xref: /freebsd/sys/netpfil/pf/if_pfsync.c (revision 031beb4e239bfce798af17f5fe8dba8bcaf13d99)
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);
615 	st->expire = time_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);
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 					return (EINVAL);
689 				return (0);	/* skip this state */
690 			}
691 			break;
692 		default:
693 			panic("%s: Unsupported pfsync_msg_version %d",
694 			    __func__, msg_version);
695 	}
696 
697 	st->id = sp->pfs_1301.id;
698 	st->creatorid = sp->pfs_1301.creatorid;
699 	pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src);
700 	pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
701 
702 	st->rule.ptr = r;
703 	st->nat_rule.ptr = NULL;
704 	st->anchor.ptr = NULL;
705 
706 	st->pfsync_time = time_uptime;
707 	st->sync_state = PFSYNC_S_NONE;
708 
709 	if (!(flags & PFSYNC_SI_IOCTL))
710 		st->state_flags |= PFSTATE_NOSYNC;
711 
712 	if ((error = pf_state_insert(kif, kif, skw, sks, st)) != 0)
713 		goto cleanup_state;
714 
715 	/* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
716 	counter_u64_add(r->states_cur, 1);
717 	counter_u64_add(r->states_tot, 1);
718 
719 	if (!(flags & PFSYNC_SI_IOCTL)) {
720 		st->state_flags &= ~PFSTATE_NOSYNC;
721 		if (st->state_flags & PFSTATE_ACK) {
722 			struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
723 			PFSYNC_BUCKET_LOCK(b);
724 			pfsync_q_ins(st, PFSYNC_S_IACK, true);
725 			PFSYNC_BUCKET_UNLOCK(b);
726 
727 			pfsync_push_all(sc);
728 		}
729 	}
730 	st->state_flags &= ~PFSTATE_ACK;
731 	PF_STATE_UNLOCK(st);
732 
733 	return (0);
734 
735 cleanup:
736 	error = ENOMEM;
737 	if (skw == sks)
738 		sks = NULL;
739 	uma_zfree(V_pf_state_key_z, skw);
740 	uma_zfree(V_pf_state_key_z, sks);
741 
742 cleanup_state:	/* pf_state_insert() frees the state keys. */
743 	if (st) {
744 		st->timeout = PFTM_UNLINKED; /* appease an assert */
745 		pf_free_state(st);
746 	}
747 	return (error);
748 }
749 
750 #ifdef INET
751 static int
752 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
753 {
754 	struct pfsync_softc *sc = V_pfsyncif;
755 	struct mbuf *m = *mp;
756 	struct ip *ip = mtod(m, struct ip *);
757 	struct pfsync_header *ph;
758 	struct pfsync_subheader subh;
759 
760 	int offset, len, flags = 0;
761 	int rv;
762 	uint16_t count;
763 
764 	PF_RULES_RLOCK_TRACKER;
765 
766 	*mp = NULL;
767 	V_pfsyncstats.pfsyncs_ipackets++;
768 
769 	/* Verify that we have a sync interface configured. */
770 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
771 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
772 		goto done;
773 
774 	/* verify that the packet came in on the right interface */
775 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
776 		V_pfsyncstats.pfsyncs_badif++;
777 		goto done;
778 	}
779 
780 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
781 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
782 	/* verify that the IP TTL is 255. */
783 	if (ip->ip_ttl != PFSYNC_DFLTTL) {
784 		V_pfsyncstats.pfsyncs_badttl++;
785 		goto done;
786 	}
787 
788 	offset = ip->ip_hl << 2;
789 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
790 		V_pfsyncstats.pfsyncs_hdrops++;
791 		goto done;
792 	}
793 
794 	if (offset + sizeof(*ph) > m->m_len) {
795 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
796 			V_pfsyncstats.pfsyncs_hdrops++;
797 			return (IPPROTO_DONE);
798 		}
799 		ip = mtod(m, struct ip *);
800 	}
801 	ph = (struct pfsync_header *)((char *)ip + offset);
802 
803 	/* verify the version */
804 	if (ph->version != PFSYNC_VERSION) {
805 		V_pfsyncstats.pfsyncs_badver++;
806 		goto done;
807 	}
808 
809 	len = ntohs(ph->len) + offset;
810 	if (m->m_pkthdr.len < len) {
811 		V_pfsyncstats.pfsyncs_badlen++;
812 		goto done;
813 	}
814 
815 	/*
816 	 * Trusting pf_chksum during packet processing, as well as seeking
817 	 * in interface name tree, require holding PF_RULES_RLOCK().
818 	 */
819 	PF_RULES_RLOCK();
820 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
821 		flags = PFSYNC_SI_CKSUM;
822 
823 	offset += sizeof(*ph);
824 	while (offset <= len - sizeof(subh)) {
825 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
826 		offset += sizeof(subh);
827 
828 		if (subh.action >= PFSYNC_ACT_MAX) {
829 			V_pfsyncstats.pfsyncs_badact++;
830 			PF_RULES_RUNLOCK();
831 			goto done;
832 		}
833 
834 		count = ntohs(subh.count);
835 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
836 		rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action);
837 		if (rv == -1) {
838 			PF_RULES_RUNLOCK();
839 			return (IPPROTO_DONE);
840 		}
841 
842 		offset += rv;
843 	}
844 	PF_RULES_RUNLOCK();
845 
846 done:
847 	m_freem(m);
848 	return (IPPROTO_DONE);
849 }
850 #endif
851 
852 #ifdef INET6
853 static int
854 pfsync6_input(struct mbuf **mp, int *offp __unused, int proto __unused)
855 {
856 	struct pfsync_softc *sc = V_pfsyncif;
857 	struct mbuf *m = *mp;
858 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
859 	struct pfsync_header *ph;
860 	struct pfsync_subheader subh;
861 
862 	int offset, len, flags = 0;
863 	int rv;
864 	uint16_t count;
865 
866 	PF_RULES_RLOCK_TRACKER;
867 
868 	*mp = NULL;
869 	V_pfsyncstats.pfsyncs_ipackets++;
870 
871 	/* Verify that we have a sync interface configured. */
872 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
873 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
874 		goto done;
875 
876 	/* verify that the packet came in on the right interface */
877 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
878 		V_pfsyncstats.pfsyncs_badif++;
879 		goto done;
880 	}
881 
882 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
883 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
884 	/* verify that the IP TTL is 255. */
885 	if (ip6->ip6_hlim != PFSYNC_DFLTTL) {
886 		V_pfsyncstats.pfsyncs_badttl++;
887 		goto done;
888 	}
889 
890 
891 	offset = sizeof(*ip6);
892 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
893 		V_pfsyncstats.pfsyncs_hdrops++;
894 		goto done;
895 	}
896 
897 	if (offset + sizeof(*ph) > m->m_len) {
898 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
899 			V_pfsyncstats.pfsyncs_hdrops++;
900 			return (IPPROTO_DONE);
901 		}
902 		ip6 = mtod(m, struct ip6_hdr *);
903 	}
904 	ph = (struct pfsync_header *)((char *)ip6 + offset);
905 
906 	/* verify the version */
907 	if (ph->version != PFSYNC_VERSION) {
908 		V_pfsyncstats.pfsyncs_badver++;
909 		goto done;
910 	}
911 
912 	len = ntohs(ph->len) + offset;
913 	if (m->m_pkthdr.len < len) {
914 		V_pfsyncstats.pfsyncs_badlen++;
915 		goto done;
916 	}
917 
918 	/*
919 	 * Trusting pf_chksum during packet processing, as well as seeking
920 	 * in interface name tree, require holding PF_RULES_RLOCK().
921 	 */
922 	PF_RULES_RLOCK();
923 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
924 		flags = PFSYNC_SI_CKSUM;
925 
926 	offset += sizeof(*ph);
927 	while (offset <= len - sizeof(subh)) {
928 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
929 		offset += sizeof(subh);
930 
931 		if (subh.action >= PFSYNC_ACT_MAX) {
932 			V_pfsyncstats.pfsyncs_badact++;
933 			PF_RULES_RUNLOCK();
934 			goto done;
935 		}
936 
937 		count = ntohs(subh.count);
938 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
939 		rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action);
940 		if (rv == -1) {
941 			PF_RULES_RUNLOCK();
942 			return (IPPROTO_DONE);
943 		}
944 
945 		offset += rv;
946 	}
947 	PF_RULES_RUNLOCK();
948 
949 done:
950 	m_freem(m);
951 	return (IPPROTO_DONE);
952 }
953 #endif
954 
955 static int
956 pfsync_in_clr(struct mbuf *m, int offset, int count, int flags, int action)
957 {
958 	struct pfsync_clr *clr;
959 	struct mbuf *mp;
960 	int len = sizeof(*clr) * count;
961 	int i, offp;
962 	u_int32_t creatorid;
963 
964 	mp = m_pulldown(m, offset, len, &offp);
965 	if (mp == NULL) {
966 		V_pfsyncstats.pfsyncs_badlen++;
967 		return (-1);
968 	}
969 	clr = (struct pfsync_clr *)(mp->m_data + offp);
970 
971 	for (i = 0; i < count; i++) {
972 		creatorid = clr[i].creatorid;
973 
974 		if (clr[i].ifname[0] != '\0' &&
975 		    pfi_kkif_find(clr[i].ifname) == NULL)
976 			continue;
977 
978 		for (int i = 0; i <= pf_hashmask; i++) {
979 			struct pf_idhash *ih = &V_pf_idhash[i];
980 			struct pf_kstate *s;
981 relock:
982 			PF_HASHROW_LOCK(ih);
983 			LIST_FOREACH(s, &ih->states, entry) {
984 				if (s->creatorid == creatorid) {
985 					s->state_flags |= PFSTATE_NOSYNC;
986 					pf_unlink_state(s);
987 					goto relock;
988 				}
989 			}
990 			PF_HASHROW_UNLOCK(ih);
991 		}
992 	}
993 
994 	return (len);
995 }
996 
997 static int
998 pfsync_in_ins(struct mbuf *m, int offset, int count, int flags, int action)
999 {
1000 	struct mbuf *mp;
1001 	union pfsync_state_union *sa, *sp;
1002 	int i, offp, len, msg_version;
1003 
1004 	switch (action) {
1005 		case PFSYNC_ACT_INS_1301:
1006 			len = sizeof(struct pfsync_state_1301) * count;
1007 			msg_version = PFSYNC_MSG_VERSION_1301;
1008 			break;
1009 		case PFSYNC_ACT_INS_1400:
1010 			len = sizeof(struct pfsync_state_1400) * count;
1011 			msg_version = PFSYNC_MSG_VERSION_1400;
1012 			break;
1013 		default:
1014 			V_pfsyncstats.pfsyncs_badact++;
1015 			return (-1);
1016 	}
1017 
1018 	mp = m_pulldown(m, offset, len, &offp);
1019 	if (mp == NULL) {
1020 		V_pfsyncstats.pfsyncs_badlen++;
1021 		return (-1);
1022 	}
1023 	sa = (union pfsync_state_union *)(mp->m_data + offp);
1024 
1025 	for (i = 0; i < count; i++) {
1026 		sp = &sa[i];
1027 
1028 		/* Check for invalid values. */
1029 		if (sp->pfs_1301.timeout >= PFTM_MAX ||
1030 		    sp->pfs_1301.src.state > PF_TCPS_PROXY_DST ||
1031 		    sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST ||
1032 		    sp->pfs_1301.direction > PF_OUT ||
1033 		    (sp->pfs_1301.af != AF_INET &&
1034 		    sp->pfs_1301.af != AF_INET6)) {
1035 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1036 				printf("%s: invalid value\n", __func__);
1037 			V_pfsyncstats.pfsyncs_badval++;
1038 			continue;
1039 		}
1040 
1041 		if (pfsync_state_import(sp, flags, msg_version) == ENOMEM)
1042 			/* Drop out, but process the rest of the actions. */
1043 			break;
1044 	}
1045 
1046 	return (len);
1047 }
1048 
1049 static int
1050 pfsync_in_iack(struct mbuf *m, int offset, int count, int flags, int action)
1051 {
1052 	struct pfsync_ins_ack *ia, *iaa;
1053 	struct pf_kstate *st;
1054 
1055 	struct mbuf *mp;
1056 	int len = count * sizeof(*ia);
1057 	int offp, i;
1058 
1059 	mp = m_pulldown(m, offset, len, &offp);
1060 	if (mp == NULL) {
1061 		V_pfsyncstats.pfsyncs_badlen++;
1062 		return (-1);
1063 	}
1064 	iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
1065 
1066 	for (i = 0; i < count; i++) {
1067 		ia = &iaa[i];
1068 
1069 		st = pf_find_state_byid(ia->id, ia->creatorid);
1070 		if (st == NULL)
1071 			continue;
1072 
1073 		if (st->state_flags & PFSTATE_ACK) {
1074 			pfsync_undefer_state(st, 0);
1075 		}
1076 		PF_STATE_UNLOCK(st);
1077 	}
1078 	/*
1079 	 * XXX this is not yet implemented, but we know the size of the
1080 	 * message so we can skip it.
