xref: /freebsd/sys/netpfil/pf/pf_if.c (revision 4530e0c3e78d0616367d37273d6c1f47f627839b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2001 Daniel Hartmeier
5  * Copyright (c) 2003 Cedric Berger
6  * Copyright (c) 2005 Henning Brauer <henning@openbsd.org>
7  * Copyright (c) 2005 Ryan McBride <mcbride@openbsd.org>
8  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
9  * All rights reserved.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  *
15  *    - Redistributions of source code must retain the above copyright
16  *      notice, this list of conditions and the following disclaimer.
17  *    - Redistributions in binary form must reproduce the above
18  *      copyright notice, this list of conditions and the following
19  *      disclaimer in the documentation and/or other materials provided
20  *      with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
26  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
28  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
30  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
32  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33  * POSSIBILITY OF SUCH DAMAGE.
34  *
35  *	$OpenBSD: pf_if.c,v 1.54 2008/06/14 16:55:28 mk Exp $
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_inet.h"
42 #include "opt_inet6.h"
43 
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/eventhandler.h>
47 #include <sys/lock.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 
51 #include <net/if.h>
52 #include <net/if_var.h>
53 #include <net/if_private.h>
54 #include <net/vnet.h>
55 #include <net/pfvar.h>
56 #include <net/route.h>
57 
58 VNET_DEFINE(struct pfi_kkif *,	 pfi_all);
59 VNET_DEFINE_STATIC(long, pfi_update);
60 #define	V_pfi_update	VNET(pfi_update)
61 #define PFI_BUFFER_MAX	0x10000
62 
63 VNET_DECLARE(int, pf_vnet_active);
64 #define V_pf_vnet_active	VNET(pf_vnet_active)
65 
66 VNET_DEFINE_STATIC(struct pfr_addr *, pfi_buffer);
67 VNET_DEFINE_STATIC(int, pfi_buffer_cnt);
68 VNET_DEFINE_STATIC(int,	pfi_buffer_max);
69 #define	V_pfi_buffer		 VNET(pfi_buffer)
70 #define	V_pfi_buffer_cnt	 VNET(pfi_buffer_cnt)
71 #define	V_pfi_buffer_max	 VNET(pfi_buffer_max)
72 
73 #ifdef PF_WANT_32_TO_64_COUNTER
74 VNET_DEFINE(struct allkiflist_head, pf_allkiflist);
75 VNET_DEFINE(size_t, pf_allkifcount);
76 VNET_DEFINE(struct pfi_kkif *, pf_kifmarker);
77 #endif
78 
79 eventhandler_tag	 pfi_attach_cookie;
80 eventhandler_tag	 pfi_detach_cookie;
81 eventhandler_tag	 pfi_attach_group_cookie;
82 eventhandler_tag	 pfi_change_group_cookie;
83 eventhandler_tag	 pfi_detach_group_cookie;
84 eventhandler_tag	 pfi_ifaddr_event_cookie;
85 
86 static void	 pfi_attach_ifnet(struct ifnet *, struct pfi_kkif *);
87 static void	 pfi_attach_ifgroup(struct ifg_group *, struct pfi_kkif *);
88 
89 static void	 pfi_kkif_update(struct pfi_kkif *);
90 static void	 pfi_dynaddr_update(struct pfi_dynaddr *dyn);
91 static void	 pfi_table_update(struct pfr_ktable *, struct pfi_kkif *, int,
92 		    int);
93 static void	 pfi_instance_add(struct ifnet *, int, int);
94 static void	 pfi_address_add(struct sockaddr *, int, int);
95 static int	 pfi_kkif_compare(struct pfi_kkif *, struct pfi_kkif *);
96 static int	 pfi_skip_if(const char *, struct pfi_kkif *);
97 static int	 pfi_unmask(void *);
98 static void	 pfi_attach_ifnet_event(void * __unused, struct ifnet *);
99 static void	 pfi_detach_ifnet_event(void * __unused, struct ifnet *);
100 static void	 pfi_attach_group_event(void * __unused, struct ifg_group *);
101 static void	 pfi_change_group_event(void * __unused, char *);
102 static void	 pfi_detach_group_event(void * __unused, struct ifg_group *);
103 static void	 pfi_ifaddr_event(void * __unused, struct ifnet *);
104 
105 RB_HEAD(pfi_ifhead, pfi_kkif);
106 static RB_PROTOTYPE(pfi_ifhead, pfi_kkif, pfik_tree, pfi_kkif_compare);
107 static RB_GENERATE(pfi_ifhead, pfi_kkif, pfik_tree, pfi_kkif_compare);
108 VNET_DEFINE_STATIC(struct pfi_ifhead, pfi_ifs);
109 #define	V_pfi_ifs	VNET(pfi_ifs)
110 
111 #define	PFI_BUFFER_MAX		0x10000
112 MALLOC_DEFINE(PFI_MTYPE, "pf_ifnet", "pf(4) interface database");
