2 * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND ISC)
4 * Copyright (c) 2002 Michael Shalayeff
5 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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.
31 * Copyright (c) 2009 David Gwynne <dlg@openbsd.org>
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.
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.
47 * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $
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
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
62 #include <sys/cdefs.h>
63 __FBSDID("$FreeBSD$");
66 #include "opt_inet6.h"
69 #include <sys/param.h>
71 #include <sys/endian.h>
72 #include <sys/interrupt.h>
73 #include <sys/kernel.h>
76 #include <sys/module.h>
77 #include <sys/mutex.h>
79 #include <sys/protosw.h>
81 #include <sys/socket.h>
82 #include <sys/sockio.h>
83 #include <sys/sysctl.h>
84 #include <sys/syslog.h>
88 #include <net/if_var.h>
89 #include <net/if_clone.h>
90 #include <net/if_types.h>
92 #include <net/pfvar.h>
93 #include <net/if_pfsync.h>
95 #include <netinet/if_ether.h>
96 #include <netinet/in.h>
97 #include <netinet/in_var.h>
98 #include <netinet/ip.h>
99 #include <netinet/ip_carp.h>
100 #include <netinet/ip_var.h>
101 #include <netinet/tcp.h>
102 #include <netinet/tcp_fsm.h>
103 #include <netinet/tcp_seq.h>
105 #define PFSYNC_MINPKT ( \
106 sizeof(struct ip) + \
107 sizeof(struct pfsync_header) + \
108 sizeof(struct pfsync_subheader) )
110 struct pfsync_bucket;
118 static int pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *,
119 struct pfsync_state_peer *);
120 static int pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int);
121 static int pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int);
122 static int pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int);
123 static int pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int);
124 static int pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int);
125 static int pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int);
126 static int pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int);
127 static int pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int);
128 static int pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int);
129 static int pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int);
130 static int pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int);
131 static int pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int);
133 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = {
134 pfsync_in_clr, /* PFSYNC_ACT_CLR */
135 pfsync_in_ins, /* PFSYNC_ACT_INS */
136 pfsync_in_iack, /* PFSYNC_ACT_INS_ACK */
137 pfsync_in_upd, /* PFSYNC_ACT_UPD */
138 pfsync_in_upd_c, /* PFSYNC_ACT_UPD_C */
139 pfsync_in_ureq, /* PFSYNC_ACT_UPD_REQ */
140 pfsync_in_del, /* PFSYNC_ACT_DEL */
141 pfsync_in_del_c, /* PFSYNC_ACT_DEL_C */
142 pfsync_in_error, /* PFSYNC_ACT_INS_F */
143 pfsync_in_error, /* PFSYNC_ACT_DEL_F */
144 pfsync_in_bus, /* PFSYNC_ACT_BUS */
145 pfsync_in_tdb, /* PFSYNC_ACT_TDB */
146 pfsync_in_eof /* PFSYNC_ACT_EOF */
150 void (*write)(struct pf_state *, void *);
155 /* we have one of these for every PFSYNC_S_ */
156 static void pfsync_out_state(struct pf_state *, void *);
157 static void pfsync_out_iack(struct pf_state *, void *);
158 static void pfsync_out_upd_c(struct pf_state *, void *);
159 static void pfsync_out_del(struct pf_state *, void *);
161 static struct pfsync_q pfsync_qs[] = {
162 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_INS },
163 { pfsync_out_iack, sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
164 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_UPD },
165 { pfsync_out_upd_c, sizeof(struct pfsync_upd_c), PFSYNC_ACT_UPD_C },
166 { pfsync_out_del, sizeof(struct pfsync_del_c), PFSYNC_ACT_DEL_C }
169 static void pfsync_q_ins(struct pf_state *, int, bool);
170 static void pfsync_q_del(struct pf_state *, bool, struct pfsync_bucket *);
172 static void pfsync_update_state(struct pf_state *);
174 struct pfsync_upd_req_item {
175 TAILQ_ENTRY(pfsync_upd_req_item) ur_entry;
176 struct pfsync_upd_req ur_msg;
179 struct pfsync_deferral {
180 struct pfsync_softc *pd_sc;
181 TAILQ_ENTRY(pfsync_deferral) pd_entry;
183 struct callout pd_tmo;
185 struct pf_state *pd_st;
194 struct pfsync_softc *b_sc;
196 struct callout b_tmo;
198 #define PFSYNCF_BUCKET_PUSH 0x00000001
201 TAILQ_HEAD(, pf_state) b_qs[PFSYNC_S_COUNT];
202 TAILQ_HEAD(, pfsync_upd_req_item) b_upd_req_list;
203 TAILQ_HEAD(, pfsync_deferral) b_deferrals;
211 struct pfsync_softc {
213 struct ifnet *sc_ifp;
214 struct ifnet *sc_sync_if;
215 struct ip_moptions sc_imo;
216 struct in_addr sc_sync_peer;
218 #define PFSYNCF_OK 0x00000001
219 #define PFSYNCF_DEFER 0x00000002
220 uint8_t sc_maxupdates;
221 struct ip sc_template;
225 struct pfsync_bucket *sc_buckets;
227 /* Bulk update info */
228 struct mtx sc_bulk_mtx;
229 uint32_t sc_ureq_sent;
231 uint32_t sc_ureq_received;
233 uint64_t sc_bulk_stateid;
234 uint32_t sc_bulk_creatorid;
235 struct callout sc_bulk_tmo;
236 struct callout sc_bulkfail_tmo;
239 #define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx)
240 #define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx)
241 #define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED)
243 #define PFSYNC_BUCKET_LOCK(b) mtx_lock(&(b)->b_mtx)
244 #define PFSYNC_BUCKET_UNLOCK(b) mtx_unlock(&(b)->b_mtx)
245 #define PFSYNC_BUCKET_LOCK_ASSERT(b) mtx_assert(&(b)->b_mtx, MA_OWNED)
247 #define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx)
248 #define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx)
249 #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
251 static const char pfsyncname[] = "pfsync";
252 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
253 VNET_DEFINE_STATIC(struct pfsync_softc *, pfsyncif) = NULL;
254 #define V_pfsyncif VNET(pfsyncif)
255 VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL;
256 #define V_pfsync_swi_cookie VNET(pfsync_swi_cookie)
257 VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats);
258 #define V_pfsyncstats VNET(pfsyncstats)
259 VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW;
