xref: /freebsd/sys/netpfil/pf/if_pfsync.c (revision ec965063070e5753c166cf592c9336444b74720a)
1 /*-
2  * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND ISC)
3  *
4  * Copyright (c) 2002 Michael Shalayeff
5  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
21  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
22  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
23  * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
25  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
26  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
27  * THE POSSIBILITY OF SUCH DAMAGE.
28  */
29 
30 /*-
31  * Copyright (c) 2009 David Gwynne <dlg@openbsd.org>
32  *
33  * Permission to use, copy, modify, and distribute this software for any
34  * purpose with or without fee is hereby granted, provided that the above
35  * copyright notice and this permission notice appear in all copies.
36  *
37  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
38  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
39  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
40  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
41  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
42  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
43  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
44  */
45 
46 /*
47  * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $
48  *
49  * Revisions picked from OpenBSD after revision 1.110 import:
50  * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input()
51  * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates
52  * 1.120, 1.175 - use monotonic time_uptime
53  * 1.122 - reduce number of updates for non-TCP sessions
54  * 1.125, 1.127 - rewrite merge or stale processing
55  * 1.128 - cleanups
56  * 1.146 - bzero() mbuf before sparsely filling it with data
57  * 1.170 - SIOCSIFMTU checks
58  * 1.126, 1.142 - deferred packets processing
59  * 1.173 - correct expire time processing
60  */
61 
62 #include <sys/cdefs.h>
63 __FBSDID("$FreeBSD$");
64 
65 #include "opt_inet.h"
66 #include "opt_inet6.h"
67 #include "opt_pf.h"
68 
69 #include <sys/param.h>
70 #include <sys/bus.h>
71 #include <sys/endian.h>
72 #include <sys/interrupt.h>
73 #include <sys/kernel.h>
74 #include <sys/lock.h>
75 #include <sys/mbuf.h>
76 #include <sys/module.h>
77 #include <sys/mutex.h>
78 #include <sys/nv.h>
79 #include <sys/priv.h>
80 #include <sys/smp.h>
81 #include <sys/socket.h>
82 #include <sys/sockio.h>
83 #include <sys/sysctl.h>
84 #include <sys/syslog.h>
85 
86 #include <net/bpf.h>
87 #include <net/if.h>
88 #include <net/if_var.h>
89 #include <net/if_clone.h>
90 #include <net/if_types.h>
91 #include <net/vnet.h>
92 #include <net/pfvar.h>
93 #include <net/if_pfsync.h>
94 
95 #include <netinet/if_ether.h>
96 #include <netinet/in.h>
97 #include <netinet/in_var.h>
98 #include <netinet/ip.h>
99 #include <netinet/ip_carp.h>
100 #include <netinet/ip_var.h>
101 #include <netinet/tcp.h>
102 #include <netinet/tcp_fsm.h>
103 #include <netinet/tcp_seq.h>
104 
105 #include <netpfil/pf/pfsync_nv.h>
106 
107 struct pfsync_bucket;
108 
109 union inet_template {
110 	struct ip      ipv4;
111 };
112 
113 #define PFSYNC_MINPKT ( \
114 	sizeof(union inet_template) + \
115 	sizeof(struct pfsync_header) + \
116 	sizeof(struct pfsync_subheader) )
117 
118 static int	pfsync_upd_tcp(struct pf_kstate *, struct pfsync_state_peer *,
119 		    struct pfsync_state_peer *);
120 static int	pfsync_in_clr(struct mbuf *, int, int, int);
121 static int	pfsync_in_ins(struct mbuf *, int, int, int);
122 static int	pfsync_in_iack(struct mbuf *, int, int, int);
123 static int	pfsync_in_upd(struct mbuf *, int, int, int);
124 static int	pfsync_in_upd_c(struct mbuf *, int, int, int);
125 static int	pfsync_in_ureq(struct mbuf *, int, int, int);
126 static int	pfsync_in_del(struct mbuf *, int, int, int);
127 static int	pfsync_in_del_c(struct mbuf *, int, int, int);
128 static int	pfsync_in_bus(struct mbuf *, int, int, int);
129 static int	pfsync_in_tdb(struct mbuf *, int, int, int);
130 static int	pfsync_in_eof(struct mbuf *, int, int, int);
131 static int	pfsync_in_error(struct mbuf *, int, int, int);
132 
133 static int (*pfsync_acts[])(struct mbuf *, int, int, int) = {
134 	pfsync_in_clr,			/* PFSYNC_ACT_CLR */
135 	pfsync_in_ins,			/* PFSYNC_ACT_INS */
136 	pfsync_in_iack,			/* PFSYNC_ACT_INS_ACK */
137 	pfsync_in_upd,			/* PFSYNC_ACT_UPD */
138 	pfsync_in_upd_c,		/* PFSYNC_ACT_UPD_C */
139 	pfsync_in_ureq,			/* PFSYNC_ACT_UPD_REQ */
140 	pfsync_in_del,			/* PFSYNC_ACT_DEL */
141 	pfsync_in_del_c,		/* PFSYNC_ACT_DEL_C */
142 	pfsync_in_error,		/* PFSYNC_ACT_INS_F */
143 	pfsync_in_error,		/* PFSYNC_ACT_DEL_F */
144 	pfsync_in_bus,			/* PFSYNC_ACT_BUS */
145 	pfsync_in_tdb,			/* PFSYNC_ACT_TDB */
146 	pfsync_in_eof			/* PFSYNC_ACT_EOF */
147 };
148 
149 struct pfsync_q {
150 	void		(*write)(struct pf_kstate *, void *);
151 	size_t		len;
152 	u_int8_t	action;
153 };
154 
155 /* we have one of these for every PFSYNC_S_ */
156 static void	pfsync_out_state(struct pf_kstate *, void *);
157 static void	pfsync_out_iack(struct pf_kstate *, void *);
158 static void	pfsync_out_upd_c(struct pf_kstate *, void *);
159 static void	pfsync_out_del(struct pf_kstate *, void *);
160 
161 static struct pfsync_q pfsync_qs[] = {
162 	{ pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_INS },
163 	{ pfsync_out_iack,  sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
164 	{ pfsync_out_state, sizeof(struct pfsync_state),   PFSYNC_ACT_UPD },
165 	{ pfsync_out_upd_c, sizeof(struct pfsync_upd_c),   PFSYNC_ACT_UPD_C },
166 	{ pfsync_out_del,   sizeof(struct pfsync_del_c),   PFSYNC_ACT_DEL_C }
167 };
168 
169 static void	pfsync_q_ins(struct pf_kstate *, int, bool);
170 static void	pfsync_q_del(struct pf_kstate *, bool, struct pfsync_bucket *);
171 
172 static void	pfsync_update_state(struct pf_kstate *);
173 
174 struct pfsync_upd_req_item {
175 	TAILQ_ENTRY(pfsync_upd_req_item)	ur_entry;
176 	struct pfsync_upd_req			ur_msg;
177 };
178 
179 struct pfsync_deferral {
180 	struct pfsync_softc		*pd_sc;
181 	TAILQ_ENTRY(pfsync_deferral)	pd_entry;
182 	u_int				pd_refs;
183 	struct callout			pd_tmo;
184 
185 	struct pf_kstate		*pd_st;
186 	struct mbuf			*pd_m;
187 };
188 
189 struct pfsync_sofct;
190 
191 struct pfsync_bucket
192 {
193 	int			b_id;
194 	struct pfsync_softc	*b_sc;
195 	struct mtx		b_mtx;
196 	struct callout		b_tmo;
197 	int			b_flags;
198 #define	PFSYNCF_BUCKET_PUSH	0x00000001
199 
200 	size_t			b_len;
201 	TAILQ_HEAD(, pf_kstate)			b_qs[PFSYNC_S_COUNT];
202 	TAILQ_HEAD(, pfsync_upd_req_item)	b_upd_req_list;
203 	TAILQ_HEAD(, pfsync_deferral)		b_deferrals;
204 	u_int			b_deferred;
205 	void			*b_plus;
206 	size_t			b_pluslen;
207 
208 	struct  ifaltq b_snd;
209 };
210 
211 struct pfsync_softc {
212 	/* Configuration */
213 	struct ifnet		*sc_ifp;
214 	struct ifnet		*sc_sync_if;
215 	struct ip_moptions	sc_imo;
216 	struct sockaddr_storage	sc_sync_peer;
217 	uint32_t		sc_flags;
218 	uint8_t			sc_maxupdates;
219 	union inet_template     sc_template;
220 	struct mtx		sc_mtx;
221 
222 	/* Queued data */
223 	struct pfsync_bucket	*sc_buckets;
224 
225 	/* Bulk update info */
226 	struct mtx		sc_bulk_mtx;
227 	uint32_t		sc_ureq_sent;
228 	int			sc_bulk_tries;
229 	uint32_t		sc_ureq_received;
230 	int			sc_bulk_hashid;
231 	uint64_t		sc_bulk_stateid;
232 	uint32_t		sc_bulk_creatorid;
233 	struct callout		sc_bulk_tmo;
234 	struct callout		sc_bulkfail_tmo;
235 };
236 
237 #define	PFSYNC_LOCK(sc)		mtx_lock(&(sc)->sc_mtx)
238 #define	PFSYNC_UNLOCK(sc)	mtx_unlock(&(sc)->sc_mtx)
239 #define	PFSYNC_LOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_mtx, MA_OWNED)
240 
241 #define PFSYNC_BUCKET_LOCK(b)		mtx_lock(&(b)->b_mtx)
242 #define PFSYNC_BUCKET_UNLOCK(b)		mtx_unlock(&(b)->b_mtx)
243 #define PFSYNC_BUCKET_LOCK_ASSERT(b)	mtx_assert(&(b)->b_mtx, MA_OWNED)
244 
245 #define	PFSYNC_BLOCK(sc)	mtx_lock(&(sc)->sc_bulk_mtx)
246 #define	PFSYNC_BUNLOCK(sc)	mtx_unlock(&(sc)->sc_bulk_mtx)
247 #define	PFSYNC_BLOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
248 
249 static const char pfsyncname[] = "pfsync";
250 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
251 VNET_DEFINE_STATIC(struct pfsync_softc	*, pfsyncif) = NULL;
252 #define	V_pfsyncif		VNET(pfsyncif)
253 VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL;
254 #define	V_pfsync_swi_cookie	VNET(pfsync_swi_cookie)
255 VNET_DEFINE_STATIC(struct intr_event *, pfsync_swi_ie);
256 #define	V_pfsync_swi_ie		VNET(pfsync_swi_ie)
257 VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats);
258 #define	V_pfsyncstats		VNET(pfsyncstats)
259 VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW;
260 #define	V_pfsync_carp_adj	VNET(pfsync_carp_adj)
261 
262 static void	pfsync_timeout(void *);
263 static void	pfsync_push(struct pfsync_bucket *);
264 static void	pfsync_push_all(struct pfsync_softc *);
265 static void	pfsyncintr(void *);
266 static int	pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
267 		    struct in_mfilter *imf);
268 static void	pfsync_multicast_cleanup(struct pfsync_softc *);
269 static void	pfsync_pointers_init(void);
270 static void	pfsync_pointers_uninit(void);
271 static int	pfsync_init(void);
272 static void	pfsync_uninit(void);
273 
274 static unsigned long pfsync_buckets;
275 
276 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
277     "PFSYNC");
278 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW,
279     &VNET_NAME(pfsyncstats), pfsyncstats,
280     "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
281 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_VNET | CTLFLAG_RW,
282     &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
283 SYSCTL_ULONG(_net_pfsync, OID_AUTO, pfsync_buckets, CTLFLAG_RDTUN,
284     &pfsync_buckets, 0, "Number of pfsync hash buckets");
285 
286 static int	pfsync_clone_create(struct if_clone *, int, caddr_t);
287 static void	pfsync_clone_destroy(struct ifnet *);
288 static int	pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
289 		    struct pf_state_peer *);
290 static int	pfsyncoutput(struct ifnet *, struct mbuf *,
291 		    const struct sockaddr *, struct route *);
292 static int	pfsyncioctl(struct ifnet *, u_long, caddr_t);
293 
294 static int	pfsync_defer(struct pf_kstate *, struct mbuf *);
295 static void	pfsync_undefer(struct pfsync_deferral *, int);
296 static void	pfsync_undefer_state(struct pf_kstate *, int);
297 static void	pfsync_defer_tmo(void *);
298 
299 static void	pfsync_request_update(u_int32_t, u_int64_t);
300 static bool	pfsync_update_state_req(struct pf_kstate *);
301 
302 static void	pfsync_drop(struct pfsync_softc *);
303 static void	pfsync_sendout(int, int);
304 static void	pfsync_send_plus(void *, size_t);
305 
306 static void	pfsync_bulk_start(void);
307 static void	pfsync_bulk_status(u_int8_t);
308 static void	pfsync_bulk_update(void *);
309 static void	pfsync_bulk_fail(void *);
310 
311 static void	pfsync_detach_ifnet(struct ifnet *);
312 
313 static int pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *,
