xref: /freebsd/sys/netpfil/pf/if_pfsync.c (revision 6580f5c38dd5b01aeeaed16b370f1a12423437f0)
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 	uint8_t			*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 		free(b->b_plus, M_PFSYNC);
478 		b->b_plus = NULL;
479 		b->b_pluslen = 0;
480 
481 		callout_drain(&b->b_tmo);
482 	}
483 
484 	callout_drain(&sc->sc_bulkfail_tmo);
485 	callout_drain(&sc->sc_bulk_tmo);
486 
487 	if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
488 		(*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
489 	bpfdetach(ifp);
490 	if_detach(ifp);
491 
492 	pfsync_drop(sc);
493 
494 	if_free(ifp);
495 	pfsync_multicast_cleanup(sc);
496 	mtx_destroy(&sc->sc_mtx);
497 	mtx_destroy(&sc->sc_bulk_mtx);
498 
499 	free(sc->sc_buckets, M_PFSYNC);
500 	free(sc, M_PFSYNC);
501 
502 	V_pfsyncif = NULL;
503 }
504 
505 static int
506 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
507     struct pf_state_peer *d)
508 {
509 	if (s->scrub.scrub_flag && d->scrub == NULL) {
510 		d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
511 		if (d->scrub == NULL)
512 			return (ENOMEM);
513 	}
514 
515 	return (0);
516 }
517 
518 static int
519 pfsync_state_import(union pfsync_state_union *sp, int flags, int msg_version)
520 {
521 	struct pfsync_softc *sc = V_pfsyncif;
522 #ifndef	__NO_STRICT_ALIGNMENT
523 	struct pfsync_state_key key[2];
524 #endif
525 	struct pfsync_state_key *kw, *ks;
526 	struct pf_kstate	*st = NULL;
527 	struct pf_state_key *skw = NULL, *sks = NULL;
528 	struct pf_krule *r = NULL;
529 	struct pfi_kkif	*kif;
530 	int error;
531 
532 	PF_RULES_RASSERT();
533 
534 	if (sp->pfs_1301.creatorid == 0) {
535 		if (V_pf_status.debug >= PF_DEBUG_MISC)
536 			printf("%s: invalid creator id: %08x\n", __func__,
537 			    ntohl(sp->pfs_1301.creatorid));
538 		return (EINVAL);
539 	}
540 
541 	if ((kif = pfi_kkif_find(sp->pfs_1301.ifname)) == NULL) {
542 		if (V_pf_status.debug >= PF_DEBUG_MISC)
543 			printf("%s: unknown interface: %s\n", __func__,
544 			    sp->pfs_1301.ifname);
545 		if (flags & PFSYNC_SI_IOCTL)
546 			return (EINVAL);
547 		return (0);	/* skip this state */
548 	}
549 
550 	/*
551 	 * If the ruleset checksums match or the state is coming from the ioctl,
552 	 * it's safe to associate the state with the rule of that number.
553 	 */
554 	if (sp->pfs_1301.rule != htonl(-1) && sp->pfs_1301.anchor == htonl(-1) &&
555 	    (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->pfs_1301.rule) <
556 	    pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
557 		r = pf_main_ruleset.rules[
558 		    PF_RULESET_FILTER].active.ptr_array[ntohl(sp->pfs_1301.rule)];
559 	else
560 		r = &V_pf_default_rule;
561 
562 	if ((r->max_states &&
563 	    counter_u64_fetch(r->states_cur) >= r->max_states))
564 		goto cleanup;
565 
566 	/*
567 	 * XXXGL: consider M_WAITOK in ioctl path after.
568 	 */
569 	st = pf_alloc_state(M_NOWAIT);
570 	if (__predict_false(st == NULL))
571 		goto cleanup;
572 
573 	if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
574 		goto cleanup;
575 
576 #ifndef	__NO_STRICT_ALIGNMENT
577 	bcopy(&sp->pfs_1301.key, key, sizeof(struct pfsync_state_key) * 2);
578 	kw = &key[PF_SK_WIRE];
579 	ks = &key[PF_SK_STACK];
580 #else
581 	kw = &sp->pfs_1301.key[PF_SK_WIRE];
582 	ks = &sp->pfs_1301.key[PF_SK_STACK];
583 #endif
584 
585 	if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->pfs_1301.af) ||
586 	    PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->pfs_1301.af) ||
587 	    kw->port[0] != ks->port[0] ||
588 	    kw->port[1] != ks->port[1]) {
589 		sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
590 		if (sks == NULL)
591 			goto cleanup;
592 	} else
593 		sks = skw;
594 
595 	/* allocate memory for scrub info */
596 	if (pfsync_alloc_scrub_memory(&sp->pfs_1301.src, &st->src) ||
597 	    pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst))
598 		goto cleanup;
599 
600 	/* Copy to state key(s). */
601 	skw->addr[0] = kw->addr[0];
602 	skw->addr[1] = kw->addr[1];
603 	skw->port[0] = kw->port[0];
604 	skw->port[1] = kw->port[1];
605 	skw->proto = sp->pfs_1301.proto;
606 	skw->af = sp->pfs_1301.af;
607 	if (sks != skw) {
608 		sks->addr[0] = ks->addr[0];
609 		sks->addr[1] = ks->addr[1];
610 		sks->port[0] = ks->port[0];
611 		sks->port[1] = ks->port[1];
612 		sks->proto = sp->pfs_1301.proto;
613 		sks->af = sp->pfs_1301.af;
614 	}
615 
616 	/* copy to state */
617 	bcopy(&sp->pfs_1301.rt_addr, &st->rt_addr, sizeof(st->rt_addr));
618 	st->creation = (time_uptime - ntohl(sp->pfs_1301.creation)) * 1000;
619 	st->expire = pf_get_uptime();
620 	if (sp->pfs_1301.expire) {
621 		uint32_t timeout;
622 
623 		timeout = r->timeout[sp->pfs_1301.timeout];
624 		if (!timeout)
625 			timeout = V_pf_default_rule.timeout[sp->pfs_1301.timeout];
626 
627 		/* sp->expire may have been adaptively scaled by export. */
628 		st->expire -= (timeout - ntohl(sp->pfs_1301.expire)) * 1000;
629 	}
630 
631 	st->direction = sp->pfs_1301.direction;
632 	st->act.log = sp->pfs_1301.log;
633 	st->timeout = sp->pfs_1301.timeout;
634 
635 	switch (msg_version) {
636 		case PFSYNC_MSG_VERSION_1301:
637 			st->state_flags = sp->pfs_1301.state_flags;
638 			/*
639 			 * In FreeBSD 13 pfsync lacks many attributes. Copy them
640 			 * from the rule if possible. If rule can't be matched
641 			 * clear any set options as we can't recover their
642 			 * parameters.
643 			*/
644 			if (r == &V_pf_default_rule) {
645 				st->state_flags &= ~PFSTATE_SETMASK;
646 			} else {
647 				/*
648 				 * Similar to pf_rule_to_actions(). This code
649 				 * won't set the actions properly if they come
650 				 * from multiple "match" rules as only rule
651 				 * creating the state is send over pfsync.
652 				 */
653 				st->act.qid = r->qid;
654 				st->act.pqid = r->pqid;
655 				st->act.rtableid = r->rtableid;
656 				if (r->scrub_flags & PFSTATE_SETTOS)
657 					st->act.set_tos = r->set_tos;
658 				st->act.min_ttl = r->min_ttl;
659 				st->act.max_mss = r->max_mss;
660 				st->state_flags |= (r->scrub_flags &
661 				    (PFSTATE_NODF|PFSTATE_RANDOMID|
662 				    PFSTATE_SETTOS|PFSTATE_SCRUB_TCP|
663 				    PFSTATE_SETPRIO));
664 				if (r->dnpipe || r->dnrpipe) {
665 					if (r->free_flags & PFRULE_DN_IS_PIPE)
666 						st->state_flags |= PFSTATE_DN_IS_PIPE;
667 					else
668 						st->state_flags &= ~PFSTATE_DN_IS_PIPE;
669 				}
670 				st->act.dnpipe = r->dnpipe;
671 				st->act.dnrpipe = r->dnrpipe;
672 			}
673 			break;
674 		case PFSYNC_MSG_VERSION_1400:
675 			st->state_flags = ntohs(sp->pfs_1400.state_flags);
676 			st->act.qid = ntohs(sp->pfs_1400.qid);
677 			st->act.pqid = ntohs(sp->pfs_1400.pqid);
678 			st->act.dnpipe = ntohs(sp->pfs_1400.dnpipe);
679 			st->act.dnrpipe = ntohs(sp->pfs_1400.dnrpipe);
680 			st->act.rtableid = ntohl(sp->pfs_1400.rtableid);
681 			st->act.min_ttl = sp->pfs_1400.min_ttl;
682 			st->act.set_tos = sp->pfs_1400.set_tos;
683 			st->act.max_mss = ntohs(sp->pfs_1400.max_mss);
684 			st->act.set_prio[0] = sp->pfs_1400.set_prio[0];
685 			st->act.set_prio[1] = sp->pfs_1400.set_prio[1];
686 			st->rt = sp->pfs_1400.rt;
687 			if (st->rt && (st->rt_kif = pfi_kkif_find(sp->pfs_1400.rt_ifname)) == NULL) {
688 				if (V_pf_status.debug >= PF_DEBUG_MISC)
689 					printf("%s: unknown route interface: %s\n",
690 					    __func__, sp->pfs_1400.rt_ifname);
691 				if (flags & PFSYNC_SI_IOCTL)
692 					error = EINVAL;
693 				else
694 					error = 0;
695 				goto cleanup_keys;
696 			}
697 			break;
698 		default:
699 			panic("%s: Unsupported pfsync_msg_version %d",
700 			    __func__, msg_version);
701 	}
702 
703 	st->id = sp->pfs_1301.id;
704 	st->creatorid = sp->pfs_1301.creatorid;
705 	pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src);
706 	pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
707 
708 	st->rule.ptr = r;
709 	st->nat_rule.ptr = NULL;
710 	st->anchor.ptr = NULL;
711 
712 	st->pfsync_time = time_uptime;
713 	st->sync_state = PFSYNC_S_NONE;
714 
715 	if (!(flags & PFSYNC_SI_IOCTL))
716 		st->state_flags |= PFSTATE_NOSYNC;
717 
718 	if ((error = pf_state_insert(kif, kif, skw, sks, st)) != 0)
719 		goto cleanup_state;
720 
721 	/* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
722 	counter_u64_add(r->states_cur, 1);
723 	counter_u64_add(r->states_tot, 1);
724 
725 	if (!(flags & PFSYNC_SI_IOCTL)) {
726 		st->state_flags &= ~PFSTATE_NOSYNC;
727 		if (st->state_flags & PFSTATE_ACK) {
728 			struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
729 			PFSYNC_BUCKET_LOCK(b);
730 			pfsync_q_ins(st, PFSYNC_S_IACK, true);
731 			PFSYNC_BUCKET_UNLOCK(b);
732 
733 			pfsync_push_all(sc);
734 		}
735 	}
736 	st->state_flags &= ~PFSTATE_ACK;
737 	PF_STATE_UNLOCK(st);
738 
739 	return (0);
740 
741 cleanup:
742 	error = ENOMEM;
743 cleanup_keys:
744 	if (skw == sks)
745 		sks = NULL;
746 	uma_zfree(V_pf_state_key_z, skw);
747 	uma_zfree(V_pf_state_key_z, sks);
748 
749 cleanup_state:	/* pf_state_insert() frees the state keys. */
750 	if (st) {
751 		st->timeout = PFTM_UNLINKED; /* appease an assert */
752 		pf_free_state(st);
753 	}
754 	return (error);
755 }
756 
757 #ifdef INET
758 static int
759 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
760 {
761 	struct pfsync_softc *sc = V_pfsyncif;
762 	struct mbuf *m = *mp;
763 	struct ip *ip = mtod(m, struct ip *);
764 	struct pfsync_header *ph;
765 	struct pfsync_subheader subh;
766 
767 	int offset, len, flags = 0;
768 	int rv;
769 	uint16_t count;
770 
771 	PF_RULES_RLOCK_TRACKER;
772 
773 	*mp = NULL;
774 	V_pfsyncstats.pfsyncs_ipackets++;
775 
776 	/* Verify that we have a sync interface configured. */
777 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
778 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
779 		goto done;
780 
781 	/* verify that the packet came in on the right interface */
782 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
783 		V_pfsyncstats.pfsyncs_badif++;
784 		goto done;
785 	}
786 
787 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
788 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
789 	/* verify that the IP TTL is 255. */
790 	if (ip->ip_ttl != PFSYNC_DFLTTL) {
791 		V_pfsyncstats.pfsyncs_badttl++;
792 		goto done;
793 	}
794 
795 	offset = ip->ip_hl << 2;
796 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
797 		V_pfsyncstats.pfsyncs_hdrops++;
798 		goto done;
799 	}
800 
801 	if (offset + sizeof(*ph) > m->m_len) {
802 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
803 			V_pfsyncstats.pfsyncs_hdrops++;
804 			return (IPPROTO_DONE);
805 		}
806 		ip = mtod(m, struct ip *);
807 	}
808 	ph = (struct pfsync_header *)((char *)ip + offset);
809 
810 	/* verify the version */
811 	if (ph->version != PFSYNC_VERSION) {
812 		V_pfsyncstats.pfsyncs_badver++;
813 		goto done;
814 	}
815 
816 	len = ntohs(ph->len) + offset;
817 	if (m->m_pkthdr.len < len) {
818 		V_pfsyncstats.pfsyncs_badlen++;
819 		goto done;
820 	}
821 
822 	/*
823 	 * Trusting pf_chksum during packet processing, as well as seeking
824 	 * in interface name tree, require holding PF_RULES_RLOCK().
