xref: /freebsd/sys/netpfil/pf/if_pfsync.c (revision 1f4bcc459a76b7aa664f3fd557684cd0ba6da352)
1 /*-
2  * Copyright (c) 2002 Michael Shalayeff
3  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
19  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
20  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
21  * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
23  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
24  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
25  * THE POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 /*-
29  * Copyright (c) 2009 David Gwynne <dlg@openbsd.org>
30  *
31  * Permission to use, copy, modify, and distribute this software for any
32  * purpose with or without fee is hereby granted, provided that the above
33  * copyright notice and this permission notice appear in all copies.
34  *
35  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
36  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
37  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
38  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
39  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
40  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
41  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
42  */
43 
44 /*
45  * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $
46  *
47  * Revisions picked from OpenBSD after revision 1.110 import:
48  * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input()
49  * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates
50  * 1.120, 1.175 - use monotonic time_uptime
51  * 1.122 - reduce number of updates for non-TCP sessions
52  * 1.125, 1.127 - rewrite merge or stale processing
53  * 1.128 - cleanups
54  * 1.146 - bzero() mbuf before sparsely filling it with data
55  * 1.170 - SIOCSIFMTU checks
56  * 1.126, 1.142 - deferred packets processing
57  * 1.173 - correct expire time processing
58  */
59 
60 #include <sys/cdefs.h>
61 __FBSDID("$FreeBSD$");
62 
63 #include "opt_inet.h"
64 #include "opt_inet6.h"
65 #include "opt_pf.h"
66 
67 #include <sys/param.h>
68 #include <sys/bus.h>
69 #include <sys/endian.h>
70 #include <sys/interrupt.h>
71 #include <sys/kernel.h>
72 #include <sys/lock.h>
73 #include <sys/mbuf.h>
74 #include <sys/module.h>
75 #include <sys/mutex.h>
76 #include <sys/priv.h>
77 #include <sys/protosw.h>
78 #include <sys/socket.h>
79 #include <sys/sockio.h>
80 #include <sys/sysctl.h>
81 
82 #include <net/bpf.h>
83 #include <net/if.h>
84 #include <net/if_var.h>
85 #include <net/if_clone.h>
86 #include <net/if_types.h>
87 #include <net/vnet.h>
88 #include <net/pfvar.h>
89 #include <net/if_pfsync.h>
90 
91 #include <netinet/if_ether.h>
92 #include <netinet/in.h>
93 #include <netinet/in_var.h>
94 #include <netinet/ip.h>
95 #include <netinet/ip_carp.h>
96 #include <netinet/ip_var.h>
97 #include <netinet/tcp.h>
98 #include <netinet/tcp_fsm.h>
99 #include <netinet/tcp_seq.h>
100 
101 #define PFSYNC_MINPKT ( \
102 	sizeof(struct ip) + \
103 	sizeof(struct pfsync_header) + \
104 	sizeof(struct pfsync_subheader) )
105 
106 struct pfsync_pkt {
107 	struct ip *ip;
108 	struct in_addr src;
109 	u_int8_t flags;
110 };
111 
112 static int	pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *,
113 		    struct pfsync_state_peer *);
114 static int	pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int);
115 static int	pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int);
116 static int	pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int);
117 static int	pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int);
118 static int	pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int);
119 static int	pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int);
120 static int	pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int);
121 static int	pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int);
122 static int	pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int);
123 static int	pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int);
124 static int	pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int);
125 static int	pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int);
126 
127 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = {
128 	pfsync_in_clr,			/* PFSYNC_ACT_CLR */
129 	pfsync_in_ins,			/* PFSYNC_ACT_INS */
130 	pfsync_in_iack,			/* PFSYNC_ACT_INS_ACK */
131 	pfsync_in_upd,			/* PFSYNC_ACT_UPD */
132 	pfsync_in_upd_c,		/* PFSYNC_ACT_UPD_C */
133 	pfsync_in_ureq,			/* PFSYNC_ACT_UPD_REQ */
134 	pfsync_in_del,			/* PFSYNC_ACT_DEL */
135 	pfsync_in_del_c,		/* PFSYNC_ACT_DEL_C */
136 	pfsync_in_error,		/* PFSYNC_ACT_INS_F */
137 	pfsync_in_error,		/* PFSYNC_ACT_DEL_F */
138 	pfsync_in_bus,			/* PFSYNC_ACT_BUS */
139 	pfsync_in_tdb,			/* PFSYNC_ACT_TDB */
140 	pfsync_in_eof			/* PFSYNC_ACT_EOF */
141 };
142 
143 struct pfsync_q {
144 	void		(*write)(struct pf_state *, void *);
145 	size_t		len;
146 	u_int8_t	action;
147 };
148 
149 /* we have one of these for every PFSYNC_S_ */
150 static void	pfsync_out_state(struct pf_state *, void *);
151 static void	pfsync_out_iack(struct pf_state *, void *);
152 static void	pfsync_out_upd_c(struct pf_state *, void *);
153 static void	pfsync_out_del(struct pf_state *, void *);
154 
155 static struct pfsync_q pfsync_qs[] = {
156 	{ pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_INS },
157 	{ pfsync_out_iack,  sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
158 	{ pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_UPD },
159 	{ pfsync_out_upd_c, sizeof(struct pfsync_upd_c),   PFSYNC_ACT_UPD_C },
160 	{ pfsync_out_del,   sizeof(struct pfsync_del_c),   PFSYNC_ACT_DEL_C }
161 };
162 
163 static void	pfsync_q_ins(struct pf_state *, int);
164 static void	pfsync_q_del(struct pf_state *);
165 
166 static void	pfsync_update_state(struct pf_state *);
167 
168 struct pfsync_upd_req_item {
169 	TAILQ_ENTRY(pfsync_upd_req_item)	ur_entry;
170 	struct pfsync_upd_req			ur_msg;
171 };
172 
173 struct pfsync_deferral {
174 	struct pfsync_softc		*pd_sc;
175 	TAILQ_ENTRY(pfsync_deferral)	pd_entry;
176 	u_int				pd_refs;
177 	struct callout			pd_tmo;
178 
179 	struct pf_state			*pd_st;
180 	struct mbuf			*pd_m;
181 };
182 
183 struct pfsync_softc {
184 	/* Configuration */
185 	struct ifnet		*sc_ifp;
186 	struct ifnet		*sc_sync_if;
187 	struct ip_moptions	sc_imo;
188 	struct in_addr		sc_sync_peer;
189 	uint32_t		sc_flags;
190 #define	PFSYNCF_OK		0x00000001
191 #define	PFSYNCF_DEFER		0x00000002
192 #define	PFSYNCF_PUSH		0x00000004
193 	uint8_t			sc_maxupdates;
194 	struct ip		sc_template;
195 	struct callout		sc_tmo;
196 	struct mtx		sc_mtx;
197 
198 	/* Queued data */
199 	size_t			sc_len;
200 	TAILQ_HEAD(, pf_state)			sc_qs[PFSYNC_S_COUNT];
201 	TAILQ_HEAD(, pfsync_upd_req_item)	sc_upd_req_list;
202 	TAILQ_HEAD(, pfsync_deferral)		sc_deferrals;
203 	u_int			sc_deferred;
204 	void			*sc_plus;
205 	size_t			sc_pluslen;
206 
207 	/* Bulk update info */
208 	struct mtx		sc_bulk_mtx;
209 	uint32_t		sc_ureq_sent;
210 	int			sc_bulk_tries;
211 	uint32_t		sc_ureq_received;
212 	int			sc_bulk_hashid;
213 	uint64_t		sc_bulk_stateid;
214 	uint32_t		sc_bulk_creatorid;
215 	struct callout		sc_bulk_tmo;
216 	struct callout		sc_bulkfail_tmo;
217 };
218 
219 #define	PFSYNC_LOCK(sc)		mtx_lock(&(sc)->sc_mtx)
220 #define	PFSYNC_UNLOCK(sc)	mtx_unlock(&(sc)->sc_mtx)
221 #define	PFSYNC_LOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_mtx, MA_OWNED)
222 
223 #define	PFSYNC_BLOCK(sc)	mtx_lock(&(sc)->sc_bulk_mtx)
224 #define	PFSYNC_BUNLOCK(sc)	mtx_unlock(&(sc)->sc_bulk_mtx)
225 #define	PFSYNC_BLOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
226 
227 static const char pfsyncname[] = "pfsync";
228 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
229 static VNET_DEFINE(struct pfsync_softc	*, pfsyncif) = NULL;
230 #define	V_pfsyncif		VNET(pfsyncif)
231 static VNET_DEFINE(void *, pfsync_swi_cookie) = NULL;
232 #define	V_pfsync_swi_cookie	VNET(pfsync_swi_cookie)
233 static VNET_DEFINE(struct pfsyncstats, pfsyncstats);
234 #define	V_pfsyncstats		VNET(pfsyncstats)
235 static VNET_DEFINE(int, pfsync_carp_adj) = CARP_MAXSKEW;
236 #define	V_pfsync_carp_adj	VNET(pfsync_carp_adj)
237 
238 static void	pfsync_timeout(void *);
239 static void	pfsync_push(struct pfsync_softc *);
240 static void	pfsyncintr(void *);
241 static int	pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
242 		    void *);
243 static void	pfsync_multicast_cleanup(struct pfsync_softc *);
244 static void	pfsync_pointers_init(void);
245 static void	pfsync_pointers_uninit(void);
246 static int	pfsync_init(void);
247 static void	pfsync_uninit(void);
248 
249 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC");
250 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW,
251     &VNET_NAME(pfsyncstats), pfsyncstats,
252     "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
253 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW,
254     &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
255 
256 static int	pfsync_clone_create(struct if_clone *, int, caddr_t);
257 static void	pfsync_clone_destroy(struct ifnet *);
258 static int	pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
259 		    struct pf_state_peer *);
260 static int	pfsyncoutput(struct ifnet *, struct mbuf *,
261 		    const struct sockaddr *, struct route *);
262 static int	pfsyncioctl(struct ifnet *, u_long, caddr_t);
263 
264 static int	pfsync_defer(struct pf_state *, struct mbuf *);
265 static void	pfsync_undefer(struct pfsync_deferral *, int);
266 static void	pfsync_undefer_state(struct pf_state *, int);
267 static void	pfsync_defer_tmo(void *);
268 
269 static void	pfsync_request_update(u_int32_t, u_int64_t);
270 static void	pfsync_update_state_req(struct pf_state *);
271 
272 static void	pfsync_drop(struct pfsync_softc *);
273 static void	pfsync_sendout(int);
274 static void	pfsync_send_plus(void *, size_t);
275 
276 static void	pfsync_bulk_start(void);
277 static void	pfsync_bulk_status(u_int8_t);
278 static void	pfsync_bulk_update(void *);
279 static void	pfsync_bulk_fail(void *);
280 
281 #ifdef IPSEC
282 static void	pfsync_update_net_tdb(struct pfsync_tdb *);
283 #endif
284 
285 #define PFSYNC_MAX_BULKTRIES	12
286 
287 VNET_DEFINE(struct if_clone *, pfsync_cloner);
288 #define	V_pfsync_cloner	VNET(pfsync_cloner)
289 
290 static int
291 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
292 {
293 	struct pfsync_softc *sc;
294 	struct ifnet *ifp;
295 	int q;
296 
297 	if (unit != 0)
298 		return (EINVAL);
299 
300 	sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
301 	sc->sc_flags |= PFSYNCF_OK;
302 
303 	for (q = 0; q < PFSYNC_S_COUNT; q++)
304 		TAILQ_INIT(&sc->sc_qs[q]);
305 
306 	TAILQ_INIT(&sc->sc_upd_req_list);
307 	TAILQ_INIT(&sc->sc_deferrals);
308 
309 	sc->sc_len = PFSYNC_MINPKT;
310 	sc->sc_maxupdates = 128;
311 
312 	ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
313 	if (ifp == NULL) {
314 		free(sc, M_PFSYNC);
315 		return (ENOSPC);
316 	}
317 	if_initname(ifp, pfsyncname, unit);
318 	ifp->if_softc = sc;
319 	ifp->if_ioctl = pfsyncioctl;
320 	ifp->if_output = pfsyncoutput;
321 	ifp->if_type = IFT_PFSYNC;
322 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
323 	ifp->if_hdrlen = sizeof(struct pfsync_header);
324 	ifp->if_mtu = ETHERMTU;
325 	mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
326 	mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
327 	callout_init(&sc->sc_tmo, 1);
328 	callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
329 	callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
330 
331 	if_attach(ifp);
332 
333 	bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
334 
335 	V_pfsyncif = sc;
336 
337 	return (0);
338 }
339 
340 static void
341 pfsync_clone_destroy(struct ifnet *ifp)
342 {
343 	struct pfsync_softc *sc = ifp->if_softc;
344 
345 	/*
346 	 * At this stage, everything should have already been
347 	 * cleared by pfsync_uninit(), and we have only to
348 	 * drain callouts.
