xref: /dragonfly/sys/net/altq/altq_subr.c (revision f746689a)
1 /*	$KAME: altq_subr.c,v 1.23 2004/04/20 16:10:06 itojun Exp $	*/
2 /*	$DragonFly: src/sys/net/altq/altq_subr.c,v 1.12 2008/05/14 11:59:23 sephe Exp $ */
3 
4 /*
5  * Copyright (C) 1997-2003
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 "opt_altq.h"
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 
34 #include <sys/param.h>
35 #include <sys/malloc.h>
36 #include <sys/mbuf.h>
37 #include <sys/systm.h>
38 #include <sys/proc.h>
39 #include <sys/socket.h>
40 #include <sys/socketvar.h>
41 #include <sys/kernel.h>
42 #include <sys/callout.h>
43 #include <sys/errno.h>
44 #include <sys/syslog.h>
45 #include <sys/sysctl.h>
46 #include <sys/queue.h>
47 #include <sys/thread2.h>
48 
49 #include <net/if.h>
50 #include <net/if_dl.h>
51 #include <net/if_types.h>
52 #include <net/ifq_var.h>
53 
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/ip.h>
57 #ifdef INET6
58 #include <netinet/ip6.h>
59 #endif
60 #include <netinet/tcp.h>
61 #include <netinet/udp.h>
62 
63 #include <net/pf/pfvar.h>
64 #include <net/altq/altq.h>
65 
66 /* machine dependent clock related includes */
67 #if defined(__i386__)
68 #include <machine/clock.h>		/* for tsc_frequency */
69 #include <machine/md_var.h>		/* for cpu_feature */
70 #include <machine/specialreg.h>		/* for CPUID_TSC */
71 #endif /* __i386__ */
72 
73 /*
74  * internal function prototypes
75  */
76 static void	tbr_timeout(void *);
77 static int	altq_enable_locked(struct ifaltq *);
78 static int	altq_disable_locked(struct ifaltq *);
79 static int	altq_detach_locked(struct ifaltq *);
80 static int	tbr_set_locked(struct ifaltq *, struct tb_profile *);
81 
82 int (*altq_input)(struct mbuf *, int) = NULL;
83 static int tbr_timer = 0;	/* token bucket regulator timer */
84 static struct callout tbr_callout;
85 
86 int pfaltq_running;	/* keep track of running state */
87 
88 MALLOC_DEFINE(M_ALTQ, "altq", "ALTQ structures");
89 
90 /*
91  * alternate queueing support routines
92  */
93 
94 /* look up the queue state by the interface name and the queueing type. */
95 void *
96 altq_lookup(const char *name, int type)
97 {
98 	struct ifnet *ifp;
99 
100 	if ((ifp = ifunit(name)) != NULL) {
101 		if (type != ALTQT_NONE && ifp->if_snd.altq_type == type)
102 			return (ifp->if_snd.altq_disc);
103 	}
104 
105 	return (NULL);
106 }
107 
108 int
109 altq_attach(struct ifaltq *ifq, int type, void *discipline,
110 	    int (*enqueue)(struct ifaltq *, struct mbuf *, struct altq_pktattr *),
111 	    struct mbuf *(*dequeue)(struct ifaltq *, struct mbuf *, int),
112 	    int (*request)(struct ifaltq *, int, void *),
113 	    void *clfier,
114 	    void *(*classify)(struct ifaltq *, struct mbuf *,
115 			      struct altq_pktattr *))
116 {
117 	if (!ifq_is_ready(ifq))
118 		return ENXIO;
119 
120 	ifq->altq_type     = type;
121 	ifq->altq_disc     = discipline;
122 	ifq->altq_enqueue  = enqueue;
123 	ifq->altq_dequeue  = dequeue;
124 	ifq->altq_request  = request;
125 	ifq->altq_clfier   = clfier;
126 	ifq->altq_classify = classify;
127 	ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED);
128 	return 0;
129 }
130 
131 static int
132 altq_detach_locked(struct ifaltq *ifq)
133 {
134 	if (!ifq_is_ready(ifq))
135 		return ENXIO;
