u%% ", bw->bw_percent);
216		} else
217			printf("bandwidth %s ", rate2str((double)a->bandwidth));
218	}
219	if (a->priority != DEFAULT_PRIORITY)
220		printf("priority %u ", a->priority);
221	if (a->qlimit != DEFAULT_QLIMIT)
222		printf("qlimit %u ", a->qlimit);
223	switch (a->scheduler) {
224	case ALTQT_CBQ:
225		print_cbq_opts(a);
226		break;
227	case ALTQT_PRIQ:
228		print_priq_opts(a);
229		break;
230	case ALTQT_HFSC:
231		print_hfsc_opts(a, qopts);
232		break;
233	}
234}
235
236/*
237 * eval_pfaltq computes the discipline parameters.
238 */
239int
240eval_pfaltq(struct pfctl *pf, struct pf_altq *pa, struct node_queue_bw *bw,
241    struct node_queue_opt *opts)
242{
243	u_int	rate, size, errors = 0;
244
245	if (bw->bw_absolute > 0)
246		pa->ifbandwidth = bw->bw_absolute;
247	else
248#ifdef __FreeBSD__
249		if ((rate = getifspeed(pf->dev, pa->ifname)) == 0) {
250#else
251		if ((rate = getifspeed(pa->ifname)) == 0) {
252#endif
253			fprintf(stderr, "cannot determine interface bandwidth "
254			    "for %s, specify an absolute bandwidth\n",
255			    pa->ifname);
256			errors++;
257		} else if ((pa->ifbandwidth = eval_bwspec(bw, rate)) == 0)
258			pa->ifbandwidth = rate;
259
260	errors += eval_queue_opts(pa, opts, pa->ifbandwidth);
261
262	/* if tbrsize is not specified, use heuristics */
263	if (pa->tbrsize == 0) {
264		rate = pa->ifbandwidth;
265		if (rate <= 1 * 1000 * 1000)
266			size = 1;
267		else if (rate <= 10 * 1000 * 1000)
268			size = 4;
269		else if (rate <= 200 * 1000 * 1000)
270			size = 8;
271		else
272			size = 24;
273		size = size * getifmtu(pa->ifname);
274		if (size > 0xffff)
275			size = 0xffff;
276		pa->tbrsize = size;
277	}
278	return (errors);
279}
280
281/*
282 * check_commit_altq does consistency check for each interface
283 */
284int
285check_commit_altq(int dev, int opts)
286{
287	struct pf_altq	*altq;
288	int		 error = 0;
289
290	/* call the discipline check for each interface. */
291	TAILQ_FOREACH(altq, &altqs, entries) {
292		if (altq->qname[0] == 0) {
293			switch (altq->scheduler) {
294			case ALTQT_CBQ:
295				error = check_commit_cbq(dev, opts, altq);
296				break;
297			case ALTQT_PRIQ:
298				error = check_commit_priq(dev, opts, altq);
299				break;
300			case ALTQT_HFSC:
301				error = check_commit_hfsc(dev, opts, altq);
302				break;
303			default:
304				break;
305			}
306		}
307	}
308	return (error);
309}
310
311/*
312 * eval_pfqueue computes the queue parameters.
313 */
314int
315eval_pfqueue(struct pfctl *pf, struct pf_altq *pa, struct node_queue_bw *bw,
316    struct node_queue_opt *opts)
317{
318	/* should be merged with expand_queue */
319	struct pf_altq	*if_pa, *parent, *altq;
320	u_int32_t	 bwsum;
321	int		 error = 0;
322
323	/* find the corresponding interface and copy fields used by queues */
324	if ((if_pa = pfaltq_lookup(pa->ifname)) == NULL) {
325		fprintf(stderr, "altq not defined on %s\n", pa->ifname);
326		return (1);
327	}
328	pa->scheduler = if_pa->scheduler;
329	pa->ifbandwidth = if_pa->ifbandwidth;
330
331	if (qname_to_pfaltq(pa->qname, pa->ifname) != NULL) {
332		fprintf(stderr, "queue %s already exists on interface %s\n",
333		    pa->qname, pa->ifname);
334		return (1);
335	}
336	pa->qid = qname_to_qid(pa->qname);
337
338	parent = NULL;
339	if (pa->parent[0] != 0) {
340		parent = qname_to_pfaltq(pa->parent, pa->ifname);
341		if (parent == NULL) {
342			fprintf(stderr, "parent %s not found for %s\n",
343			    pa->parent, pa->qname);
344			return (1);
345		}
346		pa->parent_qid = parent->qid;
