1 /*	$OpenBSD: altq_subr.c,v 1.19 2004/07/28 17:15:12 tholo Exp $	*/
2 /*	$KAME: altq_subr.c,v 1.11 2002/01/11 08:11:49 kjc Exp $	*/
3 
4 /*
5  * Copyright (C) 1997-2002
6  *	Sony Computer Science Laboratories Inc.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/param.h>
31 #include <sys/malloc.h>
32 #include <sys/mbuf.h>
33 #include <sys/systm.h>
34 #include <sys/proc.h>
35 #include <sys/socket.h>
36 #include <sys/socketvar.h>
37 #include <sys/kernel.h>
38 #include <sys/errno.h>
39 #include <sys/syslog.h>
40 #include <sys/sysctl.h>
41 #include <sys/queue.h>
42 
43 #include <net/if.h>
44 #include <net/if_dl.h>
45 #include <net/if_types.h>
46 
47 #include <netinet/in.h>
48 #include <netinet/in_systm.h>
49 #include <netinet/ip.h>
50 #ifdef INET6
51 #include <netinet/ip6.h>
52 #endif
53 #include <netinet/tcp.h>
54 #include <netinet/udp.h>
55 
56 #include <net/pfvar.h>
57 #include <altq/altq.h>
58 
59 /* machine dependent clock related includes */
60 #if defined(__i386__)
61 #include <machine/cpufunc.h>		/* for pentium tsc */
62 #include <machine/specialreg.h>		/* for CPUID_TSC */
63 #endif /* __i386__ */
64 
65 /*
66  * internal function prototypes
67  */
68 static void	tbr_timeout(void *);
69 int (*altq_input)(struct mbuf *, int) = NULL;
70 static int tbr_timer = 0;	/* token bucket regulator timer */
71 static struct callout tbr_callout = CALLOUT_INITIALIZER;
72 
73 /*
74  * alternate queueing support routines
75  */
76 
77 /* look up the queue state by the interface name and the queueing type. */
78 void *
altq_lookup(name,type)79 altq_lookup(name, type)
80 	char *name;
81 	int type;
82 {
83 	struct ifnet *ifp;
84 
85 	if ((ifp = ifunit(name)) != NULL) {
86 		if (type != ALTQT_NONE && ifp->if_snd.altq_type == type)
87 			return (ifp->if_snd.altq_disc);
88 	}
89 
90 	return NULL;
91 }
92 
93 int
altq_attach(ifq,type,discipline,enqueue,dequeue,request,clfier,classify)94 altq_attach(ifq, type, discipline, enqueue, dequeue, request, clfier, classify)
95 	struct ifaltq *ifq;
96 	int type;
97 	void *discipline;
98 	int (*enqueue)(struct ifaltq *, struct mbuf *, struct altq_pktattr *);
99 	struct mbuf *(*dequeue)(struct ifaltq *, int);
100 	int (*request)(struct ifaltq *, int, void *);
101 	void *clfier;
102 	void *(*classify)(void *, struct mbuf *, int);
103 {
104 	if (!ALTQ_IS_READY(ifq))
105 		return ENXIO;
106 
107 #if 0	/* pfaltq can override the existing discipline */
108 	if (ALTQ_IS_ENABLED(ifq))
109 		return EBUSY;
110 	if (ALTQ_IS_ATTACHED(ifq))
111 		return EEXIST;
112 #endif
113 	ifq->altq_type     = type;
114 	ifq->altq_disc     = discipline;
115 	ifq->altq_enqueue  = enqueue;
116 	ifq->altq_dequeue  = dequeue;
117 	ifq->altq_request  = request;
118 	ifq->altq_clfier   = clfier;
119 	ifq->altq_classify = classify;
120 	ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED);
121 
122 	return 0;
123 }
124 
125 int
altq_detach(ifq)126 altq_detach(ifq)
127 	struct ifaltq *ifq;
128 {
129 	if (!ALTQ_IS_READY(ifq))
130 		return ENXIO;
131 	if (ALTQ_IS_ENABLED(ifq))
132 		return EBUSY;