260 #define V_pfsync_carp_adj VNET(pfsync_carp_adj)
262 static void pfsync_timeout(void *);
263 static void pfsync_push(struct pfsync_bucket *);
264 static void pfsync_push_all(struct pfsync_softc *);
265 static void pfsyncintr(void *);
266 static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
267 struct in_mfilter *imf);
268 static void pfsync_multicast_cleanup(struct pfsync_softc *);
269 static void pfsync_pointers_init(void);
270 static void pfsync_pointers_uninit(void);
271 static int pfsync_init(void);
272 static void pfsync_uninit(void);
274 static unsigned long pfsync_buckets;
276 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
278 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW,
279 &VNET_NAME(pfsyncstats), pfsyncstats,
280 "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
281 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW,
282 &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
283 SYSCTL_ULONG(_net_pfsync, OID_AUTO, pfsync_buckets, CTLFLAG_RDTUN,
284 &pfsync_buckets, 0, "Number of pfsync hash buckets");
286 static int pfsync_clone_create(struct if_clone *, int, caddr_t);
287 static void pfsync_clone_destroy(struct ifnet *);
288 static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
289 struct pf_state_peer *);
290 static int pfsyncoutput(struct ifnet *, struct mbuf *,
291 const struct sockaddr *, struct route *);
292 static int pfsyncioctl(struct ifnet *, u_long, caddr_t);
294 static int pfsync_defer(struct pf_state *, struct mbuf *);
295 static void pfsync_undefer(struct pfsync_deferral *, int);
296 static void pfsync_undefer_state(struct pf_state *, int);
297 static void pfsync_defer_tmo(void *);
299 static void pfsync_request_update(u_int32_t, u_int64_t);
300 static bool pfsync_update_state_req(struct pf_state *);
302 static void pfsync_drop(struct pfsync_softc *);
303 static void pfsync_sendout(int, int);
304 static void pfsync_send_plus(void *, size_t);
306 static void pfsync_bulk_start(void);
307 static void pfsync_bulk_status(u_int8_t);
308 static void pfsync_bulk_update(void *);
309 static void pfsync_bulk_fail(void *);
311 static void pfsync_detach_ifnet(struct ifnet *);
313 static void pfsync_update_net_tdb(struct pfsync_tdb *);
315 static struct pfsync_bucket *pfsync_get_bucket(struct pfsync_softc *,
319 #define PFSYNC_MAX_BULKTRIES 12
321 VNET_DEFINE(struct if_clone *, pfsync_cloner);
322 #define V_pfsync_cloner VNET(pfsync_cloner)
325 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
327 struct pfsync_softc *sc;
329 struct pfsync_bucket *b;
335 if (! pfsync_buckets)
336 pfsync_buckets = mp_ncpus * 2;
338 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
339 sc->sc_flags |= PFSYNCF_OK;
340 sc->sc_maxupdates = 128;
342 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
347 if_initname(ifp, pfsyncname, unit);
349 ifp->if_ioctl = pfsyncioctl;
350 ifp->if_output = pfsyncoutput;
351 ifp->if_type = IFT_PFSYNC;
352 ifp->if_hdrlen = sizeof(struct pfsync_header);
353 ifp->if_mtu = ETHERMTU;
354 mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
355 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
356 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
357 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
361 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
363 sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets),
364 M_PFSYNC, M_ZERO | M_WAITOK);
365 for (c = 0; c < pfsync_buckets; c++) {
366 b = &sc->sc_buckets[c];
367 mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF);
371 b->b_len = PFSYNC_MINPKT;
373 for (q = 0; q < PFSYNC_S_COUNT; q++)
374 TAILQ_INIT(&b->b_qs[q]);
376 TAILQ_INIT(&b->b_upd_req_list);
377 TAILQ_INIT(&b->b_deferrals);
379 callout_init(&b->b_tmo, 1);
381 b->b_snd.ifq_maxlen = ifqmaxlen;
390 pfsync_clone_destroy(struct ifnet *ifp)
392 struct pfsync_softc *sc = ifp->if_softc;
393 struct pfsync_bucket *b;
396 for (c = 0; c < pfsync_buckets; c++) {
397 b = &sc->sc_buckets[c];
399 * At this stage, everything should have already been
400 * cleared by pfsync_uninit(), and we have only to
403 while (b->b_deferred > 0) {
404 struct pfsync_deferral *pd =
405 TAILQ_FIRST(&b->b_deferrals);
407 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
409 if (callout_stop(&pd->pd_tmo) > 0) {
410 pf_release_state(pd->pd_st);
415 callout_drain(&pd->pd_tmo);
420 callout_drain(&b->b_tmo);
423 callout_drain(&sc->sc_bulkfail_tmo);
424 callout_drain(&sc->sc_bulk_tmo);
426 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
427 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
434 pfsync_multicast_cleanup(sc);
435 mtx_destroy(&sc->sc_mtx);
436 mtx_destroy(&sc->sc_bulk_mtx);
438 free(sc->sc_buckets, M_PFSYNC);
445 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
446 struct pf_state_peer *d)
448 if (s->scrub.scrub_flag && d->scrub == NULL) {
449 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
450 if (d->scrub == NULL)
459 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
461 struct pfsync_softc *sc = V_pfsyncif;
462 #ifndef __NO_STRICT_ALIGNMENT
463 struct pfsync_state_key key[2];
465 struct pfsync_state_key *kw, *ks;
466 struct pf_state *st = NULL;
467 struct pf_state_key *skw = NULL, *sks = NULL;
468 struct pf_rule *r = NULL;
474 if (sp->creatorid == 0) {
475 if (V_pf_status.debug >= PF_DEBUG_MISC)
476 printf("%s: invalid creator id: %08x\n", __func__,
477 ntohl(sp->creatorid));
481 if ((kif = pfi_kif_find(sp->ifname)) == NULL) {
482 if (V_pf_status.debug >= PF_DEBUG_MISC)
483 printf("%s: unknown interface: %s\n", __func__,
485 if (flags & PFSYNC_SI_IOCTL)
487 return (0); /* skip this state */
491 * If the ruleset checksums match or the state is coming from the ioctl,
492 * it's safe to associate the state with the rule of that number.
494 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
495 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
496 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
497 r = pf_main_ruleset.rules[
498 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
500 r = &V_pf_default_rule;
502 if ((r->max_states &&
503 counter_u64_fetch(r->states_cur) >= r->max_states))
507 * XXXGL: consider M_WAITOK in ioctl path after.
