1 /* $OpenBSD: pfctl_altq.c,v 1.94 2008/07/25 17:43:44 martynas Exp $ */
5 * Sony Computer Science Laboratories Inc.
6 * Copyright (c) 2002, 2003 Henning Brauer <henning@openbsd.org>
8 * Permission to use, copy, modify, and distribute this software for any
9 * purpose with or without fee is hereby granted, provided that the above
10 * copyright notice and this permission notice appear in all copies.
12 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
21 #define _KERNEL_STRUCTURES
22 #include <sys/param.h>
23 #include <sys/ioctl.h>
24 #include <sys/socket.h>
25 #include <sys/sysctl.h>
28 #include <net/if_mib.h>
29 #include <netinet/in.h>
30 #include <net/pf/pfvar.h>
41 #include <net/altq/altq.h>
42 #include <net/altq/altq_cbq.h>
43 #include <net/altq/altq_priq.h>
44 #include <net/altq/altq_hfsc.h>
45 #include <net/altq/altq_fairq.h>
47 #include "pfctl_parser.h"
50 #define is_sc_null(sc) (((sc) == NULL) || ((sc)->m1 == 0 && (sc)->m2 == 0))
52 TAILQ_HEAD(altqs, pf_altq) altqs = TAILQ_HEAD_INITIALIZER(altqs);
53 LIST_HEAD(gen_sc, segment) rtsc, lssc;
55 struct pf_altq *qname_to_pfaltq(const char *, const char *);
56 u_int32_t qname_to_qid(const char *);
58 static int eval_pfqueue_cbq(struct pfctl *, struct pf_altq *);
59 static int cbq_compute_idletime(struct pfctl *, struct pf_altq *);
60 static int check_commit_cbq(int, int, struct pf_altq *);
61 static int print_cbq_opts(const struct pf_altq *);
63 static int eval_pfqueue_priq(struct pfctl *, struct pf_altq *);
64 static int check_commit_priq(int, int, struct pf_altq *);
65 static int print_priq_opts(const struct pf_altq *);
67 static int eval_pfqueue_hfsc(struct pfctl *, struct pf_altq *);
68 static int check_commit_hfsc(int, int, struct pf_altq *);
69 static int print_hfsc_opts(const struct pf_altq *,
70 const struct node_queue_opt *);
72 static int eval_pfqueue_fairq(struct pfctl *, struct pf_altq *);
73 static int print_fairq_opts(const struct pf_altq *,
74 const struct node_queue_opt *);
75 static int check_commit_fairq(int, int, struct pf_altq *);
77 static void gsc_add_sc(struct gen_sc *, struct service_curve *);
78 static int is_gsc_under_sc(struct gen_sc *,
79 struct service_curve *);
80 static void gsc_destroy(struct gen_sc *);
81 static struct segment *gsc_getentry(struct gen_sc *, double);
82 static int gsc_add_seg(struct gen_sc *, double, double, double,
84 static double sc_x2y(struct service_curve *, double);
86 u_int32_t getifspeed(const char *);
87 u_long getifmtu(char *);
88 int eval_queue_opts(struct pf_altq *, struct node_queue_opt *,
90 u_int32_t eval_bwspec(struct node_queue_bw *, u_int32_t);
91 void print_hfsc_sc(const char *, u_int, u_int, u_int,
92 const struct node_hfsc_sc *);
93 void print_fairq_sc(const char *, u_int, u_int, u_int,
94 const struct node_fairq_sc *);
97 pfaltq_store(struct pf_altq *a)
101 if ((altq = malloc(sizeof(*altq))) == NULL)
103 memcpy(altq, a, sizeof(struct pf_altq));
104 TAILQ_INSERT_TAIL(&altqs, altq, entries);
108 pfaltq_lookup(const char *ifname)
110 struct pf_altq *altq;
112 TAILQ_FOREACH(altq, &altqs, entries) {
113 if (strncmp(ifname, altq->ifname, IFNAMSIZ) == 0 &&
121 qname_to_pfaltq(const char *qname, const char *ifname)
123 struct pf_altq *altq;
125 TAILQ_FOREACH(altq, &altqs, entries) {
