xref: /freebsd/sys/netpfil/pf/if_pfsync.c (revision 7cd2dcf07629713e5a3d60472cfe4701b705a167)
1 /*	$OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $	*/
2 
3 /*
4  * Copyright (c) 2002 Michael Shalayeff
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
20  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
21  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
22  * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
24  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
25  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
26  * THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*
30  * Copyright (c) 2009 David Gwynne <dlg@openbsd.org>
31  *
32  * Permission to use, copy, modify, and distribute this software for any
33  * purpose with or without fee is hereby granted, provided that the above
34  * copyright notice and this permission notice appear in all copies.
35  *
36  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
37  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
38  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
39  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
40  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
41  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
42  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
43  */
44 
45 /*
46  * Revisions picked from OpenBSD after revision 1.110 import:
47  * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates
48  * 1.120, 1.175 - use monotonic time_uptime
49  * 1.122 - reduce number of updates for non-TCP sessions
50  * 1.128 - cleanups
51  * 1.146 - bzero() mbuf before sparsely filling it with data
52  * 1.170 - SIOCSIFMTU checks
53  * 1.126, 1.142 - deferred packets processing
54  * 1.173 - correct expire time processing
55  */
56 
57 #include <sys/cdefs.h>
58 __FBSDID("$FreeBSD$");
59 
60 #include "opt_inet.h"
61 #include "opt_inet6.h"
62 #include "opt_pf.h"
63 
64 #include <sys/param.h>
65 #include <sys/bus.h>
66 #include <sys/endian.h>
67 #include <sys/interrupt.h>
68 #include <sys/kernel.h>
69 #include <sys/lock.h>
70 #include <sys/mbuf.h>
71 #include <sys/module.h>
72 #include <sys/mutex.h>
73 #include <sys/priv.h>
74 #include <sys/protosw.h>
75 #include <sys/socket.h>
76 #include <sys/sockio.h>
77 #include <sys/sysctl.h>
78 
79 #include <net/bpf.h>
80 #include <net/if.h>
81 #include <net/if_clone.h>
82 #include <net/if_types.h>
83 #include <net/pfvar.h>
84 #include <net/if_pfsync.h>
85 
86 #include <netinet/if_ether.h>
87 #include <netinet/in.h>
88 #include <netinet/in_var.h>
89 #include <netinet/ip.h>
90 #include <netinet/ip_carp.h>
91 #include <netinet/ip_var.h>
92 #include <netinet/tcp.h>
93 #include <netinet/tcp_fsm.h>
94 #include <netinet/tcp_seq.h>
95 
96 #define PFSYNC_MINPKT ( \
97 	sizeof(struct ip) + \
98 	sizeof(struct pfsync_header) + \
99 	sizeof(struct pfsync_subheader) + \
100 	sizeof(struct pfsync_eof))
101 
102 struct pfsync_pkt {
103 	struct ip *ip;
104 	struct in_addr src;
105 	u_int8_t flags;
106 };
107 
108 static int	pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *,
109 		    struct pfsync_state_peer *);
110 static int	pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int);
111 static int	pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int);
112 static int	pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int);
113 static int	pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int);
114 static int	pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int);
115 static int	pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int);
116 static int	pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int);
117 static int	pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int);
118 static int	pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int);
119 static int	pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int);
120 static int	pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int);
121 static int	pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int);
122 
123 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = {
124 	pfsync_in_clr,			/* PFSYNC_ACT_CLR */
125 	pfsync_in_ins,			/* PFSYNC_ACT_INS */
126 	pfsync_in_iack,			/* PFSYNC_ACT_INS_ACK */
127 	pfsync_in_upd,			/* PFSYNC_ACT_UPD */
128 	pfsync_in_upd_c,		/* PFSYNC_ACT_UPD_C */
129 	pfsync_in_ureq,			/* PFSYNC_ACT_UPD_REQ */
130 	pfsync_in_del,			/* PFSYNC_ACT_DEL */
131 	pfsync_in_del_c,		/* PFSYNC_ACT_DEL_C */
132 	pfsync_in_error,		/* PFSYNC_ACT_INS_F */
133 	pfsync_in_error,		/* PFSYNC_ACT_DEL_F */
134 	pfsync_in_bus,			/* PFSYNC_ACT_BUS */
135 	pfsync_in_tdb,			/* PFSYNC_ACT_TDB */
136 	pfsync_in_eof			/* PFSYNC_ACT_EOF */
137 };
138 
139 struct pfsync_q {
140 	void		(*write)(struct pf_state *, void *);
141 	size_t		len;
142 	u_int8_t	action;
143 };
144 
145 /* we have one of these for every PFSYNC_S_ */
146 static void	pfsync_out_state(struct pf_state *, void *);
147 static void	pfsync_out_iack(struct pf_state *, void *);
148 static void	pfsync_out_upd_c(struct pf_state *, void *);
149 static void	pfsync_out_del(struct pf_state *, void *);
150 
151 static struct pfsync_q pfsync_qs[] = {
152 	{ pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_INS },
153 	{ pfsync_out_iack,  sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
154 	{ pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_UPD },
155 	{ pfsync_out_upd_c, sizeof(struct pfsync_upd_c),   PFSYNC_ACT_UPD_C },
156 	{ pfsync_out_del,   sizeof(struct pfsync_del_c),   PFSYNC_ACT_DEL_C }
157 };
158 
159 static void	pfsync_q_ins(struct pf_state *, int);
160 static void	pfsync_q_del(struct pf_state *);
161 
162 static void	pfsync_update_state(struct pf_state *);
163 
164 struct pfsync_upd_req_item {
165 	TAILQ_ENTRY(pfsync_upd_req_item)	ur_entry;
166 	struct pfsync_upd_req			ur_msg;
167 };
168 
169 struct pfsync_deferral {
170 	struct pfsync_softc		*pd_sc;
171 	TAILQ_ENTRY(pfsync_deferral)	pd_entry;
172 	u_int				pd_refs;
173 	struct callout			pd_tmo;
174 
175 	struct pf_state			*pd_st;
176 	struct mbuf			*pd_m;
177 };
178 
179 struct pfsync_softc {
180 	/* Configuration */
181 	struct ifnet		*sc_ifp;
182 	struct ifnet		*sc_sync_if;
183 	struct ip_moptions	sc_imo;
184 	struct in_addr		sc_sync_peer;
185 	uint32_t		sc_flags;
186 #define	PFSYNCF_OK		0x00000001
187 #define	PFSYNCF_DEFER		0x00000002
188 #define	PFSYNCF_PUSH		0x00000004
189 	uint8_t			sc_maxupdates;
190 	struct ip		sc_template;
191 	struct callout		sc_tmo;
192 	struct mtx		sc_mtx;
193 
194 	/* Queued data */
195 	size_t			sc_len;
196 	TAILQ_HEAD(, pf_state)			sc_qs[PFSYNC_S_COUNT];
197 	TAILQ_HEAD(, pfsync_upd_req_item)	sc_upd_req_list;
198 	TAILQ_HEAD(, pfsync_deferral)		sc_deferrals;
199 	u_int			sc_deferred;
200 	void			*sc_plus;
201 	size_t			sc_pluslen;
202 
203 	/* Bulk update info */
204 	struct mtx		sc_bulk_mtx;
205 	uint32_t		sc_ureq_sent;
206 	int			sc_bulk_tries;
207 	uint32_t		sc_ureq_received;
208 	int			sc_bulk_hashid;
209 	uint64_t		sc_bulk_stateid;
210 	uint32_t		sc_bulk_creatorid;
211 	struct callout		sc_bulk_tmo;
212 	struct callout		sc_bulkfail_tmo;
213 };
214 
215 #define	PFSYNC_LOCK(sc)		mtx_lock(&(sc)->sc_mtx)
216 #define	PFSYNC_UNLOCK(sc)	mtx_unlock(&(sc)->sc_mtx)
217 #define	PFSYNC_LOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_mtx, MA_OWNED)
218 
219 #define	PFSYNC_BLOCK(sc)	mtx_lock(&(sc)->sc_bulk_mtx)
220 #define	PFSYNC_BUNLOCK(sc)	mtx_unlock(&(sc)->sc_bulk_mtx)
221 #define	PFSYNC_BLOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
222 
223 static const char pfsyncname[] = "pfsync";
224 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
225 static VNET_DEFINE(struct pfsync_softc	*, pfsyncif) = NULL;
226 #define	V_pfsyncif		VNET(pfsyncif)
227 static VNET_DEFINE(void *, pfsync_swi_cookie) = NULL;
228 #define	V_pfsync_swi_cookie	VNET(pfsync_swi_cookie)
229 static VNET_DEFINE(struct pfsyncstats, pfsyncstats);
230 #define	V_pfsyncstats		VNET(pfsyncstats)
231 static VNET_DEFINE(int, pfsync_carp_adj) = CARP_MAXSKEW;
232 #define	V_pfsync_carp_adj	VNET(pfsync_carp_adj)
233 
234 static void	pfsync_timeout(void *);
235 static void	pfsync_push(struct pfsync_softc *);
236 static void	pfsyncintr(void *);
237 static int	pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
238 		    void *);
239 static void	pfsync_multicast_cleanup(struct pfsync_softc *);
240 static void	pfsync_pointers_init(void);
241 static void	pfsync_pointers_uninit(void);
242 static int	pfsync_init(void);
243 static void	pfsync_uninit(void);
244 
245 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC");
246 SYSCTL_VNET_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_RW,
247     &VNET_NAME(pfsyncstats), pfsyncstats,
248     "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
249 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW,
250     &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
251 
252 static int	pfsync_clone_create(struct if_clone *, int, caddr_t);
253 static void	pfsync_clone_destroy(struct ifnet *);
254 static int	pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
255 		    struct pf_state_peer *);
256 static int	pfsyncoutput(struct ifnet *, struct mbuf *, struct sockaddr *,
257 		    struct route *);
258 static int	pfsyncioctl(struct ifnet *, u_long, caddr_t);
259 
260 static int	pfsync_defer(struct pf_state *, struct mbuf *);
261 static void	pfsync_undefer(struct pfsync_deferral *, int);
262 static void	pfsync_undefer_state(struct pf_state *, int);
263 static void	pfsync_defer_tmo(void *);
264 
265 static void	pfsync_request_update(u_int32_t, u_int64_t);
266 static void	pfsync_update_state_req(struct pf_state *);
267 
268 static void	pfsync_drop(struct pfsync_softc *);
269 static void	pfsync_sendout(int);
270 static void	pfsync_send_plus(void *, size_t);
271 
272 static void	pfsync_bulk_start(void);
273 static void	pfsync_bulk_status(u_int8_t);
274 static void	pfsync_bulk_update(void *);
275 static void	pfsync_bulk_fail(void *);
276 
277 #ifdef IPSEC
278 static void	pfsync_update_net_tdb(struct pfsync_tdb *);
279 #endif
280 
281 #define PFSYNC_MAX_BULKTRIES	12
282 
283 VNET_DEFINE(struct if_clone *, pfsync_cloner);
284 #define	V_pfsync_cloner	VNET(pfsync_cloner)
285 
286 static int
287 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
288 {
289 	struct pfsync_softc *sc;
290 	struct ifnet *ifp;
291 	int q;
292 
293 	if (unit != 0)
294 		return (EINVAL);
295 
296 	sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
297 	sc->sc_flags |= PFSYNCF_OK;
298 
299 	for (q = 0; q < PFSYNC_S_COUNT; q++)
300 		TAILQ_INIT(&sc->sc_qs[q]);
301 
302 	TAILQ_INIT(&sc->sc_upd_req_list);
303 	TAILQ_INIT(&sc->sc_deferrals);
304 
305 	sc->sc_len = PFSYNC_MINPKT;
306 	sc->sc_maxupdates = 128;
307 
308 	ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
309 	if (ifp == NULL) {
310 		free(sc, M_PFSYNC);
311 		return (ENOSPC);
312 	}
313 	if_initname(ifp, pfsyncname, unit);
314 	ifp->if_softc = sc;
315 	ifp->if_ioctl = pfsyncioctl;
316 	ifp->if_output = pfsyncoutput;
317 	ifp->if_type = IFT_PFSYNC;
318 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
319 	ifp->if_hdrlen = sizeof(struct pfsync_header);
320 	ifp->if_mtu = ETHERMTU;
321 	mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
322 	mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
323 	callout_init(&sc->sc_tmo, CALLOUT_MPSAFE);
324 	callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
325 	callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
326 
327 	if_attach(ifp);
328 
329 	bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
330 
331 	V_pfsyncif = sc;
332 
333 	return (0);
334 }
335 
336 static void
337 pfsync_clone_destroy(struct ifnet *ifp)
338 {
339 	struct pfsync_softc *sc = ifp->if_softc;
340 
341 	/*
342 	 * At this stage, everything should have already been
343 	 * cleared by pfsync_uninit(), and we have only to
344 	 * drain callouts.
