2 * Copyright (c) 2002 Michael Shalayeff
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
15 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
16 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
17 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
18 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
19 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
20 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
22 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
23 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
24 * THE POSSIBILITY OF SUCH DAMAGE.
26 * $OpenBSD: if_pfsync.c,v 1.98 2008/06/29 08:42:15 mcbride Exp $
30 #include "opt_inet6.h"
34 #include <sys/param.h>
35 #include <sys/endian.h>
38 #include <sys/systm.h>
41 #include <sys/socket.h>
42 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/module.h>
45 #include <sys/msgport2.h>
46 #include <sys/sockio.h>
47 #include <sys/thread2.h>
49 #include <machine/inttypes.h>
52 #include <net/if_types.h>
53 #include <net/ifq_var.h>
54 #include <net/route.h>
56 #include <net/netisr2.h>
57 #include <net/netmsg2.h>
58 #include <netinet/in.h>
59 #include <netinet/if_ether.h>
60 #include <netinet/ip_carp.h>
61 #include <netinet/tcp.h>
62 #include <netinet/tcp_seq.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip_var.h>
72 #include <netinet6/nd6.h>
75 #include <net/pf/pfvar.h>
76 #include <net/pf/if_pfsync.h>
78 #define PFSYNCNAME "pfsync"
80 #define PFSYNC_MINMTU \
81 (sizeof(struct pfsync_header) + sizeof(struct pf_state))
84 #define DPRINTF(x) do { if (pfsyncdebug) kprintf x ; } while (0)
90 struct pfsync_softc *pfsyncif = NULL;
91 struct pfsyncstats pfsyncstats;
93 void pfsyncattach(int);
94 static int pfsync_clone_destroy(struct ifnet *);
95 static int pfsync_clone_create(struct if_clone *, int, caddr_t);
96 void pfsync_setmtu(struct pfsync_softc *, int);
97 int pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
98 struct pf_state_peer *);
99 int pfsyncoutput(struct ifnet *, struct mbuf *, struct sockaddr *,
101 int pfsyncioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
102 void pfsyncstart(struct ifnet *, struct ifaltq_subque *);
104 struct mbuf *pfsync_get_mbuf(struct pfsync_softc *, u_int8_t, void **);
105 int pfsync_request_update(struct pfsync_state_upd *, struct in_addr *);
106 int pfsync_sendout(struct pfsync_softc *);
107 int pfsync_sendout_mbuf(struct pfsync_softc *, struct mbuf *);
108 void pfsync_timeout(void *);
109 void pfsync_send_bus(struct pfsync_softc *, u_int8_t);
110 void pfsync_bulk_update(void *);
111 void pfsync_bulkfail(void *);
113 static struct in_multi *pfsync_in_addmulti(struct ifnet *);
114 static void pfsync_in_delmulti(struct in_multi *);
116 static MALLOC_DEFINE(M_PFSYNC, PFSYNCNAME, "Packet Filter State Sync. Interface");
117 static LIST_HEAD(pfsync_list, pfsync_softc) pfsync_list;
121 struct if_clone pfsync_cloner =
122 IF_CLONE_INITIALIZER("pfsync", pfsync_clone_create, pfsync_clone_destroy, 1 ,1);
125 pfsyncattach(int npfsync)
127 if_clone_attach(&pfsync_cloner);
130 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param __unused)
132 struct pfsync_softc *sc;
135 lwkt_gettoken(&pf_token);
139 sc = kmalloc(sizeof(*sc), M_PFSYNC, M_WAITOK | M_ZERO);
141 sc->sc_mbuf_net = NULL;
142 sc->sc_mbuf_tdb = NULL;
143 sc->sc_statep.s = NULL;
144 sc->sc_statep_net.s = NULL;
145 sc->sc_statep_tdb.t = NULL;
146 sc->sc_maxupdates = 128;
147 sc->sc_sync_peer.s_addr =htonl(INADDR_PFSYNC_GROUP);
148 sc->sc_sendaddr.s_addr = htonl(INADDR_PFSYNC_GROUP);
149 sc->sc_ureq_received = 0;
150 sc->sc_ureq_sent = 0;
151 sc->sc_bulk_send_next = NULL;
152 sc->sc_bulk_terminator = NULL;
153 sc->sc_bulk_send_cpu = 0;
154 sc->sc_bulk_terminator_cpu = 0;
