xref: /freebsd/sys/net/altq/altq_subr.c (revision cc68614da8232d8baaca0ae0d0dd8f890f06623e)
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  * $FreeBSD$
28  */
29 
30 #include "opt_altq.h"
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 
34 #include <sys/param.h>
35 #include <sys/malloc.h>
36 #include <sys/mbuf.h>
37 #include <sys/systm.h>
38 #include <sys/proc.h>
39 #include <sys/socket.h>
40 #include <sys/socketvar.h>
41 #include <sys/kernel.h>
42 #include <sys/errno.h>
43 #include <sys/syslog.h>
44 #include <sys/sysctl.h>
45 #include <sys/queue.h>
46 
47 #include <net/if.h>
48 #include <net/if_var.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 int (*altq_input)(struct mbuf *, int) = NULL;
82 static struct mbuf *tbr_dequeue(struct ifaltq *, int);
83 static int tbr_timer = 0;	/* token bucket regulator timer */
84 #if !defined(__FreeBSD__) || (__FreeBSD_version < 600000)
85 static struct callout tbr_callout = CALLOUT_INITIALIZER;
86 #else
87 static struct callout tbr_callout;
88 #endif
89 
90 #ifdef ALTQ3_CLFIER_COMPAT
91 static int 	extract_ports4(struct mbuf *, struct ip *, struct flowinfo_in *);
92 #ifdef INET6
93 static int 	extract_ports6(struct mbuf *, struct ip6_hdr *,
94 			       struct flowinfo_in6 *);
95 #endif
96 static int	apply_filter4(u_int32_t, struct flow_filter *,
97 			      struct flowinfo_in *);
98 static int	apply_ppfilter4(u_int32_t, struct flow_filter *,
99 				struct flowinfo_in *);
100 #ifdef INET6
101 static int	apply_filter6(u_int32_t, struct flow_filter6 *,
102 			      struct flowinfo_in6 *);
103 #endif
104 static int	apply_tosfilter4(u_int32_t, struct flow_filter *,
105 				 struct flowinfo_in *);
106 static u_long	get_filt_handle(struct acc_classifier *, int);
107 static struct acc_filter *filth_to_filtp(struct acc_classifier *, u_long);
108 static u_int32_t filt2fibmask(struct flow_filter *);
109 
110 static void 	ip4f_cache(struct ip *, struct flowinfo_in *);
111 static int 	ip4f_lookup(struct ip *, struct flowinfo_in *);
112 static int 	ip4f_init(void);
113 static struct ip4_frag	*ip4f_alloc(void);
114 static void 	ip4f_free(struct ip4_frag *);
115 #endif /* ALTQ3_CLFIER_COMPAT */
116 
117 #ifdef ALTQ
118 SYSCTL_NODE(_kern_features, OID_AUTO, altq, CTLFLAG_RD | CTLFLAG_CAPRD, 0,
119     "ALTQ packet queuing");
120 
121 #define	ALTQ_FEATURE(name, desc)					\
122 	SYSCTL_INT_WITH_LABEL(_kern_features_altq, OID_AUTO, name,	\
123 	    CTLFLAG_RD | CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, 1,		\
124 	    desc, "feature")
125 
126 #ifdef ALTQ_CBQ
127 ALTQ_FEATURE(cbq, "ALTQ Class Based Queuing discipline");
128 #endif
129 #ifdef ALTQ_CODEL
130 ALTQ_FEATURE(codel, "ALTQ Controlled Delay discipline");
131 #endif
132 #ifdef ALTQ_RED
133 ALTQ_FEATURE(red, "ALTQ Random Early Detection discipline");
134 #endif
135 #ifdef ALTQ_RIO
136 ALTQ_FEATURE(rio, "ALTQ Random Early Drop discipline");
137 #endif
138 #ifdef ALTQ_HFSC
139 ALTQ_FEATURE(hfsc, "ALTQ Hierarchical Packet Scheduler discipline");
140 #endif
141 #ifdef ALTQ_PRIQ
142 ALTQ_FEATURE(priq, "ATLQ Priority Queuing discipline");
143 #endif
144 #ifdef ALTQ_FAIRQ
145 ALTQ_FEATURE(fairq, "ALTQ Fair Queuing discipline");
146 #endif
147 #endif
148 
149 /*
150  * alternate queueing support routines
151  */
152 
153 /* look up the queue state by the interface name and the queueing type. */
154 void *
155 altq_lookup(name, type)
156 	char *name;
157 	int type;
158 {
159 	struct ifnet *ifp;
160 
161 	if ((ifp = ifunit(name)) != NULL) {
162 		/* read if_snd unlocked */
163 		if (type != ALTQT_NONE && ifp->if_snd.altq_type == type)
164 			return (ifp->if_snd.altq_disc);
165 	}
166 
167 	return NULL;
168 }
169 
170 int
171 altq_attach(ifq, type, discipline, enqueue, dequeue, request)
172 	struct ifaltq *ifq;
173 	int type;
174 	void *discipline;
175 	int (*enqueue)(struct ifaltq *, struct mbuf *, struct altq_pktattr *);
176 	struct mbuf *(*dequeue)(struct ifaltq *, int);
177 	int (*request)(struct ifaltq *, int, void *);
178 {
179 	IFQ_LOCK(ifq);
180 	if (!ALTQ_IS_READY(ifq)) {
181 		IFQ_UNLOCK(ifq);
182 		return ENXIO;
183 	}
184 
185 	ifq->altq_type     = type;
186 	ifq->altq_disc     = discipline;
187 	ifq->altq_enqueue  = enqueue;
188 	ifq->altq_dequeue  = dequeue;
189 	ifq->altq_request  = request;
190 	ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED);
191 	IFQ_UNLOCK(ifq);
192 	return 0;
193 }
194 
195 int
196 altq_detach(ifq)
197 	struct ifaltq *ifq;
198 {
199 	IFQ_LOCK(ifq);
200 
201 	if (!ALTQ_IS_READY(ifq)) {
202 		IFQ_UNLOCK(ifq);
203 		return ENXIO;
204 	}
205 	if (ALTQ_IS_ENABLED(ifq)) {
206 		IFQ_UNLOCK(ifq);
207 		return EBUSY;
208 	}
209 	if (!ALTQ_IS_ATTACHED(ifq)) {
210 		IFQ_UNLOCK(ifq);
211 		return (0);
212 	}
213 
214 	ifq->altq_type     = ALTQT_NONE;
215 	ifq->altq_disc     = NULL;
216 	ifq->altq_enqueue  = NULL;
217 	ifq->altq_dequeue  = NULL;
218 	ifq->altq_request  = NULL;
219 	ifq->altq_flags &= ALTQF_CANTCHANGE;
220 
221 	IFQ_UNLOCK(ifq);
222 	return 0;
223 }
224 
225 int
226 altq_enable(ifq)
227 	struct ifaltq *ifq;
228 {
229 	int s;
230 
231 	IFQ_LOCK(ifq);
232 
233 	if (!ALTQ_IS_READY(ifq)) {
234 		IFQ_UNLOCK(ifq);
235 		return ENXIO;
236 	}
237 	if (ALTQ_IS_ENABLED(ifq)) {
238 		IFQ_UNLOCK(ifq);
239 		return 0;
240 	}
241 
242 	s = splnet();
243 	IFQ_PURGE_NOLOCK(ifq);
244 	ASSERT(ifq->ifq_len == 0);
245 	ifq->ifq_drv_maxlen = 0;		/* disable bulk dequeue */
246 	ifq->altq_flags |= ALTQF_ENABLED;
247 	splx(s);
248 
249 	IFQ_UNLOCK(ifq);
250 	return 0;
251 }
252 
253 int
254 altq_disable(ifq)
255 	struct ifaltq *ifq;
256 {
257 	int s;
258 
259 	IFQ_LOCK(ifq);
260 	if (!ALTQ_IS_ENABLED(ifq)) {
261 		IFQ_UNLOCK(ifq);
262 		return 0;
263 	}
264 
265 	s = splnet();
266 	IFQ_PURGE_NOLOCK(ifq);
267 	ASSERT(ifq->ifq_len == 0);
268 	ifq->altq_flags &= ~(ALTQF_ENABLED);
269 	splx(s);
270 
271 	IFQ_UNLOCK(ifq);
272 	return 0;
273 }
274 
275 #ifdef ALTQ_DEBUG
276 void
277 altq_assert(file, line, failedexpr)
278 	const char *file, *failedexpr;
279 	int line;
280 {
281 	(void)printf("altq assertion \"%s\" failed: file \"%s\", line %d\n",
282 		     failedexpr, file, line);
283 	panic("altq assertion");
284 	/* NOTREACHED */
285 }
286 #endif
287 
288 /*
289  * internal representation of token bucket parameters
290  *	rate:	(byte_per_unittime << TBR_SHIFT)  / machclk_freq
291  *		(((bits_per_sec) / 8) << TBR_SHIFT) / machclk_freq
292  *	depth:	byte << TBR_SHIFT
293  *
294  */
295 #define	TBR_SHIFT	29
296 #define	TBR_SCALE(x)	((int64_t)(x) << TBR_SHIFT)
297 #define	TBR_UNSCALE(x)	((x) >> TBR_SHIFT)
298 
299 static struct mbuf *
300 tbr_dequeue(ifq, op)
301 	struct ifaltq *ifq;
302 	int op;
303 {
304 	struct tb_regulator *tbr;
305 	struct mbuf *m;
306 	int64_t interval;
307 	u_int64_t now;
308 
309 	IFQ_LOCK_ASSERT(ifq);
310 	tbr = ifq->altq_tbr;
311 	if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) {
312 		/* if this is a remove after poll, bypass tbr check */
313 	} else {
314 		/* update token only when it is negative */
315 		if (tbr->tbr_token <= 0) {
316 			now = read_machclk();
317 			interval = now - tbr->tbr_last;
318 			if (interval >= tbr->tbr_filluptime)
319 				tbr->tbr_token = tbr->tbr_depth;
320 			else {
321 				tbr->tbr_token += interval * tbr->tbr_rate;
322 				if (tbr->tbr_token > tbr->tbr_depth)
323 					tbr->tbr_token = tbr->tbr_depth;
324 			}
325 			tbr->tbr_last = now;
326 		}
327 		/* if token is still negative, don't allow dequeue */
328 		if (tbr->tbr_token <= 0)
329 			return (NULL);
330 	}
331 
332 	if (ALTQ_IS_ENABLED(ifq))
333 		m = (*ifq->altq_dequeue)(ifq, op);
334 	else {
335 		if (op == ALTDQ_POLL)
336 			_IF_POLL(ifq, m);
337 		else
338 			_IF_DEQUEUE(ifq, m);
339 	}
340 
341 	if (m != NULL && op == ALTDQ_REMOVE)
342 		tbr->tbr_token -= TBR_SCALE(m_pktlen(m));
343 	tbr->tbr_lastop = op;
344 	return (m);
345 }
346 
347 /*
348  * set a token bucket regulator.