509 if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL)
512 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
515 #ifndef __NO_STRICT_ALIGNMENT
516 bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2);
517 kw = &key[PF_SK_WIRE];
518 ks = &key[PF_SK_STACK];
520 kw = &sp->key[PF_SK_WIRE];
521 ks = &sp->key[PF_SK_STACK];
524 if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) ||
525 PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) ||
526 kw->port[0] != ks->port[0] ||
527 kw->port[1] != ks->port[1]) {
528 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
534 /* allocate memory for scrub info */
535 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
536 pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
539 /* Copy to state key(s). */
540 skw->addr[0] = kw->addr[0];
541 skw->addr[1] = kw->addr[1];
542 skw->port[0] = kw->port[0];
543 skw->port[1] = kw->port[1];
544 skw->proto = sp->proto;
547 sks->addr[0] = ks->addr[0];
548 sks->addr[1] = ks->addr[1];
549 sks->port[0] = ks->port[0];
550 sks->port[1] = ks->port[1];
551 sks->proto = sp->proto;
556 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
557 st->creation = time_uptime - ntohl(sp->creation);
558 st->expire = time_uptime;
562 timeout = r->timeout[sp->timeout];
564 timeout = V_pf_default_rule.timeout[sp->timeout];
566 /* sp->expire may have been adaptively scaled by export. */
567 st->expire -= timeout - ntohl(sp->expire);
570 st->direction = sp->direction;
572 st->timeout = sp->timeout;
573 st->state_flags = sp->state_flags;
576 st->creatorid = sp->creatorid;
577 pf_state_peer_ntoh(&sp->src, &st->src);
578 pf_state_peer_ntoh(&sp->dst, &st->dst);
581 st->nat_rule.ptr = NULL;
582 st->anchor.ptr = NULL;
585 st->pfsync_time = time_uptime;
586 st->sync_state = PFSYNC_S_NONE;
588 if (!(flags & PFSYNC_SI_IOCTL))
589 st->state_flags |= PFSTATE_NOSYNC;
591 if ((error = pf_state_insert(kif, skw, sks, st)) != 0)
594 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
595 counter_u64_add(r->states_cur, 1);
596 counter_u64_add(r->states_tot, 1);
598 if (!(flags & PFSYNC_SI_IOCTL)) {
599 st->state_flags &= ~PFSTATE_NOSYNC;
600 if (st->state_flags & PFSTATE_ACK) {
601 pfsync_q_ins(st, PFSYNC_S_IACK, true);
605 st->state_flags &= ~PFSTATE_ACK;
615 uma_zfree(V_pf_state_key_z, skw);
617 uma_zfree(V_pf_state_key_z, sks);
619 cleanup_state: /* pf_state_insert() frees the state keys. */
622 uma_zfree(V_pf_state_scrub_z, st->dst.scrub);
624 uma_zfree(V_pf_state_scrub_z, st->src.scrub);
625 uma_zfree(V_pf_state_z, st);
631 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
633 struct pfsync_softc *sc = V_pfsyncif;
634 struct pfsync_pkt pkt;
635 struct mbuf *m = *mp;
636 struct ip *ip = mtod(m, struct ip *);
637 struct pfsync_header *ph;
638 struct pfsync_subheader subh;
644 PF_RULES_RLOCK_TRACKER;
647 V_pfsyncstats.pfsyncs_ipackets++;
649 /* Verify that we have a sync interface configured. */
650 if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
651 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
654 /* verify that the packet came in on the right interface */
655 if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
656 V_pfsyncstats.pfsyncs_badif++;
660 if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
661 if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
662 /* verify that the IP TTL is 255. */
663 if (ip->ip_ttl != PFSYNC_DFLTTL) {
664 V_pfsyncstats.pfsyncs_badttl++;
668 offset = ip->ip_hl << 2;
669 if (m->m_pkthdr.len < offset + sizeof(*ph)) {
670 V_pfsyncstats.pfsyncs_hdrops++;
674 if (offset + sizeof(*ph) > m->m_len) {
675 if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
676 V_pfsyncstats.pfsyncs_hdrops++;
677 return (IPPROTO_DONE);
679 ip = mtod(m, struct ip *);
681 ph = (struct pfsync_header *)((char *)ip + offset);
683 /* verify the version */
684 if (ph->version != PFSYNC_VERSION) {
685 V_pfsyncstats.pfsyncs_badver++;
689 len = ntohs(ph->len) + offset;
690 if (m->m_pkthdr.len < len) {
691 V_pfsyncstats.pfsyncs_badlen++;
695 /* Cheaper to grab this now than having to mess with mbufs later */
697 pkt.src = ip->ip_src;
701 * Trusting pf_chksum during packet processing, as well as seeking
702 * in interface name tree, require holding PF_RULES_RLOCK().
705 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
706 pkt.flags |= PFSYNC_SI_CKSUM;
708 offset += sizeof(*ph);
709 while (offset <= len - sizeof(subh)) {
710 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
711 offset += sizeof(subh);
713 if (subh.action >= PFSYNC_ACT_MAX) {
714 V_pfsyncstats.pfsyncs_badact++;
719 count = ntohs(subh.count);
720 V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
721 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count);
724 return (IPPROTO_DONE);
733 return (IPPROTO_DONE);
737 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
739 struct pfsync_clr *clr;
741 int len = sizeof(*clr) * count;
745 mp = m_pulldown(m, offset, len, &offp);
747 V_pfsyncstats.pfsyncs_badlen++;
750 clr = (struct pfsync_clr *)(mp->m_data + offp);
752 for (i = 0; i < count; i++) {
753 creatorid = clr[i].creatorid;
755 if (clr[i].ifname[0] != '\0' &&
756 pfi_kif_find(clr[i].ifname) == NULL)
759 for (int i = 0; i <= pf_hashmask; i++) {
760 struct pf_idhash *ih = &V_pf_idhash[i];
764 LIST_FOREACH(s, &ih->states, entry) {
765 if (s->creatorid == creatorid) {
766 s->state_flags |= PFSTATE_NOSYNC;
767 pf_unlink_state(s, PF_ENTER_LOCKED);
771 PF_HASHROW_UNLOCK(ih);
779 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
782 struct pfsync_state *sa, *sp;
783 int len = sizeof(*sp) * count;
786 mp = m_pulldown(m, offset, len, &offp);
788 V_pfsyncstats.pfsyncs_badlen++;
791 sa = (struct pfsync_state *)(mp->m_data + offp);
793 for (i = 0; i < count; i++) {
796 /* Check for invalid values. */
797 if (sp->timeout >= PFTM_MAX ||
798 sp->src.state > PF_TCPS_PROXY_DST ||
799 sp->dst.state > PF_TCPS_PROXY_DST ||
800 sp->direction > PF_OUT ||
801 (sp->af != AF_INET && sp->af != AF_INET6)) {
802 if (V_pf_status.debug >= PF_DEBUG_MISC)
803 printf("%s: invalid value\n", __func__);
804 V_pfsyncstats.pfsyncs_badval++;
808 if (pfsync_state_import(sp, pkt->flags) == ENOMEM)
809 /* Drop out, but process the rest of the actions. */
817 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
819 struct pfsync_ins_ack *ia, *iaa;
823 int len = count * sizeof(*ia);
826 mp = m_pulldown(m, offset, len, &offp);
828 V_pfsyncstats.pfsyncs_badlen++;
831 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
833 for (i = 0; i < count; i++) {
836 st = pf_find_state_byid(ia->id, ia->creatorid);
840 if (st->state_flags & PFSTATE_ACK) {
841 pfsync_undefer_state(st, 0);
846 * XXX this is not yet implemented, but we know the size of the
847 * message so we can skip it.
850 return (count * sizeof(struct pfsync_ins_ack));
854 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src,
855 struct pfsync_state_peer *dst)
859 PF_STATE_LOCK_ASSERT(st);
862 * The state should never go backwards except
863 * for syn-proxy states. Neither should the
864 * sequence window slide backwards.
