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 *,
318 #define PFSYNC_MAX_BULKTRIES 12
320 VNET_DEFINE(struct if_clone *, pfsync_cloner);
321 #define V_pfsync_cloner VNET(pfsync_cloner)
324 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
326 struct pfsync_softc *sc;
328 struct pfsync_bucket *b;
334 if (! pfsync_buckets)
335 pfsync_buckets = mp_ncpus * 2;
337 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
338 sc->sc_flags |= PFSYNCF_OK;
339 sc->sc_maxupdates = 128;
341 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
346 if_initname(ifp, pfsyncname, unit);
348 ifp->if_ioctl = pfsyncioctl;
349 ifp->if_output = pfsyncoutput;
350 ifp->if_type = IFT_PFSYNC;
351 ifp->if_hdrlen = sizeof(struct pfsync_header);
352 ifp->if_mtu = ETHERMTU;
353 mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
354 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
355 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
356 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
360 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
362 sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets),
363 M_PFSYNC, M_ZERO | M_WAITOK);
364 for (c = 0; c < pfsync_buckets; c++) {
365 b = &sc->sc_buckets[c];
366 mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF);
370 b->b_len = PFSYNC_MINPKT;
372 for (q = 0; q < PFSYNC_S_COUNT; q++)
373 TAILQ_INIT(&b->b_qs[q]);
375 TAILQ_INIT(&b->b_upd_req_list);
376 TAILQ_INIT(&b->b_deferrals);
378 callout_init(&b->b_tmo, 1);
380 b->b_snd.ifq_maxlen = ifqmaxlen;
389 pfsync_clone_destroy(struct ifnet *ifp)
391 struct pfsync_softc *sc = ifp->if_softc;
392 struct pfsync_bucket *b;
395 for (c = 0; c < pfsync_buckets; c++) {
396 b = &sc->sc_buckets[c];
398 * At this stage, everything should have already been
399 * cleared by pfsync_uninit(), and we have only to
402 while (b->b_deferred > 0) {
403 struct pfsync_deferral *pd =
404 TAILQ_FIRST(&b->b_deferrals);
406 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
408 if (callout_stop(&pd->pd_tmo) > 0) {
409 pf_release_state(pd->pd_st);
414 callout_drain(&pd->pd_tmo);
419 callout_drain(&b->b_tmo);
422 callout_drain(&sc->sc_bulkfail_tmo);
423 callout_drain(&sc->sc_bulk_tmo);
425 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
426 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
433 pfsync_multicast_cleanup(sc);
434 mtx_destroy(&sc->sc_mtx);
435 mtx_destroy(&sc->sc_bulk_mtx);
437 free(sc->sc_buckets, M_PFSYNC);
444 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
445 struct pf_state_peer *d)
447 if (s->scrub.scrub_flag && d->scrub == NULL) {
448 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
449 if (d->scrub == NULL)
457 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
459 struct pfsync_softc *sc = V_pfsyncif;
460 #ifndef __NO_STRICT_ALIGNMENT
461 struct pfsync_state_key key[2];
463 struct pfsync_state_key *kw, *ks;
464 struct pf_state *st = NULL;
465 struct pf_state_key *skw = NULL, *sks = NULL;
466 struct pf_krule *r = NULL;
467 struct pfi_kkif *kif;
472 if (sp->creatorid == 0) {
473 if (V_pf_status.debug >= PF_DEBUG_MISC)
474 printf("%s: invalid creator id: %08x\n", __func__,
475 ntohl(sp->creatorid));
479 if ((kif = pfi_kkif_find(sp->ifname)) == NULL) {
480 if (V_pf_status.debug >= PF_DEBUG_MISC)
481 printf("%s: unknown interface: %s\n", __func__,
483 if (flags & PFSYNC_SI_IOCTL)
485 return (0); /* skip this state */
489 * If the ruleset checksums match or the state is coming from the ioctl,
490 * it's safe to associate the state with the rule of that number.
492 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
493 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
494 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
495 r = pf_main_ruleset.rules[
496 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
498 r = &V_pf_default_rule;
500 if ((r->max_states &&
501 counter_u64_fetch(r->states_cur) >= r->max_states))
505 * XXXGL: consider M_WAITOK in ioctl path after.
507 if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL)
510 for (int i = 0; i < 2; i++) {
511 st->packets[i] = counter_u64_alloc(M_NOWAIT);
512 st->bytes[i] = counter_u64_alloc(M_NOWAIT);
513 if (st->packets[i] == NULL || st->bytes[i] == NULL)
517 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
520 #ifndef __NO_STRICT_ALIGNMENT
521 bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2);
522 kw = &key[PF_SK_WIRE];
523 ks = &key[PF_SK_STACK];
525 kw = &sp->key[PF_SK_WIRE];
526 ks = &sp->key[PF_SK_STACK];
529 if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) ||
530 PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) ||
531 kw->port[0] != ks->port[0] ||
532 kw->port[1] != ks->port[1]) {
533 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
539 /* allocate memory for scrub info */
540 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
541 pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
544 /* Copy to state key(s). */
545 skw->addr[0] = kw->addr[0];
546 skw->addr[1] = kw->addr[1];
547 skw->port[0] = kw->port[0];
548 skw->port[1] = kw->port[1];
549 skw->proto = sp->proto;
552 sks->addr[0] = ks->addr[0];
553 sks->addr[1] = ks->addr[1];
554 sks->port[0] = ks->port[0];
555 sks->port[1] = ks->port[1];
556 sks->proto = sp->proto;
561 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
562 st->creation = time_uptime - ntohl(sp->creation);
563 st->expire = time_uptime;
567 timeout = r->timeout[sp->timeout];
569 timeout = V_pf_default_rule.timeout[sp->timeout];
571 /* sp->expire may have been adaptively scaled by export. */
572 st->expire -= timeout - ntohl(sp->expire);
575 st->direction = sp->direction;
577 st->timeout = sp->timeout;
578 st->state_flags = sp->state_flags;
