1 /*-
2 * Copyright (C) 1997-2003
3 * Sony Computer Science Laboratories Inc. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $KAME: altq_subr.c,v 1.21 2003/11/06 06:32:53 kjc Exp $
27 */
28
29 #include "opt_altq.h"
30 #include "opt_inet.h"
31 #include "opt_inet6.h"
32
33 #include <sys/param.h>
34 #include <sys/malloc.h>
35 #include <sys/mbuf.h>
36 #include <sys/systm.h>
37 #include <sys/proc.h>
38 #include <sys/socket.h>
39 #include <sys/socketvar.h>
40 #include <sys/kernel.h>
41 #include <sys/errno.h>
42 #include <sys/syslog.h>
43 #include <sys/sysctl.h>
44 #include <sys/queue.h>
45
46 #include <net/if.h>
47 #include <net/if_var.h>
48 #include <net/if_dl.h>
49 #include <net/if_types.h>
50 #include <net/vnet.h>
51
52 #include <netinet/in.h>
53 #include <netinet/in_systm.h>
54 #include <netinet/ip.h>
55 #ifdef INET6
56 #include <netinet/ip6.h>
57 #endif
58 #include <netinet/tcp.h>
59 #include <netinet/udp.h>
60
61 #include <netpfil/pf/pf.h>
62 #include <netpfil/pf/pf_altq.h>
63 #include <net/altq/altq.h>
64
65 /* machine dependent clock related includes */
66 #include <sys/bus.h>
67 #include <sys/cpu.h>
68 #include <sys/eventhandler.h>
69 #include <machine/clock.h>
70 #if defined(__amd64__) || defined(__i386__)
71 #include <machine/cpufunc.h> /* for pentium tsc */
72 #include <machine/specialreg.h> /* for CPUID_TSC */
73 #include <machine/md_var.h> /* for cpu_feature */
74 #endif /* __amd64 || __i386__ */
75
76 /*
77 * internal function prototypes
78 */
79 static void tbr_timeout(void *);
80 int (*altq_input)(struct mbuf *, int) = NULL;
81 static struct mbuf *tbr_dequeue(struct ifaltq *, int);
82 static int tbr_timer = 0; /* token bucket regulator timer */
83 #if !defined(__FreeBSD__) || (__FreeBSD_version < 600000)
84 static struct callout tbr_callout = CALLOUT_INITIALIZER;
85 #else
86 static struct callout tbr_callout;
87 #endif
88
89 #ifdef ALTQ3_CLFIER_COMPAT
90 static int extract_ports4(struct mbuf *, struct ip *, struct flowinfo_in *);
91 #ifdef INET6
92 static int extract_ports6(struct mbuf *, struct ip6_hdr *,
93 struct flowinfo_in6 *);
94 #endif
95 static int apply_filter4(u_int32_t, struct flow_filter *,
96 struct flowinfo_in *);
97 static int apply_ppfilter4(u_int32_t, struct flow_filter *,
98 struct flowinfo_in *);
99 #ifdef INET6
100 static int apply_filter6(u_int32_t, struct flow_filter6 *,
101 struct flowinfo_in6 *);
102 #endif
103 static int apply_tosfilter4(u_int32_t, struct flow_filter *,
104 struct flowinfo_in *);
105 static u_long get_filt_handle(struct acc_classifier *, int);
106 static struct acc_filter *filth_to_filtp(struct acc_classifier *, u_long);
107 static u_int32_t filt2fibmask(struct flow_filter *);
108
109 static void ip4f_cache(struct ip *, struct flowinfo_in *);
110 static int ip4f_lookup(struct ip *, struct flowinfo_in *);
111 static int ip4f_init(void);
112 static struct ip4_frag *ip4f_alloc(void);
113 static void ip4f_free(struct ip4_frag *);
114 #endif /* ALTQ3_CLFIER_COMPAT */
115
116 #ifdef ALTQ
117 SYSCTL_NODE(_kern_features, OID_AUTO, altq, CTLFLAG_RD | CTLFLAG_CAPRD, 0,
118 "ALTQ packet queuing");
119
120 #define ALTQ_FEATURE(name, desc) \
121 SYSCTL_INT_WITH_LABEL(_kern_features_altq, OID_AUTO, name, \
122 CTLFLAG_RD | CTLFLAG_CAPRD, SYSCTL_NULL_INT_PTR, 1, \
123 desc, "feature")
124
125 #ifdef ALTQ_CBQ
126 ALTQ_FEATURE(cbq, "ALTQ Class Based Queuing discipline");
127 #endif
128 #ifdef ALTQ_CODEL
129 ALTQ_FEATURE(codel, "ALTQ Controlled Delay discipline");
130 #endif
131 #ifdef ALTQ_RED
132 ALTQ_FEATURE(red, "ALTQ Random Early Detection discipline");
133 #endif
134 #ifdef ALTQ_RIO
135 ALTQ_FEATURE(rio, "ALTQ Random Early Drop discipline");
136 #endif
137 #ifdef ALTQ_HFSC
138 ALTQ_FEATURE(hfsc, "ALTQ Hierarchical Packet Scheduler discipline");
139 #endif
140 #ifdef ALTQ_PRIQ
141 ALTQ_FEATURE(priq, "ATLQ Priority Queuing discipline");
142 #endif
143 #ifdef ALTQ_FAIRQ
144 ALTQ_FEATURE(fairq, "ALTQ Fair Queuing discipline");
145 #endif
146 #endif
147
148 /*
149 * alternate queueing support routines
150 */
151
152 /* look up the queue state by the interface name and the queueing type. */
153 void *
altq_lookup(name,type)154 altq_lookup(name, type)
155 char *name;
156 int type;
157 {
158 struct ifnet *ifp;
159
160 if ((ifp = ifunit(name)) != NULL) {
161 /* read if_snd unlocked */
162 if (type != ALTQT_NONE && ifp->if_snd.altq_type == type)
163 return (ifp->if_snd.altq_disc);
164 }
165
166 return NULL;
167 }
168
169 int
altq_attach(ifq,type,discipline,enqueue,dequeue,request,clfier,classify)170 altq_attach(ifq, type, discipline, enqueue, dequeue, request, clfier, classify)
171 struct ifaltq *ifq;
172 int type;
173 void *discipline;
174 int (*enqueue)(struct ifaltq *, struct mbuf *, struct altq_pktattr *);
175 struct mbuf *(*dequeue)(struct ifaltq *, int);
176 int (*request)(struct ifaltq *, int, void *);
177 void *clfier;
178 void *(*classify)(void *, struct mbuf *, int);
179 {
180 IFQ_LOCK(ifq);
181 if (!ALTQ_IS_READY(ifq)) {
182 IFQ_UNLOCK(ifq);
183 return ENXIO;
184 }
185
186 ifq->altq_type = type;
187 ifq->altq_disc = discipline;
188 ifq->altq_enqueue = enqueue;
189 ifq->altq_dequeue = dequeue;
190 ifq->altq_request = request;
191 ifq->altq_clfier = clfier;
192 ifq->altq_classify = classify;
193 ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED);
194 IFQ_UNLOCK(ifq);
195 return 0;
196 }
197
198 int
altq_detach(ifq)199 altq_detach(ifq)
200 struct ifaltq *ifq;
201 {
202 IFQ_LOCK(ifq);
203
204 if (!ALTQ_IS_READY(ifq)) {
205 IFQ_UNLOCK(ifq);
206 return ENXIO;
207 }
208 if (ALTQ_IS_ENABLED(ifq)) {
209 IFQ_UNLOCK(ifq);
210 return EBUSY;
211 }
212 if (!ALTQ_IS_ATTACHED(ifq)) {
213 IFQ_UNLOCK(ifq);
214 return (0);
215 }
216
217 ifq->altq_type = ALTQT_NONE;
218 ifq->altq_disc = NULL;
219 ifq->altq_enqueue = NULL;
220 ifq->altq_dequeue = NULL;
221 ifq->altq_request = NULL;
222 ifq->altq_clfier = NULL;
223 ifq->altq_classify = NULL;
224 ifq->altq_flags &= ALTQF_CANTCHANGE;
225
226 IFQ_UNLOCK(ifq);
227 return 0;
228 }
229
230 int
altq_enable(ifq)231 altq_enable(ifq)
232 struct ifaltq *ifq;
233 {
234 int s;
235
236 IFQ_LOCK(ifq);
237
238 if (!ALTQ_IS_READY(ifq)) {
239 IFQ_UNLOCK(ifq);
240 return ENXIO;
241 }
