xref: /freebsd/sys/net/altq/altq_subr.c (revision ae41709ab46305df80f7f35bb478a3c8ebf22ebb)
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 	struct epoch_tracker et;
414 	int active;
415 
416 	active = 0;
417 	NET_EPOCH_ENTER(et);
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 	NET_EPOCH_EXIT(et);
435 	if (active > 0)
436 		CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
437 	else
438 		tbr_timer = 0;	/* don't need tbr_timer anymore */
439 }
440 
441 /*
442  * attach a discipline to the interface.  if one already exists, it is
443  * overridden.
444  * Locking is done in the discipline specific attach functions. Basically
445  * they call back to altq_attach which takes care of the attach and locking.
446  */
447 int
448 altq_pfattach(struct pf_altq *a)
449 {
450 	int error = 0;
451 
452 	switch (a->scheduler) {
453 	case ALTQT_NONE:
454 		break;
455 #ifdef ALTQ_CBQ
456 	case ALTQT_CBQ:
457 		error = cbq_pfattach(a);
458 		break;
459 #endif
460 #ifdef ALTQ_PRIQ
461 	case ALTQT_PRIQ:
462 		error = priq_pfattach(a);
463 		break;
464 #endif
465 #ifdef ALTQ_HFSC
466 	case ALTQT_HFSC:
467 		error = hfsc_pfattach(a);
468 		break;
469 #endif
470 #ifdef ALTQ_FAIRQ
471 	case ALTQT_FAIRQ:
472 		error = fairq_pfattach(a);
473 		break;
474 #endif
475 #ifdef ALTQ_CODEL
476 	case ALTQT_CODEL:
477 		error = codel_pfattach(a);
478 		break;
479 #endif
480 	default:
481 		error = ENXIO;
482 	}
483 
484 	return (error);
485 }
486 
487 /*
488  * detach a discipline from the interface.
489  * it is possible that the discipline was already overridden by another
490  * discipline.
491  */
492 int
493 altq_pfdetach(struct pf_altq *a)
494 {
495 	struct ifnet *ifp;
496 	int s, error = 0;
497 
498 	if ((ifp = ifunit(a->ifname)) == NULL)
499 		return (EINVAL);
500 
501 	/* if this discipline is no longer referenced, just return */
502 	/* read unlocked from if_snd */
503 	if (a->altq_disc == NULL || a->altq_disc != ifp->if_snd.altq_disc)
504 		return (0);
505 
506 	s = splnet();
507 	/* read unlocked from if_snd, _disable and _detach take care */
508 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
509 		error = altq_disable(&ifp->if_snd);
510 	if (error == 0)
511 		error = altq_detach(&ifp->if_snd);
512 	splx(s);
513 
514 	return (error);
515 }
516 
517 /*
518  * add a discipline or a queue
519  * Locking is done in the discipline specific functions with regards to
520  * malloc with WAITOK, also it is not yet clear which lock to use.
521  */
522 int
523 altq_add(struct ifnet *ifp, struct pf_altq *a)
524 {
525 	int error = 0;
526 
527 	if (a->qname[0] != 0)
528 		return (altq_add_queue(a));
529 
530 	if (machclk_freq == 0)
531 		init_machclk();
532 	if (machclk_freq == 0)
533 		panic("altq_add: no cpu clock");
534 
535 	switch (a->scheduler) {
536 #ifdef ALTQ_CBQ
537 	case ALTQT_CBQ:
538 		error = cbq_add_altq(ifp, a);
539 		break;
540 #endif
541 #ifdef ALTQ_PRIQ
542 	case ALTQT_PRIQ:
543 		error = priq_add_altq(ifp, a);
544 		break;
545 #endif
546 #ifdef ALTQ_HFSC
547 	case ALTQT_HFSC:
548 		error = hfsc_add_altq(ifp, a);
549 		break;
550 #endif
551 #ifdef ALTQ_FAIRQ
552         case ALTQT_FAIRQ:
553                 error = fairq_add_altq(ifp, a);
554                 break;
555 #endif
556 #ifdef ALTQ_CODEL
557 	case ALTQT_CODEL:
558 		error = codel_add_altq(ifp, a);
559 		break;
560 #endif
561 	default:
562 		error = ENXIO;
563 	}
564 
565 	return (error);
566 }
567 
568 /*
569  * remove a discipline or a queue
570  * It is yet unclear what lock to use to protect this operation, the
571  * discipline specific functions will determine and grab it
572  */
573 int
574 altq_remove(struct pf_altq *a)
575 {
576 	int error = 0;
577 
578 	if (a->qname[0] != 0)
579 		return (altq_remove_queue(a));
580 
581 	switch (a->scheduler) {
582 #ifdef ALTQ_CBQ
583 	case ALTQT_CBQ:
584 		error = cbq_remove_altq(a);
585 		break;
586 #endif
587 #ifdef ALTQ_PRIQ
588 	case ALTQT_PRIQ:
589 		error = priq_remove_altq(a);
590 		break;
591 #endif
592 #ifdef ALTQ_HFSC
593 	case ALTQT_HFSC:
594 		error = hfsc_remove_altq(a);
595 		break;
596 #endif
597 #ifdef ALTQ_FAIRQ
598         case ALTQT_FAIRQ:
599                 error = fairq_remove_altq(a);
600                 break;
601 #endif
602 #ifdef ALTQ_CODEL
603 	case ALTQT_CODEL:
604 		error = codel_remove_altq(a);
605 		break;
606 #endif
607 	default:
608 		error = ENXIO;
609 	}
610 
611 	return (error);
612 }
613 
614 /*
615  * add a queue to the discipline
616  * It is yet unclear what lock to use to protect this operation, the
617  * discipline specific functions will determine and grab it
618  */
619 int
620 altq_add_queue(struct pf_altq *a)
621 {
622 	int error = 0;
623 
624 	switch (a->scheduler) {
625 #ifdef ALTQ_CBQ
626 	case ALTQT_CBQ:
627 		error = cbq_add_queue(a);
628 		break;
629 #endif
630 #ifdef ALTQ_PRIQ
631 	case ALTQT_PRIQ:
632 		error = priq_add_queue(a);
633 		break;
634 #endif
635 #ifdef ALTQ_HFSC
636 	case ALTQT_HFSC:
637 		error = hfsc_add_queue(a);
638 		break;
639 #endif
640 #ifdef ALTQ_FAIRQ
641         case ALTQT_FAIRQ:
642                 error = fairq_add_queue(a);
643                 break;
644 #endif
645 	default:
646 		error = ENXIO;
647 	}
648 
649 	return (error);
650 }
651 
652 /*
653  * remove a queue from the discipline
654  * It is yet unclear what lock to use to protect this operation, the
655  * discipline specific functions will determine and grab it
656  */
657 int
658 altq_remove_queue(struct pf_altq *a)
659 {
660 	int error = 0;
661 
662 	switch (a->scheduler) {
663 #ifdef ALTQ_CBQ
664 	case ALTQT_CBQ:
665 		error = cbq_remove_queue(a);
666 		break;
667 #endif
668 #ifdef ALTQ_PRIQ
669 	case ALTQT_PRIQ:
670 		error = priq_remove_queue(a);
671 		break;
672 #endif
673 #ifdef ALTQ_HFSC
674 	case ALTQT_HFSC:
675 		error = hfsc_remove_queue(a);
676 		break;
677 #endif
678 #ifdef ALTQ_FAIRQ
679         case ALTQT_FAIRQ:
680                 error = fairq_remove_queue(a);
681                 break;
682 #endif
683 	default:
684 		error = ENXIO;
685 	}
686 
687 	return (error);
688 }
689 
690 /*
691  * get queue statistics
692  * Locking is done in the discipline specific functions with regards to
693  * copyout operations, also it is not yet clear which lock to use.
