xref: /netbsd-src/sys/altq/altq_subr.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: altq_subr.c,v 1.24 2007/10/19 12:16:36 ad Exp $	*/
2 /*	$KAME: altq_subr.c,v 1.24 2005/04/13 03:44:25 suz 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 <sys/cdefs.h>
31 __KERNEL_RCSID(0, "$NetBSD: altq_subr.c,v 1.24 2007/10/19 12:16:36 ad Exp $");
32 
33 #ifdef _KERNEL_OPT
34 #include "opt_altq.h"
35 #include "opt_inet.h"
36 #include "pf.h"
37 #endif
38 
39 #include <sys/param.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/systm.h>
43 #include <sys/proc.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/kernel.h>
47 #include <sys/errno.h>
48 #include <sys/syslog.h>
49 #include <sys/sysctl.h>
50 #include <sys/queue.h>
51 
52 #include <net/if.h>
53 #include <net/if_dl.h>
54 #include <net/if_types.h>
55 
56 #include <netinet/in.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/ip.h>
59 #ifdef INET6
60 #include <netinet/ip6.h>
61 #endif
62 #include <netinet/tcp.h>
63 #include <netinet/udp.h>
64 
65 #if NPF > 0
66 #include <net/pfvar.h>
67 #endif
68 #include <altq/altq.h>
69 #ifdef ALTQ3_COMPAT
70 #include <altq/altq_conf.h>
71 #endif
72 
73 /* machine dependent clock related includes */
74 #ifdef __FreeBSD__
75 #include "opt_cpu.h"	/* for FreeBSD-2.2.8 to get i586_ctr_freq */
76 #include <machine/clock.h>
77 #endif
78 #if defined(__i386__)
79 #include <machine/cpufunc.h>		/* for pentium tsc */
80 #include <machine/specialreg.h>		/* for CPUID_TSC */
81 #ifdef __FreeBSD__
82 #include <machine/md_var.h>		/* for cpu_feature */
83 #elif defined(__NetBSD__) || defined(__OpenBSD__)
84 #include <sys/cpu.h>		/* for cpu_feature */
85 #endif
86 #endif /* __i386__ */
87 
88 /*
89  * internal function prototypes
90  */
91 static void	tbr_timeout(void *);
92 int (*altq_input)(struct mbuf *, int) = NULL;
93 static int tbr_timer = 0;	/* token bucket regulator timer */
94 static struct callout tbr_callout;
95 
96 #ifdef ALTQ3_CLFIER_COMPAT
97 static int 	extract_ports4(struct mbuf *, struct ip *, struct flowinfo_in *);
98 #ifdef INET6
99 static int 	extract_ports6(struct mbuf *, struct ip6_hdr *,
100 			       struct flowinfo_in6 *);
101 #endif
102 static int	apply_filter4(u_int32_t, struct flow_filter *,
103 			      struct flowinfo_in *);
104 static int	apply_ppfilter4(u_int32_t, struct flow_filter *,
105 				struct flowinfo_in *);
106 #ifdef INET6
107 static int	apply_filter6(u_int32_t, struct flow_filter6 *,
108 			      struct flowinfo_in6 *);
109 #endif
110 static int	apply_tosfilter4(u_int32_t, struct flow_filter *,
111 				 struct flowinfo_in *);
112 static u_long	get_filt_handle(struct acc_classifier *, int);
113 static struct acc_filter *filth_to_filtp(struct acc_classifier *, u_long);
114 static u_int32_t filt2fibmask(struct flow_filter *);
115 
116 static void 	ip4f_cache(struct ip *, struct flowinfo_in *);
117 static int 	ip4f_lookup(struct ip *, struct flowinfo_in *);
118 static int 	ip4f_init(void);
119 static struct ip4_frag	*ip4f_alloc(void);
120 static void 	ip4f_free(struct ip4_frag *);
121 #endif /* ALTQ3_CLFIER_COMPAT */
122 
123 /*
124  * alternate queueing support routines
125  */
126 
127 /* look up the queue state by the interface name and the queueing type. */
128 void *
129 altq_lookup(char *name, int type)
130 {
131 	struct ifnet *ifp;
132 
133 	if ((ifp = ifunit(name)) != NULL) {
134 		if (type != ALTQT_NONE && ifp->if_snd.altq_type == type)
135 			return (ifp->if_snd.altq_disc);
136 	}
137 
138 	return NULL;
139 }
140 
141 int
142 altq_attach(struct ifaltq *ifq, int type, void *discipline,
143     int (*enqueue)(struct ifaltq *, struct mbuf *, struct altq_pktattr *),
144     struct mbuf *(*dequeue)(struct ifaltq *, int),
145     int (*request)(struct ifaltq *, int, void *),
146     void *clfier, void *(*classify)(void *, struct mbuf *, int))
147 {
148 	if (!ALTQ_IS_READY(ifq))
149 		return ENXIO;
150 
151 #ifdef ALTQ3_COMPAT
152 	/*
153 	 * pfaltq can override the existing discipline, but altq3 cannot.
154 	 * check these if clfier is not NULL (which implies altq3).
155 	 */
156 	if (clfier != NULL) {
157 		if (ALTQ_IS_ENABLED(ifq))
158 			return EBUSY;
159 		if (ALTQ_IS_ATTACHED(ifq))
160 			return EEXIST;
161 	}
162 #endif
163 	ifq->altq_type     = type;
164 	ifq->altq_disc     = discipline;
165 	ifq->altq_enqueue  = enqueue;
166 	ifq->altq_dequeue  = dequeue;
167 	ifq->altq_request  = request;
168 	ifq->altq_clfier   = clfier;
169 	ifq->altq_classify = classify;
170 	ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED);
171 #ifdef ALTQ3_COMPAT
172 #ifdef ALTQ_KLD
173 	altq_module_incref(type);
174 #endif
175 #endif
176 	return 0;
177 }
178 
179 int
180 altq_detach(struct ifaltq *ifq)
181 {
182 	if (!ALTQ_IS_READY(ifq))
183 		return ENXIO;
184 	if (ALTQ_IS_ENABLED(ifq))
185 		return EBUSY;
186 	if (!ALTQ_IS_ATTACHED(ifq))
187 		return (0);
188 #ifdef ALTQ3_COMPAT
189 #ifdef ALTQ_KLD
190 	altq_module_declref(ifq->altq_type);
191 #endif
192 #endif
193 
194 	ifq->altq_type     = ALTQT_NONE;
195 	ifq->altq_disc     = NULL;
196 	ifq->altq_enqueue  = NULL;
197 	ifq->altq_dequeue  = NULL;
198 	ifq->altq_request  = NULL;
199 	ifq->altq_clfier   = NULL;
200 	ifq->altq_classify = NULL;
201 	ifq->altq_flags &= ALTQF_CANTCHANGE;
202 	return 0;
203 }
204 
205 int
206 altq_enable(struct ifaltq *ifq)
207 {
208 	int s;
209 
210 	if (!ALTQ_IS_READY(ifq))
211 		return ENXIO;
212 	if (ALTQ_IS_ENABLED(ifq))
213 		return 0;
214 
215 	s = splnet();
216 	IFQ_PURGE(ifq);
217 	ASSERT(ifq->ifq_len == 0);
218 	ifq->altq_flags |= ALTQF_ENABLED;
219 	if (ifq->altq_clfier != NULL)
220 		ifq->altq_flags |= ALTQF_CLASSIFY;
221 	splx(s);
222 
223 	return 0;
224 }
225 
226 int
227 altq_disable(struct ifaltq *ifq)
228 {
