xref: /freebsd/sys/netpfil/pf/if_pfsync.c (revision f126890ac5386406dadf7c4cfa9566cbb56537c5)
1 /*-
2  * SPDX-License-Identifier: (BSD-2-Clause 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 #include "opt_inet.h"
64 #include "opt_inet6.h"
65 #include "opt_pf.h"
66 
67 #include <sys/param.h>
68 #include <sys/bus.h>
69 #include <sys/endian.h>
70 #include <sys/interrupt.h>
71 #include <sys/kernel.h>
72 #include <sys/lock.h>
73 #include <sys/mbuf.h>
74 #include <sys/module.h>
75 #include <sys/mutex.h>
76 #include <sys/nv.h>
77 #include <sys/priv.h>
78 #include <sys/smp.h>
79 #include <sys/socket.h>
80 #include <sys/sockio.h>
81 #include <sys/sysctl.h>
82 #include <sys/syslog.h>
83 
84 #include <net/bpf.h>
85 #include <net/if.h>
86 #include <net/if_var.h>
87 #include <net/if_clone.h>
88 #include <net/if_private.h>
89 #include <net/if_types.h>
90 #include <net/vnet.h>
91 #include <net/pfvar.h>
92 #include <net/route.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 <netinet6/in6_var.h>
99 #include <netinet/ip.h>
100 #include <netinet/ip6.h>
101 #include <netinet/ip_carp.h>
102 #include <netinet/ip_var.h>
103 #include <netinet/tcp.h>
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_seq.h>
106 
107 #include <netinet/ip6.h>
108 #include <netinet6/ip6_var.h>
109 #include <netinet6/scope6_var.h>
110 
111 #include <netpfil/pf/pfsync_nv.h>
112 
113 struct pfsync_bucket;
114 struct pfsync_softc;
115 
116 union inet_template {
117 	struct ip	ipv4;
118 	struct ip6_hdr	ipv6;
119 };
120 
121 #define PFSYNC_MINPKT ( \
122 	sizeof(union inet_template) + \
123 	sizeof(struct pfsync_header) + \
124 	sizeof(struct pfsync_subheader) )
125 
126 static int	pfsync_upd_tcp(struct pf_kstate *, struct pfsync_state_peer *,
127 		    struct pfsync_state_peer *);
128 static int	pfsync_in_clr(struct mbuf *, int, int, int, int);
129 static int	pfsync_in_ins(struct mbuf *, int, int, int, int);
130 static int	pfsync_in_iack(struct mbuf *, int, int, int, int);
131 static int	pfsync_in_upd(struct mbuf *, int, int, int, int);
132 static int	pfsync_in_upd_c(struct mbuf *, int, int, int, int);
133 static int	pfsync_in_ureq(struct mbuf *, int, int, int, int);
134 static int	pfsync_in_del_c(struct mbuf *, int, int, int, int);
135 static int	pfsync_in_bus(struct mbuf *, int, int, int, int);
136 static int	pfsync_in_tdb(struct mbuf *, int, int, int, int);
137 static int	pfsync_in_eof(struct mbuf *, int, int, int, int);
138 static int	pfsync_in_error(struct mbuf *, int, int, int, int);
139 
140 static int (*pfsync_acts[])(struct mbuf *, int, int, int, int) = {
141 	pfsync_in_clr,			/* PFSYNC_ACT_CLR */
142 	pfsync_in_ins,			/* PFSYNC_ACT_INS_1301 */
143 	pfsync_in_iack,			/* PFSYNC_ACT_INS_ACK */
144 	pfsync_in_upd,			/* PFSYNC_ACT_UPD_1301 */
145 	pfsync_in_upd_c,		/* PFSYNC_ACT_UPD_C */
146 	pfsync_in_ureq,			/* PFSYNC_ACT_UPD_REQ */
147 	pfsync_in_error,		/* PFSYNC_ACT_DEL */
148 	pfsync_in_del_c,		/* PFSYNC_ACT_DEL_C */
149 	pfsync_in_error,		/* PFSYNC_ACT_INS_F */
150 	pfsync_in_error,		/* PFSYNC_ACT_DEL_F */
151 	pfsync_in_bus,			/* PFSYNC_ACT_BUS */
152 	pfsync_in_tdb,			/* PFSYNC_ACT_TDB */
153 	pfsync_in_eof,			/* PFSYNC_ACT_EOF */
154 	pfsync_in_ins,			/* PFSYNC_ACT_INS_1400 */
155 	pfsync_in_upd,			/* PFSYNC_ACT_UPD_1400 */
156 };
157 
158 struct pfsync_q {
159 	void		(*write)(struct pf_kstate *, void *);
160 	size_t		len;
161 	u_int8_t	action;
162 };
163 
164 /* We have the following sync queues */
165 enum pfsync_q_id {
166 	PFSYNC_Q_INS_1301,
167 	PFSYNC_Q_INS_1400,
168 	PFSYNC_Q_IACK,
169 	PFSYNC_Q_UPD_1301,
170 	PFSYNC_Q_UPD_1400,
171 	PFSYNC_Q_UPD_C,
172 	PFSYNC_Q_DEL_C,
173 	PFSYNC_Q_COUNT,
174 };
175 
176 /* Functions for building messages for given queue */
177 static void	pfsync_out_state_1301(struct pf_kstate *, void *);
178 static void	pfsync_out_state_1400(struct pf_kstate *, void *);
179 static void	pfsync_out_iack(struct pf_kstate *, void *);
180 static void	pfsync_out_upd_c(struct pf_kstate *, void *);
181 static void	pfsync_out_del_c(struct pf_kstate *, void *);
182 
183 /* Attach those functions to queue */
184 static struct pfsync_q pfsync_qs[] = {
185 	{ pfsync_out_state_1301, sizeof(struct pfsync_state_1301), PFSYNC_ACT_INS_1301 },
186 	{ pfsync_out_state_1400, sizeof(struct pfsync_state_1400), PFSYNC_ACT_INS_1400 },
187 	{ pfsync_out_iack,       sizeof(struct pfsync_ins_ack),    PFSYNC_ACT_INS_ACK },
188 	{ pfsync_out_state_1301, sizeof(struct pfsync_state_1301), PFSYNC_ACT_UPD_1301 },
189 	{ pfsync_out_state_1400, sizeof(struct pfsync_state_1400), PFSYNC_ACT_UPD_1400 },
190 	{ pfsync_out_upd_c,      sizeof(struct pfsync_upd_c),      PFSYNC_ACT_UPD_C },
191 	{ pfsync_out_del_c,      sizeof(struct pfsync_del_c),      PFSYNC_ACT_DEL_C }
192 };
193 
194 /* Map queue to pf_kstate->sync_state */
195 static u_int8_t pfsync_qid_sstate[] = {
196 	PFSYNC_S_INS,   /* PFSYNC_Q_INS_1301 */
197 	PFSYNC_S_INS,   /* PFSYNC_Q_INS_1400 */
198 	PFSYNC_S_IACK,  /* PFSYNC_Q_IACK */
199 	PFSYNC_S_UPD,   /* PFSYNC_Q_UPD_1301 */
200 	PFSYNC_S_UPD,   /* PFSYNC_Q_UPD_1400 */
201 	PFSYNC_S_UPD_C, /* PFSYNC_Q_UPD_C */
202 	PFSYNC_S_DEL_C, /* PFSYNC_Q_DEL_C */
203 };
204 
205 /* Map pf_kstate->sync_state to queue */
206 static enum pfsync_q_id pfsync_sstate_to_qid(u_int8_t);
207 
208 static void	pfsync_q_ins(struct pf_kstate *, int sync_state, bool);
209 static void	pfsync_q_del(struct pf_kstate *, bool, struct pfsync_bucket *);
210 
211 static void	pfsync_update_state(struct pf_kstate *);
212 static void	pfsync_tx(struct pfsync_softc *, struct mbuf *);
213 
214 struct pfsync_upd_req_item {
215 	TAILQ_ENTRY(pfsync_upd_req_item)	ur_entry;
216 	struct pfsync_upd_req			ur_msg;
217 };
218 
219 struct pfsync_deferral {
220 	struct pfsync_softc		*pd_sc;
221 	TAILQ_ENTRY(pfsync_deferral)	pd_entry;
222 	struct callout			pd_tmo;
223 
224 	struct pf_kstate		*pd_st;
225 	struct mbuf			*pd_m;
226 };
227 
228 struct pfsync_bucket
229 {
230 	int			b_id;
231 	struct pfsync_softc	*b_sc;
232 	struct mtx		b_mtx;
233 	struct callout		b_tmo;
234 	int			b_flags;
235 #define	PFSYNCF_BUCKET_PUSH	0x00000001
236 
237 	size_t			b_len;
238 	TAILQ_HEAD(, pf_kstate)			b_qs[PFSYNC_Q_COUNT];
239 	TAILQ_HEAD(, pfsync_upd_req_item)	b_upd_req_list;
240 	TAILQ_HEAD(, pfsync_deferral)		b_deferrals;
241 	u_int			b_deferred;
242 	void			*b_plus;
243 	size_t			b_pluslen;
244 
245 	struct  ifaltq b_snd;
246 };
247 
248 struct pfsync_softc {
249 	/* Configuration */
250 	struct ifnet		*sc_ifp;
251 	struct ifnet		*sc_sync_if;
252 	struct ip_moptions	sc_imo;
253 	struct ip6_moptions	sc_im6o;
254 	struct sockaddr_storage	sc_sync_peer;
255 	uint32_t		sc_flags;
256 	uint8_t			sc_maxupdates;
257 	union inet_template     sc_template;
258 	struct mtx		sc_mtx;
259 	uint32_t		sc_version;
260 
261 	/* Queued data */
262 	struct pfsync_bucket	*sc_buckets;
263 
264 	/* Bulk update info */
265 	struct mtx		sc_bulk_mtx;
266 	uint32_t		sc_ureq_sent;
267 	int			sc_bulk_tries;
268 	uint32_t		sc_ureq_received;
269 	int			sc_bulk_hashid;
270 	uint64_t		sc_bulk_stateid;
271 	uint32_t		sc_bulk_creatorid;
272 	struct callout		sc_bulk_tmo;
273 	struct callout		sc_bulkfail_tmo;
274 };
275 
276 #define	PFSYNC_LOCK(sc)		mtx_lock(&(sc)->sc_mtx)
277 #define	PFSYNC_UNLOCK(sc)	mtx_unlock(&(sc)->sc_mtx)
278 #define	PFSYNC_LOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_mtx, MA_OWNED)
279 
280 #define PFSYNC_BUCKET_LOCK(b)		mtx_lock(&(b)->b_mtx)
281 #define PFSYNC_BUCKET_UNLOCK(b)		mtx_unlock(&(b)->b_mtx)
282 #define PFSYNC_BUCKET_LOCK_ASSERT(b)	mtx_assert(&(b)->b_mtx, MA_OWNED)
283 
284 #define	PFSYNC_BLOCK(sc)	mtx_lock(&(sc)->sc_bulk_mtx)
285 #define	PFSYNC_BUNLOCK(sc)	mtx_unlock(&(sc)->sc_bulk_mtx)
286 #define	PFSYNC_BLOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
287 
288 #define PFSYNC_DEFER_TIMEOUT	20
289 
290 static const char pfsyncname[] = "pfsync";
291 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
292 VNET_DEFINE_STATIC(struct pfsync_softc	*, pfsyncif) = NULL;
293 #define	V_pfsyncif		VNET(pfsyncif)
294 VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL;
295 #define	V_pfsync_swi_cookie	VNET(pfsync_swi_cookie)
296 VNET_DEFINE_STATIC(struct intr_event *, pfsync_swi_ie);
297 #define	V_pfsync_swi_ie		VNET(pfsync_swi_ie)
298 VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats);
299 #define	V_pfsyncstats		VNET(pfsyncstats)
300 VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW;
301 #define	V_pfsync_carp_adj	VNET(pfsync_carp_adj)
302 VNET_DEFINE_STATIC(unsigned int, pfsync_defer_timeout) = PFSYNC_DEFER_TIMEOUT;
303 #define	V_pfsync_defer_timeout	VNET(pfsync_defer_timeout)
304 
305 static void	pfsync_timeout(void *);
306 static void	pfsync_push(struct pfsync_bucket *);
307 static void	pfsync_push_all(struct pfsync_softc *);
308 static void	pfsyncintr(void *);
309 static int	pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
310 		    struct in_mfilter *, struct in6_mfilter *);
311 static void	pfsync_multicast_cleanup(struct pfsync_softc *);
312 static void	pfsync_pointers_init(void);
313 static void	pfsync_pointers_uninit(void);
314 static int	pfsync_init(void);
315 static void	pfsync_uninit(void);
316 
317 static unsigned long pfsync_buckets;
318 
319 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
320     "PFSYNC");
321 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW,
322     &VNET_NAME(pfsyncstats), pfsyncstats,
323     "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
324 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_VNET | CTLFLAG_RW,
325     &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
326 SYSCTL_ULONG(_net_pfsync, OID_AUTO, pfsync_buckets, CTLFLAG_RDTUN,
327     &pfsync_buckets, 0, "Number of pfsync hash buckets");
328 SYSCTL_UINT(_net_pfsync, OID_AUTO, defer_delay, CTLFLAG_VNET | CTLFLAG_RW,
329     &VNET_NAME(pfsync_defer_timeout), 0, "Deferred packet timeout (in ms)");
330 
331 static int	pfsync_clone_create(struct if_clone *, int, caddr_t);
332 static void	pfsync_clone_destroy(struct ifnet *);
333 static int	pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
334 		    struct pf_state_peer *);
335 static int	pfsyncoutput(struct ifnet *, struct mbuf *,
336 		    const struct sockaddr *, struct route *);
337 static int	pfsyncioctl(struct ifnet *, u_long, caddr_t);
338 
339 static int	pfsync_defer(struct pf_kstate *, struct mbuf *);
340 static void	pfsync_undefer(struct pfsync_deferral *, int);
341 static void	pfsync_undefer_state_locked(struct pf_kstate *, int);
342 static void	pfsync_undefer_state(struct pf_kstate *, int);
343 static void	pfsync_defer_tmo(void *);
344 
345 static void	pfsync_request_update(u_int32_t, u_int64_t);
346 static bool	pfsync_update_state_req(struct pf_kstate *);
347 
348 static void	pfsync_drop(struct pfsync_softc *);
349 static void	pfsync_sendout(int, int);
350 static void	pfsync_send_plus(void *, size_t);
351 
352 static void	pfsync_bulk_start(void);
353 static void	pfsync_bulk_status(u_int8_t);
354 static void	pfsync_bulk_update(void *);
355 static void	pfsync_bulk_fail(void *);
356 
357 static void	pfsync_detach_ifnet(struct ifnet *);
358 
359 static int pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *,
360     struct pfsync_kstatus *);
361 static int pfsync_kstatus_to_softc(struct pfsync_kstatus *,
362     struct pfsync_softc *);
363 
364 #ifdef IPSEC
365 static void	pfsync_update_net_tdb(struct pfsync_tdb *);
366 #endif
367 static struct pfsync_bucket	*pfsync_get_bucket(struct pfsync_softc *,
368 		    struct pf_kstate *);
369 
370 #define PFSYNC_MAX_BULKTRIES	12
371 
372 VNET_DEFINE(struct if_clone *, pfsync_cloner);
373 #define	V_pfsync_cloner	VNET(pfsync_cloner)
374 
375 const struct in6_addr in6addr_linklocal_pfsync_group =
376 	{{{ 0xff, 0x12, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
377 	    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xf0 }}};
378 static int
379 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
380 {
381 	struct pfsync_softc *sc;
382 	struct ifnet *ifp;
383 	struct pfsync_bucket *b;
384 	int c;
385 	enum pfsync_q_id q;
386 
387 	if (unit != 0)
388 		return (EINVAL);
389 
390 	if (! pfsync_buckets)
391 		pfsync_buckets = mp_ncpus * 2;
392 
393 	sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
394 	sc->sc_flags |= PFSYNCF_OK;
395 	sc->sc_maxupdates = 128;
396 	sc->sc_version = PFSYNC_MSG_VERSION_DEFAULT;
397 
398 	ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
399 	if (ifp == NULL) {
400 		free(sc, M_PFSYNC);
401 		return (ENOSPC);
402 	}
403 	if_initname(ifp, pfsyncname, unit);
404 	ifp->if_softc = sc;
405 	ifp->if_ioctl = pfsyncioctl;
406 	ifp->if_output = pfsyncoutput;
407 	ifp->if_type = IFT_PFSYNC;
408 	ifp->if_hdrlen = sizeof(struct pfsync_header);
409 	ifp->if_mtu = ETHERMTU;
410 	mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
411 	mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
412 	callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
413 	callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
414 
415 	if_attach(ifp);
416 
417 	bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
418 
419 	sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets),
420 	    M_PFSYNC, M_ZERO | M_WAITOK);
421 	for (c = 0; c < pfsync_buckets; c++) {
422 		b = &sc->sc_buckets[c];
423 		mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF);
424 
425 		b->b_id = c;
426 		b->b_sc = sc;
427 		b->b_len = PFSYNC_MINPKT;
428 
429 		for (q = 0; q < PFSYNC_Q_COUNT; q++)
430 			TAILQ_INIT(&b->b_qs[q]);
431 
432 		TAILQ_INIT(&b->b_upd_req_list);
433 		TAILQ_INIT(&b->b_deferrals);
434 
435 		callout_init(&b->b_tmo, 1);
436 
437 		b->b_snd.ifq_maxlen = ifqmaxlen;
438 	}
439 
440 	V_pfsyncif = sc;
441 
442 	return (0);
443 }
444 
445 static void
446 pfsync_clone_destroy(struct ifnet *ifp)
447 {
448 	struct pfsync_softc *sc = ifp->if_softc;
449 	struct pfsync_bucket *b;
450 	int c, ret;
451 
452 	for (c = 0; c < pfsync_buckets; c++) {
453 		b = &sc->sc_buckets[c];
454 		/*
455 		 * At this stage, everything should have already been
456 		 * cleared by pfsync_uninit(), and we have only to
457 		 * drain callouts.
