2 * Copyright (c) 2002 Michael Shalayeff
3 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
19 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
20 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
21 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
23 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
24 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
25 * THE POSSIBILITY OF SUCH DAMAGE.
29 * Copyright (c) 2009 David Gwynne <dlg@openbsd.org>
31 * Permission to use, copy, modify, and distribute this software for any
32 * purpose with or without fee is hereby granted, provided that the above
33 * copyright notice and this permission notice appear in all copies.
35 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
36 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
37 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
38 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
39 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
40 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
41 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
45 * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $
47 * Revisions picked from OpenBSD after revision 1.110 import:
48 * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input()
49 * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates
50 * 1.120, 1.175 - use monotonic time_uptime
51 * 1.122 - reduce number of updates for non-TCP sessions
52 * 1.125, 1.127 - rewrite merge or stale processing
54 * 1.146 - bzero() mbuf before sparsely filling it with data
55 * 1.170 - SIOCSIFMTU checks
56 * 1.126, 1.142 - deferred packets processing
57 * 1.173 - correct expire time processing
60 #include <sys/cdefs.h>
61 __FBSDID("$FreeBSD$");
64 #include "opt_inet6.h"
67 #include <sys/param.h>
69 #include <sys/endian.h>
70 #include <sys/interrupt.h>
71 #include <sys/kernel.h>
74 #include <sys/module.h>
75 #include <sys/mutex.h>
77 #include <sys/protosw.h>
78 #include <sys/socket.h>
79 #include <sys/sockio.h>
80 #include <sys/sysctl.h>
81 #include <sys/syslog.h>
85 #include <net/if_var.h>
86 #include <net/if_clone.h>
87 #include <net/if_types.h>
89 #include <net/pfvar.h>
90 #include <net/if_pfsync.h>
92 #include <netinet/if_ether.h>
93 #include <netinet/in.h>
94 #include <netinet/in_var.h>
95 #include <netinet/ip.h>
96 #include <netinet/ip_carp.h>
97 #include <netinet/ip_var.h>
98 #include <netinet/tcp.h>
99 #include <netinet/tcp_fsm.h>
100 #include <netinet/tcp_seq.h>
102 #define PFSYNC_MINPKT ( \
103 sizeof(struct ip) + \
104 sizeof(struct pfsync_header) + \
105 sizeof(struct pfsync_subheader) )
113 static int pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *,
114 struct pfsync_state_peer *);
115 static int pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int);
116 static int pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int);
117 static int pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int);
118 static int pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int);
119 static int pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int);
120 static int pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int);
121 static int pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int);
122 static int pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int);
123 static int pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int);
124 static int pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int);
125 static int pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int);
126 static int pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int);
128 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = {
129 pfsync_in_clr, /* PFSYNC_ACT_CLR */
130 pfsync_in_ins, /* PFSYNC_ACT_INS */
131 pfsync_in_iack, /* PFSYNC_ACT_INS_ACK */
132 pfsync_in_upd, /* PFSYNC_ACT_UPD */
133 pfsync_in_upd_c, /* PFSYNC_ACT_UPD_C */
134 pfsync_in_ureq, /* PFSYNC_ACT_UPD_REQ */
135 pfsync_in_del, /* PFSYNC_ACT_DEL */
136 pfsync_in_del_c, /* PFSYNC_ACT_DEL_C */
137 pfsync_in_error, /* PFSYNC_ACT_INS_F */
138 pfsync_in_error, /* PFSYNC_ACT_DEL_F */
139 pfsync_in_bus, /* PFSYNC_ACT_BUS */
140 pfsync_in_tdb, /* PFSYNC_ACT_TDB */
141 pfsync_in_eof /* PFSYNC_ACT_EOF */
145 void (*write)(struct pf_state *, void *);
150 /* we have one of these for every PFSYNC_S_ */
151 static void pfsync_out_state(struct pf_state *, void *);
152 static void pfsync_out_iack(struct pf_state *, void *);
153 static void pfsync_out_upd_c(struct pf_state *, void *);
154 static void pfsync_out_del(struct pf_state *, void *);
156 static struct pfsync_q pfsync_qs[] = {
157 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_INS },
158 { pfsync_out_iack, sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
159 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_UPD },
160 { pfsync_out_upd_c, sizeof(struct pfsync_upd_c), PFSYNC_ACT_UPD_C },
161 { pfsync_out_del, sizeof(struct pfsync_del_c), PFSYNC_ACT_DEL_C }
164 static void pfsync_q_ins(struct pf_state *, int, bool);
165 static void pfsync_q_del(struct pf_state *, bool);
167 static void pfsync_update_state(struct pf_state *);
169 struct pfsync_upd_req_item {
170 TAILQ_ENTRY(pfsync_upd_req_item) ur_entry;
171 struct pfsync_upd_req ur_msg;
174 struct pfsync_deferral {
175 struct pfsync_softc *pd_sc;
176 TAILQ_ENTRY(pfsync_deferral) pd_entry;
178 struct callout pd_tmo;
180 struct pf_state *pd_st;
184 struct pfsync_softc {
186 struct ifnet *sc_ifp;
187 struct ifnet *sc_sync_if;
188 struct ip_moptions sc_imo;
189 struct in_addr sc_sync_peer;
191 #define PFSYNCF_OK 0x00000001
192 #define PFSYNCF_DEFER 0x00000002
193 #define PFSYNCF_PUSH 0x00000004
194 uint8_t sc_maxupdates;
195 struct ip sc_template;
196 struct callout sc_tmo;
201 TAILQ_HEAD(, pf_state) sc_qs[PFSYNC_S_COUNT];
202 TAILQ_HEAD(, pfsync_upd_req_item) sc_upd_req_list;
203 TAILQ_HEAD(, pfsync_deferral) sc_deferrals;
208 /* Bulk update info */
209 struct mtx sc_bulk_mtx;
210 uint32_t sc_ureq_sent;
212 uint32_t sc_ureq_received;
214 uint64_t sc_bulk_stateid;
215 uint32_t sc_bulk_creatorid;
216 struct callout sc_bulk_tmo;
217 struct callout sc_bulkfail_tmo;
220 #define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx)