126 if (strncmp(ifname, altq->ifname, IFNAMSIZ) == 0 &&
127 strncmp(qname, altq->qname, PF_QNAME_SIZE) == 0)
134 qname_to_qid(const char *qname)
136 struct pf_altq *altq;
139 * We guarantee that same named queues on different interfaces
140 * have the same qid, so we do NOT need to limit matching on
144 TAILQ_FOREACH(altq, &altqs, entries) {
145 if (strncmp(qname, altq->qname, PF_QNAME_SIZE) == 0)
152 print_altq(const struct pf_altq *a, unsigned int level,
153 struct node_queue_bw *bw, struct node_queue_opt *qopts)
155 if (a->qname[0] != 0) {
156 print_queue(a, level, bw, 1, qopts);
160 printf("altq on %s ", a->ifname);
162 switch (a->scheduler) {
164 if (!print_cbq_opts(a))
168 if (!print_priq_opts(a))
172 if (!print_hfsc_opts(a, qopts))
176 if (!print_fairq_opts(a, qopts))
181 if (bw != NULL && bw->bw_percent > 0) {
182 if (bw->bw_percent < 100)
183 printf("bandwidth %u%% ", bw->bw_percent);
185 printf("bandwidth %s ", rate2str((double)a->ifbandwidth));
187 if (a->qlimit != DEFAULT_QLIMIT)
188 printf("qlimit %u ", a->qlimit);
189 printf("tbrsize %u ", a->tbrsize);
193 print_queue(const struct pf_altq *a, unsigned int level,
194 struct node_queue_bw *bw, int print_interface,
195 struct node_queue_opt *qopts)
200 for (i = 0; i < level; ++i)
202 printf("%s ", a->qname);
204 printf("on %s ", a->ifname);
205 if (a->scheduler == ALTQT_CBQ || a->scheduler == ALTQT_HFSC ||
206 a->scheduler == ALTQT_FAIRQ) {
207 if (bw != NULL && bw->bw_percent > 0) {
208 if (bw->bw_percent < 100)
209 printf("bandwidth %u%% ", bw->bw_percent);
211 printf("bandwidth %s ", rate2str((double)a->bandwidth));
213 if (a->priority != DEFAULT_PRIORITY)
214 printf("priority %u ", a->priority);
215 if (a->qlimit != DEFAULT_QLIMIT)
216 printf("qlimit %u ", a->qlimit);
217 switch (a->scheduler) {
225 print_hfsc_opts(a, qopts);
228 print_fairq_opts(a, qopts);
234 * eval_pfaltq computes the discipline parameters.
237 eval_pfaltq(struct pfctl *pf __unused, struct pf_altq *pa, struct node_queue_bw *bw,
238 struct node_queue_opt *opts)
240 u_int rate, size, errors = 0;
242 if (bw->bw_absolute > 0)
243 pa->ifbandwidth = bw->bw_absolute;
245 if ((rate = getifspeed(pa->ifname)) == 0) {
246 fprintf(stderr, "interface %s does not know its bandwidth, "
247 "please specify an absolute bandwidth\n",
250 } else if ((pa->ifbandwidth = eval_bwspec(bw, rate)) == 0)
251 pa->ifbandwidth = rate;
253 errors += eval_queue_opts(pa, opts, pa->ifbandwidth);
255 /* if tbrsize is not specified, use heuristics */
256 if (pa->tbrsize == 0) {
257 rate = pa->ifbandwidth;
258 if (rate <= 1 * 1000 * 1000)
260 else if (rate <= 10 * 1000 * 1000)
262 else if (rate <= 200 * 1000 * 1000)
266 size = size * getifmtu(pa->ifname);
275 * check_commit_altq does consistency check for each interface
278 check_commit_altq(int dev, int opts)
280 struct pf_altq *altq;
283 /* call the discipline check for each interface. */
284 TAILQ_FOREACH(altq, &altqs, entries) {
285 if (altq->qname[0] == 0) {
286 switch (altq->scheduler) {
288 error = check_commit_cbq(dev, opts, altq);
291 error = check_commit_priq(dev, opts, altq);
294 error = check_commit_hfsc(dev, opts, altq);
297 error = check_commit_fairq(dev, opts, altq);
308 * eval_pfqueue computes the queue parameters.