349 	 */
350 	while (sc->sc_deferred > 0) {
351 		struct pfsync_deferral *pd = TAILQ_FIRST(&sc->sc_deferrals);
352 
353 		TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
354 		sc->sc_deferred--;
355 		if (callout_stop(&pd->pd_tmo) > 0) {
356 			pf_release_state(pd->pd_st);
357 			m_freem(pd->pd_m);
358 			free(pd, M_PFSYNC);
359 		} else {
360 			pd->pd_refs++;
361 			callout_drain(&pd->pd_tmo);
362 			free(pd, M_PFSYNC);
363 		}
364 	}
365 
366 	callout_drain(&sc->sc_tmo);
367 	callout_drain(&sc->sc_bulkfail_tmo);
368 	callout_drain(&sc->sc_bulk_tmo);
369 
370 	if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
371 		(*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
372 	bpfdetach(ifp);
373 	if_detach(ifp);
374 
375 	pfsync_drop(sc);
376 
377 	if_free(ifp);
378 	if (sc->sc_imo.imo_membership)
379 		pfsync_multicast_cleanup(sc);
380 	mtx_destroy(&sc->sc_mtx);
381 	mtx_destroy(&sc->sc_bulk_mtx);
382 	free(sc, M_PFSYNC);
383 
384 	V_pfsyncif = NULL;
385 }
386 
387 static int
388 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
389     struct pf_state_peer *d)
390 {
391 	if (s->scrub.scrub_flag && d->scrub == NULL) {
392 		d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
393 		if (d->scrub == NULL)
394 			return (ENOMEM);
395 	}
396 
397 	return (0);
398 }
399 
400 
401 static int
402 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
403 {
404 	struct pfsync_softc *sc = V_pfsyncif;
405 #ifndef	__NO_STRICT_ALIGNMENT
406 	struct pfsync_state_key key[2];
407 #endif
408 	struct pfsync_state_key *kw, *ks;
409 	struct pf_state	*st = NULL;
410 	struct pf_state_key *skw = NULL, *sks = NULL;
411 	struct pf_rule *r = NULL;
412 	struct pfi_kif	*kif;
413 	int error;
414 
415 	PF_RULES_RASSERT();
416 
417 	if (sp->creatorid == 0) {
418 		if (V_pf_status.debug >= PF_DEBUG_MISC)
419 			printf("%s: invalid creator id: %08x\n", __func__,
420 			    ntohl(sp->creatorid));
421 		return (EINVAL);
422 	}
423 
424 	if ((kif = pfi_kif_find(sp->ifname)) == NULL) {
425 		if (V_pf_status.debug >= PF_DEBUG_MISC)
426 			printf("%s: unknown interface: %s\n", __func__,
427 			    sp->ifname);
428 		if (flags & PFSYNC_SI_IOCTL)
429 			return (EINVAL);
430 		return (0);	/* skip this state */
431 	}
432 
433 	/*
434 	 * If the ruleset checksums match or the state is coming from the ioctl,
435 	 * it's safe to associate the state with the rule of that number.
436 	 */
437 	if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
438 	    (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
439 	    pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
440 		r = pf_main_ruleset.rules[
441 		    PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
442 	else
443 		r = &V_pf_default_rule;
444 
445 	if ((r->max_states &&
446 	    counter_u64_fetch(r->states_cur) >= r->max_states))
447 		goto cleanup;
448 
449 	/*
450 	 * XXXGL: consider M_WAITOK in ioctl path after.
451 	 */
452 	if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL)
453 		goto cleanup;
454 
455 	if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
456 		goto cleanup;
457 
458 #ifndef	__NO_STRICT_ALIGNMENT
459 	bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2);
460 	kw = &key[PF_SK_WIRE];
461 	ks = &key[PF_SK_STACK];
462 #else
463 	kw = &sp->key[PF_SK_WIRE];
464 	ks = &sp->key[PF_SK_STACK];
465 #endif
466 
467 	if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) ||
468 	    PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) ||
469 	    kw->port[0] != ks->port[0] ||
470 	    kw->port[1] != ks->port[1]) {
471 		sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
472 		if (sks == NULL)
473 			goto cleanup;
474 	} else
475 		sks = skw;
476 
477 	/* allocate memory for scrub info */
478 	if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
479 	    pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
480 		goto cleanup;
481 
482 	/* Copy to state key(s). */
483 	skw->addr[0] = kw->addr[0];
484 	skw->addr[1] = kw->addr[1];
485 	skw->port[0] = kw->port[0];
486 	skw->port[1] = kw->port[1];
487 	skw->proto = sp->proto;
488 	skw->af = sp->af;
489 	if (sks != skw) {
490 		sks->addr[0] = ks->addr[0];
491 		sks->addr[1] = ks->addr[1];
492 		sks->port[0] = ks->port[0];
493 		sks->port[1] = ks->port[1];
494 		sks->proto = sp->proto;
495 		sks->af = sp->af;
496 	}
497 
498 	/* copy to state */
499 	bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
500 	st->creation = time_uptime - ntohl(sp->creation);
501 	st->expire = time_uptime;
502 	if (sp->expire) {
503 		uint32_t timeout;
504 
505 		timeout = r->timeout[sp->timeout];
506 		if (!timeout)
507 			timeout = V_pf_default_rule.timeout[sp->timeout];
508 
509 		/* sp->expire may have been adaptively scaled by export. */
510 		st->expire -= timeout - ntohl(sp->expire);
511 	}
512 
513 	st->direction = sp->direction;
514 	st->log = sp->log;
515 	st->timeout = sp->timeout;
516 	st->state_flags = sp->state_flags;
517 
518 	st->id = sp->id;
519 	st->creatorid = sp->creatorid;
520 	pf_state_peer_ntoh(&sp->src, &st->src);
521 	pf_state_peer_ntoh(&sp->dst, &st->dst);
522 
523 	st->rule.ptr = r;
524 	st->nat_rule.ptr = NULL;
525 	st->anchor.ptr = NULL;
526 	st->rt_kif = NULL;
527 
528 	st->pfsync_time = time_uptime;
529 	st->sync_state = PFSYNC_S_NONE;
530 
531 	if (!(flags & PFSYNC_SI_IOCTL))
532 		st->state_flags |= PFSTATE_NOSYNC;
533 
534 	if ((error = pf_state_insert(kif, skw, sks, st)) != 0)
535 		goto cleanup_state;
536 
537 	/* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
538 	counter_u64_add(r->states_cur, 1);
539 	counter_u64_add(r->states_tot, 1);
540 
541 	if (!(flags & PFSYNC_SI_IOCTL)) {
542 		st->state_flags &= ~PFSTATE_NOSYNC;
543 		if (st->state_flags & PFSTATE_ACK) {
544 			pfsync_q_ins(st, PFSYNC_S_IACK);
545 			pfsync_push(sc);
546 		}
547 	}
548 	st->state_flags &= ~PFSTATE_ACK;
549 	PF_STATE_UNLOCK(st);
550 
551 	return (0);
552 
553 cleanup:
554 	error = ENOMEM;
555 	if (skw == sks)
556 		sks = NULL;
557 	if (skw != NULL)
558 		uma_zfree(V_pf_state_key_z, skw);
559 	if (sks != NULL)
560 		uma_zfree(V_pf_state_key_z, sks);
561 
562 cleanup_state:	/* pf_state_insert() frees the state keys. */
563 	if (st) {
564 		if (st->dst.scrub)
565 			uma_zfree(V_pf_state_scrub_z, st->dst.scrub);
566 		if (st->src.scrub)
567 			uma_zfree(V_pf_state_scrub_z, st->src.scrub);
568 		uma_zfree(V_pf_state_z, st);
569 	}
570 	return (error);
571 }
572 
573 static int
574 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
575 {
576 	struct pfsync_softc *sc = V_pfsyncif;
577 	struct pfsync_pkt pkt;
578 	struct mbuf *m = *mp;
579 	struct ip *ip = mtod(m, struct ip *);
580 	struct pfsync_header *ph;
581 	struct pfsync_subheader subh;
582 
583 	int offset, len;
584 	int rv;
585 	uint16_t count;
586 
587 	*mp = NULL;
588 	V_pfsyncstats.pfsyncs_ipackets++;
589 
590 	/* Verify that we have a sync interface configured. */
591 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
592 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
593 		goto done;
594 
595 	/* verify that the packet came in on the right interface */
596 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
597 		V_pfsyncstats.pfsyncs_badif++;
598 		goto done;
599 	}
600 
601 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
602 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
603 	/* verify that the IP TTL is 255. */
604 	if (ip->ip_ttl != PFSYNC_DFLTTL) {
605 		V_pfsyncstats.pfsyncs_badttl++;
606 		goto done;
607 	}
608 
609 	offset = ip->ip_hl << 2;
610 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
611 		V_pfsyncstats.pfsyncs_hdrops++;
612 		goto done;
613 	}
614 
615 	if (offset + sizeof(*ph) > m->m_len) {
616 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
617 			V_pfsyncstats.pfsyncs_hdrops++;
618 			return (IPPROTO_DONE);
619 		}
620 		ip = mtod(m, struct ip *);
621 	}
622 	ph = (struct pfsync_header *)((char *)ip + offset);
623 
624 	/* verify the version */
625 	if (ph->version != PFSYNC_VERSION) {
626 		V_pfsyncstats.pfsyncs_badver++;
627 		goto done;
628 	}
629 
630 	len = ntohs(ph->len) + offset;
631 	if (m->m_pkthdr.len < len) {
632 		V_pfsyncstats.pfsyncs_badlen++;
633 		goto done;
634 	}
635 
636 	/* Cheaper to grab this now than having to mess with mbufs later */
637 	pkt.ip = ip;
638 	pkt.src = ip->ip_src;
639 	pkt.flags = 0;
640 
641 	/*
642 	 * Trusting pf_chksum during packet processing, as well as seeking
643 	 * in interface name tree, require holding PF_RULES_RLOCK().