581 st->creatorid = sp->creatorid;
582 pf_state_peer_ntoh(&sp->src, &st->src);
583 pf_state_peer_ntoh(&sp->dst, &st->dst);
586 st->nat_rule.ptr = NULL;
587 st->anchor.ptr = NULL;
590 st->pfsync_time = time_uptime;
591 st->sync_state = PFSYNC_S_NONE;
593 if (!(flags & PFSYNC_SI_IOCTL))
594 st->state_flags |= PFSTATE_NOSYNC;
596 if ((error = pf_state_insert(kif, skw, sks, st)) != 0)
599 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
600 counter_u64_add(r->states_cur, 1);
601 counter_u64_add(r->states_tot, 1);
603 if (!(flags & PFSYNC_SI_IOCTL)) {
604 st->state_flags &= ~PFSTATE_NOSYNC;
605 if (st->state_flags & PFSTATE_ACK) {
606 pfsync_q_ins(st, PFSYNC_S_IACK, true);
610 st->state_flags &= ~PFSTATE_ACK;
620 uma_zfree(V_pf_state_key_z, skw);
622 uma_zfree(V_pf_state_key_z, sks);
624 cleanup_state: /* pf_state_insert() frees the state keys. */
626 for (int i = 0; i < 2; i++) {
627 if (st->packets[i] != NULL)
628 counter_u64_free(st->packets[i]);
629 if (st->bytes[i] != NULL)
630 counter_u64_free(st->bytes[i]);
633 uma_zfree(V_pf_state_scrub_z, st->dst.scrub);
635 uma_zfree(V_pf_state_scrub_z, st->src.scrub);
636 uma_zfree(V_pf_state_z, st);
642 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
644 struct pfsync_softc *sc = V_pfsyncif;
645 struct pfsync_pkt pkt;
646 struct mbuf *m = *mp;
647 struct ip *ip = mtod(m, struct ip *);
648 struct pfsync_header *ph;
649 struct pfsync_subheader subh;
655 PF_RULES_RLOCK_TRACKER;
658 V_pfsyncstats.pfsyncs_ipackets++;
660 /* Verify that we have a sync interface configured. */
661 if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
662 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
665 /* verify that the packet came in on the right interface */
666 if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
667 V_pfsyncstats.pfsyncs_badif++;
671 if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
672 if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
673 /* verify that the IP TTL is 255. */
674 if (ip->ip_ttl != PFSYNC_DFLTTL) {
675 V_pfsyncstats.pfsyncs_badttl++;
679 offset = ip->ip_hl << 2;
680 if (m->m_pkthdr.len < offset + sizeof(*ph)) {
681 V_pfsyncstats.pfsyncs_hdrops++;
685 if (offset + sizeof(*ph) > m->m_len) {
686 if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
687 V_pfsyncstats.pfsyncs_hdrops++;
688 return (IPPROTO_DONE);
690 ip = mtod(m, struct ip *);
692 ph = (struct pfsync_header *)((char *)ip + offset);
694 /* verify the version */
695 if (ph->version != PFSYNC_VERSION) {
696 V_pfsyncstats.pfsyncs_badver++;
700 len = ntohs(ph->len) + offset;
701 if (m->m_pkthdr.len < len) {
702 V_pfsyncstats.pfsyncs_badlen++;
706 /* Cheaper to grab this now than having to mess with mbufs later */
708 pkt.src = ip->ip_src;
712 * Trusting pf_chksum during packet processing, as well as seeking
713 * in interface name tree, require holding PF_RULES_RLOCK().
716 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
717 pkt.flags |= PFSYNC_SI_CKSUM;
719 offset += sizeof(*ph);
720 while (offset <= len - sizeof(subh)) {
721 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
722 offset += sizeof(subh);
724 if (subh.action >= PFSYNC_ACT_MAX) {
725 V_pfsyncstats.pfsyncs_badact++;
730 count = ntohs(subh.count);
731 V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
732 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count);
735 return (IPPROTO_DONE);
744 return (IPPROTO_DONE);
748 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
750 struct pfsync_clr *clr;
752 int len = sizeof(*clr) * count;
756 mp = m_pulldown(m, offset, len, &offp);
758 V_pfsyncstats.pfsyncs_badlen++;
761 clr = (struct pfsync_clr *)(mp->m_data + offp);
763 for (i = 0; i < count; i++) {
764 creatorid = clr[i].creatorid;
766 if (clr[i].ifname[0] != '\0' &&
767 pfi_kkif_find(clr[i].ifname) == NULL)
770 for (int i = 0; i <= pf_hashmask; i++) {
771 struct pf_idhash *ih = &V_pf_idhash[i];
775 LIST_FOREACH(s, &ih->states, entry) {
776 if (s->creatorid == creatorid) {
777 s->state_flags |= PFSTATE_NOSYNC;
778 pf_unlink_state(s, PF_ENTER_LOCKED);
782 PF_HASHROW_UNLOCK(ih);
790 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
793 struct pfsync_state *sa, *sp;
794 int len = sizeof(*sp) * count;
797 mp = m_pulldown(m, offset, len, &offp);
799 V_pfsyncstats.pfsyncs_badlen++;
802 sa = (struct pfsync_state *)(mp->m_data + offp);
804 for (i = 0; i < count; i++) {
807 /* Check for invalid values. */
808 if (sp->timeout >= PFTM_MAX ||
809 sp->src.state > PF_TCPS_PROXY_DST ||
810 sp->dst.state > PF_TCPS_PROXY_DST ||
811 sp->direction > PF_OUT ||
812 (sp->af != AF_INET && sp->af != AF_INET6)) {
813 if (V_pf_status.debug >= PF_DEBUG_MISC)
814 printf("%s: invalid value\n", __func__);
815 V_pfsyncstats.pfsyncs_badval++;
819 if (pfsync_state_import(sp, pkt->flags) == ENOMEM)
820 /* Drop out, but process the rest of the actions. */
828 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
830 struct pfsync_ins_ack *ia, *iaa;
834 int len = count * sizeof(*ia);
837 mp = m_pulldown(m, offset, len, &offp);
839 V_pfsyncstats.pfsyncs_badlen++;
842 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
844 for (i = 0; i < count; i++) {
847 st = pf_find_state_byid(ia->id, ia->creatorid);
851 if (st->state_flags & PFSTATE_ACK) {
852 pfsync_undefer_state(st, 0);
857 * XXX this is not yet implemented, but we know the size of the
858 * message so we can skip it.
861 return (count * sizeof(struct pfsync_ins_ack));
865 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src,
866 struct pfsync_state_peer *dst)
870 PF_STATE_LOCK_ASSERT(st);
873 * The state should never go backwards except
874 * for syn-proxy states. Neither should the
875 * sequence window slide backwards.