155 sc->sc_imo.imo_max_memberships = IP_MAX_MEMBERSHIPS;
156 lwkt_reltoken(&pf_token);
159 if_initname(ifp, ifc->ifc_name, unit);
160 ifp->if_ioctl = pfsyncioctl;
161 ifp->if_output = pfsyncoutput;
162 ifp->if_start = pfsyncstart;
163 ifp->if_type = IFT_PFSYNC;
164 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
165 ifp->if_hdrlen = PFSYNC_HDRLEN;
166 ifp->if_baudrate = IF_Mbps(100);
169 pfsync_setmtu(sc, MCLBYTES);
170 callout_init(&sc->sc_tmo);
171 /* callout_init(&sc->sc_tdb_tmo); XXX we don't support tdb (yet) */
172 callout_init(&sc->sc_bulk_tmo);
173 callout_init(&sc->sc_bulkfail_tmo);
175 if_attach(ifp, NULL);
177 bpfattach(&sc->sc_if, DLT_PFSYNC, PFSYNC_HDRLEN);
181 if_addgroup(ifp, "carp");
184 lwkt_gettoken(&pf_token);
185 LIST_INSERT_HEAD(&pfsync_list, sc, sc_next);
186 lwkt_reltoken(&pf_token);
192 pfsync_clone_destroy(struct ifnet *ifp)
194 struct netmsg_base msg;
196 lwkt_gettoken(&pf_token);
197 lwkt_reltoken(&pf_token);
199 struct pfsync_softc *sc = ifp->if_softc;
200 callout_stop(&sc->sc_tmo);
201 /* callout_stop(&sc->sc_tdb_tmo); XXX we don't support tdb (yet) */
202 callout_stop(&sc->sc_bulk_tmo);
203 callout_stop(&sc->sc_bulkfail_tmo);
206 carp_group_demote_adj(&sc->sc_if, -1);
209 /* Unpend async sendouts. */
210 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, netmsg_sync_handler);
211 netisr_domsg(&msg, 0);
217 lwkt_gettoken(&pf_token);
218 LIST_REMOVE(sc, sc_next);
220 lwkt_reltoken(&pf_token);
226 * Start output on the pfsync interface.
229 pfsyncstart(struct ifnet *ifp, struct ifaltq_subque *ifsq)
231 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
236 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
237 struct pf_state_peer *d)
239 if (s->scrub.scrub_flag && d->scrub == NULL) {
240 d->scrub = kmalloc(sizeof(struct pf_state_scrub), M_PFSYNC, M_NOWAIT|M_ZERO);
242 if (d->scrub == NULL)
250 pfsync_state_export(struct pfsync_state *sp, struct pf_state *st)
252 bzero(sp, sizeof(struct pfsync_state));
254 /* copy from state key */
255 sp->key[PF_SK_WIRE].addr[0] = st->key[PF_SK_WIRE]->addr[0];
256 sp->key[PF_SK_WIRE].addr[1] = st->key[PF_SK_WIRE]->addr[1];
257 sp->key[PF_SK_WIRE].port[0] = st->key[PF_SK_WIRE]->port[0];
258 sp->key[PF_SK_WIRE].port[1] = st->key[PF_SK_WIRE]->port[1];
259 sp->key[PF_SK_STACK].addr[0] = st->key[PF_SK_STACK]->addr[0];
260 sp->key[PF_SK_STACK].addr[1] = st->key[PF_SK_STACK]->addr[1];
261 sp->key[PF_SK_STACK].port[0] = st->key[PF_SK_STACK]->port[0];
262 sp->key[PF_SK_STACK].port[1] = st->key[PF_SK_STACK]->port[1];
263 sp->proto = st->key[PF_SK_WIRE]->proto;
264 sp->af = st->key[PF_SK_WIRE]->af;
266 /* copy from state */
267 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname));
268 bcopy(&st->rt_addr, &sp->rt_addr, sizeof(sp->rt_addr));
269 sp->creation = htonl(time_second - st->creation);
270 sp->expire = pf_state_expires(st);
271 if (sp->expire <= time_second)
272 sp->expire = htonl(0);
274 sp->expire = htonl(sp->expire - time_second);
276 sp->direction = st->direction;
278 sp->cpuid = st->cpuid;
279 sp->pickup_mode = st->pickup_mode;
280 sp->timeout = st->timeout;
281 sp->state_flags = st->state_flags;
283 sp->sync_flags |= PFSYNC_FLAG_SRCNODE;
284 if (st->nat_src_node)
285 sp->sync_flags |= PFSYNC_FLAG_NATSRCNODE;
287 bcopy(&st->id, &sp->id, sizeof(sp->id));
288 sp->creatorid = st->creatorid;
289 pf_state_peer_hton(&st->src, &sp->src);
290 pf_state_peer_hton(&st->dst, &sp->dst);
292 if (st->rule.ptr == NULL)
293 sp->rule = htonl(-1);
295 sp->rule = htonl(st->rule.ptr->nr);
296 if (st->anchor.ptr == NULL)
297 sp->anchor = htonl(-1);
299 sp->anchor = htonl(st->anchor.ptr->nr);
300 if (st->nat_rule.ptr == NULL)
301 sp->nat_rule = htonl(-1);
303 sp->nat_rule = htonl(st->nat_rule.ptr->nr);
