xref: /freebsd/sys/net/altq/altq_subr.c (revision 3750ccefb8629a08890bfbae894dd6bc6a7483b4)
1 /*-
2  * Copyright (C) 1997-2003
3  *	Sony Computer Science Laboratories Inc.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $KAME: altq_subr.c,v 1.21 2003/11/06 06:32:53 kjc Exp $
27  */
28 
29 #include "opt_altq.h"
30 #include "opt_inet.h"
31 #include "opt_inet6.h"
32 
33 #include <sys/param.h>
34 #include <sys/malloc.h>
35 #include <sys/mbuf.h>
36 #include <sys/systm.h>
37 #include <sys/proc.h>
38 #include <sys/socket.h>
39 #include <sys/socketvar.h>
40 #include <sys/kernel.h>
41 #include <sys/errno.h>
42 #include <sys/syslog.h>
43 #include <sys/sysctl.h>
44 #include <sys/queue.h>
45 
46 #include <net/if.h>
47 #include <net/if_var.h>
48 #include <net/if_private.h>
49 #include <net/if_dl.h>
50 #include <net/if_types.h>
51 #include <net/vnet.h>
52 
53 #include <netinet/in.h>
54 #include <netinet/in_systm.h>
55 #include <netinet/ip.h>
56 #ifdef INET6
57 #include <netinet/ip6.h>
58 #endif
59 #include <netinet/tcp.h>
60 #include <netinet/udp.h>
61 
62 #include <netpfil/pf/pf.h>
63 #include <netpfil/pf/pf_altq.h>
64 #include <net/altq/altq.h>
65 
66 /* machine dependent clock related includes */
67 #include <sys/bus.h>
68 #include <sys/cpu.h>
69 #include <sys/eventhandler.h>
70 #include <machine/clock.h>
71 #if defined(__amd64__) || defined(__i386__)
72 #include <machine/cpufunc.h>		/* for pentium tsc */
73 #include <machine/specialreg.h>		/* for CPUID_TSC */
74 #include <machine/md_var.h>		/* for cpu_feature */
75 #endif /* __amd64 || __i386__ */
76 
77 /*
78  * internal function prototypes
79  */
80 static void	tbr_timeout(void *);
81 static struct mbuf *tbr_dequeue(struct ifaltq *, int);
82 static int tbr_timer = 0;	/* token bucket regulator timer */
83 static struct callout tbr_callout;
84 
85 #ifdef ALTQ3_CLFIER_COMPAT
86 static int 	extract_ports4(struct mbuf *, struct ip *, struct flowinfo_in *);
87 #ifdef INET6
88 static int 	extract_ports6(struct mbuf *, struct ip6_hdr *,
89 			       struct flowinfo_in6 *);
90 #endif
91 static int	apply_filter4(u_int32_t, struct flow_filter *,
92 			      struct flowinfo_in *);
93 static int	apply_ppfilter4(u_int32_t, struct flow_filter *,
94 				struct flowinfo_in *);
95 #ifdef INET6
96 static int	apply_filter6(u_int32_t, struct flow_filter6 *,
97 			      struct flowinfo_in6 *);
98 #endif
99 static int	apply_tosfilter4(u_int32_t, struct flow_filter *,
100 				 struct flowinfo_in *);
101 static u_long	get_filt_handle(struct acc_classifier *, int);
102 static struct acc_filter *filth_to_filtp(struct acc_classifier *, u_long);
103 static u_int32_t filt2fibmask(struct flow_filter *);
104 
105 static void 	ip4f_cache(struct ip *, struct flowinfo_in *);
106 static int 	ip4f_lookup(struct ip *, struct flowinfo_in *);
107 static int 	ip4f_init(void);
108 static struct ip4_frag	*ip4f_alloc(void);
109 static void 	ip4f_free(struct ip4_frag *);
110 #endif /* ALTQ3_CLFIER_COMPAT */
111 
112 #ifdef ALTQ
113 SYSCTL_NODE(_kern_features, OID_AUTO, altq, CTLFLAG_RD | CTLFLAG_CAPRD, 0,
114     "ALTQ packet queuing");
115 
116 #define	ALTQ_FEATURE(name, desc)					\
117 	SYSCTL_INT_WITH_LABEL(_kern_features_altq, OID_AUTO, name,	\
118 	    CTLFLAG_RD | CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, 1,		\
119 	    desc, "feature")
120 
121 #ifdef ALTQ_CBQ
122 ALTQ_FEATURE(cbq, "ALTQ Class Based Queuing discipline");
123 #endif
124 #ifdef ALTQ_CODEL
125 ALTQ_FEATURE(codel, "ALTQ Controlled Delay discipline");
126 #endif
127 #ifdef ALTQ_RED
128 ALTQ_FEATURE(red, "ALTQ Random Early Detection discipline");
129 #endif
130 #ifdef ALTQ_RIO
131 ALTQ_FEATURE(rio, "ALTQ Random Early Drop discipline");
132 #endif
133 #ifdef ALTQ_HFSC
134 ALTQ_FEATURE(hfsc, "ALTQ Hierarchical Packet Scheduler discipline");
135 #endif
136 #ifdef ALTQ_PRIQ
137 ALTQ_FEATURE(priq, "ATLQ Priority Queuing discipline");
138 #endif
139 #ifdef ALTQ_FAIRQ
140 ALTQ_FEATURE(fairq, "ALTQ Fair Queuing discipline");
141 #endif
142 #endif
143 
144 /*
145  * alternate queueing support routines
146  */
147 
148 /* look up the queue state by the interface name and the queueing type. */
149 void *
150 altq_lookup(char *name, int type)
151 {
152 	struct ifnet *ifp;
153 
154 	if ((ifp = ifunit(name)) != NULL) {
155 		/* read if_snd unlocked */
156 		if (type != ALTQT_NONE && ifp->if_snd.altq_type == type)
157 			return (ifp->if_snd.altq_disc);
158 	}
159 
160 	return NULL;
161 }
162 
163 int
164 altq_attach(struct ifaltq *ifq, int type, void *discipline,
165 	int (*enqueue)(struct ifaltq *, struct mbuf *, struct altq_pktattr *),
166 	struct mbuf *(*dequeue)(struct ifaltq *, int),
167 	int (*request)(struct ifaltq *, int, void *))
168 {
169 	IFQ_LOCK(ifq);
170 	if (!ALTQ_IS_READY(ifq)) {
171 		IFQ_UNLOCK(ifq);
172 		return ENXIO;
173 	}
174 
175 	ifq->altq_type     = type;
176 	ifq->altq_disc     = discipline;
177 	ifq->altq_enqueue  = enqueue;
178 	ifq->altq_dequeue  = dequeue;
179 	ifq->altq_request  = request;
180 	ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED);
181 	IFQ_UNLOCK(ifq);
182 	return 0;
183 }
184 
185 int
186 altq_detach(struct ifaltq *ifq)
187 {
188 	IFQ_LOCK(ifq);
189 
190 	if (!ALTQ_IS_READY(ifq)) {
191 		IFQ_UNLOCK(ifq);
192 		return ENXIO;
193 	}
194 	if (ALTQ_IS_ENABLED(ifq)) {
195 		IFQ_UNLOCK(ifq);
196 		return EBUSY;
197 	}
198 	if (!ALTQ_IS_ATTACHED(ifq)) {
199 		IFQ_UNLOCK(ifq);
200 		return (0);
201 	}
202 
203 	ifq->altq_type     = ALTQT_NONE;
204 	ifq->altq_disc     = NULL;
205 	ifq->altq_enqueue  = NULL;
206 	ifq->altq_dequeue  = NULL;
207 	ifq->altq_request  = NULL;
208 	ifq->altq_flags &= ALTQF_CANTCHANGE;
209 
210 	IFQ_UNLOCK(ifq);
211 	return 0;
212 }
213 
214 int
215 altq_enable(struct ifaltq *ifq)
216 {
217 	int s;
218 
219 	IFQ_LOCK(ifq);
220 
221 	if (!ALTQ_IS_READY(ifq)) {
222 		IFQ_UNLOCK(ifq);
223 		return ENXIO;
224 	}
225 	if (ALTQ_IS_ENABLED(ifq)) {
226 		IFQ_UNLOCK(ifq);
227 		return 0;
228 	}
229 
230 	s = splnet();
231 	IFQ_PURGE_NOLOCK(ifq);
232 	ASSERT(ifq->ifq_len == 0);
233 	ifq->ifq_drv_maxlen = 0;		/* disable bulk dequeue */
234 	ifq->altq_flags |= ALTQF_ENABLED;
235 	splx(s);
236 
237 	IFQ_UNLOCK(ifq);
238 	return 0;
239 }
240 
241 int
242 altq_disable(struct ifaltq *ifq)
243 {
244 	int s;
245 
246 	IFQ_LOCK(ifq);
247 	if (!ALTQ_IS_ENABLED(ifq)) {
248 		IFQ_UNLOCK(ifq);
249 		return 0;
250 	}
251 
252 	s = splnet();
253 	IFQ_PURGE_NOLOCK(ifq);
254 	ASSERT(ifq->ifq_len == 0);