877 if ((st->src.state > src->state &&
878 (st->src.state < PF_TCPS_PROXY_SRC ||
879 src->state >= PF_TCPS_PROXY_SRC)) ||
881 (st->src.state == src->state &&
882 SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
885 pf_state_peer_ntoh(src, &st->src);
887 if ((st->dst.state > dst->state) ||
889 (st->dst.state >= TCPS_SYN_SENT &&
890 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
893 pf_state_peer_ntoh(dst, &st->dst);
899 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
901 struct pfsync_softc *sc = V_pfsyncif;
902 struct pfsync_state *sa, *sp;
907 int len = count * sizeof(*sp);
910 mp = m_pulldown(m, offset, len, &offp);
912 V_pfsyncstats.pfsyncs_badlen++;
915 sa = (struct pfsync_state *)(mp->m_data + offp);
917 for (i = 0; i < count; i++) {
920 /* check for invalid values */
921 if (sp->timeout >= PFTM_MAX ||
922 sp->src.state > PF_TCPS_PROXY_DST ||
923 sp->dst.state > PF_TCPS_PROXY_DST) {
924 if (V_pf_status.debug >= PF_DEBUG_MISC) {
925 printf("pfsync_input: PFSYNC_ACT_UPD: "
928 V_pfsyncstats.pfsyncs_badval++;
932 st = pf_find_state_byid(sp->id, sp->creatorid);
934 /* insert the update */
935 if (pfsync_state_import(sp, pkt->flags))
936 V_pfsyncstats.pfsyncs_badstate++;
940 if (st->state_flags & PFSTATE_ACK) {
941 pfsync_undefer_state(st, 1);
944 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
945 sync = pfsync_upd_tcp(st, &sp->src, &sp->dst);
950 * Non-TCP protocol state machine always go
953 if (st->src.state > sp->src.state)
956 pf_state_peer_ntoh(&sp->src, &st->src);
957 if (st->dst.state > sp->dst.state)
960 pf_state_peer_ntoh(&sp->dst, &st->dst);
963 pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
964 pf_state_peer_ntoh(&sp->dst, &st->dst);
965 st->expire = time_uptime;
966 st->timeout = sp->timeout;
968 st->pfsync_time = time_uptime;
971 V_pfsyncstats.pfsyncs_stale++;
973 pfsync_update_state(st);
985 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
987 struct pfsync_softc *sc = V_pfsyncif;
988 struct pfsync_upd_c *ua, *up;
990 int len = count * sizeof(*up);
995 mp = m_pulldown(m, offset, len, &offp);
997 V_pfsyncstats.pfsyncs_badlen++;
1000 ua = (struct pfsync_upd_c *)(mp->m_data + offp);
1002 for (i = 0; i < count; i++) {
1005 /* check for invalid values */
1006 if (up->timeout >= PFTM_MAX ||
1007 up->src.state > PF_TCPS_PROXY_DST ||
1008 up->dst.state > PF_TCPS_PROXY_DST) {
1009 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1010 printf("pfsync_input: "
1011 "PFSYNC_ACT_UPD_C: "
1014 V_pfsyncstats.pfsyncs_badval++;
1018 st = pf_find_state_byid(up->id, up->creatorid);
1020 /* We don't have this state. Ask for it. */
1021 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1022 pfsync_request_update(up->creatorid, up->id);
1023 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1027 if (st->state_flags & PFSTATE_ACK) {
1028 pfsync_undefer_state(st, 1);
1031 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
1032 sync = pfsync_upd_tcp(st, &up->src, &up->dst);
1037 * Non-TCP protocol state machine always go
1040 if (st->src.state > up->src.state)
1043 pf_state_peer_ntoh(&up->src, &st->src);
1044 if (st->dst.state > up->dst.state)
1047 pf_state_peer_ntoh(&up->dst, &st->dst);
1050 pfsync_alloc_scrub_memory(&up->dst, &st->dst);
1051 pf_state_peer_ntoh(&up->dst, &st->dst);
1052 st->expire = time_uptime;
1053 st->timeout = up->timeout;
1055 st->pfsync_time = time_uptime;
1058 V_pfsyncstats.pfsyncs_stale++;
1060 pfsync_update_state(st);
1061 PF_STATE_UNLOCK(st);
1062 pfsync_push_all(sc);
1065 PF_STATE_UNLOCK(st);
1072 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1074 struct pfsync_upd_req *ur, *ura;
1076 int len = count * sizeof(*ur);
1079 struct pf_state *st;
1081 mp = m_pulldown(m, offset, len, &offp);
1083 V_pfsyncstats.pfsyncs_badlen++;
1086 ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1088 for (i = 0; i < count; i++) {
1091 if (ur->id == 0 && ur->creatorid == 0)
1092 pfsync_bulk_start();
1094 st = pf_find_state_byid(ur->id, ur->creatorid);
1096 V_pfsyncstats.pfsyncs_badstate++;
1099 if (st->state_flags & PFSTATE_NOSYNC) {
1100 PF_STATE_UNLOCK(st);
1104 pfsync_update_state_req(st);
1105 PF_STATE_UNLOCK(st);
1113 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1116 struct pfsync_state *sa, *sp;
1117 struct pf_state *st;
1118 int len = count * sizeof(*sp);
1121 mp = m_pulldown(m, offset, len, &offp);
1123 V_pfsyncstats.pfsyncs_badlen++;
1126 sa = (struct pfsync_state *)(mp->m_data + offp);
1128 for (i = 0; i < count; i++) {
1131 st = pf_find_state_byid(sp->id, sp->creatorid);
1133 V_pfsyncstats.pfsyncs_badstate++;
1136 st->state_flags |= PFSTATE_NOSYNC;
1137 pf_unlink_state(st, PF_ENTER_LOCKED);
1144 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1147 struct pfsync_del_c *sa, *sp;
1148 struct pf_state *st;
1149 int len = count * sizeof(*sp);
1152 mp = m_pulldown(m, offset, len, &offp);
1154 V_pfsyncstats.pfsyncs_badlen++;
1157 sa = (struct pfsync_del_c *)(mp->m_data + offp);
1159 for (i = 0; i < count; i++) {
1162 st = pf_find_state_byid(sp->id, sp->creatorid);
1164 V_pfsyncstats.pfsyncs_badstate++;
1168 st->state_flags |= PFSTATE_NOSYNC;
1169 pf_unlink_state(st, PF_ENTER_LOCKED);
1176 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1178 struct pfsync_softc *sc = V_pfsyncif;
1179 struct pfsync_bus *bus;
1181 int len = count * sizeof(*bus);
1186 /* If we're not waiting for a bulk update, who cares. */
1187 if (sc->sc_ureq_sent == 0) {
1192 mp = m_pulldown(m, offset, len, &offp);
1195 V_pfsyncstats.pfsyncs_badlen++;
1198 bus = (struct pfsync_bus *)(mp->m_data + offp);
1200 switch (bus->status) {