825 	 */
826 	PF_RULES_RLOCK();
827 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
828 		flags = PFSYNC_SI_CKSUM;
829 
830 	offset += sizeof(*ph);
831 	while (offset <= len - sizeof(subh)) {
832 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
833 		offset += sizeof(subh);
834 
835 		if (subh.action >= PFSYNC_ACT_MAX) {
836 			V_pfsyncstats.pfsyncs_badact++;
837 			PF_RULES_RUNLOCK();
838 			goto done;
839 		}
840 
841 		count = ntohs(subh.count);
842 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
843 		rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action);
844 		if (rv == -1) {
845 			PF_RULES_RUNLOCK();
846 			return (IPPROTO_DONE);
847 		}
848 
849 		offset += rv;
850 	}
851 	PF_RULES_RUNLOCK();
852 
853 done:
854 	m_freem(m);
855 	return (IPPROTO_DONE);
856 }
857 #endif
858 
859 #ifdef INET6
860 static int
861 pfsync6_input(struct mbuf **mp, int *offp __unused, int proto __unused)
862 {
863 	struct pfsync_softc *sc = V_pfsyncif;
864 	struct mbuf *m = *mp;
865 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
866 	struct pfsync_header *ph;
867 	struct pfsync_subheader subh;
868 
869 	int offset, len, flags = 0;
870 	int rv;
871 	uint16_t count;
872 
873 	PF_RULES_RLOCK_TRACKER;
874 
875 	*mp = NULL;
876 	V_pfsyncstats.pfsyncs_ipackets++;
877 
878 	/* Verify that we have a sync interface configured. */
879 	if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
880 	    (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
881 		goto done;
882 
883 	/* verify that the packet came in on the right interface */
884 	if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
885 		V_pfsyncstats.pfsyncs_badif++;
886 		goto done;
887 	}
888 
889 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
890 	if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
891 	/* verify that the IP TTL is 255. */
892 	if (ip6->ip6_hlim != PFSYNC_DFLTTL) {
893 		V_pfsyncstats.pfsyncs_badttl++;
894 		goto done;
895 	}
896 
897 
898 	offset = sizeof(*ip6);
899 	if (m->m_pkthdr.len < offset + sizeof(*ph)) {
900 		V_pfsyncstats.pfsyncs_hdrops++;
901 		goto done;
902 	}
903 
904 	if (offset + sizeof(*ph) > m->m_len) {
905 		if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
906 			V_pfsyncstats.pfsyncs_hdrops++;
907 			return (IPPROTO_DONE);
908 		}
909 		ip6 = mtod(m, struct ip6_hdr *);
910 	}
911 	ph = (struct pfsync_header *)((char *)ip6 + offset);
912 
913 	/* verify the version */
914 	if (ph->version != PFSYNC_VERSION) {
915 		V_pfsyncstats.pfsyncs_badver++;
916 		goto done;
917 	}
918 
919 	len = ntohs(ph->len) + offset;
920 	if (m->m_pkthdr.len < len) {
921 		V_pfsyncstats.pfsyncs_badlen++;
922 		goto done;
923 	}
924 
925 	/*
926 	 * Trusting pf_chksum during packet processing, as well as seeking
927 	 * in interface name tree, require holding PF_RULES_RLOCK().
928 	 */
929 	PF_RULES_RLOCK();
930 	if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
931 		flags = PFSYNC_SI_CKSUM;
932 
933 	offset += sizeof(*ph);
934 	while (offset <= len - sizeof(subh)) {
935 		m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
936 		offset += sizeof(subh);
937 
938 		if (subh.action >= PFSYNC_ACT_MAX) {
939 			V_pfsyncstats.pfsyncs_badact++;
940 			PF_RULES_RUNLOCK();
941 			goto done;
942 		}
943 
944 		count = ntohs(subh.count);
945 		V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
946 		rv = (*pfsync_acts[subh.action])(m, offset, count, flags, subh.action);
947 		if (rv == -1) {
948 			PF_RULES_RUNLOCK();
949 			return (IPPROTO_DONE);
950 		}
951 
952 		offset += rv;
953 	}
954 	PF_RULES_RUNLOCK();
955 
956 done:
957 	m_freem(m);
958 	return (IPPROTO_DONE);
959 }
960 #endif
961 
962 static int
963 pfsync_in_clr(struct mbuf *m, int offset, int count, int flags, int action)
964 {
965 	struct pfsync_clr *clr;
966 	struct mbuf *mp;
967 	int len = sizeof(*clr) * count;
968 	int i, offp;
969 	u_int32_t creatorid;
970 
971 	mp = m_pulldown(m, offset, len, &offp);
972 	if (mp == NULL) {
973 		V_pfsyncstats.pfsyncs_badlen++;
974 		return (-1);
975 	}
976 	clr = (struct pfsync_clr *)(mp->m_data + offp);
977 
978 	for (i = 0; i < count; i++) {
979 		creatorid = clr[i].creatorid;
980 
981 		if (clr[i].ifname[0] != '\0' &&
982 		    pfi_kkif_find(clr[i].ifname) == NULL)
983 			continue;
984 
985 		for (int i = 0; i <= pf_hashmask; i++) {
986 			struct pf_idhash *ih = &V_pf_idhash[i];
987 			struct pf_kstate *s;
988 relock:
989 			PF_HASHROW_LOCK(ih);
990 			LIST_FOREACH(s, &ih->states, entry) {
991 				if (s->creatorid == creatorid) {
992 					s->state_flags |= PFSTATE_NOSYNC;
993 					pf_unlink_state(s);
994 					goto relock;
995 				}
996 			}
997 			PF_HASHROW_UNLOCK(ih);
998 		}
999 	}
1000 
1001 	return (len);
1002 }
1003 
1004 static int
1005 pfsync_in_ins(struct mbuf *m, int offset, int count, int flags, int action)
1006 {
1007 	struct mbuf *mp;
1008 	union pfsync_state_union *sa, *sp;
1009 	int i, offp, total_len, msg_version, msg_len;
1010 
1011 	switch (action) {
1012 		case PFSYNC_ACT_INS_1301:
1013 			msg_len = sizeof(struct pfsync_state_1301);
1014 			total_len = msg_len * count;
1015 			msg_version = PFSYNC_MSG_VERSION_1301;
1016 			break;
1017 		case PFSYNC_ACT_INS_1400:
1018 			msg_len = sizeof(struct pfsync_state_1400);
1019 			total_len = msg_len * count;
1020 			msg_version = PFSYNC_MSG_VERSION_1400;
1021 			break;
1022 		default:
1023 			V_pfsyncstats.pfsyncs_badact++;
1024 			return (-1);
1025 	}
1026 
1027 	mp = m_pulldown(m, offset, total_len, &offp);
1028 	if (mp == NULL) {
1029 		V_pfsyncstats.pfsyncs_badlen++;
1030 		return (-1);
1031 	}
1032 	sa = (union pfsync_state_union *)(mp->m_data + offp);
1033 
1034 	for (i = 0; i < count; i++) {
1035 		sp = (union pfsync_state_union *)((char *)sa + msg_len * i);
1036 
1037 		/* Check for invalid values. */
1038 		if (sp->pfs_1301.timeout >= PFTM_MAX ||
1039 		    sp->pfs_1301.src.state > PF_TCPS_PROXY_DST ||
1040 		    sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST ||
1041 		    sp->pfs_1301.direction > PF_OUT ||
1042 		    (sp->pfs_1301.af != AF_INET &&
1043 		    sp->pfs_1301.af != AF_INET6)) {
1044 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1045 				printf("%s: invalid value\n", __func__);
1046 			V_pfsyncstats.pfsyncs_badval++;
1047 			continue;
1048 		}
1049 
1050 		if (pfsync_state_import(sp, flags, msg_version) == ENOMEM)
1051 			/* Drop out, but process the rest of the actions. */
1052 			break;
1053 	}
1054 
1055 	return (total_len);
1056 }
1057 
1058 static int
1059 pfsync_in_iack(struct mbuf *m, int offset, int count, int flags, int action)
1060 {
1061 	struct pfsync_ins_ack *ia, *iaa;
1062 	struct pf_kstate *st;
1063 
1064 	struct mbuf *mp;
1065 	int len = count * sizeof(*ia);
1066 	int offp, i;
1067 
1068 	mp = m_pulldown(m, offset, len, &offp);
1069 	if (mp == NULL) {
1070 		V_pfsyncstats.pfsyncs_badlen++;
1071 		return (-1);
1072 	}
1073 	iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
1074 
1075 	for (i = 0; i < count; i++) {
1076 		ia = &iaa[i];
1077 
1078 		st = pf_find_state_byid(ia->id, ia->creatorid);
1079 		if (st == NULL)
1080 			continue;
1081 
1082 		if (st->state_flags & PFSTATE_ACK) {
1083 			pfsync_undefer_state(st, 0);
1084 		}
1085 		PF_STATE_UNLOCK(st);
1086 	}
1087 	/*
1088 	 * XXX this is not yet implemented, but we know the size of the
1089 	 * message so we can skip it.
1090 	 */
1091 
1092 	return (count * sizeof(struct pfsync_ins_ack));
1093 }
1094 
1095 static int
1096 pfsync_upd_tcp(struct pf_kstate *st, struct pfsync_state_peer *src,
1097     struct pfsync_state_peer *dst)
1098 {
1099 	int sync = 0;
1100 
1101 	PF_STATE_LOCK_ASSERT(st);
1102 
1103 	/*
1104 	 * The state should never go backwards except
1105 	 * for syn-proxy states.  Neither should the
1106 	 * sequence window slide backwards.