1081 	 */
1082 
1083 	return (count * sizeof(struct pfsync_ins_ack));
1084 }
1085 
1086 static int
1087 pfsync_upd_tcp(struct pf_kstate *st, struct pfsync_state_peer *src,
1088     struct pfsync_state_peer *dst)
1089 {
1090 	int sync = 0;
1091 
1092 	PF_STATE_LOCK_ASSERT(st);
1093 
1094 	/*
1095 	 * The state should never go backwards except
1096 	 * for syn-proxy states.  Neither should the
1097 	 * sequence window slide backwards.
1098 	 */
1099 	if ((st->src.state > src->state &&
1100 	    (st->src.state < PF_TCPS_PROXY_SRC ||
1101 	    src->state >= PF_TCPS_PROXY_SRC)) ||
1102 
1103 	    (st->src.state == src->state &&
1104 	    SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
1105 		sync++;
1106 	else
1107 		pf_state_peer_ntoh(src, &st->src);
1108 
1109 	if ((st->dst.state > dst->state) ||
1110 
1111 	    (st->dst.state >= TCPS_SYN_SENT &&
1112 	    SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
1113 		sync++;
1114 	else
1115 		pf_state_peer_ntoh(dst, &st->dst);
1116 
1117 	return (sync);
1118 }
1119 
1120 static int
1121 pfsync_in_upd(struct mbuf *m, int offset, int count, int flags, int action)
1122 {
1123 	struct pfsync_softc *sc = V_pfsyncif;
1124 	union pfsync_state_union *sa, *sp;
1125 	struct pf_kstate *st;
1126 	struct mbuf *mp;
1127 	int sync, offp, i, len, msg_version;
1128 
1129 	switch (action) {
1130 		case PFSYNC_ACT_UPD_1301:
1131 			len = sizeof(struct pfsync_state_1301) * count;
1132 			msg_version = PFSYNC_MSG_VERSION_1301;
1133 			break;
1134 		case PFSYNC_ACT_UPD_1400:
1135 			len = sizeof(struct pfsync_state_1400) * count;
1136 			msg_version = PFSYNC_MSG_VERSION_1400;
1137 			break;
1138 		default:
1139 			V_pfsyncstats.pfsyncs_badact++;
1140 			return (-1);
1141 	}
1142 
1143 	mp = m_pulldown(m, offset, len, &offp);
1144 	if (mp == NULL) {
1145 		V_pfsyncstats.pfsyncs_badlen++;
1146 		return (-1);
1147 	}
1148 	sa = (union pfsync_state_union *)(mp->m_data + offp);
1149 
1150 	for (i = 0; i < count; i++) {
1151 		sp = &sa[i];
1152 
1153 		/* check for invalid values */
1154 		if (sp->pfs_1301.timeout >= PFTM_MAX ||
1155 		    sp->pfs_1301.src.state > PF_TCPS_PROXY_DST ||
1156 		    sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST) {
1157 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1158 				printf("pfsync_input: PFSYNC_ACT_UPD: "
1159 				    "invalid value\n");
1160 			}
1161 			V_pfsyncstats.pfsyncs_badval++;
1162 			continue;
1163 		}
1164 
1165 		st = pf_find_state_byid(sp->pfs_1301.id, sp->pfs_1301.creatorid);
1166 		if (st == NULL) {
1167 			/* insert the update */
1168 			if (pfsync_state_import(sp, flags, msg_version))
1169 				V_pfsyncstats.pfsyncs_badstate++;
1170 			continue;
1171 		}
1172 
1173 		if (st->state_flags & PFSTATE_ACK) {
1174 			pfsync_undefer_state(st, 1);
1175 		}
1176 
1177 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1178 			sync = pfsync_upd_tcp(st, &sp->pfs_1301.src, &sp->pfs_1301.dst);
1179 		else {
1180 			sync = 0;
1181 
1182 			/*
1183 			 * Non-TCP protocol state machine always go
1184 			 * forwards
1185 			 */
1186 			if (st->src.state > sp->pfs_1301.src.state)
1187 				sync++;
1188 			else
1189 				pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src);
1190 			if (st->dst.state > sp->pfs_1301.dst.state)
1191 				sync++;
1192 			else
1193 				pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
1194 		}
1195 		if (sync < 2) {
1196 			pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst);
1197 			pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
1198 			st->expire = time_uptime;
1199 			st->timeout = sp->pfs_1301.timeout;
1200 		}
1201 		st->pfsync_time = time_uptime;
1202 
1203 		if (sync) {
1204 			V_pfsyncstats.pfsyncs_stale++;
1205 
1206 			pfsync_update_state(st);
1207 			PF_STATE_UNLOCK(st);
1208 			pfsync_push_all(sc);
1209 			continue;
1210 		}
1211 		PF_STATE_UNLOCK(st);
1212 	}
1213 
1214 	return (len);
1215 }
1216 
1217 static int
1218 pfsync_in_upd_c(struct mbuf *m, int offset, int count, int flags, int action)
1219 {
1220 	struct pfsync_softc *sc = V_pfsyncif;
1221 	struct pfsync_upd_c *ua, *up;
1222 	struct pf_kstate *st;
1223 	int len = count * sizeof(*up);
1224 	int sync;
1225 	struct mbuf *mp;
1226 	int offp, i;
1227 
1228 	mp = m_pulldown(m, offset, len, &offp);
1229 	if (mp == NULL) {
1230 		V_pfsyncstats.pfsyncs_badlen++;
1231 		return (-1);
1232 	}
1233 	ua = (struct pfsync_upd_c *)(mp->m_data + offp);
1234 
1235 	for (i = 0; i < count; i++) {
1236 		up = &ua[i];
1237 
1238 		/* check for invalid values */
1239 		if (up->timeout >= PFTM_MAX ||
1240 		    up->src.state > PF_TCPS_PROXY_DST ||
1241 		    up->dst.state > PF_TCPS_PROXY_DST) {
1242 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1243 				printf("pfsync_input: "
1244 				    "PFSYNC_ACT_UPD_C: "
1245 				    "invalid value\n");
1246 			}
1247 			V_pfsyncstats.pfsyncs_badval++;
1248 			continue;
1249 		}
1250 
1251 		st = pf_find_state_byid(up->id, up->creatorid);
1252 		if (st == NULL) {
1253 			/* We don't have this state. Ask for it. */
1254 			PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1255 			pfsync_request_update(up->creatorid, up->id);
1256 			PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1257 			continue;
1258 		}
1259 
1260 		if (st->state_flags & PFSTATE_ACK) {
1261 			pfsync_undefer_state(st, 1);
1262 		}
1263 
1264 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1265 			sync = pfsync_upd_tcp(st, &up->src, &up->dst);
1266 		else {
1267 			sync = 0;
1268 
1269 			/*
1270 			 * Non-TCP protocol state machine always go
1271 			 * forwards
1272 			 */
1273 			if (st->src.state > up->src.state)
1274 				sync++;
1275 			else
1276 				pf_state_peer_ntoh(&up->src, &st->src);
1277 			if (st->dst.state > up->dst.state)
1278 				sync++;
1279 			else
1280 				pf_state_peer_ntoh(&up->dst, &st->dst);
1281 		}
1282 		if (sync < 2) {
1283 			pfsync_alloc_scrub_memory(&up->dst, &st->dst);
1284 			pf_state_peer_ntoh(&up->dst, &st->dst);
1285 			st->expire = time_uptime;
1286 			st->timeout = up->timeout;
1287 		}
1288 		st->pfsync_time = time_uptime;
1289 
1290 		if (sync) {
1291 			V_pfsyncstats.pfsyncs_stale++;
1292 
1293 			pfsync_update_state(st);
1294 			PF_STATE_UNLOCK(st);
1295 			pfsync_push_all(sc);
1296 			continue;
1297 		}
1298 		PF_STATE_UNLOCK(st);
1299 	}
1300 
1301 	return (len);
1302 }
1303 
1304 static int
1305 pfsync_in_ureq(struct mbuf *m, int offset, int count, int flags, int action)
1306 {
1307 	struct pfsync_upd_req *ur, *ura;
1308 	struct mbuf *mp;
1309 	int len = count * sizeof(*ur);
1310 	int i, offp;
1311 
1312 	struct pf_kstate *st;
1313 
1314 	mp = m_pulldown(m, offset, len, &offp);
1315 	if (mp == NULL) {
1316 		V_pfsyncstats.pfsyncs_badlen++;
1317 		return (-1);
1318 	}
1319 	ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1320 
1321 	for (i = 0; i < count; i++) {
1322 		ur = &ura[i];
1323 
1324 		if (ur->id == 0 && ur->creatorid == 0)
1325 			pfsync_bulk_start();
1326 		else {
1327 			st = pf_find_state_byid(ur->id, ur->creatorid);
1328 			if (st == NULL) {
1329 				V_pfsyncstats.pfsyncs_badstate++;
1330 				continue;
1331 			}
1332 			if (st->state_flags & PFSTATE_NOSYNC) {
1333 				PF_STATE_UNLOCK(st);
1334 				continue;
1335 			}
1336 
1337 			pfsync_update_state_req(st);
1338 			PF_STATE_UNLOCK(st);
1339 		}
1340 	}
1341 
1342 	return (len);
1343 }
1344 
1345 static int
1346 pfsync_in_del_c(struct mbuf *m, int offset, int count, int flags, int action)
1347 {
1348 	struct mbuf *mp;
1349 	struct pfsync_del_c *sa, *sp;
1350 	struct pf_kstate *st;
1351 	int len = count * sizeof(*sp);
1352 	int offp, i;
1353 
1354 	mp = m_pulldown(m, offset, len, &offp);
1355 	if (mp == NULL) {
1356 		V_pfsyncstats.pfsyncs_badlen++;
1357 		return (-1);
1358 	}
1359 	sa = (struct pfsync_del_c *)(mp->m_data + offp);
1360 
1361 	for (i = 0; i < count; i++) {
1362 		sp = &sa[i];
1363 
1364 		st = pf_find_state_byid(sp->id, sp->creatorid);
1365 		if (st == NULL) {
1366 			V_pfsyncstats.pfsyncs_badstate++;
1367 			continue;
1368 		}
1369 
1370 		st->state_flags |= PFSTATE_NOSYNC;
1371 		pf_unlink_state(st);
1372 	}
1373 
1374 	return (len);
1375 }
1376 
1377 static int
1378 pfsync_in_bus(struct mbuf *m, int offset, int count, int flags, int action)
1379 {
1380 	struct pfsync_softc *sc = V_pfsyncif;
1381 	struct pfsync_bus *bus;
1382 	struct mbuf *mp;
1383 	int len = count * sizeof(*bus);
1384 	int offp;
1385 
1386 	PFSYNC_BLOCK(sc);
1387 
1388 	/* If we're not waiting for a bulk update, who cares. */
1389 	if (sc->sc_ureq_sent == 0) {
1390 		PFSYNC_BUNLOCK(sc);
