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>
84 #include <net/if_clone.h>
85 #include <net/if_types.h>
86 #include <net/pfvar.h>
87 #include <net/if_pfsync.h>
89 #include <netinet/if_ether.h>
90 #include <netinet/in.h>
91 #include <netinet/in_var.h>
92 #include <netinet/ip.h>
93 #include <netinet/ip_carp.h>
94 #include <netinet/ip_var.h>
95 #include <netinet/tcp.h>
96 #include <netinet/tcp_fsm.h>
97 #include <netinet/tcp_seq.h>
99 #define PFSYNC_MINPKT ( \
100 sizeof(struct ip) + \
101 sizeof(struct pfsync_header) + \
102 sizeof(struct pfsync_subheader) )
110 static int pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *,
111 struct pfsync_state_peer *);
112 static int pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int);
113 static int pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int);
114 static int pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int);
115 static int pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int);
116 static int pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int);
117 static int pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int);
118 static int pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int);
119 static int pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int);
120 static int pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int);
121 static int pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int);
122 static int pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int);
123 static int pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int);
125 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = {
126 pfsync_in_clr, /* PFSYNC_ACT_CLR */
127 pfsync_in_ins, /* PFSYNC_ACT_INS */
128 pfsync_in_iack, /* PFSYNC_ACT_INS_ACK */
129 pfsync_in_upd, /* PFSYNC_ACT_UPD */
130 pfsync_in_upd_c, /* PFSYNC_ACT_UPD_C */
131 pfsync_in_ureq, /* PFSYNC_ACT_UPD_REQ */
132 pfsync_in_del, /* PFSYNC_ACT_DEL */
133 pfsync_in_del_c, /* PFSYNC_ACT_DEL_C */
134 pfsync_in_error, /* PFSYNC_ACT_INS_F */
135 pfsync_in_error, /* PFSYNC_ACT_DEL_F */
136 pfsync_in_bus, /* PFSYNC_ACT_BUS */
137 pfsync_in_tdb, /* PFSYNC_ACT_TDB */
138 pfsync_in_eof /* PFSYNC_ACT_EOF */
142 void (*write)(struct pf_state *, void *);
147 /* we have one of these for every PFSYNC_S_ */
148 static void pfsync_out_state(struct pf_state *, void *);
149 static void pfsync_out_iack(struct pf_state *, void *);
150 static void pfsync_out_upd_c(struct pf_state *, void *);
151 static void pfsync_out_del(struct pf_state *, void *);
153 static struct pfsync_q pfsync_qs[] = {
154 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_INS },
155 { pfsync_out_iack, sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK },
156 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_UPD },
157 { pfsync_out_upd_c, sizeof(struct pfsync_upd_c), PFSYNC_ACT_UPD_C },
158 { pfsync_out_del, sizeof(struct pfsync_del_c), PFSYNC_ACT_DEL_C }
161 static void pfsync_q_ins(struct pf_state *, int);
162 static void pfsync_q_del(struct pf_state *);
164 static void pfsync_update_state(struct pf_state *);
166 struct pfsync_upd_req_item {
167 TAILQ_ENTRY(pfsync_upd_req_item) ur_entry;
168 struct pfsync_upd_req ur_msg;
171 struct pfsync_deferral {
172 struct pfsync_softc *pd_sc;
173 TAILQ_ENTRY(pfsync_deferral) pd_entry;
175 struct callout pd_tmo;
177 struct pf_state *pd_st;
181 struct pfsync_softc {
183 struct ifnet *sc_ifp;
184 struct ifnet *sc_sync_if;
185 struct ip_moptions sc_imo;
186 struct in_addr sc_sync_peer;
188 #define PFSYNCF_OK 0x00000001
189 #define PFSYNCF_DEFER 0x00000002
190 #define PFSYNCF_PUSH 0x00000004
191 uint8_t sc_maxupdates;
192 struct ip sc_template;
193 struct callout sc_tmo;
198 TAILQ_HEAD(, pf_state) sc_qs[PFSYNC_S_COUNT];
199 TAILQ_HEAD(, pfsync_upd_req_item) sc_upd_req_list;
200 TAILQ_HEAD(, pfsync_deferral) sc_deferrals;
205 /* Bulk update info */
206 struct mtx sc_bulk_mtx;
207 uint32_t sc_ureq_sent;
209 uint32_t sc_ureq_received;
211 uint64_t sc_bulk_stateid;
212 uint32_t sc_bulk_creatorid;
213 struct callout sc_bulk_tmo;
214 struct callout sc_bulkfail_tmo;
217 #define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx)
218 #define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx)
219 #define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED)
221 #define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx)
222 #define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx)
223 #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED)
225 static const char pfsyncname[] = "pfsync";
226 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data");
227 static VNET_DEFINE(struct pfsync_softc *, pfsyncif) = NULL;
228 #define V_pfsyncif VNET(pfsyncif)
229 static VNET_DEFINE(void *, pfsync_swi_cookie) = NULL;
230 #define V_pfsync_swi_cookie VNET(pfsync_swi_cookie)
231 static VNET_DEFINE(struct pfsyncstats, pfsyncstats);
232 #define V_pfsyncstats VNET(pfsyncstats)
233 static VNET_DEFINE(int, pfsync_carp_adj) = CARP_MAXSKEW;
234 #define V_pfsync_carp_adj VNET(pfsync_carp_adj)
236 static void pfsync_timeout(void *);
237 static void pfsync_push(struct pfsync_softc *);
238 static void pfsyncintr(void *);
239 static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *,
241 static void pfsync_multicast_cleanup(struct pfsync_softc *);
242 static void pfsync_pointers_init(void);
243 static void pfsync_pointers_uninit(void);
244 static int pfsync_init(void);
245 static void pfsync_uninit(void);
247 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC");
248 SYSCTL_VNET_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_RW,
249 &VNET_NAME(pfsyncstats), pfsyncstats,
250 "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)");
251 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW,
252 &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment");
254 static int pfsync_clone_create(struct if_clone *, int, caddr_t);
255 static void pfsync_clone_destroy(struct ifnet *);
256 static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
257 struct pf_state_peer *);
258 static int pfsyncoutput(struct ifnet *, struct mbuf *,
259 const struct sockaddr *, struct route *);
260 static int pfsyncioctl(struct ifnet *, u_long, caddr_t);
262 static int pfsync_defer(struct pf_state *, struct mbuf *);
263 static void pfsync_undefer(struct pfsync_deferral *, int);
264 static void pfsync_undefer_state(struct pf_state *, int);
265 static void pfsync_defer_tmo(void *);
267 static void pfsync_request_update(u_int32_t, u_int64_t);
268 static void pfsync_update_state_req(struct pf_state *);
270 static void pfsync_drop(struct pfsync_softc *);
271 static void pfsync_sendout(int);
272 static void pfsync_send_plus(void *, size_t);
274 static void pfsync_bulk_start(void);
275 static void pfsync_bulk_status(u_int8_t);
276 static void pfsync_bulk_update(void *);
277 static void pfsync_bulk_fail(void *);
280 static void pfsync_update_net_tdb(struct pfsync_tdb *);
283 #define PFSYNC_MAX_BULKTRIES 12
285 VNET_DEFINE(struct if_clone *, pfsync_cloner);
286 #define V_pfsync_cloner VNET(pfsync_cloner)
289 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param)
291 struct pfsync_softc *sc;
298 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO);
299 sc->sc_flags |= PFSYNCF_OK;