305 pf_state_counter_hton(st->packets[0], sp->packets[0]);
306 pf_state_counter_hton(st->packets[1], sp->packets[1]);
307 pf_state_counter_hton(st->bytes[0], sp->bytes[0]);
308 pf_state_counter_hton(st->bytes[1], sp->bytes[1]);
313 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
315 struct pf_state *st = NULL;
316 struct pf_state_key *skw = NULL, *sks = NULL;
317 struct pf_rule *r = NULL;
322 if (sp->creatorid == 0 && pf_status.debug >= PF_DEBUG_MISC) {
323 kprintf("pfsync_insert_net_state: invalid creator id:"
324 " %08x\n", ntohl(sp->creatorid));
328 if ((kif = pfi_kif_get(sp->ifname)) == NULL) {
329 if (pf_status.debug >= PF_DEBUG_MISC)
330 kprintf("pfsync_insert_net_state: "
331 "unknown interface: %s\n", sp->ifname);
332 if (flags & PFSYNC_SI_IOCTL)
334 return (0); /* skip this state */
338 * If the ruleset checksums match or the state is coming from the ioctl,
339 * it's safe to associate the state with the rule of that number.
341 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
342 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
343 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
344 r = pf_main_ruleset.rules[
345 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
347 r = &pf_default_rule;
349 if ((r->max_states && r->states_cur >= r->max_states))
352 if (flags & PFSYNC_SI_IOCTL)
353 pool_flags = M_WAITOK | M_NULLOK | M_ZERO;
355 pool_flags = M_WAITOK | M_ZERO;
357 if ((st = kmalloc(sizeof(struct pf_state), M_PFSYNC, pool_flags)) == NULL)
359 lockinit(&st->lk, "pfstlk", 0, 0);
361 if ((skw = pf_alloc_state_key(pool_flags)) == NULL)
364 if (PF_ANEQ(&sp->key[PF_SK_WIRE].addr[0],
365 &sp->key[PF_SK_STACK].addr[0], sp->af) ||
366 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[1],
367 &sp->key[PF_SK_STACK].addr[1], sp->af) ||
368 sp->key[PF_SK_WIRE].port[0] != sp->key[PF_SK_STACK].port[0] ||
369 sp->key[PF_SK_WIRE].port[1] != sp->key[PF_SK_STACK].port[1]) {
370 if ((sks = pf_alloc_state_key(pool_flags)) == NULL)
375 /* allocate memory for scrub info */
376 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
377 pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
380 /* copy to state key(s) */
381 skw->addr[0] = sp->key[PF_SK_WIRE].addr[0];
382 skw->addr[1] = sp->key[PF_SK_WIRE].addr[1];
383 skw->port[0] = sp->key[PF_SK_WIRE].port[0];
384 skw->port[1] = sp->key[PF_SK_WIRE].port[1];
385 skw->proto = sp->proto;
388 sks->addr[0] = sp->key[PF_SK_STACK].addr[0];
389 sks->addr[1] = sp->key[PF_SK_STACK].addr[1];
390 sks->port[0] = sp->key[PF_SK_STACK].port[0];
391 sks->port[1] = sp->key[PF_SK_STACK].port[1];
392 sks->proto = sp->proto;
397 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
398 st->creation = time_second - ntohl(sp->creation);
399 st->expire = time_second;
401 /* XXX No adaptive scaling. */
402 st->expire -= r->timeout[sp->timeout] - ntohl(sp->expire);
405 st->expire = ntohl(sp->expire) + time_second;
406 st->direction = sp->direction;
408 st->timeout = sp->timeout;
409 st->state_flags = sp->state_flags;
410 if (!(flags & PFSYNC_SI_IOCTL))
411 st->sync_flags = PFSTATE_FROMSYNC;
413 bcopy(sp->id, &st->id, sizeof(st->id));
414 st->creatorid = sp->creatorid;
415 pf_state_peer_ntoh(&sp->src, &st->src);
416 pf_state_peer_ntoh(&sp->dst, &st->dst);
419 st->nat_rule.ptr = NULL;
420 st->anchor.ptr = NULL;
426 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
430 if ((error = pf_state_insert(kif, skw, sks, st)) != 0) {
431 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */
443 kfree(skw, M_PFSYNC);
445 kfree(sks, M_PFSYNC);
447 cleanup_state: /* pf_state_insert frees the state keys */
450 kfree(st->dst.scrub, M_PFSYNC);
452 kfree(st->src.scrub, M_PFSYNC);
459 pfsync_input(struct mbuf *m, ...)