136 	if (ifq_is_enabled(ifq))
137 		return EBUSY;
138 	if (!ifq_is_attached(ifq))
139 		return (0);
140 
141 	ifq_set_classic(ifq);
142 	ifq->altq_type     = ALTQT_NONE;
143 	ifq->altq_disc     = NULL;
144 	ifq->altq_clfier   = NULL;
145 	ifq->altq_classify = NULL;
146 	ifq->altq_flags &= ALTQF_CANTCHANGE;
147 	return 0;
148 }
149 
150 int
151 altq_detach(struct ifaltq *ifq)
152 {
153 	int error;
154 
155 	ALTQ_LOCK(ifq);
156 	error = altq_detach_locked(ifq);
157 	ALTQ_UNLOCK(ifq);
158 	return error;
159 }
160 
161 static int
162 altq_enable_locked(struct ifaltq *ifq)
163 {
164 	if (!ifq_is_ready(ifq))
165 		return ENXIO;
166 	if (ifq_is_enabled(ifq))
167 		return 0;
168 
169 	ifq_purge_locked(ifq);
170 	KKASSERT(ifq->ifq_len == 0);
171 
172 	ifq->altq_flags |= ALTQF_ENABLED;
173 	if (ifq->altq_clfier != NULL)
174 		ifq->altq_flags |= ALTQF_CLASSIFY;
175 	return 0;
176 }
177 
178 int
179 altq_enable(struct ifaltq *ifq)
180 {
181 	int error;
182 
183 	ALTQ_LOCK(ifq);
184 	error = altq_enable_locked(ifq);
185 	ALTQ_UNLOCK(ifq);
186 	return error;
187 }
188 
189 static int
190 altq_disable_locked(struct ifaltq *ifq)
191 {
192 	if (!ifq_is_enabled(ifq))
193 		return 0;
194 
195 	ifq_purge_locked(ifq);
196 	KKASSERT(ifq->ifq_len == 0);
197 	ifq->altq_flags &= ~(ALTQF_ENABLED|ALTQF_CLASSIFY);
198 	return 0;
199 }
200 
201 int
202 altq_disable(struct ifaltq *ifq)
203 {
204 	int error;
205 
206 	ALTQ_LOCK(ifq);
207 	error = altq_disable_locked(ifq);
208 	ALTQ_UNLOCK(ifq);
209 	return error;
210 }
211 
212 /*
213  * internal representation of token bucket parameters
214  *	rate:	byte_per_unittime << 32
215  *		(((bits_per_sec) / 8) << 32) / machclk_freq
216  *	depth:	byte << 32
217  *
218  */
219 #define	TBR_SHIFT	32
220 #define	TBR_SCALE(x)	((int64_t)(x) << TBR_SHIFT)
221 #define	TBR_UNSCALE(x)	((x) >> TBR_SHIFT)
222 
223 struct mbuf *
224 tbr_dequeue(struct ifaltq *ifq, struct mbuf *mpolled, int op)
225 {
226 	struct tb_regulator *tbr;
227 	struct mbuf *m;
228 	int64_t interval;
229 	uint64_t now;
230 
231 	crit_enter();
232 	tbr = ifq->altq_tbr;
233 	if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) {
234 		/* if this is a remove after poll, bypass tbr check */
235 	} else {
236 		/* update token only when it is negative */
237 		if (tbr->tbr_token <= 0) {
238 			now = read_machclk();
239 			interval = now - tbr->tbr_last;
240 			if (interval >= tbr->tbr_filluptime)
241 				tbr->tbr_token = tbr->tbr_depth;
242 			else {
243 				tbr->tbr_token += interval * tbr->tbr_rate;
244 				if (tbr->tbr_token > tbr->tbr_depth)
245 					tbr->tbr_token = tbr->tbr_depth;
246 			}
247 			tbr->tbr_last = now;
248 		}
249 		/* if token is still negative, don't allow dequeue */
250 		if (tbr->tbr_token <= 0) {
251 			crit_exit();
252 			return (NULL);
253 		}
254 	}
255 
256 	if (ifq_is_enabled(ifq)) {
257 		m = (*ifq->altq_dequeue)(ifq, mpolled, op);
258 	} else if (op == ALTDQ_POLL) {
259 		IF_POLL(ifq, m);
260 	} else {
261 		IF_DEQUEUE(ifq, m);
262 		KKASSERT(mpolled == NULL || mpolled == m);
263 	}
264 
265 	if (m != NULL && op == ALTDQ_REMOVE)
266 		tbr->tbr_token -= TBR_SCALE(m_pktlen(m));
267 	tbr->tbr_lastop = op;
268 	crit_exit();
269 	return (m);
270 }
271 
272 /*
273  * set a token bucket regulator.