113 
114 LIST_HEAD(pfi_list, pfi_kkif);
115 VNET_DEFINE_STATIC(struct pfi_list, pfi_unlinked_kifs);
116 #define	V_pfi_unlinked_kifs	VNET(pfi_unlinked_kifs)
117 static struct mtx pfi_unlnkdkifs_mtx;
118 MTX_SYSINIT(pfi_unlnkdkifs_mtx, &pfi_unlnkdkifs_mtx, "pf unlinked interfaces",
119     MTX_DEF);
120 
121 void
122 pfi_initialize_vnet(void)
123 {
124 	struct pfi_list kifs = LIST_HEAD_INITIALIZER();
125 	struct epoch_tracker et;
126 	struct pfi_kkif *kif;
127 	struct ifg_group *ifg;
128 	struct ifnet *ifp;
129 	int nkifs;
130 
131 	V_pfi_buffer_max = 64;
132 	V_pfi_buffer = malloc(V_pfi_buffer_max * sizeof(*V_pfi_buffer),
133 	    PFI_MTYPE, M_WAITOK);
134 
135 	nkifs = 1;	/* one for V_pfi_all */
136 	IFNET_RLOCK();
137 	CK_STAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
138 		nkifs++;
139 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link)
140 		nkifs++;
141 
142 	for (int n = 0; n < nkifs; n++) {
143 		kif = pf_kkif_create(M_WAITOK);
144 		LIST_INSERT_HEAD(&kifs, kif, pfik_list);
145 	}
146 
147 	NET_EPOCH_ENTER(et);
148 	PF_RULES_WLOCK();
149 	kif = LIST_FIRST(&kifs);
150 	LIST_REMOVE(kif, pfik_list);
151 	V_pfi_all = pfi_kkif_attach(kif, IFG_ALL);
152 	CK_STAILQ_FOREACH(ifg, &V_ifg_head, ifg_next) {
153 		kif = LIST_FIRST(&kifs);
154 		LIST_REMOVE(kif, pfik_list);
155 		pfi_attach_ifgroup(ifg, kif);
156 	}
157 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
158 		kif = LIST_FIRST(&kifs);
159 		LIST_REMOVE(kif, pfik_list);
160 		pfi_attach_ifnet(ifp, kif);
161 	}
162 	PF_RULES_WUNLOCK();
163 	NET_EPOCH_EXIT(et);
164 	IFNET_RUNLOCK();
165 
166 	MPASS(LIST_EMPTY(&kifs));
167 }
168 
169 void
170 pfi_initialize(void)
171 {
172 
173 	pfi_attach_cookie = EVENTHANDLER_REGISTER(ifnet_arrival_event,
174 	    pfi_attach_ifnet_event, NULL, EVENTHANDLER_PRI_ANY);
175 	pfi_detach_cookie = EVENTHANDLER_REGISTER(ifnet_departure_event,
176 	    pfi_detach_ifnet_event, NULL, EVENTHANDLER_PRI_ANY);
177 	pfi_attach_group_cookie = EVENTHANDLER_REGISTER(group_attach_event,
178 	    pfi_attach_group_event, NULL, EVENTHANDLER_PRI_ANY);
179 	pfi_change_group_cookie = EVENTHANDLER_REGISTER(group_change_event,
180 	    pfi_change_group_event, NULL, EVENTHANDLER_PRI_ANY);
181 	pfi_detach_group_cookie = EVENTHANDLER_REGISTER(group_detach_event,
182 	    pfi_detach_group_event, NULL, EVENTHANDLER_PRI_ANY);
183 	pfi_ifaddr_event_cookie = EVENTHANDLER_REGISTER(ifaddr_event,
184 	    pfi_ifaddr_event, NULL, EVENTHANDLER_PRI_ANY);
185 }
186 
187 void
188 pfi_cleanup_vnet(void)
189 {
190 	struct pfi_kkif *kif;
191 
192 	PF_RULES_WASSERT();
193 
194 	V_pfi_all = NULL;
195 	while ((kif = RB_MIN(pfi_ifhead, &V_pfi_ifs))) {
196 		RB_REMOVE(pfi_ifhead, &V_pfi_ifs, kif);
197 		if (kif->pfik_group)
198 			kif->pfik_group->ifg_pf_kif = NULL;
199 		if (kif->pfik_ifp) {
200 			if_rele(kif->pfik_ifp);
201 			kif->pfik_ifp->if_pf_kif = NULL;
202 		}
203 		pf_kkif_free(kif);
204 	}
205 
206 	mtx_lock(&pfi_unlnkdkifs_mtx);
207 	while ((kif = LIST_FIRST(&V_pfi_unlinked_kifs))) {
208 		LIST_REMOVE(kif, pfik_list);
209 		pf_kkif_free(kif);
210 	}
211 	mtx_unlock(&pfi_unlnkdkifs_mtx);
212 
213 	free(V_pfi_buffer, PFI_MTYPE);
214 }
215 
216 void
217 pfi_cleanup(void)
218 {
219 
220 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event, pfi_attach_cookie);
221 	EVENTHANDLER_DEREGISTER(ifnet_departure_event, pfi_detach_cookie);
222 	EVENTHANDLER_DEREGISTER(group_attach_event, pfi_attach_group_cookie);
223 	EVENTHANDLER_DEREGISTER(group_change_event, pfi_change_group_cookie);
224 	EVENTHANDLER_DEREGISTER(group_detach_event, pfi_detach_group_cookie);
225 	EVENTHANDLER_DEREGISTER(ifaddr_event, pfi_ifaddr_event_cookie);
226 }
227 
228 struct pfi_kkif*
229 pf_kkif_create(int flags)
230 {
231 	struct pfi_kkif *kif;
232 #ifdef PF_WANT_32_TO_64_COUNTER
233 	bool wowned;
234 #endif
235 
236 	kif = malloc(sizeof(*kif), PFI_MTYPE, flags | M_ZERO);
237 	if (! kif)
238 		return (kif);
239 
240 	for (int i = 0; i < 2; i++) {
241 		for (int j = 0; j < 2; j++) {
242 			for (int k = 0; k < 2; k++) {
