xref: /freebsd/sys/netpfil/pf/if_pfsync.c (revision f0cfa1b168014f56c02b83e5f28412cc5f78d117)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD AND MIT
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/socket.h>
81 #include <sys/sockio.h>
82 #include <sys/sysctl.h>
83 #include <sys/syslog.h>
84 
85 #include <net/bpf.h>
86 #include <net/if.h>
87 #include <net/if_var.h>
88 #include <net/if_clone.h>
89 #include <net/if_types.h>
90 #include <net/vnet.h>
91 #include <net/pfvar.h>
92 #include <net/if_pfsync.h>
93 
94 #include <netinet/if_ether.h>
95 #include <netinet/in.h>
96 #include <netinet/in_var.h>
97 #include <netinet/ip.h>
98 #include <netinet/ip_carp.h>
99 #include <netinet/ip_var.h>
100 #include <netinet/tcp.h>
101 #include <netinet/tcp_fsm.h>
102 #include <netinet/tcp_seq.h>
103 
104 #define PFSYNC_MINPKT ( \
105 	sizeof(struct ip) + \
106 	sizeof(struct pfsync_header) + \
107 	sizeof(struct pfsync_subheader) )
108 
109 struct pfsync_pkt {
110 	struct ip *ip;
111 	struct in_addr src;
112 	u_int8_t flags;
113 };
114 
115 static int	pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *,
116 		    struct pfsync_state_peer *);
117 static int	pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int);
118 static int	pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int);
119 static int	pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int);
120 static int	pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int);
121 static int	pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int);
122 static int	pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int);
123 static int	pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int);
124 static int	pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int);
125 static int	pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int);
126 static int	pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int);
127 static int	pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int);
128 static int	pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int);
129 
130 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = {
131 	pfsync_in_clr,			/* PFSYNC_ACT_CLR */
132 	pfsync_in_ins,			/* PFSYNC_ACT_INS */
133 	pfsync_in_iack,			/* PFSYNC_ACT_INS_ACK */
134 	pfsync_in_upd,			/* PFSYNC_ACT_UPD */
135 	pfsync_in_upd_c,		/* PFSYNC_ACT_UPD_C */
136 	pfsync_in_ureq,			/* PFSYNC_ACT_UPD_REQ */
137 	pfsync_in_del,			/* PFSYNC_ACT_DEL */
138 	pfsync_in_del_c,		/* PFSYNC_ACT_DEL_C */
139 	pfsync_in_error,		/* PFSYNC_ACT_INS_F */
140 	pfsync_in_error,		/* PFSYNC_ACT_DEL_F */
141 	pfsync_in_bus,			/* PFSYNC_ACT_BUS */
142 	pfsync_in_tdb,			/* PFSYNC_ACT_TDB */
143 	pfsync_in_eof			/* PFSYNC_ACT_EOF */
144 };
145 
146 struct pfsync_q {
147 	void		(*write)(struct pf_state *, void *);
148 	size_t		len;
149 	u_int8_t	action;
150 };
151 
152 /* we have one of these for every PFSYNC_S_ */
153 static void	pfsync_out_state(struct pf_state *, void *);
154 static void	pfsync_out_iack(struct pf_state *, void *);
155 static void	pfsync_out_upd_c(struct pf_state *, void *);
156 static void	pfsync_out_del(struct pf_state *, void *);
157 
158 static struct pfsync_q pfsync_qs[] = {
159 	{ pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_INS },
160 	{ pfsync_out_iack,  sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
161 	{ pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_UPD },
162 	{ pfsync_out_upd_c, sizeof(struct pfsync_upd_c),   PFSYNC_ACT_UPD_C },
163 	{ pfsync_out_del,   sizeof(struct pfsync_del_c),   PFSYNC_ACT_DEL_C }
164 };
165 
166 static void	pfsync_q_ins(struct pf_state *, int, bool);
167 static void	pfsync_q_del(struct pf_state *, bool);
168 
169 static void	pfsync_update_state(struct pf_state *);
170 
171 struct pfsync_upd_req_item {
172 	TAILQ_ENTRY(pfsync_upd_req_item)	ur_entry;
173 	struct pfsync_upd_req			ur_msg;
174 };
175 
176 struct pfsync_deferral {
177 	struct pfsync_softc		*pd_sc;
178 	TAILQ_ENTRY(pfsync_deferral)	pd_entry;
179 	u_int				pd_refs;
180 	struct callout			pd_tmo;
181 
182 	struct pf_state			*pd_st;
183 	struct mbuf			*pd_m;
184 };
185 
186 struct pfsync_softc {
187 	/* Configuration */
188 	struct ifnet		*sc_ifp;
189 	struct ifnet		*sc_sync_if;
190 	struct ip_moptions	sc_imo;
191 	struct in_addr		sc_sync_peer;
192 	uint32_t		sc_flags;
193 #define	PFSYNCF_OK		0x00000001
194 #define	PFSYNCF_DEFER		0x00000002
195 #define	PFSYNCF_PUSH		0x00000004
196 	uint8_t			sc_maxupdates;
197 	struct ip		sc_template;
198 	struct callout		sc_tmo;
199 	struct mtx		sc_mtx;
200 
201 	/* Queued data */
202 	size_t			sc_len;
203 	TAILQ_HEAD(, pf_state)			sc_qs[PFSYNC_S_COUNT];
204 	TAILQ_HEAD(, pfsync_upd_req_item)	sc_upd_req_list;
205 	TAILQ_HEAD(, pfsync_deferral)		sc_deferrals;
206 	u_int			sc_deferred;
207 	void			*sc_plus;
208 	size_t			sc_pluslen;
209 
210 	/* Bulk update info */
211 	struct mtx		sc_bulk_mtx;
212 	uint32_t		sc_ureq_sent;
213 	int			sc_bulk_tries;
214 	uint32_t		sc_ureq_received;
215 	int			sc_bulk_hashid;
216 	uint64_t		sc_bulk_stateid;
217 	uint32_t		sc_bulk_creatorid;
218 	struct callout		sc_bulk_tmo;
219 	struct callout		sc_bulkfail_tmo;
220 };
221 
222 #define	PFSYNC_LOCK(sc)		mtx_lock(&(sc)->sc_mtx)
223 #define	PFSYNC_UNLOCK(sc)	mtx_unlock(&(sc)->sc_mtx)
224 #define	PFSYNC_LOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_mtx, MA_OWNED)
225 
226 #define	PFSYNC_BLOCK(sc)	mtx_lock(&(sc)->sc_bulk_mtx)
227 #define	PFSYNC_BUNLOCK(sc)	mtx_unlock(&(sc)->sc_bulk_mtx)
228 #define	PFSYNC_BLOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
229 
230 static const char pfsyncname[] = "pfsync";
231 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
232 static VNET_DEFINE(struct pfsync_softc	*, pfsyncif) = NULL;
233 #define	V_pfsyncif		VNET(pfsyncif)
234 static VNET_DEFINE(void *, pfsync_swi_cookie) = NULL;
235 #define	V_pfsync_swi_cookie	VNET(pfsync_swi_cookie)
236 static VNET_DEFINE(struct pfsyncstats, pfsyncstats);
237 #define	V_pfsyncstats		VNET(pfsyncstats)
238 static VNET_DEFINE(int, pfsync_carp_adj) = CARP_MAXSKEW;
239 #define	V_pfsync_carp_adj	VNET(pfsync_carp_adj)
240 
241 static void	pfsync_timeout(void *);
242 static void	pfsync_push(struct pfsync_softc *);
243 static void	pfsyncintr(void *);
244 static int	pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
245 		    void *);
246 static void	pfsync_multicast_cleanup(struct pfsync_softc *);
247 static void	pfsync_pointers_init(void);
248 static void	pfsync_pointers_uninit(void);
249 static int	pfsync_init(void);
250 static void	pfsync_uninit(void);
251 
252 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC");
253 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW,
254     &VNET_NAME(pfsyncstats), pfsyncstats,
255     "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
256 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW,
257     &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
258 
259 static int	pfsync_clone_create(struct if_clone *, int, caddr_t);
260 static void	pfsync_clone_destroy(struct ifnet *);
261 static int	pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
262 		    struct pf_state_peer *);
263 static int	pfsyncoutput(struct ifnet *, struct mbuf *,
264 		    const struct sockaddr *, struct route *);
265 static int	pfsyncioctl(struct ifnet *, u_long, caddr_t);
266 
267 static int	pfsync_defer(struct pf_state *, struct mbuf *);
268 static void	pfsync_undefer(struct pfsync_deferral *, int);
269 static void	pfsync_undefer_state(struct pf_state *, int);
270 static void	pfsync_defer_tmo(void *);
271 
272 static void	pfsync_request_update(u_int32_t, u_int64_t);
273 static void	pfsync_update_state_req(struct pf_state *);
274 
275 static void	pfsync_drop(struct pfsync_softc *);
276 static void	pfsync_sendout(int);
277 static void	pfsync_send_plus(void *, size_t);
278 
279 static void	pfsync_bulk_start(void);
280 static void	pfsync_bulk_status(u_int8_t);
281 static void	pfsync_bulk_update(void *);
282 static void	pfsync_bulk_fail(void *);
283 
284 #ifdef IPSEC
285 static void	pfsync_update_net_tdb(struct pfsync_tdb *);
286 #endif
287 
288 #define PFSYNC_MAX_BULKTRIES	12
289 
290 VNET_DEFINE(struct if_clone *, pfsync_cloner);
291 #define	V_pfsync_cloner	VNET(pfsync_cloner)
292 
293 static int
294 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
295 {
296 	struct pfsync_softc *sc;
297 	struct ifnet *ifp;
298 	int q;
299 
300 	if (unit != 0)
301 		return (EINVAL);
302 
303 	sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
304 	sc->sc_flags |= PFSYNCF_OK;
305 
306 	for (q = 0; q < PFSYNC_S_COUNT; q++)
307 		TAILQ_INIT(&sc->sc_qs[q]);
308 
309 	TAILQ_INIT(&sc->sc_upd_req_list);
310 	TAILQ_INIT(&sc->sc_deferrals);
311 
312 	sc->sc_len = PFSYNC_MINPKT;
313 	sc->sc_maxupdates = 128;
314 
315 	ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
316 	if (ifp == NULL) {
317 		free(sc, M_PFSYNC);
318 		return (ENOSPC);
319 	}
320 	if_initname(ifp, pfsyncname, unit);
321 	ifp->if_softc = sc;
322 	ifp->if_ioctl = pfsyncioctl;
323 	ifp->if_output = pfsyncoutput;
324 	ifp->if_type = IFT_PFSYNC;
325 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
326 	ifp->if_hdrlen = sizeof(struct pfsync_header);
327 	ifp->if_mtu = ETHERMTU;
328 	mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
329 	mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
330 	callout_init(&sc->sc_tmo, 1);
331 	callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
332 	callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
333 
334 	if_attach(ifp);
335 
336 	bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
337 
338 	V_pfsyncif = sc;
339 
340 	return (0);
341 }
342 
343 static void
344 pfsync_clone_destroy(struct ifnet *ifp)
345 {
346 	struct pfsync_softc *sc = ifp->if_softc;
347 
348 	/*
349 	 * At this stage, everything should have already been
350 	 * cleared by pfsync_uninit(), and we have only to
351 	 * drain callouts.
