1 /* $OpenBSD: pfctl_altq.c,v 1.91 2006/11/28 00:08:50 henning 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 #include <sys/param.h>
22 #include <sys/ioctl.h>
23 #include <sys/socket.h>
24 #include <sys/sysctl.h>
27 #include <net/if_mib.h>
28 #include <netinet/in.h>
29 #include <net/pf/pfvar.h>
40 #include <net/altq/altq.h>
41 #include <net/altq/altq_cbq.h>
42 #include <net/altq/altq_priq.h>
43 #include <net/altq/altq_hfsc.h>
44 #include <net/altq/altq_fairq.h>
46 #include "pfctl_parser.h"
49 #define is_sc_null(sc) (((sc) == NULL) || ((sc)->m1 == 0 && (sc)->m2 == 0))
51 TAILQ_HEAD(altqs
, pf_altq
) altqs
= TAILQ_HEAD_INITIALIZER(altqs
);
52 LIST_HEAD(gen_sc
, segment
) rtsc
, lssc
;
54 struct pf_altq
*qname_to_pfaltq(const char *, const char *);
55 u_int32_t
qname_to_qid(const char *);
57 static int eval_pfqueue_cbq(struct pfctl
*, struct pf_altq
*);
58 static int cbq_compute_idletime(struct pfctl
*, struct pf_altq
*);
59 static int check_commit_cbq(int, int, struct pf_altq
*);
60 static int print_cbq_opts(const struct pf_altq
*);
62 static int eval_pfqueue_priq(struct pfctl
*, struct pf_altq
*);
63 static int check_commit_priq(int, int, struct pf_altq
*);
64 static int print_priq_opts(const struct pf_altq
*);
66 static int eval_pfqueue_hfsc(struct pfctl
*, struct pf_altq
*);
67 static int check_commit_hfsc(int, int, struct pf_altq
*);
68 static int print_hfsc_opts(const struct pf_altq
*,
69 const struct node_queue_opt
*);
71 static int eval_pfqueue_fairq(struct pfctl
*, struct pf_altq
*);
72 static int print_fairq_opts(const struct pf_altq
*,
73 const struct node_queue_opt
*);
74 static int check_commit_fairq(int, int, struct pf_altq
*);
76 static void gsc_add_sc(struct gen_sc
*, struct service_curve
*);
77 static int is_gsc_under_sc(struct gen_sc
*,
78 struct service_curve
*);
79 static void gsc_destroy(struct gen_sc
*);
80 static struct segment
*gsc_getentry(struct gen_sc
*, double);
81 static int gsc_add_seg(struct gen_sc
*, double, double, double,
83 static double sc_x2y(struct service_curve
*, double);
85 u_int32_t
getifspeed(const char *);
86 u_long
getifmtu(char *);
87 int eval_queue_opts(struct pf_altq
*, struct node_queue_opt
*,
89 u_int32_t
eval_bwspec(struct node_queue_bw
*, u_int32_t
);
90 void print_hfsc_sc(const char *, u_int
, u_int
, u_int
,
91 const struct node_hfsc_sc
*);
92 void print_fairq_sc(const char *, u_int
, u_int
, u_int
,
93 const struct node_fairq_sc
*);
96 pfaltq_store(struct pf_altq
*a
)
100 if ((altq
= malloc(sizeof(*altq
))) == NULL
)
102 memcpy(altq
, a
, sizeof(struct pf_altq
));
103 TAILQ_INSERT_TAIL(&altqs
, altq
, entries
);
107 pfaltq_lookup(const char *ifname
)
109 struct pf_altq
*altq
;
111 TAILQ_FOREACH(altq
, &altqs
, entries
) {
112 if (strncmp(ifname
, altq
->ifname
, IFNAMSIZ
) == 0 &&
120 qname_to_pfaltq(const char *qname
, const char *ifname
)
122 struct pf_altq
*altq
;
124 TAILQ_FOREACH(altq
, &altqs
, entries
) {
125 if (strncmp(ifname
, altq
->ifname
, IFNAMSIZ
) == 0 &&
126 strncmp(qname
, altq
->qname
, PF_QNAME_SIZE
) == 0)
133 qname_to_qid(const char *qname
)
135 struct pf_altq
*altq
;
138 * We guarantee that same named queues on different interfaces