347	}
348	if (pa->qlimit == 0)
349		pa->qlimit = DEFAULT_QLIMIT;
350
351	if (pa->scheduler == ALTQT_CBQ || pa->scheduler == ALTQT_HFSC) {
352		pa->bandwidth = eval_bwspec(bw,
353		    parent == NULL ? 0 : parent->bandwidth);
354
355		if (pa->bandwidth > pa->ifbandwidth) {
356			fprintf(stderr, "bandwidth for %s higher than "
357			    "interface\n", pa->qname);
358			return (1);
359		}
360		/* check the sum of the child bandwidth is under parent's */
361		if (parent != NULL) {
362			if (pa->bandwidth > parent->bandwidth) {
363				warnx("bandwidth for %s higher than parent",
364				    pa->qname);
365				return (1);
366			}
367			bwsum = 0;
368			TAILQ_FOREACH(altq, &altqs, entries) {
369				if (strncmp(altq->ifname, pa->ifname,
370				    IFNAMSIZ) == 0 &&
371				    altq->qname[0] != 0 &&
372				    strncmp(altq->parent, pa->parent,
373				    PF_QNAME_SIZE) == 0)
374					bwsum += altq->bandwidth;
375			}
376			bwsum += pa->bandwidth;
377			if (bwsum > parent->bandwidth) {
378				warnx("the sum of the child bandwidth higher"
379				    " than parent \"%s\"", parent->qname);
380			}
381		}
382	}
383
384	if (eval_queue_opts(pa, opts, parent == NULL? 0 : parent->bandwidth))
385		return (1);
386
387	switch (pa->scheduler) {
388	case ALTQT_CBQ:
389		error = eval_pfqueue_cbq(pf, pa);
390		break;
391	case ALTQT_PRIQ:
392		error = eval_pfqueue_priq(pf, pa);
393		break;
394	case ALTQT_HFSC:
395		error = eval_pfqueue_hfsc(pf, pa);
396		break;
397	default:
398		break;
399	}
400	return (error);
401}
402
403/*
404 * CBQ support functions
405 */
406#define	RM_FILTER_GAIN	5	/* log2 of gain, e.g., 5 => 31/32 */
407#define	RM_NS_PER_SEC	(1000000000)
408
409static int
410eval_pfqueue_cbq(struct pfctl *pf, struct pf_altq *pa)
411{
412	struct cbq_opts	*opts;
413	u_int		 ifmtu;
414
415	if (pa->priority >= CBQ_MAXPRI) {
416		warnx("priority out of range: max %d", CBQ_MAXPRI - 1);
417		return (-1);
418	}
419
420	ifmtu = getifmtu(pa->ifname);
421	opts = &pa->pq_u.cbq_opts;
422
423	if (opts->pktsize == 0) {	/* use default */
424		opts->pktsize = ifmtu;
425		if (opts->pktsize > MCLBYTES)	/* do what TCP does */
426			opts->pktsize &= ~MCLBYTES;
427	} else if (opts->pktsize > ifmtu)
428		opts->pktsize = ifmtu;
429	if (opts->maxpktsize == 0)	/* use default */
430		opts->maxpktsize = ifmtu;
431	else if (opts->maxpktsize > ifmtu)
432		opts->pktsize = ifmtu;
433
434	if (opts->pktsize > opts->maxpktsize)
435		opts->pktsize = opts->maxpktsize;
436
437	if (pa->parent[0] == 0)
438		opts->flags |= (CBQCLF_ROOTCLASS | CBQCLF_WRR);
439
440	cbq_compute_idletime(pf, pa);
441	return (0);
442}
443
444/*
445 * compute ns_per_byte, maxidle, minidle, and offtime
446 */
447static int
448cbq_compute_idletime(struct pfctl *pf, struct pf_altq *pa)
449{
450	struct cbq_opts	*opts;
451	double		 maxidle_s, maxidle, minidle;
452	double		 offtime, nsPerByte, ifnsPerByte, ptime, cptime;
453	double		 z, g, f, gton, gtom;
454	u_int		 minburst, maxburst;
455
456	opts = &pa->pq_u.cbq_opts;
457	ifnsPerByte = (1.0 / (double)pa->ifbandwidth) * RM_NS_PER_SEC * 8;
458	minburst = opts->minburst;
459	maxburst = opts->maxburst;
460
461	if (pa->bandwidth == 0)
462		f = 0.0001;	/* small enough? */
463	else
464		f = ((double) pa->bandwidth / (double) pa->ifbandwidth);
465
466	nsPerByte = ifnsPerByte / f;
467	ptime = (double)opts->pktsize * ifnsPerByte;
468	cptime = ptime * (1.0 - f) / f;
469
470	if (nsPerByte * (double)opts->maxpktsize > (double)INT_MAX) {
471		/*
472		 * this causes integer overflow in kernel!