242 if (ALTQ_IS_ENABLED(ifq)) {
243 IFQ_UNLOCK(ifq);
244 return 0;
245 }
246
247 s = splnet();
248 IFQ_PURGE_NOLOCK(ifq);
249 ASSERT(ifq->ifq_len == 0);
250 ifq->ifq_drv_maxlen = 0; /* disable bulk dequeue */
251 ifq->altq_flags |= ALTQF_ENABLED;
252 if (ifq->altq_clfier != NULL)
253 ifq->altq_flags |= ALTQF_CLASSIFY;
254 splx(s);
255
256 IFQ_UNLOCK(ifq);
257 return 0;
258 }
259
260 int
altq_disable(ifq)261 altq_disable(ifq)
262 struct ifaltq *ifq;
263 {
264 int s;
265
266 IFQ_LOCK(ifq);
267 if (!ALTQ_IS_ENABLED(ifq)) {
268 IFQ_UNLOCK(ifq);
269 return 0;
270 }
271
272 s = splnet();
273 IFQ_PURGE_NOLOCK(ifq);
274 ASSERT(ifq->ifq_len == 0);
275 ifq->altq_flags &= ~(ALTQF_ENABLED|ALTQF_CLASSIFY);
276 splx(s);
277
278 IFQ_UNLOCK(ifq);
279 return 0;
280 }
281
282 #ifdef ALTQ_DEBUG
283 void
altq_assert(file,line,failedexpr)284 altq_assert(file, line, failedexpr)
285 const char *file, *failedexpr;
286 int line;
287 {
288 (void)printf("altq assertion \"%s\" failed: file \"%s\", line %d\n",
289 failedexpr, file, line);
290 panic("altq assertion");
291 /* NOTREACHED */
292 }
293 #endif
294
295 /*
296 * internal representation of token bucket parameters
297 * rate: (byte_per_unittime << TBR_SHIFT) / machclk_freq
298 * (((bits_per_sec) / 8) << TBR_SHIFT) / machclk_freq
299 * depth: byte << TBR_SHIFT
300 *
301 */
302 #define TBR_SHIFT 29
303 #define TBR_SCALE(x) ((int64_t)(x) << TBR_SHIFT)
304 #define TBR_UNSCALE(x) ((x) >> TBR_SHIFT)
305
306 static struct mbuf *
tbr_dequeue(ifq,op)307 tbr_dequeue(ifq, op)
308 struct ifaltq *ifq;
309 int op;
310 {
311 struct tb_regulator *tbr;
312 struct mbuf *m;
313 int64_t interval;
314 u_int64_t now;
315
316 IFQ_LOCK_ASSERT(ifq);
317 tbr = ifq->altq_tbr;
318 if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) {
319 /* if this is a remove after poll, bypass tbr check */
320 } else {
321 /* update token only when it is negative */
322 if (tbr->tbr_token <= 0) {
323 now = read_machclk();
324 interval = now - tbr->tbr_last;
325 if (interval >= tbr->tbr_filluptime)
326 tbr->tbr_token = tbr->tbr_depth;
327 else {
328 tbr->tbr_token += interval * tbr->tbr_rate;
329 if (tbr->tbr_token > tbr->tbr_depth)
330 tbr->tbr_token = tbr->tbr_depth;
331 }
332 tbr->tbr_last = now;
333 }
334 /* if token is still negative, don't allow dequeue */
335 if (tbr->tbr_token <= 0)
336 return (NULL);
337 }
338
339 if (ALTQ_IS_ENABLED(ifq))
340 m = (*ifq->altq_dequeue)(ifq, op);
341 else {
342 if (op == ALTDQ_POLL)
343 _IF_POLL(ifq, m);
344 else
345 _IF_DEQUEUE(ifq, m);
346 }
347
348 if (m != NULL && op == ALTDQ_REMOVE)
349 tbr->tbr_token -= TBR_SCALE(m_pktlen(m));
350 tbr->tbr_lastop = op;
351 return (m);
352 }
353
354 /*
355 * set a token bucket regulator.
356 * if the specified rate is zero, the token bucket regulator is deleted.
357 */
358 int
tbr_set(ifq,profile)359 tbr_set(ifq, profile)
360 struct ifaltq *ifq;
361 struct tb_profile *profile;
362 {
363 struct tb_regulator *tbr, *otbr;
364
365 if (tbr_dequeue_ptr == NULL)
366 tbr_dequeue_ptr = tbr_dequeue;
367
368 if (machclk_freq == 0)
369 init_machclk();
370 if (machclk_freq == 0) {
371 printf("tbr_set: no cpu clock available!\n");
372 return (ENXIO);
373 }
374
375 IFQ_LOCK(ifq);
376 if (profile->rate == 0) {
377 /* delete this tbr */
378 if ((tbr = ifq->altq_tbr) == NULL) {
379 IFQ_UNLOCK(ifq);
380 return (ENOENT);
381 }
382 ifq->altq_tbr = NULL;
383 free(tbr, M_DEVBUF);
384 IFQ_UNLOCK(ifq);
385 return (0);
386 }
387
388 tbr = malloc(sizeof(struct tb_regulator), M_DEVBUF, M_NOWAIT | M_ZERO);
389 if (tbr == NULL) {
390 IFQ_UNLOCK(ifq);
391 return (ENOMEM);
392 }
393
394 tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq;
395 tbr->tbr_depth = TBR_SCALE(profile->depth);
396 if (tbr->tbr_rate > 0)
397 tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate;
398 else
399 tbr->tbr_filluptime = LLONG_MAX;
400 /*
401 * The longest time between tbr_dequeue() calls will be about 1
402 * system tick, as the callout that drives it is scheduled once per
403 * tick. The refill-time detection logic in tbr_dequeue() can only
404 * properly detect the passage of up to LLONG_MAX machclk ticks.
405 * Therefore, in order for this logic to function properly in the
406 * extreme case, the maximum value of tbr_filluptime should be
407 * LLONG_MAX less one system tick's worth of machclk ticks less
408 * some additional slop factor (here one more system tick's worth
409 * of machclk ticks).
410 */
411 if (tbr->tbr_filluptime > (LLONG_MAX - 2 * machclk_per_tick))
412 tbr->tbr_filluptime = LLONG_MAX - 2 * machclk_per_tick;
413 tbr->tbr_token = tbr->tbr_depth;
414 tbr->tbr_last = read_machclk();
415 tbr->tbr_lastop = ALTDQ_REMOVE;
416
417 otbr = ifq->altq_tbr;
418 ifq->altq_tbr = tbr; /* set the new tbr */
419
420 if (otbr != NULL)
421 free(otbr, M_DEVBUF);
422 else {
423 if (tbr_timer == 0) {
424 CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
425 tbr_timer = 1;
426 }
427 }
428 IFQ_UNLOCK(ifq);
429 return (0);
430 }
431
432 /*
433 * tbr_timeout goes through the interface list, and kicks the drivers
434 * if necessary.
435 *
436 * MPSAFE
437 */
438 static void
tbr_timeout(arg)439 tbr_timeout(arg)
440 void *arg;
441 {
442 VNET_ITERATOR_DECL(vnet_iter);
443 struct ifnet *ifp;
444 struct epoch_tracker et;
445 int active;
446
447 active = 0;
448 NET_EPOCH_ENTER(et);
449 VNET_LIST_RLOCK_NOSLEEP();
450 VNET_FOREACH(vnet_iter) {
451 CURVNET_SET(vnet_iter);
452 for (ifp = CK_STAILQ_FIRST(&V_ifnet); ifp;
453 ifp = CK_STAILQ_NEXT(ifp, if_link)) {
454 /* read from if_snd unlocked */
455 if (!TBR_IS_ENABLED(&ifp->if_snd))
456 continue;
457 active++;
458 if (!IFQ_IS_EMPTY(&ifp->if_snd) &&
459 ifp->if_start != NULL)
460 (*ifp->if_start)(ifp);
461 }
462 CURVNET_RESTORE();
463 }
464 VNET_LIST_RUNLOCK_NOSLEEP();
465 NET_EPOCH_EXIT(et);
466 if (active > 0)
467 CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0);
468 else
469 tbr_timer = 0; /* don't need tbr_timer anymore */
470 }
471
472 /*
473 * attach a discipline to the interface. if one already exists, it is
474 * overridden.
475 * Locking is done in the discipline specific attach functions. Basically
476 * they call back to altq_attach which takes care of the attach and locking.