301 for (q = 0; q < PFSYNC_S_COUNT; q++)
302 TAILQ_INIT(&sc->sc_qs[q]);
304 TAILQ_INIT(&sc->sc_upd_req_list);
305 TAILQ_INIT(&sc->sc_deferrals);
307 sc->sc_len = PFSYNC_MINPKT;
308 sc->sc_maxupdates = 128;
310 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC);
315 if_initname(ifp, pfsyncname, unit);
317 ifp->if_ioctl = pfsyncioctl;
318 ifp->if_output = pfsyncoutput;
319 ifp->if_type = IFT_PFSYNC;
320 ifp->if_snd.ifq_maxlen = ifqmaxlen;
321 ifp->if_hdrlen = sizeof(struct pfsync_header);
322 ifp->if_mtu = ETHERMTU;
323 mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF);
324 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF);
325 callout_init(&sc->sc_tmo, CALLOUT_MPSAFE);
326 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0);
327 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0);
331 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN);
339 pfsync_clone_destroy(struct ifnet *ifp)
341 struct pfsync_softc *sc = ifp->if_softc;
344 * At this stage, everything should have already been
345 * cleared by pfsync_uninit(), and we have only to
348 while (sc->sc_deferred > 0) {
349 struct pfsync_deferral *pd = TAILQ_FIRST(&sc->sc_deferrals);
351 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
353 if (callout_stop(&pd->pd_tmo)) {
354 pf_release_state(pd->pd_st);
359 callout_drain(&pd->pd_tmo);
364 callout_drain(&sc->sc_tmo);
365 callout_drain(&sc->sc_bulkfail_tmo);
366 callout_drain(&sc->sc_bulk_tmo);
368 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
369 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy");
376 if (sc->sc_imo.imo_membership)
377 pfsync_multicast_cleanup(sc);
378 mtx_destroy(&sc->sc_mtx);
379 mtx_destroy(&sc->sc_bulk_mtx);
386 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
387 struct pf_state_peer *d)
389 if (s->scrub.scrub_flag && d->scrub == NULL) {
390 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO);
391 if (d->scrub == NULL)
400 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
402 struct pfsync_softc *sc = V_pfsyncif;
403 #ifndef __NO_STRICT_ALIGNMENT
404 struct pfsync_state_key key[2];
406 struct pfsync_state_key *kw, *ks;
407 struct pf_state *st = NULL;
408 struct pf_state_key *skw = NULL, *sks = NULL;
409 struct pf_rule *r = NULL;
415 if (sp->creatorid == 0) {
416 if (V_pf_status.debug >= PF_DEBUG_MISC)
417 printf("%s: invalid creator id: %08x\n", __func__,
418 ntohl(sp->creatorid));
422 if ((kif = pfi_kif_find(sp->ifname)) == NULL) {
423 if (V_pf_status.debug >= PF_DEBUG_MISC)
424 printf("%s: unknown interface: %s\n", __func__,
426 if (flags & PFSYNC_SI_IOCTL)
428 return (0); /* skip this state */
432 * If the ruleset checksums match or the state is coming from the ioctl,
433 * it's safe to associate the state with the rule of that number.
435 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
436 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
437 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
438 r = pf_main_ruleset.rules[
439 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
441 r = &V_pf_default_rule;
443 if ((r->max_states &&
444 counter_u64_fetch(r->states_cur) >= r->max_states))
448 * XXXGL: consider M_WAITOK in ioctl path after.
450 if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL)
453 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL)
456 #ifndef __NO_STRICT_ALIGNMENT
457 bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2);
458 kw = &key[PF_SK_WIRE];
459 ks = &key[PF_SK_STACK];
461 kw = &sp->key[PF_SK_WIRE];
462 ks = &sp->key[PF_SK_STACK];
465 if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) ||
466 PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) ||
467 kw->port[0] != ks->port[0] ||
468 kw->port[1] != ks->port[1]) {
469 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
475 /* allocate memory for scrub info */
476 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
477 pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
480 /* Copy to state key(s). */
481 skw->addr[0] = kw->addr[0];
482 skw->addr[1] = kw->addr[1];
483 skw->port[0] = kw->port[0];
484 skw->port[1] = kw->port[1];
485 skw->proto = sp->proto;
488 sks->addr[0] = ks->addr[0];
489 sks->addr[1] = ks->addr[1];
490 sks->port[0] = ks->port[0];
491 sks->port[1] = ks->port[1];
492 sks->proto = sp->proto;
497 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
498 st->creation = time_uptime - ntohl(sp->creation);
499 st->expire = time_uptime;
503 timeout = r->timeout[sp->timeout];
505 timeout = V_pf_default_rule.timeout[sp->timeout];
507 /* sp->expire may have been adaptively scaled by export. */
508 st->expire -= timeout - ntohl(sp->expire);
511 st->direction = sp->direction;
513 st->timeout = sp->timeout;
514 st->state_flags = sp->state_flags;
517 st->creatorid = sp->creatorid;
518 pf_state_peer_ntoh(&sp->src, &st->src);
519 pf_state_peer_ntoh(&sp->dst, &st->dst);
522 st->nat_rule.ptr = NULL;
523 st->anchor.ptr = NULL;
526 st->pfsync_time = time_uptime;
527 st->sync_state = PFSYNC_S_NONE;
529 if (!(flags & PFSYNC_SI_IOCTL))
530 st->state_flags |= PFSTATE_NOSYNC;
532 if ((error = pf_state_insert(kif, skw, sks, st)) != 0)
535 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
536 counter_u64_add(r->states_cur, 1);
537 counter_u64_add(r->states_tot, 1);
539 if (!(flags & PFSYNC_SI_IOCTL)) {
540 st->state_flags &= ~PFSTATE_NOSYNC;
541 if (st->state_flags & PFSTATE_ACK) {
542 pfsync_q_ins(st, PFSYNC_S_IACK);
546 st->state_flags &= ~PFSTATE_ACK;
556 uma_zfree(V_pf_state_key_z, skw);
558 uma_zfree(V_pf_state_key_z, sks);
560 cleanup_state: /* pf_state_insert() frees the state keys. */
563 uma_zfree(V_pf_state_scrub_z, st->dst.scrub);
565 uma_zfree(V_pf_state_scrub_z, st->src.scrub);
566 uma_zfree(V_pf_state_z, st);
572 pfsync_input(struct mbuf *m, __unused int off)
574 struct pfsync_softc *sc = V_pfsyncif;
575 struct pfsync_pkt pkt;
576 struct ip *ip = mtod(m, struct ip *);
577 struct pfsync_header *ph;
578 struct pfsync_subheader subh;
584 V_pfsyncstats.pfsyncs_ipackets++;
586 /* Verify that we have a sync interface configured. */
587 if (!sc || !sc->sc_sync_if || !V_pf_status.running ||
588 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
591 /* verify that the packet came in on the right interface */
592 if (sc->sc_sync_if != m->m_pkthdr.rcvif) {
593 V_pfsyncstats.pfsyncs_badif++;
597 sc->sc_ifp->if_ipackets++;
598 sc->sc_ifp->if_ibytes += m->m_pkthdr.len;
599 /* verify that the IP TTL is 255. */
600 if (ip->ip_ttl != PFSYNC_DFLTTL) {
601 V_pfsyncstats.pfsyncs_badttl++;
605 offset = ip->ip_hl << 2;
606 if (m->m_pkthdr.len < offset + sizeof(*ph)) {
607 V_pfsyncstats.pfsyncs_hdrops++;
611 if (offset + sizeof(*ph) > m->m_len) {
612 if (m_pullup(m, offset + sizeof(*ph)) == NULL) {
613 V_pfsyncstats.pfsyncs_hdrops++;
616 ip = mtod(m, struct ip *);
618 ph = (struct pfsync_header *)((char *)ip + offset);
620 /* verify the version */
621 if (ph->version != PFSYNC_VERSION) {
622 V_pfsyncstats.pfsyncs_badver++;
626 len = ntohs(ph->len) + offset;
627 if (m->m_pkthdr.len < len) {
628 V_pfsyncstats.pfsyncs_badlen++;
632 /* Cheaper to grab this now than having to mess with mbufs later */
634 pkt.src = ip->ip_src;
638 * Trusting pf_chksum during packet processing, as well as seeking
639 * in interface name tree, require holding PF_RULES_RLOCK().