243 				if (pf_counter_u64_init(&kif->pfik_packets[i][j][k], flags) != 0) {
244 					pf_kkif_free(kif);
245 					return (NULL);
246 				}
247 
248 				if (pf_counter_u64_init(&kif->pfik_bytes[i][j][k], flags) != 0) {
249 					pf_kkif_free(kif);
250 					return (NULL);
251 				}
252 			}
253 		}
254 	}
255 
256 #ifdef PF_WANT_32_TO_64_COUNTER
257 	wowned = PF_RULES_WOWNED();
258 	if (!wowned)
259 		PF_RULES_WLOCK();
260 	LIST_INSERT_HEAD(&V_pf_allkiflist, kif, pfik_allkiflist);
261 	V_pf_allkifcount++;
262 	if (!wowned)
263 		PF_RULES_WUNLOCK();
264 #endif
265 
266 	return (kif);
267 }
268 
269 void
270 pf_kkif_free(struct pfi_kkif *kif)
271 {
272 #ifdef PF_WANT_32_TO_64_COUNTER
273 	bool wowned;
274 #endif
275 
276 	if (! kif)
277 		return;
278 
279 #ifdef PF_WANT_32_TO_64_COUNTER
280 	wowned = PF_RULES_WOWNED();
281 	if (!wowned)
282 		PF_RULES_WLOCK();
283 	LIST_REMOVE(kif, pfik_allkiflist);
284 	V_pf_allkifcount--;
285 	if (!wowned)
286 		PF_RULES_WUNLOCK();
287 #endif
288 
289 	for (int i = 0; i < 2; i++) {
290 		for (int j = 0; j < 2; j++) {
291 			for (int k = 0; k < 2; k++) {
292 				pf_counter_u64_deinit(&kif->pfik_packets[i][j][k]);
293 				pf_counter_u64_deinit(&kif->pfik_bytes[i][j][k]);
294 			}
295 		}
296 	}
297 
298 	free(kif, PFI_MTYPE);
299 }
300 
301 void
302 pf_kkif_zero(struct pfi_kkif *kif)
303 {
304 
305 	for (int i = 0; i < 2; i++) {
306 		for (int j = 0; j < 2; j++) {
307 			for (int k = 0; k < 2; k++) {
308 				pf_counter_u64_zero(&kif->pfik_packets[i][j][k]);
309 				pf_counter_u64_zero(&kif->pfik_bytes[i][j][k]);
310 			}
311 		}
312 	}
313 	kif->pfik_tzero = time_second;
314 }
315 
316 struct pfi_kkif *
317 pfi_kkif_find(const char *kif_name)
318 {
319 	struct pfi_kif_cmp s;
320 
321 	PF_RULES_ASSERT();
322 
323 	memset(&s, 0, sizeof(s));
324 	strlcpy(s.pfik_name, kif_name, sizeof(s.pfik_name));
325 
326 	return (RB_FIND(pfi_ifhead, &V_pfi_ifs, (struct pfi_kkif *)&s));
327 }
328 
329 struct pfi_kkif *
330 pfi_kkif_attach(struct pfi_kkif *kif, const char *kif_name)
331 {
332 	struct pfi_kkif *kif1;
333 
334 	PF_RULES_WASSERT();
335 	KASSERT(kif != NULL, ("%s: null kif", __func__));
336 
337 	kif1 = pfi_kkif_find(kif_name);
338 	if (kif1 != NULL) {
339 		pf_kkif_free(kif);
340 		return (kif1);
341 	}
342 
343 	pf_kkif_zero(kif);
344 	strlcpy(kif->pfik_name, kif_name, sizeof(kif->pfik_name));
345 	/*
346 	 * It seems that the value of time_second is in unintialzied state
347 	 * when pf sets interface statistics clear time in boot phase if pf
348 	 * was statically linked to kernel. Instead of setting the bogus
349 	 * time value have pfi_get_ifaces handle this case. In
350 	 * pfi_get_ifaces it uses time_second if it sees the time is 0.
351 	 */
352 	kif->pfik_tzero = time_second > 1 ? time_second : 0;
353 	TAILQ_INIT(&kif->pfik_dynaddrs);
354 
355 	RB_INSERT(pfi_ifhead, &V_pfi_ifs, kif);
356 
357 	return (kif);
358 }
359 
360 void
361 pfi_kkif_ref(struct pfi_kkif *kif)
362 {
363 
364 	PF_RULES_WASSERT();
365 	kif->pfik_rulerefs++;
366 }
367 
368 static void
369 pfi_kkif_remove_if_unref(struct pfi_kkif *kif)
370 {
371 
372 	PF_RULES_WASSERT();
373 
374 	if (kif->pfik_rulerefs > 0)
375 		return;
376 
377 	/* kif referencing an existing ifnet or group or holding flags should
378 	 * exist. */
379 	if (kif->pfik_ifp != NULL || kif->pfik_group != NULL ||
380 	    kif == V_pfi_all || kif->pfik_flags != 0)
381 		return;
382 
383 	RB_REMOVE(pfi_ifhead, &V_pfi_ifs, kif);
384 
385 	kif->pfik_flags |= PFI_IFLAG_REFS;
386 
387 	mtx_lock(&pfi_unlnkdkifs_mtx);
388 	LIST_INSERT_HEAD(&V_pfi_unlinked_kifs, kif, pfik_list);
389 	mtx_unlock(&pfi_unlnkdkifs_mtx);
390 }
391 
392 void
393 pfi_kkif_unref(struct pfi_kkif *kif)
394 {
395 
396 	PF_RULES_WASSERT();
397 	KASSERT(kif->pfik_rulerefs > 0, ("%s: %p has zero refs", __func__, kif));
398 
399 	kif->pfik_rulerefs--;
400 
401 	pfi_kkif_remove_if_unref(kif);
402 }
403 
404 void
405 pfi_kkif_purge(void)
406 {
407 	struct pfi_kkif *kif, *kif1;
408 
409 	/*
410 	 * Do naive mark-and-sweep garbage collecting of old kifs.
411 	 * Reference flag is raised by pf_purge_expired_states().
