xref: /dragonfly/sys/net/altq/altq_subr.c (revision 66e72101)
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 #include <machine/clock.h>		/* for tsc_frequency */
68 #include <machine/md_var.h>		/* for cpu_feature */
69 #include <machine/specialreg.h>		/* for CPUID_TSC */
70 
71 /*
72  * internal function prototypes
73  */
74 static void	tbr_timeout(void *);
75 static int	altq_enable_locked(struct ifaltq *);
76 static int	altq_disable_locked(struct ifaltq *);
77 static int	altq_detach_locked(struct ifaltq *);
78 static int	tbr_set_locked(struct ifaltq *, struct tb_profile *);
79 
80 int (*altq_input)(struct mbuf *, int) = NULL;
81 static int tbr_timer = 0;	/* token bucket regulator timer */
82 static struct callout tbr_callout;
83 
84 int pfaltq_running;	/* keep track of running state */
85 
86 MALLOC_DEFINE(M_ALTQ, "altq", "ALTQ structures");
87 
88 /*
89  * alternate queueing support routines
90  */
91 
92 /* look up the queue state by the interface name and the queueing type. */
93 void *
94 altq_lookup(const char *name, int type)
95 {
96 	struct ifnet *ifp;
97 
98 	if ((ifp = ifunit(name)) != NULL) {
99 		if (type != ALTQT_NONE && ifp->if_snd.altq_type == type)
100 			return (ifp->if_snd.altq_disc);
101 	}
102 
103 	return (NULL);
104 }
105 
106 int
107 altq_attach(struct ifaltq *ifq, int type, void *discipline,
108     altq_mapsubq_t mapsubq,
109     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request,
110     void *clfier,
111     void *(*classify)(struct ifaltq *, struct mbuf *, struct altq_pktattr *))
112 {
113 	if (!ifq_is_ready(ifq))
114 		return ENXIO;
115 
116 	ifq->altq_type     = type;
117 	ifq->altq_disc     = discipline;
118 	ifq->altq_clfier   = clfier;
119 	ifq->altq_classify = classify;
120 	ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED);
121 	ifq_set_methods(ifq, mapsubq, enqueue, dequeue, request);
122 	return 0;
123 }
124 
125 static int
126 altq_detach_locked(struct ifaltq *ifq)
127 {
128 	if (!ifq_is_ready(ifq))
129 		return ENXIO;
130 	if (ifq_is_enabled(ifq))
131 		return EBUSY;
132 	if (!ifq_is_attached(ifq))
133 		return (0);
134 
135 	ifq_set_classic(ifq);
136 	ifq->altq_type     = ALTQT_NONE;
137 	ifq->altq_disc     = NULL;
138 	ifq->altq_clfier   = NULL;
139 	ifq->altq_classify = NULL;
140 	ifq->altq_flags &= ALTQF_CANTCHANGE;
141 	return 0;
142 }
143 
144 int
145 altq_detach(struct ifaltq *ifq)
146 {
147 	int error;
148 
149 	ifq_lock_all(ifq);
150 	error = altq_detach_locked(ifq);
151 	ifq_unlock_all(ifq);
152 	return error;
153 }
154 
155 static int
156 altq_enable_locked(struct ifaltq *ifq)
157 {
158 	if (!ifq_is_ready(ifq))
159 		return ENXIO;
160 	if (ifq_is_enabled(ifq))
161 		return 0;
162 
163 	ifq_purge_all_locked(ifq);
164 
165 	ifq->altq_flags |= ALTQF_ENABLED;
166 	if (ifq->altq_clfier != NULL)
167 		ifq->altq_flags |= ALTQF_CLASSIFY;
168 	return 0;
169 }
170 
171 int
172 altq_enable(struct ifaltq *ifq)
173 {
174 	int error;
175 
176 	ifq_lock_all(ifq);
177 	error = altq_enable_locked(ifq);
178 	ifq_unlock_all(ifq);
179 	return error;
180 }
181 
182 static int
183 altq_disable_locked(struct ifaltq *ifq)
184 {
185 	if (!ifq_is_enabled(ifq))
186 		return 0;
187 
188 	ifq_purge_all_locked(ifq);
189 	ifq->altq_flags &= ~(ALTQF_ENABLED|ALTQF_CLASSIFY);
190 	return 0;
191 }
192 
193 int
194 altq_disable(struct ifaltq *ifq)
195 {
196 	int error;
197 
198 	ifq_lock_all(ifq);
199 	error = altq_disable_locked(ifq);
200 	ifq_unlock_all(ifq);
201 	return error;
202 }
203 
204 /*
205  * internal representation of token bucket parameters
206  *	rate:	byte_per_unittime << 32
207  *		(((bits_per_sec) / 8) << 32) / machclk_freq
208  *	depth:	byte << 32
209  *
210  */
211 #define	TBR_SHIFT	32
212 #define	TBR_SCALE(x)	((int64_t)(x) << TBR_SHIFT)
213 #define	TBR_UNSCALE(x)	((x) >> TBR_SHIFT)
214 
215 struct mbuf *
216 tbr_dequeue(struct ifaltq_subque *ifsq, struct mbuf *mpolled, int op)
217 {
218 	struct ifaltq *ifq = ifsq->ifsq_altq;
219 	struct tb_regulator *tbr;
220 	struct mbuf *m;
221 	int64_t interval;
222 	uint64_t now;
223 
224 	if (ifsq_get_index(ifsq) != ALTQ_SUBQ_INDEX_DEFAULT) {
225 		/*
226 		 * Race happened, the unrelated subqueue was
227 		 * picked during the packet scheduler transition.