229 	int s;
230 
231 	if (!ALTQ_IS_ENABLED(ifq))
232 		return 0;
233 
234 	s = splnet();
235 	IFQ_PURGE(ifq);
236 	ASSERT(ifq->ifq_len == 0);
237 	ifq->altq_flags &= ~(ALTQF_ENABLED|ALTQF_CLASSIFY);
238 	splx(s);
239 	return 0;
240 }
241 
242 #ifdef ALTQ_DEBUG
243 void
244 altq_assert(const char *file, int line, const char *failedexpr)
245 {
246 	(void)printf("altq assertion \"%s\" failed: file \"%s\", line %d\n",
247 		     failedexpr, file, line);
248 	panic("altq assertion");
249 	/* NOTREACHED */
250 }
251 #endif
252 
253 /*
254  * internal representation of token bucket parameters
255  *	rate:	byte_per_unittime << 32
256  *		(((bits_per_sec) / 8) << 32) / machclk_freq
257  *	depth:	byte << 32
258  *
259  */
260 #define	TBR_SHIFT	32
261 #define	TBR_SCALE(x)	((int64_t)(x) << TBR_SHIFT)
262 #define	TBR_UNSCALE(x)	((x) >> TBR_SHIFT)
263 
264 struct mbuf *
265 tbr_dequeue(struct ifaltq *ifq, int op)
266 {
267 	struct tb_regulator *tbr;
268 	struct mbuf *m;
269 	int64_t interval;
270 	u_int64_t now;
271 
272 	tbr = ifq->altq_tbr;
273 	if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) {
274 		/* if this is a remove after poll, bypass tbr check */
275 	} else {
276 		/* update token only when it is negative */
277 		if (tbr->tbr_token <= 0) {
278 			now = read_machclk();
279 			interval = now - tbr->tbr_last;
280 			if (interval >= tbr->tbr_filluptime)
281 				tbr->tbr_token = tbr->tbr_depth;
282 			else {
283 				tbr->tbr_token += interval * tbr->tbr_rate;
284 				if (tbr->tbr_token > tbr->tbr_depth)
285 					tbr->tbr_token = tbr->tbr_depth;
286 			}
287 			tbr->tbr_last = now;
288 		}
289 		/* if token is still negative, don't allow dequeue */
290 		if (tbr->tbr_token <= 0)
291 			return (NULL);
292 	}
293 
294 	if (ALTQ_IS_ENABLED(ifq))
295 		m = (*ifq->altq_dequeue)(ifq, op);
296 	else {
297 		if (op == ALTDQ_POLL)
298 			IF_POLL(ifq, m);
299 		else
300 			IF_DEQUEUE(ifq, m);
301 	}
302 
303 	if (m != NULL && op == ALTDQ_REMOVE)
304 		tbr->tbr_token -= TBR_SCALE(m_pktlen(m));
305 	tbr->tbr_lastop = op;
306 	return (m);
307 }
308 
309 /*
310  * set a token bucket regulator.
311  * if the specified rate is zero, the token bucket regulator is deleted.
312  */
313 int
314 tbr_set(struct ifaltq *ifq, struct tb_profile *profile)
315 {
316 	struct tb_regulator *tbr, *otbr;
317 
318 	if (machclk_freq == 0)
319 		init_machclk();
320 	if (machclk_freq == 0) {
321 		printf("tbr_set: no CPU clock available!\n");
322 		return (ENXIO);
323 	}
324 
325 	if (profile->rate == 0) {
326 		/* delete this tbr */
327 		if ((tbr = ifq->altq_tbr) == NULL)
328 			return (ENOENT);
329 		ifq->altq_tbr = NULL;
330 		free(tbr, M_DEVBUF);
331 		return (0);
332 	}
333 
334 	tbr = malloc(sizeof(struct tb_regulator), M_DEVBUF, M_WAITOK|M_ZERO);
335 	if (tbr == NULL)
336 		return (ENOMEM);
337 
338 	tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq;
339 	tbr->tbr_depth = TBR_SCALE(profile->depth);
340 	if (tbr->tbr_rate > 0)
341 		tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate;
342 	else
343 		tbr->tbr_filluptime = 0xffffffffffffffffLL;
344 	tbr->tbr_token = tbr->tbr_depth;
345 	tbr->tbr_last = read_machclk();
346 	tbr->tbr_lastop = ALTDQ_REMOVE;
347 
348 	otbr = ifq->altq_tbr;
349 	ifq->altq_tbr = tbr;	/* set the new tbr */
350 
351 	if (otbr != NULL)
352 		free(otbr, M_DEVBUF);
353 	else {
354 		if (tbr_timer == 0) {
355 			CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
356 			tbr_timer = 1;
357 		}
358 	}
359 	return (0);
360 }
361 
362 /*
363  * tbr_timeout goes through the interface list, and kicks the drivers
364  * if necessary.
365  */
366 static void
367 tbr_timeout(void *arg)
368 {
369 	struct ifnet *ifp;
370 	int active, s;
371 
372 	active = 0;
373 	s = splnet();
374 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
375 		if (!TBR_IS_ENABLED(&ifp->if_snd))
376 			continue;
377 		active++;
378 		if (!IFQ_IS_EMPTY(&ifp->if_snd) && ifp->if_start != NULL)
379 			(*ifp->if_start)(ifp);
380 	}
381 	splx(s);
382 	if (active > 0)
383 		CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
384 	else
385 		tbr_timer = 0;	/* don't need tbr_timer anymore */
386 #if defined(__alpha__) && !defined(ALTQ_NOPCC)
387 	{
388 		/*
389 		 * XXX read out the machine dependent clock once a second
390 		 * to detect counter wrap-around.
391 		 */
392 		static u_int cnt;
393 
394 		if (++cnt >= hz) {
395 			(void)read_machclk();
396 			cnt = 0;
397 		}
398 	}
399 #endif /* __alpha__ && !ALTQ_NOPCC */
400 }
401 
402 /*
403  * get token bucket regulator profile
404  */
405 int
406 tbr_get(struct ifaltq *ifq, struct tb_profile *profile)
407 {
408 	struct tb_regulator *tbr;
409 
410 	if ((tbr = ifq->altq_tbr) == NULL) {
411 		profile->rate = 0;
412 		profile->depth = 0;
413 	} else {
414 		profile->rate =
415 		    (u_int)TBR_UNSCALE(tbr->tbr_rate * 8 * machclk_freq);
416 		profile->depth = (u_int)TBR_UNSCALE(tbr->tbr_depth);
417 	}
418 	return (0);
419 }
420 
421 #if NPF > 0
422 /*
423  * attach a discipline to the interface.  if one already exists, it is
424  * overridden.
425  */
426 int
427 altq_pfattach(struct pf_altq *a)
428 {
429 	int error = 0;
430 
431 	switch (a->scheduler) {
432 	case ALTQT_NONE:
433 		break;
434 #ifdef ALTQ_CBQ
435 	case ALTQT_CBQ:
436 		error = cbq_pfattach(a);
437 		break;
438 #endif
439 #ifdef ALTQ_PRIQ
440 	case ALTQT_PRIQ:
441 		error = priq_pfattach(a);
442 		break;
443 #endif
444 #ifdef ALTQ_HFSC
445 	case ALTQT_HFSC:
446 		error = hfsc_pfattach(a);
447 		break;
448 #endif
449 	default:
450 		error = ENXIO;
451 	}
452 
453 	return (error);
454 }
455 
456 /*
457  * detach a discipline from the interface.
458  * it is possible that the discipline was already overridden by another
459  * discipline.