352 	 */
353 	while (sc->sc_deferred > 0) {
354 		struct pfsync_deferral *pd = TAILQ_FIRST(&sc->sc_deferrals);
355 
356 		TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
357 		sc->sc_deferred--;
358 		if (callout_stop(&pd->pd_tmo) > 0) {
359 			pf_release_state(pd->pd_st);
360 			m_freem(pd->pd_m);
361 			free(pd, M_PFSYNC);
362 		} else {
363 			pd->pd_refs++;
364 			callout_drain(&pd->pd_tmo);
365 			free(pd, M_PFSYNC);
366 		}
367 	}
368 
369 	callout_drain(&sc->sc_tmo);
370 	callout_drain(&sc->sc_bulkfail_tmo);
371 	callout_drain(&sc->sc_bulk_tmo);
372 
373 	if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
374 		(*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
375 	bpfdetach(ifp);
376 	if_detach(ifp);
377 
378 	pfsync_drop(sc);
379 
380 	if_free(ifp);
381 	if (sc->sc_imo.imo_membership)
382 		pfsync_multicast_cleanup(sc);
383 	mtx_destroy(&sc->sc_mtx);
384 	mtx_destroy(&sc->sc_bulk_mtx);
385 	free(sc, M_PFSYNC);
386 
387 	V_pfsyncif = NULL;
388 }
389 
390 static int
391 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
392     struct pf_state_peer *d)
393 {
394 	if (s->scrub.scrub_flag && d->scrub == NULL) {
395 		d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
396 		if (d->scrub == NULL)
397 			return (ENOMEM);
398 	}
399 
400 	return (0);
401 }
402 
403 
404 static int
405 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
406 {
407 	struct pfsync_softc *sc = V_pfsyncif;
408 #ifndef	__NO_STRICT_ALIGNMENT
409 	struct pfsync_state_key key[2];
410 #endif
411 	struct pfsync_state_key *kw, *ks;
412 	struct pf_state	*st = NULL;
413 	struct pf_state_key *skw = NULL, *sks = NULL;
414 	struct pf_rule *r = NULL;
415 	struct pfi_kif	*kif;
416 	int error;
417 
418 	PF_RULES_RASSERT();
419 
420 	if (sp->creatorid == 0) {
421 		if (V_pf_status.debug >= PF_DEBUG_MISC)
422 			printf("%s: invalid creator id: %08x\n", __func__,
423 			    ntohl(sp->creatorid));
424 		return (EINVAL);
425 	}
426 
427 	if ((kif = pfi_kif_find(sp->ifname)) == NULL) {
428 		if (V_pf_status.debug >= PF_DEBUG_MISC)
429 			printf("%s: unknown interface: %s\n", __func__,
430 			    sp->ifname);
431 		if (flags & PFSYNC_SI_IOCTL)
432 			return (EINVAL);
433 		return (0);	/* skip this state */
434 	}
435 
436 	/*
437 	 * If the ruleset checksums match or the state is coming from the ioctl,
438 	 * it's safe to associate the state with the rule of that number.
439 	 */
440 	if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
441 	    (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
442 	    pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
443 		r = pf_main_ruleset.rules[
444 		    PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
445 	else
446 		r = &V_pf_default_rule;
447 
448 	if ((r->max_states &&
449 	    counter_u64_fetch(r->states_cur) >= r->max_states))
450 		goto cleanup;
451 
452 	/*
453 	 * XXXGL: consider M_WAITOK in ioctl path after.
454 	 */
455 	if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL)
456 		goto cleanup;
457 
458 	if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
459 		goto cleanup;
460 
461 #ifndef	__NO_STRICT_ALIGNMENT
462 	bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2);
463 	kw = &key[PF_SK_WIRE];
464 	ks = &key[PF_SK_STACK];
465 #else
466 	kw = &sp->key[PF_SK_WIRE];
467 	ks = &sp->key[PF_SK_STACK];
468 #endif
469 
470 	if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) ||
471 	    PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) ||
472 	    kw->port[0] != ks->port[0] ||
473 	    kw->port[1] != ks->port[1]) {
474 		sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
475 		if (sks == NULL)
476 			goto cleanup;
477 	} else
478 		sks = skw;
479 
480 	/* allocate memory for scrub info */
481 	if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
482 	    pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
483 		goto cleanup;
484 
485 	/* Copy to state key(s). */
486 	skw->addr[0] = kw->addr[0];
487 	skw->addr[1] = kw->addr[1];
488 	skw->port[0] = kw->port[0];
489 	skw->port[1] = kw->port[1];
490 	skw->proto = sp->proto;
491 	skw->af = sp->af;
492 	if (sks != skw) {
493 		sks->addr[0] = ks->addr[0];
494 		sks->addr[1] = ks->addr[1];
495 		sks->port[0] = ks->port[0];
496 		sks->port[1] = ks->port[1];
497 		sks->proto = sp->proto;
498 		sks->af = sp->af;
499 	}
500 
501 	/* copy to state */
502 	bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
503 	st->creation = time_uptime - ntohl(sp->creation);
504 	st->expire = time_uptime;
505 	if (sp->expire) {
506 		uint32_t timeout;
507 
508 		timeout = r->timeout[sp->timeout];
509 		if (!timeout)
510 			timeout = V_pf_default_rule.timeout[sp->timeout];
511 
512 		/* sp->expire may have been adaptively scaled by export. */
513 		st->expire -= timeout - ntohl(sp->expire);
514 	}
515 
516 	st->direction = sp->direction;
517 	st->log = sp->log;
518 	st->timeout = sp->timeout;
519 	st->state_flags = sp->state_flags;
520 
521 	st->id = sp->id;
522 	st->creatorid = sp->creatorid;
523 	pf_state_peer_ntoh(&sp->src, &st->src);
524 	pf_state_peer_ntoh(&sp->dst, &st->dst);
525 
526 	st->rule.ptr = r;
527 	st->nat_rule.ptr = NULL;
528 	st->anchor.ptr = NULL;
529 	st->rt_kif = NULL;
530 
531 	st->pfsync_time = time_uptime;
532 	st->sync_state = PFSYNC_S_NONE;
533 
534 	if (!(flags & PFSYNC_SI_IOCTL))
535 		st->state_flags |= PFSTATE_NOSYNC;
536 
537 	if ((error = pf_state_insert(kif, skw, sks, st)) != 0)
538 		goto cleanup_state;
539 
540 	/* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
541 	counter_u64_add(r->states_cur, 1);
542 	counter_u64_add(r->states_tot, 1);
543 
544 	if (!(flags & PFSYNC_SI_IOCTL)) {
545 		st->state_flags &= ~PFSTATE_NOSYNC;
546 		if (st->state_flags & PFSTATE_ACK) {
547 			pfsync_q_ins(st, PFSYNC_S_IACK, true);
548 			pfsync_push(sc);
549 		}
550 	}
551 	st->state_flags &= ~PFSTATE_ACK;
552 	PF_STATE_UNLOCK(st);
553 
554 	return (0);
555 
556 cleanup:
557 	error = ENOMEM;
558 	if (skw == sks)
559 		sks = NULL;
560 	if (skw != NULL)
561 		uma_zfree(V_pf_state_key_z, skw);
562 	if (sks != NULL)
563 		uma_zfree(V_pf_state_key_z, sks);
564 
565 cleanup_state:	/* pf_state_insert() frees the state keys. */
566 	if (st) {
567 		if (st->dst.scrub)
568 			uma_zfree(V_pf_state_scrub_z, st->dst.scrub);
569 		if (st->src.scrub)
570 			uma_zfree(V_pf_state_scrub_z, st->src.scrub);
571 		uma_zfree(V_pf_state_z, st);
572 	}
573 	return (error);
574 }
575 
576 static int
577 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
578 {
579 	struct pfsync_softc *sc = V_pfsyncif;
580 	struct pfsync_pkt pkt;
581 	struct mbuf *m = *mp;
582 	struct ip *ip = mtod(m, struct ip *);
583 	struct pfsync_header *ph;
584 	struct pfsync_subheader subh;
585 
586 	int offset, len;
587 	int rv;
588 	uint16_t count;
589 
590 	*mp = NULL;
591 	V_pfsyncstats.pfsyncs_ipackets++;
592 
593 	/* Verify that we have a sync interface configured. */
594 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
595 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
596 		goto done;
597 
598 	/* verify that the packet came in on the right interface */
599 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
600 		V_pfsyncstats.pfsyncs_badif++;
601 		goto done;
602 	}
603 
604 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
605 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
606 	/* verify that the IP TTL is 255. */
607 	if (ip->ip_ttl != PFSYNC_DFLTTL) {
608 		V_pfsyncstats.pfsyncs_badttl++;
609 		goto done;
610 	}
611 
612 	offset = ip->ip_hl << 2;
613 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
614 		V_pfsyncstats.pfsyncs_hdrops++;
615 		goto done;
616 	}
617 
618 	if (offset + sizeof(*ph) > m->m_len) {
619 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
620 			V_pfsyncstats.pfsyncs_hdrops++;
621 			return (IPPROTO_DONE);
622 		}
623 		ip = mtod(m, struct ip *);
624 	}
625 	ph = (struct pfsync_header *)((char *)ip + offset);
626 
627 	/* verify the version */
628 	if (ph->version != PFSYNC_VERSION) {
629 		V_pfsyncstats.pfsyncs_badver++;
630 		goto done;
631 	}
632 
633 	len = ntohs(ph->len) + offset;
634 	if (m->m_pkthdr.len < len) {
635 		V_pfsyncstats.pfsyncs_badlen++;
636 		goto done;
637 	}
638 
639 	/* Cheaper to grab this now than having to mess with mbufs later */
640 	pkt.ip = ip;
641 	pkt.src = ip->ip_src;
642 	pkt.flags = 0;
643 
644 	/*
645 	 * Trusting pf_chksum during packet processing, as well as seeking
646 	 * in interface name tree, require holding PF_RULES_RLOCK().