694  */
695 int
696 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version)
697 {
698 	int error = 0;
699 
700 	switch (a->scheduler) {
701 #ifdef ALTQ_CBQ
702 	case ALTQT_CBQ:
703 		error = cbq_getqstats(a, ubuf, nbytes, version);
704 		break;
705 #endif
706 #ifdef ALTQ_PRIQ
707 	case ALTQT_PRIQ:
708 		error = priq_getqstats(a, ubuf, nbytes, version);
709 		break;
710 #endif
711 #ifdef ALTQ_HFSC
712 	case ALTQT_HFSC:
713 		error = hfsc_getqstats(a, ubuf, nbytes, version);
714 		break;
715 #endif
716 #ifdef ALTQ_FAIRQ
717         case ALTQT_FAIRQ:
718                 error = fairq_getqstats(a, ubuf, nbytes, version);
719                 break;
720 #endif
721 #ifdef ALTQ_CODEL
722 	case ALTQT_CODEL:
723 		error = codel_getqstats(a, ubuf, nbytes, version);
724 		break;
725 #endif
726 	default:
727 		error = ENXIO;
728 	}
729 
730 	return (error);
731 }
732 
733 /*
734  * read and write diffserv field in IPv4 or IPv6 header
735  */
736 u_int8_t
737 read_dsfield(m, pktattr)
738 	struct mbuf *m;
739 	struct altq_pktattr *pktattr;
740 {
741 	struct mbuf *m0;
742 	u_int8_t ds_field = 0;
743 
744 	if (pktattr == NULL ||
745 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
746 		return ((u_int8_t)0);
747 
748 	/* verify that pattr_hdr is within the mbuf data */
749 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
750 		if ((pktattr->pattr_hdr >= m0->m_data) &&
751 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
752 			break;
753 	if (m0 == NULL) {
754 		/* ick, pattr_hdr is stale */
755 		pktattr->pattr_af = AF_UNSPEC;
756 #ifdef ALTQ_DEBUG
757 		printf("read_dsfield: can't locate header!\n");
758 #endif
759 		return ((u_int8_t)0);
760 	}
761 
762 	if (pktattr->pattr_af == AF_INET) {
763 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
764 
765 		if (ip->ip_v != 4)
766 			return ((u_int8_t)0);	/* version mismatch! */
767 		ds_field = ip->ip_tos;
768 	}
769 #ifdef INET6
770 	else if (pktattr->pattr_af == AF_INET6) {
771 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
772 		u_int32_t flowlabel;
773 
774 		flowlabel = ntohl(ip6->ip6_flow);
775 		if ((flowlabel >> 28) != 6)
776 			return ((u_int8_t)0);	/* version mismatch! */
777 		ds_field = (flowlabel >> 20) & 0xff;
778 	}
779 #endif
780 	return (ds_field);
781 }
782 
783 void
784 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, u_int8_t dsfield)
785 {
786 	struct mbuf *m0;
787 
788 	if (pktattr == NULL ||
789 	    (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
790 		return;
791 
792 	/* verify that pattr_hdr is within the mbuf data */
793 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
794 		if ((pktattr->pattr_hdr >= m0->m_data) &&
795 		    (pktattr->pattr_hdr < m0->m_data + m0->m_len))
796 			break;
797 	if (m0 == NULL) {
798 		/* ick, pattr_hdr is stale */
799 		pktattr->pattr_af = AF_UNSPEC;
800 #ifdef ALTQ_DEBUG
801 		printf("write_dsfield: can't locate header!\n");
802 #endif
803 		return;
804 	}
805 
806 	if (pktattr->pattr_af == AF_INET) {
807 		struct ip *ip = (struct ip *)pktattr->pattr_hdr;
808 		u_int8_t old;
809 		int32_t sum;
810 
811 		if (ip->ip_v != 4)
812 			return;		/* version mismatch! */
813 		old = ip->ip_tos;
814 		dsfield |= old & 3;	/* leave CU bits */
815 		if (old == dsfield)
816 			return;
817 		ip->ip_tos = dsfield;
818 		/*
819 		 * update checksum (from RFC1624)
820 		 *	   HC' = ~(~HC + ~m + m')
821 		 */
822 		sum = ~ntohs(ip->ip_sum) & 0xffff;
823 		sum += 0xff00 + (~old & 0xff) + dsfield;
824 		sum = (sum >> 16) + (sum & 0xffff);
825 		sum += (sum >> 16);  /* add carry */
826 
827 		ip->ip_sum = htons(~sum & 0xffff);
828 	}
829 #ifdef INET6
830 	else if (pktattr->pattr_af == AF_INET6) {
831 		struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
832 		u_int32_t flowlabel;
833 
834 		flowlabel = ntohl(ip6->ip6_flow);
835 		if ((flowlabel >> 28) != 6)
836 			return;		/* version mismatch! */
837 		flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20);
838 		ip6->ip6_flow = htonl(flowlabel);
839 	}
840 #endif
841 	return;
842 }
843 
844 /*
845  * high resolution clock support taking advantage of a machine dependent
846  * high resolution time counter (e.g., timestamp counter of intel pentium).