458 		 */
459 		PFSYNC_BUCKET_LOCK(b);
460 		while (b->b_deferred > 0) {
461 			struct pfsync_deferral *pd =
462 			    TAILQ_FIRST(&b->b_deferrals);
463 
464 			ret = callout_stop(&pd->pd_tmo);
465 			PFSYNC_BUCKET_UNLOCK(b);
466 			if (ret > 0) {
467 				pfsync_undefer(pd, 1);
468 			} else {
469 				callout_drain(&pd->pd_tmo);
470 			}
471 			PFSYNC_BUCKET_LOCK(b);
472 		}
473 		MPASS(b->b_deferred == 0);
474 		MPASS(TAILQ_EMPTY(&b->b_deferrals));
475 		PFSYNC_BUCKET_UNLOCK(b);
476 
477 		callout_drain(&b->b_tmo);
478 	}
479 
480 	callout_drain(&sc->sc_bulkfail_tmo);
481 	callout_drain(&sc->sc_bulk_tmo);
482 
483 	if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
484 		(*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
485 	bpfdetach(ifp);
486 	if_detach(ifp);
487 
488 	pfsync_drop(sc);
489 
490 	if_free(ifp);
491 	pfsync_multicast_cleanup(sc);
492 	mtx_destroy(&sc->sc_mtx);
493 	mtx_destroy(&sc->sc_bulk_mtx);
494 
495 	free(sc->sc_buckets, M_PFSYNC);
496 	free(sc, M_PFSYNC);
497 
498 	V_pfsyncif = NULL;
499 }
500 
501 static int
502 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
503     struct pf_state_peer *d)
504 {
505 	if (s->scrub.scrub_flag && d->scrub == NULL) {
506 		d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
507 		if (d->scrub == NULL)
508 			return (ENOMEM);
509 	}
510 
511 	return (0);
512 }
513 
514 static int
515 pfsync_state_import(union pfsync_state_union *sp, int flags, int msg_version)
516 {
517 	struct pfsync_softc *sc = V_pfsyncif;
518 #ifndef	__NO_STRICT_ALIGNMENT
519 	struct pfsync_state_key key[2];
520 #endif
521 	struct pfsync_state_key *kw, *ks;
522 	struct pf_kstate	*st = NULL;
523 	struct pf_state_key *skw = NULL, *sks = NULL;
524 	struct pf_krule *r = NULL;
525 	struct pfi_kkif	*kif;
526 	int error;
527 
528 	PF_RULES_RASSERT();
529 
530 	if (sp->pfs_1301.creatorid == 0) {
531 		if (V_pf_status.debug >= PF_DEBUG_MISC)
532 			printf("%s: invalid creator id: %08x\n", __func__,
533 			    ntohl(sp->pfs_1301.creatorid));
534 		return (EINVAL);
535 	}
536 
537 	if ((kif = pfi_kkif_find(sp->pfs_1301.ifname)) == NULL) {
538 		if (V_pf_status.debug >= PF_DEBUG_MISC)
539 			printf("%s: unknown interface: %s\n", __func__,
540 			    sp->pfs_1301.ifname);
541 		if (flags & PFSYNC_SI_IOCTL)
542 			return (EINVAL);
543 		return (0);	/* skip this state */
544 	}
545 
546 	/*
547 	 * If the ruleset checksums match or the state is coming from the ioctl,
548 	 * it's safe to associate the state with the rule of that number.
549 	 */
550 	if (sp->pfs_1301.rule != htonl(-1) && sp->pfs_1301.anchor == htonl(-1) &&
551 	    (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->pfs_1301.rule) <
552 	    pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
553 		r = pf_main_ruleset.rules[
554 		    PF_RULESET_FILTER].active.ptr_array[ntohl(sp->pfs_1301.rule)];
555 	else
556 		r = &V_pf_default_rule;
557 
558 	if ((r->max_states &&
559 	    counter_u64_fetch(r->states_cur) >= r->max_states))
560 		goto cleanup;
561 
562 	/*
563 	 * XXXGL: consider M_WAITOK in ioctl path after.
564 	 */
565 	st = pf_alloc_state(M_NOWAIT);
566 	if (__predict_false(st == NULL))
567 		goto cleanup;
568 
569 	if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
570 		goto cleanup;
571 
572 #ifndef	__NO_STRICT_ALIGNMENT
573 	bcopy(&sp->pfs_1301.key, key, sizeof(struct pfsync_state_key) * 2);
574 	kw = &key[PF_SK_WIRE];
575 	ks = &key[PF_SK_STACK];
576 #else
577 	kw = &sp->pfs_1301.key[PF_SK_WIRE];
578 	ks = &sp->pfs_1301.key[PF_SK_STACK];
579 #endif
580 
581 	if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->pfs_1301.af) ||
582 	    PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->pfs_1301.af) ||
583 	    kw->port[0] != ks->port[0] ||
584 	    kw->port[1] != ks->port[1]) {
585 		sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
586 		if (sks == NULL)
587 			goto cleanup;
588 	} else
589 		sks = skw;
590 
591 	/* allocate memory for scrub info */
592 	if (pfsync_alloc_scrub_memory(&sp->pfs_1301.src, &st->src) ||
593 	    pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst))
594 		goto cleanup;
595 
596 	/* Copy to state key(s). */
597 	skw->addr[0] = kw->addr[0];
598 	skw->addr[1] = kw->addr[1];
599 	skw->port[0] = kw->port[0];
600 	skw->port[1] = kw->port[1];
601 	skw->proto = sp->pfs_1301.proto;
602 	skw->af = sp->pfs_1301.af;
603 	if (sks != skw) {
604 		sks->addr[0] = ks->addr[0];
605 		sks->addr[1] = ks->addr[1];
606 		sks->port[0] = ks->port[0];
607 		sks->port[1] = ks->port[1];
608 		sks->proto = sp->pfs_1301.proto;
609 		sks->af = sp->pfs_1301.af;
610 	}
611 
612 	/* copy to state */
613 	bcopy(&sp->pfs_1301.rt_addr, &st->rt_addr, sizeof(st->rt_addr));
614 	st->creation = (time_uptime - ntohl(sp->pfs_1301.creation)) * 1000;
615 	st->expire = pf_get_uptime();
616 	if (sp->pfs_1301.expire) {
617 		uint32_t timeout;
618 
619 		timeout = r->timeout[sp->pfs_1301.timeout];
620 		if (!timeout)
621 			timeout = V_pf_default_rule.timeout[sp->pfs_1301.timeout];
622 
623 		/* sp->expire may have been adaptively scaled by export. */
624 		st->expire -= (timeout - ntohl(sp->pfs_1301.expire)) * 1000;
625 	}
626 
627 	st->direction = sp->pfs_1301.direction;
628 	st->act.log = sp->pfs_1301.log;
629 	st->timeout = sp->pfs_1301.timeout;
630 
631 	switch (msg_version) {
632 		case PFSYNC_MSG_VERSION_1301:
633 			st->state_flags = sp->pfs_1301.state_flags;
634 			/*
635 			 * In FreeBSD 13 pfsync lacks many attributes. Copy them
636 			 * from the rule if possible. If rule can't be matched
637 			 * clear any set options as we can't recover their
638 			 * parameters.
639 			*/
640 			if (r == &V_pf_default_rule) {
641 				st->state_flags &= ~PFSTATE_SETMASK;
642 			} else {
643 				/*
644 				 * Similar to pf_rule_to_actions(). This code
645 				 * won't set the actions properly if they come
646 				 * from multiple "match" rules as only rule
647 				 * creating the state is send over pfsync.
648 				 */
649 				st->act.qid = r->qid;
650 				st->act.pqid = r->pqid;
651 				st->act.rtableid = r->rtableid;
652 				if (r->scrub_flags & PFSTATE_SETTOS)
653 					st->act.set_tos = r->set_tos;
654 				st->act.min_ttl = r->min_ttl;
655 				st->act.max_mss = r->max_mss;
656 				st->state_flags |= (r->scrub_flags &
657 				    (PFSTATE_NODF|PFSTATE_RANDOMID|
658 				    PFSTATE_SETTOS|PFSTATE_SCRUB_TCP|
659 				    PFSTATE_SETPRIO));
660 				if (r->dnpipe || r->dnrpipe) {
661 					if (r->free_flags & PFRULE_DN_IS_PIPE)
662 						st->state_flags |= PFSTATE_DN_IS_PIPE;
663 					else
664 						st->state_flags &= ~PFSTATE_DN_IS_PIPE;
665 				}
666 				st->act.dnpipe = r->dnpipe;
667 				st->act.dnrpipe = r->dnrpipe;
668 			}
669 			break;
670 		case PFSYNC_MSG_VERSION_1400:
671 			st->state_flags = ntohs(sp->pfs_1400.state_flags);
672 			st->act.qid = ntohs(sp->pfs_1400.qid);
673 			st->act.pqid = ntohs(sp->pfs_1400.pqid);
674 			st->act.dnpipe = ntohs(sp->pfs_1400.dnpipe);
675 			st->act.dnrpipe = ntohs(sp->pfs_1400.dnrpipe);
676 			st->act.rtableid = ntohl(sp->pfs_1400.rtableid);
677 			st->act.min_ttl = sp->pfs_1400.min_ttl;
678 			st->act.set_tos = sp->pfs_1400.set_tos;
679 			st->act.max_mss = ntohs(sp->pfs_1400.max_mss);
680 			st->act.set_prio[0] = sp->pfs_1400.set_prio[0];
681 			st->act.set_prio[1] = sp->pfs_1400.set_prio[1];
682 			st->rt = sp->pfs_1400.rt;
683 			if (st->rt && (st->rt_kif = pfi_kkif_find(sp->pfs_1400.rt_ifname)) == NULL) {
684 				if (V_pf_status.debug >= PF_DEBUG_MISC)
685 					printf("%s: unknown route interface: %s\n",
686 					    __func__, sp->pfs_1400.rt_ifname);
687 				if (flags & PFSYNC_SI_IOCTL)
688 					error = EINVAL;
689 				else
690 					error = 0;
691 				goto cleanup_keys;
692 			}
693 			break;
694 		default:
695 			panic("%s: Unsupported pfsync_msg_version %d",
696 			    __func__, msg_version);
697 	}
698 
699 	st->id = sp->pfs_1301.id;
700 	st->creatorid = sp->pfs_1301.creatorid;
701 	pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src);
702 	pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
703 
704 	st->rule.ptr = r;
705 	st->nat_rule.ptr = NULL;
706 	st->anchor.ptr = NULL;
707 
708 	st->pfsync_time = time_uptime;
709 	st->sync_state = PFSYNC_S_NONE;
710 
711 	if (!(flags & PFSYNC_SI_IOCTL))
712 		st->state_flags |= PFSTATE_NOSYNC;
713 
714 	if ((error = pf_state_insert(kif, kif, skw, sks, st)) != 0)
715 		goto cleanup_state;
716 
717 	/* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
718 	counter_u64_add(r->states_cur, 1);
719 	counter_u64_add(r->states_tot, 1);
720 
721 	if (!(flags & PFSYNC_SI_IOCTL)) {
722 		st->state_flags &= ~PFSTATE_NOSYNC;
723 		if (st->state_flags & PFSTATE_ACK) {
724 			struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
725 			PFSYNC_BUCKET_LOCK(b);
726 			pfsync_q_ins(st, PFSYNC_S_IACK, true);
727 			PFSYNC_BUCKET_UNLOCK(b);
728 
729 			pfsync_push_all(sc);
730 		}
731 	}
732 	st->state_flags &= ~PFSTATE_ACK;
733 	PF_STATE_UNLOCK(st);
734 
735 	return (0);
736 
737 cleanup:
738 	error = ENOMEM;
739 cleanup_keys:
740 	if (skw == sks)
741 		sks = NULL;
742 	uma_zfree(V_pf_state_key_z, skw);
743 	uma_zfree(V_pf_state_key_z, sks);
744 
745 cleanup_state:	/* pf_state_insert() frees the state keys. */
746 	if (st) {
747 		st->timeout = PFTM_UNLINKED; /* appease an assert */
748 		pf_free_state(st);
749 	}
750 	return (error);
751 }
752 
753 #ifdef INET
754 static int
755 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
756 {
757 	struct pfsync_softc *sc = V_pfsyncif;
758 	struct mbuf *m = *mp;
759 	struct ip *ip = mtod(m, struct ip *);
760 	struct pfsync_header *ph;
761 	struct pfsync_subheader subh;
762 
763 	int offset, len, flags = 0;
764 	int rv;
765 	uint16_t count;
766 
767 	PF_RULES_RLOCK_TRACKER;
768 
769 	*mp = NULL;
770 	V_pfsyncstats.pfsyncs_ipackets++;
771 
772 	/* Verify that we have a sync interface configured. */
773 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
774 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
775 		goto done;
776 
777 	/* verify that the packet came in on the right interface */
778 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
779 		V_pfsyncstats.pfsyncs_badif++;
780 		goto done;
781 	}
782 
783 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
784 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
785 	/* verify that the IP TTL is 255. */
786 	if (ip->ip_ttl != PFSYNC_DFLTTL) {
787 		V_pfsyncstats.pfsyncs_badttl++;
788 		goto done;
789 	}
790 
791 	offset = ip->ip_hl << 2;
792 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
793 		V_pfsyncstats.pfsyncs_hdrops++;
794 		goto done;
795 	}
796 
797 	if (offset + sizeof(*ph) > m->m_len) {
798 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
799 			V_pfsyncstats.pfsyncs_hdrops++;
800 			return (IPPROTO_DONE);
801 		}
802 		ip = mtod(m, struct ip *);
803 	}
804 	ph = (struct pfsync_header *)((char *)ip + offset);
805 
806 	/* verify the version */
807 	if (ph->version != PFSYNC_VERSION) {
808 		V_pfsyncstats.pfsyncs_badver++;
809 		goto done;
810 	}
811 
812 	len = ntohs(ph->len) + offset;
813 	if (m->m_pkthdr.len < len) {
814 		V_pfsyncstats.pfsyncs_badlen++;
815 		goto done;
816 	}
817 
818 	/*
819 	 * Trusting pf_chksum during packet processing, as well as seeking
820 	 * in interface name tree, require holding PF_RULES_RLOCK().