133 	if (!ALTQ_IS_ATTACHED(ifq))
134 		return (0);
135 
136 	ifq->altq_type     = ALTQT_NONE;
137 	ifq->altq_disc     = NULL;
138 	ifq->altq_enqueue  = NULL;
139 	ifq->altq_dequeue  = NULL;
140 	ifq->altq_request  = NULL;
141 	ifq->altq_clfier   = NULL;
142 	ifq->altq_classify = NULL;
143 	ifq->altq_flags &= ALTQF_CANTCHANGE;
144 	return 0;
145 }
146 
147 int
altq_enable(ifq)148 altq_enable(ifq)
149 	struct ifaltq *ifq;
150 {
151 	int s;
152 
153 	if (!ALTQ_IS_READY(ifq))
154 		return ENXIO;
155 	if (ALTQ_IS_ENABLED(ifq))
156 		return 0;
157 
158 	s = splimp();
159 	IFQ_PURGE(ifq);
160 	ASSERT(ifq->ifq_len == 0);
161 	ifq->altq_flags |= ALTQF_ENABLED;
162 	if (ifq->altq_clfier != NULL)
163 		ifq->altq_flags |= ALTQF_CLASSIFY;
164 	splx(s);
165 
166 	return 0;
167 }
168 
169 int
altq_disable(ifq)170 altq_disable(ifq)
171 	struct ifaltq *ifq;
172 {
173 	int s;
174 
175 	if (!ALTQ_IS_ENABLED(ifq))
176 		return 0;
177 
178 	s = splimp();
179 	IFQ_PURGE(ifq);
180 	ASSERT(ifq->ifq_len == 0);
181 	ifq->altq_flags &= ~(ALTQF_ENABLED|ALTQF_CLASSIFY);
182 	splx(s);
183 	return 0;
184 }
185 
186 void
altq_assert(file,line,failedexpr)187 altq_assert(file, line, failedexpr)
188 	const char *file, *failedexpr;
189 	int line;
190 {
191 	(void)printf("altq assertion \"%s\" failed: file \"%s\", line %d\n",
192 		     failedexpr, file, line);
193 	panic("altq assertion");
194 	/* NOTREACHED */
195 }
196 
197 /*
198  * internal representation of token bucket parameters
199  *	rate:	byte_per_unittime << 32
200  *		(((bits_per_sec) / 8) << 32) / machclk_freq
201  *	depth:	byte << 32
202  *
203  */
204 #define	TBR_SHIFT	32
205 #define	TBR_SCALE(x)	((int64_t)(x) << TBR_SHIFT)
206 #define	TBR_UNSCALE(x)	((x) >> TBR_SHIFT)
207 
208 struct mbuf *
tbr_dequeue(ifq,op)209 tbr_dequeue(ifq, op)
210 	struct ifaltq *ifq;
211 	int op;
212 {
213 	struct tb_regulator *tbr;
214 	struct mbuf *m;
215 	int64_t interval;
216 	u_int64_t now;
217 
218 	tbr = ifq->altq_tbr;
219 	if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) {
220 		/* if this is a remove after poll, bypass tbr check */
221 	} else {
222 		/* update token only when it is negative */
223 		if (tbr->tbr_token <= 0) {
224 			now = read_machclk();
225 			interval = now - tbr->tbr_last;
226 			if (interval >= tbr->tbr_filluptime)
227 				tbr->tbr_token = tbr->tbr_depth;
228 			else {
229 				tbr->tbr_token += interval * tbr->tbr_rate;
230 				if (tbr->tbr_token > tbr->tbr_depth)
231 					tbr->tbr_token = tbr->tbr_depth;
232 			}
233 			tbr->tbr_last = now;
234 		}
235 		/* if token is still negative, don't allow dequeue */
236 		if (tbr->tbr_token <= 0)
237 			return (NULL);
238 	}
239 
240 	if (ALTQ_IS_ENABLED(ifq))
241 		m = (*ifq->altq_dequeue)(ifq, op);
242 	else {
243 		if (op == ALTDQ_POLL)
244 			IF_POLL(ifq, m);
245 		else
246 			IF_DEQUEUE(ifq, m);
247 	}
248 
249 	if (m != NULL && op == ALTDQ_REMOVE)
250 		tbr->tbr_token -= TBR_SCALE(m_pktlen(m));
251 	tbr->tbr_lastop = op;
252 	return (m);
253 }
254 
255 /*
256  * set a token bucket regulator.