847  * we assume
848  *  - 64-bit-long monotonically-increasing counter
849  *  - frequency range is 100M-4GHz (CPU speed)
850  */
851 /* if pcc is not available or disabled, emulate 256MHz using microtime() */
852 #define	MACHCLK_SHIFT	8
853 
854 int machclk_usepcc;
855 u_int32_t machclk_freq;
856 u_int32_t machclk_per_tick;
857 
858 #if defined(__i386__) && defined(__NetBSD__)
859 extern u_int64_t cpu_tsc_freq;
860 #endif
861 
862 #if (__FreeBSD_version >= 700035)
863 /* Update TSC freq with the value indicated by the caller. */
864 static void
865 tsc_freq_changed(void *arg, const struct cf_level *level, int status)
866 {
867 	/* If there was an error during the transition, don't do anything. */
868 	if (status != 0)
869 		return;
870 
871 #if (__FreeBSD_version >= 701102) && (defined(__amd64__) || defined(__i386__))
872 	/* If TSC is P-state invariant, don't do anything. */
873 	if (tsc_is_invariant)
874 		return;
875 #endif
876 
877 	/* Total setting for this level gives the new frequency in MHz. */
878 	init_machclk();
879 }
880 EVENTHANDLER_DEFINE(cpufreq_post_change, tsc_freq_changed, NULL,
881     EVENTHANDLER_PRI_LAST);
882 #endif /* __FreeBSD_version >= 700035 */
883 
884 static void
885 init_machclk_setup(void)
886 {
887 #if (__FreeBSD_version >= 600000)
888 	callout_init(&tbr_callout, 0);
889 #endif
890 
891 	machclk_usepcc = 1;
892 
893 #if (!defined(__amd64__) && !defined(__i386__)) || defined(ALTQ_NOPCC)
894 	machclk_usepcc = 0;
895 #endif
896 #if defined(__FreeBSD__) && defined(SMP)
897 	machclk_usepcc = 0;
898 #endif
899 #if defined(__NetBSD__) && defined(MULTIPROCESSOR)
900 	machclk_usepcc = 0;
901 #endif
902 #if defined(__amd64__) || defined(__i386__)
903 	/* check if TSC is available */
904 	if ((cpu_feature & CPUID_TSC) == 0 ||
905 	    atomic_load_acq_64(&tsc_freq) == 0)
906 		machclk_usepcc = 0;
907 #endif
908 }
909 
910 void
911 init_machclk(void)
912 {
913 	static int called;
914 
915 	/* Call one-time initialization function. */
916 	if (!called) {
917 		init_machclk_setup();
918 		called = 1;
919 	}
920 
921 	if (machclk_usepcc == 0) {
922 		/* emulate 256MHz using microtime() */
923 		machclk_freq = 1000000 << MACHCLK_SHIFT;
924 		machclk_per_tick = machclk_freq / hz;
925 #ifdef ALTQ_DEBUG
926 		printf("altq: emulate %uHz cpu clock\n", machclk_freq);
927 #endif
928 		return;
929 	}
930 
931 	/*
932 	 * if the clock frequency (of Pentium TSC or Alpha PCC) is
933 	 * accessible, just use it.
934 	 */
935 #if defined(__amd64__) || defined(__i386__)
936 	machclk_freq = atomic_load_acq_64(&tsc_freq);
937 #endif
938 
939 	/*
940 	 * if we don't know the clock frequency, measure it.
941 	 */
942 	if (machclk_freq == 0) {
943 		static int	wait;
944 		struct timeval	tv_start, tv_end;
945 		u_int64_t	start, end, diff;
946 		int		timo;
947 
948 		microtime(&tv_start);
949 		start = read_machclk();
950 		timo = hz;	/* 1 sec */
951 		(void)tsleep(&wait, PWAIT | PCATCH, "init_machclk", timo);
952 		microtime(&tv_end);
953 		end = read_machclk();
954 		diff = (u_int64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000
955 		    + tv_end.tv_usec - tv_start.tv_usec;
956 		if (diff != 0)
957 			machclk_freq = (u_int)((end - start) * 1000000 / diff);
958 	}
959 
960 	machclk_per_tick = machclk_freq / hz;
961 
962 #ifdef ALTQ_DEBUG
963 	printf("altq: CPU clock: %uHz\n", machclk_freq);
964 #endif
965 }
966 
967 #if defined(__OpenBSD__) && defined(__i386__)
968 static __inline u_int64_t
969 rdtsc(void)
970 {
971 	u_int64_t rv;
972 	__asm __volatile(".byte 0x0f, 0x31" : "=A" (rv));
973 	return (rv);
974 }
975 #endif /* __OpenBSD__ && __i386__ */
976 
977 u_int64_t
978 read_machclk(void)
979 {
980 	u_int64_t val;
981 
982 	if (machclk_usepcc) {
983 #if defined(__amd64__) || defined(__i386__)
984 		val = rdtsc();
985 #else
986 		panic("read_machclk");
987 #endif
988 	} else {
989 		struct timeval tv, boottime;
990 
991 		microtime(&tv);
992 		getboottime(&boottime);
993 		val = (((u_int64_t)(tv.tv_sec - boottime.tv_sec) * 1000000
994 		    + tv.tv_usec) << MACHCLK_SHIFT);
995 	}
996 	return (val);
997 }
998 
999 #ifdef ALTQ3_CLFIER_COMPAT
1000 
1001 #ifndef IPPROTO_ESP
1002 #define	IPPROTO_ESP	50		/* encapsulating security payload */
1003 #endif
1004 #ifndef IPPROTO_AH
1005 #define	IPPROTO_AH	51		/* authentication header */
1006 #endif
1007 
1008 /*
1009  * extract flow information from a given packet.
1010  * filt_mask shows flowinfo fields required.
1011  * we assume the ip header is in one mbuf, and addresses and ports are
1012  * in network byte order.
1013  */
1014 int
1015 altq_extractflow(m, af, flow, filt_bmask)
1016 	struct mbuf *m;
1017 	int af;
1018 	struct flowinfo *flow;
1019 	u_int32_t	filt_bmask;
1020 {
1021 
1022 	switch (af) {
1023 	case PF_INET: {
1024 		struct flowinfo_in *fin;
1025 		struct ip *ip;
1026 
1027 		ip = mtod(m, struct ip *);
1028 
1029 		if (ip->ip_v != 4)
1030 			break;
1031 
1032 		fin = (struct flowinfo_in *)flow;
1033 		fin->fi_len = sizeof(struct flowinfo_in);
1034 		fin->fi_family = AF_INET;
1035 
1036 		fin->fi_proto = ip->ip_p;
1037 		fin->fi_tos = ip->ip_tos;
1038 
1039 		fin->fi_src.s_addr = ip->ip_src.s_addr;
1040 		fin->fi_dst.s_addr = ip->ip_dst.s_addr;
1041 
1042 		if (filt_bmask & FIMB4_PORTS)
1043 			/* if port info is required, extract port numbers */
1044 			extract_ports4(m, ip, fin);
1045 		else {
1046 			fin->fi_sport = 0;
1047 			fin->fi_dport = 0;
1048 			fin->fi_gpi = 0;
1049 		}
1050 		return (1);
1051 	}
1052 
1053 #ifdef INET6
1054 	case PF_INET6: {
1055 		struct flowinfo_in6 *fin6;
1056 		struct ip6_hdr *ip6;
1057 
1058 		ip6 = mtod(m, struct ip6_hdr *);
1059 		/* should we check the ip version? */
1060 
1061 		fin6 = (struct flowinfo_in6 *)flow;
1062 		fin6->fi6_len = sizeof(struct flowinfo_in6);
1063 		fin6->fi6_family = AF_INET6;
1064 
1065 		fin6->fi6_proto = ip6->ip6_nxt;
1066 		fin6->fi6_tclass   = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
1067 
1068 		fin6->fi6_flowlabel = ip6->ip6_flow & htonl(0x000fffff);
1069 		fin6->fi6_src = ip6->ip6_src;
1070 		fin6->fi6_dst = ip6->ip6_dst;
1071 
1072 		if ((filt_bmask & FIMB6_PORTS) ||
1073 		    ((filt_bmask & FIMB6_PROTO)
1074 		     && ip6->ip6_nxt > IPPROTO_IPV6))
1075 			/*
1076 			 * if port info is required, or proto is required
1077 			 * but there are option headers, extract port
1078 			 * and protocol numbers.