647 	 */
648 	PF_RULES_RLOCK();
649 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
650 		pkt.flags |= PFSYNC_SI_CKSUM;
651 
652 	offset += sizeof(*ph);
653 	while (offset <= len - sizeof(subh)) {
654 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
655 		offset += sizeof(subh);
656 
657 		if (subh.action >= PFSYNC_ACT_MAX) {
658 			V_pfsyncstats.pfsyncs_badact++;
659 			PF_RULES_RUNLOCK();
660 			goto done;
661 		}
662 
663 		count = ntohs(subh.count);
664 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
665 		rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count);
666 		if (rv == -1) {
667 			PF_RULES_RUNLOCK();
668 			return (IPPROTO_DONE);
669 		}
670 
671 		offset += rv;
672 	}
673 	PF_RULES_RUNLOCK();
674 
675 done:
676 	m_freem(m);
677 	return (IPPROTO_DONE);
678 }
679 
680 static int
681 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
682 {
683 	struct pfsync_clr *clr;
684 	struct mbuf *mp;
685 	int len = sizeof(*clr) * count;
686 	int i, offp;
687 	u_int32_t creatorid;
688 
689 	mp = m_pulldown(m, offset, len, &offp);
690 	if (mp == NULL) {
691 		V_pfsyncstats.pfsyncs_badlen++;
692 		return (-1);
693 	}
694 	clr = (struct pfsync_clr *)(mp->m_data + offp);
695 
696 	for (i = 0; i < count; i++) {
697 		creatorid = clr[i].creatorid;
698 
699 		if (clr[i].ifname[0] != '\0' &&
700 		    pfi_kif_find(clr[i].ifname) == NULL)
701 			continue;
702 
703 		for (int i = 0; i <= pf_hashmask; i++) {
704 			struct pf_idhash *ih = &V_pf_idhash[i];
705 			struct pf_state *s;
706 relock:
707 			PF_HASHROW_LOCK(ih);
708 			LIST_FOREACH(s, &ih->states, entry) {
709 				if (s->creatorid == creatorid) {
710 					s->state_flags |= PFSTATE_NOSYNC;
711 					pf_unlink_state(s, PF_ENTER_LOCKED);
712 					goto relock;
713 				}
714 			}
715 			PF_HASHROW_UNLOCK(ih);
716 		}
717 	}
718 
719 	return (len);
720 }
721 
722 static int
723 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
724 {
725 	struct mbuf *mp;
726 	struct pfsync_state *sa, *sp;
727 	int len = sizeof(*sp) * count;
728 	int i, offp;
729 
730 	mp = m_pulldown(m, offset, len, &offp);
731 	if (mp == NULL) {
732 		V_pfsyncstats.pfsyncs_badlen++;
733 		return (-1);
734 	}
735 	sa = (struct pfsync_state *)(mp->m_data + offp);
736 
737 	for (i = 0; i < count; i++) {
738 		sp = &sa[i];
739 
740 		/* Check for invalid values. */
741 		if (sp->timeout >= PFTM_MAX ||
742 		    sp->src.state > PF_TCPS_PROXY_DST ||
743 		    sp->dst.state > PF_TCPS_PROXY_DST ||
744 		    sp->direction > PF_OUT ||
745 		    (sp->af != AF_INET && sp->af != AF_INET6)) {
746 			if (V_pf_status.debug >= PF_DEBUG_MISC)
747 				printf("%s: invalid value\n", __func__);
748 			V_pfsyncstats.pfsyncs_badval++;
749 			continue;
750 		}
751 
752 		if (pfsync_state_import(sp, pkt->flags) == ENOMEM)
753 			/* Drop out, but process the rest of the actions. */
754 			break;
755 	}
756 
757 	return (len);
758 }
759 
760 static int
761 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
762 {
763 	struct pfsync_ins_ack *ia, *iaa;
764 	struct pf_state *st;
765 
766 	struct mbuf *mp;
767 	int len = count * sizeof(*ia);
768 	int offp, i;
769 
770 	mp = m_pulldown(m, offset, len, &offp);
771 	if (mp == NULL) {
772 		V_pfsyncstats.pfsyncs_badlen++;
773 		return (-1);
774 	}
775 	iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
776 
777 	for (i = 0; i < count; i++) {
778 		ia = &iaa[i];
779 
780 		st = pf_find_state_byid(ia->id, ia->creatorid);
781 		if (st == NULL)
782 			continue;
783 
784 		if (st->state_flags & PFSTATE_ACK) {
785 			PFSYNC_LOCK(V_pfsyncif);
786 			pfsync_undefer_state(st, 0);
787 			PFSYNC_UNLOCK(V_pfsyncif);
788 		}
789 		PF_STATE_UNLOCK(st);
790 	}
791 	/*
792 	 * XXX this is not yet implemented, but we know the size of the
793 	 * message so we can skip it.
794 	 */
795 
796 	return (count * sizeof(struct pfsync_ins_ack));
797 }
798 
799 static int
800 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src,
801     struct pfsync_state_peer *dst)
802 {
803 	int sync = 0;
804 
805 	PF_STATE_LOCK_ASSERT(st);
806 
807 	/*
808 	 * The state should never go backwards except
809 	 * for syn-proxy states.  Neither should the
810 	 * sequence window slide backwards.
811 	 */
812 	if ((st->src.state > src->state &&
813 	    (st->src.state < PF_TCPS_PROXY_SRC ||
814 	    src->state >= PF_TCPS_PROXY_SRC)) ||
815 
816 	    (st->src.state == src->state &&
817 	    SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
818 		sync++;
819 	else
820 		pf_state_peer_ntoh(src, &st->src);
821 
822 	if ((st->dst.state > dst->state) ||
823 
824 	    (st->dst.state >= TCPS_SYN_SENT &&
825 	    SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
826 		sync++;
827 	else
828 		pf_state_peer_ntoh(dst, &st->dst);
829 
830 	return (sync);
831 }
832 
833 static int
834 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
835 {
836 	struct pfsync_softc *sc = V_pfsyncif;
837 	struct pfsync_state *sa, *sp;
838 	struct pf_state *st;
839 	int sync;
840 
841 	struct mbuf *mp;
842 	int len = count * sizeof(*sp);
843 	int offp, i;
844 
845 	mp = m_pulldown(m, offset, len, &offp);
846 	if (mp == NULL) {
847 		V_pfsyncstats.pfsyncs_badlen++;
848 		return (-1);
849 	}
850 	sa = (struct pfsync_state *)(mp->m_data + offp);
851 
852 	for (i = 0; i < count; i++) {
853 		sp = &sa[i];
854 
855 		/* check for invalid values */
856 		if (sp->timeout >= PFTM_MAX ||
857 		    sp->src.state > PF_TCPS_PROXY_DST ||
858 		    sp->dst.state > PF_TCPS_PROXY_DST) {
859 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
860 				printf("pfsync_input: PFSYNC_ACT_UPD: "
861 				    "invalid value\n");
862 			}
863 			V_pfsyncstats.pfsyncs_badval++;
864 			continue;
865 		}
866 
867 		st = pf_find_state_byid(sp->id, sp->creatorid);
868 		if (st == NULL) {
869 			/* insert the update */
870 			if (pfsync_state_import(sp, 0))
871 				V_pfsyncstats.pfsyncs_badstate++;
872 			continue;
873 		}
874 
875 		if (st->state_flags & PFSTATE_ACK) {
876 			PFSYNC_LOCK(sc);
877 			pfsync_undefer_state(st, 1);
878 			PFSYNC_UNLOCK(sc);
879 		}
880 
881 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
882 			sync = pfsync_upd_tcp(st, &sp->src, &sp->dst);
883 		else {
884 			sync = 0;
885 
886 			/*
887 			 * Non-TCP protocol state machine always go
888 			 * forwards
889 			 */
890 			if (st->src.state > sp->src.state)
891 				sync++;
892 			else
893 				pf_state_peer_ntoh(&sp->src, &st->src);
894 			if (st->dst.state > sp->dst.state)
895 				sync++;
896 			else
897 				pf_state_peer_ntoh(&sp->dst, &st->dst);
898 		}
899 		if (sync < 2) {
900 			pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
901 			pf_state_peer_ntoh(&sp->dst, &st->dst);
902 			st->expire = time_uptime;
903 			st->timeout = sp->timeout;
904 		}
905 		st->pfsync_time = time_uptime;
906 
907 		if (sync) {
908 			V_pfsyncstats.pfsyncs_stale++;
909 
910 			pfsync_update_state(st);
911 			PF_STATE_UNLOCK(st);
912 			PFSYNC_LOCK(sc);
913 			pfsync_push(sc);
914 			PFSYNC_UNLOCK(sc);
915 			continue;
916 		}
917 		PF_STATE_UNLOCK(st);
918 	}
919 
920 	return (len);
921 }
922 
923 static int
924 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
925 {
926 	struct pfsync_softc *sc = V_pfsyncif;
927 	struct pfsync_upd_c *ua, *up;
928 	struct pf_state *st;
929 	int len = count * sizeof(*up);
930 	int sync;
931 	struct mbuf *mp;
932 	int offp, i;
933 
934 	mp = m_pulldown(m, offset, len, &offp);
935 	if (mp == NULL) {
936 		V_pfsyncstats.pfsyncs_badlen++;
937 		return (-1);
938 	}
939 	ua = (struct pfsync_upd_c *)(mp->m_data + offp);
940 
941 	for (i = 0; i < count; i++) {
942 		up = &ua[i];
943 
944 		/* check for invalid values */
945 		if (up->timeout >= PFTM_MAX ||
946 		    up->src.state > PF_TCPS_PROXY_DST ||
947 		    up->dst.state > PF_TCPS_PROXY_DST) {
948 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
949 				printf("pfsync_input: "
950 				    "PFSYNC_ACT_UPD_C: "
951 				    "invalid value\n");
952 			}
953 			V_pfsyncstats.pfsyncs_badval++;
954 			continue;
955 		}
956 
957 		st = pf_find_state_byid(up->id, up->creatorid);
958 		if (st == NULL) {
959 			/* We don't have this state. Ask for it. */
960 			PFSYNC_LOCK(sc);
961 			pfsync_request_update(up->creatorid, up->id);
962 			PFSYNC_UNLOCK(sc);
963 			continue;
964 		}