866 if ((st->src.state > src->state &&
867 (st->src.state < PF_TCPS_PROXY_SRC ||
868 src->state >= PF_TCPS_PROXY_SRC)) ||
870 (st->src.state == src->state &&
871 SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
874 pf_state_peer_ntoh(src, &st->src);
876 if ((st->dst.state > dst->state) ||
878 (st->dst.state >= TCPS_SYN_SENT &&
879 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
882 pf_state_peer_ntoh(dst, &st->dst);
888 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
890 struct pfsync_softc *sc = V_pfsyncif;
891 struct pfsync_state *sa, *sp;
896 int len = count * sizeof(*sp);
899 mp = m_pulldown(m, offset, len, &offp);
901 V_pfsyncstats.pfsyncs_badlen++;
904 sa = (struct pfsync_state *)(mp->m_data + offp);
906 for (i = 0; i < count; i++) {
909 /* check for invalid values */
910 if (sp->timeout >= PFTM_MAX ||
911 sp->src.state > PF_TCPS_PROXY_DST ||
912 sp->dst.state > PF_TCPS_PROXY_DST) {
913 if (V_pf_status.debug >= PF_DEBUG_MISC) {
914 printf("pfsync_input: PFSYNC_ACT_UPD: "
917 V_pfsyncstats.pfsyncs_badval++;
921 st = pf_find_state_byid(sp->id, sp->creatorid);
923 /* insert the update */
924 if (pfsync_state_import(sp, pkt->flags))
925 V_pfsyncstats.pfsyncs_badstate++;
929 if (st->state_flags & PFSTATE_ACK) {
930 pfsync_undefer_state(st, 1);
933 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
934 sync = pfsync_upd_tcp(st, &sp->src, &sp->dst);
939 * Non-TCP protocol state machine always go
942 if (st->src.state > sp->src.state)
945 pf_state_peer_ntoh(&sp->src, &st->src);
946 if (st->dst.state > sp->dst.state)
949 pf_state_peer_ntoh(&sp->dst, &st->dst);
952 pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
953 pf_state_peer_ntoh(&sp->dst, &st->dst);
954 st->expire = time_uptime;
955 st->timeout = sp->timeout;
957 st->pfsync_time = time_uptime;
960 V_pfsyncstats.pfsyncs_stale++;
962 pfsync_update_state(st);
974 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
976 struct pfsync_softc *sc = V_pfsyncif;
977 struct pfsync_upd_c *ua, *up;
979 int len = count * sizeof(*up);
984 mp = m_pulldown(m, offset, len, &offp);
986 V_pfsyncstats.pfsyncs_badlen++;
989 ua = (struct pfsync_upd_c *)(mp->m_data + offp);
991 for (i = 0; i < count; i++) {
994 /* check for invalid values */
995 if (up->timeout >= PFTM_MAX ||
996 up->src.state > PF_TCPS_PROXY_DST ||
997 up->dst.state > PF_TCPS_PROXY_DST) {
998 if (V_pf_status.debug >= PF_DEBUG_MISC) {
999 printf("pfsync_input: "
1000 "PFSYNC_ACT_UPD_C: "
1003 V_pfsyncstats.pfsyncs_badval++;
1007 st = pf_find_state_byid(up->id, up->creatorid);
1009 /* We don't have this state. Ask for it. */
1010 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1011 pfsync_request_update(up->creatorid, up->id);
1012 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1016 if (st->state_flags & PFSTATE_ACK) {
1017 pfsync_undefer_state(st, 1);
1020 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1021 sync = pfsync_upd_tcp(st, &up->src, &up->dst);
1026 * Non-TCP protocol state machine always go
1029 if (st->src.state > up->src.state)
1032 pf_state_peer_ntoh(&up->src, &st->src);
1033 if (st->dst.state > up->dst.state)
1036 pf_state_peer_ntoh(&up->dst, &st->dst);
1039 pfsync_alloc_scrub_memory(&up->dst, &st->dst);
1040 pf_state_peer_ntoh(&up->dst, &st->dst);
1041 st->expire = time_uptime;
1042 st->timeout = up->timeout;
1044 st->pfsync_time = time_uptime;
1047 V_pfsyncstats.pfsyncs_stale++;
1049 pfsync_update_state(st);
1050 PF_STATE_UNLOCK(st);
1051 pfsync_push_all(sc);
1054 PF_STATE_UNLOCK(st);
1061 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1063 struct pfsync_upd_req *ur, *ura;
1065 int len = count * sizeof(*ur);
1068 struct pf_state *st;
1070 mp = m_pulldown(m, offset, len, &offp);
1072 V_pfsyncstats.pfsyncs_badlen++;
1075 ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1077 for (i = 0; i < count; i++) {
1080 if (ur->id == 0 && ur->creatorid == 0)
1081 pfsync_bulk_start();
1083 st = pf_find_state_byid(ur->id, ur->creatorid);
1085 V_pfsyncstats.pfsyncs_badstate++;
1088 if (st->state_flags & PFSTATE_NOSYNC) {
1089 PF_STATE_UNLOCK(st);
1093 pfsync_update_state_req(st);
1094 PF_STATE_UNLOCK(st);
1102 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1105 struct pfsync_state *sa, *sp;
1106 struct pf_state *st;
1107 int len = count * sizeof(*sp);
1110 mp = m_pulldown(m, offset, len, &offp);
1112 V_pfsyncstats.pfsyncs_badlen++;
1115 sa = (struct pfsync_state *)(mp->m_data + offp);
1117 for (i = 0; i < count; i++) {
1120 st = pf_find_state_byid(sp->id, sp->creatorid);
1122 V_pfsyncstats.pfsyncs_badstate++;
1125 st->state_flags |= PFSTATE_NOSYNC;
1126 pf_unlink_state(st, PF_ENTER_LOCKED);
1133 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1136 struct pfsync_del_c *sa, *sp;
1137 struct pf_state *st;
1138 int len = count * sizeof(*sp);
1141 mp = m_pulldown(m, offset, len, &offp);
1143 V_pfsyncstats.pfsyncs_badlen++;
1146 sa = (struct pfsync_del_c *)(mp->m_data + offp);
1148 for (i = 0; i < count; i++) {
1151 st = pf_find_state_byid(sp->id, sp->creatorid);
1153 V_pfsyncstats.pfsyncs_badstate++;
1157 st->state_flags |= PFSTATE_NOSYNC;
1158 pf_unlink_state(st, PF_ENTER_LOCKED);
1165 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1167 struct pfsync_softc *sc = V_pfsyncif;
1168 struct pfsync_bus *bus;
1170 int len = count * sizeof(*bus);
1175 /* If we're not waiting for a bulk update, who cares. */
1176 if (sc->sc_ureq_sent == 0) {
1181 mp = m_pulldown(m, offset, len, &offp);
1184 V_pfsyncstats.pfsyncs_badlen++;
1187 bus = (struct pfsync_bus *)(mp->m_data + offp);
1189 switch (bus->status) {
1190 case PFSYNC_BUS_START:
1191 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1192 V_pf_limits[PF_LIMIT_STATES].limit /
1193 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1194 sizeof(struct pfsync_state)),
1195 pfsync_bulk_fail, sc);
1196 if (V_pf_status.debug >= PF_DEBUG_MISC)
1197 printf("pfsync: received bulk update start\n");
1200 case PFSYNC_BUS_END:
1201 if (time_uptime - ntohl(bus->endtime) >=
1203 /* that's it, we're happy */
1204 sc->sc_ureq_sent = 0;
1205 sc->sc_bulk_tries = 0;
1206 callout_stop(&sc->sc_bulkfail_tmo);
1207 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1208 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
1209 "pfsync bulk done");
1210 sc->sc_flags |= PFSYNCF_OK;
1211 if (V_pf_status.debug >= PF_DEBUG_MISC)
1212 printf("pfsync: received valid "
1213 "bulk update end\n");
1215 if (V_pf_status.debug >= PF_DEBUG_MISC)
1216 printf("pfsync: received invalid "
1217 "bulk update end: bad timestamp\n");
1227 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1229 int len = count * sizeof(struct pfsync_tdb);
1232 struct pfsync_tdb *tp;
1238 mp = m_pulldown(m, offset, len, &offp);
1240 V_pfsyncstats.pfsyncs_badlen++;
1243 tp = (struct pfsync_tdb *)(mp->m_data + offp);
1245 for (i = 0; i < count; i++)
1246 pfsync_update_net_tdb(&tp[i]);
1253 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1255 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1260 /* check for invalid values */
1261 if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1262 (pt->dst.sa.sa_family != AF_INET &&
1263 pt->dst.sa.sa_family != AF_INET6))