1079 			 */
1080 			extract_ports6(m, ip6, fin6);
1081 		else {
1082 			fin6->fi6_sport = 0;
1083 			fin6->fi6_dport = 0;
1084 			fin6->fi6_gpi = 0;
1085 		}
1086 		return (1);
1087 	}
1088 #endif /* INET6 */
1089 
1090 	default:
1091 		break;
1092 	}
1093 
1094 	/* failed */
1095 	flow->fi_len = sizeof(struct flowinfo);
1096 	flow->fi_family = AF_UNSPEC;
1097 	return (0);
1098 }
1099 
1100 /*
1101  * helper routine to extract port numbers
1102  */
1103 /* structure for ipsec and ipv6 option header template */
1104 struct _opt6 {
1105 	u_int8_t	opt6_nxt;	/* next header */
1106 	u_int8_t	opt6_hlen;	/* header extension length */
1107 	u_int16_t	_pad;
1108 	u_int32_t	ah_spi;		/* security parameter index
1109 					   for authentication header */
1110 };
1111 
1112 /*
1113  * extract port numbers from a ipv4 packet.
1114  */
1115 static int
1116 extract_ports4(m, ip, fin)
1117 	struct mbuf *m;
1118 	struct ip *ip;
1119 	struct flowinfo_in *fin;
1120 {
1121 	struct mbuf *m0;
1122 	u_short ip_off;
1123 	u_int8_t proto;
1124 	int 	off;
1125 
1126 	fin->fi_sport = 0;
1127 	fin->fi_dport = 0;
1128 	fin->fi_gpi = 0;
1129 
1130 	ip_off = ntohs(ip->ip_off);
1131 	/* if it is a fragment, try cached fragment info */
1132 	if (ip_off & IP_OFFMASK) {
1133 		ip4f_lookup(ip, fin);
1134 		return (1);
1135 	}
1136 
1137 	/* locate the mbuf containing the protocol header */
1138 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
1139 		if (((caddr_t)ip >= m0->m_data) &&
1140 		    ((caddr_t)ip < m0->m_data + m0->m_len))
1141 			break;
1142 	if (m0 == NULL) {
1143 #ifdef ALTQ_DEBUG
1144 		printf("extract_ports4: can't locate header! ip=%p\n", ip);
1145 #endif
1146 		return (0);
1147 	}
1148 	off = ((caddr_t)ip - m0->m_data) + (ip->ip_hl << 2);
1149 	proto = ip->ip_p;
1150 
1151 #ifdef ALTQ_IPSEC
1152  again:
1153 #endif
1154 	while (off >= m0->m_len) {
1155 		off -= m0->m_len;
1156 		m0 = m0->m_next;
1157 		if (m0 == NULL)
1158 			return (0);  /* bogus ip_hl! */
1159 	}
1160 	if (m0->m_len < off + 4)
1161 		return (0);
1162 
1163 	switch (proto) {
1164 	case IPPROTO_TCP:
1165 	case IPPROTO_UDP: {
1166 		struct udphdr *udp;
1167 
1168 		udp = (struct udphdr *)(mtod(m0, caddr_t) + off);
1169 		fin->fi_sport = udp->uh_sport;
1170 		fin->fi_dport = udp->uh_dport;
1171 		fin->fi_proto = proto;
1172 		}
1173 		break;
1174 
1175 #ifdef ALTQ_IPSEC
1176 	case IPPROTO_ESP:
1177 		if (fin->fi_gpi == 0){
1178 			u_int32_t *gpi;
1179 
1180 			gpi = (u_int32_t *)(mtod(m0, caddr_t) + off);
1181 			fin->fi_gpi   = *gpi;
1182 		}
1183 		fin->fi_proto = proto;
1184 		break;
1185 
1186 	case IPPROTO_AH: {
1187 			/* get next header and header length */
1188 			struct _opt6 *opt6;
1189 
1190 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1191 			proto = opt6->opt6_nxt;
1192 			off += 8 + (opt6->opt6_hlen * 4);
1193 			if (fin->fi_gpi == 0 && m0->m_len >= off + 8)
1194 				fin->fi_gpi = opt6->ah_spi;
1195 		}
1196 		/* goto the next header */
1197 		goto again;
1198 #endif  /* ALTQ_IPSEC */
1199 
1200 	default:
1201 		fin->fi_proto = proto;
1202 		return (0);
1203 	}
1204 
1205 	/* if this is a first fragment, cache it. */
1206 	if (ip_off & IP_MF)
1207 		ip4f_cache(ip, fin);
1208 
1209 	return (1);
1210 }
1211 
1212 #ifdef INET6
1213 static int
1214 extract_ports6(m, ip6, fin6)
1215 	struct mbuf *m;
1216 	struct ip6_hdr *ip6;
1217 	struct flowinfo_in6 *fin6;
1218 {
1219 	struct mbuf *m0;
1220 	int	off;
1221 	u_int8_t proto;
1222 
1223 	fin6->fi6_gpi   = 0;
1224 	fin6->fi6_sport = 0;
1225 	fin6->fi6_dport = 0;
1226 
1227 	/* locate the mbuf containing the protocol header */
1228 	for (m0 = m; m0 != NULL; m0 = m0->m_next)
1229 		if (((caddr_t)ip6 >= m0->m_data) &&
1230 		    ((caddr_t)ip6 < m0->m_data + m0->m_len))
1231 			break;
1232 	if (m0 == NULL) {
1233 #ifdef ALTQ_DEBUG
1234 		printf("extract_ports6: can't locate header! ip6=%p\n", ip6);
1235 #endif
1236 		return (0);
1237 	}
1238 	off = ((caddr_t)ip6 - m0->m_data) + sizeof(struct ip6_hdr);
1239 
1240 	proto = ip6->ip6_nxt;
1241 	do {
1242 		while (off >= m0->m_len) {
1243 			off -= m0->m_len;
1244 			m0 = m0->m_next;
1245 			if (m0 == NULL)
1246 				return (0);
1247 		}
1248 		if (m0->m_len < off + 4)
1249 			return (0);
1250 
1251 		switch (proto) {
1252 		case IPPROTO_TCP:
1253 		case IPPROTO_UDP: {
1254 			struct udphdr *udp;
1255 
1256 			udp = (struct udphdr *)(mtod(m0, caddr_t) + off);
1257 			fin6->fi6_sport = udp->uh_sport;
1258 			fin6->fi6_dport = udp->uh_dport;
1259 			fin6->fi6_proto = proto;
1260 			}
1261 			return (1);
1262 
1263 		case IPPROTO_ESP:
1264 			if (fin6->fi6_gpi == 0) {
1265 				u_int32_t *gpi;
1266 
1267 				gpi = (u_int32_t *)(mtod(m0, caddr_t) + off);
1268 				fin6->fi6_gpi   = *gpi;
1269 			}