821 	 */
822 	PF_RULES_RLOCK();
823 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
824 		flags = PFSYNC_SI_CKSUM;
825 
826 	offset += sizeof(*ph);
827 	while (offset <= len - sizeof(subh)) {
828 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
829 		offset += sizeof(subh);
830 
831 		if (subh.action >= PFSYNC_ACT_MAX) {
832 			V_pfsyncstats.pfsyncs_badact++;
833 			PF_RULES_RUNLOCK();
834 			goto done;
835 		}
836 
837 		count = ntohs(subh.count);
838 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
839 		rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action);
840 		if (rv == -1) {
841 			PF_RULES_RUNLOCK();
842 			return (IPPROTO_DONE);
843 		}
844 
845 		offset += rv;
846 	}
847 	PF_RULES_RUNLOCK();
848 
849 done:
850 	m_freem(m);
851 	return (IPPROTO_DONE);
852 }
853 #endif
854 
855 #ifdef INET6
856 static int
857 pfsync6_input(struct mbuf **mp, int *offp __unused, int proto __unused)
858 {
859 	struct pfsync_softc *sc = V_pfsyncif;
860 	struct mbuf *m = *mp;
861 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
862 	struct pfsync_header *ph;
863 	struct pfsync_subheader subh;
864 
865 	int offset, len, flags = 0;
866 	int rv;
867 	uint16_t count;
868 
869 	PF_RULES_RLOCK_TRACKER;
870 
871 	*mp = NULL;
872 	V_pfsyncstats.pfsyncs_ipackets++;
873 
874 	/* Verify that we have a sync interface configured. */
875 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
876 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
877 		goto done;
878 
879 	/* verify that the packet came in on the right interface */
880 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
881 		V_pfsyncstats.pfsyncs_badif++;
882 		goto done;
883 	}
884 
885 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
886 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
887 	/* verify that the IP TTL is 255. */
888 	if (ip6->ip6_hlim != PFSYNC_DFLTTL) {
889 		V_pfsyncstats.pfsyncs_badttl++;
890 		goto done;
891 	}
892 
893 
894 	offset = sizeof(*ip6);
895 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
896 		V_pfsyncstats.pfsyncs_hdrops++;
897 		goto done;
898 	}
899 
900 	if (offset + sizeof(*ph) > m->m_len) {
901 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
902 			V_pfsyncstats.pfsyncs_hdrops++;
903 			return (IPPROTO_DONE);
904 		}
905 		ip6 = mtod(m, struct ip6_hdr *);
906 	}
907 	ph = (struct pfsync_header *)((char *)ip6 + offset);
908 
909 	/* verify the version */
910 	if (ph->version != PFSYNC_VERSION) {
911 		V_pfsyncstats.pfsyncs_badver++;
912 		goto done;
913 	}
914 
915 	len = ntohs(ph->len) + offset;
916 	if (m->m_pkthdr.len < len) {
917 		V_pfsyncstats.pfsyncs_badlen++;
918 		goto done;
919 	}
920 
921 	/*
922 	 * Trusting pf_chksum during packet processing, as well as seeking
923 	 * in interface name tree, require holding PF_RULES_RLOCK().
924 	 */
925 	PF_RULES_RLOCK();
926 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
927 		flags = PFSYNC_SI_CKSUM;
928 
929 	offset += sizeof(*ph);
930 	while (offset <= len - sizeof(subh)) {
931 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
932 		offset += sizeof(subh);
933 
934 		if (subh.action >= PFSYNC_ACT_MAX) {
935 			V_pfsyncstats.pfsyncs_badact++;
936 			PF_RULES_RUNLOCK();
937 			goto done;
938 		}
939 
940 		count = ntohs(subh.count);
941 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
942 		rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action);
943 		if (rv == -1) {
944 			PF_RULES_RUNLOCK();
945 			return (IPPROTO_DONE);
946 		}
947 
948 		offset += rv;
949 	}
950 	PF_RULES_RUNLOCK();
951 
952 done:
953 	m_freem(m);
954 	return (IPPROTO_DONE);
955 }
956 #endif
957 
958 static int
959 pfsync_in_clr(struct mbuf *m, int offset, int count, int flags, int action)
960 {
961 	struct pfsync_clr *clr;
962 	struct mbuf *mp;
963 	int len = sizeof(*clr) * count;
964 	int i, offp;
965 	u_int32_t creatorid;
966 
967 	mp = m_pulldown(m, offset, len, &offp);
968 	if (mp == NULL) {
969 		V_pfsyncstats.pfsyncs_badlen++;
970 		return (-1);
971 	}
972 	clr = (struct pfsync_clr *)(mp->m_data + offp);
973 
974 	for (i = 0; i < count; i++) {
975 		creatorid = clr[i].creatorid;
976 
977 		if (clr[i].ifname[0] != '\0' &&
978 		    pfi_kkif_find(clr[i].ifname) == NULL)
979 			continue;
980 
981 		for (int i = 0; i <= pf_hashmask; i++) {
982 			struct pf_idhash *ih = &V_pf_idhash[i];
983 			struct pf_kstate *s;
984 relock:
985 			PF_HASHROW_LOCK(ih);
986 			LIST_FOREACH(s, &ih->states, entry) {
987 				if (s->creatorid == creatorid) {
988 					s->state_flags |= PFSTATE_NOSYNC;
989 					pf_unlink_state(s);
990 					goto relock;
991 				}
992 			}
993 			PF_HASHROW_UNLOCK(ih);
994 		}
995 	}
996 
997 	return (len);
998 }
999 
1000 static int
1001 pfsync_in_ins(struct mbuf *m, int offset, int count, int flags, int action)
1002 {
1003 	struct mbuf *mp;
1004 	union pfsync_state_union *sa, *sp;
1005 	int i, offp, total_len, msg_version, msg_len;
1006 
1007 	switch (action) {
1008 		case PFSYNC_ACT_INS_1301:
1009 			msg_len = sizeof(struct pfsync_state_1301);
1010 			total_len = msg_len * count;
1011 			msg_version = PFSYNC_MSG_VERSION_1301;
1012 			break;
1013 		case PFSYNC_ACT_INS_1400:
1014 			msg_len = sizeof(struct pfsync_state_1400);
1015 			total_len = msg_len * count;
1016 			msg_version = PFSYNC_MSG_VERSION_1400;
1017 			break;
1018 		default:
1019 			V_pfsyncstats.pfsyncs_badact++;
1020 			return (-1);
1021 	}
1022 
1023 	mp = m_pulldown(m, offset, total_len, &offp);
1024 	if (mp == NULL) {
1025 		V_pfsyncstats.pfsyncs_badlen++;
1026 		return (-1);
1027 	}
1028 	sa = (union pfsync_state_union *)(mp->m_data + offp);
1029 
1030 	for (i = 0; i < count; i++) {
1031 		sp = (union pfsync_state_union *)((char *)sa + msg_len * i);
1032 
1033 		/* Check for invalid values. */
1034 		if (sp->pfs_1301.timeout >= PFTM_MAX ||
1035 		    sp->pfs_1301.src.state > PF_TCPS_PROXY_DST ||
1036 		    sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST ||
1037 		    sp->pfs_1301.direction > PF_OUT ||
1038 		    (sp->pfs_1301.af != AF_INET &&
1039 		    sp->pfs_1301.af != AF_INET6)) {
1040 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1041 				printf("%s: invalid value\n", __func__);
1042 			V_pfsyncstats.pfsyncs_badval++;
1043 			continue;
1044 		}
1045 
1046 		if (pfsync_state_import(sp, flags, msg_version) == ENOMEM)
1047 			/* Drop out, but process the rest of the actions. */
1048 			break;
1049 	}
1050 
1051 	return (total_len);
1052 }
1053 
1054 static int
1055 pfsync_in_iack(struct mbuf *m, int offset, int count, int flags, int action)
1056 {
1057 	struct pfsync_ins_ack *ia, *iaa;
1058 	struct pf_kstate *st;
1059 
1060 	struct mbuf *mp;
1061 	int len = count * sizeof(*ia);
1062 	int offp, i;
1063 
1064 	mp = m_pulldown(m, offset, len, &offp);
1065 	if (mp == NULL) {
1066 		V_pfsyncstats.pfsyncs_badlen++;
1067 		return (-1);
1068 	}
1069 	iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
1070 
1071 	for (i = 0; i < count; i++) {
1072 		ia = &iaa[i];
1073 
1074 		st = pf_find_state_byid(ia->id, ia->creatorid);
1075 		if (st == NULL)
1076 			continue;
1077 
1078 		if (st->state_flags & PFSTATE_ACK) {
1079 			pfsync_undefer_state(st, 0);
1080 		}
1081 		PF_STATE_UNLOCK(st);
1082 	}
1083 	/*
1084 	 * XXX this is not yet implemented, but we know the size of the
1085 	 * message so we can skip it.
1086 	 */
1087 
1088 	return (count * sizeof(struct pfsync_ins_ack));
1089 }
1090 
1091 static int
1092 pfsync_upd_tcp(struct pf_kstate *st, struct pfsync_state_peer *src,
1093     struct pfsync_state_peer *dst)
1094 {
1095 	int sync = 0;
1096 
1097 	PF_STATE_LOCK_ASSERT(st);
1098 
1099 	/*
1100 	 * The state should never go backwards except
1101 	 * for syn-proxy states.  Neither should the
1102 	 * sequence window slide backwards.
