xref: /freebsd/sys/net/altq/altq_subr.c (revision 08c4a937a6685f05667996228898521fc453f8f3)
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 /*
118  * alternate queueing support routines
119  */
120 
121 /* look up the queue state by the interface name and the queueing type. */
122 void *
123 altq_lookup(name, type)
124 	char *name;
125 	int type;
126 {
127 	struct ifnet *ifp;
128 
129 	if ((ifp = ifunit(name)) != NULL) {
130 		/* read if_snd unlocked */
131 		if (type != ALTQT_NONE && ifp->if_snd.altq_type == type)
132 			return (ifp->if_snd.altq_disc);
133 	}
134 
135 	return NULL;
136 }
137 
138 int
139 altq_attach(ifq, type, discipline, enqueue, dequeue, request, clfier, classify)
140 	struct ifaltq *ifq;
141 	int type;
142 	void *discipline;
143 	int (*enqueue)(struct ifaltq *, struct mbuf *, struct altq_pktattr *);
144 	struct mbuf *(*dequeue)(struct ifaltq *, int);
145 	int (*request)(struct ifaltq *, int, void *);
146 	void *clfier;
147 	void *(*classify)(void *, struct mbuf *, int);
148 {
149 	IFQ_LOCK(ifq);
150 	if (!ALTQ_IS_READY(ifq)) {
151 		IFQ_UNLOCK(ifq);
152 		return ENXIO;
153 	}
154 
155 	ifq->altq_type     = type;
156 	ifq->altq_disc     = discipline;
157 	ifq->altq_enqueue  = enqueue;
158 	ifq->altq_dequeue  = dequeue;
159 	ifq->altq_request  = request;
160 	ifq->altq_clfier   = clfier;
161 	ifq->altq_classify = classify;
162 	ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED);
163 	IFQ_UNLOCK(ifq);
164 	return 0;
165 }
166 
167 int
168 altq_detach(ifq)
169 	struct ifaltq *ifq;
170 {
171 	IFQ_LOCK(ifq);
172 
173 	if (!ALTQ_IS_READY(ifq)) {
174 		IFQ_UNLOCK(ifq);
175 		return ENXIO;
176 	}
177 	if (ALTQ_IS_ENABLED(ifq)) {
178 		IFQ_UNLOCK(ifq);
179 		return EBUSY;
180 	}
181 	if (!ALTQ_IS_ATTACHED(ifq)) {
182 		IFQ_UNLOCK(ifq);
183 		return (0);
184 	}
185 
186 	ifq->altq_type     = ALTQT_NONE;
187 	ifq->altq_disc     = NULL;
188 	ifq->altq_enqueue  = NULL;
189 	ifq->altq_dequeue  = NULL;
190 	ifq->altq_request  = NULL;
191 	ifq->altq_clfier   = NULL;
192 	ifq->altq_classify = NULL;
193 	ifq->altq_flags &= ALTQF_CANTCHANGE;
194 
195 	IFQ_UNLOCK(ifq);
196 	return 0;
197 }
198 
199 int
200 altq_enable(ifq)
201 	struct ifaltq *ifq;
202 {
203 	int s;
204 
205 	IFQ_LOCK(ifq);
206 
207 	if (!ALTQ_IS_READY(ifq)) {
208 		IFQ_UNLOCK(ifq);
209 		return ENXIO;
210 	}
211 	if (ALTQ_IS_ENABLED(ifq)) {
212 		IFQ_UNLOCK(ifq);
213 		return 0;
214 	}
215 
216 	s = splnet();
217 	IFQ_PURGE_NOLOCK(ifq);
218 	ASSERT(ifq->ifq_len == 0);
219 	ifq->ifq_drv_maxlen = 0;		/* disable bulk dequeue */
220 	ifq->altq_flags |= ALTQF_ENABLED;
221 	if (ifq->altq_clfier != NULL)
222 		ifq->altq_flags |= ALTQF_CLASSIFY;
223 	splx(s);
224 
225 	IFQ_UNLOCK(ifq);
226 	return 0;
227 }
228 
229 int
230 altq_disable(ifq)
231 	struct ifaltq *ifq;
232 {
233 	int s;
234 
235 	IFQ_LOCK(ifq);
236 	if (!ALTQ_IS_ENABLED(ifq)) {
237 		IFQ_UNLOCK(ifq);
238 		return 0;
239 	}
240 
241 	s = splnet();
242 	IFQ_PURGE_NOLOCK(ifq);
243 	ASSERT(ifq->ifq_len == 0);
244 	ifq->altq_flags &= ~(ALTQF_ENABLED|ALTQF_CLASSIFY);
245 	splx(s);
246 
247 	IFQ_UNLOCK(ifq);
248 	return 0;
249 }
250 
251 #ifdef ALTQ_DEBUG
252 void
253 altq_assert(file, line, failedexpr)
254 	const char *file, *failedexpr;
255 	int line;
256 {
257 	(void)printf("altq assertion \"%s\" failed: file \"%s\", line %d\n",
258 		     failedexpr, file, line);
259 	panic("altq assertion");
260 	/* NOTREACHED */
261 }
262 #endif
263 
264 /*
265  * internal representation of token bucket parameters
266  *	rate:	(byte_per_unittime << TBR_SHIFT)  / machclk_freq
267  *		(((bits_per_sec) / 8) << TBR_SHIFT) / machclk_freq
268  *	depth:	byte << TBR_SHIFT
269  *
270  */
271 #define	TBR_SHIFT	29
272 #define	TBR_SCALE(x)	((int64_t)(x) << TBR_SHIFT)
273 #define	TBR_UNSCALE(x)	((x) >> TBR_SHIFT)
274 
275 static struct mbuf *
276 tbr_dequeue(ifq, op)
277 	struct ifaltq *ifq;
278 	int op;
279 {
280 	struct tb_regulator *tbr;
281 	struct mbuf *m;
282 	int64_t interval;
283 	u_int64_t now;
284 
285 	IFQ_LOCK_ASSERT(ifq);
286 	tbr = ifq->altq_tbr;
287 	if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) {
288 		/* if this is a remove after poll, bypass tbr check */
289 	} else {
290 		/* update token only when it is negative */
291 		if (tbr->tbr_token <= 0) {
292 			now = read_machclk();
293 			interval = now - tbr->tbr_last;
294 			if (interval >= tbr->tbr_filluptime)
295 				tbr->tbr_token = tbr->tbr_depth;
296 			else {
297 				tbr->tbr_token += interval * tbr->tbr_rate;
298 				if (tbr->tbr_token > tbr->tbr_depth)
299 					tbr->tbr_token = tbr->tbr_depth;
300 			}
301 			tbr->tbr_last = now;
302 		}
303 		/* if token is still negative, don't allow dequeue */
304 		if (tbr->tbr_token <= 0)
305 			return (NULL);
306 	}
307 
308 	if (ALTQ_IS_ENABLED(ifq))
309 		m = (*ifq->altq_dequeue)(ifq, op);
310 	else {
311 		if (op == ALTDQ_POLL)
312 			_IF_POLL(ifq, m);
313 		else
314 			_IF_DEQUEUE(ifq, m);
315 	}
316 
317 	if (m != NULL && op == ALTDQ_REMOVE)
318 		tbr->tbr_token -= TBR_SCALE(m_pktlen(m));
319 	tbr->tbr_lastop = op;
320 	return (m);
321 }
322 
323 /*
324  * set a token bucket regulator.
325  * if the specified rate is zero, the token bucket regulator is deleted.