477 */
478 int
altq_pfattach(struct pf_altq * a)479 altq_pfattach(struct pf_altq *a)
480 {
481 int error = 0;
482
483 switch (a->scheduler) {
484 case ALTQT_NONE:
485 break;
486 #ifdef ALTQ_CBQ
487 case ALTQT_CBQ:
488 error = cbq_pfattach(a);
489 break;
490 #endif
491 #ifdef ALTQ_PRIQ
492 case ALTQT_PRIQ:
493 error = priq_pfattach(a);
494 break;
495 #endif
496 #ifdef ALTQ_HFSC
497 case ALTQT_HFSC:
498 error = hfsc_pfattach(a);
499 break;
500 #endif
501 #ifdef ALTQ_FAIRQ
502 case ALTQT_FAIRQ:
503 error = fairq_pfattach(a);
504 break;
505 #endif
506 #ifdef ALTQ_CODEL
507 case ALTQT_CODEL:
508 error = codel_pfattach(a);
509 break;
510 #endif
511 default:
512 error = ENXIO;
513 }
514
515 return (error);
516 }
517
518 /*
519 * detach a discipline from the interface.
520 * it is possible that the discipline was already overridden by another
521 * discipline.
522 */
523 int
altq_pfdetach(struct pf_altq * a)524 altq_pfdetach(struct pf_altq *a)
525 {
526 struct ifnet *ifp;
527 int s, error = 0;
528
529 if ((ifp = ifunit(a->ifname)) == NULL)
530 return (EINVAL);
531
532 /* if this discipline is no longer referenced, just return */
533 /* read unlocked from if_snd */
534 if (a->altq_disc == NULL || a->altq_disc != ifp->if_snd.altq_disc)
535 return (0);
536
537 s = splnet();
538 /* read unlocked from if_snd, _disable and _detach take care */
539 if (ALTQ_IS_ENABLED(&ifp->if_snd))
540 error = altq_disable(&ifp->if_snd);
541 if (error == 0)
542 error = altq_detach(&ifp->if_snd);
543 splx(s);
544
545 return (error);
546 }
547
548 /*
549 * add a discipline or a queue
550 * Locking is done in the discipline specific functions with regards to
551 * malloc with WAITOK, also it is not yet clear which lock to use.
552 */
553 int
altq_add(struct ifnet * ifp,struct pf_altq * a)554 altq_add(struct ifnet *ifp, struct pf_altq *a)
555 {
556 int error = 0;
557
558 if (a->qname[0] != 0)
559 return (altq_add_queue(a));
560
561 if (machclk_freq == 0)
562 init_machclk();
563 if (machclk_freq == 0)
564 panic("altq_add: no cpu clock");
565
566 switch (a->scheduler) {
567 #ifdef ALTQ_CBQ
568 case ALTQT_CBQ:
569 error = cbq_add_altq(ifp, a);
570 break;
571 #endif
572 #ifdef ALTQ_PRIQ
573 case ALTQT_PRIQ:
574 error = priq_add_altq(ifp, a);
575 break;
576 #endif
577 #ifdef ALTQ_HFSC
578 case ALTQT_HFSC:
579 error = hfsc_add_altq(ifp, a);
580 break;
581 #endif
582 #ifdef ALTQ_FAIRQ
583 case ALTQT_FAIRQ:
584 error = fairq_add_altq(ifp, a);
585 break;
586 #endif
587 #ifdef ALTQ_CODEL
588 case ALTQT_CODEL:
589 error = codel_add_altq(ifp, a);
590 break;
591 #endif
592 default:
593 error = ENXIO;
594 }
595
596 return (error);
597 }
598
599 /*
600 * remove a discipline or a queue
601 * It is yet unclear what lock to use to protect this operation, the
602 * discipline specific functions will determine and grab it
603 */
604 int
altq_remove(struct pf_altq * a)605 altq_remove(struct pf_altq *a)
606 {
607 int error = 0;
608
609 if (a->qname[0] != 0)
610 return (altq_remove_queue(a));
611
612 switch (a->scheduler) {
613 #ifdef ALTQ_CBQ
614 case ALTQT_CBQ:
615 error = cbq_remove_altq(a);
616 break;
617 #endif
618 #ifdef ALTQ_PRIQ
619 case ALTQT_PRIQ:
620 error = priq_remove_altq(a);
621 break;
622 #endif
623 #ifdef ALTQ_HFSC
624 case ALTQT_HFSC:
625 error = hfsc_remove_altq(a);
626 break;
627 #endif
628 #ifdef ALTQ_FAIRQ
629 case ALTQT_FAIRQ:
630 error = fairq_remove_altq(a);
631 break;
632 #endif
633 #ifdef ALTQ_CODEL
634 case ALTQT_CODEL:
635 error = codel_remove_altq(a);
636 break;
637 #endif
638 default:
639 error = ENXIO;
640 }
641
642 return (error);
643 }
644
645 /*
646 * add a queue to the discipline
647 * It is yet unclear what lock to use to protect this operation, the
648 * discipline specific functions will determine and grab it
649 */
650 int
altq_add_queue(struct pf_altq * a)651 altq_add_queue(struct pf_altq *a)
652 {
653 int error = 0;
654
655 switch (a->scheduler) {
656 #ifdef ALTQ_CBQ
657 case ALTQT_CBQ:
658 error = cbq_add_queue(a);
659 break;
660 #endif
661 #ifdef ALTQ_PRIQ
662 case ALTQT_PRIQ:
663 error = priq_add_queue(a);
664 break;
665 #endif
666 #ifdef ALTQ_HFSC
667 case ALTQT_HFSC:
668 error = hfsc_add_queue(a);
669 break;
670 #endif
671 #ifdef ALTQ_FAIRQ
672 case ALTQT_FAIRQ:
673 error = fairq_add_queue(a);
674 break;
675 #endif
676 default:
677 error = ENXIO;
678 }
679
680 return (error);
681 }
682
683 /*
684 * remove a queue from the discipline
685 * It is yet unclear what lock to use to protect this operation, the
686 * discipline specific functions will determine and grab it
687 */
688 int
altq_remove_queue(struct pf_altq * a)689 altq_remove_queue(struct pf_altq *a)
690 {
691 int error = 0;
692
693 switch (a->scheduler) {
694 #ifdef ALTQ_CBQ
695 case ALTQT_CBQ:
696 error = cbq_remove_queue(a);
697 break;
698 #endif
699 #ifdef ALTQ_PRIQ
700 case ALTQT_PRIQ:
701 error = priq_remove_queue(a);
702 break;
703 #endif
704 #ifdef ALTQ_HFSC
705 case ALTQT_HFSC:
706 error = hfsc_remove_queue(a);
707 break;
708 #endif
709 #ifdef ALTQ_FAIRQ
710 case ALTQT_FAIRQ:
711 error = fairq_remove_queue(a);
712 break;
713 #endif
714 default:
715 error = ENXIO;
716 }
717
718 return (error);
719 }
720
721 /*
722 * get queue statistics
723 * Locking is done in the discipline specific functions with regards to
724 * copyout operations, also it is not yet clear which lock to use.