642 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
643 pkt.flags |= PFSYNC_SI_CKSUM;
645 offset += sizeof(*ph);
646 while (offset <= len - sizeof(subh)) {
647 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh);
648 offset += sizeof(subh);
650 if (subh.action >= PFSYNC_ACT_MAX) {
651 V_pfsyncstats.pfsyncs_badact++;
656 count = ntohs(subh.count);
657 V_pfsyncstats.pfsyncs_iacts[subh.action] += count;
658 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count);
673 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
675 struct pfsync_clr *clr;
677 int len = sizeof(*clr) * count;
681 mp = m_pulldown(m, offset, len, &offp);
683 V_pfsyncstats.pfsyncs_badlen++;
686 clr = (struct pfsync_clr *)(mp->m_data + offp);
688 for (i = 0; i < count; i++) {
689 creatorid = clr[i].creatorid;
691 if (clr[i].ifname[0] != '\0' &&
692 pfi_kif_find(clr[i].ifname) == NULL)
695 for (int i = 0; i <= pf_hashmask; i++) {
696 struct pf_idhash *ih = &V_pf_idhash[i];
700 LIST_FOREACH(s, &ih->states, entry) {
701 if (s->creatorid == creatorid) {
702 s->state_flags |= PFSTATE_NOSYNC;
703 pf_unlink_state(s, PF_ENTER_LOCKED);
707 PF_HASHROW_UNLOCK(ih);
715 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
718 struct pfsync_state *sa, *sp;
719 int len = sizeof(*sp) * count;
722 mp = m_pulldown(m, offset, len, &offp);
724 V_pfsyncstats.pfsyncs_badlen++;
727 sa = (struct pfsync_state *)(mp->m_data + offp);
729 for (i = 0; i < count; i++) {
732 /* Check for invalid values. */
733 if (sp->timeout >= PFTM_MAX ||
734 sp->src.state > PF_TCPS_PROXY_DST ||
735 sp->dst.state > PF_TCPS_PROXY_DST ||
736 sp->direction > PF_OUT ||
737 (sp->af != AF_INET && sp->af != AF_INET6)) {
738 if (V_pf_status.debug >= PF_DEBUG_MISC)
739 printf("%s: invalid value\n", __func__);
740 V_pfsyncstats.pfsyncs_badval++;
744 if (pfsync_state_import(sp, pkt->flags) == ENOMEM)
745 /* Drop out, but process the rest of the actions. */
753 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
755 struct pfsync_ins_ack *ia, *iaa;
759 int len = count * sizeof(*ia);
762 mp = m_pulldown(m, offset, len, &offp);
764 V_pfsyncstats.pfsyncs_badlen++;
767 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp);
769 for (i = 0; i < count; i++) {
772 st = pf_find_state_byid(ia->id, ia->creatorid);
776 if (st->state_flags & PFSTATE_ACK) {
777 PFSYNC_LOCK(V_pfsyncif);
778 pfsync_undefer_state(st, 0);
779 PFSYNC_UNLOCK(V_pfsyncif);
784 * XXX this is not yet implemented, but we know the size of the
785 * message so we can skip it.
788 return (count * sizeof(struct pfsync_ins_ack));
792 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src,
793 struct pfsync_state_peer *dst)
797 PF_STATE_LOCK_ASSERT(st);
800 * The state should never go backwards except
801 * for syn-proxy states. Neither should the
802 * sequence window slide backwards.
804 if ((st->src.state > src->state &&
805 (st->src.state < PF_TCPS_PROXY_SRC ||
806 src->state >= PF_TCPS_PROXY_SRC)) ||
808 (st->src.state == src->state &&
809 SEQ_GT(st->src.seqlo, ntohl(src->seqlo))))
812 pf_state_peer_ntoh(src, &st->src);
814 if ((st->dst.state > dst->state) ||
816 (st->dst.state >= TCPS_SYN_SENT &&
817 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo))))
820 pf_state_peer_ntoh(dst, &st->dst);
826 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
828 struct pfsync_softc *sc = V_pfsyncif;
829 struct pfsync_state *sa, *sp;
834 int len = count * sizeof(*sp);
837 mp = m_pulldown(m, offset, len, &offp);
839 V_pfsyncstats.pfsyncs_badlen++;
842 sa = (struct pfsync_state *)(mp->m_data + offp);
844 for (i = 0; i < count; i++) {
847 /* check for invalid values */
848 if (sp->timeout >= PFTM_MAX ||
849 sp->src.state > PF_TCPS_PROXY_DST ||
850 sp->dst.state > PF_TCPS_PROXY_DST) {
851 if (V_pf_status.debug >= PF_DEBUG_MISC) {
852 printf("pfsync_input: PFSYNC_ACT_UPD: "
855 V_pfsyncstats.pfsyncs_badval++;
859 st = pf_find_state_byid(sp->id, sp->creatorid);
861 /* insert the update */
862 if (pfsync_state_import(sp, 0))
863 V_pfsyncstats.pfsyncs_badstate++;
867 if (st->state_flags & PFSTATE_ACK) {
869 pfsync_undefer_state(st, 1);
873 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
874 sync = pfsync_upd_tcp(st, &sp->src, &sp->dst);
879 * Non-TCP protocol state machine always go
882 if (st->src.state > sp->src.state)
885 pf_state_peer_ntoh(&sp->src, &st->src);
886 if (st->dst.state > sp->dst.state)
889 pf_state_peer_ntoh(&sp->dst, &st->dst);
892 pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
893 pf_state_peer_ntoh(&sp->dst, &st->dst);
894 st->expire = time_uptime;
895 st->timeout = sp->timeout;
897 st->pfsync_time = time_uptime;
900 V_pfsyncstats.pfsyncs_stale++;
902 pfsync_update_state(st);
916 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
918 struct pfsync_softc *sc = V_pfsyncif;
919 struct pfsync_upd_c *ua, *up;
921 int len = count * sizeof(*up);
926 mp = m_pulldown(m, offset, len, &offp);
928 V_pfsyncstats.pfsyncs_badlen++;
931 ua = (struct pfsync_upd_c *)(mp->m_data + offp);
933 for (i = 0; i < count; i++) {
936 /* check for invalid values */
937 if (up->timeout >= PFTM_MAX ||
938 up->src.state > PF_TCPS_PROXY_DST ||
939 up->dst.state > PF_TCPS_PROXY_DST) {
940 if (V_pf_status.debug >= PF_DEBUG_MISC) {
941 printf("pfsync_input: "
945 V_pfsyncstats.pfsyncs_badval++;
949 st = pf_find_state_byid(up->id, up->creatorid);
951 /* We don't have this state. Ask for it. */
953 pfsync_request_update(up->creatorid, up->id);
958 if (st->state_flags & PFSTATE_ACK) {
960 pfsync_undefer_state(st, 1);
964 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP)
965 sync = pfsync_upd_tcp(st, &up->src, &up->dst);
970 * Non-TCP protocol state machine always go