473		 * (bandwidth < 6Kbps when max_pkt_size=1500)
474		 */
475		if (pa->bandwidth != 0 && (pf->opts & PF_OPT_QUIET) == 0)
476			warnx("queue bandwidth must be larger than %s",
477			    rate2str(ifnsPerByte * (double)opts->maxpktsize /
478			    (double)INT_MAX * (double)pa->ifbandwidth));
479			fprintf(stderr, "cbq: queue %s is too slow!\n",
480			    pa->qname);
481		nsPerByte = (double)(INT_MAX / opts->maxpktsize);
482	}
483
484	if (maxburst == 0) {  /* use default */
485		if (cptime > 10.0 * 1000000)
486			maxburst = 4;
487		else
488			maxburst = 16;
489	}
490	if (minburst == 0)  /* use default */
491		minburst = 2;
492	if (minburst > maxburst)
493		minburst = maxburst;
494
495	z = (double)(1 << RM_FILTER_GAIN);
496	g = (1.0 - 1.0 / z);
497	gton = pow(g, (double)maxburst);
498	gtom = pow(g, (double)(minburst-1));
499	maxidle = ((1.0 / f - 1.0) * ((1.0 - gton) / gton));
500	maxidle_s = (1.0 - g);
501	if (maxidle > maxidle_s)
502		maxidle = ptime * maxidle;
503	else
504		maxidle = ptime * maxidle_s;
505	if (minburst)
506		offtime = cptime * (1.0 + 1.0/(1.0 - g) * (1.0 - gtom) / gtom);
507	else
508		offtime = cptime;
509	minidle = -((double)opts->maxpktsize * (double)nsPerByte);
510
511	/* scale parameters */
512	maxidle = ((maxidle * 8.0) / nsPerByte) *
513	    pow(2.0, (double)RM_FILTER_GAIN);
514	offtime = (offtime * 8.0) / nsPerByte *
515	    pow(2.0, (double)RM_FILTER_GAIN);
516	minidle = ((minidle * 8.0) / nsPerByte) *
517	    pow(2.0, (double)RM_FILTER_GAIN);
518
519	maxidle = maxidle / 1000.0;
520	offtime = offtime / 1000.0;
521	minidle = minidle / 1000.0;
522
523	opts->minburst = minburst;
524	opts->maxburst = maxburst;
525	opts->ns_per_byte = (u_int)nsPerByte;
526	opts->maxidle = (u_int)fabs(maxidle);
527	opts->minidle = (int)minidle;
528	opts->offtime = (u_int)fabs(offtime);
529
530	return (0);
531}
532
533static int
534check_commit_cbq(int dev, int opts, struct pf_altq *pa)
535{
536	struct pf_altq	*altq;
537	int		 root_class, default_class;
538	int		 error = 0;
539
540	/*
541	 * check if cbq has one root queue and one default queue
542	 * for this interface
543	 */
544	root_class = default_class = 0;
545	TAILQ_FOREACH(altq, &altqs, entries) {
546		if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
547			continue;
548		if (altq->qname[0] == 0)  /* this is for interface */
549			continue;
550		if (altq->pq_u.cbq_opts.flags & CBQCLF_ROOTCLASS)
551			root_class++;
552		if (altq->pq_u.cbq_opts.flags & CBQCLF_DEFCLASS)
553			default_class++;
554	}
555	if (root_class != 1) {
556		warnx("should have one root queue on %s", pa->ifname);
557		error++;
558	}
559	if (default_class != 1) {
560		warnx("should have one default queue on %s", pa->ifname);
561		error++;
562	}
563	return (error);
564}
565
566static int
567print_cbq_opts(const struct pf_altq *a)
568{
569	const struct cbq_opts	*opts;
570
571	opts = &a->pq_u.cbq_opts;
572	if (opts->flags) {
573		printf("cbq(");
574		if (opts->flags & CBQCLF_RED)
575			printf(" red");
576		if (opts->flags & CBQCLF_ECN)
577			printf(" ecn");
578		if (opts->flags & CBQCLF_RIO)
579			printf(" rio");
580		if (opts->flags & CBQCLF_CLEARDSCP)
581			printf(" cleardscp");
582		if (opts->flags & CBQCLF_FLOWVALVE)
583			printf(" flowvalve");
584		if (opts->flags & CBQCLF_BORROW)
585			printf(" borrow");
586		if (opts->flags & CBQCLF_WRR)
587			printf(" wrr");
588		if (opts->flags & CBQCLF_EFFICIENT)
589			printf(" efficient");
590		if (opts->flags & CBQCLF_ROOTCLASS)
591			printf(" root");
592		if (opts->flags & CBQCLF_DEFCLASS)
593			printf(" default");
594		printf(" ) ");
595
596		return (1);
597	} else