725 */
726 int
altq_getqstats(struct pf_altq * a,void * ubuf,int * nbytes,int version)727 altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version)
728 {
729 int error = 0;
730
731 switch (a->scheduler) {
732 #ifdef ALTQ_CBQ
733 case ALTQT_CBQ:
734 error = cbq_getqstats(a, ubuf, nbytes, version);
735 break;
736 #endif
737 #ifdef ALTQ_PRIQ
738 case ALTQT_PRIQ:
739 error = priq_getqstats(a, ubuf, nbytes, version);
740 break;
741 #endif
742 #ifdef ALTQ_HFSC
743 case ALTQT_HFSC:
744 error = hfsc_getqstats(a, ubuf, nbytes, version);
745 break;
746 #endif
747 #ifdef ALTQ_FAIRQ
748 case ALTQT_FAIRQ:
749 error = fairq_getqstats(a, ubuf, nbytes, version);
750 break;
751 #endif
752 #ifdef ALTQ_CODEL
753 case ALTQT_CODEL:
754 error = codel_getqstats(a, ubuf, nbytes, version);
755 break;
756 #endif
757 default:
758 error = ENXIO;
759 }
760
761 return (error);
762 }
763
764 /*
765 * read and write diffserv field in IPv4 or IPv6 header
766 */
767 u_int8_t
read_dsfield(m,pktattr)768 read_dsfield(m, pktattr)
769 struct mbuf *m;
770 struct altq_pktattr *pktattr;
771 {
772 struct mbuf *m0;
773 u_int8_t ds_field = 0;
774
775 if (pktattr == NULL ||
776 (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
777 return ((u_int8_t)0);
778
779 /* verify that pattr_hdr is within the mbuf data */
780 for (m0 = m; m0 != NULL; m0 = m0->m_next)
781 if ((pktattr->pattr_hdr >= m0->m_data) &&
782 (pktattr->pattr_hdr < m0->m_data + m0->m_len))
783 break;
784 if (m0 == NULL) {
785 /* ick, pattr_hdr is stale */
786 pktattr->pattr_af = AF_UNSPEC;
787 #ifdef ALTQ_DEBUG
788 printf("read_dsfield: can't locate header!\n");
789 #endif
790 return ((u_int8_t)0);
791 }
792
793 if (pktattr->pattr_af == AF_INET) {
794 struct ip *ip = (struct ip *)pktattr->pattr_hdr;
795
796 if (ip->ip_v != 4)
797 return ((u_int8_t)0); /* version mismatch! */
798 ds_field = ip->ip_tos;
799 }
800 #ifdef INET6
801 else if (pktattr->pattr_af == AF_INET6) {
802 struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
803 u_int32_t flowlabel;
804
805 flowlabel = ntohl(ip6->ip6_flow);
806 if ((flowlabel >> 28) != 6)
807 return ((u_int8_t)0); /* version mismatch! */
808 ds_field = (flowlabel >> 20) & 0xff;
809 }
810 #endif
811 return (ds_field);
812 }
813
814 void
write_dsfield(struct mbuf * m,struct altq_pktattr * pktattr,u_int8_t dsfield)815 write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, u_int8_t dsfield)
816 {
817 struct mbuf *m0;
818
819 if (pktattr == NULL ||
820 (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6))
821 return;
822
823 /* verify that pattr_hdr is within the mbuf data */
824 for (m0 = m; m0 != NULL; m0 = m0->m_next)
825 if ((pktattr->pattr_hdr >= m0->m_data) &&
826 (pktattr->pattr_hdr < m0->m_data + m0->m_len))
827 break;
828 if (m0 == NULL) {
829 /* ick, pattr_hdr is stale */
830 pktattr->pattr_af = AF_UNSPEC;
831 #ifdef ALTQ_DEBUG
832 printf("write_dsfield: can't locate header!\n");
833 #endif
834 return;
835 }
836
837 if (pktattr->pattr_af == AF_INET) {
838 struct ip *ip = (struct ip *)pktattr->pattr_hdr;
839 u_int8_t old;
840 int32_t sum;
841
842 if (ip->ip_v != 4)
843 return; /* version mismatch! */
844 old = ip->ip_tos;
845 dsfield |= old & 3; /* leave CU bits */
846 if (old == dsfield)
847 return;
848 ip->ip_tos = dsfield;
849 /*
850 * update checksum (from RFC1624)
851 * HC' = ~(~HC + ~m + m')
852 */
853 sum = ~ntohs(ip->ip_sum) & 0xffff;
854 sum += 0xff00 + (~old & 0xff) + dsfield;
855 sum = (sum >> 16) + (sum & 0xffff);
856 sum += (sum >> 16); /* add carry */
857
858 ip->ip_sum = htons(~sum & 0xffff);
859 }
860 #ifdef INET6
861 else if (pktattr->pattr_af == AF_INET6) {
862 struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr;
863 u_int32_t flowlabel;
864
865 flowlabel = ntohl(ip6->ip6_flow);
866 if ((flowlabel >> 28) != 6)
867 return; /* version mismatch! */
868 flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20);
869 ip6->ip6_flow = htonl(flowlabel);
870 }
871 #endif
872 return;
873 }
874
875 /*
876 * high resolution clock support taking advantage of a machine dependent
877 * high resolution time counter (e.g., timestamp counter of intel pentium).
878 * we assume
879 * - 64-bit-long monotonically-increasing counter
880 * - frequency range is 100M-4GHz (CPU speed)
881 */
882 /* if pcc is not available or disabled, emulate 256MHz using microtime() */
883 #define MACHCLK_SHIFT 8
884
885 int machclk_usepcc;
886 u_int32_t machclk_freq;
887 u_int32_t machclk_per_tick;
888
889 #if defined(__i386__) && defined(__NetBSD__)
890 extern u_int64_t cpu_tsc_freq;
891 #endif
892
893 #if (__FreeBSD_version >= 700035)
894 /* Update TSC freq with the value indicated by the caller. */
895 static void
tsc_freq_changed(void * arg,const struct cf_level * level,int status)896 tsc_freq_changed(void *arg, const struct cf_level *level, int status)
897 {
898 /* If there was an error during the transition, don't do anything. */
899 if (status != 0)
900 return;
901
902 #if (__FreeBSD_version >= 701102) && (defined(__amd64__) || defined(__i386__))
903 /* If TSC is P-state invariant, don't do anything. */
904 if (tsc_is_invariant)
905 return;
906 #endif
907
908 /* Total setting for this level gives the new frequency in MHz. */
909 init_machclk();
910 }
911 EVENTHANDLER_DEFINE(cpufreq_post_change, tsc_freq_changed, NULL,
912 EVENTHANDLER_PRI_LAST);
913 #endif /* __FreeBSD_version >= 700035 */
914
915 static void
init_machclk_setup(void)916 init_machclk_setup(void)
917 {
918 #if (__FreeBSD_version >= 600000)
919 callout_init(&tbr_callout, 0);
920 #endif
921
922 machclk_usepcc = 1;
923
924 #if (!defined(__amd64__) && !defined(__i386__)) || defined(ALTQ_NOPCC)
925 machclk_usepcc = 0;
926 #endif
927 #if defined(__FreeBSD__) && defined(SMP)
928 machclk_usepcc = 0;
929 #endif
930 #if defined(__NetBSD__) && defined(MULTIPROCESSOR)
931 machclk_usepcc = 0;
932 #endif
933 #if defined(__amd64__) || defined(__i386__)
934 /* check if TSC is available */
935 if ((cpu_feature & CPUID_TSC) == 0 ||
936 atomic_load_acq_64(&tsc_freq) == 0)
937 machclk_usepcc = 0;
938 #endif
939 }
940
941 void
init_machclk(void)942 init_machclk(void)
943 {
944 static int called;
945
946 /* Call one-time initialization function. */
947 if (!called) {
948 init_machclk_setup();
949 called = 1;
950 }
951
952 if (machclk_usepcc == 0) {
953 /* emulate 256MHz using microtime() */
954 machclk_freq = 1000000 << MACHCLK_SHIFT;
955 machclk_per_tick = machclk_freq / hz;
956 #ifdef ALTQ_DEBUG
957 printf("altq: emulate %uHz cpu clock\n", machclk_freq);
958 #endif
959 return;
960 }
961
962 /*
963 * if the clock frequency (of Pentium TSC or Alpha PCC) is
964 * accessible, just use it.
965 */
966 #if defined(__amd64__) || defined(__i386__)
967 machclk_freq = atomic_load_acq_64(&tsc_freq);
968 #endif
969
970 /*
971 * if we don't know the clock frequency, measure it.
972 */
973 if (machclk_freq == 0) {
974 static int wait;
975 struct timeval tv_start, tv_end;
976 u_int64_t start, end, diff;
977 int timo;
978
979 microtime(&tv_start);
980 start = read_machclk();
981 timo = hz; /* 1 sec */
982 (void)tsleep(&wait, PWAIT | PCATCH, "init_machclk", timo);
983 microtime(&tv_end);
984 end = read_machclk();
985 diff = (u_int64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000
986 + tv_end.tv_usec - tv_start.tv_usec;
987 if (diff != 0)
988 machclk_freq = (u_int)((end - start) * 1000000 / diff);
989 }
990
991 machclk_per_tick = machclk_freq / hz;
992
993 #ifdef ALTQ_DEBUG
994 printf("altq: CPU clock: %uHz\n", machclk_freq);
995 #endif
996 }
997
998 #if defined(__OpenBSD__) && defined(__i386__)
999 static __inline u_int64_t
rdtsc(void)1000 rdtsc(void)
1001 {
1002 u_int64_t rv;
1003 __asm __volatile(".byte 0x0f, 0x31" : "=A" (rv));
1004 return (rv);
1005 }
1006 #endif /* __OpenBSD__ && __i386__ */
1007
1008 u_int64_t
read_machclk(void)1009 read_machclk(void)
1010 {
1011 u_int64_t val;
1012
1013 if (machclk_usepcc) {
1014 #if defined(__amd64__) || defined(__i386__)
1015 val = rdtsc();
1016 #else
1017 panic("read_machclk");
1018 #endif
1019 } else {
1020 struct timeval tv, boottime;
1021
1022 microtime(&tv);
1023 getboottime(&boottime);
1024 val = (((u_int64_t)(tv.tv_sec - boottime.tv_sec) * 1000000
1025 + tv.tv_usec) << MACHCLK_SHIFT);
1026 }
1027 return (val);
1028 }
1029
1030 #ifdef ALTQ3_CLFIER_COMPAT
1031
1032 #ifndef IPPROTO_ESP
1033 #define IPPROTO_ESP 50 /* encapsulating security payload */
1034 #endif
1035 #ifndef IPPROTO_AH
1036 #define IPPROTO_AH 51 /* authentication header */
1037 #endif
1038
1039 /*
1040 * extract flow information from a given packet.