1107 	 */
1108 	if ((st->src.state > src->state &&
1109 	    (st->src.state < PF_TCPS_PROXY_SRC ||
1110 	    src->state >= PF_TCPS_PROXY_SRC)) ||
1111 
1112 	    (st->src.state == src->state &&
1113 	    SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
1114 		sync++;
1115 	else
1116 		pf_state_peer_ntoh(src, &st->src);
1117 
1118 	if ((st->dst.state > dst->state) ||
1119 
1120 	    (st->dst.state >= TCPS_SYN_SENT &&
1121 	    SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
1122 		sync++;
1123 	else
1124 		pf_state_peer_ntoh(dst, &st->dst);
1125 
1126 	return (sync);
1127 }
1128 
1129 static int
1130 pfsync_in_upd(struct mbuf *m, int offset, int count, int flags, int action)
1131 {
1132 	struct pfsync_softc *sc = V_pfsyncif;
1133 	union pfsync_state_union *sa, *sp;
1134 	struct pf_kstate *st;
1135 	struct mbuf *mp;
1136 	int sync, offp, i, total_len, msg_len, msg_version;
1137 
1138 	switch (action) {
1139 		case PFSYNC_ACT_UPD_1301:
1140 			msg_len = sizeof(struct pfsync_state_1301);
1141 			total_len = msg_len * count;
1142 			msg_version = PFSYNC_MSG_VERSION_1301;
1143 			break;
1144 		case PFSYNC_ACT_UPD_1400:
1145 			msg_len = sizeof(struct pfsync_state_1400);
1146 			total_len = msg_len * count;
1147 			msg_version = PFSYNC_MSG_VERSION_1400;
1148 			break;
1149 		default:
1150 			V_pfsyncstats.pfsyncs_badact++;
1151 			return (-1);
1152 	}
1153 
1154 	mp = m_pulldown(m, offset, total_len, &offp);
1155 	if (mp == NULL) {
1156 		V_pfsyncstats.pfsyncs_badlen++;
1157 		return (-1);
1158 	}
1159 	sa = (union pfsync_state_union *)(mp->m_data + offp);
1160 
1161 	for (i = 0; i < count; i++) {
1162 		sp = (union pfsync_state_union *)((char *)sa + msg_len * i);
1163 
1164 		/* check for invalid values */
1165 		if (sp->pfs_1301.timeout >= PFTM_MAX ||
1166 		    sp->pfs_1301.src.state > PF_TCPS_PROXY_DST ||
1167 		    sp->pfs_1301.dst.state > PF_TCPS_PROXY_DST) {
1168 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1169 				printf("pfsync_input: PFSYNC_ACT_UPD: "
1170 				    "invalid value\n");
1171 			}
1172 			V_pfsyncstats.pfsyncs_badval++;
1173 			continue;
1174 		}
1175 
1176 		st = pf_find_state_byid(sp->pfs_1301.id, sp->pfs_1301.creatorid);
1177 		if (st == NULL) {
1178 			/* insert the update */
1179 			if (pfsync_state_import(sp, flags, msg_version))
1180 				V_pfsyncstats.pfsyncs_badstate++;
1181 			continue;
1182 		}
1183 
1184 		if (st->state_flags & PFSTATE_ACK) {
1185 			pfsync_undefer_state(st, 1);
1186 		}
1187 
1188 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1189 			sync = pfsync_upd_tcp(st, &sp->pfs_1301.src, &sp->pfs_1301.dst);
1190 		else {
1191 			sync = 0;
1192 
1193 			/*
1194 			 * Non-TCP protocol state machine always go
1195 			 * forwards
1196 			 */
1197 			if (st->src.state > sp->pfs_1301.src.state)
1198 				sync++;
1199 			else
1200 				pf_state_peer_ntoh(&sp->pfs_1301.src, &st->src);
1201 			if (st->dst.state > sp->pfs_1301.dst.state)
1202 				sync++;
1203 			else
1204 				pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
1205 		}
1206 		if (sync < 2) {
1207 			pfsync_alloc_scrub_memory(&sp->pfs_1301.dst, &st->dst);
1208 			pf_state_peer_ntoh(&sp->pfs_1301.dst, &st->dst);
1209 			st->expire = pf_get_uptime();
1210 			st->timeout = sp->pfs_1301.timeout;
1211 		}
1212 		st->pfsync_time = time_uptime;
1213 
1214 		if (sync) {
1215 			V_pfsyncstats.pfsyncs_stale++;
1216 
1217 			pfsync_update_state(st);
1218 			PF_STATE_UNLOCK(st);
1219 			pfsync_push_all(sc);
1220 			continue;
1221 		}
1222 		PF_STATE_UNLOCK(st);
1223 	}
1224 
1225 	return (total_len);
1226 }
1227 
1228 static int
1229 pfsync_in_upd_c(struct mbuf *m, int offset, int count, int flags, int action)
1230 {
1231 	struct pfsync_softc *sc = V_pfsyncif;
1232 	struct pfsync_upd_c *ua, *up;
1233 	struct pf_kstate *st;
1234 	int len = count * sizeof(*up);
1235 	int sync;
1236 	struct mbuf *mp;
1237 	int offp, i;
1238 
1239 	mp = m_pulldown(m, offset, len, &offp);
1240 	if (mp == NULL) {
1241 		V_pfsyncstats.pfsyncs_badlen++;
1242 		return (-1);
1243 	}
1244 	ua = (struct pfsync_upd_c *)(mp->m_data + offp);
1245 
1246 	for (i = 0; i < count; i++) {
1247 		up = &ua[i];
1248 
1249 		/* check for invalid values */
1250 		if (up->timeout >= PFTM_MAX ||
1251 		    up->src.state > PF_TCPS_PROXY_DST ||
1252 		    up->dst.state > PF_TCPS_PROXY_DST) {
1253 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1254 				printf("pfsync_input: "
1255 				    "PFSYNC_ACT_UPD_C: "
1256 				    "invalid value\n");
1257 			}
1258 			V_pfsyncstats.pfsyncs_badval++;
1259 			continue;
1260 		}
1261 
1262 		st = pf_find_state_byid(up->id, up->creatorid);
1263 		if (st == NULL) {
1264 			/* We don't have this state. Ask for it. */
1265 			PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1266 			pfsync_request_update(up->creatorid, up->id);
1267 			PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1268 			continue;
1269 		}
1270 
1271 		if (st->state_flags & PFSTATE_ACK) {
1272 			pfsync_undefer_state(st, 1);
1273 		}
1274 
1275 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1276 			sync = pfsync_upd_tcp(st, &up->src, &up->dst);
1277 		else {
1278 			sync = 0;
1279 
1280 			/*
1281 			 * Non-TCP protocol state machine always go
1282 			 * forwards
1283 			 */
1284 			if (st->src.state > up->src.state)
1285 				sync++;
1286 			else
1287 				pf_state_peer_ntoh(&up->src, &st->src);
1288 			if (st->dst.state > up->dst.state)
1289 				sync++;
1290 			else
1291 				pf_state_peer_ntoh(&up->dst, &st->dst);
1292 		}
1293 		if (sync < 2) {
1294 			pfsync_alloc_scrub_memory(&up->dst, &st->dst);
1295 			pf_state_peer_ntoh(&up->dst, &st->dst);
1296 			st->expire = pf_get_uptime();
1297 			st->timeout = up->timeout;
1298 		}
1299 		st->pfsync_time = time_uptime;
1300 
1301 		if (sync) {
1302 			V_pfsyncstats.pfsyncs_stale++;
1303 
1304 			pfsync_update_state(st);
1305 			PF_STATE_UNLOCK(st);
1306 			pfsync_push_all(sc);
1307 			continue;
1308 		}
1309 		PF_STATE_UNLOCK(st);
1310 	}
1311 
1312 	return (len);
1313 }
1314 
1315 static int
1316 pfsync_in_ureq(struct mbuf *m, int offset, int count, int flags, int action)
1317 {
1318 	struct pfsync_upd_req *ur, *ura;
1319 	struct mbuf *mp;
1320 	int len = count * sizeof(*ur);
1321 	int i, offp;
1322 
1323 	struct pf_kstate *st;
1324 
1325 	mp = m_pulldown(m, offset, len, &offp);
1326 	if (mp == NULL) {
1327 		V_pfsyncstats.pfsyncs_badlen++;
1328 		return (-1);
1329 	}
1330 	ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1331 
1332 	for (i = 0; i < count; i++) {
1333 		ur = &ura[i];
1334 
1335 		if (ur->id == 0 && ur->creatorid == 0)
1336 			pfsync_bulk_start();
1337 		else {
1338 			st = pf_find_state_byid(ur->id, ur->creatorid);
1339 			if (st == NULL) {
1340 				V_pfsyncstats.pfsyncs_badstate++;
1341 				continue;
1342 			}
1343 			if (st->state_flags & PFSTATE_NOSYNC) {
1344 				PF_STATE_UNLOCK(st);
1345 				continue;
1346 			}
1347 
1348 			pfsync_update_state_req(st);
1349 			PF_STATE_UNLOCK(st);
1350 		}
1351 	}
1352 
1353 	return (len);
1354 }
1355 
1356 static int
1357 pfsync_in_del_c(struct mbuf *m, int offset, int count, int flags, int action)
1358 {
1359 	struct mbuf *mp;
1360 	struct pfsync_del_c *sa, *sp;
1361 	struct pf_kstate *st;
1362 	int len = count * sizeof(*sp);
1363 	int offp, i;
1364 
1365 	mp = m_pulldown(m, offset, len, &offp);
1366 	if (mp == NULL) {
1367 		V_pfsyncstats.pfsyncs_badlen++;
1368 		return (-1);
1369 	}
1370 	sa = (struct pfsync_del_c *)(mp->m_data + offp);
1371 
1372 	for (i = 0; i < count; i++) {
1373 		sp = &sa[i];
1374 
1375 		st = pf_find_state_byid(sp->id, sp->creatorid);
1376 		if (st == NULL) {
1377 			V_pfsyncstats.pfsyncs_badstate++;
1378 			continue;
1379 		}
1380 
1381 		st->state_flags |= PFSTATE_NOSYNC;
1382 		pf_unlink_state(st);
1383 	}
1384 
1385 	return (len);
1386 }
1387 
1388 static int
1389 pfsync_in_bus(struct mbuf *m, int offset, int count, int flags, int action)
1390 {
1391 	struct pfsync_softc *sc = V_pfsyncif;
1392 	struct pfsync_bus *bus;
1393 	struct mbuf *mp;
1394 	int len = count * sizeof(*bus);
1395 	int offp;
1396 
1397 	PFSYNC_BLOCK(sc);
1398 
1399 	/* If we're not waiting for a bulk update, who cares. */
1400 	if (sc->sc_ureq_sent == 0) {
1401 		PFSYNC_BUNLOCK(sc);
1402 		return (len);
1403 	}
1404 
1405 	mp = m_pulldown(m, offset, len, &offp);
1406 	if (mp == NULL) {
1407 		PFSYNC_BUNLOCK(sc);
1408 		V_pfsyncstats.pfsyncs_badlen++;
1409 		return (-1);
1410 	}
1411 	bus = (struct pfsync_bus *)(mp->m_data + offp);
1412 
1413 	switch (bus->status) {
1414 	case PFSYNC_BUS_START:
1415 		callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1416 		    V_pf_limits[PF_LIMIT_STATES].limit /
1417 		    ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1418 		    sizeof(union pfsync_state_union)),
1419 		    pfsync_bulk_fail, sc);
1420 		if (V_pf_status.debug >= PF_DEBUG_MISC)
1421 			printf("pfsync: received bulk update start\n");
1422 		break;
1423 
1424 	case PFSYNC_BUS_END:
1425 		if (time_uptime - ntohl(bus->endtime) >=
1426 		    sc->sc_ureq_sent) {
1427 			/* that's it, we're happy */
1428 			sc->sc_ureq_sent = 0;
1429 			sc->sc_bulk_tries = 0;
1430 			callout_stop(&sc->sc_bulkfail_tmo);
1431 			if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1432 				(*carp_demote_adj_p)(-V_pfsync_carp_adj,
1433 				    "pfsync bulk done");
1434 			sc->sc_flags |= PFSYNCF_OK;
1435 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1436 				printf("pfsync: received valid "
1437 				    "bulk update end\n");
1438 		} else {
1439 			if (V_pf_status.debug >= PF_DEBUG_MISC)
1440 				printf("pfsync: received invalid "
1441 				    "bulk update end: bad timestamp\n");
1442 		}
1443 		break;
1444 	}
1445 	PFSYNC_BUNLOCK(sc);
1446 
1447 	return (len);
1448 }
1449 
1450 static int
1451 pfsync_in_tdb(struct mbuf *m, int offset, int count, int flags, int action)
1452 {
1453 	int len = count * sizeof(struct pfsync_tdb);
1454 
1455 #if defined(IPSEC)
1456 	struct pfsync_tdb *tp;
1457 	struct mbuf *mp;
1458 	int offp;
1459 	int i;
1460 	int s;
1461 
1462 	mp = m_pulldown(m, offset, len, &offp);
1463 	if (mp == NULL) {
1464 		V_pfsyncstats.pfsyncs_badlen++;
1465 		return (-1);
1466 	}
1467 	tp = (struct pfsync_tdb *)(mp->m_data + offp);
1468 
1469 	for (i = 0; i < count; i++)
1470 		pfsync_update_net_tdb(&tp[i]);
