xref: /freebsd/sys/netpfil/pf/if_pfsync.c (revision cd0d51baaa4509a1db83251a601d34404d20c990)
1 /*-
2  * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND ISC)
3  *
4  * Copyright (c) 2002 Michael Shalayeff
5  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
21  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
22  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
23  * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
25  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
26  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
27  * THE POSSIBILITY OF SUCH DAMAGE.
28  */
29 
30 /*-
31  * Copyright (c) 2009 David Gwynne <dlg@openbsd.org>
32  *
33  * Permission to use, copy, modify, and distribute this software for any
34  * purpose with or without fee is hereby granted, provided that the above
35  * copyright notice and this permission notice appear in all copies.
36  *
37  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
38  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
39  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
40  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
41  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
42  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
43  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
44  */
45 
46 /*
47  * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $
48  *
49  * Revisions picked from OpenBSD after revision 1.110 import:
50  * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input()
51  * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates
52  * 1.120, 1.175 - use monotonic time_uptime
53  * 1.122 - reduce number of updates for non-TCP sessions
54  * 1.125, 1.127 - rewrite merge or stale processing
55  * 1.128 - cleanups
56  * 1.146 - bzero() mbuf before sparsely filling it with data
57  * 1.170 - SIOCSIFMTU checks
58  * 1.126, 1.142 - deferred packets processing
59  * 1.173 - correct expire time processing
60  */
61 
62 #include <sys/cdefs.h>
63 __FBSDID("$FreeBSD$");
64 
65 #include "opt_inet.h"
66 #include "opt_inet6.h"
67 #include "opt_pf.h"
68 
69 #include <sys/param.h>
70 #include <sys/bus.h>
71 #include <sys/endian.h>
72 #include <sys/interrupt.h>
73 #include <sys/kernel.h>
74 #include <sys/lock.h>
75 #include <sys/mbuf.h>
76 #include <sys/module.h>
77 #include <sys/mutex.h>
78 #include <sys/priv.h>
79 #include <sys/protosw.h>
80 #include <sys/smp.h>
81 #include <sys/socket.h>
82 #include <sys/sockio.h>
83 #include <sys/sysctl.h>
84 #include <sys/syslog.h>
85 
86 #include <net/bpf.h>
87 #include <net/if.h>
88 #include <net/if_var.h>
89 #include <net/if_clone.h>
90 #include <net/if_types.h>
91 #include <net/vnet.h>
92 #include <net/pfvar.h>
93 #include <net/if_pfsync.h>
94 
95 #include <netinet/if_ether.h>
96 #include <netinet/in.h>
97 #include <netinet/in_var.h>
98 #include <netinet/ip.h>
99 #include <netinet/ip_carp.h>
100 #include <netinet/ip_var.h>
101 #include <netinet/tcp.h>
102 #include <netinet/tcp_fsm.h>
103 #include <netinet/tcp_seq.h>
104 
105 #define PFSYNC_MINPKT ( \
106 	sizeof(struct ip) + \
107 	sizeof(struct pfsync_header) + \
108 	sizeof(struct pfsync_subheader) )
109 
110 struct pfsync_bucket;
111 
112 struct pfsync_pkt {
113 	struct ip *ip;
114 	struct in_addr src;
115 	u_int8_t flags;
116 };
117 
118 static int	pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *,
119 		    struct pfsync_state_peer *);
120 static int	pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int);
121 static int	pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int);
122 static int	pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int);
123 static int	pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int);
124 static int	pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int);
125 static int	pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int);
126 static int	pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int);
127 static int	pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int);
128 static int	pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int);
129 static int	pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int);
130 static int	pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int);
131 static int	pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int);
132 
133 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = {
134 	pfsync_in_clr,			/* PFSYNC_ACT_CLR */
135 	pfsync_in_ins,			/* PFSYNC_ACT_INS */
136 	pfsync_in_iack,			/* PFSYNC_ACT_INS_ACK */
137 	pfsync_in_upd,			/* PFSYNC_ACT_UPD */
138 	pfsync_in_upd_c,		/* PFSYNC_ACT_UPD_C */
139 	pfsync_in_ureq,			/* PFSYNC_ACT_UPD_REQ */
140 	pfsync_in_del,			/* PFSYNC_ACT_DEL */
141 	pfsync_in_del_c,		/* PFSYNC_ACT_DEL_C */
142 	pfsync_in_error,		/* PFSYNC_ACT_INS_F */
143 	pfsync_in_error,		/* PFSYNC_ACT_DEL_F */
144 	pfsync_in_bus,			/* PFSYNC_ACT_BUS */
145 	pfsync_in_tdb,			/* PFSYNC_ACT_TDB */
146 	pfsync_in_eof			/* PFSYNC_ACT_EOF */
147 };
148 
149 struct pfsync_q {
150 	void		(*write)(struct pf_state *, void *);
151 	size_t		len;
152 	u_int8_t	action;
153 };
154 
155 /* we have one of these for every PFSYNC_S_ */
156 static void	pfsync_out_state(struct pf_state *, void *);
157 static void	pfsync_out_iack(struct pf_state *, void *);
158 static void	pfsync_out_upd_c(struct pf_state *, void *);
159 static void	pfsync_out_del(struct pf_state *, void *);
160 
161 static struct pfsync_q pfsync_qs[] = {
162 	{ pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_INS },
163 	{ pfsync_out_iack,  sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
164 	{ pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_UPD },
165 	{ pfsync_out_upd_c, sizeof(struct pfsync_upd_c),   PFSYNC_ACT_UPD_C },
166 	{ pfsync_out_del,   sizeof(struct pfsync_del_c),   PFSYNC_ACT_DEL_C }
167 };
168 
169 static void	pfsync_q_ins(struct pf_state *, int, bool);
170 static void	pfsync_q_del(struct pf_state *, bool, struct pfsync_bucket *);
171 
172 static void	pfsync_update_state(struct pf_state *);
173 
174 struct pfsync_upd_req_item {
175 	TAILQ_ENTRY(pfsync_upd_req_item)	ur_entry;
176 	struct pfsync_upd_req			ur_msg;
177 };
178 
179 struct pfsync_deferral {
180 	struct pfsync_softc		*pd_sc;
181 	TAILQ_ENTRY(pfsync_deferral)	pd_entry;
182 	u_int				pd_refs;
183 	struct callout			pd_tmo;
184 
185 	struct pf_state			*pd_st;
186 	struct mbuf			*pd_m;
187 };
188 
189 struct pfsync_sofct;
190 
191 struct pfsync_bucket
192 {
193 	int			b_id;
194 	struct pfsync_softc	*b_sc;
195 	struct mtx		b_mtx;
196 	struct callout		b_tmo;
197 	int			b_flags;
198 #define	PFSYNCF_BUCKET_PUSH	0x00000001
199 
200 	size_t			b_len;
201 	TAILQ_HEAD(, pf_state)			b_qs[PFSYNC_S_COUNT];
202 	TAILQ_HEAD(, pfsync_upd_req_item)	b_upd_req_list;
203 	TAILQ_HEAD(, pfsync_deferral)		b_deferrals;
204 	u_int			b_deferred;
205 	void			*b_plus;
206 	size_t			b_pluslen;
207 
208 	struct  ifaltq b_snd;
209 };
210 
211 struct pfsync_softc {
212 	/* Configuration */
213 	struct ifnet		*sc_ifp;
214 	struct ifnet		*sc_sync_if;
215 	struct ip_moptions	sc_imo;
216 	struct in_addr		sc_sync_peer;
217 	uint32_t		sc_flags;
218 #define	PFSYNCF_OK		0x00000001
219 #define	PFSYNCF_DEFER		0x00000002
220 	uint8_t			sc_maxupdates;
221 	struct ip		sc_template;
222 	struct mtx		sc_mtx;
223 
224 	/* Queued data */
225 	struct pfsync_bucket	*sc_buckets;
226 
227 	/* Bulk update info */
228 	struct mtx		sc_bulk_mtx;
229 	uint32_t		sc_ureq_sent;
230 	int			sc_bulk_tries;
231 	uint32_t		sc_ureq_received;
232 	int			sc_bulk_hashid;
233 	uint64_t		sc_bulk_stateid;
234 	uint32_t		sc_bulk_creatorid;
235 	struct callout		sc_bulk_tmo;
236 	struct callout		sc_bulkfail_tmo;
237 };
238 
239 #define	PFSYNC_LOCK(sc)		mtx_lock(&(sc)->sc_mtx)
240 #define	PFSYNC_UNLOCK(sc)	mtx_unlock(&(sc)->sc_mtx)
241 #define	PFSYNC_LOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_mtx, MA_OWNED)
242 
243 #define PFSYNC_BUCKET_LOCK(b)		mtx_lock(&(b)->b_mtx)
244 #define PFSYNC_BUCKET_UNLOCK(b)		mtx_unlock(&(b)->b_mtx)
245 #define PFSYNC_BUCKET_LOCK_ASSERT(b)	mtx_assert(&(b)->b_mtx, MA_OWNED)
246 
247 #define	PFSYNC_BLOCK(sc)	mtx_lock(&(sc)->sc_bulk_mtx)
248 #define	PFSYNC_BUNLOCK(sc)	mtx_unlock(&(sc)->sc_bulk_mtx)
249 #define	PFSYNC_BLOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