1391 		return (len);
1392 	}
1393 
1394 	mp = m_pulldown(m, offset, len, &offp);
1395 	if (mp == NULL) {
1396 		PFSYNC_BUNLOCK(sc);
1397 		V_pfsyncstats.pfsyncs_badlen++;
1398 		return (-1);
1399 	}
1400 	bus = (struct pfsync_bus *)(mp->m_data + offp);
1401 
1402 	switch (bus->status) {
1403 	case PFSYNC_BUS_START:
1404 		callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1405 		    V_pf_limits[PF_LIMIT_STATES].limit /
1406 		    ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1407 		    sizeof(union pfsync_state_union)),
1408 		    pfsync_bulk_fail, sc);
1409 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1410 			printf("pfsync: received bulk update start\n");
1411 		break;
1412 
1413 	case PFSYNC_BUS_END:
1414 		if (time_uptime - ntohl(bus->endtime) >=
1415 		    sc->sc_ureq_sent) {
1416 			/* that's it, we're happy */
1417 			sc->sc_ureq_sent = 0;
1418 			sc->sc_bulk_tries = 0;
1419 			callout_stop(&sc->sc_bulkfail_tmo);
1420 			if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1421 				(*carp_demote_adj_p)(-V_pfsync_carp_adj,
1422 				    "pfsync bulk done");
1423 			sc->sc_flags |= PFSYNCF_OK;
1424 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1425 				printf("pfsync: received valid "
1426 				    "bulk update end\n");
1427 		} else {
1428 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1429 				printf("pfsync: received invalid "
1430 				    "bulk update end: bad timestamp\n");
1431 		}
1432 		break;
1433 	}
1434 	PFSYNC_BUNLOCK(sc);
1435 
1436 	return (len);
1437 }
1438 
1439 static int
1440 pfsync_in_tdb(struct mbuf *m, int offset, int count, int flags, int action)
1441 {
1442 	int len = count * sizeof(struct pfsync_tdb);
1443 
1444 #if defined(IPSEC)
1445 	struct pfsync_tdb *tp;
1446 	struct mbuf *mp;
1447 	int offp;
1448 	int i;
1449 	int s;
1450 
1451 	mp = m_pulldown(m, offset, len, &offp);
1452 	if (mp == NULL) {
1453 		V_pfsyncstats.pfsyncs_badlen++;
1454 		return (-1);
1455 	}
1456 	tp = (struct pfsync_tdb *)(mp->m_data + offp);
1457 
1458 	for (i = 0; i < count; i++)
1459 		pfsync_update_net_tdb(&tp[i]);
1460 #endif
1461 
1462 	return (len);
1463 }
1464 
1465 #if defined(IPSEC)
1466 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1467 static void
1468 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1469 {
1470 	struct tdb		*tdb;
1471 	int			 s;
1472 
1473 	/* check for invalid values */
1474 	if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1475 	    (pt->dst.sa.sa_family != AF_INET &&
1476 	    pt->dst.sa.sa_family != AF_INET6))
1477 		goto bad;
1478 
1479 	tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1480 	if (tdb) {
1481 		pt->rpl = ntohl(pt->rpl);
1482 		pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1483 
1484 		/* Neither replay nor byte counter should ever decrease. */
1485 		if (pt->rpl < tdb->tdb_rpl ||
1486 		    pt->cur_bytes < tdb->tdb_cur_bytes) {
1487 			goto bad;
1488 		}
1489 
1490 		tdb->tdb_rpl = pt->rpl;
1491 		tdb->tdb_cur_bytes = pt->cur_bytes;
1492 	}
1493 	return;
1494 
1495 bad:
1496 	if (V_pf_status.debug >= PF_DEBUG_MISC)
1497 		printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1498 		    "invalid value\n");
1499 	V_pfsyncstats.pfsyncs_badstate++;
1500 	return;
1501 }
1502 #endif
1503 
1504 static int
1505 pfsync_in_eof(struct mbuf *m, int offset, int count, int flags, int action)
1506 {
1507 	/* check if we are at the right place in the packet */
1508 	if (offset != m->m_pkthdr.len)
1509 		V_pfsyncstats.pfsyncs_badlen++;
1510 
1511 	/* we're done. free and let the caller return */
1512 	m_freem(m);
1513 	return (-1);
1514 }
1515 
1516 static int
1517 pfsync_in_error(struct mbuf *m, int offset, int count, int flags, int action)
1518 {
1519 	V_pfsyncstats.pfsyncs_badact++;
1520 
1521 	m_freem(m);
1522 	return (-1);
1523 }
1524 
1525 static int
1526 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1527 	struct route *rt)
1528 {
1529 	m_freem(m);
1530 	return (0);
1531 }
1532 
1533 /* ARGSUSED */
1534 static int
1535 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1536 {
1537 	struct pfsync_softc *sc = ifp->if_softc;
1538 	struct ifreq *ifr = (struct ifreq *)data;
1539 	struct pfsyncreq pfsyncr;
1540 	size_t nvbuflen;
1541 	int error;
1542 	int c;
1543 
1544 	switch (cmd) {
1545 	case SIOCSIFFLAGS:
1546 		PFSYNC_LOCK(sc);
1547 		if (ifp->if_flags & IFF_UP) {
1548 			ifp->if_drv_flags |= IFF_DRV_RUNNING;
1549 			PFSYNC_UNLOCK(sc);
1550 			pfsync_pointers_init();
1551 		} else {
1552 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1553 			PFSYNC_UNLOCK(sc);
1554 			pfsync_pointers_uninit();
1555 		}
1556 		break;
1557 	case SIOCSIFMTU:
1558 		if (!sc->sc_sync_if ||
1559 		    ifr->ifr_mtu <= PFSYNC_MINPKT ||
1560 		    ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1561 			return (EINVAL);
1562 		if (ifr->ifr_mtu < ifp->if_mtu) {
1563 			for (c = 0; c < pfsync_buckets; c++) {
1564 				PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1565 				if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT)
1566 					pfsync_sendout(1, c);
1567 				PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1568 			}
1569 		}
1570 		ifp->if_mtu = ifr->ifr_mtu;
1571 		break;
1572 	case SIOCGETPFSYNC:
1573 		bzero(&pfsyncr, sizeof(pfsyncr));
1574 		PFSYNC_LOCK(sc);
1575 		if (sc->sc_sync_if) {
1576 			strlcpy(pfsyncr.pfsyncr_syncdev,
1577 			    sc->sc_sync_if->if_xname, IFNAMSIZ);
1578 		}
1579 		pfsyncr.pfsyncr_syncpeer = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
1580 		pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1581 		pfsyncr.pfsyncr_defer = sc->sc_flags;
1582 		PFSYNC_UNLOCK(sc);
1583 		return (copyout(&pfsyncr, ifr_data_get_ptr(ifr),
1584 		    sizeof(pfsyncr)));
1585 
1586 	case SIOCGETPFSYNCNV:
1587 	    {
1588 		nvlist_t *nvl_syncpeer;
1589 		nvlist_t *nvl = nvlist_create(0);
1590 
1591 		if (nvl == NULL)
1592 			return (ENOMEM);
1593 
1594 		if (sc->sc_sync_if)
1595 			nvlist_add_string(nvl, "syncdev", sc->sc_sync_if->if_xname);
1596 		nvlist_add_number(nvl, "maxupdates", sc->sc_maxupdates);
1597 		nvlist_add_number(nvl, "flags", sc->sc_flags);
1598 		nvlist_add_number(nvl, "version", sc->sc_version);
1599 		if ((nvl_syncpeer = pfsync_sockaddr_to_syncpeer_nvlist(&sc->sc_sync_peer)) != NULL)
1600 			nvlist_add_nvlist(nvl, "syncpeer", nvl_syncpeer);
1601 
1602 		void *packed = NULL;
1603 		packed = nvlist_pack(nvl, &nvbuflen);
1604 		if (packed == NULL) {
1605 			free(packed, M_NVLIST);
1606 			nvlist_destroy(nvl);
1607 			return (ENOMEM);
1608 		}
1609 
1610 		if (nvbuflen > ifr->ifr_cap_nv.buf_length) {
1611 			ifr->ifr_cap_nv.length = nvbuflen;
1612 			ifr->ifr_cap_nv.buffer = NULL;
1613 			free(packed, M_NVLIST);
1614 			nvlist_destroy(nvl);
1615 			return (EFBIG);
1616 		}
1617 
1618 		ifr->ifr_cap_nv.length = nvbuflen;
1619 		error = copyout(packed, ifr->ifr_cap_nv.buffer, nvbuflen);
1620 
1621 		nvlist_destroy(nvl);
1622 		nvlist_destroy(nvl_syncpeer);
1623 		free(packed, M_NVLIST);
1624 		break;
1625 	    }
1626 
1627 	case SIOCSETPFSYNC:
1628 	    {
1629 		struct pfsync_kstatus status;
1630 
1631 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1632 			return (error);
1633 		if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr,
1634 		    sizeof(pfsyncr))))
1635 			return (error);
1636 
1637 		memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1638 		pfsync_pfsyncreq_to_kstatus(&pfsyncr, &status);
1639 
1640 		error = pfsync_kstatus_to_softc(&status, sc);
1641 		return (error);
1642 	    }
1643 	case SIOCSETPFSYNCNV:
1644 	    {
1645 		struct pfsync_kstatus status;
1646 		void *data;
1647 		nvlist_t *nvl;
1648 
1649 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1650 			return (error);
1651 		if (ifr->ifr_cap_nv.length > IFR_CAP_NV_MAXBUFSIZE)
1652 			return (EINVAL);
1653 
1654 		data = malloc(ifr->ifr_cap_nv.length, M_TEMP, M_WAITOK);
1655 
1656 		if ((error = copyin(ifr->ifr_cap_nv.buffer, data,
1657 		    ifr->ifr_cap_nv.length)) != 0) {
1658 			free(data, M_TEMP);
1659 			return (error);
1660 		}
1661 
1662 		if ((nvl = nvlist_unpack(data, ifr->ifr_cap_nv.length, 0)) == NULL) {
1663 			free(data, M_TEMP);
1664 			return (EINVAL);
1665 		}
1666 
1667 		memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1668 		pfsync_nvstatus_to_kstatus(nvl, &status);
1669 
1670 		nvlist_destroy(nvl);
1671 		free(data, M_TEMP);
1672 
1673 		error = pfsync_kstatus_to_softc(&status, sc);
1674 		return (error);
1675 	    }
1676 	default:
1677 		return (ENOTTY);
1678 	}
1679 
1680 	return (0);
1681 }
1682 