345 	 */
346 	while (sc->sc_deferred > 0) {
347 		struct pfsync_deferral *pd = TAILQ_FIRST(&sc->sc_deferrals);
348 
349 		TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
350 		sc->sc_deferred--;
351 		if (callout_stop(&pd->pd_tmo)) {
352 			pf_release_state(pd->pd_st);
353 			m_freem(pd->pd_m);
354 			free(pd, M_PFSYNC);
355 		} else {
356 			pd->pd_refs++;
357 			callout_drain(&pd->pd_tmo);
358 			free(pd, M_PFSYNC);
359 		}
360 	}
361 
362 	callout_drain(&sc->sc_tmo);
363 	callout_drain(&sc->sc_bulkfail_tmo);
364 	callout_drain(&sc->sc_bulk_tmo);
365 
366 	if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
367 		(*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
368 	bpfdetach(ifp);
369 	if_detach(ifp);
370 
371 	pfsync_drop(sc);
372 
373 	if_free(ifp);
374 	if (sc->sc_imo.imo_membership)
375 		pfsync_multicast_cleanup(sc);
376 	mtx_destroy(&sc->sc_mtx);
377 	mtx_destroy(&sc->sc_bulk_mtx);
378 	free(sc, M_PFSYNC);
379 
380 	V_pfsyncif = NULL;
381 }
382 
383 static int
384 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
385     struct pf_state_peer *d)
386 {
387 	if (s->scrub.scrub_flag && d->scrub == NULL) {
388 		d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
389 		if (d->scrub == NULL)
390 			return (ENOMEM);
391 	}
392 
393 	return (0);
394 }
395 
396 
397 static int
398 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
399 {
400 	struct pfsync_softc *sc = V_pfsyncif;
401 	struct pf_state	*st = NULL;
402 	struct pf_state_key *skw = NULL, *sks = NULL;
403 	struct pf_rule *r = NULL;
404 	struct pfi_kif	*kif;
405 	int error;
406 
407 	PF_RULES_RASSERT();
408 
409 	if (sp->creatorid == 0 && V_pf_status.debug >= PF_DEBUG_MISC) {
410 		printf("%s: invalid creator id: %08x\n", __func__,
411 		    ntohl(sp->creatorid));
412 		return (EINVAL);
413 	}
414 
415 	if ((kif = pfi_kif_find(sp->ifname)) == NULL) {
416 		if (V_pf_status.debug >= PF_DEBUG_MISC)
417 			printf("%s: unknown interface: %s\n", __func__,
418 			    sp->ifname);
419 		if (flags & PFSYNC_SI_IOCTL)
420 			return (EINVAL);
421 		return (0);	/* skip this state */
422 	}
423 
424 	/*
425 	 * If the ruleset checksums match or the state is coming from the ioctl,
426 	 * it's safe to associate the state with the rule of that number.
427 	 */
428 	if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
429 	    (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
430 	    pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
431 		r = pf_main_ruleset.rules[
432 		    PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
433 	else
434 		r = &V_pf_default_rule;
435 
436 	if ((r->max_states && r->states_cur >= r->max_states))
437 		goto cleanup;
438 
439 	/*
440 	 * XXXGL: consider M_WAITOK in ioctl path after.
441 	 */
442 	if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL)
443 		goto cleanup;
444 
445 	if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
446 		goto cleanup;
447 
448 	if (PF_ANEQ(&sp->key[PF_SK_WIRE].addr[0],
449 	    &sp->key[PF_SK_STACK].addr[0], sp->af) ||
450 	    PF_ANEQ(&sp->key[PF_SK_WIRE].addr[1],
451 	    &sp->key[PF_SK_STACK].addr[1], sp->af) ||
452 	    sp->key[PF_SK_WIRE].port[0] != sp->key[PF_SK_STACK].port[0] ||
453 	    sp->key[PF_SK_WIRE].port[1] != sp->key[PF_SK_STACK].port[1]) {
454 		sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
455 		if (sks == NULL)
456 			goto cleanup;
457 	} else
458 		sks = skw;
459 
460 	/* allocate memory for scrub info */
461 	if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
462 	    pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
463 		goto cleanup;
464 
465 	/* copy to state key(s) */
466 	skw->addr[0] = sp->key[PF_SK_WIRE].addr[0];
467 	skw->addr[1] = sp->key[PF_SK_WIRE].addr[1];
468 	skw->port[0] = sp->key[PF_SK_WIRE].port[0];
469 	skw->port[1] = sp->key[PF_SK_WIRE].port[1];
470 	skw->proto = sp->proto;
471 	skw->af = sp->af;
472 	if (sks != skw) {
473 		sks->addr[0] = sp->key[PF_SK_STACK].addr[0];
474 		sks->addr[1] = sp->key[PF_SK_STACK].addr[1];
475 		sks->port[0] = sp->key[PF_SK_STACK].port[0];
476 		sks->port[1] = sp->key[PF_SK_STACK].port[1];
477 		sks->proto = sp->proto;
478 		sks->af = sp->af;
479 	}
480 
481 	/* copy to state */
482 	bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
483 	st->creation = time_uptime - ntohl(sp->creation);
484 	st->expire = time_uptime;
485 	if (sp->expire) {
486 		uint32_t timeout;
487 
488 		timeout = r->timeout[sp->timeout];
489 		if (!timeout)
490 			timeout = V_pf_default_rule.timeout[sp->timeout];
491 
492 		/* sp->expire may have been adaptively scaled by export. */
493 		st->expire -= timeout - ntohl(sp->expire);
494 	}
495 
496 	st->direction = sp->direction;
497 	st->log = sp->log;
498 	st->timeout = sp->timeout;
499 	st->state_flags = sp->state_flags;
500 
501 	st->id = sp->id;
502 	st->creatorid = sp->creatorid;
503 	pf_state_peer_ntoh(&sp->src, &st->src);
504 	pf_state_peer_ntoh(&sp->dst, &st->dst);
505 
506 	st->rule.ptr = r;
507 	st->nat_rule.ptr = NULL;
508 	st->anchor.ptr = NULL;
509 	st->rt_kif = NULL;
510 
511 	st->pfsync_time = time_uptime;
512 	st->sync_state = PFSYNC_S_NONE;
513 
514 	/* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
515 	r->states_cur++;
516 	r->states_tot++;
517 
518 	if (!(flags & PFSYNC_SI_IOCTL))
519 		st->state_flags |= PFSTATE_NOSYNC;
520 
521 	if ((error = pf_state_insert(kif, skw, sks, st)) != 0) {
522 		/* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */
523 		r->states_cur--;
524 		goto cleanup_state;
525 	}
526 
527 	if (!(flags & PFSYNC_SI_IOCTL)) {
528 		st->state_flags &= ~PFSTATE_NOSYNC;
529 		if (st->state_flags & PFSTATE_ACK) {
530 			pfsync_q_ins(st, PFSYNC_S_IACK);
531 			pfsync_push(sc);
532 		}
533 	}
534 	st->state_flags &= ~PFSTATE_ACK;
535 	PF_STATE_UNLOCK(st);
536 
537 	return (0);
538 
539 cleanup:
540 	error = ENOMEM;
541 	if (skw == sks)
542 		sks = NULL;
543 	if (skw != NULL)
544 		uma_zfree(V_pf_state_key_z, skw);
545 	if (sks != NULL)
546 		uma_zfree(V_pf_state_key_z, sks);
547 
548 cleanup_state:	/* pf_state_insert() frees the state keys. */
549 	if (st) {
550 		if (st->dst.scrub)
551 			uma_zfree(V_pf_state_scrub_z, st->dst.scrub);
552 		if (st->src.scrub)
553 			uma_zfree(V_pf_state_scrub_z, st->src.scrub);
554 		uma_zfree(V_pf_state_z, st);
555 	}
556 	return (error);
557 }
558 
559 static void
560 pfsync_input(struct mbuf *m, __unused int off)
561 {
562 	struct pfsync_softc *sc = V_pfsyncif;
563 	struct pfsync_pkt pkt;
564 	struct ip *ip = mtod(m, struct ip *);
565 	struct pfsync_header *ph;
566 	struct pfsync_subheader subh;
567 
568 	int offset;
569 	int rv;
570 	uint16_t count;
571 
572 	V_pfsyncstats.pfsyncs_ipackets++;
573 
574 	/* Verify that we have a sync interface configured. */
575 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
576 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
577 		goto done;
578 
579 	/* verify that the packet came in on the right interface */
580 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
581 		V_pfsyncstats.pfsyncs_badif++;
582 		goto done;
583 	}
584 
585 	sc->sc_ifp->if_ipackets++;
586 	sc->sc_ifp->if_ibytes += m->m_pkthdr.len;
587 	/* verify that the IP TTL is 255. */
588 	if (ip->ip_ttl != PFSYNC_DFLTTL) {
589 		V_pfsyncstats.pfsyncs_badttl++;
590 		goto done;
591 	}
592 
593 	offset = ip->ip_hl << 2;
594 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
595 		V_pfsyncstats.pfsyncs_hdrops++;
596 		goto done;
597 	}
598 
599 	if (offset + sizeof(*ph) > m->m_len) {
600 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
601 			V_pfsyncstats.pfsyncs_hdrops++;
602 			return;
603 		}
604 		ip = mtod(m, struct ip *);
605 	}
606 	ph = (struct pfsync_header *)((char *)ip + offset);
607 
608 	/* verify the version */
609 	if (ph->version != PFSYNC_VERSION) {
610 		V_pfsyncstats.pfsyncs_badver++;
611 		goto done;
612 	}
613 
614 	/* Cheaper to grab this now than having to mess with mbufs later */
615 	pkt.ip = ip;
616 	pkt.src = ip->ip_src;
617 	pkt.flags = 0;
618 
619 	/*
620 	 * Trusting pf_chksum during packet processing, as well as seeking
621 	 * in interface name tree, require holding PF_RULES_RLOCK().