274  * if the specified rate is zero, the token bucket regulator is deleted.
275  */
276 static int
277 tbr_set_locked(struct ifaltq *ifq, struct tb_profile *profile)
278 {
279 	struct tb_regulator *tbr, *otbr;
280 
281 	if (machclk_freq == 0)
282 		init_machclk();
283 	if (machclk_freq == 0) {
284 		kprintf("%s: no cpu clock available!\n", __func__);
285 		return (ENXIO);
286 	}
287 
288 	if (profile->rate == 0) {
289 		/* delete this tbr */
290 		if ((tbr = ifq->altq_tbr) == NULL)
291 			return (ENOENT);
292 		ifq->altq_tbr = NULL;
293 		kfree(tbr, M_ALTQ);
294 		return (0);
295 	}
296 
297 	tbr = kmalloc(sizeof(*tbr), M_ALTQ, M_WAITOK | M_ZERO);
298 	tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq;
299 	tbr->tbr_depth = TBR_SCALE(profile->depth);
300 	if (tbr->tbr_rate > 0)
301 		tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate;
302 	else
303 		tbr->tbr_filluptime = 0xffffffffffffffffLL;
304 	tbr->tbr_token = tbr->tbr_depth;
305 	tbr->tbr_last = read_machclk();
306 	tbr->tbr_lastop = ALTDQ_REMOVE;
307 
308 	otbr = ifq->altq_tbr;
309 	ifq->altq_tbr = tbr;	/* set the new tbr */
310 
311 	if (otbr != NULL)
312 		kfree(otbr, M_ALTQ);
313 	else if (tbr_timer == 0) {
314 		callout_reset(&tbr_callout, 1, tbr_timeout, NULL);
315 		tbr_timer = 1;
316 	}
317 	return (0);
318 }
319 
320 int
321 tbr_set(struct ifaltq *ifq, struct tb_profile *profile)
322 {
323 	int error;
324 
325 	ALTQ_LOCK(ifq);
326 	error = tbr_set_locked(ifq, profile);
327 	ALTQ_UNLOCK(ifq);
328 	return error;
329 }
330 
331 /*
332  * tbr_timeout goes through the interface list, and kicks the drivers
333  * if necessary.
334  */
335 static void
336 tbr_timeout(void *arg)
337 {
338 	struct ifnet *ifp;
339 	int active;
340 
341 	active = 0;
342 	crit_enter();
343 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
344 		if (ifp->if_snd.altq_tbr == NULL)
345 			continue;
346 		active++;
347 		if (!ifq_is_empty(&ifp->if_snd) && ifp->if_start != NULL) {
348 			ifnet_serialize_tx(ifp);
349 			(*ifp->if_start)(ifp);
350 			ifnet_deserialize_tx(ifp);
351 		}
352 	}
353 	crit_exit();
354 	if (active > 0)
355 		callout_reset(&tbr_callout, 1, tbr_timeout, NULL);
356 	else
357 		tbr_timer = 0;	/* don't need tbr_timer anymore */
358 }
359 
360 /*
361  * get token bucket regulator profile
362  */
363 int
364 tbr_get(struct ifaltq *ifq, struct tb_profile *profile)
365 {
366 	struct tb_regulator *tbr;
367 
368 	if ((tbr = ifq->altq_tbr) == NULL) {
369 		profile->rate = 0;
370 		profile->depth = 0;
371 	} else {
372 		profile->rate =
373 		    (u_int)TBR_UNSCALE(tbr->tbr_rate * 8 * machclk_freq);
374 		profile->depth = (u_int)TBR_UNSCALE(tbr->tbr_depth);
375 	}
376 	return (0);
377 }
378 
379 /*
380  * attach a discipline to the interface.  if one already exists, it is
381  * overridden.