965 
966 		if (st->state_flags & PFSTATE_ACK) {
967 			PFSYNC_LOCK(sc);
968 			pfsync_undefer_state(st, 1);
969 			PFSYNC_UNLOCK(sc);
970 		}
971 
972 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
973 			sync = pfsync_upd_tcp(st, &up->src, &up->dst);
974 		else {
975 			sync = 0;
976 
977 			/*
978 			 * Non-TCP protocol state machine always go
979 			 * forwards
980 			 */
981 			if (st->src.state > up->src.state)
982 				sync++;
983 			else
984 				pf_state_peer_ntoh(&up->src, &st->src);
985 			if (st->dst.state > up->dst.state)
986 				sync++;
987 			else
988 				pf_state_peer_ntoh(&up->dst, &st->dst);
989 		}
990 		if (sync < 2) {
991 			pfsync_alloc_scrub_memory(&up->dst, &st->dst);
992 			pf_state_peer_ntoh(&up->dst, &st->dst);
993 			st->expire = time_uptime;
994 			st->timeout = up->timeout;
995 		}
996 		st->pfsync_time = time_uptime;
997 
998 		if (sync) {
999 			V_pfsyncstats.pfsyncs_stale++;
1000 
1001 			pfsync_update_state(st);
1002 			PF_STATE_UNLOCK(st);
1003 			PFSYNC_LOCK(sc);
1004 			pfsync_push(sc);
1005 			PFSYNC_UNLOCK(sc);
1006 			continue;
1007 		}
1008 		PF_STATE_UNLOCK(st);
1009 	}
1010 
1011 	return (len);
1012 }
1013 
1014 static int
1015 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1016 {
1017 	struct pfsync_upd_req *ur, *ura;
1018 	struct mbuf *mp;
1019 	int len = count * sizeof(*ur);
1020 	int i, offp;
1021 
1022 	struct pf_state *st;
1023 
1024 	mp = m_pulldown(m, offset, len, &offp);
1025 	if (mp == NULL) {
1026 		V_pfsyncstats.pfsyncs_badlen++;
1027 		return (-1);
1028 	}
1029 	ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1030 
1031 	for (i = 0; i < count; i++) {
1032 		ur = &ura[i];
1033 
1034 		if (ur->id == 0 && ur->creatorid == 0)
1035 			pfsync_bulk_start();
1036 		else {
1037 			st = pf_find_state_byid(ur->id, ur->creatorid);
1038 			if (st == NULL) {
1039 				V_pfsyncstats.pfsyncs_badstate++;
1040 				continue;
1041 			}
1042 			if (st->state_flags & PFSTATE_NOSYNC) {
1043 				PF_STATE_UNLOCK(st);
1044 				continue;
1045 			}
1046 
1047 			pfsync_update_state_req(st);
1048 			PF_STATE_UNLOCK(st);
1049 		}
1050 	}
1051 
1052 	return (len);
1053 }
1054 
1055 static int
1056 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1057 {
1058 	struct mbuf *mp;
1059 	struct pfsync_state *sa, *sp;
1060 	struct pf_state *st;
1061 	int len = count * sizeof(*sp);
1062 	int offp, i;
1063 
1064 	mp = m_pulldown(m, offset, len, &offp);
1065 	if (mp == NULL) {
1066 		V_pfsyncstats.pfsyncs_badlen++;
1067 		return (-1);
1068 	}
1069 	sa = (struct pfsync_state *)(mp->m_data + offp);
1070 
1071 	for (i = 0; i < count; i++) {
1072 		sp = &sa[i];
1073 
1074 		st = pf_find_state_byid(sp->id, sp->creatorid);
1075 		if (st == NULL) {
1076 			V_pfsyncstats.pfsyncs_badstate++;
1077 			continue;
1078 		}
1079 		st->state_flags |= PFSTATE_NOSYNC;
1080 		pf_unlink_state(st, PF_ENTER_LOCKED);
1081 	}
1082 
1083 	return (len);
1084 }
1085 
1086 static int
1087 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1088 {
1089 	struct mbuf *mp;
1090 	struct pfsync_del_c *sa, *sp;
1091 	struct pf_state *st;
1092 	int len = count * sizeof(*sp);
1093 	int offp, i;
1094 
1095 	mp = m_pulldown(m, offset, len, &offp);
1096 	if (mp == NULL) {
1097 		V_pfsyncstats.pfsyncs_badlen++;
1098 		return (-1);
1099 	}
1100 	sa = (struct pfsync_del_c *)(mp->m_data + offp);
1101 
1102 	for (i = 0; i < count; i++) {
1103 		sp = &sa[i];
1104 
1105 		st = pf_find_state_byid(sp->id, sp->creatorid);
1106 		if (st == NULL) {
1107 			V_pfsyncstats.pfsyncs_badstate++;
1108 			continue;
1109 		}
1110 
1111 		st->state_flags |= PFSTATE_NOSYNC;
1112 		pf_unlink_state(st, PF_ENTER_LOCKED);
1113 	}
1114 
1115 	return (len);
1116 }
1117 
1118 static int
1119 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1120 {
1121 	struct pfsync_softc *sc = V_pfsyncif;
1122 	struct pfsync_bus *bus;
1123 	struct mbuf *mp;
1124 	int len = count * sizeof(*bus);
1125 	int offp;
1126 
1127 	PFSYNC_BLOCK(sc);
1128 
1129 	/* If we're not waiting for a bulk update, who cares. */
1130 	if (sc->sc_ureq_sent == 0) {
1131 		PFSYNC_BUNLOCK(sc);
1132 		return (len);
1133 	}
1134 
1135 	mp = m_pulldown(m, offset, len, &offp);
1136 	if (mp == NULL) {
1137 		PFSYNC_BUNLOCK(sc);
1138 		V_pfsyncstats.pfsyncs_badlen++;
1139 		return (-1);
1140 	}
1141 	bus = (struct pfsync_bus *)(mp->m_data + offp);
1142 
1143 	switch (bus->status) {
1144 	case PFSYNC_BUS_START:
1145 		callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1146 		    V_pf_limits[PF_LIMIT_STATES].limit /
1147 		    ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1148 		    sizeof(struct pfsync_state)),
1149 		    pfsync_bulk_fail, sc);
1150 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1151 			printf("pfsync: received bulk update start\n");
1152 		break;
1153 
1154 	case PFSYNC_BUS_END:
1155 		if (time_uptime - ntohl(bus->endtime) >=
1156 		    sc->sc_ureq_sent) {
1157 			/* that's it, we're happy */
1158 			sc->sc_ureq_sent = 0;
1159 			sc->sc_bulk_tries = 0;
1160 			callout_stop(&sc->sc_bulkfail_tmo);
1161 			if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1162 				(*carp_demote_adj_p)(-V_pfsync_carp_adj,
1163 				    "pfsync bulk done");
1164 			sc->sc_flags |= PFSYNCF_OK;
1165 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1166 				printf("pfsync: received valid "
1167 				    "bulk update end\n");
1168 		} else {
1169 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1170 				printf("pfsync: received invalid "
1171 				    "bulk update end: bad timestamp\n");
1172 		}
1173 		break;
1174 	}
1175 	PFSYNC_BUNLOCK(sc);
1176 
1177 	return (len);
1178 }
1179 
1180 static int
1181 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1182 {
1183 	int len = count * sizeof(struct pfsync_tdb);
1184 
1185 #if defined(IPSEC)
1186 	struct pfsync_tdb *tp;
1187 	struct mbuf *mp;
1188 	int offp;
1189 	int i;
1190 	int s;
1191 
1192 	mp = m_pulldown(m, offset, len, &offp);
1193 	if (mp == NULL) {
1194 		V_pfsyncstats.pfsyncs_badlen++;
1195 		return (-1);
1196 	}
1197 	tp = (struct pfsync_tdb *)(mp->m_data + offp);
1198 
1199 	for (i = 0; i < count; i++)
1200 		pfsync_update_net_tdb(&tp[i]);
1201 #endif
1202 
1203 	return (len);
1204 }
1205 
1206 #if defined(IPSEC)
1207 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1208 static void
1209 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1210 {
1211 	struct tdb		*tdb;
1212 	int			 s;
1213 
1214 	/* check for invalid values */
1215 	if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1216 	    (pt->dst.sa.sa_family != AF_INET &&
1217 	    pt->dst.sa.sa_family != AF_INET6))
1218 		goto bad;
1219 
1220 	tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1221 	if (tdb) {
1222 		pt->rpl = ntohl(pt->rpl);
1223 		pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1224 
1225 		/* Neither replay nor byte counter should ever decrease. */
1226 		if (pt->rpl < tdb->tdb_rpl ||
1227 		    pt->cur_bytes < tdb->tdb_cur_bytes) {
1228 			goto bad;
1229 		}
1230 
1231 		tdb->tdb_rpl = pt->rpl;
1232 		tdb->tdb_cur_bytes = pt->cur_bytes;
1233 	}
1234 	return;
1235 
1236 bad:
1237 	if (V_pf_status.debug >= PF_DEBUG_MISC)
1238 		printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1239 		    "invalid value\n");
1240 	V_pfsyncstats.pfsyncs_badstate++;
1241 	return;
1242 }
1243 #endif
1244 
1245 
1246 static int
1247 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1248 {
1249 	/* check if we are at the right place in the packet */
1250 	if (offset != m->m_pkthdr.len)
1251 		V_pfsyncstats.pfsyncs_badlen++;
1252 
1253 	/* we're done. free and let the caller return */
1254 	m_freem(m);
1255 	return (-1);
1256 }
1257 
1258 static int
1259 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1260 {
1261 	V_pfsyncstats.pfsyncs_badact++;
1262 
1263 	m_freem(m);
1264 	return (-1);
1265 }
1266 
1267 static int
1268 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1269 	struct route *rt)
1270 {
1271 	m_freem(m);
1272 	return (0);
1273 }
1274 
1275 /* ARGSUSED */
1276 static int
1277 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1278 {
1279 	struct pfsync_softc *sc = ifp->if_softc;
1280 	struct ifreq *ifr = (struct ifreq *)data;
1281 	struct pfsyncreq pfsyncr;
1282 	int error;
1283 