460  */
461 int
462 altq_pfdetach(struct pf_altq *a)
463 {
464 	struct ifnet *ifp;
465 	int s, error = 0;
466 
467 	if ((ifp = ifunit(a->ifname)) == NULL)
468 		return (EINVAL);
469 
470 	/* if this discipline is no longer referenced, just return */
471 	if (a->altq_disc == NULL || a->altq_disc != ifp->if_snd.altq_disc)
472 		return (0);
473 
474 	s = splnet();
475 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
476 		error = altq_disable(&ifp->if_snd);
477 	if (error == 0)
478 		error = altq_detach(&ifp->if_snd);
479 	splx(s);
480 
481 	return (error);
482 }
483 
484 /*
485  * add a discipline or a queue
486  */
487 int
488 altq_add(struct pf_altq *a)
489 {
490 	int error = 0;
491 
492 	if (a->qname[0] != 0)
493 		return (altq_add_queue(a));
494 
495 	if (machclk_freq == 0)
496 		init_machclk();
497 	if (machclk_freq == 0)
498 		panic("altq_add: no CPU clock");
499 
500 	switch (a->scheduler) {
501 #ifdef ALTQ_CBQ
502 	case ALTQT_CBQ:
503 		error = cbq_add_altq(a);
504 		break;
505 #endif
506 #ifdef ALTQ_PRIQ
507 	case ALTQT_PRIQ:
508 		error = priq_add_altq(a);
509 		break;
510 #endif
511 #ifdef ALTQ_HFSC
512 	case ALTQT_HFSC:
513 		error = hfsc_add_altq(a);
514 		break;
515 #endif
516 	default:
517 		error = ENXIO;
518 	}
519 
520 	return (error);
521 }
522 
523 /*
524  * remove a discipline or a queue
525  */
526 int
527 altq_remove(struct pf_altq *a)
528 {
529 	int error = 0;
530 
531 	if (a->qname[0] != 0)
532 		return (altq_remove_queue(a));
533 
534 	switch (a->scheduler) {
535 #ifdef ALTQ_CBQ
536 	case ALTQT_CBQ:
537 		error = cbq_remove_altq(a);
538 		break;
539 #endif
540 #ifdef ALTQ_PRIQ
541 	case ALTQT_PRIQ:
542 		error = priq_remove_altq(a);
543 		break;
544 #endif
545 #ifdef ALTQ_HFSC
546 	case ALTQT_HFSC:
547 		error = hfsc_remove_altq(a);
548 		break;
549 #endif
550 	default:
551 		error = ENXIO;
552 	}
553 
554 	return (error);
555 }
556 
557 /*
558  * add a queue to the discipline
559  */
560 int
561 altq_add_queue(struct pf_altq *a)
562 {
563 	int error = 0;
564 
565 	switch (a->scheduler) {
566 #ifdef ALTQ_CBQ
567 	case ALTQT_CBQ:
568 		error = cbq_add_queue(a);
569 		break;
570 #endif
571 #ifdef ALTQ_PRIQ
572 	case ALTQT_PRIQ:
573 		error = priq_add_queue(a);
574 		break;
575 #endif
576 #ifdef ALTQ_HFSC
577 	case ALTQT_HFSC:
578 		error = hfsc_add_queue(a);
579 		break;
580 #endif
581 	default:
582 		error = ENXIO;
583 	}
584 
585 	return (error);
586 }
587 
588 /*
589  * remove a queue from the discipline
590  */
591 int
592 altq_remove_queue(struct pf_altq *a)
593 {
594 	int error = 0;
595 
596 	switch (a->scheduler) {
597 #ifdef ALTQ_CBQ
598 	case ALTQT_CBQ:
599 		error = cbq_remove_queue(a);
600 		break;
601 #endif
602 #ifdef ALTQ_PRIQ
603 	case ALTQT_PRIQ:
604 		error = priq_remove_queue(a);
605 		break;
606 #endif
607 #ifdef ALTQ_HFSC
608 	case ALTQT_HFSC:
609 		error = hfsc_remove_queue(a);
610 		break;
611 #endif
612 	default:
613 		error = ENXIO;
614 	}
615 
616 	return (error);
617 }
618 
619 /*
620  * get queue statistics
621  */
622 int
623 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes)
624 {
625 	int error = 0;
626 
627 	switch (a->scheduler) {
628 #ifdef ALTQ_CBQ
629 	case ALTQT_CBQ:
630 		error = cbq_getqstats(a, ubuf, nbytes);
631 		break;
632 #endif
633 #ifdef ALTQ_PRIQ
634 	case ALTQT_PRIQ:
635 		error = priq_getqstats(a, ubuf, nbytes);
636 		break;
637 #endif
638 #ifdef ALTQ_HFSC
639 	case ALTQT_HFSC:
640 		error = hfsc_getqstats(a, ubuf, nbytes);
641 		break;
642 #endif
643 	default:
644 		error = ENXIO;
645 	}
646 
647 	return (error);
648 }
649 #endif /* NPF > 0 */
650 
651 /*
652  * read and write diffserv field in IPv4 or IPv6 header
653  */
654 u_int8_t
655 read_dsfield(struct mbuf *m, struct altq_pktattr *pktattr)
656 {
657 	struct mbuf *m0;
658 	u_int8_t ds_field = 0;
659 
660 	if (pktattr == NULL ||
661 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
662 		return ((u_int8_t)0);
663 
664 	/* verify that pattr_hdr is within the mbuf data */
665 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
666 		if (((char *)pktattr->pattr_hdr >= m0->m_data) &&
667 		    ((char *)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("read_dsfield: can't locate header!\n");
674 #endif
675 		return ((u_int8_t)0);
676 	}
677 
678 	if (pktattr->pattr_af == AF_INET) {
679 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
680 
681 		if (ip->ip_v != 4)
682 			return ((u_int8_t)0);	/* version mismatch! */
683 		ds_field = ip->ip_tos;
684 	}
685 #ifdef INET6
686 	else if (pktattr->pattr_af == AF_INET6) {
687 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
688 		u_int32_t flowlabel;
689 
690 		flowlabel = ntohl(ip6->ip6_flow);
691 		if ((flowlabel >> 28) != 6)
692 			return ((u_int8_t)0);	/* version mismatch! */
693 		ds_field = (flowlabel >> 20) & 0xff;
694 	}
695 #endif
696 	return (ds_field);
697 }
698 
699 void
700 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, u_int8_t dsfield)
701 {
702 	struct mbuf *m0;
703 
704 	if (pktattr == NULL ||
705 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
706 		return;
707 
708 	/* verify that pattr_hdr is within the mbuf data */
709 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
710 		if (((char *)pktattr->pattr_hdr >= m0->m_data) &&
711 		    ((char *)pktattr->pattr_hdr < m0->m_data + m0->m_len))
712 			break;
713 	if (m0 == NULL) {
714 		/* ick, pattr_hdr is stale */
715 		pktattr->pattr_af = AF_UNSPEC;
716 #ifdef ALTQ_DEBUG
717 		printf("write_dsfield: can't locate header!\n");
718 #endif
719 		return;
720 	}
721 
722 	if (pktattr->pattr_af == AF_INET) {
723 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
724 		u_int8_t old;
725 		int32_t sum;
726 
727 		if (ip->ip_v != 4)
728 			return;		/* version mismatch! */
729 		old = ip->ip_tos;
730 		dsfield |= old & 3;	/* leave CU bits */
731 		if (old == dsfield)
732 			return;
733 		ip->ip_tos = dsfield;
734 		/*
735 		 * update checksum (from RFC1624)
736 		 *	   HC' = ~(~HC + ~m + m')
737 		 */
738 		sum = ~ntohs(ip->ip_sum) & 0xffff;
739 		sum += 0xff00 + (~old & 0xff) + dsfield;
740 		sum = (sum >> 16) + (sum & 0xffff);
741 		sum += (sum >> 16);  /* add carry */
742 
743 		ip->ip_sum = htons(~sum & 0xffff);
744 	}
745 #ifdef INET6
746 	else if (pktattr->pattr_af == AF_INET6) {
747 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
748 		u_int32_t flowlabel;
749 
750 		flowlabel = ntohl(ip6->ip6_flow);
751 		if ((flowlabel >> 28) != 6)
752 			return;		/* version mismatch! */
753 		flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20);
754 		ip6->ip6_flow = htonl(flowlabel);
755 	}
756 #endif
757 	return;
758 }
759 
760 
761 /*
762  * high resolution clock support taking advantage of a machine dependent
763  * high resolution time counter (e.g., timestamp counter of intel pentium).