973 if (st->src.state > up->src.state)
976 pf_state_peer_ntoh(&up->src, &st->src);
977 if (st->dst.state > up->dst.state)
980 pf_state_peer_ntoh(&up->dst, &st->dst);
983 pfsync_alloc_scrub_memory(&up->dst, &st->dst);
984 pf_state_peer_ntoh(&up->dst, &st->dst);
985 st->expire = time_uptime;
986 st->timeout = up->timeout;
988 st->pfsync_time = time_uptime;
991 V_pfsyncstats.pfsyncs_stale++;
993 pfsync_update_state(st);
1000 PF_STATE_UNLOCK(st);
1007 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1009 struct pfsync_upd_req *ur, *ura;
1011 int len = count * sizeof(*ur);
1014 struct pf_state *st;
1016 mp = m_pulldown(m, offset, len, &offp);
1018 V_pfsyncstats.pfsyncs_badlen++;
1021 ura = (struct pfsync_upd_req *)(mp->m_data + offp);
1023 for (i = 0; i < count; i++) {
1026 if (ur->id == 0 && ur->creatorid == 0)
1027 pfsync_bulk_start();
1029 st = pf_find_state_byid(ur->id, ur->creatorid);
1031 V_pfsyncstats.pfsyncs_badstate++;
1034 if (st->state_flags & PFSTATE_NOSYNC) {
1035 PF_STATE_UNLOCK(st);
1039 pfsync_update_state_req(st);
1040 PF_STATE_UNLOCK(st);
1048 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1051 struct pfsync_state *sa, *sp;
1052 struct pf_state *st;
1053 int len = count * sizeof(*sp);
1056 mp = m_pulldown(m, offset, len, &offp);
1058 V_pfsyncstats.pfsyncs_badlen++;
1061 sa = (struct pfsync_state *)(mp->m_data + offp);
1063 for (i = 0; i < count; i++) {
1066 st = pf_find_state_byid(sp->id, sp->creatorid);
1068 V_pfsyncstats.pfsyncs_badstate++;
1071 st->state_flags |= PFSTATE_NOSYNC;
1072 pf_unlink_state(st, PF_ENTER_LOCKED);
1079 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1082 struct pfsync_del_c *sa, *sp;
1083 struct pf_state *st;
1084 int len = count * sizeof(*sp);
1087 mp = m_pulldown(m, offset, len, &offp);
1089 V_pfsyncstats.pfsyncs_badlen++;
1092 sa = (struct pfsync_del_c *)(mp->m_data + offp);
1094 for (i = 0; i < count; i++) {
1097 st = pf_find_state_byid(sp->id, sp->creatorid);
1099 V_pfsyncstats.pfsyncs_badstate++;
1103 st->state_flags |= PFSTATE_NOSYNC;
1104 pf_unlink_state(st, PF_ENTER_LOCKED);
1111 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1113 struct pfsync_softc *sc = V_pfsyncif;
1114 struct pfsync_bus *bus;
1116 int len = count * sizeof(*bus);
1121 /* If we're not waiting for a bulk update, who cares. */
1122 if (sc->sc_ureq_sent == 0) {
1127 mp = m_pulldown(m, offset, len, &offp);
1130 V_pfsyncstats.pfsyncs_badlen++;
1133 bus = (struct pfsync_bus *)(mp->m_data + offp);
1135 switch (bus->status) {
1136 case PFSYNC_BUS_START:
1137 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz +
1138 V_pf_limits[PF_LIMIT_STATES].limit /
1139 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) /
1140 sizeof(struct pfsync_state)),
1141 pfsync_bulk_fail, sc);
1142 if (V_pf_status.debug >= PF_DEBUG_MISC)
1143 printf("pfsync: received bulk update start\n");
1146 case PFSYNC_BUS_END:
1147 if (time_uptime - ntohl(bus->endtime) >=
1149 /* that's it, we're happy */
1150 sc->sc_ureq_sent = 0;
1151 sc->sc_bulk_tries = 0;
1152 callout_stop(&sc->sc_bulkfail_tmo);
1153 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1154 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
1155 "pfsync bulk done");
1156 sc->sc_flags |= PFSYNCF_OK;
1157 if (V_pf_status.debug >= PF_DEBUG_MISC)
1158 printf("pfsync: received valid "
1159 "bulk update end\n");
1161 if (V_pf_status.debug >= PF_DEBUG_MISC)
1162 printf("pfsync: received invalid "
1163 "bulk update end: bad timestamp\n");
1173 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1175 int len = count * sizeof(struct pfsync_tdb);
1178 struct pfsync_tdb *tp;
1184 mp = m_pulldown(m, offset, len, &offp);
1186 V_pfsyncstats.pfsyncs_badlen++;
1189 tp = (struct pfsync_tdb *)(mp->m_data + offp);
1191 for (i = 0; i < count; i++)
1192 pfsync_update_net_tdb(&tp[i]);
1199 /* Update an in-kernel tdb. Silently fail if no tdb is found. */
1201 pfsync_update_net_tdb(struct pfsync_tdb *pt)
1206 /* check for invalid values */
1207 if (ntohl(pt->spi) <= SPI_RESERVED_MAX ||
1208 (pt->dst.sa.sa_family != AF_INET &&
1209 pt->dst.sa.sa_family != AF_INET6))
1212 tdb = gettdb(pt->spi, &pt->dst, pt->sproto);
1214 pt->rpl = ntohl(pt->rpl);
1215 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes);
1217 /* Neither replay nor byte counter should ever decrease. */
1218 if (pt->rpl < tdb->tdb_rpl ||
1219 pt->cur_bytes < tdb->tdb_cur_bytes) {
1223 tdb->tdb_rpl = pt->rpl;
1224 tdb->tdb_cur_bytes = pt->cur_bytes;
1229 if (V_pf_status.debug >= PF_DEBUG_MISC)
1230 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: "
1232 V_pfsyncstats.pfsyncs_badstate++;
1239 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1241 /* check if we are at the right place in the packet */
1242 if (offset != m->m_pkthdr.len)
1243 V_pfsyncstats.pfsyncs_badlen++;
1245 /* we're done. free and let the caller return */
1251 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count)
1253 V_pfsyncstats.pfsyncs_badact++;
1260 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
1269 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1271 struct pfsync_softc *sc = ifp->if_softc;
1272 struct ifreq *ifr = (struct ifreq *)data;
1273 struct pfsyncreq pfsyncr;
1279 if (ifp->if_flags & IFF_UP) {
1280 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1282 pfsync_pointers_init();
1284 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1286 pfsync_pointers_uninit();
1290 if (!sc->sc_sync_if ||
1291 ifr->ifr_mtu <= PFSYNC_MINPKT ||
1292 ifr->ifr_mtu > sc->sc_sync_if->if_mtu)
1294 if (ifr->ifr_mtu < ifp->if_mtu) {
1296 if (sc->sc_len > PFSYNC_MINPKT)
1300 ifp->if_mtu = ifr->ifr_mtu;
1303 bzero(&pfsyncr, sizeof(pfsyncr));