250 
251 static const char pfsyncname[] = "pfsync";
252 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
253 VNET_DEFINE_STATIC(struct pfsync_softc	*, pfsyncif) = NULL;
254 #define	V_pfsyncif		VNET(pfsyncif)
255 VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL;
256 #define	V_pfsync_swi_cookie	VNET(pfsync_swi_cookie)
257 VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats);
258 #define	V_pfsyncstats		VNET(pfsyncstats)
259 VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW;
260 #define	V_pfsync_carp_adj	VNET(pfsync_carp_adj)
261 
262 static void	pfsync_timeout(void *);
263 static void	pfsync_push(struct pfsync_bucket *);
264 static void	pfsync_push_all(struct pfsync_softc *);
265 static void	pfsyncintr(void *);
266 static int	pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
267 		    struct in_mfilter *imf);
268 static void	pfsync_multicast_cleanup(struct pfsync_softc *);
269 static void	pfsync_pointers_init(void);
270 static void	pfsync_pointers_uninit(void);
271 static int	pfsync_init(void);
272 static void	pfsync_uninit(void);
273 
274 static unsigned long pfsync_buckets;
275 
276 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC");
277 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW,
278     &VNET_NAME(pfsyncstats), pfsyncstats,
279     "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
280 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW,
281     &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
282 SYSCTL_ULONG(_net_pfsync, OID_AUTO, pfsync_buckets, CTLFLAG_RDTUN,
283     &pfsync_buckets, 0, "Number of pfsync hash buckets");
284 
285 static int	pfsync_clone_create(struct if_clone *, int, caddr_t);
286 static void	pfsync_clone_destroy(struct ifnet *);
287 static int	pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
288 		    struct pf_state_peer *);
289 static int	pfsyncoutput(struct ifnet *, struct mbuf *,
290 		    const struct sockaddr *, struct route *);
291 static int	pfsyncioctl(struct ifnet *, u_long, caddr_t);
292 
293 static int	pfsync_defer(struct pf_state *, struct mbuf *);
294 static void	pfsync_undefer(struct pfsync_deferral *, int);
295 static void	pfsync_undefer_state(struct pf_state *, int);
296 static void	pfsync_defer_tmo(void *);
297 
298 static void	pfsync_request_update(u_int32_t, u_int64_t);
299 static bool	pfsync_update_state_req(struct pf_state *);
300 
301 static void	pfsync_drop(struct pfsync_softc *);
302 static void	pfsync_sendout(int, int);
303 static void	pfsync_send_plus(void *, size_t);
304 
305 static void	pfsync_bulk_start(void);
306 static void	pfsync_bulk_status(u_int8_t);
307 static void	pfsync_bulk_update(void *);
308 static void	pfsync_bulk_fail(void *);
309 
310 static void	pfsync_detach_ifnet(struct ifnet *);
311 #ifdef IPSEC
312 static void	pfsync_update_net_tdb(struct pfsync_tdb *);
313 #endif
314 static struct pfsync_bucket	*pfsync_get_bucket(struct pfsync_softc *,
315 		    struct pf_state *);
316 
317 
318 #define PFSYNC_MAX_BULKTRIES	12
319 
320 VNET_DEFINE(struct if_clone *, pfsync_cloner);
321 #define	V_pfsync_cloner	VNET(pfsync_cloner)
322 
323 static int
324 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
325 {
326 	struct pfsync_softc *sc;
327 	struct ifnet *ifp;
328 	struct pfsync_bucket *b;
329 	int c, q;
330 
331 	if (unit != 0)
332 		return (EINVAL);
333 
334 	if (! pfsync_buckets)
335 		pfsync_buckets = mp_ncpus * 2;
336 
337 	sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
338 	sc->sc_flags |= PFSYNCF_OK;
339 	sc->sc_maxupdates = 128;
340 
341 	ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
342 	if (ifp == NULL) {
343 		free(sc, M_PFSYNC);
344 		return (ENOSPC);
345 	}
346 	if_initname(ifp, pfsyncname, unit);
347 	ifp->if_softc = sc;
348 	ifp->if_ioctl = pfsyncioctl;
349 	ifp->if_output = pfsyncoutput;
350 	ifp->if_type = IFT_PFSYNC;
351 	ifp->if_hdrlen = sizeof(struct pfsync_header);
352 	ifp->if_mtu = ETHERMTU;
353 	mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
354 	mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
355 	callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
356 	callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
357 
358 	if_attach(ifp);
359 
360 	bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
361 
362 	sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets),
363 	    M_PFSYNC, M_ZERO | M_WAITOK);
364 	for (c = 0; c < pfsync_buckets; c++) {
365 		b = &sc->sc_buckets[c];
366 		mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF);
367 
368 		b->b_id = c;
369 		b->b_sc = sc;
370 		b->b_len = PFSYNC_MINPKT;
371 
372 		for (q = 0; q < PFSYNC_S_COUNT; q++)
373 			TAILQ_INIT(&b->b_qs[q]);
374 
375 		TAILQ_INIT(&b->b_upd_req_list);
376 		TAILQ_INIT(&b->b_deferrals);
377 
378 		callout_init(&b->b_tmo, 1);
379 
380 		b->b_snd.ifq_maxlen = ifqmaxlen;
381 	}
382 
383 	V_pfsyncif = sc;
384 
385 	return (0);
386 }
387 
388 static void
389 pfsync_clone_destroy(struct ifnet *ifp)
390 {
391 	struct pfsync_softc *sc = ifp->if_softc;
392 	struct pfsync_bucket *b;
393 	int c;
394 
395 	for (c = 0; c < pfsync_buckets; c++) {
396 		b = &sc->sc_buckets[c];
397 		/*
398 		 * At this stage, everything should have already been
399 		 * cleared by pfsync_uninit(), and we have only to
400 		 * drain callouts.
401 		 */
402 		while (b->b_deferred > 0) {
403 			struct pfsync_deferral *pd =
404 			    TAILQ_FIRST(&b->b_deferrals);
405 
406 			TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
407 			b->b_deferred--;
408 			if (callout_stop(&pd->pd_tmo) > 0) {
409 				pf_release_state(pd->pd_st);
410 				m_freem(pd->pd_m);
411 				free(pd, M_PFSYNC);
412 			} else {
413 				pd->pd_refs++;
414 				callout_drain(&pd->pd_tmo);
415 				free(pd, M_PFSYNC);
416 			}
417 		}
418 
419 		callout_drain(&b->b_tmo);
420 	}
421 
422 	callout_drain(&sc->sc_bulkfail_tmo);
423 	callout_drain(&sc->sc_bulk_tmo);
424 
425 	if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
426 		(*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
427 	bpfdetach(ifp);
428 	if_detach(ifp);
429 
430 	pfsync_drop(sc);
431 
432 	if_free(ifp);
433 	pfsync_multicast_cleanup(sc);
434 	mtx_destroy(&sc->sc_mtx);
435 	mtx_destroy(&sc->sc_bulk_mtx);
436 
437 	free(sc->sc_buckets, M_PFSYNC);
438 	free(sc, M_PFSYNC);
439 
440 	V_pfsyncif = NULL;
441 }
442 
443 static int
444 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
445     struct pf_state_peer *d)
446 {
447 	if (s->scrub.scrub_flag && d->scrub == NULL) {
448 		d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
449 		if (d->scrub == NULL)
450 			return (ENOMEM);
451 	}
452 
453 	return (0);
454 }
455 
456 
457 static int
458 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
459 {
460 	struct pfsync_softc *sc = V_pfsyncif;
461 #ifndef	__NO_STRICT_ALIGNMENT
462 	struct pfsync_state_key key[2];
463 #endif
464 	struct pfsync_state_key *kw, *ks;
465 	struct pf_state	*st = NULL;
466 	struct pf_state_key *skw = NULL, *sks = NULL;
467 	struct pf_rule *r = NULL;
468 	struct pfi_kif	*kif;
469 	int error;
470 
471 	PF_RULES_RASSERT();
472 
473 	if (sp->creatorid == 0) {
474 		if (V_pf_status.debug >= PF_DEBUG_MISC)
475 			printf("%s: invalid creator id: %08x\n", __func__,
476 			    ntohl(sp->creatorid));
477 		return (EINVAL);
478 	}
479 
480 	if ((kif = pfi_kif_find(sp->ifname)) == NULL) {
481 		if (V_pf_status.debug >= PF_DEBUG_MISC)
482 			printf("%s: unknown interface: %s\n", __func__,
483 			    sp->ifname);
484 		if (flags & PFSYNC_SI_IOCTL)
485 			return (EINVAL);
486 		return (0);	/* skip this state */
487 	}
488 
489 	/*
490 	 * If the ruleset checksums match or the state is coming from the ioctl,
491 	 * it's safe to associate the state with the rule of that number.
492 	 */
493 	if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
494 	    (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
495 	    pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
496 		r = pf_main_ruleset.rules[
497 		    PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
498 	else
499 		r = &V_pf_default_rule;
500 
501 	if ((r->max_states &&
502 	    counter_u64_fetch(r->states_cur) >= r->max_states))
503 		goto cleanup;
504 
505 	/*
506 	 * XXXGL: consider M_WAITOK in ioctl path after.