349  * if the specified rate is zero, the token bucket regulator is deleted.
350  */
351 int
352 tbr_set(ifq, profile)
353 	struct ifaltq *ifq;
354 	struct tb_profile *profile;
355 {
356 	struct tb_regulator *tbr, *otbr;
357 
358 	if (tbr_dequeue_ptr == NULL)
359 		tbr_dequeue_ptr = tbr_dequeue;
360 
361 	if (machclk_freq == 0)
362 		init_machclk();
363 	if (machclk_freq == 0) {
364 		printf("tbr_set: no cpu clock available!\n");
365 		return (ENXIO);
366 	}
367 
368 	IFQ_LOCK(ifq);
369 	if (profile->rate == 0) {
370 		/* delete this tbr */
371 		if ((tbr = ifq->altq_tbr) == NULL) {
372 			IFQ_UNLOCK(ifq);
373 			return (ENOENT);
374 		}
375 		ifq->altq_tbr = NULL;
376 		free(tbr, M_DEVBUF);
377 		IFQ_UNLOCK(ifq);
378 		return (0);
379 	}
380 
381 	tbr = malloc(sizeof(struct tb_regulator), M_DEVBUF, M_NOWAIT | M_ZERO);
382 	if (tbr == NULL) {
383 		IFQ_UNLOCK(ifq);
384 		return (ENOMEM);
385 	}
386 
387 	tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq;
388 	tbr->tbr_depth = TBR_SCALE(profile->depth);
389 	if (tbr->tbr_rate > 0)
390 		tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate;
391 	else
392 		tbr->tbr_filluptime = LLONG_MAX;
393 	/*
394 	 *  The longest time between tbr_dequeue() calls will be about 1
395 	 *  system tick, as the callout that drives it is scheduled once per
396 	 *  tick.  The refill-time detection logic in tbr_dequeue() can only
397 	 *  properly detect the passage of up to LLONG_MAX machclk ticks.
398 	 *  Therefore, in order for this logic to function properly in the
399 	 *  extreme case, the maximum value of tbr_filluptime should be
400 	 *  LLONG_MAX less one system tick's worth of machclk ticks less
401 	 *  some additional slop factor (here one more system tick's worth
402 	 *  of machclk ticks).
403 	 */
404 	if (tbr->tbr_filluptime > (LLONG_MAX - 2 * machclk_per_tick))
405 		tbr->tbr_filluptime = LLONG_MAX - 2 * machclk_per_tick;
406 	tbr->tbr_token = tbr->tbr_depth;
407 	tbr->tbr_last = read_machclk();
408 	tbr->tbr_lastop = ALTDQ_REMOVE;
409 
410 	otbr = ifq->altq_tbr;
411 	ifq->altq_tbr = tbr;	/* set the new tbr */
412 
413 	if (otbr != NULL)
414 		free(otbr, M_DEVBUF);
415 	else {
416 		if (tbr_timer == 0) {
417 			CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
418 			tbr_timer = 1;
419 		}
420 	}
421 	IFQ_UNLOCK(ifq);
422 	return (0);
423 }
424 
425 /*
426  * tbr_timeout goes through the interface list, and kicks the drivers
427  * if necessary.
428  *
429  * MPSAFE
430  */
431 static void
432 tbr_timeout(arg)
433 	void *arg;
434 {
435 	VNET_ITERATOR_DECL(vnet_iter);
436 	struct ifnet *ifp;
437 	struct epoch_tracker et;
438 	int active;
439 
440 	active = 0;
441 	NET_EPOCH_ENTER(et);
442 	VNET_LIST_RLOCK_NOSLEEP();
443 	VNET_FOREACH(vnet_iter) {
444 		CURVNET_SET(vnet_iter);
445 		for (ifp = CK_STAILQ_FIRST(&V_ifnet); ifp;
446 		    ifp = CK_STAILQ_NEXT(ifp, if_link)) {
447 			/* read from if_snd unlocked */
448 			if (!TBR_IS_ENABLED(&ifp->if_snd))
449 				continue;
450 			active++;
451 			if (!IFQ_IS_EMPTY(&ifp->if_snd) &&
452 			    ifp->if_start != NULL)
453 				(*ifp->if_start)(ifp);
454 		}
455 		CURVNET_RESTORE();
456 	}
457 	VNET_LIST_RUNLOCK_NOSLEEP();
458 	NET_EPOCH_EXIT(et);
459 	if (active > 0)
460 		CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
461 	else
462 		tbr_timer = 0;	/* don't need tbr_timer anymore */
463 }
464 
465 /*
466  * attach a discipline to the interface.  if one already exists, it is
467  * overridden.
468  * Locking is done in the discipline specific attach functions. Basically
469  * they call back to altq_attach which takes care of the attach and locking.
470  */
471 int
472 altq_pfattach(struct pf_altq *a)
473 {
474 	int error = 0;
475 
476 	switch (a->scheduler) {
477 	case ALTQT_NONE:
478 		break;
479 #ifdef ALTQ_CBQ
480 	case ALTQT_CBQ:
481 		error = cbq_pfattach(a);
482 		break;
483 #endif
484 #ifdef ALTQ_PRIQ
485 	case ALTQT_PRIQ:
486 		error = priq_pfattach(a);
487 		break;
488 #endif
489 #ifdef ALTQ_HFSC
490 	case ALTQT_HFSC:
491 		error = hfsc_pfattach(a);
492 		break;
493 #endif
494 #ifdef ALTQ_FAIRQ
495 	case ALTQT_FAIRQ:
496 		error = fairq_pfattach(a);
497 		break;
498 #endif
499 #ifdef ALTQ_CODEL
500 	case ALTQT_CODEL:
501 		error = codel_pfattach(a);
502 		break;
503 #endif
504 	default:
505 		error = ENXIO;
506 	}
507 
508 	return (error);
509 }
510 
511 /*
512  * detach a discipline from the interface.
513  * it is possible that the discipline was already overridden by another
514  * discipline.