255 	ifq->altq_flags &= ~(ALTQF_ENABLED);
256 	splx(s);
257 
258 	IFQ_UNLOCK(ifq);
259 	return 0;
260 }
261 
262 #ifdef ALTQ_DEBUG
263 void
264 altq_assert(const char *file, int line, const char *failedexpr)
265 {
266 	(void)printf("altq assertion \"%s\" failed: file \"%s\", line %d\n",
267 		     failedexpr, file, line);
268 	panic("altq assertion");
269 	/* NOTREACHED */
270 }
271 #endif
272 
273 /*
274  * internal representation of token bucket parameters
275  *	rate:	(byte_per_unittime << TBR_SHIFT)  / machclk_freq
276  *		(((bits_per_sec) / 8) << TBR_SHIFT) / machclk_freq
277  *	depth:	byte << TBR_SHIFT
278  *
279  */
280 #define	TBR_SHIFT	29
281 #define	TBR_SCALE(x)	((int64_t)(x) << TBR_SHIFT)
282 #define	TBR_UNSCALE(x)	((x) >> TBR_SHIFT)
283 
284 static struct mbuf *
285 tbr_dequeue(struct ifaltq *ifq, int op)
286 {
287 	struct tb_regulator *tbr;
288 	struct mbuf *m;
289 	int64_t interval;
290 	u_int64_t now;
291 
292 	IFQ_LOCK_ASSERT(ifq);
293 	tbr = ifq->altq_tbr;
294 	if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) {
295 		/* if this is a remove after poll, bypass tbr check */
296 	} else {
297 		/* update token only when it is negative */
298 		if (tbr->tbr_token <= 0) {
299 			now = read_machclk();
300 			interval = now - tbr->tbr_last;
301 			if (interval >= tbr->tbr_filluptime)
302 				tbr->tbr_token = tbr->tbr_depth;
303 			else {
304 				tbr->tbr_token += interval * tbr->tbr_rate;
305 				if (tbr->tbr_token > tbr->tbr_depth)
306 					tbr->tbr_token = tbr->tbr_depth;
307 			}
308 			tbr->tbr_last = now;
309 		}
310 		/* if token is still negative, don't allow dequeue */
311 		if (tbr->tbr_token <= 0)
312 			return (NULL);
313 	}
314 
315 	if (ALTQ_IS_ENABLED(ifq))
316 		m = (*ifq->altq_dequeue)(ifq, op);
317 	else {
318 		if (op == ALTDQ_POLL)
319 			_IF_POLL(ifq, m);
320 		else
321 			_IF_DEQUEUE(ifq, m);
322 	}
323 
324 	if (m != NULL && op == ALTDQ_REMOVE)
325 		tbr->tbr_token -= TBR_SCALE(m_pktlen(m));
326 	tbr->tbr_lastop = op;
327 	return (m);
328 }
329 
330 /*
331  * set a token bucket regulator.
332  * if the specified rate is zero, the token bucket regulator is deleted.
333  */
334 int
335 tbr_set(struct ifaltq *ifq, struct tb_profile *profile)
336 {
337 	struct tb_regulator *tbr, *otbr;
338 
339 	if (tbr_dequeue_ptr == NULL)
340 		tbr_dequeue_ptr = tbr_dequeue;
341 
342 	if (machclk_freq == 0)
343 		init_machclk();
344 	if (machclk_freq == 0) {
345 		printf("tbr_set: no cpu clock available!\n");
346 		return (ENXIO);
347 	}
348 
349 	IFQ_LOCK(ifq);
350 	if (profile->rate == 0) {
351 		/* delete this tbr */
352 		if ((tbr = ifq->altq_tbr) == NULL) {
353 			IFQ_UNLOCK(ifq);
354 			return (ENOENT);
355 		}
356 		ifq->altq_tbr = NULL;
357 		free(tbr, M_DEVBUF);
358 		IFQ_UNLOCK(ifq);
359 		return (0);
360 	}
361 
362 	tbr = malloc(sizeof(struct tb_regulator), M_DEVBUF, M_NOWAIT | M_ZERO);
363 	if (tbr == NULL) {
364 		IFQ_UNLOCK(ifq);
365 		return (ENOMEM);
366 	}
367 
368 	tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq;
369 	tbr->tbr_depth = TBR_SCALE(profile->depth);
370 	if (tbr->tbr_rate > 0)
371 		tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate;
372 	else
373 		tbr->tbr_filluptime = LLONG_MAX;
374 	/*
375 	 *  The longest time between tbr_dequeue() calls will be about 1
376 	 *  system tick, as the callout that drives it is scheduled once per
377 	 *  tick.  The refill-time detection logic in tbr_dequeue() can only
378 	 *  properly detect the passage of up to LLONG_MAX machclk ticks.
379 	 *  Therefore, in order for this logic to function properly in the
380 	 *  extreme case, the maximum value of tbr_filluptime should be
381 	 *  LLONG_MAX less one system tick's worth of machclk ticks less
382 	 *  some additional slop factor (here one more system tick's worth
383 	 *  of machclk ticks).
384 	 */
385 	if (tbr->tbr_filluptime > (LLONG_MAX - 2 * machclk_per_tick))
386 		tbr->tbr_filluptime = LLONG_MAX - 2 * machclk_per_tick;
387 	tbr->tbr_token = tbr->tbr_depth;
388 	tbr->tbr_last = read_machclk();
389 	tbr->tbr_lastop = ALTDQ_REMOVE;
390 
391 	otbr = ifq->altq_tbr;
392 	ifq->altq_tbr = tbr;	/* set the new tbr */
393 
394 	if (otbr != NULL)
395 		free(otbr, M_DEVBUF);
396 	else {
397 		if (tbr_timer == 0) {
398 			CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
399 			tbr_timer = 1;
400 		}
401 	}
402 	IFQ_UNLOCK(ifq);
403 	return (0);
404 }
405 
406 /*
407  * tbr_timeout goes through the interface list, and kicks the drivers
408  * if necessary.
409  *
410  * MPSAFE
411  */
412 static void
413 tbr_timeout(void *arg)
414 {
415 	VNET_ITERATOR_DECL(vnet_iter);
416 	struct ifnet *ifp;
417 	struct epoch_tracker et;
418 	int active;
419 
420 	active = 0;
421 	NET_EPOCH_ENTER(et);
422 	VNET_LIST_RLOCK_NOSLEEP();
423 	VNET_FOREACH(vnet_iter) {
424 		CURVNET_SET(vnet_iter);
425 		for (ifp = CK_STAILQ_FIRST(&V_ifnet); ifp;
426 		    ifp = CK_STAILQ_NEXT(ifp, if_link)) {
427 			/* read from if_snd unlocked */
428 			if (!TBR_IS_ENABLED(&ifp->if_snd))
429 				continue;
430 			active++;
431 			if (!IFQ_IS_EMPTY(&ifp->if_snd) &&
432 			    ifp->if_start != NULL)
433 				(*ifp->if_start)(ifp);
434 		}
435 		CURVNET_RESTORE();
436 	}
437 	VNET_LIST_RUNLOCK_NOSLEEP();
438 	NET_EPOCH_EXIT(et);
439 	if (active > 0)
440 		CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
441 	else
442 		tbr_timer = 0;	/* don't need tbr_timer anymore */
443 }
444 
445 /*
446  * attach a discipline to the interface.  if one already exists, it is
447  * overridden.
448  * Locking is done in the discipline specific attach functions. Basically
449  * they call back to altq_attach which takes care of the attach and locking.
450  */
451 int
452 altq_pfattach(struct pf_altq *a)
453 {
454 	int error = 0;
455 
456 	switch (a->scheduler) {
457 	case ALTQT_NONE:
458 		break;
459 #ifdef ALTQ_CBQ
460 	case ALTQT_CBQ:
461 		error = cbq_pfattach(a);
462 		break;
463 #endif
464 #ifdef ALTQ_PRIQ
465 	case ALTQT_PRIQ:
466 		error = priq_pfattach(a);
467 		break;
468 #endif
469 #ifdef ALTQ_HFSC
470 	case ALTQT_HFSC:
471 		error = hfsc_pfattach(a);
472 		break;
473 #endif
474 #ifdef ALTQ_FAIRQ
475 	case ALTQT_FAIRQ:
476 		error = fairq_pfattach(a);
477 		break;
478 #endif
479 #ifdef ALTQ_CODEL
480 	case ALTQT_CODEL:
481 		error = codel_pfattach(a);
482 		break;
483 #endif
484 	default:
485 		error = ENXIO;
486 	}
487 
488 	return (error);
489 }
490 
491 /*
492  * detach a discipline from the interface.