257  * if the specified rate is zero, the token bucket regulator is deleted.
258  */
259 int
tbr_set(ifq,profile)260 tbr_set(ifq, profile)
261 	struct ifaltq *ifq;
262 	struct tb_profile *profile;
263 {
264 	struct tb_regulator *tbr, *otbr;
265 
266 	if (machclk_freq == 0)
267 		init_machclk();
268 	if (machclk_freq == 0) {
269 		printf("tbr_set: no cpu clock available!\n");
270 		return (ENXIO);
271 	}
272 
273 	if (profile->rate == 0) {
274 		/* delete this tbr */
275 		if ((tbr = ifq->altq_tbr) == NULL)
276 			return (ENOENT);
277 		ifq->altq_tbr = NULL;
278 		FREE(tbr, M_DEVBUF);
279 		return (0);
280 	}
281 
282 	MALLOC(tbr, struct tb_regulator *, sizeof(struct tb_regulator),
283 	       M_DEVBUF, M_WAITOK);
284 	if (tbr == NULL)
285 		return (ENOMEM);
286 	bzero(tbr, sizeof(struct tb_regulator));
287 
288 	tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq;
289 	tbr->tbr_depth = TBR_SCALE(profile->depth);
290 	if (tbr->tbr_rate > 0)
291 		tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate;
292 	else
293 		tbr->tbr_filluptime = 0xffffffffffffffffLL;
294 	tbr->tbr_token = tbr->tbr_depth;
295 	tbr->tbr_last = read_machclk();
296 	tbr->tbr_lastop = ALTDQ_REMOVE;
297 
298 	otbr = ifq->altq_tbr;
299 	ifq->altq_tbr = tbr;	/* set the new tbr */
300 
301 	if (otbr != NULL)
302 		FREE(otbr, M_DEVBUF);
303 	else {
304 		if (tbr_timer == 0) {
305 			CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
306 			tbr_timer = 1;
307 		}
308 	}
309 	return (0);
310 }
311 
312 /*
313  * tbr_timeout goes through the interface list, and kicks the drivers
314  * if necessary.
315  */
316 static void
tbr_timeout(arg)317 tbr_timeout(arg)
318 	void *arg;
319 {
320 	struct ifnet *ifp;
321 	int active, s;
322 
323 	active = 0;
324 	s = splimp();
325 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
326 		if (!TBR_IS_ENABLED(&ifp->if_snd))
327 			continue;
328 		active++;
329 		if (!IFQ_IS_EMPTY(&ifp->if_snd) && ifp->if_start != NULL)
330 			(*ifp->if_start)(ifp);
331 	}
332 	splx(s);
333 	if (active > 0)
334 		CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
335 	else
336 		tbr_timer = 0;	/* don't need tbr_timer anymore */
337 #if defined(__alpha__) && !defined(ALTQ_NOPCC)
338 	{
339 		/*
340 		 * XXX read out the machine dependent clock once a second
341 		 * to detect counter wrap-around.
342 		 */
343 		static u_int cnt;
344 
345 		if (++cnt >= hz) {
346 			(void)read_machclk();
347 			cnt = 0;
348 		}
349 	}
350 #endif /* __alpha__ && !ALTQ_NOPCC */
351 }
352 
353 /*
354  * get token bucket regulator profile
355  */
356 int
tbr_get(ifq,profile)357 tbr_get(ifq, profile)
358 	struct ifaltq *ifq;
359 	struct tb_profile *profile;
360 {
361 	struct tb_regulator *tbr;
362 
363 	if ((tbr = ifq->altq_tbr) == NULL) {
364 		profile->rate = 0;
365 		profile->depth = 0;
366 	} else {
367 		profile->rate =
368 		    (u_int)TBR_UNSCALE(tbr->tbr_rate * 8 * machclk_freq);
369 		profile->depth = (u_int)TBR_UNSCALE(tbr->tbr_depth);
370 	}
371 	return (0);
372 }
373 
374 /*
375  * attach a discipline to the interface.  if one already exists, it is
376  * overridden.