515  */
516 int
517 altq_pfdetach(struct pf_altq *a)
518 {
519 	struct ifnet *ifp;
520 	int s, error = 0;
521 
522 	if ((ifp = ifunit(a->ifname)) == NULL)
523 		return (EINVAL);
524 
525 	/* if this discipline is no longer referenced, just return */
526 	/* read unlocked from if_snd */
527 	if (a->altq_disc == NULL || a->altq_disc != ifp->if_snd.altq_disc)
528 		return (0);
529 
530 	s = splnet();
531 	/* read unlocked from if_snd, _disable and _detach take care */
532 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
533 		error = altq_disable(&ifp->if_snd);
534 	if (error == 0)
535 		error = altq_detach(&ifp->if_snd);
536 	splx(s);
537 
538 	return (error);
539 }
540 
541 /*
542  * add a discipline or a queue
543  * Locking is done in the discipline specific functions with regards to
544  * malloc with WAITOK, also it is not yet clear which lock to use.
545  */
546 int
547 altq_add(struct ifnet *ifp, struct pf_altq *a)
548 {
549 	int error = 0;
550 
551 	if (a->qname[0] != 0)
552 		return (altq_add_queue(a));
553 
554 	if (machclk_freq == 0)
555 		init_machclk();
556 	if (machclk_freq == 0)
557 		panic("altq_add: no cpu clock");
558 
559 	switch (a->scheduler) {
560 #ifdef ALTQ_CBQ
561 	case ALTQT_CBQ:
562 		error = cbq_add_altq(ifp, a);
563 		break;
564 #endif
565 #ifdef ALTQ_PRIQ
566 	case ALTQT_PRIQ:
567 		error = priq_add_altq(ifp, a);
568 		break;
569 #endif
570 #ifdef ALTQ_HFSC
571 	case ALTQT_HFSC:
572 		error = hfsc_add_altq(ifp, a);
573 		break;
574 #endif
575 #ifdef ALTQ_FAIRQ
576         case ALTQT_FAIRQ:
577                 error = fairq_add_altq(ifp, a);
578                 break;
579 #endif
580 #ifdef ALTQ_CODEL
581 	case ALTQT_CODEL:
582 		error = codel_add_altq(ifp, a);
583 		break;
584 #endif
585 	default:
586 		error = ENXIO;
587 	}
588 
589 	return (error);
590 }
591 
592 /*
593  * remove a discipline or a queue
594  * It is yet unclear what lock to use to protect this operation, the
595  * discipline specific functions will determine and grab it
596  */
597 int
598 altq_remove(struct pf_altq *a)
599 {
600 	int error = 0;
601 
602 	if (a->qname[0] != 0)
603 		return (altq_remove_queue(a));
604 
605 	switch (a->scheduler) {
606 #ifdef ALTQ_CBQ
607 	case ALTQT_CBQ:
608 		error = cbq_remove_altq(a);
609 		break;
610 #endif
611 #ifdef ALTQ_PRIQ
612 	case ALTQT_PRIQ:
613 		error = priq_remove_altq(a);
614 		break;
615 #endif
616 #ifdef ALTQ_HFSC
617 	case ALTQT_HFSC:
618 		error = hfsc_remove_altq(a);
619 		break;
620 #endif
621 #ifdef ALTQ_FAIRQ
622         case ALTQT_FAIRQ:
623                 error = fairq_remove_altq(a);
624                 break;
625 #endif
626 #ifdef ALTQ_CODEL
627 	case ALTQT_CODEL:
628 		error = codel_remove_altq(a);
629 		break;
630 #endif
631 	default:
632 		error = ENXIO;
633 	}
634 
635 	return (error);
636 }
637 
638 /*
639  * add a queue to the discipline
640  * It is yet unclear what lock to use to protect this operation, the
641  * discipline specific functions will determine and grab it
642  */
643 int
644 altq_add_queue(struct pf_altq *a)
645 {
646 	int error = 0;
647 
648 	switch (a->scheduler) {
649 #ifdef ALTQ_CBQ
650 	case ALTQT_CBQ:
651 		error = cbq_add_queue(a);
652 		break;
653 #endif
654 #ifdef ALTQ_PRIQ
655 	case ALTQT_PRIQ:
656 		error = priq_add_queue(a);
657 		break;
658 #endif
659 #ifdef ALTQ_HFSC
660 	case ALTQT_HFSC:
661 		error = hfsc_add_queue(a);
662 		break;
663 #endif
664 #ifdef ALTQ_FAIRQ
665         case ALTQT_FAIRQ:
666                 error = fairq_add_queue(a);
667                 break;
668 #endif
669 	default:
670 		error = ENXIO;
671 	}
672 
673 	return (error);
674 }
675 
676 /*
677  * remove a queue from the discipline
678  * It is yet unclear what lock to use to protect this operation, the
679  * discipline specific functions will determine and grab it
680  */
681 int
682 altq_remove_queue(struct pf_altq *a)
683 {
684 	int error = 0;
685 
686 	switch (a->scheduler) {
687 #ifdef ALTQ_CBQ
688 	case ALTQT_CBQ:
689 		error = cbq_remove_queue(a);
690 		break;
691 #endif
692 #ifdef ALTQ_PRIQ
693 	case ALTQT_PRIQ:
694 		error = priq_remove_queue(a);
695 		break;
696 #endif
697 #ifdef ALTQ_HFSC
698 	case ALTQT_HFSC:
699 		error = hfsc_remove_queue(a);
700 		break;
701 #endif
702 #ifdef ALTQ_FAIRQ
703         case ALTQT_FAIRQ:
704                 error = fairq_remove_queue(a);
705                 break;
706 #endif
707 	default:
708 		error = ENXIO;
709 	}
710 
711 	return (error);
712 }
713 
714 /*
715  * get queue statistics
716  * Locking is done in the discipline specific functions with regards to
717  * copyout operations, also it is not yet clear which lock to use.
718  */
719 int
720 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version)
721 {
722 	int error = 0;
723 
724 	switch (a->scheduler) {
725 #ifdef ALTQ_CBQ
726 	case ALTQT_CBQ:
727 		error = cbq_getqstats(a, ubuf, nbytes, version);
728 		break;
729 #endif
730 #ifdef ALTQ_PRIQ
731 	case ALTQT_PRIQ:
732 		error = priq_getqstats(a, ubuf, nbytes, version);
733 		break;
734 #endif
735 #ifdef ALTQ_HFSC
736 	case ALTQT_HFSC:
737 		error = hfsc_getqstats(a, ubuf, nbytes, version);
738 		break;
739 #endif
740 #ifdef ALTQ_FAIRQ
741         case ALTQT_FAIRQ:
742                 error = fairq_getqstats(a, ubuf, nbytes, version);
743                 break;
744 #endif
745 #ifdef ALTQ_CODEL
746 	case ALTQT_CODEL:
747 		error = codel_getqstats(a, ubuf, nbytes, version);
748 		break;
749 #endif
750 	default:
751 		error = ENXIO;
752 	}
753 
754 	return (error);
755 }
756 
757 /*
758  * read and write diffserv field in IPv4 or IPv6 header
759  */
760 u_int8_t
761 read_dsfield(m, pktattr)
762 	struct mbuf *m;
763 	struct altq_pktattr *pktattr;
764 {
765 	struct mbuf *m0;
766 	u_int8_t ds_field = 0;
767 
768 	if (pktattr == NULL ||
769 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
770 		return ((u_int8_t)0);
771 
772 	/* verify that pattr_hdr is within the mbuf data */
773 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
774 		if ((pktattr->pattr_hdr >= m0->m_data) &&
775 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
776 			break;
777 	if (m0 == NULL) {
778 		/* ick, pattr_hdr is stale */
779 		pktattr->pattr_af = AF_UNSPEC;
780 #ifdef ALTQ_DEBUG
781 		printf("read_dsfield: can't locate header!\n");
782 #endif
783 		return ((u_int8_t)0);
784 	}
785 
786 	if (pktattr->pattr_af == AF_INET) {
787 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
788 
789 		if (ip->ip_v != 4)
790 			return ((u_int8_t)0);	/* version mismatch! */
791 		ds_field = ip->ip_tos;
792 	}
793 #ifdef INET6
794 	else if (pktattr->pattr_af == AF_INET6) {
795 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
796 		u_int32_t flowlabel;
797 
798 		flowlabel = ntohl(ip6->ip6_flow);
799 		if ((flowlabel >> 28) != 6)
800 			return ((u_int8_t)0);	/* version mismatch! */
801 		ds_field = (flowlabel >> 20) & 0xff;
802 	}
803 #endif
804 	return (ds_field);
805 }
806 
807 void
808 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, u_int8_t dsfield)
809 {
810 	struct mbuf *m0;
811 
812 	if (pktattr == NULL ||
813 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
814 		return;
815 
816 	/* verify that pattr_hdr is within the mbuf data */
817 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
818 		if ((pktattr->pattr_hdr >= m0->m_data) &&
819 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
820 			break;
821 	if (m0 == NULL) {
822 		/* ick, pattr_hdr is stale */
823 		pktattr->pattr_af = AF_UNSPEC;
824 #ifdef ALTQ_DEBUG
825 		printf("write_dsfield: can't locate header!\n");
826 #endif
827 		return;
828 	}
829 
830 	if (pktattr->pattr_af == AF_INET) {
831 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
832 		u_int8_t old;
833 		int32_t sum;
834 
835 		if (ip->ip_v != 4)
836 			return;		/* version mismatch! */
837 		old = ip->ip_tos;
838 		dsfield |= old & 3;	/* leave CU bits */
839 		if (old == dsfield)
840 			return;
841 		ip->ip_tos = dsfield;
842 		/*
843 		 * update checksum (from RFC1624)
844 		 *	   HC' = ~(~HC + ~m + m')
845 		 */
846 		sum = ~ntohs(ip->ip_sum) & 0xffff;
847 		sum += 0xff00 + (~old & 0xff) + dsfield;
848 		sum = (sum >> 16) + (sum & 0xffff);
849 		sum += (sum >> 16);  /* add carry */
850 
851 		ip->ip_sum = htons(~sum & 0xffff);
852 	}
853 #ifdef INET6
854 	else if (pktattr->pattr_af == AF_INET6) {
855 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
856 		u_int32_t flowlabel;
857 
858 		flowlabel = ntohl(ip6->ip6_flow);
859 		if ((flowlabel >> 28) != 6)
860 			return;		/* version mismatch! */
861 		flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20);
862 		ip6->ip6_flow = htonl(flowlabel);
863 	}
864 #endif
865 	return;
866 }
867 
868 /*
869  * high resolution clock support taking advantage of a machine dependent
870  * high resolution time counter (e.g., timestamp counter of intel pentium).