1266 tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1268 pt->rpl = ntohl(pt->rpl);
1269 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1271 /* Neither replay nor byte counter should ever decrease. */
1272 if (pt->rpl < tdb->tdb_rpl ||
1273 pt->cur_bytes < tdb->tdb_cur_bytes) {
1277 tdb->tdb_rpl = pt->rpl;
1278 tdb->tdb_cur_bytes = pt->cur_bytes;
1283 if (V_pf_status.debug >= PF_DEBUG_MISC)
1284 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1286 V_pfsyncstats.pfsyncs_badstate++;
1293 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1295 /* check if we are at the right place in the packet */
1296 if (offset != m->m_pkthdr.len)
1297 V_pfsyncstats.pfsyncs_badlen++;
1299 /* we're done. free and let the caller return */
1305 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1307 V_pfsyncstats.pfsyncs_badact++;
1314 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1323 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1325 struct pfsync_softc *sc = ifp->if_softc;
1326 struct ifreq *ifr = (struct ifreq *)data;
1327 struct pfsyncreq pfsyncr;
1334 if (ifp->if_flags & IFF_UP) {
1335 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1337 pfsync_pointers_init();
1339 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1341 pfsync_pointers_uninit();
1345 if (!sc->sc_sync_if ||
1346 ifr->ifr_mtu <= PFSYNC_MINPKT ||
1347 ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1349 if (ifr->ifr_mtu < ifp->if_mtu) {
1350 for (c = 0; c < pfsync_buckets; c++) {
1351 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1352 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT)
1353 pfsync_sendout(1, c);
1354 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1357 ifp->if_mtu = ifr->ifr_mtu;
1360 bzero(&pfsyncr, sizeof(pfsyncr));
1362 if (sc->sc_sync_if) {
1363 strlcpy(pfsyncr.pfsyncr_syncdev,
1364 sc->sc_sync_if->if_xname, IFNAMSIZ);
1366 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1367 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1368 pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER ==
1369 (sc->sc_flags & PFSYNCF_DEFER));
1371 return (copyout(&pfsyncr, ifr_data_get_ptr(ifr),
1376 struct in_mfilter *imf = NULL;
1380 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1382 if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr,
1386 if (pfsyncr.pfsyncr_maxupdates > 255)
1389 if (pfsyncr.pfsyncr_syncdev[0] == 0)
1391 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL)
1394 if (sifp != NULL && (
1395 pfsyncr.pfsyncr_syncpeer.s_addr == 0 ||
1396 pfsyncr.pfsyncr_syncpeer.s_addr ==
1397 htonl(INADDR_PFSYNC_GROUP)))
1398 imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
1401 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1402 sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP);
1404 sc->sc_sync_peer.s_addr =
1405 pfsyncr.pfsyncr_syncpeer.s_addr;
1407 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1408 if (pfsyncr.pfsyncr_defer) {
1409 sc->sc_flags |= PFSYNCF_DEFER;
1410 V_pfsync_defer_ptr = pfsync_defer;
1412 sc->sc_flags &= ~PFSYNCF_DEFER;
1413 V_pfsync_defer_ptr = NULL;
1418 if_rele(sc->sc_sync_if);
1419 sc->sc_sync_if = NULL;
1420 pfsync_multicast_cleanup(sc);
1425 for (c = 0; c < pfsync_buckets; c++) {
1426 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1427 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT &&
1428 (sifp->if_mtu < sc->sc_ifp->if_mtu ||
1429 (sc->sc_sync_if != NULL &&
1430 sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
1431 sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
1432 pfsync_sendout(1, c);
1433 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1436 pfsync_multicast_cleanup(sc);
1438 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
1439 error = pfsync_multicast_setup(sc, sifp, imf);
1442 ip_mfilter_free(imf);
1448 if_rele(sc->sc_sync_if);
1449 sc->sc_sync_if = sifp;
1451 ip = &sc->sc_template;
1452 bzero(ip, sizeof(*ip));
1453 ip->ip_v = IPVERSION;
1454 ip->ip_hl = sizeof(sc->sc_template) >> 2;
1455 ip->ip_tos = IPTOS_LOWDELAY;
1456 /* len and id are set later. */
1457 ip->ip_off = htons(IP_DF);
1458 ip->ip_ttl = PFSYNC_DFLTTL;
1459 ip->ip_p = IPPROTO_PFSYNC;
1460 ip->ip_src.s_addr = INADDR_ANY;
1461 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr;
1463 /* Request a full state table update. */
1464 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1465 (*carp_demote_adj_p)(V_pfsync_carp_adj,
1466 "pfsync bulk start");
1467 sc->sc_flags &= ~PFSYNCF_OK;
1468 if (V_pf_status.debug >= PF_DEBUG_MISC)
1469 printf("pfsync: requesting bulk update\n");
1471 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1472 pfsync_request_update(0, 0);
1473 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1475 sc->sc_ureq_sent = time_uptime;
1476 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail,
1490 pfsync_out_state(struct pf_state *st, void *buf)
1492 struct pfsync_state *sp = buf;
1494 pfsync_state_export(sp, st);
1498 pfsync_out_iack(struct pf_state *st, void *buf)
1500 struct pfsync_ins_ack *iack = buf;
1503 iack->creatorid = st->creatorid;
1507 pfsync_out_upd_c(struct pf_state *st, void *buf)
1509 struct pfsync_upd_c *up = buf;
1511 bzero(up, sizeof(*up));
1513 pf_state_peer_hton(&st->src, &up->src);
1514 pf_state_peer_hton(&st->dst, &up->dst);
1515 up->creatorid = st->creatorid;
1516 up->timeout = st->timeout;
1520 pfsync_out_del(struct pf_state *st, void *buf)
1522 struct pfsync_del_c *dp = buf;
1525 dp->creatorid = st->creatorid;
1526 st->state_flags |= PFSTATE_NOSYNC;
1530 pfsync_drop(struct pfsync_softc *sc)
1532 struct pf_state *st, *next;
1533 struct pfsync_upd_req_item *ur;
1534 struct pfsync_bucket *b;
1537 for (c = 0; c < pfsync_buckets; c++) {
1538 b = &sc->sc_buckets[c];
1539 for (q = 0; q < PFSYNC_S_COUNT; q++) {
1540 if (TAILQ_EMPTY(&b->b_qs[q]))
1543 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) {
1544 KASSERT(st->sync_state == q,
1545 ("%s: st->sync_state == q",
1547 st->sync_state = PFSYNC_S_NONE;
1548 pf_release_state(st);
1550 TAILQ_INIT(&b->b_qs[q]);
1553 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1554 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1558 b->b_len = PFSYNC_MINPKT;
1564 pfsync_sendout(int schedswi, int c)
1566 struct pfsync_softc *sc = V_pfsyncif;
1567 struct ifnet *ifp = sc->sc_ifp;
1570 struct pfsync_header *ph;
1571 struct pfsync_subheader *subh;
1572 struct pf_state *st, *st_next;
1573 struct pfsync_upd_req_item *ur;
1574 struct pfsync_bucket *b = &sc->sc_buckets[c];
1578 KASSERT(sc != NULL, ("%s: null sc", __func__));
1579 KASSERT(b->b_len > PFSYNC_MINPKT,
1580 ("%s: sc_len %zu", __func__, b->b_len));
1581 PFSYNC_BUCKET_LOCK_ASSERT(b);
1583 if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1588 m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1590 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1591 V_pfsyncstats.pfsyncs_onomem++;
1594 m->m_data += max_linkhdr;
1595 m->m_len = m->m_pkthdr.len = b->b_len;
1597 /* build the ip header */
1598 ip = (struct ip *)m->m_data;
1599 bcopy(&sc->sc_template, ip, sizeof(*ip));
1600 offset = sizeof(*ip);
1602 ip->ip_len = htons(m->m_pkthdr.len);
1605 /* build the pfsync header */