764  * we assume
765  *  - 64-bit-long monotonically-increasing counter
766  *  - frequency range is 100M-4GHz (CPU speed)
767  */
768 /* if pcc is not available or disabled, emulate 256MHz using microtime() */
769 #define	MACHCLK_SHIFT	8
770 
771 int machclk_usepcc;
772 u_int32_t machclk_freq = 0;
773 u_int32_t machclk_per_tick = 0;
774 
775 #ifdef __alpha__
776 #ifdef __FreeBSD__
777 extern u_int32_t cycles_per_sec;	/* alpha cpu clock frequency */
778 #elif defined(__NetBSD__) || defined(__OpenBSD__)
779 extern u_int64_t cycles_per_usec;	/* alpha cpu clock frequency */
780 #endif
781 #endif /* __alpha__ */
782 
783 void
784 init_machclk(void)
785 {
786 
787 	callout_init(&tbr_callout, 0);
788 
789 	machclk_usepcc = 1;
790 
791 #if (!defined(__i386__) && !defined(__alpha__)) || defined(ALTQ_NOPCC)
792 	machclk_usepcc = 0;
793 #endif
794 #if defined(__FreeBSD__) && defined(SMP)
795 	machclk_usepcc = 0;
796 #endif
797 #if defined(__NetBSD__) && defined(MULTIPROCESSOR)
798 	machclk_usepcc = 0;
799 #endif
800 #ifdef __i386__
801 	/* check if TSC is available */
802 	if (machclk_usepcc == 1 && (cpu_feature & CPUID_TSC) == 0)
803 		machclk_usepcc = 0;
804 #endif
805 
806 	if (machclk_usepcc == 0) {
807 		/* emulate 256MHz using microtime() */
808 		machclk_freq = 1000000 << MACHCLK_SHIFT;
809 		machclk_per_tick = machclk_freq / hz;
810 #ifdef ALTQ_DEBUG
811 		printf("altq: emulate %uHz CPU clock\n", machclk_freq);
812 #endif
813 		return;
814 	}
815 
816 	/*
817 	 * if the clock frequency (of Pentium TSC or Alpha PCC) is
818 	 * accessible, just use it.
819 	 */
820 #ifdef __i386__
821 #ifdef __FreeBSD__
822 #if (__FreeBSD_version > 300000)
823 	machclk_freq = tsc_freq;
824 #else
825 	machclk_freq = i586_ctr_freq;
826 #endif
827 #elif defined(__NetBSD__)
828 	machclk_freq = (u_int32_t)curcpu()->ci_tsc_freq;
829 #elif defined(__OpenBSD__) && (defined(I586_CPU) || defined(I686_CPU))
830 	machclk_freq = pentium_mhz * 1000000;
831 #endif
832 #elif defined(__alpha__)
833 #ifdef __FreeBSD__
834 	machclk_freq = cycles_per_sec;
835 #elif defined(__NetBSD__) || defined(__OpenBSD__)
836 	machclk_freq = (u_int32_t)(cycles_per_usec * 1000000);
837 #endif
838 #endif /* __alpha__ */
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 		u_int64_t	start, end, diff;
847 		int		timo;
848 
849 		microtime(&tv_start);
850 		start = read_machclk();
851 		timo = hz;	/* 1 sec */
852 		(void)tsleep(&wait, PWAIT | PCATCH, "init_machclk", timo);
853 		microtime(&tv_end);
854 		end = read_machclk();
855 		diff = (u_int64_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 = (u_int)((end - start) * 1000000 / diff);
859 	}
860 
861 	machclk_per_tick = machclk_freq / hz;
862 
863 #ifdef ALTQ_DEBUG
864 	printf("altq: CPU clock: %uHz\n", machclk_freq);
865 #endif
866 }
867 
868 #if defined(__OpenBSD__) && defined(__i386__)
869 static inline u_int64_t
870 rdtsc(void)
871 {
872 	u_int64_t rv;
873 	__asm __volatile(".byte 0x0f, 0x31" : "=A" (rv));
874 	return (rv);
875 }
876 #endif /* __OpenBSD__ && __i386__ */
877 
878 u_int64_t
879 read_machclk(void)
880 {
881 	u_int64_t val;
882 
883 	if (machclk_usepcc) {
884 #if defined(__i386__)
885 		val = rdtsc();
886 #elif defined(__alpha__)
887 		static u_int32_t last_pcc, upper;
888 		u_int32_t pcc;
889 
890 		/*
891 		 * for alpha, make a 64bit counter value out of the 32bit
892 		 * alpha processor cycle counter.
893 		 * read_machclk must be called within a half of its
894 		 * wrap-around cycle (about 5 sec for 400MHz cpu) to properly
895 		 * detect a counter wrap-around.
896 		 * tbr_timeout calls read_machclk once a second.
897 		 */
898 		pcc = (u_int32_t)alpha_rpcc();
899 		if (pcc <= last_pcc)
900 			upper++;
901 		last_pcc = pcc;
902 		val = ((u_int64_t)upper << 32) + pcc;
903 #else
904 		panic("read_machclk");
905 #endif
906 	} else {
907 		struct timeval tv;
908 
909 		microtime(&tv);
910 		val = (((u_int64_t)(tv.tv_sec - boottime.tv_sec) * 1000000
911 		    + tv.tv_usec) << MACHCLK_SHIFT);
912 	}
913 	return (val);
914 }
915 
916 #ifdef ALTQ3_CLFIER_COMPAT
917 
918 #ifndef IPPROTO_ESP
919 #define	IPPROTO_ESP	50		/* encapsulating security payload */
920 #endif
921 #ifndef IPPROTO_AH
922 #define	IPPROTO_AH	51		/* authentication header */
923 #endif
924 
925 /*
926  * extract flow information from a given packet.
927  * filt_mask shows flowinfo fields required.
928  * we assume the ip header is in one mbuf, and addresses and ports are
929  * in network byte order.
930  */
931 int
932 altq_extractflow(struct mbuf *m, int af, struct flowinfo *flow,
933     u_int32_t filt_bmask)
934 {
935 
936 	switch (af) {
937 	case PF_INET: {
938 		struct flowinfo_in *fin;
939 		struct ip *ip;
940 
941 		ip = mtod(m, struct ip *);
942 
943 		if (ip->ip_v != 4)
944 			break;
945 
946 		fin = (struct flowinfo_in *)flow;
947 		fin->fi_len = sizeof(struct flowinfo_in);
948 		fin->fi_family = AF_INET;
949 
950 		fin->fi_proto = ip->ip_p;
951 		fin->fi_tos = ip->ip_tos;
952 
953 		fin->fi_src.s_addr = ip->ip_src.s_addr;
954 		fin->fi_dst.s_addr = ip->ip_dst.s_addr;
955 
956 		if (filt_bmask & FIMB4_PORTS)
957 			/* if port info is required, extract port numbers */
958 			extract_ports4(m, ip, fin);
959 		else {
960 			fin->fi_sport = 0;
961 			fin->fi_dport = 0;
962 			fin->fi_gpi = 0;
963 		}
964 		return (1);
965 	}
966 
967 #ifdef INET6
968 	case PF_INET6: {
969 		struct flowinfo_in6 *fin6;
970 		struct ip6_hdr *ip6;
971 
972 		ip6 = mtod(m, struct ip6_hdr *);
973 		/* should we check the ip version? */
974 
975 		fin6 = (struct flowinfo_in6 *)flow;
976 		fin6->fi6_len = sizeof(struct flowinfo_in6);
977 		fin6->fi6_family = AF_INET6;
978 
979 		fin6->fi6_proto = ip6->ip6_nxt;
980 		fin6->fi6_tclass   = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
981 
982 		fin6->fi6_flowlabel = ip6->ip6_flow & htonl(0x000fffff);
983 		fin6->fi6_src = ip6->ip6_src;
984 		fin6->fi6_dst = ip6->ip6_dst;
985 
986 		if ((filt_bmask & FIMB6_PORTS) ||
987 		    ((filt_bmask & FIMB6_PROTO)
988 		     && ip6->ip6_nxt > IPPROTO_IPV6))
989 			/*
990 			 * if port info is required, or proto is required
991 			 * but there are option headers, extract port
992 			 * and protocol numbers.