644 	 */
645 	PF_RULES_RLOCK();
646 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
647 		pkt.flags |= PFSYNC_SI_CKSUM;
648 
649 	offset += sizeof(*ph);
650 	while (offset <= len - sizeof(subh)) {
651 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
652 		offset += sizeof(subh);
653 
654 		if (subh.action >= PFSYNC_ACT_MAX) {
655 			V_pfsyncstats.pfsyncs_badact++;
656 			PF_RULES_RUNLOCK();
657 			goto done;
658 		}
659 
660 		count = ntohs(subh.count);
661 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
662 		rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count);
663 		if (rv == -1) {
664 			PF_RULES_RUNLOCK();
665 			return (IPPROTO_DONE);
666 		}
667 
668 		offset += rv;
669 	}
670 	PF_RULES_RUNLOCK();
671 
672 done:
673 	m_freem(m);
674 	return (IPPROTO_DONE);
675 }
676 
677 static int
678 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
679 {
680 	struct pfsync_clr *clr;
681 	struct mbuf *mp;
682 	int len = sizeof(*clr) * count;
683 	int i, offp;
684 	u_int32_t creatorid;
685 
686 	mp = m_pulldown(m, offset, len, &offp);
687 	if (mp == NULL) {
688 		V_pfsyncstats.pfsyncs_badlen++;
689 		return (-1);
690 	}
691 	clr = (struct pfsync_clr *)(mp->m_data + offp);
692 
693 	for (i = 0; i < count; i++) {
694 		creatorid = clr[i].creatorid;
695 
696 		if (clr[i].ifname[0] != '\0' &&
697 		    pfi_kif_find(clr[i].ifname) == NULL)
698 			continue;
699 
700 		for (int i = 0; i <= pf_hashmask; i++) {
701 			struct pf_idhash *ih = &V_pf_idhash[i];
702 			struct pf_state *s;
703 relock:
704 			PF_HASHROW_LOCK(ih);
705 			LIST_FOREACH(s, &ih->states, entry) {
706 				if (s->creatorid == creatorid) {
707 					s->state_flags |= PFSTATE_NOSYNC;
708 					pf_unlink_state(s, PF_ENTER_LOCKED);
709 					goto relock;
710 				}
711 			}
712 			PF_HASHROW_UNLOCK(ih);
713 		}
714 	}
715 
716 	return (len);
717 }
718 
719 static int
720 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
721 {
722 	struct mbuf *mp;
723 	struct pfsync_state *sa, *sp;
724 	int len = sizeof(*sp) * count;
725 	int i, offp;
726 
727 	mp = m_pulldown(m, offset, len, &offp);
728 	if (mp == NULL) {
729 		V_pfsyncstats.pfsyncs_badlen++;
730 		return (-1);
731 	}
732 	sa = (struct pfsync_state *)(mp->m_data + offp);
733 
734 	for (i = 0; i < count; i++) {
735 		sp = &sa[i];
736 
737 		/* Check for invalid values. */
738 		if (sp->timeout >= PFTM_MAX ||
739 		    sp->src.state > PF_TCPS_PROXY_DST ||
740 		    sp->dst.state > PF_TCPS_PROXY_DST ||
741 		    sp->direction > PF_OUT ||
742 		    (sp->af != AF_INET && sp->af != AF_INET6)) {
743 			if (V_pf_status.debug >= PF_DEBUG_MISC)
744 				printf("%s: invalid value\n", __func__);
745 			V_pfsyncstats.pfsyncs_badval++;
746 			continue;
747 		}
748 
749 		if (pfsync_state_import(sp, pkt->flags) == ENOMEM)
750 			/* Drop out, but process the rest of the actions. */
751 			break;
752 	}
753 
754 	return (len);
755 }
756 
757 static int
758 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
759 {
760 	struct pfsync_ins_ack *ia, *iaa;
761 	struct pf_state *st;
762 
763 	struct mbuf *mp;
764 	int len = count * sizeof(*ia);
765 	int offp, i;
766 
767 	mp = m_pulldown(m, offset, len, &offp);
768 	if (mp == NULL) {
769 		V_pfsyncstats.pfsyncs_badlen++;
770 		return (-1);
771 	}
772 	iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
773 
774 	for (i = 0; i < count; i++) {
775 		ia = &iaa[i];
776 
777 		st = pf_find_state_byid(ia->id, ia->creatorid);
778 		if (st == NULL)
779 			continue;
780 
781 		if (st->state_flags & PFSTATE_ACK) {
782 			PFSYNC_LOCK(V_pfsyncif);
783 			pfsync_undefer_state(st, 0);
784 			PFSYNC_UNLOCK(V_pfsyncif);
785 		}
786 		PF_STATE_UNLOCK(st);
787 	}
788 	/*
789 	 * XXX this is not yet implemented, but we know the size of the
790 	 * message so we can skip it.
791 	 */
792 
793 	return (count * sizeof(struct pfsync_ins_ack));
794 }
795 
796 static int
797 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src,
798     struct pfsync_state_peer *dst)
799 {
800 	int sync = 0;
801 
802 	PF_STATE_LOCK_ASSERT(st);
803 
804 	/*
805 	 * The state should never go backwards except
806 	 * for syn-proxy states.  Neither should the
807 	 * sequence window slide backwards.
808 	 */
809 	if ((st->src.state > src->state &&
810 	    (st->src.state < PF_TCPS_PROXY_SRC ||
811 	    src->state >= PF_TCPS_PROXY_SRC)) ||
812 
813 	    (st->src.state == src->state &&
814 	    SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
815 		sync++;
816 	else
817 		pf_state_peer_ntoh(src, &st->src);
818 
819 	if ((st->dst.state > dst->state) ||
820 
821 	    (st->dst.state >= TCPS_SYN_SENT &&
822 	    SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
823 		sync++;
824 	else
825 		pf_state_peer_ntoh(dst, &st->dst);
826 
827 	return (sync);
828 }
829 
830 static int
831 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
832 {
833 	struct pfsync_softc *sc = V_pfsyncif;
834 	struct pfsync_state *sa, *sp;
835 	struct pf_state *st;
836 	int sync;
837 
838 	struct mbuf *mp;
839 	int len = count * sizeof(*sp);
840 	int offp, i;
841 
842 	mp = m_pulldown(m, offset, len, &offp);
843 	if (mp == NULL) {
844 		V_pfsyncstats.pfsyncs_badlen++;
845 		return (-1);
846 	}
847 	sa = (struct pfsync_state *)(mp->m_data + offp);
848 
849 	for (i = 0; i < count; i++) {
850 		sp = &sa[i];
851 
852 		/* check for invalid values */
853 		if (sp->timeout >= PFTM_MAX ||
854 		    sp->src.state > PF_TCPS_PROXY_DST ||
855 		    sp->dst.state > PF_TCPS_PROXY_DST) {
856 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
857 				printf("pfsync_input: PFSYNC_ACT_UPD: "
858 				    "invalid value\n");
859 			}
860 			V_pfsyncstats.pfsyncs_badval++;
861 			continue;
862 		}
863 
864 		st = pf_find_state_byid(sp->id, sp->creatorid);
865 		if (st == NULL) {
866 			/* insert the update */
867 			if (pfsync_state_import(sp, 0))
868 				V_pfsyncstats.pfsyncs_badstate++;
869 			continue;
870 		}
871 
872 		if (st->state_flags & PFSTATE_ACK) {
873 			PFSYNC_LOCK(sc);
874 			pfsync_undefer_state(st, 1);
875 			PFSYNC_UNLOCK(sc);
876 		}
877 
878 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
879 			sync = pfsync_upd_tcp(st, &sp->src, &sp->dst);
880 		else {
881 			sync = 0;
882 
883 			/*
884 			 * Non-TCP protocol state machine always go
885 			 * forwards
886 			 */
887 			if (st->src.state > sp->src.state)
888 				sync++;
889 			else
890 				pf_state_peer_ntoh(&sp->src, &st->src);
891 			if (st->dst.state > sp->dst.state)
892 				sync++;
893 			else
894 				pf_state_peer_ntoh(&sp->dst, &st->dst);
895 		}
896 		if (sync < 2) {
897 			pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
898 			pf_state_peer_ntoh(&sp->dst, &st->dst);
899 			st->expire = time_uptime;
900 			st->timeout = sp->timeout;
901 		}
902 		st->pfsync_time = time_uptime;
903 
904 		if (sync) {
905 			V_pfsyncstats.pfsyncs_stale++;
906 
907 			pfsync_update_state(st);
908 			PF_STATE_UNLOCK(st);
909 			PFSYNC_LOCK(sc);
910 			pfsync_push(sc);
911 			PFSYNC_UNLOCK(sc);
912 			continue;
913 		}
914 		PF_STATE_UNLOCK(st);
915 	}
916 
917 	return (len);
918 }
919 
920 static int
921 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
922 {
923 	struct pfsync_softc *sc = V_pfsyncif;
924 	struct pfsync_upd_c *ua, *up;
925 	struct pf_state *st;
926 	int len = count * sizeof(*up);
927 	int sync;
928 	struct mbuf *mp;
929 	int offp, i;
930 
931 	mp = m_pulldown(m, offset, len, &offp);
932 	if (mp == NULL) {
933 		V_pfsyncstats.pfsyncs_badlen++;
934 		return (-1);
935 	}
936 	ua = (struct pfsync_upd_c *)(mp->m_data + offp);
937 
938 	for (i = 0; i < count; i++) {
939 		up = &ua[i];
940 
941 		/* check for invalid values */
942 		if (up->timeout >= PFTM_MAX ||
943 		    up->src.state > PF_TCPS_PROXY_DST ||
944 		    up->dst.state > PF_TCPS_PROXY_DST) {
945 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
946 				printf("pfsync_input: "
947 				    "PFSYNC_ACT_UPD_C: "
948 				    "invalid value\n");
949 			}
950 			V_pfsyncstats.pfsyncs_badval++;
951 			continue;
952 		}
953 
954 		st = pf_find_state_byid(up->id, up->creatorid);
955 		if (st == NULL) {
956 			/* We don't have this state. Ask for it. */
957 			PFSYNC_LOCK(sc);
958 			pfsync_request_update(up->creatorid, up->id);
959 			PFSYNC_UNLOCK(sc);