493  * it is possible that the discipline was already overridden by another
494  * discipline.
495  */
496 int
497 altq_pfdetach(struct pf_altq *a)
498 {
499 	struct ifnet *ifp;
500 	int s, error = 0;
501 
502 	if ((ifp = ifunit(a->ifname)) == NULL)
503 		return (EINVAL);
504 
505 	/* if this discipline is no longer referenced, just return */
506 	/* read unlocked from if_snd */
507 	if (a->altq_disc == NULL || a->altq_disc != ifp->if_snd.altq_disc)
508 		return (0);
509 
510 	s = splnet();
511 	/* read unlocked from if_snd, _disable and _detach take care */
512 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
513 		error = altq_disable(&ifp->if_snd);
514 	if (error == 0)
515 		error = altq_detach(&ifp->if_snd);
516 	splx(s);
517 
518 	return (error);
519 }
520 
521 /*
522  * add a discipline or a queue
523  * Locking is done in the discipline specific functions with regards to
524  * malloc with WAITOK, also it is not yet clear which lock to use.
525  */
526 int
527 altq_add(struct ifnet *ifp, struct pf_altq *a)
528 {
529 	int error = 0;
530 
531 	if (a->qname[0] != 0)
532 		return (altq_add_queue(a));
533 
534 	if (machclk_freq == 0)
535 		init_machclk();
536 	if (machclk_freq == 0)
537 		panic("altq_add: no cpu clock");
538 
539 	switch (a->scheduler) {
540 #ifdef ALTQ_CBQ
541 	case ALTQT_CBQ:
542 		error = cbq_add_altq(ifp, a);
543 		break;
544 #endif
545 #ifdef ALTQ_PRIQ
546 	case ALTQT_PRIQ:
547 		error = priq_add_altq(ifp, a);
548 		break;
549 #endif
550 #ifdef ALTQ_HFSC
551 	case ALTQT_HFSC:
552 		error = hfsc_add_altq(ifp, a);
553 		break;
554 #endif
555 #ifdef ALTQ_FAIRQ
556         case ALTQT_FAIRQ:
557                 error = fairq_add_altq(ifp, a);
558                 break;
559 #endif
560 #ifdef ALTQ_CODEL
561 	case ALTQT_CODEL:
562 		error = codel_add_altq(ifp, a);
563 		break;
564 #endif
565 	default:
566 		error = ENXIO;
567 	}
568 
569 	return (error);
570 }
571 
572 /*
573  * remove a discipline or a queue
574  * It is yet unclear what lock to use to protect this operation, the
575  * discipline specific functions will determine and grab it
576  */
577 int
578 altq_remove(struct pf_altq *a)
579 {
580 	int error = 0;
581 
582 	if (a->qname[0] != 0)
583 		return (altq_remove_queue(a));
584 
585 	switch (a->scheduler) {
586 #ifdef ALTQ_CBQ
587 	case ALTQT_CBQ:
588 		error = cbq_remove_altq(a);
589 		break;
590 #endif
591 #ifdef ALTQ_PRIQ
592 	case ALTQT_PRIQ:
593 		error = priq_remove_altq(a);
594 		break;
595 #endif
596 #ifdef ALTQ_HFSC
597 	case ALTQT_HFSC:
598 		error = hfsc_remove_altq(a);
599 		break;
600 #endif
601 #ifdef ALTQ_FAIRQ
602         case ALTQT_FAIRQ:
603                 error = fairq_remove_altq(a);
604                 break;
605 #endif
606 #ifdef ALTQ_CODEL
607 	case ALTQT_CODEL:
608 		error = codel_remove_altq(a);
609 		break;
610 #endif
611 	default:
612 		error = ENXIO;
613 	}
614 
615 	return (error);
616 }
617 
618 /*
619  * add a queue to the discipline
620  * It is yet unclear what lock to use to protect this operation, the
621  * discipline specific functions will determine and grab it
622  */
623 int
624 altq_add_queue(struct pf_altq *a)
625 {
626 	int error = 0;
627 
628 	switch (a->scheduler) {
629 #ifdef ALTQ_CBQ
630 	case ALTQT_CBQ:
631 		error = cbq_add_queue(a);
632 		break;
633 #endif
634 #ifdef ALTQ_PRIQ
635 	case ALTQT_PRIQ:
636 		error = priq_add_queue(a);
637 		break;
638 #endif
639 #ifdef ALTQ_HFSC
640 	case ALTQT_HFSC:
641 		error = hfsc_add_queue(a);
642 		break;
643 #endif
644 #ifdef ALTQ_FAIRQ
645         case ALTQT_FAIRQ:
646                 error = fairq_add_queue(a);
647                 break;
648 #endif
649 	default:
650 		error = ENXIO;
651 	}
652 
653 	return (error);
654 }
655 
656 /*
657  * remove a queue from the discipline
658  * It is yet unclear what lock to use to protect this operation, the
659  * discipline specific functions will determine and grab it
660  */
661 int
662 altq_remove_queue(struct pf_altq *a)
663 {
664 	int error = 0;
665 
666 	switch (a->scheduler) {
667 #ifdef ALTQ_CBQ
668 	case ALTQT_CBQ:
669 		error = cbq_remove_queue(a);
670 		break;
671 #endif
672 #ifdef ALTQ_PRIQ
673 	case ALTQT_PRIQ:
674 		error = priq_remove_queue(a);
675 		break;
676 #endif
677 #ifdef ALTQ_HFSC
678 	case ALTQT_HFSC:
679 		error = hfsc_remove_queue(a);
680 		break;
681 #endif
682 #ifdef ALTQ_FAIRQ
683         case ALTQT_FAIRQ:
684                 error = fairq_remove_queue(a);
685                 break;
686 #endif
687 	default:
688 		error = ENXIO;
689 	}
690 
691 	return (error);
692 }
693 
694 /*
695  * get queue statistics
696  * Locking is done in the discipline specific functions with regards to
697  * copyout operations, also it is not yet clear which lock to use.