598		return (0);
599}
600
601/*
602 * PRIQ support functions
603 */
604static int
605eval_pfqueue_priq(struct pfctl *pf, struct pf_altq *pa)
606{
607	struct pf_altq	*altq;
608
609	if (pa->priority >= PRIQ_MAXPRI) {
610		warnx("priority out of range: max %d", PRIQ_MAXPRI - 1);
611		return (-1);
612	}
613	/* the priority should be unique for the interface */
614	TAILQ_FOREACH(altq, &altqs, entries) {
615		if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) == 0 &&
616		    altq->qname[0] != 0 && altq->priority == pa->priority) {
617			warnx("%s and %s have the same priority",
618			    altq->qname, pa->qname);
619			return (-1);
620		}
621	}
622
623	return (0);
624}
625
626static int
627check_commit_priq(int dev, int opts, struct pf_altq *pa)
628{
629	struct pf_altq	*altq;
630	int		 default_class;
631	int		 error = 0;
632
633	/*
634	 * check if priq has one default class for this interface
635	 */
636	default_class = 0;
637	TAILQ_FOREACH(altq, &altqs, entries) {
638		if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
639			continue;
640		if (altq->qname[0] == 0)  /* this is for interface */
641			continue;
642		if (altq->pq_u.priq_opts.flags & PRCF_DEFAULTCLASS)
643			default_class++;
644	}
645	if (default_class != 1) {
646		warnx("should have one default queue on %s", pa->ifname);
647		error++;
648	}
649	return (error);
650}
651
652static int
653print_priq_opts(const struct pf_altq *a)
654{
655	const struct priq_opts	*opts;
656
657	opts = &a->pq_u.priq_opts;
658
659	if (opts->flags) {
660		printf("priq(");
661		if (opts->flags & PRCF_RED)
662			printf(" red");
663		if (opts->flags & PRCF_ECN)
664			printf(" ecn");
665		if (opts->flags & PRCF_RIO)
666			printf(" rio");
667		if (opts->flags & PRCF_CLEARDSCP)
668			printf(" cleardscp");
669		if (opts->flags & PRCF_DEFAULTCLASS)
670			printf(" default");
671		printf(" ) ");
672
673		return (1);
674	} else
675		return (0);
676}
677
678/*
679 * HFSC support functions
680 */
681static int
682eval_pfqueue_hfsc(struct pfctl *pf, struct pf_altq *pa)
683{
684	struct pf_altq		*altq, *parent;
685	struct hfsc_opts	*opts;
686	struct service_curve	 sc;
687
688	opts = &pa->pq_u.hfsc_opts;
689
690	if (pa->parent[0] == 0) {
691		/* root queue */
692		opts->lssc_m1 = pa->ifbandwidth;
693		opts->lssc_m2 = pa->ifbandwidth;
694		opts->lssc_d = 0;
695		return (0);
696	}
697
698	LIST_INIT(&rtsc);
699	LIST_INIT(&lssc);
700
701	/* if link_share is not specified, use bandwidth */
702	if (opts->lssc_m2 == 0)
703		opts->lssc_m2 = pa->bandwidth;
704
705	if ((opts->rtsc_m1 > 0 && opts->rtsc_m2 == 0) ||
706	    (opts->lssc_m1 > 0 && opts->lssc_m2 == 0) ||
707	    (opts->ulsc_m1 > 0 && opts->ulsc_m2 == 0)) {
708		warnx("m2 is zero for %s", pa->qname);
709		return (-1);
710	}
711
712	if ((opts->rtsc_m1 < opts->rtsc_m2 && opts->rtsc_m1 != 0) ||
713	    (opts->lssc_m1 < opts->lssc_m2 && opts->lssc_m1 != 0) ||
714	    (opts->ulsc_m1 < opts->ulsc_m2 && opts->ulsc_m1 != 0)) {
715		warnx("m1 must be zero for convex curve: %s", pa->qname);
716		return (-1);
717	}
718
719	/*
720	 * admission control:
721	 * for the real-time service curve, the sum of the service curves
722	 * should not exceed 80% of the interface bandwidth.  20% is reserved
723	 * not to over-commit the actual interface bandwidth.
724	 * for the linkshare service curve, the sum of the child service
725	 * curve should not exceed the parent service curve.
726	 * for the upper-limit service curve, the assigned bandwidth should
727	 * be smaller than the interface bandwidth, and the upper-limit should
728	 * be larger than the real-time service curve when both are defined.