1201 case PFSYNC_BUS_START:
1202 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1203 V_pf_limits[PF_LIMIT_STATES].limit /
1204 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1205 sizeof(struct pfsync_state)),
1206 pfsync_bulk_fail, sc);
1207 if (V_pf_status.debug >= PF_DEBUG_MISC)
1208 printf("pfsync: received bulk update start\n");
1211 case PFSYNC_BUS_END:
1212 if (time_uptime - ntohl(bus->endtime) >=
1214 /* that's it, we're happy */
1215 sc->sc_ureq_sent = 0;
1216 sc->sc_bulk_tries = 0;
1217 callout_stop(&sc->sc_bulkfail_tmo);
1218 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1219 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
1220 "pfsync bulk done");
1221 sc->sc_flags |= PFSYNCF_OK;
1222 if (V_pf_status.debug >= PF_DEBUG_MISC)
1223 printf("pfsync: received valid "
1224 "bulk update end\n");
1226 if (V_pf_status.debug >= PF_DEBUG_MISC)
1227 printf("pfsync: received invalid "
1228 "bulk update end: bad timestamp\n");
1238 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1240 int len = count * sizeof(struct pfsync_tdb);
1243 struct pfsync_tdb *tp;
1249 mp = m_pulldown(m, offset, len, &offp);
1251 V_pfsyncstats.pfsyncs_badlen++;
1254 tp = (struct pfsync_tdb *)(mp->m_data + offp);
1256 for (i = 0; i < count; i++)
1257 pfsync_update_net_tdb(&tp[i]);
1264 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1266 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1271 /* check for invalid values */
1272 if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1273 (pt->dst.sa.sa_family != AF_INET &&
1274 pt->dst.sa.sa_family != AF_INET6))
1277 tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1279 pt->rpl = ntohl(pt->rpl);
1280 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1282 /* Neither replay nor byte counter should ever decrease. */
1283 if (pt->rpl < tdb->tdb_rpl ||
1284 pt->cur_bytes < tdb->tdb_cur_bytes) {
1288 tdb->tdb_rpl = pt->rpl;
1289 tdb->tdb_cur_bytes = pt->cur_bytes;
1294 if (V_pf_status.debug >= PF_DEBUG_MISC)
1295 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1297 V_pfsyncstats.pfsyncs_badstate++;
1303 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1305 /* check if we are at the right place in the packet */
1306 if (offset != m->m_pkthdr.len)
1307 V_pfsyncstats.pfsyncs_badlen++;
1309 /* we're done. free and let the caller return */
1315 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1317 V_pfsyncstats.pfsyncs_badact++;
1324 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1333 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1335 struct pfsync_softc *sc = ifp->if_softc;
1336 struct ifreq *ifr = (struct ifreq *)data;
1337 struct pfsyncreq pfsyncr;
1344 if (ifp->if_flags & IFF_UP) {
1345 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1347 pfsync_pointers_init();
1349 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1351 pfsync_pointers_uninit();
1355 if (!sc->sc_sync_if ||
1356 ifr->ifr_mtu <= PFSYNC_MINPKT ||
1357 ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1359 if (ifr->ifr_mtu < ifp->if_mtu) {
1360 for (c = 0; c < pfsync_buckets; c++) {
1361 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1362 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT)
1363 pfsync_sendout(1, c);
1364 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1367 ifp->if_mtu = ifr->ifr_mtu;
1370 bzero(&pfsyncr, sizeof(pfsyncr));
1372 if (sc->sc_sync_if) {
1373 strlcpy(pfsyncr.pfsyncr_syncdev,
1374 sc->sc_sync_if->if_xname, IFNAMSIZ);
1376 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1377 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1378 pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER ==
1379 (sc->sc_flags & PFSYNCF_DEFER));
1381 return (copyout(&pfsyncr, ifr_data_get_ptr(ifr),
1386 struct in_mfilter *imf = NULL;
1390 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1392 if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr,
1396 if (pfsyncr.pfsyncr_maxupdates > 255)
1399 if (pfsyncr.pfsyncr_syncdev[0] == 0)
1401 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL)
1404 if (sifp != NULL && (
1405 pfsyncr.pfsyncr_syncpeer.s_addr == 0 ||
1406 pfsyncr.pfsyncr_syncpeer.s_addr ==
1407 htonl(INADDR_PFSYNC_GROUP)))
1408 imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
1411 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1412 sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP);
1414 sc->sc_sync_peer.s_addr =
1415 pfsyncr.pfsyncr_syncpeer.s_addr;
1417 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1418 if (pfsyncr.pfsyncr_defer) {
1419 sc->sc_flags |= PFSYNCF_DEFER;
1420 V_pfsync_defer_ptr = pfsync_defer;
1422 sc->sc_flags &= ~PFSYNCF_DEFER;
1423 V_pfsync_defer_ptr = NULL;
1428 if_rele(sc->sc_sync_if);
1429 sc->sc_sync_if = NULL;
1430 pfsync_multicast_cleanup(sc);
1435 for (c = 0; c < pfsync_buckets; c++) {
1436 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]);
1437 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT &&
1438 (sifp->if_mtu < sc->sc_ifp->if_mtu ||
1439 (sc->sc_sync_if != NULL &&
1440 sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
1441 sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
1442 pfsync_sendout(1, c);
1443 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]);
1446 pfsync_multicast_cleanup(sc);
1448 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
1449 error = pfsync_multicast_setup(sc, sifp, imf);
1452 ip_mfilter_free(imf);
1458 if_rele(sc->sc_sync_if);
1459 sc->sc_sync_if = sifp;
1461 ip = &sc->sc_template;
1462 bzero(ip, sizeof(*ip));
1463 ip->ip_v = IPVERSION;