139 * have the same qid, so we do NOT need to limit matching on
143 TAILQ_FOREACH(altq
, &altqs
, entries
) {
144 if (strncmp(qname
, altq
->qname
, PF_QNAME_SIZE
) == 0)
151 print_altq(const struct pf_altq
*a
, unsigned level
, struct node_queue_bw
*bw
,
152 struct node_queue_opt
*qopts
)
154 if (a
->qname
[0] != 0) {
155 print_queue(a
, level
, bw
, 1, qopts
);
159 printf("altq on %s ", a
->ifname
);
161 switch (a
->scheduler
) {
163 if (!print_cbq_opts(a
))
167 if (!print_priq_opts(a
))
171 if (!print_hfsc_opts(a
, qopts
))
175 if (!print_fairq_opts(a
, qopts
))
180 if (bw
!= NULL
&& bw
->bw_percent
> 0) {
181 if (bw
->bw_percent
< 100)
182 printf("bandwidth %u%% ", bw
->bw_percent
);
184 printf("bandwidth %s ", rate2str((double)a
->ifbandwidth
));
186 if (a
->qlimit
!= DEFAULT_QLIMIT
)
187 printf("qlimit %u ", a
->qlimit
);
188 printf("tbrsize %u ", a
->tbrsize
);
192 print_queue(const struct pf_altq
*a
, unsigned level
, struct node_queue_bw
*bw
,
193 int print_interface
, struct node_queue_opt
*qopts
)
198 for (i
= 0; i
< level
; ++i
)
200 printf("%s ", a
->qname
);
202 printf("on %s ", a
->ifname
);
203 if (a
->scheduler
== ALTQT_CBQ
|| a
->scheduler
== ALTQT_HFSC
||
204 a
->scheduler
== ALTQT_FAIRQ
) {
205 if (bw
!= NULL
&& bw
->bw_percent
> 0) {
206 if (bw
->bw_percent
< 100)
207 printf("bandwidth %u%% ", bw
->bw_percent
);
209 printf("bandwidth %s ", rate2str((double)a
->bandwidth
));
211 if (a
->priority
!= DEFAULT_PRIORITY
)
212 printf("priority %u ", a
->priority
);
213 if (a
->qlimit
!= DEFAULT_QLIMIT
)
214 printf("qlimit %u ", a
->qlimit
);
215 switch (a
->scheduler
) {
223 print_hfsc_opts(a
, qopts
);
226 print_fairq_opts(a
, qopts
);
232 * eval_pfaltq computes the discipline parameters.
235 eval_pfaltq(struct pfctl
*pf __unused
, struct pf_altq
*pa
, struct node_queue_bw
*bw
,
236 struct node_queue_opt
*opts
)
238 u_int rate
, size
, errors
= 0;
240 if (bw
->bw_absolute
> 0)
241 pa
->ifbandwidth
= bw
->bw_absolute
;
243 if ((rate
= getifspeed(pa
->ifname
)) == 0) {
244 fprintf(stderr
, "interface %s does not know its bandwidth, "
245 "please specify an absolute bandwidth\n",
248 } else if ((pa
->ifbandwidth
= eval_bwspec(bw
, rate
)) == 0)
249 pa
->ifbandwidth
= rate
;
251 errors
+= eval_queue_opts(pa
, opts
, pa
->ifbandwidth
);
253 /* if tbrsize is not specified, use heuristics */
254 if (pa
->tbrsize
== 0) {
255 rate
= pa
->ifbandwidth
;
256 if (rate
<= 1 * 1000 * 1000)
258 else if (rate
<= 10 * 1000 * 1000)
260 else if (rate
<= 200 * 1000 * 1000)
264 size
= size
* getifmtu(pa
->ifname
);
273 * check_commit_altq does consistency check for each interface
276 check_commit_altq(int dev
, int opts
)
278 struct pf_altq
*altq
;
281 /* call the discipline check for each interface. */
282 TAILQ_FOREACH(altq
, &altqs
, entries
) {
283 if (altq
->qname
[0] == 0) {
284 switch (altq
->scheduler
) {
286 error
= check_commit_cbq(dev
, opts
, altq
);
289 error
= check_commit_priq(dev
, opts
, altq
);
292 error
= check_commit_hfsc(dev
, opts
, altq
);
295 error
= check_commit_fairq(dev
, opts
, altq
);
306 * eval_pfqueue computes the queue parameters.