622 	 */
623 	PF_RULES_RLOCK();
624 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
625 		pkt.flags |= PFSYNC_SI_CKSUM;
626 
627 	offset += sizeof(*ph);
628 	for (;;) {
629 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
630 		offset += sizeof(subh);
631 
632 		if (subh.action >= PFSYNC_ACT_MAX) {
633 			V_pfsyncstats.pfsyncs_badact++;
634 			PF_RULES_RUNLOCK();
635 			goto done;
636 		}
637 
638 		count = ntohs(subh.count);
639 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
640 		rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count);
641 		if (rv == -1) {
642 			PF_RULES_RUNLOCK();
643 			return;
644 		}
645 
646 		offset += rv;
647 	}
648 	PF_RULES_RUNLOCK();
649 
650 done:
651 	m_freem(m);
652 }
653 
654 static int
655 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
656 {
657 	struct pfsync_clr *clr;
658 	struct mbuf *mp;
659 	int len = sizeof(*clr) * count;
660 	int i, offp;
661 	u_int32_t creatorid;
662 
663 	mp = m_pulldown(m, offset, len, &offp);
664 	if (mp == NULL) {
665 		V_pfsyncstats.pfsyncs_badlen++;
666 		return (-1);
667 	}
668 	clr = (struct pfsync_clr *)(mp->m_data + offp);
669 
670 	for (i = 0; i < count; i++) {
671 		creatorid = clr[i].creatorid;
672 
673 		if (clr[i].ifname[0] != '\0' &&
674 		    pfi_kif_find(clr[i].ifname) == NULL)
675 			continue;
676 
677 		for (int i = 0; i <= V_pf_hashmask; i++) {
678 			struct pf_idhash *ih = &V_pf_idhash[i];
679 			struct pf_state *s;
680 relock:
681 			PF_HASHROW_LOCK(ih);
682 			LIST_FOREACH(s, &ih->states, entry) {
683 				if (s->creatorid == creatorid) {
684 					s->state_flags |= PFSTATE_NOSYNC;
685 					pf_unlink_state(s, PF_ENTER_LOCKED);
686 					goto relock;
687 				}
688 			}
689 			PF_HASHROW_UNLOCK(ih);
690 		}
691 	}
692 
693 	return (len);
694 }
695 
696 static int
697 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
698 {
699 	struct mbuf *mp;
700 	struct pfsync_state *sa, *sp;
701 	int len = sizeof(*sp) * count;
702 	int i, offp;
703 
704 	mp = m_pulldown(m, offset, len, &offp);
705 	if (mp == NULL) {
706 		V_pfsyncstats.pfsyncs_badlen++;
707 		return (-1);
708 	}
709 	sa = (struct pfsync_state *)(mp->m_data + offp);
710 
711 	for (i = 0; i < count; i++) {
712 		sp = &sa[i];
713 
714 		/* Check for invalid values. */
715 		if (sp->timeout >= PFTM_MAX ||
716 		    sp->src.state > PF_TCPS_PROXY_DST ||
717 		    sp->dst.state > PF_TCPS_PROXY_DST ||
718 		    sp->direction > PF_OUT ||
719 		    (sp->af != AF_INET && sp->af != AF_INET6)) {
720 			if (V_pf_status.debug >= PF_DEBUG_MISC)
721 				printf("%s: invalid value\n", __func__);
722 			V_pfsyncstats.pfsyncs_badval++;
723 			continue;
724 		}
725 
726 		if (pfsync_state_import(sp, pkt->flags) == ENOMEM)
727 			/* Drop out, but process the rest of the actions. */
728 			break;
729 	}
730 
731 	return (len);
732 }
733 
734 static int
735 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
736 {
737 	struct pfsync_ins_ack *ia, *iaa;
738 	struct pf_state *st;
739 
740 	struct mbuf *mp;
741 	int len = count * sizeof(*ia);
742 	int offp, i;
743 
744 	mp = m_pulldown(m, offset, len, &offp);
745 	if (mp == NULL) {
746 		V_pfsyncstats.pfsyncs_badlen++;
747 		return (-1);
748 	}
749 	iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
750 
751 	for (i = 0; i < count; i++) {
752 		ia = &iaa[i];
753 
754 		st = pf_find_state_byid(ia->id, ia->creatorid);
755 		if (st == NULL)
756 			continue;
757 
758 		if (st->state_flags & PFSTATE_ACK) {
759 			PFSYNC_LOCK(V_pfsyncif);
760 			pfsync_undefer_state(st, 0);
761 			PFSYNC_UNLOCK(V_pfsyncif);
762 		}
763 		PF_STATE_UNLOCK(st);
764 	}
765 	/*
766 	 * XXX this is not yet implemented, but we know the size of the
767 	 * message so we can skip it.
768 	 */
769 
770 	return (count * sizeof(struct pfsync_ins_ack));
771 }
772 
773 static int
774 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src,
775     struct pfsync_state_peer *dst)
776 {
777 	int sfail = 0;
778 
779 	PF_STATE_LOCK_ASSERT(st);
780 
781 	/*
782 	 * The state should never go backwards except
783 	 * for syn-proxy states.  Neither should the
784 	 * sequence window slide backwards.
785 	 */
786 	if (st->src.state > src->state &&
787 	    (st->src.state < PF_TCPS_PROXY_SRC ||
788 	    src->state >= PF_TCPS_PROXY_SRC))
789 		sfail = 1;
790 	else if (SEQ_GT(st->src.seqlo, ntohl(src->seqlo)))
791 		sfail = 3;
792 	else if (st->dst.state > dst->state) {
793 		/* There might still be useful
794 		 * information about the src state here,
795 		 * so import that part of the update,
796 		 * then "fail" so we send the updated
797 		 * state back to the peer who is missing
798 		 * our what we know. */
799 		pf_state_peer_ntoh(src, &st->src);
800 		/* XXX do anything with timeouts? */
801 		sfail = 7;
802 	} else if (st->dst.state >= TCPS_SYN_SENT &&
803 	    SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo)))
804 		sfail = 4;
805 
806 	return (sfail);
807 }
808 
809 static int
810 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
811 {
812 	struct pfsync_softc *sc = V_pfsyncif;
813 	struct pfsync_state *sa, *sp;
814 	struct pf_state_key *sk;
815 	struct pf_state *st;
816 	int sfail;
817 
818 	struct mbuf *mp;
819 	int len = count * sizeof(*sp);
820 	int offp, i;
821 
822 	mp = m_pulldown(m, offset, len, &offp);
823 	if (mp == NULL) {
824 		V_pfsyncstats.pfsyncs_badlen++;
825 		return (-1);
826 	}
827 	sa = (struct pfsync_state *)(mp->m_data + offp);
828 
829 	for (i = 0; i < count; i++) {
830 		sp = &sa[i];
831 
832 		/* check for invalid values */
833 		if (sp->timeout >= PFTM_MAX ||
834 		    sp->src.state > PF_TCPS_PROXY_DST ||
835 		    sp->dst.state > PF_TCPS_PROXY_DST) {
836 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
837 				printf("pfsync_input: PFSYNC_ACT_UPD: "
838 				    "invalid value\n");
839 			}
840 			V_pfsyncstats.pfsyncs_badval++;
841 			continue;
842 		}
843 
844 		st = pf_find_state_byid(sp->id, sp->creatorid);
845 		if (st == NULL) {
846 			/* insert the update */
847 			if (pfsync_state_import(sp, 0))
848 				V_pfsyncstats.pfsyncs_badstate++;
849 			continue;
850 		}
851 
852 		if (st->state_flags & PFSTATE_ACK) {
853 			PFSYNC_LOCK(sc);
854 			pfsync_undefer_state(st, 1);
855 			PFSYNC_UNLOCK(sc);
856 		}
857 
858 		sk = st->key[PF_SK_WIRE];	/* XXX right one? */
859 		sfail = 0;
860 		if (sk->proto == IPPROTO_TCP)
861 			sfail = pfsync_upd_tcp(st, &sp->src, &sp->dst);
862 		else {
863 			/*
864 			 * Non-TCP protocol state machine always go
865 			 * forwards
866 			 */
867 			if (st->src.state > sp->src.state)
868 				sfail = 5;
869 			else if (st->dst.state > sp->dst.state)
870 				sfail = 6;
871 		}
872 
873 		if (sfail) {
874 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
875 				printf("pfsync: %s stale update (%d)"
876 				    " id: %016llx creatorid: %08x\n",
877 				    (sfail < 7 ?  "ignoring" : "partial"),
878 				    sfail, (unsigned long long)be64toh(st->id),
879 				    ntohl(st->creatorid));
880 			}
881 			V_pfsyncstats.pfsyncs_stale++;
882 
883 			pfsync_update_state(st);
884 			PF_STATE_UNLOCK(st);
885 			PFSYNC_LOCK(sc);
886 			pfsync_push(sc);
887 			PFSYNC_UNLOCK(sc);
888 			continue;
889 		}
890 		pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
891 		pf_state_peer_ntoh(&sp->src, &st->src);
892 		pf_state_peer_ntoh(&sp->dst, &st->dst);
893 		st->expire = time_uptime;
894 		st->timeout = sp->timeout;
895 		st->pfsync_time = time_uptime;
896 		PF_STATE_UNLOCK(st);
897 	}
898 
899 	return (len);
900 }
901 
902 static int
903 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
904 {
905 	struct pfsync_softc *sc = V_pfsyncif;
906 	struct pfsync_upd_c *ua, *up;
907 	struct pf_state_key *sk;
908 	struct pf_state *st;
909 
910 	int len = count * sizeof(*up);
911 	int sfail;
912 
913 	struct mbuf *mp;
914 	int offp, i;
915 
916 	mp = m_pulldown(m, offset, len, &offp);
917 	if (mp == NULL) {
918 		V_pfsyncstats.pfsyncs_badlen++;
919 		return (-1);
920 	}
921 	ua = (struct pfsync_upd_c *)(mp->m_data + offp);
922 
923 	for (i = 0; i < count; i++) {
924 		up = &ua[i];
925 
926 		/* check for invalid values */
927 		if (up->timeout >= PFTM_MAX ||
928 		    up->src.state > PF_TCPS_PROXY_DST ||
929 		    up->dst.state > PF_TCPS_PROXY_DST) {
930 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
931 				printf("pfsync_input: "
932 				    "PFSYNC_ACT_UPD_C: "