1103 	 */
1104 	if ((st->src.state > src->state &&
1105 	    (st->src.state < PF_TCPS_PROXY_SRC ||
1106 	    src->state >= PF_TCPS_PROXY_SRC)) ||
1107 
1108 	    (st->src.state == src->state &&
1109 	    SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
1110 		sync++;
1111 	else
1112 		pf_state_peer_ntoh(src, &st->src);
1113 
1114 	if ((st->dst.state > dst->state) ||
1115 
1116 	    (st->dst.state >= TCPS_SYN_SENT &&
1117 	    SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
1118 		sync++;
1119 	else
1120 		pf_state_peer_ntoh(dst, &st->dst);
1121 
1122 	return (sync);
1123 }
1124 
1125 static int
1126 pfsync_in_upd(struct mbuf *m, int offset, int count, int flags, int action)
1127 {
1128 	struct pfsync_softc *sc = V_pfsyncif;
1129 	union pfsync_state_union *sa, *sp;
1130 	struct pf_kstate *st;
1131 	struct mbuf *mp;
1132 	int sync, offp, i, total_len, msg_len, msg_version;
1133 
1134 	switch (action) {
1135 		case PFSYNC_ACT_UPD_1301:
1136 			msg_len = sizeof(struct pfsync_state_1301);
1137 			total_len = msg_len * count;
1138 			msg_version = PFSYNC_MSG_VERSION_1301;
1139 			break;
1140 		case PFSYNC_ACT_UPD_1400:
1141 			msg_len = sizeof(struct pfsync_state_1400);
1142 			total_len = msg_len * count;
1143 			msg_version = PFSYNC_MSG_VERSION_1400;
1144 			break;
1145 		default:
1146 			V_pfsyncstats.pfsyncs_badact++;
1147 			return (-1);
1148 	}
1149 
1150 	mp = m_pulldown(m, offset, total_len, &offp);
1151 	if (mp == NULL) {
1152 		V_pfsyncstats.pfsyncs_badlen++;
1153 		return (-1);
1154 	}
1155 	sa = (union pfsync_state_union *)(mp->m_data + offp);
1156 
1157 	for (i = 0; i < count; i++) {
1158 		sp = (union pfsync_state_union *)((char *)sa + msg_len * i);
1159 
1160 		/* check for invalid values */
1161 		if (sp->pfs_1301.timeout >= PFTM_MAX ||
1162 		    sp->pfs_1301.src.state > PF_TCPS_PROXY_DST ||
1163 		    sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST) {
1164 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1165 				printf("pfsync_input: PFSYNC_ACT_UPD: "
1166 				    "invalid value\n");
1167 			}
1168 			V_pfsyncstats.pfsyncs_badval++;
1169 			continue;
1170 		}
1171 
1172 		st = pf_find_state_byid(sp->pfs_1301.id, sp->pfs_1301.creatorid);
1173 		if (st == NULL) {
1174 			/* insert the update */
1175 			if (pfsync_state_import(sp, flags, msg_version))
1176 				V_pfsyncstats.pfsyncs_badstate++;
1177 			continue;
1178 		}
1179 
1180 		if (st->state_flags & PFSTATE_ACK) {
1181 			pfsync_undefer_state(st, 1);
1182 		}
1183 
1184 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1185 			sync = pfsync_upd_tcp(st, &sp->pfs_1301.src, &sp->pfs_1301.dst);
1186 		else {
1187 			sync = 0;
1188 
1189 			/*
1190 			 * Non-TCP protocol state machine always go
1191 			 * forwards
1192 			 */
1193 			if (st->src.state > sp->pfs_1301.src.state)
1194 				sync++;
1195 			else
1196 				pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src);
1197 			if (st->dst.state > sp->pfs_1301.dst.state)
1198 				sync++;
1199 			else
1200 				pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
1201 		}
1202 		if (sync < 2) {
1203 			pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst);
1204 			pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
1205 			st->expire = pf_get_uptime();
1206 			st->timeout = sp->pfs_1301.timeout;
1207 		}
1208 		st->pfsync_time = time_uptime;
1209 
1210 		if (sync) {
1211 			V_pfsyncstats.pfsyncs_stale++;
1212 
1213 			pfsync_update_state(st);
1214 			PF_STATE_UNLOCK(st);
1215 			pfsync_push_all(sc);
1216 			continue;
1217 		}
1218 		PF_STATE_UNLOCK(st);
1219 	}
1220 
1221 	return (total_len);
1222 }
1223 
1224 static int
1225 pfsync_in_upd_c(struct mbuf *m, int offset, int count, int flags, int action)
1226 {
1227 	struct pfsync_softc *sc = V_pfsyncif;
1228 	struct pfsync_upd_c *ua, *up;
1229 	struct pf_kstate *st;
1230 	int len = count * sizeof(*up);
1231 	int sync;
1232 	struct mbuf *mp;
1233 	int offp, i;
1234 
1235 	mp = m_pulldown(m, offset, len, &offp);
1236 	if (mp == NULL) {
1237 		V_pfsyncstats.pfsyncs_badlen++;
1238 		return (-1);
1239 	}
1240 	ua = (struct pfsync_upd_c *)(mp->m_data + offp);
1241 
1242 	for (i = 0; i < count; i++) {
1243 		up = &ua[i];
1244 
1245 		/* check for invalid values */
1246 		if (up->timeout >= PFTM_MAX ||
1247 		    up->src.state > PF_TCPS_PROXY_DST ||
1248 		    up->dst.state > PF_TCPS_PROXY_DST) {
1249 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1250 				printf("pfsync_input: "
1251 				    "PFSYNC_ACT_UPD_C: "
1252 				    "invalid value\n");
1253 			}
1254 			V_pfsyncstats.pfsyncs_badval++;
1255 			continue;
1256 		}
1257 
1258 		st = pf_find_state_byid(up->id, up->creatorid);
1259 		if (st == NULL) {
1260 			/* We don't have this state. Ask for it. */
1261 			PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1262 			pfsync_request_update(up->creatorid, up->id);
1263 			PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1264 			continue;
1265 		}
1266 
1267 		if (st->state_flags & PFSTATE_ACK) {
1268 			pfsync_undefer_state(st, 1);
1269 		}
1270 
1271 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1272 			sync = pfsync_upd_tcp(st, &up->src, &up->dst);
1273 		else {
1274 			sync = 0;
1275 
1276 			/*
1277 			 * Non-TCP protocol state machine always go
1278 			 * forwards
1279 			 */
1280 			if (st->src.state > up->src.state)
1281 				sync++;
1282 			else
1283 				pf_state_peer_ntoh(&up->src, &st->src);
1284 			if (st->dst.state > up->dst.state)
1285 				sync++;
1286 			else
1287 				pf_state_peer_ntoh(&up->dst, &st->dst);
1288 		}
1289 		if (sync < 2) {
1290 			pfsync_alloc_scrub_memory(&up->dst, &st->dst);
1291 			pf_state_peer_ntoh(&up->dst, &st->dst);
1292 			st->expire = pf_get_uptime();
1293 			st->timeout = up->timeout;
1294 		}
1295 		st->pfsync_time = time_uptime;
1296 
1297 		if (sync) {
1298 			V_pfsyncstats.pfsyncs_stale++;
1299 
1300 			pfsync_update_state(st);
1301 			PF_STATE_UNLOCK(st);
1302 			pfsync_push_all(sc);
1303 			continue;
1304 		}
1305 		PF_STATE_UNLOCK(st);
1306 	}
1307 
1308 	return (len);
1309 }
1310 
1311 static int
1312 pfsync_in_ureq(struct mbuf *m, int offset, int count, int flags, int action)
1313 {
1314 	struct pfsync_upd_req *ur, *ura;
1315 	struct mbuf *mp;
1316 	int len = count * sizeof(*ur);
1317 	int i, offp;
1318 
1319 	struct pf_kstate *st;
1320 
1321 	mp = m_pulldown(m, offset, len, &offp);
1322 	if (mp == NULL) {
1323 		V_pfsyncstats.pfsyncs_badlen++;
1324 		return (-1);
1325 	}
1326 	ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1327 
1328 	for (i = 0; i < count; i++) {
1329 		ur = &ura[i];
1330 
1331 		if (ur->id == 0 && ur->creatorid == 0)
1332 			pfsync_bulk_start();
1333 		else {
1334 			st = pf_find_state_byid(ur->id, ur->creatorid);
1335 			if (st == NULL) {
1336 				V_pfsyncstats.pfsyncs_badstate++;
1337 				continue;
1338 			}
1339 			if (st->state_flags & PFSTATE_NOSYNC) {
1340 				PF_STATE_UNLOCK(st);
1341 				continue;
1342 			}
1343 
1344 			pfsync_update_state_req(st);
1345 			PF_STATE_UNLOCK(st);
1346 		}
1347 	}
1348 
1349 	return (len);
1350 }
1351 
1352 static int
1353 pfsync_in_del_c(struct mbuf *m, int offset, int count, int flags, int action)
1354 {
1355 	struct mbuf *mp;
1356 	struct pfsync_del_c *sa, *sp;
1357 	struct pf_kstate *st;
1358 	int len = count * sizeof(*sp);
1359 	int offp, i;
1360 
1361 	mp = m_pulldown(m, offset, len, &offp);
1362 	if (mp == NULL) {
1363 		V_pfsyncstats.pfsyncs_badlen++;
1364 		return (-1);
1365 	}
1366 	sa = (struct pfsync_del_c *)(mp->m_data + offp);
1367 
1368 	for (i = 0; i < count; i++) {
1369 		sp = &sa[i];
1370 
1371 		st = pf_find_state_byid(sp->id, sp->creatorid);
1372 		if (st == NULL) {
1373 			V_pfsyncstats.pfsyncs_badstate++;
1374 			continue;
1375 		}
1376 
1377 		st->state_flags |= PFSTATE_NOSYNC;
1378 		pf_unlink_state(st);
1379 	}
1380 
1381 	return (len);
1382 }
1383 
1384 static int
1385 pfsync_in_bus(struct mbuf *m, int offset, int count, int flags, int action)
1386 {
1387 	struct pfsync_softc *sc = V_pfsyncif;
1388 	struct pfsync_bus *bus;
1389 	struct mbuf *mp;
1390 	int len = count * sizeof(*bus);
1391 	int offp;
1392 
1393 	PFSYNC_BLOCK(sc);
1394 
1395 	/* If we're not waiting for a bulk update, who cares. */
1396 	if (sc->sc_ureq_sent == 0) {
1397 		PFSYNC_BUNLOCK(sc);
1398 		return (len);
1399 	}
1400 
1401 	mp = m_pulldown(m, offset, len, &offp);
1402 	if (mp == NULL) {
1403 		PFSYNC_BUNLOCK(sc);
1404 		V_pfsyncstats.pfsyncs_badlen++;
1405 		return (-1);
1406 	}
1407 	bus = (struct pfsync_bus *)(mp->m_data + offp);
1408 
1409 	switch (bus->status) {
1410 	case PFSYNC_BUS_START:
1411 		callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1412 		    V_pf_limits[PF_LIMIT_STATES].limit /
1413 		    ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1414 		    sizeof(union pfsync_state_union)),
1415 		    pfsync_bulk_fail, sc);
1416 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1417 			printf("pfsync: received bulk update start\n");
1418 		break;
1419 
1420 	case PFSYNC_BUS_END:
1421 		if (time_uptime - ntohl(bus->endtime) >=
1422 		    sc->sc_ureq_sent) {
1423 			/* that's it, we're happy */
1424 			sc->sc_ureq_sent = 0;
1425 			sc->sc_bulk_tries = 0;
1426 			callout_stop(&sc->sc_bulkfail_tmo);
1427 			if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1428 				(*carp_demote_adj_p)(-V_pfsync_carp_adj,
1429 				    "pfsync bulk done");
1430 			sc->sc_flags |= PFSYNCF_OK;
1431 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1432 				printf("pfsync: received valid "
1433 				    "bulk update end\n");
1434 		} else {
1435 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1436 				printf("pfsync: received invalid "
1437 				    "bulk update end: bad timestamp\n");
1438 		}
1439 		break;
1440 	}
1441 	PFSYNC_BUNLOCK(sc);
1442 
1443 	return (len);
1444 }
1445 
1446 static int
1447 pfsync_in_tdb(struct mbuf *m, int offset, int count, int flags, int action)
1448 {
1449 	int len = count * sizeof(struct pfsync_tdb);
1450 
1451 #if defined(IPSEC)
1452 	struct pfsync_tdb *tp;
1453 	struct mbuf *mp;
1454 	int offp;
1455 	int i;
1456 	int s;
1457 
1458 	mp = m_pulldown(m, offset, len, &offp);
1459 	if (mp == NULL) {
1460 		V_pfsyncstats.pfsyncs_badlen++;
1461 		return (-1);
1462 	}
1463 	tp = (struct pfsync_tdb *)(mp->m_data + offp);
1464 
1465 	for (i = 0; i < count; i++)
1466 		pfsync_update_net_tdb(&tp[i]);
1467 #endif
1468 
1469 	return (len);
1470 }
1471 
1472 #if defined(IPSEC)
1473 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1474 static void
1475 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1476 {
1477 	struct tdb		*tdb;
1478 	int			 s;
1479 
1480 	/* check for invalid values */
1481 	if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1482 	    (pt->dst.sa.sa_family != AF_INET &&
1483 	    pt->dst.sa.sa_family != AF_INET6))
1484 		goto bad;
1485 
1486 	tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1487 	if (tdb) {
1488 		pt->rpl = ntohl(pt->rpl);
1489 		pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1490 
1491 		/* Neither replay nor byte counter should ever decrease. */
1492 		if (pt->rpl < tdb->tdb_rpl ||
1493 		    pt->cur_bytes < tdb->tdb_cur_bytes) {
1494 			goto bad;
1495 		}
1496 
1497 		tdb->tdb_rpl = pt->rpl;
1498 		tdb->tdb_cur_bytes = pt->cur_bytes;
1499 	}
1500 	return;
1501 
1502 bad:
1503 	if (V_pf_status.debug >= PF_DEBUG_MISC)
1504 		printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1505 		    "invalid value\n");
1506 	V_pfsyncstats.pfsyncs_badstate++;
1507 	return;
1508 }
1509 #endif
1510 
1511 static int
1512 pfsync_in_eof(struct mbuf *m, int offset, int count, int flags, int action)
1513 {
1514 	/* check if we are at the right place in the packet */
1515 	if (offset != m->m_pkthdr.len)
1516 		V_pfsyncstats.pfsyncs_badlen++;
1517 
1518 	/* we're done. free and let the caller return */
1519 	m_freem(m);
1520 	return (-1);
1521 }
1522 
1523 static int
1524 pfsync_in_error(struct mbuf *m, int offset, int count, int flags, int action)
1525 {
1526 	V_pfsyncstats.pfsyncs_badact++;
1527 
1528 	m_freem(m);
1529 	return (-1);
1530 }
1531 
1532 static int
1533 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1534 	struct route *rt)
1535 {
1536 	m_freem(m);
1537 	return (0);
1538 }
1539 
1540 /* ARGSUSED */
1541 static int
1542 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1543 {
1544 	struct pfsync_softc *sc = ifp->if_softc;
1545 	struct ifreq *ifr = (struct ifreq *)data;
1546 	struct pfsyncreq pfsyncr;
1547 	size_t nvbuflen;
1548 	int error;
1549 	int c;
1550 
1551 	switch (cmd) {
1552 	case SIOCSIFFLAGS:
1553 		PFSYNC_LOCK(sc);
1554 		if (ifp->if_flags & IFF_UP) {