412 	 */
413 	mtx_lock(&pfi_unlnkdkifs_mtx);
414 	LIST_FOREACH_SAFE(kif, &V_pfi_unlinked_kifs, pfik_list, kif1) {
415 		if (!(kif->pfik_flags & PFI_IFLAG_REFS)) {
416 			LIST_REMOVE(kif, pfik_list);
417 			pf_kkif_free(kif);
418 		} else
419 			kif->pfik_flags &= ~PFI_IFLAG_REFS;
420 	}
421 	mtx_unlock(&pfi_unlnkdkifs_mtx);
422 }
423 
424 int
425 pfi_kkif_match(struct pfi_kkif *rule_kif, struct pfi_kkif *packet_kif)
426 {
427 	struct ifg_list	*p;
428 
429 	NET_EPOCH_ASSERT();
430 
431 	if (rule_kif == NULL || rule_kif == packet_kif)
432 		return (1);
433 
434 	if (rule_kif->pfik_group != NULL) {
435 		CK_STAILQ_FOREACH(p, &packet_kif->pfik_ifp->if_groups, ifgl_next)
436 			if (p->ifgl_group == rule_kif->pfik_group)
437 				return (1);
438 	}
439 
440 	return (0);
441 }
442 
443 static void
444 pfi_attach_ifnet(struct ifnet *ifp, struct pfi_kkif *kif)
445 {
446 
447 	PF_RULES_WASSERT();
448 
449 	V_pfi_update++;
450 	kif = pfi_kkif_attach(kif, ifp->if_xname);
451 	if_ref(ifp);
452 	kif->pfik_ifp = ifp;
453 	ifp->if_pf_kif = kif;
454 	pfi_kkif_update(kif);
455 }
456 
457 static void
458 pfi_attach_ifgroup(struct ifg_group *ifg, struct pfi_kkif *kif)
459 {
460 
461 	PF_RULES_WASSERT();
462 
463 	V_pfi_update++;
464 	kif = pfi_kkif_attach(kif, ifg->ifg_group);
465 	kif->pfik_group = ifg;
466 	ifg->ifg_pf_kif = kif;
467 }
468 
469 int
470 pfi_match_addr(struct pfi_dynaddr *dyn, struct pf_addr *a, sa_family_t af)
471 {
472 	switch (af) {
473 #ifdef INET
474 	case AF_INET:
475 		switch (dyn->pfid_acnt4) {
476 		case 0:
477 			return (0);
478 		case 1:
479 			return (PF_MATCHA(0, &dyn->pfid_addr4,
480 			    &dyn->pfid_mask4, a, AF_INET));
481 		default:
482 			return (pfr_match_addr(dyn->pfid_kt, a, AF_INET));
483 		}
484 		break;
485 #endif /* INET */
486 #ifdef INET6
487 	case AF_INET6:
488 		switch (dyn->pfid_acnt6) {
489 		case 0:
490 			return (0);
491 		case 1:
492 			return (PF_MATCHA(0, &dyn->pfid_addr6,
493 			    &dyn->pfid_mask6, a, AF_INET6));
494 		default:
495 			return (pfr_match_addr(dyn->pfid_kt, a, AF_INET6));
496 		}
497 		break;
498 #endif /* INET6 */
499 	default:
500 		return (0);
501 	}
502 }
503 
504 int
505 pfi_dynaddr_setup(struct pf_addr_wrap *aw, sa_family_t af)
506 {
507 	struct epoch_tracker	 et;
508 	struct pfi_dynaddr	*dyn;
509 	char			 tblname[PF_TABLE_NAME_SIZE];
510 	struct pf_kruleset	*ruleset = NULL;
511 	struct pfi_kkif		*kif;
512 	int			 rv = 0;
513 
514 	PF_RULES_WASSERT();
515 	KASSERT(aw->type == PF_ADDR_DYNIFTL, ("%s: type %u",
516 	    __func__, aw->type));
517 	KASSERT(aw->p.dyn == NULL, ("%s: dyn is %p", __func__, aw->p.dyn));
518 
519 	if ((dyn = malloc(sizeof(*dyn), PFI_MTYPE, M_NOWAIT | M_ZERO)) == NULL)
520 		return (ENOMEM);
521 
522 	if ((kif = pf_kkif_create(M_NOWAIT)) == NULL) {
523 		free(dyn, PFI_MTYPE);
524 		return (ENOMEM);
525 	}
526 
527 	if (!strcmp(aw->v.ifname, "self"))
528 		dyn->pfid_kif = pfi_kkif_attach(kif, IFG_ALL);
529 	else
530 		dyn->pfid_kif = pfi_kkif_attach(kif, aw->v.ifname);
531 	kif = NULL;
532 	pfi_kkif_ref(dyn->pfid_kif);
533 
534 	dyn->pfid_net = pfi_unmask(&aw->v.a.mask);
535 	if (af == AF_INET && dyn->pfid_net == 32)
536 		dyn->pfid_net = 128;
537 	strlcpy(tblname, aw->v.ifname, sizeof(tblname));
538 	if (aw->iflags & PFI_AFLAG_NETWORK)
539 		strlcat(tblname, ":network", sizeof(tblname));
540 	if (aw->iflags & PFI_AFLAG_BROADCAST)
541 		strlcat(tblname, ":broadcast", sizeof(tblname));
542 	if (aw->iflags & PFI_AFLAG_PEER)
543 		strlcat(tblname, ":peer", sizeof(tblname));
544 	if (aw->iflags & PFI_AFLAG_NOALIAS)
545 		strlcat(tblname, ":0", sizeof(tblname));
546 	if (dyn->pfid_net != 128)
547 		snprintf(tblname + strlen(tblname),
548 		    sizeof(tblname) - strlen(tblname), "/%d", dyn->pfid_net);