933 				    "invalid value\n");
934 			}
935 			V_pfsyncstats.pfsyncs_badval++;
936 			continue;
937 		}
938 
939 		st = pf_find_state_byid(up->id, up->creatorid);
940 		if (st == NULL) {
941 			/* We don't have this state. Ask for it. */
942 			PFSYNC_LOCK(sc);
943 			pfsync_request_update(up->creatorid, up->id);
944 			PFSYNC_UNLOCK(sc);
945 			continue;
946 		}
947 
948 		if (st->state_flags & PFSTATE_ACK) {
949 			PFSYNC_LOCK(sc);
950 			pfsync_undefer_state(st, 1);
951 			PFSYNC_UNLOCK(sc);
952 		}
953 
954 		sk = st->key[PF_SK_WIRE]; /* XXX right one? */
955 		sfail = 0;
956 		if (sk->proto == IPPROTO_TCP)
957 			sfail = pfsync_upd_tcp(st, &up->src, &up->dst);
958 		else {
959 			/*
960 			 * Non-TCP protocol state machine always go forwards
961 			 */
962 			if (st->src.state > up->src.state)
963 				sfail = 5;
964 			else if (st->dst.state > up->dst.state)
965 				sfail = 6;
966 		}
967 
968 		if (sfail) {
969 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
970 				printf("pfsync: ignoring stale update "
971 				    "(%d) id: %016llx "
972 				    "creatorid: %08x\n", sfail,
973 				    (unsigned long long)be64toh(st->id),
974 				    ntohl(st->creatorid));
975 			}
976 			V_pfsyncstats.pfsyncs_stale++;
977 
978 			pfsync_update_state(st);
979 			PF_STATE_UNLOCK(st);
980 			PFSYNC_LOCK(sc);
981 			pfsync_push(sc);
982 			PFSYNC_UNLOCK(sc);
983 			continue;
984 		}
985 		pfsync_alloc_scrub_memory(&up->dst, &st->dst);
986 		pf_state_peer_ntoh(&up->src, &st->src);
987 		pf_state_peer_ntoh(&up->dst, &st->dst);
988 		st->expire = time_uptime;
989 		st->timeout = up->timeout;
990 		st->pfsync_time = time_uptime;
991 		PF_STATE_UNLOCK(st);
992 	}
993 
994 	return (len);
995 }
996 
997 static int
998 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
999 {
1000 	struct pfsync_upd_req *ur, *ura;
1001 	struct mbuf *mp;
1002 	int len = count * sizeof(*ur);
1003 	int i, offp;
1004 
1005 	struct pf_state *st;
1006 
1007 	mp = m_pulldown(m, offset, len, &offp);
1008 	if (mp == NULL) {
1009 		V_pfsyncstats.pfsyncs_badlen++;
1010 		return (-1);
1011 	}
1012 	ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1013 
1014 	for (i = 0; i < count; i++) {
1015 		ur = &ura[i];
1016 
1017 		if (ur->id == 0 && ur->creatorid == 0)
1018 			pfsync_bulk_start();
1019 		else {
1020 			st = pf_find_state_byid(ur->id, ur->creatorid);
1021 			if (st == NULL) {
1022 				V_pfsyncstats.pfsyncs_badstate++;
1023 				continue;
1024 			}
1025 			if (st->state_flags & PFSTATE_NOSYNC) {
1026 				PF_STATE_UNLOCK(st);
1027 				continue;
1028 			}
1029 
1030 			pfsync_update_state_req(st);
1031 			PF_STATE_UNLOCK(st);
1032 		}
1033 	}
1034 
1035 	return (len);
1036 }
1037 
1038 static int
1039 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1040 {
1041 	struct mbuf *mp;
1042 	struct pfsync_state *sa, *sp;
1043 	struct pf_state *st;
1044 	int len = count * sizeof(*sp);
1045 	int offp, i;
1046 
1047 	mp = m_pulldown(m, offset, len, &offp);
1048 	if (mp == NULL) {
1049 		V_pfsyncstats.pfsyncs_badlen++;
1050 		return (-1);
1051 	}
1052 	sa = (struct pfsync_state *)(mp->m_data + offp);
1053 
1054 	for (i = 0; i < count; i++) {
1055 		sp = &sa[i];
1056 
1057 		st = pf_find_state_byid(sp->id, sp->creatorid);
1058 		if (st == NULL) {
1059 			V_pfsyncstats.pfsyncs_badstate++;
1060 			continue;
1061 		}
1062 		st->state_flags |= PFSTATE_NOSYNC;
1063 		pf_unlink_state(st, PF_ENTER_LOCKED);
1064 	}
1065 
1066 	return (len);
1067 }
1068 
1069 static int
1070 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1071 {
1072 	struct mbuf *mp;
1073 	struct pfsync_del_c *sa, *sp;
1074 	struct pf_state *st;
1075 	int len = count * sizeof(*sp);
1076 	int offp, i;
1077 
1078 	mp = m_pulldown(m, offset, len, &offp);
1079 	if (mp == NULL) {
1080 		V_pfsyncstats.pfsyncs_badlen++;
1081 		return (-1);
1082 	}
1083 	sa = (struct pfsync_del_c *)(mp->m_data + offp);
1084 
1085 	for (i = 0; i < count; i++) {
1086 		sp = &sa[i];
1087 
1088 		st = pf_find_state_byid(sp->id, sp->creatorid);
1089 		if (st == NULL) {
1090 			V_pfsyncstats.pfsyncs_badstate++;
1091 			continue;
1092 		}
1093 
1094 		st->state_flags |= PFSTATE_NOSYNC;
1095 		pf_unlink_state(st, PF_ENTER_LOCKED);
1096 	}
1097 
1098 	return (len);
1099 }
1100 
1101 static int
1102 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1103 {
1104 	struct pfsync_softc *sc = V_pfsyncif;
1105 	struct pfsync_bus *bus;
1106 	struct mbuf *mp;
1107 	int len = count * sizeof(*bus);
1108 	int offp;
1109 
1110 	PFSYNC_BLOCK(sc);
1111 
1112 	/* If we're not waiting for a bulk update, who cares. */
1113 	if (sc->sc_ureq_sent == 0) {
1114 		PFSYNC_BUNLOCK(sc);
1115 		return (len);
1116 	}
1117 
1118 	mp = m_pulldown(m, offset, len, &offp);
1119 	if (mp == NULL) {
1120 		PFSYNC_BUNLOCK(sc);
1121 		V_pfsyncstats.pfsyncs_badlen++;
1122 		return (-1);
1123 	}
1124 	bus = (struct pfsync_bus *)(mp->m_data + offp);
1125 
1126 	switch (bus->status) {
1127 	case PFSYNC_BUS_START:
1128 		callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1129 		    V_pf_limits[PF_LIMIT_STATES].limit /
1130 		    ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1131 		    sizeof(struct pfsync_state)),
1132 		    pfsync_bulk_fail, sc);
1133 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1134 			printf("pfsync: received bulk update start\n");
1135 		break;
1136 
1137 	case PFSYNC_BUS_END:
1138 		if (time_uptime - ntohl(bus->endtime) >=
1139 		    sc->sc_ureq_sent) {
1140 			/* that's it, we're happy */
1141 			sc->sc_ureq_sent = 0;
1142 			sc->sc_bulk_tries = 0;
1143 			callout_stop(&sc->sc_bulkfail_tmo);
1144 			if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1145 				(*carp_demote_adj_p)(-V_pfsync_carp_adj,
1146 				    "pfsync bulk done");
1147 			sc->sc_flags |= PFSYNCF_OK;
1148 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1149 				printf("pfsync: received valid "
1150 				    "bulk update end\n");
1151 		} else {
1152 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1153 				printf("pfsync: received invalid "
1154 				    "bulk update end: bad timestamp\n");
1155 		}
1156 		break;
1157 	}
1158 	PFSYNC_BUNLOCK(sc);
1159 
1160 	return (len);
1161 }
1162 
1163 static int
1164 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1165 {
1166 	int len = count * sizeof(struct pfsync_tdb);
1167 
1168 #if defined(IPSEC)
1169 	struct pfsync_tdb *tp;
1170 	struct mbuf *mp;
1171 	int offp;
1172 	int i;
1173 	int s;
1174 
1175 	mp = m_pulldown(m, offset, len, &offp);
1176 	if (mp == NULL) {
1177 		V_pfsyncstats.pfsyncs_badlen++;
1178 		return (-1);
1179 	}
1180 	tp = (struct pfsync_tdb *)(mp->m_data + offp);
1181 
1182 	for (i = 0; i < count; i++)
1183 		pfsync_update_net_tdb(&tp[i]);
1184 #endif
1185 
1186 	return (len);
1187 }
1188 
1189 #if defined(IPSEC)
1190 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1191 static void
1192 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1193 {
1194 	struct tdb		*tdb;
1195 	int			 s;
1196 
1197 	/* check for invalid values */
1198 	if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1199 	    (pt->dst.sa.sa_family != AF_INET &&
1200 	    pt->dst.sa.sa_family != AF_INET6))
1201 		goto bad;
1202 
1203 	tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1204 	if (tdb) {
1205 		pt->rpl = ntohl(pt->rpl);
1206 		pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1207 
1208 		/* Neither replay nor byte counter should ever decrease. */
1209 		if (pt->rpl < tdb->tdb_rpl ||
1210 		    pt->cur_bytes < tdb->tdb_cur_bytes) {
1211 			goto bad;
1212 		}
1213 
1214 		tdb->tdb_rpl = pt->rpl;
1215 		tdb->tdb_cur_bytes = pt->cur_bytes;
1216 	}
1217 	return;
1218 
1219 bad:
1220 	if (V_pf_status.debug >= PF_DEBUG_MISC)
1221 		printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1222 		    "invalid value\n");
1223 	V_pfsyncstats.pfsyncs_badstate++;
1224 	return;
1225 }
1226 #endif
1227 
1228 
1229 static int
1230 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1231 {
1232 	/* check if we are at the right place in the packet */
1233 	if (offset != m->m_pkthdr.len - sizeof(struct pfsync_eof))
1234 		V_pfsyncstats.pfsyncs_badact++;
1235 
1236 	/* we're done. free and let the caller return */
1237 	m_freem(m);
1238 	return (-1);
1239 }
1240 
1241 static int
1242 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1243 {
1244 	V_pfsyncstats.pfsyncs_badact++;
1245 