326  */
327 int
328 tbr_set(ifq, profile)
329 	struct ifaltq *ifq;
330 	struct tb_profile *profile;
331 {
332 	struct tb_regulator *tbr, *otbr;
333 
334 	if (tbr_dequeue_ptr == NULL)
335 		tbr_dequeue_ptr = tbr_dequeue;
336 
337 	if (machclk_freq == 0)
338 		init_machclk();
339 	if (machclk_freq == 0) {
340 		printf("tbr_set: no cpu clock available!\n");
341 		return (ENXIO);
342 	}
343 
344 	IFQ_LOCK(ifq);
345 	if (profile->rate == 0) {
346 		/* delete this tbr */
347 		if ((tbr = ifq->altq_tbr) == NULL) {
348 			IFQ_UNLOCK(ifq);
349 			return (ENOENT);
350 		}
351 		ifq->altq_tbr = NULL;
352 		free(tbr, M_DEVBUF);
353 		IFQ_UNLOCK(ifq);
354 		return (0);
355 	}
356 
357 	tbr = malloc(sizeof(struct tb_regulator), M_DEVBUF, M_NOWAIT | M_ZERO);
358 	if (tbr == NULL) {
359 		IFQ_UNLOCK(ifq);
360 		return (ENOMEM);
361 	}
362 
363 	tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq;
364 	tbr->tbr_depth = TBR_SCALE(profile->depth);
365 	if (tbr->tbr_rate > 0)
366 		tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate;
367 	else
368 		tbr->tbr_filluptime = LLONG_MAX;
369 	/*
370 	 *  The longest time between tbr_dequeue() calls will be about 1
371 	 *  system tick, as the callout that drives it is scheduled once per
372 	 *  tick.  The refill-time detection logic in tbr_dequeue() can only
373 	 *  properly detect the passage of up to LLONG_MAX machclk ticks.
374 	 *  Therefore, in order for this logic to function properly in the
375 	 *  extreme case, the maximum value of tbr_filluptime should be
376 	 *  LLONG_MAX less one system tick's worth of machclk ticks less
377 	 *  some additional slop factor (here one more system tick's worth
378 	 *  of machclk ticks).
379 	 */
380 	if (tbr->tbr_filluptime > (LLONG_MAX - 2 * machclk_per_tick))
381 		tbr->tbr_filluptime = LLONG_MAX - 2 * machclk_per_tick;
382 	tbr->tbr_token = tbr->tbr_depth;
383 	tbr->tbr_last = read_machclk();
384 	tbr->tbr_lastop = ALTDQ_REMOVE;
385 
386 	otbr = ifq->altq_tbr;
387 	ifq->altq_tbr = tbr;	/* set the new tbr */
388 
389 	if (otbr != NULL)
390 		free(otbr, M_DEVBUF);
391 	else {
392 		if (tbr_timer == 0) {
393 			CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
394 			tbr_timer = 1;
395 		}
396 	}
397 	IFQ_UNLOCK(ifq);
398 	return (0);
399 }
400 
401 /*
402  * tbr_timeout goes through the interface list, and kicks the drivers
403  * if necessary.
404  *
405  * MPSAFE
406  */
407 static void
408 tbr_timeout(arg)
409 	void *arg;
410 {
411 	VNET_ITERATOR_DECL(vnet_iter);
412 	struct ifnet *ifp;
413 	int active, s;
414 
415 	active = 0;
416 	s = splnet();
417 	IFNET_RLOCK_NOSLEEP();
418 	VNET_LIST_RLOCK_NOSLEEP();
419 	VNET_FOREACH(vnet_iter) {
420 		CURVNET_SET(vnet_iter);
421 		for (ifp = CK_STAILQ_FIRST(&V_ifnet); ifp;
422 		    ifp = CK_STAILQ_NEXT(ifp, if_link)) {
423 			/* read from if_snd unlocked */
424 			if (!TBR_IS_ENABLED(&ifp->if_snd))
425 				continue;
426 			active++;
427 			if (!IFQ_IS_EMPTY(&ifp->if_snd) &&
428 			    ifp->if_start != NULL)
429 				(*ifp->if_start)(ifp);
430 		}
431 		CURVNET_RESTORE();
432 	}
433 	VNET_LIST_RUNLOCK_NOSLEEP();
434 	IFNET_RUNLOCK_NOSLEEP();
435 	splx(s);
436 	if (active > 0)
437 		CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
438 	else
439 		tbr_timer = 0;	/* don't need tbr_timer anymore */
440 }
441 
442 /*
443  * attach a discipline to the interface.  if one already exists, it is
444  * overridden.
445  * Locking is done in the discipline specific attach functions. Basically
446  * they call back to altq_attach which takes care of the attach and locking.
447  */
448 int
449 altq_pfattach(struct pf_altq *a)
450 {
451 	int error = 0;
452 
453 	switch (a->scheduler) {
454 	case ALTQT_NONE:
455 		break;
456 #ifdef ALTQ_CBQ
457 	case ALTQT_CBQ:
458 		error = cbq_pfattach(a);
459 		break;
460 #endif
461 #ifdef ALTQ_PRIQ
462 	case ALTQT_PRIQ:
463 		error = priq_pfattach(a);
464 		break;
465 #endif
466 #ifdef ALTQ_HFSC
467 	case ALTQT_HFSC:
468 		error = hfsc_pfattach(a);
469 		break;
470 #endif
471 #ifdef ALTQ_FAIRQ
472 	case ALTQT_FAIRQ:
473 		error = fairq_pfattach(a);
474 		break;
475 #endif
476 #ifdef ALTQ_CODEL
477 	case ALTQT_CODEL:
478 		error = codel_pfattach(a);
479 		break;
480 #endif
481 	default:
482 		error = ENXIO;
483 	}
484 
485 	return (error);
486 }
487 
488 /*
489  * detach a discipline from the interface.
490  * it is possible that the discipline was already overridden by another
491  * discipline.
492  */
493 int
494 altq_pfdetach(struct pf_altq *a)
495 {
496 	struct ifnet *ifp;
497 	int s, error = 0;
498 
499 	if ((ifp = ifunit(a->ifname)) == NULL)
500 		return (EINVAL);
501 
502 	/* if this discipline is no longer referenced, just return */
503 	/* read unlocked from if_snd */
504 	if (a->altq_disc == NULL || a->altq_disc != ifp->if_snd.altq_disc)
505 		return (0);
506 
507 	s = splnet();
508 	/* read unlocked from if_snd, _disable and _detach take care */
509 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
510 		error = altq_disable(&ifp->if_snd);
511 	if (error == 0)
512 		error = altq_detach(&ifp->if_snd);
513 	splx(s);
514 
515 	return (error);
516 }
517 
518 /*
519  * add a discipline or a queue
520  * Locking is done in the discipline specific functions with regards to
521  * malloc with WAITOK, also it is not yet clear which lock to use.