314     struct pfsync_kstatus *);
315 static int pfsync_kstatus_to_softc(struct pfsync_kstatus *,
316     struct pfsync_softc *);
317 
318 #ifdef IPSEC
319 static void	pfsync_update_net_tdb(struct pfsync_tdb *);
320 #endif
321 static struct pfsync_bucket	*pfsync_get_bucket(struct pfsync_softc *,
322 		    struct pf_kstate *);
323 
324 #define PFSYNC_MAX_BULKTRIES	12
325 #define PFSYNC_DEFER_TIMEOUT	((20 * hz) / 1000)
326 
327 VNET_DEFINE(struct if_clone *, pfsync_cloner);
328 #define	V_pfsync_cloner	VNET(pfsync_cloner)
329 
330 static int
331 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
332 {
333 	struct pfsync_softc *sc;
334 	struct ifnet *ifp;
335 	struct pfsync_bucket *b;
336 	int c, q;
337 
338 	if (unit != 0)
339 		return (EINVAL);
340 
341 	if (! pfsync_buckets)
342 		pfsync_buckets = mp_ncpus * 2;
343 
344 	sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
345 	sc->sc_flags |= PFSYNCF_OK;
346 	sc->sc_maxupdates = 128;
347 
348 	ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
349 	if (ifp == NULL) {
350 		free(sc, M_PFSYNC);
351 		return (ENOSPC);
352 	}
353 	if_initname(ifp, pfsyncname, unit);
354 	ifp->if_softc = sc;
355 	ifp->if_ioctl = pfsyncioctl;
356 	ifp->if_output = pfsyncoutput;
357 	ifp->if_type = IFT_PFSYNC;
358 	ifp->if_hdrlen = sizeof(struct pfsync_header);
359 	ifp->if_mtu = ETHERMTU;
360 	mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
361 	mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
362 	callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
363 	callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
364 
365 	if_attach(ifp);
366 
367 	bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
368 
369 	sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets),
370 	    M_PFSYNC, M_ZERO | M_WAITOK);
371 	for (c = 0; c < pfsync_buckets; c++) {
372 		b = &sc->sc_buckets[c];
373 		mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF);
374 
375 		b->b_id = c;
376 		b->b_sc = sc;
377 		b->b_len = PFSYNC_MINPKT;
378 
379 		for (q = 0; q < PFSYNC_S_COUNT; q++)
380 			TAILQ_INIT(&b->b_qs[q]);
381 
382 		TAILQ_INIT(&b->b_upd_req_list);
383 		TAILQ_INIT(&b->b_deferrals);
384 
385 		callout_init(&b->b_tmo, 1);
386 
387 		b->b_snd.ifq_maxlen = ifqmaxlen;
388 	}
389 
390 	V_pfsyncif = sc;
391 
392 	return (0);
393 }
394 
395 static void
396 pfsync_clone_destroy(struct ifnet *ifp)
397 {
398 	struct pfsync_softc *sc = ifp->if_softc;
399 	struct pfsync_bucket *b;
400 	int c;
401 
402 	for (c = 0; c < pfsync_buckets; c++) {
403 		b = &sc->sc_buckets[c];
404 		/*
405 		 * At this stage, everything should have already been
406 		 * cleared by pfsync_uninit(), and we have only to
407 		 * drain callouts.
408 		 */
409 		while (b->b_deferred > 0) {
410 			struct pfsync_deferral *pd =
411 			    TAILQ_FIRST(&b->b_deferrals);
412 
413 			TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
414 			b->b_deferred--;
415 			if (callout_stop(&pd->pd_tmo) > 0) {
416 				pf_release_state(pd->pd_st);
417 				m_freem(pd->pd_m);
418 				free(pd, M_PFSYNC);
419 			} else {
420 				pd->pd_refs++;
421 				callout_drain(&pd->pd_tmo);
422 				free(pd, M_PFSYNC);
423 			}
424 		}
425 
426 		callout_drain(&b->b_tmo);
427 	}
428 
429 	callout_drain(&sc->sc_bulkfail_tmo);
430 	callout_drain(&sc->sc_bulk_tmo);
431 
432 	if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
433 		(*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
434 	bpfdetach(ifp);
435 	if_detach(ifp);
436 
437 	pfsync_drop(sc);
438 
439 	if_free(ifp);
440 	pfsync_multicast_cleanup(sc);
441 	mtx_destroy(&sc->sc_mtx);
442 	mtx_destroy(&sc->sc_bulk_mtx);
443 
444 	free(sc->sc_buckets, M_PFSYNC);
445 	free(sc, M_PFSYNC);
446 
447 	V_pfsyncif = NULL;
448 }
449 
450 static int
451 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
452     struct pf_state_peer *d)
453 {
454 	if (s->scrub.scrub_flag && d->scrub == NULL) {
455 		d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
456 		if (d->scrub == NULL)
457 			return (ENOMEM);
458 	}
459 
460 	return (0);
461 }
462 
463 static int
464 pfsync_state_import(struct pfsync_state *sp, int flags)
465 {
466 	struct pfsync_softc *sc = V_pfsyncif;
467 #ifndef	__NO_STRICT_ALIGNMENT
468 	struct pfsync_state_key key[2];
469 #endif
470 	struct pfsync_state_key *kw, *ks;
471 	struct pf_kstate	*st = NULL;
472 	struct pf_state_key *skw = NULL, *sks = NULL;
473 	struct pf_krule *r = NULL;
474 	struct pfi_kkif	*kif;
475 	int error;
476 
477 	PF_RULES_RASSERT();
478 
479 	if (sp->creatorid == 0) {
480 		if (V_pf_status.debug >= PF_DEBUG_MISC)
481 			printf("%s: invalid creator id: %08x\n", __func__,
482 			    ntohl(sp->creatorid));
483 		return (EINVAL);
484 	}
485 
486 	if ((kif = pfi_kkif_find(sp->ifname)) == NULL) {
487 		if (V_pf_status.debug >= PF_DEBUG_MISC)
488 			printf("%s: unknown interface: %s\n", __func__,
489 			    sp->ifname);
490 		if (flags & PFSYNC_SI_IOCTL)
491 			return (EINVAL);
492 		return (0);	/* skip this state */
493 	}
494 
495 	/*
496 	 * If the ruleset checksums match or the state is coming from the ioctl,
497 	 * it's safe to associate the state with the rule of that number.
498 	 */
499 	if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
500 	    (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
501 	    pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
502 		r = pf_main_ruleset.rules[
503 		    PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
504 	else
505 		r = &V_pf_default_rule;
506 
507 	if ((r->max_states &&
508 	    counter_u64_fetch(r->states_cur) >= r->max_states))
509 		goto cleanup;
510 
511 	/*
512 	 * XXXGL: consider M_WAITOK in ioctl path after.
513 	 */
514 	st = pf_alloc_state(M_NOWAIT);
515 	if (__predict_false(st == NULL))
516 		goto cleanup;
517 
518 	if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
519 		goto cleanup;
520 
521 #ifndef	__NO_STRICT_ALIGNMENT
522 	bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2);
523 	kw = &key[PF_SK_WIRE];
524 	ks = &key[PF_SK_STACK];
525 #else
526 	kw = &sp->key[PF_SK_WIRE];
527 	ks = &sp->key[PF_SK_STACK];
528 #endif
529 
530 	if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) ||
531 	    PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) ||
532 	    kw->port[0] != ks->port[0] ||
533 	    kw->port[1] != ks->port[1]) {
534 		sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
535 		if (sks == NULL)
536 			goto cleanup;
537 	} else
538 		sks = skw;
539 
540 	/* allocate memory for scrub info */
541 	if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
542 	    pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
543 		goto cleanup;
544 
545 	/* Copy to state key(s). */
546 	skw->addr[0] = kw->addr[0];
547 	skw->addr[1] = kw->addr[1];
548 	skw->port[0] = kw->port[0];
549 	skw->port[1] = kw->port[1];
550 	skw->proto = sp->proto;
551 	skw->af = sp->af;
552 	if (sks != skw) {
553 		sks->addr[0] = ks->addr[0];
554 		sks->addr[1] = ks->addr[1];
555 		sks->port[0] = ks->port[0];
556 		sks->port[1] = ks->port[1];
557 		sks->proto = sp->proto;
558 		sks->af = sp->af;
559 	}
560 
561 	/* copy to state */
562 	bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
563 	st->creation = time_uptime - ntohl(sp->creation);
564 	st->expire = time_uptime;
565 	if (sp->expire) {
566 		uint32_t timeout;
567 
568 		timeout = r->timeout[sp->timeout];
569 		if (!timeout)
570 			timeout = V_pf_default_rule.timeout[sp->timeout];
571 
572 		/* sp->expire may have been adaptively scaled by export. */
573 		st->expire -= timeout - ntohl(sp->expire);
574 	}
575 
576 	st->direction = sp->direction;
577 	st->log = sp->log;
578 	st->timeout = sp->timeout;
579 	st->state_flags = sp->state_flags;
580 
581 	st->id = sp->id;
582 	st->creatorid = sp->creatorid;
583 	pf_state_peer_ntoh(&sp->src, &st->src);
584 	pf_state_peer_ntoh(&sp->dst, &st->dst);
585 
586 	st->rule.ptr = r;
587 	st->nat_rule.ptr = NULL;
588 	st->anchor.ptr = NULL;
589 	st->rt_kif = NULL;
590 
591 	st->pfsync_time = time_uptime;
592 	st->sync_state = PFSYNC_S_NONE;
593 
594 	if (!(flags & PFSYNC_SI_IOCTL))
595 		st->state_flags |= PFSTATE_NOSYNC;
596 
597 	if ((error = pf_state_insert(kif, kif, skw, sks, st)) != 0)
598 		goto cleanup_state;
599 
600 	/* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
601 	counter_u64_add(r->states_cur, 1);
602 	counter_u64_add(r->states_tot, 1);
603 
604 	if (!(flags & PFSYNC_SI_IOCTL)) {
605 		st->state_flags &= ~PFSTATE_NOSYNC;
606 		if (st->state_flags & PFSTATE_ACK) {
607 			pfsync_q_ins(st, PFSYNC_S_IACK, true);
608 			pfsync_push_all(sc);
609 		}
610 	}
611 	st->state_flags &= ~PFSTATE_ACK;
612 	PF_STATE_UNLOCK(st);
613 
614 	return (0);
615 
616 cleanup:
617 	error = ENOMEM;
618 	if (skw == sks)
619 		sks = NULL;
620 	if (skw != NULL)
621 		uma_zfree(V_pf_state_key_z, skw);
622 	if (sks != NULL)
623 		uma_zfree(V_pf_state_key_z, sks);
624 
625 cleanup_state:	/* pf_state_insert() frees the state keys. */
626 	if (st) {
627 		st->timeout = PFTM_UNLINKED; /* appease an assert */
628 		pf_free_state(st);
629 	}
630 	return (error);
631 }
632 
633 #ifdef INET
634 static int
635 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
636 {
637 	struct pfsync_softc *sc = V_pfsyncif;
638 	struct mbuf *m = *mp;
639 	struct ip *ip = mtod(m, struct ip *);
640 	struct pfsync_header *ph;
641 	struct pfsync_subheader subh;
642 
643 	int offset, len, flags = 0;
644 	int rv;
645 	uint16_t count;
646 
647 	PF_RULES_RLOCK_TRACKER;
648 
649 	*mp = NULL;
650 	V_pfsyncstats.pfsyncs_ipackets++;
651 
652 	/* Verify that we have a sync interface configured. */
653 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
654 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
655 		goto done;
656 
657 	/* verify that the packet came in on the right interface */
658 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
659 		V_pfsyncstats.pfsyncs_badif++;
660 		goto done;
661 	}
662 
663 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
664 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
665 	/* verify that the IP TTL is 255. */
666 	if (ip->ip_ttl != PFSYNC_DFLTTL) {
667 		V_pfsyncstats.pfsyncs_badttl++;
668 		goto done;
669 	}
670 
671 	offset = ip->ip_hl << 2;
672 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
673 		V_pfsyncstats.pfsyncs_hdrops++;
674 		goto done;
675 	}
676 
677 	if (offset + sizeof(*ph) > m->m_len) {
678 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
679 			V_pfsyncstats.pfsyncs_hdrops++;
680 			return (IPPROTO_DONE);
681 		}
682 		ip = mtod(m, struct ip *);
683 	}
684 	ph = (struct pfsync_header *)((char *)ip + offset);
685 
686 	/* verify the version */
687 	if (ph->version != PFSYNC_VERSION) {
688 		V_pfsyncstats.pfsyncs_badver++;
689 		goto done;
690 	}
691 
692 	len = ntohs(ph->len) + offset;
693 	if (m->m_pkthdr.len < len) {
694 		V_pfsyncstats.pfsyncs_badlen++;
695 		goto done;
696 	}
697 
698 	/*
699 	 * Trusting pf_chksum during packet processing, as well as seeking
700 	 * in interface name tree, require holding PF_RULES_RLOCK().