729	 */
730	parent = qname_to_pfaltq(pa->parent, pa->ifname);
731	if (parent == NULL)
732		errx(1, "parent %s not found for %s", pa->parent, pa->qname);
733
734	TAILQ_FOREACH(altq, &altqs, entries) {
735		if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
736			continue;
737		if (altq->qname[0] == 0)  /* this is for interface */
738			continue;
739
740		/* if the class has a real-time service curve, add it. */
741		if (opts->rtsc_m2 != 0 && altq->pq_u.hfsc_opts.rtsc_m2 != 0) {
742			sc.m1 = altq->pq_u.hfsc_opts.rtsc_m1;
743			sc.d = altq->pq_u.hfsc_opts.rtsc_d;
744			sc.m2 = altq->pq_u.hfsc_opts.rtsc_m2;
745			gsc_add_sc(&rtsc, &sc);
746		}
747
748		if (strncmp(altq->parent, pa->parent, PF_QNAME_SIZE) != 0)
749			continue;
750
751		/* if the class has a linkshare service curve, add it. */
752		if (opts->lssc_m2 != 0 && altq->pq_u.hfsc_opts.lssc_m2 != 0) {
753			sc.m1 = altq->pq_u.hfsc_opts.lssc_m1;
754			sc.d = altq->pq_u.hfsc_opts.lssc_d;
755			sc.m2 = altq->pq_u.hfsc_opts.lssc_m2;
756			gsc_add_sc(&lssc, &sc);
757		}
758	}
759
760	/* check the real-time service curve.  reserve 20% of interface bw */
761	if (opts->rtsc_m2 != 0) {
762		/* add this queue to the sum */
763		sc.m1 = opts->rtsc_m1;
764		sc.d = opts->rtsc_d;
765		sc.m2 = opts->rtsc_m2;
766		gsc_add_sc(&rtsc, &sc);
767		/* compare the sum with 80% of the interface */
768		sc.m1 = 0;
769		sc.d = 0;
770		sc.m2 = pa->ifbandwidth / 100 * 80;
771		if (!is_gsc_under_sc(&rtsc, &sc)) {
772			warnx("real-time sc exceeds 80%% of the interface "
773			    "bandwidth (%s)", rate2str((double)sc.m2));
774			goto err_ret;
775		}
776	}
777
778	/* check the linkshare service curve. */
779	if (opts->lssc_m2 != 0) {
780		/* add this queue to the child sum */
781		sc.m1 = opts->lssc_m1;
782		sc.d = opts->lssc_d;
783		sc.m2 = opts->lssc_m2;
784		gsc_add_sc(&lssc, &sc);
785		/* compare the sum of the children with parent's sc */
786		sc.m1 = parent->pq_u.hfsc_opts.lssc_m1;
787		sc.d = parent->pq_u.hfsc_opts.lssc_d;
788		sc.m2 = parent->pq_u.hfsc_opts.lssc_m2;
789		if (!is_gsc_under_sc(&lssc, &sc)) {
790			warnx("linkshare sc exceeds parent's sc");
791			goto err_ret;
792		}
793	}
794
795	/* check the upper-limit service curve. */
796	if (opts->ulsc_m2 != 0) {
797		if (opts->ulsc_m1 > pa->ifbandwidth ||
798		    opts->ulsc_m2 > pa->ifbandwidth) {
799			warnx("upper-limit larger than interface bandwidth");
800			goto err_ret;
801		}
802		if (opts->rtsc_m2 != 0 && opts->rtsc_m2 > opts->ulsc_m2) {
803			warnx("upper-limit sc smaller than real-time sc");
804			goto err_ret;
805		}
806	}
807
808	gsc_destroy(&rtsc);
809	gsc_destroy(&lssc);
810
811	return (0);
812
813err_ret:
814	gsc_destroy(&rtsc);
815	gsc_destroy(&lssc);
816	return (-1);
817}
818
819static int
820check_commit_hfsc(int dev, int opts, struct pf_altq *pa)
821{
822	struct pf_altq	*altq, *def = NULL;
823	int		 default_class;
824	int		 error = 0;
825
826	/* check if hfsc has one default queue for this interface */
827	default_class = 0;
828	TAILQ_FOREACH(altq, &altqs, entries) {
829		if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
830			continue;
831		if (altq->qname[0] == 0)  /* this is for interface */
832			continue;
833		if (altq->parent[0] == 0)  /* dummy root */
834			continue;
835		if (altq->pq_u.hfsc_opts.flags & HFCF_DEFAULTCLASS) {
836			default_class++;
837			def = altq;
838		}
839	}
840	if (default_class != 1) {
841		warnx("should have one default queue on %s", pa->ifname);
842		return (1);
843	}
844	/* make sure the default queue is a leaf */
845	TAILQ_FOREACH(altq, &altqs, entries) {
846		if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
847			continue;