309 eval_pfqueue(struct pfctl
*pf
, struct pf_altq
*pa
, struct node_queue_bw
*bw
,
310 struct node_queue_opt
*opts
)
312 /* should be merged with expand_queue */
313 struct pf_altq
*if_pa
, *parent
, *altq
;
317 /* find the corresponding interface and copy fields used by queues */
318 if ((if_pa
= pfaltq_lookup(pa
->ifname
)) == NULL
) {
319 fprintf(stderr
, "altq not defined on %s\n", pa
->ifname
);
322 pa
->scheduler
= if_pa
->scheduler
;
323 pa
->ifbandwidth
= if_pa
->ifbandwidth
;
325 if (qname_to_pfaltq(pa
->qname
, pa
->ifname
) != NULL
) {
326 fprintf(stderr
, "queue %s already exists on interface %s\n",
327 pa
->qname
, pa
->ifname
);
330 pa
->qid
= qname_to_qid(pa
->qname
);
333 if (pa
->parent
[0] != 0) {
334 parent
= qname_to_pfaltq(pa
->parent
, pa
->ifname
);
335 if (parent
== NULL
) {
336 fprintf(stderr
, "parent %s not found for %s\n",
337 pa
->parent
, pa
->qname
);
340 pa
->parent_qid
= parent
->qid
;
343 pa
->qlimit
= DEFAULT_QLIMIT
;
345 if (pa
->scheduler
== ALTQT_CBQ
|| pa
->scheduler
== ALTQT_HFSC
||
346 pa
->scheduler
== ALTQT_FAIRQ
) {
347 pa
->bandwidth
= eval_bwspec(bw
,
348 parent
== NULL
? 0 : parent
->bandwidth
);
350 if (pa
->bandwidth
> pa
->ifbandwidth
) {
351 fprintf(stderr
, "bandwidth for %s higher than "
352 "interface\n", pa
->qname
);
355 /* check the sum of the child bandwidth is under parent's */
356 if (parent
!= NULL
) {
357 if (pa
->bandwidth
> parent
->bandwidth
) {
358 warnx("bandwidth for %s higher than parent",
363 TAILQ_FOREACH(altq
, &altqs
, entries
) {
364 if (strncmp(altq
->ifname
, pa
->ifname
,
366 altq
->qname
[0] != 0 &&
367 strncmp(altq
->parent
, pa
->parent
,
369 bwsum
+= altq
->bandwidth
;
371 bwsum
+= pa
->bandwidth
;
372 if (bwsum
> parent
->bandwidth
) {
373 warnx("the sum of the child bandwidth higher"
374 " than parent \"%s\"", parent
->qname
);
379 if (eval_queue_opts(pa
, opts
, parent
== NULL
? 0 : parent
->bandwidth
))
382 switch (pa
->scheduler
) {
384 error
= eval_pfqueue_cbq(pf
, pa
);
387 error
= eval_pfqueue_priq(pf
, pa
);
390 error
= eval_pfqueue_hfsc(pf
, pa
);
393 error
= eval_pfqueue_fairq(pf
, pa
);
402 * CBQ support functions
404 #define RM_FILTER_GAIN 5 /* log2 of gain, e.g., 5 => 31/32 */
405 #define RM_NS_PER_SEC (1000000000)
408 eval_pfqueue_cbq(struct pfctl
*pf
, struct pf_altq
*pa
)
410 struct cbq_opts
*opts
;
413 if (pa
->priority
>= CBQ_MAXPRI
) {
414 warnx("priority out of range: max %d", CBQ_MAXPRI
- 1);
418 ifmtu
= getifmtu(pa
->ifname
);
419 opts
= &pa
->pq_u
.cbq_opts
;
421 if (opts
->pktsize
== 0) { /* use default */
422 opts
->pktsize
= ifmtu
;
423 if (opts
->pktsize
> MCLBYTES
) /* do what TCP does */
424 opts
->pktsize
&= ~MCLBYTES
;
425 } else if (opts
->pktsize
> ifmtu
)
426 opts
->pktsize
= ifmtu
;
427 if (opts
->maxpktsize
== 0) /* use default */
428 opts
->maxpktsize
= ifmtu
;
429 else if (opts
->maxpktsize
> ifmtu
)
430 opts
->pktsize
= ifmtu
;
432 if (opts
->pktsize
> opts
->maxpktsize
)
433 opts
->pktsize
= opts
->maxpktsize
;
435 if (pa
->parent
[0] == 0)
436 opts
->flags
|= (CBQCLF_ROOTCLASS
| CBQCLF_WRR
);
438 cbq_compute_idletime(pf
, pa
);
443 * compute ns_per_byte, maxidle, minidle, and offtime
446 cbq_compute_idletime(struct pfctl
*pf
, struct pf_altq
*pa
)
448 struct cbq_opts
*opts
;
449 double maxidle_s
, maxidle
, minidle
;
450 double offtime
, nsPerByte
, ifnsPerByte
, ptime
, cptime
;
451 double z
, g
, f
, gton
, gtom
;
452 u_int minburst
, maxburst
;
454 opts
= &pa
->pq_u
.cbq_opts
;
455 ifnsPerByte
= (1.0 / (double)pa
->ifbandwidth
) * RM_NS_PER_SEC
* 8;
456 minburst
= opts
->minburst
;
457 maxburst
= opts
->maxburst
;
459 if (pa
->bandwidth
== 0)
460 f
= 0.0001; /* small enough? */
462 f
= ((double) pa
->bandwidth
/ (double) pa
->ifbandwidth
);
464 nsPerByte
= ifnsPerByte
/ f
;
465 ptime
= (double)opts
->pktsize
* ifnsPerByte
;
466 cptime
= ptime
* (1.0 - f
) / f
;
468 if (nsPerByte
* (double)opts
->maxpktsize
> (double)INT_MAX
) {
470 * this causes integer overflow in kernel!