960 			continue;
961 		}
962 
963 		if (st->state_flags & PFSTATE_ACK) {
964 			PFSYNC_LOCK(sc);
965 			pfsync_undefer_state(st, 1);
966 			PFSYNC_UNLOCK(sc);
967 		}
968 
969 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
970 			sync = pfsync_upd_tcp(st, &up->src, &up->dst);
971 		else {
972 			sync = 0;
973 
974 			/*
975 			 * Non-TCP protocol state machine always go
976 			 * forwards
977 			 */
978 			if (st->src.state > up->src.state)
979 				sync++;
980 			else
981 				pf_state_peer_ntoh(&up->src, &st->src);
982 			if (st->dst.state > up->dst.state)
983 				sync++;
984 			else
985 				pf_state_peer_ntoh(&up->dst, &st->dst);
986 		}
987 		if (sync < 2) {
988 			pfsync_alloc_scrub_memory(&up->dst, &st->dst);
989 			pf_state_peer_ntoh(&up->dst, &st->dst);
990 			st->expire = time_uptime;
991 			st->timeout = up->timeout;
992 		}
993 		st->pfsync_time = time_uptime;
994 
995 		if (sync) {
996 			V_pfsyncstats.pfsyncs_stale++;
997 
998 			pfsync_update_state(st);
999 			PF_STATE_UNLOCK(st);
1000 			PFSYNC_LOCK(sc);
1001 			pfsync_push(sc);
1002 			PFSYNC_UNLOCK(sc);
1003 			continue;
1004 		}
1005 		PF_STATE_UNLOCK(st);
1006 	}
1007 
1008 	return (len);
1009 }
1010 
1011 static int
1012 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1013 {
1014 	struct pfsync_upd_req *ur, *ura;
1015 	struct mbuf *mp;
1016 	int len = count * sizeof(*ur);
1017 	int i, offp;
1018 
1019 	struct pf_state *st;
1020 
1021 	mp = m_pulldown(m, offset, len, &offp);
1022 	if (mp == NULL) {
1023 		V_pfsyncstats.pfsyncs_badlen++;
1024 		return (-1);
1025 	}
1026 	ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1027 
1028 	for (i = 0; i < count; i++) {
1029 		ur = &ura[i];
1030 
1031 		if (ur->id == 0 && ur->creatorid == 0)
1032 			pfsync_bulk_start();
1033 		else {
1034 			st = pf_find_state_byid(ur->id, ur->creatorid);
1035 			if (st == NULL) {
1036 				V_pfsyncstats.pfsyncs_badstate++;
1037 				continue;
1038 			}
1039 			if (st->state_flags & PFSTATE_NOSYNC) {
1040 				PF_STATE_UNLOCK(st);
1041 				continue;
1042 			}
1043 
1044 			pfsync_update_state_req(st);
1045 			PF_STATE_UNLOCK(st);
1046 		}
1047 	}
1048 
1049 	return (len);
1050 }
1051 
1052 static int
1053 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1054 {
1055 	struct mbuf *mp;
1056 	struct pfsync_state *sa, *sp;
1057 	struct pf_state *st;
1058 	int len = count * sizeof(*sp);
1059 	int offp, i;
1060 
1061 	mp = m_pulldown(m, offset, len, &offp);
1062 	if (mp == NULL) {
1063 		V_pfsyncstats.pfsyncs_badlen++;
1064 		return (-1);
1065 	}
1066 	sa = (struct pfsync_state *)(mp->m_data + offp);
1067 
1068 	for (i = 0; i < count; i++) {
1069 		sp = &sa[i];
1070 
1071 		st = pf_find_state_byid(sp->id, sp->creatorid);
1072 		if (st == NULL) {
1073 			V_pfsyncstats.pfsyncs_badstate++;
1074 			continue;
1075 		}
1076 		st->state_flags |= PFSTATE_NOSYNC;
1077 		pf_unlink_state(st, PF_ENTER_LOCKED);
1078 	}
1079 
1080 	return (len);
1081 }
1082 
1083 static int
1084 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1085 {
1086 	struct mbuf *mp;
1087 	struct pfsync_del_c *sa, *sp;
1088 	struct pf_state *st;
1089 	int len = count * sizeof(*sp);
1090 	int offp, i;
1091 
1092 	mp = m_pulldown(m, offset, len, &offp);
1093 	if (mp == NULL) {
1094 		V_pfsyncstats.pfsyncs_badlen++;
1095 		return (-1);
1096 	}
1097 	sa = (struct pfsync_del_c *)(mp->m_data + offp);
1098 
1099 	for (i = 0; i < count; i++) {
1100 		sp = &sa[i];
1101 
1102 		st = pf_find_state_byid(sp->id, sp->creatorid);
1103 		if (st == NULL) {
1104 			V_pfsyncstats.pfsyncs_badstate++;
1105 			continue;
1106 		}
1107 
1108 		st->state_flags |= PFSTATE_NOSYNC;
1109 		pf_unlink_state(st, PF_ENTER_LOCKED);
1110 	}
1111 
1112 	return (len);
1113 }
1114 
1115 static int
1116 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1117 {
1118 	struct pfsync_softc *sc = V_pfsyncif;
1119 	struct pfsync_bus *bus;
1120 	struct mbuf *mp;
1121 	int len = count * sizeof(*bus);
1122 	int offp;
1123 
1124 	PFSYNC_BLOCK(sc);
1125 
1126 	/* If we're not waiting for a bulk update, who cares. */
1127 	if (sc->sc_ureq_sent == 0) {
1128 		PFSYNC_BUNLOCK(sc);
1129 		return (len);
1130 	}
1131 
1132 	mp = m_pulldown(m, offset, len, &offp);
1133 	if (mp == NULL) {
1134 		PFSYNC_BUNLOCK(sc);
1135 		V_pfsyncstats.pfsyncs_badlen++;
1136 		return (-1);
1137 	}
1138 	bus = (struct pfsync_bus *)(mp->m_data + offp);
1139 
1140 	switch (bus->status) {
1141 	case PFSYNC_BUS_START:
1142 		callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1143 		    V_pf_limits[PF_LIMIT_STATES].limit /
1144 		    ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1145 		    sizeof(struct pfsync_state)),
1146 		    pfsync_bulk_fail, sc);
1147 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1148 			printf("pfsync: received bulk update start\n");
1149 		break;
1150 
1151 	case PFSYNC_BUS_END:
1152 		if (time_uptime - ntohl(bus->endtime) >=
1153 		    sc->sc_ureq_sent) {
1154 			/* that's it, we're happy */
1155 			sc->sc_ureq_sent = 0;
1156 			sc->sc_bulk_tries = 0;
1157 			callout_stop(&sc->sc_bulkfail_tmo);
1158 			if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1159 				(*carp_demote_adj_p)(-V_pfsync_carp_adj,
1160 				    "pfsync bulk done");
1161 			sc->sc_flags |= PFSYNCF_OK;
1162 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1163 				printf("pfsync: received valid "
1164 				    "bulk update end\n");
1165 		} else {
1166 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1167 				printf("pfsync: received invalid "
1168 				    "bulk update end: bad timestamp\n");
1169 		}
1170 		break;
1171 	}
1172 	PFSYNC_BUNLOCK(sc);
1173 
1174 	return (len);
1175 }
1176 
1177 static int
1178 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1179 {
1180 	int len = count * sizeof(struct pfsync_tdb);
1181 
1182 #if defined(IPSEC)
1183 	struct pfsync_tdb *tp;
1184 	struct mbuf *mp;
1185 	int offp;
1186 	int i;
1187 	int s;
1188 
1189 	mp = m_pulldown(m, offset, len, &offp);
1190 	if (mp == NULL) {
1191 		V_pfsyncstats.pfsyncs_badlen++;
1192 		return (-1);
1193 	}
1194 	tp = (struct pfsync_tdb *)(mp->m_data + offp);
1195 
1196 	for (i = 0; i < count; i++)
1197 		pfsync_update_net_tdb(&tp[i]);
1198 #endif
1199 
1200 	return (len);
1201 }
1202 
1203 #if defined(IPSEC)
1204 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1205 static void
1206 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1207 {
1208 	struct tdb		*tdb;
1209 	int			 s;
1210 
1211 	/* check for invalid values */
1212 	if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1213 	    (pt->dst.sa.sa_family != AF_INET &&
1214 	    pt->dst.sa.sa_family != AF_INET6))
1215 		goto bad;
1216 
1217 	tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1218 	if (tdb) {
1219 		pt->rpl = ntohl(pt->rpl);
1220 		pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1221 
1222 		/* Neither replay nor byte counter should ever decrease. */
1223 		if (pt->rpl < tdb->tdb_rpl ||
1224 		    pt->cur_bytes < tdb->tdb_cur_bytes) {
1225 			goto bad;
1226 		}
1227 
1228 		tdb->tdb_rpl = pt->rpl;
1229 		tdb->tdb_cur_bytes = pt->cur_bytes;
1230 	}
1231 	return;
1232 
1233 bad:
1234 	if (V_pf_status.debug >= PF_DEBUG_MISC)
1235 		printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1236 		    "invalid value\n");
1237 	V_pfsyncstats.pfsyncs_badstate++;
1238 	return;
1239 }
1240 #endif
1241 
1242 
1243 static int
1244 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1245 {
1246 	/* check if we are at the right place in the packet */
1247 	if (offset != m->m_pkthdr.len)
1248 		V_pfsyncstats.pfsyncs_badlen++;
1249 
1250 	/* we're done. free and let the caller return */
1251 	m_freem(m);
1252 	return (-1);
1253 }
1254 
1255 static int
1256 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1257 {
1258 	V_pfsyncstats.pfsyncs_badact++;
1259 
1260 	m_freem(m);
1261 	return (-1);
1262 }
1263 
1264 static int
1265 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1266 	struct route *rt)
1267 {
1268 	m_freem(m);
1269 	return (0);
1270 }
1271 
1272 /* ARGSUSED */
1273 static int
1274 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1275 {
1276 	struct pfsync_softc *sc = ifp->if_softc;
1277 	struct ifreq *ifr = (struct ifreq *)data;