221 #define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx)
222 #define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED)
224 #define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx)
225 #define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx)
226 #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
228 static const char pfsyncname[] = "pfsync";
229 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
230 static VNET_DEFINE(struct pfsync_softc *, pfsyncif) = NULL;
231 #define V_pfsyncif VNET(pfsyncif)
232 static VNET_DEFINE(void *, pfsync_swi_cookie) = NULL;
233 #define V_pfsync_swi_cookie VNET(pfsync_swi_cookie)
234 static VNET_DEFINE(struct pfsyncstats, pfsyncstats);
235 #define V_pfsyncstats VNET(pfsyncstats)
236 static VNET_DEFINE(int, pfsync_carp_adj) = CARP_MAXSKEW;
237 #define V_pfsync_carp_adj VNET(pfsync_carp_adj)
239 static void pfsync_timeout(void *);
240 static void pfsync_push(struct pfsync_softc *);
241 static void pfsyncintr(void *);
242 static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
244 static void pfsync_multicast_cleanup(struct pfsync_softc *);
245 static void pfsync_pointers_init(void);
246 static void pfsync_pointers_uninit(void);
247 static int pfsync_init(void);
248 static void pfsync_uninit(void);
250 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC");
251 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW,
252 &VNET_NAME(pfsyncstats), pfsyncstats,
253 "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
254 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW,
255 &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
257 static int pfsync_clone_create(struct if_clone *, int, caddr_t);
258 static void pfsync_clone_destroy(struct ifnet *);
259 static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
260 struct pf_state_peer *);
261 static int pfsyncoutput(struct ifnet *, struct mbuf *,
262 const struct sockaddr *, struct route *);
263 static int pfsyncioctl(struct ifnet *, u_long, caddr_t);
265 static int pfsync_defer(struct pf_state *, struct mbuf *);
266 static void pfsync_undefer(struct pfsync_deferral *, int);
267 static void pfsync_undefer_state(struct pf_state *, int);
268 static void pfsync_defer_tmo(void *);
270 static void pfsync_request_update(u_int32_t, u_int64_t);
271 static void pfsync_update_state_req(struct pf_state *);
273 static void pfsync_drop(struct pfsync_softc *);
274 static void pfsync_sendout(int);
275 static void pfsync_send_plus(void *, size_t);
277 static void pfsync_bulk_start(void);
278 static void pfsync_bulk_status(u_int8_t);
279 static void pfsync_bulk_update(void *);
280 static void pfsync_bulk_fail(void *);
283 static void pfsync_update_net_tdb(struct pfsync_tdb *);
286 #define PFSYNC_MAX_BULKTRIES 12
288 VNET_DEFINE(struct if_clone *, pfsync_cloner);
289 #define V_pfsync_cloner VNET(pfsync_cloner)
292 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
294 struct pfsync_softc *sc;
301 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
302 sc->sc_flags |= PFSYNCF_OK;
304 for (q = 0; q < PFSYNC_S_COUNT; q++)
305 TAILQ_INIT(&sc->sc_qs[q]);
307 TAILQ_INIT(&sc->sc_upd_req_list);
308 TAILQ_INIT(&sc->sc_deferrals);
310 sc->sc_len = PFSYNC_MINPKT;
311 sc->sc_maxupdates = 128;
313 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
318 if_initname(ifp, pfsyncname, unit);
320 ifp->if_ioctl = pfsyncioctl;
321 ifp->if_output = pfsyncoutput;
322 ifp->if_type = IFT_PFSYNC;
323 ifp->if_snd.ifq_maxlen = ifqmaxlen;
324 ifp->if_hdrlen = sizeof(struct pfsync_header);
325 ifp->if_mtu = ETHERMTU;
326 mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
327 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
328 callout_init(&sc->sc_tmo, 1);
329 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
330 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
334 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
342 pfsync_clone_destroy(struct ifnet *ifp)
344 struct pfsync_softc *sc = ifp->if_softc;
347 * At this stage, everything should have already been
348 * cleared by pfsync_uninit(), and we have only to
351 while (sc->sc_deferred > 0) {
352 struct pfsync_deferral *pd = TAILQ_FIRST(&sc->sc_deferrals);
354 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
356 if (callout_stop(&pd->pd_tmo) > 0) {
357 pf_release_state(pd->pd_st);
362 callout_drain(&pd->pd_tmo);
367 callout_drain(&sc->sc_tmo);
368 callout_drain(&sc->sc_bulkfail_tmo);
369 callout_drain(&sc->sc_bulk_tmo);
371 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
372 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
379 if (sc->sc_imo.imo_membership)
380 pfsync_multicast_cleanup(sc);
381 mtx_destroy(&sc->sc_mtx);
382 mtx_destroy(&sc->sc_bulk_mtx);
389 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
390 struct pf_state_peer *d)
392 if (s->scrub.scrub_flag && d->scrub == NULL) {
393 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
394 if (d->scrub == NULL)
403 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
405 struct pfsync_softc *sc = V_pfsyncif;
406 #ifndef __NO_STRICT_ALIGNMENT
407 struct pfsync_state_key key[2];
409 struct pfsync_state_key *kw, *ks;
410 struct pf_state *st = NULL;
411 struct pf_state_key *skw = NULL, *sks = NULL;
412 struct pf_rule *r = NULL;
418 if (sp->creatorid == 0) {
419 if (V_pf_status.debug >= PF_DEBUG_MISC)
420 printf("%s: invalid creator id: %08x\n", __func__,
421 ntohl(sp->creatorid));
425 if ((kif = pfi_kif_find(sp->ifname)) == NULL) {
426 if (V_pf_status.debug >= PF_DEBUG_MISC)
427 printf("%s: unknown interface: %s\n", __func__,
429 if (flags & PFSYNC_SI_IOCTL)
431 return (0); /* skip this state */
435 * If the ruleset checksums match or the state is coming from the ioctl,
436 * it's safe to associate the state with the rule of that number.
438 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
439 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
440 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
441 r = pf_main_ruleset.rules[
442 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
444 r = &V_pf_default_rule;
446 if ((r->max_states &&
447 counter_u64_fetch(r->states_cur) >= r->max_states))
451 * XXXGL: consider M_WAITOK in ioctl path after.