311 eval_pfqueue(struct pfctl *pf, struct pf_altq *pa, struct node_queue_bw *bw,
312 struct node_queue_opt *opts)
314 /* should be merged with expand_queue */
315 struct pf_altq *if_pa, *parent, *altq;
319 /* find the corresponding interface and copy fields used by queues */
320 if ((if_pa = pfaltq_lookup(pa->ifname)) == NULL) {
321 fprintf(stderr, "altq not defined on %s\n", pa->ifname);
324 pa->scheduler = if_pa->scheduler;
325 pa->ifbandwidth = if_pa->ifbandwidth;
327 if (qname_to_pfaltq(pa->qname, pa->ifname) != NULL) {
328 fprintf(stderr, "queue %s already exists on interface %s\n",
329 pa->qname, pa->ifname);
332 pa->qid = qname_to_qid(pa->qname);
335 if (pa->parent[0] != 0) {
336 parent = qname_to_pfaltq(pa->parent, pa->ifname);
337 if (parent == NULL) {
338 fprintf(stderr, "parent %s not found for %s\n",
339 pa->parent, pa->qname);
342 pa->parent_qid = parent->qid;
345 pa->qlimit = DEFAULT_QLIMIT;
347 if (pa->scheduler == ALTQT_CBQ || pa->scheduler == ALTQT_HFSC ||
348 pa->scheduler == ALTQT_FAIRQ) {
349 pa->bandwidth = eval_bwspec(bw,
350 parent == NULL ? 0 : parent->bandwidth);
352 if (pa->bandwidth > pa->ifbandwidth) {
353 fprintf(stderr, "bandwidth for %s higher than "
354 "interface\n", pa->qname);
357 /* check the sum of the child bandwidth is under parent's */
358 if (parent != NULL) {
359 if (pa->bandwidth > parent->bandwidth) {
360 warnx("bandwidth for %s higher than parent",
365 TAILQ_FOREACH(altq, &altqs, entries) {
366 if (strncmp(altq->ifname, pa->ifname,
368 altq->qname[0] != 0 &&
369 strncmp(altq->parent, pa->parent,
371 bwsum += altq->bandwidth;
373 bwsum += pa->bandwidth;
374 if (bwsum > parent->bandwidth) {
375 warnx("the sum of the child bandwidth higher"
376 " than parent \"%s\"", parent->qname);
381 if (eval_queue_opts(pa, opts, parent == NULL? 0 : parent->bandwidth))
384 switch (pa->scheduler) {
386 error = eval_pfqueue_cbq(pf, pa);
389 error = eval_pfqueue_priq(pf, pa);
392 error = eval_pfqueue_hfsc(pf, pa);
395 error = eval_pfqueue_fairq(pf, pa);
404 * CBQ support functions
406 #define RM_FILTER_GAIN 5 /* log2 of gain, e.g., 5 => 31/32 */
407 #define RM_NS_PER_SEC (1000000000)
410 eval_pfqueue_cbq(struct pfctl *pf, struct pf_altq *pa)
412 struct cbq_opts *opts;
415 if (pa->priority >= CBQ_MAXPRI) {
416 warnx("priority out of range: max %d", CBQ_MAXPRI - 1);
420 ifmtu = getifmtu(pa->ifname);
421 opts = &pa->pq_u.cbq_opts;
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;
434 if (opts->pktsize > opts->maxpktsize)
435 opts->pktsize = opts->maxpktsize;
437 if (pa->parent[0] == 0)
438 opts->flags |= (CBQCLF_ROOTCLASS | CBQCLF_WRR);
440 cbq_compute_idletime(pf, pa);
445 * compute ns_per_byte, maxidle, minidle, and offtime
448 cbq_compute_idletime(struct pfctl *pf, struct pf_altq *pa)
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;
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;
461 if (pa->bandwidth == 0)
462 f = 0.0001; /* small enough? */
464 f = ((double) pa->bandwidth / (double) pa->ifbandwidth);
466 nsPerByte = ifnsPerByte / f;
467 ptime = (double)opts->pktsize * ifnsPerByte;
468 cptime = ptime * (1.0 - f) / f;
470 if (nsPerByte * (double)opts->maxpktsize > (double)INT_MAX) {
472 * this causes integer overflow in kernel!
473 * (bandwidth < 6Kbps when max_pkt_size=1500)
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",
482 nsPerByte = (double)(INT_MAX / opts->maxpktsize);
485 if (maxburst == 0) { /* use default */
486 if (cptime > 10.0 * 1000000)
491 if (minburst == 0) /* use default */
493 if (minburst > maxburst)
496 z = (double)(1 << RM_FILTER_GAIN);
498 gton = pow(g, (double)maxburst);
499 gtom = pow(g, (double)(minburst-1));
500 maxidle = ((1.0 / f - 1.0) * ((1.0 - gton) / gton));
501 maxidle_s = (1.0 - g);
502 if (maxidle > maxidle_s)
503 maxidle = ptime * maxidle;
505 maxidle = ptime * maxidle_s;
506 offtime = cptime * (1.0 + 1.0/(1.0 - g) * (1.0 - gtom) / gtom);
507 minidle = -((double)opts->maxpktsize * (double)nsPerByte);
509 /* scale parameters */
510 maxidle = ((maxidle * 8.0) / nsPerByte) *
511 pow(2.0, (double)RM_FILTER_GAIN);
512 offtime = (offtime * 8.0) / nsPerByte *
513 pow(2.0, (double)RM_FILTER_GAIN);
514 minidle = ((minidle * 8.0) / nsPerByte) *
515 pow(2.0, (double)RM_FILTER_GAIN);