507 	 */
508 	if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL)
509 		goto cleanup;
510 
511 	if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
512 		goto cleanup;
513 
514 #ifndef	__NO_STRICT_ALIGNMENT
515 	bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2);
516 	kw = &key[PF_SK_WIRE];
517 	ks = &key[PF_SK_STACK];
518 #else
519 	kw = &sp->key[PF_SK_WIRE];
520 	ks = &sp->key[PF_SK_STACK];
521 #endif
522 
523 	if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) ||
524 	    PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) ||
525 	    kw->port[0] != ks->port[0] ||
526 	    kw->port[1] != ks->port[1]) {
527 		sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
528 		if (sks == NULL)
529 			goto cleanup;
530 	} else
531 		sks = skw;
532 
533 	/* allocate memory for scrub info */
534 	if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
535 	    pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
536 		goto cleanup;
537 
538 	/* Copy to state key(s). */
539 	skw->addr[0] = kw->addr[0];
540 	skw->addr[1] = kw->addr[1];
541 	skw->port[0] = kw->port[0];
542 	skw->port[1] = kw->port[1];
543 	skw->proto = sp->proto;
544 	skw->af = sp->af;
545 	if (sks != skw) {
546 		sks->addr[0] = ks->addr[0];
547 		sks->addr[1] = ks->addr[1];
548 		sks->port[0] = ks->port[0];
549 		sks->port[1] = ks->port[1];
550 		sks->proto = sp->proto;
551 		sks->af = sp->af;
552 	}
553 
554 	/* copy to state */
555 	bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
556 	st->creation = time_uptime - ntohl(sp->creation);
557 	st->expire = time_uptime;
558 	if (sp->expire) {
559 		uint32_t timeout;
560 
561 		timeout = r->timeout[sp->timeout];
562 		if (!timeout)
563 			timeout = V_pf_default_rule.timeout[sp->timeout];
564 
565 		/* sp->expire may have been adaptively scaled by export. */
566 		st->expire -= timeout - ntohl(sp->expire);
567 	}
568 
569 	st->direction = sp->direction;
570 	st->log = sp->log;
571 	st->timeout = sp->timeout;
572 	st->state_flags = sp->state_flags;
573 
574 	st->id = sp->id;
575 	st->creatorid = sp->creatorid;
576 	pf_state_peer_ntoh(&sp->src, &st->src);
577 	pf_state_peer_ntoh(&sp->dst, &st->dst);
578 
579 	st->rule.ptr = r;
580 	st->nat_rule.ptr = NULL;
581 	st->anchor.ptr = NULL;
582 	st->rt_kif = NULL;
583 
584 	st->pfsync_time = time_uptime;
585 	st->sync_state = PFSYNC_S_NONE;
586 
587 	if (!(flags & PFSYNC_SI_IOCTL))
588 		st->state_flags |= PFSTATE_NOSYNC;
589 
590 	if ((error = pf_state_insert(kif, skw, sks, st)) != 0)
591 		goto cleanup_state;
592 
593 	/* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
594 	counter_u64_add(r->states_cur, 1);
595 	counter_u64_add(r->states_tot, 1);
596 
597 	if (!(flags & PFSYNC_SI_IOCTL)) {
598 		st->state_flags &= ~PFSTATE_NOSYNC;
599 		if (st->state_flags & PFSTATE_ACK) {
600 			pfsync_q_ins(st, PFSYNC_S_IACK, true);
601 			pfsync_push_all(sc);
602 		}
603 	}
604 	st->state_flags &= ~PFSTATE_ACK;
605 	PF_STATE_UNLOCK(st);
606 
607 	return (0);
608 
609 cleanup:
610 	error = ENOMEM;
611 	if (skw == sks)
612 		sks = NULL;
613 	if (skw != NULL)
614 		uma_zfree(V_pf_state_key_z, skw);
615 	if (sks != NULL)
616 		uma_zfree(V_pf_state_key_z, sks);
617 
618 cleanup_state:	/* pf_state_insert() frees the state keys. */
619 	if (st) {
620 		if (st->dst.scrub)
621 			uma_zfree(V_pf_state_scrub_z, st->dst.scrub);
622 		if (st->src.scrub)
623 			uma_zfree(V_pf_state_scrub_z, st->src.scrub);
624 		uma_zfree(V_pf_state_z, st);
625 	}
626 	return (error);
627 }
628 
629 static int
630 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
631 {
632 	struct pfsync_softc *sc = V_pfsyncif;
633 	struct pfsync_pkt pkt;
634 	struct mbuf *m = *mp;
635 	struct ip *ip = mtod(m, struct ip *);
636 	struct pfsync_header *ph;
637 	struct pfsync_subheader subh;
638 
639 	int offset, len;
640 	int rv;
641 	uint16_t count;
642 
643 	PF_RULES_RLOCK_TRACKER;
644 
645 	*mp = NULL;
646 	V_pfsyncstats.pfsyncs_ipackets++;
647 
648 	/* Verify that we have a sync interface configured. */
649 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
650 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
651 		goto done;
652 
653 	/* verify that the packet came in on the right interface */
654 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
655 		V_pfsyncstats.pfsyncs_badif++;
656 		goto done;
657 	}
658 
659 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
660 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
661 	/* verify that the IP TTL is 255. */
662 	if (ip->ip_ttl != PFSYNC_DFLTTL) {
663 		V_pfsyncstats.pfsyncs_badttl++;
664 		goto done;
665 	}
666 
667 	offset = ip->ip_hl << 2;
668 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
669 		V_pfsyncstats.pfsyncs_hdrops++;
670 		goto done;
671 	}
672 
673 	if (offset + sizeof(*ph) > m->m_len) {
674 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
675 			V_pfsyncstats.pfsyncs_hdrops++;
676 			return (IPPROTO_DONE);
677 		}
678 		ip = mtod(m, struct ip *);
679 	}
680 	ph = (struct pfsync_header *)((char *)ip + offset);
681 
682 	/* verify the version */
683 	if (ph->version != PFSYNC_VERSION) {
684 		V_pfsyncstats.pfsyncs_badver++;
685 		goto done;
686 	}
687 
688 	len = ntohs(ph->len) + offset;
689 	if (m->m_pkthdr.len < len) {
690 		V_pfsyncstats.pfsyncs_badlen++;
691 		goto done;
692 	}
693 
694 	/* Cheaper to grab this now than having to mess with mbufs later */
695 	pkt.ip = ip;
696 	pkt.src = ip->ip_src;
697 	pkt.flags = 0;
698 
699 	/*
700 	 * Trusting pf_chksum during packet processing, as well as seeking
701 	 * in interface name tree, require holding PF_RULES_RLOCK().
702 	 */
703 	PF_RULES_RLOCK();
704 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
705 		pkt.flags |= PFSYNC_SI_CKSUM;
706 
707 	offset += sizeof(*ph);
708 	while (offset <= len - sizeof(subh)) {
709 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
710 		offset += sizeof(subh);
711 
712 		if (subh.action >= PFSYNC_ACT_MAX) {
713 			V_pfsyncstats.pfsyncs_badact++;
714 			PF_RULES_RUNLOCK();
715 			goto done;
716 		}
717 
718 		count = ntohs(subh.count);
719 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
720 		rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count);
721 		if (rv == -1) {
722 			PF_RULES_RUNLOCK();
723 			return (IPPROTO_DONE);
724 		}
725 
726 		offset += rv;
727 	}
728 	PF_RULES_RUNLOCK();
729 
730 done:
731 	m_freem(m);
732 	return (IPPROTO_DONE);
733 }
734 
735 static int
736 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
737 {
738 	struct pfsync_clr *clr;
739 	struct mbuf *mp;
740 	int len = sizeof(*clr) * count;
741 	int i, offp;
742 	u_int32_t creatorid;
743 
744 	mp = m_pulldown(m, offset, len, &offp);
745 	if (mp == NULL) {
746 		V_pfsyncstats.pfsyncs_badlen++;
747 		return (-1);
748 	}
749 	clr = (struct pfsync_clr *)(mp->m_data + offp);
750 
751 	for (i = 0; i < count; i++) {
752 		creatorid = clr[i].creatorid;
753 
754 		if (clr[i].ifname[0] != '\0' &&
755 		    pfi_kif_find(clr[i].ifname) == NULL)
756 			continue;
757 
758 		for (int i = 0; i <= pf_hashmask; i++) {
759 			struct pf_idhash *ih = &V_pf_idhash[i];
760 			struct pf_state *s;
761 relock:
762 			PF_HASHROW_LOCK(ih);
763 			LIST_FOREACH(s, &ih->states, entry) {
764 				if (s->creatorid == creatorid) {
765 					s->state_flags |= PFSTATE_NOSYNC;
766 					pf_unlink_state(s, PF_ENTER_LOCKED);
767 					goto relock;
768 				}
769 			}
770 			PF_HASHROW_UNLOCK(ih);
771 		}
772 	}
773 
774 	return (len);
775 }
776 
777 static int
778 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
779 {
780 	struct mbuf *mp;
781 	struct pfsync_state *sa, *sp;
782 	int len = sizeof(*sp) * count;
783 	int i, offp;
784 
785 	mp = m_pulldown(m, offset, len, &offp);
786 	if (mp == NULL) {
787 		V_pfsyncstats.pfsyncs_badlen++;
788 		return (-1);
789 	}
790 	sa = (struct pfsync_state *)(mp->m_data + offp);
791 
792 	for (i = 0; i < count; i++) {
793 		sp = &sa[i];
794 
795 		/* Check for invalid values. */
796 		if (sp->timeout >= PFTM_MAX ||
797 		    sp->src.state > PF_TCPS_PROXY_DST ||
798 		    sp->dst.state > PF_TCPS_PROXY_DST ||
799 		    sp->direction > PF_OUT ||
800 		    (sp->af != AF_INET && sp->af != AF_INET6)) {
801 			if (V_pf_status.debug >= PF_DEBUG_MISC)
802 				printf("%s: invalid value\n", __func__);
803 			V_pfsyncstats.pfsyncs_badval++;
804 			continue;
805 		}
806 
807 		if (pfsync_state_import(sp, pkt->flags) == ENOMEM)
808 			/* Drop out, but process the rest of the actions. */
809 			break;
810 	}
811 
812 	return (len);
813 }
814 
815 static int
816 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
817 {
818 	struct pfsync_ins_ack *ia, *iaa;
819 	struct pf_state *st;
820 
821 	struct mbuf *mp;
822 	int len = count * sizeof(*ia);
823 	int offp, i;
824 
825 	mp = m_pulldown(m, offset, len, &offp);
826 	if (mp == NULL) {
827 		V_pfsyncstats.pfsyncs_badlen++;
828 		return (-1);
829 	}
830 	iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
831 
832 	for (i = 0; i < count; i++) {
833 		ia = &iaa[i];
834 
835 		st = pf_find_state_byid(ia->id, ia->creatorid);
836 		if (st == NULL)
837 			continue;
838 
839 		if (st->state_flags & PFSTATE_ACK) {
840 			pfsync_undefer_state(st, 0);
841 		}
842 		PF_STATE_UNLOCK(st);
843 	}
844 	/*
845 	 * XXX this is not yet implemented, but we know the size of the
846 	 * message so we can skip it.
847 	 */
848 
849 	return (count * sizeof(struct pfsync_ins_ack));
850 }
851 
852 static int
853 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src,
854     struct pfsync_state_peer *dst)
855 {
856 	int sync = 0;
857 
858 	PF_STATE_LOCK_ASSERT(st);
859 
860 	/*
861 	 * The state should never go backwards except
862 	 * for syn-proxy states.  Neither should the
863 	 * sequence window slide backwards.