871  * we assume
872  *  - 64-bit-long monotonically-increasing counter
873  *  - frequency range is 100M-4GHz (CPU speed)
874  */
875 /* if pcc is not available or disabled, emulate 256MHz using microtime() */
876 #define	MACHCLK_SHIFT	8
877 
878 int machclk_usepcc;
879 u_int32_t machclk_freq;
880 u_int32_t machclk_per_tick;
881 
882 #if defined(__i386__) && defined(__NetBSD__)
883 extern u_int64_t cpu_tsc_freq;
884 #endif
885 
886 #if (__FreeBSD_version >= 700035)
887 /* Update TSC freq with the value indicated by the caller. */
888 static void
889 tsc_freq_changed(void *arg, const struct cf_level *level, int status)
890 {
891 	/* If there was an error during the transition, don't do anything. */
892 	if (status != 0)
893 		return;
894 
895 #if (__FreeBSD_version >= 701102) && (defined(__amd64__) || defined(__i386__))
896 	/* If TSC is P-state invariant, don't do anything. */
897 	if (tsc_is_invariant)
898 		return;
899 #endif
900 
901 	/* Total setting for this level gives the new frequency in MHz. */
902 	init_machclk();
903 }
904 EVENTHANDLER_DEFINE(cpufreq_post_change, tsc_freq_changed, NULL,
905     EVENTHANDLER_PRI_LAST);
906 #endif /* __FreeBSD_version >= 700035 */
907 
908 static void
909 init_machclk_setup(void)
910 {
911 	callout_init(&tbr_callout, 1);
912 
913 	machclk_usepcc = 1;
914 
915 #if (!defined(__amd64__) && !defined(__i386__)) || defined(ALTQ_NOPCC)
916 	machclk_usepcc = 0;
917 #endif
918 #if defined(__FreeBSD__) && defined(SMP)
919 	machclk_usepcc = 0;
920 #endif
921 #if defined(__NetBSD__) && defined(MULTIPROCESSOR)
922 	machclk_usepcc = 0;
923 #endif
924 #if defined(__amd64__) || defined(__i386__)
925 	/* check if TSC is available */
926 	if ((cpu_feature & CPUID_TSC) == 0 ||
927 	    atomic_load_acq_64(&tsc_freq) == 0)
928 		machclk_usepcc = 0;
929 #endif
930 }
931 
932 void
933 init_machclk(void)
934 {
935 	static int called;
936 
937 	/* Call one-time initialization function. */
938 	if (!called) {
939 		init_machclk_setup();
940 		called = 1;
941 	}
942 
943 	if (machclk_usepcc == 0) {
944 		/* emulate 256MHz using microtime() */
945 		machclk_freq = 1000000 << MACHCLK_SHIFT;
946 		machclk_per_tick = machclk_freq / hz;
947 #ifdef ALTQ_DEBUG
948 		printf("altq: emulate %uHz cpu clock\n", machclk_freq);
949 #endif
950 		return;
951 	}
952 
953 	/*
954 	 * if the clock frequency (of Pentium TSC or Alpha PCC) is
955 	 * accessible, just use it.
956 	 */
957 #if defined(__amd64__) || defined(__i386__)
958 	machclk_freq = atomic_load_acq_64(&tsc_freq);
959 #endif
960 
961 	/*
962 	 * if we don't know the clock frequency, measure it.
963 	 */
964 	if (machclk_freq == 0) {
965 		static int	wait;
966 		struct timeval	tv_start, tv_end;
967 		u_int64_t	start, end, diff;
968 		int		timo;
969 
970 		microtime(&tv_start);
971 		start = read_machclk();
972 		timo = hz;	/* 1 sec */
973 		(void)tsleep(&wait, PWAIT | PCATCH, "init_machclk", timo);
974 		microtime(&tv_end);
975 		end = read_machclk();
976 		diff = (u_int64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000
977 		    + tv_end.tv_usec - tv_start.tv_usec;
978 		if (diff != 0)
979 			machclk_freq = (u_int)((end - start) * 1000000 / diff);
980 	}
981 
982 	machclk_per_tick = machclk_freq / hz;
983 
984 #ifdef ALTQ_DEBUG
985 	printf("altq: CPU clock: %uHz\n", machclk_freq);
986 #endif
987 }
988 
989 #if defined(__OpenBSD__) && defined(__i386__)
990 static __inline u_int64_t
991 rdtsc(void)
992 {
993 	u_int64_t rv;
994 	__asm __volatile(".byte 0x0f, 0x31" : "=A" (rv));
995 	return (rv);
996 }
997 #endif /* __OpenBSD__ && __i386__ */
998 
999 u_int64_t
1000 read_machclk(void)
1001 {
1002 	u_int64_t val;
1003 
1004 	if (machclk_usepcc) {
1005 #if defined(__amd64__) || defined(__i386__)
1006 		val = rdtsc();
1007 #else
1008 		panic("read_machclk");
1009 #endif
1010 	} else {
1011 		struct timeval tv, boottime;
1012 
1013 		microtime(&tv);
1014 		getboottime(&boottime);
1015 		val = (((u_int64_t)(tv.tv_sec - boottime.tv_sec) * 1000000
1016 		    + tv.tv_usec) << MACHCLK_SHIFT);
1017 	}
1018 	return (val);
1019 }
1020 
1021 #ifdef ALTQ3_CLFIER_COMPAT
1022 
1023 #ifndef IPPROTO_ESP
1024 #define	IPPROTO_ESP	50		/* encapsulating security payload */
1025 #endif
1026 #ifndef IPPROTO_AH
1027 #define	IPPROTO_AH	51		/* authentication header */
1028 #endif
1029 
1030 /*
1031  * extract flow information from a given packet.
1032  * filt_mask shows flowinfo fields required.
1033  * we assume the ip header is in one mbuf, and addresses and ports are
1034  * in network byte order.