701 	 */
702 	PF_RULES_RLOCK();
703 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
704 		flags = PFSYNC_SI_CKSUM;
705 
706 	offset += sizeof(*ph);
707 	while (offset <= len - sizeof(subh)) {
708 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
709 		offset += sizeof(subh);
710 
711 		if (subh.action >= PFSYNC_ACT_MAX) {
712 			V_pfsyncstats.pfsyncs_badact++;
713 			PF_RULES_RUNLOCK();
714 			goto done;
715 		}
716 
717 		count = ntohs(subh.count);
718 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
719 		rv = (*pfsync_acts[subh.action])(m, offset, count, flags);
720 		if (rv == -1) {
721 			PF_RULES_RUNLOCK();
722 			return (IPPROTO_DONE);
723 		}
724 
725 		offset += rv;
726 	}
727 	PF_RULES_RUNLOCK();
728 
729 done:
730 	m_freem(m);
731 	return (IPPROTO_DONE);
732 }
733 #endif
734 
735 static int
736 pfsync_in_clr(struct mbuf *m, int offset, int count, int flags)
737 {
738 	struct pfsync_clr *clr;
739 	struct mbuf *mp;
740 	int len = sizeof(*clr) * count;
741 	int i, offp;
742 	u_int32_t creatorid;
743 
744 	mp = m_pulldown(m, offset, len, &offp);
745 	if (mp == NULL) {
746 		V_pfsyncstats.pfsyncs_badlen++;
747 		return (-1);
748 	}
749 	clr = (struct pfsync_clr *)(mp->m_data + offp);
750 
751 	for (i = 0; i < count; i++) {
752 		creatorid = clr[i].creatorid;
753 
754 		if (clr[i].ifname[0] != '\0' &&
755 		    pfi_kkif_find(clr[i].ifname) == NULL)
756 			continue;
757 
758 		for (int i = 0; i <= pf_hashmask; i++) {
759 			struct pf_idhash *ih = &V_pf_idhash[i];
760 			struct pf_kstate *s;
761 relock:
762 			PF_HASHROW_LOCK(ih);
763 			LIST_FOREACH(s, &ih->states, entry) {
764 				if (s->creatorid == creatorid) {
765 					s->state_flags |= PFSTATE_NOSYNC;
766 					pf_unlink_state(s);
767 					goto relock;
768 				}
769 			}
770 			PF_HASHROW_UNLOCK(ih);
771 		}
772 	}
773 
774 	return (len);
775 }
776 
777 static int
778 pfsync_in_ins(struct mbuf *m, int offset, int count, int flags)
779 {
780 	struct mbuf *mp;
781 	struct pfsync_state *sa, *sp;
782 	int len = sizeof(*sp) * count;
783 	int i, offp;
784 
785 	mp = m_pulldown(m, offset, len, &offp);
786 	if (mp == NULL) {
787 		V_pfsyncstats.pfsyncs_badlen++;
788 		return (-1);
789 	}
790 	sa = (struct pfsync_state *)(mp->m_data + offp);
791 
792 	for (i = 0; i < count; i++) {
793 		sp = &sa[i];
794 
795 		/* Check for invalid values. */
796 		if (sp->timeout >= PFTM_MAX ||
797 		    sp->src.state > PF_TCPS_PROXY_DST ||
798 		    sp->dst.state > PF_TCPS_PROXY_DST ||
799 		    sp->direction > PF_OUT ||
800 		    (sp->af != AF_INET && sp->af != AF_INET6)) {
801 			if (V_pf_status.debug >= PF_DEBUG_MISC)
802 				printf("%s: invalid value\n", __func__);
803 			V_pfsyncstats.pfsyncs_badval++;
804 			continue;
805 		}
806 
807 		if (pfsync_state_import(sp, flags) == ENOMEM)
808 			/* Drop out, but process the rest of the actions. */
809 			break;
810 	}
811 
812 	return (len);
813 }
814 
815 static int
816 pfsync_in_iack(struct mbuf *m, int offset, int count, int flags)
817 {
818 	struct pfsync_ins_ack *ia, *iaa;
819 	struct pf_kstate *st;
820 
821 	struct mbuf *mp;
822 	int len = count * sizeof(*ia);
823 	int offp, i;
824 
825 	mp = m_pulldown(m, offset, len, &offp);
826 	if (mp == NULL) {
827 		V_pfsyncstats.pfsyncs_badlen++;
828 		return (-1);
829 	}
830 	iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
831 
832 	for (i = 0; i < count; i++) {
833 		ia = &iaa[i];
834 
835 		st = pf_find_state_byid(ia->id, ia->creatorid);
836 		if (st == NULL)
837 			continue;
838 
839 		if (st->state_flags & PFSTATE_ACK) {
840 			pfsync_undefer_state(st, 0);
841 		}
842 		PF_STATE_UNLOCK(st);
843 	}
844 	/*
845 	 * XXX this is not yet implemented, but we know the size of the
846 	 * message so we can skip it.
847 	 */
848 
849 	return (count * sizeof(struct pfsync_ins_ack));
850 }
851 
852 static int
853 pfsync_upd_tcp(struct pf_kstate *st, struct pfsync_state_peer *src,
854     struct pfsync_state_peer *dst)
855 {
856 	int sync = 0;
857 
858 	PF_STATE_LOCK_ASSERT(st);
859 
860 	/*
861 	 * The state should never go backwards except
862 	 * for syn-proxy states.  Neither should the
863 	 * sequence window slide backwards.
864 	 */
865 	if ((st->src.state > src->state &&
866 	    (st->src.state < PF_TCPS_PROXY_SRC ||
867 	    src->state >= PF_TCPS_PROXY_SRC)) ||
868 
869 	    (st->src.state == src->state &&
870 	    SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
871 		sync++;
872 	else
873 		pf_state_peer_ntoh(src, &st->src);
874 
875 	if ((st->dst.state > dst->state) ||
876 
877 	    (st->dst.state >= TCPS_SYN_SENT &&
878 	    SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
879 		sync++;
880 	else
881 		pf_state_peer_ntoh(dst, &st->dst);
882 
883 	return (sync);
884 }
885 
886 static int
887 pfsync_in_upd(struct mbuf *m, int offset, int count, int flags)
888 {
889 	struct pfsync_softc *sc = V_pfsyncif;
890 	struct pfsync_state *sa, *sp;
891 	struct pf_kstate *st;
892 	int sync;
893 
894 	struct mbuf *mp;
895 	int len = count * sizeof(*sp);
896 	int offp, i;
897 
898 	mp = m_pulldown(m, offset, len, &offp);
899 	if (mp == NULL) {
900 		V_pfsyncstats.pfsyncs_badlen++;
901 		return (-1);
902 	}
903 	sa = (struct pfsync_state *)(mp->m_data + offp);
904 
905 	for (i = 0; i < count; i++) {
906 		sp = &sa[i];
907 
908 		/* check for invalid values */
909 		if (sp->timeout >= PFTM_MAX ||
910 		    sp->src.state > PF_TCPS_PROXY_DST ||
911 		    sp->dst.state > PF_TCPS_PROXY_DST) {
912 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
913 				printf("pfsync_input: PFSYNC_ACT_UPD: "
914 				    "invalid value\n");
915 			}
916 			V_pfsyncstats.pfsyncs_badval++;
917 			continue;
918 		}
919 
920 		st = pf_find_state_byid(sp->id, sp->creatorid);
921 		if (st == NULL) {
922 			/* insert the update */
923 			if (pfsync_state_import(sp, flags))
924 				V_pfsyncstats.pfsyncs_badstate++;
925 			continue;
926 		}
927 
928 		if (st->state_flags & PFSTATE_ACK) {
929 			pfsync_undefer_state(st, 1);
930 		}
931 
932 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
933 			sync = pfsync_upd_tcp(st, &sp->src, &sp->dst);
934 		else {
935 			sync = 0;
936 
937 			/*
938 			 * Non-TCP protocol state machine always go
939 			 * forwards
940 			 */
941 			if (st->src.state > sp->src.state)
942 				sync++;
943 			else
944 				pf_state_peer_ntoh(&sp->src, &st->src);
945 			if (st->dst.state > sp->dst.state)
946 				sync++;
947 			else
948 				pf_state_peer_ntoh(&sp->dst, &st->dst);
949 		}
950 		if (sync < 2) {
951 			pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
952 			pf_state_peer_ntoh(&sp->dst, &st->dst);
953 			st->expire = time_uptime;
954 			st->timeout = sp->timeout;
955 		}
956 		st->pfsync_time = time_uptime;
957 
958 		if (sync) {
959 			V_pfsyncstats.pfsyncs_stale++;
960 
961 			pfsync_update_state(st);
962 			PF_STATE_UNLOCK(st);
963 			pfsync_push_all(sc);
964 			continue;
965 		}
966 		PF_STATE_UNLOCK(st);
967 	}
968 
969 	return (len);
970 }
971 
972 static int
973 pfsync_in_upd_c(struct mbuf *m, int offset, int count, int flags)
974 {
975 	struct pfsync_softc *sc = V_pfsyncif;
976 	struct pfsync_upd_c *ua, *up;
977 	struct pf_kstate *st;
978 	int len = count * sizeof(*up);
979 	int sync;
980 	struct mbuf *mp;
981 	int offp, i;
982 
983 	mp = m_pulldown(m, offset, len, &offp);
984 	if (mp == NULL) {
985 		V_pfsyncstats.pfsyncs_badlen++;
986 		return (-1);
987 	}
988 	ua = (struct pfsync_upd_c *)(mp->m_data + offp);
989 
990 	for (i = 0; i < count; i++) {
991 		up = &ua[i];
992 
993 		/* check for invalid values */
994 		if (up->timeout >= PFTM_MAX ||
995 		    up->src.state > PF_TCPS_PROXY_DST ||
996 		    up->dst.state > PF_TCPS_PROXY_DST) {
997 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
998 				printf("pfsync_input: "
999 				    "PFSYNC_ACT_UPD_C: "
1000 				    "invalid value\n");
1001 			}
1002 			V_pfsyncstats.pfsyncs_badval++;
1003 			continue;
1004 		}
1005 
1006 		st = pf_find_state_byid(up->id, up->creatorid);
1007 		if (st == NULL) {
1008 			/* We don't have this state. Ask for it. */
1009 			PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1010 			pfsync_request_update(up->creatorid, up->id);
1011 			PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1012 			continue;
1013 		}
1014 
1015 		if (st->state_flags & PFSTATE_ACK) {
1016 			pfsync_undefer_state(st, 1);
1017 		}
1018 
1019 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1020 			sync = pfsync_upd_tcp(st, &up->src, &up->dst);
1021 		else {
1022 			sync = 0;