522  */
523 int
524 altq_add(struct pf_altq *a)
525 {
526 	int error = 0;
527 
528 	if (a->qname[0] != 0)
529 		return (altq_add_queue(a));
530 
531 	if (machclk_freq == 0)
532 		init_machclk();
533 	if (machclk_freq == 0)
534 		panic("altq_add: no cpu clock");
535 
536 	switch (a->scheduler) {
537 #ifdef ALTQ_CBQ
538 	case ALTQT_CBQ:
539 		error = cbq_add_altq(a);
540 		break;
541 #endif
542 #ifdef ALTQ_PRIQ
543 	case ALTQT_PRIQ:
544 		error = priq_add_altq(a);
545 		break;
546 #endif
547 #ifdef ALTQ_HFSC
548 	case ALTQT_HFSC:
549 		error = hfsc_add_altq(a);
550 		break;
551 #endif
552 #ifdef ALTQ_FAIRQ
553         case ALTQT_FAIRQ:
554                 error = fairq_add_altq(a);
555                 break;
556 #endif
557 #ifdef ALTQ_CODEL
558 	case ALTQT_CODEL:
559 		error = codel_add_altq(a);
560 		break;
561 #endif
562 	default:
563 		error = ENXIO;
564 	}
565 
566 	return (error);
567 }
568 
569 /*
570  * remove a discipline or a queue
571  * It is yet unclear what lock to use to protect this operation, the
572  * discipline specific functions will determine and grab it
573  */
574 int
575 altq_remove(struct pf_altq *a)
576 {
577 	int error = 0;
578 
579 	if (a->qname[0] != 0)
580 		return (altq_remove_queue(a));
581 
582 	switch (a->scheduler) {
583 #ifdef ALTQ_CBQ
584 	case ALTQT_CBQ:
585 		error = cbq_remove_altq(a);
586 		break;
587 #endif
588 #ifdef ALTQ_PRIQ
589 	case ALTQT_PRIQ:
590 		error = priq_remove_altq(a);
591 		break;
592 #endif
593 #ifdef ALTQ_HFSC
594 	case ALTQT_HFSC:
595 		error = hfsc_remove_altq(a);
596 		break;
597 #endif
598 #ifdef ALTQ_FAIRQ
599         case ALTQT_FAIRQ:
600                 error = fairq_remove_altq(a);
601                 break;
602 #endif
603 #ifdef ALTQ_CODEL
604 	case ALTQT_CODEL:
605 		error = codel_remove_altq(a);
606 		break;
607 #endif
608 	default:
609 		error = ENXIO;
610 	}
611 
612 	return (error);
613 }
614 
615 /*
616  * add a queue to the discipline
617  * It is yet unclear what lock to use to protect this operation, the
618  * discipline specific functions will determine and grab it
619  */
620 int
621 altq_add_queue(struct pf_altq *a)
622 {
623 	int error = 0;
624 
625 	switch (a->scheduler) {
626 #ifdef ALTQ_CBQ
627 	case ALTQT_CBQ:
628 		error = cbq_add_queue(a);
629 		break;
630 #endif
631 #ifdef ALTQ_PRIQ
632 	case ALTQT_PRIQ:
633 		error = priq_add_queue(a);
634 		break;
635 #endif
636 #ifdef ALTQ_HFSC
637 	case ALTQT_HFSC:
638 		error = hfsc_add_queue(a);
639 		break;
640 #endif
641 #ifdef ALTQ_FAIRQ
642         case ALTQT_FAIRQ:
643                 error = fairq_add_queue(a);
644                 break;
645 #endif
646 	default:
647 		error = ENXIO;
648 	}
649 
650 	return (error);
651 }
652 
653 /*
654  * remove a queue from the discipline
655  * It is yet unclear what lock to use to protect this operation, the
656  * discipline specific functions will determine and grab it
657  */
658 int
659 altq_remove_queue(struct pf_altq *a)
660 {
661 	int error = 0;
662 
663 	switch (a->scheduler) {
664 #ifdef ALTQ_CBQ
665 	case ALTQT_CBQ:
666 		error = cbq_remove_queue(a);
667 		break;
668 #endif
669 #ifdef ALTQ_PRIQ
670 	case ALTQT_PRIQ:
671 		error = priq_remove_queue(a);
672 		break;
673 #endif
674 #ifdef ALTQ_HFSC
675 	case ALTQT_HFSC:
676 		error = hfsc_remove_queue(a);
677 		break;
678 #endif
679 #ifdef ALTQ_FAIRQ
680         case ALTQT_FAIRQ:
681                 error = fairq_remove_queue(a);
682                 break;
683 #endif
684 	default:
685 		error = ENXIO;
686 	}
687 
688 	return (error);
689 }
690 
691 /*
692  * get queue statistics
693  * Locking is done in the discipline specific functions with regards to
694  * copyout operations, also it is not yet clear which lock to use.
695  */
696 int
697 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version)
698 {
699 	int error = 0;
700 
701 	switch (a->scheduler) {
702 #ifdef ALTQ_CBQ
703 	case ALTQT_CBQ:
704 		error = cbq_getqstats(a, ubuf, nbytes, version);
705 		break;
706 #endif
707 #ifdef ALTQ_PRIQ
708 	case ALTQT_PRIQ:
709 		error = priq_getqstats(a, ubuf, nbytes, version);
710 		break;
711 #endif
712 #ifdef ALTQ_HFSC
713 	case ALTQT_HFSC:
714 		error = hfsc_getqstats(a, ubuf, nbytes, version);
715 		break;
716 #endif
717 #ifdef ALTQ_FAIRQ
718         case ALTQT_FAIRQ:
719                 error = fairq_getqstats(a, ubuf, nbytes, version);
720                 break;
721 #endif
722 #ifdef ALTQ_CODEL
723 	case ALTQT_CODEL:
724 		error = codel_getqstats(a, ubuf, nbytes, version);
725 		break;
726 #endif
727 	default:
728 		error = ENXIO;
729 	}
730 
731 	return (error);
732 }
733 
734 /*
735  * read and write diffserv field in IPv4 or IPv6 header
736  */
737 u_int8_t
738 read_dsfield(m, pktattr)
739 	struct mbuf *m;
740 	struct altq_pktattr *pktattr;
741 {
742 	struct mbuf *m0;
743 	u_int8_t ds_field = 0;
744 
745 	if (pktattr == NULL ||
746 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
747 		return ((u_int8_t)0);
748 
749 	/* verify that pattr_hdr is within the mbuf data */
750 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
751 		if ((pktattr->pattr_hdr >= m0->m_data) &&
752 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
753 			break;
754 	if (m0 == NULL) {
755 		/* ick, pattr_hdr is stale */
756 		pktattr->pattr_af = AF_UNSPEC;
757 #ifdef ALTQ_DEBUG
758 		printf("read_dsfield: can't locate header!\n");
759 #endif
760 		return ((u_int8_t)0);
761 	}
762 
763 	if (pktattr->pattr_af == AF_INET) {
764 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
765 
766 		if (ip->ip_v != 4)
767 			return ((u_int8_t)0);	/* version mismatch! */
768 		ds_field = ip->ip_tos;
769 	}
770 #ifdef INET6
771 	else if (pktattr->pattr_af == AF_INET6) {
772 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
773 		u_int32_t flowlabel;
774 
775 		flowlabel = ntohl(ip6->ip6_flow);
776 		if ((flowlabel >> 28) != 6)
777 			return ((u_int8_t)0);	/* version mismatch! */
778 		ds_field = (flowlabel >> 20) & 0xff;
779 	}
780 #endif
781 	return (ds_field);
782 }
783 
784 void
785 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, u_int8_t dsfield)
786 {
787 	struct mbuf *m0;
788 
789 	if (pktattr == NULL ||
790 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
791 		return;
792 
793 	/* verify that pattr_hdr is within the mbuf data */
794 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
795 		if ((pktattr->pattr_hdr >= m0->m_data) &&
796 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
797 			break;
798 	if (m0 == NULL) {
799 		/* ick, pattr_hdr is stale */
800 		pktattr->pattr_af = AF_UNSPEC;
801 #ifdef ALTQ_DEBUG
802 		printf("write_dsfield: can't locate header!\n");
803 #endif
804 		return;
805 	}
806 
807 	if (pktattr->pattr_af == AF_INET) {
808 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
809 		u_int8_t old;
810 		int32_t sum;
811 
812 		if (ip->ip_v != 4)
813 			return;		/* version mismatch! */