228 		 */
229 		ifsq_classic_request(ifsq, ALTRQ_PURGE, NULL);
230 		return NULL;
231 	}
232 
233 	crit_enter();
234 	tbr = ifq->altq_tbr;
235 	if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) {
236 		/* if this is a remove after poll, bypass tbr check */
237 	} else {
238 		/* update token only when it is negative */
239 		if (tbr->tbr_token <= 0) {
240 			now = read_machclk();
241 			interval = now - tbr->tbr_last;
242 			if (interval >= tbr->tbr_filluptime)
243 				tbr->tbr_token = tbr->tbr_depth;
244 			else {
245 				tbr->tbr_token += interval * tbr->tbr_rate;
246 				if (tbr->tbr_token > tbr->tbr_depth)
247 					tbr->tbr_token = tbr->tbr_depth;
248 			}
249 			tbr->tbr_last = now;
250 		}
251 		/* if token is still negative, don't allow dequeue */
252 		if (tbr->tbr_token <= 0) {
253 			crit_exit();
254 			return (NULL);
255 		}
256 	}
257 
258 	if (ifq_is_enabled(ifq)) {
259 		m = (*ifsq->ifsq_dequeue)(ifsq, mpolled, op);
260 	} else if (op == ALTDQ_POLL) {
261 		IF_POLL(ifsq, m);
262 	} else {
263 		IF_DEQUEUE(ifsq, m);
264 		KKASSERT(mpolled == NULL || mpolled == m);
265 	}
266 
267 	if (m != NULL && op == ALTDQ_REMOVE)
268 		tbr->tbr_token -= TBR_SCALE(m_pktlen(m));
269 	tbr->tbr_lastop = op;
270 	crit_exit();
271 	return (m);
272 }
273 
274 /*
275  * set a token bucket regulator.
276  * if the specified rate is zero, the token bucket regulator is deleted.
277  */
278 static int
279 tbr_set_locked(struct ifaltq *ifq, struct tb_profile *profile)
280 {
281 	struct tb_regulator *tbr, *otbr;
282 
283 	if (machclk_freq == 0)
284 		init_machclk();
285 	if (machclk_freq == 0) {
286 		kprintf("%s: no cpu clock available!\n", __func__);
287 		return (ENXIO);
288 	}
289 
290 	if (profile->rate == 0) {
291 		/* delete this tbr */
292 		if ((tbr = ifq->altq_tbr) == NULL)
293 			return (ENOENT);
294 		ifq->altq_tbr = NULL;
295 		kfree(tbr, M_ALTQ);
296 		return (0);
297 	}
298 
299 	tbr = kmalloc(sizeof(*tbr), M_ALTQ, M_WAITOK | M_ZERO);
300 	tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq;
301 	tbr->tbr_depth = TBR_SCALE(profile->depth);
302 	if (tbr->tbr_rate > 0)
303 		tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate;
304 	else
305 		tbr->tbr_filluptime = 0xffffffffffffffffLL;
306 	tbr->tbr_token = tbr->tbr_depth;
307 	tbr->tbr_last = read_machclk();
308 	tbr->tbr_lastop = ALTDQ_REMOVE;
309 
310 	otbr = ifq->altq_tbr;
311 	ifq->altq_tbr = tbr;	/* set the new tbr */
312 
313 	if (otbr != NULL)
314 		kfree(otbr, M_ALTQ);
315 	else if (tbr_timer == 0) {
316 		callout_reset(&tbr_callout, 1, tbr_timeout, NULL);
317 		tbr_timer = 1;
318 	}
319 	return (0);
320 }
321 
322 int
323 tbr_set(struct ifaltq *ifq, struct tb_profile *profile)
324 {
325 	int error;
326 
327 	ifq_lock_all(ifq);
328 	error = tbr_set_locked(ifq, profile);
329 	ifq_unlock_all(ifq);
330 	return error;
331 }
332 
333 /*
334  * tbr_timeout goes through the interface list, and kicks the drivers
335  * if necessary.