1023 
1024 			/*
1025 			 * Non-TCP protocol state machine always go
1026 			 * forwards
1027 			 */
1028 			if (st->src.state > up->src.state)
1029 				sync++;
1030 			else
1031 				pf_state_peer_ntoh(&up->src, &st->src);
1032 			if (st->dst.state > up->dst.state)
1033 				sync++;
1034 			else
1035 				pf_state_peer_ntoh(&up->dst, &st->dst);
1036 		}
1037 		if (sync < 2) {
1038 			pfsync_alloc_scrub_memory(&up->dst, &st->dst);
1039 			pf_state_peer_ntoh(&up->dst, &st->dst);
1040 			st->expire = time_uptime;
1041 			st->timeout = up->timeout;
1042 		}
1043 		st->pfsync_time = time_uptime;
1044 
1045 		if (sync) {
1046 			V_pfsyncstats.pfsyncs_stale++;
1047 
1048 			pfsync_update_state(st);
1049 			PF_STATE_UNLOCK(st);
1050 			pfsync_push_all(sc);
1051 			continue;
1052 		}
1053 		PF_STATE_UNLOCK(st);
1054 	}
1055 
1056 	return (len);
1057 }
1058 
1059 static int
1060 pfsync_in_ureq(struct mbuf *m, int offset, int count, int flags)
1061 {
1062 	struct pfsync_upd_req *ur, *ura;
1063 	struct mbuf *mp;
1064 	int len = count * sizeof(*ur);
1065 	int i, offp;
1066 
1067 	struct pf_kstate *st;
1068 
1069 	mp = m_pulldown(m, offset, len, &offp);
1070 	if (mp == NULL) {
1071 		V_pfsyncstats.pfsyncs_badlen++;
1072 		return (-1);
1073 	}
1074 	ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1075 
1076 	for (i = 0; i < count; i++) {
1077 		ur = &ura[i];
1078 
1079 		if (ur->id == 0 && ur->creatorid == 0)
1080 			pfsync_bulk_start();
1081 		else {
1082 			st = pf_find_state_byid(ur->id, ur->creatorid);
1083 			if (st == NULL) {
1084 				V_pfsyncstats.pfsyncs_badstate++;
1085 				continue;
1086 			}
1087 			if (st->state_flags & PFSTATE_NOSYNC) {
1088 				PF_STATE_UNLOCK(st);
1089 				continue;
1090 			}
1091 
1092 			pfsync_update_state_req(st);
1093 			PF_STATE_UNLOCK(st);
1094 		}
1095 	}
1096 
1097 	return (len);
1098 }
1099 
1100 static int
1101 pfsync_in_del(struct mbuf *m, int offset, int count, int flags)
1102 {
1103 	struct mbuf *mp;
1104 	struct pfsync_state *sa, *sp;
1105 	struct pf_kstate *st;
1106 	int len = count * sizeof(*sp);
1107 	int offp, i;
1108 
1109 	mp = m_pulldown(m, offset, len, &offp);
1110 	if (mp == NULL) {
1111 		V_pfsyncstats.pfsyncs_badlen++;
1112 		return (-1);
1113 	}
1114 	sa = (struct pfsync_state *)(mp->m_data + offp);
1115 
1116 	for (i = 0; i < count; i++) {
1117 		sp = &sa[i];
1118 
1119 		st = pf_find_state_byid(sp->id, sp->creatorid);
1120 		if (st == NULL) {
1121 			V_pfsyncstats.pfsyncs_badstate++;
1122 			continue;
1123 		}
1124 		st->state_flags |= PFSTATE_NOSYNC;
1125 		pf_unlink_state(st);
1126 	}
1127 
1128 	return (len);
1129 }
1130 
1131 static int
1132 pfsync_in_del_c(struct mbuf *m, int offset, int count, int flags)
1133 {
1134 	struct mbuf *mp;
1135 	struct pfsync_del_c *sa, *sp;
1136 	struct pf_kstate *st;
1137 	int len = count * sizeof(*sp);
1138 	int offp, i;
1139 
1140 	mp = m_pulldown(m, offset, len, &offp);
1141 	if (mp == NULL) {
1142 		V_pfsyncstats.pfsyncs_badlen++;
1143 		return (-1);
1144 	}
1145 	sa = (struct pfsync_del_c *)(mp->m_data + offp);
1146 
1147 	for (i = 0; i < count; i++) {
1148 		sp = &sa[i];
1149 
1150 		st = pf_find_state_byid(sp->id, sp->creatorid);
1151 		if (st == NULL) {
1152 			V_pfsyncstats.pfsyncs_badstate++;
1153 			continue;
1154 		}
1155 
1156 		st->state_flags |= PFSTATE_NOSYNC;
1157 		pf_unlink_state(st);
1158 	}
1159 
1160 	return (len);
1161 }
1162 
1163 static int
1164 pfsync_in_bus(struct mbuf *m, int offset, int count, int flags)
1165 {
1166 	struct pfsync_softc *sc = V_pfsyncif;
1167 	struct pfsync_bus *bus;
1168 	struct mbuf *mp;
1169 	int len = count * sizeof(*bus);
1170 	int offp;
1171 
1172 	PFSYNC_BLOCK(sc);
1173 
1174 	/* If we're not waiting for a bulk update, who cares. */
1175 	if (sc->sc_ureq_sent == 0) {
1176 		PFSYNC_BUNLOCK(sc);
1177 		return (len);
1178 	}
1179 
1180 	mp = m_pulldown(m, offset, len, &offp);
1181 	if (mp == NULL) {
1182 		PFSYNC_BUNLOCK(sc);
1183 		V_pfsyncstats.pfsyncs_badlen++;
1184 		return (-1);
1185 	}
1186 	bus = (struct pfsync_bus *)(mp->m_data + offp);
1187 
1188 	switch (bus->status) {
1189 	case PFSYNC_BUS_START:
1190 		callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1191 		    V_pf_limits[PF_LIMIT_STATES].limit /
1192 		    ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1193 		    sizeof(struct pfsync_state)),
1194 		    pfsync_bulk_fail, sc);
1195 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1196 			printf("pfsync: received bulk update start\n");
1197 		break;
1198 
1199 	case PFSYNC_BUS_END:
1200 		if (time_uptime - ntohl(bus->endtime) >=
1201 		    sc->sc_ureq_sent) {
1202 			/* that's it, we're happy */
1203 			sc->sc_ureq_sent = 0;
1204 			sc->sc_bulk_tries = 0;
1205 			callout_stop(&sc->sc_bulkfail_tmo);
1206 			if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1207 				(*carp_demote_adj_p)(-V_pfsync_carp_adj,
1208 				    "pfsync bulk done");
1209 			sc->sc_flags |= PFSYNCF_OK;
1210 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1211 				printf("pfsync: received valid "
1212 				    "bulk update end\n");
1213 		} else {
1214 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1215 				printf("pfsync: received invalid "
1216 				    "bulk update end: bad timestamp\n");
1217 		}
1218 		break;
1219 	}
1220 	PFSYNC_BUNLOCK(sc);
1221 
1222 	return (len);
1223 }
1224 
1225 static int
1226 pfsync_in_tdb(struct mbuf *m, int offset, int count, int flags)
1227 {
1228 	int len = count * sizeof(struct pfsync_tdb);
1229 
1230 #if defined(IPSEC)
1231 	struct pfsync_tdb *tp;
1232 	struct mbuf *mp;
1233 	int offp;
1234 	int i;
1235 	int s;
1236 
1237 	mp = m_pulldown(m, offset, len, &offp);
1238 	if (mp == NULL) {
1239 		V_pfsyncstats.pfsyncs_badlen++;
1240 		return (-1);
1241 	}
1242 	tp = (struct pfsync_tdb *)(mp->m_data + offp);
1243 
1244 	for (i = 0; i < count; i++)
1245 		pfsync_update_net_tdb(&tp[i]);
1246 #endif
1247 
1248 	return (len);
1249 }
1250 
1251 #if defined(IPSEC)
1252 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1253 static void
1254 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1255 {
1256 	struct tdb		*tdb;
1257 	int			 s;
1258 
1259 	/* check for invalid values */
1260 	if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1261 	    (pt->dst.sa.sa_family != AF_INET &&
1262 	    pt->dst.sa.sa_family != AF_INET6))
1263 		goto bad;
1264 
1265 	tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1266 	if (tdb) {
1267 		pt->rpl = ntohl(pt->rpl);
1268 		pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1269 
1270 		/* Neither replay nor byte counter should ever decrease. */
1271 		if (pt->rpl < tdb->tdb_rpl ||
1272 		    pt->cur_bytes < tdb->tdb_cur_bytes) {
1273 			goto bad;
1274 		}
1275 
1276 		tdb->tdb_rpl = pt->rpl;
1277 		tdb->tdb_cur_bytes = pt->cur_bytes;
1278 	}
1279 	return;
1280 
1281 bad:
1282 	if (V_pf_status.debug >= PF_DEBUG_MISC)
1283 		printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1284 		    "invalid value\n");
1285 	V_pfsyncstats.pfsyncs_badstate++;
1286 	return;
1287 }
1288 #endif
1289 
1290 static int
1291 pfsync_in_eof(struct mbuf *m, int offset, int count, int flags)
1292 {
1293 	/* check if we are at the right place in the packet */
1294 	if (offset != m->m_pkthdr.len)
1295 		V_pfsyncstats.pfsyncs_badlen++;
1296 
1297 	/* we're done. free and let the caller return */
1298 	m_freem(m);
1299 	return (-1);
1300 }
1301 
1302 static int
1303 pfsync_in_error(struct mbuf *m, int offset, int count, int flags)
1304 {
1305 	V_pfsyncstats.pfsyncs_badact++;
1306 
1307 	m_freem(m);
1308 	return (-1);
1309 }
1310 
1311 static int
1312 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1313 	struct route *rt)
1314 {
1315 	m_freem(m);
1316 	return (0);
1317 }
1318 
1319 /* ARGSUSED */
1320 static int
1321 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1322 {
1323 	struct pfsync_softc *sc = ifp->if_softc;
1324 	struct ifreq *ifr = (struct ifreq *)data;
1325 	struct pfsyncreq pfsyncr;
1326 	size_t nvbuflen;
1327 	int error;
1328 	int c;
1329 
1330 	switch (cmd) {
1331 	case SIOCSIFFLAGS:
1332 		PFSYNC_LOCK(sc);
1333 		if (ifp->if_flags & IFF_UP) {
1334 			ifp->if_drv_flags |= IFF_DRV_RUNNING;
1335 			PFSYNC_UNLOCK(sc);
1336 			pfsync_pointers_init();
1337 		} else {
1338 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1339 			PFSYNC_UNLOCK(sc);
1340 			pfsync_pointers_uninit();
1341 		}
1342 		break;
1343 	case SIOCSIFMTU:
1344 		if (!sc->sc_sync_if ||
1345 		    ifr->ifr_mtu <= PFSYNC_MINPKT ||