814 		old = ip->ip_tos;
815 		dsfield |= old & 3;	/* leave CU bits */
816 		if (old == dsfield)
817 			return;
818 		ip->ip_tos = dsfield;
819 		/*
820 		 * update checksum (from RFC1624)
821 		 *	   HC' = ~(~HC + ~m + m')
822 		 */
823 		sum = ~ntohs(ip->ip_sum) & 0xffff;
824 		sum += 0xff00 + (~old & 0xff) + dsfield;
825 		sum = (sum >> 16) + (sum & 0xffff);
826 		sum += (sum >> 16);  /* add carry */
827 
828 		ip->ip_sum = htons(~sum & 0xffff);
829 	}
830 #ifdef INET6
831 	else if (pktattr->pattr_af == AF_INET6) {
832 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
833 		u_int32_t flowlabel;
834 
835 		flowlabel = ntohl(ip6->ip6_flow);
836 		if ((flowlabel >> 28) != 6)
837 			return;		/* version mismatch! */
838 		flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20);
839 		ip6->ip6_flow = htonl(flowlabel);
840 	}
841 #endif
842 	return;
843 }
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 #if (__FreeBSD_version >= 700035)
865 /* Update TSC freq with the value indicated by the caller. */
866 static void
867 tsc_freq_changed(void *arg, const struct cf_level *level, int status)
868 {
869 	/* If there was an error during the transition, don't do anything. */
870 	if (status != 0)
871 		return;
872 
873 #if (__FreeBSD_version >= 701102) && (defined(__amd64__) || defined(__i386__))
874 	/* If TSC is P-state invariant, don't do anything. */
875 	if (tsc_is_invariant)
876 		return;
877 #endif
878 
879 	/* Total setting for this level gives the new frequency in MHz. */
880 	init_machclk();
881 }
882 EVENTHANDLER_DEFINE(cpufreq_post_change, tsc_freq_changed, NULL,
883     EVENTHANDLER_PRI_LAST);
884 #endif /* __FreeBSD_version >= 700035 */
885 
886 static void
887 init_machclk_setup(void)
888 {
889 #if (__FreeBSD_version >= 600000)
890 	callout_init(&tbr_callout, 0);
891 #endif
892 
893 	machclk_usepcc = 1;
894 
895 #if (!defined(__amd64__) && !defined(__i386__)) || defined(ALTQ_NOPCC)
896 	machclk_usepcc = 0;
897 #endif
898 #if defined(__FreeBSD__) && defined(SMP)
899 	machclk_usepcc = 0;
900 #endif
901 #if defined(__NetBSD__) && defined(MULTIPROCESSOR)
902 	machclk_usepcc = 0;
903 #endif
904 #if defined(__amd64__) || defined(__i386__)
905 	/* check if TSC is available */
906 	if ((cpu_feature & CPUID_TSC) == 0 ||
907 	    atomic_load_acq_64(&tsc_freq) == 0)
908 		machclk_usepcc = 0;
909 #endif
910 }
911 
912 void
913 init_machclk(void)
914 {
915 	static int called;
916 
917 	/* Call one-time initialization function. */
918 	if (!called) {
919 		init_machclk_setup();
920 		called = 1;
921 	}
922 
923 	if (machclk_usepcc == 0) {
924 		/* emulate 256MHz using microtime() */
925 		machclk_freq = 1000000 << MACHCLK_SHIFT;
926 		machclk_per_tick = machclk_freq / hz;
927 #ifdef ALTQ_DEBUG
928 		printf("altq: emulate %uHz cpu clock\n", machclk_freq);
929 #endif
930 		return;
931 	}
932 
933 	/*
934 	 * if the clock frequency (of Pentium TSC or Alpha PCC) is
935 	 * accessible, just use it.
936 	 */
937 #if defined(__amd64__) || defined(__i386__)
938 	machclk_freq = atomic_load_acq_64(&tsc_freq);
939 #endif
940 
941 	/*
942 	 * if we don't know the clock frequency, measure it.
943 	 */
944 	if (machclk_freq == 0) {
945 		static int	wait;
946 		struct timeval	tv_start, tv_end;
947 		u_int64_t	start, end, diff;
948 		int		timo;
949 
950 		microtime(&tv_start);
951 		start = read_machclk();
952 		timo = hz;	/* 1 sec */
953 		(void)tsleep(&wait, PWAIT | PCATCH, "init_machclk", timo);
954 		microtime(&tv_end);
955 		end = read_machclk();
956 		diff = (u_int64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000
957 		    + tv_end.tv_usec - tv_start.tv_usec;
958 		if (diff != 0)
959 			machclk_freq = (u_int)((end - start) * 1000000 / diff);
960 	}
961 
962 	machclk_per_tick = machclk_freq / hz;
963 
964 #ifdef ALTQ_DEBUG
965 	printf("altq: CPU clock: %uHz\n", machclk_freq);
966 #endif
967 }
968 
969 #if defined(__OpenBSD__) && defined(__i386__)
970 static __inline u_int64_t
971 rdtsc(void)
972 {
973 	u_int64_t rv;
974 	__asm __volatile(".byte 0x0f, 0x31" : "=A" (rv));
975 	return (rv);
976 }
977 #endif /* __OpenBSD__ && __i386__ */
978 
979 u_int64_t
980 read_machclk(void)
981 {
982 	u_int64_t val;
983 
984 	if (machclk_usepcc) {
985 #if defined(__amd64__) || defined(__i386__)
986 		val = rdtsc();
987 #else
988 		panic("read_machclk");
989 #endif
990 	} else {
991 		struct timeval tv, boottime;
992 
993 		microtime(&tv);
994 		getboottime(&boottime);
995 		val = (((u_int64_t)(tv.tv_sec - boottime.tv_sec) * 1000000
996 		    + tv.tv_usec) << MACHCLK_SHIFT);
997 	}
998 	return (val);
999 }
1000 
1001 #ifdef ALTQ3_CLFIER_COMPAT
1002 
1003 #ifndef IPPROTO_ESP
1004 #define	IPPROTO_ESP	50		/* encapsulating security payload */
1005 #endif
1006 #ifndef IPPROTO_AH
1007 #define	IPPROTO_AH	51		/* authentication header */
1008 #endif
1009 
1010 /*
1011  * extract flow information from a given packet.
1012  * filt_mask shows flowinfo fields required.
1013  * we assume the ip header is in one mbuf, and addresses and ports are
1014  * in network byte order.
1015  */
1016 int
1017 altq_extractflow(m, af, flow, filt_bmask)
1018 	struct mbuf *m;
1019 	int af;
1020 	struct flowinfo *flow;
1021 	u_int32_t	filt_bmask;
1022 {
1023 
1024 	switch (af) {
1025 	case PF_INET: {
1026 		struct flowinfo_in *fin;
1027 		struct ip *ip;
1028 
1029 		ip = mtod(m, struct ip *);
1030 
1031 		if (ip->ip_v != 4)
1032 			break;
1033 
1034 		fin = (struct flowinfo_in *)flow;
1035 		fin->fi_len = sizeof(struct flowinfo_in);
1036 		fin->fi_family = AF_INET;
1037 
1038 		fin->fi_proto = ip->ip_p;
1039 		fin->fi_tos = ip->ip_tos;
1040 
1041 		fin->fi_src.s_addr = ip->ip_src.s_addr;
1042 		fin->fi_dst.s_addr = ip->ip_dst.s_addr;
1043 
1044 		if (filt_bmask & FIMB4_PORTS)
1045 			/* if port info is required, extract port numbers */
1046 			extract_ports4(m, ip, fin);
1047 		else {
1048 			fin->fi_sport = 0;
1049 			fin->fi_dport = 0;
1050 			fin->fi_gpi = 0;
1051 		}
1052 		return (1);
1053 	}
1054 
1055 #ifdef INET6
1056 	case PF_INET6: {
1057 		struct flowinfo_in6 *fin6;
1058 		struct ip6_hdr *ip6;
1059 
1060 		ip6 = mtod(m, struct ip6_hdr *);
1061 		/* should we check the ip version? */
1062 
1063 		fin6 = (struct flowinfo_in6 *)flow;
1064 		fin6->fi6_len = sizeof(struct flowinfo_in6);
1065 		fin6->fi6_family = AF_INET6;
1066 
1067 		fin6->fi6_proto = ip6->ip6_nxt;
1068 		fin6->fi6_tclass   = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
1069 
1070 		fin6->fi6_flowlabel = ip6->ip6_flow & htonl(0x000fffff);
1071 		fin6->fi6_src = ip6->ip6_src;
1072 		fin6->fi6_dst = ip6->ip6_dst;
1073 
1074 		if ((filt_bmask & FIMB6_PORTS) ||
1075 		    ((filt_bmask & FIMB6_PROTO)
1076 		     && ip6->ip6_nxt > IPPROTO_IPV6))
1077 			/*
1078 			 * if port info is required, or proto is required
1079 			 * but there are option headers, extract port
1080 			 * and protocol numbers.