453 if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL)
456 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
459 #ifndef __NO_STRICT_ALIGNMENT
460 bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2);
461 kw = &key[PF_SK_WIRE];
462 ks = &key[PF_SK_STACK];
464 kw = &sp->key[PF_SK_WIRE];
465 ks = &sp->key[PF_SK_STACK];
468 if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) ||
469 PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) ||
470 kw->port[0] != ks->port[0] ||
471 kw->port[1] != ks->port[1]) {
472 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
478 /* allocate memory for scrub info */
479 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
480 pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
483 /* Copy to state key(s). */
484 skw->addr[0] = kw->addr[0];
485 skw->addr[1] = kw->addr[1];
486 skw->port[0] = kw->port[0];
487 skw->port[1] = kw->port[1];
488 skw->proto = sp->proto;
491 sks->addr[0] = ks->addr[0];
492 sks->addr[1] = ks->addr[1];
493 sks->port[0] = ks->port[0];
494 sks->port[1] = ks->port[1];
495 sks->proto = sp->proto;
500 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
501 st->creation = time_uptime - ntohl(sp->creation);
502 st->expire = time_uptime;
506 timeout = r->timeout[sp->timeout];
508 timeout = V_pf_default_rule.timeout[sp->timeout];
510 /* sp->expire may have been adaptively scaled by export. */
511 st->expire -= timeout - ntohl(sp->expire);
514 st->direction = sp->direction;
516 st->timeout = sp->timeout;
517 st->state_flags = sp->state_flags;
520 st->creatorid = sp->creatorid;
521 pf_state_peer_ntoh(&sp->src, &st->src);
522 pf_state_peer_ntoh(&sp->dst, &st->dst);
525 st->nat_rule.ptr = NULL;
526 st->anchor.ptr = NULL;
529 st->pfsync_time = time_uptime;
530 st->sync_state = PFSYNC_S_NONE;
532 if (!(flags & PFSYNC_SI_IOCTL))
533 st->state_flags |= PFSTATE_NOSYNC;
535 if ((error = pf_state_insert(kif, skw, sks, st)) != 0)
538 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
539 counter_u64_add(r->states_cur, 1);
540 counter_u64_add(r->states_tot, 1);
542 if (!(flags & PFSYNC_SI_IOCTL)) {
543 st->state_flags &= ~PFSTATE_NOSYNC;
544 if (st->state_flags & PFSTATE_ACK) {
545 pfsync_q_ins(st, PFSYNC_S_IACK, true);
549 st->state_flags &= ~PFSTATE_ACK;
559 uma_zfree(V_pf_state_key_z, skw);
561 uma_zfree(V_pf_state_key_z, sks);
563 cleanup_state: /* pf_state_insert() frees the state keys. */
566 uma_zfree(V_pf_state_scrub_z, st->dst.scrub);
568 uma_zfree(V_pf_state_scrub_z, st->src.scrub);
569 uma_zfree(V_pf_state_z, st);
575 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused)
577 struct pfsync_softc *sc = V_pfsyncif;
578 struct pfsync_pkt pkt;
579 struct mbuf *m = *mp;
580 struct ip *ip = mtod(m, struct ip *);
581 struct pfsync_header *ph;
582 struct pfsync_subheader subh;
589 V_pfsyncstats.pfsyncs_ipackets++;
591 /* Verify that we have a sync interface configured. */
592 if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
593 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
596 /* verify that the packet came in on the right interface */
597 if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
598 V_pfsyncstats.pfsyncs_badif++;
602 if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1);
603 if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
604 /* verify that the IP TTL is 255. */
605 if (ip->ip_ttl != PFSYNC_DFLTTL) {
606 V_pfsyncstats.pfsyncs_badttl++;
610 offset = ip->ip_hl << 2;
611 if (m->m_pkthdr.len < offset + sizeof(*ph)) {
612 V_pfsyncstats.pfsyncs_hdrops++;
616 if (offset + sizeof(*ph) > m->m_len) {
617 if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
618 V_pfsyncstats.pfsyncs_hdrops++;
619 return (IPPROTO_DONE);
621 ip = mtod(m, struct ip *);
623 ph = (struct pfsync_header *)((char *)ip + offset);
625 /* verify the version */
626 if (ph->version != PFSYNC_VERSION) {
627 V_pfsyncstats.pfsyncs_badver++;
631 len = ntohs(ph->len) + offset;
632 if (m->m_pkthdr.len < len) {
633 V_pfsyncstats.pfsyncs_badlen++;
637 /* Cheaper to grab this now than having to mess with mbufs later */
639 pkt.src = ip->ip_src;
643 * Trusting pf_chksum during packet processing, as well as seeking
644 * in interface name tree, require holding PF_RULES_RLOCK().
647 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
648 pkt.flags |= PFSYNC_SI_CKSUM;
650 offset += sizeof(*ph);
651 while (offset <= len - sizeof(subh)) {
652 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
653 offset += sizeof(subh);
655 if (subh.action >= PFSYNC_ACT_MAX) {
656 V_pfsyncstats.pfsyncs_badact++;
661 count = ntohs(subh.count);
662 V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
663 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count);
666 return (IPPROTO_DONE);
675 return (IPPROTO_DONE);
679 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
681 struct pfsync_clr *clr;
683 int len = sizeof(*clr) * count;
687 mp = m_pulldown(m, offset, len, &offp);
689 V_pfsyncstats.pfsyncs_badlen++;
692 clr = (struct pfsync_clr *)(mp->m_data + offp);
694 for (i = 0; i < count; i++) {
695 creatorid = clr[i].creatorid;
697 if (clr[i].ifname[0] != '\0' &&
698 pfi_kif_find(clr[i].ifname) == NULL)
701 for (int i = 0; i <= pf_hashmask; i++) {
702 struct pf_idhash *ih = &V_pf_idhash[i];
706 LIST_FOREACH(s, &ih->states, entry) {
707 if (s->creatorid == creatorid) {
708 s->state_flags |= PFSTATE_NOSYNC;
709 pf_unlink_state(s, PF_ENTER_LOCKED);
713 PF_HASHROW_UNLOCK(ih);
721 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
724 struct pfsync_state *sa, *sp;
725 int len = sizeof(*sp) * count;
728 mp = m_pulldown(m, offset, len, &offp);
730 V_pfsyncstats.pfsyncs_badlen++;
733 sa = (struct pfsync_state *)(mp->m_data + offp);
735 for (i = 0; i < count; i++) {
738 /* Check for invalid values. */
739 if (sp->timeout >= PFTM_MAX ||
740 sp->src.state > PF_TCPS_PROXY_DST ||
741 sp->dst.state > PF_TCPS_PROXY_DST ||
742 sp->direction > PF_OUT ||
743 (sp->af != AF_INET && sp->af != AF_INET6)) {
744 if (V_pf_status.debug >= PF_DEBUG_MISC)
745 printf("%s: invalid value\n", __func__);
746 V_pfsyncstats.pfsyncs_badval++;
750 if (pfsync_state_import(sp, pkt->flags) == ENOMEM)
751 /* Drop out, but process the rest of the actions. */
759 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
761 struct pfsync_ins_ack *ia, *iaa;
765 int len = count * sizeof(*ia);
768 mp = m_pulldown(m, offset, len, &offp);
770 V_pfsyncstats.pfsyncs_badlen++;
773 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
775 for (i = 0; i < count; i++) {
778 st = pf_find_state_byid(ia->id, ia->creatorid);
782 if (st->state_flags & PFSTATE_ACK) {
783 PFSYNC_LOCK(V_pfsyncif);
784 pfsync_undefer_state(st, 0);
785 PFSYNC_UNLOCK(V_pfsyncif);
790 * XXX this is not yet implemented, but we know the size of the
791 * message so we can skip it.
794 return (count * sizeof(struct pfsync_ins_ack));
798 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src,
799 struct pfsync_state_peer *dst)
803 PF_STATE_LOCK_ASSERT(st);
806 * The state should never go backwards except
807 * for syn-proxy states. Neither should the
808 * sequence window slide backwards.