1035  */
1036 int
1037 altq_extractflow(m, af, flow, filt_bmask)
1038 	struct mbuf *m;
1039 	int af;
1040 	struct flowinfo *flow;
1041 	u_int32_t	filt_bmask;
1042 {
1043 
1044 	switch (af) {
1045 	case PF_INET: {
1046 		struct flowinfo_in *fin;
1047 		struct ip *ip;
1048 
1049 		ip = mtod(m, struct ip *);
1050 
1051 		if (ip->ip_v != 4)
1052 			break;
1053 
1054 		fin = (struct flowinfo_in *)flow;
1055 		fin->fi_len = sizeof(struct flowinfo_in);
1056 		fin->fi_family = AF_INET;
1057 
1058 		fin->fi_proto = ip->ip_p;
1059 		fin->fi_tos = ip->ip_tos;
1060 
1061 		fin->fi_src.s_addr = ip->ip_src.s_addr;
1062 		fin->fi_dst.s_addr = ip->ip_dst.s_addr;
1063 
1064 		if (filt_bmask & FIMB4_PORTS)
1065 			/* if port info is required, extract port numbers */
1066 			extract_ports4(m, ip, fin);
1067 		else {
1068 			fin->fi_sport = 0;
1069 			fin->fi_dport = 0;
1070 			fin->fi_gpi = 0;
1071 		}
1072 		return (1);
1073 	}
1074 
1075 #ifdef INET6
1076 	case PF_INET6: {
1077 		struct flowinfo_in6 *fin6;
1078 		struct ip6_hdr *ip6;
1079 
1080 		ip6 = mtod(m, struct ip6_hdr *);
1081 		/* should we check the ip version? */
1082 
1083 		fin6 = (struct flowinfo_in6 *)flow;
1084 		fin6->fi6_len = sizeof(struct flowinfo_in6);
1085 		fin6->fi6_family = AF_INET6;
1086 
1087 		fin6->fi6_proto = ip6->ip6_nxt;
1088 		fin6->fi6_tclass   = IPV6_TRAFFIC_CLASS(ip6);
1089 
1090 		fin6->fi6_flowlabel = ip6->ip6_flow & htonl(0x000fffff);
1091 		fin6->fi6_src = ip6->ip6_src;
1092 		fin6->fi6_dst = ip6->ip6_dst;
1093 
1094 		if ((filt_bmask & FIMB6_PORTS) ||
1095 		    ((filt_bmask & FIMB6_PROTO)
1096 		     && ip6->ip6_nxt > IPPROTO_IPV6))
1097 			/*
1098 			 * if port info is required, or proto is required
1099 			 * but there are option headers, extract port
1100 			 * and protocol numbers.
1101 			 */
1102 			extract_ports6(m, ip6, fin6);
1103 		else {
1104 			fin6->fi6_sport = 0;
1105 			fin6->fi6_dport = 0;
1106 			fin6->fi6_gpi = 0;
1107 		}
1108 		return (1);
1109 	}
1110 #endif /* INET6 */
1111 
1112 	default:
1113 		break;
1114 	}
1115 
1116 	/* failed */
1117 	flow->fi_len = sizeof(struct flowinfo);
1118 	flow->fi_family = AF_UNSPEC;
1119 	return (0);
1120 }
1121 
1122 /*
1123  * helper routine to extract port numbers
1124  */
1125 /* structure for ipsec and ipv6 option header template */
1126 struct _opt6 {
1127 	u_int8_t	opt6_nxt;	/* next header */
1128 	u_int8_t	opt6_hlen;	/* header extension length */
1129 	u_int16_t	_pad;
1130 	u_int32_t	ah_spi;		/* security parameter index
1131 					   for authentication header */
1132 };
1133 
1134 /*
1135  * extract port numbers from a ipv4 packet.
1136  */
1137 static int
1138 extract_ports4(m, ip, fin)
1139 	struct mbuf *m;
1140 	struct ip *ip;
1141 	struct flowinfo_in *fin;
1142 {
1143 	struct mbuf *m0;
1144 	u_short ip_off;
1145 	u_int8_t proto;
1146 	int 	off;
1147 
1148 	fin->fi_sport = 0;
1149 	fin->fi_dport = 0;
1150 	fin->fi_gpi = 0;
1151 
1152 	ip_off = ntohs(ip->ip_off);
1153 	/* if it is a fragment, try cached fragment info */
1154 	if (ip_off & IP_OFFMASK) {
1155 		ip4f_lookup(ip, fin);
1156 		return (1);
1157 	}
1158 
1159 	/* locate the mbuf containing the protocol header */
1160 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
1161 		if (((caddr_t)ip >= m0->m_data) &&
1162 		    ((caddr_t)ip < m0->m_data + m0->m_len))
1163 			break;
1164 	if (m0 == NULL) {
1165 #ifdef ALTQ_DEBUG
1166 		printf("extract_ports4: can't locate header! ip=%p\n", ip);
1167 #endif
1168 		return (0);
1169 	}
1170 	off = ((caddr_t)ip - m0->m_data) + (ip->ip_hl << 2);
1171 	proto = ip->ip_p;
1172 
1173 #ifdef ALTQ_IPSEC
1174  again:
1175 #endif
1176 	while (off >= m0->m_len) {
1177 		off -= m0->m_len;
1178 		m0 = m0->m_next;
1179 		if (m0 == NULL)
1180 			return (0);  /* bogus ip_hl! */
1181 	}
1182 	if (m0->m_len < off + 4)
1183 		return (0);
1184 
1185 	switch (proto) {
1186 	case IPPROTO_TCP:
1187 	case IPPROTO_UDP: {
1188 		struct udphdr *udp;
1189 
1190 		udp = (struct udphdr *)(mtod(m0, caddr_t) + off);
1191 		fin->fi_sport = udp->uh_sport;
1192 		fin->fi_dport = udp->uh_dport;
1193 		fin->fi_proto = proto;
1194 		}
1195 		break;
1196 
1197 #ifdef ALTQ_IPSEC
1198 	case IPPROTO_ESP:
1199 		if (fin->fi_gpi == 0){
1200 			u_int32_t *gpi;
1201 
1202 			gpi = (u_int32_t *)(mtod(m0, caddr_t) + off);
1203 			fin->fi_gpi   = *gpi;
1204 		}
1205 		fin->fi_proto = proto;
1206 		break;
1207 
1208 	case IPPROTO_AH: {
1209 			/* get next header and header length */
1210 			struct _opt6 *opt6;
1211 
1212 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1213 			proto = opt6->opt6_nxt;
1214 			off += 8 + (opt6->opt6_hlen * 4);
1215 			if (fin->fi_gpi == 0 && m0->m_len >= off + 8)
1216 				fin->fi_gpi = opt6->ah_spi;
1217 		}
1218 		/* goto the next header */
1219 		goto again;
1220 #endif  /* ALTQ_IPSEC */
1221 
1222 	default:
1223 		fin->fi_proto = proto;
1224 		return (0);
1225 	}
1226 
1227 	/* if this is a first fragment, cache it. */
1228 	if (ip_off & IP_MF)
1229 		ip4f_cache(ip, fin);
1230 
1231 	return (1);
1232 }
1233 
1234 #ifdef INET6
1235 static int
1236 extract_ports6(m, ip6, fin6)
1237 	struct mbuf *m;
1238 	struct ip6_hdr *ip6;
1239 	struct flowinfo_in6 *fin6;
1240 {
1241 	struct mbuf *m0;
1242 	int	off;
1243 	u_int8_t proto;
1244 
1245 	fin6->fi6_gpi   = 0;
1246 	fin6->fi6_sport = 0;
1247 	fin6->fi6_dport = 0;
1248 
1249 	/* locate the mbuf containing the protocol header */
1250 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
1251 		if (((caddr_t)ip6 >= m0->m_data) &&
1252 		    ((caddr_t)ip6 < m0->m_data + m0->m_len))
1253 			break;
1254 	if (m0 == NULL) {
1255 #ifdef ALTQ_DEBUG
1256 		printf("extract_ports6: can't locate header! ip6=%p\n", ip6);
1257 #endif
1258 		return (0);
1259 	}
1260 	off = ((caddr_t)ip6 - m0->m_data) + sizeof(struct ip6_hdr);
1261 
1262 	proto = ip6->ip6_nxt;
1263 	do {
1264 		while (off >= m0->m_len) {
1265 			off -= m0->m_len;
1266 			m0 = m0->m_next;
1267 			if (m0 == NULL)
1268 				return (0);
1269 		}
1270 		if (m0->m_len < off + 4)
1271 			return (0);
1272 
1273 		switch (proto) {
1274 		case IPPROTO_TCP:
1275 		case IPPROTO_UDP: {
1276 			struct udphdr *udp;
1277 
1278 			udp = (struct udphdr *)(mtod(m0, caddr_t) + off);