1683 static void
1684 pfsync_out_state_1301(struct pf_kstate *st, void *buf)
1685 {
1686 	union pfsync_state_union *sp = buf;
1687 
1688 	pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1301);
1689 }
1690 
1691 static void
1692 pfsync_out_state_1400(struct pf_kstate *st, void *buf)
1693 {
1694 	union pfsync_state_union *sp = buf;
1695 
1696 	pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1400);
1697 }
1698 
1699 static void
1700 pfsync_out_iack(struct pf_kstate *st, void *buf)
1701 {
1702 	struct pfsync_ins_ack *iack = buf;
1703 
1704 	iack->id = st->id;
1705 	iack->creatorid = st->creatorid;
1706 }
1707 
1708 static void
1709 pfsync_out_upd_c(struct pf_kstate *st, void *buf)
1710 {
1711 	struct pfsync_upd_c *up = buf;
1712 
1713 	bzero(up, sizeof(*up));
1714 	up->id = st->id;
1715 	pf_state_peer_hton(&st->src, &up->src);
1716 	pf_state_peer_hton(&st->dst, &up->dst);
1717 	up->creatorid = st->creatorid;
1718 	up->timeout = st->timeout;
1719 }
1720 
1721 static void
1722 pfsync_out_del_c(struct pf_kstate *st, void *buf)
1723 {
1724 	struct pfsync_del_c *dp = buf;
1725 
1726 	dp->id = st->id;
1727 	dp->creatorid = st->creatorid;
1728 	st->state_flags |= PFSTATE_NOSYNC;
1729 }
1730 
1731 static void
1732 pfsync_drop(struct pfsync_softc *sc)
1733 {
1734 	struct pf_kstate *st, *next;
1735 	struct pfsync_upd_req_item *ur;
1736 	struct pfsync_bucket *b;
1737 	int c;
1738 	enum pfsync_q_id q;
1739 
1740 	for (c = 0; c < pfsync_buckets; c++) {
1741 		b = &sc->sc_buckets[c];
1742 		for (q = 0; q < PFSYNC_Q_COUNT; q++) {
1743 			if (TAILQ_EMPTY(&b->b_qs[q]))
1744 				continue;
1745 
1746 			TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) {
1747 				KASSERT(st->sync_state == pfsync_qid_sstate[q],
1748 					("%s: st->sync_state == q",
1749 						__func__));
1750 				st->sync_state = PFSYNC_S_NONE;
1751 				pf_release_state(st);
1752 			}
1753 			TAILQ_INIT(&b->b_qs[q]);
1754 		}
1755 
1756 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1757 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1758 			free(ur, M_PFSYNC);
1759 		}
1760 
1761 		b->b_len = PFSYNC_MINPKT;
1762 		b->b_plus = NULL;
1763 	}
1764 }
1765 
1766 static void
1767 pfsync_sendout(int schedswi, int c)
1768 {
1769 	struct pfsync_softc *sc = V_pfsyncif;
1770 	struct ifnet *ifp = sc->sc_ifp;
1771 	struct mbuf *m;
1772 	struct pfsync_header *ph;
1773 	struct pfsync_subheader *subh;
1774 	struct pf_kstate *st, *st_next;
1775 	struct pfsync_upd_req_item *ur;
1776 	struct pfsync_bucket *b = &sc->sc_buckets[c];
1777 	int aflen, offset, count = 0;
1778 	enum pfsync_q_id q;
1779 
1780 	KASSERT(sc != NULL, ("%s: null sc", __func__));
1781 	KASSERT(b->b_len > PFSYNC_MINPKT,
1782 	    ("%s: sc_len %zu", __func__, b->b_len));
1783 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1784 
1785 	if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1786 		pfsync_drop(sc);
1787 		return;
1788 	}
1789 
1790 	m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1791 	if (m == NULL) {
1792 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1793 		V_pfsyncstats.pfsyncs_onomem++;
1794 		return;
1795 	}
1796 	m->m_data += max_linkhdr;
1797 	m->m_len = m->m_pkthdr.len = b->b_len;
1798 
1799 	/* build the ip header */
1800 	switch (sc->sc_sync_peer.ss_family) {
1801 #ifdef INET
1802 	case AF_INET:
1803 	    {
1804 		struct ip *ip;
1805 
1806 		ip = mtod(m, struct ip *);
1807 		bcopy(&sc->sc_template.ipv4, ip, sizeof(*ip));
1808 		aflen = offset = sizeof(*ip);
1809 
1810 		ip->ip_len = htons(m->m_pkthdr.len);
1811 		ip_fillid(ip);
1812 		break;
1813 	    }
1814 #endif
1815 #ifdef INET6
1816 	case AF_INET6:
1817 		{
1818 		struct ip6_hdr *ip6;
1819 
1820 		ip6 = mtod(m, struct ip6_hdr *);
1821 		bcopy(&sc->sc_template.ipv6, ip6, sizeof(*ip6));
1822 		aflen = offset = sizeof(*ip6);
1823 
1824 		ip6->ip6_plen = htons(m->m_pkthdr.len);
1825 		break;
1826 		}
1827 #endif
1828 	default:
1829 		m_freem(m);
1830 		return;
1831 	}
1832 
1833 	/* build the pfsync header */
1834 	ph = (struct pfsync_header *)(m->m_data + offset);
1835 	bzero(ph, sizeof(*ph));
1836 	offset += sizeof(*ph);
1837 
1838 	ph->version = PFSYNC_VERSION;
1839 	ph->len = htons(b->b_len - aflen);
1840 	bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1841 
1842 	/* walk the queues */
1843 	for (q = 0; q < PFSYNC_Q_COUNT; q++) {
1844 		if (TAILQ_EMPTY(&b->b_qs[q]))
1845 			continue;
1846 
1847 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1848 		offset += sizeof(*subh);
1849 
1850 		count = 0;
1851 		TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) {
1852 			KASSERT(st->sync_state == pfsync_qid_sstate[q],
1853 				("%s: st->sync_state == q",
1854 					__func__));
1855 			/*
1856 			 * XXXGL: some of write methods do unlocked reads
1857 			 * of state data :(
1858 			 */
1859 			pfsync_qs[q].write(st, m->m_data + offset);
1860 			offset += pfsync_qs[q].len;
1861 			st->sync_state = PFSYNC_S_NONE;
1862 			pf_release_state(st);
1863 			count++;
1864 		}
1865 		TAILQ_INIT(&b->b_qs[q]);
1866 
1867 		bzero(subh, sizeof(*subh));
1868 		subh->action = pfsync_qs[q].action;
1869 		subh->count = htons(count);
1870 		V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1871 	}
1872 
1873 	if (!TAILQ_EMPTY(&b->b_upd_req_list)) {
1874 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1875 		offset += sizeof(*subh);
1876 
1877 		count = 0;
1878 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1879 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1880 
1881 			bcopy(&ur->ur_msg, m->m_data + offset,
1882 			    sizeof(ur->ur_msg));
1883 			offset += sizeof(ur->ur_msg);
1884 			free(ur, M_PFSYNC);
1885 			count++;
1886 		}
1887 
1888 		bzero(subh, sizeof(*subh));
1889 		subh->action = PFSYNC_ACT_UPD_REQ;
1890 		subh->count = htons(count);
1891 		V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1892 	}
1893 
1894 	/* has someone built a custom region for us to add? */
1895 	if (b->b_plus != NULL) {
1896 		bcopy(b->b_plus, m->m_data + offset, b->b_pluslen);
1897 		offset += b->b_pluslen;
1898 
1899 		b->b_plus = NULL;
1900 	}
1901 
1902 	subh = (struct pfsync_subheader *)(m->m_data + offset);
1903 	offset += sizeof(*subh);
1904 
1905 	bzero(subh, sizeof(*subh));
1906 	subh->action = PFSYNC_ACT_EOF;
1907 	subh->count = htons(1);
1908 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1909 
1910 	/* we're done, let's put it on the wire */
1911 	if (ifp->if_bpf) {
1912 		m->m_data += aflen;
1913 		m->m_len = m->m_pkthdr.len = b->b_len - aflen;
1914 		BPF_MTAP(ifp, m);
1915 		m->m_data -= aflen;
1916 		m->m_len = m->m_pkthdr.len = b->b_len;
1917 	}
1918 
1919 	if (sc->sc_sync_if == NULL) {
1920 		b->b_len = PFSYNC_MINPKT;
1921 		m_freem(m);
1922 		return;
1923 	}
1924 
1925 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1926 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1927 	b->b_len = PFSYNC_MINPKT;
1928 
1929 	if (!_IF_QFULL(&b->b_snd))
1930 		_IF_ENQUEUE(&b->b_snd, m);
1931 	else {
1932 		m_freem(m);
1933 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1934 	}
1935 	if (schedswi)
1936 		swi_sched(V_pfsync_swi_cookie, 0);
1937 }
1938 
1939 static void
1940 pfsync_insert_state(struct pf_kstate *st)
1941 {
1942 	struct pfsync_softc *sc = V_pfsyncif;
1943 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1944 
1945 	if (st->state_flags & PFSTATE_NOSYNC)
1946 		return;
1947 
1948 	if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1949 	    st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1950 		st->state_flags |= PFSTATE_NOSYNC;
1951 		return;
1952 	}
1953 
1954 	KASSERT(st->sync_state == PFSYNC_S_NONE,
1955 		("%s: st->sync_state %u", __func__, st->sync_state));
1956 
1957 	PFSYNC_BUCKET_LOCK(b);
1958 	if (b->b_len == PFSYNC_MINPKT)
1959 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1960 
1961 	pfsync_q_ins(st, PFSYNC_S_INS, true);
1962 	PFSYNC_BUCKET_UNLOCK(b);
1963 
1964 	st->sync_updates = 0;
1965 }
1966 
1967 static int
1968 pfsync_defer(struct pf_kstate *st, struct mbuf *m)
1969 {
1970 	struct pfsync_softc *sc = V_pfsyncif;
1971 	struct pfsync_deferral *pd;
1972 	struct pfsync_bucket *b;
1973 
1974 	if (m->m_flags & (M_BCAST|M_MCAST))
1975 		return (0);
1976 
1977 	if (sc == NULL)
1978 		return (0);
1979 
1980 	b = pfsync_get_bucket(sc, st);
1981 
1982 	PFSYNC_LOCK(sc);
1983 
1984 	if (!(sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) ||
1985 	    !(sc->sc_flags & PFSYNCF_DEFER)) {
1986 		PFSYNC_UNLOCK(sc);
1987 		return (0);
1988 	}
1989 
1990 	PFSYNC_BUCKET_LOCK(b);
1991 	PFSYNC_UNLOCK(sc);
1992 