810 if ((st->src.state > src->state &&
811 (st->src.state < PF_TCPS_PROXY_SRC ||
812 src->state >= PF_TCPS_PROXY_SRC)) ||
814 (st->src.state == src->state &&
815 SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
818 pf_state_peer_ntoh(src, &st->src);
820 if ((st->dst.state > dst->state) ||
822 (st->dst.state >= TCPS_SYN_SENT &&
823 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
826 pf_state_peer_ntoh(dst, &st->dst);
832 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
834 struct pfsync_softc *sc = V_pfsyncif;
835 struct pfsync_state *sa, *sp;
840 int len = count * sizeof(*sp);
843 mp = m_pulldown(m, offset, len, &offp);
845 V_pfsyncstats.pfsyncs_badlen++;
848 sa = (struct pfsync_state *)(mp->m_data + offp);
850 for (i = 0; i < count; i++) {
853 /* check for invalid values */
854 if (sp->timeout >= PFTM_MAX ||
855 sp->src.state > PF_TCPS_PROXY_DST ||
856 sp->dst.state > PF_TCPS_PROXY_DST) {
857 if (V_pf_status.debug >= PF_DEBUG_MISC) {
858 printf("pfsync_input: PFSYNC_ACT_UPD: "
861 V_pfsyncstats.pfsyncs_badval++;
865 st = pf_find_state_byid(sp->id, sp->creatorid);
867 /* insert the update */
868 if (pfsync_state_import(sp, 0))
869 V_pfsyncstats.pfsyncs_badstate++;
873 if (st->state_flags & PFSTATE_ACK) {
875 pfsync_undefer_state(st, 1);
879 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
880 sync = pfsync_upd_tcp(st, &sp->src, &sp->dst);
885 * Non-TCP protocol state machine always go
888 if (st->src.state > sp->src.state)
891 pf_state_peer_ntoh(&sp->src, &st->src);
892 if (st->dst.state > sp->dst.state)
895 pf_state_peer_ntoh(&sp->dst, &st->dst);
898 pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
899 pf_state_peer_ntoh(&sp->dst, &st->dst);
900 st->expire = time_uptime;
901 st->timeout = sp->timeout;
903 st->pfsync_time = time_uptime;
906 V_pfsyncstats.pfsyncs_stale++;
908 pfsync_update_state(st);
922 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
924 struct pfsync_softc *sc = V_pfsyncif;
925 struct pfsync_upd_c *ua, *up;
927 int len = count * sizeof(*up);
932 mp = m_pulldown(m, offset, len, &offp);
934 V_pfsyncstats.pfsyncs_badlen++;
937 ua = (struct pfsync_upd_c *)(mp->m_data + offp);
939 for (i = 0; i < count; i++) {
942 /* check for invalid values */
943 if (up->timeout >= PFTM_MAX ||
944 up->src.state > PF_TCPS_PROXY_DST ||
945 up->dst.state > PF_TCPS_PROXY_DST) {
946 if (V_pf_status.debug >= PF_DEBUG_MISC) {
947 printf("pfsync_input: "
951 V_pfsyncstats.pfsyncs_badval++;
955 st = pf_find_state_byid(up->id, up->creatorid);
957 /* We don't have this state. Ask for it. */
959 pfsync_request_update(up->creatorid, up->id);
964 if (st->state_flags & PFSTATE_ACK) {
966 pfsync_undefer_state(st, 1);
970 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
971 sync = pfsync_upd_tcp(st, &up->src, &up->dst);
976 * Non-TCP protocol state machine always go
979 if (st->src.state > up->src.state)
982 pf_state_peer_ntoh(&up->src, &st->src);
983 if (st->dst.state > up->dst.state)
986 pf_state_peer_ntoh(&up->dst, &st->dst);
989 pfsync_alloc_scrub_memory(&up->dst, &st->dst);
990 pf_state_peer_ntoh(&up->dst, &st->dst);
991 st->expire = time_uptime;
992 st->timeout = up->timeout;
994 st->pfsync_time = time_uptime;
997 V_pfsyncstats.pfsyncs_stale++;
999 pfsync_update_state(st);
1000 PF_STATE_UNLOCK(st);
1006 PF_STATE_UNLOCK(st);
1013 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1015 struct pfsync_upd_req *ur, *ura;
1017 int len = count * sizeof(*ur);
1020 struct pf_state *st;
1022 mp = m_pulldown(m, offset, len, &offp);
1024 V_pfsyncstats.pfsyncs_badlen++;
1027 ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1029 for (i = 0; i < count; i++) {
1032 if (ur->id == 0 && ur->creatorid == 0)
1033 pfsync_bulk_start();
1035 st = pf_find_state_byid(ur->id, ur->creatorid);
1037 V_pfsyncstats.pfsyncs_badstate++;
1040 if (st->state_flags & PFSTATE_NOSYNC) {
1041 PF_STATE_UNLOCK(st);
1045 pfsync_update_state_req(st);
1046 PF_STATE_UNLOCK(st);
1054 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1057 struct pfsync_state *sa, *sp;
1058 struct pf_state *st;
1059 int len = count * sizeof(*sp);
1062 mp = m_pulldown(m, offset, len, &offp);
1064 V_pfsyncstats.pfsyncs_badlen++;
1067 sa = (struct pfsync_state *)(mp->m_data + offp);
1069 for (i = 0; i < count; i++) {
1072 st = pf_find_state_byid(sp->id, sp->creatorid);
1074 V_pfsyncstats.pfsyncs_badstate++;
1077 st->state_flags |= PFSTATE_NOSYNC;
1078 pf_unlink_state(st, PF_ENTER_LOCKED);
1085 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1088 struct pfsync_del_c *sa, *sp;
1089 struct pf_state *st;
1090 int len = count * sizeof(*sp);
1093 mp = m_pulldown(m, offset, len, &offp);
1095 V_pfsyncstats.pfsyncs_badlen++;
1098 sa = (struct pfsync_del_c *)(mp->m_data + offp);
1100 for (i = 0; i < count; i++) {
1103 st = pf_find_state_byid(sp->id, sp->creatorid);
1105 V_pfsyncstats.pfsyncs_badstate++;
1109 st->state_flags |= PFSTATE_NOSYNC;
1110 pf_unlink_state(st, PF_ENTER_LOCKED);
1117 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1119 struct pfsync_softc *sc = V_pfsyncif;
1120 struct pfsync_bus *bus;
1122 int len = count * sizeof(*bus);
1127 /* If we're not waiting for a bulk update, who cares. */
1128 if (sc->sc_ureq_sent == 0) {
1133 mp = m_pulldown(m, offset, len, &offp);
1136 V_pfsyncstats.pfsyncs_badlen++;
1139 bus = (struct pfsync_bus *)(mp->m_data + offp);
1141 switch (bus->status) {
1142 case PFSYNC_BUS_START:
1143 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1144 V_pf_limits[PF_LIMIT_STATES].limit /
1145 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1146 sizeof(struct pfsync_state)),
1147 pfsync_bulk_fail, sc);
1148 if (V_pf_status.debug >= PF_DEBUG_MISC)
1149 printf("pfsync: received bulk update start\n");
1152 case PFSYNC_BUS_END:
1153 if (time_uptime - ntohl(bus->endtime) >=
1155 /* that's it, we're happy */
1156 sc->sc_ureq_sent = 0;
1157 sc->sc_bulk_tries = 0;
1158 callout_stop(&sc->sc_bulkfail_tmo);
1159 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1160 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
1161 "pfsync bulk done");
1162 sc->sc_flags |= PFSYNCF_OK;
1163 if (V_pf_status.debug >= PF_DEBUG_MISC)
1164 printf("pfsync: received valid "
1165 "bulk update end\n");
1167 if (V_pf_status.debug >= PF_DEBUG_MISC)
1168 printf("pfsync: received invalid "
1169 "bulk update end: bad timestamp\n");
1179 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1181 int len = count * sizeof(struct pfsync_tdb);
1184 struct pfsync_tdb *tp;
1190 mp = m_pulldown(m, offset, len, &offp);
1192 V_pfsyncstats.pfsyncs_badlen++;
1195 tp = (struct pfsync_tdb *)(mp->m_data + offp);
1197 for (i = 0; i < count; i++)
1198 pfsync_update_net_tdb(&tp[i]);