549 	if ((ruleset = pf_find_or_create_kruleset(PF_RESERVED_ANCHOR)) == NULL) {
550 		rv = ENOMEM;
551 		goto _bad;
552 	}
553 
554 	if ((dyn->pfid_kt = pfr_attach_table(ruleset, tblname)) == NULL) {
555 		rv = ENOMEM;
556 		goto _bad;
557 	}
558 
559 	dyn->pfid_kt->pfrkt_flags |= PFR_TFLAG_ACTIVE;
560 	dyn->pfid_iflags = aw->iflags;
561 	dyn->pfid_af = af;
562 
563 	TAILQ_INSERT_TAIL(&dyn->pfid_kif->pfik_dynaddrs, dyn, entry);
564 	aw->p.dyn = dyn;
565 	NET_EPOCH_ENTER(et);
566 	pfi_kkif_update(dyn->pfid_kif);
567 	NET_EPOCH_EXIT(et);
568 
569 	return (0);
570 
571 _bad:
572 	if (dyn->pfid_kt != NULL)
573 		pfr_detach_table(dyn->pfid_kt);
574 	if (ruleset != NULL)
575 		pf_remove_if_empty_kruleset(ruleset);
576 	pfi_kkif_unref(dyn->pfid_kif);
577 	free(dyn, PFI_MTYPE);
578 
579 	return (rv);
580 }
581 
582 static void
583 pfi_kkif_update(struct pfi_kkif *kif)
584 {
585 	struct ifg_list		*ifgl;
586 	struct ifg_member	*ifgm;
587 	struct pfi_dynaddr	*p;
588 	struct pfi_kkif		*tmpkif;
589 
590 	NET_EPOCH_ASSERT();
591 	PF_RULES_WASSERT();
592 
593 	/* update all dynaddr */
594 	TAILQ_FOREACH(p, &kif->pfik_dynaddrs, entry)
595 		pfi_dynaddr_update(p);
596 
597 	/* Apply group flags to new members. */
598 	if (kif->pfik_group != NULL) {
599 		CK_STAILQ_FOREACH(ifgm, &kif->pfik_group->ifg_members,
600 		    ifgm_next) {
601 			tmpkif = (struct pfi_kkif *)ifgm->ifgm_ifp->if_pf_kif;
602 			if (tmpkif == NULL)
603 				continue;
604 
605 			tmpkif->pfik_flags |= kif->pfik_flags;
606 		}
607 	}
608 
609 	/* again for all groups kif is member of */
610 	if (kif->pfik_ifp != NULL) {
611 		CK_STAILQ_FOREACH(ifgl, &kif->pfik_ifp->if_groups, ifgl_next)
612 			pfi_kkif_update((struct pfi_kkif *)
613 			    ifgl->ifgl_group->ifg_pf_kif);
614 	}
615 }
616 
617 static void
618 pfi_dynaddr_update(struct pfi_dynaddr *dyn)
619 {
620 	struct pfi_kkif		*kif;
621 	struct pfr_ktable	*kt;
622 
623 	PF_RULES_WASSERT();
624 	KASSERT(dyn && dyn->pfid_kif && dyn->pfid_kt,
625 	    ("%s: bad argument", __func__));
626 
627 	kif = dyn->pfid_kif;
628 	kt = dyn->pfid_kt;
629 
630 	if (kt->pfrkt_larg != V_pfi_update) {
631 		/* this table needs to be brought up-to-date */
632 		pfi_table_update(kt, kif, dyn->pfid_net, dyn->pfid_iflags);
633 		kt->pfrkt_larg = V_pfi_update;
634 	}
635 	pfr_dynaddr_update(kt, dyn);
636 }
637 
638 static void
639 pfi_table_update(struct pfr_ktable *kt, struct pfi_kkif *kif, int net, int flags)
640 {
641 	int			 e, size2 = 0;
642 	struct ifg_member	*ifgm;
643 
644 	NET_EPOCH_ASSERT();
645 
646 	V_pfi_buffer_cnt = 0;
647 
648 	if (kif->pfik_ifp != NULL)
649 		pfi_instance_add(kif->pfik_ifp, net, flags);
650 	else if (kif->pfik_group != NULL) {
651 		CK_STAILQ_FOREACH(ifgm, &kif->pfik_group->ifg_members, ifgm_next)
652 			pfi_instance_add(ifgm->ifgm_ifp, net, flags);
653 	}
654 
655 	if ((e = pfr_set_addrs(&kt->pfrkt_t, V_pfi_buffer, V_pfi_buffer_cnt, &size2,
656 	    NULL, NULL, NULL, 0, PFR_TFLAG_ALLMASK)))
657 		printf("%s: cannot set %d new addresses into table %s: %d\n",
658 		    __func__, V_pfi_buffer_cnt, kt->pfrkt_name, e);
659 }
660 
661 static void
662 pfi_instance_add(struct ifnet *ifp, int net, int flags)
663 {
664 	struct ifaddr	*ia;
665 	int		 got4 = 0, got6 = 0;
666 	int		 net2, af;
667 
668 	NET_EPOCH_ASSERT();
669 
670 	CK_STAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
671 		if (ia->ifa_addr == NULL)
672 			continue;
673 		af = ia->ifa_addr->sa_family;
674 		if (af != AF_INET && af != AF_INET6)
675 			continue;
676 		/*
677 		 * XXX: For point-to-point interfaces, (ifname:0) and IPv4,
678 		 *      jump over addresses without a proper route to work
679 		 *      around a problem with ppp not fully removing the
680 		 *      address used during IPCP.