1346 		    ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1347 			return (EINVAL);
1348 		if (ifr->ifr_mtu < ifp->if_mtu) {
1349 			for (c = 0; c < pfsync_buckets; c++) {
1350 				PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1351 				if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT)
1352 					pfsync_sendout(1, c);
1353 				PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1354 			}
1355 		}
1356 		ifp->if_mtu = ifr->ifr_mtu;
1357 		break;
1358 	case SIOCGETPFSYNC:
1359 		bzero(&pfsyncr, sizeof(pfsyncr));
1360 		PFSYNC_LOCK(sc);
1361 		if (sc->sc_sync_if) {
1362 			strlcpy(pfsyncr.pfsyncr_syncdev,
1363 			    sc->sc_sync_if->if_xname, IFNAMSIZ);
1364 		}
1365 		pfsyncr.pfsyncr_syncpeer = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
1366 		pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1367 		pfsyncr.pfsyncr_defer = sc->sc_flags;
1368 		PFSYNC_UNLOCK(sc);
1369 		return (copyout(&pfsyncr, ifr_data_get_ptr(ifr),
1370 		    sizeof(pfsyncr)));
1371 
1372 	case SIOCGETPFSYNCNV:
1373 	    {
1374 		nvlist_t *nvl_syncpeer;
1375 		nvlist_t *nvl = nvlist_create(0);
1376 
1377 		if (nvl == NULL)
1378 			return (ENOMEM);
1379 
1380 		if (sc->sc_sync_if)
1381 			nvlist_add_string(nvl, "syncdev", sc->sc_sync_if->if_xname);
1382 		nvlist_add_number(nvl, "maxupdates", sc->sc_maxupdates);
1383 		nvlist_add_number(nvl, "flags", sc->sc_flags);
1384 		if ((nvl_syncpeer = pfsync_sockaddr_to_syncpeer_nvlist(&sc->sc_sync_peer)) != NULL)
1385 			nvlist_add_nvlist(nvl, "syncpeer", nvl_syncpeer);
1386 
1387 		void *packed = NULL;
1388 		packed = nvlist_pack(nvl, &nvbuflen);
1389 		if (packed == NULL) {
1390 			free(packed, M_NVLIST);
1391 			nvlist_destroy(nvl);
1392 			return (ENOMEM);
1393 		}
1394 
1395 		if (nvbuflen > ifr->ifr_cap_nv.buf_length) {
1396 			ifr->ifr_cap_nv.length = nvbuflen;
1397 			ifr->ifr_cap_nv.buffer = NULL;
1398 			free(packed, M_NVLIST);
1399 			nvlist_destroy(nvl);
1400 			return (EFBIG);
1401 		}
1402 
1403 		ifr->ifr_cap_nv.length = nvbuflen;
1404 		error = copyout(packed, ifr->ifr_cap_nv.buffer, nvbuflen);
1405 
1406 		nvlist_destroy(nvl);
1407 		nvlist_destroy(nvl_syncpeer);
1408 		free(packed, M_NVLIST);
1409 		break;
1410 	    }
1411 
1412 	case SIOCSETPFSYNC:
1413 	    {
1414 		struct pfsync_kstatus status;
1415 
1416 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1417 			return (error);
1418 		if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr,
1419 		    sizeof(pfsyncr))))
1420 			return (error);
1421 
1422 		memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1423 		pfsync_pfsyncreq_to_kstatus(&pfsyncr, &status);
1424 
1425 		error = pfsync_kstatus_to_softc(&status, sc);
1426 		return (error);
1427 	    }
1428 	case SIOCSETPFSYNCNV:
1429 	    {
1430 		struct pfsync_kstatus status;
1431 		void *data;
1432 		nvlist_t *nvl;
1433 
1434 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1435 			return (error);
1436 		if (ifr->ifr_cap_nv.length > IFR_CAP_NV_MAXBUFSIZE)
1437 			return (EINVAL);
1438 
1439 		data = malloc(ifr->ifr_cap_nv.length, M_TEMP, M_WAITOK);
1440 
1441 		if ((error = copyin(ifr->ifr_cap_nv.buffer, data,
1442 		    ifr->ifr_cap_nv.length)) != 0) {
1443 			free(data, M_TEMP);
1444 			return (error);
1445 		}
1446 
1447 		if ((nvl = nvlist_unpack(data, ifr->ifr_cap_nv.length, 0)) == NULL) {
1448 			free(data, M_TEMP);
1449 			return (EINVAL);
1450 		}
1451 
1452 		memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1453 		pfsync_nvstatus_to_kstatus(nvl, &status);
1454 
1455 		nvlist_destroy(nvl);
1456 		free(data, M_TEMP);
1457 
1458 		error = pfsync_kstatus_to_softc(&status, sc);
1459 		return (error);
1460 	    }
1461 	default:
1462 		return (ENOTTY);
1463 	}
1464 
1465 	return (0);
1466 }
1467 
1468 static void
1469 pfsync_out_state(struct pf_kstate *st, void *buf)
1470 {
1471 	struct pfsync_state *sp = buf;
1472 
1473 	pfsync_state_export(sp, st);
1474 }
1475 
1476 static void
1477 pfsync_out_iack(struct pf_kstate *st, void *buf)
1478 {
1479 	struct pfsync_ins_ack *iack = buf;
1480 
1481 	iack->id = st->id;
1482 	iack->creatorid = st->creatorid;
1483 }
1484 
1485 static void
1486 pfsync_out_upd_c(struct pf_kstate *st, void *buf)
1487 {
1488 	struct pfsync_upd_c *up = buf;
1489 
1490 	bzero(up, sizeof(*up));
1491 	up->id = st->id;
1492 	pf_state_peer_hton(&st->src, &up->src);
1493 	pf_state_peer_hton(&st->dst, &up->dst);
1494 	up->creatorid = st->creatorid;
1495 	up->timeout = st->timeout;
1496 }
1497 
1498 static void
1499 pfsync_out_del(struct pf_kstate *st, void *buf)
1500 {
1501 	struct pfsync_del_c *dp = buf;
1502 
1503 	dp->id = st->id;
1504 	dp->creatorid = st->creatorid;
1505 	st->state_flags |= PFSTATE_NOSYNC;
1506 }
1507 
1508 static void
1509 pfsync_drop(struct pfsync_softc *sc)
1510 {
1511 	struct pf_kstate *st, *next;
1512 	struct pfsync_upd_req_item *ur;
1513 	struct pfsync_bucket *b;
1514 	int c, q;
1515 
1516 	for (c = 0; c < pfsync_buckets; c++) {
1517 		b = &sc->sc_buckets[c];
1518 		for (q = 0; q < PFSYNC_S_COUNT; q++) {
1519 			if (TAILQ_EMPTY(&b->b_qs[q]))
1520 				continue;
1521 
1522 			TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) {
1523 				KASSERT(st->sync_state == q,
1524 					("%s: st->sync_state == q",
1525 						__func__));
1526 				st->sync_state = PFSYNC_S_NONE;
1527 				pf_release_state(st);
1528 			}
1529 			TAILQ_INIT(&b->b_qs[q]);
1530 		}
1531 
1532 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1533 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1534 			free(ur, M_PFSYNC);
1535 		}
1536 
1537 		b->b_len = PFSYNC_MINPKT;
1538 		b->b_plus = NULL;
1539 	}
1540 }
1541 
1542 static void
1543 pfsync_sendout(int schedswi, int c)
1544 {
1545 	struct pfsync_softc *sc = V_pfsyncif;
1546 	struct ifnet *ifp = sc->sc_ifp;
1547 	struct mbuf *m;
1548 	struct pfsync_header *ph;
1549 	struct pfsync_subheader *subh;
1550 	struct pf_kstate *st, *st_next;
1551 	struct pfsync_upd_req_item *ur;
1552 	struct pfsync_bucket *b = &sc->sc_buckets[c];
1553 	int aflen, offset;
1554 	int q, count = 0;
1555 
1556 	KASSERT(sc != NULL, ("%s: null sc", __func__));
1557 	KASSERT(b->b_len > PFSYNC_MINPKT,
1558 	    ("%s: sc_len %zu", __func__, b->b_len));
1559 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1560 
1561 	if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1562 		pfsync_drop(sc);
1563 		return;
1564 	}
1565 
1566 	m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1567 	if (m == NULL) {
1568 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1569 		V_pfsyncstats.pfsyncs_onomem++;
1570 		return;
1571 	}
1572 	m->m_data += max_linkhdr;
1573 	m->m_len = m->m_pkthdr.len = b->b_len;
1574 
1575 	/* build the ip header */
1576 	switch (sc->sc_sync_peer.ss_family) {
1577 #ifdef INET
1578 	case AF_INET:
1579 	    {
1580 		struct ip *ip;
1581 
1582 		ip = mtod(m, struct ip *);
1583 		bcopy(&sc->sc_template.ipv4, ip, sizeof(*ip));
1584 		aflen = offset = sizeof(*ip);
1585 
1586 		ip->ip_len = htons(m->m_pkthdr.len);
1587 		ip_fillid(ip);
1588 		break;
1589 	    }
1590 #endif
1591 	default:
1592 		m_freem(m);
1593 		return;
1594 	}
1595 
1596 
1597 	/* build the pfsync header */
1598 	ph = (struct pfsync_header *)(m->m_data + offset);
1599 	bzero(ph, sizeof(*ph));
1600 	offset += sizeof(*ph);
1601 
1602 	ph->version = PFSYNC_VERSION;
1603 	ph->len = htons(b->b_len - aflen);
1604 	bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1605 
1606 	/* walk the queues */
1607 	for (q = 0; q < PFSYNC_S_COUNT; q++) {
1608 		if (TAILQ_EMPTY(&b->b_qs[q]))
1609 			continue;
1610 
1611 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1612 		offset += sizeof(*subh);
1613 
1614 		count = 0;
1615 		TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) {
1616 			KASSERT(st->sync_state == q,
1617 				("%s: st->sync_state == q",
1618 					__func__));
1619 			/*
1620 			 * XXXGL: some of write methods do unlocked reads
1621 			 * of state data :(
1622 			 */
1623 			pfsync_qs[q].write(st, m->m_data + offset);
1624 			offset += pfsync_qs[q].len;
1625 			st->sync_state = PFSYNC_S_NONE;
1626 			pf_release_state(st);
1627 			count++;
1628 		}
1629 		TAILQ_INIT(&b->b_qs[q]);
1630 
1631 		bzero(subh, sizeof(*subh));
1632 		subh->action = pfsync_qs[q].action;
1633 		subh->count = htons(count);
1634 		V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1635 	}
1636 
1637 	if (!TAILQ_EMPTY(&b->b_upd_req_list)) {
1638 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1639 		offset += sizeof(*subh);