1270 			fin6->fi6_proto = proto;
1271 			return (1);
1272 
1273 		case IPPROTO_AH: {
1274 			/* get next header and header length */
1275 			struct _opt6 *opt6;
1276 
1277 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1278 			if (fin6->fi6_gpi == 0 && m0->m_len >= off + 8)
1279 				fin6->fi6_gpi = opt6->ah_spi;
1280 			proto = opt6->opt6_nxt;
1281 			off += 8 + (opt6->opt6_hlen * 4);
1282 			/* goto the next header */
1283 			break;
1284 			}
1285 
1286 		case IPPROTO_HOPOPTS:
1287 		case IPPROTO_ROUTING:
1288 		case IPPROTO_DSTOPTS: {
1289 			/* get next header and header length */
1290 			struct _opt6 *opt6;
1291 
1292 			opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1293 			proto = opt6->opt6_nxt;
1294 			off += (opt6->opt6_hlen + 1) * 8;
1295 			/* goto the next header */
1296 			break;
1297 			}
1298 
1299 		case IPPROTO_FRAGMENT:
1300 			/* ipv6 fragmentations are not supported yet */
1301 		default:
1302 			fin6->fi6_proto = proto;
1303 			return (0);
1304 		}
1305 	} while (1);
1306 	/*NOTREACHED*/
1307 }
1308 #endif /* INET6 */
1309 
1310 /*
1311  * altq common classifier
1312  */
1313 int
1314 acc_add_filter(classifier, filter, class, phandle)
1315 	struct acc_classifier *classifier;
1316 	struct flow_filter *filter;
1317 	void	*class;
1318 	u_long	*phandle;
1319 {
1320 	struct acc_filter *afp, *prev, *tmp;
1321 	int	i, s;
1322 
1323 #ifdef INET6
1324 	if (filter->ff_flow.fi_family != AF_INET &&
1325 	    filter->ff_flow.fi_family != AF_INET6)
1326 		return (EINVAL);
1327 #else
1328 	if (filter->ff_flow.fi_family != AF_INET)
1329 		return (EINVAL);
1330 #endif
1331 
1332 	afp = malloc(sizeof(struct acc_filter),
1333 	       M_DEVBUF, M_WAITOK);
1334 	if (afp == NULL)
1335 		return (ENOMEM);
1336 	bzero(afp, sizeof(struct acc_filter));
1337 
1338 	afp->f_filter = *filter;
1339 	afp->f_class = class;
1340 
1341 	i = ACC_WILDCARD_INDEX;
1342 	if (filter->ff_flow.fi_family == AF_INET) {
1343 		struct flow_filter *filter4 = &afp->f_filter;
1344 
1345 		/*
1346 		 * if address is 0, it's a wildcard.  if address mask
1347 		 * isn't set, use full mask.
1348 		 */
1349 		if (filter4->ff_flow.fi_dst.s_addr == 0)
1350 			filter4->ff_mask.mask_dst.s_addr = 0;
1351 		else if (filter4->ff_mask.mask_dst.s_addr == 0)
1352 			filter4->ff_mask.mask_dst.s_addr = 0xffffffff;
1353 		if (filter4->ff_flow.fi_src.s_addr == 0)
1354 			filter4->ff_mask.mask_src.s_addr = 0;
1355 		else if (filter4->ff_mask.mask_src.s_addr == 0)
1356 			filter4->ff_mask.mask_src.s_addr = 0xffffffff;
1357 
1358 		/* clear extra bits in addresses  */
1359 		   filter4->ff_flow.fi_dst.s_addr &=
1360 		       filter4->ff_mask.mask_dst.s_addr;
1361 		   filter4->ff_flow.fi_src.s_addr &=
1362 		       filter4->ff_mask.mask_src.s_addr;
1363 
1364 		/*
1365 		 * if dst address is a wildcard, use hash-entry
1366 		 * ACC_WILDCARD_INDEX.
1367 		 */
1368 		if (filter4->ff_mask.mask_dst.s_addr != 0xffffffff)
1369 			i = ACC_WILDCARD_INDEX;
1370 		else
1371 			i = ACC_GET_HASH_INDEX(filter4->ff_flow.fi_dst.s_addr);
1372 	}
1373 #ifdef INET6
1374 	else if (filter->ff_flow.fi_family == AF_INET6) {
1375 		struct flow_filter6 *filter6 =
1376 			(struct flow_filter6 *)&afp->f_filter;
1377 #ifndef IN6MASK0 /* taken from kame ipv6 */
1378 #define	IN6MASK0	{{{ 0, 0, 0, 0 }}}
1379 #define	IN6MASK128	{{{ 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff }}}
1380 		const struct in6_addr in6mask0 = IN6MASK0;
1381 		const struct in6_addr in6mask128 = IN6MASK128;
1382 #endif
1383 
1384 		if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_dst))
1385 			filter6->ff_mask6.mask6_dst = in6mask0;
1386 		else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_dst))
1387 			filter6->ff_mask6.mask6_dst = in6mask128;
1388 		if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_src))
1389 			filter6->ff_mask6.mask6_src = in6mask0;
1390 		else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_src))
1391 			filter6->ff_mask6.mask6_src = in6mask128;
1392 
1393 		/* clear extra bits in addresses  */
1394 		for (i = 0; i < 16; i++)
1395 			filter6->ff_flow6.fi6_dst.s6_addr[i] &=
1396 			    filter6->ff_mask6.mask6_dst.s6_addr[i];
1397 		for (i = 0; i < 16; i++)
1398 			filter6->ff_flow6.fi6_src.s6_addr[i] &=
1399 			    filter6->ff_mask6.mask6_src.s6_addr[i];
1400 
1401 		if (filter6->ff_flow6.fi6_flowlabel == 0)
1402 			i = ACC_WILDCARD_INDEX;
1403 		else
1404 			i = ACC_GET_HASH_INDEX(filter6->ff_flow6.fi6_flowlabel);
1405 	}
1406 #endif /* INET6 */
1407 
1408 	afp->f_handle = get_filt_handle(classifier, i);
1409 
1410 	/* update filter bitmask */
1411 	afp->f_fbmask = filt2fibmask(filter);
1412 	classifier->acc_fbmask |= afp->f_fbmask;
1413 
1414 	/*
1415 	 * add this filter to the filter list.