993 			 */
994 			extract_ports6(m, ip6, fin6);
995 		else {
996 			fin6->fi6_sport = 0;
997 			fin6->fi6_dport = 0;
998 			fin6->fi6_gpi = 0;
999 		}
1000 		return (1);
1001 	}
1002 #endif /* INET6 */
1003 
1004 	default:
1005 		break;
1006 	}
1007 
1008 	/* failed */
1009 	flow->fi_len = sizeof(struct flowinfo);
1010 	flow->fi_family = AF_UNSPEC;
1011 	return (0);
1012 }
1013 
1014 /*
1015  * helper routine to extract port numbers
1016  */
1017 /* structure for ipsec and ipv6 option header template */
1018 struct _opt6 {
1019 	u_int8_t	opt6_nxt;	/* next header */
1020 	u_int8_t	opt6_hlen;	/* header extension length */
1021 	u_int16_t	_pad;
1022 	u_int32_t	ah_spi;		/* security parameter index
1023 					   for authentication header */
1024 };
1025 
1026 /*
1027  * extract port numbers from a ipv4 packet.
1028  */
1029 static int
1030 extract_ports4(struct mbuf *m, struct ip *ip, struct flowinfo_in *fin)
1031 {
1032 	struct mbuf *m0;
1033 	u_short ip_off;
1034 	u_int8_t proto;
1035 	int 	off;
1036 
1037 	fin->fi_sport = 0;
1038 	fin->fi_dport = 0;
1039 	fin->fi_gpi = 0;
1040 
1041 	ip_off = ntohs(ip->ip_off);
1042 	/* if it is a fragment, try cached fragment info */
1043 	if (ip_off & IP_OFFMASK) {
1044 		ip4f_lookup(ip, fin);
1045 		return (1);
1046 	}
1047 
1048 	/* locate the mbuf containing the protocol header */
1049 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
1050 		if (((char *)ip >= m0->m_data) &&
1051 		    ((char *)ip < m0->m_data + m0->m_len))
1052 			break;
1053 	if (m0 == NULL) {
1054 #ifdef ALTQ_DEBUG
1055 		printf("extract_ports4: can't locate header! ip=%p\n", ip);
1056 #endif
1057 		return (0);
1058 	}
1059 	off = ((char *)ip - m0->m_data) + (ip->ip_hl << 2);
1060 	proto = ip->ip_p;
1061 
1062 #ifdef ALTQ_IPSEC
1063  again:
1064 #endif
1065 	while (off >= m0->m_len) {
1066 		off -= m0->m_len;
1067 		m0 = m0->m_next;
1068 		if (m0 == NULL)
1069 			return (0);  /* bogus ip_hl! */
1070 	}
1071 	if (m0->m_len < off + 4)
1072 		return (0);
1073 
1074 	switch (proto) {
1075 	case IPPROTO_TCP:
1076 	case IPPROTO_UDP: {
1077 		struct udphdr *udp;
1078 
1079 		udp = (struct udphdr *)(mtod(m0, char *) + off);
1080 		fin->fi_sport = udp->uh_sport;
1081 		fin->fi_dport = udp->uh_dport;
1082 		fin->fi_proto = proto;
1083 		}
1084 		break;
1085 
1086 #ifdef ALTQ_IPSEC
1087 	case IPPROTO_ESP:
1088 		if (fin->fi_gpi == 0){
1089 			u_int32_t *gpi;
1090 
1091 			gpi = (u_int32_t *)(mtod(m0, char *) + off);
1092 			fin->fi_gpi   = *gpi;
1093 		}
1094 		fin->fi_proto = proto;
1095 		break;
1096 
1097 	case IPPROTO_AH: {
1098 			/* get next header and header length */
1099 			struct _opt6 *opt6;
1100 
1101 			opt6 = (struct _opt6 *)(mtod(m0, char *) + off);
1102 			proto = opt6->opt6_nxt;
1103 			off += 8 + (opt6->opt6_hlen * 4);
1104 			if (fin->fi_gpi == 0 && m0->m_len >= off + 8)
1105 				fin->fi_gpi = opt6->ah_spi;
1106 		}
1107 		/* goto the next header */
1108 		goto again;
1109 #endif  /* ALTQ_IPSEC */
1110 
1111 	default:
1112 		fin->fi_proto = proto;
1113 		return (0);
1114 	}
1115 
1116 	/* if this is a first fragment, cache it. */
1117 	if (ip_off & IP_MF)
1118 		ip4f_cache(ip, fin);
1119 
1120 	return (1);
1121 }
1122 
1123 #ifdef INET6
1124 static int
1125 extract_ports6(struct mbuf *m, struct ip6_hdr *ip6, struct flowinfo_in6 *fin6)
1126 {
1127 	struct mbuf *m0;
1128 	int	off;
1129 	u_int8_t proto;
1130 
1131 	fin6->fi6_gpi   = 0;
1132 	fin6->fi6_sport = 0;
1133 	fin6->fi6_dport = 0;
1134 
1135 	/* locate the mbuf containing the protocol header */
1136 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
1137 		if (((char *)ip6 >= m0->m_data) &&
1138 		    ((char *)ip6 < m0->m_data + m0->m_len))
1139 			break;
1140 	if (m0 == NULL) {
1141 #ifdef ALTQ_DEBUG
1142 		printf("extract_ports6: can't locate header! ip6=%p\n", ip6);
1143 #endif
1144 		return (0);
1145 	}
1146 	off = ((char *)ip6 - m0->m_data) + sizeof(struct ip6_hdr);
1147 
1148 	proto = ip6->ip6_nxt;
1149 	do {
1150 		while (off >= m0->m_len) {
1151 			off -= m0->m_len;
1152 			m0 = m0->m_next;
1153 			if (m0 == NULL)
1154 				return (0);
1155 		}
1156 		if (m0->m_len < off + 4)
1157 			return (0);
1158 
1159 		switch (proto) {
1160 		case IPPROTO_TCP:
1161 		case IPPROTO_UDP: {
1162 			struct udphdr *udp;
1163 
1164 			udp = (struct udphdr *)(mtod(m0, char *) + off);
1165 			fin6->fi6_sport = udp->uh_sport;
1166 			fin6->fi6_dport = udp->uh_dport;
1167 			fin6->fi6_proto = proto;
1168 			}
1169 			return (1);
1170 
1171 		case IPPROTO_ESP:
1172 			if (fin6->fi6_gpi == 0) {
1173 				u_int32_t *gpi;
1174 
1175 				gpi = (u_int32_t *)(mtod(m0, char *) + off);
1176 				fin6->fi6_gpi   = *gpi;
1177 			}
1178 			fin6->fi6_proto = proto;
1179 			return (1);
1180 
1181 		case IPPROTO_AH: {
1182 			/* get next header and header length */
1183 			struct _opt6 *opt6;
1184 
1185 			opt6 = (struct _opt6 *)(mtod(m0, char *) + off);
1186 			if (fin6->fi6_gpi == 0 && m0->m_len >= off + 8)
1187 				fin6->fi6_gpi = opt6->ah_spi;
1188 			proto = opt6->opt6_nxt;
1189 			off += 8 + (opt6->opt6_hlen * 4);
1190 			/* goto the next header */
1191 			break;
1192 			}
1193 
1194 		case IPPROTO_HOPOPTS:
1195 		case IPPROTO_ROUTING:
1196 		case IPPROTO_DSTOPTS: {
1197 			/* get next header and header length */
1198 			struct _opt6 *opt6;
1199 
1200 			opt6 = (struct _opt6 *)(mtod(m0, char *) + off);
1201 			proto = opt6->opt6_nxt;
1202 			off += (opt6->opt6_hlen + 1) * 8;
1203 			/* goto the next header */
1204 			break;
1205 			}
1206 
1207 		case IPPROTO_FRAGMENT:
1208 			/* ipv6 fragmentations are not supported yet */
1209 		default:
1210 			fin6->fi6_proto = proto;
1211 			return (0);
1212 		}
1213 	} while (1);
1214 	/*NOTREACHED*/
1215 }
1216 #endif /* INET6 */
1217 
1218 /*
1219  * altq common classifier
1220  */
1221 int
1222 acc_add_filter(struct acc_classifier *classifier, struct flow_filter *filter,
1223     void *class, u_long *phandle)
1224 {
1225 	struct acc_filter *afp, *prev, *tmp;
1226 	int	i, s;
1227 
1228 #ifdef INET6
1229 	if (filter->ff_flow.fi_family != AF_INET &&
1230 	    filter->ff_flow.fi_family != AF_INET6)
1231 		return (EINVAL);
1232 #else
1233 	if (filter->ff_flow.fi_family != AF_INET)
1234 		return (EINVAL);
1235 #endif
1236 
1237 	afp = malloc(sizeof(struct acc_filter), M_DEVBUF, M_WAITOK|M_ZERO);
1238 	if (afp == NULL)
1239 		return (ENOMEM);
1240 
1241 	afp->f_filter = *filter;
1242 	afp->f_class = class;
1243 
1244 	i = ACC_WILDCARD_INDEX;
1245 	if (filter->ff_flow.fi_family == AF_INET) {
1246 		struct flow_filter *filter4 = &afp->f_filter;
1247 
1248 		/*
1249 		 * if address is 0, it's a wildcard.  if address mask
1250 		 * isn't set, use full mask.