1246 	m_freem(m);
1247 	return (-1);
1248 }
1249 
1250 static int
1251 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
1252 	struct route *rt)
1253 {
1254 	m_freem(m);
1255 	return (0);
1256 }
1257 
1258 /* ARGSUSED */
1259 static int
1260 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1261 {
1262 	struct pfsync_softc *sc = ifp->if_softc;
1263 	struct ifreq *ifr = (struct ifreq *)data;
1264 	struct pfsyncreq pfsyncr;
1265 	int error;
1266 
1267 	switch (cmd) {
1268 	case SIOCSIFFLAGS:
1269 		PFSYNC_LOCK(sc);
1270 		if (ifp->if_flags & IFF_UP) {
1271 			ifp->if_drv_flags |= IFF_DRV_RUNNING;
1272 			PFSYNC_UNLOCK(sc);
1273 			pfsync_pointers_init();
1274 		} else {
1275 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1276 			PFSYNC_UNLOCK(sc);
1277 			pfsync_pointers_uninit();
1278 		}
1279 		break;
1280 	case SIOCSIFMTU:
1281 		if (!sc->sc_sync_if ||
1282 		    ifr->ifr_mtu <= PFSYNC_MINPKT ||
1283 		    ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1284 			return (EINVAL);
1285 		if (ifr->ifr_mtu < ifp->if_mtu) {
1286 			PFSYNC_LOCK(sc);
1287 			if (sc->sc_len > PFSYNC_MINPKT)
1288 				pfsync_sendout(1);
1289 			PFSYNC_UNLOCK(sc);
1290 		}
1291 		ifp->if_mtu = ifr->ifr_mtu;
1292 		break;
1293 	case SIOCGETPFSYNC:
1294 		bzero(&pfsyncr, sizeof(pfsyncr));
1295 		PFSYNC_LOCK(sc);
1296 		if (sc->sc_sync_if) {
1297 			strlcpy(pfsyncr.pfsyncr_syncdev,
1298 			    sc->sc_sync_if->if_xname, IFNAMSIZ);
1299 		}
1300 		pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1301 		pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1302 		pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER ==
1303 		    (sc->sc_flags & PFSYNCF_DEFER));
1304 		PFSYNC_UNLOCK(sc);
1305 		return (copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)));
1306 
1307 	case SIOCSETPFSYNC:
1308 	    {
1309 		struct ip_moptions *imo = &sc->sc_imo;
1310 		struct ifnet *sifp;
1311 		struct ip *ip;
1312 		void *mship = NULL;
1313 
1314 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1315 			return (error);
1316 		if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr))))
1317 			return (error);
1318 
1319 		if (pfsyncr.pfsyncr_maxupdates > 255)
1320 			return (EINVAL);
1321 
1322 		if (pfsyncr.pfsyncr_syncdev[0] == 0)
1323 			sifp = NULL;
1324 		else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL)
1325 			return (EINVAL);
1326 
1327 		if (pfsyncr.pfsyncr_syncpeer.s_addr == 0 && sifp != NULL)
1328 			mship = malloc((sizeof(struct in_multi *) *
1329 			    IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO);
1330 
1331 		PFSYNC_LOCK(sc);
1332 		if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1333 			sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP);
1334 		else
1335 			sc->sc_sync_peer.s_addr =
1336 			    pfsyncr.pfsyncr_syncpeer.s_addr;
1337 
1338 		sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1339 		if (pfsyncr.pfsyncr_defer) {
1340 			sc->sc_flags |= PFSYNCF_DEFER;
1341 			pfsync_defer_ptr = pfsync_defer;
1342 		} else {
1343 			sc->sc_flags &= ~PFSYNCF_DEFER;
1344 			pfsync_defer_ptr = NULL;
1345 		}
1346 
1347 		if (sifp == NULL) {
1348 			if (sc->sc_sync_if)
1349 				if_rele(sc->sc_sync_if);
1350 			sc->sc_sync_if = NULL;
1351 			if (imo->imo_membership)
1352 				pfsync_multicast_cleanup(sc);
1353 			PFSYNC_UNLOCK(sc);
1354 			break;
1355 		}
1356 
1357 		if (sc->sc_len > PFSYNC_MINPKT &&
1358 		    (sifp->if_mtu < sc->sc_ifp->if_mtu ||
1359 		    (sc->sc_sync_if != NULL &&
1360 		    sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
1361 		    sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
1362 			pfsync_sendout(1);
1363 
1364 		if (imo->imo_membership)
1365 			pfsync_multicast_cleanup(sc);
1366 
1367 		if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
1368 			error = pfsync_multicast_setup(sc, sifp, mship);
1369 			if (error) {
1370 				if_rele(sifp);
1371 				free(mship, M_PFSYNC);
1372 				return (error);
1373 			}
1374 		}
1375 		if (sc->sc_sync_if)
1376 			if_rele(sc->sc_sync_if);
1377 		sc->sc_sync_if = sifp;
1378 
1379 		ip = &sc->sc_template;
1380 		bzero(ip, sizeof(*ip));
1381 		ip->ip_v = IPVERSION;
1382 		ip->ip_hl = sizeof(sc->sc_template) >> 2;
1383 		ip->ip_tos = IPTOS_LOWDELAY;
1384 		/* len and id are set later. */
1385 		ip->ip_off = htons(IP_DF);
1386 		ip->ip_ttl = PFSYNC_DFLTTL;
1387 		ip->ip_p = IPPROTO_PFSYNC;
1388 		ip->ip_src.s_addr = INADDR_ANY;
1389 		ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr;
1390 
1391 		/* Request a full state table update. */
1392 		if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1393 			(*carp_demote_adj_p)(V_pfsync_carp_adj,
1394 			    "pfsync bulk start");
1395 		sc->sc_flags &= ~PFSYNCF_OK;
1396 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1397 			printf("pfsync: requesting bulk update\n");
1398 		pfsync_request_update(0, 0);
1399 		PFSYNC_UNLOCK(sc);
1400 		PFSYNC_BLOCK(sc);
1401 		sc->sc_ureq_sent = time_uptime;
1402 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail,
1403 		    sc);
1404 		PFSYNC_BUNLOCK(sc);
1405 
1406 		break;
1407 	    }
1408 	default:
1409 		return (ENOTTY);
1410 	}
1411 
1412 	return (0);
1413 }
1414 
1415 static void
1416 pfsync_out_state(struct pf_state *st, void *buf)
1417 {
1418 	struct pfsync_state *sp = buf;
1419 
1420 	pfsync_state_export(sp, st);
1421 }
1422 
1423 static void
1424 pfsync_out_iack(struct pf_state *st, void *buf)
1425 {
1426 	struct pfsync_ins_ack *iack = buf;
1427 
1428 	iack->id = st->id;
1429 	iack->creatorid = st->creatorid;
1430 }
1431 
1432 static void
1433 pfsync_out_upd_c(struct pf_state *st, void *buf)
1434 {
1435 	struct pfsync_upd_c *up = buf;
1436 
1437 	bzero(up, sizeof(*up));
1438 	up->id = st->id;
1439 	pf_state_peer_hton(&st->src, &up->src);
1440 	pf_state_peer_hton(&st->dst, &up->dst);
1441 	up->creatorid = st->creatorid;
1442 	up->timeout = st->timeout;
1443 }
1444 
1445 static void
1446 pfsync_out_del(struct pf_state *st, void *buf)
1447 {
1448 	struct pfsync_del_c *dp = buf;
1449 
1450 	dp->id = st->id;
1451 	dp->creatorid = st->creatorid;
1452 	st->state_flags |= PFSTATE_NOSYNC;
1453 }
1454 
1455 static void
1456 pfsync_drop(struct pfsync_softc *sc)
1457 {
1458 	struct pf_state *st, *next;
1459 	struct pfsync_upd_req_item *ur;
1460 	int q;
1461 
1462 	for (q = 0; q < PFSYNC_S_COUNT; q++) {
1463 		if (TAILQ_EMPTY(&sc->sc_qs[q]))
1464 			continue;
1465 
1466 		TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, next) {
1467 			KASSERT(st->sync_state == q,
1468 				("%s: st->sync_state == q",
1469 					__func__));
1470 			st->sync_state = PFSYNC_S_NONE;
1471 			pf_release_state(st);
1472 		}
1473 		TAILQ_INIT(&sc->sc_qs[q]);
1474 	}
1475 
1476 	while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1477 		TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1478 		free(ur, M_PFSYNC);
1479 	}
1480 
1481 	sc->sc_plus = NULL;
1482 	sc->sc_len = PFSYNC_MINPKT;
1483 }
1484 
1485 static void
1486 pfsync_sendout(int schedswi)
1487 {
1488 	struct pfsync_softc *sc = V_pfsyncif;
1489 	struct ifnet *ifp = sc->sc_ifp;
1490 	struct mbuf *m;
1491 	struct ip *ip;
1492 	struct pfsync_header *ph;
1493 	struct pfsync_subheader *subh;
1494 	struct pf_state *st;
1495 	struct pfsync_upd_req_item *ur;
1496 	int offset;
1497 	int q, count = 0;
1498 
1499 	KASSERT(sc != NULL, ("%s: null sc", __func__));
1500 	KASSERT(sc->sc_len > PFSYNC_MINPKT,
1501 	    ("%s: sc_len %zu", __func__, sc->sc_len));
1502 	PFSYNC_LOCK_ASSERT(sc);
1503 
1504 	if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1505 		pfsync_drop(sc);
1506 		return;
1507 	}
1508 
1509 	m = m_get2(M_NOWAIT, MT_DATA, M_PKTHDR, max_linkhdr + sc->sc_len);
1510 	if (m == NULL) {
1511 		sc->sc_ifp->if_oerrors++;
1512 		V_pfsyncstats.pfsyncs_onomem++;
1513 		return;
1514 	}
1515 	m->m_data += max_linkhdr;
1516 	m->m_len = m->m_pkthdr.len = sc->sc_len;
1517 
1518 	/* build the ip header */
1519 	ip = (struct ip *)m->m_data;
1520 	bcopy(&sc->sc_template, ip, sizeof(*ip));
1521 	offset = sizeof(*ip);
1522 
1523 	ip->ip_len = htons(m->m_pkthdr.len);
1524 	ip->ip_id = htons(ip_randomid());
1525 
1526 	/* build the pfsync header */
1527 	ph = (struct pfsync_header *)(m->m_data + offset);
1528 	bzero(ph, sizeof(*ph));
1529 	offset += sizeof(*ph);
1530 
1531 	ph->version = PFSYNC_VERSION;
1532 	ph->len = htons(sc->sc_len - sizeof(*ip));