1606 ph = (struct pfsync_header *)(m->m_data + offset);
1607 bzero(ph, sizeof(*ph));
1608 offset += sizeof(*ph);
1610 ph->version = PFSYNC_VERSION;
1611 ph->len = htons(b->b_len - sizeof(*ip));
1612 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1614 /* walk the queues */
1615 for (q = 0; q < PFSYNC_S_COUNT; q++) {
1616 if (TAILQ_EMPTY(&b->b_qs[q]))
1619 subh = (struct pfsync_subheader *)(m->m_data + offset);
1620 offset += sizeof(*subh);
1623 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) {
1624 KASSERT(st->sync_state == q,
1625 ("%s: st->sync_state == q",
1628 * XXXGL: some of write methods do unlocked reads
1631 pfsync_qs[q].write(st, m->m_data + offset);
1632 offset += pfsync_qs[q].len;
1633 st->sync_state = PFSYNC_S_NONE;
1634 pf_release_state(st);
1637 TAILQ_INIT(&b->b_qs[q]);
1639 bzero(subh, sizeof(*subh));
1640 subh->action = pfsync_qs[q].action;
1641 subh->count = htons(count);
1642 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1645 if (!TAILQ_EMPTY(&b->b_upd_req_list)) {
1646 subh = (struct pfsync_subheader *)(m->m_data + offset);
1647 offset += sizeof(*subh);
1650 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1651 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1653 bcopy(&ur->ur_msg, m->m_data + offset,
1654 sizeof(ur->ur_msg));
1655 offset += sizeof(ur->ur_msg);
1660 bzero(subh, sizeof(*subh));
1661 subh->action = PFSYNC_ACT_UPD_REQ;
1662 subh->count = htons(count);
1663 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1666 /* has someone built a custom region for us to add? */
1667 if (b->b_plus != NULL) {
1668 bcopy(b->b_plus, m->m_data + offset, b->b_pluslen);
1669 offset += b->b_pluslen;
1674 subh = (struct pfsync_subheader *)(m->m_data + offset);
1675 offset += sizeof(*subh);
1677 bzero(subh, sizeof(*subh));
1678 subh->action = PFSYNC_ACT_EOF;
1679 subh->count = htons(1);
1680 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1682 /* we're done, let's put it on the wire */
1684 m->m_data += sizeof(*ip);
1685 m->m_len = m->m_pkthdr.len = b->b_len - sizeof(*ip);
1687 m->m_data -= sizeof(*ip);
1688 m->m_len = m->m_pkthdr.len = b->b_len;
1691 if (sc->sc_sync_if == NULL) {
1692 b->b_len = PFSYNC_MINPKT;
1697 if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1698 if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1699 b->b_len = PFSYNC_MINPKT;
1701 if (!_IF_QFULL(&b->b_snd))
1702 _IF_ENQUEUE(&b->b_snd, m);
1705 if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1708 swi_sched(V_pfsync_swi_cookie, 0);
1712 pfsync_insert_state(struct pf_state *st)
1714 struct pfsync_softc *sc = V_pfsyncif;
1715 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1717 if (st->state_flags & PFSTATE_NOSYNC)
1720 if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1721 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1722 st->state_flags |= PFSTATE_NOSYNC;
1726 KASSERT(st->sync_state == PFSYNC_S_NONE,
1727 ("%s: st->sync_state %u", __func__, st->sync_state));
1729 PFSYNC_BUCKET_LOCK(b);
1730 if (b->b_len == PFSYNC_MINPKT)
1731 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1733 pfsync_q_ins(st, PFSYNC_S_INS, true);
1734 PFSYNC_BUCKET_UNLOCK(b);
1736 st->sync_updates = 0;
1740 pfsync_defer(struct pf_state *st, struct mbuf *m)
1742 struct pfsync_softc *sc = V_pfsyncif;
1743 struct pfsync_deferral *pd;
1744 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1746 if (m->m_flags & (M_BCAST|M_MCAST))
1751 if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) ||
1752 !(sc->sc_flags & PFSYNCF_DEFER)) {
1757 if (b->b_deferred >= 128)
1758 pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0);
1760 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1765 m->m_flags |= M_SKIP_FIREWALL;
1766 st->state_flags |= PFSTATE_ACK;
1774 TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry);
1775 callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED);
1776 callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd);
1784 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1786 struct pfsync_softc *sc = pd->pd_sc;
1787 struct mbuf *m = pd->pd_m;
1788 struct pf_state *st = pd->pd_st;
1789 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1791 PFSYNC_BUCKET_LOCK_ASSERT(b);
1793 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
1795 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1797 pf_release_state(st);
1802 _IF_ENQUEUE(&b->b_snd, m);
1808 pfsync_defer_tmo(void *arg)
1810 struct epoch_tracker et;
1811 struct pfsync_deferral *pd = arg;
1812 struct pfsync_softc *sc = pd->pd_sc;
1813 struct mbuf *m = pd->pd_m;
1814 struct pf_state *st = pd->pd_st;
1815 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1817 PFSYNC_BUCKET_LOCK_ASSERT(b);
1819 NET_EPOCH_ENTER(et);
1820 CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
1822 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
1824 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1825 if (pd->pd_refs == 0)
1829 ip_output(m, NULL, NULL, 0, NULL, NULL);
1831 pf_release_state(st);
1838 pfsync_undefer_state(struct pf_state *st, int drop)
1840 struct pfsync_softc *sc = V_pfsyncif;
1841 struct pfsync_deferral *pd;
1842 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1844 PFSYNC_BUCKET_LOCK(b);
1846 TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) {
1847 if (pd->pd_st == st) {
1848 if (callout_stop(&pd->pd_tmo) > 0)
1849 pfsync_undefer(pd, drop);
1851 PFSYNC_BUCKET_UNLOCK(b);
1855 PFSYNC_BUCKET_UNLOCK(b);
1857 panic("%s: unable to find deferred state", __func__);
1860 static struct pfsync_bucket*
1861 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_state *st)
1863 int c = PF_IDHASH(st) % pfsync_buckets;
1864 return &sc->sc_buckets[c];
1868 pfsync_update_state(struct pf_state *st)
1870 struct pfsync_softc *sc = V_pfsyncif;
1871 bool sync = false, ref = true;
1872 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1874 PF_STATE_LOCK_ASSERT(st);
1875 PFSYNC_BUCKET_LOCK(b);
1877 if (st->state_flags & PFSTATE_ACK)
1878 pfsync_undefer_state(st, 0);
1879 if (st->state_flags & PFSTATE_NOSYNC) {
1880 if (st->sync_state != PFSYNC_S_NONE)
1881 pfsync_q_del(st, true, b);
1882 PFSYNC_BUCKET_UNLOCK(b);
1886 if (b->b_len == PFSYNC_MINPKT)
1887 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1889 switch (st->sync_state) {
1890 case PFSYNC_S_UPD_C:
1893 /* we're already handling it */
1895 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1897 if (st->sync_updates >= sc->sc_maxupdates)
1903 pfsync_q_del(st, false, b);
1908 pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
1909 st->sync_updates = 0;
1913 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1916 if (sync || (time_uptime - st->pfsync_time) < 2)
1919 PFSYNC_BUCKET_UNLOCK(b);
1923 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1925 struct pfsync_softc *sc = V_pfsyncif;
1926 struct pfsync_bucket *b = &sc->sc_buckets[0];
1927 struct pfsync_upd_req_item *item;
1928 size_t nlen = sizeof(struct pfsync_upd_req);
1930 PFSYNC_BUCKET_LOCK_ASSERT(b);
1933 * This code does a bit to prevent multiple update requests for the
1934 * same state being generated. It searches current subheader queue,
1935 * but it doesn't lookup into queue of already packed datagrams.