1278 	struct pfsyncreq pfsyncr;
1279 	int error;
1280 
1281 	switch (cmd) {
1282 	case SIOCSIFFLAGS:
1283 		PFSYNC_LOCK(sc);
1284 		if (ifp->if_flags & IFF_UP) {
1285 			ifp->if_drv_flags |= IFF_DRV_RUNNING;
1286 			PFSYNC_UNLOCK(sc);
1287 			pfsync_pointers_init();
1288 		} else {
1289 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1290 			PFSYNC_UNLOCK(sc);
1291 			pfsync_pointers_uninit();
1292 		}
1293 		break;
1294 	case SIOCSIFMTU:
1295 		if (!sc->sc_sync_if ||
1296 		    ifr->ifr_mtu <= PFSYNC_MINPKT ||
1297 		    ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1298 			return (EINVAL);
1299 		if (ifr->ifr_mtu < ifp->if_mtu) {
1300 			PFSYNC_LOCK(sc);
1301 			if (sc->sc_len > PFSYNC_MINPKT)
1302 				pfsync_sendout(1);
1303 			PFSYNC_UNLOCK(sc);
1304 		}
1305 		ifp->if_mtu = ifr->ifr_mtu;
1306 		break;
1307 	case SIOCGETPFSYNC:
1308 		bzero(&pfsyncr, sizeof(pfsyncr));
1309 		PFSYNC_LOCK(sc);
1310 		if (sc->sc_sync_if) {
1311 			strlcpy(pfsyncr.pfsyncr_syncdev,
1312 			    sc->sc_sync_if->if_xname, IFNAMSIZ);
1313 		}
1314 		pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1315 		pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1316 		pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER ==
1317 		    (sc->sc_flags & PFSYNCF_DEFER));
1318 		PFSYNC_UNLOCK(sc);
1319 		return (copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)));
1320 
1321 	case SIOCSETPFSYNC:
1322 	    {
1323 		struct ip_moptions *imo = &sc->sc_imo;
1324 		struct ifnet *sifp;
1325 		struct ip *ip;
1326 		void *mship = NULL;
1327 
1328 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1329 			return (error);
1330 		if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr))))
1331 			return (error);
1332 
1333 		if (pfsyncr.pfsyncr_maxupdates > 255)
1334 			return (EINVAL);
1335 
1336 		if (pfsyncr.pfsyncr_syncdev[0] == 0)
1337 			sifp = NULL;
1338 		else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL)
1339 			return (EINVAL);
1340 
1341 		if (sifp != NULL && (
1342 		    pfsyncr.pfsyncr_syncpeer.s_addr == 0 ||
1343 		    pfsyncr.pfsyncr_syncpeer.s_addr ==
1344 		    htonl(INADDR_PFSYNC_GROUP)))
1345 			mship = malloc((sizeof(struct in_multi *) *
1346 			    IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO);
1347 
1348 		PFSYNC_LOCK(sc);
1349 		if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1350 			sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP);
1351 		else
1352 			sc->sc_sync_peer.s_addr =
1353 			    pfsyncr.pfsyncr_syncpeer.s_addr;
1354 
1355 		sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1356 		if (pfsyncr.pfsyncr_defer) {
1357 			sc->sc_flags |= PFSYNCF_DEFER;
1358 			pfsync_defer_ptr = pfsync_defer;
1359 		} else {
1360 			sc->sc_flags &= ~PFSYNCF_DEFER;
1361 			pfsync_defer_ptr = NULL;
1362 		}
1363 
1364 		if (sifp == NULL) {
1365 			if (sc->sc_sync_if)
1366 				if_rele(sc->sc_sync_if);
1367 			sc->sc_sync_if = NULL;
1368 			if (imo->imo_membership)
1369 				pfsync_multicast_cleanup(sc);
1370 			PFSYNC_UNLOCK(sc);
1371 			break;
1372 		}
1373 
1374 		if (sc->sc_len > PFSYNC_MINPKT &&
1375 		    (sifp->if_mtu < sc->sc_ifp->if_mtu ||
1376 		    (sc->sc_sync_if != NULL &&
1377 		    sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
1378 		    sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
1379 			pfsync_sendout(1);
1380 
1381 		if (imo->imo_membership)
1382 			pfsync_multicast_cleanup(sc);
1383 
1384 		if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
1385 			error = pfsync_multicast_setup(sc, sifp, mship);
1386 			if (error) {
1387 				if_rele(sifp);
1388 				free(mship, M_PFSYNC);
1389 				return (error);
1390 			}
1391 		}
1392 		if (sc->sc_sync_if)
1393 			if_rele(sc->sc_sync_if);
1394 		sc->sc_sync_if = sifp;
1395 
1396 		ip = &sc->sc_template;
1397 		bzero(ip, sizeof(*ip));
1398 		ip->ip_v = IPVERSION;
1399 		ip->ip_hl = sizeof(sc->sc_template) >> 2;
1400 		ip->ip_tos = IPTOS_LOWDELAY;
1401 		/* len and id are set later. */
1402 		ip->ip_off = htons(IP_DF);
1403 		ip->ip_ttl = PFSYNC_DFLTTL;
1404 		ip->ip_p = IPPROTO_PFSYNC;
1405 		ip->ip_src.s_addr = INADDR_ANY;
1406 		ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr;
1407 
1408 		/* Request a full state table update. */
1409 		if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1410 			(*carp_demote_adj_p)(V_pfsync_carp_adj,
1411 			    "pfsync bulk start");
1412 		sc->sc_flags &= ~PFSYNCF_OK;
1413 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1414 			printf("pfsync: requesting bulk update\n");
1415 		pfsync_request_update(0, 0);
1416 		PFSYNC_UNLOCK(sc);
1417 		PFSYNC_BLOCK(sc);
1418 		sc->sc_ureq_sent = time_uptime;
1419 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail,
1420 		    sc);
1421 		PFSYNC_BUNLOCK(sc);
1422 
1423 		break;
1424 	    }
1425 	default:
1426 		return (ENOTTY);
1427 	}
1428 
1429 	return (0);
1430 }
1431 
1432 static void
1433 pfsync_out_state(struct pf_state *st, void *buf)
1434 {
1435 	struct pfsync_state *sp = buf;
1436 
1437 	pfsync_state_export(sp, st);
1438 }
1439 
1440 static void
1441 pfsync_out_iack(struct pf_state *st, void *buf)
1442 {
1443 	struct pfsync_ins_ack *iack = buf;
1444 
1445 	iack->id = st->id;
1446 	iack->creatorid = st->creatorid;
1447 }
1448 
1449 static void
1450 pfsync_out_upd_c(struct pf_state *st, void *buf)
1451 {
1452 	struct pfsync_upd_c *up = buf;
1453 
1454 	bzero(up, sizeof(*up));
1455 	up->id = st->id;
1456 	pf_state_peer_hton(&st->src, &up->src);
1457 	pf_state_peer_hton(&st->dst, &up->dst);
1458 	up->creatorid = st->creatorid;
1459 	up->timeout = st->timeout;
1460 }
1461 
1462 static void
1463 pfsync_out_del(struct pf_state *st, void *buf)
1464 {
1465 	struct pfsync_del_c *dp = buf;
1466 
1467 	dp->id = st->id;
1468 	dp->creatorid = st->creatorid;
1469 	st->state_flags |= PFSTATE_NOSYNC;
1470 }
1471 
1472 static void
1473 pfsync_drop(struct pfsync_softc *sc)
1474 {
1475 	struct pf_state *st, *next;
1476 	struct pfsync_upd_req_item *ur;
1477 	int q;
1478 
1479 	for (q = 0; q < PFSYNC_S_COUNT; q++) {
1480 		if (TAILQ_EMPTY(&sc->sc_qs[q]))
1481 			continue;
1482 
1483 		TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, next) {
1484 			KASSERT(st->sync_state == q,
1485 				("%s: st->sync_state == q",
1486 					__func__));
1487 			st->sync_state = PFSYNC_S_NONE;
1488 			pf_release_state(st);
1489 		}
1490 		TAILQ_INIT(&sc->sc_qs[q]);
1491 	}
1492 
1493 	while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1494 		TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1495 		free(ur, M_PFSYNC);
1496 	}
1497 
1498 	sc->sc_plus = NULL;
1499 	sc->sc_len = PFSYNC_MINPKT;
1500 }
1501 
1502 static void
1503 pfsync_sendout(int schedswi)
1504 {
1505 	struct pfsync_softc *sc = V_pfsyncif;
1506 	struct ifnet *ifp = sc->sc_ifp;
1507 	struct mbuf *m;
1508 	struct ip *ip;
1509 	struct pfsync_header *ph;
1510 	struct pfsync_subheader *subh;
1511 	struct pf_state *st;
1512 	struct pfsync_upd_req_item *ur;
1513 	int offset;
1514 	int q, count = 0;
1515 
1516 	KASSERT(sc != NULL, ("%s: null sc", __func__));
1517 	KASSERT(sc->sc_len > PFSYNC_MINPKT,
1518 	    ("%s: sc_len %zu", __func__, sc->sc_len));
1519 	PFSYNC_LOCK_ASSERT(sc);
1520 
1521 	if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1522 		pfsync_drop(sc);
1523 		return;
1524 	}
1525 
1526 	m = m_get2(max_linkhdr + sc->sc_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1527 	if (m == NULL) {
1528 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1529 		V_pfsyncstats.pfsyncs_onomem++;
1530 		return;
1531 	}
1532 	m->m_data += max_linkhdr;
1533 	m->m_len = m->m_pkthdr.len = sc->sc_len;
1534 
1535 	/* build the ip header */
1536 	ip = (struct ip *)m->m_data;
1537 	bcopy(&sc->sc_template, ip, sizeof(*ip));
1538 	offset = sizeof(*ip);
1539 
1540 	ip->ip_len = htons(m->m_pkthdr.len);
1541 	ip_fillid(ip);
1542 
1543 	/* build the pfsync header */
1544 	ph = (struct pfsync_header *)(m->m_data + offset);
1545 	bzero(ph, sizeof(*ph));
1546 	offset += sizeof(*ph);
1547 
1548 	ph->version = PFSYNC_VERSION;
1549 	ph->len = htons(sc->sc_len - sizeof(*ip));
1550 	bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1551 
1552 	/* walk the queues */
1553 	for (q = 0; q < PFSYNC_S_COUNT; q++) {