336  */
337 static void
338 tbr_timeout(void *arg)
339 {
340 	struct ifnet *ifp;
341 	int active;
342 
343 	active = 0;
344 	crit_enter();
345 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
346 		struct ifaltq_subque *ifsq;
347 
348 		if (ifp->if_snd.altq_tbr == NULL)
349 			continue;
350 
351 		ifsq = &ifp->if_snd.altq_subq[ALTQ_SUBQ_INDEX_DEFAULT];
352 		active++;
353 		if (!ifsq_is_empty(ifsq) && ifp->if_start != NULL) {
354 			ifsq_serialize_hw(ifsq);
355 			(*ifp->if_start)(ifp, ifsq);
356 			ifsq_deserialize_hw(ifsq);
357 		}
358 	}
359 	crit_exit();
360 	if (active > 0)
361 		callout_reset(&tbr_callout, 1, tbr_timeout, NULL);
362 	else
363 		tbr_timer = 0;	/* don't need tbr_timer anymore */
364 }
365 
366 /*
367  * get token bucket regulator profile
368  */
369 int
370 tbr_get(struct ifaltq *ifq, struct tb_profile *profile)
371 {
372 	struct tb_regulator *tbr;
373 
374 	if ((tbr = ifq->altq_tbr) == NULL) {
375 		profile->rate = 0;
376 		profile->depth = 0;
377 	} else {
378 		profile->rate =
379 		    (u_int)TBR_UNSCALE(tbr->tbr_rate * 8 * machclk_freq);
380 		profile->depth = (u_int)TBR_UNSCALE(tbr->tbr_depth);
381 	}
382 	return (0);
383 }
384 
385 /*
386  * attach a discipline to the interface.  if one already exists, it is
387  * overridden.
388  */
389 int
390 altq_pfattach(struct pf_altq *a)
391 {
392 	struct ifaltq *ifq;
393 	struct ifnet *ifp;
394 	int error;
395 
396 	if (a->scheduler == ALTQT_NONE)
397 		return 0;
398 
399 	if (a->altq_disc == NULL)
400 		return EINVAL;
401 
402 	ifp = ifunit(a->ifname);
403 	if (ifp == NULL)
404 		return EINVAL;
405 	ifq = &ifp->if_snd;
406 
407 	ifq_lock_all(ifq);
408 
409 	switch (a->scheduler) {
410 #ifdef ALTQ_CBQ
411 	case ALTQT_CBQ:
412 		error = cbq_pfattach(a, ifq);
413 		break;
414 #endif
415 #ifdef ALTQ_PRIQ
416 	case ALTQT_PRIQ:
417 		error = priq_pfattach(a, ifq);
418 		break;
419 #endif
420 #ifdef ALTQ_HFSC
421 	case ALTQT_HFSC:
422 		error = hfsc_pfattach(a, ifq);
423 		break;
424 #endif
425 #ifdef ALTQ_FAIRQ
426 	case ALTQT_FAIRQ:
427 		error = fairq_pfattach(a, ifq);
428 		break;
429 #endif
430 	default:
431 		error = ENXIO;
432 		goto back;
433 	}
434 
435 	/* if the state is running, enable altq */
436 	if (error == 0 && pfaltq_running && ifq->altq_type != ALTQT_NONE &&
437 	    !ifq_is_enabled(ifq))
438 		error = altq_enable_locked(ifq);
439 
440 	/* if altq is already enabled, reset set tokenbucket regulator */
441 	if (error == 0 && ifq_is_enabled(ifq)) {
442 		struct tb_profile tb;
443 
444 		tb.rate = a->ifbandwidth;
445 		tb.depth = a->tbrsize;
446 		error = tbr_set_locked(ifq, &tb);
447 	}
448 back:
449 	ifq_unlock_all(ifq);
450 	return (error);
451 }
452 
453 /*
454  * detach a discipline from the interface.