1937 TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry)
1938 if (item->ur_msg.id == id &&
1939 item->ur_msg.creatorid == creatorid)
1942 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1944 return; /* XXX stats */
1946 item->ur_msg.id = id;
1947 item->ur_msg.creatorid = creatorid;
1949 if (TAILQ_EMPTY(&b->b_upd_req_list))
1950 nlen += sizeof(struct pfsync_subheader);
1952 if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
1953 pfsync_sendout(1, 0);
1955 nlen = sizeof(struct pfsync_subheader) +
1956 sizeof(struct pfsync_upd_req);
1959 TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry);
1964 pfsync_update_state_req(struct pf_state *st)
1966 struct pfsync_softc *sc = V_pfsyncif;
1967 bool ref = true, full = false;
1968 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1970 PF_STATE_LOCK_ASSERT(st);
1971 PFSYNC_BUCKET_LOCK(b);
1973 if (st->state_flags & PFSTATE_NOSYNC) {
1974 if (st->sync_state != PFSYNC_S_NONE)
1975 pfsync_q_del(st, true, b);
1976 PFSYNC_BUCKET_UNLOCK(b);
1980 switch (st->sync_state) {
1981 case PFSYNC_S_UPD_C:
1983 pfsync_q_del(st, false, b);
1988 pfsync_q_ins(st, PFSYNC_S_UPD, ref);
1995 /* we're already handling it */
1999 panic("%s: unexpected sync state %d", __func__, st->sync_state);
2002 if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(struct pfsync_state))
2005 PFSYNC_BUCKET_UNLOCK(b);
2011 pfsync_delete_state(struct pf_state *st)
2013 struct pfsync_softc *sc = V_pfsyncif;
2014 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2017 PFSYNC_BUCKET_LOCK(b);
2018 if (st->state_flags & PFSTATE_ACK)
2019 pfsync_undefer_state(st, 1);
2020 if (st->state_flags & PFSTATE_NOSYNC) {
2021 if (st->sync_state != PFSYNC_S_NONE)
2022 pfsync_q_del(st, true, b);
2023 PFSYNC_BUCKET_UNLOCK(b);
2027 if (b->b_len == PFSYNC_MINPKT)
2028 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2030 switch (st->sync_state) {
2032 /* We never got to tell the world so just forget about it. */
2033 pfsync_q_del(st, true, b);
2036 case PFSYNC_S_UPD_C:
2039 pfsync_q_del(st, false, b);
2044 pfsync_q_ins(st, PFSYNC_S_DEL, ref);
2048 panic("%s: unexpected sync state %d", __func__, st->sync_state);
2051 PFSYNC_BUCKET_UNLOCK(b);
2055 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
2058 struct pfsync_subheader subh;
2059 struct pfsync_clr clr;
2062 bzero(&r, sizeof(r));
2064 r.subh.action = PFSYNC_ACT_CLR;
2065 r.subh.count = htons(1);
2066 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
2068 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
2069 r.clr.creatorid = creatorid;
2071 pfsync_send_plus(&r, sizeof(r));
2075 pfsync_q_ins(struct pf_state *st, int q, bool ref)
2077 struct pfsync_softc *sc = V_pfsyncif;
2078 size_t nlen = pfsync_qs[q].len;
2079 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2081 PFSYNC_BUCKET_LOCK_ASSERT(b);
2083 KASSERT(st->sync_state == PFSYNC_S_NONE,
2084 ("%s: st->sync_state %u", __func__, st->sync_state));
2085 KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2088 if (TAILQ_EMPTY(&b->b_qs[q]))
2089 nlen += sizeof(struct pfsync_subheader);
2091 if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2092 pfsync_sendout(1, b->b_id);
2094 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2098 TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list);
2105 pfsync_q_del(struct pf_state *st, bool unref, struct pfsync_bucket *b)
2107 int q = st->sync_state;
2109 PFSYNC_BUCKET_LOCK_ASSERT(b);
2110 KASSERT(st->sync_state != PFSYNC_S_NONE,
2111 ("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2113 b->b_len -= pfsync_qs[q].len;
2114 TAILQ_REMOVE(&b->b_qs[q], st, sync_list);
2115 st->sync_state = PFSYNC_S_NONE;
2117 pf_release_state(st);
2119 if (TAILQ_EMPTY(&b->b_qs[q]))
2120 b->b_len -= sizeof(struct pfsync_subheader);
2124 pfsync_bulk_start(void)
2126 struct pfsync_softc *sc = V_pfsyncif;
2128 if (V_pf_status.debug >= PF_DEBUG_MISC)
2129 printf("pfsync: received bulk update request\n");
2133 sc->sc_ureq_received = time_uptime;
2134 sc->sc_bulk_hashid = 0;
2135 sc->sc_bulk_stateid = 0;
2136 pfsync_bulk_status(PFSYNC_BUS_START);
2137 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2142 pfsync_bulk_update(void *arg)
2144 struct pfsync_softc *sc = arg;
2148 PFSYNC_BLOCK_ASSERT(sc);
2149 CURVNET_SET(sc->sc_ifp->if_vnet);
2152 * Start with last state from previous invocation.