1464 ip->ip_hl = sizeof(sc->sc_template) >> 2;
1465 ip->ip_tos = IPTOS_LOWDELAY;
1466 /* len and id are set later. */
1467 ip->ip_off = htons(IP_DF);
1468 ip->ip_ttl = PFSYNC_DFLTTL;
1469 ip->ip_p = IPPROTO_PFSYNC;
1470 ip->ip_src.s_addr = INADDR_ANY;
1471 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr;
1473 /* Request a full state table update. */
1474 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1475 (*carp_demote_adj_p)(V_pfsync_carp_adj,
1476 "pfsync bulk start");
1477 sc->sc_flags &= ~PFSYNCF_OK;
1478 if (V_pf_status.debug >= PF_DEBUG_MISC)
1479 printf("pfsync: requesting bulk update\n");
1481 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]);
1482 pfsync_request_update(0, 0);
1483 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]);
1485 sc->sc_ureq_sent = time_uptime;
1486 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail,
1500 pfsync_out_state(struct pf_state *st, void *buf)
1502 struct pfsync_state *sp = buf;
1504 pfsync_state_export(sp, st);
1508 pfsync_out_iack(struct pf_state *st, void *buf)
1510 struct pfsync_ins_ack *iack = buf;
1513 iack->creatorid = st->creatorid;
1517 pfsync_out_upd_c(struct pf_state *st, void *buf)
1519 struct pfsync_upd_c *up = buf;
1521 bzero(up, sizeof(*up));
1523 pf_state_peer_hton(&st->src, &up->src);
1524 pf_state_peer_hton(&st->dst, &up->dst);
1525 up->creatorid = st->creatorid;
1526 up->timeout = st->timeout;
1530 pfsync_out_del(struct pf_state *st, void *buf)
1532 struct pfsync_del_c *dp = buf;
1535 dp->creatorid = st->creatorid;
1536 st->state_flags |= PFSTATE_NOSYNC;
1540 pfsync_drop(struct pfsync_softc *sc)
1542 struct pf_state *st, *next;
1543 struct pfsync_upd_req_item *ur;
1544 struct pfsync_bucket *b;
1547 for (c = 0; c < pfsync_buckets; c++) {
1548 b = &sc->sc_buckets[c];
1549 for (q = 0; q < PFSYNC_S_COUNT; q++) {
1550 if (TAILQ_EMPTY(&b->b_qs[q]))
1553 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) {
1554 KASSERT(st->sync_state == q,
1555 ("%s: st->sync_state == q",
1557 st->sync_state = PFSYNC_S_NONE;
1558 pf_release_state(st);
1560 TAILQ_INIT(&b->b_qs[q]);
1563 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1564 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1568 b->b_len = PFSYNC_MINPKT;
1574 pfsync_sendout(int schedswi, int c)
1576 struct pfsync_softc *sc = V_pfsyncif;
1577 struct ifnet *ifp = sc->sc_ifp;
1580 struct pfsync_header *ph;
1581 struct pfsync_subheader *subh;
1582 struct pf_state *st, *st_next;
1583 struct pfsync_upd_req_item *ur;
1584 struct pfsync_bucket *b = &sc->sc_buckets[c];
1588 KASSERT(sc != NULL, ("%s: null sc", __func__));
1589 KASSERT(b->b_len > PFSYNC_MINPKT,
1590 ("%s: sc_len %zu", __func__, b->b_len));
1591 PFSYNC_BUCKET_LOCK_ASSERT(b);
1593 if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1598 m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1600 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1601 V_pfsyncstats.pfsyncs_onomem++;
1604 m->m_data += max_linkhdr;
1605 m->m_len = m->m_pkthdr.len = b->b_len;
1607 /* build the ip header */
1608 ip = (struct ip *)m->m_data;
1609 bcopy(&sc->sc_template, ip, sizeof(*ip));
1610 offset = sizeof(*ip);
1612 ip->ip_len = htons(m->m_pkthdr.len);
1615 /* build the pfsync header */
1616 ph = (struct pfsync_header *)(m->m_data + offset);
1617 bzero(ph, sizeof(*ph));
1618 offset += sizeof(*ph);
1620 ph->version = PFSYNC_VERSION;
1621 ph->len = htons(b->b_len - sizeof(*ip));
1622 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1624 /* walk the queues */
1625 for (q = 0; q < PFSYNC_S_COUNT; q++) {
1626 if (TAILQ_EMPTY(&b->b_qs[q]))
1629 subh = (struct pfsync_subheader *)(m->m_data + offset);
1630 offset += sizeof(*subh);
1633 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) {
1634 KASSERT(st->sync_state == q,
1635 ("%s: st->sync_state == q",
1638 * XXXGL: some of write methods do unlocked reads
1641 pfsync_qs[q].write(st, m->m_data + offset);
1642 offset += pfsync_qs[q].len;
1643 st->sync_state = PFSYNC_S_NONE;
1644 pf_release_state(st);
1647 TAILQ_INIT(&b->b_qs[q]);
1649 bzero(subh, sizeof(*subh));
1650 subh->action = pfsync_qs[q].action;
1651 subh->count = htons(count);
1652 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1655 if (!TAILQ_EMPTY(&b->b_upd_req_list)) {
1656 subh = (struct pfsync_subheader *)(m->m_data + offset);
1657 offset += sizeof(*subh);
1660 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) {
1661 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry);
1663 bcopy(&ur->ur_msg, m->m_data + offset,
1664 sizeof(ur->ur_msg));
1665 offset += sizeof(ur->ur_msg);
1670 bzero(subh, sizeof(*subh));
1671 subh->action = PFSYNC_ACT_UPD_REQ;
1672 subh->count = htons(count);
1673 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1676 /* has someone built a custom region for us to add? */
1677 if (b->b_plus != NULL) {
1678 bcopy(b->b_plus, m->m_data + offset, b->b_pluslen);
1679 offset += b->b_pluslen;
1684 subh = (struct pfsync_subheader *)(m->m_data + offset);
1685 offset += sizeof(*subh);
1687 bzero(subh, sizeof(*subh));
1688 subh->action = PFSYNC_ACT_EOF;
1689 subh->count = htons(1);
1690 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1692 /* we're done, let's put it on the wire */
1694 m->m_data += sizeof(*ip);
1695 m->m_len = m->m_pkthdr.len = b->b_len - sizeof(*ip);
1697 m->m_data -= sizeof(*ip);
1698 m->m_len = m->m_pkthdr.len = b->b_len;
1701 if (sc->sc_sync_if == NULL) {
1702 b->b_len = PFSYNC_MINPKT;
1707 if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1708 if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1709 b->b_len = PFSYNC_MINPKT;