1533 	bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1534 
1535 	/* walk the queues */
1536 	for (q = 0; q < PFSYNC_S_COUNT; q++) {
1537 		if (TAILQ_EMPTY(&sc->sc_qs[q]))
1538 			continue;
1539 
1540 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1541 		offset += sizeof(*subh);
1542 
1543 		count = 0;
1544 		TAILQ_FOREACH(st, &sc->sc_qs[q], sync_list) {
1545 			KASSERT(st->sync_state == q,
1546 				("%s: st->sync_state == q",
1547 					__func__));
1548 			/*
1549 			 * XXXGL: some of write methods do unlocked reads
1550 			 * of state data :(
1551 			 */
1552 			pfsync_qs[q].write(st, m->m_data + offset);
1553 			offset += pfsync_qs[q].len;
1554 			st->sync_state = PFSYNC_S_NONE;
1555 			pf_release_state(st);
1556 			count++;
1557 		}
1558 		TAILQ_INIT(&sc->sc_qs[q]);
1559 
1560 		bzero(subh, sizeof(*subh));
1561 		subh->action = pfsync_qs[q].action;
1562 		subh->count = htons(count);
1563 		V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1564 	}
1565 
1566 	if (!TAILQ_EMPTY(&sc->sc_upd_req_list)) {
1567 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1568 		offset += sizeof(*subh);
1569 
1570 		count = 0;
1571 		while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1572 			TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1573 
1574 			bcopy(&ur->ur_msg, m->m_data + offset,
1575 			    sizeof(ur->ur_msg));
1576 			offset += sizeof(ur->ur_msg);
1577 			free(ur, M_PFSYNC);
1578 			count++;
1579 		}
1580 
1581 		bzero(subh, sizeof(*subh));
1582 		subh->action = PFSYNC_ACT_UPD_REQ;
1583 		subh->count = htons(count);
1584 		V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1585 	}
1586 
1587 	/* has someone built a custom region for us to add? */
1588 	if (sc->sc_plus != NULL) {
1589 		bcopy(sc->sc_plus, m->m_data + offset, sc->sc_pluslen);
1590 		offset += sc->sc_pluslen;
1591 
1592 		sc->sc_plus = NULL;
1593 	}
1594 
1595 	subh = (struct pfsync_subheader *)(m->m_data + offset);
1596 	offset += sizeof(*subh);
1597 
1598 	bzero(subh, sizeof(*subh));
1599 	subh->action = PFSYNC_ACT_EOF;
1600 	subh->count = htons(1);
1601 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1602 
1603 	/* XXX write checksum in EOF here */
1604 
1605 	/* we're done, let's put it on the wire */
1606 	if (ifp->if_bpf) {
1607 		m->m_data += sizeof(*ip);
1608 		m->m_len = m->m_pkthdr.len = sc->sc_len - sizeof(*ip);
1609 		BPF_MTAP(ifp, m);
1610 		m->m_data -= sizeof(*ip);
1611 		m->m_len = m->m_pkthdr.len = sc->sc_len;
1612 	}
1613 
1614 	if (sc->sc_sync_if == NULL) {
1615 		sc->sc_len = PFSYNC_MINPKT;
1616 		m_freem(m);
1617 		return;
1618 	}
1619 
1620 	sc->sc_ifp->if_opackets++;
1621 	sc->sc_ifp->if_obytes += m->m_pkthdr.len;
1622 	sc->sc_len = PFSYNC_MINPKT;
1623 
1624 	if (!_IF_QFULL(&sc->sc_ifp->if_snd))
1625 		_IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1626 	else {
1627 		m_freem(m);
1628 		sc->sc_ifp->if_snd.ifq_drops++;
1629 	}
1630 	if (schedswi)
1631 		swi_sched(V_pfsync_swi_cookie, 0);
1632 }
1633 
1634 static void
1635 pfsync_insert_state(struct pf_state *st)
1636 {
1637 	struct pfsync_softc *sc = V_pfsyncif;
1638 
1639 	if (st->state_flags & PFSTATE_NOSYNC)
1640 		return;
1641 
1642 	if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1643 	    st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1644 		st->state_flags |= PFSTATE_NOSYNC;
1645 		return;
1646 	}
1647 
1648 	KASSERT(st->sync_state == PFSYNC_S_NONE,
1649 		("%s: st->sync_state == PFSYNC_S_NONE", __func__));
1650 
1651 	PFSYNC_LOCK(sc);
1652 	if (sc->sc_len == PFSYNC_MINPKT)
1653 		callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1654 
1655 	pfsync_q_ins(st, PFSYNC_S_INS);
1656 	PFSYNC_UNLOCK(sc);
1657 
1658 	st->sync_updates = 0;
1659 }
1660 
1661 static int
1662 pfsync_defer(struct pf_state *st, struct mbuf *m)
1663 {
1664 	struct pfsync_softc *sc = V_pfsyncif;
1665 	struct pfsync_deferral *pd;
1666 
1667 	if (m->m_flags & (M_BCAST|M_MCAST))
1668 		return (0);
1669 
1670 	PFSYNC_LOCK(sc);
1671 
1672 	if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) ||
1673 	    !(sc->sc_flags & PFSYNCF_DEFER)) {
1674 		PFSYNC_UNLOCK(sc);
1675 		return (0);
1676 	}
1677 
1678 	 if (sc->sc_deferred >= 128)
1679 		pfsync_undefer(TAILQ_FIRST(&sc->sc_deferrals), 0);
1680 
1681 	pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1682 	if (pd == NULL)
1683 		return (0);
1684 	sc->sc_deferred++;
1685 
1686 	m->m_flags |= M_SKIP_FIREWALL;
1687 	st->state_flags |= PFSTATE_ACK;
1688 
1689 	pd->pd_sc = sc;
1690 	pd->pd_refs = 0;
1691 	pd->pd_st = st;
1692 	pf_ref_state(st);
1693 	pd->pd_m = m;
1694 
1695 	TAILQ_INSERT_TAIL(&sc->sc_deferrals, pd, pd_entry);
1696 	callout_init_mtx(&pd->pd_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1697 	callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd);
1698 
1699 	pfsync_push(sc);
1700 
1701 	return (1);
1702 }
1703 
1704 static void
1705 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1706 {
1707 	struct pfsync_softc *sc = pd->pd_sc;
1708 	struct mbuf *m = pd->pd_m;
1709 	struct pf_state *st = pd->pd_st;
1710 
1711 	PFSYNC_LOCK_ASSERT(sc);
1712 
1713 	TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1714 	sc->sc_deferred--;
1715 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
1716 	free(pd, M_PFSYNC);
1717 	pf_release_state(st);
1718 
1719 	if (drop)
1720 		m_freem(m);
1721 	else {
1722 		_IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1723 		pfsync_push(sc);
1724 	}
1725 }
1726 
1727 static void
1728 pfsync_defer_tmo(void *arg)
1729 {
1730 	struct pfsync_deferral *pd = arg;
1731 	struct pfsync_softc *sc = pd->pd_sc;
1732 	struct mbuf *m = pd->pd_m;
1733 	struct pf_state *st = pd->pd_st;
1734 
1735 	PFSYNC_LOCK_ASSERT(sc);
1736 
1737 	CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
1738 
1739 	TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1740 	sc->sc_deferred--;
1741 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
1742 	if (pd->pd_refs == 0)
1743 		free(pd, M_PFSYNC);
1744 	PFSYNC_UNLOCK(sc);
1745 
1746 	ip_output(m, NULL, NULL, 0, NULL, NULL);
1747 
1748 	pf_release_state(st);
1749 
1750 	CURVNET_RESTORE();
1751 }
1752 
1753 static void
1754 pfsync_undefer_state(struct pf_state *st, int drop)
1755 {
1756 	struct pfsync_softc *sc = V_pfsyncif;
1757 	struct pfsync_deferral *pd;
1758 
1759 	PFSYNC_LOCK_ASSERT(sc);
1760 
1761 	TAILQ_FOREACH(pd, &sc->sc_deferrals, pd_entry) {
1762 		 if (pd->pd_st == st) {
1763 			if (callout_stop(&pd->pd_tmo))
1764 				pfsync_undefer(pd, drop);
1765 			return;
1766 		}
1767 	}
1768 
1769 	panic("%s: unable to find deferred state", __func__);
1770 }
1771 
1772 static void
1773 pfsync_update_state(struct pf_state *st)
1774 {
1775 	struct pfsync_softc *sc = V_pfsyncif;
1776 	int sync = 0;
1777 
1778 	PF_STATE_LOCK_ASSERT(st);
1779 	PFSYNC_LOCK(sc);
1780 
1781 	if (st->state_flags & PFSTATE_ACK)
1782 		pfsync_undefer_state(st, 0);
1783 	if (st->state_flags & PFSTATE_NOSYNC) {
1784 		if (st->sync_state != PFSYNC_S_NONE)
1785 			pfsync_q_del(st);
1786 		PFSYNC_UNLOCK(sc);
1787 		return;
1788 	}
1789 
1790 	if (sc->sc_len == PFSYNC_MINPKT)
1791 		callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1792 
1793 	switch (st->sync_state) {
1794 	case PFSYNC_S_UPD_C:
1795 	case PFSYNC_S_UPD:
1796 	case PFSYNC_S_INS:
1797 		/* we're already handling it */
1798 
1799 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1800 			st->sync_updates++;
1801 			if (st->sync_updates >= sc->sc_maxupdates)
1802 				sync = 1;
1803 		}
1804 		break;
1805 
1806 	case PFSYNC_S_IACK:
1807 		pfsync_q_del(st);
1808 	case PFSYNC_S_NONE:
1809 		pfsync_q_ins(st, PFSYNC_S_UPD_C);
1810 		st->sync_updates = 0;
1811 		break;
1812 
1813 	default:
1814 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1815 	}
1816 
1817 	if (sync || (time_uptime - st->pfsync_time) < 2)
1818 		pfsync_push(sc);
1819 
1820 	PFSYNC_UNLOCK(sc);
1821 }
1822 
1823 static void
1824 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1825 {
1826 	struct pfsync_softc *sc = V_pfsyncif;
1827 	struct pfsync_upd_req_item *item;
1828 	size_t nlen = sizeof(struct pfsync_upd_req);
1829 
1830 	PFSYNC_LOCK_ASSERT(sc);
1831 
1832 	/*
1833 	 * This code does a bit to prevent multiple update requests for the
1834 	 * same state being generated. It searches current subheader queue,
1835 	 * but it doesn't lookup into queue of already packed datagrams.