1554 		if (TAILQ_EMPTY(&sc->sc_qs[q]))
1555 			continue;
1556 
1557 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1558 		offset += sizeof(*subh);
1559 
1560 		count = 0;
1561 		TAILQ_FOREACH(st, &sc->sc_qs[q], sync_list) {
1562 			KASSERT(st->sync_state == q,
1563 				("%s: st->sync_state == q",
1564 					__func__));
1565 			/*
1566 			 * XXXGL: some of write methods do unlocked reads
1567 			 * of state data :(
1568 			 */
1569 			pfsync_qs[q].write(st, m->m_data + offset);
1570 			offset += pfsync_qs[q].len;
1571 			st->sync_state = PFSYNC_S_NONE;
1572 			pf_release_state(st);
1573 			count++;
1574 		}
1575 		TAILQ_INIT(&sc->sc_qs[q]);
1576 
1577 		bzero(subh, sizeof(*subh));
1578 		subh->action = pfsync_qs[q].action;
1579 		subh->count = htons(count);
1580 		V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1581 	}
1582 
1583 	if (!TAILQ_EMPTY(&sc->sc_upd_req_list)) {
1584 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1585 		offset += sizeof(*subh);
1586 
1587 		count = 0;
1588 		while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1589 			TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1590 
1591 			bcopy(&ur->ur_msg, m->m_data + offset,
1592 			    sizeof(ur->ur_msg));
1593 			offset += sizeof(ur->ur_msg);
1594 			free(ur, M_PFSYNC);
1595 			count++;
1596 		}
1597 
1598 		bzero(subh, sizeof(*subh));
1599 		subh->action = PFSYNC_ACT_UPD_REQ;
1600 		subh->count = htons(count);
1601 		V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1602 	}
1603 
1604 	/* has someone built a custom region for us to add? */
1605 	if (sc->sc_plus != NULL) {
1606 		bcopy(sc->sc_plus, m->m_data + offset, sc->sc_pluslen);
1607 		offset += sc->sc_pluslen;
1608 
1609 		sc->sc_plus = NULL;
1610 	}
1611 
1612 	subh = (struct pfsync_subheader *)(m->m_data + offset);
1613 	offset += sizeof(*subh);
1614 
1615 	bzero(subh, sizeof(*subh));
1616 	subh->action = PFSYNC_ACT_EOF;
1617 	subh->count = htons(1);
1618 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1619 
1620 	/* we're done, let's put it on the wire */
1621 	if (ifp->if_bpf) {
1622 		m->m_data += sizeof(*ip);
1623 		m->m_len = m->m_pkthdr.len = sc->sc_len - sizeof(*ip);
1624 		BPF_MTAP(ifp, m);
1625 		m->m_data -= sizeof(*ip);
1626 		m->m_len = m->m_pkthdr.len = sc->sc_len;
1627 	}
1628 
1629 	if (sc->sc_sync_if == NULL) {
1630 		sc->sc_len = PFSYNC_MINPKT;
1631 		m_freem(m);
1632 		return;
1633 	}
1634 
1635 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1636 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1637 	sc->sc_len = PFSYNC_MINPKT;
1638 
1639 	if (!_IF_QFULL(&sc->sc_ifp->if_snd))
1640 		_IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1641 	else {
1642 		m_freem(m);
1643 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1644 	}
1645 	if (schedswi)
1646 		swi_sched(V_pfsync_swi_cookie, 0);
1647 }
1648 
1649 static void
1650 pfsync_insert_state(struct pf_state *st)
1651 {
1652 	struct pfsync_softc *sc = V_pfsyncif;
1653 
1654 	if (st->state_flags & PFSTATE_NOSYNC)
1655 		return;
1656 
1657 	if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1658 	    st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1659 		st->state_flags |= PFSTATE_NOSYNC;
1660 		return;
1661 	}
1662 
1663 	KASSERT(st->sync_state == PFSYNC_S_NONE,
1664 		("%s: st->sync_state %u", __func__, st->sync_state));
1665 
1666 	PFSYNC_LOCK(sc);
1667 	if (sc->sc_len == PFSYNC_MINPKT)
1668 		callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1669 
1670 	pfsync_q_ins(st, PFSYNC_S_INS);
1671 	PFSYNC_UNLOCK(sc);
1672 
1673 	st->sync_updates = 0;
1674 }
1675 
1676 static int
1677 pfsync_defer(struct pf_state *st, struct mbuf *m)
1678 {
1679 	struct pfsync_softc *sc = V_pfsyncif;
1680 	struct pfsync_deferral *pd;
1681 
1682 	if (m->m_flags & (M_BCAST|M_MCAST))
1683 		return (0);
1684 
1685 	PFSYNC_LOCK(sc);
1686 
1687 	if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) ||
1688 	    !(sc->sc_flags & PFSYNCF_DEFER)) {
1689 		PFSYNC_UNLOCK(sc);
1690 		return (0);
1691 	}
1692 
1693 	 if (sc->sc_deferred >= 128)
1694 		pfsync_undefer(TAILQ_FIRST(&sc->sc_deferrals), 0);
1695 
1696 	pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1697 	if (pd == NULL)
1698 		return (0);
1699 	sc->sc_deferred++;
1700 
1701 	m->m_flags |= M_SKIP_FIREWALL;
1702 	st->state_flags |= PFSTATE_ACK;
1703 
1704 	pd->pd_sc = sc;
1705 	pd->pd_refs = 0;
1706 	pd->pd_st = st;
1707 	pf_ref_state(st);
1708 	pd->pd_m = m;
1709 
1710 	TAILQ_INSERT_TAIL(&sc->sc_deferrals, pd, pd_entry);
1711 	callout_init_mtx(&pd->pd_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1712 	callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd);
1713 
1714 	pfsync_push(sc);
1715 
1716 	return (1);
1717 }
1718 
1719 static void
1720 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1721 {
1722 	struct pfsync_softc *sc = pd->pd_sc;
1723 	struct mbuf *m = pd->pd_m;
1724 	struct pf_state *st = pd->pd_st;
1725 
1726 	PFSYNC_LOCK_ASSERT(sc);
1727 
1728 	TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1729 	sc->sc_deferred--;
1730 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
1731 	free(pd, M_PFSYNC);
1732 	pf_release_state(st);
1733 
1734 	if (drop)
1735 		m_freem(m);
1736 	else {
1737 		_IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1738 		pfsync_push(sc);
1739 	}
1740 }
1741 
1742 static void
1743 pfsync_defer_tmo(void *arg)
1744 {
1745 	struct pfsync_deferral *pd = arg;
1746 	struct pfsync_softc *sc = pd->pd_sc;
1747 	struct mbuf *m = pd->pd_m;
1748 	struct pf_state *st = pd->pd_st;
1749 
1750 	PFSYNC_LOCK_ASSERT(sc);
1751 
1752 	CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
1753 
1754 	TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1755 	sc->sc_deferred--;
1756 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
1757 	if (pd->pd_refs == 0)
1758 		free(pd, M_PFSYNC);
1759 	PFSYNC_UNLOCK(sc);
1760 
1761 	ip_output(m, NULL, NULL, 0, NULL, NULL);
1762 
1763 	pf_release_state(st);
1764 
1765 	CURVNET_RESTORE();
1766 }
1767 
1768 static void
1769 pfsync_undefer_state(struct pf_state *st, int drop)
1770 {
1771 	struct pfsync_softc *sc = V_pfsyncif;
1772 	struct pfsync_deferral *pd;
1773 
1774 	PFSYNC_LOCK_ASSERT(sc);
1775 
1776 	TAILQ_FOREACH(pd, &sc->sc_deferrals, pd_entry) {
1777 		 if (pd->pd_st == st) {
1778 			if (callout_stop(&pd->pd_tmo) > 0)
1779 				pfsync_undefer(pd, drop);
1780 			return;
1781 		}
1782 	}
1783 
1784 	panic("%s: unable to find deferred state", __func__);
1785 }
1786 
1787 static void
1788 pfsync_update_state(struct pf_state *st)
1789 {
1790 	struct pfsync_softc *sc = V_pfsyncif;
1791 	int sync = 0;
1792 
1793 	PF_STATE_LOCK_ASSERT(st);
1794 	PFSYNC_LOCK(sc);
1795 
1796 	if (st->state_flags & PFSTATE_ACK)
1797 		pfsync_undefer_state(st, 0);
1798 	if (st->state_flags & PFSTATE_NOSYNC) {
1799 		if (st->sync_state != PFSYNC_S_NONE)
1800 			pfsync_q_del(st);
1801 		PFSYNC_UNLOCK(sc);
1802 		return;
1803 	}
1804 
1805 	if (sc->sc_len == PFSYNC_MINPKT)
1806 		callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1807 
1808 	switch (st->sync_state) {
1809 	case PFSYNC_S_UPD_C:
1810 	case PFSYNC_S_UPD:
1811 	case PFSYNC_S_INS:
1812 		/* we're already handling it */
1813 
1814 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1815 			st->sync_updates++;
1816 			if (st->sync_updates >= sc->sc_maxupdates)
1817 				sync = 1;
1818 		}
1819 		break;
1820 
1821 	case PFSYNC_S_IACK:
1822 		pfsync_q_del(st);
1823 	case PFSYNC_S_NONE:
1824 		pfsync_q_ins(st, PFSYNC_S_UPD_C);
1825 		st->sync_updates = 0;
1826 		break;
1827 
1828 	default:
1829 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1830 	}
1831 
1832 	if (sync || (time_uptime - st->pfsync_time) < 2)
1833 		pfsync_push(sc);
1834 
1835 	PFSYNC_UNLOCK(sc);
1836 }
1837 
1838 static void
1839 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1840 {
1841 	struct pfsync_softc *sc = V_pfsyncif;
1842 	struct pfsync_upd_req_item *item;
1843 	size_t nlen = sizeof(struct pfsync_upd_req);
1844 
1845 	PFSYNC_LOCK_ASSERT(sc);
1846 
1847 	/*
1848 	 * This code does a bit to prevent multiple update requests for the
1849 	 * same state being generated. It searches current subheader queue,
1850 	 * but it doesn't lookup into queue of already packed datagrams.