698  */
699 int
700 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version)
701 {
702 	int error = 0;
703 
704 	switch (a->scheduler) {
705 #ifdef ALTQ_CBQ
706 	case ALTQT_CBQ:
707 		error = cbq_getqstats(a, ubuf, nbytes, version);
708 		break;
709 #endif
710 #ifdef ALTQ_PRIQ
711 	case ALTQT_PRIQ:
712 		error = priq_getqstats(a, ubuf, nbytes, version);
713 		break;
714 #endif
715 #ifdef ALTQ_HFSC
716 	case ALTQT_HFSC:
717 		error = hfsc_getqstats(a, ubuf, nbytes, version);
718 		break;
719 #endif
720 #ifdef ALTQ_FAIRQ
721         case ALTQT_FAIRQ:
722                 error = fairq_getqstats(a, ubuf, nbytes, version);
723                 break;
724 #endif
725 #ifdef ALTQ_CODEL
726 	case ALTQT_CODEL:
727 		error = codel_getqstats(a, ubuf, nbytes, version);
728 		break;
729 #endif
730 	default:
731 		error = ENXIO;
732 	}
733 
734 	return (error);
735 }
736 
737 /*
738  * read and write diffserv field in IPv4 or IPv6 header
739  */
740 u_int8_t
741 read_dsfield(struct mbuf *m, struct altq_pktattr *pktattr)
742 {
743 	struct mbuf *m0;
744 	u_int8_t ds_field = 0;
745 
746 	if (pktattr == NULL ||
747 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
748 		return ((u_int8_t)0);
749 
750 	/* verify that pattr_hdr is within the mbuf data */
751 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
752 		if ((pktattr->pattr_hdr >= m0->m_data) &&
753 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
754 			break;
755 	if (m0 == NULL) {
756 		/* ick, pattr_hdr is stale */
757 		pktattr->pattr_af = AF_UNSPEC;
758 #ifdef ALTQ_DEBUG
759 		printf("read_dsfield: can't locate header!\n");
760 #endif
761 		return ((u_int8_t)0);
762 	}
763 
764 	if (pktattr->pattr_af == AF_INET) {
765 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
766 
767 		if (ip->ip_v != 4)
768 			return ((u_int8_t)0);	/* version mismatch! */
769 		ds_field = ip->ip_tos;
770 	}
771 #ifdef INET6
772 	else if (pktattr->pattr_af == AF_INET6) {
773 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
774 		u_int32_t flowlabel;
775 
776 		flowlabel = ntohl(ip6->ip6_flow);
777 		if ((flowlabel >> 28) != 6)
778 			return ((u_int8_t)0);	/* version mismatch! */
779 		ds_field = (flowlabel >> 20) & 0xff;
780 	}
781 #endif
782 	return (ds_field);
783 }
784 
785 void
786 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, u_int8_t dsfield)
787 {
788 	struct mbuf *m0;
789 
790 	if (pktattr == NULL ||
791 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
792 		return;
793 
794 	/* verify that pattr_hdr is within the mbuf data */
795 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
796 		if ((pktattr->pattr_hdr >= m0->m_data) &&
797 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
798 			break;
799 	if (m0 == NULL) {
800 		/* ick, pattr_hdr is stale */
801 		pktattr->pattr_af = AF_UNSPEC;
802 #ifdef ALTQ_DEBUG
803 		printf("write_dsfield: can't locate header!\n");
804 #endif
805 		return;
806 	}
807 
808 	if (pktattr->pattr_af == AF_INET) {
809 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
810 		u_int8_t old;
811 		int32_t sum;
812 
813 		if (ip->ip_v != 4)
814 			return;		/* version mismatch! */
815 		old = ip->ip_tos;
816 		dsfield |= old & 3;	/* leave CU bits */
817 		if (old == dsfield)
818 			return;
819 		ip->ip_tos = dsfield;
820 		/*
821 		 * update checksum (from RFC1624)
822 		 *	   HC' = ~(~HC + ~m + m')
823 		 */
824 		sum = ~ntohs(ip->ip_sum) & 0xffff;
825 		sum += 0xff00 + (~old & 0xff) + dsfield;
826 		sum = (sum >> 16) + (sum & 0xffff);
827 		sum += (sum >> 16);  /* add carry */
828 
829 		ip->ip_sum = htons(~sum & 0xffff);
830 	}
831 #ifdef INET6
832 	else if (pktattr->pattr_af == AF_INET6) {
833 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
834 		u_int32_t flowlabel;
835 
836 		flowlabel = ntohl(ip6->ip6_flow);
837 		if ((flowlabel >> 28) != 6)
838 			return;		/* version mismatch! */
839 		flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20);
840 		ip6->ip6_flow = htonl(flowlabel);
841 	}
842 #endif
843 	return;
844 }
845 
846 /*
847  * high resolution clock support taking advantage of a machine dependent
848  * high resolution time counter (e.g., timestamp counter of intel pentium).
849  * we assume
850  *  - 64-bit-long monotonically-increasing counter
851  *  - frequency range is 100M-4GHz (CPU speed)
852  */
853 /* if pcc is not available or disabled, emulate 256MHz using microtime() */
854 #define	MACHCLK_SHIFT	8
855 
856 int machclk_usepcc;
857 u_int32_t machclk_freq;
858 u_int32_t machclk_per_tick;
859 
860 #if defined(__i386__) && defined(__NetBSD__)
861 extern u_int64_t cpu_tsc_freq;
862 #endif
863 
864 /* Update TSC freq with the value indicated by the caller. */
865 static void
866 tsc_freq_changed(void *arg, const struct cf_level *level, int status)
867 {
868 	/* If there was an error during the transition, don't do anything. */
869 	if (status != 0)
870 		return;
871 
872 #if defined(__amd64__) || defined(__i386__)
873 	/* If TSC is P-state invariant, don't do anything. */
874 	if (tsc_is_invariant)
875 		return;
876 #endif
877 
878 	/* Total setting for this level gives the new frequency in MHz. */
879 	init_machclk();
880 }
881 EVENTHANDLER_DEFINE(cpufreq_post_change, tsc_freq_changed, NULL,
882     EVENTHANDLER_PRI_LAST);
883 
884 static void
885 init_machclk_setup(void)
886 {
887 	callout_init(&tbr_callout, 1);
888 
889 	machclk_usepcc = 1;
890 
891 #if (!defined(__amd64__) && !defined(__i386__)) || defined(ALTQ_NOPCC)
892 	machclk_usepcc = 0;
893 #endif
894 #if defined(__FreeBSD__) && defined(SMP)
895 	machclk_usepcc = 0;
896 #endif
897 #if defined(__NetBSD__) && defined(MULTIPROCESSOR)
898 	machclk_usepcc = 0;
899 #endif
900 #if defined(__amd64__) || defined(__i386__)
901 	/* check if TSC is available */
902 	if ((cpu_feature & CPUID_TSC) == 0 ||
903 	    atomic_load_acq_64(&tsc_freq) == 0)
904 		machclk_usepcc = 0;
905 #endif
906 }
907 
908 void
909 init_machclk(void)
910 {
911 	static int called;
912 
913 	/* Call one-time initialization function. */
914 	if (!called) {
915 		init_machclk_setup();
916 		called = 1;
917 	}
918 
919 	if (machclk_usepcc == 0) {
920 		/* emulate 256MHz using microtime() */
921 		machclk_freq = 1000000 << MACHCLK_SHIFT;
922 		machclk_per_tick = machclk_freq / hz;
923 #ifdef ALTQ_DEBUG
924 		printf("altq: emulate %uHz cpu clock\n", machclk_freq);
925 #endif
926 		return;
927 	}
928 
929 	/*
930 	 * if the clock frequency (of Pentium TSC or Alpha PCC) is
931 	 * accessible, just use it.
932 	 */
933 #if defined(__amd64__) || defined(__i386__)
934 	machclk_freq = atomic_load_acq_64(&tsc_freq);
935 #endif
936 
937 	/*
938 	 * if we don't know the clock frequency, measure it.
939 	 */
940 	if (machclk_freq == 0) {
941 		static int	wait;
942 		struct timeval	tv_start, tv_end;
943 		u_int64_t	start, end, diff;
944 		int		timo;
945 
946 		microtime(&tv_start);
947 		start = read_machclk();
948 		timo = hz;	/* 1 sec */
949 		(void)tsleep(&wait, PWAIT | PCATCH, "init_machclk", timo);
950 		microtime(&tv_end);
951 		end = read_machclk();
952 		diff = (u_int64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000
953 		    + tv_end.tv_usec - tv_start.tv_usec;
954 		if (diff != 0)
955 			machclk_freq = (u_int)((end - start) * 1000000 / diff);
956 	}
957 
958 	machclk_per_tick = machclk_freq / hz;
959 
960 #ifdef ALTQ_DEBUG
961 	printf("altq: CPU clock: %uHz\n", machclk_freq);
962 #endif
963 }
964 
965 #if defined(__OpenBSD__) && defined(__i386__)
966 static __inline u_int64_t
967 rdtsc(void)
968 {
969 	u_int64_t rv;
970 	__asm __volatile(".byte 0x0f, 0x31" : "=A" (rv));
971 	return (rv);
972 }
973 #endif /* __OpenBSD__ && __i386__ */
974 
975 u_int64_t
976 read_machclk(void)
977 {
978 	u_int64_t val;
979 
980 	if (machclk_usepcc) {
981 #if defined(__amd64__) || defined(__i386__)
982 		val = rdtsc();
983 #else
984 		panic("read_machclk");
985 #endif
986 	} else {
987 		struct timeval tv, boottime;
988 
989 		microtime(&tv);
990 		getboottime(&boottime);
991 		val = (((u_int64_t)(tv.tv_sec - boottime.tv_sec) * 1000000
992 		    + tv.tv_usec) << MACHCLK_SHIFT);
993 	}
994 	return (val);
995 }
996 
997 #ifdef ALTQ3_CLFIER_COMPAT
998 
999 #ifndef IPPROTO_ESP
1000 #define	IPPROTO_ESP	50		/* encapsulating security payload */
1001 #endif
1002 #ifndef IPPROTO_AH
1003 #define	IPPROTO_AH	51		/* authentication header */
1004 #endif
1005 
1006 /*
1007  * extract flow information from a given packet.
1008  * filt_mask shows flowinfo fields required.
1009  * we assume the ip header is in one mbuf, and addresses and ports are
1010  * in network byte order.