1416 	 * filters are ordered from the highest rule number.
1417 	 */
1418 	s = splnet();
1419 	prev = NULL;
1420 	LIST_FOREACH(tmp, &classifier->acc_filters[i], f_chain) {
1421 		if (tmp->f_filter.ff_ruleno > afp->f_filter.ff_ruleno)
1422 			prev = tmp;
1423 		else
1424 			break;
1425 	}
1426 	if (prev == NULL)
1427 		LIST_INSERT_HEAD(&classifier->acc_filters[i], afp, f_chain);
1428 	else
1429 		LIST_INSERT_AFTER(prev, afp, f_chain);
1430 	splx(s);
1431 
1432 	*phandle = afp->f_handle;
1433 	return (0);
1434 }
1435 
1436 int
1437 acc_delete_filter(classifier, handle)
1438 	struct acc_classifier *classifier;
1439 	u_long handle;
1440 {
1441 	struct acc_filter *afp;
1442 	int	s;
1443 
1444 	if ((afp = filth_to_filtp(classifier, handle)) == NULL)
1445 		return (EINVAL);
1446 
1447 	s = splnet();
1448 	LIST_REMOVE(afp, f_chain);
1449 	splx(s);
1450 
1451 	free(afp, M_DEVBUF);
1452 
1453 	/* todo: update filt_bmask */
1454 
1455 	return (0);
1456 }
1457 
1458 /*
1459  * delete filters referencing to the specified class.
1460  * if the all flag is not 0, delete all the filters.
1461  */
1462 int
1463 acc_discard_filters(classifier, class, all)
1464 	struct acc_classifier *classifier;
1465 	void	*class;
1466 	int	all;
1467 {
1468 	struct acc_filter *afp;
1469 	int	i, s;
1470 
1471 	s = splnet();
1472 	for (i = 0; i < ACC_FILTER_TABLESIZE; i++) {
1473 		do {
1474 			LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1475 				if (all || afp->f_class == class) {
1476 					LIST_REMOVE(afp, f_chain);
1477 					free(afp, M_DEVBUF);
1478 					/* start again from the head */
1479 					break;
1480 				}
1481 		} while (afp != NULL);
1482 	}
1483 	splx(s);
1484 
1485 	if (all)
1486 		classifier->acc_fbmask = 0;
1487 
1488 	return (0);
1489 }
1490 
1491 void *
1492 acc_classify(clfier, m, af)
1493 	void *clfier;
1494 	struct mbuf *m;
1495 	int af;
1496 {
1497 	struct acc_classifier *classifier;
1498 	struct flowinfo flow;
1499 	struct acc_filter *afp;
1500 	int	i;
1501 
1502 	classifier = (struct acc_classifier *)clfier;
1503 	altq_extractflow(m, af, &flow, classifier->acc_fbmask);
1504 
1505 	if (flow.fi_family == AF_INET) {
1506 		struct flowinfo_in *fp = (struct flowinfo_in *)&flow;
1507 
1508 		if ((classifier->acc_fbmask & FIMB4_ALL) == FIMB4_TOS) {
1509 			/* only tos is used */
1510 			LIST_FOREACH(afp,
1511 				 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1512 				 f_chain)
1513 				if (apply_tosfilter4(afp->f_fbmask,
1514 						     &afp->f_filter, fp))
1515 					/* filter matched */
1516 					return (afp->f_class);
1517 		} else if ((classifier->acc_fbmask &
1518 			(~(FIMB4_PROTO|FIMB4_SPORT|FIMB4_DPORT) & FIMB4_ALL))
1519 		    == 0) {
1520 			/* only proto and ports are used */
1521 			LIST_FOREACH(afp,
1522 				 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1523 				 f_chain)
1524 				if (apply_ppfilter4(afp->f_fbmask,
1525 						    &afp->f_filter, fp))
1526 					/* filter matched */
1527 					return (afp->f_class);
1528 		} else {
1529 			/* get the filter hash entry from its dest address */
1530 			i = ACC_GET_HASH_INDEX(fp->fi_dst.s_addr);
1531 			do {
1532 				/*
1533 				 * go through this loop twice.  first for dst
1534 				 * hash, second for wildcards.
1535 				 */
1536 				LIST_FOREACH(afp, &classifier->acc_filters[i],
1537 					     f_chain)
1538 					if (apply_filter4(afp->f_fbmask,
1539 							  &afp->f_filter, fp))
1540 						/* filter matched */
1541 						return (afp->f_class);
1542 
1543 				/*
1544 				 * check again for filters with a dst addr
1545 				 * wildcard.
1546 				 * (daddr == 0 || dmask != 0xffffffff).
1547 				 */
1548 				if (i != ACC_WILDCARD_INDEX)
1549 					i = ACC_WILDCARD_INDEX;
1550 				else
1551 					break;
1552 			} while (1);
1553 		}
1554 	}
1555 #ifdef INET6
1556 	else if (flow.fi_family == AF_INET6) {
1557 		struct flowinfo_in6 *fp6 = (struct flowinfo_in6 *)&flow;
1558 
1559 		/* get the filter hash entry from its flow ID */
1560 		if (fp6->fi6_flowlabel != 0)
1561 			i = ACC_GET_HASH_INDEX(fp6->fi6_flowlabel);
1562 		else
1563 			/* flowlable can be zero */
1564 			i = ACC_WILDCARD_INDEX;
1565 
1566 		/* go through this loop twice.  first for flow hash, second
1567 		   for wildcards. */
1568 		do {
1569 			LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1570 				if (apply_filter6(afp->f_fbmask,
1571 					(struct flow_filter6 *)&afp->f_filter,
1572 					fp6))
1573 					/* filter matched */
1574 					return (afp->f_class);
1575 
1576 			/*
1577 			 * check again for filters with a wildcard.