1555 			ifp->if_drv_flags |= IFF_DRV_RUNNING;
1556 			PFSYNC_UNLOCK(sc);
1557 			pfsync_pointers_init();
1558 		} else {
1559 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1560 			PFSYNC_UNLOCK(sc);
1561 			pfsync_pointers_uninit();
1562 		}
1563 		break;
1564 	case SIOCSIFMTU:
1565 		if (!sc->sc_sync_if ||
1566 		    ifr->ifr_mtu <= PFSYNC_MINPKT ||
1567 		    ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1568 			return (EINVAL);
1569 		if (ifr->ifr_mtu < ifp->if_mtu) {
1570 			for (c = 0; c < pfsync_buckets; c++) {
1571 				PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1572 				if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT)
1573 					pfsync_sendout(1, c);
1574 				PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1575 			}
1576 		}
1577 		ifp->if_mtu = ifr->ifr_mtu;
1578 		break;
1579 	case SIOCGETPFSYNC:
1580 		bzero(&pfsyncr, sizeof(pfsyncr));
1581 		PFSYNC_LOCK(sc);
1582 		if (sc->sc_sync_if) {
1583 			strlcpy(pfsyncr.pfsyncr_syncdev,
1584 			    sc->sc_sync_if->if_xname, IFNAMSIZ);
1585 		}
1586 		pfsyncr.pfsyncr_syncpeer = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
1587 		pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1588 		pfsyncr.pfsyncr_defer = sc->sc_flags;
1589 		PFSYNC_UNLOCK(sc);
1590 		return (copyout(&pfsyncr, ifr_data_get_ptr(ifr),
1591 		    sizeof(pfsyncr)));
1592 
1593 	case SIOCGETPFSYNCNV:
1594 	    {
1595 		nvlist_t *nvl_syncpeer;
1596 		nvlist_t *nvl = nvlist_create(0);
1597 
1598 		if (nvl == NULL)
1599 			return (ENOMEM);
1600 
1601 		if (sc->sc_sync_if)
1602 			nvlist_add_string(nvl, "syncdev", sc->sc_sync_if->if_xname);
1603 		nvlist_add_number(nvl, "maxupdates", sc->sc_maxupdates);
1604 		nvlist_add_number(nvl, "flags", sc->sc_flags);
1605 		nvlist_add_number(nvl, "version", sc->sc_version);
1606 		if ((nvl_syncpeer = pfsync_sockaddr_to_syncpeer_nvlist(&sc->sc_sync_peer)) != NULL)
1607 			nvlist_add_nvlist(nvl, "syncpeer", nvl_syncpeer);
1608 
1609 		void *packed = NULL;
1610 		packed = nvlist_pack(nvl, &nvbuflen);
1611 		if (packed == NULL) {
1612 			free(packed, M_NVLIST);
1613 			nvlist_destroy(nvl);
1614 			return (ENOMEM);
1615 		}
1616 
1617 		if (nvbuflen > ifr->ifr_cap_nv.buf_length) {
1618 			ifr->ifr_cap_nv.length = nvbuflen;
1619 			ifr->ifr_cap_nv.buffer = NULL;
1620 			free(packed, M_NVLIST);
1621 			nvlist_destroy(nvl);
1622 			return (EFBIG);
1623 		}
1624 
1625 		ifr->ifr_cap_nv.length = nvbuflen;
1626 		error = copyout(packed, ifr->ifr_cap_nv.buffer, nvbuflen);
1627 
1628 		nvlist_destroy(nvl);
1629 		nvlist_destroy(nvl_syncpeer);
1630 		free(packed, M_NVLIST);
1631 		break;
1632 	    }
1633 
1634 	case SIOCSETPFSYNC:
1635 	    {
1636 		struct pfsync_kstatus status;
1637 
1638 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1639 			return (error);
1640 		if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr,
1641 		    sizeof(pfsyncr))))
1642 			return (error);
1643 
1644 		memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1645 		pfsync_pfsyncreq_to_kstatus(&pfsyncr, &status);
1646 
1647 		error = pfsync_kstatus_to_softc(&status, sc);
1648 		return (error);
1649 	    }
1650 	case SIOCSETPFSYNCNV:
1651 	    {
1652 		struct pfsync_kstatus status;
1653 		void *data;
1654 		nvlist_t *nvl;
1655 
1656 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1657 			return (error);
1658 		if (ifr->ifr_cap_nv.length > IFR_CAP_NV_MAXBUFSIZE)
1659 			return (EINVAL);
1660 
1661 		data = malloc(ifr->ifr_cap_nv.length, M_TEMP, M_WAITOK);
1662 
1663 		if ((error = copyin(ifr->ifr_cap_nv.buffer, data,
1664 		    ifr->ifr_cap_nv.length)) != 0) {
1665 			free(data, M_TEMP);
1666 			return (error);
1667 		}
1668 
1669 		if ((nvl = nvlist_unpack(data, ifr->ifr_cap_nv.length, 0)) == NULL) {
1670 			free(data, M_TEMP);
1671 			return (EINVAL);
1672 		}
1673 
1674 		memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1675 		pfsync_nvstatus_to_kstatus(nvl, &status);
1676 
1677 		nvlist_destroy(nvl);
1678 		free(data, M_TEMP);
1679 
1680 		error = pfsync_kstatus_to_softc(&status, sc);
1681 		return (error);
1682 	    }
1683 	default:
1684 		return (ENOTTY);
1685 	}
1686 
1687 	return (0);
1688 }
1689 
1690 static void
1691 pfsync_out_state_1301(struct pf_kstate *st, void *buf)
1692 {
1693 	union pfsync_state_union *sp = buf;
1694 
1695 	pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1301);
1696 }
1697 
1698 static void
1699 pfsync_out_state_1400(struct pf_kstate *st, void *buf)
1700 {
1701 	union pfsync_state_union *sp = buf;
1702 
1703 	pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1400);
1704 }
1705 
1706 static void
1707 pfsync_out_iack(struct pf_kstate *st, void *buf)
1708 {
1709 	struct pfsync_ins_ack *iack = buf;
1710 
1711 	iack->id = st->id;
1712 	iack->creatorid = st->creatorid;
1713 }
1714 
1715 static void
1716 pfsync_out_upd_c(struct pf_kstate *st, void *buf)
1717 {
1718 	struct pfsync_upd_c *up = buf;
1719 
1720 	bzero(up, sizeof(*up));
1721 	up->id = st->id;
1722 	pf_state_peer_hton(&st->src, &up->src);
1723 	pf_state_peer_hton(&st->dst, &up->dst);
1724 	up->creatorid = st->creatorid;
1725 	up->timeout = st->timeout;
1726 }
1727 
1728 static void
1729 pfsync_out_del_c(struct pf_kstate *st, void *buf)
1730 {
1731 	struct pfsync_del_c *dp = buf;
1732 
1733 	dp->id = st->id;
1734 	dp->creatorid = st->creatorid;
1735 	st->state_flags |= PFSTATE_NOSYNC;
1736 }
1737 
1738 static void
1739 pfsync_drop(struct pfsync_softc *sc)
1740 {
1741 	struct pf_kstate *st, *next;
1742 	struct pfsync_upd_req_item *ur;
1743 	struct pfsync_bucket *b;
1744 	int c;
1745 	enum pfsync_q_id q;
1746 
1747 	for (c = 0; c < pfsync_buckets; c++) {
1748 		b = &sc->sc_buckets[c];
1749 		for (q = 0; q < PFSYNC_Q_COUNT; q++) {
1750 			if (TAILQ_EMPTY(&b->b_qs[q]))
1751 				continue;
1752 
1753 			TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) {
1754 				KASSERT(st->sync_state == pfsync_qid_sstate[q],
1755 					("%s: st->sync_state == q",
1756 						__func__));
1757 				st->sync_state = PFSYNC_S_NONE;
1758 				pf_release_state(st);
1759 			}
1760 			TAILQ_INIT(&b->b_qs[q]);
1761 		}
1762 
1763 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1764 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1765 			free(ur, M_PFSYNC);
1766 		}
1767 
1768 		b->b_len = PFSYNC_MINPKT;
1769 		b->b_plus = NULL;
1770 	}
1771 }
1772 
1773 static void
1774 pfsync_sendout(int schedswi, int c)
1775 {
1776 	struct pfsync_softc *sc = V_pfsyncif;
1777 	struct ifnet *ifp = sc->sc_ifp;
1778 	struct mbuf *m;
1779 	struct pfsync_header *ph;
1780 	struct pfsync_subheader *subh;
1781 	struct pf_kstate *st, *st_next;
1782 	struct pfsync_upd_req_item *ur;
1783 	struct pfsync_bucket *b = &sc->sc_buckets[c];
1784 	size_t len;
1785 	int aflen, offset, count = 0;
1786 	enum pfsync_q_id q;
1787 
1788 	KASSERT(sc != NULL, ("%s: null sc", __func__));
1789 	KASSERT(b->b_len > PFSYNC_MINPKT,
1790 	    ("%s: sc_len %zu", __func__, b->b_len));
1791 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1792 
1793 	if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1794 		pfsync_drop(sc);
1795 		return;
1796 	}
1797 
1798 	m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1799 	if (m == NULL) {
1800 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1801 		V_pfsyncstats.pfsyncs_onomem++;
1802 		return;
1803 	}
1804 	m->m_data += max_linkhdr;
1805 	bzero(m->m_data, b->b_len);
1806 
1807 	len = b->b_len;
1808 
1809 	/* build the ip header */
1810 	switch (sc->sc_sync_peer.ss_family) {
1811 #ifdef INET
1812 	case AF_INET:
1813 	    {
1814 		struct ip *ip;
1815 
1816 		ip = mtod(m, struct ip *);
1817 		bcopy(&sc->sc_template.ipv4, ip, sizeof(*ip));
1818 		aflen = offset = sizeof(*ip);
1819 
1820 		len -= sizeof(union inet_template) - sizeof(struct ip);
1821 		ip->ip_len = htons(len);
1822 		ip_fillid(ip);
1823 		break;
1824 	    }
1825 #endif
1826 #ifdef INET6
1827 	case AF_INET6:
1828 		{
1829 		struct ip6_hdr *ip6;
1830 
1831 		ip6 = mtod(m, struct ip6_hdr *);
1832 		bcopy(&sc->sc_template.ipv6, ip6, sizeof(*ip6));
1833 		aflen = offset = sizeof(*ip6);
1834 
1835 		len -= sizeof(union inet_template) - sizeof(struct ip6_hdr);
1836 		ip6->ip6_plen = htons(len);
1837 		break;
1838 		}
1839 #endif
1840 	default:
1841 		m_freem(m);
1842 		return;
1843 	}
1844 	m->m_len = m->m_pkthdr.len = len;
1845 
1846 	/* build the pfsync header */
1847 	ph = (struct pfsync_header *)(m->m_data + offset);
1848 	offset += sizeof(*ph);
1849 
1850 	ph->version = PFSYNC_VERSION;
1851 	ph->len = htons(len - aflen);
1852 	bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1853 
1854 	/* walk the queues */
1855 	for (q = 0; q < PFSYNC_Q_COUNT; q++) {
1856 		if (TAILQ_EMPTY(&b->b_qs[q]))
1857 			continue;
1858 
1859 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1860 		offset += sizeof(*subh);
1861 
1862 		count = 0;
1863 		TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) {
1864 			KASSERT(st->sync_state == pfsync_qid_sstate[q],
1865 				("%s: st->sync_state == q",
1866 					__func__));
1867 			/*
1868 			 * XXXGL: some of write methods do unlocked reads
1869 			 * of state data :(
1870 			 */
1871 			pfsync_qs[q].write(st, m->m_data + offset);
1872 			offset += pfsync_qs[q].len;
1873 			st->sync_state = PFSYNC_S_NONE;
1874 			pf_release_state(st);
1875 			count++;
1876 		}
1877 		TAILQ_INIT(&b->b_qs[q]);
1878 
1879 		subh->action = pfsync_qs[q].action;
1880 		subh->count = htons(count);
1881 		V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1882 	}
1883 
1884 	if (!TAILQ_EMPTY(&b->b_upd_req_list)) {
1885 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1886 		offset += sizeof(*subh);
1887 
1888 		count = 0;
1889 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1890 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1891 
1892 			bcopy(&ur->ur_msg, m->m_data + offset,
1893 			    sizeof(ur->ur_msg));
1894 			offset += sizeof(ur->ur_msg);
1895 			free(ur, M_PFSYNC);
1896 			count++;
1897 		}
1898 
1899 		subh->action = PFSYNC_ACT_UPD_REQ;
1900 		subh->count = htons(count);
1901 		V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1902 	}
1903 
1904 	/* has someone built a custom region for us to add? */
1905 	if (b->b_plus != NULL) {
1906 		bcopy(b->b_plus, m->m_data + offset, b->b_pluslen);
1907 		offset += b->b_pluslen;
1908 
1909 		b->b_plus = NULL;
1910 	}
1911 
1912 	subh = (struct pfsync_subheader *)(m->m_data + offset);
1913 	offset += sizeof(*subh);
1914 
1915 	subh->action = PFSYNC_ACT_EOF;
1916 	subh->count = htons(1);
1917 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1918 
1919 	/* we're done, let's put it on the wire */
1920 	if (ifp->if_bpf) {
1921 		m->m_data += aflen;
1922 		m->m_len = m->m_pkthdr.len = len - aflen;
1923 		BPF_MTAP(ifp, m);
1924 		m->m_data -= aflen;
1925 		m->m_len = m->m_pkthdr.len = len;
1926 	}
1927 
1928 	if (sc->sc_sync_if == NULL) {
1929 		b->b_len = PFSYNC_MINPKT;
1930 		m_freem(m);
1931 		return;
1932 	}
1933 
1934 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1935 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1936 	b->b_len = PFSYNC_MINPKT;
1937 
1938 	if (!_IF_QFULL(&b->b_snd))
1939 		_IF_ENQUEUE(&b->b_snd, m);
1940 	else {
1941 		m_freem(m);
1942 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1943 	}
1944 	if (schedswi)
1945 		swi_sched(V_pfsync_swi_cookie, 0);
1946 }
1947 
1948 static void
1949 pfsync_insert_state(struct pf_kstate *st)
1950 {
1951 	struct pfsync_softc *sc = V_pfsyncif;
1952 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1953 
1954 	if (st->state_flags & PFSTATE_NOSYNC)
1955 		return;
1956 
1957 	if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1958 	    st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1959 		st->state_flags |= PFSTATE_NOSYNC;
1960 		return;
1961 	}
1962 
1963 	KASSERT(st->sync_state == PFSYNC_S_NONE,
1964 		("%s: st->sync_state %u", __func__, st->sync_state));
1965 
1966 	PFSYNC_BUCKET_LOCK(b);
1967 	if (b->b_len == PFSYNC_MINPKT)
1968 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1969 
1970 	pfsync_q_ins(st, PFSYNC_S_INS, true);
1971 	PFSYNC_BUCKET_UNLOCK(b);
1972 
1973 	st->sync_updates = 0;
1974 }
1975 
1976 static int
1977 pfsync_defer(struct pf_kstate *st, struct mbuf *m)
1978 {
1979 	struct pfsync_softc *sc = V_pfsyncif;
1980 	struct pfsync_deferral *pd;
1981 	struct pfsync_bucket *b;
1982 
1983 	if (m->m_flags & (M_BCAST|M_MCAST))
1984 		return (0);
1985 
1986 	if (sc == NULL)
1987 		return (0);
1988 
1989 	b = pfsync_get_bucket(sc, st);
1990 
1991 	PFSYNC_LOCK(sc);
1992 
1993 	if (!(sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) ||
1994 	    !(sc->sc_flags & PFSYNCF_DEFER)) {
1995 		PFSYNC_UNLOCK(sc);
1996 		return (0);