1041 * filt_mask shows flowinfo fields required.
1042 * we assume the ip header is in one mbuf, and addresses and ports are
1043 * in network byte order.
1044 */
1045 int
altq_extractflow(m,af,flow,filt_bmask)1046 altq_extractflow(m, af, flow, filt_bmask)
1047 struct mbuf *m;
1048 int af;
1049 struct flowinfo *flow;
1050 u_int32_t filt_bmask;
1051 {
1052
1053 switch (af) {
1054 case PF_INET: {
1055 struct flowinfo_in *fin;
1056 struct ip *ip;
1057
1058 ip = mtod(m, struct ip *);
1059
1060 if (ip->ip_v != 4)
1061 break;
1062
1063 fin = (struct flowinfo_in *)flow;
1064 fin->fi_len = sizeof(struct flowinfo_in);
1065 fin->fi_family = AF_INET;
1066
1067 fin->fi_proto = ip->ip_p;
1068 fin->fi_tos = ip->ip_tos;
1069
1070 fin->fi_src.s_addr = ip->ip_src.s_addr;
1071 fin->fi_dst.s_addr = ip->ip_dst.s_addr;
1072
1073 if (filt_bmask & FIMB4_PORTS)
1074 /* if port info is required, extract port numbers */
1075 extract_ports4(m, ip, fin);
1076 else {
1077 fin->fi_sport = 0;
1078 fin->fi_dport = 0;
1079 fin->fi_gpi = 0;
1080 }
1081 return (1);
1082 }
1083
1084 #ifdef INET6
1085 case PF_INET6: {
1086 struct flowinfo_in6 *fin6;
1087 struct ip6_hdr *ip6;
1088
1089 ip6 = mtod(m, struct ip6_hdr *);
1090 /* should we check the ip version? */
1091
1092 fin6 = (struct flowinfo_in6 *)flow;
1093 fin6->fi6_len = sizeof(struct flowinfo_in6);
1094 fin6->fi6_family = AF_INET6;
1095
1096 fin6->fi6_proto = ip6->ip6_nxt;
1097 fin6->fi6_tclass = IPV6_TRAFFIC_CLASS(ip6);
1098
1099 fin6->fi6_flowlabel = ip6->ip6_flow & htonl(0x000fffff);
1100 fin6->fi6_src = ip6->ip6_src;
1101 fin6->fi6_dst = ip6->ip6_dst;
1102
1103 if ((filt_bmask & FIMB6_PORTS) ||
1104 ((filt_bmask & FIMB6_PROTO)
1105 && ip6->ip6_nxt > IPPROTO_IPV6))
1106 /*
1107 * if port info is required, or proto is required
1108 * but there are option headers, extract port
1109 * and protocol numbers.
1110 */
1111 extract_ports6(m, ip6, fin6);
1112 else {
1113 fin6->fi6_sport = 0;
1114 fin6->fi6_dport = 0;
1115 fin6->fi6_gpi = 0;
1116 }
1117 return (1);
1118 }
1119 #endif /* INET6 */
1120
1121 default:
1122 break;
1123 }
1124
1125 /* failed */
1126 flow->fi_len = sizeof(struct flowinfo);
1127 flow->fi_family = AF_UNSPEC;
1128 return (0);
1129 }
1130
1131 /*
1132 * helper routine to extract port numbers
1133 */
1134 /* structure for ipsec and ipv6 option header template */
1135 struct _opt6 {
1136 u_int8_t opt6_nxt; /* next header */
1137 u_int8_t opt6_hlen; /* header extension length */
1138 u_int16_t _pad;
1139 u_int32_t ah_spi; /* security parameter index
1140 for authentication header */
1141 };
1142
1143 /*
1144 * extract port numbers from a ipv4 packet.
1145 */
1146 static int
extract_ports4(m,ip,fin)1147 extract_ports4(m, ip, fin)
1148 struct mbuf *m;
1149 struct ip *ip;
1150 struct flowinfo_in *fin;
1151 {
1152 struct mbuf *m0;
1153 u_short ip_off;
1154 u_int8_t proto;
1155 int off;
1156
1157 fin->fi_sport = 0;
1158 fin->fi_dport = 0;
1159 fin->fi_gpi = 0;
1160
1161 ip_off = ntohs(ip->ip_off);
1162 /* if it is a fragment, try cached fragment info */
1163 if (ip_off & IP_OFFMASK) {
1164 ip4f_lookup(ip, fin);
1165 return (1);
1166 }
1167
1168 /* locate the mbuf containing the protocol header */
1169 for (m0 = m; m0 != NULL; m0 = m0->m_next)
1170 if (((caddr_t)ip >= m0->m_data) &&
1171 ((caddr_t)ip < m0->m_data + m0->m_len))
1172 break;
1173 if (m0 == NULL) {
1174 #ifdef ALTQ_DEBUG
1175 printf("extract_ports4: can't locate header! ip=%p\n", ip);
1176 #endif
1177 return (0);
1178 }
1179 off = ((caddr_t)ip - m0->m_data) + (ip->ip_hl << 2);
1180 proto = ip->ip_p;
1181
1182 #ifdef ALTQ_IPSEC
1183 again:
1184 #endif
1185 while (off >= m0->m_len) {
1186 off -= m0->m_len;
1187 m0 = m0->m_next;
1188 if (m0 == NULL)
1189 return (0); /* bogus ip_hl! */
1190 }
1191 if (m0->m_len < off + 4)
1192 return (0);
1193
1194 switch (proto) {
1195 case IPPROTO_TCP:
1196 case IPPROTO_UDP: {
1197 struct udphdr *udp;
1198
1199 udp = (struct udphdr *)(mtod(m0, caddr_t) + off);
1200 fin->fi_sport = udp->uh_sport;
1201 fin->fi_dport = udp->uh_dport;
1202 fin->fi_proto = proto;
1203 }
1204 break;
1205
1206 #ifdef ALTQ_IPSEC
1207 case IPPROTO_ESP:
1208 if (fin->fi_gpi == 0){
1209 u_int32_t *gpi;
1210
1211 gpi = (u_int32_t *)(mtod(m0, caddr_t) + off);
1212 fin->fi_gpi = *gpi;
1213 }
1214 fin->fi_proto = proto;
1215 break;
1216
1217 case IPPROTO_AH: {
1218 /* get next header and header length */
1219 struct _opt6 *opt6;
1220
1221 opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1222 proto = opt6->opt6_nxt;
1223 off += 8 + (opt6->opt6_hlen * 4);
1224 if (fin->fi_gpi == 0 && m0->m_len >= off + 8)
1225 fin->fi_gpi = opt6->ah_spi;
1226 }
1227 /* goto the next header */
1228 goto again;
1229 #endif /* ALTQ_IPSEC */
1230
1231 default:
1232 fin->fi_proto = proto;
1233 return (0);
1234 }
1235
1236 /* if this is a first fragment, cache it. */
1237 if (ip_off & IP_MF)
1238 ip4f_cache(ip, fin);
1239
1240 return (1);
1241 }
1242
1243 #ifdef INET6
1244 static int
extract_ports6(m,ip6,fin6)1245 extract_ports6(m, ip6, fin6)
1246 struct mbuf *m;
1247 struct ip6_hdr *ip6;
1248 struct flowinfo_in6 *fin6;
1249 {
1250 struct mbuf *m0;
1251 int off;
1252 u_int8_t proto;
1253
1254 fin6->fi6_gpi = 0;
1255 fin6->fi6_sport = 0;
1256 fin6->fi6_dport = 0;
1257
1258 /* locate the mbuf containing the protocol header */
1259 for (m0 = m; m0 != NULL; m0 = m0->m_next)
1260 if (((caddr_t)ip6 >= m0->m_data) &&
1261 ((caddr_t)ip6 < m0->m_data + m0->m_len))
1262 break;
1263 if (m0 == NULL) {
1264 #ifdef ALTQ_DEBUG
1265 printf("extract_ports6: can't locate header! ip6=%p\n", ip6);
1266 #endif
1267 return (0);
1268 }
1269 off = ((caddr_t)ip6 - m0->m_data) + sizeof(struct ip6_hdr);
1270
1271 proto = ip6->ip6_nxt;
1272 do {
1273 while (off >= m0->m_len) {
1274 off -= m0->m_len;
1275 m0 = m0->m_next;
1276 if (m0 == NULL)
1277 return (0);
1278 }
1279 if (m0->m_len < off + 4)
1280 return (0);
1281
1282 switch (proto) {