382  */
383 int
384 altq_pfattach(struct pf_altq *a)
385 {
386 	struct ifaltq *ifq;
387 	struct ifnet *ifp;
388 	int error;
389 
390 	if (a->scheduler == ALTQT_NONE)
391 		return 0;
392 
393 	if (a->altq_disc == NULL)
394 		return EINVAL;
395 
396 	ifp = ifunit(a->ifname);
397 	if (ifp == NULL)
398 		return EINVAL;
399 	ifq = &ifp->if_snd;
400 
401 	ALTQ_LOCK(ifq);
402 
403 	switch (a->scheduler) {
404 #ifdef ALTQ_CBQ
405 	case ALTQT_CBQ:
406 		error = cbq_pfattach(a, ifq);
407 		break;
408 #endif
409 #ifdef ALTQ_PRIQ
410 	case ALTQT_PRIQ:
411 		error = priq_pfattach(a, ifq);
412 		break;
413 #endif
414 #ifdef ALTQ_HFSC
415 	case ALTQT_HFSC:
416 		error = hfsc_pfattach(a, ifq);
417 		break;
418 #endif
419 #ifdef ALTQ_FAIRQ
420 	case ALTQT_FAIRQ:
421 		error = fairq_pfattach(a, ifq);
422 		break;
423 #endif
424 	default:
425 		error = ENXIO;
426 		goto back;
427 	}
428 
429 	/* if the state is running, enable altq */
430 	if (error == 0 && pfaltq_running && ifq->altq_type != ALTQT_NONE &&
431 	    !ifq_is_enabled(ifq))
432 		error = altq_enable_locked(ifq);
433 
434 	/* if altq is already enabled, reset set tokenbucket regulator */
435 	if (error == 0 && ifq_is_enabled(ifq)) {
436 		struct tb_profile tb;
437 
438 		tb.rate = a->ifbandwidth;
439 		tb.depth = a->tbrsize;
440 		error = tbr_set_locked(ifq, &tb);
441 	}
442 back:
443 	ALTQ_UNLOCK(ifq);
444 	return (error);
445 }
446 
447 /*
448  * detach a discipline from the interface.
449  * it is possible that the discipline was already overridden by another
450  * discipline.
451  */
452 int
453 altq_pfdetach(struct pf_altq *a)
454 {
455 	struct ifnet *ifp;
456 	struct ifaltq *ifq;
457 	int error = 0;
458 
459 	ifp = ifunit(a->ifname);
460 	if (ifp == NULL)
461 		return (EINVAL);
462 	ifq = &ifp->if_snd;
463 
464 	/* if this discipline is no longer referenced, just return */
465 	if (a->altq_disc == NULL)
466 		return (0);
467 
468 	ALTQ_LOCK(ifq);
469 
470 	if (a->altq_disc != ifq->altq_disc)
471 		goto back;
472 
473 	if (ifq_is_enabled(ifq))
474 		error = altq_disable_locked(ifq);
475 	if (error == 0)
476 		error = altq_detach_locked(ifq);
477 
478 back:
479 	ALTQ_UNLOCK(ifq);
480 	return (error);
481 }
482 
483 /*
484  * add a discipline or a queue
485  */
486 int
487 altq_add(struct pf_altq *a)
488 {
489 	int error = 0;
490 
491 	if (a->qname[0] != 0)
492 		return (altq_add_queue(a));
493 
494 	if (machclk_freq == 0)
495 		init_machclk();
496 	if (machclk_freq == 0)
497 		panic("altq_add: no cpu clock");
498 
499 	switch (a->scheduler) {
500 #ifdef ALTQ_CBQ
501 	case ALTQT_CBQ:
502 		error = cbq_add_altq(a);
503 		break;
504 #endif
505 #ifdef ALTQ_PRIQ
506 	case ALTQT_PRIQ:
507 		error = priq_add_altq(a);
508 		break;
509 #endif
510 #ifdef ALTQ_HFSC
511 	case ALTQT_HFSC:
512 		error = hfsc_add_altq(a);
513 		break;
514 #endif
515 #ifdef ALTQ_FAIRQ
516 	case ALTQT_FAIRQ:
517 		error = fairq_add_altq(a);
518 		break;
519 #endif
520 	default:
521 		error = ENXIO;
522 	}
523 
524 	return (error);