1471 #endif
1472 
1473 	return (len);
1474 }
1475 
1476 #if defined(IPSEC)
1477 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1478 static void
1479 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1480 {
1481 	struct tdb		*tdb;
1482 	int			 s;
1483 
1484 	/* check for invalid values */
1485 	if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1486 	    (pt->dst.sa.sa_family != AF_INET &&
1487 	    pt->dst.sa.sa_family != AF_INET6))
1488 		goto bad;
1489 
1490 	tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1491 	if (tdb) {
1492 		pt->rpl = ntohl(pt->rpl);
1493 		pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1494 
1495 		/* Neither replay nor byte counter should ever decrease. */
1496 		if (pt->rpl < tdb->tdb_rpl ||
1497 		    pt->cur_bytes < tdb->tdb_cur_bytes) {
1498 			goto bad;
1499 		}
1500 
1501 		tdb->tdb_rpl = pt->rpl;
1502 		tdb->tdb_cur_bytes = pt->cur_bytes;
1503 	}
1504 	return;
1505 
1506 bad:
1507 	if (V_pf_status.debug >= PF_DEBUG_MISC)
1508 		printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1509 		    "invalid value\n");
1510 	V_pfsyncstats.pfsyncs_badstate++;
1511 	return;
1512 }
1513 #endif
1514 
1515 static int
1516 pfsync_in_eof(struct mbuf *m, int offset, int count, int flags, int action)
1517 {
1518 	/* check if we are at the right place in the packet */
1519 	if (offset != m->m_pkthdr.len)
1520 		V_pfsyncstats.pfsyncs_badlen++;
1521 
1522 	/* we're done. free and let the caller return */
1523 	m_freem(m);
1524 	return (-1);
1525 }
1526 
1527 static int
1528 pfsync_in_error(struct mbuf *m, int offset, int count, int flags, int action)
1529 {
1530 	V_pfsyncstats.pfsyncs_badact++;
1531 
1532 	m_freem(m);
1533 	return (-1);
1534 }
1535 
1536 static int
1537 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1538 	struct route *rt)
1539 {
1540 	m_freem(m);
1541 	return (0);
1542 }
1543 
1544 /* ARGSUSED */
1545 static int
1546 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1547 {
1548 	struct pfsync_softc *sc = ifp->if_softc;
1549 	struct ifreq *ifr = (struct ifreq *)data;
1550 	struct pfsyncreq pfsyncr;
1551 	size_t nvbuflen;
1552 	int error;
1553 	int c;
1554 
1555 	switch (cmd) {
1556 	case SIOCSIFFLAGS:
1557 		PFSYNC_LOCK(sc);
1558 		if (ifp->if_flags & IFF_UP) {
1559 			ifp->if_drv_flags |= IFF_DRV_RUNNING;
1560 			PFSYNC_UNLOCK(sc);
1561 			pfsync_pointers_init();
1562 		} else {
1563 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1564 			PFSYNC_UNLOCK(sc);
1565 			pfsync_pointers_uninit();
1566 		}
1567 		break;
1568 	case SIOCSIFMTU:
1569 		if (!sc->sc_sync_if ||
1570 		    ifr->ifr_mtu <= PFSYNC_MINPKT ||
1571 		    ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1572 			return (EINVAL);
1573 		if (ifr->ifr_mtu < ifp->if_mtu) {
1574 			for (c = 0; c < pfsync_buckets; c++) {
1575 				PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1576 				if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT)
1577 					pfsync_sendout(1, c);
1578 				PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1579 			}
1580 		}
1581 		ifp->if_mtu = ifr->ifr_mtu;
1582 		break;
1583 	case SIOCGETPFSYNC:
1584 		bzero(&pfsyncr, sizeof(pfsyncr));
1585 		PFSYNC_LOCK(sc);
1586 		if (sc->sc_sync_if) {
1587 			strlcpy(pfsyncr.pfsyncr_syncdev,
1588 			    sc->sc_sync_if->if_xname, IFNAMSIZ);
1589 		}
1590 		pfsyncr.pfsyncr_syncpeer = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
1591 		pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1592 		pfsyncr.pfsyncr_defer = sc->sc_flags;
1593 		PFSYNC_UNLOCK(sc);
1594 		return (copyout(&pfsyncr, ifr_data_get_ptr(ifr),
1595 		    sizeof(pfsyncr)));
1596 
1597 	case SIOCGETPFSYNCNV:
1598 	    {
1599 		nvlist_t *nvl_syncpeer;
1600 		nvlist_t *nvl = nvlist_create(0);
1601 
1602 		if (nvl == NULL)
1603 			return (ENOMEM);
1604 
1605 		if (sc->sc_sync_if)
1606 			nvlist_add_string(nvl, "syncdev", sc->sc_sync_if->if_xname);
1607 		nvlist_add_number(nvl, "maxupdates", sc->sc_maxupdates);
1608 		nvlist_add_number(nvl, "flags", sc->sc_flags);
1609 		nvlist_add_number(nvl, "version", sc->sc_version);
1610 		if ((nvl_syncpeer = pfsync_sockaddr_to_syncpeer_nvlist(&sc->sc_sync_peer)) != NULL)
1611 			nvlist_add_nvlist(nvl, "syncpeer", nvl_syncpeer);
1612 
1613 		void *packed = NULL;
1614 		packed = nvlist_pack(nvl, &nvbuflen);
1615 		if (packed == NULL) {
1616 			free(packed, M_NVLIST);
1617 			nvlist_destroy(nvl);
1618 			return (ENOMEM);
1619 		}
1620 
1621 		if (nvbuflen > ifr->ifr_cap_nv.buf_length) {
1622 			ifr->ifr_cap_nv.length = nvbuflen;
1623 			ifr->ifr_cap_nv.buffer = NULL;
1624 			free(packed, M_NVLIST);
1625 			nvlist_destroy(nvl);
1626 			return (EFBIG);
1627 		}
1628 
1629 		ifr->ifr_cap_nv.length = nvbuflen;
1630 		error = copyout(packed, ifr->ifr_cap_nv.buffer, nvbuflen);
1631 
1632 		nvlist_destroy(nvl);
1633 		nvlist_destroy(nvl_syncpeer);
1634 		free(packed, M_NVLIST);
1635 		break;
1636 	    }
1637 
1638 	case SIOCSETPFSYNC:
1639 	    {
1640 		struct pfsync_kstatus status;
1641 
1642 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1643 			return (error);
1644 		if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr,
1645 		    sizeof(pfsyncr))))
1646 			return (error);
1647 
1648 		memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1649 		pfsync_pfsyncreq_to_kstatus(&pfsyncr, &status);
1650 
1651 		error = pfsync_kstatus_to_softc(&status, sc);
1652 		return (error);
1653 	    }
1654 	case SIOCSETPFSYNCNV:
1655 	    {
1656 		struct pfsync_kstatus status;
1657 		void *data;
1658 		nvlist_t *nvl;
1659 
1660 		if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1661 			return (error);
1662 		if (ifr->ifr_cap_nv.length > IFR_CAP_NV_MAXBUFSIZE)
1663 			return (EINVAL);
1664 
1665 		data = malloc(ifr->ifr_cap_nv.length, M_TEMP, M_WAITOK);
1666 
1667 		if ((error = copyin(ifr->ifr_cap_nv.buffer, data,
1668 		    ifr->ifr_cap_nv.length)) != 0) {
1669 			free(data, M_TEMP);
1670 			return (error);
1671 		}
1672 
1673 		if ((nvl = nvlist_unpack(data, ifr->ifr_cap_nv.length, 0)) == NULL) {
1674 			free(data, M_TEMP);
1675 			return (EINVAL);
1676 		}
1677 
1678 		memset((char *)&status, 0, sizeof(struct pfsync_kstatus));
1679 		pfsync_nvstatus_to_kstatus(nvl, &status);
1680 
1681 		nvlist_destroy(nvl);
1682 		free(data, M_TEMP);
1683 
1684 		error = pfsync_kstatus_to_softc(&status, sc);
1685 		return (error);
1686 	    }
1687 	default:
1688 		return (ENOTTY);
1689 	}
1690 
1691 	return (0);
1692 }
1693 
1694 static void
1695 pfsync_out_state_1301(struct pf_kstate *st, void *buf)
1696 {
1697 	union pfsync_state_union *sp = buf;
1698 
1699 	pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1301);
1700 }
1701 
1702 static void
1703 pfsync_out_state_1400(struct pf_kstate *st, void *buf)
1704 {
1705 	union pfsync_state_union *sp = buf;
1706 
1707 	pfsync_state_export(sp, st, PFSYNC_MSG_VERSION_1400);
1708 }
1709 
1710 static void
1711 pfsync_out_iack(struct pf_kstate *st, void *buf)
1712 {
1713 	struct pfsync_ins_ack *iack = buf;
1714 
1715 	iack->id = st->id;
1716 	iack->creatorid = st->creatorid;
1717 }
1718 
1719 static void
1720 pfsync_out_upd_c(struct pf_kstate *st, void *buf)
1721 {
1722 	struct pfsync_upd_c *up = buf;
1723 
1724 	bzero(up, sizeof(*up));
1725 	up->id = st->id;
1726 	pf_state_peer_hton(&st->src, &up->src);
1727 	pf_state_peer_hton(&st->dst, &up->dst);
1728 	up->creatorid = st->creatorid;
1729 	up->timeout = st->timeout;
1730 }
1731 
1732 static void
1733 pfsync_out_del_c(struct pf_kstate *st, void *buf)
1734 {
1735 	struct pfsync_del_c *dp = buf;
1736 
1737 	dp->id = st->id;
1738 	dp->creatorid = st->creatorid;
1739 	st->state_flags |= PFSTATE_NOSYNC;
1740 }
1741 
1742 static void
1743 pfsync_drop(struct pfsync_softc *sc)
1744 {
1745 	struct pf_kstate *st, *next;
1746 	struct pfsync_upd_req_item *ur;
1747 	struct pfsync_bucket *b;
1748 	int c;
1749 	enum pfsync_q_id q;
1750 
1751 	for (c = 0; c < pfsync_buckets; c++) {
1752 		b = &sc->sc_buckets[c];
1753 		for (q = 0; q < PFSYNC_Q_COUNT; q++) {
1754 			if (TAILQ_EMPTY(&b->b_qs[q]))
1755 				continue;
1756 
1757 			TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) {
1758 				KASSERT(st->sync_state == pfsync_qid_sstate[q],
1759 					("%s: st->sync_state == q",
1760 						__func__));
1761 				st->sync_state = PFSYNC_S_NONE;
1762 				pf_release_state(st);
1763 			}
1764 			TAILQ_INIT(&b->b_qs[q]);
1765 		}
1766 
1767 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1768 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1769 			free(ur, M_PFSYNC);
1770 		}
1771 
1772 		b->b_len = PFSYNC_MINPKT;
1773 		free(b->b_plus, M_PFSYNC);
1774 		b->b_plus = NULL;
1775 		b->b_pluslen = 0;
1776 	}
1777 }
1778 
1779 static void
1780 pfsync_sendout(int schedswi, int c)
1781 {
1782 	struct pfsync_softc *sc = V_pfsyncif;
1783 	struct ifnet *ifp = sc->sc_ifp;
1784 	struct mbuf *m;
1785 	struct pfsync_header *ph;
1786 	struct pfsync_subheader *subh;
1787 	struct pf_kstate *st, *st_next;
1788 	struct pfsync_upd_req_item *ur;
1789 	struct pfsync_bucket *b = &sc->sc_buckets[c];
1790 	size_t len;
1791 	int aflen, offset, count = 0;
1792 	enum pfsync_q_id q;
1793 
1794 	KASSERT(sc != NULL, ("%s: null sc", __func__));
1795 	KASSERT(b->b_len > PFSYNC_MINPKT,
1796 	    ("%s: sc_len %zu", __func__, b->b_len));
1797 	PFSYNC_BUCKET_LOCK_ASSERT(b);
1798 
1799 	if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1800 		pfsync_drop(sc);
1801 		return;
1802 	}
1803 
1804 	m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1805 	if (m == NULL) {
1806 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1807 		V_pfsyncstats.pfsyncs_onomem++;
1808 		return;
1809 	}
1810 	m->m_data += max_linkhdr;
1811 	bzero(m->m_data, b->b_len);
1812 
1813 	len = b->b_len;
1814 
1815 	/* build the ip header */
1816 	switch (sc->sc_sync_peer.ss_family) {
1817 #ifdef INET
1818 	case AF_INET:
1819 	    {
1820 		struct ip *ip;
1821 
1822 		ip = mtod(m, struct ip *);
1823 		bcopy(&sc->sc_template.ipv4, ip, sizeof(*ip));
1824 		aflen = offset = sizeof(*ip);
1825 
1826 		len -= sizeof(union inet_template) - sizeof(struct ip);
1827 		ip->ip_len = htons(len);
1828 		ip_fillid(ip);
1829 		break;
1830 	    }
1831 #endif
1832 #ifdef INET6
1833 	case AF_INET6:
1834 		{
1835 		struct ip6_hdr *ip6;
1836 
1837 		ip6 = mtod(m, struct ip6_hdr *);
1838 		bcopy(&sc->sc_template.ipv6, ip6, sizeof(*ip6));