1993 	if (b->b_deferred >= 128)
1994 		pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0);
1995 
1996 	pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1997 	if (pd == NULL) {
1998 		PFSYNC_BUCKET_UNLOCK(b);
1999 		return (0);
2000 	}
2001 	b->b_deferred++;
2002 
2003 	m->m_flags |= M_SKIP_FIREWALL;
2004 	st->state_flags |= PFSTATE_ACK;
2005 
2006 	pd->pd_sc = sc;
2007 	pd->pd_st = st;
2008 	pf_ref_state(st);
2009 	pd->pd_m = m;
2010 
2011 	TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry);
2012 	callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED);
2013 	callout_reset(&pd->pd_tmo, (V_pfsync_defer_timeout * hz) / 1000,
2014 	    pfsync_defer_tmo, pd);
2015 
2016 	pfsync_push(b);
2017 	PFSYNC_BUCKET_UNLOCK(b);
2018 
2019 	return (1);
2020 }
2021 
2022 static void
2023 pfsync_undefer(struct pfsync_deferral *pd, int drop)
2024 {
2025 	struct pfsync_softc *sc = pd->pd_sc;
2026 	struct mbuf *m = pd->pd_m;
2027 	struct pf_kstate *st = pd->pd_st;
2028 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2029 
2030 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2031 
2032 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
2033 	b->b_deferred--;
2034 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
2035 	free(pd, M_PFSYNC);
2036 	pf_release_state(st);
2037 
2038 	if (drop)
2039 		m_freem(m);
2040 	else {
2041 		_IF_ENQUEUE(&b->b_snd, m);
2042 		pfsync_push(b);
2043 	}
2044 }
2045 
2046 static void
2047 pfsync_defer_tmo(void *arg)
2048 {
2049 	struct epoch_tracker et;
2050 	struct pfsync_deferral *pd = arg;
2051 	struct pfsync_softc *sc = pd->pd_sc;
2052 	struct mbuf *m = pd->pd_m;
2053 	struct pf_kstate *st = pd->pd_st;
2054 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2055 
2056 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2057 
2058 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
2059 	b->b_deferred--;
2060 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
2061 	PFSYNC_BUCKET_UNLOCK(b);
2062 	free(pd, M_PFSYNC);
2063 
2064 	if (sc->sc_sync_if == NULL) {
2065 		pf_release_state(st);
2066 		m_freem(m);
2067 		return;
2068 	}
2069 
2070 	NET_EPOCH_ENTER(et);
2071 	CURVNET_SET(sc->sc_sync_if->if_vnet);
2072 
2073 	pfsync_tx(sc, m);
2074 
2075 	pf_release_state(st);
2076 
2077 	CURVNET_RESTORE();
2078 	NET_EPOCH_EXIT(et);
2079 }
2080 
2081 static void
2082 pfsync_undefer_state_locked(struct pf_kstate *st, int drop)
2083 {
2084 	struct pfsync_softc *sc = V_pfsyncif;
2085 	struct pfsync_deferral *pd;
2086 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2087 
2088 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2089 
2090 	TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) {
2091 		 if (pd->pd_st == st) {
2092 			if (callout_stop(&pd->pd_tmo) > 0)
2093 				pfsync_undefer(pd, drop);
2094 
2095 			return;
2096 		}
2097 	}
2098 
2099 	panic("%s: unable to find deferred state", __func__);
2100 }
2101 
2102 static void
2103 pfsync_undefer_state(struct pf_kstate *st, int drop)
2104 {
2105 	struct pfsync_softc *sc = V_pfsyncif;
2106 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2107 
2108 	PFSYNC_BUCKET_LOCK(b);
2109 	pfsync_undefer_state_locked(st, drop);
2110 	PFSYNC_BUCKET_UNLOCK(b);
2111 }
2112 
2113 static struct pfsync_bucket*
2114 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_kstate *st)
2115 {
2116 	int c = PF_IDHASH(st) % pfsync_buckets;
2117 	return &sc->sc_buckets[c];
2118 }
2119 
2120 static void
2121 pfsync_update_state(struct pf_kstate *st)
2122 {
2123 	struct pfsync_softc *sc = V_pfsyncif;
2124 	bool sync = false, ref = true;
2125 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2126 
2127 	PF_STATE_LOCK_ASSERT(st);
2128 	PFSYNC_BUCKET_LOCK(b);
2129 
2130 	if (st->state_flags & PFSTATE_ACK)
2131 		pfsync_undefer_state_locked(st, 0);
2132 	if (st->state_flags & PFSTATE_NOSYNC) {
2133 		if (st->sync_state != PFSYNC_S_NONE)
2134 			pfsync_q_del(st, true, b);
2135 		PFSYNC_BUCKET_UNLOCK(b);
2136 		return;
2137 	}
2138 
2139 	if (b->b_len == PFSYNC_MINPKT)
2140 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2141 
2142 	switch (st->sync_state) {
2143 	case PFSYNC_S_UPD_C:
2144 	case PFSYNC_S_UPD:
2145 	case PFSYNC_S_INS:
2146 		/* we're already handling it */
2147 
2148 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
2149 			st->sync_updates++;
2150 			if (st->sync_updates >= sc->sc_maxupdates)
2151 				sync = true;
2152 		}
2153 		break;
2154 
2155 	case PFSYNC_S_IACK:
2156 		pfsync_q_del(st, false, b);
2157 		ref = false;
2158 		/* FALLTHROUGH */
2159 
2160 	case PFSYNC_S_NONE:
2161 		pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
2162 		st->sync_updates = 0;
2163 		break;
2164 
2165 	default:
2166 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2167 	}
2168 
2169 	if (sync || (time_uptime - st->pfsync_time) < 2)
2170 		pfsync_push(b);
2171 
2172 	PFSYNC_BUCKET_UNLOCK(b);
2173 }
2174 
2175 static void
2176 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
2177 {
2178 	struct pfsync_softc *sc = V_pfsyncif;
2179 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2180 	struct pfsync_upd_req_item *item;
2181 	size_t nlen = sizeof(struct pfsync_upd_req);
2182 
2183 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2184 
2185 	/*
2186 	 * This code does a bit to prevent multiple update requests for the
2187 	 * same state being generated. It searches current subheader queue,
2188 	 * but it doesn't lookup into queue of already packed datagrams.
2189 	 */
2190 	TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry)
2191 		if (item->ur_msg.id == id &&
2192 		    item->ur_msg.creatorid == creatorid)
2193 			return;
2194 
2195 	item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
2196 	if (item == NULL)
2197 		return; /* XXX stats */
2198 
2199 	item->ur_msg.id = id;
2200 	item->ur_msg.creatorid = creatorid;
2201 
2202 	if (TAILQ_EMPTY(&b->b_upd_req_list))
2203 		nlen += sizeof(struct pfsync_subheader);
2204 
2205 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2206 		pfsync_sendout(0, 0);
2207 
2208 		nlen = sizeof(struct pfsync_subheader) +
2209 		    sizeof(struct pfsync_upd_req);
2210 	}
2211 
2212 	TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry);
2213 	b->b_len += nlen;
2214 
2215 	pfsync_push(b);
2216 }
2217 
2218 static bool
2219 pfsync_update_state_req(struct pf_kstate *st)
2220 {
2221 	struct pfsync_softc *sc = V_pfsyncif;
2222 	bool ref = true, full = false;
2223 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2224 
2225 	PF_STATE_LOCK_ASSERT(st);
2226 	PFSYNC_BUCKET_LOCK(b);
2227 
2228 	if (st->state_flags & PFSTATE_NOSYNC) {
2229 		if (st->sync_state != PFSYNC_S_NONE)
2230 			pfsync_q_del(st, true, b);
2231 		PFSYNC_BUCKET_UNLOCK(b);
2232 		return (full);
2233 	}
2234 
2235 	switch (st->sync_state) {
2236 	case PFSYNC_S_UPD_C:
2237 	case PFSYNC_S_IACK:
2238 		pfsync_q_del(st, false, b);
2239 		ref = false;
2240 		/* FALLTHROUGH */
2241 
2242 	case PFSYNC_S_NONE:
2243 		pfsync_q_ins(st, PFSYNC_S_UPD, ref);
2244 		pfsync_push(b);
2245 		break;
2246 
2247 	case PFSYNC_S_INS:
2248 	case PFSYNC_S_UPD:
2249 	case PFSYNC_S_DEL_C:
2250 		/* we're already handling it */
2251 		break;
2252 
2253 	default:
2254 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2255 	}
2256 
2257 	if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(union pfsync_state_union))
2258 		full = true;
2259 
2260 	PFSYNC_BUCKET_UNLOCK(b);
2261 
2262 	return (full);
2263 }
2264 
2265 static void
2266 pfsync_delete_state(struct pf_kstate *st)
2267 {
2268 	struct pfsync_softc *sc = V_pfsyncif;
2269 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2270 	bool ref = true;
2271 
2272 	PFSYNC_BUCKET_LOCK(b);
2273 	if (st->state_flags & PFSTATE_ACK)
2274 		pfsync_undefer_state_locked(st, 1);
2275 	if (st->state_flags & PFSTATE_NOSYNC) {
2276 		if (st->sync_state != PFSYNC_S_NONE)
2277 			pfsync_q_del(st, true, b);
2278 		PFSYNC_BUCKET_UNLOCK(b);
2279 		return;
2280 	}
2281 
2282 	if (b->b_len == PFSYNC_MINPKT)
2283 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2284 
2285 	switch (st->sync_state) {
2286 	case PFSYNC_S_INS:
2287 		/* We never got to tell the world so just forget about it. */
2288 		pfsync_q_del(st, true, b);
2289 		break;
2290 
2291 	case PFSYNC_S_UPD_C:
2292 	case PFSYNC_S_UPD:
2293 	case PFSYNC_S_IACK:
2294 		pfsync_q_del(st, false, b);
2295 		ref = false;
2296 		/* FALLTHROUGH */
2297 
2298 	case PFSYNC_S_NONE:
2299 		pfsync_q_ins(st, PFSYNC_S_DEL_C, ref);
2300 		break;
2301 
2302 	default:
2303 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2304 	}