1279 			fin6->fi6_sport = udp->uh_sport;
1280 			fin6->fi6_dport = udp->uh_dport;
1281 			fin6->fi6_proto = proto;
1282 			}
1283 			return (1);
1284 
1285 		case IPPROTO_ESP:
1286 			if (fin6->fi6_gpi == 0) {
1287 				u_int32_t *gpi;
1288 
1289 				gpi = (u_int32_t *)(mtod(m0, caddr_t) + off);
1290 				fin6->fi6_gpi   = *gpi;
1291 			}
1292 			fin6->fi6_proto = proto;
1293 			return (1);
1294 
1295 		case IPPROTO_AH: {
1296 			/* get next header and header length */
1297 			struct _opt6 *opt6;
1298 
1299 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1300 			if (fin6->fi6_gpi == 0 && m0->m_len >= off + 8)
1301 				fin6->fi6_gpi = opt6->ah_spi;
1302 			proto = opt6->opt6_nxt;
1303 			off += 8 + (opt6->opt6_hlen * 4);
1304 			/* goto the next header */
1305 			break;
1306 			}
1307 
1308 		case IPPROTO_HOPOPTS:
1309 		case IPPROTO_ROUTING:
1310 		case IPPROTO_DSTOPTS: {
1311 			/* get next header and header length */
1312 			struct _opt6 *opt6;
1313 
1314 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1315 			proto = opt6->opt6_nxt;
1316 			off += (opt6->opt6_hlen + 1) * 8;
1317 			/* goto the next header */
1318 			break;
1319 			}
1320 
1321 		case IPPROTO_FRAGMENT:
1322 			/* ipv6 fragmentations are not supported yet */
1323 		default:
1324 			fin6->fi6_proto = proto;
1325 			return (0);
1326 		}
1327 	} while (1);
1328 	/*NOTREACHED*/
1329 }
1330 #endif /* INET6 */
1331 
1332 /*
1333  * altq common classifier
1334  */
1335 int
1336 acc_add_filter(classifier, filter, class, phandle)
1337 	struct acc_classifier *classifier;
1338 	struct flow_filter *filter;
1339 	void	*class;
1340 	u_long	*phandle;
1341 {
1342 	struct acc_filter *afp, *prev, *tmp;
1343 	int	i, s;
1344 
1345 #ifdef INET6
1346 	if (filter->ff_flow.fi_family != AF_INET &&
1347 	    filter->ff_flow.fi_family != AF_INET6)
1348 		return (EINVAL);
1349 #else
1350 	if (filter->ff_flow.fi_family != AF_INET)
1351 		return (EINVAL);
1352 #endif
1353 
1354 	afp = malloc(sizeof(struct acc_filter),
1355 	       M_DEVBUF, M_WAITOK);
1356 	if (afp == NULL)
1357 		return (ENOMEM);
1358 	bzero(afp, sizeof(struct acc_filter));
1359 
1360 	afp->f_filter = *filter;
1361 	afp->f_class = class;
1362 
1363 	i = ACC_WILDCARD_INDEX;
1364 	if (filter->ff_flow.fi_family == AF_INET) {
1365 		struct flow_filter *filter4 = &afp->f_filter;
1366 
1367 		/*
1368 		 * if address is 0, it's a wildcard.  if address mask
1369 		 * isn't set, use full mask.
1370 		 */
1371 		if (filter4->ff_flow.fi_dst.s_addr == 0)
1372 			filter4->ff_mask.mask_dst.s_addr = 0;
1373 		else if (filter4->ff_mask.mask_dst.s_addr == 0)
1374 			filter4->ff_mask.mask_dst.s_addr = 0xffffffff;
1375 		if (filter4->ff_flow.fi_src.s_addr == 0)
1376 			filter4->ff_mask.mask_src.s_addr = 0;
1377 		else if (filter4->ff_mask.mask_src.s_addr == 0)
1378 			filter4->ff_mask.mask_src.s_addr = 0xffffffff;
1379 
1380 		/* clear extra bits in addresses  */
1381 		   filter4->ff_flow.fi_dst.s_addr &=
1382 		       filter4->ff_mask.mask_dst.s_addr;
1383 		   filter4->ff_flow.fi_src.s_addr &=
1384 		       filter4->ff_mask.mask_src.s_addr;
1385 
1386 		/*
1387 		 * if dst address is a wildcard, use hash-entry
1388 		 * ACC_WILDCARD_INDEX.
1389 		 */
1390 		if (filter4->ff_mask.mask_dst.s_addr != 0xffffffff)
1391 			i = ACC_WILDCARD_INDEX;
1392 		else
1393 			i = ACC_GET_HASH_INDEX(filter4->ff_flow.fi_dst.s_addr);
1394 	}
1395 #ifdef INET6
1396 	else if (filter->ff_flow.fi_family == AF_INET6) {
1397 		struct flow_filter6 *filter6 =
1398 			(struct flow_filter6 *)&afp->f_filter;
1399 #ifndef IN6MASK0 /* taken from kame ipv6 */
1400 #define	IN6MASK0	{{{ 0, 0, 0, 0 }}}
1401 #define	IN6MASK128	{{{ 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff }}}
1402 		const struct in6_addr in6mask0 = IN6MASK0;
1403 		const struct in6_addr in6mask128 = IN6MASK128;
1404 #endif
1405 
1406 		if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_dst))
1407 			filter6->ff_mask6.mask6_dst = in6mask0;
1408 		else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_dst))
1409 			filter6->ff_mask6.mask6_dst = in6mask128;
1410 		if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_src))
1411 			filter6->ff_mask6.mask6_src = in6mask0;
1412 		else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_src))
1413 			filter6->ff_mask6.mask6_src = in6mask128;
1414 
1415 		/* clear extra bits in addresses  */
1416 		for (i = 0; i < 16; i++)
1417 			filter6->ff_flow6.fi6_dst.s6_addr[i] &=
1418 			    filter6->ff_mask6.mask6_dst.s6_addr[i];
1419 		for (i = 0; i < 16; i++)
1420 			filter6->ff_flow6.fi6_src.s6_addr[i] &=
1421 			    filter6->ff_mask6.mask6_src.s6_addr[i];
1422 
1423 		if (filter6->ff_flow6.fi6_flowlabel == 0)
1424 			i = ACC_WILDCARD_INDEX;
1425 		else
1426 			i = ACC_GET_HASH_INDEX(filter6->ff_flow6.fi6_flowlabel);
1427 	}
1428 #endif /* INET6 */
1429 
1430 	afp->f_handle = get_filt_handle(classifier, i);
1431 
1432 	/* update filter bitmask */
1433 	afp->f_fbmask = filt2fibmask(filter);
1434 	classifier->acc_fbmask |= afp->f_fbmask;
1435 
1436 	/*
1437 	 * add this filter to the filter list.
1438 	 * filters are ordered from the highest rule number.
1439 	 */
1440 	s = splnet();
1441 	prev = NULL;
1442 	LIST_FOREACH(tmp, &classifier->acc_filters[i], f_chain) {
1443 		if (tmp->f_filter.ff_ruleno > afp->f_filter.ff_ruleno)
1444 			prev = tmp;
1445 		else
1446 			break;
1447 	}
1448 	if (prev == NULL)
1449 		LIST_INSERT_HEAD(&classifier->acc_filters[i], afp, f_chain);
1450 	else
1451 		LIST_INSERT_AFTER(prev, afp, f_chain);
1452 	splx(s);
1453 
1454 	*phandle = afp->f_handle;
1455 	return (0);
1456 }
1457 
1458 int
1459 acc_delete_filter(classifier, handle)
1460 	struct acc_classifier *classifier;
1461 	u_long handle;
1462 {
1463 	struct acc_filter *afp;
1464 	int	s;
1465 
1466 	if ((afp = filth_to_filtp(classifier, handle)) == NULL)
1467 		return (EINVAL);
1468 
1469 	s = splnet();
1470 	LIST_REMOVE(afp, f_chain);
1471 	splx(s);
1472 
1473 	free(afp, M_DEVBUF);
1474 
1475 	/* todo: update filt_bmask */
1476 
1477 	return (0);
1478 }
1479 
1480 /*
1481  * delete filters referencing to the specified class.
1482  * if the all flag is not 0, delete all the filters.