1836 	 */
1837 	TAILQ_FOREACH(item, &sc->sc_upd_req_list, ur_entry)
1838 		if (item->ur_msg.id == id &&
1839 		    item->ur_msg.creatorid == creatorid)
1840 			return;
1841 
1842 	item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1843 	if (item == NULL)
1844 		return; /* XXX stats */
1845 
1846 	item->ur_msg.id = id;
1847 	item->ur_msg.creatorid = creatorid;
1848 
1849 	if (TAILQ_EMPTY(&sc->sc_upd_req_list))
1850 		nlen += sizeof(struct pfsync_subheader);
1851 
1852 	if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
1853 		pfsync_sendout(1);
1854 
1855 		nlen = sizeof(struct pfsync_subheader) +
1856 		    sizeof(struct pfsync_upd_req);
1857 	}
1858 
1859 	TAILQ_INSERT_TAIL(&sc->sc_upd_req_list, item, ur_entry);
1860 	sc->sc_len += nlen;
1861 }
1862 
1863 static void
1864 pfsync_update_state_req(struct pf_state *st)
1865 {
1866 	struct pfsync_softc *sc = V_pfsyncif;
1867 
1868 	PF_STATE_LOCK_ASSERT(st);
1869 	PFSYNC_LOCK(sc);
1870 
1871 	if (st->state_flags & PFSTATE_NOSYNC) {
1872 		if (st->sync_state != PFSYNC_S_NONE)
1873 			pfsync_q_del(st);
1874 		PFSYNC_UNLOCK(sc);
1875 		return;
1876 	}
1877 
1878 	switch (st->sync_state) {
1879 	case PFSYNC_S_UPD_C:
1880 	case PFSYNC_S_IACK:
1881 		pfsync_q_del(st);
1882 	case PFSYNC_S_NONE:
1883 		pfsync_q_ins(st, PFSYNC_S_UPD);
1884 		pfsync_push(sc);
1885 		break;
1886 
1887 	case PFSYNC_S_INS:
1888 	case PFSYNC_S_UPD:
1889 	case PFSYNC_S_DEL:
1890 		/* we're already handling it */
1891 		break;
1892 
1893 	default:
1894 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1895 	}
1896 
1897 	PFSYNC_UNLOCK(sc);
1898 }
1899 
1900 static void
1901 pfsync_delete_state(struct pf_state *st)
1902 {
1903 	struct pfsync_softc *sc = V_pfsyncif;
1904 
1905 	PFSYNC_LOCK(sc);
1906 	if (st->state_flags & PFSTATE_ACK)
1907 		pfsync_undefer_state(st, 1);
1908 	if (st->state_flags & PFSTATE_NOSYNC) {
1909 		if (st->sync_state != PFSYNC_S_NONE)
1910 			pfsync_q_del(st);
1911 		PFSYNC_UNLOCK(sc);
1912 		return;
1913 	}
1914 
1915 	if (sc->sc_len == PFSYNC_MINPKT)
1916 		callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1917 
1918 	switch (st->sync_state) {
1919 	case PFSYNC_S_INS:
1920 		/* We never got to tell the world so just forget about it. */
1921 		pfsync_q_del(st);
1922 		break;
1923 
1924 	case PFSYNC_S_UPD_C:
1925 	case PFSYNC_S_UPD:
1926 	case PFSYNC_S_IACK:
1927 		pfsync_q_del(st);
1928 		/* FALLTHROUGH to putting it on the del list */
1929 
1930 	case PFSYNC_S_NONE:
1931 		pfsync_q_ins(st, PFSYNC_S_DEL);
1932 		break;
1933 
1934 	default:
1935 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1936 	}
1937 	PFSYNC_UNLOCK(sc);
1938 }
1939 
1940 static void
1941 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
1942 {
1943 	struct pfsync_softc *sc = V_pfsyncif;
1944 	struct {
1945 		struct pfsync_subheader subh;
1946 		struct pfsync_clr clr;
1947 	} __packed r;
1948 
1949 	bzero(&r, sizeof(r));
1950 
1951 	r.subh.action = PFSYNC_ACT_CLR;
1952 	r.subh.count = htons(1);
1953 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
1954 
1955 	strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
1956 	r.clr.creatorid = creatorid;
1957 
1958 	PFSYNC_LOCK(sc);
1959 	pfsync_send_plus(&r, sizeof(r));
1960 	PFSYNC_UNLOCK(sc);
1961 }
1962 
1963 static void
1964 pfsync_q_ins(struct pf_state *st, int q)
1965 {
1966 	struct pfsync_softc *sc = V_pfsyncif;
1967 	size_t nlen = pfsync_qs[q].len;
1968 
1969 	PFSYNC_LOCK_ASSERT(sc);
1970 
1971 	KASSERT(st->sync_state == PFSYNC_S_NONE,
1972 		("%s: st->sync_state == PFSYNC_S_NONE", __func__));
1973 	KASSERT(sc->sc_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
1974 	    sc->sc_len));
1975 
1976 	if (TAILQ_EMPTY(&sc->sc_qs[q]))
1977 		nlen += sizeof(struct pfsync_subheader);
1978 
1979 	if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
1980 		pfsync_sendout(1);
1981 
1982 		nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
1983 	}
1984 
1985 	sc->sc_len += nlen;
1986 	TAILQ_INSERT_TAIL(&sc->sc_qs[q], st, sync_list);
1987 	st->sync_state = q;
1988 	pf_ref_state(st);
1989 }
1990 
1991 static void
1992 pfsync_q_del(struct pf_state *st)
1993 {
1994 	struct pfsync_softc *sc = V_pfsyncif;
1995 	int q = st->sync_state;
1996 
1997 	PFSYNC_LOCK_ASSERT(sc);
1998 	KASSERT(st->sync_state != PFSYNC_S_NONE,
1999 		("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2000 
2001 	sc->sc_len -= pfsync_qs[q].len;
2002 	TAILQ_REMOVE(&sc->sc_qs[q], st, sync_list);
2003 	st->sync_state = PFSYNC_S_NONE;
2004 	pf_release_state(st);
2005 
2006 	if (TAILQ_EMPTY(&sc->sc_qs[q]))
2007 		sc->sc_len -= sizeof(struct pfsync_subheader);
2008 }
2009 
2010 static void
2011 pfsync_bulk_start(void)
2012 {
2013 	struct pfsync_softc *sc = V_pfsyncif;
2014 
2015 	if (V_pf_status.debug >= PF_DEBUG_MISC)
2016 		printf("pfsync: received bulk update request\n");
2017 
2018 	PFSYNC_BLOCK(sc);
2019 
2020 	sc->sc_ureq_received = time_uptime;
2021 	sc->sc_bulk_hashid = 0;
2022 	sc->sc_bulk_stateid = 0;
2023 	pfsync_bulk_status(PFSYNC_BUS_START);
2024 	callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2025 	PFSYNC_BUNLOCK(sc);
2026 }
2027 
2028 static void
2029 pfsync_bulk_update(void *arg)
2030 {
2031 	struct pfsync_softc *sc = arg;
2032 	struct pf_state *s;
2033 	int i, sent = 0;
2034 
2035 	PFSYNC_BLOCK_ASSERT(sc);
2036 	CURVNET_SET(sc->sc_ifp->if_vnet);
2037 
2038 	/*
2039 	 * Start with last state from previous invocation.
2040 	 * It may had gone, in this case start from the
2041 	 * hash slot.