1997 	}
1998 
1999 	PFSYNC_BUCKET_LOCK(b);
2000 	PFSYNC_UNLOCK(sc);
2001 
2002 	if (b->b_deferred >= 128)
2003 		pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0);
2004 
2005 	pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
2006 	if (pd == NULL) {
2007 		PFSYNC_BUCKET_UNLOCK(b);
2008 		return (0);
2009 	}
2010 	b->b_deferred++;
2011 
2012 	m->m_flags |= M_SKIP_FIREWALL;
2013 	st->state_flags |= PFSTATE_ACK;
2014 
2015 	pd->pd_sc = sc;
2016 	pd->pd_st = st;
2017 	pf_ref_state(st);
2018 	pd->pd_m = m;
2019 
2020 	TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry);
2021 	callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED);
2022 	callout_reset(&pd->pd_tmo, (V_pfsync_defer_timeout * hz) / 1000,
2023 	    pfsync_defer_tmo, pd);
2024 
2025 	pfsync_push(b);
2026 	PFSYNC_BUCKET_UNLOCK(b);
2027 
2028 	return (1);
2029 }
2030 
2031 static void
2032 pfsync_undefer(struct pfsync_deferral *pd, int drop)
2033 {
2034 	struct pfsync_softc *sc = pd->pd_sc;
2035 	struct mbuf *m = pd->pd_m;
2036 	struct pf_kstate *st = pd->pd_st;
2037 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2038 
2039 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2040 
2041 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
2042 	b->b_deferred--;
2043 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
2044 	free(pd, M_PFSYNC);
2045 	pf_release_state(st);
2046 
2047 	if (drop)
2048 		m_freem(m);
2049 	else {
2050 		_IF_ENQUEUE(&b->b_snd, m);
2051 		pfsync_push(b);
2052 	}
2053 }
2054 
2055 static void
2056 pfsync_defer_tmo(void *arg)
2057 {
2058 	struct epoch_tracker et;
2059 	struct pfsync_deferral *pd = arg;
2060 	struct pfsync_softc *sc = pd->pd_sc;
2061 	struct mbuf *m = pd->pd_m;
2062 	struct pf_kstate *st = pd->pd_st;
2063 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2064 
2065 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2066 
2067 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
2068 	b->b_deferred--;
2069 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
2070 	PFSYNC_BUCKET_UNLOCK(b);
2071 	free(pd, M_PFSYNC);
2072 
2073 	if (sc->sc_sync_if == NULL) {
2074 		pf_release_state(st);
2075 		m_freem(m);
2076 		return;
2077 	}
2078 
2079 	NET_EPOCH_ENTER(et);
2080 	CURVNET_SET(sc->sc_sync_if->if_vnet);
2081 
2082 	pfsync_tx(sc, m);
2083 
2084 	pf_release_state(st);
2085 
2086 	CURVNET_RESTORE();
2087 	NET_EPOCH_EXIT(et);
2088 }
2089 
2090 static void
2091 pfsync_undefer_state_locked(struct pf_kstate *st, int drop)
2092 {
2093 	struct pfsync_softc *sc = V_pfsyncif;
2094 	struct pfsync_deferral *pd;
2095 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2096 
2097 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2098 
2099 	TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) {
2100 		 if (pd->pd_st == st) {
2101 			if (callout_stop(&pd->pd_tmo) > 0)
2102 				pfsync_undefer(pd, drop);
2103 
2104 			return;
2105 		}
2106 	}
2107 
2108 	panic("%s: unable to find deferred state", __func__);
2109 }
2110 
2111 static void
2112 pfsync_undefer_state(struct pf_kstate *st, int drop)
2113 {
2114 	struct pfsync_softc *sc = V_pfsyncif;
2115 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2116 
2117 	PFSYNC_BUCKET_LOCK(b);
2118 	pfsync_undefer_state_locked(st, drop);
2119 	PFSYNC_BUCKET_UNLOCK(b);
2120 }
2121 
2122 static struct pfsync_bucket*
2123 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_kstate *st)
2124 {
2125 	int c = PF_IDHASH(st) % pfsync_buckets;
2126 	return &sc->sc_buckets[c];
2127 }
2128 
2129 static void
2130 pfsync_update_state(struct pf_kstate *st)
2131 {
2132 	struct pfsync_softc *sc = V_pfsyncif;
2133 	bool sync = false, ref = true;
2134 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2135 
2136 	PF_STATE_LOCK_ASSERT(st);
2137 	PFSYNC_BUCKET_LOCK(b);
2138 
2139 	if (st->state_flags & PFSTATE_ACK)
2140 		pfsync_undefer_state_locked(st, 0);
2141 	if (st->state_flags & PFSTATE_NOSYNC) {
2142 		if (st->sync_state != PFSYNC_S_NONE)
2143 			pfsync_q_del(st, true, b);
2144 		PFSYNC_BUCKET_UNLOCK(b);
2145 		return;
2146 	}
2147 
2148 	if (b->b_len == PFSYNC_MINPKT)
2149 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2150 
2151 	switch (st->sync_state) {
2152 	case PFSYNC_S_UPD_C:
2153 	case PFSYNC_S_UPD:
2154 	case PFSYNC_S_INS:
2155 		/* we're already handling it */
2156 
2157 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
2158 			st->sync_updates++;
2159 			if (st->sync_updates >= sc->sc_maxupdates)
2160 				sync = true;
2161 		}
2162 		break;
2163 
2164 	case PFSYNC_S_IACK:
2165 		pfsync_q_del(st, false, b);
2166 		ref = false;
2167 		/* FALLTHROUGH */
2168 
2169 	case PFSYNC_S_NONE:
2170 		pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
2171 		st->sync_updates = 0;
2172 		break;
2173 
2174 	default:
2175 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2176 	}
2177 
2178 	if (sync || (time_uptime - st->pfsync_time) < 2)
2179 		pfsync_push(b);
2180 
2181 	PFSYNC_BUCKET_UNLOCK(b);
2182 }
2183 
2184 static void
2185 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
2186 {
2187 	struct pfsync_softc *sc = V_pfsyncif;
2188 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2189 	struct pfsync_upd_req_item *item;
2190 	size_t nlen = sizeof(struct pfsync_upd_req);
2191 
2192 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2193 
2194 	/*
2195 	 * This code does a bit to prevent multiple update requests for the
2196 	 * same state being generated. It searches current subheader queue,
2197 	 * but it doesn't lookup into queue of already packed datagrams.
2198 	 */
2199 	TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry)
2200 		if (item->ur_msg.id == id &&
2201 		    item->ur_msg.creatorid == creatorid)
2202 			return;
2203 
2204 	item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
2205 	if (item == NULL)
2206 		return; /* XXX stats */
2207 
2208 	item->ur_msg.id = id;
2209 	item->ur_msg.creatorid = creatorid;
2210 
2211 	if (TAILQ_EMPTY(&b->b_upd_req_list))
2212 		nlen += sizeof(struct pfsync_subheader);
2213 
2214 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2215 		pfsync_sendout(0, 0);
2216 
2217 		nlen = sizeof(struct pfsync_subheader) +
2218 		    sizeof(struct pfsync_upd_req);
2219 	}
2220 
2221 	TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry);
2222 	b->b_len += nlen;
2223 
2224 	pfsync_push(b);
2225 }
2226 
2227 static bool
2228 pfsync_update_state_req(struct pf_kstate *st)
2229 {
2230 	struct pfsync_softc *sc = V_pfsyncif;
2231 	bool ref = true, full = false;
2232 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2233 
2234 	PF_STATE_LOCK_ASSERT(st);
2235 	PFSYNC_BUCKET_LOCK(b);
2236 
2237 	if (st->state_flags & PFSTATE_NOSYNC) {
2238 		if (st->sync_state != PFSYNC_S_NONE)
2239 			pfsync_q_del(st, true, b);
2240 		PFSYNC_BUCKET_UNLOCK(b);
2241 		return (full);
2242 	}
2243 
2244 	switch (st->sync_state) {
2245 	case PFSYNC_S_UPD_C:
2246 	case PFSYNC_S_IACK:
2247 		pfsync_q_del(st, false, b);
2248 		ref = false;
2249 		/* FALLTHROUGH */
2250 
2251 	case PFSYNC_S_NONE:
2252 		pfsync_q_ins(st, PFSYNC_S_UPD, ref);
2253 		pfsync_push(b);
2254 		break;
2255 
2256 	case PFSYNC_S_INS:
2257 	case PFSYNC_S_UPD:
2258 	case PFSYNC_S_DEL_C:
2259 		/* we're already handling it */
2260 		break;
2261 
2262 	default:
2263 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2264 	}
2265 
2266 	if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(union pfsync_state_union))
2267 		full = true;
2268 
2269 	PFSYNC_BUCKET_UNLOCK(b);
2270 
2271 	return (full);
2272 }
2273 
2274 static void
2275 pfsync_delete_state(struct pf_kstate *st)
2276 {
2277 	struct pfsync_softc *sc = V_pfsyncif;
2278 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2279 	bool ref = true;
2280 
2281 	PFSYNC_BUCKET_LOCK(b);
2282 	if (st->state_flags & PFSTATE_ACK)
2283 		pfsync_undefer_state_locked(st, 1);
2284 	if (st->state_flags & PFSTATE_NOSYNC) {
2285 		if (st->sync_state != PFSYNC_S_NONE)
2286 			pfsync_q_del(st, true, b);
2287 		PFSYNC_BUCKET_UNLOCK(b);
2288 		return;
2289 	}
2290 
2291 	if (b->b_len == PFSYNC_MINPKT)
2292 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2293 
2294 	switch (st->sync_state) {
2295 	case PFSYNC_S_INS:
2296 		/* We never got to tell the world so just forget about it. */
2297 		pfsync_q_del(st, true, b);
2298 		break;
2299 
2300 	case PFSYNC_S_UPD_C:
2301 	case PFSYNC_S_UPD:
2302 	case PFSYNC_S_IACK:
2303 		pfsync_q_del(st, false, b);
2304 		ref = false;
2305 		/* FALLTHROUGH */
2306 
2307 	case PFSYNC_S_NONE:
2308 		pfsync_q_ins(st, PFSYNC_S_DEL_C, ref);
2309 		break;
2310 
2311 	default:
2312 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2313 	}
2314 
2315 	PFSYNC_BUCKET_UNLOCK(b);
2316 }
2317 
2318 static void
2319 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
2320 {
2321 	struct {
2322 		struct pfsync_subheader subh;
2323 		struct pfsync_clr clr;
2324 	} __packed r;
2325 
2326 	bzero(&r, sizeof(r));
2327 
2328 	r.subh.action = PFSYNC_ACT_CLR;
2329 	r.subh.count = htons(1);
2330 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
2331 
2332 	strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
2333 	r.clr.creatorid = creatorid;
2334 
2335 	pfsync_send_plus(&r, sizeof(r));
2336 }
2337 
2338 static enum pfsync_q_id
2339 pfsync_sstate_to_qid(u_int8_t sync_state)
2340 {
2341 	struct pfsync_softc *sc = V_pfsyncif;
2342 
2343 	switch (sync_state) {
2344 		case PFSYNC_S_INS:
2345 			switch (sc->sc_version) {
2346 				case PFSYNC_MSG_VERSION_1301:
2347 					return PFSYNC_Q_INS_1301;
2348 				case PFSYNC_MSG_VERSION_1400:
2349 					return PFSYNC_Q_INS_1400;
2350 			}
2351 			break;
2352 		case PFSYNC_S_IACK:
2353 			return PFSYNC_Q_IACK;
2354 		case PFSYNC_S_UPD:
2355 			switch (sc->sc_version) {
2356 				case PFSYNC_MSG_VERSION_1301:
2357 					return PFSYNC_Q_UPD_1301;
2358 				case PFSYNC_MSG_VERSION_1400:
2359 					return PFSYNC_Q_UPD_1400;
2360 			}
2361 			break;
2362 		case PFSYNC_S_UPD_C:
2363 			return PFSYNC_Q_UPD_C;
2364 		case PFSYNC_S_DEL_C:
2365 			return PFSYNC_Q_DEL_C;
2366 		default:
2367 			panic("%s: Unsupported st->sync_state 0x%02x",
2368 			__func__, sync_state);
2369 	}
2370 
2371 	panic("%s: Unsupported pfsync_msg_version %d",
2372 	    __func__, sc->sc_version);
2373 }
2374 
2375 static void
2376 pfsync_q_ins(struct pf_kstate *st, int sync_state, bool ref)
2377 {
2378 	enum pfsync_q_id q = pfsync_sstate_to_qid(sync_state);
2379 	struct pfsync_softc *sc = V_pfsyncif;
2380 	size_t nlen = pfsync_qs[q].len;
2381 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2382 
2383 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2384 
2385 	KASSERT(st->sync_state == PFSYNC_S_NONE,
2386 		("%s: st->sync_state %u", __func__, st->sync_state));
2387 	KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2388 	    b->b_len));
2389 
2390 	if (TAILQ_EMPTY(&b->b_qs[q]))
2391 		nlen += sizeof(struct pfsync_subheader);
2392 
2393 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2394 		pfsync_sendout(1, b->b_id);
2395 
2396 		nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2397 	}
2398 
2399 	b->b_len += nlen;
2400 	TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list);
2401 	st->sync_state = pfsync_qid_sstate[q];
2402 	if (ref)
2403 		pf_ref_state(st);
2404 }
2405 
2406 static void
2407 pfsync_q_del(struct pf_kstate *st, bool unref, struct pfsync_bucket *b)
2408 {
2409 	enum pfsync_q_id q;
2410 
2411 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2412 	KASSERT(st->sync_state != PFSYNC_S_NONE,
2413 		("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2414 
2415 	q =  pfsync_sstate_to_qid(st->sync_state);
2416 	b->b_len -= pfsync_qs[q].len;
2417 	TAILQ_REMOVE(&b->b_qs[q], st, sync_list);
2418 	st->sync_state = PFSYNC_S_NONE;
2419 	if (unref)
2420 		pf_release_state(st);
2421 
2422 	if (TAILQ_EMPTY(&b->b_qs[q]))
2423 		b->b_len -= sizeof(struct pfsync_subheader);
2424 }
2425 
2426 static void
2427 pfsync_bulk_start(void)
2428 {
2429 	struct pfsync_softc *sc = V_pfsyncif;
2430 
2431 	if (V_pf_status.debug >= PF_DEBUG_MISC)
2432 		printf("pfsync: received bulk update request\n");
2433 
2434 	PFSYNC_BLOCK(sc);
2435 
2436 	sc->sc_ureq_received = time_uptime;
2437 	sc->sc_bulk_hashid = 0;
2438 	sc->sc_bulk_stateid = 0;
2439 	pfsync_bulk_status(PFSYNC_BUS_START);
2440 	callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2441 	PFSYNC_BUNLOCK(sc);
2442 }
2443 
2444 static void
2445 pfsync_bulk_update(void *arg)
2446 {
2447 	struct pfsync_softc *sc = arg;
2448 	struct pf_kstate *s;
2449 	int i;
2450 
2451 	PFSYNC_BLOCK_ASSERT(sc);
2452 	CURVNET_SET(sc->sc_ifp->if_vnet);
2453 
2454 	/*
2455 	 * Start with last state from previous invocation.