1251 		 */
1252 		if (filter4->ff_flow.fi_dst.s_addr == 0)
1253 			filter4->ff_mask.mask_dst.s_addr = 0;
1254 		else if (filter4->ff_mask.mask_dst.s_addr == 0)
1255 			filter4->ff_mask.mask_dst.s_addr = 0xffffffff;
1256 		if (filter4->ff_flow.fi_src.s_addr == 0)
1257 			filter4->ff_mask.mask_src.s_addr = 0;
1258 		else if (filter4->ff_mask.mask_src.s_addr == 0)
1259 			filter4->ff_mask.mask_src.s_addr = 0xffffffff;
1260 
1261 		/* clear extra bits in addresses  */
1262 		   filter4->ff_flow.fi_dst.s_addr &=
1263 		       filter4->ff_mask.mask_dst.s_addr;
1264 		   filter4->ff_flow.fi_src.s_addr &=
1265 		       filter4->ff_mask.mask_src.s_addr;
1266 
1267 		/*
1268 		 * if dst address is a wildcard, use hash-entry
1269 		 * ACC_WILDCARD_INDEX.
1270 		 */
1271 		if (filter4->ff_mask.mask_dst.s_addr != 0xffffffff)
1272 			i = ACC_WILDCARD_INDEX;
1273 		else
1274 			i = ACC_GET_HASH_INDEX(filter4->ff_flow.fi_dst.s_addr);
1275 	}
1276 #ifdef INET6
1277 	else if (filter->ff_flow.fi_family == AF_INET6) {
1278 		struct flow_filter6 *filter6 =
1279 			(struct flow_filter6 *)&afp->f_filter;
1280 #ifndef IN6MASK0 /* taken from kame ipv6 */
1281 #define	IN6MASK0	{{{ 0, 0, 0, 0 }}}
1282 #define	IN6MASK128	{{{ 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff }}}
1283 		const struct in6_addr in6mask0 = IN6MASK0;
1284 		const struct in6_addr in6mask128 = IN6MASK128;
1285 #endif
1286 
1287 		if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_dst))
1288 			filter6->ff_mask6.mask6_dst = in6mask0;
1289 		else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_dst))
1290 			filter6->ff_mask6.mask6_dst = in6mask128;
1291 		if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_src))
1292 			filter6->ff_mask6.mask6_src = in6mask0;
1293 		else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_src))
1294 			filter6->ff_mask6.mask6_src = in6mask128;
1295 
1296 		/* clear extra bits in addresses  */
1297 		for (i = 0; i < 16; i++)
1298 			filter6->ff_flow6.fi6_dst.s6_addr[i] &=
1299 			    filter6->ff_mask6.mask6_dst.s6_addr[i];
1300 		for (i = 0; i < 16; i++)
1301 			filter6->ff_flow6.fi6_src.s6_addr[i] &=
1302 			    filter6->ff_mask6.mask6_src.s6_addr[i];
1303 
1304 		if (filter6->ff_flow6.fi6_flowlabel == 0)
1305 			i = ACC_WILDCARD_INDEX;
1306 		else
1307 			i = ACC_GET_HASH_INDEX(filter6->ff_flow6.fi6_flowlabel);
1308 	}
1309 #endif /* INET6 */
1310 
1311 	afp->f_handle = get_filt_handle(classifier, i);
1312 
1313 	/* update filter bitmask */
1314 	afp->f_fbmask = filt2fibmask(filter);
1315 	classifier->acc_fbmask |= afp->f_fbmask;
1316 
1317 	/*
1318 	 * add this filter to the filter list.
1319 	 * filters are ordered from the highest rule number.
1320 	 */
1321 	s = splnet();
1322 	prev = NULL;
1323 	LIST_FOREACH(tmp, &classifier->acc_filters[i], f_chain) {
1324 		if (tmp->f_filter.ff_ruleno > afp->f_filter.ff_ruleno)
1325 			prev = tmp;
1326 		else
1327 			break;
1328 	}
1329 	if (prev == NULL)
1330 		LIST_INSERT_HEAD(&classifier->acc_filters[i], afp, f_chain);
1331 	else
1332 		LIST_INSERT_AFTER(prev, afp, f_chain);
1333 	splx(s);
1334 
1335 	*phandle = afp->f_handle;
1336 	return (0);
1337 }
1338 
1339 int
1340 acc_delete_filter(struct acc_classifier *classifier, u_long handle)
1341 {
1342 	struct acc_filter *afp;
1343 	int	s;
1344 
1345 	if ((afp = filth_to_filtp(classifier, handle)) == NULL)
1346 		return (EINVAL);
1347 
1348 	s = splnet();
1349 	LIST_REMOVE(afp, f_chain);
1350 	splx(s);
1351 
1352 	free(afp, M_DEVBUF);
1353 
1354 	/* todo: update filt_bmask */
1355 
1356 	return (0);
1357 }
1358 
1359 /*
1360  * delete filters referencing to the specified class.
1361  * if the all flag is not 0, delete all the filters.
1362  */
1363 int
1364 acc_discard_filters(struct acc_classifier *classifier, void *class, int all)
1365 {
1366 	struct acc_filter *afp;
1367 	int	i, s;
1368 
1369 	s = splnet();
1370 	for (i = 0; i < ACC_FILTER_TABLESIZE; i++) {
1371 		do {
1372 			LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1373 				if (all || afp->f_class == class) {
1374 					LIST_REMOVE(afp, f_chain);
1375 					free(afp, M_DEVBUF);
1376 					/* start again from the head */
1377 					break;
1378 				}
1379 		} while (afp != NULL);
1380 	}
1381 	splx(s);
1382 
1383 	if (all)
1384 		classifier->acc_fbmask = 0;
1385 
1386 	return (0);
1387 }
1388 
1389 void *
1390 acc_classify(void *clfier, struct mbuf *m, int af)
1391 {
1392 	struct acc_classifier *classifier;
1393 	struct flowinfo flow;
1394 	struct acc_filter *afp;
1395 	int	i;
1396 
1397 	classifier = (struct acc_classifier *)clfier;
1398 	altq_extractflow(m, af, &flow, classifier->acc_fbmask);
1399 
1400 	if (flow.fi_family == AF_INET) {
1401 		struct flowinfo_in *fp = (struct flowinfo_in *)&flow;
1402 
1403 		if ((classifier->acc_fbmask & FIMB4_ALL) == FIMB4_TOS) {
1404 			/* only tos is used */
1405 			LIST_FOREACH(afp,
1406 				 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1407 				 f_chain)
1408 				if (apply_tosfilter4(afp->f_fbmask,
1409 						     &afp->f_filter, fp))
1410 					/* filter matched */
1411 					return (afp->f_class);
1412 		} else if ((classifier->acc_fbmask &
1413 			(~(FIMB4_PROTO|FIMB4_SPORT|FIMB4_DPORT) & FIMB4_ALL))
1414 		    == 0) {
1415 			/* only proto and ports are used */
1416 			LIST_FOREACH(afp,
1417 				 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1418 				 f_chain)
1419 				if (apply_ppfilter4(afp->f_fbmask,
1420 						    &afp->f_filter, fp))
1421 					/* filter matched */
1422 					return (afp->f_class);
1423 		} else {
1424 			/* get the filter hash entry from its dest address */
1425 			i = ACC_GET_HASH_INDEX(fp->fi_dst.s_addr);
1426 			do {
1427 				/*
1428 				 * go through this loop twice.  first for dst
1429 				 * hash, second for wildcards.