471 * (bandwidth < 6Kbps when max_pkt_size=1500)
473 if (pa
->bandwidth
!= 0 && (pf
->opts
& PF_OPT_QUIET
) == 0)
474 warnx("queue bandwidth must be larger than %s",
475 rate2str(ifnsPerByte
* (double)opts
->maxpktsize
/
476 (double)INT_MAX
* (double)pa
->ifbandwidth
));
477 fprintf(stderr
, "cbq: queue %s is too slow!\n",
479 nsPerByte
= (double)(INT_MAX
/ opts
->maxpktsize
);
482 if (maxburst
== 0) { /* use default */
483 if (cptime
> 10.0 * 1000000)
488 if (minburst
== 0) /* use default */
490 if (minburst
> maxburst
)
493 z
= (double)(1 << RM_FILTER_GAIN
);
495 gton
= pow(g
, (double)maxburst
);
496 gtom
= pow(g
, (double)(minburst
-1));
497 maxidle
= ((1.0 / f
- 1.0) * ((1.0 - gton
) / gton
));
498 maxidle_s
= (1.0 - g
);
499 if (maxidle
> maxidle_s
)
500 maxidle
= ptime
* maxidle
;
502 maxidle
= ptime
* maxidle_s
;
503 offtime
= cptime
* (1.0 + 1.0/(1.0 - g
) * (1.0 - gtom
) / gtom
);
504 minidle
= -((double)opts
->maxpktsize
* (double)nsPerByte
);
506 /* scale parameters */
507 maxidle
= ((maxidle
* 8.0) / nsPerByte
) *
508 pow(2.0, (double)RM_FILTER_GAIN
);
509 offtime
= (offtime
* 8.0) / nsPerByte
*
510 pow(2.0, (double)RM_FILTER_GAIN
);
511 minidle
= ((minidle
* 8.0) / nsPerByte
) *
512 pow(2.0, (double)RM_FILTER_GAIN
);
514 maxidle
= maxidle
/ 1000.0;
515 offtime
= offtime
/ 1000.0;
516 minidle
= minidle
/ 1000.0;
518 opts
->minburst
= minburst
;
519 opts
->maxburst
= maxburst
;
520 opts
->ns_per_byte
= (u_int
)nsPerByte
;
521 opts
->maxidle
= (u_int
)fabs(maxidle
);
522 opts
->minidle
= (int)minidle
;
523 opts
->offtime
= (u_int
)fabs(offtime
);
529 check_commit_cbq(int dev __unused
, int opts __unused
, struct pf_altq
*pa
)
531 struct pf_altq
*altq
;
532 int root_class
, default_class
;
536 * check if cbq has one root queue and one default queue
539 root_class
= default_class
= 0;
540 TAILQ_FOREACH(altq
, &altqs
, entries
) {
541 if (strncmp(altq
->ifname
, pa
->ifname
, IFNAMSIZ
) != 0)
543 if (altq
->qname
[0] == 0) /* this is for interface */
545 if (altq
->pq_u
.cbq_opts
.flags
& CBQCLF_ROOTCLASS
)
547 if (altq
->pq_u
.cbq_opts
.flags
& CBQCLF_DEFCLASS
)
550 if (root_class
!= 1) {
551 warnx("should have one root queue on %s", pa
->ifname
);
554 if (default_class
!= 1) {
555 warnx("should have one default queue on %s", pa
->ifname
);
562 print_cbq_opts(const struct pf_altq
*a
)
564 const struct cbq_opts
*opts
;
566 opts
= &a
->pq_u
.cbq_opts
;
569 if (opts
->flags
& CBQCLF_RED
)
571 if (opts
->flags
& CBQCLF_ECN
)
573 if (opts
->flags
& CBQCLF_RIO
)
575 if (opts
->flags
& CBQCLF_CLEARDSCP
)
576 printf(" cleardscp");
577 if (opts
->flags
& CBQCLF_BORROW
)
579 if (opts
->flags
& CBQCLF_WRR
)
581 if (opts
->flags
& CBQCLF_EFFICIENT
)
582 printf(" efficient");
583 if (opts
->flags
& CBQCLF_ROOTCLASS
)
585 if (opts
->flags
& CBQCLF_DEFCLASS
)
595 * PRIQ support functions
598 eval_pfqueue_priq(struct pfctl
*pf __unused
, struct pf_altq
*pa
)
600 struct pf_altq
*altq
;
602 if (pa
->priority
>= PRIQ_MAXPRI
) {
603 warnx("priority out of range: max %d", PRIQ_MAXPRI
- 1);
606 /* the priority should be unique for the interface */
607 TAILQ_FOREACH(altq
, &altqs
, entries
) {
608 if (strncmp(altq
->ifname
, pa
->ifname
, IFNAMSIZ
) == 0 &&
609 altq
->qname
[0] != 0 && altq
->priority
== pa
->priority
) {
610 warnx("%s and %s have the same priority",
611 altq
->qname
, pa
->qname
);