1284 	switch (cmd) {
1285 	case SIOCSIFFLAGS:
1286 		PFSYNC_LOCK(sc);
1287 		if (ifp->if_flags & IFF_UP) {
1288 			ifp->if_drv_flags |= IFF_DRV_RUNNING;
1289 			PFSYNC_UNLOCK(sc);
1290 			pfsync_pointers_init();
1291 		} else {
1292 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1293 			PFSYNC_UNLOCK(sc);
1294 			pfsync_pointers_uninit();
1295 		}
1296 		break;
1297 	case SIOCSIFMTU:
1298 		if (!sc->sc_sync_if ||
1299 		    ifr->ifr_mtu <= PFSYNC_MINPKT ||
1300 		    ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1301 			return (EINVAL);
1302 		if (ifr->ifr_mtu < ifp->if_mtu) {
1303 			PFSYNC_LOCK(sc);
1304 			if (sc->sc_len > PFSYNC_MINPKT)
1305 				pfsync_sendout(1);
1306 			PFSYNC_UNLOCK(sc);
1307 		}
1308 		ifp->if_mtu = ifr->ifr_mtu;
1309 		break;
1310 	case SIOCGETPFSYNC:
1311 		bzero(&pfsyncr, sizeof(pfsyncr));
1312 		PFSYNC_LOCK(sc);
1313 		if (sc->sc_sync_if) {
1314 			strlcpy(pfsyncr.pfsyncr_syncdev,
1315 			    sc->sc_sync_if->if_xname, IFNAMSIZ);
1316 		}
1317 		pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1318 		pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1319 		pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER ==
1320 		    (sc->sc_flags & PFSYNCF_DEFER));
1321 		PFSYNC_UNLOCK(sc);
1322 		return (copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)));
1323 
1324 	case SIOCSETPFSYNC:
1325 	    {
1326 		struct ip_moptions *imo = &sc->sc_imo;
1327 		struct ifnet *sifp;
1328 		struct ip *ip;
1329 		void *mship = NULL;
1330 
1331 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1332 			return (error);
1333 		if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr))))
1334 			return (error);
1335 
1336 		if (pfsyncr.pfsyncr_maxupdates > 255)
1337 			return (EINVAL);
1338 
1339 		if (pfsyncr.pfsyncr_syncdev[0] == 0)
1340 			sifp = NULL;
1341 		else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL)
1342 			return (EINVAL);
1343 
1344 		if (sifp != NULL && (
1345 		    pfsyncr.pfsyncr_syncpeer.s_addr == 0 ||
1346 		    pfsyncr.pfsyncr_syncpeer.s_addr ==
1347 		    htonl(INADDR_PFSYNC_GROUP)))
1348 			mship = malloc((sizeof(struct in_multi *) *
1349 			    IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO);
1350 
1351 		PFSYNC_LOCK(sc);
1352 		if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1353 			sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP);
1354 		else
1355 			sc->sc_sync_peer.s_addr =
1356 			    pfsyncr.pfsyncr_syncpeer.s_addr;
1357 
1358 		sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1359 		if (pfsyncr.pfsyncr_defer) {
1360 			sc->sc_flags |= PFSYNCF_DEFER;
1361 			pfsync_defer_ptr = pfsync_defer;
1362 		} else {
1363 			sc->sc_flags &= ~PFSYNCF_DEFER;
1364 			pfsync_defer_ptr = NULL;
1365 		}
1366 
1367 		if (sifp == NULL) {
1368 			if (sc->sc_sync_if)
1369 				if_rele(sc->sc_sync_if);
1370 			sc->sc_sync_if = NULL;
1371 			if (imo->imo_membership)
1372 				pfsync_multicast_cleanup(sc);
1373 			PFSYNC_UNLOCK(sc);
1374 			break;
1375 		}
1376 
1377 		if (sc->sc_len > PFSYNC_MINPKT &&
1378 		    (sifp->if_mtu < sc->sc_ifp->if_mtu ||
1379 		    (sc->sc_sync_if != NULL &&
1380 		    sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
1381 		    sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
1382 			pfsync_sendout(1);
1383 
1384 		if (imo->imo_membership)
1385 			pfsync_multicast_cleanup(sc);
1386 
1387 		if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
1388 			error = pfsync_multicast_setup(sc, sifp, mship);
1389 			if (error) {
1390 				if_rele(sifp);
1391 				free(mship, M_PFSYNC);
1392 				return (error);
1393 			}
1394 		}
1395 		if (sc->sc_sync_if)
1396 			if_rele(sc->sc_sync_if);
1397 		sc->sc_sync_if = sifp;
1398 
1399 		ip = &sc->sc_template;
1400 		bzero(ip, sizeof(*ip));
1401 		ip->ip_v = IPVERSION;
1402 		ip->ip_hl = sizeof(sc->sc_template) >> 2;
1403 		ip->ip_tos = IPTOS_LOWDELAY;
1404 		/* len and id are set later. */
1405 		ip->ip_off = htons(IP_DF);
1406 		ip->ip_ttl = PFSYNC_DFLTTL;
1407 		ip->ip_p = IPPROTO_PFSYNC;
1408 		ip->ip_src.s_addr = INADDR_ANY;
1409 		ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr;
1410 
1411 		/* Request a full state table update. */
1412 		if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1413 			(*carp_demote_adj_p)(V_pfsync_carp_adj,
1414 			    "pfsync bulk start");
1415 		sc->sc_flags &= ~PFSYNCF_OK;
1416 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1417 			printf("pfsync: requesting bulk update\n");
1418 		pfsync_request_update(0, 0);
1419 		PFSYNC_UNLOCK(sc);
1420 		PFSYNC_BLOCK(sc);
1421 		sc->sc_ureq_sent = time_uptime;
1422 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail,
1423 		    sc);
1424 		PFSYNC_BUNLOCK(sc);
1425 
1426 		break;
1427 	    }
1428 	default:
1429 		return (ENOTTY);
1430 	}
1431 
1432 	return (0);
1433 }
1434 
1435 static void
1436 pfsync_out_state(struct pf_state *st, void *buf)
1437 {
1438 	struct pfsync_state *sp = buf;
1439 
1440 	pfsync_state_export(sp, st);
1441 }
1442 
1443 static void
1444 pfsync_out_iack(struct pf_state *st, void *buf)
1445 {
1446 	struct pfsync_ins_ack *iack = buf;
1447 
1448 	iack->id = st->id;
1449 	iack->creatorid = st->creatorid;
1450 }
1451 
1452 static void
1453 pfsync_out_upd_c(struct pf_state *st, void *buf)
1454 {
1455 	struct pfsync_upd_c *up = buf;
1456 
1457 	bzero(up, sizeof(*up));
1458 	up->id = st->id;
1459 	pf_state_peer_hton(&st->src, &up->src);
1460 	pf_state_peer_hton(&st->dst, &up->dst);
1461 	up->creatorid = st->creatorid;
1462 	up->timeout = st->timeout;
1463 }
1464 
1465 static void
1466 pfsync_out_del(struct pf_state *st, void *buf)
1467 {
1468 	struct pfsync_del_c *dp = buf;
1469 
1470 	dp->id = st->id;
1471 	dp->creatorid = st->creatorid;
1472 	st->state_flags |= PFSTATE_NOSYNC;
1473 }
1474 
1475 static void
1476 pfsync_drop(struct pfsync_softc *sc)
1477 {
1478 	struct pf_state *st, *next;
1479 	struct pfsync_upd_req_item *ur;
1480 	int q;
1481 
1482 	for (q = 0; q < PFSYNC_S_COUNT; q++) {
1483 		if (TAILQ_EMPTY(&sc->sc_qs[q]))
1484 			continue;
1485 
1486 		TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, next) {
1487 			KASSERT(st->sync_state == q,
1488 				("%s: st->sync_state == q",
1489 					__func__));
1490 			st->sync_state = PFSYNC_S_NONE;
1491 			pf_release_state(st);
1492 		}
1493 		TAILQ_INIT(&sc->sc_qs[q]);
1494 	}
1495 
1496 	while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1497 		TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1498 		free(ur, M_PFSYNC);
1499 	}
1500 
1501 	sc->sc_plus = NULL;
1502 	sc->sc_len = PFSYNC_MINPKT;
1503 }
1504 
1505 static void
1506 pfsync_sendout(int schedswi)
1507 {
1508 	struct pfsync_softc *sc = V_pfsyncif;
1509 	struct ifnet *ifp = sc->sc_ifp;
1510 	struct mbuf *m;
1511 	struct ip *ip;
1512 	struct pfsync_header *ph;
1513 	struct pfsync_subheader *subh;
1514 	struct pf_state *st, *st_next;
1515 	struct pfsync_upd_req_item *ur;
1516 	int offset;
1517 	int q, count = 0;
1518 
1519 	KASSERT(sc != NULL, ("%s: null sc", __func__));
1520 	KASSERT(sc->sc_len > PFSYNC_MINPKT,
1521 	    ("%s: sc_len %zu", __func__, sc->sc_len));
1522 	PFSYNC_LOCK_ASSERT(sc);
1523 
1524 	if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1525 		pfsync_drop(sc);
1526 		return;
1527 	}
1528 
1529 	m = m_get2(max_linkhdr + sc->sc_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1530 	if (m == NULL) {
1531 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1532 		V_pfsyncstats.pfsyncs_onomem++;
1533 		return;
1534 	}
1535 	m->m_data += max_linkhdr;
1536 	m->m_len = m->m_pkthdr.len = sc->sc_len;
1537 
1538 	/* build the ip header */
1539 	ip = (struct ip *)m->m_data;
1540 	bcopy(&sc->sc_template, ip, sizeof(*ip));
1541 	offset = sizeof(*ip);
1542 
1543 	ip->ip_len = htons(m->m_pkthdr.len);
1544 	ip_fillid(ip);
1545 
1546 	/* build the pfsync header */
1547 	ph = (struct pfsync_header *)(m->m_data + offset);
1548 	bzero(ph, sizeof(*ph));
1549 	offset += sizeof(*ph);
1550 
1551 	ph->version = PFSYNC_VERSION;
1552 	ph->len = htons(sc->sc_len - sizeof(*ip));
1553 	bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1554 
1555 	/* walk the queues */
1556 	for (q = 0; q < PFSYNC_S_COUNT; q++) {
1557 		if (TAILQ_EMPTY(&sc->sc_qs[q]))
1558 			continue;
1559 
1560 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1561 		offset += sizeof(*subh);