525 }
526 
527 /*
528  * remove a discipline or a queue
529  */
530 int
531 altq_remove(struct pf_altq *a)
532 {
533 	int error = 0;
534 
535 	if (a->qname[0] != 0)
536 		return (altq_remove_queue(a));
537 
538 	switch (a->scheduler) {
539 #ifdef ALTQ_CBQ
540 	case ALTQT_CBQ:
541 		error = cbq_remove_altq(a);
542 		break;
543 #endif
544 #ifdef ALTQ_PRIQ
545 	case ALTQT_PRIQ:
546 		error = priq_remove_altq(a);
547 		break;
548 #endif
549 #ifdef ALTQ_HFSC
550 	case ALTQT_HFSC:
551 		error = hfsc_remove_altq(a);
552 		break;
553 #endif
554 #ifdef ALTQ_FAIRQ
555 	case ALTQT_FAIRQ:
556 		error = fairq_remove_altq(a);
557 		break;
558 #endif
559 	default:
560 		error = ENXIO;
561 	}
562 
563 	return (error);
564 }
565 
566 /*
567  * add a queue to the discipline
568  */
569 int
570 altq_add_queue(struct pf_altq *a)
571 {
572 	int error = 0;
573 
574 	switch (a->scheduler) {
575 #ifdef ALTQ_CBQ
576 	case ALTQT_CBQ:
577 		error = cbq_add_queue(a);
578 		break;
579 #endif
580 #ifdef ALTQ_PRIQ
581 	case ALTQT_PRIQ:
582 		error = priq_add_queue(a);
583 		break;
584 #endif
585 #ifdef ALTQ_HFSC
586 	case ALTQT_HFSC:
587 		error = hfsc_add_queue(a);
588 		break;
589 #endif
590 #ifdef ALTQ_FAIRQ
591 	case ALTQT_FAIRQ:
592 		error = fairq_add_queue(a);
593 		break;
594 #endif
595 	default:
596 		error = ENXIO;
597 	}
598 
599 	return (error);
600 }
601 
602 /*
603  * remove a queue from the discipline
604  */
605 int
606 altq_remove_queue(struct pf_altq *a)
607 {
608 	int error = 0;
609 
610 	switch (a->scheduler) {
611 #ifdef ALTQ_CBQ
612 	case ALTQT_CBQ:
613 		error = cbq_remove_queue(a);
614 		break;
615 #endif
616 #ifdef ALTQ_PRIQ
617 	case ALTQT_PRIQ:
618 		error = priq_remove_queue(a);
619 		break;
620 #endif
621 #ifdef ALTQ_HFSC
622 	case ALTQT_HFSC:
623 		error = hfsc_remove_queue(a);
624 		break;
625 #endif
626 #ifdef ALTQ_FAIRQ
627 	case ALTQT_FAIRQ:
628 		error = fairq_remove_queue(a);
629 		break;
630 #endif
631 	default:
632 		error = ENXIO;
633 	}
634 
635 	return (error);
636 }
637 
638 /*
639  * get queue statistics
640  */
641 int
642 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes)
643 {
644 	int error = 0;
645 
646 	switch (a->scheduler) {
647 #ifdef ALTQ_CBQ
648 	case ALTQT_CBQ:
649 		error = cbq_getqstats(a, ubuf, nbytes);
650 		break;
651 #endif
652 #ifdef ALTQ_PRIQ
653 	case ALTQT_PRIQ:
654 		error = priq_getqstats(a, ubuf, nbytes);
655 		break;
656 #endif
657 #ifdef ALTQ_HFSC
658 	case ALTQT_HFSC:
659 		error = hfsc_getqstats(a, ubuf, nbytes);
660 		break;
661 #endif
662 #ifdef ALTQ_FAIRQ
663 	case ALTQT_FAIRQ:
664 		error = fairq_getqstats(a, ubuf, nbytes);
665 		break;
666 #endif
667 	default:
668 		error = ENXIO;
669 	}
670 
671 	return (error);
672 }
673 
674 /*
675  * read and write diffserv field in IPv4 or IPv6 header
676  */
677 uint8_t
678 read_dsfield(struct mbuf *m, struct altq_pktattr *pktattr)
679 {
680 	struct mbuf *m0;
681 	uint8_t ds_field = 0;
682 
683 	if (pktattr == NULL ||
684 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