1839 		aflen = offset = sizeof(*ip6);
1840 
1841 		len -= sizeof(union inet_template) - sizeof(struct ip6_hdr);
1842 		ip6->ip6_plen = htons(len);
1843 		break;
1844 		}
1845 #endif
1846 	default:
1847 		m_freem(m);
1848 		return;
1849 	}
1850 	m->m_len = m->m_pkthdr.len = len;
1851 
1852 	/* build the pfsync header */
1853 	ph = (struct pfsync_header *)(m->m_data + offset);
1854 	offset += sizeof(*ph);
1855 
1856 	ph->version = PFSYNC_VERSION;
1857 	ph->len = htons(len - aflen);
1858 	bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1859 
1860 	/* walk the queues */
1861 	for (q = 0; q < PFSYNC_Q_COUNT; q++) {
1862 		if (TAILQ_EMPTY(&b->b_qs[q]))
1863 			continue;
1864 
1865 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1866 		offset += sizeof(*subh);
1867 
1868 		count = 0;
1869 		TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) {
1870 			KASSERT(st->sync_state == pfsync_qid_sstate[q],
1871 				("%s: st->sync_state == q",
1872 					__func__));
1873 			/*
1874 			 * XXXGL: some of write methods do unlocked reads
1875 			 * of state data :(
1876 			 */
1877 			pfsync_qs[q].write(st, m->m_data + offset);
1878 			offset += pfsync_qs[q].len;
1879 			st->sync_state = PFSYNC_S_NONE;
1880 			pf_release_state(st);
1881 			count++;
1882 		}
1883 		TAILQ_INIT(&b->b_qs[q]);
1884 
1885 		subh->action = pfsync_qs[q].action;
1886 		subh->count = htons(count);
1887 		V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1888 	}
1889 
1890 	if (!TAILQ_EMPTY(&b->b_upd_req_list)) {
1891 		subh = (struct pfsync_subheader *)(m->m_data + offset);
1892 		offset += sizeof(*subh);
1893 
1894 		count = 0;
1895 		while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1896 			TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1897 
1898 			bcopy(&ur->ur_msg, m->m_data + offset,
1899 			    sizeof(ur->ur_msg));
1900 			offset += sizeof(ur->ur_msg);
1901 			free(ur, M_PFSYNC);
1902 			count++;
1903 		}
1904 
1905 		subh->action = PFSYNC_ACT_UPD_REQ;
1906 		subh->count = htons(count);
1907 		V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1908 	}
1909 
1910 	/* has someone built a custom region for us to add? */
1911 	if (b->b_plus != NULL) {
1912 		bcopy(b->b_plus, m->m_data + offset, b->b_pluslen);
1913 		offset += b->b_pluslen;
1914 
1915 		free(b->b_plus, M_PFSYNC);
1916 		b->b_plus = NULL;
1917 		b->b_pluslen = 0;
1918 	}
1919 
1920 	subh = (struct pfsync_subheader *)(m->m_data + offset);
1921 	offset += sizeof(*subh);
1922 
1923 	subh->action = PFSYNC_ACT_EOF;
1924 	subh->count = htons(1);
1925 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1926 
1927 	/* we're done, let's put it on the wire */
1928 	if (ifp->if_bpf) {
1929 		m->m_data += aflen;
1930 		m->m_len = m->m_pkthdr.len = len - aflen;
1931 		BPF_MTAP(ifp, m);
1932 		m->m_data -= aflen;
1933 		m->m_len = m->m_pkthdr.len = len;
1934 	}
1935 
1936 	if (sc->sc_sync_if == NULL) {
1937 		b->b_len = PFSYNC_MINPKT;
1938 		m_freem(m);
1939 		return;
1940 	}
1941 
1942 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1943 	if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1944 	b->b_len = PFSYNC_MINPKT;
1945 
1946 	if (!_IF_QFULL(&b->b_snd))
1947 		_IF_ENQUEUE(&b->b_snd, m);
1948 	else {
1949 		m_freem(m);
1950 		if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1951 	}
1952 	if (schedswi)
1953 		swi_sched(V_pfsync_swi_cookie, 0);
1954 }
1955 
1956 static void
1957 pfsync_insert_state(struct pf_kstate *st)
1958 {
1959 	struct pfsync_softc *sc = V_pfsyncif;
1960 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1961 
1962 	if (st->state_flags & PFSTATE_NOSYNC)
1963 		return;
1964 
1965 	if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1966 	    st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1967 		st->state_flags |= PFSTATE_NOSYNC;
1968 		return;
1969 	}
1970 
1971 	KASSERT(st->sync_state == PFSYNC_S_NONE,
1972 		("%s: st->sync_state %u", __func__, st->sync_state));
1973 
1974 	PFSYNC_BUCKET_LOCK(b);
1975 	if (b->b_len == PFSYNC_MINPKT)
1976 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1977 
1978 	pfsync_q_ins(st, PFSYNC_S_INS, true);
1979 	PFSYNC_BUCKET_UNLOCK(b);
1980 
1981 	st->sync_updates = 0;
1982 }
1983 
1984 static int
1985 pfsync_defer(struct pf_kstate *st, struct mbuf *m)
1986 {
1987 	struct pfsync_softc *sc = V_pfsyncif;
1988 	struct pfsync_deferral *pd;
1989 	struct pfsync_bucket *b;
1990 
1991 	if (m->m_flags & (M_BCAST|M_MCAST))
1992 		return (0);
1993 
1994 	if (sc == NULL)
1995 		return (0);
1996 
1997 	b = pfsync_get_bucket(sc, st);
1998 
1999 	PFSYNC_LOCK(sc);
2000 
2001 	if (!(sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) ||
2002 	    !(sc->sc_flags & PFSYNCF_DEFER)) {
2003 		PFSYNC_UNLOCK(sc);
2004 		return (0);
2005 	}
2006 
2007 	PFSYNC_BUCKET_LOCK(b);
2008 	PFSYNC_UNLOCK(sc);
2009 
2010 	if (b->b_deferred >= 128)
2011 		pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0);
2012 
2013 	pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
2014 	if (pd == NULL) {
2015 		PFSYNC_BUCKET_UNLOCK(b);
2016 		return (0);
2017 	}
2018 	b->b_deferred++;
2019 
2020 	m->m_flags |= M_SKIP_FIREWALL;
2021 	st->state_flags |= PFSTATE_ACK;
2022 
2023 	pd->pd_sc = sc;
2024 	pd->pd_st = st;
2025 	pf_ref_state(st);
2026 	pd->pd_m = m;
2027 
2028 	TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry);
2029 	callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED);
2030 	callout_reset(&pd->pd_tmo, (V_pfsync_defer_timeout * hz) / 1000,
2031 	    pfsync_defer_tmo, pd);
2032 
2033 	pfsync_push(b);
2034 	PFSYNC_BUCKET_UNLOCK(b);
2035 
2036 	return (1);
2037 }
2038 
2039 static void
2040 pfsync_undefer(struct pfsync_deferral *pd, int drop)
2041 {
2042 	struct pfsync_softc *sc = pd->pd_sc;
2043 	struct mbuf *m = pd->pd_m;
2044 	struct pf_kstate *st = pd->pd_st;
2045 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2046 
2047 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2048 
2049 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
2050 	b->b_deferred--;
2051 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
2052 	free(pd, M_PFSYNC);
2053 	pf_release_state(st);
2054 
2055 	if (drop)
2056 		m_freem(m);
2057 	else {
2058 		_IF_ENQUEUE(&b->b_snd, m);
2059 		pfsync_push(b);
2060 	}
2061 }
2062 
2063 static void
2064 pfsync_defer_tmo(void *arg)
2065 {
2066 	struct epoch_tracker et;
2067 	struct pfsync_deferral *pd = arg;
2068 	struct pfsync_softc *sc = pd->pd_sc;
2069 	struct mbuf *m = pd->pd_m;
2070 	struct pf_kstate *st = pd->pd_st;
2071 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2072 
2073 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2074 
2075 	TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
2076 	b->b_deferred--;
2077 	pd->pd_st->state_flags &= ~PFSTATE_ACK;	/* XXX: locking! */
2078 	PFSYNC_BUCKET_UNLOCK(b);
2079 	free(pd, M_PFSYNC);
2080 
2081 	if (sc->sc_sync_if == NULL) {
2082 		pf_release_state(st);
2083 		m_freem(m);
2084 		return;
2085 	}
2086 
2087 	NET_EPOCH_ENTER(et);
2088 	CURVNET_SET(sc->sc_sync_if->if_vnet);
2089 
2090 	pfsync_tx(sc, m);
2091 
2092 	pf_release_state(st);
2093 
2094 	CURVNET_RESTORE();
2095 	NET_EPOCH_EXIT(et);
2096 }
2097 
2098 static void
2099 pfsync_undefer_state_locked(struct pf_kstate *st, int drop)
2100 {
2101 	struct pfsync_softc *sc = V_pfsyncif;
2102 	struct pfsync_deferral *pd;
2103 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2104 
2105 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2106 
2107 	TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) {
2108 		 if (pd->pd_st == st) {
2109 			if (callout_stop(&pd->pd_tmo) > 0)
2110 				pfsync_undefer(pd, drop);
2111 
2112 			return;
2113 		}
2114 	}
2115 
2116 	panic("%s: unable to find deferred state", __func__);
2117 }
2118 
2119 static void
2120 pfsync_undefer_state(struct pf_kstate *st, int drop)
2121 {
2122 	struct pfsync_softc *sc = V_pfsyncif;
2123 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2124 
2125 	PFSYNC_BUCKET_LOCK(b);
2126 	pfsync_undefer_state_locked(st, drop);
2127 	PFSYNC_BUCKET_UNLOCK(b);
2128 }
2129 
2130 static struct pfsync_bucket*
2131 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_kstate *st)
2132 {
2133 	int c = PF_IDHASH(st) % pfsync_buckets;
2134 	return &sc->sc_buckets[c];
2135 }
2136 
2137 static void
2138 pfsync_update_state(struct pf_kstate *st)
2139 {
2140 	struct pfsync_softc *sc = V_pfsyncif;
2141 	bool sync = false, ref = true;
2142 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2143 
2144 	PF_STATE_LOCK_ASSERT(st);
2145 	PFSYNC_BUCKET_LOCK(b);
2146 
2147 	if (st->state_flags & PFSTATE_ACK)
2148 		pfsync_undefer_state_locked(st, 0);
2149 	if (st->state_flags & PFSTATE_NOSYNC) {
2150 		if (st->sync_state != PFSYNC_S_NONE)
2151 			pfsync_q_del(st, true, b);
2152 		PFSYNC_BUCKET_UNLOCK(b);
2153 		return;
2154 	}
2155 
2156 	if (b->b_len == PFSYNC_MINPKT)
2157 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2158 
2159 	switch (st->sync_state) {
2160 	case PFSYNC_S_UPD_C:
2161 	case PFSYNC_S_UPD:
2162 	case PFSYNC_S_INS:
2163 		/* we're already handling it */
2164 
2165 		if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
2166 			st->sync_updates++;
2167 			if (st->sync_updates >= sc->sc_maxupdates)
2168 				sync = true;
2169 		}
2170 		break;
2171 
2172 	case PFSYNC_S_IACK:
2173 		pfsync_q_del(st, false, b);
2174 		ref = false;
2175 		/* FALLTHROUGH */
2176 
2177 	case PFSYNC_S_NONE:
2178 		pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
2179 		st->sync_updates = 0;
2180 		break;
2181 
2182 	default:
2183 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2184 	}
2185 
2186 	if (sync || (time_uptime - st->pfsync_time) < 2)
2187 		pfsync_push(b);
2188 
2189 	PFSYNC_BUCKET_UNLOCK(b);
2190 }
2191 
2192 static void
2193 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
2194 {
2195 	struct pfsync_softc *sc = V_pfsyncif;
2196 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2197 	struct pfsync_upd_req_item *item;
2198 	size_t nlen = sizeof(struct pfsync_upd_req);
2199 
2200 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2201 
2202 	/*
2203 	 * This code does a bit to prevent multiple update requests for the
2204 	 * same state being generated. It searches current subheader queue,
2205 	 * but it doesn't lookup into queue of already packed datagrams.