1011  */
1012 int
1013 altq_extractflow(m, af, flow, filt_bmask)
1014 	struct mbuf *m;
1015 	int af;
1016 	struct flowinfo *flow;
1017 	u_int32_t	filt_bmask;
1018 {
1019 
1020 	switch (af) {
1021 	case PF_INET: {
1022 		struct flowinfo_in *fin;
1023 		struct ip *ip;
1024 
1025 		ip = mtod(m, struct ip *);
1026 
1027 		if (ip->ip_v != 4)
1028 			break;
1029 
1030 		fin = (struct flowinfo_in *)flow;
1031 		fin->fi_len = sizeof(struct flowinfo_in);
1032 		fin->fi_family = AF_INET;
1033 
1034 		fin->fi_proto = ip->ip_p;
1035 		fin->fi_tos = ip->ip_tos;
1036 
1037 		fin->fi_src.s_addr = ip->ip_src.s_addr;
1038 		fin->fi_dst.s_addr = ip->ip_dst.s_addr;
1039 
1040 		if (filt_bmask & FIMB4_PORTS)
1041 			/* if port info is required, extract port numbers */
1042 			extract_ports4(m, ip, fin);
1043 		else {
1044 			fin->fi_sport = 0;
1045 			fin->fi_dport = 0;
1046 			fin->fi_gpi = 0;
1047 		}
1048 		return (1);
1049 	}
1050 
1051 #ifdef INET6
1052 	case PF_INET6: {
1053 		struct flowinfo_in6 *fin6;
1054 		struct ip6_hdr *ip6;
1055 
1056 		ip6 = mtod(m, struct ip6_hdr *);
1057 		/* should we check the ip version? */
1058 
1059 		fin6 = (struct flowinfo_in6 *)flow;
1060 		fin6->fi6_len = sizeof(struct flowinfo_in6);
1061 		fin6->fi6_family = AF_INET6;
1062 
1063 		fin6->fi6_proto = ip6->ip6_nxt;
1064 		fin6->fi6_tclass   = IPV6_TRAFFIC_CLASS(ip6);
1065 
1066 		fin6->fi6_flowlabel = ip6->ip6_flow & htonl(0x000fffff);
1067 		fin6->fi6_src = ip6->ip6_src;
1068 		fin6->fi6_dst = ip6->ip6_dst;
1069 
1070 		if ((filt_bmask & FIMB6_PORTS) ||
1071 		    ((filt_bmask & FIMB6_PROTO)
1072 		     && ip6->ip6_nxt > IPPROTO_IPV6))
1073 			/*
1074 			 * if port info is required, or proto is required
1075 			 * but there are option headers, extract port
1076 			 * and protocol numbers.
1077 			 */
1078 			extract_ports6(m, ip6, fin6);
1079 		else {
1080 			fin6->fi6_sport = 0;
1081 			fin6->fi6_dport = 0;
1082 			fin6->fi6_gpi = 0;
1083 		}
1084 		return (1);
1085 	}
1086 #endif /* INET6 */
1087 
1088 	default:
1089 		break;
1090 	}
1091 
1092 	/* failed */
1093 	flow->fi_len = sizeof(struct flowinfo);
1094 	flow->fi_family = AF_UNSPEC;
1095 	return (0);
1096 }
1097 
1098 /*
1099  * helper routine to extract port numbers
1100  */
1101 /* structure for ipsec and ipv6 option header template */
1102 struct _opt6 {
1103 	u_int8_t	opt6_nxt;	/* next header */
1104 	u_int8_t	opt6_hlen;	/* header extension length */
1105 	u_int16_t	_pad;
1106 	u_int32_t	ah_spi;		/* security parameter index
1107 					   for authentication header */
1108 };
1109 
1110 /*
1111  * extract port numbers from a ipv4 packet.
1112  */
1113 static int
1114 extract_ports4(m, ip, fin)
1115 	struct mbuf *m;
1116 	struct ip *ip;
1117 	struct flowinfo_in *fin;
1118 {
1119 	struct mbuf *m0;
1120 	u_short ip_off;
1121 	u_int8_t proto;
1122 	int 	off;
1123 
1124 	fin->fi_sport = 0;
1125 	fin->fi_dport = 0;
1126 	fin->fi_gpi = 0;
1127 
1128 	ip_off = ntohs(ip->ip_off);
1129 	/* if it is a fragment, try cached fragment info */
1130 	if (ip_off & IP_OFFMASK) {
1131 		ip4f_lookup(ip, fin);
1132 		return (1);
1133 	}
1134 
1135 	/* locate the mbuf containing the protocol header */
1136 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
1137 		if (((caddr_t)ip >= m0->m_data) &&
1138 		    ((caddr_t)ip < m0->m_data + m0->m_len))
1139 			break;
1140 	if (m0 == NULL) {
1141 #ifdef ALTQ_DEBUG
1142 		printf("extract_ports4: can't locate header! ip=%p\n", ip);
1143 #endif
1144 		return (0);
1145 	}
1146 	off = ((caddr_t)ip - m0->m_data) + (ip->ip_hl << 2);
1147 	proto = ip->ip_p;
1148 
1149 #ifdef ALTQ_IPSEC
1150  again:
1151 #endif
1152 	while (off >= m0->m_len) {
1153 		off -= m0->m_len;
1154 		m0 = m0->m_next;
1155 		if (m0 == NULL)
1156 			return (0);  /* bogus ip_hl! */
1157 	}
1158 	if (m0->m_len < off + 4)
1159 		return (0);
1160 
1161 	switch (proto) {
1162 	case IPPROTO_TCP:
1163 	case IPPROTO_UDP: {
1164 		struct udphdr *udp;
1165 
1166 		udp = (struct udphdr *)(mtod(m0, caddr_t) + off);
1167 		fin->fi_sport = udp->uh_sport;
1168 		fin->fi_dport = udp->uh_dport;
1169 		fin->fi_proto = proto;
1170 		}
1171 		break;
1172 
1173 #ifdef ALTQ_IPSEC
1174 	case IPPROTO_ESP:
1175 		if (fin->fi_gpi == 0){
1176 			u_int32_t *gpi;
1177 
1178 			gpi = (u_int32_t *)(mtod(m0, caddr_t) + off);
1179 			fin->fi_gpi   = *gpi;
1180 		}
1181 		fin->fi_proto = proto;
1182 		break;
1183 
1184 	case IPPROTO_AH: {
1185 			/* get next header and header length */
1186 			struct _opt6 *opt6;
1187 
1188 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1189 			proto = opt6->opt6_nxt;
1190 			off += 8 + (opt6->opt6_hlen * 4);
1191 			if (fin->fi_gpi == 0 && m0->m_len >= off + 8)
1192 				fin->fi_gpi = opt6->ah_spi;
1193 		}
1194 		/* goto the next header */
1195 		goto again;
1196 #endif  /* ALTQ_IPSEC */
1197 
1198 	default:
1199 		fin->fi_proto = proto;
1200 		return (0);
1201 	}
1202 
1203 	/* if this is a first fragment, cache it. */
1204 	if (ip_off & IP_MF)
1205 		ip4f_cache(ip, fin);
1206 
1207 	return (1);
1208 }
1209 
1210 #ifdef INET6
1211 static int
1212 extract_ports6(m, ip6, fin6)
1213 	struct mbuf *m;
1214 	struct ip6_hdr *ip6;
1215 	struct flowinfo_in6 *fin6;
1216 {
1217 	struct mbuf *m0;
1218 	int	off;
1219 	u_int8_t proto;
1220 
1221 	fin6->fi6_gpi   = 0;
1222 	fin6->fi6_sport = 0;
1223 	fin6->fi6_dport = 0;
1224 
1225 	/* locate the mbuf containing the protocol header */
1226 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
1227 		if (((caddr_t)ip6 >= m0->m_data) &&
1228 		    ((caddr_t)ip6 < m0->m_data + m0->m_len))
1229 			break;
1230 	if (m0 == NULL) {
1231 #ifdef ALTQ_DEBUG
1232 		printf("extract_ports6: can't locate header! ip6=%p\n", ip6);
1233 #endif
1234 		return (0);
1235 	}
1236 	off = ((caddr_t)ip6 - m0->m_data) + sizeof(struct ip6_hdr);
1237 
1238 	proto = ip6->ip6_nxt;
1239 	do {
1240 		while (off >= m0->m_len) {
1241 			off -= m0->m_len;
1242 			m0 = m0->m_next;
1243 			if (m0 == NULL)
1244 				return (0);
1245 		}
1246 		if (m0->m_len < off + 4)
1247 			return (0);
1248 
1249 		switch (proto) {
1250 		case IPPROTO_TCP:
1251 		case IPPROTO_UDP: {
1252 			struct udphdr *udp;