2456 	 * It may had gone, in this case start from the
2457 	 * hash slot.
2458 	 */
2459 	s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2460 
2461 	if (s != NULL)
2462 		i = PF_IDHASH(s);
2463 	else
2464 		i = sc->sc_bulk_hashid;
2465 
2466 	for (; i <= pf_hashmask; i++) {
2467 		struct pf_idhash *ih = &V_pf_idhash[i];
2468 
2469 		if (s != NULL)
2470 			PF_HASHROW_ASSERT(ih);
2471 		else {
2472 			PF_HASHROW_LOCK(ih);
2473 			s = LIST_FIRST(&ih->states);
2474 		}
2475 
2476 		for (; s; s = LIST_NEXT(s, entry)) {
2477 			if (s->sync_state == PFSYNC_S_NONE &&
2478 			    s->timeout < PFTM_MAX &&
2479 			    s->pfsync_time <= sc->sc_ureq_received) {
2480 				if (pfsync_update_state_req(s)) {
2481 					/* We've filled a packet. */
2482 					sc->sc_bulk_hashid = i;
2483 					sc->sc_bulk_stateid = s->id;
2484 					sc->sc_bulk_creatorid = s->creatorid;
2485 					PF_HASHROW_UNLOCK(ih);
2486 					callout_reset(&sc->sc_bulk_tmo, 1,
2487 					    pfsync_bulk_update, sc);
2488 					goto full;
2489 				}
2490 			}
2491 		}
2492 		PF_HASHROW_UNLOCK(ih);
2493 	}
2494 
2495 	/* We're done. */
2496 	pfsync_bulk_status(PFSYNC_BUS_END);
2497 full:
2498 	CURVNET_RESTORE();
2499 }
2500 
2501 static void
2502 pfsync_bulk_status(u_int8_t status)
2503 {
2504 	struct {
2505 		struct pfsync_subheader subh;
2506 		struct pfsync_bus bus;
2507 	} __packed r;
2508 
2509 	struct pfsync_softc *sc = V_pfsyncif;
2510 
2511 	bzero(&r, sizeof(r));
2512 
2513 	r.subh.action = PFSYNC_ACT_BUS;
2514 	r.subh.count = htons(1);
2515 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2516 
2517 	r.bus.creatorid = V_pf_status.hostid;
2518 	r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2519 	r.bus.status = status;
2520 
2521 	pfsync_send_plus(&r, sizeof(r));
2522 }
2523 
2524 static void
2525 pfsync_bulk_fail(void *arg)
2526 {
2527 	struct pfsync_softc *sc = arg;
2528 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2529 
2530 	CURVNET_SET(sc->sc_ifp->if_vnet);
2531 
2532 	PFSYNC_BLOCK_ASSERT(sc);
2533 
2534 	if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2535 		/* Try again */
2536 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2537 		    pfsync_bulk_fail, V_pfsyncif);
2538 		PFSYNC_BUCKET_LOCK(b);
2539 		pfsync_request_update(0, 0);
2540 		PFSYNC_BUCKET_UNLOCK(b);
2541 	} else {
2542 		/* Pretend like the transfer was ok. */
2543 		sc->sc_ureq_sent = 0;
2544 		sc->sc_bulk_tries = 0;
2545 		PFSYNC_LOCK(sc);
2546 		if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2547 			(*carp_demote_adj_p)(-V_pfsync_carp_adj,
2548 			    "pfsync bulk fail");
2549 		sc->sc_flags |= PFSYNCF_OK;
2550 		PFSYNC_UNLOCK(sc);
2551 		if (V_pf_status.debug >= PF_DEBUG_MISC)
2552 			printf("pfsync: failed to receive bulk update\n");
2553 	}
2554 
2555 	CURVNET_RESTORE();
2556 }
2557 
2558 static void
2559 pfsync_send_plus(void *plus, size_t pluslen)
2560 {
2561 	struct pfsync_softc *sc = V_pfsyncif;
2562 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2563 
2564 	PFSYNC_BUCKET_LOCK(b);
2565 
2566 	if (b->b_len + pluslen > sc->sc_ifp->if_mtu)
2567 		pfsync_sendout(1, b->b_id);
2568 
2569 	b->b_plus = plus;
2570 	b->b_len += (b->b_pluslen = pluslen);
2571 
2572 	pfsync_sendout(1, b->b_id);
2573 	PFSYNC_BUCKET_UNLOCK(b);
2574 }
2575 
2576 static void
2577 pfsync_timeout(void *arg)
2578 {
2579 	struct pfsync_bucket *b = arg;
2580 
2581 	CURVNET_SET(b->b_sc->sc_ifp->if_vnet);
2582 	PFSYNC_BUCKET_LOCK(b);
2583 	pfsync_push(b);
2584 	PFSYNC_BUCKET_UNLOCK(b);
2585 	CURVNET_RESTORE();
2586 }
2587 
2588 static void
2589 pfsync_push(struct pfsync_bucket *b)
2590 {
2591 
2592 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2593 
2594 	b->b_flags |= PFSYNCF_BUCKET_PUSH;
2595 	swi_sched(V_pfsync_swi_cookie, 0);
2596 }
2597 
2598 static void
2599 pfsync_push_all(struct pfsync_softc *sc)
2600 {
2601 	int c;
2602 	struct pfsync_bucket *b;
2603 
2604 	for (c = 0; c < pfsync_buckets; c++) {
2605 		b = &sc->sc_buckets[c];
2606 
2607 		PFSYNC_BUCKET_LOCK(b);
2608 		pfsync_push(b);
2609 		PFSYNC_BUCKET_UNLOCK(b);
2610 	}
2611 }
2612 
2613 static void
2614 pfsync_tx(struct pfsync_softc *sc, struct mbuf *m)
2615 {
2616 	struct ip *ip;
2617 	int af, error = 0;
2618 
2619 	ip = mtod(m, struct ip *);
2620 	MPASS(ip->ip_v == IPVERSION || ip->ip_v == (IPV6_VERSION >> 4));
2621 
2622 	af = ip->ip_v == IPVERSION ? AF_INET : AF_INET6;
2623 
2624 	/*
2625 	 * We distinguish between a deferral packet and our
2626 	 * own pfsync packet based on M_SKIP_FIREWALL
2627 	 * flag. This is XXX.
2628 	 */
2629 	switch (af) {
2630 #ifdef INET
2631 	case AF_INET:
2632 		if (m->m_flags & M_SKIP_FIREWALL) {
2633 			error = ip_output(m, NULL, NULL, 0,
2634 			    NULL, NULL);
2635 		} else {
2636 			error = ip_output(m, NULL, NULL,
2637 			    IP_RAWOUTPUT, &sc->sc_imo, NULL);
2638 		}
2639 		break;
2640 #endif
2641 #ifdef INET6
2642 	case AF_INET6:
2643 		if (m->m_flags & M_SKIP_FIREWALL) {
2644 			error = ip6_output(m, NULL, NULL, 0,
2645 			    NULL, NULL, NULL);
2646 		} else {
2647 			error = ip6_output(m, NULL, NULL, 0,
2648 				&sc->sc_im6o, NULL, NULL);
2649 		}
2650 		break;
2651 #endif
2652 	}
2653 
2654 	if (error == 0)
2655 		V_pfsyncstats.pfsyncs_opackets++;
2656 	else
2657 		V_pfsyncstats.pfsyncs_oerrors++;
2658 
2659 }
2660 
2661 static void
2662 pfsyncintr(void *arg)
2663 {
2664 	struct epoch_tracker et;
2665 	struct pfsync_softc *sc = arg;
2666 	struct pfsync_bucket *b;
2667 	struct mbuf *m, *n;
2668 	int c;
2669 
2670 	NET_EPOCH_ENTER(et);
2671 	CURVNET_SET(sc->sc_ifp->if_vnet);
2672 
2673 	for (c = 0; c < pfsync_buckets; c++) {
2674 		b = &sc->sc_buckets[c];
2675 
2676 		PFSYNC_BUCKET_LOCK(b);
2677 		if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) {
2678 			pfsync_sendout(0, b->b_id);
2679 			b->b_flags &= ~PFSYNCF_BUCKET_PUSH;
2680 		}
2681 		_IF_DEQUEUE_ALL(&b->b_snd, m);
2682 		PFSYNC_BUCKET_UNLOCK(b);
2683 
2684 		for (; m != NULL; m = n) {
2685 			n = m->m_nextpkt;
2686 			m->m_nextpkt = NULL;
2687 
2688 			pfsync_tx(sc, m);
2689 		}
2690 	}
2691 	CURVNET_RESTORE();
2692 	NET_EPOCH_EXIT(et);
2693 }
2694 
2695 static int
2696 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp,
2697     struct in_mfilter* imf, struct in6_mfilter* im6f)
2698 {
2699 #ifdef  INET
2700 	struct ip_moptions *imo = &sc->sc_imo;
2701 #endif
2702 #ifdef INET6
2703 	struct ip6_moptions *im6o = &sc->sc_im6o;
2704 	struct sockaddr_in6 *syncpeer_sa6 = NULL;
2705 #endif
2706 
2707 	if (!(ifp->if_flags & IFF_MULTICAST))
2708 		return (EADDRNOTAVAIL);
2709 
2710 	switch (sc->sc_sync_peer.ss_family) {
2711 #ifdef INET
2712 	case AF_INET:
2713 	{
2714 		int error;
2715 
2716 		ip_mfilter_init(&imo->imo_head);
2717 		imo->imo_multicast_vif = -1;
2718 		if ((error = in_joingroup(ifp,
2719 		    &((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr, NULL,
2720 		    &imf->imf_inm)) != 0)
2721 			return (error);
2722 
2723 		ip_mfilter_insert(&imo->imo_head, imf);
2724 		imo->imo_multicast_ifp = ifp;
2725 		imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2726 		imo->imo_multicast_loop = 0;
2727 		break;
2728 	}
2729 #endif
2730 #ifdef INET6
2731 	case AF_INET6:
2732 	{
2733 		int error;
2734 
2735 		syncpeer_sa6 = (struct sockaddr_in6 *)&sc->sc_sync_peer;
2736 		if ((error = in6_setscope(&syncpeer_sa6->sin6_addr, ifp, NULL)))
2737 			return (error);
2738 
2739 		ip6_mfilter_init(&im6o->im6o_head);
2740 		if ((error = in6_joingroup(ifp, &syncpeer_sa6->sin6_addr, NULL,
2741 		    &(im6f->im6f_in6m), 0)) != 0)
2742 			return (error);
2743 
2744 		ip6_mfilter_insert(&im6o->im6o_head, im6f);
2745 		im6o->im6o_multicast_ifp = ifp;
2746 		im6o->im6o_multicast_hlim = PFSYNC_DFLTTL;
2747 		im6o->im6o_multicast_loop = 0;
2748 		break;
2749 	}
2750 #endif
2751 	}
2752 
2753 	return (0);
2754 }
2755 
2756 static void
2757 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2758 {
2759 #ifdef INET
2760 	struct ip_moptions *imo = &sc->sc_imo;
2761 	struct in_mfilter *imf;
2762 
2763 	while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
2764 		ip_mfilter_remove(&imo->imo_head, imf);
2765 		in_leavegroup(imf->imf_inm, NULL);
2766 		ip_mfilter_free(imf);
2767 	}
2768 	imo->imo_multicast_ifp = NULL;
2769 #endif
2770 
2771 #ifdef INET6
2772 	struct ip6_moptions *im6o = &sc->sc_im6o;
2773 	struct in6_mfilter *im6f;
2774 
2775 	while ((im6f = ip6_mfilter_first(&im6o->im6o_head)) != NULL) {
2776 		ip6_mfilter_remove(&im6o->im6o_head, im6f);
2777 		in6_leavegroup(im6f->im6f_in6m, NULL);
2778 		ip6_mfilter_free(im6f);
2779 	}
2780 	im6o->im6o_multicast_ifp = NULL;
2781 #endif
2782 }
2783 
2784 void
2785 pfsync_detach_ifnet(struct ifnet *ifp)
2786 {
2787 	struct pfsync_softc *sc = V_pfsyncif;
2788 
2789 	if (sc == NULL)
2790 		return;
2791 
2792 	PFSYNC_LOCK(sc);
2793 
2794 	if (sc->sc_sync_if == ifp) {
2795 		/* We don't need mutlicast cleanup here, because the interface
2796 		 * is going away. We do need to ensure we don't try to do
2797 		 * cleanup later.