1305 if (sc->sc_sync_if) {
1306 strlcpy(pfsyncr.pfsyncr_syncdev,
1307 sc->sc_sync_if->if_xname, IFNAMSIZ);
1309 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1310 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1311 pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER ==
1312 (sc->sc_flags & PFSYNCF_DEFER));
1314 return (copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)));
1318 struct ip_moptions *imo = &sc->sc_imo;
1323 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0)
1325 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr))))
1328 if (pfsyncr.pfsyncr_maxupdates > 255)
1331 if (pfsyncr.pfsyncr_syncdev[0] == 0)
1333 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL)
1336 if (sifp != NULL && (
1337 pfsyncr.pfsyncr_syncpeer.s_addr == 0 ||
1338 pfsyncr.pfsyncr_syncpeer.s_addr ==
1339 htonl(INADDR_PFSYNC_GROUP)))
1340 mship = malloc((sizeof(struct in_multi *) *
1341 IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO);
1344 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1345 sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP);
1347 sc->sc_sync_peer.s_addr =
1348 pfsyncr.pfsyncr_syncpeer.s_addr;
1350 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1351 if (pfsyncr.pfsyncr_defer) {
1352 sc->sc_flags |= PFSYNCF_DEFER;
1353 pfsync_defer_ptr = pfsync_defer;
1355 sc->sc_flags &= ~PFSYNCF_DEFER;
1356 pfsync_defer_ptr = NULL;
1361 if_rele(sc->sc_sync_if);
1362 sc->sc_sync_if = NULL;
1363 if (imo->imo_membership)
1364 pfsync_multicast_cleanup(sc);
1369 if (sc->sc_len > PFSYNC_MINPKT &&
1370 (sifp->if_mtu < sc->sc_ifp->if_mtu ||
1371 (sc->sc_sync_if != NULL &&
1372 sifp->if_mtu < sc->sc_sync_if->if_mtu) ||
1373 sifp->if_mtu < MCLBYTES - sizeof(struct ip)))
1376 if (imo->imo_membership)
1377 pfsync_multicast_cleanup(sc);
1379 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) {
1380 error = pfsync_multicast_setup(sc, sifp, mship);
1383 free(mship, M_PFSYNC);
1388 if_rele(sc->sc_sync_if);
1389 sc->sc_sync_if = sifp;
1391 ip = &sc->sc_template;
1392 bzero(ip, sizeof(*ip));
1393 ip->ip_v = IPVERSION;
1394 ip->ip_hl = sizeof(sc->sc_template) >> 2;
1395 ip->ip_tos = IPTOS_LOWDELAY;
1396 /* len and id are set later. */
1397 ip->ip_off = htons(IP_DF);
1398 ip->ip_ttl = PFSYNC_DFLTTL;
1399 ip->ip_p = IPPROTO_PFSYNC;
1400 ip->ip_src.s_addr = INADDR_ANY;
1401 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr;
1403 /* Request a full state table update. */
1404 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
1405 (*carp_demote_adj_p)(V_pfsync_carp_adj,
1406 "pfsync bulk start");
1407 sc->sc_flags &= ~PFSYNCF_OK;
1408 if (V_pf_status.debug >= PF_DEBUG_MISC)
1409 printf("pfsync: requesting bulk update\n");
1410 pfsync_request_update(0, 0);
1413 sc->sc_ureq_sent = time_uptime;
1414 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail,
1428 pfsync_out_state(struct pf_state *st, void *buf)
1430 struct pfsync_state *sp = buf;
1432 pfsync_state_export(sp, st);
1436 pfsync_out_iack(struct pf_state *st, void *buf)
1438 struct pfsync_ins_ack *iack = buf;
1441 iack->creatorid = st->creatorid;
1445 pfsync_out_upd_c(struct pf_state *st, void *buf)
1447 struct pfsync_upd_c *up = buf;
1449 bzero(up, sizeof(*up));
1451 pf_state_peer_hton(&st->src, &up->src);
1452 pf_state_peer_hton(&st->dst, &up->dst);
1453 up->creatorid = st->creatorid;
1454 up->timeout = st->timeout;
1458 pfsync_out_del(struct pf_state *st, void *buf)
1460 struct pfsync_del_c *dp = buf;
1463 dp->creatorid = st->creatorid;
1464 st->state_flags |= PFSTATE_NOSYNC;
1468 pfsync_drop(struct pfsync_softc *sc)
1470 struct pf_state *st, *next;
1471 struct pfsync_upd_req_item *ur;
1474 for (q = 0; q < PFSYNC_S_COUNT; q++) {
1475 if (TAILQ_EMPTY(&sc->sc_qs[q]))
1478 TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, next) {
1479 KASSERT(st->sync_state == q,
1480 ("%s: st->sync_state == q",
1482 st->sync_state = PFSYNC_S_NONE;
1483 pf_release_state(st);
1485 TAILQ_INIT(&sc->sc_qs[q]);
1488 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1489 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1494 sc->sc_len = PFSYNC_MINPKT;
1498 pfsync_sendout(int schedswi)
1500 struct pfsync_softc *sc = V_pfsyncif;
1501 struct ifnet *ifp = sc->sc_ifp;
1504 struct pfsync_header *ph;
1505 struct pfsync_subheader *subh;
1506 struct pf_state *st;
1507 struct pfsync_upd_req_item *ur;
1511 KASSERT(sc != NULL, ("%s: null sc", __func__));
1512 KASSERT(sc->sc_len > PFSYNC_MINPKT,
1513 ("%s: sc_len %zu", __func__, sc->sc_len));
1514 PFSYNC_LOCK_ASSERT(sc);
1516 if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) {
1521 m = m_get2(max_linkhdr + sc->sc_len, M_NOWAIT, MT_DATA, M_PKTHDR);
1523 sc->sc_ifp->if_oerrors++;
1524 V_pfsyncstats.pfsyncs_onomem++;
1527 m->m_data += max_linkhdr;
1528 m->m_len = m->m_pkthdr.len = sc->sc_len;
1530 /* build the ip header */
1531 ip = (struct ip *)m->m_data;
1532 bcopy(&sc->sc_template, ip, sizeof(*ip));
1533 offset = sizeof(*ip);
1535 ip->ip_len = htons(m->m_pkthdr.len);
1536 ip->ip_id = htons(ip_randomid());
1538 /* build the pfsync header */
1539 ph = (struct pfsync_header *)(m->m_data + offset);
1540 bzero(ph, sizeof(*ph));
1541 offset += sizeof(*ph);
1543 ph->version = PFSYNC_VERSION;
1544 ph->len = htons(sc->sc_len - sizeof(*ip));
1545 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH);
1547 /* walk the queues */
1548 for (q = 0; q < PFSYNC_S_COUNT; q++) {
1549 if (TAILQ_EMPTY(&sc->sc_qs[q]))
1552 subh = (struct pfsync_subheader *)(m->m_data + offset);
1553 offset += sizeof(*subh);
1556 TAILQ_FOREACH(st, &sc->sc_qs[q], sync_list) {
1557 KASSERT(st->sync_state == q,
1558 ("%s: st->sync_state == q",