1483  */
1484 int
1485 acc_discard_filters(classifier, class, all)
1486 	struct acc_classifier *classifier;
1487 	void	*class;
1488 	int	all;
1489 {
1490 	struct acc_filter *afp;
1491 	int	i, s;
1492 
1493 	s = splnet();
1494 	for (i = 0; i < ACC_FILTER_TABLESIZE; i++) {
1495 		do {
1496 			LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1497 				if (all || afp->f_class == class) {
1498 					LIST_REMOVE(afp, f_chain);
1499 					free(afp, M_DEVBUF);
1500 					/* start again from the head */
1501 					break;
1502 				}
1503 		} while (afp != NULL);
1504 	}
1505 	splx(s);
1506 
1507 	if (all)
1508 		classifier->acc_fbmask = 0;
1509 
1510 	return (0);
1511 }
1512 
1513 void *
1514 acc_classify(clfier, m, af)
1515 	void *clfier;
1516 	struct mbuf *m;
1517 	int af;
1518 {
1519 	struct acc_classifier *classifier;
1520 	struct flowinfo flow;
1521 	struct acc_filter *afp;
1522 	int	i;
1523 
1524 	classifier = (struct acc_classifier *)clfier;
1525 	altq_extractflow(m, af, &flow, classifier->acc_fbmask);
1526 
1527 	if (flow.fi_family == AF_INET) {
1528 		struct flowinfo_in *fp = (struct flowinfo_in *)&flow;
1529 
1530 		if ((classifier->acc_fbmask & FIMB4_ALL) == FIMB4_TOS) {
1531 			/* only tos is used */
1532 			LIST_FOREACH(afp,
1533 				 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1534 				 f_chain)
1535 				if (apply_tosfilter4(afp->f_fbmask,
1536 						     &afp->f_filter, fp))
1537 					/* filter matched */
1538 					return (afp->f_class);
1539 		} else if ((classifier->acc_fbmask &
1540 			(~(FIMB4_PROTO|FIMB4_SPORT|FIMB4_DPORT) & FIMB4_ALL))
1541 		    == 0) {
1542 			/* only proto and ports are used */
1543 			LIST_FOREACH(afp,
1544 				 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1545 				 f_chain)
1546 				if (apply_ppfilter4(afp->f_fbmask,
1547 						    &afp->f_filter, fp))
1548 					/* filter matched */
1549 					return (afp->f_class);
1550 		} else {
1551 			/* get the filter hash entry from its dest address */
1552 			i = ACC_GET_HASH_INDEX(fp->fi_dst.s_addr);
1553 			do {
1554 				/*
1555 				 * go through this loop twice.  first for dst
1556 				 * hash, second for wildcards.
1557 				 */
1558 				LIST_FOREACH(afp, &classifier->acc_filters[i],
1559 					     f_chain)
1560 					if (apply_filter4(afp->f_fbmask,
1561 							  &afp->f_filter, fp))
1562 						/* filter matched */
1563 						return (afp->f_class);
1564 
1565 				/*
1566 				 * check again for filters with a dst addr
1567 				 * wildcard.
1568 				 * (daddr == 0 || dmask != 0xffffffff).
1569 				 */
1570 				if (i != ACC_WILDCARD_INDEX)
1571 					i = ACC_WILDCARD_INDEX;
1572 				else
1573 					break;
1574 			} while (1);
1575 		}
1576 	}
1577 #ifdef INET6
1578 	else if (flow.fi_family == AF_INET6) {
1579 		struct flowinfo_in6 *fp6 = (struct flowinfo_in6 *)&flow;
1580 
1581 		/* get the filter hash entry from its flow ID */
1582 		if (fp6->fi6_flowlabel != 0)
1583 			i = ACC_GET_HASH_INDEX(fp6->fi6_flowlabel);
1584 		else
1585 			/* flowlable can be zero */
1586 			i = ACC_WILDCARD_INDEX;
1587 
1588 		/* go through this loop twice.  first for flow hash, second
1589 		   for wildcards. */
1590 		do {
1591 			LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1592 				if (apply_filter6(afp->f_fbmask,
1593 					(struct flow_filter6 *)&afp->f_filter,
1594 					fp6))
1595 					/* filter matched */
1596 					return (afp->f_class);
1597 
1598 			/*
1599 			 * check again for filters with a wildcard.
1600 			 */
1601 			if (i != ACC_WILDCARD_INDEX)
1602 				i = ACC_WILDCARD_INDEX;
1603 			else
1604 				break;
1605 		} while (1);
1606 	}
1607 #endif /* INET6 */
1608 
1609 	/* no filter matched */
1610 	return (NULL);
1611 }
1612 
1613 static int
1614 apply_filter4(fbmask, filt, pkt)
1615 	u_int32_t	fbmask;
1616 	struct flow_filter *filt;
1617 	struct flowinfo_in *pkt;
1618 {
1619 	if (filt->ff_flow.fi_family != AF_INET)
1620 		return (0);
1621 	if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1622 		return (0);
1623 	if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1624 		return (0);
1625 	if ((fbmask & FIMB4_DADDR) &&
1626 	    filt->ff_flow.fi_dst.s_addr !=
1627 	    (pkt->fi_dst.s_addr & filt->ff_mask.mask_dst.s_addr))
1628 		return (0);
1629 	if ((fbmask & FIMB4_SADDR) &&
1630 	    filt->ff_flow.fi_src.s_addr !=
1631 	    (pkt->fi_src.s_addr & filt->ff_mask.mask_src.s_addr))
1632 		return (0);
1633 	if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1634 		return (0);
1635 	if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1636 	    (pkt->fi_tos & filt->ff_mask.mask_tos))
1637 		return (0);
1638 	if ((fbmask & FIMB4_GPI) && filt->ff_flow.fi_gpi != (pkt->fi_gpi))
1639 		return (0);
1640 	/* match */
1641 	return (1);
1642 }
1643 
1644 /*
1645  * filter matching function optimized for a common case that checks
1646  * only protocol and port numbers
1647  */
1648 static int
1649 apply_ppfilter4(fbmask, filt, pkt)
1650 	u_int32_t	fbmask;
1651 	struct flow_filter *filt;
1652 	struct flowinfo_in *pkt;
1653 {
1654 	if (filt->ff_flow.fi_family != AF_INET)
1655 		return (0);
1656 	if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1657 		return (0);
1658 	if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1659 		return (0);
1660 	if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1661 		return (0);
1662 	/* match */
1663 	return (1);
1664 }
1665 
1666 /*
1667  * filter matching function only for tos field.
1668  */
1669 static int
1670 apply_tosfilter4(fbmask, filt, pkt)
1671 	u_int32_t	fbmask;
1672 	struct flow_filter *filt;
1673 	struct flowinfo_in *pkt;
1674 {
1675 	if (filt->ff_flow.fi_family != AF_INET)
1676 		return (0);
1677 	if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1678 	    (pkt->fi_tos & filt->ff_mask.mask_tos))
1679 		return (0);
1680 	/* match */
1681 	return (1);
1682 }
1683 
1684 #ifdef INET6
1685 static int
1686 apply_filter6(fbmask, filt, pkt)
1687 	u_int32_t	fbmask;
1688 	struct flow_filter6 *filt;
1689 	struct flowinfo_in6 *pkt;
1690 {
1691 	int i;
1692 
1693 	if (filt->ff_flow6.fi6_family != AF_INET6)
1694 		return (0);
1695 	if ((fbmask & FIMB6_FLABEL) &&
1696 	    filt->ff_flow6.fi6_flowlabel != pkt->fi6_flowlabel)
1697 		return (0);
1698 	if ((fbmask & FIMB6_PROTO) &&
1699 	    filt->ff_flow6.fi6_proto != pkt->fi6_proto)
1700 		return (0);
1701 	if ((fbmask & FIMB6_SPORT) &&
1702 	    filt->ff_flow6.fi6_sport != pkt->fi6_sport)
1703 		return (0);
1704 	if ((fbmask & FIMB6_DPORT) &&
1705 	    filt->ff_flow6.fi6_dport != pkt->fi6_dport)
1706 		return (0);
1707 	if (fbmask & FIMB6_SADDR) {
1708 		for (i = 0; i < 4; i++)
1709 			if (filt->ff_flow6.fi6_src.s6_addr32[i] !=
1710 			    (pkt->fi6_src.s6_addr32[i] &
1711 			     filt->ff_mask6.mask6_src.s6_addr32[i]))
1712 				return (0);
1713 	}
1714 	if (fbmask & FIMB6_DADDR) {
1715 		for (i = 0; i < 4; i++)
1716 			if (filt->ff_flow6.fi6_dst.s6_addr32[i] !=
1717 			    (pkt->fi6_dst.s6_addr32[i] &
1718 			     filt->ff_mask6.mask6_dst.s6_addr32[i]))
1719 				return (0);
1720 	}
1721 	if ((fbmask & FIMB6_TCLASS) &&
1722 	    filt->ff_flow6.fi6_tclass !=
1723 	    (pkt->fi6_tclass & filt->ff_mask6.mask6_tclass))
1724 		return (0);
1725 	if ((fbmask & FIMB6_GPI) &&
1726 	    filt->ff_flow6.fi6_gpi != pkt->fi6_gpi)
1727 		return (0);
1728 	/* match */
1729 	return (1);
1730 }
1731 #endif /* INET6 */
1732 
1733 /*
1734  *  filter handle:
1735  *	bit 20-28: index to the filter hash table
1736  *	bit  0-19: unique id in the hash bucket.