1205 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1207 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1212 /* check for invalid values */
1213 if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1214 (pt->dst.sa.sa_family != AF_INET &&
1215 pt->dst.sa.sa_family != AF_INET6))
1218 tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1220 pt->rpl = ntohl(pt->rpl);
1221 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1223 /* Neither replay nor byte counter should ever decrease. */
1224 if (pt->rpl < tdb->tdb_rpl ||
1225 pt->cur_bytes < tdb->tdb_cur_bytes) {
1229 tdb->tdb_rpl = pt->rpl;
1230 tdb->tdb_cur_bytes = pt->cur_bytes;
1235 if (V_pf_status.debug >= PF_DEBUG_MISC)
1236 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1238 V_pfsyncstats.pfsyncs_badstate++;
1245 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1247 /* check if we are at the right place in the packet */
1248 if (offset != m->m_pkthdr.len)
1249 V_pfsyncstats.pfsyncs_badlen++;
1251 /* we're done. free and let the caller return */
1257 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1259 V_pfsyncstats.pfsyncs_badact++;
1266 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1275 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1277 struct pfsync_softc *sc = ifp->if_softc;
1278 struct ifreq *ifr = (struct ifreq *)data;
1279 struct pfsyncreq pfsyncr;
1285 if (ifp->if_flags & IFF_UP) {
1286 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1288 pfsync_pointers_init();
1290 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1292 pfsync_pointers_uninit();
1296 if (!sc->sc_sync_if ||
1297 ifr->ifr_mtu <= PFSYNC_MINPKT ||
1298 ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1300 if (ifr->ifr_mtu < ifp->if_mtu) {
1302 if (sc->sc_len > PFSYNC_MINPKT)
1306 ifp->if_mtu = ifr->ifr_mtu;
1309 bzero(&pfsyncr, sizeof(pfsyncr));
1311 if (sc->sc_sync_if) {
1312 strlcpy(pfsyncr.pfsyncr_syncdev,
1313 sc->sc_sync_if->if_xname, IFNAMSIZ);
1315 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1316 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1317 pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER ==
1318 (sc->sc_flags & PFSYNCF_DEFER));
1320 return (copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)));
1324 struct ip_moptions *imo = &sc->sc_imo;
1329 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1331 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr))))
1334 if (pfsyncr.pfsyncr_maxupdates > 255)
1337 if (pfsyncr.pfsyncr_syncdev[0] == 0)
1339 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL)
1342 if (sifp != NULL && (
1343 pfsyncr.pfsyncr_syncpeer.s_addr == 0 ||
1344 pfsyncr.pfsyncr_syncpeer.s_addr ==
1345 htonl(INADDR_PFSYNC_GROUP)))
1346 mship = malloc((sizeof(struct in_multi *) *
1347 IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO);
1350 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1351 sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP);
1353 sc->sc_sync_peer.s_addr =
1354 pfsyncr.pfsyncr_syncpeer.s_addr;
1356 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1357 if (pfsyncr.pfsyncr_defer) {
1358 sc->sc_flags |= PFSYNCF_DEFER;
1359 pfsync_defer_ptr = pfsync_defer;
1361 sc->sc_flags &= ~PFSYNCF_DEFER;
1362 pfsync_defer_ptr = NULL;
1367 if_rele(sc->sc_sync_if);
1368 sc->sc_sync_if = NULL;
1369 if (imo->imo_membership)
1370 pfsync_multicast_cleanup(sc);
1375 if (sc->sc_len > PFSYNC_MINPKT &&
1376 (sifp->if_mtu < sc->sc_ifp->if_mtu ||
1377 (sc->sc_sync_if != NULL &&
1378 sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
1379 sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
1382 if (imo->imo_membership)
1383 pfsync_multicast_cleanup(sc);
1385 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
1386 error = pfsync_multicast_setup(sc, sifp, mship);
1389 free(mship, M_PFSYNC);
1394 if_rele(sc->sc_sync_if);
1395 sc->sc_sync_if = sifp;
1397 ip = &sc->sc_template;
1398 bzero(ip, sizeof(*ip));
1399 ip->ip_v = IPVERSION;
1400 ip->ip_hl = sizeof(sc->sc_template) >> 2;
1401 ip->ip_tos = IPTOS_LOWDELAY;
1402 /* len and id are set later. */
1403 ip->ip_off = htons(IP_DF);
1404 ip->ip_ttl = PFSYNC_DFLTTL;
1405 ip->ip_p = IPPROTO_PFSYNC;
1406 ip->ip_src.s_addr = INADDR_ANY;
1407 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr;
1409 /* Request a full state table update. */
1410 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1411 (*carp_demote_adj_p)(V_pfsync_carp_adj,
1412 "pfsync bulk start");
1413 sc->sc_flags &= ~PFSYNCF_OK;
1414 if (V_pf_status.debug >= PF_DEBUG_MISC)
1415 printf("pfsync: requesting bulk update\n");
1416 pfsync_request_update(0, 0);
1419 sc->sc_ureq_sent = time_uptime;
1420 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail,
1434 pfsync_out_state(struct pf_state *st, void *buf)
1436 struct pfsync_state *sp = buf;
1438 pfsync_state_export(sp, st);
1442 pfsync_out_iack(struct pf_state *st, void *buf)
1444 struct pfsync_ins_ack *iack = buf;
1447 iack->creatorid = st->creatorid;
1451 pfsync_out_upd_c(struct pf_state *st, void *buf)
1453 struct pfsync_upd_c *up = buf;
1455 bzero(up, sizeof(*up));
1457 pf_state_peer_hton(&st->src, &up->src);
1458 pf_state_peer_hton(&st->dst, &up->dst);
1459 up->creatorid = st->creatorid;
1460 up->timeout = st->timeout;
1464 pfsync_out_del(struct pf_state *st, void *buf)
1466 struct pfsync_del_c *dp = buf;
1469 dp->creatorid = st->creatorid;
1470 st->state_flags |= PFSTATE_NOSYNC;
1474 pfsync_drop(struct pfsync_softc *sc)
1476 struct pf_state *st, *next;
1477 struct pfsync_upd_req_item *ur;
1480 for (q = 0; q < PFSYNC_S_COUNT; q++) {
1481 if (TAILQ_EMPTY(&sc->sc_qs[q]))
1484 TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, next) {
1485 KASSERT(st->sync_state == q,
1486 ("%s: st->sync_state == q",
1488 st->sync_state = PFSYNC_S_NONE;
1489 pf_release_state(st);
1491 TAILQ_INIT(&sc->sc_qs[q]);
1494 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1495 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1500 sc->sc_len = PFSYNC_MINPKT;
1504 pfsync_sendout(int schedswi)
1506 struct pfsync_softc *sc = V_pfsyncif;
1507 struct ifnet *ifp = sc->sc_ifp;
1510 struct pfsync_header *ph;
1511 struct pfsync_subheader *subh;
1512 struct pf_state *st, *st_next;
1513 struct pfsync_upd_req_item *ur;
1517 KASSERT(sc != NULL, ("%s: null sc", __func__));
1518 KASSERT(sc->sc_len > PFSYNC_MINPKT,
1519 ("%s: sc_len %zu", __func__, sc->sc_len));
1520 PFSYNC_LOCK_ASSERT(sc);
1522 if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1527 m = m_get2(max_linkhdr + sc->sc_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1529 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1);
1530 V_pfsyncstats.pfsyncs_onomem++;
1533 m->m_data += max_linkhdr;