455  * it is possible that the discipline was already overridden by another
456  * discipline.
457  */
458 int
459 altq_pfdetach(struct pf_altq *a)
460 {
461 	struct ifnet *ifp;
462 	struct ifaltq *ifq;
463 	int error = 0;
464 
465 	ifp = ifunit(a->ifname);
466 	if (ifp == NULL)
467 		return (EINVAL);
468 	ifq = &ifp->if_snd;
469 
470 	/* if this discipline is no longer referenced, just return */
471 	if (a->altq_disc == NULL)
472 		return (0);
473 
474 	ifq_lock_all(ifq);
475 
476 	if (a->altq_disc != ifq->altq_disc)
477 		goto back;
478 
479 	if (ifq_is_enabled(ifq))
480 		error = altq_disable_locked(ifq);
481 	if (error == 0)
482 		error = altq_detach_locked(ifq);
483 
484 back:
485 	ifq_unlock_all(ifq);
486 	return (error);
487 }
488 
489 /*
490  * add a discipline or a queue
491  */
492 int
493 altq_add(struct pf_altq *a)
494 {
495 	int error = 0;
496 
497 	if (a->qname[0] != 0)
498 		return (altq_add_queue(a));
499 
500 	if (machclk_freq == 0)
501 		init_machclk();
502 	if (machclk_freq == 0)
503 		panic("altq_add: no cpu clock");
504 
505 	switch (a->scheduler) {
506 #ifdef ALTQ_CBQ
507 	case ALTQT_CBQ:
508 		error = cbq_add_altq(a);
509 		break;
510 #endif
511 #ifdef ALTQ_PRIQ
512 	case ALTQT_PRIQ:
513 		error = priq_add_altq(a);
514 		break;
515 #endif
516 #ifdef ALTQ_HFSC
517 	case ALTQT_HFSC:
518 		error = hfsc_add_altq(a);
519 		break;
520 #endif
521 #ifdef ALTQ_FAIRQ
522 	case ALTQT_FAIRQ:
523 		error = fairq_add_altq(a);
524 		break;
525 #endif
526 	default:
527 		error = ENXIO;
528 	}
529 
530 	return (error);
531 }
532 
533 /*
534  * remove a discipline or a queue
535  */
536 int
537 altq_remove(struct pf_altq *a)
538 {
539 	int error = 0;
540 
541 	if (a->qname[0] != 0)
542 		return (altq_remove_queue(a));
543 
544 	switch (a->scheduler) {
545 #ifdef ALTQ_CBQ
546 	case ALTQT_CBQ:
547 		error = cbq_remove_altq(a);
548 		break;
549 #endif
550 #ifdef ALTQ_PRIQ
551 	case ALTQT_PRIQ:
552 		error = priq_remove_altq(a);
553 		break;
554 #endif
555 #ifdef ALTQ_HFSC
556 	case ALTQT_HFSC:
557 		error = hfsc_remove_altq(a);
558 		break;
559 #endif
560 #ifdef ALTQ_FAIRQ
561 	case ALTQT_FAIRQ:
562 		error = fairq_remove_altq(a);
563 		break;
564 #endif
565 	default:
566 		error = ENXIO;
567 	}
568 
569 	return (error);
570 }
571 
572 /*
573  * add a queue to the discipline
574  */
575 int
576 altq_add_queue(struct pf_altq *a)
577 {
578 	int error = 0;
579 
580 	switch (a->scheduler) {
581 #ifdef ALTQ_CBQ
582 	case ALTQT_CBQ:
583 		error = cbq_add_queue(a);
584 		break;
585 #endif
586 #ifdef ALTQ_PRIQ
587 	case ALTQT_PRIQ:
588 		error = priq_add_queue(a);
589 		break;
590 #endif
591 #ifdef ALTQ_HFSC
592 	case ALTQT_HFSC:
593 		error = hfsc_add_queue(a);
594 		break;
595 #endif
596 #ifdef ALTQ_FAIRQ
597 	case ALTQT_FAIRQ:
598 		error = fairq_add_queue(a);
599 		break;
600 #endif
601 	default:
602 		error = ENXIO;
603 	}
604 
605 	return (error);
606 }
607 
608 /*
609  * remove a queue from the discipline
610  */
611 int
612 altq_remove_queue(struct pf_altq *a)
613 {
614 	int error = 0;
615 
616 	switch (a->scheduler) {
617 #ifdef ALTQ_CBQ
618 	case ALTQT_CBQ:
619 		error = cbq_remove_queue(a);
620 		break;
621 #endif
622 #ifdef ALTQ_PRIQ
623 	case ALTQT_PRIQ:
624 		error = priq_remove_queue(a);
625 		break;
626 #endif
627 #ifdef ALTQ_HFSC
628 	case ALTQT_HFSC:
629 		error = hfsc_remove_queue(a);
630 		break;
631 #endif
632 #ifdef ALTQ_FAIRQ
633 	case ALTQT_FAIRQ:
634 		error = fairq_remove_queue(a);
635 		break;
636 #endif
637 	default:
638 		error = ENXIO;
639 	}
640 
641 	return (error);
642 }
643 
644 /*
645  * get queue statistics
646  */
647 int
648 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes)
649 {
650 	int error = 0;
651 
652 	switch (a->scheduler) {
653 #ifdef ALTQ_CBQ
654 	case ALTQT_CBQ:
655 		error = cbq_getqstats(a, ubuf, nbytes);
656 		break;
657 #endif
658 #ifdef ALTQ_PRIQ
659 	case ALTQT_PRIQ:
660 		error = priq_getqstats(a, ubuf, nbytes);
661 		break;
662 #endif
663 #ifdef ALTQ_HFSC
664 	case ALTQT_HFSC:
665 		error = hfsc_getqstats(a, ubuf, nbytes);
666 		break;
667 #endif
668 #ifdef ALTQ_FAIRQ
669 	case ALTQT_FAIRQ:
670 		error = fairq_getqstats(a, ubuf, nbytes);
671 		break;
672 #endif
673 	default:
674 		error = ENXIO;
675 	}
676 
677 	return (error);
678 }
679 
680 /*
681  * read and write diffserv field in IPv4 or IPv6 header
682  */
683 uint8_t
684 read_dsfield(struct mbuf *m, struct altq_pktattr *pktattr)
685 {
686 	struct mbuf *m0;
687 	uint8_t ds_field = 0;
688 
689 	if (pktattr == NULL ||
690 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
691 		return ((uint8_t)0);
692 
693 	/* verify that pattr_hdr is within the mbuf data */
694 	for (m0 = m; m0 != NULL; m0 = m0->m_next) {
695 		if ((pktattr->pattr_hdr >= m0->m_data) &&
696 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
697 			break;
698 	}
699 	if (m0 == NULL) {
700 		/* ick, pattr_hdr is stale */
701 		pktattr->pattr_af = AF_UNSPEC;
702 #ifdef ALTQ_DEBUG
703 		kprintf("read_dsfield: can't locate header!\n");
704 #endif
705 		return ((uint8_t)0);
706 	}
707 
708 	if (pktattr->pattr_af == AF_INET) {
709 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
710 
711 		if (ip->ip_v != 4)
712 			return ((uint8_t)0);	/* version mismatch! */
713 		ds_field = ip->ip_tos;
714 	}
715 #ifdef INET6
716 	else if (pktattr->pattr_af == AF_INET6) {
717 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
718 		uint32_t flowlabel;
719 
720 		flowlabel = ntohl(ip6->ip6_flow);
721 		if ((flowlabel >> 28) != 6)
722 			return ((uint8_t)0);	/* version mismatch! */
723 		ds_field = (flowlabel >> 20) & 0xff;
724 	}
725 #endif
726 	return (ds_field);
727 }
728 
729 void
730 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, uint8_t dsfield)
731 {
732 	struct mbuf *m0;
733 
734 	if (pktattr == NULL ||
735 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
736 		return;
737 
738 	/* verify that pattr_hdr is within the mbuf data */
739 	for (m0 = m; m0 != NULL; m0 = m0->m_next) {
740 		if ((pktattr->pattr_hdr >= m0->m_data) &&
741 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
742 			break;
743 	}
744 	if (m0 == NULL) {
745 		/* ick, pattr_hdr is stale */
746 		pktattr->pattr_af = AF_UNSPEC;
747 #ifdef ALTQ_DEBUG
748 		kprintf("write_dsfield: can't locate header!\n");
749 #endif
750 		return;
751 	}
752 
753 	if (pktattr->pattr_af == AF_INET) {
754 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
755 		uint8_t old;
756 		int32_t sum;
757 
758 		if (ip->ip_v != 4)
759 			return;		/* version mismatch! */
760 		old = ip->ip_tos;
761 		dsfield |= old & 3;	/* leave CU bits */
762 		if (old == dsfield)
763 			return;
764 		ip->ip_tos = dsfield;
765 		/*
766 		 * update checksum (from RFC1624)
767 		 *	   HC' = ~(~HC + ~m + m')
768 		 */
769 		sum = ~ntohs(ip->ip_sum) & 0xffff;
770 		sum += 0xff00 + (~old & 0xff) + dsfield;
771 		sum = (sum >> 16) + (sum & 0xffff);
772 		sum += (sum >> 16);  /* add carry */
773 
774 		ip->ip_sum = htons(~sum & 0xffff);
775 	}
776 #ifdef INET6
777 	else if (pktattr->pattr_af == AF_INET6) {
778 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
779 		uint32_t flowlabel;
780 
781 		flowlabel = ntohl(ip6->ip6_flow);
782 		if ((flowlabel >> 28) != 6)
783 			return;		/* version mismatch! */
784 		flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20);
785 		ip6->ip6_flow = htonl(flowlabel);
786 	}
787 #endif
788 }
789 
790 /*
791  * high resolution clock support taking advantage of a machine dependent
792  * high resolution time counter (e.g., timestamp counter of intel pentium).