848		if (altq->qname[0] == 0)  /* this is for interface */
849			continue;
850		if (strncmp(altq->parent, def->qname, PF_QNAME_SIZE) == 0) {
851			warnx("default queue is not a leaf");
852			error++;
853		}
854	}
855	return (error);
856}
857
858static int
859print_hfsc_opts(const struct pf_altq *a, const struct node_queue_opt *qopts)
860{
861	const struct hfsc_opts		*opts;
862	const struct node_hfsc_sc	*rtsc, *lssc, *ulsc;
863
864	opts = &a->pq_u.hfsc_opts;
865	if (qopts == NULL)
866		rtsc = lssc = ulsc = NULL;
867	else {
868		rtsc = &qopts->data.hfsc_opts.realtime;
869		lssc = &qopts->data.hfsc_opts.linkshare;
870		ulsc = &qopts->data.hfsc_opts.upperlimit;
871	}
872
873	if (opts->flags || opts->rtsc_m2 != 0 || opts->ulsc_m2 != 0 ||
874	    (opts->lssc_m2 != 0 && (opts->lssc_m2 != a->bandwidth ||
875	    opts->lssc_d != 0))) {
876		printf("hfsc(");
877		if (opts->flags & HFCF_RED)
878			printf(" red");
879		if (opts->flags & HFCF_ECN)
880			printf(" ecn");
881		if (opts->flags & HFCF_RIO)
882			printf(" rio");
883		if (opts->flags & HFCF_CLEARDSCP)
884			printf(" cleardscp");
885		if (opts->flags & HFCF_DEFAULTCLASS)
886			printf(" default");
887		if (opts->rtsc_m2 != 0)
888			print_hfsc_sc("realtime", opts->rtsc_m1, opts->rtsc_d,
889			    opts->rtsc_m2, rtsc);
890		if (opts->lssc_m2 != 0 && (opts->lssc_m2 != a->bandwidth ||
891		    opts->lssc_d != 0))
892			print_hfsc_sc("linkshare", opts->lssc_m1, opts->lssc_d,
893			    opts->lssc_m2, lssc);
894		if (opts->ulsc_m2 != 0)
895			print_hfsc_sc("upperlimit", opts->ulsc_m1, opts->ulsc_d,
896			    opts->ulsc_m2, ulsc);
897		printf(" ) ");
898
899		return (1);
900	} else
901		return (0);
902}
903
904/*
905 * admission control using generalized service curve
906 */
907#ifndef INFINITY
908#define	INFINITY	HUGE_VAL  /* positive infinity defined in <math.h> */
909#endif
910
911/* add a new service curve to a generalized service curve */
912static void
913gsc_add_sc(struct gen_sc *gsc, struct service_curve *sc)
914{
915	if (is_sc_null(sc))
916		return;
917	if (sc->d != 0)
918		gsc_add_seg(gsc, 0.0, 0.0, (double)sc->d, (double)sc->m1);
919	gsc_add_seg(gsc, (double)sc->d, 0.0, INFINITY, (double)sc->m2);
920}
921
922/*
923 * check whether all points of a generalized service curve have
924 * their y-coordinates no larger than a given two-piece linear
925 * service curve.
926 */
927static int
928is_gsc_under_sc(struct gen_sc *gsc, struct service_curve *sc)
929{
930	struct segment	*s, *last, *end;
931	double		 y;
932
933	if (is_sc_null(sc)) {
934		if (LIST_EMPTY(gsc))
935			return (1);
936		LIST_FOREACH(s, gsc, _next) {
937			if (s->m != 0)
938				return (0);
939		}
940		return (1);
941	}
942	/*
943	 * gsc has a dummy entry at the end with x = INFINITY.
944	 * loop through up to this dummy entry.
945	 */
946	end = gsc_getentry(gsc, INFINITY);
947	if (end == NULL)
948		return (1);
949	last = NULL;
950	for (s = LIST_FIRST(gsc); s != end; s = LIST_NEXT(s, _next)) {
951		if (s->y > sc_x2y(sc, s->x))
952			return (0);
953		last = s;
954	}
955	/* last now holds the real last segment */
956	if (last == NULL)
957		return (1);
958	if (last->m > sc->m2)
959		return (0);
960	if (last->x < sc->d && last->m > sc->m1) {
961		y = last->y + (sc->d - last->x) * last->m;
962		if (y > sc_x2y(sc, sc->d))
963			return (0);
964	}
965	return (1);
966}
967
968static void
969gsc_destroy(struct gen_sc *gsc)
970{
971	struct segment	*s;
972
973	while ((s = LIST_FIRST(gsc)) != NULL) {
974		LIST_REMOVE(s, _next);
975		free(s);
976	}
977}
978
979/*
980 * return a segment entry starting at x.
981 * if gsc has no entry starting at x, a new entry is created at x.