681 		 */
682 		if ((ifp->if_flags & IFF_POINTOPOINT) &&
683 		    !(ia->ifa_flags & IFA_ROUTE) &&
684 		    (flags & PFI_AFLAG_NOALIAS) && (af == AF_INET))
685 			continue;
686 		if ((flags & PFI_AFLAG_BROADCAST) && af == AF_INET6)
687 			continue;
688 		if ((flags & PFI_AFLAG_BROADCAST) &&
689 		    !(ifp->if_flags & IFF_BROADCAST))
690 			continue;
691 		if ((flags & PFI_AFLAG_PEER) &&
692 		    !(ifp->if_flags & IFF_POINTOPOINT))
693 			continue;
694 		if ((flags & (PFI_AFLAG_NETWORK | PFI_AFLAG_NOALIAS)) &&
695 		    af == AF_INET6 &&
696 		    IN6_IS_ADDR_LINKLOCAL(
697 		    &((struct sockaddr_in6 *)ia->ifa_addr)->sin6_addr))
698 			continue;
699 		if (flags & PFI_AFLAG_NOALIAS) {
700 			if (af == AF_INET && got4)
701 				continue;
702 			if (af == AF_INET6 && got6)
703 				continue;
704 		}
705 		if (af == AF_INET)
706 			got4 = 1;
707 		else if (af == AF_INET6)
708 			got6 = 1;
709 		net2 = net;
710 		if (net2 == 128 && (flags & PFI_AFLAG_NETWORK)) {
711 			if (af == AF_INET)
712 				net2 = pfi_unmask(&((struct sockaddr_in *)
713 				    ia->ifa_netmask)->sin_addr);
714 			else if (af == AF_INET6)
715 				net2 = pfi_unmask(&((struct sockaddr_in6 *)
716 				    ia->ifa_netmask)->sin6_addr);
717 		}
718 		if (af == AF_INET && net2 > 32)
719 			net2 = 32;
720 		if (flags & PFI_AFLAG_BROADCAST)
721 			pfi_address_add(ia->ifa_broadaddr, af, net2);
722 		else if (flags & PFI_AFLAG_PEER)
723 			pfi_address_add(ia->ifa_dstaddr, af, net2);
724 		else
725 			pfi_address_add(ia->ifa_addr, af, net2);
726 	}
727 }
728 
729 static void
730 pfi_address_add(struct sockaddr *sa, int af, int net)
731 {
732 	struct pfr_addr	*p;
733 	int		 i;
734 
735 	if (V_pfi_buffer_cnt >= V_pfi_buffer_max) {
736 		int		 new_max = V_pfi_buffer_max * 2;
737 
738 		if (new_max > PFI_BUFFER_MAX) {
739 			printf("%s: address buffer full (%d/%d)\n", __func__,
740 			    V_pfi_buffer_cnt, PFI_BUFFER_MAX);
741 			return;
742 		}
743 		p = malloc(new_max * sizeof(*V_pfi_buffer), PFI_MTYPE,
744 		    M_NOWAIT);
745 		if (p == NULL) {
746 			printf("%s: no memory to grow buffer (%d/%d)\n",
747 			    __func__, V_pfi_buffer_cnt, PFI_BUFFER_MAX);
748 			return;
749 		}
750 		memcpy(p, V_pfi_buffer, V_pfi_buffer_max * sizeof(*V_pfi_buffer));
751 		/* no need to zero buffer */
752 		free(V_pfi_buffer, PFI_MTYPE);
753 		V_pfi_buffer = p;
754 		V_pfi_buffer_max = new_max;
755 	}
756 	if (af == AF_INET && net > 32)
757 		net = 128;
758 	p = V_pfi_buffer + V_pfi_buffer_cnt++;
759 	memset(p, 0, sizeof(*p));
760 	p->pfra_af = af;
761 	p->pfra_net = net;
762 	if (af == AF_INET)
763 		p->pfra_ip4addr = ((struct sockaddr_in *)sa)->sin_addr;
764 	else if (af == AF_INET6) {
765 		p->pfra_ip6addr = ((struct sockaddr_in6 *)sa)->sin6_addr;
766 		if (IN6_IS_SCOPE_EMBED(&p->pfra_ip6addr))
767 			p->pfra_ip6addr.s6_addr16[1] = 0;
768 	}
769 	/* mask network address bits */
770 	if (net < 128)
771 		((caddr_t)p)[p->pfra_net/8] &= ~(0xFF >> (p->pfra_net%8));
772 	for (i = (p->pfra_net+7)/8; i < sizeof(p->pfra_u); i++)
773 		((caddr_t)p)[i] = 0;
774 }
775 
776 void
777 pfi_dynaddr_remove(struct pfi_dynaddr *dyn)
778 {
779 
780 	KASSERT(dyn->pfid_kif != NULL, ("%s: null pfid_kif", __func__));
781 	KASSERT(dyn->pfid_kt != NULL, ("%s: null pfid_kt", __func__));
782 
783 	TAILQ_REMOVE(&dyn->pfid_kif->pfik_dynaddrs, dyn, entry);
784 	pfi_kkif_unref(dyn->pfid_kif);
785 	pfr_detach_table(dyn->pfid_kt);
786 	free(dyn, PFI_MTYPE);
787 }
788 
789 void
790 pfi_dynaddr_copyout(struct pf_addr_wrap *aw)
791 {
792 
793 	KASSERT(aw->type == PF_ADDR_DYNIFTL,
794 	    ("%s: type %u", __func__, aw->type));
795 
796 	if (aw->p.dyn == NULL || aw->p.dyn->pfid_kif == NULL)
797 		return;
798 	aw->p.dyncnt = aw->p.dyn->pfid_acnt4 + aw->p.dyn->pfid_acnt6;
799 }
800 
801 static int
802 pfi_kkif_compare(struct pfi_kkif *p, struct pfi_kkif *q)
803 {
804 	return (strncmp(p->pfik_name, q->pfik_name, IFNAMSIZ));
805 }
806 
807 void
808 pfi_update_status(const char *name, struct pf_status *pfs)
809 {
810 	struct pfi_kkif		*p;
811 	struct pfi_kif_cmp	 key;
812 	struct ifg_member	 p_member, *ifgm;
813 	CK_STAILQ_HEAD(, ifg_member) ifg_members;
814 	int			 i, j, k;
815 
816 	if (pfs) {
817 		memset(pfs->pcounters, 0, sizeof(pfs->pcounters));