1578 			 */
1579 			if (i != ACC_WILDCARD_INDEX)
1580 				i = ACC_WILDCARD_INDEX;
1581 			else
1582 				break;
1583 		} while (1);
1584 	}
1585 #endif /* INET6 */
1586 
1587 	/* no filter matched */
1588 	return (NULL);
1589 }
1590 
1591 static int
1592 apply_filter4(fbmask, filt, pkt)
1593 	u_int32_t	fbmask;
1594 	struct flow_filter *filt;
1595 	struct flowinfo_in *pkt;
1596 {
1597 	if (filt->ff_flow.fi_family != AF_INET)
1598 		return (0);
1599 	if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1600 		return (0);
1601 	if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1602 		return (0);
1603 	if ((fbmask & FIMB4_DADDR) &&
1604 	    filt->ff_flow.fi_dst.s_addr !=
1605 	    (pkt->fi_dst.s_addr & filt->ff_mask.mask_dst.s_addr))
1606 		return (0);
1607 	if ((fbmask & FIMB4_SADDR) &&
1608 	    filt->ff_flow.fi_src.s_addr !=
1609 	    (pkt->fi_src.s_addr & filt->ff_mask.mask_src.s_addr))
1610 		return (0);
1611 	if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1612 		return (0);
1613 	if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1614 	    (pkt->fi_tos & filt->ff_mask.mask_tos))
1615 		return (0);
1616 	if ((fbmask & FIMB4_GPI) && filt->ff_flow.fi_gpi != (pkt->fi_gpi))
1617 		return (0);
1618 	/* match */
1619 	return (1);
1620 }
1621 
1622 /*
1623  * filter matching function optimized for a common case that checks
1624  * only protocol and port numbers
1625  */
1626 static int
1627 apply_ppfilter4(fbmask, filt, pkt)
1628 	u_int32_t	fbmask;
1629 	struct flow_filter *filt;
1630 	struct flowinfo_in *pkt;
1631 {
1632 	if (filt->ff_flow.fi_family != AF_INET)
1633 		return (0);
1634 	if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1635 		return (0);
1636 	if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1637 		return (0);
1638 	if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1639 		return (0);
1640 	/* match */
1641 	return (1);
1642 }
1643 
1644 /*
1645  * filter matching function only for tos field.
1646  */
1647 static int
1648 apply_tosfilter4(fbmask, filt, pkt)
1649 	u_int32_t	fbmask;
1650 	struct flow_filter *filt;
1651 	struct flowinfo_in *pkt;
1652 {
1653 	if (filt->ff_flow.fi_family != AF_INET)
1654 		return (0);
1655 	if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1656 	    (pkt->fi_tos & filt->ff_mask.mask_tos))
1657 		return (0);
1658 	/* match */
1659 	return (1);
1660 }
1661 
1662 #ifdef INET6
1663 static int
1664 apply_filter6(fbmask, filt, pkt)
1665 	u_int32_t	fbmask;
1666 	struct flow_filter6 *filt;
1667 	struct flowinfo_in6 *pkt;
1668 {
1669 	int i;
1670 
1671 	if (filt->ff_flow6.fi6_family != AF_INET6)
1672 		return (0);
1673 	if ((fbmask & FIMB6_FLABEL) &&
1674 	    filt->ff_flow6.fi6_flowlabel != pkt->fi6_flowlabel)
1675 		return (0);
1676 	if ((fbmask & FIMB6_PROTO) &&
1677 	    filt->ff_flow6.fi6_proto != pkt->fi6_proto)
1678 		return (0);
1679 	if ((fbmask & FIMB6_SPORT) &&
1680 	    filt->ff_flow6.fi6_sport != pkt->fi6_sport)
1681 		return (0);
1682 	if ((fbmask & FIMB6_DPORT) &&
1683 	    filt->ff_flow6.fi6_dport != pkt->fi6_dport)
1684 		return (0);
1685 	if (fbmask & FIMB6_SADDR) {
1686 		for (i = 0; i < 4; i++)
1687 			if (filt->ff_flow6.fi6_src.s6_addr32[i] !=
1688 			    (pkt->fi6_src.s6_addr32[i] &
1689 			     filt->ff_mask6.mask6_src.s6_addr32[i]))
1690 				return (0);
1691 	}
1692 	if (fbmask & FIMB6_DADDR) {
1693 		for (i = 0; i < 4; i++)
1694 			if (filt->ff_flow6.fi6_dst.s6_addr32[i] !=
1695 			    (pkt->fi6_dst.s6_addr32[i] &
1696 			     filt->ff_mask6.mask6_dst.s6_addr32[i]))
1697 				return (0);
1698 	}
1699 	if ((fbmask & FIMB6_TCLASS) &&
1700 	    filt->ff_flow6.fi6_tclass !=
1701 	    (pkt->fi6_tclass & filt->ff_mask6.mask6_tclass))
1702 		return (0);
1703 	if ((fbmask & FIMB6_GPI) &&
1704 	    filt->ff_flow6.fi6_gpi != pkt->fi6_gpi)
1705 		return (0);
1706 	/* match */
1707 	return (1);
1708 }
1709 #endif /* INET6 */
1710 
1711 /*
1712  *  filter handle:
1713  *	bit 20-28: index to the filter hash table
1714  *	bit  0-19: unique id in the hash bucket.