793  * we assume
794  *  - 64-bit-long monotonically-increasing counter
795  *  - frequency range is 100M-4GHz (CPU speed)
796  */
797 /* if pcc is not available or disabled, emulate 256MHz using microtime() */
798 #define	MACHCLK_SHIFT	8
799 
800 int machclk_usepcc;
801 uint64_t machclk_freq = 0;
802 uint32_t machclk_per_tick = 0;
803 
804 void
805 init_machclk(void)
806 {
807 	callout_init(&tbr_callout);
808 
809 	machclk_usepcc = 1;
810 
811 #if !defined(__i386__) || defined(ALTQ_NOPCC)
812 	machclk_usepcc = 0;
813 #elif defined(__DragonFly__)
814 	machclk_usepcc = 0;
815 #elif defined(__i386__)
816 	/* check if TSC is available */
817 	if (machclk_usepcc == 1 && (cpu_feature & CPUID_TSC) == 0)
818 		machclk_usepcc = 0;
819 #endif
820 
821 	if (machclk_usepcc == 0) {
822 		/* emulate 256MHz using microtime() */
823 		machclk_freq = 1000000LLU << MACHCLK_SHIFT;
824 		machclk_per_tick = machclk_freq / hz;
825 #ifdef ALTQ_DEBUG
826 		kprintf("altq: emulate %juHz cpu clock\n", (uintmax_t)machclk_freq);
827 #endif
828 		return;
829 	}
830 
831 	/*
832 	 * if the clock frequency (of Pentium TSC or Alpha PCC) is
833 	 * accessible, just use it.
834 	 */
835 #ifdef _RDTSC_SUPPORTED_
836 	if (cpu_feature & CPUID_TSC)
837 		machclk_freq = (uint64_t)tsc_frequency;
838 #endif
839 
840 	/*
841 	 * if we don't know the clock frequency, measure it.
842 	 */
843 	if (machclk_freq == 0) {
844 		static int	wait;
845 		struct timeval	tv_start, tv_end;
846 		uint64_t	start, end, diff;
847 		int		timo;
848 
849 		microtime(&tv_start);
850 		start = read_machclk();
851 		timo = hz;	/* 1 sec */
852 		tsleep(&wait, PCATCH, "init_machclk", timo);
853 		microtime(&tv_end);
854 		end = read_machclk();
855 		diff = (uint64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000
856 		    + tv_end.tv_usec - tv_start.tv_usec;
857 		if (diff != 0)
858 			machclk_freq = (end - start) * 1000000 / diff;
859 	}
860 
861 	machclk_per_tick = machclk_freq / hz;
862 
863 #ifdef ALTQ_DEBUG
864 	kprintf("altq: CPU clock: %juHz\n", (uintmax_t)machclk_freq);
865 #endif
866 }
867 
868 uint64_t
869 read_machclk(void)
870 {
871 	uint64_t val;
872 
873 	if (machclk_usepcc) {
874 #ifdef _RDTSC_SUPPORTED_
875 		val = rdtsc();
876 #else
877 		panic("read_machclk");
878 #endif
879 	} else {
880 		struct timeval tv;
881 
882 		microtime(&tv);
883 		val = (((uint64_t)(tv.tv_sec - boottime.tv_sec) * 1000000
884 		    + tv.tv_usec) << MACHCLK_SHIFT);
885 	}
886 	return (val);
887 }
888 
889