2042 	 */
2043 	s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2044 
2045 	if (s != NULL)
2046 		i = PF_IDHASH(s);
2047 	else
2048 		i = sc->sc_bulk_hashid;
2049 
2050 	for (; i <= V_pf_hashmask; i++) {
2051 		struct pf_idhash *ih = &V_pf_idhash[i];
2052 
2053 		if (s != NULL)
2054 			PF_HASHROW_ASSERT(ih);
2055 		else {
2056 			PF_HASHROW_LOCK(ih);
2057 			s = LIST_FIRST(&ih->states);
2058 		}
2059 
2060 		for (; s; s = LIST_NEXT(s, entry)) {
2061 
2062 			if (sent > 1 && (sc->sc_ifp->if_mtu - sc->sc_len) <
2063 			    sizeof(struct pfsync_state)) {
2064 				/* We've filled a packet. */
2065 				sc->sc_bulk_hashid = i;
2066 				sc->sc_bulk_stateid = s->id;
2067 				sc->sc_bulk_creatorid = s->creatorid;
2068 				PF_HASHROW_UNLOCK(ih);
2069 				callout_reset(&sc->sc_bulk_tmo, 1,
2070 				    pfsync_bulk_update, sc);
2071 				goto full;
2072 			}
2073 
2074 			if (s->sync_state == PFSYNC_S_NONE &&
2075 			    s->timeout < PFTM_MAX &&
2076 			    s->pfsync_time <= sc->sc_ureq_received) {
2077 				PFSYNC_LOCK(sc);
2078 				pfsync_update_state_req(s);
2079 				PFSYNC_UNLOCK(sc);
2080 				sent++;
2081 			}
2082 		}
2083 		PF_HASHROW_UNLOCK(ih);
2084 	}
2085 
2086 	/* We're done. */
2087 	pfsync_bulk_status(PFSYNC_BUS_END);
2088 
2089 full:
2090 	CURVNET_RESTORE();
2091 }
2092 
2093 static void
2094 pfsync_bulk_status(u_int8_t status)
2095 {
2096 	struct {
2097 		struct pfsync_subheader subh;
2098 		struct pfsync_bus bus;
2099 	} __packed r;
2100 
2101 	struct pfsync_softc *sc = V_pfsyncif;
2102 
2103 	bzero(&r, sizeof(r));
2104 
2105 	r.subh.action = PFSYNC_ACT_BUS;
2106 	r.subh.count = htons(1);
2107 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2108 
2109 	r.bus.creatorid = V_pf_status.hostid;
2110 	r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2111 	r.bus.status = status;
2112 
2113 	PFSYNC_LOCK(sc);
2114 	pfsync_send_plus(&r, sizeof(r));
2115 	PFSYNC_UNLOCK(sc);
2116 }
2117 
2118 static void
2119 pfsync_bulk_fail(void *arg)
2120 {
2121 	struct pfsync_softc *sc = arg;
2122 
2123 	CURVNET_SET(sc->sc_ifp->if_vnet);
2124 
2125 	PFSYNC_BLOCK_ASSERT(sc);
2126 
2127 	if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2128 		/* Try again */
2129 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2130 		    pfsync_bulk_fail, V_pfsyncif);
2131 		PFSYNC_LOCK(sc);
2132 		pfsync_request_update(0, 0);
2133 		PFSYNC_UNLOCK(sc);
2134 	} else {
2135 		/* Pretend like the transfer was ok. */
2136 		sc->sc_ureq_sent = 0;
2137 		sc->sc_bulk_tries = 0;
2138 		PFSYNC_LOCK(sc);
2139 		if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2140 			(*carp_demote_adj_p)(-V_pfsync_carp_adj,
2141 			    "pfsync bulk fail");
2142 		sc->sc_flags |= PFSYNCF_OK;
2143 		PFSYNC_UNLOCK(sc);
2144 		if (V_pf_status.debug >= PF_DEBUG_MISC)
2145 			printf("pfsync: failed to receive bulk update\n");
2146 	}
2147 
2148 	CURVNET_RESTORE();
2149 }
2150 
2151 static void
2152 pfsync_send_plus(void *plus, size_t pluslen)
2153 {
2154 	struct pfsync_softc *sc = V_pfsyncif;
2155 
2156 	PFSYNC_LOCK_ASSERT(sc);
2157 
2158 	if (sc->sc_len + pluslen > sc->sc_ifp->if_mtu)
2159 		pfsync_sendout(1);
2160 
2161 	sc->sc_plus = plus;
2162 	sc->sc_len += (sc->sc_pluslen = pluslen);
2163 
2164 	pfsync_sendout(1);
2165 }
2166 
2167 static void
2168 pfsync_timeout(void *arg)
2169 {
2170 	struct pfsync_softc *sc = arg;
2171 
2172 	CURVNET_SET(sc->sc_ifp->if_vnet);
2173 	PFSYNC_LOCK(sc);
2174 	pfsync_push(sc);
2175 	PFSYNC_UNLOCK(sc);
2176 	CURVNET_RESTORE();
2177 }
2178 
2179 static void
2180 pfsync_push(struct pfsync_softc *sc)
2181 {
2182 
2183 	PFSYNC_LOCK_ASSERT(sc);
2184 
2185 	sc->sc_flags |= PFSYNCF_PUSH;
2186 	swi_sched(V_pfsync_swi_cookie, 0);
2187 }
2188 
2189 static void
2190 pfsyncintr(void *arg)
2191 {
2192 	struct pfsync_softc *sc = arg;
2193 	struct mbuf *m, *n;
2194 
2195 	CURVNET_SET(sc->sc_ifp->if_vnet);
2196 
2197 	PFSYNC_LOCK(sc);
2198 	if ((sc->sc_flags & PFSYNCF_PUSH) && sc->sc_len > PFSYNC_MINPKT) {
2199 		pfsync_sendout(0);
2200 		sc->sc_flags &= ~PFSYNCF_PUSH;
2201 	}
2202 	_IF_DEQUEUE_ALL(&sc->sc_ifp->if_snd, m);
2203 	PFSYNC_UNLOCK(sc);
2204 
2205 	for (; m != NULL; m = n) {
2206 
2207 		n = m->m_nextpkt;
2208 		m->m_nextpkt = NULL;
2209 
2210 		/*
2211 		 * We distinguish between a deferral packet and our
2212 		 * own pfsync packet based on M_SKIP_FIREWALL
2213 		 * flag. This is XXX.
2214 		 */
2215 		if (m->m_flags & M_SKIP_FIREWALL)
2216 			ip_output(m, NULL, NULL, 0, NULL, NULL);
2217 		else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo,
2218 		    NULL) == 0)
2219 			V_pfsyncstats.pfsyncs_opackets++;
2220 		else
2221 			V_pfsyncstats.pfsyncs_oerrors++;
2222 	}
2223 	CURVNET_RESTORE();
2224 }
2225 
2226 static int
2227 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship)
2228 {
2229 	struct ip_moptions *imo = &sc->sc_imo;
2230 	int error;
2231 
2232 	if (!(ifp->if_flags & IFF_MULTICAST))
2233 		return (EADDRNOTAVAIL);
2234 
2235 	imo->imo_membership = (struct in_multi **)mship;
2236 	imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
2237 	imo->imo_multicast_vif = -1;
2238 
2239 	if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL,
2240 	    &imo->imo_membership[0])) != 0) {
2241 		imo->imo_membership = NULL;
2242 		return (error);
2243 	}
2244 	imo->imo_num_memberships++;
2245 	imo->imo_multicast_ifp = ifp;
2246 	imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2247 	imo->imo_multicast_loop = 0;
2248 
2249 	return (0);
2250 }
2251 
2252 static void
2253 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2254 {
2255 	struct ip_moptions *imo = &sc->sc_imo;
2256 
2257 	in_leavegroup(imo->imo_membership[0], NULL);
2258 	free(imo->imo_membership, M_PFSYNC);
2259 	imo->imo_membership = NULL;
2260 	imo->imo_multicast_ifp = NULL;
2261 }
2262 
2263 #ifdef INET
2264 extern  struct domain inetdomain;
2265 static struct protosw in_pfsync_protosw = {
2266 	.pr_type =		SOCK_RAW,
2267 	.pr_domain =		&inetdomain,
2268 	.pr_protocol =		IPPROTO_PFSYNC,
2269 	.pr_flags =		PR_ATOMIC|PR_ADDR,
2270 	.pr_input =		pfsync_input,
2271 	.pr_output =		(pr_output_t *)rip_output,
2272 	.pr_ctloutput =		rip_ctloutput,
2273 	.pr_usrreqs =		&rip_usrreqs
2274 };
2275 #endif
2276 
2277 static void
2278 pfsync_pointers_init()
2279 {
2280 
2281 	PF_RULES_WLOCK();
2282 	pfsync_state_import_ptr = pfsync_state_import;
2283 	pfsync_insert_state_ptr = pfsync_insert_state;
2284 	pfsync_update_state_ptr = pfsync_update_state;
2285 	pfsync_delete_state_ptr = pfsync_delete_state;
2286 	pfsync_clear_states_ptr = pfsync_clear_states;
2287 	pfsync_defer_ptr = pfsync_defer;
2288 	PF_RULES_WUNLOCK();
2289 }
2290 
2291 static void
2292 pfsync_pointers_uninit()
2293 {
2294 
2295 	PF_RULES_WLOCK();
2296 	pfsync_state_import_ptr = NULL;
2297 	pfsync_insert_state_ptr = NULL;
2298 	pfsync_update_state_ptr = NULL;
2299 	pfsync_delete_state_ptr = NULL;
2300 	pfsync_clear_states_ptr = NULL;
2301 	pfsync_defer_ptr = NULL;
2302 	PF_RULES_WUNLOCK();
2303 }
2304 
2305 static int
2306 pfsync_init()
2307 {
2308 	VNET_ITERATOR_DECL(vnet_iter);
2309 	int error = 0;
2310 
2311 	VNET_LIST_RLOCK();
2312 	VNET_FOREACH(vnet_iter) {
2313 		CURVNET_SET(vnet_iter);
2314 		V_pfsync_cloner = if_clone_simple(pfsyncname,
2315 		    pfsync_clone_create, pfsync_clone_destroy, 1);
2316 		error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif,
2317 		    SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2318 		CURVNET_RESTORE();
2319 		if (error)
2320 			goto fail_locked;
2321 	}
2322 	VNET_LIST_RUNLOCK();
2323 #ifdef INET
2324 	error = pf_proto_register(PF_INET, &in_pfsync_protosw);
2325 	if (error)
2326 		goto fail;
2327 	error = ipproto_register(IPPROTO_PFSYNC);
2328 	if (error) {
2329 		pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2330 		goto fail;
2331 	}
2332 #endif
2333 	pfsync_pointers_init();
2334 
2335 	return (0);
2336 
2337 fail:
2338 	VNET_LIST_RLOCK();
2339 fail_locked:
2340 	VNET_FOREACH(vnet_iter) {
2341 		CURVNET_SET(vnet_iter);
2342 		if (V_pfsync_swi_cookie) {
2343 			swi_remove(V_pfsync_swi_cookie);
2344 			if_clone_detach(V_pfsync_cloner);
2345 		}
2346 		CURVNET_RESTORE();
2347 	}
2348 	VNET_LIST_RUNLOCK();
2349 
2350 	return (error);
2351 }
2352 
2353 static void
2354 pfsync_uninit()
2355 {
2356 	VNET_ITERATOR_DECL(vnet_iter);
2357 
2358 	pfsync_pointers_uninit();
2359 
2360 	ipproto_unregister(IPPROTO_PFSYNC);
2361 	pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2362 	VNET_LIST_RLOCK();
2363 	VNET_FOREACH(vnet_iter) {
2364 		CURVNET_SET(vnet_iter);
2365 		if_clone_detach(V_pfsync_cloner);
2366 		swi_remove(V_pfsync_swi_cookie);
2367 		CURVNET_RESTORE();
2368 	}
2369 	VNET_LIST_RUNLOCK();
2370 }
2371 
2372 static int
2373 pfsync_modevent(module_t mod, int type, void *data)
2374 {
2375 	int error = 0;
2376 
2377 	switch (type) {
2378 	case MOD_LOAD:
2379 		error = pfsync_init();
2380 		break;
2381 	case MOD_QUIESCE:
2382 		/*
2383 		 * Module should not be unloaded due to race conditions.
2384 		 */
2385 		error = EBUSY;
2386 		break;
2387 	case MOD_UNLOAD:
2388 		pfsync_uninit();
2389 		break;
2390 	default:
2391 		error = EINVAL;
2392 		break;
2393 	}
2394 
2395 	return (error);
2396 }
2397 
2398 static moduledata_t pfsync_mod = {
2399 	pfsyncname,
2400 	pfsync_modevent,
2401 	0
2402 };
2403 
2404 #define PFSYNC_MODVER 1
2405 
2406 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);
2407 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2408 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
2409