982 */
983static struct segment *
984gsc_getentry(struct gen_sc *gsc, double x)
985{
986	struct segment	*new, *prev, *s;
987
988	prev = NULL;
989	LIST_FOREACH(s, gsc, _next) {
990		if (s->x == x)
991			return (s);	/* matching entry found */
992		else if (s->x < x)
993			prev = s;
994		else
995			break;
996	}
997
998	/* we have to create a new entry */
999	if ((new = calloc(1, sizeof(struct segment))) == NULL)
1000		return (NULL);
1001
1002	new->x = x;
1003	if (x == INFINITY || s == NULL)
1004		new->d = 0;
1005	else if (s->x == INFINITY)
1006		new->d = INFINITY;
1007	else
1008		new->d = s->x - x;
1009	if (prev == NULL) {
1010		/* insert the new entry at the head of the list */
1011		new->y = 0;
1012		new->m = 0;
1013		LIST_INSERT_HEAD(gsc, new, _next);
1014	} else {
1015		/*
1016		 * the start point intersects with the segment pointed by
1017		 * prev.  divide prev into 2 segments
1018		 */
1019		if (x == INFINITY) {
1020			prev->d = INFINITY;
1021			if (prev->m == 0)
1022				new->y = prev->y;
1023			else
1024				new->y = INFINITY;
1025		} else {
1026			prev->d = x - prev->x;
1027			new->y = prev->d * prev->m + prev->y;
1028		}
1029		new->m = prev->m;
1030		LIST_INSERT_AFTER(prev, new, _next);
1031	}
1032	return (new);
1033}
1034
1035/* add a segment to a generalized service curve */
1036static int
1037gsc_add_seg(struct gen_sc *gsc, double x, double y, double d, double m)
1038{
1039	struct segment	*start, *end, *s;
1040	double		 x2;
1041
1042	if (d == INFINITY)
1043		x2 = INFINITY;
1044	else
1045		x2 = x + d;
1046	start = gsc_getentry(gsc, x);
1047	end = gsc_getentry(gsc, x2);
1048	if (start == NULL || end == NULL)
1049		return (-1);
1050
1051	for (s = start; s != end; s = LIST_NEXT(s, _next)) {
1052		s->m += m;
1053		s->y += y + (s->x - x) * m;
1054	}
1055
1056	end = gsc_getentry(gsc, INFINITY);
1057	for (; s != end; s = LIST_NEXT(s, _next)) {
1058		s->y += m * d;
1059	}
1060
1061	return (0);
1062}
1063
1064/* get y-projection of a service curve */
1065static double
1066sc_x2y(struct service_curve *sc, double x)
1067{
1068	double	y;
1069
1070	if (x <= (double)sc->d)
1071		/* y belongs to the 1st segment */
1072		y = x * (double)sc->m1;
1073	else
1074		/* y belongs to the 2nd segment */
1075		y = (double)sc->d * (double)sc->m1
1076			+ (x - (double)sc->d) * (double)sc->m2;
1077	return (y);
1078}
1079
1080/*
1081 * misc utilities
1082 */
1083#define	R2S_BUFS	8
1084#define	RATESTR_MAX	16
1085
1086char *
1087rate2str(double rate)
1088{
1089	char		*buf;
1090	static char	 r2sbuf[R2S_BUFS][RATESTR_MAX];  /* ring bufer */
1091	static int	 idx = 0;
1092	int		 i;
1093	static const char unit[] = " KMG";
1094
1095	buf = r2sbuf[idx++];
1096	if (idx == R2S_BUFS)
1097		idx = 0;
1098
1099	for (i = 0; rate >= 1000 && i <= 3; i++)
1100		rate /= 1000;
1101
1102	if ((int)(rate * 100) % 100)
1103		snprintf(buf, RATESTR_MAX, "%.2f%cb", rate, unit[i]);
1104	else
1105		snprintf(buf, RATESTR_MAX, "%d%cb", (int)rate, unit[i]);
1106
1107	return (buf);
1108}
1109
1110#ifdef __FreeBSD__
1111/*
1112 * XXX
1113 * FreeBSD does not have SIOCGIFDATA.
1114 * To emulate this, DIOCGIFSPEED ioctl added to pf.
1115 */
1116u_int32_t
1117getifspeed(int pfdev, char *ifname)
1118{
1119	struct pf_ifspeed io;
1120
1121	bzero(&io, sizeof io);
1122	if (strlcpy(io.ifname, ifname, IFNAMSIZ) >=
1123	    sizeof(io.ifname))
1124		errx(1, "getifspeed: strlcpy");
1125	if (ioctl(pfdev, DIOCGIFSPEED, &io) == -1)
1126		err(1, "DIOCGIFSPEED");
1127	return ((u_int32_t)io.baudrate);
1128}
1129#else
1130u_int32_t
1131getifspeed(char *ifname)
1132{
1133	int		s;
1134	struct ifreq	ifr;
1135	struct if_data	ifrdat;
1136
1137	if ((s = socket(AF_INET, SOCK_DGRAM, 0)) < 0)