377  */
378 int
altq_pfattach(struct pf_altq * a)379 altq_pfattach(struct pf_altq *a)
380 {
381 	int error = 0;
382 
383 	switch (a->scheduler) {
384 	case ALTQT_NONE:
385 		break;
386 #ifdef ALTQ_CBQ
387 	case ALTQT_CBQ:
388 		error = cbq_pfattach(a);
389 		break;
390 #endif
391 #ifdef ALTQ_PRIQ
392 	case ALTQT_PRIQ:
393 		error = priq_pfattach(a);
394 		break;
395 #endif
396 #ifdef ALTQ_HFSC
397 	case ALTQT_HFSC:
398 		error = hfsc_pfattach(a);
399 		break;
400 #endif
401 	default:
402 		error = ENXIO;
403 	}
404 
405 	return (error);
406 }
407 
408 /*
409  * detach a discipline from the interface.
410  * it is possible that the discipline was already overridden by another
411  * discipline.
412  */
413 int
altq_pfdetach(struct pf_altq * a)414 altq_pfdetach(struct pf_altq *a)
415 {
416 	struct ifnet *ifp;
417 	int s, error = 0;
418 
419 	if ((ifp = ifunit(a->ifname)) == NULL)
420 		return (EINVAL);
421 
422 	/* if this discipline is no longer referenced, just return */
423 	if (a->altq_disc == NULL || a->altq_disc != ifp->if_snd.altq_disc)
424 		return (0);
425 
426 	s = splimp();
427 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
428 		error = altq_disable(&ifp->if_snd);
429 	if (error == 0)
430 		error = altq_detach(&ifp->if_snd);
431 	splx(s);
432 
433 	return (error);
434 }
435 
436 /*
437  * add a discipline or a queue
438  */
439 int
altq_add(struct pf_altq * a)440 altq_add(struct pf_altq *a)
441 {
442 	int error = 0;
443 
444 	if (a->qname[0] != 0)
445 		return (altq_add_queue(a));
446 
447 	if (machclk_freq == 0)
448 		init_machclk();
449 	if (machclk_freq == 0)
450 		panic("altq_add: no cpu clock");
451 
452 	switch (a->scheduler) {
453 #ifdef ALTQ_CBQ
454 	case ALTQT_CBQ:
455 		error = cbq_add_altq(a);
456 		break;
457 #endif
458 #ifdef ALTQ_PRIQ
459 	case ALTQT_PRIQ:
460 		error = priq_add_altq(a);
461 		break;
462 #endif
463 #ifdef ALTQ_HFSC
464 	case ALTQT_HFSC:
465 		error = hfsc_add_altq(a);
466 		break;
467 #endif
468 	default:
469 		error = ENXIO;
470 	}
471 
472 	return (error);
473 }
474 
475 /*
476  * remove a discipline or a queue
477  */
478 int
altq_remove(struct pf_altq * a)479 altq_remove(struct pf_altq *a)
480 {
481 	int error = 0;
482 
483 	if (a->qname[0] != 0)
484 		return (altq_remove_queue(a));
485 
486 	switch (a->scheduler) {
487 #ifdef ALTQ_CBQ
488 	case ALTQT_CBQ:
489 		error = cbq_remove_altq(a);
490 		break;
491 #endif
492 #ifdef ALTQ_PRIQ
493 	case ALTQT_PRIQ:
494 		error = priq_remove_altq(a);
495 		break;
496 #endif
497 #ifdef ALTQ_HFSC
498 	case ALTQT_HFSC:
499 		error = hfsc_remove_altq(a);
500 		break;
501 #endif
502 	default:
503 		error = ENXIO;
504 	}
505 
506 	return (error);
507 }
508 
509 /*
510  * add a queue to the discipline
511  */
512 int
altq_add_queue(struct pf_altq * a)513 altq_add_queue(struct pf_altq *a)
514 {
515 	int error = 0;
516 
517 	switch (a->scheduler) {
518 #ifdef ALTQ_CBQ
519 	case ALTQT_CBQ:
520 		error = cbq_add_queue(a);
521 		break;
522 #endif
523 #ifdef ALTQ_PRIQ