620 check_commit_priq(int dev __unused
, int opts __unused
, struct pf_altq
*pa
)
622 struct pf_altq
*altq
;
627 * check if priq has one default class for this interface
630 TAILQ_FOREACH(altq
, &altqs
, entries
) {
631 if (strncmp(altq
->ifname
, pa
->ifname
, IFNAMSIZ
) != 0)
633 if (altq
->qname
[0] == 0) /* this is for interface */
635 if (altq
->pq_u
.priq_opts
.flags
& PRCF_DEFAULTCLASS
)
638 if (default_class
!= 1) {
639 warnx("should have one default queue on %s", pa
->ifname
);
646 print_priq_opts(const struct pf_altq
*a
)
648 const struct priq_opts
*opts
;
650 opts
= &a
->pq_u
.priq_opts
;
654 if (opts
->flags
& PRCF_RED
)
656 if (opts
->flags
& PRCF_ECN
)
658 if (opts
->flags
& PRCF_RIO
)
660 if (opts
->flags
& PRCF_CLEARDSCP
)
661 printf(" cleardscp");
662 if (opts
->flags
& PRCF_DEFAULTCLASS
)
672 * HFSC support functions
675 eval_pfqueue_hfsc(struct pfctl
*pf __unused
, struct pf_altq
*pa
)
677 struct pf_altq
*altq
, *parent
;
678 struct hfsc_opts
*opts
;
679 struct service_curve sc
;
681 opts
= &pa
->pq_u
.hfsc_opts
;
683 if (pa
->parent
[0] == 0) {
685 opts
->lssc_m1
= pa
->ifbandwidth
;
686 opts
->lssc_m2
= pa
->ifbandwidth
;
694 /* if link_share is not specified, use bandwidth */
695 if (opts
->lssc_m2
== 0)
696 opts
->lssc_m2
= pa
->bandwidth
;
698 if ((opts
->rtsc_m1
> 0 && opts
->rtsc_m2
== 0) ||
699 (opts
->lssc_m1
> 0 && opts
->lssc_m2
== 0) ||
700 (opts
->ulsc_m1
> 0 && opts
->ulsc_m2
== 0)) {
701 warnx("m2 is zero for %s", pa
->qname
);
705 if ((opts
->rtsc_m1
< opts
->rtsc_m2
&& opts
->rtsc_m1
!= 0) ||
706 (opts
->lssc_m1
< opts
->lssc_m2
&& opts
->lssc_m1
!= 0) ||
707 (opts
->ulsc_m1
< opts
->ulsc_m2
&& opts
->ulsc_m1
!= 0)) {
708 warnx("m1 must be zero for convex curve: %s", pa
->qname
);
714 * for the real-time service curve, the sum of the service curves
715 * should not exceed 80% of the interface bandwidth. 20% is reserved
716 * not to over-commit the actual interface bandwidth.
717 * for the linkshare service curve, the sum of the child service
718 * curve should not exceed the parent service curve.
719 * for the upper-limit service curve, the assigned bandwidth should
720 * be smaller than the interface bandwidth, and the upper-limit should
721 * be larger than the real-time service curve when both are defined.
723 parent
= qname_to_pfaltq(pa
->parent
, pa
->ifname
);
725 errx(1, "parent %s not found for %s", pa
->parent
, pa
->qname
);
727 TAILQ_FOREACH(altq
, &altqs
, entries
) {
728 if (strncmp(altq
->ifname
, pa
->ifname
, IFNAMSIZ
) != 0)
730 if (altq
->qname
[0] == 0) /* this is for interface */
733 /* if the class has a real-time service curve, add it. */
734 if (opts
->rtsc_m2
!= 0 && altq
->pq_u
.hfsc_opts
.rtsc_m2
!= 0) {
735 sc
.m1
= altq
->pq_u
.hfsc_opts
.rtsc_m1
;
736 sc
.d
= altq
->pq_u
.hfsc_opts
.rtsc_d
;
737 sc
.m2
= altq
->pq_u
.hfsc_opts
.rtsc_m2
;
738 gsc_add_sc(&rtsc
, &sc
);
741 if (strncmp(altq
->parent
, pa
->parent
, PF_QNAME_SIZE
) != 0)
744 /* if the class has a linkshare service curve, add it. */
745 if (opts
->lssc_m2
!= 0 && altq
->pq_u
.hfsc_opts
.lssc_m2
!= 0) {
746 sc
.m1
= altq
->pq_u
.hfsc_opts
.lssc_m1
;
747 sc
.d
= altq
->pq_u
.hfsc_opts
.lssc_d
;
748 sc
.m2
= altq
->pq_u
.hfsc_opts
.lssc_m2
;
749 gsc_add_sc(&lssc
, &sc
);
753 /* check the real-time service curve. reserve 20% of interface bw */
754 if (opts
->rtsc_m2
!= 0) {
755 /* add this queue to the sum */