2206 	 */
2207 	TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry)
2208 		if (item->ur_msg.id == id &&
2209 		    item->ur_msg.creatorid == creatorid)
2210 			return;
2211 
2212 	item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
2213 	if (item == NULL)
2214 		return; /* XXX stats */
2215 
2216 	item->ur_msg.id = id;
2217 	item->ur_msg.creatorid = creatorid;
2218 
2219 	if (TAILQ_EMPTY(&b->b_upd_req_list))
2220 		nlen += sizeof(struct pfsync_subheader);
2221 
2222 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2223 		pfsync_sendout(0, 0);
2224 
2225 		nlen = sizeof(struct pfsync_subheader) +
2226 		    sizeof(struct pfsync_upd_req);
2227 	}
2228 
2229 	TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry);
2230 	b->b_len += nlen;
2231 
2232 	pfsync_push(b);
2233 }
2234 
2235 static bool
2236 pfsync_update_state_req(struct pf_kstate *st)
2237 {
2238 	struct pfsync_softc *sc = V_pfsyncif;
2239 	bool ref = true, full = false;
2240 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2241 
2242 	PF_STATE_LOCK_ASSERT(st);
2243 	PFSYNC_BUCKET_LOCK(b);
2244 
2245 	if (st->state_flags & PFSTATE_NOSYNC) {
2246 		if (st->sync_state != PFSYNC_S_NONE)
2247 			pfsync_q_del(st, true, b);
2248 		PFSYNC_BUCKET_UNLOCK(b);
2249 		return (full);
2250 	}
2251 
2252 	switch (st->sync_state) {
2253 	case PFSYNC_S_UPD_C:
2254 	case PFSYNC_S_IACK:
2255 		pfsync_q_del(st, false, b);
2256 		ref = false;
2257 		/* FALLTHROUGH */
2258 
2259 	case PFSYNC_S_NONE:
2260 		pfsync_q_ins(st, PFSYNC_S_UPD, ref);
2261 		pfsync_push(b);
2262 		break;
2263 
2264 	case PFSYNC_S_INS:
2265 	case PFSYNC_S_UPD:
2266 	case PFSYNC_S_DEL_C:
2267 		/* we're already handling it */
2268 		break;
2269 
2270 	default:
2271 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2272 	}
2273 
2274 	if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(union pfsync_state_union))
2275 		full = true;
2276 
2277 	PFSYNC_BUCKET_UNLOCK(b);
2278 
2279 	return (full);
2280 }
2281 
2282 static void
2283 pfsync_delete_state(struct pf_kstate *st)
2284 {
2285 	struct pfsync_softc *sc = V_pfsyncif;
2286 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2287 	bool ref = true;
2288 
2289 	PFSYNC_BUCKET_LOCK(b);
2290 	if (st->state_flags & PFSTATE_ACK)
2291 		pfsync_undefer_state_locked(st, 1);
2292 	if (st->state_flags & PFSTATE_NOSYNC) {
2293 		if (st->sync_state != PFSYNC_S_NONE)
2294 			pfsync_q_del(st, true, b);
2295 		PFSYNC_BUCKET_UNLOCK(b);
2296 		return;
2297 	}
2298 
2299 	if (b->b_len == PFSYNC_MINPKT)
2300 		callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2301 
2302 	switch (st->sync_state) {
2303 	case PFSYNC_S_INS:
2304 		/* We never got to tell the world so just forget about it. */
2305 		pfsync_q_del(st, true, b);
2306 		break;
2307 
2308 	case PFSYNC_S_UPD_C:
2309 	case PFSYNC_S_UPD:
2310 	case PFSYNC_S_IACK:
2311 		pfsync_q_del(st, false, b);
2312 		ref = false;
2313 		/* FALLTHROUGH */
2314 
2315 	case PFSYNC_S_NONE:
2316 		pfsync_q_ins(st, PFSYNC_S_DEL_C, ref);
2317 		break;
2318 
2319 	default:
2320 		panic("%s: unexpected sync state %d", __func__, st->sync_state);
2321 	}
2322 
2323 	PFSYNC_BUCKET_UNLOCK(b);
2324 }
2325 
2326 static void
2327 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
2328 {
2329 	struct {
2330 		struct pfsync_subheader subh;
2331 		struct pfsync_clr clr;
2332 	} __packed r;
2333 
2334 	bzero(&r, sizeof(r));
2335 
2336 	r.subh.action = PFSYNC_ACT_CLR;
2337 	r.subh.count = htons(1);
2338 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
2339 
2340 	strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
2341 	r.clr.creatorid = creatorid;
2342 
2343 	pfsync_send_plus(&r, sizeof(r));
2344 }
2345 
2346 static enum pfsync_q_id
2347 pfsync_sstate_to_qid(u_int8_t sync_state)
2348 {
2349 	struct pfsync_softc *sc = V_pfsyncif;
2350 
2351 	switch (sync_state) {
2352 		case PFSYNC_S_INS:
2353 			switch (sc->sc_version) {
2354 				case PFSYNC_MSG_VERSION_1301:
2355 					return PFSYNC_Q_INS_1301;
2356 				case PFSYNC_MSG_VERSION_1400:
2357 					return PFSYNC_Q_INS_1400;
2358 			}
2359 			break;
2360 		case PFSYNC_S_IACK:
2361 			return PFSYNC_Q_IACK;
2362 		case PFSYNC_S_UPD:
2363 			switch (sc->sc_version) {
2364 				case PFSYNC_MSG_VERSION_1301:
2365 					return PFSYNC_Q_UPD_1301;
2366 				case PFSYNC_MSG_VERSION_1400:
2367 					return PFSYNC_Q_UPD_1400;
2368 			}
2369 			break;
2370 		case PFSYNC_S_UPD_C:
2371 			return PFSYNC_Q_UPD_C;
2372 		case PFSYNC_S_DEL_C:
2373 			return PFSYNC_Q_DEL_C;
2374 		default:
2375 			panic("%s: Unsupported st->sync_state 0x%02x",
2376 			__func__, sync_state);
2377 	}
2378 
2379 	panic("%s: Unsupported pfsync_msg_version %d",
2380 	    __func__, sc->sc_version);
2381 }
2382 
2383 static void
2384 pfsync_q_ins(struct pf_kstate *st, int sync_state, bool ref)
2385 {
2386 	enum pfsync_q_id q = pfsync_sstate_to_qid(sync_state);
2387 	struct pfsync_softc *sc = V_pfsyncif;
2388 	size_t nlen = pfsync_qs[q].len;
2389 	struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2390 
2391 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2392 
2393 	KASSERT(st->sync_state == PFSYNC_S_NONE,
2394 		("%s: st->sync_state %u", __func__, st->sync_state));
2395 	KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2396 	    b->b_len));
2397 
2398 	if (TAILQ_EMPTY(&b->b_qs[q]))
2399 		nlen += sizeof(struct pfsync_subheader);
2400 
2401 	if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2402 		pfsync_sendout(1, b->b_id);
2403 
2404 		nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2405 	}
2406 
2407 	b->b_len += nlen;
2408 	TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list);
2409 	st->sync_state = pfsync_qid_sstate[q];
2410 	if (ref)
2411 		pf_ref_state(st);
2412 }
2413 
2414 static void
2415 pfsync_q_del(struct pf_kstate *st, bool unref, struct pfsync_bucket *b)
2416 {
2417 	enum pfsync_q_id q;
2418 
2419 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2420 	KASSERT(st->sync_state != PFSYNC_S_NONE,
2421 		("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2422 
2423 	q =  pfsync_sstate_to_qid(st->sync_state);
2424 	b->b_len -= pfsync_qs[q].len;
2425 	TAILQ_REMOVE(&b->b_qs[q], st, sync_list);
2426 	st->sync_state = PFSYNC_S_NONE;
2427 	if (unref)
2428 		pf_release_state(st);
2429 
2430 	if (TAILQ_EMPTY(&b->b_qs[q]))
2431 		b->b_len -= sizeof(struct pfsync_subheader);
2432 }
2433 
2434 static void
2435 pfsync_bulk_start(void)
2436 {
2437 	struct pfsync_softc *sc = V_pfsyncif;
2438 
2439 	if (V_pf_status.debug >= PF_DEBUG_MISC)
2440 		printf("pfsync: received bulk update request\n");
2441 
2442 	PFSYNC_BLOCK(sc);
2443 
2444 	sc->sc_ureq_received = time_uptime;
2445 	sc->sc_bulk_hashid = 0;
2446 	sc->sc_bulk_stateid = 0;
2447 	pfsync_bulk_status(PFSYNC_BUS_START);
2448 	callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2449 	PFSYNC_BUNLOCK(sc);
2450 }
2451 
2452 static void
2453 pfsync_bulk_update(void *arg)
2454 {
2455 	struct pfsync_softc *sc = arg;
2456 	struct pf_kstate *s;
2457 	int i;
2458 
2459 	PFSYNC_BLOCK_ASSERT(sc);
2460 	CURVNET_SET(sc->sc_ifp->if_vnet);
2461 
2462 	/*
2463 	 * Start with last state from previous invocation.
2464 	 * It may had gone, in this case start from the
2465 	 * hash slot.
2466 	 */
2467 	s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2468 
2469 	if (s != NULL)
2470 		i = PF_IDHASH(s);
2471 	else
2472 		i = sc->sc_bulk_hashid;
2473 
2474 	for (; i <= pf_hashmask; i++) {
2475 		struct pf_idhash *ih = &V_pf_idhash[i];
2476 
2477 		if (s != NULL)
2478 			PF_HASHROW_ASSERT(ih);
2479 		else {
2480 			PF_HASHROW_LOCK(ih);
2481 			s = LIST_FIRST(&ih->states);
2482 		}
2483 
2484 		for (; s; s = LIST_NEXT(s, entry)) {
2485 			if (s->sync_state == PFSYNC_S_NONE &&
2486 			    s->timeout < PFTM_MAX &&
2487 			    s->pfsync_time <= sc->sc_ureq_received) {
2488 				if (pfsync_update_state_req(s)) {
2489 					/* We've filled a packet. */
2490 					sc->sc_bulk_hashid = i;
2491 					sc->sc_bulk_stateid = s->id;
2492 					sc->sc_bulk_creatorid = s->creatorid;
2493 					PF_HASHROW_UNLOCK(ih);
2494 					callout_reset(&sc->sc_bulk_tmo, 1,
2495 					    pfsync_bulk_update, sc);
2496 					goto full;
2497 				}
2498 			}
2499 		}
2500 		PF_HASHROW_UNLOCK(ih);
2501 	}
2502 
2503 	/* We're done. */
2504 	pfsync_bulk_status(PFSYNC_BUS_END);
2505 full:
2506 	CURVNET_RESTORE();
2507 }
2508 
2509 static void
2510 pfsync_bulk_status(u_int8_t status)
2511 {
2512 	struct {
2513 		struct pfsync_subheader subh;
2514 		struct pfsync_bus bus;
2515 	} __packed r;
2516 
2517 	struct pfsync_softc *sc = V_pfsyncif;
2518 
2519 	bzero(&r, sizeof(r));
2520 
2521 	r.subh.action = PFSYNC_ACT_BUS;
2522 	r.subh.count = htons(1);
2523 	V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2524 
2525 	r.bus.creatorid = V_pf_status.hostid;
2526 	r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2527 	r.bus.status = status;
2528 
2529 	pfsync_send_plus(&r, sizeof(r));
2530 }
2531 
2532 static void
2533 pfsync_bulk_fail(void *arg)
2534 {
2535 	struct pfsync_softc *sc = arg;
2536 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2537 
2538 	CURVNET_SET(sc->sc_ifp->if_vnet);
2539 
2540 	PFSYNC_BLOCK_ASSERT(sc);
2541 
2542 	if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2543 		/* Try again */
2544 		callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2545 		    pfsync_bulk_fail, V_pfsyncif);
2546 		PFSYNC_BUCKET_LOCK(b);
2547 		pfsync_request_update(0, 0);
2548 		PFSYNC_BUCKET_UNLOCK(b);
2549 	} else {
2550 		/* Pretend like the transfer was ok. */
2551 		sc->sc_ureq_sent = 0;
2552 		sc->sc_bulk_tries = 0;
2553 		PFSYNC_LOCK(sc);
2554 		if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2555 			(*carp_demote_adj_p)(-V_pfsync_carp_adj,
2556 			    "pfsync bulk fail");
2557 		sc->sc_flags |= PFSYNCF_OK;
2558 		PFSYNC_UNLOCK(sc);
2559 		if (V_pf_status.debug >= PF_DEBUG_MISC)
2560 			printf("pfsync: failed to receive bulk update\n");
2561 	}
2562 
2563 	CURVNET_RESTORE();
2564 }
2565 
2566 static void
2567 pfsync_send_plus(void *plus, size_t pluslen)
2568 {
2569 	struct pfsync_softc *sc = V_pfsyncif;
2570 	struct pfsync_bucket *b = &sc->sc_buckets[0];
2571 	uint8_t *newplus;
2572 
2573 	PFSYNC_BUCKET_LOCK(b);
2574 
2575 	if (b->b_len + pluslen > sc->sc_ifp->if_mtu)
2576 		pfsync_sendout(1, b->b_id);
2577 
2578 	newplus = malloc(pluslen + b->b_pluslen, M_PFSYNC, M_NOWAIT);
2579 	if (newplus == NULL)
2580 		goto out;
2581 
2582 	if (b->b_plus != NULL) {
2583 		memcpy(newplus, b->b_plus, b->b_pluslen);
2584 		free(b->b_plus, M_PFSYNC);
2585 	} else {
2586 		MPASS(b->b_pluslen == 0);
2587 	}
2588 	memcpy(newplus + b->b_pluslen, plus, pluslen);
2589 
2590 	b->b_plus = newplus;
2591 	b->b_pluslen += pluslen;
2592 	b->b_len += pluslen;
2593 
2594 	pfsync_sendout(1, b->b_id);
2595 
2596 out:
2597 	PFSYNC_BUCKET_UNLOCK(b);
2598 }
2599 
2600 static void
2601 pfsync_timeout(void *arg)
2602 {
2603 	struct pfsync_bucket *b = arg;
2604 
2605 	CURVNET_SET(b->b_sc->sc_ifp->if_vnet);
2606 	PFSYNC_BUCKET_LOCK(b);
2607 	pfsync_push(b);
2608 	PFSYNC_BUCKET_UNLOCK(b);
2609 	CURVNET_RESTORE();
2610 }
2611 
2612 static void
2613 pfsync_push(struct pfsync_bucket *b)
2614 {
2615 
2616 	PFSYNC_BUCKET_LOCK_ASSERT(b);
2617 
2618 	b->b_flags |= PFSYNCF_BUCKET_PUSH;
2619 	swi_sched(V_pfsync_swi_cookie, 0);
2620 }
2621 
2622 static void
2623 pfsync_push_all(struct pfsync_softc *sc)
2624 {
2625 	int c;
2626 	struct pfsync_bucket *b;
2627 
2628 	for (c = 0; c < pfsync_buckets; c++) {
2629 		b = &sc->sc_buckets[c];
2630 
2631 		PFSYNC_BUCKET_LOCK(b);
2632 		pfsync_push(b);
2633 		PFSYNC_BUCKET_UNLOCK(b);
2634 	}
2635 }
2636 
2637 static void
2638 pfsync_tx(struct pfsync_softc *sc, struct mbuf *m)
2639 {
2640 	struct ip *ip;
2641 	int af, error = 0;
2642 
2643 	ip = mtod(m, struct ip *);
2644 	MPASS(ip->ip_v == IPVERSION || ip->ip_v == (IPV6_VERSION >> 4));
2645 
2646 	af = ip->ip_v == IPVERSION ? AF_INET : AF_INET6;
2647 
2648 	/*
2649 	 * We distinguish between a deferral packet and our
2650 	 * own pfsync packet based on M_SKIP_FIREWALL
2651 	 * flag. This is XXX.