864 	 */
865 	if ((st->src.state > src->state &&
866 	    (st->src.state < PF_TCPS_PROXY_SRC ||
867 	    src->state >= PF_TCPS_PROXY_SRC)) ||
868 
869 	    (st->src.state == src->state &&
870 	    SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
871 		sync++;
872 	else
873 		pf_state_peer_ntoh(src, &st->src);
874 
875 	if ((st->dst.state > dst->state) ||
876 
877 	    (st->dst.state >= TCPS_SYN_SENT &&
878 	    SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
879 		sync++;
880 	else
881 		pf_state_peer_ntoh(dst, &st->dst);
882 
883 	return (sync);
884 }
885 
886 static int
887 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
888 {
889 	struct pfsync_softc *sc = V_pfsyncif;
890 	struct pfsync_state *sa, *sp;
891 	struct pf_state *st;
892 	int sync;
893 
894 	struct mbuf *mp;
895 	int len = count * sizeof(*sp);
896 	int offp, i;
897 
898 	mp = m_pulldown(m, offset, len, &offp);
899 	if (mp == NULL) {
900 		V_pfsyncstats.pfsyncs_badlen++;
901 		return (-1);
902 	}
903 	sa = (struct pfsync_state *)(mp->m_data + offp);
904 
905 	for (i = 0; i < count; i++) {
906 		sp = &sa[i];
907 
908 		/* check for invalid values */
909 		if (sp->timeout >= PFTM_MAX ||
910 		    sp->src.state > PF_TCPS_PROXY_DST ||
911 		    sp->dst.state > PF_TCPS_PROXY_DST) {
912 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
913 				printf("pfsync_input: PFSYNC_ACT_UPD: "
914 				    "invalid value\n");
915 			}
916 			V_pfsyncstats.pfsyncs_badval++;
917 			continue;
918 		}
919 
920 		st = pf_find_state_byid(sp->id, sp->creatorid);
921 		if (st == NULL) {
922 			/* insert the update */
923 			if (pfsync_state_import(sp, pkt->flags))
924 				V_pfsyncstats.pfsyncs_badstate++;
925 			continue;
926 		}
927 
928 		if (st->state_flags & PFSTATE_ACK) {
929 			pfsync_undefer_state(st, 1);
930 		}
931 
932 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
933 			sync = pfsync_upd_tcp(st, &sp->src, &sp->dst);
934 		else {
935 			sync = 0;
936 
937 			/*
938 			 * Non-TCP protocol state machine always go
939 			 * forwards
940 			 */
941 			if (st->src.state > sp->src.state)
942 				sync++;
943 			else
944 				pf_state_peer_ntoh(&sp->src, &st->src);
945 			if (st->dst.state > sp->dst.state)
946 				sync++;
947 			else
948 				pf_state_peer_ntoh(&sp->dst, &st->dst);
949 		}
950 		if (sync < 2) {
951 			pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
952 			pf_state_peer_ntoh(&sp->dst, &st->dst);
953 			st->expire = time_uptime;
954 			st->timeout = sp->timeout;
955 		}
956 		st->pfsync_time = time_uptime;
957 
958 		if (sync) {
959 			V_pfsyncstats.pfsyncs_stale++;
960 
961 			pfsync_update_state(st);
962 			PF_STATE_UNLOCK(st);
963 			pfsync_push_all(sc);
964 			continue;
965 		}
966 		PF_STATE_UNLOCK(st);
967 	}
968 
969 	return (len);
970 }
971 
972 static int
973 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
974 {
975 	struct pfsync_softc *sc = V_pfsyncif;
976 	struct pfsync_upd_c *ua, *up;
977 	struct pf_state *st;
978 	int len = count * sizeof(*up);
979 	int sync;
980 	struct mbuf *mp;
981 	int offp, i;
982 
983 	mp = m_pulldown(m, offset, len, &offp);
984 	if (mp == NULL) {
985 		V_pfsyncstats.pfsyncs_badlen++;
986 		return (-1);
987 	}
988 	ua = (struct pfsync_upd_c *)(mp->m_data + offp);
989 
990 	for (i = 0; i < count; i++) {
991 		up = &ua[i];
992 
993 		/* check for invalid values */
994 		if (up->timeout >= PFTM_MAX ||
995 		    up->src.state > PF_TCPS_PROXY_DST ||
996 		    up->dst.state > PF_TCPS_PROXY_DST) {
997 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
998 				printf("pfsync_input: "
999 				    "PFSYNC_ACT_UPD_C: "
1000 				    "invalid value\n");
1001 			}
1002 			V_pfsyncstats.pfsyncs_badval++;
1003 			continue;
1004 		}
1005 
1006 		st = pf_find_state_byid(up->id, up->creatorid);
1007 		if (st == NULL) {
1008 			/* We don't have this state. Ask for it. */
1009 			PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1010 			pfsync_request_update(up->creatorid, up->id);
1011 			PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1012 			continue;
1013 		}
1014 
1015 		if (st->state_flags & PFSTATE_ACK) {
1016 			pfsync_undefer_state(st, 1);
1017 		}
1018 
1019 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1020 			sync = pfsync_upd_tcp(st, &up->src, &up->dst);
1021 		else {
1022 			sync = 0;
1023 
1024 			/*
1025 			 * Non-TCP protocol state machine always go
1026 			 * forwards
1027 			 */
1028 			if (st->src.state > up->src.state)
1029 				sync++;
1030 			else
1031 				pf_state_peer_ntoh(&up->src, &st->src);
1032 			if (st->dst.state > up->dst.state)
1033 				sync++;
1034 			else
1035 				pf_state_peer_ntoh(&up->dst, &st->dst);
1036 		}
1037 		if (sync < 2) {
1038 			pfsync_alloc_scrub_memory(&up->dst, &st->dst);
1039 			pf_state_peer_ntoh(&up->dst, &st->dst);
1040 			st->expire = time_uptime;
1041 			st->timeout = up->timeout;
1042 		}
1043 		st->pfsync_time = time_uptime;
1044 
1045 		if (sync) {
1046 			V_pfsyncstats.pfsyncs_stale++;
1047 
1048 			pfsync_update_state(st);
1049 			PF_STATE_UNLOCK(st);
1050 			pfsync_push_all(sc);
1051 			continue;
1052 		}
1053 		PF_STATE_UNLOCK(st);
1054 	}
1055 
1056 	return (len);
1057 }
1058 
1059 static int
1060 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1061 {
1062 	struct pfsync_upd_req *ur, *ura;
1063 	struct mbuf *mp;
1064 	int len = count * sizeof(*ur);
1065 	int i, offp;
1066 
1067 	struct pf_state *st;
1068 
1069 	mp = m_pulldown(m, offset, len, &offp);
1070 	if (mp == NULL) {
1071 		V_pfsyncstats.pfsyncs_badlen++;
1072 		return (-1);
1073 	}
1074 	ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1075 
1076 	for (i = 0; i < count; i++) {
1077 		ur = &ura[i];
1078 
1079 		if (ur->id == 0 && ur->creatorid == 0)
1080 			pfsync_bulk_start();
1081 		else {
1082 			st = pf_find_state_byid(ur->id, ur->creatorid);
1083 			if (st == NULL) {
1084 				V_pfsyncstats.pfsyncs_badstate++;
1085 				continue;
1086 			}
1087 			if (st->state_flags & PFSTATE_NOSYNC) {
1088 				PF_STATE_UNLOCK(st);
1089 				continue;
1090 			}
1091 
1092 			pfsync_update_state_req(st);
1093 			PF_STATE_UNLOCK(st);
1094 		}
1095 	}
1096 
1097 	return (len);
1098 }
1099 
1100 static int
1101 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1102 {
1103 	struct mbuf *mp;
1104 	struct pfsync_state *sa, *sp;
1105 	struct pf_state *st;
1106 	int len = count * sizeof(*sp);
1107 	int offp, i;
1108 
1109 	mp = m_pulldown(m, offset, len, &offp);
1110 	if (mp == NULL) {
1111 		V_pfsyncstats.pfsyncs_badlen++;
1112 		return (-1);
1113 	}
1114 	sa = (struct pfsync_state *)(mp->m_data + offp);
1115 
1116 	for (i = 0; i < count; i++) {
1117 		sp = &sa[i];
1118 
1119 		st = pf_find_state_byid(sp->id, sp->creatorid);
1120 		if (st == NULL) {
1121 			V_pfsyncstats.pfsyncs_badstate++;
1122 			continue;
1123 		}
1124 		st->state_flags |= PFSTATE_NOSYNC;
1125 		pf_unlink_state(st, PF_ENTER_LOCKED);
1126 	}
1127 
1128 	return (len);
1129 }
1130 
1131 static int
1132 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1133 {
1134 	struct mbuf *mp;
1135 	struct pfsync_del_c *sa, *sp;
1136 	struct pf_state *st;
1137 	int len = count * sizeof(*sp);
1138 	int offp, i;
1139 
1140 	mp = m_pulldown(m, offset, len, &offp);
1141 	if (mp == NULL) {
1142 		V_pfsyncstats.pfsyncs_badlen++;
1143 		return (-1);
1144 	}
1145 	sa = (struct pfsync_del_c *)(mp->m_data + offp);
1146 
1147 	for (i = 0; i < count; i++) {
1148 		sp = &sa[i];
1149 
1150 		st = pf_find_state_byid(sp->id, sp->creatorid);
1151 		if (st == NULL) {
1152 			V_pfsyncstats.pfsyncs_badstate++;
1153 			continue;
1154 		}
1155 
1156 		st->state_flags |= PFSTATE_NOSYNC;
1157 		pf_unlink_state(st, PF_ENTER_LOCKED);
1158 	}
1159 
1160 	return (len);
1161 }
1162 
1163 static int
1164 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1165 {
1166 	struct pfsync_softc *sc = V_pfsyncif;
1167 	struct pfsync_bus *bus;
1168 	struct mbuf *mp;
1169 	int len = count * sizeof(*bus);
1170 	int offp;
1171 
1172 	PFSYNC_BLOCK(sc);
1173 
1174 	/* If we're not waiting for a bulk update, who cares. */
1175 	if (sc->sc_ureq_sent == 0) {
1176 		PFSYNC_BUNLOCK(sc);
1177 		return (len);
1178 	}
1179 
1180 	mp = m_pulldown(m, offset, len, &offp);
1181 	if (mp == NULL) {
1182 		PFSYNC_BUNLOCK(sc);
1183 		V_pfsyncstats.pfsyncs_badlen++;
1184 		return (-1);
1185 	}
1186 	bus = (struct pfsync_bus *)(mp->m_data + offp);
1187 
1188 	switch (bus->status) {
1189 	case PFSYNC_BUS_START:
1190 		callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1191 		    V_pf_limits[PF_LIMIT_STATES].limit /
1192 		    ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1193 		    sizeof(struct pfsync_state)),
1194 		    pfsync_bulk_fail, sc);
1195 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1196 			printf("pfsync: received bulk update start\n");
1197 		break;
1198 
1199 	case PFSYNC_BUS_END:
1200 		if (time_uptime - ntohl(bus->endtime) >=
1201 		    sc->sc_ureq_sent) {
1202 			/* that's it, we're happy */
1203 			sc->sc_ureq_sent = 0;
1204 			sc->sc_bulk_tries = 0;
1205 			callout_stop(&sc->sc_bulkfail_tmo);
1206 			if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1207 				(*carp_demote_adj_p)(-V_pfsync_carp_adj,
1208 				    "pfsync bulk done");
1209 			sc->sc_flags |= PFSYNCF_OK;
1210 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1211 				printf("pfsync: received valid "
1212 				    "bulk update end\n");
1213 		} else {
1214 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1215 				printf("pfsync: received invalid "
1216 				    "bulk update end: bad timestamp\n");
1217 		}
1218 		break;
1219 	}
1220 	PFSYNC_BUNLOCK(sc);
1221 
1222 	return (len);
1223 }
1224 
1225 static int
1226 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1227 {
1228 	int len = count * sizeof(struct pfsync_tdb);
1229 
1230 #if defined(IPSEC)
1231 	struct pfsync_tdb *tp;
1232 	struct mbuf *mp;
1233 	int offp;
1234 	int i;
1235 	int s;
1236 
1237 	mp = m_pulldown(m, offset, len, &offp);
1238 	if (mp == NULL) {
1239 		V_pfsyncstats.pfsyncs_badlen++;