1561 * XXXGL: some of write methods do unlocked reads
1564 pfsync_qs[q].write(st, m->m_data + offset);
1565 offset += pfsync_qs[q].len;
1566 st->sync_state = PFSYNC_S_NONE;
1567 pf_release_state(st);
1570 TAILQ_INIT(&sc->sc_qs[q]);
1572 bzero(subh, sizeof(*subh));
1573 subh->action = pfsync_qs[q].action;
1574 subh->count = htons(count);
1575 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count;
1578 if (!TAILQ_EMPTY(&sc->sc_upd_req_list)) {
1579 subh = (struct pfsync_subheader *)(m->m_data + offset);
1580 offset += sizeof(*subh);
1583 while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) {
1584 TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry);
1586 bcopy(&ur->ur_msg, m->m_data + offset,
1587 sizeof(ur->ur_msg));
1588 offset += sizeof(ur->ur_msg);
1593 bzero(subh, sizeof(*subh));
1594 subh->action = PFSYNC_ACT_UPD_REQ;
1595 subh->count = htons(count);
1596 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count;
1599 /* has someone built a custom region for us to add? */
1600 if (sc->sc_plus != NULL) {
1601 bcopy(sc->sc_plus, m->m_data + offset, sc->sc_pluslen);
1602 offset += sc->sc_pluslen;
1607 subh = (struct pfsync_subheader *)(m->m_data + offset);
1608 offset += sizeof(*subh);
1610 bzero(subh, sizeof(*subh));
1611 subh->action = PFSYNC_ACT_EOF;
1612 subh->count = htons(1);
1613 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++;
1615 /* we're done, let's put it on the wire */
1617 m->m_data += sizeof(*ip);
1618 m->m_len = m->m_pkthdr.len = sc->sc_len - sizeof(*ip);
1620 m->m_data -= sizeof(*ip);
1621 m->m_len = m->m_pkthdr.len = sc->sc_len;
1624 if (sc->sc_sync_if == NULL) {
1625 sc->sc_len = PFSYNC_MINPKT;
1630 sc->sc_ifp->if_opackets++;
1631 sc->sc_ifp->if_obytes += m->m_pkthdr.len;
1632 sc->sc_len = PFSYNC_MINPKT;
1634 if (!_IF_QFULL(&sc->sc_ifp->if_snd))
1635 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1638 sc->sc_ifp->if_snd.ifq_drops++;
1641 swi_sched(V_pfsync_swi_cookie, 0);
1645 pfsync_insert_state(struct pf_state *st)
1647 struct pfsync_softc *sc = V_pfsyncif;
1649 if (st->state_flags & PFSTATE_NOSYNC)
1652 if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) ||
1653 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) {
1654 st->state_flags |= PFSTATE_NOSYNC;
1658 KASSERT(st->sync_state == PFSYNC_S_NONE,
1659 ("%s: st->sync_state %u", __func__, st->sync_state));
1662 if (sc->sc_len == PFSYNC_MINPKT)
1663 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1665 pfsync_q_ins(st, PFSYNC_S_INS);
1668 st->sync_updates = 0;
1672 pfsync_defer(struct pf_state *st, struct mbuf *m)
1674 struct pfsync_softc *sc = V_pfsyncif;
1675 struct pfsync_deferral *pd;
1677 if (m->m_flags & (M_BCAST|M_MCAST))
1682 if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) ||
1683 !(sc->sc_flags & PFSYNCF_DEFER)) {
1688 if (sc->sc_deferred >= 128)
1689 pfsync_undefer(TAILQ_FIRST(&sc->sc_deferrals), 0);
1691 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT);
1696 m->m_flags |= M_SKIP_FIREWALL;
1697 st->state_flags |= PFSTATE_ACK;
1705 TAILQ_INSERT_TAIL(&sc->sc_deferrals, pd, pd_entry);
1706 callout_init_mtx(&pd->pd_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1707 callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd);
1715 pfsync_undefer(struct pfsync_deferral *pd, int drop)
1717 struct pfsync_softc *sc = pd->pd_sc;
1718 struct mbuf *m = pd->pd_m;
1719 struct pf_state *st = pd->pd_st;
1721 PFSYNC_LOCK_ASSERT(sc);
1723 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1725 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1727 pf_release_state(st);
1732 _IF_ENQUEUE(&sc->sc_ifp->if_snd, m);
1738 pfsync_defer_tmo(void *arg)
1740 struct pfsync_deferral *pd = arg;
1741 struct pfsync_softc *sc = pd->pd_sc;
1742 struct mbuf *m = pd->pd_m;
1743 struct pf_state *st = pd->pd_st;
1745 PFSYNC_LOCK_ASSERT(sc);
1747 CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
1749 TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry);
1751 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */
1752 if (pd->pd_refs == 0)
1756 ip_output(m, NULL, NULL, 0, NULL, NULL);
1758 pf_release_state(st);
1764 pfsync_undefer_state(struct pf_state *st, int drop)
1766 struct pfsync_softc *sc = V_pfsyncif;
1767 struct pfsync_deferral *pd;
1769 PFSYNC_LOCK_ASSERT(sc);
1771 TAILQ_FOREACH(pd, &sc->sc_deferrals, pd_entry) {
1772 if (pd->pd_st == st) {
1773 if (callout_stop(&pd->pd_tmo))
1774 pfsync_undefer(pd, drop);
1779 panic("%s: unable to find deferred state", __func__);
1783 pfsync_update_state(struct pf_state *st)
1785 struct pfsync_softc *sc = V_pfsyncif;
1788 PF_STATE_LOCK_ASSERT(st);
1791 if (st->state_flags & PFSTATE_ACK)
1792 pfsync_undefer_state(st, 0);
1793 if (st->state_flags & PFSTATE_NOSYNC) {
1794 if (st->sync_state != PFSYNC_S_NONE)
1800 if (sc->sc_len == PFSYNC_MINPKT)
1801 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1803 switch (st->sync_state) {
1804 case PFSYNC_S_UPD_C:
1807 /* we're already handling it */
1809 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) {
1811 if (st->sync_updates >= sc->sc_maxupdates)
1819 pfsync_q_ins(st, PFSYNC_S_UPD_C);
1820 st->sync_updates = 0;
1824 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1827 if (sync || (time_uptime - st->pfsync_time) < 2)
1834 pfsync_request_update(u_int32_t creatorid, u_int64_t id)
1836 struct pfsync_softc *sc = V_pfsyncif;
1837 struct pfsync_upd_req_item *item;
1838 size_t nlen = sizeof(struct pfsync_upd_req);
1840 PFSYNC_LOCK_ASSERT(sc);
1843 * This code does a bit to prevent multiple update requests for the
1844 * same state being generated. It searches current subheader queue,
1845 * but it doesn't lookup into queue of already packed datagrams.