1640 
1641 		count = 0;
1642 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1643 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1644 
1645 			bcopy(&ur->ur_msg, m->m_data + offset,
1646 			    sizeof(ur->ur_msg));
1647 			offset += sizeof(ur->ur_msg);
1648 			free(ur, M_PFSYNC);
1649 			count++;
1650 		}
1651 
1652 		bzero(subh, sizeof(*subh));
1653 		subh->action = PFSYNC_ACT_UPD_REQ;
1654 		subh->count = htons(count);
1655 		V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1656 	}
1657 
1658 	/* has someone built a custom region for us to add? */
1659 	if (b->b_plus != NULL) {
1660 		bcopy(b->b_plus, m->m_data + offset, b->b_pluslen);
1661 		offset += b->b_pluslen;
1662 
1663 		b->b_plus = NULL;
1664 	}
1665 
1666 	subh = (struct pfsync_subheader *)(m->m_data + offset);
1667 	offset += sizeof(*subh);
1668 
1669 	bzero(subh, sizeof(*subh));
1670 	subh->action = PFSYNC_ACT_EOF;
1671 	subh->count = htons(1);
1672 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1673 
1674 	/* we're done, let's put it on the wire */
1675 	if (ifp->if_bpf) {
1676 		m->m_data += aflen;
1677 		m->m_len = m->m_pkthdr.len = b->b_len - aflen;
1678 		BPF_MTAP(ifp, m);
1679 		m->m_data -= aflen;
1680 		m->m_len = m->m_pkthdr.len = b->b_len;
1681 	}
1682 
1683 	if (sc->sc_sync_if == NULL) {
1684 		b->b_len = PFSYNC_MINPKT;
1685 		m_freem(m);
1686 		return;
1687 	}
1688 
1689 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1690 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1691 	b->b_len = PFSYNC_MINPKT;
1692 
1693 	if (!_IF_QFULL(&b->b_snd))
1694 		_IF_ENQUEUE(&b->b_snd, m);
1695 	else {
1696 		m_freem(m);
1697 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1698 	}
1699 	if (schedswi)
1700 		swi_sched(V_pfsync_swi_cookie, 0);
1701 }
1702 
1703 static void
1704 pfsync_insert_state(struct pf_kstate *st)
1705 {
1706 	struct pfsync_softc *sc = V_pfsyncif;
1707 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1708 
1709 	if (st->state_flags & PFSTATE_NOSYNC)
1710 		return;
1711 
1712 	if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1713 	    st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1714 		st->state_flags |= PFSTATE_NOSYNC;
1715 		return;
1716 	}
1717 
1718 	KASSERT(st->sync_state == PFSYNC_S_NONE,
1719 		("%s: st->sync_state %u", __func__, st->sync_state));
1720 
1721 	PFSYNC_BUCKET_LOCK(b);
1722 	if (b->b_len == PFSYNC_MINPKT)
1723 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1724 
1725 	pfsync_q_ins(st, PFSYNC_S_INS, true);
1726 	PFSYNC_BUCKET_UNLOCK(b);
1727 
1728 	st->sync_updates = 0;
1729 }
1730 
1731 static int
1732 pfsync_defer(struct pf_kstate *st, struct mbuf *m)
1733 {
1734 	struct pfsync_softc *sc = V_pfsyncif;
1735 	struct pfsync_deferral *pd;
1736 	struct pfsync_bucket *b;
1737 
1738 	if (m->m_flags & (M_BCAST|M_MCAST))
1739 		return (0);
1740 
1741 	if (sc == NULL)
1742 		return (0);
1743 
1744 	b = pfsync_get_bucket(sc, st);
1745 
1746 	PFSYNC_LOCK(sc);
1747 
1748 	if (!(sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) ||
1749 	    !(sc->sc_flags & PFSYNCF_DEFER)) {
1750 		PFSYNC_UNLOCK(sc);
1751 		return (0);
1752 	}
1753 
1754 	PFSYNC_BUCKET_LOCK(b);
1755 	PFSYNC_UNLOCK(sc);
1756 
1757 	if (b->b_deferred >= 128)
1758 		pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0);
1759 
1760 	pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1761 	if (pd == NULL) {
1762 		PFSYNC_BUCKET_UNLOCK(b);
1763 		return (0);
1764 	}
1765 	b->b_deferred++;
1766 
1767 	m->m_flags |= M_SKIP_FIREWALL;
1768 	st->state_flags |= PFSTATE_ACK;
1769 
1770 	pd->pd_sc = sc;
1771 	pd->pd_refs = 0;
1772 	pd->pd_st = st;
1773 	pf_ref_state(st);
1774 	pd->pd_m = m;
1775 
1776 	TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry);
1777 	callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED);
1778 	callout_reset(&pd->pd_tmo, PFSYNC_DEFER_TIMEOUT, pfsync_defer_tmo, pd);
1779 
1780 	pfsync_push(b);
1781 	PFSYNC_BUCKET_UNLOCK(b);
1782 
1783 	return (1);
1784 }
1785 
1786 static void
1787 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1788 {
1789 	struct pfsync_softc *sc = pd->pd_sc;
1790 	struct mbuf *m = pd->pd_m;
1791 	struct pf_kstate *st = pd->pd_st;
1792 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1793 
1794 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1795 
1796 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
1797 	b->b_deferred--;
1798 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
1799 	free(pd, M_PFSYNC);
1800 	pf_release_state(st);
1801 
1802 	if (drop)
1803 		m_freem(m);
1804 	else {
1805 		_IF_ENQUEUE(&b->b_snd, m);
1806 		pfsync_push(b);
1807 	}
1808 }
1809 
1810 static void
1811 pfsync_defer_tmo(void *arg)
1812 {
1813 	struct epoch_tracker et;
1814 	struct pfsync_deferral *pd = arg;
1815 	struct pfsync_softc *sc = pd->pd_sc;
1816 	struct mbuf *m = pd->pd_m;
1817 	struct pf_kstate *st = pd->pd_st;
1818 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1819 
1820 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1821 
1822 	if (sc->sc_sync_if == NULL)
1823 		return;
1824 
1825 	NET_EPOCH_ENTER(et);
1826 	CURVNET_SET(sc->sc_sync_if->if_vnet);
1827 
1828 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
1829 	b->b_deferred--;
1830 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
1831 	if (pd->pd_refs == 0)
1832 		free(pd, M_PFSYNC);
1833 	PFSYNC_BUCKET_UNLOCK(b);
1834 
1835 	switch (sc->sc_sync_peer.ss_family) {
1836 #ifdef INET
1837 	case AF_INET:
1838 		ip_output(m, NULL, NULL, 0, NULL, NULL);
1839 		break;
1840 #endif
1841 	}
1842 
1843 	pf_release_state(st);
1844 
1845 	CURVNET_RESTORE();
1846 	NET_EPOCH_EXIT(et);
1847 }
1848 
1849 static void
1850 pfsync_undefer_state(struct pf_kstate *st, int drop)
1851 {
1852 	struct pfsync_softc *sc = V_pfsyncif;
1853 	struct pfsync_deferral *pd;
1854 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1855 
1856 	PFSYNC_BUCKET_LOCK(b);
1857 
1858 	TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) {
1859 		 if (pd->pd_st == st) {
1860 			if (callout_stop(&pd->pd_tmo) > 0)
1861 				pfsync_undefer(pd, drop);
1862 
1863 			PFSYNC_BUCKET_UNLOCK(b);
1864 			return;
1865 		}
1866 	}
1867 	PFSYNC_BUCKET_UNLOCK(b);
1868 
1869 	panic("%s: unable to find deferred state", __func__);
1870 }
1871 
1872 static struct pfsync_bucket*
1873 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_kstate *st)
1874 {
1875 	int c = PF_IDHASH(st) % pfsync_buckets;
1876 	return &sc->sc_buckets[c];
1877 }
1878 
1879 static void
1880 pfsync_update_state(struct pf_kstate *st)
1881 {
1882 	struct pfsync_softc *sc = V_pfsyncif;
1883 	bool sync = false, ref = true;
1884 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1885 
1886 	PF_STATE_LOCK_ASSERT(st);
1887 	PFSYNC_BUCKET_LOCK(b);
1888 
1889 	if (st->state_flags & PFSTATE_ACK)
1890 		pfsync_undefer_state(st, 0);
1891 	if (st->state_flags & PFSTATE_NOSYNC) {
1892 		if (st->sync_state != PFSYNC_S_NONE)
1893 			pfsync_q_del(st, true, b);
1894 		PFSYNC_BUCKET_UNLOCK(b);
1895 		return;
1896 	}
1897 
1898 	if (b->b_len == PFSYNC_MINPKT)
1899 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1900 
1901 	switch (st->sync_state) {
1902 	case PFSYNC_S_UPD_C:
1903 	case PFSYNC_S_UPD:
1904 	case PFSYNC_S_INS:
1905 		/* we're already handling it */
1906 
1907 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1908 			st->sync_updates++;
1909 			if (st->sync_updates >= sc->sc_maxupdates)
1910 				sync = true;
1911 		}
1912 		break;
1913 
1914 	case PFSYNC_S_IACK:
1915 		pfsync_q_del(st, false, b);
1916 		ref = false;
1917 		/* FALLTHROUGH */
1918 
1919 	case PFSYNC_S_NONE:
1920 		pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
1921 		st->sync_updates = 0;
1922 		break;
1923 
1924 	default:
1925 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
1926 	}
1927 
1928 	if (sync || (time_uptime - st->pfsync_time) < 2)
1929 		pfsync_push(b);
1930 
1931 	PFSYNC_BUCKET_UNLOCK(b);
1932 }
1933 
1934 static void
1935 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1936 {
1937 	struct pfsync_softc *sc = V_pfsyncif;
1938 	struct pfsync_bucket *b = &sc->sc_buckets[0];
1939 	struct pfsync_upd_req_item *item;
1940 	size_t nlen = sizeof(struct pfsync_upd_req);
1941 
1942 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1943 
1944 	/*
1945 	 * This code does a bit to prevent multiple update requests for the
1946 	 * same state being generated. It searches current subheader queue,
1947 	 * but it doesn't lookup into queue of already packed datagrams.