1562 
1563 		count = 0;
1564 		TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, st_next) {
1565 			KASSERT(st->sync_state == q,
1566 				("%s: st->sync_state == q",
1567 					__func__));
1568 			/*
1569 			 * XXXGL: some of write methods do unlocked reads
1570 			 * of state data :(
1571 			 */
1572 			pfsync_qs[q].write(st, m->m_data + offset);
1573 			offset += pfsync_qs[q].len;
1574 			st->sync_state = PFSYNC_S_NONE;
1575 			pf_release_state(st);
1576 			count++;
1577 		}
1578 		TAILQ_INIT(&sc->sc_qs[q]);
1579 
1580 		bzero(subh, sizeof(*subh));
1581 		subh->action = pfsync_qs[q].action;
1582 		subh->count = htons(count);
1583 		V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1584 	}
1585 
1586 	if (!TAILQ_EMPTY(&sc->sc_upd_req_list)) {
1587 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1588 		offset += sizeof(*subh);
1589 
1590 		count = 0;
1591 		while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1592 			TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1593 
1594 			bcopy(&ur->ur_msg, m->m_data + offset,
1595 			    sizeof(ur->ur_msg));
1596 			offset += sizeof(ur->ur_msg);
1597 			free(ur, M_PFSYNC);
1598 			count++;
1599 		}
1600 
1601 		bzero(subh, sizeof(*subh));
1602 		subh->action = PFSYNC_ACT_UPD_REQ;
1603 		subh->count = htons(count);
1604 		V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1605 	}
1606 
1607 	/* has someone built a custom region for us to add? */
1608 	if (sc->sc_plus != NULL) {
1609 		bcopy(sc->sc_plus, m->m_data + offset, sc->sc_pluslen);
1610 		offset += sc->sc_pluslen;
1611 
1612 		sc->sc_plus = NULL;
1613 	}
1614 
1615 	subh = (struct pfsync_subheader *)(m->m_data + offset);
1616 	offset += sizeof(*subh);
1617 
1618 	bzero(subh, sizeof(*subh));
1619 	subh->action = PFSYNC_ACT_EOF;
1620 	subh->count = htons(1);
1621 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1622 
1623 	/* we're done, let's put it on the wire */
1624 	if (ifp->if_bpf) {
1625 		m->m_data += sizeof(*ip);
1626 		m->m_len = m->m_pkthdr.len = sc->sc_len - sizeof(*ip);
1627 		BPF_MTAP(ifp, m);
1628 		m->m_data -= sizeof(*ip);
1629 		m->m_len = m->m_pkthdr.len = sc->sc_len;
1630 	}
1631 
1632 	if (sc->sc_sync_if == NULL) {
1633 		sc->sc_len = PFSYNC_MINPKT;
1634 		m_freem(m);
1635 		return;
1636 	}
1637 
1638 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1639 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1640 	sc->sc_len = PFSYNC_MINPKT;
1641 
1642 	if (!_IF_QFULL(&sc->sc_ifp->if_snd))
1643 		_IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1644 	else {
1645 		m_freem(m);
1646 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1647 	}
1648 	if (schedswi)
1649 		swi_sched(V_pfsync_swi_cookie, 0);
1650 }
1651 
1652 static void
1653 pfsync_insert_state(struct pf_state *st)
1654 {
1655 	struct pfsync_softc *sc = V_pfsyncif;
1656 
1657 	if (st->state_flags & PFSTATE_NOSYNC)
1658 		return;
1659 
1660 	if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1661 	    st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1662 		st->state_flags |= PFSTATE_NOSYNC;
1663 		return;
1664 	}
1665 
1666 	KASSERT(st->sync_state == PFSYNC_S_NONE,
1667 		("%s: st->sync_state %u", __func__, st->sync_state));
1668 
1669 	PFSYNC_LOCK(sc);
1670 	if (sc->sc_len == PFSYNC_MINPKT)
1671 		callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1672 
1673 	pfsync_q_ins(st, PFSYNC_S_INS, true);
1674 	PFSYNC_UNLOCK(sc);
1675 
1676 	st->sync_updates = 0;
1677 }
1678 
1679 static int
1680 pfsync_defer(struct pf_state *st, struct mbuf *m)
1681 {
1682 	struct pfsync_softc *sc = V_pfsyncif;
1683 	struct pfsync_deferral *pd;
1684 
1685 	if (m->m_flags & (M_BCAST|M_MCAST))
1686 		return (0);
1687 
1688 	PFSYNC_LOCK(sc);
1689 
1690 	if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) ||
1691 	    !(sc->sc_flags & PFSYNCF_DEFER)) {
1692 		PFSYNC_UNLOCK(sc);
1693 		return (0);
1694 	}
1695 
1696 	 if (sc->sc_deferred >= 128)
1697 		pfsync_undefer(TAILQ_FIRST(&sc->sc_deferrals), 0);
1698 
1699 	pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1700 	if (pd == NULL)
1701 		return (0);
1702 	sc->sc_deferred++;
1703 
1704 	m->m_flags |= M_SKIP_FIREWALL;
1705 	st->state_flags |= PFSTATE_ACK;
1706 
1707 	pd->pd_sc = sc;
1708 	pd->pd_refs = 0;
1709 	pd->pd_st = st;
1710 	pf_ref_state(st);
1711 	pd->pd_m = m;
1712 
1713 	TAILQ_INSERT_TAIL(&sc->sc_deferrals, pd, pd_entry);
1714 	callout_init_mtx(&pd->pd_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1715 	callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd);
1716 
1717 	pfsync_push(sc);
1718 
1719 	return (1);
1720 }
1721 
1722 static void
1723 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1724 {
1725 	struct pfsync_softc *sc = pd->pd_sc;
1726 	struct mbuf *m = pd->pd_m;
1727 	struct pf_state *st = pd->pd_st;
1728 
1729 	PFSYNC_LOCK_ASSERT(sc);
1730 
1731 	TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1732 	sc->sc_deferred--;
1733 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
1734 	free(pd, M_PFSYNC);
1735 	pf_release_state(st);
1736 
1737 	if (drop)
1738 		m_freem(m);
1739 	else {
1740 		_IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1741 		pfsync_push(sc);
1742 	}
1743 }
1744 
1745 static void
1746 pfsync_defer_tmo(void *arg)
1747 {
1748 	struct pfsync_deferral *pd = arg;
1749 	struct pfsync_softc *sc = pd->pd_sc;
1750 	struct mbuf *m = pd->pd_m;
1751 	struct pf_state *st = pd->pd_st;
1752 
1753 	PFSYNC_LOCK_ASSERT(sc);
1754 
1755 	CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
1756 
1757 	TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1758 	sc->sc_deferred--;
1759 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
1760 	if (pd->pd_refs == 0)
1761 		free(pd, M_PFSYNC);
1762 	PFSYNC_UNLOCK(sc);
1763 
1764 	ip_output(m, NULL, NULL, 0, NULL, NULL);
1765 
1766 	pf_release_state(st);
1767 
1768 	CURVNET_RESTORE();
1769 }
1770 
1771 static void
1772 pfsync_undefer_state(struct pf_state *st, int drop)
1773 {
1774 	struct pfsync_softc *sc = V_pfsyncif;
1775 	struct pfsync_deferral *pd;
1776 
1777 	PFSYNC_LOCK_ASSERT(sc);
1778 
1779 	TAILQ_FOREACH(pd, &sc->sc_deferrals, pd_entry) {
1780 		 if (pd->pd_st == st) {
1781 			if (callout_stop(&pd->pd_tmo) > 0)
1782 				pfsync_undefer(pd, drop);
1783 			return;
1784 		}
1785 	}
1786 
1787 	panic("%s: unable to find deferred state", __func__);
1788 }
1789 
1790 static void
1791 pfsync_update_state(struct pf_state *st)
1792 {
1793 	struct pfsync_softc *sc = V_pfsyncif;
1794 	bool sync = false, ref = true;
1795 
1796 	PF_STATE_LOCK_ASSERT(st);
1797 	PFSYNC_LOCK(sc);
1798 
1799 	if (st->state_flags & PFSTATE_ACK)
1800 		pfsync_undefer_state(st, 0);
1801 	if (st->state_flags & PFSTATE_NOSYNC) {
1802 		if (st->sync_state != PFSYNC_S_NONE)
1803 			pfsync_q_del(st, true);
1804 		PFSYNC_UNLOCK(sc);
1805 		return;
1806 	}
1807 
1808 	if (sc->sc_len == PFSYNC_MINPKT)
1809 		callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1810 
1811 	switch (st->sync_state) {
1812 	case PFSYNC_S_UPD_C:
1813 	case PFSYNC_S_UPD:
1814 	case PFSYNC_S_INS:
1815 		/* we're already handling it */
1816 
1817 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1818 			st->sync_updates++;
1819 			if (st->sync_updates >= sc->sc_maxupdates)
1820 				sync = true;
1821 		}
1822 		break;
1823 
1824 	case PFSYNC_S_IACK:
1825 		pfsync_q_del(st, false);
1826 		ref = false;
1827 		/* FALLTHROUGH */
1828 
1829 	case PFSYNC_S_NONE:
1830 		pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
1831 		st->sync_updates = 0;
1832 		break;
1833 
1834 	default:
1835 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1836 	}
1837 
1838 	if (sync || (time_uptime - st->pfsync_time) < 2)
1839 		pfsync_push(sc);
1840 
1841 	PFSYNC_UNLOCK(sc);
1842 }
1843 
1844 static void
1845 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1846 {
1847 	struct pfsync_softc *sc = V_pfsyncif;
1848 	struct pfsync_upd_req_item *item;
1849 	size_t nlen = sizeof(struct pfsync_upd_req);
1850 
1851 	PFSYNC_LOCK_ASSERT(sc);
1852 
1853 	/*
1854 	 * This code does a bit to prevent multiple update requests for the
1855 	 * same state being generated. It searches current subheader queue,
1856 	 * but it doesn't lookup into queue of already packed datagrams.