517 maxidle = maxidle / 1000.0;
518 offtime = offtime / 1000.0;
519 minidle = minidle / 1000.0;
521 opts->minburst = minburst;
522 opts->maxburst = maxburst;
523 opts->ns_per_byte = (u_int)nsPerByte;
524 opts->maxidle = (u_int)fabs(maxidle);
525 opts->minidle = (int)minidle;
526 opts->offtime = (u_int)fabs(offtime);
532 check_commit_cbq(int dev __unused, int opts __unused, struct pf_altq *pa)
534 struct pf_altq *altq;
535 int root_class, default_class;
539 * check if cbq has one root queue and one default queue
542 root_class = default_class = 0;
543 TAILQ_FOREACH(altq, &altqs, entries) {
544 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
546 if (altq->qname[0] == 0) /* this is for interface */
548 if (altq->pq_u.cbq_opts.flags & CBQCLF_ROOTCLASS)
550 if (altq->pq_u.cbq_opts.flags & CBQCLF_DEFCLASS)
553 if (root_class != 1) {
554 warnx("should have one root queue on %s", pa->ifname);
557 if (default_class != 1) {
558 warnx("should have one default queue on %s", pa->ifname);
565 print_cbq_opts(const struct pf_altq *a)
567 const struct cbq_opts *opts;
569 opts = &a->pq_u.cbq_opts;
572 if (opts->flags & CBQCLF_RED)
574 if (opts->flags & CBQCLF_ECN)
576 if (opts->flags & CBQCLF_RIO)
578 if (opts->flags & CBQCLF_CLEARDSCP)
579 printf(" cleardscp");
580 if (opts->flags & CBQCLF_BORROW)
582 if (opts->flags & CBQCLF_WRR)
584 if (opts->flags & CBQCLF_EFFICIENT)
585 printf(" efficient");
586 if (opts->flags & CBQCLF_ROOTCLASS)
588 if (opts->flags & CBQCLF_DEFCLASS)
598 * PRIQ support functions
601 eval_pfqueue_priq(struct pfctl *pf __unused, struct pf_altq *pa)
603 struct pf_altq *altq;
605 if (pa->priority >= PRIQ_MAXPRI) {
606 warnx("priority out of range: max %d", PRIQ_MAXPRI - 1);
609 /* the priority should be unique for the interface */
610 TAILQ_FOREACH(altq, &altqs, entries) {
611 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) == 0 &&
612 altq->qname[0] != 0 && altq->priority == pa->priority) {
613 warnx("%s and %s have the same priority",
614 altq->qname, pa->qname);
623 check_commit_priq(int dev __unused, int opts __unused, struct pf_altq *pa)
625 struct pf_altq *altq;
630 * check if priq has one default class for this interface
633 TAILQ_FOREACH(altq, &altqs, entries) {
634 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
636 if (altq->qname[0] == 0) /* this is for interface */
638 if (altq->pq_u.priq_opts.flags & PRCF_DEFAULTCLASS)
641 if (default_class != 1) {
642 warnx("should have one default queue on %s", pa->ifname);
649 print_priq_opts(const struct pf_altq *a)
651 const struct priq_opts *opts;
653 opts = &a->pq_u.priq_opts;
657 if (opts->flags & PRCF_RED)
659 if (opts->flags & PRCF_ECN)
661 if (opts->flags & PRCF_RIO)
663 if (opts->flags & PRCF_CLEARDSCP)
664 printf(" cleardscp");
665 if (opts->flags & PRCF_DEFAULTCLASS)
675 * HFSC support functions
678 eval_pfqueue_hfsc(struct pfctl *pf __unused, struct pf_altq *pa)
680 struct pf_altq *altq, *parent;
681 struct hfsc_opts *opts;
682 struct service_curve sc;
684 opts = &pa->pq_u.hfsc_opts;
686 if (pa->parent[0] == 0) {
688 opts->lssc_m1 = pa->ifbandwidth;
689 opts->lssc_m2 = pa->ifbandwidth;
697 /* if link_share is not specified, use bandwidth */
698 if (opts->lssc_m2 == 0)
699 opts->lssc_m2 = pa->bandwidth;
701 if ((opts->rtsc_m1 > 0 && opts->rtsc_m2 == 0) ||
702 (opts->lssc_m1 > 0 && opts->lssc_m2 == 0) ||
703 (opts->ulsc_m1 > 0 && opts->ulsc_m2 == 0)) {
704 warnx("m2 is zero for %s", pa->qname);
708 if ((opts->rtsc_m1 < opts->rtsc_m2 && opts->rtsc_m1 != 0) ||
709 (opts->lssc_m1 < opts->lssc_m2 && opts->lssc_m1 != 0) ||
710 (opts->ulsc_m1 < opts->ulsc_m2 && opts->ulsc_m1 != 0)) {
711 warnx("m1 must be zero for convex curve: %s", pa->qname);
717 * for the real-time service curve, the sum of the service curves
718 * should not exceed 80% of the interface bandwidth. 20% is reserved
719 * not to over-commit the actual interface bandwidth.
720 * for the linkshare service curve, the sum of the child service
721 * curve should not exceed the parent service curve.
722 * for the upper-limit service curve, the assigned bandwidth should
723 * be smaller than the interface bandwidth, and the upper-limit should
724 * be larger than the real-time service curve when both are defined.