1948 	 */
1949 	TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry)
1950 		if (item->ur_msg.id == id &&
1951 		    item->ur_msg.creatorid == creatorid)
1952 			return;
1953 
1954 	item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1955 	if (item == NULL)
1956 		return; /* XXX stats */
1957 
1958 	item->ur_msg.id = id;
1959 	item->ur_msg.creatorid = creatorid;
1960 
1961 	if (TAILQ_EMPTY(&b->b_upd_req_list))
1962 		nlen += sizeof(struct pfsync_subheader);
1963 
1964 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
1965 		pfsync_sendout(0, 0);
1966 
1967 		nlen = sizeof(struct pfsync_subheader) +
1968 		    sizeof(struct pfsync_upd_req);
1969 	}
1970 
1971 	TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry);
1972 	b->b_len += nlen;
1973 
1974 	pfsync_push(b);
1975 }
1976 
1977 static bool
1978 pfsync_update_state_req(struct pf_kstate *st)
1979 {
1980 	struct pfsync_softc *sc = V_pfsyncif;
1981 	bool ref = true, full = false;
1982 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1983 
1984 	PF_STATE_LOCK_ASSERT(st);
1985 	PFSYNC_BUCKET_LOCK(b);
1986 
1987 	if (st->state_flags & PFSTATE_NOSYNC) {
1988 		if (st->sync_state != PFSYNC_S_NONE)
1989 			pfsync_q_del(st, true, b);
1990 		PFSYNC_BUCKET_UNLOCK(b);
1991 		return (full);
1992 	}
1993 
1994 	switch (st->sync_state) {
1995 	case PFSYNC_S_UPD_C:
1996 	case PFSYNC_S_IACK:
1997 		pfsync_q_del(st, false, b);
1998 		ref = false;
1999 		/* FALLTHROUGH */
2000 
2001 	case PFSYNC_S_NONE:
2002 		pfsync_q_ins(st, PFSYNC_S_UPD, ref);
2003 		pfsync_push(b);
2004 		break;
2005 
2006 	case PFSYNC_S_INS:
2007 	case PFSYNC_S_UPD:
2008 	case PFSYNC_S_DEL:
2009 		/* we're already handling it */
2010 		break;
2011 
2012 	default:
2013 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2014 	}
2015 
2016 	if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(struct pfsync_state))
2017 		full = true;
2018 
2019 	PFSYNC_BUCKET_UNLOCK(b);
2020 
2021 	return (full);
2022 }
2023 
2024 static void
2025 pfsync_delete_state(struct pf_kstate *st)
2026 {
2027 	struct pfsync_softc *sc = V_pfsyncif;
2028 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2029 	bool ref = true;
2030 
2031 	PFSYNC_BUCKET_LOCK(b);
2032 	if (st->state_flags & PFSTATE_ACK)
2033 		pfsync_undefer_state(st, 1);
2034 	if (st->state_flags & PFSTATE_NOSYNC) {
2035 		if (st->sync_state != PFSYNC_S_NONE)
2036 			pfsync_q_del(st, true, b);
2037 		PFSYNC_BUCKET_UNLOCK(b);
2038 		return;
2039 	}
2040 
2041 	if (b->b_len == PFSYNC_MINPKT)
2042 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2043 
2044 	switch (st->sync_state) {
2045 	case PFSYNC_S_INS:
2046 		/* We never got to tell the world so just forget about it. */
2047 		pfsync_q_del(st, true, b);
2048 		break;
2049 
2050 	case PFSYNC_S_UPD_C:
2051 	case PFSYNC_S_UPD:
2052 	case PFSYNC_S_IACK:
2053 		pfsync_q_del(st, false, b);
2054 		ref = false;
2055 		/* FALLTHROUGH */
2056 
2057 	case PFSYNC_S_NONE:
2058 		pfsync_q_ins(st, PFSYNC_S_DEL, ref);
2059 		break;
2060 
2061 	default:
2062 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2063 	}
2064 
2065 	PFSYNC_BUCKET_UNLOCK(b);
2066 }
2067 
2068 static void
2069 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
2070 {
2071 	struct {
2072 		struct pfsync_subheader subh;
2073 		struct pfsync_clr clr;
2074 	} __packed r;
2075 
2076 	bzero(&r, sizeof(r));
2077 
2078 	r.subh.action = PFSYNC_ACT_CLR;
2079 	r.subh.count = htons(1);
2080 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
2081 
2082 	strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
2083 	r.clr.creatorid = creatorid;
2084 
2085 	pfsync_send_plus(&r, sizeof(r));
2086 }
2087 
2088 static void
2089 pfsync_q_ins(struct pf_kstate *st, int q, bool ref)
2090 {
2091 	struct pfsync_softc *sc = V_pfsyncif;
2092 	size_t nlen = pfsync_qs[q].len;
2093 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2094 
2095 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2096 
2097 	KASSERT(st->sync_state == PFSYNC_S_NONE,
2098 		("%s: st->sync_state %u", __func__, st->sync_state));
2099 	KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2100 	    b->b_len));
2101 
2102 	if (TAILQ_EMPTY(&b->b_qs[q]))
2103 		nlen += sizeof(struct pfsync_subheader);
2104 
2105 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2106 		pfsync_sendout(1, b->b_id);
2107 
2108 		nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2109 	}
2110 
2111 	b->b_len += nlen;
2112 	TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list);
2113 	st->sync_state = q;
2114 	if (ref)
2115 		pf_ref_state(st);
2116 }
2117 
2118 static void
2119 pfsync_q_del(struct pf_kstate *st, bool unref, struct pfsync_bucket *b)
2120 {
2121 	int q = st->sync_state;
2122 
2123 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2124 	KASSERT(st->sync_state != PFSYNC_S_NONE,
2125 		("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2126 
2127 	b->b_len -= pfsync_qs[q].len;
2128 	TAILQ_REMOVE(&b->b_qs[q], st, sync_list);
2129 	st->sync_state = PFSYNC_S_NONE;
2130 	if (unref)
2131 		pf_release_state(st);
2132 
2133 	if (TAILQ_EMPTY(&b->b_qs[q]))
2134 		b->b_len -= sizeof(struct pfsync_subheader);
2135 }
2136 
2137 static void
2138 pfsync_bulk_start(void)
2139 {
2140 	struct pfsync_softc *sc = V_pfsyncif;
2141 
2142 	if (V_pf_status.debug >= PF_DEBUG_MISC)
2143 		printf("pfsync: received bulk update request\n");
2144 
2145 	PFSYNC_BLOCK(sc);
2146 
2147 	sc->sc_ureq_received = time_uptime;
2148 	sc->sc_bulk_hashid = 0;
2149 	sc->sc_bulk_stateid = 0;
2150 	pfsync_bulk_status(PFSYNC_BUS_START);
2151 	callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2152 	PFSYNC_BUNLOCK(sc);
2153 }
2154 
2155 static void
2156 pfsync_bulk_update(void *arg)
2157 {
2158 	struct pfsync_softc *sc = arg;
2159 	struct pf_kstate *s;
2160 	int i;
2161 
2162 	PFSYNC_BLOCK_ASSERT(sc);
2163 	CURVNET_SET(sc->sc_ifp->if_vnet);
2164 
2165 	/*
2166 	 * Start with last state from previous invocation.
2167 	 * It may had gone, in this case start from the
2168 	 * hash slot.
2169 	 */
2170 	s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2171 
2172 	if (s != NULL)
2173 		i = PF_IDHASH(s);
2174 	else
2175 		i = sc->sc_bulk_hashid;
2176 
2177 	for (; i <= pf_hashmask; i++) {
2178 		struct pf_idhash *ih = &V_pf_idhash[i];
2179 
2180 		if (s != NULL)
2181 			PF_HASHROW_ASSERT(ih);
2182 		else {
2183 			PF_HASHROW_LOCK(ih);
2184 			s = LIST_FIRST(&ih->states);
2185 		}
2186 
2187 		for (; s; s = LIST_NEXT(s, entry)) {
2188 			if (s->sync_state == PFSYNC_S_NONE &&
2189 			    s->timeout < PFTM_MAX &&
2190 			    s->pfsync_time <= sc->sc_ureq_received) {
2191 				if (pfsync_update_state_req(s)) {
2192 					/* We've filled a packet. */
2193 					sc->sc_bulk_hashid = i;
2194 					sc->sc_bulk_stateid = s->id;
2195 					sc->sc_bulk_creatorid = s->creatorid;
2196 					PF_HASHROW_UNLOCK(ih);
2197 					callout_reset(&sc->sc_bulk_tmo, 1,
2198 					    pfsync_bulk_update, sc);
2199 					goto full;
2200 				}
2201 			}
2202 		}
2203 		PF_HASHROW_UNLOCK(ih);
2204 	}
2205 
2206 	/* We're done. */
2207 	pfsync_bulk_status(PFSYNC_BUS_END);
2208 full:
2209 	CURVNET_RESTORE();
2210 }
2211 
2212 static void
2213 pfsync_bulk_status(u_int8_t status)
2214 {
2215 	struct {
2216 		struct pfsync_subheader subh;
2217 		struct pfsync_bus bus;
2218 	} __packed r;
2219 
2220 	struct pfsync_softc *sc = V_pfsyncif;
2221 
2222 	bzero(&r, sizeof(r));
2223 
2224 	r.subh.action = PFSYNC_ACT_BUS;
2225 	r.subh.count = htons(1);
2226 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2227 
2228 	r.bus.creatorid = V_pf_status.hostid;
2229 	r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2230 	r.bus.status = status;
2231 
2232 	pfsync_send_plus(&r, sizeof(r));
2233 }
2234 
2235 static void
2236 pfsync_bulk_fail(void *arg)
2237 {
2238 	struct pfsync_softc *sc = arg;
2239 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2240 
2241 	CURVNET_SET(sc->sc_ifp->if_vnet);
2242 
2243 	PFSYNC_BLOCK_ASSERT(sc);
2244 
2245 	if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2246 		/* Try again */
2247 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2248 		    pfsync_bulk_fail, V_pfsyncif);
2249 		PFSYNC_BUCKET_LOCK(b);
2250 		pfsync_request_update(0, 0);
2251 		PFSYNC_BUCKET_UNLOCK(b);
2252 	} else {
2253 		/* Pretend like the transfer was ok. */
2254 		sc->sc_ureq_sent = 0;
2255 		sc->sc_bulk_tries = 0;
2256 		PFSYNC_LOCK(sc);
2257 		if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2258 			(*carp_demote_adj_p)(-V_pfsync_carp_adj,
2259 			    "pfsync bulk fail");
2260 		sc->sc_flags |= PFSYNCF_OK;
2261 		PFSYNC_UNLOCK(sc);
2262 		if (V_pf_status.debug >= PF_DEBUG_MISC)
2263 			printf("pfsync: failed to receive bulk update\n");
2264 	}
2265 
2266 	CURVNET_RESTORE();
2267 }
2268 
2269 static void
2270 pfsync_send_plus(void *plus, size_t pluslen)
2271 {
2272 	struct pfsync_softc *sc = V_pfsyncif;
2273 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2274 
2275 	PFSYNC_BUCKET_LOCK(b);
2276 
2277 	if (b->b_len + pluslen > sc->sc_ifp->if_mtu)
2278 		pfsync_sendout(1, b->b_id);
2279 
2280 	b->b_plus = plus;
2281 	b->b_len += (b->b_pluslen = pluslen);
2282 
2283 	pfsync_sendout(1, b->b_id);
2284 	PFSYNC_BUCKET_UNLOCK(b);
2285 }
2286 
2287 static void
2288 pfsync_timeout(void *arg)
2289 {
2290 	struct pfsync_bucket *b = arg;
2291 
2292 	CURVNET_SET(b->b_sc->sc_ifp->if_vnet);
2293 	PFSYNC_BUCKET_LOCK(b);
2294 	pfsync_push(b);
2295 	PFSYNC_BUCKET_UNLOCK(b);
2296 	CURVNET_RESTORE();
2297 }
2298 
2299 static void
2300 pfsync_push(struct pfsync_bucket *b)
2301 {
2302 
2303 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2304 
2305 	b->b_flags |= PFSYNCF_BUCKET_PUSH;
2306 	swi_sched(V_pfsync_swi_cookie, 0);
2307 }
2308 
2309 static void
2310 pfsync_push_all(struct pfsync_softc *sc)
2311 {
2312 	int c;
2313 	struct pfsync_bucket *b;
2314 
2315 	for (c = 0; c < pfsync_buckets; c++) {
2316 		b = &sc->sc_buckets[c];
2317 
2318 		PFSYNC_BUCKET_LOCK(b);
2319 		pfsync_push(b);
2320 		PFSYNC_BUCKET_UNLOCK(b);
2321 	}
2322 }
2323 
2324 static void
2325 pfsyncintr(void *arg)
2326 {
2327 	struct epoch_tracker et;
2328 	struct pfsync_softc *sc = arg;
2329 	struct pfsync_bucket *b;
2330 	struct mbuf *m, *n;
2331 	int c, error;
2332 
2333 	NET_EPOCH_ENTER(et);
2334 	CURVNET_SET(sc->sc_ifp->if_vnet);
2335 
2336 	for (c = 0; c < pfsync_buckets; c++) {
2337 		b = &sc->sc_buckets[c];
2338 
2339 		PFSYNC_BUCKET_LOCK(b);
2340 		if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) {
2341 			pfsync_sendout(0, b->b_id);
2342 			b->b_flags &= ~PFSYNCF_BUCKET_PUSH;
2343 		}
2344 		_IF_DEQUEUE_ALL(&b->b_snd, m);
2345 		PFSYNC_BUCKET_UNLOCK(b);
2346 
2347 		for (; m != NULL; m = n) {
2348 			n = m->m_nextpkt;
2349 			m->m_nextpkt = NULL;
2350 
2351 			/*
2352 			 * We distinguish between a deferral packet and our
2353 			 * own pfsync packet based on M_SKIP_FIREWALL
2354 			 * flag. This is XXX.