1534 m->m_len = m->m_pkthdr.len = sc->sc_len;
1536 /* build the ip header */
1537 ip = (struct ip *)m->m_data;
1538 bcopy(&sc->sc_template, ip, sizeof(*ip));
1539 offset = sizeof(*ip);
1541 ip->ip_len = htons(m->m_pkthdr.len);
1544 /* build the pfsync header */
1545 ph = (struct pfsync_header *)(m->m_data + offset);
1546 bzero(ph, sizeof(*ph));
1547 offset += sizeof(*ph);
1549 ph->version = PFSYNC_VERSION;
1550 ph->len = htons(sc->sc_len - sizeof(*ip));
1551 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1553 /* walk the queues */
1554 for (q = 0; q < PFSYNC_S_COUNT; q++) {
1555 if (TAILQ_EMPTY(&sc->sc_qs[q]))
1558 subh = (struct pfsync_subheader *)(m->m_data + offset);
1559 offset += sizeof(*subh);
1562 TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, st_next) {
1563 KASSERT(st->sync_state == q,
1564 ("%s: st->sync_state == q",
1567 * XXXGL: some of write methods do unlocked reads
1570 pfsync_qs[q].write(st, m->m_data + offset);
1571 offset += pfsync_qs[q].len;
1572 st->sync_state = PFSYNC_S_NONE;
1573 pf_release_state(st);
1576 TAILQ_INIT(&sc->sc_qs[q]);
1578 bzero(subh, sizeof(*subh));
1579 subh->action = pfsync_qs[q].action;
1580 subh->count = htons(count);
1581 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1584 if (!TAILQ_EMPTY(&sc->sc_upd_req_list)) {
1585 subh = (struct pfsync_subheader *)(m->m_data + offset);
1586 offset += sizeof(*subh);
1589 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1590 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1592 bcopy(&ur->ur_msg, m->m_data + offset,
1593 sizeof(ur->ur_msg));
1594 offset += sizeof(ur->ur_msg);
1599 bzero(subh, sizeof(*subh));
1600 subh->action = PFSYNC_ACT_UPD_REQ;
1601 subh->count = htons(count);
1602 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1605 /* has someone built a custom region for us to add? */
1606 if (sc->sc_plus != NULL) {
1607 bcopy(sc->sc_plus, m->m_data + offset, sc->sc_pluslen);
1608 offset += sc->sc_pluslen;
1613 subh = (struct pfsync_subheader *)(m->m_data + offset);
1614 offset += sizeof(*subh);
1616 bzero(subh, sizeof(*subh));
1617 subh->action = PFSYNC_ACT_EOF;
1618 subh->count = htons(1);
1619 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1621 /* we're done, let's put it on the wire */
1623 m->m_data += sizeof(*ip);
1624 m->m_len = m->m_pkthdr.len = sc->sc_len - sizeof(*ip);
1626 m->m_data -= sizeof(*ip);
1627 m->m_len = m->m_pkthdr.len = sc->sc_len;
1630 if (sc->sc_sync_if == NULL) {
1631 sc->sc_len = PFSYNC_MINPKT;
1636 if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1);
1637 if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
1638 sc->sc_len = PFSYNC_MINPKT;
1640 if (!_IF_QFULL(&sc->sc_ifp->if_snd))
1641 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1644 if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1);
1647 swi_sched(V_pfsync_swi_cookie, 0);
1651 pfsync_insert_state(struct pf_state *st)
1653 struct pfsync_softc *sc = V_pfsyncif;
1655 if (st->state_flags & PFSTATE_NOSYNC)
1658 if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1659 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1660 st->state_flags |= PFSTATE_NOSYNC;
1664 KASSERT(st->sync_state == PFSYNC_S_NONE,
1665 ("%s: st->sync_state %u", __func__, st->sync_state));
1668 if (sc->sc_len == PFSYNC_MINPKT)
1669 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1671 pfsync_q_ins(st, PFSYNC_S_INS, true);
1674 st->sync_updates = 0;
1678 pfsync_defer(struct pf_state *st, struct mbuf *m)
1680 struct pfsync_softc *sc = V_pfsyncif;
1681 struct pfsync_deferral *pd;
1683 if (m->m_flags & (M_BCAST|M_MCAST))
1688 if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) ||
1689 !(sc->sc_flags & PFSYNCF_DEFER)) {
1694 if (sc->sc_deferred >= 128)
1695 pfsync_undefer(TAILQ_FIRST(&sc->sc_deferrals), 0);
1697 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1702 m->m_flags |= M_SKIP_FIREWALL;
1703 st->state_flags |= PFSTATE_ACK;
1711 TAILQ_INSERT_TAIL(&sc->sc_deferrals, pd, pd_entry);
1712 callout_init_mtx(&pd->pd_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1713 callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd);
1721 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1723 struct pfsync_softc *sc = pd->pd_sc;
1724 struct mbuf *m = pd->pd_m;
1725 struct pf_state *st = pd->pd_st;
1727 PFSYNC_LOCK_ASSERT(sc);
1729 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1731 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1733 pf_release_state(st);
1738 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1744 pfsync_defer_tmo(void *arg)
1746 struct pfsync_deferral *pd = arg;
1747 struct pfsync_softc *sc = pd->pd_sc;
1748 struct mbuf *m = pd->pd_m;
1749 struct pf_state *st = pd->pd_st;
1751 PFSYNC_LOCK_ASSERT(sc);
1753 CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
1755 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1757 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1758 if (pd->pd_refs == 0)
1762 ip_output(m, NULL, NULL, 0, NULL, NULL);
1764 pf_release_state(st);
1770 pfsync_undefer_state(struct pf_state *st, int drop)
1772 struct pfsync_softc *sc = V_pfsyncif;
1773 struct pfsync_deferral *pd;
1775 PFSYNC_LOCK_ASSERT(sc);
1777 TAILQ_FOREACH(pd, &sc->sc_deferrals, pd_entry) {
1778 if (pd->pd_st == st) {
1779 if (callout_stop(&pd->pd_tmo) > 0)
1780 pfsync_undefer(pd, drop);
1785 panic("%s: unable to find deferred state", __func__);
1789 pfsync_update_state(struct pf_state *st)
1791 struct pfsync_softc *sc = V_pfsyncif;
1792 bool sync = false, ref = true;
1794 PF_STATE_LOCK_ASSERT(st);
1797 if (st->state_flags & PFSTATE_ACK)
1798 pfsync_undefer_state(st, 0);
1799 if (st->state_flags & PFSTATE_NOSYNC) {
1800 if (st->sync_state != PFSYNC_S_NONE)
1801 pfsync_q_del(st, true);
1806 if (sc->sc_len == PFSYNC_MINPKT)
1807 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1809 switch (st->sync_state) {
1810 case PFSYNC_S_UPD_C:
1813 /* we're already handling it */
1815 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1817 if (st->sync_updates >= sc->sc_maxupdates)
1823 pfsync_q_del(st, false);
1828 pfsync_q_ins(st, PFSYNC_S_UPD_C, ref);
1829 st->sync_updates = 0;
1833 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1836 if (sync || (time_uptime - st->pfsync_time) < 2)
1843 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1845 struct pfsync_softc *sc = V_pfsyncif;
1846 struct pfsync_upd_req_item *item;
1847 size_t nlen = sizeof(struct pfsync_upd_req);
1849 PFSYNC_LOCK_ASSERT(sc);
1852 * This code does a bit to prevent multiple update requests for the
1853 * same state being generated. It searches current subheader queue,
1854 * but it doesn't lookup into queue of already packed datagrams.