2153 * It may had gone, in this case start from the
2156 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2161 i = sc->sc_bulk_hashid;
2163 for (; i <= pf_hashmask; i++) {
2164 struct pf_idhash *ih = &V_pf_idhash[i];
2167 PF_HASHROW_ASSERT(ih);
2169 PF_HASHROW_LOCK(ih);
2170 s = LIST_FIRST(&ih->states);
2173 for (; s; s = LIST_NEXT(s, entry)) {
2174 if (s->sync_state == PFSYNC_S_NONE &&
2175 s->timeout < PFTM_MAX &&
2176 s->pfsync_time <= sc->sc_ureq_received) {
2177 if (pfsync_update_state_req(s)) {
2178 /* We've filled a packet. */
2179 sc->sc_bulk_hashid = i;
2180 sc->sc_bulk_stateid = s->id;
2181 sc->sc_bulk_creatorid = s->creatorid;
2182 PF_HASHROW_UNLOCK(ih);
2183 callout_reset(&sc->sc_bulk_tmo, 1,
2184 pfsync_bulk_update, sc);
2190 PF_HASHROW_UNLOCK(ih);
2194 pfsync_bulk_status(PFSYNC_BUS_END);
2200 pfsync_bulk_status(u_int8_t status)
2203 struct pfsync_subheader subh;
2204 struct pfsync_bus bus;
2207 struct pfsync_softc *sc = V_pfsyncif;
2209 bzero(&r, sizeof(r));
2211 r.subh.action = PFSYNC_ACT_BUS;
2212 r.subh.count = htons(1);
2213 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2215 r.bus.creatorid = V_pf_status.hostid;
2216 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2217 r.bus.status = status;
2219 pfsync_send_plus(&r, sizeof(r));
2223 pfsync_bulk_fail(void *arg)
2225 struct pfsync_softc *sc = arg;
2226 struct pfsync_bucket *b = &sc->sc_buckets[0];
2228 CURVNET_SET(sc->sc_ifp->if_vnet);
2230 PFSYNC_BLOCK_ASSERT(sc);
2232 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2234 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2235 pfsync_bulk_fail, V_pfsyncif);
2236 PFSYNC_BUCKET_LOCK(b);
2237 pfsync_request_update(0, 0);
2238 PFSYNC_BUCKET_UNLOCK(b);
2240 /* Pretend like the transfer was ok. */
2241 sc->sc_ureq_sent = 0;
2242 sc->sc_bulk_tries = 0;
2244 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2245 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
2246 "pfsync bulk fail");
2247 sc->sc_flags |= PFSYNCF_OK;
2249 if (V_pf_status.debug >= PF_DEBUG_MISC)
2250 printf("pfsync: failed to receive bulk update\n");
2257 pfsync_send_plus(void *plus, size_t pluslen)
2259 struct pfsync_softc *sc = V_pfsyncif;
2260 struct pfsync_bucket *b = &sc->sc_buckets[0];
2262 PFSYNC_BUCKET_LOCK(b);
2264 if (b->b_len + pluslen > sc->sc_ifp->if_mtu)
2265 pfsync_sendout(1, b->b_id);
2268 b->b_len += (b->b_pluslen = pluslen);
2270 pfsync_sendout(1, b->b_id);
2271 PFSYNC_BUCKET_UNLOCK(b);
2275 pfsync_timeout(void *arg)
2277 struct pfsync_bucket *b = arg;
2279 CURVNET_SET(b->b_sc->sc_ifp->if_vnet);
2280 PFSYNC_BUCKET_LOCK(b);
2282 PFSYNC_BUCKET_UNLOCK(b);
2287 pfsync_push(struct pfsync_bucket *b)
2290 PFSYNC_BUCKET_LOCK_ASSERT(b);
2292 b->b_flags |= PFSYNCF_BUCKET_PUSH;
2293 swi_sched(V_pfsync_swi_cookie, 0);
2297 pfsync_push_all(struct pfsync_softc *sc)
2300 struct pfsync_bucket *b;
2302 for (c = 0; c < pfsync_buckets; c++) {
2303 b = &sc->sc_buckets[c];
2305 PFSYNC_BUCKET_LOCK(b);
2307 PFSYNC_BUCKET_UNLOCK(b);
2312 pfsyncintr(void *arg)
2314 struct epoch_tracker et;
2315 struct pfsync_softc *sc = arg;
2316 struct pfsync_bucket *b;
2320 NET_EPOCH_ENTER(et);
2321 CURVNET_SET(sc->sc_ifp->if_vnet);
2323 for (c = 0; c < pfsync_buckets; c++) {
2324 b = &sc->sc_buckets[c];
2326 PFSYNC_BUCKET_LOCK(b);
2327 if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) {
2328 pfsync_sendout(0, b->b_id);
2329 b->b_flags &= ~PFSYNCF_BUCKET_PUSH;
2331 _IF_DEQUEUE_ALL(&b->b_snd, m);
2332 PFSYNC_BUCKET_UNLOCK(b);
2334 for (; m != NULL; m = n) {
2337 m->m_nextpkt = NULL;
2340 * We distinguish between a deferral packet and our
2341 * own pfsync packet based on M_SKIP_FIREWALL
2342 * flag. This is XXX.
2344 if (m->m_flags & M_SKIP_FIREWALL)
2345 ip_output(m, NULL, NULL, 0, NULL, NULL);
2346 else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo,
2348 V_pfsyncstats.pfsyncs_opackets++;
2350 V_pfsyncstats.pfsyncs_oerrors++;
2358 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp,
2359 struct in_mfilter *imf)
2361 struct ip_moptions *imo = &sc->sc_imo;
2364 if (!(ifp->if_flags & IFF_MULTICAST))
2365 return (EADDRNOTAVAIL);
2367 imo->imo_multicast_vif = -1;
2369 if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL,
2370 &imf->imf_inm)) != 0)
2373 ip_mfilter_init(&imo->imo_head);
2374 ip_mfilter_insert(&imo->imo_head, imf);
2375 imo->imo_multicast_ifp = ifp;
2376 imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2377 imo->imo_multicast_loop = 0;
2383 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2385 struct ip_moptions *imo = &sc->sc_imo;
2386 struct in_mfilter *imf;
2388 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
2389 ip_mfilter_remove(&imo->imo_head, imf);
2390 in_leavegroup(imf->imf_inm, NULL);
2391 ip_mfilter_free(imf);
2393 imo->imo_multicast_ifp = NULL;
2397 pfsync_detach_ifnet(struct ifnet *ifp)
2399 struct pfsync_softc *sc = V_pfsyncif;
2406 if (sc->sc_sync_if == ifp) {
2407 /* We don't need mutlicast cleanup here, because the interface
2408 * is going away. We do need to ensure we don't try to do
2411 ip_mfilter_init(&sc->sc_imo.imo_head);
2412 sc->sc_imo.imo_multicast_ifp = NULL;
2413 sc->sc_sync_if = NULL;
2420 extern struct domain inetdomain;
2421 static struct protosw in_pfsync_protosw = {
2422 .pr_type = SOCK_RAW,
2423 .pr_domain = &inetdomain,
2424 .pr_protocol = IPPROTO_PFSYNC,
2425 .pr_flags = PR_ATOMIC|PR_ADDR,
2426 .pr_input = pfsync_input,
2427 .pr_output = rip_output,
2428 .pr_ctloutput = rip_ctloutput,
2429 .pr_usrreqs = &rip_usrreqs
2434 pfsync_pointers_init()
2438 V_pfsync_state_import_ptr = pfsync_state_import;
2439 V_pfsync_insert_state_ptr = pfsync_insert_state;
2440 V_pfsync_update_state_ptr = pfsync_update_state;
2441 V_pfsync_delete_state_ptr = pfsync_delete_state;
2442 V_pfsync_clear_states_ptr = pfsync_clear_states;
2443 V_pfsync_defer_ptr = pfsync_defer;
2448 pfsync_pointers_uninit()
2452 V_pfsync_state_import_ptr = NULL;
2453 V_pfsync_insert_state_ptr = NULL;
2454 V_pfsync_update_state_ptr = NULL;
2455 V_pfsync_delete_state_ptr = NULL;
2456 V_pfsync_clear_states_ptr = NULL;
2457 V_pfsync_defer_ptr = NULL;
2462 vnet_pfsync_init(const void *unused __unused)
2466 V_pfsync_cloner = if_clone_simple(pfsyncname,
2467 pfsync_clone_create, pfsync_clone_destroy, 1);
2468 error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif,
2469 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2471 if_clone_detach(V_pfsync_cloner);
2472 log(LOG_INFO, "swi_add() failed in %s\n", __func__);
2475 pfsync_pointers_init();
2477 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
2478 vnet_pfsync_init, NULL);
2481 vnet_pfsync_uninit(const void *unused __unused)
2484 pfsync_pointers_uninit();
2486 if_clone_detach(V_pfsync_cloner);
2487 swi_remove(V_pfsync_swi_cookie);
2490 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH,
2491 vnet_pfsync_uninit, NULL);
2499 pfsync_detach_ifnet_ptr = pfsync_detach_ifnet;
2501 error = pf_proto_register(PF_INET, &in_pfsync_protosw);
2504 error = ipproto_register(IPPROTO_PFSYNC);
2506 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2517 pfsync_detach_ifnet_ptr = NULL;
2520 ipproto_unregister(IPPROTO_PFSYNC);
2521 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2526 pfsync_modevent(module_t mod, int type, void *data)
2532 error = pfsync_init();
2545 static moduledata_t pfsync_mod = {
2551 #define PFSYNC_MODVER 1
2553 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
2554 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
2555 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2556 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);