1240 		return (-1);
1241 	}
1242 	tp = (struct pfsync_tdb *)(mp->m_data + offp);
1243 
1244 	for (i = 0; i < count; i++)
1245 		pfsync_update_net_tdb(&tp[i]);
1246 #endif
1247 
1248 	return (len);
1249 }
1250 
1251 #if defined(IPSEC)
1252 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1253 static void
1254 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1255 {
1256 	struct tdb		*tdb;
1257 	int			 s;
1258 
1259 	/* check for invalid values */
1260 	if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1261 	    (pt->dst.sa.sa_family != AF_INET &&
1262 	    pt->dst.sa.sa_family != AF_INET6))
1263 		goto bad;
1264 
1265 	tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1266 	if (tdb) {
1267 		pt->rpl = ntohl(pt->rpl);
1268 		pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1269 
1270 		/* Neither replay nor byte counter should ever decrease. */
1271 		if (pt->rpl < tdb->tdb_rpl ||
1272 		    pt->cur_bytes < tdb->tdb_cur_bytes) {
1273 			goto bad;
1274 		}
1275 
1276 		tdb->tdb_rpl = pt->rpl;
1277 		tdb->tdb_cur_bytes = pt->cur_bytes;
1278 	}
1279 	return;
1280 
1281 bad:
1282 	if (V_pf_status.debug >= PF_DEBUG_MISC)
1283 		printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1284 		    "invalid value\n");
1285 	V_pfsyncstats.pfsyncs_badstate++;
1286 	return;
1287 }
1288 #endif
1289 
1290 
1291 static int
1292 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1293 {
1294 	/* check if we are at the right place in the packet */
1295 	if (offset != m->m_pkthdr.len)
1296 		V_pfsyncstats.pfsyncs_badlen++;
1297 
1298 	/* we're done. free and let the caller return */
1299 	m_freem(m);
1300 	return (-1);
1301 }
1302 
1303 static int
1304 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1305 {
1306 	V_pfsyncstats.pfsyncs_badact++;
1307 
1308 	m_freem(m);
1309 	return (-1);
1310 }
1311 
1312 static int
1313 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1314 	struct route *rt)
1315 {
1316 	m_freem(m);
1317 	return (0);
1318 }
1319 
1320 /* ARGSUSED */
1321 static int
1322 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1323 {
1324 	struct pfsync_softc *sc = ifp->if_softc;
1325 	struct ifreq *ifr = (struct ifreq *)data;
1326 	struct pfsyncreq pfsyncr;
1327 	int error;
1328 	int c;
1329 
1330 	switch (cmd) {
1331 	case SIOCSIFFLAGS:
1332 		PFSYNC_LOCK(sc);
1333 		if (ifp->if_flags & IFF_UP) {
1334 			ifp->if_drv_flags |= IFF_DRV_RUNNING;
1335 			PFSYNC_UNLOCK(sc);
1336 			pfsync_pointers_init();
1337 		} else {
1338 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1339 			PFSYNC_UNLOCK(sc);
1340 			pfsync_pointers_uninit();
1341 		}
1342 		break;
1343 	case SIOCSIFMTU:
1344 		if (!sc->sc_sync_if ||
1345 		    ifr->ifr_mtu <= PFSYNC_MINPKT ||
1346 		    ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1347 			return (EINVAL);
1348 		if (ifr->ifr_mtu < ifp->if_mtu) {
1349 			for (c = 0; c < pfsync_buckets; c++) {
1350 				PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1351 				if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT)
1352 					pfsync_sendout(1, c);
1353 				PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1354 			}
1355 		}
1356 		ifp->if_mtu = ifr->ifr_mtu;
1357 		break;
1358 	case SIOCGETPFSYNC:
1359 		bzero(&pfsyncr, sizeof(pfsyncr));
1360 		PFSYNC_LOCK(sc);
1361 		if (sc->sc_sync_if) {
1362 			strlcpy(pfsyncr.pfsyncr_syncdev,
1363 			    sc->sc_sync_if->if_xname, IFNAMSIZ);
1364 		}
1365 		pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1366 		pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1367 		pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER ==
1368 		    (sc->sc_flags & PFSYNCF_DEFER));
1369 		PFSYNC_UNLOCK(sc);
1370 		return (copyout(&pfsyncr, ifr_data_get_ptr(ifr),
1371 		    sizeof(pfsyncr)));
1372 
1373 	case SIOCSETPFSYNC:
1374 	    {
1375 		struct in_mfilter *imf = NULL;
1376 		struct ifnet *sifp;
1377 		struct ip *ip;
1378 
1379 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1380 			return (error);
1381 		if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr,
1382 		    sizeof(pfsyncr))))
1383 			return (error);
1384 
1385 		if (pfsyncr.pfsyncr_maxupdates > 255)
1386 			return (EINVAL);
1387 
1388 		if (pfsyncr.pfsyncr_syncdev[0] == 0)
1389 			sifp = NULL;
1390 		else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL)
1391 			return (EINVAL);
1392 
1393 		if (sifp != NULL && (
1394 		    pfsyncr.pfsyncr_syncpeer.s_addr == 0 ||
1395 		    pfsyncr.pfsyncr_syncpeer.s_addr ==
1396 		    htonl(INADDR_PFSYNC_GROUP)))
1397 			imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
1398 
1399 		PFSYNC_LOCK(sc);
1400 		if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1401 			sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP);
1402 		else
1403 			sc->sc_sync_peer.s_addr =
1404 			    pfsyncr.pfsyncr_syncpeer.s_addr;
1405 
1406 		sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1407 		if (pfsyncr.pfsyncr_defer) {
1408 			sc->sc_flags |= PFSYNCF_DEFER;
1409 			V_pfsync_defer_ptr = pfsync_defer;
1410 		} else {
1411 			sc->sc_flags &= ~PFSYNCF_DEFER;
1412 			V_pfsync_defer_ptr = NULL;
1413 		}
1414 
1415 		if (sifp == NULL) {
1416 			if (sc->sc_sync_if)
1417 				if_rele(sc->sc_sync_if);
1418 			sc->sc_sync_if = NULL;
1419 			pfsync_multicast_cleanup(sc);
1420 			PFSYNC_UNLOCK(sc);
1421 			break;
1422 		}
1423 
1424 		for (c = 0; c < pfsync_buckets; c++) {
1425 			PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1426 			if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT &&
1427 			    (sifp->if_mtu < sc->sc_ifp->if_mtu ||
1428 			    (sc->sc_sync_if != NULL &&
1429 			    sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
1430 			    sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
1431 				pfsync_sendout(1, c);
1432 			PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1433 		}
1434 
1435 		pfsync_multicast_cleanup(sc);
1436 
1437 		if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
1438 			error = pfsync_multicast_setup(sc, sifp, imf);
1439 			if (error) {
1440 				if_rele(sifp);
1441 				ip_mfilter_free(imf);
1442 				PFSYNC_UNLOCK(sc);
1443 				return (error);
1444 			}
1445 		}
1446 		if (sc->sc_sync_if)
1447 			if_rele(sc->sc_sync_if);
1448 		sc->sc_sync_if = sifp;
1449 
1450 		ip = &sc->sc_template;
1451 		bzero(ip, sizeof(*ip));
1452 		ip->ip_v = IPVERSION;
1453 		ip->ip_hl = sizeof(sc->sc_template) >> 2;
1454 		ip->ip_tos = IPTOS_LOWDELAY;
1455 		/* len and id are set later. */
1456 		ip->ip_off = htons(IP_DF);
1457 		ip->ip_ttl = PFSYNC_DFLTTL;
1458 		ip->ip_p = IPPROTO_PFSYNC;
1459 		ip->ip_src.s_addr = INADDR_ANY;
1460 		ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr;
1461 
1462 		/* Request a full state table update. */
1463 		if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1464 			(*carp_demote_adj_p)(V_pfsync_carp_adj,
1465 			    "pfsync bulk start");
1466 		sc->sc_flags &= ~PFSYNCF_OK;
1467 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1468 			printf("pfsync: requesting bulk update\n");
1469 		PFSYNC_UNLOCK(sc);
1470 		PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1471 		pfsync_request_update(0, 0);
1472 		PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1473 		PFSYNC_BLOCK(sc);
1474 		sc->sc_ureq_sent = time_uptime;
1475 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail,
1476 		    sc);
1477 		PFSYNC_BUNLOCK(sc);
1478 
1479 		break;
1480 	    }
1481 	default:
1482 		return (ENOTTY);
1483 	}
1484 
1485 	return (0);
1486 }
1487 
1488 static void
1489 pfsync_out_state(struct pf_state *st, void *buf)
1490 {
1491 	struct pfsync_state *sp = buf;
1492 
1493 	pfsync_state_export(sp, st);
1494 }
1495 
1496 static void
1497 pfsync_out_iack(struct pf_state *st, void *buf)
1498 {
1499 	struct pfsync_ins_ack *iack = buf;
1500 
1501 	iack->id = st->id;
1502 	iack->creatorid = st->creatorid;
1503 }
1504 
1505 static void
1506 pfsync_out_upd_c(struct pf_state *st, void *buf)
1507 {
1508 	struct pfsync_upd_c *up = buf;
1509 
1510 	bzero(up, sizeof(*up));
1511 	up->id = st->id;
1512 	pf_state_peer_hton(&st->src, &up->src);
1513 	pf_state_peer_hton(&st->dst, &up->dst);
1514 	up->creatorid = st->creatorid;
1515 	up->timeout = st->timeout;
1516 }
1517 
1518 static void
1519 pfsync_out_del(struct pf_state *st, void *buf)
1520 {
1521 	struct pfsync_del_c *dp = buf;
1522 
1523 	dp->id = st->id;
1524 	dp->creatorid = st->creatorid;
1525 	st->state_flags |= PFSTATE_NOSYNC;
1526 }
1527 
1528 static void
1529 pfsync_drop(struct pfsync_softc *sc)
1530 {
1531 	struct pf_state *st, *next;
1532 	struct pfsync_upd_req_item *ur;
1533 	struct pfsync_bucket *b;
1534 	int c, q;
1535 
1536 	for (c = 0; c < pfsync_buckets; c++) {
1537 		b = &sc->sc_buckets[c];
1538 		for (q = 0; q < PFSYNC_S_COUNT; q++) {
1539 			if (TAILQ_EMPTY(&b->b_qs[q]))
1540 				continue;
1541 
1542 			TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) {
1543 				KASSERT(st->sync_state == q,
1544 					("%s: st->sync_state == q",
1545 						__func__));
1546 				st->sync_state = PFSYNC_S_NONE;
1547 				pf_release_state(st);
1548 			}
1549 			TAILQ_INIT(&b->b_qs[q]);
1550 		}
1551 
1552 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1553 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1554 			free(ur, M_PFSYNC);
1555 		}
1556 
1557 		b->b_len = PFSYNC_MINPKT;
1558 		b->b_plus = NULL;
1559 	}
1560 }
1561 
1562 static void
1563 pfsync_sendout(int schedswi, int c)
1564 {
1565 	struct pfsync_softc *sc = V_pfsyncif;
1566 	struct ifnet *ifp = sc->sc_ifp;
1567 	struct mbuf *m;
1568 	struct ip *ip;
1569 	struct pfsync_header *ph;
1570 	struct pfsync_subheader *subh;
1571 	struct pf_state *st, *st_next;
1572 	struct pfsync_upd_req_item *ur;
1573 	struct pfsync_bucket *b = &sc->sc_buckets[c];
1574 	int offset;
1575 	int q, count = 0;
1576 
1577 	KASSERT(sc != NULL, ("%s: null sc", __func__));
1578 	KASSERT(b->b_len > PFSYNC_MINPKT,
1579 	    ("%s: sc_len %zu", __func__, b->b_len));
1580 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1581 