756 sc
.m1
= opts
->rtsc_m1
;
758 sc
.m2
= opts
->rtsc_m2
;
759 gsc_add_sc(&rtsc
, &sc
);
760 /* compare the sum with 80% of the interface */
763 sc
.m2
= pa
->ifbandwidth
/ 100 * 80;
764 if (!is_gsc_under_sc(&rtsc
, &sc
)) {
765 warnx("real-time sc exceeds 80%% of the interface "
766 "bandwidth (%s)", rate2str((double)sc
.m2
));
771 /* check the linkshare service curve. */
772 if (opts
->lssc_m2
!= 0) {
773 /* add this queue to the child sum */
774 sc
.m1
= opts
->lssc_m1
;
776 sc
.m2
= opts
->lssc_m2
;
777 gsc_add_sc(&lssc
, &sc
);
778 /* compare the sum of the children with parent's sc */
779 sc
.m1
= parent
->pq_u
.hfsc_opts
.lssc_m1
;
780 sc
.d
= parent
->pq_u
.hfsc_opts
.lssc_d
;
781 sc
.m2
= parent
->pq_u
.hfsc_opts
.lssc_m2
;
782 if (!is_gsc_under_sc(&lssc
, &sc
)) {
783 warnx("linkshare sc exceeds parent's sc");
788 /* check the upper-limit service curve. */
789 if (opts
->ulsc_m2
!= 0) {
790 if (opts
->ulsc_m1
> pa
->ifbandwidth
||
791 opts
->ulsc_m2
> pa
->ifbandwidth
) {
792 warnx("upper-limit larger than interface bandwidth");
795 if (opts
->rtsc_m2
!= 0 && opts
->rtsc_m2
> opts
->ulsc_m2
) {
796 warnx("upper-limit sc smaller than real-time sc");
813 * FAIRQ support functions
816 eval_pfqueue_fairq(struct pfctl
*pf __unused
, struct pf_altq
*pa
)
818 struct pf_altq
*altq
, *parent
;
819 struct fairq_opts
*opts
;
820 struct service_curve sc
;
822 opts
= &pa
->pq_u
.fairq_opts
;
824 if (pa
->parent
[0] == 0) {
826 opts
->lssc_m1
= pa
->ifbandwidth
;
827 opts
->lssc_m2
= pa
->ifbandwidth
;
834 /* if link_share is not specified, use bandwidth */
835 if (opts
->lssc_m2
== 0)
836 opts
->lssc_m2
= pa
->bandwidth
;
840 * for the real-time service curve, the sum of the service curves
841 * should not exceed 80% of the interface bandwidth. 20% is reserved
842 * not to over-commit the actual interface bandwidth.
843 * for the link-sharing service curve, the sum of the child service
844 * curve should not exceed the parent service curve.
845 * for the upper-limit service curve, the assigned bandwidth should
846 * be smaller than the interface bandwidth, and the upper-limit should
847 * be larger than the real-time service curve when both are defined.
849 parent
= qname_to_pfaltq(pa
->parent
, pa
->ifname
);
851 errx(1, "parent %s not found for %s", pa
->parent
, pa
->qname
);
853 TAILQ_FOREACH(altq
, &altqs
, entries
) {
854 if (strncmp(altq
->ifname
, pa
->ifname
, IFNAMSIZ
) != 0)
856 if (altq
->qname
[0] == 0) /* this is for interface */
859 if (strncmp(altq
->parent
, pa
->parent
, PF_QNAME_SIZE
) != 0)
862 /* if the class has a link-sharing service curve, add it. */
863 if (opts
->lssc_m2
!= 0 && altq
->pq_u
.fairq_opts
.lssc_m2
!= 0) {
864 sc
.m1
= altq
->pq_u
.fairq_opts
.lssc_m1
;
865 sc
.d
= altq
->pq_u
.fairq_opts
.lssc_d
;
866 sc
.m2
= altq
->pq_u
.fairq_opts
.lssc_m2
;
867 gsc_add_sc(&lssc
, &sc
);
871 /* check the link-sharing service curve. */
872 if (opts
->lssc_m2
!= 0) {
873 sc
.m1
= parent
->pq_u
.fairq_opts
.lssc_m1
;
874 sc
.d
= parent
->pq_u
.fairq_opts
.lssc_d
;
875 sc
.m2
= parent
->pq_u
.fairq_opts
.lssc_m2
;
876 if (!is_gsc_under_sc(&lssc
, &sc
)) {
877 warnx("link-sharing sc exceeds parent's sc");
892 check_commit_hfsc(int dev __unused
, int opts __unused
, struct pf_altq
*pa
)
894 struct pf_altq
*altq
, *def
= NULL
;
898 /* check if hfsc has one default queue for this interface */