2305 
2306 	PFSYNC_BUCKET_UNLOCK(b);
2307 }
2308 
2309 static void
2310 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
2311 {
2312 	struct {
2313 		struct pfsync_subheader subh;
2314 		struct pfsync_clr clr;
2315 	} __packed r;
2316 
2317 	bzero(&r, sizeof(r));
2318 
2319 	r.subh.action = PFSYNC_ACT_CLR;
2320 	r.subh.count = htons(1);
2321 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
2322 
2323 	strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
2324 	r.clr.creatorid = creatorid;
2325 
2326 	pfsync_send_plus(&r, sizeof(r));
2327 }
2328 
2329 static enum pfsync_q_id
2330 pfsync_sstate_to_qid(u_int8_t sync_state)
2331 {
2332 	struct pfsync_softc *sc = V_pfsyncif;
2333 
2334 	switch (sync_state) {
2335 		case PFSYNC_S_INS:
2336 			switch (sc->sc_version) {
2337 				case PFSYNC_MSG_VERSION_1301:
2338 					return PFSYNC_Q_INS_1301;
2339 				case PFSYNC_MSG_VERSION_1400:
2340 					return PFSYNC_Q_INS_1400;
2341 			}
2342 			break;
2343 		case PFSYNC_S_IACK:
2344 			return PFSYNC_Q_IACK;
2345 		case PFSYNC_S_UPD:
2346 			switch (sc->sc_version) {
2347 				case PFSYNC_MSG_VERSION_1301:
2348 					return PFSYNC_Q_UPD_1301;
2349 				case PFSYNC_MSG_VERSION_1400:
2350 					return PFSYNC_Q_UPD_1400;
2351 			}
2352 			break;
2353 		case PFSYNC_S_UPD_C:
2354 			return PFSYNC_Q_UPD_C;
2355 		case PFSYNC_S_DEL_C:
2356 			return PFSYNC_Q_DEL_C;
2357 		default:
2358 			panic("%s: Unsupported st->sync_state 0x%02x",
2359 			__func__, sync_state);
2360 	}
2361 
2362 	panic("%s: Unsupported pfsync_msg_version %d",
2363 	    __func__, sc->sc_version);
2364 }
2365 
2366 static void
2367 pfsync_q_ins(struct pf_kstate *st, int sync_state, bool ref)
2368 {
2369 	enum pfsync_q_id q = pfsync_sstate_to_qid(sync_state);
2370 	struct pfsync_softc *sc = V_pfsyncif;
2371 	size_t nlen = pfsync_qs[q].len;
2372 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2373 
2374 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2375 
2376 	KASSERT(st->sync_state == PFSYNC_S_NONE,
2377 		("%s: st->sync_state %u", __func__, st->sync_state));
2378 	KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2379 	    b->b_len));
2380 
2381 	if (TAILQ_EMPTY(&b->b_qs[q]))
2382 		nlen += sizeof(struct pfsync_subheader);
2383 
2384 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2385 		pfsync_sendout(1, b->b_id);
2386 
2387 		nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2388 	}
2389 
2390 	b->b_len += nlen;
2391 	TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list);
2392 	st->sync_state = pfsync_qid_sstate[q];
2393 	if (ref)
2394 		pf_ref_state(st);
2395 }
2396 
2397 static void
2398 pfsync_q_del(struct pf_kstate *st, bool unref, struct pfsync_bucket *b)
2399 {
2400 	enum pfsync_q_id q;
2401 
2402 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2403 	KASSERT(st->sync_state != PFSYNC_S_NONE,
2404 		("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2405 
2406 	q =  pfsync_sstate_to_qid(st->sync_state);
2407 	b->b_len -= pfsync_qs[q].len;
2408 	TAILQ_REMOVE(&b->b_qs[q], st, sync_list);
2409 	st->sync_state = PFSYNC_S_NONE;
2410 	if (unref)
2411 		pf_release_state(st);
2412 
2413 	if (TAILQ_EMPTY(&b->b_qs[q]))
2414 		b->b_len -= sizeof(struct pfsync_subheader);
2415 }
2416 
2417 static void
2418 pfsync_bulk_start(void)
2419 {
2420 	struct pfsync_softc *sc = V_pfsyncif;
2421 
2422 	if (V_pf_status.debug >= PF_DEBUG_MISC)
2423 		printf("pfsync: received bulk update request\n");
2424 
2425 	PFSYNC_BLOCK(sc);
2426 
2427 	sc->sc_ureq_received = time_uptime;
2428 	sc->sc_bulk_hashid = 0;
2429 	sc->sc_bulk_stateid = 0;
2430 	pfsync_bulk_status(PFSYNC_BUS_START);
2431 	callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2432 	PFSYNC_BUNLOCK(sc);
2433 }
2434 
2435 static void
2436 pfsync_bulk_update(void *arg)
2437 {
2438 	struct pfsync_softc *sc = arg;
2439 	struct pf_kstate *s;
2440 	int i;
2441 
2442 	PFSYNC_BLOCK_ASSERT(sc);
2443 	CURVNET_SET(sc->sc_ifp->if_vnet);
2444 
2445 	/*
2446 	 * Start with last state from previous invocation.
2447 	 * It may had gone, in this case start from the
2448 	 * hash slot.
2449 	 */
2450 	s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2451 
2452 	if (s != NULL)
2453 		i = PF_IDHASH(s);
2454 	else
2455 		i = sc->sc_bulk_hashid;
2456 
2457 	for (; i <= pf_hashmask; i++) {
2458 		struct pf_idhash *ih = &V_pf_idhash[i];
2459 
2460 		if (s != NULL)
2461 			PF_HASHROW_ASSERT(ih);
2462 		else {
2463 			PF_HASHROW_LOCK(ih);
2464 			s = LIST_FIRST(&ih->states);
2465 		}
2466 
2467 		for (; s; s = LIST_NEXT(s, entry)) {
2468 			if (s->sync_state == PFSYNC_S_NONE &&
2469 			    s->timeout < PFTM_MAX &&
2470 			    s->pfsync_time <= sc->sc_ureq_received) {
2471 				if (pfsync_update_state_req(s)) {
2472 					/* We've filled a packet. */
2473 					sc->sc_bulk_hashid = i;
2474 					sc->sc_bulk_stateid = s->id;
2475 					sc->sc_bulk_creatorid = s->creatorid;
2476 					PF_HASHROW_UNLOCK(ih);
2477 					callout_reset(&sc->sc_bulk_tmo, 1,
2478 					    pfsync_bulk_update, sc);
2479 					goto full;
2480 				}
2481 			}
2482 		}
2483 		PF_HASHROW_UNLOCK(ih);
2484 	}
2485 
2486 	/* We're done. */
2487 	pfsync_bulk_status(PFSYNC_BUS_END);
2488 full:
2489 	CURVNET_RESTORE();
2490 }
2491 
2492 static void
2493 pfsync_bulk_status(u_int8_t status)
2494 {
2495 	struct {
2496 		struct pfsync_subheader subh;
2497 		struct pfsync_bus bus;
2498 	} __packed r;
2499 
2500 	struct pfsync_softc *sc = V_pfsyncif;
2501 
2502 	bzero(&r, sizeof(r));
2503 
2504 	r.subh.action = PFSYNC_ACT_BUS;
2505 	r.subh.count = htons(1);
2506 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2507 
2508 	r.bus.creatorid = V_pf_status.hostid;
2509 	r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2510 	r.bus.status = status;
2511 
2512 	pfsync_send_plus(&r, sizeof(r));
2513 }
2514 
2515 static void
2516 pfsync_bulk_fail(void *arg)
2517 {
2518 	struct pfsync_softc *sc = arg;
2519 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2520 
2521 	CURVNET_SET(sc->sc_ifp->if_vnet);
2522 
2523 	PFSYNC_BLOCK_ASSERT(sc);
2524 
2525 	if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2526 		/* Try again */
2527 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2528 		    pfsync_bulk_fail, V_pfsyncif);
2529 		PFSYNC_BUCKET_LOCK(b);
2530 		pfsync_request_update(0, 0);
2531 		PFSYNC_BUCKET_UNLOCK(b);
2532 	} else {
2533 		/* Pretend like the transfer was ok. */
2534 		sc->sc_ureq_sent = 0;
2535 		sc->sc_bulk_tries = 0;
2536 		PFSYNC_LOCK(sc);
2537 		if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2538 			(*carp_demote_adj_p)(-V_pfsync_carp_adj,
2539 			    "pfsync bulk fail");
2540 		sc->sc_flags |= PFSYNCF_OK;
2541 		PFSYNC_UNLOCK(sc);
2542 		if (V_pf_status.debug >= PF_DEBUG_MISC)
2543 			printf("pfsync: failed to receive bulk update\n");
2544 	}
2545 
2546 	CURVNET_RESTORE();
2547 }
2548 
2549 static void
2550 pfsync_send_plus(void *plus, size_t pluslen)
2551 {
2552 	struct pfsync_softc *sc = V_pfsyncif;
2553 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2554 
2555 	PFSYNC_BUCKET_LOCK(b);
2556 
2557 	if (b->b_len + pluslen > sc->sc_ifp->if_mtu)
2558 		pfsync_sendout(1, b->b_id);
2559 
2560 	b->b_plus = plus;
2561 	b->b_len += (b->b_pluslen = pluslen);
2562 
2563 	pfsync_sendout(1, b->b_id);
2564 	PFSYNC_BUCKET_UNLOCK(b);
2565 }
2566 
2567 static void
2568 pfsync_timeout(void *arg)
2569 {
2570 	struct pfsync_bucket *b = arg;
2571 
2572 	CURVNET_SET(b->b_sc->sc_ifp->if_vnet);
2573 	PFSYNC_BUCKET_LOCK(b);
2574 	pfsync_push(b);
2575 	PFSYNC_BUCKET_UNLOCK(b);
2576 	CURVNET_RESTORE();
2577 }
2578 
2579 static void
2580 pfsync_push(struct pfsync_bucket *b)
2581 {
2582 
2583 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2584 
2585 	b->b_flags |= PFSYNCF_BUCKET_PUSH;
2586 	swi_sched(V_pfsync_swi_cookie, 0);
2587 }
2588 
2589 static void
2590 pfsync_push_all(struct pfsync_softc *sc)
2591 {
2592 	int c;
2593 	struct pfsync_bucket *b;
2594 
2595 	for (c = 0; c < pfsync_buckets; c++) {
2596 		b = &sc->sc_buckets[c];
2597 
2598 		PFSYNC_BUCKET_LOCK(b);
2599 		pfsync_push(b);
2600 		PFSYNC_BUCKET_UNLOCK(b);
2601 	}
2602 }
2603 
2604 static void
2605 pfsync_tx(struct pfsync_softc *sc, struct mbuf *m)
2606 {
2607 	struct ip *ip;
2608 	int af, error = 0;
2609 
2610 	ip = mtod(m, struct ip *);
2611 	MPASS(ip->ip_v == IPVERSION || ip->ip_v == (IPV6_VERSION >> 4));
2612 
2613 	af = ip->ip_v == IPVERSION ? AF_INET : AF_INET6;
2614 
2615 	/*
2616 	 * We distinguish between a deferral packet and our
2617 	 * own pfsync packet based on M_SKIP_FIREWALL
2618 	 * flag. This is XXX.