818 		memset(pfs->bcounters, 0, sizeof(pfs->bcounters));
819 	}
820 
821 	strlcpy(key.pfik_name, name, sizeof(key.pfik_name));
822 	p = RB_FIND(pfi_ifhead, &V_pfi_ifs, (struct pfi_kkif *)&key);
823 	if (p == NULL) {
824 		return;
825 	}
826 
827 	if (p->pfik_group != NULL) {
828 		memcpy(&ifg_members, &p->pfik_group->ifg_members,
829 		    sizeof(ifg_members));
830 	} else {
831 		/* build a temporary list for p only */
832 		memset(&p_member, 0, sizeof(p_member));
833 		p_member.ifgm_ifp = p->pfik_ifp;
834 		CK_STAILQ_INIT(&ifg_members);
835 		CK_STAILQ_INSERT_TAIL(&ifg_members, &p_member, ifgm_next);
836 	}
837 	CK_STAILQ_FOREACH(ifgm, &ifg_members, ifgm_next) {
838 		if (ifgm->ifgm_ifp == NULL || ifgm->ifgm_ifp->if_pf_kif == NULL)
839 			continue;
840 		p = (struct pfi_kkif *)ifgm->ifgm_ifp->if_pf_kif;
841 
842 		/* just clear statistics */
843 		if (pfs == NULL) {
844 			pf_kkif_zero(p);
845 			continue;
846 		}
847 		for (i = 0; i < 2; i++)
848 			for (j = 0; j < 2; j++)
849 				for (k = 0; k < 2; k++) {
850 					pfs->pcounters[i][j][k] +=
851 					    pf_counter_u64_fetch(&p->pfik_packets[i][j][k]);
852 					pfs->bcounters[i][j] +=
853 					    pf_counter_u64_fetch(&p->pfik_bytes[i][j][k]);
854 				}
855 	}
856 }
857 
858 static void
859 pf_kkif_to_kif(struct pfi_kkif *kkif, struct pfi_kif *kif)
860 {
861 
862 	memset(kif, 0, sizeof(*kif));
863 	strlcpy(kif->pfik_name, kkif->pfik_name, sizeof(kif->pfik_name));
864 	for (int i = 0; i < 2; i++) {
865 		for (int j = 0; j < 2; j++) {
866 			for (int k = 0; k < 2; k++) {
867 				kif->pfik_packets[i][j][k] =
868 				    pf_counter_u64_fetch(&kkif->pfik_packets[i][j][k]);
869 				kif->pfik_bytes[i][j][k] =
870 				    pf_counter_u64_fetch(&kkif->pfik_bytes[i][j][k]);
871 			}
872 		}
873 	}
874 	kif->pfik_flags = kkif->pfik_flags;
875 	kif->pfik_tzero = kkif->pfik_tzero;
876 	kif->pfik_rulerefs = kkif->pfik_rulerefs;
877 	/*
878 	 * Userspace relies on this pointer to decide if this is a group or
879 	 * not. We don't want to share the actual pointer, because it's
880 	 * useless to userspace and leaks kernel memory layout information.
881 	 * So instead we provide 0xfeedcode as 'true' and NULL as 'false'.
882 	 */
883 	kif->pfik_group =
884 	    kkif->pfik_group ? (struct ifg_group *)0xfeedc0de : NULL;
885 }
886 
887 void
888 pfi_get_ifaces(const char *name, struct pfi_kif *buf, int *size)
889 {
890 	struct epoch_tracker et;
891 	struct pfi_kkif	*p, *nextp;
892 	int		 n = 0;
893 
894 	NET_EPOCH_ENTER(et);
895 	for (p = RB_MIN(pfi_ifhead, &V_pfi_ifs); p; p = nextp) {
896 		nextp = RB_NEXT(pfi_ifhead, &V_pfi_ifs, p);
897 		if (pfi_skip_if(name, p))
898 			continue;
899 		if (*size <= n++)
900 			break;
901 		if (!p->pfik_tzero)
902 			p->pfik_tzero = time_second;
903 		pf_kkif_to_kif(p, buf++);
904 		nextp = RB_NEXT(pfi_ifhead, &V_pfi_ifs, p);
905 	}
906 	*size = n;
907 	NET_EPOCH_EXIT(et);
908 }
909 
910 static int
911 pfi_skip_if(const char *filter, struct pfi_kkif *p)
912 {
913 	struct ifg_list *i;
914 	int	n;
915 
916 	NET_EPOCH_ASSERT();
917 
918 	if (filter == NULL || !*filter)
919 		return (0);
920 	if (!strcmp(p->pfik_name, filter))
921 		return (0);	/* exact match */
922 	n = strlen(filter);
923 	if (n < 1 || n >= IFNAMSIZ)
924 		return (1);	/* sanity check */
925 	if (filter[n-1] >= '0' && filter[n-1] <= '9')
926 		return (1);	/* group names may not end in a digit */
927 	if (p->pfik_ifp == NULL)
928 		return (1);
929 	CK_STAILQ_FOREACH(i, &p->pfik_ifp->if_groups, ifgl_next)
930 		if (!strncmp(i->ifgl_group->ifg_group, filter, IFNAMSIZ))
931 			return (0); /* iface is in group "filter" */
932 	return (1);
933 }
934 
935 int
936 pfi_set_flags(const char *name, int flags)
937 {
938 	struct epoch_tracker et;
939 	struct pfi_kkif	*p, *kif;
940 
941 	kif = pf_kkif_create(M_NOWAIT);
942 	if (kif == NULL)
943 		return (ENOMEM);
944 
945 	NET_EPOCH_ENTER(et);
946 
947 	kif = pfi_kkif_attach(kif, name);
948 
949 	RB_FOREACH(p, pfi_ifhead, &V_pfi_ifs) {
950 		if (pfi_skip_if(name, p))
951 			continue;
952 		p->pfik_flags |= flags;
953 	}
954 	NET_EPOCH_EXIT(et);
955 	return (0);
956 }
957 
958 int
959 pfi_clear_flags(const char *name, int flags)
960 {
961 	struct epoch_tracker et;
962 	struct pfi_kkif *p, *tmp;
963 
964 	NET_EPOCH_ENTER(et);