1711 if (!_IF_QFULL(&b->b_snd))
1712 _IF_ENQUEUE(&b->b_snd, m);
1715 if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1718 swi_sched(V_pfsync_swi_cookie, 0);
1722 pfsync_insert_state(struct pf_state *st)
1724 struct pfsync_softc *sc = V_pfsyncif;
1725 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1727 if (st->state_flags & PFSTATE_NOSYNC)
1730 if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1731 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1732 st->state_flags |= PFSTATE_NOSYNC;
1736 KASSERT(st->sync_state == PFSYNC_S_NONE,
1737 ("%s: st->sync_state %u", __func__, st->sync_state));
1739 PFSYNC_BUCKET_LOCK(b);
1740 if (b->b_len == PFSYNC_MINPKT)
1741 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1743 pfsync_q_ins(st, PFSYNC_S_INS, true);
1744 PFSYNC_BUCKET_UNLOCK(b);
1746 st->sync_updates = 0;
1750 pfsync_defer(struct pf_state *st, struct mbuf *m)
1752 struct pfsync_softc *sc = V_pfsyncif;
1753 struct pfsync_deferral *pd;
1754 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1756 if (m->m_flags & (M_BCAST|M_MCAST))
1761 if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) ||
1762 !(sc->sc_flags & PFSYNCF_DEFER)) {
1767 if (b->b_deferred >= 128)
1768 pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0);
1770 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1775 m->m_flags |= M_SKIP_FIREWALL;
1776 st->state_flags |= PFSTATE_ACK;
1784 TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry);
1785 callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED);
1786 callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd);
1794 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1796 struct pfsync_softc *sc = pd->pd_sc;
1797 struct mbuf *m = pd->pd_m;
1798 struct pf_state *st = pd->pd_st;
1799 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1801 PFSYNC_BUCKET_LOCK_ASSERT(b);
1803 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
1805 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1807 pf_release_state(st);
1812 _IF_ENQUEUE(&b->b_snd, m);
1818 pfsync_defer_tmo(void *arg)
1820 struct epoch_tracker et;
1821 struct pfsync_deferral *pd = arg;
1822 struct pfsync_softc *sc = pd->pd_sc;
1823 struct mbuf *m = pd->pd_m;
1824 struct pf_state *st = pd->pd_st;
1825 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1827 PFSYNC_BUCKET_LOCK_ASSERT(b);
1829 NET_EPOCH_ENTER(et);
1830 CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
1832 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry);
1834 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1835 if (pd->pd_refs == 0)
1839 ip_output(m, NULL, NULL, 0, NULL, NULL);
1841 pf_release_state(st);
1848 pfsync_undefer_state(struct pf_state *st, int drop)
1850 struct pfsync_softc *sc = V_pfsyncif;
1851 struct pfsync_deferral *pd;
1852 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1854 PFSYNC_BUCKET_LOCK(b);
1856 TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) {
1857 if (pd->pd_st == st) {
1858 if (callout_stop(&pd->pd_tmo) > 0)
1859 pfsync_undefer(pd, drop);
1861 PFSYNC_BUCKET_UNLOCK(b);
1865 PFSYNC_BUCKET_UNLOCK(b);
1867 panic("%s: unable to find deferred state", __func__);
1870 static struct pfsync_bucket*
1871 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_state *st)
1873 int c = PF_IDHASH(st) % pfsync_buckets;
1874 return &sc->sc_buckets[c];
1878 pfsync_update_state(struct pf_state *st)
1880 struct pfsync_softc *sc = V_pfsyncif;
1881 bool sync = false, ref = true;
1882 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1884 PF_STATE_LOCK_ASSERT(st);
1885 PFSYNC_BUCKET_LOCK(b);
1887 if (st->state_flags & PFSTATE_ACK)
1888 pfsync_undefer_state(st, 0);
1889 if (st->state_flags & PFSTATE_NOSYNC) {
1890 if (st->sync_state != PFSYNC_S_NONE)
1891 pfsync_q_del(st, true, b);
1892 PFSYNC_BUCKET_UNLOCK(b);
1896 if (b->b_len == PFSYNC_MINPKT)
1897 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
1899 switch (st->sync_state) {
1900 case PFSYNC_S_UPD_C:
1903 /* we're already handling it */
1905 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1907 if (st->sync_updates >= sc->sc_maxupdates)
1913 pfsync_q_del(st, false, b);
1918 pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
1919 st->sync_updates = 0;
1923 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1926 if (sync || (time_uptime - st->pfsync_time) < 2)
1929 PFSYNC_BUCKET_UNLOCK(b);
1933 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1935 struct pfsync_softc *sc = V_pfsyncif;
1936 struct pfsync_bucket *b = &sc->sc_buckets[0];
1937 struct pfsync_upd_req_item *item;
1938 size_t nlen = sizeof(struct pfsync_upd_req);
1940 PFSYNC_BUCKET_LOCK_ASSERT(b);
1943 * This code does a bit to prevent multiple update requests for the
1944 * same state being generated. It searches current subheader queue,
1945 * but it doesn't lookup into queue of already packed datagrams.
1947 TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry)
1948 if (item->ur_msg.id == id &&
1949 item->ur_msg.creatorid == creatorid)
1952 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1954 return; /* XXX stats */
1956 item->ur_msg.id = id;
1957 item->ur_msg.creatorid = creatorid;
1959 if (TAILQ_EMPTY(&b->b_upd_req_list))
1960 nlen += sizeof(struct pfsync_subheader);
1962 if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
1963 pfsync_sendout(1, 0);
1965 nlen = sizeof(struct pfsync_subheader) +
1966 sizeof(struct pfsync_upd_req);
1969 TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry);
1974 pfsync_update_state_req(struct pf_state *st)
1976 struct pfsync_softc *sc = V_pfsyncif;
1977 bool ref = true, full = false;