685 		return ((uint8_t)0);
686 
687 	/* verify that pattr_hdr is within the mbuf data */
688 	for (m0 = m; m0 != NULL; m0 = m0->m_next) {
689 		if ((pktattr->pattr_hdr >= m0->m_data) &&
690 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
691 			break;
692 	}
693 	if (m0 == NULL) {
694 		/* ick, pattr_hdr is stale */
695 		pktattr->pattr_af = AF_UNSPEC;
696 #ifdef ALTQ_DEBUG
697 		kprintf("read_dsfield: can't locate header!\n");
698 #endif
699 		return ((uint8_t)0);
700 	}
701 
702 	if (pktattr->pattr_af == AF_INET) {
703 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
704 
705 		if (ip->ip_v != 4)
706 			return ((uint8_t)0);	/* version mismatch! */
707 		ds_field = ip->ip_tos;
708 	}
709 #ifdef INET6
710 	else if (pktattr->pattr_af == AF_INET6) {
711 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
712 		uint32_t flowlabel;
713 
714 		flowlabel = ntohl(ip6->ip6_flow);
715 		if ((flowlabel >> 28) != 6)
716 			return ((uint8_t)0);	/* version mismatch! */
717 		ds_field = (flowlabel >> 20) & 0xff;
718 	}
719 #endif
720 	return (ds_field);
721 }
722 
723 void
724 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, uint8_t dsfield)
725 {
726 	struct mbuf *m0;
727 
728 	if (pktattr == NULL ||
729 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
730 		return;
731 
732 	/* verify that pattr_hdr is within the mbuf data */
733 	for (m0 = m; m0 != NULL; m0 = m0->m_next) {
734 		if ((pktattr->pattr_hdr >= m0->m_data) &&
735 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
736 			break;
737 	}
738 	if (m0 == NULL) {
739 		/* ick, pattr_hdr is stale */
740 		pktattr->pattr_af = AF_UNSPEC;
741 #ifdef ALTQ_DEBUG
742 		kprintf("write_dsfield: can't locate header!\n");
743 #endif
744 		return;
745 	}
746 
747 	if (pktattr->pattr_af == AF_INET) {
748 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
749 		uint8_t old;
750 		int32_t sum;
751 
752 		if (ip->ip_v != 4)
753 			return;		/* version mismatch! */
754 		old = ip->ip_tos;
755 		dsfield |= old & 3;	/* leave CU bits */
756 		if (old == dsfield)
757 			return;
758 		ip->ip_tos = dsfield;
759 		/*
760 		 * update checksum (from RFC1624)
761 		 *	   HC' = ~(~HC + ~m + m')
762 		 */
763 		sum = ~ntohs(ip->ip_sum) & 0xffff;
764 		sum += 0xff00 + (~old & 0xff) + dsfield;
765 		sum = (sum >> 16) + (sum & 0xffff);
766 		sum += (sum >> 16);  /* add carry */
767 
768 		ip->ip_sum = htons(~sum & 0xffff);
769 	}
770 #ifdef INET6
771 	else if (pktattr->pattr_af == AF_INET6) {
772 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
773 		uint32_t flowlabel;
774 
775 		flowlabel = ntohl(ip6->ip6_flow);
776 		if ((flowlabel >> 28) != 6)
777 			return;		/* version mismatch! */
778 		flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20);
779 		ip6->ip6_flow = htonl(flowlabel);
780 	}
781 #endif
782 }
783 
784 /*
785  * high resolution clock support taking advantage of a machine dependent
786  * high resolution time counter (e.g., timestamp counter of intel pentium).