2619 	 */
2620 	switch (af) {
2621 #ifdef INET
2622 	case AF_INET:
2623 		if (m->m_flags & M_SKIP_FIREWALL) {
2624 			error = ip_output(m, NULL, NULL, 0,
2625 			    NULL, NULL);
2626 		} else {
2627 			error = ip_output(m, NULL, NULL,
2628 			    IP_RAWOUTPUT, &sc->sc_imo, NULL);
2629 		}
2630 		break;
2631 #endif
2632 #ifdef INET6
2633 	case AF_INET6:
2634 		if (m->m_flags & M_SKIP_FIREWALL) {
2635 			error = ip6_output(m, NULL, NULL, 0,
2636 			    NULL, NULL, NULL);
2637 		} else {
2638 			error = ip6_output(m, NULL, NULL, 0,
2639 				&sc->sc_im6o, NULL, NULL);
2640 		}
2641 		break;
2642 #endif
2643 	}
2644 
2645 	if (error == 0)
2646 		V_pfsyncstats.pfsyncs_opackets++;
2647 	else
2648 		V_pfsyncstats.pfsyncs_oerrors++;
2649 
2650 }
2651 
2652 static void
2653 pfsyncintr(void *arg)
2654 {
2655 	struct epoch_tracker et;
2656 	struct pfsync_softc *sc = arg;
2657 	struct pfsync_bucket *b;
2658 	struct mbuf *m, *n;
2659 	int c;
2660 
2661 	NET_EPOCH_ENTER(et);
2662 	CURVNET_SET(sc->sc_ifp->if_vnet);
2663 
2664 	for (c = 0; c < pfsync_buckets; c++) {
2665 		b = &sc->sc_buckets[c];
2666 
2667 		PFSYNC_BUCKET_LOCK(b);
2668 		if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) {
2669 			pfsync_sendout(0, b->b_id);
2670 			b->b_flags &= ~PFSYNCF_BUCKET_PUSH;
2671 		}
2672 		_IF_DEQUEUE_ALL(&b->b_snd, m);
2673 		PFSYNC_BUCKET_UNLOCK(b);
2674 
2675 		for (; m != NULL; m = n) {
2676 			n = m->m_nextpkt;
2677 			m->m_nextpkt = NULL;
2678 
2679 			pfsync_tx(sc, m);
2680 		}
2681 	}
2682 	CURVNET_RESTORE();
2683 	NET_EPOCH_EXIT(et);
2684 }
2685 
2686 static int
2687 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp,
2688     struct in_mfilter* imf, struct in6_mfilter* im6f)
2689 {
2690 #ifdef  INET
2691 	struct ip_moptions *imo = &sc->sc_imo;
2692 #endif
2693 #ifdef INET6
2694 	struct ip6_moptions *im6o = &sc->sc_im6o;
2695 	struct sockaddr_in6 *syncpeer_sa6 = NULL;
2696 #endif
2697 
2698 	if (!(ifp->if_flags & IFF_MULTICAST))
2699 		return (EADDRNOTAVAIL);
2700 
2701 	switch (sc->sc_sync_peer.ss_family) {
2702 #ifdef INET
2703 	case AF_INET:
2704 	{
2705 		int error;
2706 
2707 		ip_mfilter_init(&imo->imo_head);
2708 		imo->imo_multicast_vif = -1;
2709 		if ((error = in_joingroup(ifp,
2710 		    &((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr, NULL,
2711 		    &imf->imf_inm)) != 0)
2712 			return (error);
2713 
2714 		ip_mfilter_insert(&imo->imo_head, imf);
2715 		imo->imo_multicast_ifp = ifp;
2716 		imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2717 		imo->imo_multicast_loop = 0;
2718 		break;
2719 	}
2720 #endif
2721 #ifdef INET6
2722 	case AF_INET6:
2723 	{
2724 		int error;
2725 
2726 		syncpeer_sa6 = (struct sockaddr_in6 *)&sc->sc_sync_peer;
2727 		if ((error = in6_setscope(&syncpeer_sa6->sin6_addr, ifp, NULL)))
2728 			return (error);
2729 
2730 		ip6_mfilter_init(&im6o->im6o_head);
2731 		if ((error = in6_joingroup(ifp, &syncpeer_sa6->sin6_addr, NULL,
2732 		    &(im6f->im6f_in6m), 0)) != 0)
2733 			return (error);
2734 
2735 		ip6_mfilter_insert(&im6o->im6o_head, im6f);
2736 		im6o->im6o_multicast_ifp = ifp;
2737 		im6o->im6o_multicast_hlim = PFSYNC_DFLTTL;
2738 		im6o->im6o_multicast_loop = 0;
2739 		break;
2740 	}
2741 #endif
2742 	}
2743 
2744 	return (0);
2745 }
2746 
2747 static void
2748 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2749 {
2750 #ifdef INET
2751 	struct ip_moptions *imo = &sc->sc_imo;
2752 	struct in_mfilter *imf;
2753 
2754 	while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
2755 		ip_mfilter_remove(&imo->imo_head, imf);
2756 		in_leavegroup(imf->imf_inm, NULL);
2757 		ip_mfilter_free(imf);
2758 	}
2759 	imo->imo_multicast_ifp = NULL;
2760 #endif
2761 
2762 #ifdef INET6
2763 	struct ip6_moptions *im6o = &sc->sc_im6o;
2764 	struct in6_mfilter *im6f;
2765 
2766 	while ((im6f = ip6_mfilter_first(&im6o->im6o_head)) != NULL) {
2767 		ip6_mfilter_remove(&im6o->im6o_head, im6f);
2768 		in6_leavegroup(im6f->im6f_in6m, NULL);
2769 		ip6_mfilter_free(im6f);
2770 	}
2771 	im6o->im6o_multicast_ifp = NULL;
2772 #endif
2773 }
2774 
2775 void
2776 pfsync_detach_ifnet(struct ifnet *ifp)
2777 {
2778 	struct pfsync_softc *sc = V_pfsyncif;
2779 
2780 	if (sc == NULL)
2781 		return;
2782 
2783 	PFSYNC_LOCK(sc);
2784 
2785 	if (sc->sc_sync_if == ifp) {
2786 		/* We don't need mutlicast cleanup here, because the interface
2787 		 * is going away. We do need to ensure we don't try to do
2788 		 * cleanup later.
2789 		 */
2790 		ip_mfilter_init(&sc->sc_imo.imo_head);
2791 		sc->sc_imo.imo_multicast_ifp = NULL;
2792 		sc->sc_im6o.im6o_multicast_ifp = NULL;
2793 		sc->sc_sync_if = NULL;
2794 	}
2795 
2796 	PFSYNC_UNLOCK(sc);
2797 }
2798 
2799 static int
2800 pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *pfsyncr, struct pfsync_kstatus *status)
2801 {
2802 	struct sockaddr_storage sa;
2803 	status->maxupdates = pfsyncr->pfsyncr_maxupdates;
2804 	status->flags = pfsyncr->pfsyncr_defer;
2805 
2806 	strlcpy(status->syncdev, pfsyncr->pfsyncr_syncdev, IFNAMSIZ);
2807 
2808 	memset(&sa, 0, sizeof(sa));
2809 	if (pfsyncr->pfsyncr_syncpeer.s_addr != 0) {
2810 		struct sockaddr_in *in = (struct sockaddr_in *)&sa;
2811 		in->sin_family = AF_INET;
2812 		in->sin_len = sizeof(*in);
2813 		in->sin_addr.s_addr = pfsyncr->pfsyncr_syncpeer.s_addr;
2814 	}
2815 	status->syncpeer = sa;
2816 
2817 	return 0;
2818 }
2819 
2820 static int
2821 pfsync_kstatus_to_softc(struct pfsync_kstatus *status, struct pfsync_softc *sc)
2822 {
2823 	struct ifnet *sifp;
2824 	struct in_mfilter *imf = NULL;
2825 	struct in6_mfilter *im6f = NULL;
2826 	int error;
2827 	int c;
2828 
2829 	if ((status->maxupdates < 0) || (status->maxupdates > 255))
2830 		return (EINVAL);
2831 
2832 	if (status->syncdev[0] == '\0')
2833 		sifp = NULL;
2834 	else if ((sifp = ifunit_ref(status->syncdev)) == NULL)
2835 		return (EINVAL);
2836 
2837 	switch (status->syncpeer.ss_family) {
2838 #ifdef INET
2839 	case AF_UNSPEC:
2840 	case AF_INET: {
2841 		struct sockaddr_in *status_sin;
2842 		status_sin = (struct sockaddr_in *)&(status->syncpeer);
2843 		if (sifp != NULL) {
2844 			if (status_sin->sin_addr.s_addr == 0 ||
2845 			    status_sin->sin_addr.s_addr ==
2846 			    htonl(INADDR_PFSYNC_GROUP)) {
2847 				status_sin->sin_family = AF_INET;
2848 				status_sin->sin_len = sizeof(*status_sin);
2849 				status_sin->sin_addr.s_addr =
2850 				    htonl(INADDR_PFSYNC_GROUP);
2851 			}
2852 
2853 			if (IN_MULTICAST(ntohl(status_sin->sin_addr.s_addr))) {
2854 				imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
2855 			}
2856 		}
2857 		break;
2858 	}
2859 #endif
2860 #ifdef INET6
2861 	case AF_INET6: {
2862 		struct sockaddr_in6 *status_sin6;
2863 		status_sin6 = (struct sockaddr_in6*)&(status->syncpeer);
2864 		if (sifp != NULL) {
2865 			if (IN6_IS_ADDR_UNSPECIFIED(&status_sin6->sin6_addr) ||
2866 			    IN6_ARE_ADDR_EQUAL(&status_sin6->sin6_addr,
2867 				&in6addr_linklocal_pfsync_group)) {
2868 				status_sin6->sin6_family = AF_INET6;
2869 				status_sin6->sin6_len = sizeof(*status_sin6);
2870 				status_sin6->sin6_addr =
2871 				    in6addr_linklocal_pfsync_group;
2872 			}
2873 
2874 			if (IN6_IS_ADDR_MULTICAST(&status_sin6->sin6_addr)) {
2875 				im6f = ip6_mfilter_alloc(M_WAITOK, 0, 0);
2876 			}
2877 		}
2878 		break;
2879 	}
2880 #endif
2881 	}
2882 
2883 	PFSYNC_LOCK(sc);
2884 
2885 	switch (status->version) {
2886 		case PFSYNC_MSG_VERSION_UNSPECIFIED:
2887 			sc->sc_version = PFSYNC_MSG_VERSION_DEFAULT;
2888 			break;
2889 		case PFSYNC_MSG_VERSION_1301:
2890 		case PFSYNC_MSG_VERSION_1400:
2891 			sc->sc_version = status->version;
2892 			break;
2893 		default:
2894 			PFSYNC_UNLOCK(sc);
2895 			return (EINVAL);
2896 	}
2897 
2898 	switch (status->syncpeer.ss_family) {
2899 	case AF_INET: {
2900 		struct sockaddr_in *status_sin = (struct sockaddr_in *)&(status->syncpeer);