461 struct ip *ip = mtod(m, struct ip *);
462 struct pfsync_header *ph;
463 struct pfsync_softc *sc = pfsyncif;
465 struct pf_state_key *sk;
466 struct pf_state_item *si;
467 struct pf_state_cmp id_key;
468 struct pfsync_state *sp;
469 struct pfsync_state_upd *up;
470 struct pfsync_state_del *dp;
471 struct pfsync_state_clr *cp;
472 struct pfsync_state_upd_req *rup;
473 struct pfsync_state_bus *bus;
476 int iplen, action, error, i, count, offp, sfail, stale = 0;
479 /* This function is not yet called from anywhere */
480 /* Still we assume for safety that pf_token must be held */
481 ASSERT_LWKT_TOKEN_HELD(&pf_token);
483 pfsyncstats.pfsyncs_ipackets++;
485 /* verify that we have a sync interface configured */
486 if (!sc || !sc->sc_sync_ifp || !pf_status.running)
489 /* verify that the packet came in on the right interface */
490 if (sc->sc_sync_ifp != m->m_pkthdr.rcvif) {
491 pfsyncstats.pfsyncs_badif++;
495 /* verify that the IP TTL is 255. */
496 if (ip->ip_ttl != PFSYNC_DFLTTL) {
497 pfsyncstats.pfsyncs_badttl++;
501 iplen = ip->ip_hl << 2;
503 if (m->m_pkthdr.len < iplen + sizeof(*ph)) {
504 pfsyncstats.pfsyncs_hdrops++;
508 if (iplen + sizeof(*ph) > m->m_len) {
509 if ((m = m_pullup(m, iplen + sizeof(*ph))) == NULL) {
510 pfsyncstats.pfsyncs_hdrops++;
513 ip = mtod(m, struct ip *);
515 ph = (struct pfsync_header *)((char *)ip + iplen);
517 /* verify the version */
518 if (ph->version != PFSYNC_VERSION) {
519 pfsyncstats.pfsyncs_badver++;
526 /* make sure it's a valid action code */
527 if (action >= PFSYNC_ACT_MAX) {
528 pfsyncstats.pfsyncs_badact++;
532 /* Cheaper to grab this now than having to mess with mbufs later */
535 if (!bcmp(&ph->pf_chksum, &pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
536 flags |= PFSYNC_SI_CKSUM;
539 case PFSYNC_ACT_CLR: {
540 struct pf_state *nexts;
541 struct pf_state_key *nextsk;
543 globaldata_t save_gd = mycpu;
547 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
548 sizeof(*cp), &offp)) == NULL) {
549 pfsyncstats.pfsyncs_badlen++;
552 cp = (struct pfsync_state_clr *)(mp->m_data + offp);
553 creatorid = cp->creatorid;
556 if (cp->ifname[0] == '\0') {
557 lwkt_gettoken(&pf_token);
558 for (nn = 0; nn < ncpus; ++nn) {
559 lwkt_setcpu_self(globaldata_find(nn));
560 for (st = RB_MIN(pf_state_tree_id,
563 nexts = RB_NEXT(pf_state_tree_id,
565 if (st->creatorid == creatorid) {
572 lwkt_setcpu_self(save_gd);
573 lwkt_reltoken(&pf_token);
575 if ((kif = pfi_kif_get(cp->ifname)) == NULL) {
580 lwkt_gettoken(&pf_token);
581 for (nn = 0; nn < ncpus; ++nn) {
582 lwkt_setcpu_self(globaldata_find(nn));
583 for (sk = RB_MIN(pf_state_tree,
587 nextsk = RB_NEXT(pf_state_tree,
588 &pf_statetbl[n], sk);
589 TAILQ_FOREACH(si, &sk->states, entry) {
590 if (si->s->creatorid ==
594 pf_unlink_state(si->s);
599 lwkt_setcpu_self(save_gd);
600 lwkt_reltoken(&pf_token);
607 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
608 count * sizeof(*sp), &offp)) == NULL) {
609 pfsyncstats.pfsyncs_badlen++;
614 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
615 i < count; i++, sp++) {
616 /* check for invalid values */
617 if (sp->timeout >= PFTM_MAX ||
618 sp->src.state > PF_TCPS_PROXY_DST ||
619 sp->dst.state > PF_TCPS_PROXY_DST ||
620 sp->direction > PF_OUT ||
621 (sp->af != AF_INET && sp->af != AF_INET6)) {
622 if (pf_status.debug >= PF_DEBUG_MISC)
623 kprintf("pfsync_insert: PFSYNC_ACT_INS: "
625 pfsyncstats.pfsyncs_badval++;
629 if ((error = pfsync_state_import(sp, flags))) {
630 if (error == ENOMEM) {
639 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
640 count * sizeof(*sp), &offp)) == NULL) {
641 pfsyncstats.pfsyncs_badlen++;
646 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
647 i < count; i++, sp++) {
648 int flags = PFSYNC_FLAG_STALE;