524 	case ALTQT_PRIQ:
525 		error = priq_add_queue(a);
526 		break;
527 #endif
528 #ifdef ALTQ_HFSC
529 	case ALTQT_HFSC:
530 		error = hfsc_add_queue(a);
531 		break;
532 #endif
533 	default:
534 		error = ENXIO;
535 	}
536 
537 	return (error);
538 }
539 
540 /*
541  * remove a queue from the discipline
542  */
543 int
altq_remove_queue(struct pf_altq * a)544 altq_remove_queue(struct pf_altq *a)
545 {
546 	int error = 0;
547 
548 	switch (a->scheduler) {
549 #ifdef ALTQ_CBQ
550 	case ALTQT_CBQ:
551 		error = cbq_remove_queue(a);
552 		break;
553 #endif
554 #ifdef ALTQ_PRIQ
555 	case ALTQT_PRIQ:
556 		error = priq_remove_queue(a);
557 		break;
558 #endif
559 #ifdef ALTQ_HFSC
560 	case ALTQT_HFSC:
561 		error = hfsc_remove_queue(a);
562 		break;
563 #endif
564 	default:
565 		error = ENXIO;
566 	}
567 
568 	return (error);
569 }
570 
571 /*
572  * get queue statistics
573  */
574 int
altq_getqstats(struct pf_altq * a,void * ubuf,int * nbytes)575 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes)
576 {
577 	int error = 0;
578 
579 	switch (a->scheduler) {
580 #ifdef ALTQ_CBQ
581 	case ALTQT_CBQ:
582 		error = cbq_getqstats(a, ubuf, nbytes);
583 		break;
584 #endif
585 #ifdef ALTQ_PRIQ
586 	case ALTQT_PRIQ:
587 		error = priq_getqstats(a, ubuf, nbytes);
588 		break;
589 #endif
590 #ifdef ALTQ_HFSC
591 	case ALTQT_HFSC:
592 		error = hfsc_getqstats(a, ubuf, nbytes);
593 		break;
594 #endif
595 	default:
596 		error = ENXIO;
597 	}
598 
599 	return (error);
600 }
601 
602 /*
603  * read and write diffserv field in IPv4 or IPv6 header
604  */
605 u_int8_t
read_dsfield(m,pktattr)606 read_dsfield(m, pktattr)
607 	struct mbuf *m;
608 	struct altq_pktattr *pktattr;
609 {
610 	struct mbuf *m0;
611 	u_int8_t ds_field = 0;
612 
613 	if (pktattr == NULL ||
614 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
615 		return ((u_int8_t)0);
616 
617 	/* verify that pattr_hdr is within the mbuf data */
618 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
619 		if ((pktattr->pattr_hdr >= m0->m_data) &&
620 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
621 			break;
622 	if (m0 == NULL) {
623 		/* ick, pattr_hdr is stale */
624 		pktattr->pattr_af = AF_UNSPEC;
625 #ifdef ALTQ_DEBUG
626 		printf("read_dsfield: can't locate header!\n");
627 #endif
628 		return ((u_int8_t)0);
629 	}
630 
631 	if (pktattr->pattr_af == AF_INET) {
632 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
633 
634 		if (ip->ip_v != 4)
635 			return ((u_int8_t)0);	/* version mismatch! */
636 		ds_field = ip->ip_tos;
637 	}
638 #ifdef INET6
639 	else if (pktattr->pattr_af == AF_INET6) {
640 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
641 		u_int32_t flowlabel;
642 
643 		flowlabel = ntohl(ip6->ip6_flow);
644 		if ((flowlabel >> 28) != 6)
645 			return ((u_int8_t)0);	/* version mismatch! */
646 		ds_field = (flowlabel >> 20) & 0xff;
647 	}
648 #endif
649 	return (ds_field);
650 }
651 
652 void
write_dsfield(m,pktattr,dsfield)653 write_dsfield(m, pktattr, dsfield)