1856 TAILQ_FOREACH(item, &sc->sc_upd_req_list, ur_entry)
1857 if (item->ur_msg.id == id &&
1858 item->ur_msg.creatorid == creatorid)
1861 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1863 return; /* XXX stats */
1865 item->ur_msg.id = id;
1866 item->ur_msg.creatorid = creatorid;
1868 if (TAILQ_EMPTY(&sc->sc_upd_req_list))
1869 nlen += sizeof(struct pfsync_subheader);
1871 if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
1874 nlen = sizeof(struct pfsync_subheader) +
1875 sizeof(struct pfsync_upd_req);
1878 TAILQ_INSERT_TAIL(&sc->sc_upd_req_list, item, ur_entry);
1883 pfsync_update_state_req(struct pf_state *st)
1885 struct pfsync_softc *sc = V_pfsyncif;
1888 PF_STATE_LOCK_ASSERT(st);
1891 if (st->state_flags & PFSTATE_NOSYNC) {
1892 if (st->sync_state != PFSYNC_S_NONE)
1893 pfsync_q_del(st, true);
1898 switch (st->sync_state) {
1899 case PFSYNC_S_UPD_C:
1901 pfsync_q_del(st, false);
1906 pfsync_q_ins(st, PFSYNC_S_UPD, ref);
1913 /* we're already handling it */
1917 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1924 pfsync_delete_state(struct pf_state *st)
1926 struct pfsync_softc *sc = V_pfsyncif;
1930 if (st->state_flags & PFSTATE_ACK)
1931 pfsync_undefer_state(st, 1);
1932 if (st->state_flags & PFSTATE_NOSYNC) {
1933 if (st->sync_state != PFSYNC_S_NONE)
1934 pfsync_q_del(st, true);
1939 if (sc->sc_len == PFSYNC_MINPKT)
1940 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1942 switch (st->sync_state) {
1944 /* We never got to tell the world so just forget about it. */
1945 pfsync_q_del(st, true);
1948 case PFSYNC_S_UPD_C:
1951 pfsync_q_del(st, false);
1956 pfsync_q_ins(st, PFSYNC_S_DEL, ref);
1960 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1967 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
1969 struct pfsync_softc *sc = V_pfsyncif;
1971 struct pfsync_subheader subh;
1972 struct pfsync_clr clr;
1975 bzero(&r, sizeof(r));
1977 r.subh.action = PFSYNC_ACT_CLR;
1978 r.subh.count = htons(1);
1979 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
1981 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
1982 r.clr.creatorid = creatorid;
1985 pfsync_send_plus(&r, sizeof(r));
1990 pfsync_q_ins(struct pf_state *st, int q, bool ref)
1992 struct pfsync_softc *sc = V_pfsyncif;
1993 size_t nlen = pfsync_qs[q].len;
1995 PFSYNC_LOCK_ASSERT(sc);
1997 KASSERT(st->sync_state == PFSYNC_S_NONE,
1998 ("%s: st->sync_state %u", __func__, st->sync_state));
1999 KASSERT(sc->sc_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
2002 if (TAILQ_EMPTY(&sc->sc_qs[q]))
2003 nlen += sizeof(struct pfsync_subheader);
2005 if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
2008 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
2012 TAILQ_INSERT_TAIL(&sc->sc_qs[q], st, sync_list);
2019 pfsync_q_del(struct pf_state *st, bool unref)
2021 struct pfsync_softc *sc = V_pfsyncif;
2022 int q = st->sync_state;
2024 PFSYNC_LOCK_ASSERT(sc);
2025 KASSERT(st->sync_state != PFSYNC_S_NONE,
2026 ("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2028 sc->sc_len -= pfsync_qs[q].len;
2029 TAILQ_REMOVE(&sc->sc_qs[q], st, sync_list);
2030 st->sync_state = PFSYNC_S_NONE;
2032 pf_release_state(st);
2034 if (TAILQ_EMPTY(&sc->sc_qs[q]))
2035 sc->sc_len -= sizeof(struct pfsync_subheader);
2039 pfsync_bulk_start(void)
2041 struct pfsync_softc *sc = V_pfsyncif;
2043 if (V_pf_status.debug >= PF_DEBUG_MISC)
2044 printf("pfsync: received bulk update request\n");
2048 sc->sc_ureq_received = time_uptime;
2049 sc->sc_bulk_hashid = 0;
2050 sc->sc_bulk_stateid = 0;
2051 pfsync_bulk_status(PFSYNC_BUS_START);
2052 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2057 pfsync_bulk_update(void *arg)
2059 struct pfsync_softc *sc = arg;
2063 PFSYNC_BLOCK_ASSERT(sc);
2064 CURVNET_SET(sc->sc_ifp->if_vnet);
2067 * Start with last state from previous invocation.