2355 			 */
2356 			switch (sc->sc_sync_peer.ss_family) {
2357 #ifdef INET
2358 			case AF_INET:
2359 				if (m->m_flags & M_SKIP_FIREWALL) {
2360 					error = ip_output(m, NULL, NULL, 0,
2361 					    NULL, NULL);
2362 				} else {
2363 					error = ip_output(m, NULL, NULL,
2364 					    IP_RAWOUTPUT, &sc->sc_imo, NULL);
2365 				}
2366 				break;
2367 #endif
2368 			}
2369 
2370 			if (error == 0)
2371 				V_pfsyncstats.pfsyncs_opackets++;
2372 			else
2373 				V_pfsyncstats.pfsyncs_oerrors++;
2374 		}
2375 	}
2376 	CURVNET_RESTORE();
2377 	NET_EPOCH_EXIT(et);
2378 }
2379 
2380 static int
2381 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp,
2382     struct in_mfilter *imf)
2383 {
2384 	struct ip_moptions *imo = &sc->sc_imo;
2385 	int error;
2386 
2387 	if (!(ifp->if_flags & IFF_MULTICAST))
2388 		return (EADDRNOTAVAIL);
2389 
2390 	switch (sc->sc_sync_peer.ss_family) {
2391 #ifdef INET
2392 	case AF_INET:
2393 	    {
2394 		ip_mfilter_init(&imo->imo_head);
2395 		imo->imo_multicast_vif = -1;
2396 		if ((error = in_joingroup(ifp, &((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr, NULL,
2397 		    &imf->imf_inm)) != 0)
2398 			return (error);
2399 
2400 		ip_mfilter_insert(&imo->imo_head, imf);
2401 		imo->imo_multicast_ifp = ifp;
2402 		imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2403 		imo->imo_multicast_loop = 0;
2404 		break;
2405 	    }
2406 #endif
2407 	}
2408 
2409 	return (0);
2410 }
2411 
2412 static void
2413 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2414 {
2415 	struct ip_moptions *imo = &sc->sc_imo;
2416 	struct in_mfilter *imf;
2417 
2418 	while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
2419 		ip_mfilter_remove(&imo->imo_head, imf);
2420 		in_leavegroup(imf->imf_inm, NULL);
2421 		ip_mfilter_free(imf);
2422 	}
2423 	imo->imo_multicast_ifp = NULL;
2424 }
2425 
2426 void
2427 pfsync_detach_ifnet(struct ifnet *ifp)
2428 {
2429 	struct pfsync_softc *sc = V_pfsyncif;
2430 
2431 	if (sc == NULL)
2432 		return;
2433 
2434 	PFSYNC_LOCK(sc);
2435 
2436 	if (sc->sc_sync_if == ifp) {
2437 		/* We don't need mutlicast cleanup here, because the interface
2438 		 * is going away. We do need to ensure we don't try to do
2439 		 * cleanup later.
2440 		 */
2441 		ip_mfilter_init(&sc->sc_imo.imo_head);
2442 		sc->sc_imo.imo_multicast_ifp = NULL;
2443 		sc->sc_sync_if = NULL;
2444 	}
2445 
2446 	PFSYNC_UNLOCK(sc);
2447 }
2448 
2449 static int
2450 pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *pfsyncr, struct pfsync_kstatus *status)
2451 {
2452 	struct sockaddr_storage sa;
2453 	status->maxupdates = pfsyncr->pfsyncr_maxupdates;
2454 	status->flags = pfsyncr->pfsyncr_defer;
2455 
2456 	strlcpy(status->syncdev, pfsyncr->pfsyncr_syncdev, IFNAMSIZ);
2457 
2458 	memset(&sa, 0, sizeof(sa));
2459 	if (pfsyncr->pfsyncr_syncpeer.s_addr != 0) {
2460 		struct sockaddr_in *in = (struct sockaddr_in *)&sa;
2461 		in->sin_family = AF_INET;
2462 		in->sin_len = sizeof(*in);
2463 		in->sin_addr.s_addr = pfsyncr->pfsyncr_syncpeer.s_addr;
2464 	}
2465 	status->syncpeer = sa;
2466 
2467 	return 0;
2468 }
2469 
2470 static int
2471 pfsync_kstatus_to_softc(struct pfsync_kstatus *status, struct pfsync_softc *sc)
2472 {
2473 	struct in_mfilter *imf = NULL;
2474 	struct ifnet *sifp;
2475 	struct ip *ip;
2476 	int error;
2477 	int c;
2478 
2479 	if ((status->maxupdates < 0) || (status->maxupdates > 255))
2480 		return (EINVAL);
2481 
2482 	if (status->syncdev[0] == '\0')
2483 		sifp = NULL;
2484 	else if ((sifp = ifunit_ref(status->syncdev)) == NULL)
2485 		return (EINVAL);
2486 
2487 	struct sockaddr_in *status_sin =
2488 	    (struct sockaddr_in *)&(status->syncpeer);
2489 	if (sifp != NULL && (status_sin->sin_addr.s_addr == 0 ||
2490 				status_sin->sin_addr.s_addr ==
2491 				    htonl(INADDR_PFSYNC_GROUP)))
2492 		imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
2493 
2494 	PFSYNC_LOCK(sc);
2495 	struct sockaddr_in *sc_sin = (struct sockaddr_in *)&sc->sc_sync_peer;
2496 	sc_sin->sin_family = AF_INET;
2497 	sc_sin->sin_len = sizeof(*sc_sin);
2498 	if (status_sin->sin_addr.s_addr == 0) {
2499 		sc_sin->sin_addr.s_addr = htonl(INADDR_PFSYNC_GROUP);
2500 	} else {
2501 		sc_sin->sin_addr.s_addr = status_sin->sin_addr.s_addr;
2502 	}
2503 
2504 	sc->sc_maxupdates = status->maxupdates;
2505 	if (status->flags & PFSYNCF_DEFER) {
2506 		sc->sc_flags |= PFSYNCF_DEFER;
2507 		V_pfsync_defer_ptr = pfsync_defer;
2508 	} else {
2509 		sc->sc_flags &= ~PFSYNCF_DEFER;
2510 		V_pfsync_defer_ptr = NULL;
2511 	}
2512 
2513 	if (sifp == NULL) {
2514 		if (sc->sc_sync_if)
2515 			if_rele(sc->sc_sync_if);
2516 		sc->sc_sync_if = NULL;
2517 		pfsync_multicast_cleanup(sc);
2518 		PFSYNC_UNLOCK(sc);
2519 		return (0);
2520 	}
2521 
2522 	for (c = 0; c < pfsync_buckets; c++) {
2523 		PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
2524 		if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT &&
2525 		    (sifp->if_mtu < sc->sc_ifp->if_mtu ||
2526 			(sc->sc_sync_if != NULL &&
2527 			    sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
2528 			sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
2529 			pfsync_sendout(1, c);
2530 		PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
2531 	}
2532 
2533 	pfsync_multicast_cleanup(sc);
2534 
2535 	if (sc_sin->sin_addr.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
2536 		error = pfsync_multicast_setup(sc, sifp, imf);
2537 		if (error) {
2538 			if_rele(sifp);
2539 			ip_mfilter_free(imf);
2540 			PFSYNC_UNLOCK(sc);
2541 			return (error);
2542 		}
2543 	}
2544 	if (sc->sc_sync_if)
2545 		if_rele(sc->sc_sync_if);
2546 	sc->sc_sync_if = sifp;
2547 
2548 	ip = &sc->sc_template.ipv4;
2549 	bzero(ip, sizeof(*ip));
2550 	ip->ip_v = IPVERSION;
2551 	ip->ip_hl = sizeof(sc->sc_template.ipv4) >> 2;
2552 	ip->ip_tos = IPTOS_LOWDELAY;
2553 	/* len and id are set later. */
2554 	ip->ip_off = htons(IP_DF);
2555 	ip->ip_ttl = PFSYNC_DFLTTL;
2556 	ip->ip_p = IPPROTO_PFSYNC;
2557 	ip->ip_src.s_addr = INADDR_ANY;
2558 	ip->ip_dst.s_addr = sc_sin->sin_addr.s_addr;
2559 
2560 	/* Request a full state table update. */
2561 	if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2562 		(*carp_demote_adj_p)(V_pfsync_carp_adj,
2563 		    "pfsync bulk start");
2564 	sc->sc_flags &= ~PFSYNCF_OK;
2565 	if (V_pf_status.debug >= PF_DEBUG_MISC)
2566 		printf("pfsync: requesting bulk update\n");
2567 	PFSYNC_UNLOCK(sc);
2568 	PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
2569 	pfsync_request_update(0, 0);
2570 	PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
2571 	PFSYNC_BLOCK(sc);
2572 	sc->sc_ureq_sent = time_uptime;
2573 	callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, sc);
2574 	PFSYNC_BUNLOCK(sc);
2575 	return (0);
2576 }
2577 
2578 static void
2579 pfsync_pointers_init(void)
2580 {
2581 
2582 	PF_RULES_WLOCK();
2583 	V_pfsync_state_import_ptr = pfsync_state_import;
2584 	V_pfsync_insert_state_ptr = pfsync_insert_state;
2585 	V_pfsync_update_state_ptr = pfsync_update_state;
2586 	V_pfsync_delete_state_ptr = pfsync_delete_state;
2587 	V_pfsync_clear_states_ptr = pfsync_clear_states;
2588 	V_pfsync_defer_ptr = pfsync_defer;
2589 	PF_RULES_WUNLOCK();
2590 }
2591 
2592 static void
2593 pfsync_pointers_uninit(void)
2594 {
2595 
2596 	PF_RULES_WLOCK();
2597 	V_pfsync_state_import_ptr = NULL;
2598 	V_pfsync_insert_state_ptr = NULL;
2599 	V_pfsync_update_state_ptr = NULL;
2600 	V_pfsync_delete_state_ptr = NULL;
2601 	V_pfsync_clear_states_ptr = NULL;
2602 	V_pfsync_defer_ptr = NULL;
2603 	PF_RULES_WUNLOCK();
2604 }
2605 
2606 static void
2607 vnet_pfsync_init(const void *unused __unused)
2608 {
2609 	int error;
2610 
2611 	V_pfsync_cloner = if_clone_simple(pfsyncname,
2612 	    pfsync_clone_create, pfsync_clone_destroy, 1);
2613 	error = swi_add(&V_pfsync_swi_ie, pfsyncname, pfsyncintr, V_pfsyncif,
2614 	    SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2615 	if (error) {
2616 		if_clone_detach(V_pfsync_cloner);
2617 		log(LOG_INFO, "swi_add() failed in %s\n", __func__);
2618 	}
2619 
2620 	pfsync_pointers_init();
2621 }
2622 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
2623     vnet_pfsync_init, NULL);
2624 
2625 static void
2626 vnet_pfsync_uninit(const void *unused __unused)
2627 {
2628 	int ret __diagused;
2629 
2630 	pfsync_pointers_uninit();
2631 
2632 	if_clone_detach(V_pfsync_cloner);
2633 	ret = swi_remove(V_pfsync_swi_cookie);
2634 	MPASS(ret == 0);
2635 	ret = intr_event_destroy(V_pfsync_swi_ie);
2636 	MPASS(ret == 0);
2637 }
2638 
2639 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH,
2640     vnet_pfsync_uninit, NULL);
2641 
2642 static int
2643 pfsync_init(void)
2644 {
2645 #ifdef INET
2646 	int error;
2647 
2648 	pfsync_detach_ifnet_ptr = pfsync_detach_ifnet;
2649 
2650 	error = ipproto_register(IPPROTO_PFSYNC, pfsync_input, NULL);
2651 	if (error)
2652 		return (error);
2653 #endif
2654 
2655 	return (0);
2656 }
2657 
2658 static void
2659 pfsync_uninit(void)
2660 {
2661 	pfsync_detach_ifnet_ptr = NULL;
2662 
2663 #ifdef INET
2664 	ipproto_unregister(IPPROTO_PFSYNC);
2665 #endif
2666 }
2667 
2668 static int
2669 pfsync_modevent(module_t mod, int type, void *data)
2670 {
2671 	int error = 0;
2672 
2673 	switch (type) {
2674 	case MOD_LOAD:
2675 		error = pfsync_init();
2676 		break;
2677 	case MOD_UNLOAD:
2678 		pfsync_uninit();
2679 		break;
2680 	default:
2681 		error = EINVAL;
2682 		break;
2683 	}
2684 
2685 	return (error);
2686 }
2687 
2688 static moduledata_t pfsync_mod = {
2689 	pfsyncname,
2690 	pfsync_modevent,
2691 	0
2692 };
2693 
2694 #define PFSYNC_MODVER 1
2695 
2696 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
2697 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
2698 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2699 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
2700