650 /* check for invalid values */
651 if (sp->timeout >= PFTM_MAX ||
652 sp->src.state > PF_TCPS_PROXY_DST ||
653 sp->dst.state > PF_TCPS_PROXY_DST) {
654 if (pf_status.debug >= PF_DEBUG_MISC)
655 kprintf("pfsync_insert: PFSYNC_ACT_UPD: "
657 pfsyncstats.pfsyncs_badval++;
661 bcopy(sp->id, &id_key.id, sizeof(id_key.id));
662 id_key.creatorid = sp->creatorid;
664 st = pf_find_state_byid(&id_key);
666 /* insert the update */
667 if (pfsync_state_import(sp, flags))
668 pfsyncstats.pfsyncs_badstate++;
671 sk = st->key[PF_SK_WIRE]; /* XXX right one? */
673 if (sk->proto == IPPROTO_TCP) {
675 * The state should never go backwards except
676 * for syn-proxy states. Neither should the
677 * sequence window slide backwards.
679 if (st->src.state > sp->src.state &&
680 (st->src.state < PF_TCPS_PROXY_SRC ||
681 sp->src.state >= PF_TCPS_PROXY_SRC))
683 else if (SEQ_GT(st->src.seqlo,
684 ntohl(sp->src.seqlo)))
686 else if (st->dst.state > sp->dst.state) {
687 /* There might still be useful
688 * information about the src state here,
689 * so import that part of the update,
690 * then "fail" so we send the updated
691 * state back to the peer who is missing
692 * our what we know. */
693 pf_state_peer_ntoh(&sp->src, &st->src);
694 /* XXX do anything with timeouts? */
697 } else if (st->dst.state >= TCPS_SYN_SENT &&
698 SEQ_GT(st->dst.seqlo, ntohl(sp->dst.seqlo)))
702 * Non-TCP protocol state machine always go
705 if (st->src.state > sp->src.state)
707 else if (st->dst.state > sp->dst.state)
711 if (pf_status.debug >= PF_DEBUG_MISC)
712 kprintf("pfsync: %s stale update "
715 (sfail < 7 ? "ignoring"
717 (uintmax_t)be64toh(st->id),
718 ntohl(st->creatorid));
719 pfsyncstats.pfsyncs_stale++;
721 if (!(sp->sync_flags & PFSTATE_STALE)) {
722 /* we have a better state, send it */
723 if (sc->sc_mbuf != NULL && !stale)
728 PFSYNC_ACT_UPD, st, flags);
732 pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
733 pf_state_peer_ntoh(&sp->src, &st->src);
734 pf_state_peer_ntoh(&sp->dst, &st->dst);
735 st->expire = ntohl(sp->expire) + time_second;
736 st->timeout = sp->timeout;
738 if (stale && sc->sc_mbuf != NULL)
743 * It's not strictly necessary for us to support the "uncompressed"
744 * delete action, but it's relatively simple and maintains consistency.
747 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
748 count * sizeof(*sp), &offp)) == NULL) {
749 pfsyncstats.pfsyncs_badlen++;
754 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
755 i < count; i++, sp++) {
756 bcopy(sp->id, &id_key.id, sizeof(id_key.id));
757 id_key.creatorid = sp->creatorid;
759 st = pf_find_state_byid(&id_key);
761 pfsyncstats.pfsyncs_badstate++;
764 st->sync_flags |= PFSTATE_FROMSYNC;
769 case PFSYNC_ACT_UPD_C: {
770 int update_requested = 0;
772 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
773 count * sizeof(*up), &offp)) == NULL) {
774 pfsyncstats.pfsyncs_badlen++;
779 for (i = 0, up = (struct pfsync_state_upd *)(mp->m_data + offp);
780 i < count; i++, up++) {
781 /* check for invalid values */
782 if (up->timeout >= PFTM_MAX ||
783 up->src.state > PF_TCPS_PROXY_DST ||
784 up->dst.state > PF_TCPS_PROXY_DST) {
785 if (pf_status.debug >= PF_DEBUG_MISC)
786 kprintf("pfsync_insert: "
789 pfsyncstats.pfsyncs_badval++;
793 bcopy(up->id, &id_key.id, sizeof(id_key.id));
794 id_key.creatorid = up->creatorid;
796 st = pf_find_state_byid(&id_key);
798 /* We don't have this state. Ask for it. */
799 error = pfsync_request_update(up, &src);
800 if (error == ENOMEM) {
804 update_requested = 1;
805 pfsyncstats.pfsyncs_badstate++;
808 sk = st->key[PF_SK_WIRE]; /* XXX right one? */
810 if (sk->proto == IPPROTO_TCP) {
812 * The state should never go backwards except
813 * for syn-proxy states. Neither should the
814 * sequence window slide backwards.