787  * we assume
788  *  - 64-bit-long monotonically-increasing counter
789  *  - frequency range is 100M-4GHz (CPU speed)
790  */
791 /* if pcc is not available or disabled, emulate 256MHz using microtime() */
792 #define	MACHCLK_SHIFT	8
793 
794 int machclk_usepcc;
795 uint64_t machclk_freq = 0;
796 uint32_t machclk_per_tick = 0;
797 
798 void
799 init_machclk(void)
800 {
801 	callout_init(&tbr_callout);
802 
803 	machclk_usepcc = 1;
804 
805 #if !defined(__i386__) || defined(ALTQ_NOPCC)
806 	machclk_usepcc = 0;
807 #elif defined(__DragonFly__) && defined(SMP)
808 	machclk_usepcc = 0;
809 #elif defined(__i386__)
810 	/* check if TSC is available */
811 	if (machclk_usepcc == 1 && (cpu_feature & CPUID_TSC) == 0)
812 		machclk_usepcc = 0;
813 #endif
814 
815 	if (machclk_usepcc == 0) {
816 		/* emulate 256MHz using microtime() */
817 		machclk_freq = 1000000LLU << MACHCLK_SHIFT;
818 		machclk_per_tick = machclk_freq / hz;
819 #ifdef ALTQ_DEBUG
820 		kprintf("altq: emulate %lluHz cpu clock\n", machclk_freq);
821 #endif
822 		return;
823 	}
824 
825 	/*
826 	 * if the clock frequency (of Pentium TSC or Alpha PCC) is
827 	 * accessible, just use it.
828 	 */
829 #if _RDTSC_SUPPORTED
830 	if (cpu_feature & CPUID_TSC)
831 		machclk_freq = (uint64_t)tsc_frequency;
832 #endif
833 
834 	/*
835 	 * if we don't know the clock frequency, measure it.
836 	 */
837 	if (machclk_freq == 0) {
838 		static int	wait;
839 		struct timeval	tv_start, tv_end;
840 		uint64_t	start, end, diff;
841 		int		timo;
842 
843 		microtime(&tv_start);
844 		start = read_machclk();
845 		timo = hz;	/* 1 sec */
846 		tsleep(&wait, PCATCH, "init_machclk", timo);
847 		microtime(&tv_end);
848 		end = read_machclk();
849 		diff = (uint64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000
850 		    + tv_end.tv_usec - tv_start.tv_usec;
851 		if (diff != 0)
852 			machclk_freq = (end - start) * 1000000 / diff;
853 	}
854 
855 	machclk_per_tick = machclk_freq / hz;
856 
857 #ifdef ALTQ_DEBUG
858 	kprintf("altq: CPU clock: %lluHz\n", machclk_freq);
859 #endif
860 }
861 
862 uint64_t
863 read_machclk(void)
864 {
865 	uint64_t val;
866 
867 	if (machclk_usepcc) {
868 #ifdef _RDTSC_SUPPORTED_
869 		val = rdtsc();
870 #else
871 		panic("read_machclk");
872 #endif
873 	} else {
874 		struct timeval tv;
875 
876 		microtime(&tv);
877 		val = (((uint64_t)(tv.tv_sec - boottime.tv_sec) * 1000000
878 		    + tv.tv_usec) << MACHCLK_SHIFT);
879 	}
880 	return (val);
881 }
882