1430 				 */
1431 				LIST_FOREACH(afp, &classifier->acc_filters[i],
1432 					     f_chain)
1433 					if (apply_filter4(afp->f_fbmask,
1434 							  &afp->f_filter, fp))
1435 						/* filter matched */
1436 						return (afp->f_class);
1437 
1438 				/*
1439 				 * check again for filters with a dst addr
1440 				 * wildcard.
1441 				 * (daddr == 0 || dmask != 0xffffffff).
1442 				 */
1443 				if (i != ACC_WILDCARD_INDEX)
1444 					i = ACC_WILDCARD_INDEX;
1445 				else
1446 					break;
1447 			} while (1);
1448 		}
1449 	}
1450 #ifdef INET6
1451 	else if (flow.fi_family == AF_INET6) {
1452 		struct flowinfo_in6 *fp6 = (struct flowinfo_in6 *)&flow;
1453 
1454 		/* get the filter hash entry from its flow ID */
1455 		if (fp6->fi6_flowlabel != 0)
1456 			i = ACC_GET_HASH_INDEX(fp6->fi6_flowlabel);
1457 		else
1458 			/* flowlable can be zero */
1459 			i = ACC_WILDCARD_INDEX;
1460 
1461 		/* go through this loop twice.  first for flow hash, second
1462 		   for wildcards. */
1463 		do {
1464 			LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1465 				if (apply_filter6(afp->f_fbmask,
1466 					(struct flow_filter6 *)&afp->f_filter,
1467 					fp6))
1468 					/* filter matched */
1469 					return (afp->f_class);
1470 
1471 			/*
1472 			 * check again for filters with a wildcard.
1473 			 */
1474 			if (i != ACC_WILDCARD_INDEX)
1475 				i = ACC_WILDCARD_INDEX;
1476 			else
1477 				break;
1478 		} while (1);
1479 	}
1480 #endif /* INET6 */
1481 
1482 	/* no filter matched */
1483 	return (NULL);
1484 }
1485 
1486 static int
1487 apply_filter4(u_int32_t fbmask, struct flow_filter *filt,
1488     struct flowinfo_in *pkt)
1489 {
1490 	if (filt->ff_flow.fi_family != AF_INET)
1491 		return (0);
1492 	if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1493 		return (0);
1494 	if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1495 		return (0);
1496 	if ((fbmask & FIMB4_DADDR) &&
1497 	    filt->ff_flow.fi_dst.s_addr !=
1498 	    (pkt->fi_dst.s_addr & filt->ff_mask.mask_dst.s_addr))
1499 		return (0);
1500 	if ((fbmask & FIMB4_SADDR) &&
1501 	    filt->ff_flow.fi_src.s_addr !=
1502 	    (pkt->fi_src.s_addr & filt->ff_mask.mask_src.s_addr))
1503 		return (0);
1504 	if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1505 		return (0);
1506 	if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1507 	    (pkt->fi_tos & filt->ff_mask.mask_tos))
1508 		return (0);
1509 	if ((fbmask & FIMB4_GPI) && filt->ff_flow.fi_gpi != (pkt->fi_gpi))
1510 		return (0);
1511 	/* match */
1512 	return (1);
1513 }
1514 
1515 /*
1516  * filter matching function optimized for a common case that checks
1517  * only protocol and port numbers
1518  */
1519 static int
1520 apply_ppfilter4(u_int32_t fbmask, struct flow_filter *filt,
1521     struct flowinfo_in *pkt)
1522 {
1523 	if (filt->ff_flow.fi_family != AF_INET)
1524 		return (0);
1525 	if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1526 		return (0);
1527 	if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1528 		return (0);
1529 	if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1530 		return (0);
1531 	/* match */
1532 	return (1);
1533 }
1534 
1535 /*
1536  * filter matching function only for tos field.
1537  */
1538 static int
1539 apply_tosfilter4(u_int32_t fbmask, struct flow_filter *filt,
1540     struct flowinfo_in *pkt)
1541 {
1542 	if (filt->ff_flow.fi_family != AF_INET)
1543 		return (0);
1544 	if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1545 	    (pkt->fi_tos & filt->ff_mask.mask_tos))
1546 		return (0);
1547 	/* match */
1548 	return (1);
1549 }
1550 
1551 #ifdef INET6
1552 static int
1553 apply_filter6(u_int32_t fbmask, struct flow_filter6 *filt,
1554     struct flowinfo_in6 *pkt)
1555 {
1556 	int i;
1557 
1558 	if (filt->ff_flow6.fi6_family != AF_INET6)
1559 		return (0);
1560 	if ((fbmask & FIMB6_FLABEL) &&
1561 	    filt->ff_flow6.fi6_flowlabel != pkt->fi6_flowlabel)
1562 		return (0);
1563 	if ((fbmask & FIMB6_PROTO) &&
1564 	    filt->ff_flow6.fi6_proto != pkt->fi6_proto)
1565 		return (0);
1566 	if ((fbmask & FIMB6_SPORT) &&
1567 	    filt->ff_flow6.fi6_sport != pkt->fi6_sport)
1568 		return (0);
1569 	if ((fbmask & FIMB6_DPORT) &&
1570 	    filt->ff_flow6.fi6_dport != pkt->fi6_dport)
1571 		return (0);
1572 	if (fbmask & FIMB6_SADDR) {
1573 		for (i = 0; i < 4; i++)
1574 			if (filt->ff_flow6.fi6_src.s6_addr32[i] !=
1575 			    (pkt->fi6_src.s6_addr32[i] &
1576 			     filt->ff_mask6.mask6_src.s6_addr32[i]))
1577 				return (0);
1578 	}
1579 	if (fbmask & FIMB6_DADDR) {
1580 		for (i = 0; i < 4; i++)
1581 			if (filt->ff_flow6.fi6_dst.s6_addr32[i] !=
1582 			    (pkt->fi6_dst.s6_addr32[i] &
1583 			     filt->ff_mask6.mask6_dst.s6_addr32[i]))
1584 				return (0);
1585 	}
1586 	if ((fbmask & FIMB6_TCLASS) &&
1587 	    filt->ff_flow6.fi6_tclass !=
1588 	    (pkt->fi6_tclass & filt->ff_mask6.mask6_tclass))
1589 		return (0);
1590 	if ((fbmask & FIMB6_GPI) &&
1591 	    filt->ff_flow6.fi6_gpi != pkt->fi6_gpi)
1592 		return (0);
1593 	/* match */
1594 	return (1);
1595 }
1596 #endif /* INET6 */
1597 
1598 /*
1599  *  filter handle:
1600  *	bit 20-28: index to the filter hash table
1601  *	bit  0-19: unique id in the hash bucket.