816 if (st->src.state > up->src.state &&
817 (st->src.state < PF_TCPS_PROXY_SRC ||
818 up->src.state >= PF_TCPS_PROXY_SRC))
820 else if (st->dst.state > up->dst.state)
822 else if (SEQ_GT(st->src.seqlo,
823 ntohl(up->src.seqlo)))
825 else if (st->dst.state >= TCPS_SYN_SENT &&
826 SEQ_GT(st->dst.seqlo, ntohl(up->dst.seqlo)))
830 * Non-TCP protocol state machine always go
833 if (st->src.state > up->src.state)
835 else if (st->dst.state > up->dst.state)
839 if (pf_status.debug >= PF_DEBUG_MISC)
840 kprintf("pfsync: ignoring stale update "
841 "(%d) id: %016" PRIx64 " "
842 "creatorid: %08x\n", sfail,
844 ntohl(st->creatorid));
845 pfsyncstats.pfsyncs_stale++;
847 /* we have a better state, send it out */
848 if ((!stale || update_requested) &&
849 sc->sc_mbuf != NULL) {
851 update_requested = 0;
855 pfsync_pack_state(PFSYNC_ACT_UPD, st,
859 pfsync_alloc_scrub_memory(&up->dst, &st->dst);
860 pf_state_peer_ntoh(&up->src, &st->src);
861 pf_state_peer_ntoh(&up->dst, &st->dst);
862 st->expire = ntohl(up->expire) + time_second;
863 st->timeout = up->timeout;
865 if ((update_requested || stale) && sc->sc_mbuf)
870 case PFSYNC_ACT_DEL_C:
871 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
872 count * sizeof(*dp), &offp)) == NULL) {
873 pfsyncstats.pfsyncs_badlen++;
878 for (i = 0, dp = (struct pfsync_state_del *)(mp->m_data + offp);
879 i < count; i++, dp++) {
880 bcopy(dp->id, &id_key.id, sizeof(id_key.id));
881 id_key.creatorid = dp->creatorid;
883 st = pf_find_state_byid(&id_key);
885 pfsyncstats.pfsyncs_badstate++;
888 st->sync_flags |= PFSTATE_FROMSYNC;
893 case PFSYNC_ACT_INS_F:
894 case PFSYNC_ACT_DEL_F:
895 /* not implemented */
897 case PFSYNC_ACT_UREQ:
898 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
899 count * sizeof(*rup), &offp)) == NULL) {
900 pfsyncstats.pfsyncs_badlen++;
905 if (sc->sc_mbuf != NULL)
908 rup = (struct pfsync_state_upd_req *)(mp->m_data + offp);
909 i < count; i++, rup++) {
910 bcopy(rup->id, &id_key.id, sizeof(id_key.id));
911 id_key.creatorid = rup->creatorid;
913 if (id_key.id == 0 && id_key.creatorid == 0) {
914 sc->sc_ureq_received = mycpu->gd_time_seconds;
915 if (sc->sc_bulk_send_next == NULL) {
916 if (++sc->sc_bulk_send_cpu >= ncpus)
917 sc->sc_bulk_send_cpu = 0;
918 sc->sc_bulk_send_next =
919 TAILQ_FIRST(&state_list[sc->sc_bulk_send_cpu]);
921 sc->sc_bulk_terminator =
922 sc->sc_bulk_send_next;
923 sc->sc_bulk_terminator_cpu =
924 sc->sc_bulk_send_cpu;
925 if (pf_status.debug >= PF_DEBUG_MISC)
926 kprintf("pfsync: received "
927 "bulk update request\n");
928 pfsync_send_bus(sc, PFSYNC_BUS_START);
929 lwkt_reltoken(&pf_token);
930 callout_init(&sc->sc_bulk_tmo);
931 lwkt_gettoken(&pf_token);
933 st = pf_find_state_byid(&id_key);
935 pfsyncstats.pfsyncs_badstate++;
939 pfsync_pack_state(PFSYNC_ACT_UPD,
943 if (sc->sc_mbuf != NULL)
948 /* If we're not waiting for a bulk update, who cares. */
949 if (sc->sc_ureq_sent == 0)
952 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
953 sizeof(*bus), &offp)) == NULL) {
954 pfsyncstats.pfsyncs_badlen++;
957 bus = (struct pfsync_state_bus *)(mp->m_data + offp);
958 switch (bus->status) {
959 case PFSYNC_BUS_START:
960 lwkt_reltoken(&pf_token);