1851 	 */
1852 	TAILQ_FOREACH(item, &sc->sc_upd_req_list, ur_entry)
1853 		if (item->ur_msg.id == id &&
1854 		    item->ur_msg.creatorid == creatorid)
1855 			return;
1856 
1857 	item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1858 	if (item == NULL)
1859 		return; /* XXX stats */
1860 
1861 	item->ur_msg.id = id;
1862 	item->ur_msg.creatorid = creatorid;
1863 
1864 	if (TAILQ_EMPTY(&sc->sc_upd_req_list))
1865 		nlen += sizeof(struct pfsync_subheader);
1866 
1867 	if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
1868 		pfsync_sendout(1);
1869 
1870 		nlen = sizeof(struct pfsync_subheader) +
1871 		    sizeof(struct pfsync_upd_req);
1872 	}
1873 
1874 	TAILQ_INSERT_TAIL(&sc->sc_upd_req_list, item, ur_entry);
1875 	sc->sc_len += nlen;
1876 }
1877 
1878 static void
1879 pfsync_update_state_req(struct pf_state *st)
1880 {
1881 	struct pfsync_softc *sc = V_pfsyncif;
1882 
1883 	PF_STATE_LOCK_ASSERT(st);
1884 	PFSYNC_LOCK(sc);
1885 
1886 	if (st->state_flags & PFSTATE_NOSYNC) {
1887 		if (st->sync_state != PFSYNC_S_NONE)
1888 			pfsync_q_del(st);
1889 		PFSYNC_UNLOCK(sc);
1890 		return;
1891 	}
1892 
1893 	switch (st->sync_state) {
1894 	case PFSYNC_S_UPD_C:
1895 	case PFSYNC_S_IACK:
1896 		pfsync_q_del(st);
1897 	case PFSYNC_S_NONE:
1898 		pfsync_q_ins(st, PFSYNC_S_UPD);
1899 		pfsync_push(sc);
1900 		break;
1901 
1902 	case PFSYNC_S_INS:
1903 	case PFSYNC_S_UPD:
1904 	case PFSYNC_S_DEL:
1905 		/* we're already handling it */
1906 		break;
1907 
1908 	default:
1909 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1910 	}
1911 
1912 	PFSYNC_UNLOCK(sc);
1913 }
1914 
1915 static void
1916 pfsync_delete_state(struct pf_state *st)
1917 {
1918 	struct pfsync_softc *sc = V_pfsyncif;
1919 
1920 	PFSYNC_LOCK(sc);
1921 	if (st->state_flags & PFSTATE_ACK)
1922 		pfsync_undefer_state(st, 1);
1923 	if (st->state_flags & PFSTATE_NOSYNC) {
1924 		if (st->sync_state != PFSYNC_S_NONE)
1925 			pfsync_q_del(st);
1926 		PFSYNC_UNLOCK(sc);
1927 		return;
1928 	}
1929 
1930 	if (sc->sc_len == PFSYNC_MINPKT)
1931 		callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1932 
1933 	switch (st->sync_state) {
1934 	case PFSYNC_S_INS:
1935 		/* We never got to tell the world so just forget about it. */
1936 		pfsync_q_del(st);
1937 		break;
1938 
1939 	case PFSYNC_S_UPD_C:
1940 	case PFSYNC_S_UPD:
1941 	case PFSYNC_S_IACK:
1942 		pfsync_q_del(st);
1943 		/* FALLTHROUGH to putting it on the del list */
1944 
1945 	case PFSYNC_S_NONE:
1946 		pfsync_q_ins(st, PFSYNC_S_DEL);
1947 		break;
1948 
1949 	default:
1950 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1951 	}
1952 	PFSYNC_UNLOCK(sc);
1953 }
1954 
1955 static void
1956 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
1957 {
1958 	struct pfsync_softc *sc = V_pfsyncif;
1959 	struct {
1960 		struct pfsync_subheader subh;
1961 		struct pfsync_clr clr;
1962 	} __packed r;
1963 
1964 	bzero(&r, sizeof(r));
1965 
1966 	r.subh.action = PFSYNC_ACT_CLR;
1967 	r.subh.count = htons(1);
1968 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
1969 
1970 	strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
1971 	r.clr.creatorid = creatorid;
1972 
1973 	PFSYNC_LOCK(sc);
1974 	pfsync_send_plus(&r, sizeof(r));
1975 	PFSYNC_UNLOCK(sc);
1976 }
1977 
1978 static void
1979 pfsync_q_ins(struct pf_state *st, int q)
1980 {
1981 	struct pfsync_softc *sc = V_pfsyncif;
1982 	size_t nlen = pfsync_qs[q].len;
1983 
1984 	PFSYNC_LOCK_ASSERT(sc);
1985 
1986 	KASSERT(st->sync_state == PFSYNC_S_NONE,
1987 		("%s: st->sync_state %u", __func__, st->sync_state));
1988 	KASSERT(sc->sc_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
1989 	    sc->sc_len));
1990 
1991 	if (TAILQ_EMPTY(&sc->sc_qs[q]))
1992 		nlen += sizeof(struct pfsync_subheader);
1993 
1994 	if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
1995 		pfsync_sendout(1);
1996 
1997 		nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
1998 	}
1999 
2000 	sc->sc_len += nlen;
2001 	TAILQ_INSERT_TAIL(&sc->sc_qs[q], st, sync_list);
2002 	st->sync_state = q;
2003 	pf_ref_state(st);
2004 }
2005 
2006 static void
2007 pfsync_q_del(struct pf_state *st)
2008 {
2009 	struct pfsync_softc *sc = V_pfsyncif;
2010 	int q = st->sync_state;
2011 
2012 	PFSYNC_LOCK_ASSERT(sc);
2013 	KASSERT(st->sync_state != PFSYNC_S_NONE,
2014 		("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2015 
2016 	sc->sc_len -= pfsync_qs[q].len;
2017 	TAILQ_REMOVE(&sc->sc_qs[q], st, sync_list);
2018 	st->sync_state = PFSYNC_S_NONE;
2019 	pf_release_state(st);
2020 
2021 	if (TAILQ_EMPTY(&sc->sc_qs[q]))
2022 		sc->sc_len -= sizeof(struct pfsync_subheader);
2023 }
2024 
2025 static void
2026 pfsync_bulk_start(void)
2027 {
2028 	struct pfsync_softc *sc = V_pfsyncif;
2029 
2030 	if (V_pf_status.debug >= PF_DEBUG_MISC)
2031 		printf("pfsync: received bulk update request\n");
2032 
2033 	PFSYNC_BLOCK(sc);
2034 
2035 	sc->sc_ureq_received = time_uptime;
2036 	sc->sc_bulk_hashid = 0;
2037 	sc->sc_bulk_stateid = 0;
2038 	pfsync_bulk_status(PFSYNC_BUS_START);
2039 	callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2040 	PFSYNC_BUNLOCK(sc);
2041 }
2042 
2043 static void
2044 pfsync_bulk_update(void *arg)
2045 {
2046 	struct pfsync_softc *sc = arg;
2047 	struct pf_state *s;
2048 	int i, sent = 0;
2049 
2050 	PFSYNC_BLOCK_ASSERT(sc);
2051 	CURVNET_SET(sc->sc_ifp->if_vnet);
2052 
2053 	/*
2054 	 * Start with last state from previous invocation.
2055 	 * It may had gone, in this case start from the
2056 	 * hash slot.