1582 	if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1583 		pfsync_drop(sc);
1584 		return;
1585 	}
1586 
1587 	m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1588 	if (m == NULL) {
1589 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1590 		V_pfsyncstats.pfsyncs_onomem++;
1591 		return;
1592 	}
1593 	m->m_data += max_linkhdr;
1594 	m->m_len = m->m_pkthdr.len = b->b_len;
1595 
1596 	/* build the ip header */
1597 	ip = (struct ip *)m->m_data;
1598 	bcopy(&sc->sc_template, ip, sizeof(*ip));
1599 	offset = sizeof(*ip);
1600 
1601 	ip->ip_len = htons(m->m_pkthdr.len);
1602 	ip_fillid(ip);
1603 
1604 	/* build the pfsync header */
1605 	ph = (struct pfsync_header *)(m->m_data + offset);
1606 	bzero(ph, sizeof(*ph));
1607 	offset += sizeof(*ph);
1608 
1609 	ph->version = PFSYNC_VERSION;
1610 	ph->len = htons(b->b_len - sizeof(*ip));
1611 	bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1612 
1613 	/* walk the queues */
1614 	for (q = 0; q < PFSYNC_S_COUNT; q++) {
1615 		if (TAILQ_EMPTY(&b->b_qs[q]))
1616 			continue;
1617 
1618 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1619 		offset += sizeof(*subh);
1620 
1621 		count = 0;
1622 		TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) {
1623 			KASSERT(st->sync_state == q,
1624 				("%s: st->sync_state == q",
1625 					__func__));
1626 			/*
1627 			 * XXXGL: some of write methods do unlocked reads
1628 			 * of state data :(
1629 			 */
1630 			pfsync_qs[q].write(st, m->m_data + offset);
1631 			offset += pfsync_qs[q].len;
1632 			st->sync_state = PFSYNC_S_NONE;
1633 			pf_release_state(st);
1634 			count++;
1635 		}
1636 		TAILQ_INIT(&b->b_qs[q]);
1637 
1638 		bzero(subh, sizeof(*subh));
1639 		subh->action = pfsync_qs[q].action;
1640 		subh->count = htons(count);
1641 		V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1642 	}
1643 
1644 	if (!TAILQ_EMPTY(&b->b_upd_req_list)) {
1645 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1646 		offset += sizeof(*subh);
1647 
1648 		count = 0;
1649 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1650 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1651 
1652 			bcopy(&ur->ur_msg, m->m_data + offset,
1653 			    sizeof(ur->ur_msg));
1654 			offset += sizeof(ur->ur_msg);
1655 			free(ur, M_PFSYNC);
1656 			count++;
1657 		}
1658 
1659 		bzero(subh, sizeof(*subh));
1660 		subh->action = PFSYNC_ACT_UPD_REQ;
1661 		subh->count = htons(count);
1662 		V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1663 	}
1664 
1665 	/* has someone built a custom region for us to add? */
1666 	if (b->b_plus != NULL) {
1667 		bcopy(b->b_plus, m->m_data + offset, b->b_pluslen);
1668 		offset += b->b_pluslen;
1669 
1670 		b->b_plus = NULL;
1671 	}
1672 
1673 	subh = (struct pfsync_subheader *)(m->m_data + offset);
1674 	offset += sizeof(*subh);
1675 
1676 	bzero(subh, sizeof(*subh));
1677 	subh->action = PFSYNC_ACT_EOF;
1678 	subh->count = htons(1);
1679 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1680 
1681 	/* we're done, let's put it on the wire */
1682 	if (ifp->if_bpf) {
1683 		m->m_data += sizeof(*ip);
1684 		m->m_len = m->m_pkthdr.len = b->b_len - sizeof(*ip);
1685 		BPF_MTAP(ifp, m);
1686 		m->m_data -= sizeof(*ip);
1687 		m->m_len = m->m_pkthdr.len = b->b_len;
1688 	}
1689 
1690 	if (sc->sc_sync_if == NULL) {
1691 		b->b_len = PFSYNC_MINPKT;
1692 		m_freem(m);
1693 		return;
1694 	}
1695 
1696 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1697 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1698 	b->b_len = PFSYNC_MINPKT;
1699 
1700 	if (!_IF_QFULL(&b->b_snd))
1701 		_IF_ENQUEUE(&b->b_snd, m);
1702 	else {
1703 		m_freem(m);
1704 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1705 	}
1706 	if (schedswi)
1707 		swi_sched(V_pfsync_swi_cookie, 0);
1708 }
1709 
1710 static void
1711 pfsync_insert_state(struct pf_state *st)
1712 {
1713 	struct pfsync_softc *sc = V_pfsyncif;
1714 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1715 
1716 	if (st->state_flags & PFSTATE_NOSYNC)
1717 		return;
1718 
1719 	if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1720 	    st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1721 		st->state_flags |= PFSTATE_NOSYNC;
1722 		return;
1723 	}
1724 
1725 	KASSERT(st->sync_state == PFSYNC_S_NONE,
1726 		("%s: st->sync_state %u", __func__, st->sync_state));
1727 
1728 	PFSYNC_BUCKET_LOCK(b);
1729 	if (b->b_len == PFSYNC_MINPKT)
1730 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1731 
1732 	pfsync_q_ins(st, PFSYNC_S_INS, true);
1733 	PFSYNC_BUCKET_UNLOCK(b);
1734 
1735 	st->sync_updates = 0;
1736 }
1737 
1738 static int
1739 pfsync_defer(struct pf_state *st, struct mbuf *m)
1740 {
1741 	struct pfsync_softc *sc = V_pfsyncif;
1742 	struct pfsync_deferral *pd;
1743 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1744 
1745 	if (m->m_flags & (M_BCAST|M_MCAST))
1746 		return (0);
1747 
1748 	PFSYNC_LOCK(sc);
1749 
1750 	if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) ||
1751 	    !(sc->sc_flags & PFSYNCF_DEFER)) {
1752 		PFSYNC_UNLOCK(sc);
1753 		return (0);
1754 	}
1755 
1756 	if (b->b_deferred >= 128)
1757 		pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0);
1758 
1759 	pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1760 	if (pd == NULL)
1761 		return (0);
1762 	b->b_deferred++;
1763 
1764 	m->m_flags |= M_SKIP_FIREWALL;
1765 	st->state_flags |= PFSTATE_ACK;
1766 
1767 	pd->pd_sc = sc;
1768 	pd->pd_refs = 0;
1769 	pd->pd_st = st;
1770 	pf_ref_state(st);
1771 	pd->pd_m = m;
1772 
1773 	TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry);
1774 	callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED);
1775 	callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd);
1776 
1777 	pfsync_push(b);
1778 
1779 	return (1);
1780 }
1781 
1782 static void
1783 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1784 {
1785 	struct pfsync_softc *sc = pd->pd_sc;
1786 	struct mbuf *m = pd->pd_m;
1787 	struct pf_state *st = pd->pd_st;
1788 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1789 
1790 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1791 
1792 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
1793 	b->b_deferred--;
1794 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
1795 	free(pd, M_PFSYNC);
1796 	pf_release_state(st);
1797 
1798 	if (drop)
1799 		m_freem(m);
1800 	else {
1801 		_IF_ENQUEUE(&b->b_snd, m);
1802 		pfsync_push(b);
1803 	}
1804 }
1805 
1806 static void
1807 pfsync_defer_tmo(void *arg)
1808 {
1809 	struct pfsync_deferral *pd = arg;
1810 	struct pfsync_softc *sc = pd->pd_sc;
1811 	struct mbuf *m = pd->pd_m;
1812 	struct pf_state *st = pd->pd_st;
1813 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1814 
1815 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1816 
1817 	CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
1818 
1819 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
1820 	b->b_deferred--;
1821 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
1822 	if (pd->pd_refs == 0)
1823 		free(pd, M_PFSYNC);
1824 	PFSYNC_UNLOCK(sc);
1825 
1826 	ip_output(m, NULL, NULL, 0, NULL, NULL);
1827 
1828 	pf_release_state(st);
1829 
1830 	CURVNET_RESTORE();
1831 }
1832 
1833 static void
1834 pfsync_undefer_state(struct pf_state *st, int drop)
1835 {
1836 	struct pfsync_softc *sc = V_pfsyncif;
1837 	struct pfsync_deferral *pd;
1838 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1839 
1840 	PFSYNC_BUCKET_LOCK(b);
1841 
1842 	TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) {
1843 		 if (pd->pd_st == st) {
1844 			if (callout_stop(&pd->pd_tmo) > 0)
1845 				pfsync_undefer(pd, drop);
1846 
1847 			PFSYNC_BUCKET_UNLOCK(b);
1848 			return;
1849 		}
1850 	}
1851 	PFSYNC_BUCKET_UNLOCK(b);
1852 
1853 	panic("%s: unable to find deferred state", __func__);
1854 }
1855 
1856 static struct pfsync_bucket*
1857 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_state *st)
1858 {
1859 	int c = PF_IDHASH(st) % pfsync_buckets;
1860 	return &sc->sc_buckets[c];
1861 }
1862 
1863 static void
1864 pfsync_update_state(struct pf_state *st)
1865 {
1866 	struct pfsync_softc *sc = V_pfsyncif;
1867 	bool sync = false, ref = true;
1868 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1869 
1870 	PF_STATE_LOCK_ASSERT(st);
1871 	PFSYNC_BUCKET_LOCK(b);
1872 
1873 	if (st->state_flags & PFSTATE_ACK)
1874 		pfsync_undefer_state(st, 0);
1875 	if (st->state_flags & PFSTATE_NOSYNC) {
1876 		if (st->sync_state != PFSYNC_S_NONE)
1877 			pfsync_q_del(st, true, b);
1878 		PFSYNC_BUCKET_UNLOCK(b);
1879 		return;
1880 	}
1881 
1882 	if (b->b_len == PFSYNC_MINPKT)
1883 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1884 
1885 	switch (st->sync_state) {
1886 	case PFSYNC_S_UPD_C:
1887 	case PFSYNC_S_UPD:
1888 	case PFSYNC_S_INS:
1889 		/* we're already handling it */
1890 
1891 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1892 			st->sync_updates++;
1893 			if (st->sync_updates >= sc->sc_maxupdates)
1894 				sync = true;
1895 		}
1896 		break;
1897 
1898 	case PFSYNC_S_IACK:
1899 		pfsync_q_del(st, false, b);
1900 		ref = false;
1901 		/* FALLTHROUGH */
1902 
1903 	case PFSYNC_S_NONE:
1904 		pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
1905 		st->sync_updates = 0;
1906 		break;
1907 
1908 	default:
1909 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1910 	}
1911 
1912 	if (sync || (time_uptime - st->pfsync_time) < 2)
1913 		pfsync_push(b);
1914 
1915 	PFSYNC_BUCKET_UNLOCK(b);
1916 }
1917 
1918 static void
1919 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1920 {
1921 	struct pfsync_softc *sc = V_pfsyncif;
1922 	struct pfsync_bucket *b = &sc->sc_buckets[0];
1923 	struct pfsync_upd_req_item *item;
1924 	size_t nlen = sizeof(struct pfsync_upd_req);
1925 
1926 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1927 
1928 	/*
1929 	 * This code does a bit to prevent multiple update requests for the
1930 	 * same state being generated. It searches current subheader queue,
1931 	 * but it doesn't lookup into queue of already packed datagrams.