726 parent = qname_to_pfaltq(pa->parent, pa->ifname);
728 errx(1, "parent %s not found for %s", pa->parent, pa->qname);
730 TAILQ_FOREACH(altq, &altqs, entries) {
731 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
733 if (altq->qname[0] == 0) /* this is for interface */
736 /* if the class has a real-time service curve, add it. */
737 if (opts->rtsc_m2 != 0 && altq->pq_u.hfsc_opts.rtsc_m2 != 0) {
738 sc.m1 = altq->pq_u.hfsc_opts.rtsc_m1;
739 sc.d = altq->pq_u.hfsc_opts.rtsc_d;
740 sc.m2 = altq->pq_u.hfsc_opts.rtsc_m2;
741 gsc_add_sc(&rtsc, &sc);
744 if (strncmp(altq->parent, pa->parent, PF_QNAME_SIZE) != 0)
747 /* if the class has a linkshare service curve, add it. */
748 if (opts->lssc_m2 != 0 && altq->pq_u.hfsc_opts.lssc_m2 != 0) {
749 sc.m1 = altq->pq_u.hfsc_opts.lssc_m1;
750 sc.d = altq->pq_u.hfsc_opts.lssc_d;
751 sc.m2 = altq->pq_u.hfsc_opts.lssc_m2;
752 gsc_add_sc(&lssc, &sc);
756 /* check the real-time service curve. reserve 20% of interface bw */
757 if (opts->rtsc_m2 != 0) {
758 /* add this queue to the sum */
759 sc.m1 = opts->rtsc_m1;
761 sc.m2 = opts->rtsc_m2;
762 gsc_add_sc(&rtsc, &sc);
763 /* compare the sum with 80% of the interface */
766 sc.m2 = pa->ifbandwidth / 100 * 80;
767 if (!is_gsc_under_sc(&rtsc, &sc)) {
768 warnx("real-time sc exceeds 80%% of the interface "
769 "bandwidth (%s)", rate2str((double)sc.m2));
774 /* check the linkshare service curve. */
775 if (opts->lssc_m2 != 0) {
776 /* add this queue to the child sum */
777 sc.m1 = opts->lssc_m1;
779 sc.m2 = opts->lssc_m2;
780 gsc_add_sc(&lssc, &sc);
781 /* compare the sum of the children with parent's sc */
782 sc.m1 = parent->pq_u.hfsc_opts.lssc_m1;
783 sc.d = parent->pq_u.hfsc_opts.lssc_d;
784 sc.m2 = parent->pq_u.hfsc_opts.lssc_m2;
785 if (!is_gsc_under_sc(&lssc, &sc)) {
786 warnx("linkshare sc exceeds parent's sc");
791 /* check the upper-limit service curve. */
792 if (opts->ulsc_m2 != 0) {
793 if (opts->ulsc_m1 > pa->ifbandwidth ||
794 opts->ulsc_m2 > pa->ifbandwidth) {
795 warnx("upper-limit larger than interface bandwidth");
798 if (opts->rtsc_m2 != 0 && opts->rtsc_m2 > opts->ulsc_m2) {
799 warnx("upper-limit sc smaller than real-time sc");
816 * FAIRQ support functions
819 eval_pfqueue_fairq(struct pfctl *pf __unused, struct pf_altq *pa)
821 struct pf_altq *altq, *parent;
822 struct fairq_opts *opts;
823 struct service_curve sc;
825 opts = &pa->pq_u.fairq_opts;
827 if (pa->parent[0] == 0) {
829 opts->lssc_m1 = pa->ifbandwidth;
830 opts->lssc_m2 = pa->ifbandwidth;
837 /* if link_share is not specified, use bandwidth */
838 if (opts->lssc_m2 == 0)
839 opts->lssc_m2 = pa->bandwidth;
843 * for the real-time service curve, the sum of the service curves
844 * should not exceed 80% of the interface bandwidth. 20% is reserved
845 * not to over-commit the actual interface bandwidth.
846 * for the link-sharing service curve, the sum of the child service
847 * curve should not exceed the parent service curve.
848 * for the upper-limit service curve, the assigned bandwidth should
849 * be smaller than the interface bandwidth, and the upper-limit should
850 * be larger than the real-time service curve when both are defined.