1081 			 */
1082 			extract_ports6(m, ip6, fin6);
1083 		else {
1084 			fin6->fi6_sport = 0;
1085 			fin6->fi6_dport = 0;
1086 			fin6->fi6_gpi = 0;
1087 		}
1088 		return (1);
1089 	}
1090 #endif /* INET6 */
1091 
1092 	default:
1093 		break;
1094 	}
1095 
1096 	/* failed */
1097 	flow->fi_len = sizeof(struct flowinfo);
1098 	flow->fi_family = AF_UNSPEC;
1099 	return (0);
1100 }
1101 
1102 /*
1103  * helper routine to extract port numbers
1104  */
1105 /* structure for ipsec and ipv6 option header template */
1106 struct _opt6 {
1107 	u_int8_t	opt6_nxt;	/* next header */
1108 	u_int8_t	opt6_hlen;	/* header extension length */
1109 	u_int16_t	_pad;
1110 	u_int32_t	ah_spi;		/* security parameter index
1111 					   for authentication header */
1112 };
1113 
1114 /*
1115  * extract port numbers from a ipv4 packet.
1116  */
1117 static int
1118 extract_ports4(m, ip, fin)
1119 	struct mbuf *m;
1120 	struct ip *ip;
1121 	struct flowinfo_in *fin;
1122 {
1123 	struct mbuf *m0;
1124 	u_short ip_off;
1125 	u_int8_t proto;
1126 	int 	off;
1127 
1128 	fin->fi_sport = 0;
1129 	fin->fi_dport = 0;
1130 	fin->fi_gpi = 0;
1131 
1132 	ip_off = ntohs(ip->ip_off);
1133 	/* if it is a fragment, try cached fragment info */
1134 	if (ip_off & IP_OFFMASK) {
1135 		ip4f_lookup(ip, fin);
1136 		return (1);
1137 	}
1138 
1139 	/* locate the mbuf containing the protocol header */
1140 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
1141 		if (((caddr_t)ip >= m0->m_data) &&
1142 		    ((caddr_t)ip < m0->m_data + m0->m_len))
1143 			break;
1144 	if (m0 == NULL) {
1145 #ifdef ALTQ_DEBUG
1146 		printf("extract_ports4: can't locate header! ip=%p\n", ip);
1147 #endif
1148 		return (0);
1149 	}
1150 	off = ((caddr_t)ip - m0->m_data) + (ip->ip_hl << 2);
1151 	proto = ip->ip_p;
1152 
1153 #ifdef ALTQ_IPSEC
1154  again:
1155 #endif
1156 	while (off >= m0->m_len) {
1157 		off -= m0->m_len;
1158 		m0 = m0->m_next;
1159 		if (m0 == NULL)
1160 			return (0);  /* bogus ip_hl! */
1161 	}
1162 	if (m0->m_len < off + 4)
1163 		return (0);
1164 
1165 	switch (proto) {
1166 	case IPPROTO_TCP:
1167 	case IPPROTO_UDP: {
1168 		struct udphdr *udp;
1169 
1170 		udp = (struct udphdr *)(mtod(m0, caddr_t) + off);
1171 		fin->fi_sport = udp->uh_sport;
1172 		fin->fi_dport = udp->uh_dport;
1173 		fin->fi_proto = proto;
1174 		}
1175 		break;
1176 
1177 #ifdef ALTQ_IPSEC
1178 	case IPPROTO_ESP:
1179 		if (fin->fi_gpi == 0){
1180 			u_int32_t *gpi;
1181 
1182 			gpi = (u_int32_t *)(mtod(m0, caddr_t) + off);
1183 			fin->fi_gpi   = *gpi;
1184 		}
1185 		fin->fi_proto = proto;
1186 		break;
1187 
1188 	case IPPROTO_AH: {
1189 			/* get next header and header length */
1190 			struct _opt6 *opt6;
1191 
1192 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1193 			proto = opt6->opt6_nxt;
1194 			off += 8 + (opt6->opt6_hlen * 4);
1195 			if (fin->fi_gpi == 0 && m0->m_len >= off + 8)
1196 				fin->fi_gpi = opt6->ah_spi;
1197 		}
1198 		/* goto the next header */
1199 		goto again;
1200 #endif  /* ALTQ_IPSEC */
1201 
1202 	default:
1203 		fin->fi_proto = proto;
1204 		return (0);
1205 	}
1206 
1207 	/* if this is a first fragment, cache it. */
1208 	if (ip_off & IP_MF)
1209 		ip4f_cache(ip, fin);
1210 
1211 	return (1);
1212 }
1213 
1214 #ifdef INET6
1215 static int
1216 extract_ports6(m, ip6, fin6)
1217 	struct mbuf *m;
1218 	struct ip6_hdr *ip6;
1219 	struct flowinfo_in6 *fin6;
1220 {
1221 	struct mbuf *m0;
1222 	int	off;
1223 	u_int8_t proto;
1224 
1225 	fin6->fi6_gpi   = 0;
1226 	fin6->fi6_sport = 0;
1227 	fin6->fi6_dport = 0;
1228 
1229 	/* locate the mbuf containing the protocol header */
1230 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
1231 		if (((caddr_t)ip6 >= m0->m_data) &&
1232 		    ((caddr_t)ip6 < m0->m_data + m0->m_len))
1233 			break;
1234 	if (m0 == NULL) {
1235 #ifdef ALTQ_DEBUG
1236 		printf("extract_ports6: can't locate header! ip6=%p\n", ip6);
1237 #endif
1238 		return (0);
1239 	}
1240 	off = ((caddr_t)ip6 - m0->m_data) + sizeof(struct ip6_hdr);
1241 
1242 	proto = ip6->ip6_nxt;
1243 	do {
1244 		while (off >= m0->m_len) {
1245 			off -= m0->m_len;
1246 			m0 = m0->m_next;
1247 			if (m0 == NULL)
1248 				return (0);
1249 		}
1250 		if (m0->m_len < off + 4)
1251 			return (0);
1252 
1253 		switch (proto) {
1254 		case IPPROTO_TCP:
1255 		case IPPROTO_UDP: {
1256 			struct udphdr *udp;
1257 
1258 			udp = (struct udphdr *)(mtod(m0, caddr_t) + off);
1259 			fin6->fi6_sport = udp->uh_sport;
1260 			fin6->fi6_dport = udp->uh_dport;
1261 			fin6->fi6_proto = proto;
1262 			}
1263 			return (1);
1264 
1265 		case IPPROTO_ESP:
1266 			if (fin6->fi6_gpi == 0) {
1267 				u_int32_t *gpi;
1268 
1269 				gpi = (u_int32_t *)(mtod(m0, caddr_t) + off);
1270 				fin6->fi6_gpi   = *gpi;
1271 			}
1272 			fin6->fi6_proto = proto;
1273 			return (1);
1274 
1275 		case IPPROTO_AH: {
1276 			/* get next header and header length */
1277 			struct _opt6 *opt6;
1278 
1279 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1280 			if (fin6->fi6_gpi == 0 && m0->m_len >= off + 8)
1281 				fin6->fi6_gpi = opt6->ah_spi;
1282 			proto = opt6->opt6_nxt;
1283 			off += 8 + (opt6->opt6_hlen * 4);
1284 			/* goto the next header */
1285 			break;
1286 			}
1287 
1288 		case IPPROTO_HOPOPTS:
1289 		case IPPROTO_ROUTING:
1290 		case IPPROTO_DSTOPTS: {
1291 			/* get next header and header length */
1292 			struct _opt6 *opt6;
1293 
1294 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1295 			proto = opt6->opt6_nxt;
1296 			off += (opt6->opt6_hlen + 1) * 8;
1297 			/* goto the next header */
1298 			break;
1299 			}
1300 
1301 		case IPPROTO_FRAGMENT:
1302 			/* ipv6 fragmentations are not supported yet */
1303 		default:
1304 			fin6->fi6_proto = proto;
1305 			return (0);
1306 		}
1307 	} while (1);
1308 	/*NOTREACHED*/
1309 }
1310 #endif /* INET6 */
1311 
1312 /*
1313  * altq common classifier
1314  */
1315 int
1316 acc_add_filter(classifier, filter, class, phandle)
1317 	struct acc_classifier *classifier;
1318 	struct flow_filter *filter;
1319 	void	*class;
1320 	u_long	*phandle;
1321 {
1322 	struct acc_filter *afp, *prev, *tmp;
1323 	int	i, s;
1324 
1325 #ifdef INET6
1326 	if (filter->ff_flow.fi_family != AF_INET &&
1327 	    filter->ff_flow.fi_family != AF_INET6)
1328 		return (EINVAL);
1329 #else
1330 	if (filter->ff_flow.fi_family != AF_INET)
1331 		return (EINVAL);
1332 #endif
1333 
1334 	afp = malloc(sizeof(struct acc_filter),
1335 	       M_DEVBUF, M_WAITOK);
1336 	if (afp == NULL)
1337 		return (ENOMEM);
1338 	bzero(afp, sizeof(struct acc_filter));
1339 
1340 	afp->f_filter = *filter;
1341 	afp->f_class = class;
1342 
1343 	i = ACC_WILDCARD_INDEX;
1344 	if (filter->ff_flow.fi_family == AF_INET) {
1345 		struct flow_filter *filter4 = &afp->f_filter;
1346 
1347 		/*
1348 		 * if address is 0, it's a wildcard.  if address mask
1349 		 * isn't set, use full mask.