1253 
1254 			udp = (struct udphdr *)(mtod(m0, caddr_t) + off);
1255 			fin6->fi6_sport = udp->uh_sport;
1256 			fin6->fi6_dport = udp->uh_dport;
1257 			fin6->fi6_proto = proto;
1258 			}
1259 			return (1);
1260 
1261 		case IPPROTO_ESP:
1262 			if (fin6->fi6_gpi == 0) {
1263 				u_int32_t *gpi;
1264 
1265 				gpi = (u_int32_t *)(mtod(m0, caddr_t) + off);
1266 				fin6->fi6_gpi   = *gpi;
1267 			}
1268 			fin6->fi6_proto = proto;
1269 			return (1);
1270 
1271 		case IPPROTO_AH: {
1272 			/* get next header and header length */
1273 			struct _opt6 *opt6;
1274 
1275 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1276 			if (fin6->fi6_gpi == 0 && m0->m_len >= off + 8)
1277 				fin6->fi6_gpi = opt6->ah_spi;
1278 			proto = opt6->opt6_nxt;
1279 			off += 8 + (opt6->opt6_hlen * 4);
1280 			/* goto the next header */
1281 			break;
1282 			}
1283 
1284 		case IPPROTO_HOPOPTS:
1285 		case IPPROTO_ROUTING:
1286 		case IPPROTO_DSTOPTS: {
1287 			/* get next header and header length */
1288 			struct _opt6 *opt6;
1289 
1290 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1291 			proto = opt6->opt6_nxt;
1292 			off += (opt6->opt6_hlen + 1) * 8;
1293 			/* goto the next header */
1294 			break;
1295 			}
1296 
1297 		case IPPROTO_FRAGMENT:
1298 			/* ipv6 fragmentations are not supported yet */
1299 		default:
1300 			fin6->fi6_proto = proto;
1301 			return (0);
1302 		}
1303 	} while (1);
1304 	/*NOTREACHED*/
1305 }
1306 #endif /* INET6 */
1307 
1308 /*
1309  * altq common classifier
1310  */
1311 int
1312 acc_add_filter(classifier, filter, class, phandle)
1313 	struct acc_classifier *classifier;
1314 	struct flow_filter *filter;
1315 	void	*class;
1316 	u_long	*phandle;
1317 {
1318 	struct acc_filter *afp, *prev, *tmp;
1319 	int	i, s;
1320 
1321 #ifdef INET6
1322 	if (filter->ff_flow.fi_family != AF_INET &&
1323 	    filter->ff_flow.fi_family != AF_INET6)
1324 		return (EINVAL);
1325 #else
1326 	if (filter->ff_flow.fi_family != AF_INET)
1327 		return (EINVAL);
1328 #endif
1329 
1330 	afp = malloc(sizeof(*afp), M_DEVBUF, M_WAITOK | M_ZERO);
1331 	afp->f_filter = *filter;
1332 	afp->f_class = class;
1333 
1334 	i = ACC_WILDCARD_INDEX;
1335 	if (filter->ff_flow.fi_family == AF_INET) {
1336 		struct flow_filter *filter4 = &afp->f_filter;
1337 
1338 		/*
1339 		 * if address is 0, it's a wildcard.  if address mask
1340 		 * isn't set, use full mask.
1341 		 */
1342 		if (filter4->ff_flow.fi_dst.s_addr == 0)
1343 			filter4->ff_mask.mask_dst.s_addr = 0;
1344 		else if (filter4->ff_mask.mask_dst.s_addr == 0)
1345 			filter4->ff_mask.mask_dst.s_addr = 0xffffffff;
1346 		if (filter4->ff_flow.fi_src.s_addr == 0)
1347 			filter4->ff_mask.mask_src.s_addr = 0;
1348 		else if (filter4->ff_mask.mask_src.s_addr == 0)
1349 			filter4->ff_mask.mask_src.s_addr = 0xffffffff;
1350 
1351 		/* clear extra bits in addresses  */
1352 		   filter4->ff_flow.fi_dst.s_addr &=
1353 		       filter4->ff_mask.mask_dst.s_addr;
1354 		   filter4->ff_flow.fi_src.s_addr &=
1355 		       filter4->ff_mask.mask_src.s_addr;
1356 
1357 		/*
1358 		 * if dst address is a wildcard, use hash-entry
1359 		 * ACC_WILDCARD_INDEX.
1360 		 */
1361 		if (filter4->ff_mask.mask_dst.s_addr != 0xffffffff)
1362 			i = ACC_WILDCARD_INDEX;
1363 		else
1364 			i = ACC_GET_HASH_INDEX(filter4->ff_flow.fi_dst.s_addr);
1365 	}
1366 #ifdef INET6
1367 	else if (filter->ff_flow.fi_family == AF_INET6) {
1368 		struct flow_filter6 *filter6 =
1369 			(struct flow_filter6 *)&afp->f_filter;
1370 #ifndef IN6MASK0 /* taken from kame ipv6 */
1371 #define	IN6MASK0	{{{ 0, 0, 0, 0 }}}
1372 #define	IN6MASK128	{{{ 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff }}}
1373 		const struct in6_addr in6mask0 = IN6MASK0;
1374 		const struct in6_addr in6mask128 = IN6MASK128;
1375 #endif
1376 
1377 		if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_dst))
1378 			filter6->ff_mask6.mask6_dst = in6mask0;
1379 		else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_dst))
1380 			filter6->ff_mask6.mask6_dst = in6mask128;
1381 		if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_src))
1382 			filter6->ff_mask6.mask6_src = in6mask0;
1383 		else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_src))
1384 			filter6->ff_mask6.mask6_src = in6mask128;
1385 
1386 		/* clear extra bits in addresses  */
1387 		for (i = 0; i < 16; i++)
1388 			filter6->ff_flow6.fi6_dst.s6_addr[i] &=
1389 			    filter6->ff_mask6.mask6_dst.s6_addr[i];
1390 		for (i = 0; i < 16; i++)
1391 			filter6->ff_flow6.fi6_src.s6_addr[i] &=
1392 			    filter6->ff_mask6.mask6_src.s6_addr[i];
1393 
1394 		if (filter6->ff_flow6.fi6_flowlabel == 0)
1395 			i = ACC_WILDCARD_INDEX;
1396 		else
1397 			i = ACC_GET_HASH_INDEX(filter6->ff_flow6.fi6_flowlabel);
1398 	}
1399 #endif /* INET6 */
1400 
1401 	afp->f_handle = get_filt_handle(classifier, i);
1402 
1403 	/* update filter bitmask */
1404 	afp->f_fbmask = filt2fibmask(filter);
1405 	classifier->acc_fbmask |= afp->f_fbmask;
1406 
1407 	/*
1408 	 * add this filter to the filter list.
1409 	 * filters are ordered from the highest rule number.
1410 	 */
1411 	s = splnet();
1412 	prev = NULL;
1413 	LIST_FOREACH(tmp, &classifier->acc_filters[i], f_chain) {
1414 		if (tmp->f_filter.ff_ruleno > afp->f_filter.ff_ruleno)
1415 			prev = tmp;
1416 		else
1417 			break;
1418 	}
1419 	if (prev == NULL)
1420 		LIST_INSERT_HEAD(&classifier->acc_filters[i], afp, f_chain);
1421 	else
1422 		LIST_INSERT_AFTER(prev, afp, f_chain);
1423 	splx(s);
1424 
1425 	*phandle = afp->f_handle;
1426 	return (0);
1427 }
1428 
1429 int
1430 acc_delete_filter(classifier, handle)
1431 	struct acc_classifier *classifier;
1432 	u_long handle;
1433 {
1434 	struct acc_filter *afp;
1435 	int	s;
1436 
1437 	if ((afp = filth_to_filtp(classifier, handle)) == NULL)
1438 		return (EINVAL);
1439 
1440 	s = splnet();
1441 	LIST_REMOVE(afp, f_chain);
1442 	splx(s);
1443 
1444 	free(afp, M_DEVBUF);
1445 
1446 	/* todo: update filt_bmask */
1447 
1448 	return (0);
1449 }
1450 
1451 /*
1452  * delete filters referencing to the specified class.
1453  * if the all flag is not 0, delete all the filters.