1283 case IPPROTO_TCP:
1284 case IPPROTO_UDP: {
1285 struct udphdr *udp;
1286
1287 udp = (struct udphdr *)(mtod(m0, caddr_t) + off);
1288 fin6->fi6_sport = udp->uh_sport;
1289 fin6->fi6_dport = udp->uh_dport;
1290 fin6->fi6_proto = proto;
1291 }
1292 return (1);
1293
1294 case IPPROTO_ESP:
1295 if (fin6->fi6_gpi == 0) {
1296 u_int32_t *gpi;
1297
1298 gpi = (u_int32_t *)(mtod(m0, caddr_t) + off);
1299 fin6->fi6_gpi = *gpi;
1300 }
1301 fin6->fi6_proto = proto;
1302 return (1);
1303
1304 case IPPROTO_AH: {
1305 /* get next header and header length */
1306 struct _opt6 *opt6;
1307
1308 opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1309 if (fin6->fi6_gpi == 0 && m0->m_len >= off + 8)
1310 fin6->fi6_gpi = opt6->ah_spi;
1311 proto = opt6->opt6_nxt;
1312 off += 8 + (opt6->opt6_hlen * 4);
1313 /* goto the next header */
1314 break;
1315 }
1316
1317 case IPPROTO_HOPOPTS:
1318 case IPPROTO_ROUTING:
1319 case IPPROTO_DSTOPTS: {
1320 /* get next header and header length */
1321 struct _opt6 *opt6;
1322
1323 opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off);
1324 proto = opt6->opt6_nxt;
1325 off += (opt6->opt6_hlen + 1) * 8;
1326 /* goto the next header */
1327 break;
1328 }
1329
1330 case IPPROTO_FRAGMENT:
1331 /* ipv6 fragmentations are not supported yet */
1332 default:
1333 fin6->fi6_proto = proto;
1334 return (0);
1335 }
1336 } while (1);
1337 /*NOTREACHED*/
1338 }
1339 #endif /* INET6 */
1340
1341 /*
1342 * altq common classifier
1343 */
1344 int
acc_add_filter(classifier,filter,class,phandle)1345 acc_add_filter(classifier, filter, class, phandle)
1346 struct acc_classifier *classifier;
1347 struct flow_filter *filter;
1348 void *class;
1349 u_long *phandle;
1350 {
1351 struct acc_filter *afp, *prev, *tmp;
1352 int i, s;
1353
1354 #ifdef INET6
1355 if (filter->ff_flow.fi_family != AF_INET &&
1356 filter->ff_flow.fi_family != AF_INET6)
1357 return (EINVAL);
1358 #else
1359 if (filter->ff_flow.fi_family != AF_INET)
1360 return (EINVAL);
1361 #endif
1362
1363 afp = malloc(sizeof(*afp), M_DEVBUF, M_WAITOK | M_ZERO);
1364 afp->f_filter = *filter;
1365 afp->f_class = class;
1366
1367 i = ACC_WILDCARD_INDEX;
1368 if (filter->ff_flow.fi_family == AF_INET) {
1369 struct flow_filter *filter4 = &afp->f_filter;
1370
1371 /*
1372 * if address is 0, it's a wildcard. if address mask
1373 * isn't set, use full mask.
1374 */
1375 if (filter4->ff_flow.fi_dst.s_addr == 0)
1376 filter4->ff_mask.mask_dst.s_addr = 0;
1377 else if (filter4->ff_mask.mask_dst.s_addr == 0)
1378 filter4->ff_mask.mask_dst.s_addr = 0xffffffff;
1379 if (filter4->ff_flow.fi_src.s_addr == 0)
1380 filter4->ff_mask.mask_src.s_addr = 0;
1381 else if (filter4->ff_mask.mask_src.s_addr == 0)
1382 filter4->ff_mask.mask_src.s_addr = 0xffffffff;
1383
1384 /* clear extra bits in addresses */
1385 filter4->ff_flow.fi_dst.s_addr &=
1386 filter4->ff_mask.mask_dst.s_addr;
1387 filter4->ff_flow.fi_src.s_addr &=
1388 filter4->ff_mask.mask_src.s_addr;
1389
1390 /*
1391 * if dst address is a wildcard, use hash-entry
1392 * ACC_WILDCARD_INDEX.
1393 */
1394 if (filter4->ff_mask.mask_dst.s_addr != 0xffffffff)
1395 i = ACC_WILDCARD_INDEX;
1396 else
1397 i = ACC_GET_HASH_INDEX(filter4->ff_flow.fi_dst.s_addr);
1398 }
1399 #ifdef INET6
1400 else if (filter->ff_flow.fi_family == AF_INET6) {
1401 struct flow_filter6 *filter6 =
1402 (struct flow_filter6 *)&afp->f_filter;
1403 #ifndef IN6MASK0 /* taken from kame ipv6 */
1404 #define IN6MASK0 {{{ 0, 0, 0, 0 }}}
1405 #define IN6MASK128 {{{ 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff }}}
1406 const struct in6_addr in6mask0 = IN6MASK0;
1407 const struct in6_addr in6mask128 = IN6MASK128;
1408 #endif
1409
1410 if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_dst))
1411 filter6->ff_mask6.mask6_dst = in6mask0;
1412 else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_dst))
1413 filter6->ff_mask6.mask6_dst = in6mask128;
1414 if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_src))
1415 filter6->ff_mask6.mask6_src = in6mask0;
1416 else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_src))
1417 filter6->ff_mask6.mask6_src = in6mask128;
1418
1419 /* clear extra bits in addresses */
1420 for (i = 0; i < 16; i++)
1421 filter6->ff_flow6.fi6_dst.s6_addr[i] &=
1422 filter6->ff_mask6.mask6_dst.s6_addr[i];
1423 for (i = 0; i < 16; i++)
1424 filter6->ff_flow6.fi6_src.s6_addr[i] &=
1425 filter6->ff_mask6.mask6_src.s6_addr[i];
1426
1427 if (filter6->ff_flow6.fi6_flowlabel == 0)
1428 i = ACC_WILDCARD_INDEX;
1429 else
1430 i = ACC_GET_HASH_INDEX(filter6->ff_flow6.fi6_flowlabel);
1431 }
1432 #endif /* INET6 */
1433
1434 afp->f_handle = get_filt_handle(classifier, i);
1435
1436 /* update filter bitmask */
1437 afp->f_fbmask = filt2fibmask(filter);
1438 classifier->acc_fbmask |= afp->f_fbmask;
1439
1440 /*
1441 * add this filter to the filter list.
1442 * filters are ordered from the highest rule number.
1443 */
1444 s = splnet();
1445 prev = NULL;
1446 LIST_FOREACH(tmp, &classifier->acc_filters[i], f_chain) {
1447 if (tmp->f_filter.ff_ruleno > afp->f_filter.ff_ruleno)
1448 prev = tmp;
1449 else
1450 break;
1451 }
1452 if (prev == NULL)
1453 LIST_INSERT_HEAD(&classifier->acc_filters[i], afp, f_chain);
1454 else
1455 LIST_INSERT_AFTER(prev, afp, f_chain);
1456 splx(s);
1457
1458 *phandle = afp->f_handle;
1459 return (0);
1460 }
1461
1462 int
acc_delete_filter(classifier,handle)1463 acc_delete_filter(classifier, handle)
1464 struct acc_classifier *classifier;
1465 u_long handle;
1466 {
1467 struct acc_filter *afp;
1468 int s;
1469
1470 if ((afp = filth_to_filtp(classifier, handle)) == NULL)
1471 return (EINVAL);
1472
1473 s = splnet();
1474 LIST_REMOVE(afp, f_chain);
1475 splx(s);
1476
1477 free(afp, M_DEVBUF);
1478
1479 /* todo: update filt_bmask */
1480
1481 return (0);
1482 }
1483
1484 /*
1485 * delete filters referencing to the specified class.
1486 * if the all flag is not 0, delete all the filters.