1847 TAILQ_FOREACH(item, &sc->sc_upd_req_list, ur_entry)
1848 if (item->ur_msg.id == id &&
1849 item->ur_msg.creatorid == creatorid)
1852 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT);
1854 return; /* XXX stats */
1856 item->ur_msg.id = id;
1857 item->ur_msg.creatorid = creatorid;
1859 if (TAILQ_EMPTY(&sc->sc_upd_req_list))
1860 nlen += sizeof(struct pfsync_subheader);
1862 if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
1865 nlen = sizeof(struct pfsync_subheader) +
1866 sizeof(struct pfsync_upd_req);
1869 TAILQ_INSERT_TAIL(&sc->sc_upd_req_list, item, ur_entry);
1874 pfsync_update_state_req(struct pf_state *st)
1876 struct pfsync_softc *sc = V_pfsyncif;
1878 PF_STATE_LOCK_ASSERT(st);
1881 if (st->state_flags & PFSTATE_NOSYNC) {
1882 if (st->sync_state != PFSYNC_S_NONE)
1888 switch (st->sync_state) {
1889 case PFSYNC_S_UPD_C:
1893 pfsync_q_ins(st, PFSYNC_S_UPD);
1900 /* we're already handling it */
1904 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1911 pfsync_delete_state(struct pf_state *st)
1913 struct pfsync_softc *sc = V_pfsyncif;
1916 if (st->state_flags & PFSTATE_ACK)
1917 pfsync_undefer_state(st, 1);
1918 if (st->state_flags & PFSTATE_NOSYNC) {
1919 if (st->sync_state != PFSYNC_S_NONE)
1925 if (sc->sc_len == PFSYNC_MINPKT)
1926 callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif);
1928 switch (st->sync_state) {
1930 /* We never got to tell the world so just forget about it. */
1934 case PFSYNC_S_UPD_C:
1938 /* FALLTHROUGH to putting it on the del list */
1941 pfsync_q_ins(st, PFSYNC_S_DEL);
1945 panic("%s: unexpected sync state %d", __func__, st->sync_state);
1951 pfsync_clear_states(u_int32_t creatorid, const char *ifname)
1953 struct pfsync_softc *sc = V_pfsyncif;
1955 struct pfsync_subheader subh;
1956 struct pfsync_clr clr;
1959 bzero(&r, sizeof(r));
1961 r.subh.action = PFSYNC_ACT_CLR;
1962 r.subh.count = htons(1);
1963 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++;
1965 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname));
1966 r.clr.creatorid = creatorid;
1969 pfsync_send_plus(&r, sizeof(r));
1974 pfsync_q_ins(struct pf_state *st, int q)
1976 struct pfsync_softc *sc = V_pfsyncif;
1977 size_t nlen = pfsync_qs[q].len;
1979 PFSYNC_LOCK_ASSERT(sc);
1981 KASSERT(st->sync_state == PFSYNC_S_NONE,
1982 ("%s: st->sync_state %u", __func__, st->sync_state));
1983 KASSERT(sc->sc_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu",
1986 if (TAILQ_EMPTY(&sc->sc_qs[q]))
1987 nlen += sizeof(struct pfsync_subheader);
1989 if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) {
1992 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len;
1996 TAILQ_INSERT_TAIL(&sc->sc_qs[q], st, sync_list);
2002 pfsync_q_del(struct pf_state *st)
2004 struct pfsync_softc *sc = V_pfsyncif;
2005 int q = st->sync_state;
2007 PFSYNC_LOCK_ASSERT(sc);
2008 KASSERT(st->sync_state != PFSYNC_S_NONE,
2009 ("%s: st->sync_state != PFSYNC_S_NONE", __func__));
2011 sc->sc_len -= pfsync_qs[q].len;
2012 TAILQ_REMOVE(&sc->sc_qs[q], st, sync_list);
2013 st->sync_state = PFSYNC_S_NONE;
2014 pf_release_state(st);
2016 if (TAILQ_EMPTY(&sc->sc_qs[q]))
2017 sc->sc_len -= sizeof(struct pfsync_subheader);
2021 pfsync_bulk_start(void)
2023 struct pfsync_softc *sc = V_pfsyncif;
2025 if (V_pf_status.debug >= PF_DEBUG_MISC)
2026 printf("pfsync: received bulk update request\n");
2030 sc->sc_ureq_received = time_uptime;
2031 sc->sc_bulk_hashid = 0;
2032 sc->sc_bulk_stateid = 0;
2033 pfsync_bulk_status(PFSYNC_BUS_START);
2034 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc);
2039 pfsync_bulk_update(void *arg)
2041 struct pfsync_softc *sc = arg;
2045 PFSYNC_BLOCK_ASSERT(sc);
2046 CURVNET_SET(sc->sc_ifp->if_vnet);
2049 * Start with last state from previous invocation.