1454  */
1455 int
1456 acc_discard_filters(classifier, class, all)
1457 	struct acc_classifier *classifier;
1458 	void	*class;
1459 	int	all;
1460 {
1461 	struct acc_filter *afp;
1462 	int	i, s;
1463 
1464 	s = splnet();
1465 	for (i = 0; i < ACC_FILTER_TABLESIZE; i++) {
1466 		do {
1467 			LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1468 				if (all || afp->f_class == class) {
1469 					LIST_REMOVE(afp, f_chain);
1470 					free(afp, M_DEVBUF);
1471 					/* start again from the head */
1472 					break;
1473 				}
1474 		} while (afp != NULL);
1475 	}
1476 	splx(s);
1477 
1478 	if (all)
1479 		classifier->acc_fbmask = 0;
1480 
1481 	return (0);
1482 }
1483 
1484 void *
1485 acc_classify(clfier, m, af)
1486 	void *clfier;
1487 	struct mbuf *m;
1488 	int af;
1489 {
1490 	struct acc_classifier *classifier;
1491 	struct flowinfo flow;
1492 	struct acc_filter *afp;
1493 	int	i;
1494 
1495 	classifier = (struct acc_classifier *)clfier;
1496 	altq_extractflow(m, af, &flow, classifier->acc_fbmask);
1497 
1498 	if (flow.fi_family == AF_INET) {
1499 		struct flowinfo_in *fp = (struct flowinfo_in *)&flow;
1500 
1501 		if ((classifier->acc_fbmask & FIMB4_ALL) == FIMB4_TOS) {
1502 			/* only tos is used */
1503 			LIST_FOREACH(afp,
1504 				 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1505 				 f_chain)
1506 				if (apply_tosfilter4(afp->f_fbmask,
1507 						     &afp->f_filter, fp))
1508 					/* filter matched */
1509 					return (afp->f_class);
1510 		} else if ((classifier->acc_fbmask &
1511 			(~(FIMB4_PROTO|FIMB4_SPORT|FIMB4_DPORT) & FIMB4_ALL))
1512 		    == 0) {
1513 			/* only proto and ports are used */
1514 			LIST_FOREACH(afp,
1515 				 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1516 				 f_chain)
1517 				if (apply_ppfilter4(afp->f_fbmask,
1518 						    &afp->f_filter, fp))
1519 					/* filter matched */
1520 					return (afp->f_class);
1521 		} else {
1522 			/* get the filter hash entry from its dest address */
1523 			i = ACC_GET_HASH_INDEX(fp->fi_dst.s_addr);
1524 			do {
1525 				/*
1526 				 * go through this loop twice.  first for dst
1527 				 * hash, second for wildcards.
1528 				 */
1529 				LIST_FOREACH(afp, &classifier->acc_filters[i],
1530 					     f_chain)
1531 					if (apply_filter4(afp->f_fbmask,
1532 							  &afp->f_filter, fp))
1533 						/* filter matched */
1534 						return (afp->f_class);
1535 
1536 				/*
1537 				 * check again for filters with a dst addr
1538 				 * wildcard.
1539 				 * (daddr == 0 || dmask != 0xffffffff).
1540 				 */
1541 				if (i != ACC_WILDCARD_INDEX)
1542 					i = ACC_WILDCARD_INDEX;
1543 				else
1544 					break;
1545 			} while (1);
1546 		}
1547 	}
1548 #ifdef INET6
1549 	else if (flow.fi_family == AF_INET6) {
1550 		struct flowinfo_in6 *fp6 = (struct flowinfo_in6 *)&flow;
1551 
1552 		/* get the filter hash entry from its flow ID */
1553 		if (fp6->fi6_flowlabel != 0)
1554 			i = ACC_GET_HASH_INDEX(fp6->fi6_flowlabel);
1555 		else
1556 			/* flowlable can be zero */
1557 			i = ACC_WILDCARD_INDEX;
1558 
1559 		/* go through this loop twice.  first for flow hash, second
1560 		   for wildcards. */
1561 		do {
1562 			LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1563 				if (apply_filter6(afp->f_fbmask,
1564 					(struct flow_filter6 *)&afp->f_filter,
1565 					fp6))
1566 					/* filter matched */
1567 					return (afp->f_class);
1568 
1569 			/*
1570 			 * check again for filters with a wildcard.
1571 			 */
1572 			if (i != ACC_WILDCARD_INDEX)
1573 				i = ACC_WILDCARD_INDEX;
1574 			else
1575 				break;
1576 		} while (1);
1577 	}
1578 #endif /* INET6 */
1579 
1580 	/* no filter matched */
1581 	return (NULL);
1582 }
1583 
1584 static int
1585 apply_filter4(fbmask, filt, pkt)
1586 	u_int32_t	fbmask;
1587 	struct flow_filter *filt;
1588 	struct flowinfo_in *pkt;
1589 {
1590 	if (filt->ff_flow.fi_family != AF_INET)
1591 		return (0);
1592 	if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1593 		return (0);
1594 	if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1595 		return (0);
1596 	if ((fbmask & FIMB4_DADDR) &&
1597 	    filt->ff_flow.fi_dst.s_addr !=
1598 	    (pkt->fi_dst.s_addr & filt->ff_mask.mask_dst.s_addr))
1599 		return (0);
1600 	if ((fbmask & FIMB4_SADDR) &&
1601 	    filt->ff_flow.fi_src.s_addr !=
1602 	    (pkt->fi_src.s_addr & filt->ff_mask.mask_src.s_addr))
1603 		return (0);
1604 	if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1605 		return (0);
1606 	if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1607 	    (pkt->fi_tos & filt->ff_mask.mask_tos))
1608 		return (0);
1609 	if ((fbmask & FIMB4_GPI) && filt->ff_flow.fi_gpi != (pkt->fi_gpi))
1610 		return (0);
1611 	/* match */
1612 	return (1);
1613 }
1614 
1615 /*
1616  * filter matching function optimized for a common case that checks
1617  * only protocol and port numbers
1618  */
1619 static int
1620 apply_ppfilter4(fbmask, filt, pkt)
1621 	u_int32_t	fbmask;
1622 	struct flow_filter *filt;
1623 	struct flowinfo_in *pkt;
1624 {
1625 	if (filt->ff_flow.fi_family != AF_INET)
1626 		return (0);
1627 	if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1628 		return (0);
1629 	if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1630 		return (0);
1631 	if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1632 		return (0);
1633 	/* match */
1634 	return (1);
1635 }
1636 
1637 /*
1638  * filter matching function only for tos field.
1639  */
1640 static int
1641 apply_tosfilter4(fbmask, filt, pkt)
1642 	u_int32_t	fbmask;
1643 	struct flow_filter *filt;
1644 	struct flowinfo_in *pkt;
1645 {
1646 	if (filt->ff_flow.fi_family != AF_INET)
1647 		return (0);
1648 	if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1649 	    (pkt->fi_tos & filt->ff_mask.mask_tos))
1650 		return (0);
1651 	/* match */
1652 	return (1);
1653 }
1654 
1655 #ifdef INET6
1656 static int
1657 apply_filter6(fbmask, filt, pkt)
1658 	u_int32_t	fbmask;
1659 	struct flow_filter6 *filt;
1660 	struct flowinfo_in6 *pkt;
1661 {
1662 	int i;
1663 
1664 	if (filt->ff_flow6.fi6_family != AF_INET6)
1665 		return (0);
1666 	if ((fbmask & FIMB6_FLABEL) &&
1667 	    filt->ff_flow6.fi6_flowlabel != pkt->fi6_flowlabel)
1668 		return (0);
1669 	if ((fbmask & FIMB6_PROTO) &&
1670 	    filt->ff_flow6.fi6_proto != pkt->fi6_proto)
1671 		return (0);
1672 	if ((fbmask & FIMB6_SPORT) &&
1673 	    filt->ff_flow6.fi6_sport != pkt->fi6_sport)
1674 		return (0);
1675 	if ((fbmask & FIMB6_DPORT) &&
1676 	    filt->ff_flow6.fi6_dport != pkt->fi6_dport)
1677 		return (0);
1678 	if (fbmask & FIMB6_SADDR) {
1679 		for (i = 0; i < 4; i++)
1680 			if (filt->ff_flow6.fi6_src.s6_addr32[i] !=
1681 			    (pkt->fi6_src.s6_addr32[i] &
1682 			     filt->ff_mask6.mask6_src.s6_addr32[i]))
1683 				return (0);
1684 	}
1685 	if (fbmask & FIMB6_DADDR) {
1686 		for (i = 0; i < 4; i++)
1687 			if (filt->ff_flow6.fi6_dst.s6_addr32[i] !=
1688 			    (pkt->fi6_dst.s6_addr32[i] &
1689 			     filt->ff_mask6.mask6_dst.s6_addr32[i]))
1690 				return (0);
1691 	}
1692 	if ((fbmask & FIMB6_TCLASS) &&
1693 	    filt->ff_flow6.fi6_tclass !=
1694 	    (pkt->fi6_tclass & filt->ff_mask6.mask6_tclass))
1695 		return (0);
1696 	if ((fbmask & FIMB6_GPI) &&
1697 	    filt->ff_flow6.fi6_gpi != pkt->fi6_gpi)
1698 		return (0);
1699 	/* match */
1700 	return (1);
1701 }
1702 #endif /* INET6 */
1703 
1704 /*
1705  *  filter handle:
1706  *	bit 20-28: index to the filter hash table
1707  *	bit  0-19: unique id in the hash bucket.