2652 	 */
2653 	switch (af) {
2654 #ifdef INET
2655 	case AF_INET:
2656 		if (m->m_flags & M_SKIP_FIREWALL) {
2657 			error = ip_output(m, NULL, NULL, 0,
2658 			    NULL, NULL);
2659 		} else {
2660 			error = ip_output(m, NULL, NULL,
2661 			    IP_RAWOUTPUT, &sc->sc_imo, NULL);
2662 		}
2663 		break;
2664 #endif
2665 #ifdef INET6
2666 	case AF_INET6:
2667 		if (m->m_flags & M_SKIP_FIREWALL) {
2668 			error = ip6_output(m, NULL, NULL, 0,
2669 			    NULL, NULL, NULL);
2670 		} else {
2671 			error = ip6_output(m, NULL, NULL, 0,
2672 				&sc->sc_im6o, NULL, NULL);
2673 		}
2674 		break;
2675 #endif
2676 	}
2677 
2678 	if (error == 0)
2679 		V_pfsyncstats.pfsyncs_opackets++;
2680 	else
2681 		V_pfsyncstats.pfsyncs_oerrors++;
2682 
2683 }
2684 
2685 static void
2686 pfsyncintr(void *arg)
2687 {
2688 	struct epoch_tracker et;
2689 	struct pfsync_softc *sc = arg;
2690 	struct pfsync_bucket *b;
2691 	struct mbuf *m, *n;
2692 	int c;
2693 
2694 	NET_EPOCH_ENTER(et);
2695 	CURVNET_SET(sc->sc_ifp->if_vnet);
2696 
2697 	for (c = 0; c < pfsync_buckets; c++) {
2698 		b = &sc->sc_buckets[c];
2699 
2700 		PFSYNC_BUCKET_LOCK(b);
2701 		if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) {
2702 			pfsync_sendout(0, b->b_id);
2703 			b->b_flags &= ~PFSYNCF_BUCKET_PUSH;
2704 		}
2705 		_IF_DEQUEUE_ALL(&b->b_snd, m);
2706 		PFSYNC_BUCKET_UNLOCK(b);
2707 
2708 		for (; m != NULL; m = n) {
2709 			n = m->m_nextpkt;
2710 			m->m_nextpkt = NULL;
2711 
2712 			pfsync_tx(sc, m);
2713 		}
2714 	}
2715 	CURVNET_RESTORE();
2716 	NET_EPOCH_EXIT(et);
2717 }
2718 
2719 static int
2720 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp,
2721     struct in_mfilter* imf, struct in6_mfilter* im6f)
2722 {
2723 #ifdef  INET
2724 	struct ip_moptions *imo = &sc->sc_imo;
2725 #endif
2726 #ifdef INET6
2727 	struct ip6_moptions *im6o = &sc->sc_im6o;
2728 	struct sockaddr_in6 *syncpeer_sa6 = NULL;
2729 #endif
2730 
2731 	if (!(ifp->if_flags & IFF_MULTICAST))
2732 		return (EADDRNOTAVAIL);
2733 
2734 	switch (sc->sc_sync_peer.ss_family) {
2735 #ifdef INET
2736 	case AF_INET:
2737 	{
2738 		int error;
2739 
2740 		ip_mfilter_init(&imo->imo_head);
2741 		imo->imo_multicast_vif = -1;
2742 		if ((error = in_joingroup(ifp,
2743 		    &((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr, NULL,
2744 		    &imf->imf_inm)) != 0)
2745 			return (error);
2746 
2747 		ip_mfilter_insert(&imo->imo_head, imf);
2748 		imo->imo_multicast_ifp = ifp;
2749 		imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2750 		imo->imo_multicast_loop = 0;
2751 		break;
2752 	}
2753 #endif
2754 #ifdef INET6
2755 	case AF_INET6:
2756 	{
2757 		int error;
2758 
2759 		syncpeer_sa6 = (struct sockaddr_in6 *)&sc->sc_sync_peer;
2760 		if ((error = in6_setscope(&syncpeer_sa6->sin6_addr, ifp, NULL)))
2761 			return (error);
2762 
2763 		ip6_mfilter_init(&im6o->im6o_head);
2764 		if ((error = in6_joingroup(ifp, &syncpeer_sa6->sin6_addr, NULL,
2765 		    &(im6f->im6f_in6m), 0)) != 0)
2766 			return (error);
2767 
2768 		ip6_mfilter_insert(&im6o->im6o_head, im6f);
2769 		im6o->im6o_multicast_ifp = ifp;
2770 		im6o->im6o_multicast_hlim = PFSYNC_DFLTTL;
2771 		im6o->im6o_multicast_loop = 0;
2772 		break;
2773 	}
2774 #endif
2775 	}
2776 
2777 	return (0);
2778 }
2779 
2780 static void
2781 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2782 {
2783 #ifdef INET
2784 	struct ip_moptions *imo = &sc->sc_imo;
2785 	struct in_mfilter *imf;
2786 
2787 	while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
2788 		ip_mfilter_remove(&imo->imo_head, imf);
2789 		in_leavegroup(imf->imf_inm, NULL);
2790 		ip_mfilter_free(imf);
2791 	}
2792 	imo->imo_multicast_ifp = NULL;
2793 #endif
2794 
2795 #ifdef INET6
2796 	struct ip6_moptions *im6o = &sc->sc_im6o;
2797 	struct in6_mfilter *im6f;
2798 
2799 	while ((im6f = ip6_mfilter_first(&im6o->im6o_head)) != NULL) {
2800 		ip6_mfilter_remove(&im6o->im6o_head, im6f);
2801 		in6_leavegroup(im6f->im6f_in6m, NULL);
2802 		ip6_mfilter_free(im6f);
2803 	}
2804 	im6o->im6o_multicast_ifp = NULL;
2805 #endif
2806 }
2807 
2808 void
2809 pfsync_detach_ifnet(struct ifnet *ifp)
2810 {
2811 	struct pfsync_softc *sc = V_pfsyncif;
2812 
2813 	if (sc == NULL)
2814 		return;
2815 
2816 	PFSYNC_LOCK(sc);
2817 
2818 	if (sc->sc_sync_if == ifp) {
2819 		/* We don't need mutlicast cleanup here, because the interface
2820 		 * is going away. We do need to ensure we don't try to do
2821 		 * cleanup later.