1932 	 */
1933 	TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry)
1934 		if (item->ur_msg.id == id &&
1935 		    item->ur_msg.creatorid == creatorid)
1936 			return;
1937 
1938 	item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1939 	if (item == NULL)
1940 		return; /* XXX stats */
1941 
1942 	item->ur_msg.id = id;
1943 	item->ur_msg.creatorid = creatorid;
1944 
1945 	if (TAILQ_EMPTY(&b->b_upd_req_list))
1946 		nlen += sizeof(struct pfsync_subheader);
1947 
1948 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
1949 		pfsync_sendout(1, 0);
1950 
1951 		nlen = sizeof(struct pfsync_subheader) +
1952 		    sizeof(struct pfsync_upd_req);
1953 	}
1954 
1955 	TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry);
1956 	b->b_len += nlen;
1957 }
1958 
1959 static bool
1960 pfsync_update_state_req(struct pf_state *st)
1961 {
1962 	struct pfsync_softc *sc = V_pfsyncif;
1963 	bool ref = true, full = false;
1964 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1965 
1966 	PF_STATE_LOCK_ASSERT(st);
1967 	PFSYNC_BUCKET_LOCK(b);
1968 
1969 	if (st->state_flags & PFSTATE_NOSYNC) {
1970 		if (st->sync_state != PFSYNC_S_NONE)
1971 			pfsync_q_del(st, true, b);
1972 		PFSYNC_BUCKET_UNLOCK(b);
1973 		return (full);
1974 	}
1975 
1976 	switch (st->sync_state) {
1977 	case PFSYNC_S_UPD_C:
1978 	case PFSYNC_S_IACK:
1979 		pfsync_q_del(st, false, b);
1980 		ref = false;
1981 		/* FALLTHROUGH */
1982 
1983 	case PFSYNC_S_NONE:
1984 		pfsync_q_ins(st, PFSYNC_S_UPD, ref);
1985 		pfsync_push(b);
1986 		break;
1987 
1988 	case PFSYNC_S_INS:
1989 	case PFSYNC_S_UPD:
1990 	case PFSYNC_S_DEL:
1991 		/* we're already handling it */
1992 		break;
1993 
1994 	default:
1995 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1996 	}
1997 
1998 	if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(struct pfsync_state))
1999 		full = true;
2000 
2001 	PFSYNC_BUCKET_UNLOCK(b);
2002 
2003 	return (full);
2004 }
2005 
2006 static void
2007 pfsync_delete_state(struct pf_state *st)
2008 {
2009 	struct pfsync_softc *sc = V_pfsyncif;
2010 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2011 	bool ref = true;
2012 
2013 	PFSYNC_BUCKET_LOCK(b);
2014 	if (st->state_flags & PFSTATE_ACK)
2015 		pfsync_undefer_state(st, 1);
2016 	if (st->state_flags & PFSTATE_NOSYNC) {
2017 		if (st->sync_state != PFSYNC_S_NONE)
2018 			pfsync_q_del(st, true, b);
2019 		PFSYNC_BUCKET_UNLOCK(b);
2020 		return;
2021 	}
2022 
2023 	if (b->b_len == PFSYNC_MINPKT)
2024 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2025 
2026 	switch (st->sync_state) {
2027 	case PFSYNC_S_INS:
2028 		/* We never got to tell the world so just forget about it. */
2029 		pfsync_q_del(st, true, b);
2030 		break;
2031 
2032 	case PFSYNC_S_UPD_C:
2033 	case PFSYNC_S_UPD:
2034 	case PFSYNC_S_IACK:
2035 		pfsync_q_del(st, false, b);
2036 		ref = false;
2037 		/* FALLTHROUGH */
2038 
2039 	case PFSYNC_S_NONE:
2040 		pfsync_q_ins(st, PFSYNC_S_DEL, ref);
2041 		break;
2042 
2043 	default:
2044 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2045 	}
2046 
2047 	PFSYNC_BUCKET_UNLOCK(b);
2048 }
2049 
2050 static void
2051 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
2052 {
2053 	struct {
2054 		struct pfsync_subheader subh;
2055 		struct pfsync_clr clr;
2056 	} __packed r;
2057 
2058 	bzero(&r, sizeof(r));
2059 
2060 	r.subh.action = PFSYNC_ACT_CLR;
2061 	r.subh.count = htons(1);
2062 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
2063 
2064 	strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
2065 	r.clr.creatorid = creatorid;
2066 
2067 	pfsync_send_plus(&r, sizeof(r));
2068 }
2069 
2070 static void
2071 pfsync_q_ins(struct pf_state *st, int q, bool ref)
2072 {
2073 	struct pfsync_softc *sc = V_pfsyncif;
2074 	size_t nlen = pfsync_qs[q].len;
2075 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2076 
2077 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2078 
2079 	KASSERT(st->sync_state == PFSYNC_S_NONE,
2080 		("%s: st->sync_state %u", __func__, st->sync_state));
2081 	KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2082 	    b->b_len));
2083 
2084 	if (TAILQ_EMPTY(&b->b_qs[q]))
2085 		nlen += sizeof(struct pfsync_subheader);
2086 
2087 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2088 		pfsync_sendout(1, b->b_id);
2089 
2090 		nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2091 	}
2092 
2093 	b->b_len += nlen;
2094 	TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list);
2095 	st->sync_state = q;
2096 	if (ref)
2097 		pf_ref_state(st);
2098 }
2099 
2100 static void
2101 pfsync_q_del(struct pf_state *st, bool unref, struct pfsync_bucket *b)
2102 {
2103 	int q = st->sync_state;
2104 
2105 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2106 	KASSERT(st->sync_state != PFSYNC_S_NONE,
2107 		("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2108 
2109 	b->b_len -= pfsync_qs[q].len;
2110 	TAILQ_REMOVE(&b->b_qs[q], st, sync_list);
2111 	st->sync_state = PFSYNC_S_NONE;
2112 	if (unref)
2113 		pf_release_state(st);
2114 
2115 	if (TAILQ_EMPTY(&b->b_qs[q]))
2116 		b->b_len -= sizeof(struct pfsync_subheader);
2117 }
2118 
2119 static void
2120 pfsync_bulk_start(void)
2121 {
2122 	struct pfsync_softc *sc = V_pfsyncif;
2123 
2124 	if (V_pf_status.debug >= PF_DEBUG_MISC)
2125 		printf("pfsync: received bulk update request\n");
2126 
2127 	PFSYNC_BLOCK(sc);
2128 
2129 	sc->sc_ureq_received = time_uptime;
2130 	sc->sc_bulk_hashid = 0;
2131 	sc->sc_bulk_stateid = 0;
2132 	pfsync_bulk_status(PFSYNC_BUS_START);
2133 	callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2134 	PFSYNC_BUNLOCK(sc);
2135 }
2136 
2137 static void
2138 pfsync_bulk_update(void *arg)
2139 {
2140 	struct pfsync_softc *sc = arg;
2141 	struct pf_state *s;
2142 	int i, sent = 0;
2143 
2144 	PFSYNC_BLOCK_ASSERT(sc);
2145 	CURVNET_SET(sc->sc_ifp->if_vnet);
2146 
2147 	/*
2148 	 * Start with last state from previous invocation.
2149 	 * It may had gone, in this case start from the
2150 	 * hash slot.