2057 	 */
2058 	s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2059 
2060 	if (s != NULL)
2061 		i = PF_IDHASH(s);
2062 	else
2063 		i = sc->sc_bulk_hashid;
2064 
2065 	for (; i <= pf_hashmask; i++) {
2066 		struct pf_idhash *ih = &V_pf_idhash[i];
2067 
2068 		if (s != NULL)
2069 			PF_HASHROW_ASSERT(ih);
2070 		else {
2071 			PF_HASHROW_LOCK(ih);
2072 			s = LIST_FIRST(&ih->states);
2073 		}
2074 
2075 		for (; s; s = LIST_NEXT(s, entry)) {
2076 
2077 			if (sent > 1 && (sc->sc_ifp->if_mtu - sc->sc_len) <
2078 			    sizeof(struct pfsync_state)) {
2079 				/* We've filled a packet. */
2080 				sc->sc_bulk_hashid = i;
2081 				sc->sc_bulk_stateid = s->id;
2082 				sc->sc_bulk_creatorid = s->creatorid;
2083 				PF_HASHROW_UNLOCK(ih);
2084 				callout_reset(&sc->sc_bulk_tmo, 1,
2085 				    pfsync_bulk_update, sc);
2086 				goto full;
2087 			}
2088 
2089 			if (s->sync_state == PFSYNC_S_NONE &&
2090 			    s->timeout < PFTM_MAX &&
2091 			    s->pfsync_time <= sc->sc_ureq_received) {
2092 				pfsync_update_state_req(s);
2093 				sent++;
2094 			}
2095 		}
2096 		PF_HASHROW_UNLOCK(ih);
2097 	}
2098 
2099 	/* We're done. */
2100 	pfsync_bulk_status(PFSYNC_BUS_END);
2101 
2102 full:
2103 	CURVNET_RESTORE();
2104 }
2105 
2106 static void
2107 pfsync_bulk_status(u_int8_t status)
2108 {
2109 	struct {
2110 		struct pfsync_subheader subh;
2111 		struct pfsync_bus bus;
2112 	} __packed r;
2113 
2114 	struct pfsync_softc *sc = V_pfsyncif;
2115 
2116 	bzero(&r, sizeof(r));
2117 
2118 	r.subh.action = PFSYNC_ACT_BUS;
2119 	r.subh.count = htons(1);
2120 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2121 
2122 	r.bus.creatorid = V_pf_status.hostid;
2123 	r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2124 	r.bus.status = status;
2125 
2126 	PFSYNC_LOCK(sc);
2127 	pfsync_send_plus(&r, sizeof(r));
2128 	PFSYNC_UNLOCK(sc);
2129 }
2130 
2131 static void
2132 pfsync_bulk_fail(void *arg)
2133 {
2134 	struct pfsync_softc *sc = arg;
2135 
2136 	CURVNET_SET(sc->sc_ifp->if_vnet);
2137 
2138 	PFSYNC_BLOCK_ASSERT(sc);
2139 
2140 	if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2141 		/* Try again */
2142 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2143 		    pfsync_bulk_fail, V_pfsyncif);
2144 		PFSYNC_LOCK(sc);
2145 		pfsync_request_update(0, 0);
2146 		PFSYNC_UNLOCK(sc);
2147 	} else {
2148 		/* Pretend like the transfer was ok. */
2149 		sc->sc_ureq_sent = 0;
2150 		sc->sc_bulk_tries = 0;
2151 		PFSYNC_LOCK(sc);
2152 		if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2153 			(*carp_demote_adj_p)(-V_pfsync_carp_adj,
2154 			    "pfsync bulk fail");
2155 		sc->sc_flags |= PFSYNCF_OK;
2156 		PFSYNC_UNLOCK(sc);
2157 		if (V_pf_status.debug >= PF_DEBUG_MISC)
2158 			printf("pfsync: failed to receive bulk update\n");
2159 	}
2160 
2161 	CURVNET_RESTORE();
2162 }
2163 
2164 static void
2165 pfsync_send_plus(void *plus, size_t pluslen)
2166 {
2167 	struct pfsync_softc *sc = V_pfsyncif;
2168 
2169 	PFSYNC_LOCK_ASSERT(sc);
2170 
2171 	if (sc->sc_len + pluslen > sc->sc_ifp->if_mtu)
2172 		pfsync_sendout(1);
2173 
2174 	sc->sc_plus = plus;
2175 	sc->sc_len += (sc->sc_pluslen = pluslen);
2176 
2177 	pfsync_sendout(1);
2178 }
2179 
2180 static void
2181 pfsync_timeout(void *arg)
2182 {
2183 	struct pfsync_softc *sc = arg;
2184 
2185 	CURVNET_SET(sc->sc_ifp->if_vnet);
2186 	PFSYNC_LOCK(sc);
2187 	pfsync_push(sc);
2188 	PFSYNC_UNLOCK(sc);
2189 	CURVNET_RESTORE();
2190 }
2191 
2192 static void
2193 pfsync_push(struct pfsync_softc *sc)
2194 {
2195 
2196 	PFSYNC_LOCK_ASSERT(sc);
2197 
2198 	sc->sc_flags |= PFSYNCF_PUSH;
2199 	swi_sched(V_pfsync_swi_cookie, 0);
2200 }
2201 
2202 static void
2203 pfsyncintr(void *arg)
2204 {
2205 	struct pfsync_softc *sc = arg;
2206 	struct mbuf *m, *n;
2207 
2208 	CURVNET_SET(sc->sc_ifp->if_vnet);
2209 
2210 	PFSYNC_LOCK(sc);
2211 	if ((sc->sc_flags & PFSYNCF_PUSH) && sc->sc_len > PFSYNC_MINPKT) {
2212 		pfsync_sendout(0);
2213 		sc->sc_flags &= ~PFSYNCF_PUSH;
2214 	}
2215 	_IF_DEQUEUE_ALL(&sc->sc_ifp->if_snd, m);
2216 	PFSYNC_UNLOCK(sc);
2217 
2218 	for (; m != NULL; m = n) {
2219 
2220 		n = m->m_nextpkt;
2221 		m->m_nextpkt = NULL;
2222 
2223 		/*
2224 		 * We distinguish between a deferral packet and our
2225 		 * own pfsync packet based on M_SKIP_FIREWALL
2226 		 * flag. This is XXX.
2227 		 */
2228 		if (m->m_flags & M_SKIP_FIREWALL)
2229 			ip_output(m, NULL, NULL, 0, NULL, NULL);
2230 		else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo,
2231 		    NULL) == 0)
2232 			V_pfsyncstats.pfsyncs_opackets++;
2233 		else
2234 			V_pfsyncstats.pfsyncs_oerrors++;
2235 	}
2236 	CURVNET_RESTORE();
2237 }
2238 
2239 static int
2240 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship)
2241 {
2242 	struct ip_moptions *imo = &sc->sc_imo;
2243 	int error;
2244 
2245 	if (!(ifp->if_flags & IFF_MULTICAST))
2246 		return (EADDRNOTAVAIL);
2247 
2248 	imo->imo_membership = (struct in_multi **)mship;
2249 	imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
2250 	imo->imo_multicast_vif = -1;
2251 
2252 	if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL,
2253 	    &imo->imo_membership[0])) != 0) {
2254 		imo->imo_membership = NULL;
2255 		return (error);
2256 	}
2257 	imo->imo_num_memberships++;
2258 	imo->imo_multicast_ifp = ifp;
2259 	imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2260 	imo->imo_multicast_loop = 0;
2261 
2262 	return (0);
2263 }
2264 
2265 static void
2266 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2267 {
2268 	struct ip_moptions *imo = &sc->sc_imo;
2269 
2270 	in_leavegroup(imo->imo_membership[0], NULL);
2271 	free(imo->imo_membership, M_PFSYNC);
2272 	imo->imo_membership = NULL;
2273 	imo->imo_multicast_ifp = NULL;
2274 }
2275 
2276 #ifdef INET
2277 extern  struct domain inetdomain;
2278 static struct protosw in_pfsync_protosw = {
2279 	.pr_type =		SOCK_RAW,
2280 	.pr_domain =		&inetdomain,
2281 	.pr_protocol =		IPPROTO_PFSYNC,
2282 	.pr_flags =		PR_ATOMIC|PR_ADDR,
2283 	.pr_input =		pfsync_input,
2284 	.pr_output =		rip_output,
2285 	.pr_ctloutput =		rip_ctloutput,
2286 	.pr_usrreqs =		&rip_usrreqs
2287 };
2288 #endif
2289 
2290 static void
2291 pfsync_pointers_init()
2292 {
2293 
2294 	PF_RULES_WLOCK();
2295 	pfsync_state_import_ptr = pfsync_state_import;
2296 	pfsync_insert_state_ptr = pfsync_insert_state;
2297 	pfsync_update_state_ptr = pfsync_update_state;
2298 	pfsync_delete_state_ptr = pfsync_delete_state;
2299 	pfsync_clear_states_ptr = pfsync_clear_states;
2300 	pfsync_defer_ptr = pfsync_defer;
2301 	PF_RULES_WUNLOCK();
2302 }
2303 
2304 static void
2305 pfsync_pointers_uninit()
2306 {
2307 
2308 	PF_RULES_WLOCK();
2309 	pfsync_state_import_ptr = NULL;
2310 	pfsync_insert_state_ptr = NULL;
2311 	pfsync_update_state_ptr = NULL;
2312 	pfsync_delete_state_ptr = NULL;
2313 	pfsync_clear_states_ptr = NULL;
2314 	pfsync_defer_ptr = NULL;
2315 	PF_RULES_WUNLOCK();
2316 }
2317 
2318 static int
2319 pfsync_init()
2320 {
2321 	VNET_ITERATOR_DECL(vnet_iter);
2322 	int error = 0;
2323 
2324 	VNET_LIST_RLOCK();
2325 	VNET_FOREACH(vnet_iter) {
2326 		CURVNET_SET(vnet_iter);
2327 		V_pfsync_cloner = if_clone_simple(pfsyncname,
2328 		    pfsync_clone_create, pfsync_clone_destroy, 1);
2329 		error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif,
2330 		    SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2331 		CURVNET_RESTORE();
2332 		if (error)
2333 			goto fail_locked;
2334 	}
2335 	VNET_LIST_RUNLOCK();
2336 #ifdef INET
2337 	error = pf_proto_register(PF_INET, &in_pfsync_protosw);
2338 	if (error)
2339 		goto fail;
2340 	error = ipproto_register(IPPROTO_PFSYNC);
2341 	if (error) {
2342 		pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2343 		goto fail;
2344 	}
2345 #endif
2346 	pfsync_pointers_init();
2347 
2348 	return (0);
2349 
2350 fail:
2351 	VNET_LIST_RLOCK();
2352 fail_locked:
2353 	VNET_FOREACH(vnet_iter) {
2354 		CURVNET_SET(vnet_iter);
2355 		if (V_pfsync_swi_cookie) {
2356 			swi_remove(V_pfsync_swi_cookie);
2357 			if_clone_detach(V_pfsync_cloner);
2358 		}
2359 		CURVNET_RESTORE();
2360 	}
2361 	VNET_LIST_RUNLOCK();
2362 
2363 	return (error);
2364 }
2365 
2366 static void
2367 pfsync_uninit()
2368 {
2369 	VNET_ITERATOR_DECL(vnet_iter);
2370 
2371 	pfsync_pointers_uninit();
2372 
2373 	ipproto_unregister(IPPROTO_PFSYNC);
2374 	pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2375 	VNET_LIST_RLOCK();
2376 	VNET_FOREACH(vnet_iter) {
2377 		CURVNET_SET(vnet_iter);
2378 		if_clone_detach(V_pfsync_cloner);
2379 		swi_remove(V_pfsync_swi_cookie);
2380 		CURVNET_RESTORE();
2381 	}
2382 	VNET_LIST_RUNLOCK();
2383 }
2384 
2385 static int
2386 pfsync_modevent(module_t mod, int type, void *data)
2387 {
2388 	int error = 0;
2389 
2390 	switch (type) {
2391 	case MOD_LOAD:
2392 		error = pfsync_init();
2393 		break;
2394 	case MOD_QUIESCE:
2395 		/*
2396 		 * Module should not be unloaded due to race conditions.
2397 		 */
2398 		error = EBUSY;
2399 		break;
2400 	case MOD_UNLOAD:
2401 		pfsync_uninit();
2402 		break;
2403 	default:
2404 		error = EINVAL;
2405 		break;
2406 	}
2407 
2408 	return (error);
2409 }
2410 
2411 static moduledata_t pfsync_mod = {
2412 	pfsyncname,
2413 	pfsync_modevent,
2414 	0
2415 };
2416 
2417 #define PFSYNC_MODVER 1
2418 
2419 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);
2420 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2421 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
2422