2822 		 */
2823 		ip_mfilter_init(&sc->sc_imo.imo_head);
2824 		sc->sc_imo.imo_multicast_ifp = NULL;
2825 		sc->sc_im6o.im6o_multicast_ifp = NULL;
2826 		sc->sc_sync_if = NULL;
2827 	}
2828 
2829 	PFSYNC_UNLOCK(sc);
2830 }
2831 
2832 static int
2833 pfsync_pfsyncreq_to_kstatus(struct pfsyncreq *pfsyncr, struct pfsync_kstatus *status)
2834 {
2835 	struct sockaddr_storage sa;
2836 	status->maxupdates = pfsyncr->pfsyncr_maxupdates;
2837 	status->flags = pfsyncr->pfsyncr_defer;
2838 
2839 	strlcpy(status->syncdev, pfsyncr->pfsyncr_syncdev, IFNAMSIZ);
2840 
2841 	memset(&sa, 0, sizeof(sa));
2842 	if (pfsyncr->pfsyncr_syncpeer.s_addr != 0) {
2843 		struct sockaddr_in *in = (struct sockaddr_in *)&sa;
2844 		in->sin_family = AF_INET;
2845 		in->sin_len = sizeof(*in);
2846 		in->sin_addr.s_addr = pfsyncr->pfsyncr_syncpeer.s_addr;
2847 	}
2848 	status->syncpeer = sa;
2849 
2850 	return 0;
2851 }
2852 
2853 static int
2854 pfsync_kstatus_to_softc(struct pfsync_kstatus *status, struct pfsync_softc *sc)
2855 {
2856 	struct ifnet *sifp;
2857 	struct in_mfilter *imf = NULL;
2858 	struct in6_mfilter *im6f = NULL;
2859 	int error;
2860 	int c;
2861 
2862 	if ((status->maxupdates < 0) || (status->maxupdates > 255))
2863 		return (EINVAL);
2864 
2865 	if (status->syncdev[0] == '\0')
2866 		sifp = NULL;
2867 	else if ((sifp = ifunit_ref(status->syncdev)) == NULL)
2868 		return (EINVAL);
2869 
2870 	switch (status->syncpeer.ss_family) {
2871 #ifdef INET
2872 	case AF_UNSPEC:
2873 	case AF_INET: {
2874 		struct sockaddr_in *status_sin;
2875 		status_sin = (struct sockaddr_in *)&(status->syncpeer);
2876 		if (sifp != NULL) {
2877 			if (status_sin->sin_addr.s_addr == 0 ||
2878 			    status_sin->sin_addr.s_addr ==
2879 			    htonl(INADDR_PFSYNC_GROUP)) {
2880 				status_sin->sin_family = AF_INET;
2881 				status_sin->sin_len = sizeof(*status_sin);
2882 				status_sin->sin_addr.s_addr =
2883 				    htonl(INADDR_PFSYNC_GROUP);
2884 			}
2885 
2886 			if (IN_MULTICAST(ntohl(status_sin->sin_addr.s_addr))) {
2887 				imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
2888 			}
2889 		}
2890 		break;
2891 	}
2892 #endif
2893 #ifdef INET6
2894 	case AF_INET6: {
2895 		struct sockaddr_in6 *status_sin6;
2896 		status_sin6 = (struct sockaddr_in6*)&(status->syncpeer);
2897 		if (sifp != NULL) {
2898 			if (IN6_IS_ADDR_UNSPECIFIED(&status_sin6->sin6_addr) ||
2899 			    IN6_ARE_ADDR_EQUAL(&status_sin6->sin6_addr,
2900 				&in6addr_linklocal_pfsync_group)) {
2901 				status_sin6->sin6_family = AF_INET6;
2902 				status_sin6->sin6_len = sizeof(*status_sin6);
2903 				status_sin6->sin6_addr =
2904 				    in6addr_linklocal_pfsync_group;
2905 			}
2906 
2907 			if (IN6_IS_ADDR_MULTICAST(&status_sin6->sin6_addr)) {
2908 				im6f = ip6_mfilter_alloc(M_WAITOK, 0, 0);
2909 			}
2910 		}
2911 		break;
2912 	}
2913 #endif
2914 	}
2915 
2916 	PFSYNC_LOCK(sc);
2917 
2918 	switch (status->version) {
2919 		case PFSYNC_MSG_VERSION_UNSPECIFIED:
2920 			sc->sc_version = PFSYNC_MSG_VERSION_DEFAULT;
2921 			break;
2922 		case PFSYNC_MSG_VERSION_1301:
2923 		case PFSYNC_MSG_VERSION_1400:
2924 			sc->sc_version = status->version;
2925 			break;
2926 		default:
2927 			PFSYNC_UNLOCK(sc);
2928 			return (EINVAL);
2929 	}
2930 
2931 	switch (status->syncpeer.ss_family) {
2932 	case AF_INET: {
2933 		struct sockaddr_in *status_sin = (struct sockaddr_in *)&(status->syncpeer);
2934 		struct sockaddr_in *sc_sin = (struct sockaddr_in *)&sc->sc_sync_peer;
2935 		sc_sin->sin_family = AF_INET;
2936 		sc_sin->sin_len = sizeof(*sc_sin);
2937 		if (status_sin->sin_addr.s_addr == 0) {
2938 			sc_sin->sin_addr.s_addr = htonl(INADDR_PFSYNC_GROUP);
2939 		} else {
2940 			sc_sin->sin_addr.s_addr = status_sin->sin_addr.s_addr;
2941 		}
2942 		break;
2943 	}
2944 	case AF_INET6: {
2945 		struct sockaddr_in6 *status_sin = (struct sockaddr_in6 *)&(status->syncpeer);
2946 		struct sockaddr_in6 *sc_sin = (struct sockaddr_in6 *)&sc->sc_sync_peer;
2947 		sc_sin->sin6_family = AF_INET6;
2948 		sc_sin->sin6_len = sizeof(*sc_sin);
2949 		if(IN6_IS_ADDR_UNSPECIFIED(&status_sin->sin6_addr)) {
2950 			sc_sin->sin6_addr = in6addr_linklocal_pfsync_group;
2951 		} else {
2952 			sc_sin->sin6_addr = status_sin->sin6_addr;
2953 		}
2954 		break;
2955 	}
2956 	}
2957 
2958 	sc->sc_maxupdates = status->maxupdates;
2959 	if (status->flags & PFSYNCF_DEFER) {
2960 		sc->sc_flags |= PFSYNCF_DEFER;
2961 		V_pfsync_defer_ptr = pfsync_defer;
2962 	} else {
2963 		sc->sc_flags &= ~PFSYNCF_DEFER;
2964 		V_pfsync_defer_ptr = NULL;
2965 	}
2966 
2967 	if (sifp == NULL) {
2968 		if (sc->sc_sync_if)
2969 			if_rele(sc->sc_sync_if);
2970 		sc->sc_sync_if = NULL;
2971 		pfsync_multicast_cleanup(sc);
2972 		PFSYNC_UNLOCK(sc);
2973 		return (0);
2974 	}
2975 
2976 	for (c = 0; c < pfsync_buckets; c++) {
2977 		PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
2978 		if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT &&
2979 		    (sifp->if_mtu < sc->sc_ifp->if_mtu ||
2980 			(sc->sc_sync_if != NULL &&
2981 			    sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
2982 			sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
2983 			pfsync_sendout(1, c);
2984 		PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
2985 	}
2986 
2987 	pfsync_multicast_cleanup(sc);
2988 
2989 	if (((sc->sc_sync_peer.ss_family == AF_INET) &&
2990 	    IN_MULTICAST(ntohl(((struct sockaddr_in *)
2991 	        &sc->sc_sync_peer)->sin_addr.s_addr))) ||
2992 	    ((sc->sc_sync_peer.ss_family == AF_INET6) &&
2993 	    IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6*)
2994 	        &sc->sc_sync_peer)->sin6_addr))) {
2995 		error = pfsync_multicast_setup(sc, sifp, imf, im6f);
2996 		if (error) {
2997 			if_rele(sifp);
2998 			PFSYNC_UNLOCK(sc);
2999 #ifdef INET
3000 			if (imf != NULL)
3001 				ip_mfilter_free(imf);
3002 #endif
3003 #ifdef INET6
3004 			if (im6f != NULL)
3005 				ip6_mfilter_free(im6f);
3006 #endif
3007 			return (error);
3008 		}
3009 	}
3010 	if (sc->sc_sync_if)
3011 		if_rele(sc->sc_sync_if);
3012 	sc->sc_sync_if = sifp;
3013 
3014 	switch (sc->sc_sync_peer.ss_family) {
3015 #ifdef INET
3016 	case AF_INET: {
3017 		struct ip *ip;
3018 		ip = &sc->sc_template.ipv4;
3019 		bzero(ip, sizeof(*ip));
3020 		ip->ip_v = IPVERSION;
3021 		ip->ip_hl = sizeof(sc->sc_template.ipv4) >> 2;
3022 		ip->ip_tos = IPTOS_LOWDELAY;
3023 		/* len and id are set later. */
3024 		ip->ip_off = htons(IP_DF);
3025 		ip->ip_ttl = PFSYNC_DFLTTL;
3026 		ip->ip_p = IPPROTO_PFSYNC;
3027 		ip->ip_src.s_addr = INADDR_ANY;
3028 		ip->ip_dst = ((struct sockaddr_in *)&sc->sc_sync_peer)->sin_addr;
3029 		break;
3030 	}
3031 #endif
3032 #ifdef INET6
3033 	case AF_INET6: {
3034 		struct ip6_hdr *ip6;
3035 		ip6 = &sc->sc_template.ipv6;
3036 		bzero(ip6, sizeof(*ip6));
3037 		ip6->ip6_vfc = IPV6_VERSION;
3038 		ip6->ip6_hlim = PFSYNC_DFLTTL;
3039 		ip6->ip6_nxt = IPPROTO_PFSYNC;
3040 		ip6->ip6_dst = ((struct sockaddr_in6 *)&sc->sc_sync_peer)->sin6_addr;
3041 
3042 		struct epoch_tracker et;
3043 		NET_EPOCH_ENTER(et);
3044 		in6_selectsrc_addr(if_getfib(sc->sc_sync_if), &ip6->ip6_dst, 0,
3045 		    sc->sc_sync_if, &ip6->ip6_src, NULL);
3046 		NET_EPOCH_EXIT(et);
3047 		break;
3048 	}
3049 #endif
3050 	}
3051 
3052 	/* Request a full state table update. */
3053 	if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
3054 		(*carp_demote_adj_p)(V_pfsync_carp_adj,
3055 		    "pfsync bulk start");
3056 	sc->sc_flags &= ~PFSYNCF_OK;
3057 	if (V_pf_status.debug >= PF_DEBUG_MISC)
3058 		printf("pfsync: requesting bulk update\n");
3059 	PFSYNC_UNLOCK(sc);
3060 	PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
3061 	pfsync_request_update(0, 0);
3062 	PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
3063 	PFSYNC_BLOCK(sc);
3064 	sc->sc_ureq_sent = time_uptime;
3065 	callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, sc);
3066 	PFSYNC_BUNLOCK(sc);
3067 	return (0);
3068 }
3069 
3070 static void
3071 pfsync_pointers_init(void)
3072 {
3073 
3074 	PF_RULES_WLOCK();
3075 	V_pfsync_state_import_ptr = pfsync_state_import;
3076 	V_pfsync_insert_state_ptr = pfsync_insert_state;
3077 	V_pfsync_update_state_ptr = pfsync_update_state;
3078 	V_pfsync_delete_state_ptr = pfsync_delete_state;
3079 	V_pfsync_clear_states_ptr = pfsync_clear_states;
3080 	V_pfsync_defer_ptr = pfsync_defer;
3081 	PF_RULES_WUNLOCK();
3082 }
3083 
3084 static void
3085 pfsync_pointers_uninit(void)
3086 {
3087 
3088 	PF_RULES_WLOCK();
3089 	V_pfsync_state_import_ptr = NULL;
3090 	V_pfsync_insert_state_ptr = NULL;
3091 	V_pfsync_update_state_ptr = NULL;
3092 	V_pfsync_delete_state_ptr = NULL;
3093 	V_pfsync_clear_states_ptr = NULL;
3094 	V_pfsync_defer_ptr = NULL;
3095 	PF_RULES_WUNLOCK();
3096 }
3097 
3098 static void
3099 vnet_pfsync_init(const void *unused __unused)
3100 {
3101 	int error;
3102 
3103 	V_pfsync_cloner = if_clone_simple(pfsyncname,
3104 	    pfsync_clone_create, pfsync_clone_destroy, 1);
3105 	error = swi_add(&V_pfsync_swi_ie, pfsyncname, pfsyncintr, V_pfsyncif,
3106 	    SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
3107 	if (error) {
3108 		if_clone_detach(V_pfsync_cloner);
3109 		log(LOG_INFO, "swi_add() failed in %s\n", __func__);
3110 	}
3111 
3112 	pfsync_pointers_init();
3113 }
3114 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
3115     vnet_pfsync_init, NULL);
3116 
3117 static void
3118 vnet_pfsync_uninit(const void *unused __unused)
3119 {
3120 	int ret __diagused;
3121 
3122 	pfsync_pointers_uninit();
3123 
3124 	if_clone_detach(V_pfsync_cloner);
3125 	ret = swi_remove(V_pfsync_swi_cookie);
3126 	MPASS(ret == 0);
3127 	ret = intr_event_destroy(V_pfsync_swi_ie);
3128 	MPASS(ret == 0);
3129 }
3130 
3131 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH,
3132     vnet_pfsync_uninit, NULL);
3133 
3134 static int
3135 pfsync_init(void)
3136 {
3137 	int error;
3138 
3139 	pfsync_detach_ifnet_ptr = pfsync_detach_ifnet;
3140 
3141 #ifdef INET
3142 	error = ipproto_register(IPPROTO_PFSYNC, pfsync_input, NULL);
3143 	if (error)
3144 		return (error);
3145 #endif
3146 #ifdef INET6
3147 	error = ip6proto_register(IPPROTO_PFSYNC, pfsync6_input, NULL);
3148 	if (error) {
3149 		ipproto_unregister(IPPROTO_PFSYNC);
3150 		return (error);
3151 	}
3152 #endif
3153 
3154 	return (0);
3155 }
3156 
3157 static void
3158 pfsync_uninit(void)
3159 {
3160 	pfsync_detach_ifnet_ptr = NULL;
3161 
3162 #ifdef INET
3163 	ipproto_unregister(IPPROTO_PFSYNC);
3164 #endif
3165 #ifdef INET6
3166 	ip6proto_unregister(IPPROTO_PFSYNC);
3167 #endif
3168 }
3169 
3170 static int
3171 pfsync_modevent(module_t mod, int type, void *data)
3172 {
3173 	int error = 0;
3174 
3175 	switch (type) {
3176 	case MOD_LOAD:
3177 		error = pfsync_init();
3178 		break;
3179 	case MOD_UNLOAD:
3180 		pfsync_uninit();
3181 		break;
3182 	default:
3183 		error = EINVAL;
3184 		break;
3185 	}
3186 
3187 	return (error);
3188 }
3189 
3190 static moduledata_t pfsync_mod = {
3191 	pfsyncname,
3192 	pfsync_modevent,
3193 	0
3194 };
3195 
3196 #define PFSYNC_MODVER 1
3197 
3198 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
3199 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
3200 MODULE_VERSION(pfsync, PFSYNC_MODVER);
3201 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
3202