654 	struct mbuf *m;
655 	struct altq_pktattr *pktattr;
656 	u_int8_t dsfield;
657 {
658 	struct mbuf *m0;
659 
660 	if (pktattr == NULL ||
661 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
662 		return;
663 
664 	/* verify that pattr_hdr is within the mbuf data */
665 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
666 		if ((pktattr->pattr_hdr >= m0->m_data) &&
667 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
668 			break;
669 	if (m0 == NULL) {
670 		/* ick, pattr_hdr is stale */
671 		pktattr->pattr_af = AF_UNSPEC;
672 #ifdef ALTQ_DEBUG
673 		printf("write_dsfield: can't locate header!\n");
674 #endif
675 		return;
676 	}
677 
678 	if (pktattr->pattr_af == AF_INET) {
679 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
680 		u_int8_t old;
681 		int32_t sum;
682 
683 		if (ip->ip_v != 4)
684 			return;		/* version mismatch! */
685 		old = ip->ip_tos;
686 		dsfield |= old & 3;	/* leave CU bits */
687 		if (old == dsfield)
688 			return;
689 		ip->ip_tos = dsfield;
690 		/*
691 		 * update checksum (from RFC1624)
692 		 *	   HC' = ~(~HC + ~m + m')
693 		 */
694 		sum = ~ntohs(ip->ip_sum) & 0xffff;
695 		sum += 0xff00 + (~old & 0xff) + dsfield;
696 		sum = (sum >> 16) + (sum & 0xffff);
697 		sum += (sum >> 16);  /* add carry */
698 
699 		ip->ip_sum = htons(~sum & 0xffff);
700 	}
701 #ifdef INET6
702 	else if (pktattr->pattr_af == AF_INET6) {
703 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
704 		u_int32_t flowlabel;
705 
706 		flowlabel = ntohl(ip6->ip6_flow);
707 		if ((flowlabel >> 28) != 6)
708 			return;		/* version mismatch! */
709 		flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20);
710 		ip6->ip6_flow = htonl(flowlabel);
711 	}
712 #endif
713 	return;
714 }
715 
716 
717 /*
718  * high resolution clock support taking advantage of a machine dependent
719  * high resolution time counter (e.g., timestamp counter of intel pentium).
720  * we assume
721  *  - 64-bit-long monotonically-increasing counter
722  *  - frequency range is 100M-4GHz (CPU speed)
723  */
724 /* if pcc is not available or disabled, emulate 256MHz using microtime() */
725 #define	MACHCLK_SHIFT	8
726 
727 int machclk_usepcc;
728 u_int32_t machclk_freq = 0;
729 u_int32_t machclk_per_tick = 0;
730 
731 #ifdef __alpha__
732 extern u_int64_t cycles_per_usec;	/* alpha cpu clock frequency */
733 #endif /* __alpha__ */
734 
735 void
init_machclk(void)736 init_machclk(void)
737 {
738 	machclk_usepcc = 1;
739 
740 #if (!defined(__i386__) && !defined(__alpha__)) || defined(ALTQ_NOPCC)
741 	machclk_usepcc = 0;
742 #endif
743 #if defined(__FreeBSD__) && defined(SMP)
744 	machclk_usepcc = 0;
745 #endif
746 #if defined(__NetBSD__) && defined(MULTIPROCESSOR)
747 	machclk_usepcc = 0;
748 #endif
749 #if defined(__OpenBSD__) && defined(__HAVE_TIMECOUNTER)
750 	/*
751 	 * If we have timecounters, microtime is good enough and we can
752 	 * avoid problems on machines with variable cycle counter
753 	 * frequencies.