2068 * It may had gone, in this case start from the
2071 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2076 i = sc->sc_bulk_hashid;
2078 for (; i <= pf_hashmask; i++) {
2079 struct pf_idhash *ih = &V_pf_idhash[i];
2082 PF_HASHROW_ASSERT(ih);
2084 PF_HASHROW_LOCK(ih);
2085 s = LIST_FIRST(&ih->states);
2088 for (; s; s = LIST_NEXT(s, entry)) {
2090 if (sent > 1 && (sc->sc_ifp->if_mtu - sc->sc_len) <
2091 sizeof(struct pfsync_state)) {
2092 /* We've filled a packet. */
2093 sc->sc_bulk_hashid = i;
2094 sc->sc_bulk_stateid = s->id;
2095 sc->sc_bulk_creatorid = s->creatorid;
2096 PF_HASHROW_UNLOCK(ih);
2097 callout_reset(&sc->sc_bulk_tmo, 1,
2098 pfsync_bulk_update, sc);
2102 if (s->sync_state == PFSYNC_S_NONE &&
2103 s->timeout < PFTM_MAX &&
2104 s->pfsync_time <= sc->sc_ureq_received) {
2105 pfsync_update_state_req(s);
2109 PF_HASHROW_UNLOCK(ih);
2113 pfsync_bulk_status(PFSYNC_BUS_END);
2120 pfsync_bulk_status(u_int8_t status)
2123 struct pfsync_subheader subh;
2124 struct pfsync_bus bus;
2127 struct pfsync_softc *sc = V_pfsyncif;
2129 bzero(&r, sizeof(r));
2131 r.subh.action = PFSYNC_ACT_BUS;
2132 r.subh.count = htons(1);
2133 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2135 r.bus.creatorid = V_pf_status.hostid;
2136 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2137 r.bus.status = status;
2140 pfsync_send_plus(&r, sizeof(r));
2145 pfsync_bulk_fail(void *arg)
2147 struct pfsync_softc *sc = arg;
2149 CURVNET_SET(sc->sc_ifp->if_vnet);
2151 PFSYNC_BLOCK_ASSERT(sc);
2153 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2155 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2156 pfsync_bulk_fail, V_pfsyncif);
2158 pfsync_request_update(0, 0);
2161 /* Pretend like the transfer was ok. */
2162 sc->sc_ureq_sent = 0;
2163 sc->sc_bulk_tries = 0;
2165 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2166 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
2167 "pfsync bulk fail");
2168 sc->sc_flags |= PFSYNCF_OK;
2170 if (V_pf_status.debug >= PF_DEBUG_MISC)
2171 printf("pfsync: failed to receive bulk update\n");
2178 pfsync_send_plus(void *plus, size_t pluslen)
2180 struct pfsync_softc *sc = V_pfsyncif;
2182 PFSYNC_LOCK_ASSERT(sc);
2184 if (sc->sc_len + pluslen > sc->sc_ifp->if_mtu)
2188 sc->sc_len += (sc->sc_pluslen = pluslen);
2194 pfsync_timeout(void *arg)
2196 struct pfsync_softc *sc = arg;
2198 CURVNET_SET(sc->sc_ifp->if_vnet);
2206 pfsync_push(struct pfsync_softc *sc)
2209 PFSYNC_LOCK_ASSERT(sc);
2211 sc->sc_flags |= PFSYNCF_PUSH;
2212 swi_sched(V_pfsync_swi_cookie, 0);
2216 pfsyncintr(void *arg)
2218 struct pfsync_softc *sc = arg;
2221 CURVNET_SET(sc->sc_ifp->if_vnet);
2224 if ((sc->sc_flags & PFSYNCF_PUSH) && sc->sc_len > PFSYNC_MINPKT) {
2226 sc->sc_flags &= ~PFSYNCF_PUSH;
2228 _IF_DEQUEUE_ALL(&sc->sc_ifp->if_snd, m);
2231 for (; m != NULL; m = n) {
2234 m->m_nextpkt = NULL;
2237 * We distinguish between a deferral packet and our
2238 * own pfsync packet based on M_SKIP_FIREWALL
2239 * flag. This is XXX.
2241 if (m->m_flags & M_SKIP_FIREWALL)
2242 ip_output(m, NULL, NULL, 0, NULL, NULL);
2243 else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo,
2245 V_pfsyncstats.pfsyncs_opackets++;
2247 V_pfsyncstats.pfsyncs_oerrors++;
2253 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship)
2255 struct ip_moptions *imo = &sc->sc_imo;
2258 if (!(ifp->if_flags & IFF_MULTICAST))
2259 return (EADDRNOTAVAIL);
2261 imo->imo_membership = (struct in_multi **)mship;
2262 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
2263 imo->imo_multicast_vif = -1;
2265 if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL,
2266 &imo->imo_membership[0])) != 0) {
2267 imo->imo_membership = NULL;
2270 imo->imo_num_memberships++;
2271 imo->imo_multicast_ifp = ifp;
2272 imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2273 imo->imo_multicast_loop = 0;
2279 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2281 struct ip_moptions *imo = &sc->sc_imo;
2283 in_leavegroup(imo->imo_membership[0], NULL);
2284 free(imo->imo_membership, M_PFSYNC);
2285 imo->imo_membership = NULL;
2286 imo->imo_multicast_ifp = NULL;
2290 extern struct domain inetdomain;
2291 static struct protosw in_pfsync_protosw = {
2292 .pr_type = SOCK_RAW,
2293 .pr_domain = &inetdomain,
2294 .pr_protocol = IPPROTO_PFSYNC,
2295 .pr_flags = PR_ATOMIC|PR_ADDR,
2296 .pr_input = pfsync_input,
2297 .pr_output = rip_output,
2298 .pr_ctloutput = rip_ctloutput,
2299 .pr_usrreqs = &rip_usrreqs
2304 pfsync_pointers_init()
2308 pfsync_state_import_ptr = pfsync_state_import;
2309 pfsync_insert_state_ptr = pfsync_insert_state;
2310 pfsync_update_state_ptr = pfsync_update_state;
2311 pfsync_delete_state_ptr = pfsync_delete_state;
2312 pfsync_clear_states_ptr = pfsync_clear_states;
2313 pfsync_defer_ptr = pfsync_defer;
2318 pfsync_pointers_uninit()
2322 pfsync_state_import_ptr = NULL;
2323 pfsync_insert_state_ptr = NULL;
2324 pfsync_update_state_ptr = NULL;
2325 pfsync_delete_state_ptr = NULL;
2326 pfsync_clear_states_ptr = NULL;
2327 pfsync_defer_ptr = NULL;
2332 vnet_pfsync_init(const void *unused __unused)
2336 V_pfsync_cloner = if_clone_simple(pfsyncname,
2337 pfsync_clone_create, pfsync_clone_destroy, 1);
2338 error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif,
2339 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2341 if_clone_detach(V_pfsync_cloner);
2342 log(LOG_INFO, "swi_add() failed in %s\n", __func__);
2345 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY,
2346 vnet_pfsync_init, NULL);
2349 vnet_pfsync_uninit(const void *unused __unused)
2352 if_clone_detach(V_pfsync_cloner);
2353 swi_remove(V_pfsync_swi_cookie);
2356 * Detach after pf is gone; otherwise we might touch pfsync memory
2357 * from within pf after freeing pfsync.
2359 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_INIT_IF, SI_ORDER_SECOND,
2360 vnet_pfsync_uninit, NULL);
2368 error = pf_proto_register(PF_INET, &in_pfsync_protosw);
2371 error = ipproto_register(IPPROTO_PFSYNC);
2373 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2377 pfsync_pointers_init();
2386 pfsync_pointers_uninit();
2389 ipproto_unregister(IPPROTO_PFSYNC);
2390 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2395 pfsync_modevent(module_t mod, int type, void *data)
2401 error = pfsync_init();
2405 * Module should not be unloaded due to race conditions.
2420 static moduledata_t pfsync_mod = {
2426 #define PFSYNC_MODVER 1
2428 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */
2429 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY);
2430 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2431 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);