2050 * It may had gone, in this case start from the
2053 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid);
2058 i = sc->sc_bulk_hashid;
2060 for (; i <= pf_hashmask; i++) {
2061 struct pf_idhash *ih = &V_pf_idhash[i];
2064 PF_HASHROW_ASSERT(ih);
2066 PF_HASHROW_LOCK(ih);
2067 s = LIST_FIRST(&ih->states);
2070 for (; s; s = LIST_NEXT(s, entry)) {
2072 if (sent > 1 && (sc->sc_ifp->if_mtu - sc->sc_len) <
2073 sizeof(struct pfsync_state)) {
2074 /* We've filled a packet. */
2075 sc->sc_bulk_hashid = i;
2076 sc->sc_bulk_stateid = s->id;
2077 sc->sc_bulk_creatorid = s->creatorid;
2078 PF_HASHROW_UNLOCK(ih);
2079 callout_reset(&sc->sc_bulk_tmo, 1,
2080 pfsync_bulk_update, sc);
2084 if (s->sync_state == PFSYNC_S_NONE &&
2085 s->timeout < PFTM_MAX &&
2086 s->pfsync_time <= sc->sc_ureq_received) {
2087 pfsync_update_state_req(s);
2091 PF_HASHROW_UNLOCK(ih);
2095 pfsync_bulk_status(PFSYNC_BUS_END);
2102 pfsync_bulk_status(u_int8_t status)
2105 struct pfsync_subheader subh;
2106 struct pfsync_bus bus;
2109 struct pfsync_softc *sc = V_pfsyncif;
2111 bzero(&r, sizeof(r));
2113 r.subh.action = PFSYNC_ACT_BUS;
2114 r.subh.count = htons(1);
2115 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++;
2117 r.bus.creatorid = V_pf_status.hostid;
2118 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received);
2119 r.bus.status = status;
2122 pfsync_send_plus(&r, sizeof(r));
2127 pfsync_bulk_fail(void *arg)
2129 struct pfsync_softc *sc = arg;
2131 CURVNET_SET(sc->sc_ifp->if_vnet);
2133 PFSYNC_BLOCK_ASSERT(sc);
2135 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
2137 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
2138 pfsync_bulk_fail, V_pfsyncif);
2140 pfsync_request_update(0, 0);
2143 /* Pretend like the transfer was ok. */
2144 sc->sc_ureq_sent = 0;
2145 sc->sc_bulk_tries = 0;
2147 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p)
2148 (*carp_demote_adj_p)(-V_pfsync_carp_adj,
2149 "pfsync bulk fail");
2150 sc->sc_flags |= PFSYNCF_OK;
2152 if (V_pf_status.debug >= PF_DEBUG_MISC)
2153 printf("pfsync: failed to receive bulk update\n");
2160 pfsync_send_plus(void *plus, size_t pluslen)
2162 struct pfsync_softc *sc = V_pfsyncif;
2164 PFSYNC_LOCK_ASSERT(sc);
2166 if (sc->sc_len + pluslen > sc->sc_ifp->if_mtu)
2170 sc->sc_len += (sc->sc_pluslen = pluslen);
2176 pfsync_timeout(void *arg)
2178 struct pfsync_softc *sc = arg;
2180 CURVNET_SET(sc->sc_ifp->if_vnet);
2188 pfsync_push(struct pfsync_softc *sc)
2191 PFSYNC_LOCK_ASSERT(sc);
2193 sc->sc_flags |= PFSYNCF_PUSH;
2194 swi_sched(V_pfsync_swi_cookie, 0);
2198 pfsyncintr(void *arg)
2200 struct pfsync_softc *sc = arg;
2203 CURVNET_SET(sc->sc_ifp->if_vnet);
2206 if ((sc->sc_flags & PFSYNCF_PUSH) && sc->sc_len > PFSYNC_MINPKT) {
2208 sc->sc_flags &= ~PFSYNCF_PUSH;
2210 _IF_DEQUEUE_ALL(&sc->sc_ifp->if_snd, m);
2213 for (; m != NULL; m = n) {
2216 m->m_nextpkt = NULL;
2219 * We distinguish between a deferral packet and our
2220 * own pfsync packet based on M_SKIP_FIREWALL
2221 * flag. This is XXX.
2223 if (m->m_flags & M_SKIP_FIREWALL)
2224 ip_output(m, NULL, NULL, 0, NULL, NULL);
2225 else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo,
2227 V_pfsyncstats.pfsyncs_opackets++;
2229 V_pfsyncstats.pfsyncs_oerrors++;
2235 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship)
2237 struct ip_moptions *imo = &sc->sc_imo;
2240 if (!(ifp->if_flags & IFF_MULTICAST))
2241 return (EADDRNOTAVAIL);
2243 imo->imo_membership = (struct in_multi **)mship;
2244 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
2245 imo->imo_multicast_vif = -1;
2247 if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL,
2248 &imo->imo_membership[0])) != 0) {
2249 imo->imo_membership = NULL;
2252 imo->imo_num_memberships++;
2253 imo->imo_multicast_ifp = ifp;
2254 imo->imo_multicast_ttl = PFSYNC_DFLTTL;
2255 imo->imo_multicast_loop = 0;
2261 pfsync_multicast_cleanup(struct pfsync_softc *sc)
2263 struct ip_moptions *imo = &sc->sc_imo;
2265 in_leavegroup(imo->imo_membership[0], NULL);
2266 free(imo->imo_membership, M_PFSYNC);
2267 imo->imo_membership = NULL;
2268 imo->imo_multicast_ifp = NULL;
2272 extern struct domain inetdomain;
2273 static struct protosw in_pfsync_protosw = {
2274 .pr_type = SOCK_RAW,
2275 .pr_domain = &inetdomain,
2276 .pr_protocol = IPPROTO_PFSYNC,
2277 .pr_flags = PR_ATOMIC|PR_ADDR,
2278 .pr_input = pfsync_input,
2279 .pr_output = (pr_output_t *)rip_output,
2280 .pr_ctloutput = rip_ctloutput,
2281 .pr_usrreqs = &rip_usrreqs
2286 pfsync_pointers_init()
2290 pfsync_state_import_ptr = pfsync_state_import;
2291 pfsync_insert_state_ptr = pfsync_insert_state;
2292 pfsync_update_state_ptr = pfsync_update_state;
2293 pfsync_delete_state_ptr = pfsync_delete_state;
2294 pfsync_clear_states_ptr = pfsync_clear_states;
2295 pfsync_defer_ptr = pfsync_defer;
2300 pfsync_pointers_uninit()
2304 pfsync_state_import_ptr = NULL;
2305 pfsync_insert_state_ptr = NULL;
2306 pfsync_update_state_ptr = NULL;
2307 pfsync_delete_state_ptr = NULL;
2308 pfsync_clear_states_ptr = NULL;
2309 pfsync_defer_ptr = NULL;
2316 VNET_ITERATOR_DECL(vnet_iter);
2320 VNET_FOREACH(vnet_iter) {
2321 CURVNET_SET(vnet_iter);
2322 V_pfsync_cloner = if_clone_simple(pfsyncname,
2323 pfsync_clone_create, pfsync_clone_destroy, 1);
2324 error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif,
2325 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie);
2330 VNET_LIST_RUNLOCK();
2332 error = pf_proto_register(PF_INET, &in_pfsync_protosw);
2335 error = ipproto_register(IPPROTO_PFSYNC);
2337 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2341 pfsync_pointers_init();
2348 VNET_FOREACH(vnet_iter) {
2349 CURVNET_SET(vnet_iter);
2350 if (V_pfsync_swi_cookie) {
2351 swi_remove(V_pfsync_swi_cookie);
2352 if_clone_detach(V_pfsync_cloner);
2356 VNET_LIST_RUNLOCK();
2364 VNET_ITERATOR_DECL(vnet_iter);
2366 pfsync_pointers_uninit();
2368 ipproto_unregister(IPPROTO_PFSYNC);
2369 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW);
2371 VNET_FOREACH(vnet_iter) {
2372 CURVNET_SET(vnet_iter);
2373 if_clone_detach(V_pfsync_cloner);
2374 swi_remove(V_pfsync_swi_cookie);
2377 VNET_LIST_RUNLOCK();
2381 pfsync_modevent(module_t mod, int type, void *data)
2387 error = pfsync_init();
2391 * Module should not be unloaded due to race conditions.
2406 static moduledata_t pfsync_mod = {
2412 #define PFSYNC_MODVER 1
2414 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);
2415 MODULE_VERSION(pfsync, PFSYNC_MODVER);
2416 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);