1978 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
1980 PF_STATE_LOCK_ASSERT(st);
1981 PFSYNC_BUCKET_LOCK(b);
1983 if (st->state_flags & PFSTATE_NOSYNC) {
1984 if (st->sync_state != PFSYNC_S_NONE)
1985 pfsync_q_del(st, true, b);
1986 PFSYNC_BUCKET_UNLOCK(b);
1990 switch (st->sync_state) {
1991 case PFSYNC_S_UPD_C:
1993 pfsync_q_del(st, false, b);
1998 pfsync_q_ins(st, PFSYNC_S_UPD, ref);
2005 /* we're already handling it */
2009 panic("%s: unexpected sync state %d", __func__, st->sync_state);
2012 if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(struct pfsync_state))
2015 PFSYNC_BUCKET_UNLOCK(b);
2021 pfsync_delete_state(struct pf_state *st)
2023 struct pfsync_softc *sc = V_pfsyncif;
2024 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2027 PFSYNC_BUCKET_LOCK(b);
2028 if (st->state_flags & PFSTATE_ACK)
2029 pfsync_undefer_state(st, 1);
2030 if (st->state_flags & PFSTATE_NOSYNC) {
2031 if (st->sync_state != PFSYNC_S_NONE)
2032 pfsync_q_del(st, true, b);
2033 PFSYNC_BUCKET_UNLOCK(b);
2037 if (b->b_len == PFSYNC_MINPKT)
2038 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b);
2040 switch (st->sync_state) {
2042 /* We never got to tell the world so just forget about it. */
2043 pfsync_q_del(st, true, b);
2046 case PFSYNC_S_UPD_C:
2049 pfsync_q_del(st, false, b);
2054 pfsync_q_ins(st, PFSYNC_S_DEL, ref);
2058 panic("%s: unexpected sync state %d", __func__, st->sync_state);
2061 PFSYNC_BUCKET_UNLOCK(b);
2065 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
2068 struct pfsync_subheader subh;
2069 struct pfsync_clr clr;
2072 bzero(&r, sizeof(r));
2074 r.subh.action = PFSYNC_ACT_CLR;
2075 r.subh.count = htons(1);
2076 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
2078 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
2079 r.clr.creatorid = creatorid;
2081 pfsync_send_plus(&r, sizeof(r));
2085 pfsync_q_ins(struct pf_state *st, int q, bool ref)
2087 struct pfsync_softc *sc = V_pfsyncif;
2088 size_t nlen = pfsync_qs[q].len;
2089 struct pfsync_bucket *b = pfsync_get_bucket(sc, st);
2091 PFSYNC_BUCKET_LOCK_ASSERT(b);
2093 KASSERT(st->sync_state == PFSYNC_S_NONE,
2094 ("%s: st->sync_state %u", __func__, st->sync_state));
2095 KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2098 if (TAILQ_EMPTY(&b->b_qs[q]))
2099 nlen += sizeof(struct pfsync_subheader);
2101 if (b->b_len + nlen > sc->sc_ifp->if_mtu) {
2102 pfsync_sendout(1, b->b_id);
2104 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2108 TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list);
2115 pfsync_q_del(struct pf_state *st, bool unref, struct pfsync_bucket *b)
2117 int q = st->sync_state;
2119 PFSYNC_BUCKET_LOCK_ASSERT(b);
2120 KASSERT(st->sync_state != PFSYNC_S_NONE,
2121 ("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2123 b->b_len -= pfsync_qs[q].len;
2124 TAILQ_REMOVE(&b->b_qs[q], st, sync_list);
2125 st->sync_state = PFSYNC_S_NONE;
2127 pf_release_state(st);
2129 if (TAILQ_EMPTY(&b->b_qs[q]))
2130 b->b_len -= sizeof(struct pfsync_subheader);
2134 pfsync_bulk_start(void)
2136 struct pfsync_softc *sc = V_pfsyncif;
2138 if (V_pf_status.debug >= PF_DEBUG_MISC)
2139 printf("pfsync: received bulk update request\n");
2143 sc->sc_ureq_received = time_uptime;
2144 sc->sc_bulk_hashid = 0;
2145 sc->sc_bulk_stateid = 0;
2146 pfsync_bulk_status(PFSYNC_BUS_START);
2147 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2152 pfsync_bulk_update(void *arg)
2154 struct pfsync_softc *sc = arg;
2158 PFSYNC_BLOCK_ASSERT(sc);
2159 CURVNET_SET(sc->sc_ifp->if_vnet);
2162 * Start with last state from previous invocation.
2163 * It may had gone, in this case start from the
2166 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2171 i = sc->sc_bulk_hashid;
2173 for (; i <= pf_hashmask; i++) {
2174 struct pf_idhash *ih = &V_pf_idhash[i];
2177 PF_HASHROW_ASSERT(ih);
2179 PF_HASHROW_LOCK(ih);
2180 s = LIST_FIRST(&ih->states);
2183 for (; s; s = LIST_NEXT(s, entry)) {
2184 if (s->sync_state == PFSYNC_S_NONE &&
2185 s->timeout < PFTM_MAX &&
2186 s->pfsync_time <= sc->sc_ureq_received) {
2187 if (pfsync_update_state_req(s)) {
2188 /* We've filled a packet. */
2189 sc->sc_bulk_hashid = i;
2190 sc->sc_bulk_stateid = s->id;
2191 sc->sc_bulk_creatorid = s->creatorid;
2192 PF_HASHROW_UNLOCK(ih);
2193 callout_reset(&sc->sc_bulk_tmo, 1,
2194 pfsync_bulk_update, sc);
2200 PF_HASHROW_UNLOCK(ih);
2204 pfsync_bulk_status(PFSYNC_BUS_END);
2210 pfsync_bulk_status(u_int8_t status)
2213 struct pfsync_subheader subh;
2214 struct pfsync_bus bus;
2217 struct pfsync_softc *sc = V_pfsyncif;
2219 bzero(&r, sizeof(r));
2221 r.subh.action = PFSYNC_ACT_BUS;
2222 r.subh.count = htons(1);
2223 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2225 r.bus.creatorid = V_pf_status.hostid;
2226 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2227 r.bus.status = status;
2229 pfsync_send_plus(&r, sizeof(r));
2233 pfsync_bulk_fail(void *arg)
2235 struct pfsync_softc *sc = arg;
2236 struct pfsync_bucket *b = &sc->sc_buckets[0];
2238 CURVNET_SET(sc->sc_ifp->if_vnet);
2240 PFSYNC_BLOCK_ASSERT(sc);
2242 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2244 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2245 pfsync_bulk_fail, V_pfsyncif);
2246 PFSYNC_BUCKET_LOCK(b);
2247 pfsync_request_update(0, 0);
2248 PFSYNC_BUCKET_UNLOCK(b);
2250 /* Pretend like the transfer was ok. */
2251 sc->sc_ureq_sent = 0;
2252 sc->sc_bulk_tries = 0;
2254 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2255 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
2256 "pfsync bulk fail");
2257 sc->sc_flags |= PFSYNCF_OK;