1708  */
1709 static u_long
1710 get_filt_handle(classifier, i)
1711 	struct acc_classifier *classifier;
1712 	int	i;
1713 {
1714 	static u_long handle_number = 1;
1715 	u_long 	handle;
1716 	struct acc_filter *afp;
1717 
1718 	while (1) {
1719 		handle = handle_number++ & 0x000fffff;
1720 
1721 		if (LIST_EMPTY(&classifier->acc_filters[i]))
1722 			break;
1723 
1724 		LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1725 			if ((afp->f_handle & 0x000fffff) == handle)
1726 				break;
1727 		if (afp == NULL)
1728 			break;
1729 		/* this handle is already used, try again */
1730 	}
1731 
1732 	return ((i << 20) | handle);
1733 }
1734 
1735 /* convert filter handle to filter pointer */
1736 static struct acc_filter *
1737 filth_to_filtp(classifier, handle)
1738 	struct acc_classifier *classifier;
1739 	u_long handle;
1740 {
1741 	struct acc_filter *afp;
1742 	int	i;
1743 
1744 	i = ACC_GET_HINDEX(handle);
1745 
1746 	LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1747 		if (afp->f_handle == handle)
1748 			return (afp);
1749 
1750 	return (NULL);
1751 }
1752 
1753 /* create flowinfo bitmask */
1754 static u_int32_t
1755 filt2fibmask(filt)
1756 	struct flow_filter *filt;
1757 {
1758 	u_int32_t mask = 0;
1759 #ifdef INET6
1760 	struct flow_filter6 *filt6;
1761 #endif
1762 
1763 	switch (filt->ff_flow.fi_family) {
1764 	case AF_INET:
1765 		if (filt->ff_flow.fi_proto != 0)
1766 			mask |= FIMB4_PROTO;
1767 		if (filt->ff_flow.fi_tos != 0)
1768 			mask |= FIMB4_TOS;
1769 		if (filt->ff_flow.fi_dst.s_addr != 0)
1770 			mask |= FIMB4_DADDR;
1771 		if (filt->ff_flow.fi_src.s_addr != 0)
1772 			mask |= FIMB4_SADDR;
1773 		if (filt->ff_flow.fi_sport != 0)
1774 			mask |= FIMB4_SPORT;
1775 		if (filt->ff_flow.fi_dport != 0)
1776 			mask |= FIMB4_DPORT;
1777 		if (filt->ff_flow.fi_gpi != 0)
1778 			mask |= FIMB4_GPI;
1779 		break;
1780 #ifdef INET6
1781 	case AF_INET6:
1782 		filt6 = (struct flow_filter6 *)filt;
1783 
1784 		if (filt6->ff_flow6.fi6_proto != 0)
1785 			mask |= FIMB6_PROTO;
1786 		if (filt6->ff_flow6.fi6_tclass != 0)
1787 			mask |= FIMB6_TCLASS;
1788 		if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_dst))
1789 			mask |= FIMB6_DADDR;
1790 		if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_src))
1791 			mask |= FIMB6_SADDR;
1792 		if (filt6->ff_flow6.fi6_sport != 0)
1793 			mask |= FIMB6_SPORT;
1794 		if (filt6->ff_flow6.fi6_dport != 0)
1795 			mask |= FIMB6_DPORT;
1796 		if (filt6->ff_flow6.fi6_gpi != 0)
1797 			mask |= FIMB6_GPI;
1798 		if (filt6->ff_flow6.fi6_flowlabel != 0)
1799 			mask |= FIMB6_FLABEL;
1800 		break;
1801 #endif /* INET6 */
1802 	}
1803 	return (mask);
1804 }
1805 
1806 /*
1807  * helper functions to handle IPv4 fragments.
1808  * currently only in-sequence fragments are handled.
1809  *	- fragment info is cached in a LRU list.
1810  *	- when a first fragment is found, cache its flow info.
1811  *	- when a non-first fragment is found, lookup the cache.
1812  */
1813 
1814 struct ip4_frag {
1815     TAILQ_ENTRY(ip4_frag) ip4f_chain;
1816     char    ip4f_valid;
1817     u_short ip4f_id;
1818     struct flowinfo_in ip4f_info;
1819 };
1820 
1821 static TAILQ_HEAD(ip4f_list, ip4_frag) ip4f_list; /* IPv4 fragment cache */
1822 
1823 #define	IP4F_TABSIZE		16	/* IPv4 fragment cache size */
1824 
1825 static void
1826 ip4f_cache(ip, fin)
1827 	struct ip *ip;
1828 	struct flowinfo_in *fin;
1829 {
1830 	struct ip4_frag *fp;
1831 
1832 	if (TAILQ_EMPTY(&ip4f_list)) {
1833 		/* first time call, allocate fragment cache entries. */
1834 		if (ip4f_init() < 0)
1835 			/* allocation failed! */
1836 			return;
1837 	}
1838 
1839 	fp = ip4f_alloc();
1840 	fp->ip4f_id = ip->ip_id;
1841 	fp->ip4f_info.fi_proto = ip->ip_p;
1842 	fp->ip4f_info.fi_src.s_addr = ip->ip_src.s_addr;
1843 	fp->ip4f_info.fi_dst.s_addr = ip->ip_dst.s_addr;
1844 
1845 	/* save port numbers */
1846 	fp->ip4f_info.fi_sport = fin->fi_sport;
1847 	fp->ip4f_info.fi_dport = fin->fi_dport;
1848 	fp->ip4f_info.fi_gpi   = fin->fi_gpi;
1849 }
1850 
1851 static int
1852 ip4f_lookup(ip, fin)
1853 	struct ip *ip;
1854 	struct flowinfo_in *fin;
1855 {
1856 	struct ip4_frag *fp;
1857 
1858 	for (fp = TAILQ_FIRST(&ip4f_list); fp != NULL && fp->ip4f_valid;
1859 	     fp = TAILQ_NEXT(fp, ip4f_chain))
1860 		if (ip->ip_id == fp->ip4f_id &&
1861 		    ip->ip_src.s_addr == fp->ip4f_info.fi_src.s_addr &&
1862 		    ip->ip_dst.s_addr == fp->ip4f_info.fi_dst.s_addr &&
1863 		    ip->ip_p == fp->ip4f_info.fi_proto) {
1864 			/* found the matching entry */
1865 			fin->fi_sport = fp->ip4f_info.fi_sport;
1866 			fin->fi_dport = fp->ip4f_info.fi_dport;
1867 			fin->fi_gpi   = fp->ip4f_info.fi_gpi;
1868 
1869 			if ((ntohs(ip->ip_off) & IP_MF) == 0)
1870 				/* this is the last fragment,
1871 				   release the entry. */
1872 				ip4f_free(fp);
1873 
1874 			return (1);
1875 		}
1876 
1877 	/* no matching entry found */
1878 	return (0);
1879 }
1880 
1881 static int
1882 ip4f_init(void)
1883 {
1884 	struct ip4_frag *fp;
1885 	int i;
1886 
1887 	TAILQ_INIT(&ip4f_list);
1888 	for (i=0; i<IP4F_TABSIZE; i++) {
1889 		fp = malloc(sizeof(struct ip4_frag),
1890 		       M_DEVBUF, M_NOWAIT);
1891 		if (fp == NULL) {
1892 			printf("ip4f_init: can't alloc %dth entry!\n", i);
1893 			if (i == 0)
1894 				return (-1);
1895 			return (0);
1896 		}
1897 		fp->ip4f_valid = 0;
1898 		TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1899 	}
1900 	return (0);
1901 }
1902 
1903 static struct ip4_frag *
1904 ip4f_alloc(void)
1905 {
1906 	struct ip4_frag *fp;
1907 
1908 	/* reclaim an entry at the tail, put it at the head */
1909 	fp = TAILQ_LAST(&ip4f_list, ip4f_list);
1910 	TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1911 	fp->ip4f_valid = 1;
1912 	TAILQ_INSERT_HEAD(&ip4f_list, fp, ip4f_chain);
1913 	return (fp);
1914 }
1915 
1916 static void
1917 ip4f_free(fp)
1918 	struct ip4_frag *fp;
1919 {
1920 	TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1921 	fp->ip4f_valid = 0;
1922 	TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1923 }
1924 
1925 #endif /* ALTQ3_CLFIER_COMPAT */
1926