1487 */
1488 int
acc_discard_filters(classifier,class,all)1489 acc_discard_filters(classifier, class, all)
1490 struct acc_classifier *classifier;
1491 void *class;
1492 int all;
1493 {
1494 struct acc_filter *afp;
1495 int i, s;
1496
1497 s = splnet();
1498 for (i = 0; i < ACC_FILTER_TABLESIZE; i++) {
1499 do {
1500 LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1501 if (all || afp->f_class == class) {
1502 LIST_REMOVE(afp, f_chain);
1503 free(afp, M_DEVBUF);
1504 /* start again from the head */
1505 break;
1506 }
1507 } while (afp != NULL);
1508 }
1509 splx(s);
1510
1511 if (all)
1512 classifier->acc_fbmask = 0;
1513
1514 return (0);
1515 }
1516
1517 void *
acc_classify(clfier,m,af)1518 acc_classify(clfier, m, af)
1519 void *clfier;
1520 struct mbuf *m;
1521 int af;
1522 {
1523 struct acc_classifier *classifier;
1524 struct flowinfo flow;
1525 struct acc_filter *afp;
1526 int i;
1527
1528 classifier = (struct acc_classifier *)clfier;
1529 altq_extractflow(m, af, &flow, classifier->acc_fbmask);
1530
1531 if (flow.fi_family == AF_INET) {
1532 struct flowinfo_in *fp = (struct flowinfo_in *)&flow;
1533
1534 if ((classifier->acc_fbmask & FIMB4_ALL) == FIMB4_TOS) {
1535 /* only tos is used */
1536 LIST_FOREACH(afp,
1537 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1538 f_chain)
1539 if (apply_tosfilter4(afp->f_fbmask,
1540 &afp->f_filter, fp))
1541 /* filter matched */
1542 return (afp->f_class);
1543 } else if ((classifier->acc_fbmask &
1544 (~(FIMB4_PROTO|FIMB4_SPORT|FIMB4_DPORT) & FIMB4_ALL))
1545 == 0) {
1546 /* only proto and ports are used */
1547 LIST_FOREACH(afp,
1548 &classifier->acc_filters[ACC_WILDCARD_INDEX],
1549 f_chain)
1550 if (apply_ppfilter4(afp->f_fbmask,
1551 &afp->f_filter, fp))
1552 /* filter matched */
1553 return (afp->f_class);
1554 } else {
1555 /* get the filter hash entry from its dest address */
1556 i = ACC_GET_HASH_INDEX(fp->fi_dst.s_addr);
1557 do {
1558 /*
1559 * go through this loop twice. first for dst
1560 * hash, second for wildcards.
1561 */
1562 LIST_FOREACH(afp, &classifier->acc_filters[i],
1563 f_chain)
1564 if (apply_filter4(afp->f_fbmask,
1565 &afp->f_filter, fp))
1566 /* filter matched */
1567 return (afp->f_class);
1568
1569 /*
1570 * check again for filters with a dst addr
1571 * wildcard.
1572 * (daddr == 0 || dmask != 0xffffffff).
1573 */
1574 if (i != ACC_WILDCARD_INDEX)
1575 i = ACC_WILDCARD_INDEX;
1576 else
1577 break;
1578 } while (1);
1579 }
1580 }
1581 #ifdef INET6
1582 else if (flow.fi_family == AF_INET6) {
1583 struct flowinfo_in6 *fp6 = (struct flowinfo_in6 *)&flow;
1584
1585 /* get the filter hash entry from its flow ID */
1586 if (fp6->fi6_flowlabel != 0)
1587 i = ACC_GET_HASH_INDEX(fp6->fi6_flowlabel);
1588 else
1589 /* flowlable can be zero */
1590 i = ACC_WILDCARD_INDEX;
1591
1592 /* go through this loop twice. first for flow hash, second
1593 for wildcards. */
1594 do {
1595 LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1596 if (apply_filter6(afp->f_fbmask,
1597 (struct flow_filter6 *)&afp->f_filter,
1598 fp6))
1599 /* filter matched */
1600 return (afp->f_class);
1601
1602 /*
1603 * check again for filters with a wildcard.
1604 */
1605 if (i != ACC_WILDCARD_INDEX)
1606 i = ACC_WILDCARD_INDEX;
1607 else
1608 break;
1609 } while (1);
1610 }
1611 #endif /* INET6 */
1612
1613 /* no filter matched */
1614 return (NULL);
1615 }
1616
1617 static int
apply_filter4(fbmask,filt,pkt)1618 apply_filter4(fbmask, filt, pkt)
1619 u_int32_t fbmask;
1620 struct flow_filter *filt;
1621 struct flowinfo_in *pkt;
1622 {
1623 if (filt->ff_flow.fi_family != AF_INET)
1624 return (0);
1625 if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1626 return (0);
1627 if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1628 return (0);
1629 if ((fbmask & FIMB4_DADDR) &&
1630 filt->ff_flow.fi_dst.s_addr !=
1631 (pkt->fi_dst.s_addr & filt->ff_mask.mask_dst.s_addr))
1632 return (0);
1633 if ((fbmask & FIMB4_SADDR) &&
1634 filt->ff_flow.fi_src.s_addr !=
1635 (pkt->fi_src.s_addr & filt->ff_mask.mask_src.s_addr))
1636 return (0);
1637 if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1638 return (0);
1639 if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1640 (pkt->fi_tos & filt->ff_mask.mask_tos))
1641 return (0);
1642 if ((fbmask & FIMB4_GPI) && filt->ff_flow.fi_gpi != (pkt->fi_gpi))
1643 return (0);
1644 /* match */
1645 return (1);
1646 }
1647
1648 /*
1649 * filter matching function optimized for a common case that checks
1650 * only protocol and port numbers
1651 */
1652 static int
apply_ppfilter4(fbmask,filt,pkt)1653 apply_ppfilter4(fbmask, filt, pkt)
1654 u_int32_t fbmask;
1655 struct flow_filter *filt;
1656 struct flowinfo_in *pkt;
1657 {
1658 if (filt->ff_flow.fi_family != AF_INET)
1659 return (0);
1660 if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport)
1661 return (0);
1662 if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport)
1663 return (0);
1664 if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto)
1665 return (0);
1666 /* match */
1667 return (1);
1668 }
1669
1670 /*
1671 * filter matching function only for tos field.
1672 */
1673 static int
apply_tosfilter4(fbmask,filt,pkt)1674 apply_tosfilter4(fbmask, filt, pkt)
1675 u_int32_t fbmask;
1676 struct flow_filter *filt;
1677 struct flowinfo_in *pkt;
1678 {
1679 if (filt->ff_flow.fi_family != AF_INET)
1680 return (0);
1681 if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos !=
1682 (pkt->fi_tos & filt->ff_mask.mask_tos))
1683 return (0);
1684 /* match */
1685 return (1);
1686 }
1687
1688 #ifdef INET6
1689 static int
apply_filter6(fbmask,filt,pkt)1690 apply_filter6(fbmask, filt, pkt)
1691 u_int32_t fbmask;
1692 struct flow_filter6 *filt;
1693 struct flowinfo_in6 *pkt;
1694 {
1695 int i;
1696
1697 if (filt->ff_flow6.fi6_family != AF_INET6)
1698 return (0);
1699 if ((fbmask & FIMB6_FLABEL) &&
1700 filt->ff_flow6.fi6_flowlabel != pkt->fi6_flowlabel)
1701 return (0);
1702 if ((fbmask & FIMB6_PROTO) &&
1703 filt->ff_flow6.fi6_proto != pkt->fi6_proto)
1704 return (0);
1705 if ((fbmask & FIMB6_SPORT) &&
1706 filt->ff_flow6.fi6_sport != pkt->fi6_sport)
1707 return (0);
1708 if ((fbmask & FIMB6_DPORT) &&
1709 filt->ff_flow6.fi6_dport != pkt->fi6_dport)
1710 return (0);
1711 if (fbmask & FIMB6_SADDR) {
1712 for (i = 0; i < 4; i++)
1713 if (filt->ff_flow6.fi6_src.s6_addr32[i] !=
1714 (pkt->fi6_src.s6_addr32[i] &
1715 filt->ff_mask6.mask6_src.s6_addr32[i]))
1716 return (0);
1717 }
1718 if (fbmask & FIMB6_DADDR) {
1719 for (i = 0; i < 4; i++)
1720 if (filt->ff_flow6.fi6_dst.s6_addr32[i] !=
1721 (pkt->fi6_dst.s6_addr32[i] &
1722 filt->ff_mask6.mask6_dst.s6_addr32[i]))
1723 return (0);
1724 }
1725 if ((fbmask & FIMB6_TCLASS) &&
1726 filt->ff_flow6.fi6_tclass !=
1727 (pkt->fi6_tclass & filt->ff_mask6.mask6_tclass))
1728 return (0);
1729 if ((fbmask & FIMB6_GPI) &&
1730 filt->ff_flow6.fi6_gpi != pkt->fi6_gpi)
1731 return (0);
1732 /* match */
1733 return (1);
1734 }
1735 #endif /* INET6 */
1736
1737 /*
1738 * filter handle:
1739 * bit 20-28: index to the filter hash table
1740 * bit 0-19: unique id in the hash bucket.