2259 if (V_pf_status.debug >= PF_DEBUG_MISC)
2260 printf("pfsync: failed to receive bulk update\n");
2267 pfsync_send_plus(void *plus, size_t pluslen)
2269 struct pfsync_softc *sc = V_pfsyncif;
2270 struct pfsync_bucket *b = &sc->sc_buckets[0];
2272 PFSYNC_BUCKET_LOCK(b);
2274 if (b->b_len + pluslen > sc->sc_ifp->if_mtu)
2275 pfsync_sendout(1, b->b_id);
2278 b->b_len += (b->b_pluslen = pluslen);
2280 pfsync_sendout(1, b->b_id);
2281 PFSYNC_BUCKET_UNLOCK(b);
2285 pfsync_timeout(void *arg)
2287 struct pfsync_bucket *b = arg;
2289 CURVNET_SET(b->b_sc->sc_ifp->if_vnet);
2290 PFSYNC_BUCKET_LOCK(b);
2292 PFSYNC_BUCKET_UNLOCK(b);
2297 pfsync_push(struct pfsync_bucket *b)
2300 PFSYNC_BUCKET_LOCK_ASSERT(b);
2302 b->b_flags |= PFSYNCF_BUCKET_PUSH;
2303 swi_sched(V_pfsync_swi_cookie, 0);
2307 pfsync_push_all(struct pfsync_softc *sc)
2310 struct pfsync_bucket *b;
2312 for (c = 0; c < pfsync_buckets; c++) {
2313 b = &sc->sc_buckets[c];
2315 PFSYNC_BUCKET_LOCK(b);
2317 PFSYNC_BUCKET_UNLOCK(b);
2322 pfsyncintr(void *arg)
2324 struct epoch_tracker et;
2325 struct pfsync_softc *sc = arg;
2326 struct pfsync_bucket *b;
2330 NET_EPOCH_ENTER(et);
2331 CURVNET_SET(sc->sc_ifp->if_vnet);
2333 for (c = 0; c < pfsync_buckets; c++) {
2334 b = &sc->sc_buckets[c];
2336 PFSYNC_BUCKET_LOCK(b);
2337 if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) {
2338 pfsync_sendout(0, b->b_id);
2339 b->b_flags &= ~PFSYNCF_BUCKET_PUSH;
2341 _IF_DEQUEUE_ALL(&b->b_snd, m);
2342 PFSYNC_BUCKET_UNLOCK(b);
2344 for (; m != NULL; m = n) {
2346 m->m_nextpkt = NULL;
2349 * We distinguish between a deferral packet and our
2350 * own pfsync packet based on M_SKIP_FIREWALL
2351 * flag. This is XXX.
2353 if (m->m_flags & M_SKIP_FIREWALL)
2354 ip_output(m, NULL, NULL, 0, NULL, NULL);
2355 else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo,
2357 V_pfsyncstats.pfsyncs_opackets++;
2359 V_pfsyncstats.pfsyncs_oerrors++;
2367 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp,
2368 struct in_mfilter *imf)
2370 struct ip_moptions *imo = &sc->sc_imo;
2373 if (!(ifp->if_flags & IFF_MULTICAST))
2374 return (EADDRNOTAVAIL);
2376 imo->imo_multicast_vif = -1;
2378 if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL,
2379 &imf->imf_inm)) != 0)
2382 ip_mfilter_init(&imo->imo_head);
2383 ip_mfilter_insert(&imo->imo_head, imf);
2384 imo->imo_multicast_ifp = ifp;
2385 imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2386 imo->imo_multicast_loop = 0;
2392 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2394 struct ip_moptions *imo = &sc->sc_imo;
2395 struct in_mfilter *imf;
2397 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
2398 ip_mfilter_remove(&imo->imo_head, imf);
2399 in_leavegroup(imf->imf_inm, NULL);
2400 ip_mfilter_free(imf);
2402 imo->imo_multicast_ifp = NULL;
2406 pfsync_detach_ifnet(struct ifnet *ifp)
2408 struct pfsync_softc *sc = V_pfsyncif;
2415 if (sc->sc_sync_if == ifp) {
2416 /* We don't need mutlicast cleanup here, because the interface
2417 * is going away. We do need to ensure we don't try to do
2420 ip_mfilter_init(&sc->sc_imo.imo_head);
2421 sc->sc_imo.imo_multicast_ifp = NULL;
2422 sc->sc_sync_if = NULL;
2429 extern struct domain inetdomain;
2430 static struct protosw in_pfsync_protosw = {
2431 .pr_type = SOCK_RAW,
2432 .pr_domain = &inetdomain,
2433 .pr_protocol = IPPROTO_PFSYNC,
2434 .pr_flags = PR_ATOMIC|PR_ADDR,
2435 .pr_input = pfsync_input,
2436 .pr_output = rip_output,
2437 .pr_ctloutput = rip_ctloutput,
2438 .pr_usrreqs = &rip_usrreqs
2443 pfsync_pointers_init()
2447 V_pfsync_state_import_ptr = pfsync_state_import;
2448 V_pfsync_insert_state_ptr = pfsync_insert_state;
2449 V_pfsync_update_state_ptr = pfsync_update_state;
2450 V_pfsync_delete_state_ptr = pfsync_delete_state;
2451 V_pfsync_clear_states_ptr = pfsync_clear_states;
2452 V_pfsync_defer_ptr = pfsync_defer;
2457 pfsync_pointers_uninit()
2461 V_pfsync_state_import_ptr = NULL;
2462 V_pfsync_insert_state_ptr = NULL;
2463 V_pfsync_update_state_ptr = NULL;
2464 V_pfsync_delete_state_ptr = NULL;
2465 V_pfsync_clear_states_ptr = NULL;
2466 V_pfsync_defer_ptr = NULL;
2471 vnet_pfsync_init(const void *unused __unused)
2475 V_pfsync_cloner = if_clone_simple(pfsyncname,
2476 pfsync_clone_create, pfsync_clone_destroy, 1);
2477 error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif,
2478 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2480 if_clone_detach(V_pfsync_cloner);
2481 log(LOG_INFO, "swi_add() failed in %s\n", __func__);
2484 pfsync_pointers_init();
2486 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
2487 vnet_pfsync_init, NULL);
2490 vnet_pfsync_uninit(const void *unused __unused)
2493 pfsync_pointers_uninit();
2495 if_clone_detach(V_pfsync_cloner);
2496 swi_remove(V_pfsync_swi_cookie);
2499 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH,
2500 vnet_pfsync_uninit, NULL);
2508 pfsync_detach_ifnet_ptr = pfsync_detach_ifnet;
2510 error = pf_proto_register(PF_INET, &in_pfsync_protosw);
2513 error = ipproto_register(IPPROTO_PFSYNC);
2515 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2526 pfsync_detach_ifnet_ptr = NULL;
2529 ipproto_unregister(IPPROTO_PFSYNC);
2530 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2535 pfsync_modevent(module_t mod, int type, void *data)
2541 error = pfsync_init();
2554 static moduledata_t pfsync_mod = {
2560 #define PFSYNC_MODVER 1
2562 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
2563 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
2564 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2565 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);