1350 		 */
1351 		if (filter4->ff_flow.fi_dst.s_addr == 0)
1352 			filter4->ff_mask.mask_dst.s_addr = 0;
1353 		else if (filter4->ff_mask.mask_dst.s_addr == 0)
1354 			filter4->ff_mask.mask_dst.s_addr = 0xffffffff;
1355 		if (filter4->ff_flow.fi_src.s_addr == 0)
1356 			filter4->ff_mask.mask_src.s_addr = 0;
1357 		else if (filter4->ff_mask.mask_src.s_addr == 0)
1358 			filter4->ff_mask.mask_src.s_addr = 0xffffffff;
1359 
1360 		/* clear extra bits in addresses  */
1361 		   filter4->ff_flow.fi_dst.s_addr &=
1362 		       filter4->ff_mask.mask_dst.s_addr;
1363 		   filter4->ff_flow.fi_src.s_addr &=
1364 		       filter4->ff_mask.mask_src.s_addr;
1365 
1366 		/*
1367 		 * if dst address is a wildcard, use hash-entry
1368 		 * ACC_WILDCARD_INDEX.
1369 		 */
1370 		if (filter4->ff_mask.mask_dst.s_addr != 0xffffffff)
1371 			i = ACC_WILDCARD_INDEX;
1372 		else
1373 			i = ACC_GET_HASH_INDEX(filter4->ff_flow.fi_dst.s_addr);
1374 	}
1375 #ifdef INET6
1376 	else if (filter->ff_flow.fi_family == AF_INET6) {
1377 		struct flow_filter6 *filter6 =
1378 			(struct flow_filter6 *)&afp->f_filter;
1379 #ifndef IN6MASK0 /* taken from kame ipv6 */
1380 #define	IN6MASK0	{{{ 0, 0, 0, 0 }}}
1381 #define	IN6MASK128	{{{ 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff }}}
1382 		const struct in6_addr in6mask0 = IN6MASK0;
1383 		const struct in6_addr in6mask128 = IN6MASK128;
1384 #endif
1385 
1386 		if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_dst))
1387 			filter6->ff_mask6.mask6_dst = in6mask0;
1388 		else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_dst))
1389 			filter6->ff_mask6.mask6_dst = in6mask128;
1390 		if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_src))
1391 			filter6->ff_mask6.mask6_src = in6mask0;
1392 		else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_src))
1393 			filter6->ff_mask6.mask6_src = in6mask128;
1394 
1395 		/* clear extra bits in addresses  */
1396 		for (i = 0; i < 16; i++)
1397 			filter6->ff_flow6.fi6_dst.s6_addr[i] &=
1398 			    filter6->ff_mask6.mask6_dst.s6_addr[i];
1399 		for (i = 0; i < 16; i++)
1400 			filter6->ff_flow6.fi6_src.s6_addr[i] &=
1401 			    filter6->ff_mask6.mask6_src.s6_addr[i];
1402 
1403 		if (filter6->ff_flow6.fi6_flowlabel == 0)
1404 			i = ACC_WILDCARD_INDEX;
1405 		else
1406 			i = ACC_GET_HASH_INDEX(filter6->ff_flow6.fi6_flowlabel);
1407 	}
1408 #endif /* INET6 */
1409 
1410 	afp->f_handle = get_filt_handle(classifier, i);
1411 
1412 	/* update filter bitmask */
1413 	afp->f_fbmask = filt2fibmask(filter);
1414 	classifier->acc_fbmask |= afp->f_fbmask;
1415 
1416 	/*
1417 	 * add this filter to the filter list.
1418 	 * filters are ordered from the highest rule number.
1419 	 */
1420 	s = splnet();
1421 	prev = NULL;
1422 	LIST_FOREACH(tmp, &classifier->acc_filters[i], f_chain) {
1423 		if (tmp->f_filter.ff_ruleno > afp->f_filter.ff_ruleno)
1424 			prev = tmp;
1425 		else
1426 			break;
1427 	}
1428 	if (prev == NULL)
1429 		LIST_INSERT_HEAD(&classifier->acc_filters[i], afp, f_chain);
1430 	else
1431 		LIST_INSERT_AFTER(prev, afp, f_chain);
1432 	splx(s);
1433 
1434 	*phandle = afp->f_handle;
1435 	return (0);
1436 }
1437 
1438 int
1439 acc_delete_filter(classifier, handle)
1440 	struct acc_classifier *classifier;
1441 	u_long handle;
1442 {
1443 	struct acc_filter *afp;
1444 	int	s;
1445 
1446 	if ((afp = filth_to_filtp(classifier, handle)) == NULL)
1447 		return (EINVAL);
1448 
1449 	s = splnet();
1450 	LIST_REMOVE(afp, f_chain);
1451 	splx(s);
1452 
1453 	free(afp, M_DEVBUF);
1454 
1455 	/* todo: update filt_bmask */
1456 
1457 	return (0);
1458 }
1459 
1460 /*
1461  * delete filters referencing to the specified class.
1462  * if the all flag is not 0, delete all the filters.
1463  */
1464 int
1465 acc_discard_filters(classifier, class, all)
1466 	struct acc_classifier *classifier;
1467 	void	*class;
1468 	int	all;
1469 {
1470 	struct acc_filter *afp;
1471 	int	i, s;
1472 
1473 	s = splnet();
1474 	for (i = 0; i < ACC_FILTER_TABLESIZE; i++) {
1475 		do {
1476 			LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1477 				if (all || afp->f_class == class) {
1478 					LIST_REMOVE(afp, f_chain);
1479 					free(afp, M_DEVBUF);
1480 					/* start again from the head */
1481 					break;
1482 				}
1483 		} while (afp != NULL);
1484 	}
1485 	splx(s);
1486 
1487 	if (all)
1488 		classifier->acc_fbmask = 0;
1489 
1490 	return (0);
1491 }
1492 
1493 void *
1494 acc_classify(clfier, m, af)
1495 	void *clfier;
1496 	struct mbuf *m;
1497 	int af;
1498 {
1499 	struct acc_classifier *classifier;
1500 	struct flowinfo flow;
1501 	struct acc_filter *afp;
1502 	int	i;
1503 
1504 	classifier = (struct acc_classifier *)clfier;
1505 	altq_extractflow(m, af, &flow, classifier->acc_fbmask);
1506 
1507 	if (flow.fi_family == AF_INET) {
1508 		struct flowinfo_in *fp = (struct flowinfo_in *)&flow;
1509 
1510 		if ((classifier->acc_fbmask & FIMB4_ALL) == FIMB4_TOS) {
1511 			/* only tos is used */
1512 			LIST_FOREACH(afp,
1513 				 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1514 				 f_chain)
1515 				if (apply_tosfilter4(afp->f_fbmask,
1516 						     &afp->f_filter, fp))
1517 					/* filter matched */
1518 					return (afp->f_class);
1519 		} else if ((classifier->acc_fbmask &
1520 			(~(FIMB4_PROTO|FIMB4_SPORT|FIMB4_DPORT) & FIMB4_ALL))
1521 		    == 0) {
1522 			/* only proto and ports are used */
1523 			LIST_FOREACH(afp,
1524 				 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1525 				 f_chain)
1526 				if (apply_ppfilter4(afp->f_fbmask,
1527 						    &afp->f_filter, fp))
1528 					/* filter matched */
1529 					return (afp->f_class);
1530 		} else {
1531 			/* get the filter hash entry from its dest address */
1532 			i = ACC_GET_HASH_INDEX(fp->fi_dst.s_addr);
1533 			do {
1534 				/*
1535 				 * go through this loop twice.  first for dst
1536 				 * hash, second for wildcards.