852 parent = qname_to_pfaltq(pa->parent, pa->ifname);
854 errx(1, "parent %s not found for %s", pa->parent, pa->qname);
856 TAILQ_FOREACH(altq, &altqs, entries) {
857 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
859 if (altq->qname[0] == 0) /* this is for interface */
862 if (strncmp(altq->parent, pa->parent, PF_QNAME_SIZE) != 0)
865 /* if the class has a link-sharing service curve, add it. */
866 if (opts->lssc_m2 != 0 && altq->pq_u.fairq_opts.lssc_m2 != 0) {
867 sc.m1 = altq->pq_u.fairq_opts.lssc_m1;
868 sc.d = altq->pq_u.fairq_opts.lssc_d;
869 sc.m2 = altq->pq_u.fairq_opts.lssc_m2;
870 gsc_add_sc(&lssc, &sc);
874 /* check the link-sharing service curve. */
875 if (opts->lssc_m2 != 0) {
876 sc.m1 = parent->pq_u.fairq_opts.lssc_m1;
877 sc.d = parent->pq_u.fairq_opts.lssc_d;
878 sc.m2 = parent->pq_u.fairq_opts.lssc_m2;
879 if (!is_gsc_under_sc(&lssc, &sc)) {
880 warnx("link-sharing sc exceeds parent's sc");
895 check_commit_hfsc(int dev __unused, int opts __unused, struct pf_altq *pa)
897 struct pf_altq *altq, *def = NULL;
901 /* check if hfsc has one default queue for this interface */
903 TAILQ_FOREACH(altq, &altqs, entries) {
904 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
906 if (altq->qname[0] == 0) /* this is for interface */
908 if (altq->parent[0] == 0) /* dummy root */
910 if (altq->pq_u.hfsc_opts.flags & HFCF_DEFAULTCLASS) {
915 if (default_class != 1) {
916 warnx("should have one default queue on %s", pa->ifname);
919 /* make sure the default queue is a leaf */
920 TAILQ_FOREACH(altq, &altqs, entries) {
921 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
923 if (altq->qname[0] == 0) /* this is for interface */
925 if (strncmp(altq->parent, def->qname, PF_QNAME_SIZE) == 0) {
926 warnx("default queue is not a leaf");
934 check_commit_fairq(int dev __unused, int opts __unused, struct pf_altq *pa)
936 struct pf_altq *altq, *def = NULL;
940 /* check if fairq has one default queue for this interface */
942 TAILQ_FOREACH(altq, &altqs, entries) {
943 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
945 if (altq->qname[0] == 0) /* this is for interface */
947 if (altq->pq_u.fairq_opts.flags & FARF_DEFAULTCLASS) {
952 if (default_class != 1) {
953 warnx("should have one default queue on %s", pa->ifname);
956 /* make sure the default queue is a leaf */
957 TAILQ_FOREACH(altq, &altqs, entries) {
958 if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0)
960 if (altq->qname[0] == 0) /* this is for interface */
962 if (strncmp(altq->parent, def->qname, PF_QNAME_SIZE) == 0) {
963 warnx("default queue is not a leaf");
971 print_hfsc_opts(const struct pf_altq *a, const struct node_queue_opt *qopts)
973 const struct hfsc_opts *opts;
974 const struct node_hfsc_sc *loc_rtsc, *loc_lssc, *ulsc;
976 opts = &a->pq_u.hfsc_opts;
978 loc_rtsc = loc_lssc = ulsc = NULL;
980 loc_rtsc = &qopts->data.hfsc_opts.realtime;
981 loc_lssc = &qopts->data.hfsc_opts.linkshare;
982 ulsc = &qopts->data.hfsc_opts.upperlimit;
985 if (opts->flags || opts->rtsc_m2 != 0 || opts->ulsc_m2 != 0 ||
986 (opts->lssc_m2 != 0 && (opts->lssc_m2 != a->bandwidth ||
987 opts->lssc_d != 0))) {
989 if (opts->flags & HFCF_RED)
991 if (opts->flags & HFCF_ECN)
993 if (opts->flags & HFCF_RIO)
995 if (opts->flags & HFCF_CLEARDSCP)
996 printf(" cleardscp");
997 if (opts->flags & HFCF_DEFAULTCLASS)
999 if (opts->rtsc_m2 != 0)
1000 print_hfsc_sc("realtime", opts->rtsc_m1, opts->rtsc_d,
1001 opts->rtsc_m2, loc_rtsc);
1002 if (opts->lssc_m2 != 0 && (opts->lssc_m2 != a->bandwidth ||
1004 print_hfsc_sc("linkshare", opts->lssc_m1, opts->lssc_d,
1005 opts->lssc_m2, loc_lssc);
1006 if (opts->ulsc_m2 != 0)
1007 print_hfsc_sc("upperlimit", opts->ulsc_m1, opts->ulsc_d,
1008 opts->ulsc_m2, ulsc);
1017 print_fairq_opts(const struct pf_altq *a, const struct node_queue_opt *qopts)
1019 const struct fairq_opts *opts;
1020 const struct node_fairq_sc *loc_lssc;
1022 opts = &a->pq_u.fairq_opts;
1026 loc_lssc = &qopts->data.fairq_opts.linkshare;
1029 (opts->lssc_m2 != 0 && (opts->lssc_m2 != a->bandwidth ||
1030 opts->lssc_d != 0))) {
1032 if (opts->flags & FARF_RED)
1034 if (opts->flags & FARF_ECN)
1036 if (opts->flags & FARF_RIO)
1038 if (opts->flags & FARF_CLEARDSCP)
1039 printf(" cleardscp");
1040 if (opts->flags & FARF_DEFAULTCLASS)
1042 if (opts->lssc_m2 != 0 && (opts->lssc_m2 != a->bandwidth ||
1044 print_fairq_sc("linkshare", opts->lssc_m1, opts->lssc_d,
1045 opts->lssc_m2, loc_lssc);
1054 * admission control using generalized service curve
1057 /* add a new service curve to a generalized service curve */
1059 gsc_add_sc(struct gen_sc *gsc, struct service_curve *sc)
1064 gsc_add_seg(gsc, 0.0, 0.0, (double)sc->d, (double)sc->m1);
1065 gsc_add_seg(gsc, (double)sc->d, 0.0, INFINITY, (double)sc->m2);
1069 * check whether all points of a generalized service curve have
1070 * their y-coordinates no larger than a given two-piece linear
1074 is_gsc_under_sc(struct gen_sc *gsc, struct service_curve *sc)
1076 struct segment *s, *last, *end;
1079 if (is_sc_null(sc)) {
1080 if (LIST_EMPTY(gsc))
1082 LIST_FOREACH(s, gsc, _next) {
1089 * gsc has a dummy entry at the end with x = INFINITY.