1715  */
1716 static u_long
1717 get_filt_handle(classifier, i)
1718 	struct acc_classifier *classifier;
1719 	int	i;
1720 {
1721 	static u_long handle_number = 1;
1722 	u_long 	handle;
1723 	struct acc_filter *afp;
1724 
1725 	while (1) {
1726 		handle = handle_number++ & 0x000fffff;
1727 
1728 		if (LIST_EMPTY(&classifier->acc_filters[i]))
1729 			break;
1730 
1731 		LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1732 			if ((afp->f_handle & 0x000fffff) == handle)
1733 				break;
1734 		if (afp == NULL)
1735 			break;
1736 		/* this handle is already used, try again */
1737 	}
1738 
1739 	return ((i << 20) | handle);
1740 }
1741 
1742 /* convert filter handle to filter pointer */
1743 static struct acc_filter *
1744 filth_to_filtp(classifier, handle)
1745 	struct acc_classifier *classifier;
1746 	u_long handle;
1747 {
1748 	struct acc_filter *afp;
1749 	int	i;
1750 
1751 	i = ACC_GET_HINDEX(handle);
1752 
1753 	LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1754 		if (afp->f_handle == handle)
1755 			return (afp);
1756 
1757 	return (NULL);
1758 }
1759 
1760 /* create flowinfo bitmask */
1761 static u_int32_t
1762 filt2fibmask(filt)
1763 	struct flow_filter *filt;
1764 {
1765 	u_int32_t mask = 0;
1766 #ifdef INET6
1767 	struct flow_filter6 *filt6;
1768 #endif
1769 
1770 	switch (filt->ff_flow.fi_family) {
1771 	case AF_INET:
1772 		if (filt->ff_flow.fi_proto != 0)
1773 			mask |= FIMB4_PROTO;
1774 		if (filt->ff_flow.fi_tos != 0)
1775 			mask |= FIMB4_TOS;
1776 		if (filt->ff_flow.fi_dst.s_addr != 0)
1777 			mask |= FIMB4_DADDR;
1778 		if (filt->ff_flow.fi_src.s_addr != 0)
1779 			mask |= FIMB4_SADDR;
1780 		if (filt->ff_flow.fi_sport != 0)
1781 			mask |= FIMB4_SPORT;
1782 		if (filt->ff_flow.fi_dport != 0)
1783 			mask |= FIMB4_DPORT;
1784 		if (filt->ff_flow.fi_gpi != 0)
1785 			mask |= FIMB4_GPI;
1786 		break;
1787 #ifdef INET6
1788 	case AF_INET6:
1789 		filt6 = (struct flow_filter6 *)filt;
1790 
1791 		if (filt6->ff_flow6.fi6_proto != 0)
1792 			mask |= FIMB6_PROTO;
1793 		if (filt6->ff_flow6.fi6_tclass != 0)
1794 			mask |= FIMB6_TCLASS;
1795 		if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_dst))
1796 			mask |= FIMB6_DADDR;
1797 		if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_src))
1798 			mask |= FIMB6_SADDR;
1799 		if (filt6->ff_flow6.fi6_sport != 0)
1800 			mask |= FIMB6_SPORT;
1801 		if (filt6->ff_flow6.fi6_dport != 0)
1802 			mask |= FIMB6_DPORT;
1803 		if (filt6->ff_flow6.fi6_gpi != 0)
1804 			mask |= FIMB6_GPI;
1805 		if (filt6->ff_flow6.fi6_flowlabel != 0)
1806 			mask |= FIMB6_FLABEL;
1807 		break;
1808 #endif /* INET6 */
1809 	}
1810 	return (mask);
1811 }
1812 
1813 /*
1814  * helper functions to handle IPv4 fragments.
1815  * currently only in-sequence fragments are handled.
1816  *	- fragment info is cached in a LRU list.
1817  *	- when a first fragment is found, cache its flow info.
1818  *	- when a non-first fragment is found, lookup the cache.
1819  */
1820 
1821 struct ip4_frag {
1822     TAILQ_ENTRY(ip4_frag) ip4f_chain;
1823     char    ip4f_valid;
1824     u_short ip4f_id;
1825     struct flowinfo_in ip4f_info;
1826 };
1827 
1828 static TAILQ_HEAD(ip4f_list, ip4_frag) ip4f_list; /* IPv4 fragment cache */
1829 
1830 #define	IP4F_TABSIZE		16	/* IPv4 fragment cache size */
1831 
1832 static void
1833 ip4f_cache(ip, fin)
1834 	struct ip *ip;
1835 	struct flowinfo_in *fin;
1836 {
1837 	struct ip4_frag *fp;
1838 
1839 	if (TAILQ_EMPTY(&ip4f_list)) {
1840 		/* first time call, allocate fragment cache entries. */
1841 		if (ip4f_init() < 0)
1842 			/* allocation failed! */
1843 			return;
1844 	}
1845 
1846 	fp = ip4f_alloc();
1847 	fp->ip4f_id = ip->ip_id;
1848 	fp->ip4f_info.fi_proto = ip->ip_p;
1849 	fp->ip4f_info.fi_src.s_addr = ip->ip_src.s_addr;
1850 	fp->ip4f_info.fi_dst.s_addr = ip->ip_dst.s_addr;
1851 
1852 	/* save port numbers */
1853 	fp->ip4f_info.fi_sport = fin->fi_sport;
1854 	fp->ip4f_info.fi_dport = fin->fi_dport;
1855 	fp->ip4f_info.fi_gpi   = fin->fi_gpi;
1856 }
1857 
1858 static int
1859 ip4f_lookup(ip, fin)
1860 	struct ip *ip;
1861 	struct flowinfo_in *fin;
1862 {
1863 	struct ip4_frag *fp;
1864 
1865 	for (fp = TAILQ_FIRST(&ip4f_list); fp != NULL && fp->ip4f_valid;
1866 	     fp = TAILQ_NEXT(fp, ip4f_chain))
1867 		if (ip->ip_id == fp->ip4f_id &&
1868 		    ip->ip_src.s_addr == fp->ip4f_info.fi_src.s_addr &&
1869 		    ip->ip_dst.s_addr == fp->ip4f_info.fi_dst.s_addr &&
1870 		    ip->ip_p == fp->ip4f_info.fi_proto) {
1871 			/* found the matching entry */
1872 			fin->fi_sport = fp->ip4f_info.fi_sport;
1873 			fin->fi_dport = fp->ip4f_info.fi_dport;
1874 			fin->fi_gpi   = fp->ip4f_info.fi_gpi;
1875 
1876 			if ((ntohs(ip->ip_off) & IP_MF) == 0)
1877 				/* this is the last fragment,
1878 				   release the entry. */
1879 				ip4f_free(fp);
1880 
1881 			return (1);
1882 		}
1883 
1884 	/* no matching entry found */
1885 	return (0);
1886 }
1887 
1888 static int
1889 ip4f_init(void)
1890 {
1891 	struct ip4_frag *fp;
1892 	int i;
1893 
1894 	TAILQ_INIT(&ip4f_list);
1895 	for (i=0; i<IP4F_TABSIZE; i++) {
1896 		fp = malloc(sizeof(struct ip4_frag),
1897 		       M_DEVBUF, M_NOWAIT);
1898 		if (fp == NULL) {
1899 			printf("ip4f_init: can't alloc %dth entry!\n", i);
1900 			if (i == 0)
1901 				return (-1);
1902 			return (0);
1903 		}
1904 		fp->ip4f_valid = 0;
1905 		TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1906 	}
1907 	return (0);
1908 }
1909 
1910 static struct ip4_frag *
1911 ip4f_alloc(void)
1912 {
1913 	struct ip4_frag *fp;
1914 
1915 	/* reclaim an entry at the tail, put it at the head */
1916 	fp = TAILQ_LAST(&ip4f_list, ip4f_list);
1917 	TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1918 	fp->ip4f_valid = 1;
1919 	TAILQ_INSERT_HEAD(&ip4f_list, fp, ip4f_chain);
1920 	return (fp);
1921 }
1922 
1923 static void
1924 ip4f_free(fp)
1925 	struct ip4_frag *fp;
1926 {
1927 	TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1928 	fp->ip4f_valid = 0;
1929 	TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1930 }
1931 
1932 #endif /* ALTQ3_CLFIER_COMPAT */
1933