2798 		 */
2799 		ip_mfilter_init(&sc->sc_imo.imo_head);
2800 		sc->sc_imo.imo_multicast_ifp = NULL;
2801 		sc->sc_im6o.im6o_multicast_ifp = NULL;
2802 		sc->sc_sync_if = NULL;
2803 	}
2804 
2805 	PFSYNC_UNLOCK(sc);
2806 }
2807 
2808 static int
2809 pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *pfsyncr, struct pfsync_kstatus *status)
2810 {
2811 	struct sockaddr_storage sa;
2812 	status->maxupdates = pfsyncr->pfsyncr_maxupdates;
2813 	status->flags = pfsyncr->pfsyncr_defer;
2814 
2815 	strlcpy(status->syncdev, pfsyncr->pfsyncr_syncdev, IFNAMSIZ);
2816 
2817 	memset(&sa, 0, sizeof(sa));
2818 	if (pfsyncr->pfsyncr_syncpeer.s_addr != 0) {
2819 		struct sockaddr_in *in = (struct sockaddr_in *)&sa;
2820 		in->sin_family = AF_INET;
2821 		in->sin_len = sizeof(*in);
2822 		in->sin_addr.s_addr = pfsyncr->pfsyncr_syncpeer.s_addr;
2823 	}
2824 	status->syncpeer = sa;
2825 
2826 	return 0;
2827 }
2828 
2829 static int
2830 pfsync_kstatus_to_softc(struct pfsync_kstatus *status, struct pfsync_softc *sc)
2831 {
2832 	struct ifnet *sifp;
2833 	struct in_mfilter *imf = NULL;
2834 	struct in6_mfilter *im6f = NULL;
2835 	int error;
2836 	int c;
2837 
2838 	if ((status->maxupdates < 0) || (status->maxupdates > 255))
2839 		return (EINVAL);
2840 
2841 	if (status->syncdev[0] == '\0')
2842 		sifp = NULL;
2843 	else if ((sifp = ifunit_ref(status->syncdev)) == NULL)
2844 		return (EINVAL);
2845 
2846 	switch (status->syncpeer.ss_family) {
2847 #ifdef INET
2848 	case AF_UNSPEC:
2849 	case AF_INET: {
2850 		struct sockaddr_in *status_sin;
2851 		status_sin = (struct sockaddr_in *)&(status->syncpeer);
2852 		if (sifp != NULL) {
2853 			if (status_sin->sin_addr.s_addr == 0 ||
2854 			    status_sin->sin_addr.s_addr ==
2855 			    htonl(INADDR_PFSYNC_GROUP)) {
2856 				status_sin->sin_family = AF_INET;
2857 				status_sin->sin_len = sizeof(*status_sin);
2858 				status_sin->sin_addr.s_addr =
2859 				    htonl(INADDR_PFSYNC_GROUP);
2860 			}
2861 
2862 			if (IN_MULTICAST(ntohl(status_sin->sin_addr.s_addr))) {
2863 				imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
2864 			}
2865 		}
2866 		break;
2867 	}
2868 #endif
2869 #ifdef INET6
2870 	case AF_INET6: {
2871 		struct sockaddr_in6 *status_sin6;
2872 		status_sin6 = (struct sockaddr_in6*)&(status->syncpeer);
2873 		if (sifp != NULL) {
2874 			if (IN6_IS_ADDR_UNSPECIFIED(&status_sin6->sin6_addr) ||
2875 			    IN6_ARE_ADDR_EQUAL(&status_sin6->sin6_addr,
2876 				&in6addr_linklocal_pfsync_group)) {
2877 				status_sin6->sin6_family = AF_INET6;
2878 				status_sin6->sin6_len = sizeof(*status_sin6);
2879 				status_sin6->sin6_addr =
2880 				    in6addr_linklocal_pfsync_group;
2881 			}
2882 
2883 			if (IN6_IS_ADDR_MULTICAST(&status_sin6->sin6_addr)) {
2884 				im6f = ip6_mfilter_alloc(M_WAITOK, 0, 0);
2885 			}
2886 		}
2887 		break;
2888 	}
2889 #endif
2890 	}
2891 
2892 	PFSYNC_LOCK(sc);
2893 
2894 	switch (status->version) {
2895 		case PFSYNC_MSG_VERSION_UNSPECIFIED:
2896 			sc->sc_version = PFSYNC_MSG_VERSION_DEFAULT;
2897 			break;
2898 		case PFSYNC_MSG_VERSION_1301:
2899 		case PFSYNC_MSG_VERSION_1400:
2900 			sc->sc_version = status->version;
2901 			break;
2902 		default:
2903 			PFSYNC_UNLOCK(sc);
2904 			return (EINVAL);
2905 	}
2906 
2907 	switch (status->syncpeer.ss_family) {
2908 	case AF_INET: {
2909 		struct sockaddr_in *status_sin = (struct sockaddr_in *)&(status->syncpeer);
2910 		struct sockaddr_in *sc_sin = (struct sockaddr_in *)&sc->sc_sync_peer;
2911 		sc_sin->sin_family = AF_INET;
2912 		sc_sin->sin_len = sizeof(*sc_sin);
2913 		if (status_sin->sin_addr.s_addr == 0) {
2914 			sc_sin->sin_addr.s_addr = htonl(INADDR_PFSYNC_GROUP);
2915 		} else {
2916 			sc_sin->sin_addr.s_addr = status_sin->sin_addr.s_addr;
2917 		}
2918 		break;
2919 	}
2920 	case AF_INET6: {
2921 		struct sockaddr_in6 *status_sin = (struct sockaddr_in6 *)&(status->syncpeer);
2922 		struct sockaddr_in6 *sc_sin = (struct sockaddr_in6 *)&sc->sc_sync_peer;
2923 		sc_sin->sin6_family = AF_INET6;
2924 		sc_sin->sin6_len = sizeof(*sc_sin);
2925 		if(IN6_IS_ADDR_UNSPECIFIED(&status_sin->sin6_addr)) {
2926 			sc_sin->sin6_addr = in6addr_linklocal_pfsync_group;
2927 		} else {
2928 			sc_sin->sin6_addr = status_sin->sin6_addr;
2929 		}
2930 		break;
2931 	}
2932 	}
2933 
2934 	sc->sc_maxupdates = status->maxupdates;
2935 	if (status->flags & PFSYNCF_DEFER) {
2936 		sc->sc_flags |= PFSYNCF_DEFER;
2937 		V_pfsync_defer_ptr = pfsync_defer;
2938 	} else {
2939 		sc->sc_flags &= ~PFSYNCF_DEFER;
2940 		V_pfsync_defer_ptr = NULL;
2941 	}
2942 
2943 	if (sifp == NULL) {
2944 		if (sc->sc_sync_if)
2945 			if_rele(sc->sc_sync_if);
2946 		sc->sc_sync_if = NULL;
2947 		pfsync_multicast_cleanup(sc);
2948 		PFSYNC_UNLOCK(sc);
2949 		return (0);
2950 	}
2951 
2952 	for (c = 0; c < pfsync_buckets; c++) {
2953 		PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
2954 		if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT &&
2955 		    (sifp->if_mtu < sc->sc_ifp->if_mtu ||
2956 			(sc->sc_sync_if != NULL &&
2957 			    sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
2958 			sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
2959 			pfsync_sendout(1, c);
2960 		PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
2961 	}
2962 
2963 	pfsync_multicast_cleanup(sc);
2964 
2965 	if (((sc->sc_sync_peer.ss_family == AF_INET) &&
2966 	    IN_MULTICAST(ntohl(((struct sockaddr_in *)
2967 	        &sc->sc_sync_peer)->sin_addr.s_addr))) ||
2968 	    ((sc->sc_sync_peer.ss_family == AF_INET6) &&
2969 	    IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6*)
2970 	        &sc->sc_sync_peer)->sin6_addr))) {
2971 		error = pfsync_multicast_setup(sc, sifp, imf, im6f);
2972 		if (error) {
2973 			if_rele(sifp);
2974 			PFSYNC_UNLOCK(sc);
2975 #ifdef INET
2976 			if (imf != NULL)
2977 				ip_mfilter_free(imf);
2978 #endif
2979 #ifdef INET6
2980 			if (im6f != NULL)
2981 				ip6_mfilter_free(im6f);
2982 #endif
2983 			return (error);
2984 		}
2985 	}
2986 	if (sc->sc_sync_if)
2987 		if_rele(sc->sc_sync_if);
2988 	sc->sc_sync_if = sifp;
2989 
2990 	switch (sc->sc_sync_peer.ss_family) {
2991 #ifdef INET
2992 	case AF_INET: {
2993 		struct ip *ip;
2994 		ip = &sc->sc_template.ipv4;
2995 		bzero(ip, sizeof(*ip));
2996 		ip->ip_v = IPVERSION;
2997 		ip->ip_hl = sizeof(sc->sc_template.ipv4) >> 2;
2998 		ip->ip_tos = IPTOS_LOWDELAY;
2999 		/* len and id are set later. */
3000 		ip->ip_off = htons(IP_DF);
3001 		ip->ip_ttl = PFSYNC_DFLTTL;
3002 		ip->ip_p = IPPROTO_PFSYNC;
3003 		ip->ip_src.s_addr = INADDR_ANY;
3004 		ip->ip_dst = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
3005 		break;
3006 	}
3007 #endif
3008 #ifdef INET6
3009 	case AF_INET6: {
3010 		struct ip6_hdr *ip6;
3011 		ip6 = &sc->sc_template.ipv6;
3012 		bzero(ip6, sizeof(*ip6));
3013 		ip6->ip6_vfc = IPV6_VERSION;
3014 		ip6->ip6_hlim = PFSYNC_DFLTTL;
3015 		ip6->ip6_nxt = IPPROTO_PFSYNC;
3016 		ip6->ip6_dst = ((struct sockaddr_in6 *)&sc->sc_sync_peer)->sin6_addr;
3017 
3018 		struct epoch_tracker et;
3019 		NET_EPOCH_ENTER(et);
3020 		in6_selectsrc_addr(if_getfib(sc->sc_sync_if), &ip6->ip6_dst, 0,
3021 		    sc->sc_sync_if, &ip6->ip6_src, NULL);
3022 		NET_EPOCH_EXIT(et);
3023 		break;
3024 	}
3025 #endif
3026 	}
3027 
3028 	/* Request a full state table update. */
3029 	if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
3030 		(*carp_demote_adj_p)(V_pfsync_carp_adj,
3031 		    "pfsync bulk start");
3032 	sc->sc_flags &= ~PFSYNCF_OK;
3033 	if (V_pf_status.debug >= PF_DEBUG_MISC)
3034 		printf("pfsync: requesting bulk update\n");
3035 	PFSYNC_UNLOCK(sc);
3036 	PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
3037 	pfsync_request_update(0, 0);
3038 	PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
3039 	PFSYNC_BLOCK(sc);
3040 	sc->sc_ureq_sent = time_uptime;
3041 	callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, sc);
3042 	PFSYNC_BUNLOCK(sc);
3043 	return (0);
3044 }
3045 
3046 static void
3047 pfsync_pointers_init(void)
3048 {
3049 
3050 	PF_RULES_WLOCK();
3051 	V_pfsync_state_import_ptr = pfsync_state_import;
3052 	V_pfsync_insert_state_ptr = pfsync_insert_state;
3053 	V_pfsync_update_state_ptr = pfsync_update_state;
3054 	V_pfsync_delete_state_ptr = pfsync_delete_state;
3055 	V_pfsync_clear_states_ptr = pfsync_clear_states;
3056 	V_pfsync_defer_ptr = pfsync_defer;
3057 	PF_RULES_WUNLOCK();
3058 }
3059 
3060 static void
3061 pfsync_pointers_uninit(void)
3062 {
3063 
3064 	PF_RULES_WLOCK();
3065 	V_pfsync_state_import_ptr = NULL;
3066 	V_pfsync_insert_state_ptr = NULL;
3067 	V_pfsync_update_state_ptr = NULL;
3068 	V_pfsync_delete_state_ptr = NULL;
3069 	V_pfsync_clear_states_ptr = NULL;
3070 	V_pfsync_defer_ptr = NULL;
3071 	PF_RULES_WUNLOCK();
3072 }
3073 
3074 static void
3075 vnet_pfsync_init(const void *unused __unused)
3076 {
3077 	int error;
3078 
3079 	V_pfsync_cloner = if_clone_simple(pfsyncname,
3080 	    pfsync_clone_create, pfsync_clone_destroy, 1);
3081 	error = swi_add(&V_pfsync_swi_ie, pfsyncname, pfsyncintr, V_pfsyncif,
3082 	    SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
3083 	if (error) {
3084 		if_clone_detach(V_pfsync_cloner);
3085 		log(LOG_INFO, "swi_add() failed in %s\n", __func__);
3086 	}
3087 
3088 	pfsync_pointers_init();
3089 }
3090 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
3091     vnet_pfsync_init, NULL);
3092 
3093 static void
3094 vnet_pfsync_uninit(const void *unused __unused)
3095 {
3096 	int ret __diagused;
3097 
3098 	pfsync_pointers_uninit();
3099 
3100 	if_clone_detach(V_pfsync_cloner);
3101 	ret = swi_remove(V_pfsync_swi_cookie);
3102 	MPASS(ret == 0);
3103 	ret = intr_event_destroy(V_pfsync_swi_ie);
3104 	MPASS(ret == 0);
3105 }
3106 
3107 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH,
3108     vnet_pfsync_uninit, NULL);
3109 
3110 static int
3111 pfsync_init(void)
3112 {
3113 	int error;
3114 
3115 	pfsync_detach_ifnet_ptr = pfsync_detach_ifnet;
3116 
3117 #ifdef INET
3118 	error = ipproto_register(IPPROTO_PFSYNC, pfsync_input, NULL);
3119 	if (error)
3120 		return (error);
3121 #endif
3122 #ifdef INET6
3123 	error = ip6proto_register(IPPROTO_PFSYNC, pfsync6_input, NULL);
3124 	if (error) {
3125 		ipproto_unregister(IPPROTO_PFSYNC);
3126 		return (error);
3127 	}
3128 #endif
3129 
3130 	return (0);
3131 }
3132 
3133 static void
3134 pfsync_uninit(void)
3135 {
3136 	pfsync_detach_ifnet_ptr = NULL;
3137 
3138 #ifdef INET
3139 	ipproto_unregister(IPPROTO_PFSYNC);
3140 #endif
3141 #ifdef INET6
3142 	ip6proto_unregister(IPPROTO_PFSYNC);
3143 #endif
3144 }
3145 
3146 static int
3147 pfsync_modevent(module_t mod, int type, void *data)
3148 {
3149 	int error = 0;
3150 
3151 	switch (type) {
3152 	case MOD_LOAD:
3153 		error = pfsync_init();
3154 		break;
3155 	case MOD_UNLOAD:
3156 		pfsync_uninit();
3157 		break;
3158 	default:
3159 		error = EINVAL;
3160 		break;
3161 	}
3162 
3163 	return (error);
3164 }
3165 
3166 static moduledata_t pfsync_mod = {
3167 	pfsyncname,
3168 	pfsync_modevent,
3169 	0
3170 };
3171 
3172 #define PFSYNC_MODVER 1
3173 
3174 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
3175 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
3176 MODULE_VERSION(pfsync, PFSYNC_MODVER);
3177 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
3178