1537 				 */
1538 				LIST_FOREACH(afp, &classifier->acc_filters[i],
1539 					     f_chain)
1540 					if (apply_filter4(afp->f_fbmask,
1541 							  &afp->f_filter, fp))
1542 						/* filter matched */
1543 						return (afp->f_class);
1544 
1545 				/*
1546 				 * check again for filters with a dst addr
1547 				 * wildcard.
1548 				 * (daddr == 0 || dmask != 0xffffffff).
1549 				 */
1550 				if (i != ACC_WILDCARD_INDEX)
1551 					i = ACC_WILDCARD_INDEX;
1552 				else
1553 					break;
1554 			} while (1);
1555 		}
1556 	}
1557 #ifdef INET6
1558 	else if (flow.fi_family == AF_INET6) {
1559 		struct flowinfo_in6 *fp6 = (struct flowinfo_in6 *)&flow;
1560 
1561 		/* get the filter hash entry from its flow ID */
1562 		if (fp6->fi6_flowlabel != 0)
1563 			i = ACC_GET_HASH_INDEX(fp6->fi6_flowlabel);
1564 		else
1565 			/* flowlable can be zero */
1566 			i = ACC_WILDCARD_INDEX;
1567 
1568 		/* go through this loop twice.  first for flow hash, second
1569 		   for wildcards. */
1570 		do {
1571 			LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1572 				if (apply_filter6(afp->f_fbmask,
1573 					(struct flow_filter6 *)&afp->f_filter,
1574 					fp6))
1575 					/* filter matched */
1576 					return (afp->f_class);
1577 
1578 			/*
1579 			 * check again for filters with a wildcard.
1580 			 */
1581 			if (i != ACC_WILDCARD_INDEX)
1582 				i = ACC_WILDCARD_INDEX;
1583 			else
1584 				break;
1585 		} while (1);
1586 	}
1587 #endif /* INET6 */
1588 
1589 	/* no filter matched */
1590 	return (NULL);
1591 }
1592 
1593 static int
1594 apply_filter4(fbmask, filt, pkt)
1595 	u_int32_t	fbmask;
1596 	struct flow_filter *filt;
1597 	struct flowinfo_in *pkt;
1598 {
1599 	if (filt->ff_flow.fi_family != AF_INET)
1600 		return (0);
1601 	if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1602 		return (0);
1603 	if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1604 		return (0);
1605 	if ((fbmask & FIMB4_DADDR) &&
1606 	    filt->ff_flow.fi_dst.s_addr !=
1607 	    (pkt->fi_dst.s_addr & filt->ff_mask.mask_dst.s_addr))
1608 		return (0);
1609 	if ((fbmask & FIMB4_SADDR) &&
1610 	    filt->ff_flow.fi_src.s_addr !=
1611 	    (pkt->fi_src.s_addr & filt->ff_mask.mask_src.s_addr))
1612 		return (0);
1613 	if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1614 		return (0);
1615 	if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1616 	    (pkt->fi_tos & filt->ff_mask.mask_tos))
1617 		return (0);
1618 	if ((fbmask & FIMB4_GPI) && filt->ff_flow.fi_gpi != (pkt->fi_gpi))
1619 		return (0);
1620 	/* match */
1621 	return (1);
1622 }
1623 
1624 /*
1625  * filter matching function optimized for a common case that checks
1626  * only protocol and port numbers
1627  */
1628 static int
1629 apply_ppfilter4(fbmask, filt, pkt)
1630 	u_int32_t	fbmask;
1631 	struct flow_filter *filt;
1632 	struct flowinfo_in *pkt;
1633 {
1634 	if (filt->ff_flow.fi_family != AF_INET)
1635 		return (0);
1636 	if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1637 		return (0);
1638 	if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1639 		return (0);
1640 	if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1641 		return (0);
1642 	/* match */
1643 	return (1);
1644 }
1645 
1646 /*
1647  * filter matching function only for tos field.
1648  */
1649 static int
1650 apply_tosfilter4(fbmask, filt, pkt)
1651 	u_int32_t	fbmask;
1652 	struct flow_filter *filt;
1653 	struct flowinfo_in *pkt;
1654 {
1655 	if (filt->ff_flow.fi_family != AF_INET)
1656 		return (0);
1657 	if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1658 	    (pkt->fi_tos & filt->ff_mask.mask_tos))
1659 		return (0);
1660 	/* match */
1661 	return (1);
1662 }
1663 
1664 #ifdef INET6
1665 static int
1666 apply_filter6(fbmask, filt, pkt)
1667 	u_int32_t	fbmask;
1668 	struct flow_filter6 *filt;
1669 	struct flowinfo_in6 *pkt;
1670 {
1671 	int i;
1672 
1673 	if (filt->ff_flow6.fi6_family != AF_INET6)
1674 		return (0);
1675 	if ((fbmask & FIMB6_FLABEL) &&
1676 	    filt->ff_flow6.fi6_flowlabel != pkt->fi6_flowlabel)
1677 		return (0);
1678 	if ((fbmask & FIMB6_PROTO) &&
1679 	    filt->ff_flow6.fi6_proto != pkt->fi6_proto)
1680 		return (0);
1681 	if ((fbmask & FIMB6_SPORT) &&
1682 	    filt->ff_flow6.fi6_sport != pkt->fi6_sport)
1683 		return (0);
1684 	if ((fbmask & FIMB6_DPORT) &&
1685 	    filt->ff_flow6.fi6_dport != pkt->fi6_dport)
1686 		return (0);
1687 	if (fbmask & FIMB6_SADDR) {
1688 		for (i = 0; i < 4; i++)
1689 			if (filt->ff_flow6.fi6_src.s6_addr32[i] !=
1690 			    (pkt->fi6_src.s6_addr32[i] &
1691 			     filt->ff_mask6.mask6_src.s6_addr32[i]))
1692 				return (0);
1693 	}
1694 	if (fbmask & FIMB6_DADDR) {
1695 		for (i = 0; i < 4; i++)
1696 			if (filt->ff_flow6.fi6_dst.s6_addr32[i] !=
1697 			    (pkt->fi6_dst.s6_addr32[i] &
1698 			     filt->ff_mask6.mask6_dst.s6_addr32[i]))
1699 				return (0);
1700 	}
1701 	if ((fbmask & FIMB6_TCLASS) &&
1702 	    filt->ff_flow6.fi6_tclass !=
1703 	    (pkt->fi6_tclass & filt->ff_mask6.mask6_tclass))
1704 		return (0);
1705 	if ((fbmask & FIMB6_GPI) &&
1706 	    filt->ff_flow6.fi6_gpi != pkt->fi6_gpi)
1707 		return (0);
1708 	/* match */
1709 	return (1);
1710 }
1711 #endif /* INET6 */
1712 
1713 /*
1714  *  filter handle:
1715  *	bit 20-28: index to the filter hash table
1716  *	bit  0-19: unique id in the hash bucket.