1090 * loop through up to this dummy entry.
1092 end = gsc_getentry(gsc, INFINITY);
1096 for (s = LIST_FIRST(gsc); s != end; s = LIST_NEXT(s, _next)) {
1097 if (s->y > sc_x2y(sc, s->x))
1101 /* last now holds the real last segment */
1104 if (last->m > sc->m2)
1106 if (last->x < sc->d && last->m > sc->m1) {
1107 y = last->y + (sc->d - last->x) * last->m;
1108 if (y > sc_x2y(sc, sc->d))
1115 gsc_destroy(struct gen_sc *gsc)
1119 while ((s = LIST_FIRST(gsc)) != NULL) {
1120 LIST_REMOVE(s, _next);
1126 * return a segment entry starting at x.
1127 * if gsc has no entry starting at x, a new entry is created at x.
1129 static struct segment *
1130 gsc_getentry(struct gen_sc *gsc, double x)
1132 struct segment *new, *prev, *s;
1135 LIST_FOREACH(s, gsc, _next) {
1137 return (s); /* matching entry found */
1144 /* we have to create a new entry */
1145 if ((new = calloc(1, sizeof(struct segment))) == NULL)
1149 if (x == INFINITY || s == NULL)
1151 else if (s->x == INFINITY)
1156 /* insert the new entry at the head of the list */
1159 LIST_INSERT_HEAD(gsc, new, _next);
1162 * the start point intersects with the segment pointed by
1163 * prev. divide prev into 2 segments
1165 if (x == INFINITY) {
1172 prev->d = x - prev->x;
1173 new->y = prev->d * prev->m + prev->y;
1176 LIST_INSERT_AFTER(prev, new, _next);
1181 /* add a segment to a generalized service curve */
1183 gsc_add_seg(struct gen_sc *gsc, double x, double y, double d, double m)
1185 struct segment *start, *end, *s;
1192 start = gsc_getentry(gsc, x);
1193 end = gsc_getentry(gsc, x2);
1194 if (start == NULL || end == NULL)
1197 for (s = start; s != end; s = LIST_NEXT(s, _next)) {
1199 s->y += y + (s->x - x) * m;
1202 end = gsc_getentry(gsc, INFINITY);
1203 for (; s != end; s = LIST_NEXT(s, _next)) {
1210 /* get y-projection of a service curve */
1212 sc_x2y(struct service_curve *sc, double x)
1216 if (x <= (double)sc->d)
1217 /* y belongs to the 1st segment */
1218 y = x * (double)sc->m1;
1220 /* y belongs to the 2nd segment */
1221 y = (double)sc->d * (double)sc->m1
1222 + (x - (double)sc->d) * (double)sc->m2;
1230 #define RATESTR_MAX 16
1233 rate2str(double rate)
1236 static char r2sbuf[R2S_BUFS][RATESTR_MAX]; /* ring bufer */
1239 static const char unit[] = " KMG";
1241 buf = r2sbuf[idx++];
1242 if (idx == R2S_BUFS)
1245 for (i = 0; rate >= 1000 && i <= 3; i++)
1248 if ((int)(rate * 100) % 100)
1249 snprintf(buf, RATESTR_MAX, "%.2f%cb", rate, unit[i]);
1251 snprintf(buf, RATESTR_MAX, "%d%cb", (int)rate, unit[i]);
1257 getifspeed(const char *ifname)
1261 struct ifmibdata data;
1271 if ((idx = (int)if_nametoindex(ifname)) == 0)
1272 err(1, "getifspeed: if_nametoindex");
1275 datalen = sizeof(data);
1276 if (sysctl(name, 6, &data, &datalen, NULL, 0))
1277 err(1, "getifspeed: sysctl");
1279 return(data.ifmd_data.ifi_baudrate);
1283 getifmtu(char *ifname)
1288 if ((s = socket(AF_INET, SOCK_DGRAM, 0)) < 0)
1290 bzero(&ifr, sizeof(ifr));
1291 if (strlcpy(ifr.ifr_name, ifname, sizeof(ifr.ifr_name)) >=