1138		err(1, "socket");
1139	bzero(&ifr, sizeof(ifr));
1140	if (strlcpy(ifr.ifr_name, ifname, sizeof(ifr.ifr_name)) >=
1141	    sizeof(ifr.ifr_name))
1142		errx(1, "getifspeed: strlcpy");
1143	ifr.ifr_data = (caddr_t)&ifrdat;
1144	if (ioctl(s, SIOCGIFDATA, (caddr_t)&ifr) == -1)
1145		err(1, "SIOCGIFDATA");
1146	if (shutdown(s, SHUT_RDWR) == -1)
1147		err(1, "shutdown");
1148	if (close(s))
1149		err(1, "close");
1150	return ((u_int32_t)ifrdat.ifi_baudrate);
1151}
1152#endif
1153
1154u_long
1155getifmtu(char *ifname)
1156{
1157	int		s;
1158	struct ifreq	ifr;
1159
1160	if ((s = socket(AF_INET, SOCK_DGRAM, 0)) < 0)
1161		err(1, "socket");
1162	bzero(&ifr, sizeof(ifr));
1163	if (strlcpy(ifr.ifr_name, ifname, sizeof(ifr.ifr_name)) >=
1164	    sizeof(ifr.ifr_name))
1165		errx(1, "getifmtu: strlcpy");
1166	if (ioctl(s, SIOCGIFMTU, (caddr_t)&ifr) == -1)
1167		err(1, "SIOCGIFMTU");
1168	if (shutdown(s, SHUT_RDWR) == -1)
1169		err(1, "shutdown");
1170	if (close(s))
1171		err(1, "close");
1172	if (ifr.ifr_mtu > 0)
1173		return (ifr.ifr_mtu);
1174	else {
1175		warnx("could not get mtu for %s, assuming 1500", ifname);
1176		return (1500);
1177	}
1178}
1179
1180int
1181eval_queue_opts(struct pf_altq *pa, struct node_queue_opt *opts,
1182    u_int32_t ref_bw)
1183{
1184	int	errors = 0;
1185
1186	switch (pa->scheduler) {
1187	case ALTQT_CBQ:
1188		pa->pq_u.cbq_opts = opts->data.cbq_opts;
1189		break;
1190	case ALTQT_PRIQ:
1191		pa->pq_u.priq_opts = opts->data.priq_opts;
1192		break;
1193	case ALTQT_HFSC:
1194		pa->pq_u.hfsc_opts.flags = opts->data.hfsc_opts.flags;
1195		if (opts->data.hfsc_opts.linkshare.used) {
1196			pa->pq_u.hfsc_opts.lssc_m1 =
1197			    eval_bwspec(&opts->data.hfsc_opts.linkshare.m1,
1198			    ref_bw);
1199			pa->pq_u.hfsc_opts.lssc_m2 =
1200			    eval_bwspec(&opts->data.hfsc_opts.linkshare.m2,
1201			    ref_bw);
1202			pa->pq_u.hfsc_opts.lssc_d =
1203			    opts->data.hfsc_opts.linkshare.d;
1204		}
1205		if (opts->data.hfsc_opts.realtime.used) {
1206			pa->pq_u.hfsc_opts.rtsc_m1 =
1207			    eval_bwspec(&opts->data.hfsc_opts.realtime.m1,
1208			    ref_bw);
1209			pa->pq_u.hfsc_opts.rtsc_m2 =
1210			    eval_bwspec(&opts->data.hfsc_opts.realtime.m2,
1211			    ref_bw);
1212			pa->pq_u.hfsc_opts.rtsc_d =
1213			    opts->data.hfsc_opts.realtime.d;
1214		}
1215		if (opts->data.hfsc_opts.upperlimit.used) {
1216			pa->pq_u.hfsc_opts.ulsc_m1 =
1217			    eval_bwspec(&opts->data.hfsc_opts.upperlimit.m1,
1218			    ref_bw);
1219			pa->pq_u.hfsc_opts.ulsc_m2 =
1220			    eval_bwspec(&opts->data.hfsc_opts.upperlimit.m2,
1221			    ref_bw);
1222			pa->pq_u.hfsc_opts.ulsc_d =
1223			    opts->data.hfsc_opts.upperlimit.d;
1224		}
1225		break;
1226	default:
1227		warnx("eval_queue_opts: unknown scheduler type %u",
1228		    opts->qtype);
1229		errors++;
1230		break;
1231	}
1232
1233	return (errors);
1234}
1235
1236u_int32_t
1237eval_bwspec(struct node_queue_bw *bw, u_int32_t ref_bw)
1238{
1239	if (bw->bw_absolute > 0)
1240		return (bw->bw_absolute);
1241
1242	if (bw->bw_percent > 0)
1243		return (ref_bw / 100 * bw->bw_percent);
1244
1245	return (0);
1246}
1247
1248void
1249print_hfsc_sc(const char *scname, u_int m1, u_int d, u_int m2,
1250    const struct node_hfsc_sc *sc)
1251{
1252	printf(" %s", scname);
1253
1254	if (d != 0) {
1255		printf("(");
1256		if (sc != NULL && sc->m1.bw_percent > 0)
1257			printf("%u%%", sc->m1.bw_percent);
1258		else
1259			printf("%s", rate2str((double)m1));
1260		printf(" %u", d);
1261	}
1262
1263	if (sc != NULL && sc->m2.bw_percent > 0)
1264		printf(" %u%%", sc->m2.bw_percent);
1265	else
1266		printf(" %s", rate2str((double)m2));
1267
1268	if (d != 0)
1269		printf(")");
1270}
1271