1857 	 */
1858 	TAILQ_FOREACH(item, &sc->sc_upd_req_list, ur_entry)
1859 		if (item->ur_msg.id == id &&
1860 		    item->ur_msg.creatorid == creatorid)
1861 			return;
1862 
1863 	item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1864 	if (item == NULL)
1865 		return; /* XXX stats */
1866 
1867 	item->ur_msg.id = id;
1868 	item->ur_msg.creatorid = creatorid;
1869 
1870 	if (TAILQ_EMPTY(&sc->sc_upd_req_list))
1871 		nlen += sizeof(struct pfsync_subheader);
1872 
1873 	if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
1874 		pfsync_sendout(1);
1875 
1876 		nlen = sizeof(struct pfsync_subheader) +
1877 		    sizeof(struct pfsync_upd_req);
1878 	}
1879 
1880 	TAILQ_INSERT_TAIL(&sc->sc_upd_req_list, item, ur_entry);
1881 	sc->sc_len += nlen;
1882 }
1883 
1884 static void
1885 pfsync_update_state_req(struct pf_state *st)
1886 {
1887 	struct pfsync_softc *sc = V_pfsyncif;
1888 	bool ref = true;
1889 
1890 	PF_STATE_LOCK_ASSERT(st);
1891 	PFSYNC_LOCK(sc);
1892 
1893 	if (st->state_flags & PFSTATE_NOSYNC) {
1894 		if (st->sync_state != PFSYNC_S_NONE)
1895 			pfsync_q_del(st, true);
1896 		PFSYNC_UNLOCK(sc);
1897 		return;
1898 	}
1899 
1900 	switch (st->sync_state) {
1901 	case PFSYNC_S_UPD_C:
1902 	case PFSYNC_S_IACK:
1903 		pfsync_q_del(st, false);
1904 		ref = false;
1905 		/* FALLTHROUGH */
1906 
1907 	case PFSYNC_S_NONE:
1908 		pfsync_q_ins(st, PFSYNC_S_UPD, ref);
1909 		pfsync_push(sc);
1910 		break;
1911 
1912 	case PFSYNC_S_INS:
1913 	case PFSYNC_S_UPD:
1914 	case PFSYNC_S_DEL:
1915 		/* we're already handling it */
1916 		break;
1917 
1918 	default:
1919 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1920 	}
1921 
1922 	PFSYNC_UNLOCK(sc);
1923 }
1924 
1925 static void
1926 pfsync_delete_state(struct pf_state *st)
1927 {
1928 	struct pfsync_softc *sc = V_pfsyncif;
1929 	bool ref = true;
1930 
1931 	PFSYNC_LOCK(sc);
1932 	if (st->state_flags & PFSTATE_ACK)
1933 		pfsync_undefer_state(st, 1);
1934 	if (st->state_flags & PFSTATE_NOSYNC) {
1935 		if (st->sync_state != PFSYNC_S_NONE)
1936 			pfsync_q_del(st, true);
1937 		PFSYNC_UNLOCK(sc);
1938 		return;
1939 	}
1940 
1941 	if (sc->sc_len == PFSYNC_MINPKT)
1942 		callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1943 
1944 	switch (st->sync_state) {
1945 	case PFSYNC_S_INS:
1946 		/* We never got to tell the world so just forget about it. */
1947 		pfsync_q_del(st, true);
1948 		break;
1949 
1950 	case PFSYNC_S_UPD_C:
1951 	case PFSYNC_S_UPD:
1952 	case PFSYNC_S_IACK:
1953 		pfsync_q_del(st, false);
1954 		ref = false;
1955 		/* FALLTHROUGH */
1956 
1957 	case PFSYNC_S_NONE:
1958 		pfsync_q_ins(st, PFSYNC_S_DEL, ref);
1959 		break;
1960 
1961 	default:
1962 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1963 	}
1964 
1965 	PFSYNC_UNLOCK(sc);
1966 }
1967 
1968 static void
1969 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
1970 {
1971 	struct pfsync_softc *sc = V_pfsyncif;
1972 	struct {
1973 		struct pfsync_subheader subh;
1974 		struct pfsync_clr clr;
1975 	} __packed r;
1976 
1977 	bzero(&r, sizeof(r));
1978 
1979 	r.subh.action = PFSYNC_ACT_CLR;
1980 	r.subh.count = htons(1);
1981 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
1982 
1983 	strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
1984 	r.clr.creatorid = creatorid;
1985 
1986 	PFSYNC_LOCK(sc);
1987 	pfsync_send_plus(&r, sizeof(r));
1988 	PFSYNC_UNLOCK(sc);
1989 }
1990 
1991 static void
1992 pfsync_q_ins(struct pf_state *st, int q, bool ref)
1993 {
1994 	struct pfsync_softc *sc = V_pfsyncif;
1995 	size_t nlen = pfsync_qs[q].len;
1996 
1997 	PFSYNC_LOCK_ASSERT(sc);
1998 
1999 	KASSERT(st->sync_state == PFSYNC_S_NONE,
2000 		("%s: st->sync_state %u", __func__, st->sync_state));
2001 	KASSERT(sc->sc_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2002 	    sc->sc_len));
2003 
2004 	if (TAILQ_EMPTY(&sc->sc_qs[q]))
2005 		nlen += sizeof(struct pfsync_subheader);
2006 
2007 	if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
2008 		pfsync_sendout(1);
2009 
2010 		nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2011 	}
2012 
2013 	sc->sc_len += nlen;
2014 	TAILQ_INSERT_TAIL(&sc->sc_qs[q], st, sync_list);
2015 	st->sync_state = q;
2016 	if (ref)
2017 		pf_ref_state(st);
2018 }
2019 
2020 static void
2021 pfsync_q_del(struct pf_state *st, bool unref)
2022 {
2023 	struct pfsync_softc *sc = V_pfsyncif;
2024 	int q = st->sync_state;
2025 
2026 	PFSYNC_LOCK_ASSERT(sc);
2027 	KASSERT(st->sync_state != PFSYNC_S_NONE,
2028 		("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2029 
2030 	sc->sc_len -= pfsync_qs[q].len;
2031 	TAILQ_REMOVE(&sc->sc_qs[q], st, sync_list);
2032 	st->sync_state = PFSYNC_S_NONE;
2033 	if (unref)
2034 		pf_release_state(st);
2035 
2036 	if (TAILQ_EMPTY(&sc->sc_qs[q]))
2037 		sc->sc_len -= sizeof(struct pfsync_subheader);
2038 }
2039 
2040 static void
2041 pfsync_bulk_start(void)
2042 {
2043 	struct pfsync_softc *sc = V_pfsyncif;
2044 
2045 	if (V_pf_status.debug >= PF_DEBUG_MISC)
2046 		printf("pfsync: received bulk update request\n");
2047 
2048 	PFSYNC_BLOCK(sc);
2049 
2050 	sc->sc_ureq_received = time_uptime;
2051 	sc->sc_bulk_hashid = 0;
2052 	sc->sc_bulk_stateid = 0;
2053 	pfsync_bulk_status(PFSYNC_BUS_START);
2054 	callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2055 	PFSYNC_BUNLOCK(sc);
2056 }
2057 
2058 static void
2059 pfsync_bulk_update(void *arg)
2060 {
2061 	struct pfsync_softc *sc = arg;
2062 	struct pf_state *s;
2063 	int i, sent = 0;
2064 
2065 	PFSYNC_BLOCK_ASSERT(sc);
2066 	CURVNET_SET(sc->sc_ifp->if_vnet);
2067 
2068 	/*
2069 	 * Start with last state from previous invocation.
2070 	 * It may had gone, in this case start from the
2071 	 * hash slot.