1292 sizeof(ifr.ifr_name))
1293 errx(1, "getifmtu: strlcpy");
1294 if (ioctl(s, SIOCGIFMTU, (caddr_t)&ifr) == -1)
1295 err(1, "SIOCGIFMTU");
1298 if (ifr.ifr_mtu > 0)
1299 return (ifr.ifr_mtu);
1301 warnx("could not get mtu for %s, assuming 1500", ifname);
1307 eval_queue_opts(struct pf_altq *pa, struct node_queue_opt *opts,
1312 switch (pa->scheduler) {
1314 pa->pq_u.cbq_opts = opts->data.cbq_opts;
1317 pa->pq_u.priq_opts = opts->data.priq_opts;
1320 pa->pq_u.hfsc_opts.flags = opts->data.hfsc_opts.flags;
1321 if (opts->data.hfsc_opts.linkshare.used) {
1322 pa->pq_u.hfsc_opts.lssc_m1 =
1323 eval_bwspec(&opts->data.hfsc_opts.linkshare.m1,
1325 pa->pq_u.hfsc_opts.lssc_m2 =
1326 eval_bwspec(&opts->data.hfsc_opts.linkshare.m2,
1328 pa->pq_u.hfsc_opts.lssc_d =
1329 opts->data.hfsc_opts.linkshare.d;
1331 if (opts->data.hfsc_opts.realtime.used) {
1332 pa->pq_u.hfsc_opts.rtsc_m1 =
1333 eval_bwspec(&opts->data.hfsc_opts.realtime.m1,
1335 pa->pq_u.hfsc_opts.rtsc_m2 =
1336 eval_bwspec(&opts->data.hfsc_opts.realtime.m2,
1338 pa->pq_u.hfsc_opts.rtsc_d =
1339 opts->data.hfsc_opts.realtime.d;
1341 if (opts->data.hfsc_opts.upperlimit.used) {
1342 pa->pq_u.hfsc_opts.ulsc_m1 =
1343 eval_bwspec(&opts->data.hfsc_opts.upperlimit.m1,
1345 pa->pq_u.hfsc_opts.ulsc_m2 =
1346 eval_bwspec(&opts->data.hfsc_opts.upperlimit.m2,
1348 pa->pq_u.hfsc_opts.ulsc_d =
1349 opts->data.hfsc_opts.upperlimit.d;
1353 pa->pq_u.fairq_opts.flags = opts->data.fairq_opts.flags;
1354 pa->pq_u.fairq_opts.nbuckets = opts->data.fairq_opts.nbuckets;
1355 pa->pq_u.fairq_opts.hogs_m1 =
1356 eval_bwspec(&opts->data.fairq_opts.hogs_bw, ref_bw);
1358 if (opts->data.fairq_opts.linkshare.used) {
1359 pa->pq_u.fairq_opts.lssc_m1 =
1360 eval_bwspec(&opts->data.fairq_opts.linkshare.m1,
1362 pa->pq_u.fairq_opts.lssc_m2 =
1363 eval_bwspec(&opts->data.fairq_opts.linkshare.m2,
1365 pa->pq_u.fairq_opts.lssc_d =
1366 opts->data.fairq_opts.linkshare.d;
1370 warnx("eval_queue_opts: unknown scheduler type %u",
1380 eval_bwspec(struct node_queue_bw *bw, u_int32_t ref_bw)
1382 if (bw->bw_absolute > 0)
1383 return (bw->bw_absolute);
1385 if (bw->bw_percent > 0)
1386 return (ref_bw / 100 * bw->bw_percent);
1392 print_hfsc_sc(const char *scname, u_int m1, u_int d, u_int m2,
1393 const struct node_hfsc_sc *sc)
1395 printf(" %s", scname);
1399 if (sc != NULL && sc->m1.bw_percent > 0)
1400 printf("%u%%", sc->m1.bw_percent);
1402 printf("%s", rate2str((double)m1));
1406 if (sc != NULL && sc->m2.bw_percent > 0)
1407 printf(" %u%%", sc->m2.bw_percent);
1409 printf(" %s", rate2str((double)m2));
1416 print_fairq_sc(const char *scname, u_int m1, u_int d, u_int m2,
1417 const struct node_fairq_sc *sc)
1419 printf(" %s", scname);
1423 if (sc != NULL && sc->m1.bw_percent > 0)
1424 printf("%u%%", sc->m1.bw_percent);
1426 printf("%s", rate2str((double)m1));
1430 if (sc != NULL && sc->m2.bw_percent > 0)
1431 printf(" %u%%", sc->m2.bw_percent);
1433 printf(" %s", rate2str((double)m2));