2901 		struct sockaddr_in *sc_sin = (struct sockaddr_in *)&sc->sc_sync_peer;
2902 		sc_sin->sin_family = AF_INET;
2903 		sc_sin->sin_len = sizeof(*sc_sin);
2904 		if (status_sin->sin_addr.s_addr == 0) {
2905 			sc_sin->sin_addr.s_addr = htonl(INADDR_PFSYNC_GROUP);
2906 		} else {
2907 			sc_sin->sin_addr.s_addr = status_sin->sin_addr.s_addr;
2908 		}
2909 		break;
2910 	}
2911 	case AF_INET6: {
2912 		struct sockaddr_in6 *status_sin = (struct sockaddr_in6 *)&(status->syncpeer);
2913 		struct sockaddr_in6 *sc_sin = (struct sockaddr_in6 *)&sc->sc_sync_peer;
2914 		sc_sin->sin6_family = AF_INET6;
2915 		sc_sin->sin6_len = sizeof(*sc_sin);
2916 		if(IN6_IS_ADDR_UNSPECIFIED(&status_sin->sin6_addr)) {
2917 			sc_sin->sin6_addr = in6addr_linklocal_pfsync_group;
2918 		} else {
2919 			sc_sin->sin6_addr = status_sin->sin6_addr;
2920 		}
2921 		break;
2922 	}
2923 	}
2924 
2925 	sc->sc_maxupdates = status->maxupdates;
2926 	if (status->flags & PFSYNCF_DEFER) {
2927 		sc->sc_flags |= PFSYNCF_DEFER;
2928 		V_pfsync_defer_ptr = pfsync_defer;
2929 	} else {
2930 		sc->sc_flags &= ~PFSYNCF_DEFER;
2931 		V_pfsync_defer_ptr = NULL;
2932 	}
2933 
2934 	if (sifp == NULL) {
2935 		if (sc->sc_sync_if)
2936 			if_rele(sc->sc_sync_if);
2937 		sc->sc_sync_if = NULL;
2938 		pfsync_multicast_cleanup(sc);
2939 		PFSYNC_UNLOCK(sc);
2940 		return (0);
2941 	}
2942 
2943 	for (c = 0; c < pfsync_buckets; c++) {
2944 		PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
2945 		if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT &&
2946 		    (sifp->if_mtu < sc->sc_ifp->if_mtu ||
2947 			(sc->sc_sync_if != NULL &&
2948 			    sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
2949 			sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
2950 			pfsync_sendout(1, c);
2951 		PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
2952 	}
2953 
2954 	pfsync_multicast_cleanup(sc);
2955 
2956 	if (((sc->sc_sync_peer.ss_family == AF_INET) &&
2957 	    IN_MULTICAST(ntohl(((struct sockaddr_in *)
2958 	        &sc->sc_sync_peer)->sin_addr.s_addr))) ||
2959 	    ((sc->sc_sync_peer.ss_family == AF_INET6) &&
2960 	    IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6*)
2961 	        &sc->sc_sync_peer)->sin6_addr))) {
2962 		error = pfsync_multicast_setup(sc, sifp, imf, im6f);
2963 		if (error) {
2964 			if_rele(sifp);
2965 			PFSYNC_UNLOCK(sc);
2966 #ifdef INET
2967 			if (imf != NULL)
2968 				ip_mfilter_free(imf);
2969 #endif
2970 #ifdef INET6
2971 			if (im6f != NULL)
2972 				ip6_mfilter_free(im6f);
2973 #endif
2974 			return (error);
2975 		}
2976 	}
2977 	if (sc->sc_sync_if)
2978 		if_rele(sc->sc_sync_if);
2979 	sc->sc_sync_if = sifp;
2980 
2981 	switch (sc->sc_sync_peer.ss_family) {
2982 #ifdef INET
2983 	case AF_INET: {
2984 		struct ip *ip;
2985 		ip = &sc->sc_template.ipv4;
2986 		bzero(ip, sizeof(*ip));
2987 		ip->ip_v = IPVERSION;
2988 		ip->ip_hl = sizeof(sc->sc_template.ipv4) >> 2;
2989 		ip->ip_tos = IPTOS_LOWDELAY;
2990 		/* len and id are set later. */
2991 		ip->ip_off = htons(IP_DF);
2992 		ip->ip_ttl = PFSYNC_DFLTTL;
2993 		ip->ip_p = IPPROTO_PFSYNC;
2994 		ip->ip_src.s_addr = INADDR_ANY;
2995 		ip->ip_dst = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
2996 		break;
2997 	}
2998 #endif
2999 #ifdef INET6
3000 	case AF_INET6: {
3001 		struct ip6_hdr *ip6;
3002 		ip6 = &sc->sc_template.ipv6;
3003 		bzero(ip6, sizeof(*ip6));
3004 		ip6->ip6_vfc = IPV6_VERSION;
3005 		ip6->ip6_hlim = PFSYNC_DFLTTL;
3006 		ip6->ip6_nxt = IPPROTO_PFSYNC;
3007 		ip6->ip6_dst = ((struct sockaddr_in6 *)&sc->sc_sync_peer)->sin6_addr;
3008 
3009 		struct epoch_tracker et;
3010 		NET_EPOCH_ENTER(et);
3011 		in6_selectsrc_addr(if_getfib(sc->sc_sync_if), &ip6->ip6_dst, 0,
3012 		    sc->sc_sync_if, &ip6->ip6_src, NULL);
3013 		NET_EPOCH_EXIT(et);
3014 		break;
3015 	}
3016 #endif
3017 	}
3018 
3019 	/* Request a full state table update. */
3020 	if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
3021 		(*carp_demote_adj_p)(V_pfsync_carp_adj,
3022 		    "pfsync bulk start");
3023 	sc->sc_flags &= ~PFSYNCF_OK;
3024 	if (V_pf_status.debug >= PF_DEBUG_MISC)
3025 		printf("pfsync: requesting bulk update\n");
3026 	PFSYNC_UNLOCK(sc);
3027 	PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
3028 	pfsync_request_update(0, 0);
3029 	PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
3030 	PFSYNC_BLOCK(sc);
3031 	sc->sc_ureq_sent = time_uptime;
3032 	callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, sc);
3033 	PFSYNC_BUNLOCK(sc);
3034 	return (0);
3035 }
3036 
3037 static void
3038 pfsync_pointers_init(void)
3039 {
3040 
3041 	PF_RULES_WLOCK();
3042 	V_pfsync_state_import_ptr = pfsync_state_import;
3043 	V_pfsync_insert_state_ptr = pfsync_insert_state;
3044 	V_pfsync_update_state_ptr = pfsync_update_state;
3045 	V_pfsync_delete_state_ptr = pfsync_delete_state;
3046 	V_pfsync_clear_states_ptr = pfsync_clear_states;
3047 	V_pfsync_defer_ptr = pfsync_defer;
3048 	PF_RULES_WUNLOCK();
3049 }
3050 
3051 static void
3052 pfsync_pointers_uninit(void)
3053 {
3054 
3055 	PF_RULES_WLOCK();
3056 	V_pfsync_state_import_ptr = NULL;
3057 	V_pfsync_insert_state_ptr = NULL;
3058 	V_pfsync_update_state_ptr = NULL;
3059 	V_pfsync_delete_state_ptr = NULL;
3060 	V_pfsync_clear_states_ptr = NULL;
3061 	V_pfsync_defer_ptr = NULL;
3062 	PF_RULES_WUNLOCK();
3063 }
3064 
3065 static void
3066 vnet_pfsync_init(const void *unused __unused)
3067 {
3068 	int error;
3069 
3070 	V_pfsync_cloner = if_clone_simple(pfsyncname,
3071 	    pfsync_clone_create, pfsync_clone_destroy, 1);
3072 	error = swi_add(&V_pfsync_swi_ie, pfsyncname, pfsyncintr, V_pfsyncif,
3073 	    SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
3074 	if (error) {
3075 		if_clone_detach(V_pfsync_cloner);
3076 		log(LOG_INFO, "swi_add() failed in %s\n", __func__);
3077 	}
3078 
3079 	pfsync_pointers_init();
3080 }
3081 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
3082     vnet_pfsync_init, NULL);
3083 
3084 static void
3085 vnet_pfsync_uninit(const void *unused __unused)
3086 {
3087 	int ret __diagused;
3088 
3089 	pfsync_pointers_uninit();
3090 
3091 	if_clone_detach(V_pfsync_cloner);
3092 	ret = swi_remove(V_pfsync_swi_cookie);
3093 	MPASS(ret == 0);
3094 	ret = intr_event_destroy(V_pfsync_swi_ie);
3095 	MPASS(ret == 0);
3096 }
3097 
3098 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH,
3099     vnet_pfsync_uninit, NULL);
3100 
3101 static int
3102 pfsync_init(void)
3103 {
3104 	int error;
3105 
3106 	pfsync_detach_ifnet_ptr = pfsync_detach_ifnet;
3107 
3108 #ifdef INET
3109 	error = ipproto_register(IPPROTO_PFSYNC, pfsync_input, NULL);
3110 	if (error)
3111 		return (error);
3112 #endif
3113 #ifdef INET6
3114 	error = ip6proto_register(IPPROTO_PFSYNC, pfsync6_input, NULL);
3115 	if (error) {
3116 		ipproto_unregister(IPPROTO_PFSYNC);
3117 		return (error);
3118 	}
3119 #endif
3120 
3121 	return (0);
3122 }
3123 
3124 static void
3125 pfsync_uninit(void)
3126 {
3127 	pfsync_detach_ifnet_ptr = NULL;
3128 
3129 #ifdef INET
3130 	ipproto_unregister(IPPROTO_PFSYNC);
3131 #endif
3132 #ifdef INET6
3133 	ip6proto_unregister(IPPROTO_PFSYNC);
3134 #endif
3135 }
3136 
3137 static int
3138 pfsync_modevent(module_t mod, int type, void *data)
3139 {
3140 	int error = 0;
3141 
3142 	switch (type) {
3143 	case MOD_LOAD:
3144 		error = pfsync_init();
3145 		break;
3146 	case MOD_UNLOAD:
3147 		pfsync_uninit();
3148 		break;
3149 	default:
3150 		error = EINVAL;
3151 		break;
3152 	}
3153 
3154 	return (error);
3155 }
3156 
3157 static moduledata_t pfsync_mod = {
3158 	pfsyncname,
3159 	pfsync_modevent,
3160 	0
3161 };
3162 
3163 #define PFSYNC_MODVER 1
3164 
3165 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
3166 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
3167 MODULE_VERSION(pfsync, PFSYNC_MODVER);
3168 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
3169