1737  */
1738 static u_long
1739 get_filt_handle(classifier, i)
1740 	struct acc_classifier *classifier;
1741 	int	i;
1742 {
1743 	static u_long handle_number = 1;
1744 	u_long 	handle;
1745 	struct acc_filter *afp;
1746 
1747 	while (1) {
1748 		handle = handle_number++ & 0x000fffff;
1749 
1750 		if (LIST_EMPTY(&classifier->acc_filters[i]))
1751 			break;
1752 
1753 		LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1754 			if ((afp->f_handle & 0x000fffff) == handle)
1755 				break;
1756 		if (afp == NULL)
1757 			break;
1758 		/* this handle is already used, try again */
1759 	}
1760 
1761 	return ((i << 20) | handle);
1762 }
1763 
1764 /* convert filter handle to filter pointer */
1765 static struct acc_filter *
1766 filth_to_filtp(classifier, handle)
1767 	struct acc_classifier *classifier;
1768 	u_long handle;
1769 {
1770 	struct acc_filter *afp;
1771 	int	i;
1772 
1773 	i = ACC_GET_HINDEX(handle);
1774 
1775 	LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1776 		if (afp->f_handle == handle)
1777 			return (afp);
1778 
1779 	return (NULL);
1780 }
1781 
1782 /* create flowinfo bitmask */
1783 static u_int32_t
1784 filt2fibmask(filt)
1785 	struct flow_filter *filt;
1786 {
1787 	u_int32_t mask = 0;
1788 #ifdef INET6
1789 	struct flow_filter6 *filt6;
1790 #endif
1791 
1792 	switch (filt->ff_flow.fi_family) {
1793 	case AF_INET:
1794 		if (filt->ff_flow.fi_proto != 0)
1795 			mask |= FIMB4_PROTO;
1796 		if (filt->ff_flow.fi_tos != 0)
1797 			mask |= FIMB4_TOS;
1798 		if (filt->ff_flow.fi_dst.s_addr != 0)
1799 			mask |= FIMB4_DADDR;
1800 		if (filt->ff_flow.fi_src.s_addr != 0)
1801 			mask |= FIMB4_SADDR;
1802 		if (filt->ff_flow.fi_sport != 0)
1803 			mask |= FIMB4_SPORT;
1804 		if (filt->ff_flow.fi_dport != 0)
1805 			mask |= FIMB4_DPORT;
1806 		if (filt->ff_flow.fi_gpi != 0)
1807 			mask |= FIMB4_GPI;
1808 		break;
1809 #ifdef INET6
1810 	case AF_INET6:
1811 		filt6 = (struct flow_filter6 *)filt;
1812 
1813 		if (filt6->ff_flow6.fi6_proto != 0)
1814 			mask |= FIMB6_PROTO;
1815 		if (filt6->ff_flow6.fi6_tclass != 0)
1816 			mask |= FIMB6_TCLASS;
1817 		if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_dst))
1818 			mask |= FIMB6_DADDR;
1819 		if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_src))
1820 			mask |= FIMB6_SADDR;
1821 		if (filt6->ff_flow6.fi6_sport != 0)
1822 			mask |= FIMB6_SPORT;
1823 		if (filt6->ff_flow6.fi6_dport != 0)
1824 			mask |= FIMB6_DPORT;
1825 		if (filt6->ff_flow6.fi6_gpi != 0)
1826 			mask |= FIMB6_GPI;
1827 		if (filt6->ff_flow6.fi6_flowlabel != 0)
1828 			mask |= FIMB6_FLABEL;
1829 		break;
1830 #endif /* INET6 */
1831 	}
1832 	return (mask);
1833 }
1834 
1835 /*
1836  * helper functions to handle IPv4 fragments.
1837  * currently only in-sequence fragments are handled.
1838  *	- fragment info is cached in a LRU list.
1839  *	- when a first fragment is found, cache its flow info.
1840  *	- when a non-first fragment is found, lookup the cache.
1841  */
1842 
1843 struct ip4_frag {
1844     TAILQ_ENTRY(ip4_frag) ip4f_chain;
1845     char    ip4f_valid;
1846     u_short ip4f_id;
1847     struct flowinfo_in ip4f_info;
1848 };
1849 
1850 static TAILQ_HEAD(ip4f_list, ip4_frag) ip4f_list; /* IPv4 fragment cache */
1851 
1852 #define	IP4F_TABSIZE		16	/* IPv4 fragment cache size */
1853 
1854 static void
1855 ip4f_cache(ip, fin)
1856 	struct ip *ip;
1857 	struct flowinfo_in *fin;
1858 {
1859 	struct ip4_frag *fp;
1860 
1861 	if (TAILQ_EMPTY(&ip4f_list)) {
1862 		/* first time call, allocate fragment cache entries. */
1863 		if (ip4f_init() < 0)
1864 			/* allocation failed! */
1865 			return;
1866 	}
1867 
1868 	fp = ip4f_alloc();
1869 	fp->ip4f_id = ip->ip_id;
1870 	fp->ip4f_info.fi_proto = ip->ip_p;
1871 	fp->ip4f_info.fi_src.s_addr = ip->ip_src.s_addr;
1872 	fp->ip4f_info.fi_dst.s_addr = ip->ip_dst.s_addr;
1873 
1874 	/* save port numbers */
1875 	fp->ip4f_info.fi_sport = fin->fi_sport;
1876 	fp->ip4f_info.fi_dport = fin->fi_dport;
1877 	fp->ip4f_info.fi_gpi   = fin->fi_gpi;
1878 }
1879 
1880 static int
1881 ip4f_lookup(ip, fin)
1882 	struct ip *ip;
1883 	struct flowinfo_in *fin;
1884 {
1885 	struct ip4_frag *fp;
1886 
1887 	for (fp = TAILQ_FIRST(&ip4f_list); fp != NULL && fp->ip4f_valid;
1888 	     fp = TAILQ_NEXT(fp, ip4f_chain))
1889 		if (ip->ip_id == fp->ip4f_id &&
1890 		    ip->ip_src.s_addr == fp->ip4f_info.fi_src.s_addr &&
1891 		    ip->ip_dst.s_addr == fp->ip4f_info.fi_dst.s_addr &&
1892 		    ip->ip_p == fp->ip4f_info.fi_proto) {
1893 			/* found the matching entry */
1894 			fin->fi_sport = fp->ip4f_info.fi_sport;
1895 			fin->fi_dport = fp->ip4f_info.fi_dport;
1896 			fin->fi_gpi   = fp->ip4f_info.fi_gpi;
1897 
1898 			if ((ntohs(ip->ip_off) & IP_MF) == 0)
1899 				/* this is the last fragment,
1900 				   release the entry. */
1901 				ip4f_free(fp);
1902 
1903 			return (1);
1904 		}
1905 
1906 	/* no matching entry found */
1907 	return (0);
1908 }
1909 
1910 static int
1911 ip4f_init(void)
1912 {
1913 	struct ip4_frag *fp;
1914 	int i;
1915 
1916 	TAILQ_INIT(&ip4f_list);
1917 	for (i=0; i<IP4F_TABSIZE; i++) {
1918 		fp = malloc(sizeof(struct ip4_frag),
1919 		       M_DEVBUF, M_NOWAIT);
1920 		if (fp == NULL) {
1921 			printf("ip4f_init: can't alloc %dth entry!\n", i);
1922 			if (i == 0)
1923 				return (-1);
1924 			return (0);
1925 		}
1926 		fp->ip4f_valid = 0;
1927 		TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1928 	}
1929 	return (0);
1930 }
1931 
1932 static struct ip4_frag *
1933 ip4f_alloc(void)
1934 {
1935 	struct ip4_frag *fp;
1936 
1937 	/* reclaim an entry at the tail, put it at the head */
1938 	fp = TAILQ_LAST(&ip4f_list, ip4f_list);
1939 	TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1940 	fp->ip4f_valid = 1;
1941 	TAILQ_INSERT_HEAD(&ip4f_list, fp, ip4f_chain);
1942 	return (fp);
1943 }
1944 
1945 static void
1946 ip4f_free(fp)
1947 	struct ip4_frag *fp;
1948 {
1949 	TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1950 	fp->ip4f_valid = 0;
1951 	TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1952 }
1953 
1954 #endif /* ALTQ3_CLFIER_COMPAT */
1955