1741 */
1742 static u_long
get_filt_handle(classifier,i)1743 get_filt_handle(classifier, i)
1744 struct acc_classifier *classifier;
1745 int i;
1746 {
1747 static u_long handle_number = 1;
1748 u_long handle;
1749 struct acc_filter *afp;
1750
1751 while (1) {
1752 handle = handle_number++ & 0x000fffff;
1753
1754 if (LIST_EMPTY(&classifier->acc_filters[i]))
1755 break;
1756
1757 LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1758 if ((afp->f_handle & 0x000fffff) == handle)
1759 break;
1760 if (afp == NULL)
1761 break;
1762 /* this handle is already used, try again */
1763 }
1764
1765 return ((i << 20) | handle);
1766 }
1767
1768 /* convert filter handle to filter pointer */
1769 static struct acc_filter *
filth_to_filtp(classifier,handle)1770 filth_to_filtp(classifier, handle)
1771 struct acc_classifier *classifier;
1772 u_long handle;
1773 {
1774 struct acc_filter *afp;
1775 int i;
1776
1777 i = ACC_GET_HINDEX(handle);
1778
1779 LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain)
1780 if (afp->f_handle == handle)
1781 return (afp);
1782
1783 return (NULL);
1784 }
1785
1786 /* create flowinfo bitmask */
1787 static u_int32_t
filt2fibmask(filt)1788 filt2fibmask(filt)
1789 struct flow_filter *filt;
1790 {
1791 u_int32_t mask = 0;
1792 #ifdef INET6
1793 struct flow_filter6 *filt6;
1794 #endif
1795
1796 switch (filt->ff_flow.fi_family) {
1797 case AF_INET:
1798 if (filt->ff_flow.fi_proto != 0)
1799 mask |= FIMB4_PROTO;
1800 if (filt->ff_flow.fi_tos != 0)
1801 mask |= FIMB4_TOS;
1802 if (filt->ff_flow.fi_dst.s_addr != 0)
1803 mask |= FIMB4_DADDR;
1804 if (filt->ff_flow.fi_src.s_addr != 0)
1805 mask |= FIMB4_SADDR;
1806 if (filt->ff_flow.fi_sport != 0)
1807 mask |= FIMB4_SPORT;
1808 if (filt->ff_flow.fi_dport != 0)
1809 mask |= FIMB4_DPORT;
1810 if (filt->ff_flow.fi_gpi != 0)
1811 mask |= FIMB4_GPI;
1812 break;
1813 #ifdef INET6
1814 case AF_INET6:
1815 filt6 = (struct flow_filter6 *)filt;
1816
1817 if (filt6->ff_flow6.fi6_proto != 0)
1818 mask |= FIMB6_PROTO;
1819 if (filt6->ff_flow6.fi6_tclass != 0)
1820 mask |= FIMB6_TCLASS;
1821 if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_dst))
1822 mask |= FIMB6_DADDR;
1823 if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_src))
1824 mask |= FIMB6_SADDR;
1825 if (filt6->ff_flow6.fi6_sport != 0)
1826 mask |= FIMB6_SPORT;
1827 if (filt6->ff_flow6.fi6_dport != 0)
1828 mask |= FIMB6_DPORT;
1829 if (filt6->ff_flow6.fi6_gpi != 0)
1830 mask |= FIMB6_GPI;
1831 if (filt6->ff_flow6.fi6_flowlabel != 0)
1832 mask |= FIMB6_FLABEL;
1833 break;
1834 #endif /* INET6 */
1835 }
1836 return (mask);
1837 }
1838
1839 /*
1840 * helper functions to handle IPv4 fragments.
1841 * currently only in-sequence fragments are handled.
1842 * - fragment info is cached in a LRU list.
1843 * - when a first fragment is found, cache its flow info.
1844 * - when a non-first fragment is found, lookup the cache.
1845 */
1846
1847 struct ip4_frag {
1848 TAILQ_ENTRY(ip4_frag) ip4f_chain;
1849 char ip4f_valid;
1850 u_short ip4f_id;
1851 struct flowinfo_in ip4f_info;
1852 };
1853
1854 static TAILQ_HEAD(ip4f_list, ip4_frag) ip4f_list; /* IPv4 fragment cache */
1855
1856 #define IP4F_TABSIZE 16 /* IPv4 fragment cache size */
1857
1858 static void
ip4f_cache(ip,fin)1859 ip4f_cache(ip, fin)
1860 struct ip *ip;
1861 struct flowinfo_in *fin;
1862 {
1863 struct ip4_frag *fp;
1864
1865 if (TAILQ_EMPTY(&ip4f_list)) {
1866 /* first time call, allocate fragment cache entries. */
1867 if (ip4f_init() < 0)
1868 /* allocation failed! */
1869 return;
1870 }
1871
1872 fp = ip4f_alloc();
1873 fp->ip4f_id = ip->ip_id;
1874 fp->ip4f_info.fi_proto = ip->ip_p;
1875 fp->ip4f_info.fi_src.s_addr = ip->ip_src.s_addr;
1876 fp->ip4f_info.fi_dst.s_addr = ip->ip_dst.s_addr;
1877
1878 /* save port numbers */
1879 fp->ip4f_info.fi_sport = fin->fi_sport;
1880 fp->ip4f_info.fi_dport = fin->fi_dport;
1881 fp->ip4f_info.fi_gpi = fin->fi_gpi;
1882 }
1883
1884 static int
ip4f_lookup(ip,fin)1885 ip4f_lookup(ip, fin)
1886 struct ip *ip;
1887 struct flowinfo_in *fin;
1888 {
1889 struct ip4_frag *fp;
1890
1891 for (fp = TAILQ_FIRST(&ip4f_list); fp != NULL && fp->ip4f_valid;
1892 fp = TAILQ_NEXT(fp, ip4f_chain))
1893 if (ip->ip_id == fp->ip4f_id &&
1894 ip->ip_src.s_addr == fp->ip4f_info.fi_src.s_addr &&
1895 ip->ip_dst.s_addr == fp->ip4f_info.fi_dst.s_addr &&
1896 ip->ip_p == fp->ip4f_info.fi_proto) {
1897 /* found the matching entry */
1898 fin->fi_sport = fp->ip4f_info.fi_sport;
1899 fin->fi_dport = fp->ip4f_info.fi_dport;
1900 fin->fi_gpi = fp->ip4f_info.fi_gpi;
1901
1902 if ((ntohs(ip->ip_off) & IP_MF) == 0)
1903 /* this is the last fragment,
1904 release the entry. */
1905 ip4f_free(fp);
1906
1907 return (1);
1908 }
1909
1910 /* no matching entry found */
1911 return (0);
1912 }
1913
1914 static int
ip4f_init(void)1915 ip4f_init(void)
1916 {
1917 struct ip4_frag *fp;
1918 int i;
1919
1920 TAILQ_INIT(&ip4f_list);
1921 for (i=0; i<IP4F_TABSIZE; i++) {
1922 fp = malloc(sizeof(struct ip4_frag),
1923 M_DEVBUF, M_NOWAIT);
1924 if (fp == NULL) {
1925 printf("ip4f_init: can't alloc %dth entry!\n", i);
1926 if (i == 0)
1927 return (-1);
1928 return (0);
1929 }
1930 fp->ip4f_valid = 0;
1931 TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1932 }
1933 return (0);
1934 }
1935
1936 static struct ip4_frag *
ip4f_alloc(void)1937 ip4f_alloc(void)
1938 {
1939 struct ip4_frag *fp;
1940
1941 /* reclaim an entry at the tail, put it at the head */
1942 fp = TAILQ_LAST(&ip4f_list, ip4f_list);
1943 TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1944 fp->ip4f_valid = 1;
1945 TAILQ_INSERT_HEAD(&ip4f_list, fp, ip4f_chain);
1946 return (fp);
1947 }
1948
1949 static void
ip4f_free(fp)1950 ip4f_free(fp)
1951 struct ip4_frag *fp;
1952 {
1953 TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain);
1954 fp->ip4f_valid = 0;
1955 TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain);
1956 }
1957
1958 #endif /* ALTQ3_CLFIER_COMPAT */
1959