2072 	 */
2073 	s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2074 
2075 	if (s != NULL)
2076 		i = PF_IDHASH(s);
2077 	else
2078 		i = sc->sc_bulk_hashid;
2079 
2080 	for (; i <= pf_hashmask; i++) {
2081 		struct pf_idhash *ih = &V_pf_idhash[i];
2082 
2083 		if (s != NULL)
2084 			PF_HASHROW_ASSERT(ih);
2085 		else {
2086 			PF_HASHROW_LOCK(ih);
2087 			s = LIST_FIRST(&ih->states);
2088 		}
2089 
2090 		for (; s; s = LIST_NEXT(s, entry)) {
2091 
2092 			if (sent > 1 && (sc->sc_ifp->if_mtu - sc->sc_len) <
2093 			    sizeof(struct pfsync_state)) {
2094 				/* We've filled a packet. */
2095 				sc->sc_bulk_hashid = i;
2096 				sc->sc_bulk_stateid = s->id;
2097 				sc->sc_bulk_creatorid = s->creatorid;
2098 				PF_HASHROW_UNLOCK(ih);
2099 				callout_reset(&sc->sc_bulk_tmo, 1,
2100 				    pfsync_bulk_update, sc);
2101 				goto full;
2102 			}
2103 
2104 			if (s->sync_state == PFSYNC_S_NONE &&
2105 			    s->timeout < PFTM_MAX &&
2106 			    s->pfsync_time <= sc->sc_ureq_received) {
2107 				pfsync_update_state_req(s);
2108 				sent++;
2109 			}
2110 		}
2111 		PF_HASHROW_UNLOCK(ih);
2112 	}
2113 
2114 	/* We're done. */
2115 	pfsync_bulk_status(PFSYNC_BUS_END);
2116 
2117 full:
2118 	CURVNET_RESTORE();
2119 }
2120 
2121 static void
2122 pfsync_bulk_status(u_int8_t status)
2123 {
2124 	struct {
2125 		struct pfsync_subheader subh;
2126 		struct pfsync_bus bus;
2127 	} __packed r;
2128 
2129 	struct pfsync_softc *sc = V_pfsyncif;
2130 
2131 	bzero(&r, sizeof(r));
2132 
2133 	r.subh.action = PFSYNC_ACT_BUS;
2134 	r.subh.count = htons(1);
2135 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2136 
2137 	r.bus.creatorid = V_pf_status.hostid;
2138 	r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2139 	r.bus.status = status;
2140 
2141 	PFSYNC_LOCK(sc);
2142 	pfsync_send_plus(&r, sizeof(r));
2143 	PFSYNC_UNLOCK(sc);
2144 }
2145 
2146 static void
2147 pfsync_bulk_fail(void *arg)
2148 {
2149 	struct pfsync_softc *sc = arg;
2150 
2151 	CURVNET_SET(sc->sc_ifp->if_vnet);
2152 
2153 	PFSYNC_BLOCK_ASSERT(sc);
2154 
2155 	if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2156 		/* Try again */
2157 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2158 		    pfsync_bulk_fail, V_pfsyncif);
2159 		PFSYNC_LOCK(sc);
2160 		pfsync_request_update(0, 0);
2161 		PFSYNC_UNLOCK(sc);
2162 	} else {
2163 		/* Pretend like the transfer was ok. */
2164 		sc->sc_ureq_sent = 0;
2165 		sc->sc_bulk_tries = 0;
2166 		PFSYNC_LOCK(sc);
2167 		if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2168 			(*carp_demote_adj_p)(-V_pfsync_carp_adj,
2169 			    "pfsync bulk fail");
2170 		sc->sc_flags |= PFSYNCF_OK;
2171 		PFSYNC_UNLOCK(sc);
2172 		if (V_pf_status.debug >= PF_DEBUG_MISC)
2173 			printf("pfsync: failed to receive bulk update\n");
2174 	}
2175 
2176 	CURVNET_RESTORE();
2177 }
2178 
2179 static void
2180 pfsync_send_plus(void *plus, size_t pluslen)
2181 {
2182 	struct pfsync_softc *sc = V_pfsyncif;
2183 
2184 	PFSYNC_LOCK_ASSERT(sc);
2185 
2186 	if (sc->sc_len + pluslen > sc->sc_ifp->if_mtu)
2187 		pfsync_sendout(1);
2188 
2189 	sc->sc_plus = plus;
2190 	sc->sc_len += (sc->sc_pluslen = pluslen);
2191 
2192 	pfsync_sendout(1);
2193 }
2194 
2195 static void
2196 pfsync_timeout(void *arg)
2197 {
2198 	struct pfsync_softc *sc = arg;
2199 
2200 	CURVNET_SET(sc->sc_ifp->if_vnet);
2201 	PFSYNC_LOCK(sc);
2202 	pfsync_push(sc);
2203 	PFSYNC_UNLOCK(sc);
2204 	CURVNET_RESTORE();
2205 }
2206 
2207 static void
2208 pfsync_push(struct pfsync_softc *sc)
2209 {
2210 
2211 	PFSYNC_LOCK_ASSERT(sc);
2212 
2213 	sc->sc_flags |= PFSYNCF_PUSH;
2214 	swi_sched(V_pfsync_swi_cookie, 0);
2215 }
2216 
2217 static void
2218 pfsyncintr(void *arg)
2219 {
2220 	struct pfsync_softc *sc = arg;
2221 	struct mbuf *m, *n;
2222 
2223 	CURVNET_SET(sc->sc_ifp->if_vnet);
2224 
2225 	PFSYNC_LOCK(sc);
2226 	if ((sc->sc_flags & PFSYNCF_PUSH) && sc->sc_len > PFSYNC_MINPKT) {
2227 		pfsync_sendout(0);
2228 		sc->sc_flags &= ~PFSYNCF_PUSH;
2229 	}
2230 	_IF_DEQUEUE_ALL(&sc->sc_ifp->if_snd, m);
2231 	PFSYNC_UNLOCK(sc);
2232 
2233 	for (; m != NULL; m = n) {
2234 
2235 		n = m->m_nextpkt;
2236 		m->m_nextpkt = NULL;
2237 
2238 		/*
2239 		 * We distinguish between a deferral packet and our
2240 		 * own pfsync packet based on M_SKIP_FIREWALL
2241 		 * flag. This is XXX.
2242 		 */
2243 		if (m->m_flags & M_SKIP_FIREWALL)
2244 			ip_output(m, NULL, NULL, 0, NULL, NULL);
2245 		else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo,
2246 		    NULL) == 0)
2247 			V_pfsyncstats.pfsyncs_opackets++;
2248 		else
2249 			V_pfsyncstats.pfsyncs_oerrors++;
2250 	}
2251 	CURVNET_RESTORE();
2252 }
2253 
2254 static int
2255 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship)
2256 {
2257 	struct ip_moptions *imo = &sc->sc_imo;
2258 	int error;
2259 
2260 	if (!(ifp->if_flags & IFF_MULTICAST))
2261 		return (EADDRNOTAVAIL);
2262 
2263 	imo->imo_membership = (struct in_multi **)mship;
2264 	imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
2265 	imo->imo_multicast_vif = -1;
2266 
2267 	if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL,
2268 	    &imo->imo_membership[0])) != 0) {
2269 		imo->imo_membership = NULL;
2270 		return (error);
2271 	}
2272 	imo->imo_num_memberships++;
2273 	imo->imo_multicast_ifp = ifp;
2274 	imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2275 	imo->imo_multicast_loop = 0;
2276 
2277 	return (0);
2278 }
2279 
2280 static void
2281 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2282 {
2283 	struct ip_moptions *imo = &sc->sc_imo;
2284 
2285 	in_leavegroup(imo->imo_membership[0], NULL);
2286 	free(imo->imo_membership, M_PFSYNC);
2287 	imo->imo_membership = NULL;
2288 	imo->imo_multicast_ifp = NULL;
2289 }
2290 
2291 #ifdef INET
2292 extern  struct domain inetdomain;
2293 static struct protosw in_pfsync_protosw = {
2294 	.pr_type =		SOCK_RAW,
2295 	.pr_domain =		&inetdomain,
2296 	.pr_protocol =		IPPROTO_PFSYNC,
2297 	.pr_flags =		PR_ATOMIC|PR_ADDR,
2298 	.pr_input =		pfsync_input,
2299 	.pr_output =		rip_output,
2300 	.pr_ctloutput =		rip_ctloutput,
2301 	.pr_usrreqs =		&rip_usrreqs
2302 };
2303 #endif
2304 
2305 static void
2306 pfsync_pointers_init()
2307 {
2308 
2309 	PF_RULES_WLOCK();
2310 	pfsync_state_import_ptr = pfsync_state_import;
2311 	pfsync_insert_state_ptr = pfsync_insert_state;
2312 	pfsync_update_state_ptr = pfsync_update_state;
2313 	pfsync_delete_state_ptr = pfsync_delete_state;
2314 	pfsync_clear_states_ptr = pfsync_clear_states;
2315 	pfsync_defer_ptr = pfsync_defer;
2316 	PF_RULES_WUNLOCK();
2317 }
2318 
2319 static void
2320 pfsync_pointers_uninit()
2321 {
2322 
2323 	PF_RULES_WLOCK();
2324 	pfsync_state_import_ptr = NULL;
2325 	pfsync_insert_state_ptr = NULL;
2326 	pfsync_update_state_ptr = NULL;
2327 	pfsync_delete_state_ptr = NULL;
2328 	pfsync_clear_states_ptr = NULL;
2329 	pfsync_defer_ptr = NULL;
2330 	PF_RULES_WUNLOCK();
2331 }
2332 
2333 static void
2334 vnet_pfsync_init(const void *unused __unused)
2335 {
2336 	int error;
2337 
2338 	V_pfsync_cloner = if_clone_simple(pfsyncname,
2339 	    pfsync_clone_create, pfsync_clone_destroy, 1);
2340 	error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif,
2341 	    SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2342 	if (error) {
2343 		if_clone_detach(V_pfsync_cloner);
2344 		log(LOG_INFO, "swi_add() failed in %s\n", __func__);
2345 	}
2346 }
2347 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
2348     vnet_pfsync_init, NULL);
2349 
2350 static void
2351 vnet_pfsync_uninit(const void *unused __unused)
2352 {
2353 
2354 	if_clone_detach(V_pfsync_cloner);
2355 	swi_remove(V_pfsync_swi_cookie);
2356 }
2357 /*
2358  * Detach after pf is gone; otherwise we might touch pfsync memory
2359  * from within pf after freeing pfsync.
2360  */
2361 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_INIT_IF, SI_ORDER_SECOND,
2362     vnet_pfsync_uninit, NULL);
2363 
2364 static int
2365 pfsync_init()
2366 {
2367 #ifdef INET
2368 	int error;
2369 
2370 	error = pf_proto_register(PF_INET, &in_pfsync_protosw);
2371 	if (error)
2372 		return (error);
2373 	error = ipproto_register(IPPROTO_PFSYNC);
2374 	if (error) {
2375 		pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2376 		return (error);
2377 	}
2378 #endif
2379 	pfsync_pointers_init();
2380 
2381 	return (0);
2382 }
2383 
2384 static void
2385 pfsync_uninit()
2386 {
2387 
2388 	pfsync_pointers_uninit();
2389 
2390 #ifdef INET
2391 	ipproto_unregister(IPPROTO_PFSYNC);
2392 	pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2393 #endif
2394 }
2395 
2396 static int
2397 pfsync_modevent(module_t mod, int type, void *data)
2398 {
2399 	int error = 0;
2400 
2401 	switch (type) {
2402 	case MOD_LOAD:
2403 		error = pfsync_init();
2404 		break;
2405 	case MOD_QUIESCE:
2406 		/*
2407 		 * Module should not be unloaded due to race conditions.
2408 		 */
2409 		error = EBUSY;
2410 		break;
2411 	case MOD_UNLOAD:
2412 		pfsync_uninit();
2413 		break;
2414 	default:
2415 		error = EINVAL;
2416 		break;
2417 	}
2418 
2419 	return (error);
2420 }
2421 
2422 static moduledata_t pfsync_mod = {
2423 	pfsyncname,
2424 	pfsync_modevent,
2425 	0
2426 };
2427 
2428 #define PFSYNC_MODVER 1
2429 
2430 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
2431 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
2432 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2433 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
2434