1602  */
1603 static u_long
1604 get_filt_handle(struct acc_classifier *classifier, int i)
1605 {
1606 	static u_long handle_number = 1;
1607 	u_long 	handle;
1608 	struct acc_filter *afp;
1609 
1610 	while (1) {
1611 		handle = handle_number++ & 0x000fffff;
1612 
1613 		if (LIST_EMPTY(&classifier->acc_filters[i]))
1614 			break;
1615 
1616 		LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1617 			if ((afp->f_handle & 0x000fffff) == handle)
1618 				break;
1619 		if (afp == NULL)
1620 			break;
1621 		/* this handle is already used, try again */
1622 	}
1623 
1624 	return ((i << 20) | handle);
1625 }
1626 
1627 /* convert filter handle to filter pointer */
1628 static struct acc_filter *
1629 filth_to_filtp(struct acc_classifier *classifier, u_long handle)
1630 {
1631 	struct acc_filter *afp;
1632 	int	i;
1633 
1634 	i = ACC_GET_HINDEX(handle);
1635 
1636 	LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1637 		if (afp->f_handle == handle)
1638 			return (afp);
1639 
1640 	return (NULL);
1641 }
1642 
1643 /* create flowinfo bitmask */
1644 static u_int32_t
1645 filt2fibmask(struct flow_filter *filt)
1646 {
1647 	u_int32_t mask = 0;
1648 #ifdef INET6
1649 	struct flow_filter6 *filt6;
1650 #endif
1651 
1652 	switch (filt->ff_flow.fi_family) {
1653 	case AF_INET:
1654 		if (filt->ff_flow.fi_proto != 0)
1655 			mask |= FIMB4_PROTO;
1656 		if (filt->ff_flow.fi_tos != 0)
1657 			mask |= FIMB4_TOS;
1658 		if (filt->ff_flow.fi_dst.s_addr != 0)
1659 			mask |= FIMB4_DADDR;
1660 		if (filt->ff_flow.fi_src.s_addr != 0)
1661 			mask |= FIMB4_SADDR;
1662 		if (filt->ff_flow.fi_sport != 0)
1663 			mask |= FIMB4_SPORT;
1664 		if (filt->ff_flow.fi_dport != 0)
1665 			mask |= FIMB4_DPORT;
1666 		if (filt->ff_flow.fi_gpi != 0)
1667 			mask |= FIMB4_GPI;
1668 		break;
1669 #ifdef INET6
1670 	case AF_INET6:
1671 		filt6 = (struct flow_filter6 *)filt;
1672 
1673 		if (filt6->ff_flow6.fi6_proto != 0)
1674 			mask |= FIMB6_PROTO;
1675 		if (filt6->ff_flow6.fi6_tclass != 0)
1676 			mask |= FIMB6_TCLASS;
1677 		if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_dst))
1678 			mask |= FIMB6_DADDR;
1679 		if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_src))
1680 			mask |= FIMB6_SADDR;
1681 		if (filt6->ff_flow6.fi6_sport != 0)
1682 			mask |= FIMB6_SPORT;
1683 		if (filt6->ff_flow6.fi6_dport != 0)
1684 			mask |= FIMB6_DPORT;
1685 		if (filt6->ff_flow6.fi6_gpi != 0)
1686 			mask |= FIMB6_GPI;
1687 		if (filt6->ff_flow6.fi6_flowlabel != 0)
1688 			mask |= FIMB6_FLABEL;
1689 		break;
1690 #endif /* INET6 */
1691 	}
1692 	return (mask);
1693 }
1694 
1695 
1696 /*
1697  * helper functions to handle IPv4 fragments.
1698  * currently only in-sequence fragments are handled.
1699  *	- fragment info is cached in a LRU list.
1700  *	- when a first fragment is found, cache its flow info.
1701  *	- when a non-first fragment is found, lookup the cache.
1702  */
1703 
1704 struct ip4_frag {
1705     TAILQ_ENTRY(ip4_frag) ip4f_chain;
1706     char    ip4f_valid;
1707     u_short ip4f_id;
1708     struct flowinfo_in ip4f_info;
1709 };
1710 
1711 static TAILQ_HEAD(ip4f_list, ip4_frag) ip4f_list; /* IPv4 fragment cache */
1712 
1713 #define	IP4F_TABSIZE		16	/* IPv4 fragment cache size */
1714 
1715 
1716 static void
1717 ip4f_cache(struct ip *ip, struct flowinfo_in *fin)
1718 {
1719 	struct ip4_frag *fp;
1720 
1721 	if (TAILQ_EMPTY(&ip4f_list)) {
1722 		/* first time call, allocate fragment cache entries. */
1723 		if (ip4f_init() < 0)
1724 			/* allocation failed! */
1725 			return;
1726 	}
1727 
1728 	fp = ip4f_alloc();
1729 	fp->ip4f_id = ip->ip_id;
1730 	fp->ip4f_info.fi_proto = ip->ip_p;
1731 	fp->ip4f_info.fi_src.s_addr = ip->ip_src.s_addr;
1732 	fp->ip4f_info.fi_dst.s_addr = ip->ip_dst.s_addr;
1733 
1734 	/* save port numbers */
1735 	fp->ip4f_info.fi_sport = fin->fi_sport;
1736 	fp->ip4f_info.fi_dport = fin->fi_dport;
1737 	fp->ip4f_info.fi_gpi   = fin->fi_gpi;
1738 }
1739 
1740 static int
1741 ip4f_lookup(struct ip *ip, struct flowinfo_in *fin)
1742 {
1743 	struct ip4_frag *fp;
1744 
1745 	for (fp = TAILQ_FIRST(&ip4f_list); fp != NULL && fp->ip4f_valid;
1746 	     fp = TAILQ_NEXT(fp, ip4f_chain))
1747 		if (ip->ip_id == fp->ip4f_id &&
1748 		    ip->ip_src.s_addr == fp->ip4f_info.fi_src.s_addr &&
1749 		    ip->ip_dst.s_addr == fp->ip4f_info.fi_dst.s_addr &&
1750 		    ip->ip_p == fp->ip4f_info.fi_proto) {
1751 
1752 			/* found the matching entry */
1753 			fin->fi_sport = fp->ip4f_info.fi_sport;
1754 			fin->fi_dport = fp->ip4f_info.fi_dport;
1755 			fin->fi_gpi   = fp->ip4f_info.fi_gpi;
1756 
1757 			if ((ntohs(ip->ip_off) & IP_MF) == 0)
1758 				/* this is the last fragment,
1759 				   release the entry. */
1760 				ip4f_free(fp);
1761 
1762 			return (1);
1763 		}
1764 
1765 	/* no matching entry found */
1766 	return (0);
1767 }
1768 
1769 static int
1770 ip4f_init(void)
1771 {
1772 	struct ip4_frag *fp;
1773 	int i;
1774 
1775 	TAILQ_INIT(&ip4f_list);
1776 	for (i=0; i<IP4F_TABSIZE; i++) {
1777 		fp = malloc(sizeof(struct ip4_frag), M_DEVBUF, M_NOWAIT);
1778 		if (fp == NULL) {
1779 			printf("ip4f_init: can't alloc %dth entry!\n", i);
1780 			if (i == 0)
1781 				return (-1);
1782 			return (0);
1783 		}
1784 		fp->ip4f_valid = 0;
1785 		TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1786 	}
1787 	return (0);
1788 }
1789 
1790 static struct ip4_frag *
1791 ip4f_alloc(void)
1792 {
1793 	struct ip4_frag *fp;
1794 
1795 	/* reclaim an entry at the tail, put it at the head */
1796 	fp = TAILQ_LAST(&ip4f_list, ip4f_list);
1797 	TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1798 	fp->ip4f_valid = 1;
1799 	TAILQ_INSERT_HEAD(&ip4f_list, fp, ip4f_chain);
1800 	return (fp);
1801 }
1802 
1803 static void
1804 ip4f_free(struct ip4_frag *fp)
1805 {
1806 	TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1807 	fp->ip4f_valid = 0;
1808 	TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1809 }
1810 
1811 #endif /* ALTQ3_CLFIER_COMPAT */
1812