965 	RB_FOREACH_SAFE(p, pfi_ifhead, &V_pfi_ifs, tmp) {
966 		if (pfi_skip_if(name, p))
967 			continue;
968 		p->pfik_flags &= ~flags;
969 
970 		if (p->pfik_ifp == NULL && p->pfik_group == NULL &&
971 		    p->pfik_flags == 0 && p->pfik_rulerefs == 0) {
972 			/* Delete this kif. */
973 			RB_REMOVE(pfi_ifhead, &V_pfi_ifs, p);
974 			pf_kkif_free(p);
975 		}
976 	}
977 	NET_EPOCH_EXIT(et);
978 	return (0);
979 }
980 
981 /* from pf_print_state.c */
982 static int
983 pfi_unmask(void *addr)
984 {
985 	struct pf_addr *m = addr;
986 	int i = 31, j = 0, b = 0;
987 	u_int32_t tmp;
988 
989 	while (j < 4 && m->addr32[j] == 0xffffffff) {
990 		b += 32;
991 		j++;
992 	}
993 	if (j < 4) {
994 		tmp = ntohl(m->addr32[j]);
995 		for (i = 31; tmp & (1 << i); --i)
996 			b++;
997 	}
998 	return (b);
999 }
1000 
1001 static void
1002 pfi_attach_ifnet_event(void *arg __unused, struct ifnet *ifp)
1003 {
1004 	struct epoch_tracker et;
1005 	struct pfi_kkif *kif;
1006 
1007 	if (V_pf_vnet_active == 0) {
1008 		/* Avoid teardown race in the least expensive way. */
1009 		return;
1010 	}
1011 	kif = pf_kkif_create(M_NOWAIT);
1012 	NET_EPOCH_ENTER(et);
1013 	PF_RULES_WLOCK();
1014 	pfi_attach_ifnet(ifp, kif);
1015 #ifdef ALTQ
1016 	pf_altq_ifnet_event(ifp, 0);
1017 #endif
1018 	PF_RULES_WUNLOCK();
1019 	NET_EPOCH_EXIT(et);
1020 }
1021 
1022 static void
1023 pfi_detach_ifnet_event(void *arg __unused, struct ifnet *ifp)
1024 {
1025 	struct epoch_tracker et;
1026 	struct pfi_kkif *kif = (struct pfi_kkif *)ifp->if_pf_kif;
1027 
1028 	if (pfsync_detach_ifnet_ptr)
1029 		pfsync_detach_ifnet_ptr(ifp);
1030 
1031 	if (kif == NULL)
1032 		return;
1033 
1034 	if (V_pf_vnet_active == 0) {
1035 		/* Avoid teardown race in the least expensive way. */
1036 		return;
1037 	}
1038 
1039 	NET_EPOCH_ENTER(et);
1040 	PF_RULES_WLOCK();
1041 	V_pfi_update++;
1042 	pfi_kkif_update(kif);
1043 
1044 	if (kif->pfik_ifp)
1045 		if_rele(kif->pfik_ifp);
1046 
1047 	kif->pfik_ifp = NULL;
1048 	ifp->if_pf_kif = NULL;
1049 #ifdef ALTQ
1050 	pf_altq_ifnet_event(ifp, 1);
1051 #endif
1052 	pfi_kkif_remove_if_unref(kif);
1053 
1054 	PF_RULES_WUNLOCK();
1055 	NET_EPOCH_EXIT(et);
1056 }
1057 
1058 static void
1059 pfi_attach_group_event(void *arg __unused, struct ifg_group *ifg)
1060 {
1061 	struct epoch_tracker et;
1062 	struct pfi_kkif *kif;
1063 
1064 	if (V_pf_vnet_active == 0) {
1065 		/* Avoid teardown race in the least expensive way. */
1066 		return;
1067 	}
1068 	kif = pf_kkif_create(M_WAITOK);
1069 	NET_EPOCH_ENTER(et);
1070 	PF_RULES_WLOCK();
1071 	pfi_attach_ifgroup(ifg, kif);
1072 	PF_RULES_WUNLOCK();
1073 	NET_EPOCH_EXIT(et);
1074 }
1075 
1076 static void
1077 pfi_change_group_event(void *arg __unused, char *gname)
1078 {
1079 	struct epoch_tracker et;
1080 	struct pfi_kkif *kif;
1081 
1082 	if (V_pf_vnet_active == 0) {
1083 		/* Avoid teardown race in the least expensive way. */
1084 		return;
1085 	}
1086 
1087 	kif = pf_kkif_create(M_WAITOK);
1088 	NET_EPOCH_ENTER(et);
1089 	PF_RULES_WLOCK();
1090 	V_pfi_update++;
1091 	kif = pfi_kkif_attach(kif, gname);
1092 	pfi_kkif_update(kif);
1093 	PF_RULES_WUNLOCK();
1094 	NET_EPOCH_EXIT(et);
1095 }
1096 
1097 static void
1098 pfi_detach_group_event(void *arg __unused, struct ifg_group *ifg)
1099 {
1100 	struct pfi_kkif *kif = (struct pfi_kkif *)ifg->ifg_pf_kif;
1101 
1102 	if (kif == NULL)
1103 		return;
1104 
1105 	if (V_pf_vnet_active == 0) {
1106 		/* Avoid teardown race in the least expensive way. */
1107 		return;
1108 	}
1109 	PF_RULES_WLOCK();
1110 	V_pfi_update++;
1111 
1112 	kif->pfik_group = NULL;
1113 	ifg->ifg_pf_kif = NULL;
1114 
1115 	pfi_kkif_remove_if_unref(kif);
1116 	PF_RULES_WUNLOCK();
1117 }
1118 
1119 static void
1120 pfi_ifaddr_event(void *arg __unused, struct ifnet *ifp)
1121 {
1122 
1123 	KASSERT(ifp, ("ifp == NULL"));
1124 
1125 	if (ifp->if_pf_kif == NULL)
1126 		return;
1127 
1128 	if (V_pf_vnet_active == 0) {
1129 		/* Avoid teardown race in the least expensive way. */
1130 		return;
1131 	}
1132 	PF_RULES_WLOCK();
1133 	if (ifp->if_pf_kif) {
1134 		struct epoch_tracker et;
1135 
1136 		V_pfi_update++;
1137 		NET_EPOCH_ENTER(et);
1138 		pfi_kkif_update(ifp->if_pf_kif);
1139 		NET_EPOCH_EXIT(et);
1140 	}
1141 	PF_RULES_WUNLOCK();
1142 }
1143