1717  */
1718 static u_long
1719 get_filt_handle(classifier, i)
1720 	struct acc_classifier *classifier;
1721 	int	i;
1722 {
1723 	static u_long handle_number = 1;
1724 	u_long 	handle;
1725 	struct acc_filter *afp;
1726 
1727 	while (1) {
1728 		handle = handle_number++ & 0x000fffff;
1729 
1730 		if (LIST_EMPTY(&classifier->acc_filters[i]))
1731 			break;
1732 
1733 		LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1734 			if ((afp->f_handle & 0x000fffff) == handle)
1735 				break;
1736 		if (afp == NULL)
1737 			break;
1738 		/* this handle is already used, try again */
1739 	}
1740 
1741 	return ((i << 20) | handle);
1742 }
1743 
1744 /* convert filter handle to filter pointer */
1745 static struct acc_filter *
1746 filth_to_filtp(classifier, handle)
1747 	struct acc_classifier *classifier;
1748 	u_long handle;
1749 {
1750 	struct acc_filter *afp;
1751 	int	i;
1752 
1753 	i = ACC_GET_HINDEX(handle);
1754 
1755 	LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1756 		if (afp->f_handle == handle)
1757 			return (afp);
1758 
1759 	return (NULL);
1760 }
1761 
1762 /* create flowinfo bitmask */
1763 static u_int32_t
1764 filt2fibmask(filt)
1765 	struct flow_filter *filt;
1766 {
1767 	u_int32_t mask = 0;
1768 #ifdef INET6
1769 	struct flow_filter6 *filt6;
1770 #endif
1771 
1772 	switch (filt->ff_flow.fi_family) {
1773 	case AF_INET:
1774 		if (filt->ff_flow.fi_proto != 0)
1775 			mask |= FIMB4_PROTO;
1776 		if (filt->ff_flow.fi_tos != 0)
1777 			mask |= FIMB4_TOS;
1778 		if (filt->ff_flow.fi_dst.s_addr != 0)
1779 			mask |= FIMB4_DADDR;
1780 		if (filt->ff_flow.fi_src.s_addr != 0)
1781 			mask |= FIMB4_SADDR;
1782 		if (filt->ff_flow.fi_sport != 0)
1783 			mask |= FIMB4_SPORT;
1784 		if (filt->ff_flow.fi_dport != 0)
1785 			mask |= FIMB4_DPORT;
1786 		if (filt->ff_flow.fi_gpi != 0)
1787 			mask |= FIMB4_GPI;
1788 		break;
1789 #ifdef INET6
1790 	case AF_INET6:
1791 		filt6 = (struct flow_filter6 *)filt;
1792 
1793 		if (filt6->ff_flow6.fi6_proto != 0)
1794 			mask |= FIMB6_PROTO;
1795 		if (filt6->ff_flow6.fi6_tclass != 0)
1796 			mask |= FIMB6_TCLASS;
1797 		if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_dst))
1798 			mask |= FIMB6_DADDR;
1799 		if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_src))
1800 			mask |= FIMB6_SADDR;
1801 		if (filt6->ff_flow6.fi6_sport != 0)
1802 			mask |= FIMB6_SPORT;
1803 		if (filt6->ff_flow6.fi6_dport != 0)
1804 			mask |= FIMB6_DPORT;
1805 		if (filt6->ff_flow6.fi6_gpi != 0)
1806 			mask |= FIMB6_GPI;
1807 		if (filt6->ff_flow6.fi6_flowlabel != 0)
1808 			mask |= FIMB6_FLABEL;
1809 		break;
1810 #endif /* INET6 */
1811 	}
1812 	return (mask);
1813 }
1814 
1815 
1816 /*
1817  * helper functions to handle IPv4 fragments.
1818  * currently only in-sequence fragments are handled.
1819  *	- fragment info is cached in a LRU list.
1820  *	- when a first fragment is found, cache its flow info.
1821  *	- when a non-first fragment is found, lookup the cache.
1822  */
1823 
1824 struct ip4_frag {
1825     TAILQ_ENTRY(ip4_frag) ip4f_chain;
1826     char    ip4f_valid;
1827     u_short ip4f_id;
1828     struct flowinfo_in ip4f_info;
1829 };
1830 
1831 static TAILQ_HEAD(ip4f_list, ip4_frag) ip4f_list; /* IPv4 fragment cache */
1832 
1833 #define	IP4F_TABSIZE		16	/* IPv4 fragment cache size */
1834 
1835 
1836 static void
1837 ip4f_cache(ip, fin)
1838 	struct ip *ip;
1839 	struct flowinfo_in *fin;
1840 {
1841 	struct ip4_frag *fp;
1842 
1843 	if (TAILQ_EMPTY(&ip4f_list)) {
1844 		/* first time call, allocate fragment cache entries. */
1845 		if (ip4f_init() < 0)
1846 			/* allocation failed! */
1847 			return;
1848 	}
1849 
1850 	fp = ip4f_alloc();
1851 	fp->ip4f_id = ip->ip_id;
1852 	fp->ip4f_info.fi_proto = ip->ip_p;
1853 	fp->ip4f_info.fi_src.s_addr = ip->ip_src.s_addr;
1854 	fp->ip4f_info.fi_dst.s_addr = ip->ip_dst.s_addr;
1855 
1856 	/* save port numbers */
1857 	fp->ip4f_info.fi_sport = fin->fi_sport;
1858 	fp->ip4f_info.fi_dport = fin->fi_dport;
1859 	fp->ip4f_info.fi_gpi   = fin->fi_gpi;
1860 }
1861 
1862 static int
1863 ip4f_lookup(ip, fin)
1864 	struct ip *ip;
1865 	struct flowinfo_in *fin;
1866 {
1867 	struct ip4_frag *fp;
1868 
1869 	for (fp = TAILQ_FIRST(&ip4f_list); fp != NULL && fp->ip4f_valid;
1870 	     fp = TAILQ_NEXT(fp, ip4f_chain))
1871 		if (ip->ip_id == fp->ip4f_id &&
1872 		    ip->ip_src.s_addr == fp->ip4f_info.fi_src.s_addr &&
1873 		    ip->ip_dst.s_addr == fp->ip4f_info.fi_dst.s_addr &&
1874 		    ip->ip_p == fp->ip4f_info.fi_proto) {
1875 
1876 			/* found the matching entry */
1877 			fin->fi_sport = fp->ip4f_info.fi_sport;
1878 			fin->fi_dport = fp->ip4f_info.fi_dport;
1879 			fin->fi_gpi   = fp->ip4f_info.fi_gpi;
1880 
1881 			if ((ntohs(ip->ip_off) & IP_MF) == 0)
1882 				/* this is the last fragment,
1883 				   release the entry. */
1884 				ip4f_free(fp);
1885 
1886 			return (1);
1887 		}
1888 
1889 	/* no matching entry found */
1890 	return (0);
1891 }
1892 
1893 static int
1894 ip4f_init(void)
1895 {
1896 	struct ip4_frag *fp;
1897 	int i;
1898 
1899 	TAILQ_INIT(&ip4f_list);
1900 	for (i=0; i<IP4F_TABSIZE; i++) {
1901 		fp = malloc(sizeof(struct ip4_frag),
1902 		       M_DEVBUF, M_NOWAIT);
1903 		if (fp == NULL) {
1904 			printf("ip4f_init: can't alloc %dth entry!\n", i);
1905 			if (i == 0)
1906 				return (-1);
1907 			return (0);
1908 		}
1909 		fp->ip4f_valid = 0;
1910 		TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1911 	}
1912 	return (0);
1913 }
1914 
1915 static struct ip4_frag *
1916 ip4f_alloc(void)
1917 {
1918 	struct ip4_frag *fp;
1919 
1920 	/* reclaim an entry at the tail, put it at the head */
1921 	fp = TAILQ_LAST(&ip4f_list, ip4f_list);
1922 	TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1923 	fp->ip4f_valid = 1;
1924 	TAILQ_INSERT_HEAD(&ip4f_list, fp, ip4f_chain);
1925 	return (fp);
1926 }
1927 
1928 static void
1929 ip4f_free(fp)
1930 	struct ip4_frag *fp;
1931 {
1932 	TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1933 	fp->ip4f_valid = 0;
1934 	TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1935 }
1936 
1937 #endif /* ALTQ3_CLFIER_COMPAT */
1938