961 callout_reset(&sc->sc_bulkfail_tmo,
962 pf_pool_limits[PF_LIMIT_STATES].limit /
963 (PFSYNC_BULKPACKETS * sc->sc_maxcount),
964 pfsync_bulkfail, LIST_FIRST(&pfsync_list));
965 lwkt_gettoken(&pf_token);
966 if (pf_status.debug >= PF_DEBUG_MISC)
967 kprintf("pfsync: received bulk "
971 if (mycpu->gd_time_seconds - ntohl(bus->endtime) >=
973 /* that's it, we're happy */
974 sc->sc_ureq_sent = 0;
975 sc->sc_bulk_tries = 0;
976 lwkt_reltoken(&pf_token);
977 callout_stop(&sc->sc_bulkfail_tmo);
978 lwkt_gettoken(&pf_token);
980 if (!pfsync_sync_ok) {
981 lwkt_reltoken(&pf_token);
982 carp_group_demote_adj(&sc->sc_if, -1);
983 lwkt_gettoken(&pf_token);
987 if (pf_status.debug >= PF_DEBUG_MISC)
988 kprintf("pfsync: received valid "
989 "bulk update end\n");
991 if (pf_status.debug >= PF_DEBUG_MISC)
992 kprintf("pfsync: received invalid "
993 "bulk update end: bad timestamp\n");
1006 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
1015 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
1017 struct pfsync_softc *sc = ifp->if_softc;
1018 struct ifreq *ifr = (struct ifreq *)data;
1019 struct ip_moptions *imo = &sc->sc_imo;
1020 struct pfsyncreq pfsyncr;
1024 lwkt_gettoken(&pf_token);
1029 case SIOCSIFDSTADDR:
1031 if (ifp->if_flags & IFF_UP)
1032 ifp->if_flags |= IFF_RUNNING;
1034 ifp->if_flags &= ~IFF_RUNNING;
1037 if (ifr->ifr_mtu < PFSYNC_MINMTU) {
1038 lwkt_reltoken(&pf_token);
1041 if (ifr->ifr_mtu > MCLBYTES)
1042 ifr->ifr_mtu = MCLBYTES;
1044 if (ifr->ifr_mtu < ifp->if_mtu)
1046 pfsync_setmtu(sc, ifr->ifr_mtu);
1050 bzero(&pfsyncr, sizeof(pfsyncr));
1051 if (sc->sc_sync_ifp)
1052 strlcpy(pfsyncr.pfsyncr_syncdev,
1053 sc->sc_sync_ifp->if_xname, IFNAMSIZ);
1054 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1055 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1056 lwkt_reltoken(&pf_token);
1057 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr))))
1059 lwkt_gettoken(&pf_token);
1062 if ((error = priv_check_cred(cr, PRIV_ROOT, NULL_CRED_OKAY)) != 0) {
1063 lwkt_reltoken(&pf_token);
1066 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) {
1067 lwkt_reltoken(&pf_token);
1071 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1072 sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP;
1074 sc->sc_sync_peer.s_addr =
1075 pfsyncr.pfsyncr_syncpeer.s_addr;
1077 if (pfsyncr.pfsyncr_maxupdates > 255) {
1078 lwkt_reltoken(&pf_token);
1081 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1083 if (pfsyncr.pfsyncr_syncdev[0] == 0) {
1084 sc->sc_sync_ifp = NULL;
1085 if (sc->sc_mbuf_net != NULL) {
1086 /* Don't keep stale pfsync packets around. */
1088 m_freem(sc->sc_mbuf_net);
1089 sc->sc_mbuf_net = NULL;
1090 sc->sc_statep_net.s = NULL;
1093 if (imo->imo_num_memberships > 0) {
1094 pfsync_in_delmulti(imo->imo_membership[--imo->imo_num_memberships]);
1095 imo->imo_multicast_ifp = NULL;
1101 * XXX not that MPSAFE; pfsync needs serious rework
1103 ifnet_deserialize_all(ifp);
1105 sifp = ifunit(pfsyncr.pfsyncr_syncdev);
1107 ifnet_serialize_all(ifp);
1110 lwkt_reltoken(&pf_token);
1115 if (sifp->if_mtu < sc->sc_if.if_mtu ||
1116 (sc->sc_sync_ifp != NULL &&
1117 sifp->if_mtu < sc->sc_sync_ifp->if_mtu) ||