2151 	 */
2152 	s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2153 
2154 	if (s != NULL)
2155 		i = PF_IDHASH(s);
2156 	else
2157 		i = sc->sc_bulk_hashid;
2158 
2159 	for (; i <= pf_hashmask; i++) {
2160 		struct pf_idhash *ih = &V_pf_idhash[i];
2161 
2162 		if (s != NULL)
2163 			PF_HASHROW_ASSERT(ih);
2164 		else {
2165 			PF_HASHROW_LOCK(ih);
2166 			s = LIST_FIRST(&ih->states);
2167 		}
2168 
2169 		for (; s; s = LIST_NEXT(s, entry)) {
2170 			if (s->sync_state == PFSYNC_S_NONE &&
2171 			    s->timeout < PFTM_MAX &&
2172 			    s->pfsync_time <= sc->sc_ureq_received) {
2173 				if (pfsync_update_state_req(s)) {
2174 					/* We've filled a packet. */
2175 					sc->sc_bulk_hashid = i;
2176 					sc->sc_bulk_stateid = s->id;
2177 					sc->sc_bulk_creatorid = s->creatorid;
2178 					PF_HASHROW_UNLOCK(ih);
2179 					callout_reset(&sc->sc_bulk_tmo, 1,
2180 					    pfsync_bulk_update, sc);
2181 					goto full;
2182 				}
2183 				sent++;
2184 			}
2185 		}
2186 		PF_HASHROW_UNLOCK(ih);
2187 	}
2188 
2189 	/* We're done. */
2190 	pfsync_bulk_status(PFSYNC_BUS_END);
2191 full:
2192 	CURVNET_RESTORE();
2193 }
2194 
2195 static void
2196 pfsync_bulk_status(u_int8_t status)
2197 {
2198 	struct {
2199 		struct pfsync_subheader subh;
2200 		struct pfsync_bus bus;
2201 	} __packed r;
2202 
2203 	struct pfsync_softc *sc = V_pfsyncif;
2204 
2205 	bzero(&r, sizeof(r));
2206 
2207 	r.subh.action = PFSYNC_ACT_BUS;
2208 	r.subh.count = htons(1);
2209 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2210 
2211 	r.bus.creatorid = V_pf_status.hostid;
2212 	r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2213 	r.bus.status = status;
2214 
2215 	pfsync_send_plus(&r, sizeof(r));
2216 }
2217 
2218 static void
2219 pfsync_bulk_fail(void *arg)
2220 {
2221 	struct pfsync_softc *sc = arg;
2222 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2223 
2224 	CURVNET_SET(sc->sc_ifp->if_vnet);
2225 
2226 	PFSYNC_BLOCK_ASSERT(sc);
2227 
2228 	if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2229 		/* Try again */
2230 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2231 		    pfsync_bulk_fail, V_pfsyncif);
2232 		PFSYNC_BUCKET_LOCK(b);
2233 		pfsync_request_update(0, 0);
2234 		PFSYNC_BUCKET_UNLOCK(b);
2235 	} else {
2236 		/* Pretend like the transfer was ok. */
2237 		sc->sc_ureq_sent = 0;
2238 		sc->sc_bulk_tries = 0;
2239 		PFSYNC_LOCK(sc);
2240 		if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2241 			(*carp_demote_adj_p)(-V_pfsync_carp_adj,
2242 			    "pfsync bulk fail");
2243 		sc->sc_flags |= PFSYNCF_OK;
2244 		PFSYNC_UNLOCK(sc);
2245 		if (V_pf_status.debug >= PF_DEBUG_MISC)
2246 			printf("pfsync: failed to receive bulk update\n");
2247 	}
2248 
2249 	CURVNET_RESTORE();
2250 }
2251 
2252 static void
2253 pfsync_send_plus(void *plus, size_t pluslen)
2254 {
2255 	struct pfsync_softc *sc = V_pfsyncif;
2256 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2257 
2258 	PFSYNC_BUCKET_LOCK(b);
2259 
2260 	if (b->b_len + pluslen > sc->sc_ifp->if_mtu)
2261 		pfsync_sendout(1, b->b_id);
2262 
2263 	b->b_plus = plus;
2264 	b->b_len += (b->b_pluslen = pluslen);
2265 
2266 	pfsync_sendout(1, b->b_id);
2267 	PFSYNC_BUCKET_UNLOCK(b);
2268 }
2269 
2270 static void
2271 pfsync_timeout(void *arg)
2272 {
2273 	struct pfsync_bucket *b = arg;
2274 
2275 	CURVNET_SET(b->b_sc->sc_ifp->if_vnet);
2276 	PFSYNC_BUCKET_LOCK(b);
2277 	pfsync_push(b);
2278 	PFSYNC_BUCKET_UNLOCK(b);
2279 	CURVNET_RESTORE();
2280 }
2281 
2282 static void
2283 pfsync_push(struct pfsync_bucket *b)
2284 {
2285 
2286 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2287 
2288 	b->b_flags |= PFSYNCF_BUCKET_PUSH;
2289 	swi_sched(V_pfsync_swi_cookie, 0);
2290 }
2291 
2292 static void
2293 pfsync_push_all(struct pfsync_softc *sc)
2294 {
2295 	int c;
2296 	struct pfsync_bucket *b;
2297 
2298 	for (c = 0; c < pfsync_buckets; c++) {
2299 		b = &sc->sc_buckets[c];
2300 
2301 		PFSYNC_BUCKET_LOCK(b);
2302 		pfsync_push(b);
2303 		PFSYNC_BUCKET_UNLOCK(b);
2304 	}
2305 }
2306 
2307 static void
2308 pfsyncintr(void *arg)
2309 {
2310 	struct pfsync_softc *sc = arg;
2311 	struct pfsync_bucket *b;
2312 	struct mbuf *m, *n;
2313 	int c;
2314 
2315 	CURVNET_SET(sc->sc_ifp->if_vnet);
2316 
2317 	for (c = 0; c < pfsync_buckets; c++) {
2318 		b = &sc->sc_buckets[c];
2319 
2320 		PFSYNC_BUCKET_LOCK(b);
2321 		if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) {
2322 			pfsync_sendout(0, b->b_id);
2323 			b->b_flags &= ~PFSYNCF_BUCKET_PUSH;
2324 		}
2325 		_IF_DEQUEUE_ALL(&b->b_snd, m);
2326 		PFSYNC_BUCKET_UNLOCK(b);
2327 
2328 		for (; m != NULL; m = n) {
2329 
2330 			n = m->m_nextpkt;
2331 			m->m_nextpkt = NULL;
2332 
2333 			/*
2334 			 * We distinguish between a deferral packet and our
2335 			 * own pfsync packet based on M_SKIP_FIREWALL
2336 			 * flag. This is XXX.
2337 			 */
2338 			if (m->m_flags & M_SKIP_FIREWALL)
2339 				ip_output(m, NULL, NULL, 0, NULL, NULL);
2340 			else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo,
2341 			    NULL) == 0)
2342 				V_pfsyncstats.pfsyncs_opackets++;
2343 			else
2344 				V_pfsyncstats.pfsyncs_oerrors++;
2345 		}
2346 	}
2347 	CURVNET_RESTORE();
2348 }
2349 
2350 static int
2351 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp,
2352     struct in_mfilter *imf)
2353 {
2354 	struct ip_moptions *imo = &sc->sc_imo;
2355 	int error;
2356 
2357 	if (!(ifp->if_flags & IFF_MULTICAST))
2358 		return (EADDRNOTAVAIL);
2359 
2360 	imo->imo_multicast_vif = -1;
2361 
2362 	if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL,
2363 	    &imf->imf_inm)) != 0)
2364 		return (error);
2365 
2366 	ip_mfilter_init(&imo->imo_head);
2367 	ip_mfilter_insert(&imo->imo_head, imf);
2368 	imo->imo_multicast_ifp = ifp;
2369 	imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2370 	imo->imo_multicast_loop = 0;
2371 
2372 	return (0);
2373 }
2374 
2375 static void
2376 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2377 {
2378 	struct ip_moptions *imo = &sc->sc_imo;
2379 	struct in_mfilter *imf;
2380 
2381 	while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
2382 		ip_mfilter_remove(&imo->imo_head, imf);
2383 		in_leavegroup(imf->imf_inm, NULL);
2384 		ip_mfilter_free(imf);
2385 	}
2386 	imo->imo_multicast_ifp = NULL;
2387 }
2388 
2389 void
2390 pfsync_detach_ifnet(struct ifnet *ifp)
2391 {
2392 	struct pfsync_softc *sc = V_pfsyncif;
2393 
2394 	if (sc == NULL)
2395 		return;
2396 
2397 	PFSYNC_LOCK(sc);
2398 
2399 	if (sc->sc_sync_if == ifp) {
2400 		/* We don't need mutlicast cleanup here, because the interface
2401 		 * is going away. We do need to ensure we don't try to do
2402 		 * cleanup later.
2403 		 */
2404 		ip_mfilter_init(&sc->sc_imo.imo_head);
2405 		sc->sc_imo.imo_multicast_ifp = NULL;
2406 		sc->sc_sync_if = NULL;
2407 	}
2408 
2409 	PFSYNC_UNLOCK(sc);
2410 }
2411 
2412 #ifdef INET
2413 extern  struct domain inetdomain;
2414 static struct protosw in_pfsync_protosw = {
2415 	.pr_type =		SOCK_RAW,
2416 	.pr_domain =		&inetdomain,
2417 	.pr_protocol =		IPPROTO_PFSYNC,
2418 	.pr_flags =		PR_ATOMIC|PR_ADDR,
2419 	.pr_input =		pfsync_input,
2420 	.pr_output =		rip_output,
2421 	.pr_ctloutput =		rip_ctloutput,
2422 	.pr_usrreqs =		&rip_usrreqs
2423 };
2424 #endif
2425 
2426 static void
2427 pfsync_pointers_init()
2428 {
2429 
2430 	PF_RULES_WLOCK();
2431 	V_pfsync_state_import_ptr = pfsync_state_import;
2432 	V_pfsync_insert_state_ptr = pfsync_insert_state;
2433 	V_pfsync_update_state_ptr = pfsync_update_state;
2434 	V_pfsync_delete_state_ptr = pfsync_delete_state;
2435 	V_pfsync_clear_states_ptr = pfsync_clear_states;
2436 	V_pfsync_defer_ptr = pfsync_defer;
2437 	PF_RULES_WUNLOCK();
2438 }
2439 
2440 static void
2441 pfsync_pointers_uninit()
2442 {
2443 
2444 	PF_RULES_WLOCK();
2445 	V_pfsync_state_import_ptr = NULL;
2446 	V_pfsync_insert_state_ptr = NULL;
2447 	V_pfsync_update_state_ptr = NULL;
2448 	V_pfsync_delete_state_ptr = NULL;
2449 	V_pfsync_clear_states_ptr = NULL;
2450 	V_pfsync_defer_ptr = NULL;
2451 	PF_RULES_WUNLOCK();
2452 }
2453 
2454 static void
2455 vnet_pfsync_init(const void *unused __unused)
2456 {
2457 	int error;
2458 
2459 	V_pfsync_cloner = if_clone_simple(pfsyncname,
2460 	    pfsync_clone_create, pfsync_clone_destroy, 1);
2461 	error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif,
2462 	    SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2463 	if (error) {
2464 		if_clone_detach(V_pfsync_cloner);
2465 		log(LOG_INFO, "swi_add() failed in %s\n", __func__);
2466 	}
2467 
2468 	pfsync_pointers_init();
2469 }
2470 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
2471     vnet_pfsync_init, NULL);
2472 
2473 static void
2474 vnet_pfsync_uninit(const void *unused __unused)
2475 {
2476 
2477 	pfsync_pointers_uninit();
2478 
2479 	if_clone_detach(V_pfsync_cloner);
2480 	swi_remove(V_pfsync_swi_cookie);
2481 }
2482 
2483 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH,
2484     vnet_pfsync_uninit, NULL);
2485 
2486 static int
2487 pfsync_init()
2488 {
2489 #ifdef INET
2490 	int error;
2491 
2492 	pfsync_detach_ifnet_ptr = pfsync_detach_ifnet;
2493 
2494 	error = pf_proto_register(PF_INET, &in_pfsync_protosw);
2495 	if (error)
2496 		return (error);
2497 	error = ipproto_register(IPPROTO_PFSYNC);
2498 	if (error) {
2499 		pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2500 		return (error);
2501 	}
2502 #endif
2503 
2504 	return (0);
2505 }
2506 
2507 static void
2508 pfsync_uninit()
2509 {
2510 	pfsync_detach_ifnet_ptr = NULL;
2511 
2512 #ifdef INET
2513 	ipproto_unregister(IPPROTO_PFSYNC);
2514 	pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2515 #endif
2516 }
2517 
2518 static int
2519 pfsync_modevent(module_t mod, int type, void *data)
2520 {
2521 	int error = 0;
2522 
2523 	switch (type) {
2524 	case MOD_LOAD:
2525 		error = pfsync_init();
2526 		break;
2527 	case MOD_UNLOAD:
2528 		pfsync_uninit();
2529 		break;
2530 	default:
2531 		error = EINVAL;
2532 		break;
2533 	}
2534 
2535 	return (error);
2536 }
2537 
2538 static moduledata_t pfsync_mod = {
2539 	pfsyncname,
2540 	pfsync_modevent,
2541 	0
2542 };
2543 
2544 #define PFSYNC_MODVER 1
2545 
2546 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
2547 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
2548 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2549 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
2550