900 TAILQ_FOREACH(altq
, &altqs
, entries
) {
901 if (strncmp(altq
->ifname
, pa
->ifname
, IFNAMSIZ
) != 0)
903 if (altq
->qname
[0] == 0) /* this is for interface */
905 if (altq
->parent
[0] == 0) /* dummy root */
907 if (altq
->pq_u
.hfsc_opts
.flags
& HFCF_DEFAULTCLASS
) {
912 if (default_class
!= 1) {
913 warnx("should have one default queue on %s", pa
->ifname
);
916 /* make sure the default queue is a leaf */
917 TAILQ_FOREACH(altq
, &altqs
, entries
) {
918 if (strncmp(altq
->ifname
, pa
->ifname
, IFNAMSIZ
) != 0)
920 if (altq
->qname
[0] == 0) /* this is for interface */
922 if (strncmp(altq
->parent
, def
->qname
, PF_QNAME_SIZE
) == 0) {
923 warnx("default queue is not a leaf");
931 check_commit_fairq(int dev __unused
, int opts __unused
, struct pf_altq
*pa
)
933 struct pf_altq
*altq
, *def
= NULL
;
937 /* check if fairq has one default queue for this interface */
939 TAILQ_FOREACH(altq
, &altqs
, entries
) {
940 if (strncmp(altq
->ifname
, pa
->ifname
, IFNAMSIZ
) != 0)
942 if (altq
->qname
[0] == 0) /* this is for interface */
944 if (altq
->pq_u
.fairq_opts
.flags
& FARF_DEFAULTCLASS
) {
949 if (default_class
!= 1) {
950 warnx("should have one default queue on %s", pa
->ifname
);
953 /* make sure the default queue is a leaf */
954 TAILQ_FOREACH(altq
, &altqs
, entries
) {
955 if (strncmp(altq
->ifname
, pa
->ifname
, IFNAMSIZ
) != 0)
957 if (altq
->qname
[0] == 0) /* this is for interface */
959 if (strncmp(altq
->parent
, def
->qname
, PF_QNAME_SIZE
) == 0) {
960 warnx("default queue is not a leaf");
968 print_hfsc_opts(const struct pf_altq
*a
, const struct node_queue_opt
*qopts
)
970 const struct hfsc_opts
*opts
;
971 const struct node_hfsc_sc
*loc_rtsc
, *loc_lssc
, *ulsc
;
973 opts
= &a
->pq_u
.hfsc_opts
;
975 loc_rtsc
= loc_lssc
= ulsc
= NULL
;
977 loc_rtsc
= &qopts
->data
.hfsc_opts
.realtime
;
978 loc_lssc
= &qopts
->data
.hfsc_opts
.linkshare
;
979 ulsc
= &qopts
->data
.hfsc_opts
.upperlimit
;
982 if (opts
->flags
|| opts
->rtsc_m2
!= 0 || opts
->ulsc_m2
!= 0 ||
983 (opts
->lssc_m2
!= 0 && (opts
->lssc_m2
!= a
->bandwidth
||
984 opts
->lssc_d
!= 0))) {
986 if (opts
->flags
& HFCF_RED
)
988 if (opts
->flags
& HFCF_ECN
)
990 if (opts
->flags
& HFCF_RIO
)
992 if (opts
->flags
& HFCF_CLEARDSCP
)
993 printf(" cleardscp");
994 if (opts
->flags
& HFCF_DEFAULTCLASS
)
996 if (opts
->rtsc_m2
!= 0)
997 print_hfsc_sc("realtime", opts
->rtsc_m1
, opts
->rtsc_d
,
998 opts
->rtsc_m2
, loc_rtsc
);
999 if (opts
->lssc_m2
!= 0 && (opts
->lssc_m2
!= a
->bandwidth
||
1001 print_hfsc_sc("linkshare", opts
->lssc_m1
, opts
->lssc_d
,
1002 opts
->lssc_m2
, loc_lssc
);
1003 if (opts
->ulsc_m2
!= 0)
1004 print_hfsc_sc("upperlimit", opts
->ulsc_m1
, opts
->ulsc_d
,
1005 opts
->ulsc_m2
, ulsc
);
1014 print_fairq_opts(const struct pf_altq
*a
, const struct node_queue_opt
*qopts
)
1016 const struct fairq_opts
*opts
;
1017 const struct node_fairq_sc
*loc_lssc
;
1019 opts
= &a
->pq_u
.fairq_opts
;
1023 loc_lssc
= &qopts
->data
.fairq_opts
.linkshare
;
1026 (opts
->lssc_m2
!= 0 && (opts
->lssc_m2
!= a
->bandwidth
||
1027 opts
->lssc_d
!= 0))) {
1029 if (opts
->flags
& FARF_RED
)
1031 if (opts
->flags
& FARF_ECN
)
1033 if (opts
->flags
& FARF_RIO
)
1035 if (opts
->flags
& FARF_CLEARDSCP
)
1036 printf(" cleardscp");
1037 if (opts
->flags
& FARF_DEFAULTCLASS
)
1039 if (opts
->lssc_m2
!= 0 && (opts
->lssc_m2