754 	 */
755 	machclk_usepcc = 0;
756 #endif
757 #ifdef __i386__
758 	/* check if TSC is available */
759 	if (machclk_usepcc == 1 && (cpu_feature & CPUID_TSC) == 0)
760 		machclk_usepcc = 0;
761 #endif
762 
763 	if (machclk_usepcc == 0) {
764 		/* emulate 256MHz using microtime() */
765 		machclk_freq = 1000000 << MACHCLK_SHIFT;
766 		machclk_per_tick = machclk_freq / hz;
767 #ifdef ALTQ_DEBUG
768 		printf("altq: emulate %uHz cpu clock\n", machclk_freq);
769 #endif
770 		return;
771 	}
772 
773 	/*
774 	 * if the clock frequency (of Pentium TSC or Alpha PCC) is
775 	 * accessible, just use it.
776 	 */
777 #if defined(__i386__) && (defined(I586_CPU) || defined(I686_CPU))
778 	/* XXX - this will break down with variable cpu frequency. */
779 	machclk_freq = pentium_mhz * 1000000;
780 #endif
781 #if defined(__alpha__)
782 	machclk_freq = (u_int32_t)(cycles_per_usec * 1000000);
783 #endif /* __alpha__ */
784 
785 	/*
786 	 * if we don't know the clock frequency, measure it.
787 	 */
788 	if (machclk_freq == 0) {
789 		static int	wait;
790 		struct timeval	tv_start, tv_end;
791 		u_int64_t	start, end, diff;
792 		int		timo;
793 
794 		microtime(&tv_start);
795 		start = read_machclk();
796 		timo = hz;	/* 1 sec */
797 		(void)tsleep(&wait, PWAIT | PCATCH, "init_machclk", timo);
798 		microtime(&tv_end);
799 		end = read_machclk();
800 		diff = (u_int64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000
801 		    + tv_end.tv_usec - tv_start.tv_usec;
802 		if (diff != 0)
803 			machclk_freq = (u_int)((end - start) * 1000000 / diff);
804 	}
805 
806 	machclk_per_tick = machclk_freq / hz;
807 
808 #ifdef ALTQ_DEBUG
809 	printf("altq: CPU clock: %uHz\n", machclk_freq);
810 #endif
811 }
812 
813 #if defined(__OpenBSD__) && defined(__i386__)
814 static __inline u_int64_t
rdtsc(void)815 rdtsc(void)
816 {
817 	u_int64_t rv;
818 	__asm __volatile(".byte 0x0f, 0x31" : "=A" (rv));
819 	return (rv);
820 }
821 #endif /* __OpenBSD__ && __i386__ */
822 
823 u_int64_t
read_machclk(void)824 read_machclk(void)
825 {
826 	u_int64_t val;
827 
828 	if (machclk_usepcc) {
829 #if defined(__i386__)
830 		val = rdtsc();
831 #elif defined(__alpha__)
832 		static u_int32_t last_pcc, upper;
833 		u_int32_t pcc;
834 
835 		/*
836 		 * for alpha, make a 64bit counter value out of the 32bit
837 		 * alpha processor cycle counter.
838 		 * read_machclk must be called within a half of its
839 		 * wrap-around cycle (about 5 sec for 400MHz cpu) to properly
840 		 * detect a counter wrap-around.
841 		 * tbr_timeout calls read_machclk once a second.
842 		 */
843 		pcc = (u_int32_t)alpha_rpcc();
844 		if (pcc <= last_pcc)
845 			upper++;
846 		last_pcc = pcc;
847 		val = ((u_int64_t)upper << 32) + pcc;
848 #else
849 		panic("read_machclk");
850 #endif
851 	} else {
852 		struct timeval tv;
853 
854 		microtime(&tv);
855 		val = (((u_int64_t)(tv.tv_sec - boottime.tv_sec) * 1000000
856 		    + tv.tv_usec) << MACHCLK_SHIFT);
857 	}
858 	return (val);
859 }
860