!= a
->bandwidth
||
1041 print_fairq_sc("linkshare", opts
->lssc_m1
, opts
->lssc_d
,
1042 opts
->lssc_m2
, loc_lssc
);
1051 * admission control using generalized service curve
1054 #define INFINITY HUGE_VAL /* positive infinity defined in <math.h> */
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");
1296 if (shutdown(s
, SHUT_RDWR
) == -1)
1300 if (ifr
.ifr_mtu
> 0)
1301 return (ifr
.ifr_mtu
);
1303 warnx("could not get mtu for %s, assuming 1500", ifname
);
1309 eval_queue_opts(struct pf_altq
*pa
, struct node_queue_opt
*opts
,
1314 switch (pa
->scheduler
) {
1316 pa
->pq_u
.cbq_opts
= opts
->data
.cbq_opts
;
1319 pa
->pq_u
.priq_opts
= opts
->data
.priq_opts
;
1322 pa
->pq_u
.hfsc_opts
.flags
= opts
->data
.hfsc_opts
.flags
;
1323 if (opts
->data
.hfsc_opts
.linkshare
.used
) {
1324 pa
->pq_u
.hfsc_opts
.lssc_m1
=
1325 eval_bwspec(&opts
->data
.hfsc_opts
.linkshare
.m1
,
1327 pa
->pq_u
.hfsc_opts
.lssc_m2
=
1328 eval_bwspec(&opts
->data
.hfsc_opts
.linkshare
.m2
,
1330 pa
->pq_u
.hfsc_opts
.lssc_d
=
1331 opts
->data
.hfsc_opts
.linkshare
.d
;
1333 if (opts
->data
.hfsc_opts
.realtime
.used
) {
1334 pa
->pq_u
.hfsc_opts
.rtsc_m1
=
1335 eval_bwspec(&opts
->data
.hfsc_opts
.realtime
.m1
,
1337 pa
->pq_u
.hfsc_opts
.rtsc_m2
=
1338 eval_bwspec(&opts
->data
.hfsc_opts
.realtime
.m2
,
1340 pa
->pq_u
.hfsc_opts
.rtsc_d
=
1341 opts
->data
.hfsc_opts
.realtime
.d
;
1343 if (opts
->data
.hfsc_opts
.upperlimit
.used
) {
1344 pa
->pq_u
.hfsc_opts
.ulsc_m1
=
1345 eval_bwspec(&opts
->data
.hfsc_opts
.upperlimit
.m1
,
1347 pa
->pq_u
.hfsc_opts
.ulsc_m2
=
1348 eval_bwspec(&opts
->data
.hfsc_opts
.upperlimit
.m2
,
1350 pa
->pq_u
.hfsc_opts
.ulsc_d
=
1351 opts
->data
.hfsc_opts
.upperlimit
.d
;
1355 pa
->pq_u
.fairq_opts
.flags
= opts
->data
.fairq_opts
.flags
;
1356 pa
->pq_u
.fairq_opts
.nbuckets
= opts
->data
.fairq_opts
.nbuckets
;
1357 pa
->pq_u
.fairq_opts
.hogs_m1
=
1358 eval_bwspec(&opts
->data
.fairq_opts
.hogs_bw
, ref_bw
);
1360 if (opts
->data
.fairq_opts
.linkshare
.used
) {
1361 pa
->pq_u
.fairq_opts
.lssc_m1
=
1362 eval_bwspec(&opts
->data
.fairq_opts
.linkshare
.m1
,
1364 pa
->pq_u
.fairq_opts
.lssc_m2
=
1365 eval_bwspec(&opts
->data
.fairq_opts
.linkshare
.m2
,
1367 pa
->pq_u
.fairq_opts
.lssc_d
=
1368 opts
->data
.fairq_opts
.linkshare
.d
;
1372 warnx("eval_queue_opts: unknown scheduler type %u",
1382 eval_bwspec(struct node_queue_bw
*bw
, u_int32_t ref_bw
)
1384 if (bw
->bw_absolute
> 0)
1385 return (bw
->bw_absolute
);
1387 if (bw
->bw_percent
> 0)
1388 return (ref_bw
/ 100 * bw
->bw_percent
);
1394 print_hfsc_sc(const char *scname
, u_int m1
, u_int d
, u_int m2
,
1395 const struct node_hfsc_sc
*sc
)
1397 printf(" %s", scname
);
1401 if (sc
!= NULL
&& sc
->m1
.bw_percent
> 0)
1402 printf("%u%%", sc
->m1
.bw_percent
);
1404 printf("%s", rate2str((double)m1
));
1408 if (sc
!= NULL
&& sc
->m2
.bw_percent
> 0)
1409 printf(" %u%%", sc
->m2
.bw_percent
);
1411 printf(" %s", rate2str((double)m2
));
1418 print_fairq_sc(const char *scname
, u_int m1
, u_int d
, u_int m2
,
1419 const struct node_fairq_sc
*sc
)
1421 printf(" %s", scname
);
1425 if (sc
!= NULL
&& sc
->m1
.bw_percent
> 0)
1426 printf("%u%%", sc
->m1
.bw_percent
);
1428 printf("%s", rate2str((double)m1
));
1432 if (sc
!= NULL
&& sc
->m2
.bw_percent
> 0)
1433 printf(" %u%%", sc
->m2
.bw_percent
);
1435 printf(" %s", rate2str((double)m2
));