1118 sifp->if_mtu < MCLBYTES - sizeof(struct ip))
1120 sc->sc_sync_ifp = sifp;
1122 pfsync_setmtu(sc, sc->sc_if.if_mtu);
1124 if (imo->imo_num_memberships > 0) {
1125 pfsync_in_delmulti(imo->imo_membership[--imo->imo_num_memberships]);
1126 imo->imo_multicast_ifp = NULL;
1129 if (sc->sc_sync_ifp &&
1130 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) {
1131 if (!(sc->sc_sync_ifp->if_flags & IFF_MULTICAST)) {
1132 sc->sc_sync_ifp = NULL;
1133 lwkt_reltoken(&pf_token);
1135 return (EADDRNOTAVAIL);
1138 if ((imo->imo_membership[0] =
1139 pfsync_in_addmulti(sc->sc_sync_ifp)) == NULL) {
1140 sc->sc_sync_ifp = NULL;
1141 lwkt_reltoken(&pf_token);
1145 imo->imo_num_memberships++;
1146 imo->imo_multicast_ifp = sc->sc_sync_ifp;
1147 imo->imo_multicast_ttl = PFSYNC_DFLTTL;
1148 imo->imo_multicast_loop = 0;
1151 if (sc->sc_sync_ifp ||
1152 sc->sc_sendaddr.s_addr != INADDR_PFSYNC_GROUP) {
1153 /* Request a full state table update. */
1154 sc->sc_ureq_sent = mycpu->gd_time_seconds;
1157 carp_group_demote_adj(&sc->sc_if, 1);
1160 if (pf_status.debug >= PF_DEBUG_MISC)
1161 kprintf("pfsync: requesting bulk update\n");
1162 lwkt_reltoken(&pf_token);
1163 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz,
1164 pfsync_bulkfail, LIST_FIRST(&pfsync_list));
1165 lwkt_gettoken(&pf_token);
1166 error = pfsync_request_update(NULL, NULL);
1167 if (error == ENOMEM) {
1168 lwkt_reltoken(&pf_token);
1179 lwkt_reltoken(&pf_token);
1183 lwkt_reltoken(&pf_token);
1188 pfsync_setmtu(struct pfsync_softc *sc, int mtu_req)
1192 if (sc->sc_sync_ifp && sc->sc_sync_ifp->if_mtu < mtu_req)
1193 mtu = sc->sc_sync_ifp->if_mtu;
1197 sc->sc_maxcount = (mtu - sizeof(struct pfsync_header)) /
1198 sizeof(struct pfsync_state);
1199 if (sc->sc_maxcount > 254)
1200 sc->sc_maxcount = 254;
1201 sc->sc_if.if_mtu = sizeof(struct pfsync_header) +
1202 sc->sc_maxcount * sizeof(struct pfsync_state);
1206 pfsync_get_mbuf(struct pfsync_softc *sc, u_int8_t action, void **sp)
1208 struct pfsync_header *h;
1212 ASSERT_LWKT_TOKEN_HELD(&pf_token);
1214 MGETHDR(m, M_WAITOK, MT_DATA);
1216 IFNET_STAT_INC(&sc->sc_if, oerrors, 1);
1221 case PFSYNC_ACT_CLR:
1222 len = sizeof(struct pfsync_header) +
1223 sizeof(struct pfsync_state_clr);
1225 case PFSYNC_ACT_UPD_C:
1226 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd)) +
1227 sizeof(struct pfsync_header);
1229 case PFSYNC_ACT_DEL_C:
1230 len = (sc->sc_maxcount * sizeof(struct pfsync_state_del)) +
1231 sizeof(struct pfsync_header);
1233 case PFSYNC_ACT_UREQ:
1234 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd_req)) +
1235 sizeof(struct pfsync_header);
1237 case PFSYNC_ACT_BUS:
1238 len = sizeof(struct pfsync_header) +
1239 sizeof(struct pfsync_state_bus);
1241 case PFSYNC_ACT_TDB_UPD:
1242 len = (sc->sc_maxcount * sizeof(struct pfsync_tdb)) +
1243 sizeof(struct pfsync_header);
1246 len = (sc->sc_maxcount * sizeof(struct pfsync_state)) +
1247 sizeof(struct pfsync_header);
1252 MCLGET(m, M_WAITOK);
1253 if ((m->m_flags & M_EXT) == 0) {
1255 IFNET_STAT_INC(&sc->sc_if, oerrors, 1);
1258 m->m_data += (MCLBYTES - len) &~ (sizeof(long) - 1);
1262 m->m_pkthdr.rcvif = NULL;
1263 m->m_pkthdr.len = m->m_len = sizeof(struct pfsync_header);
1264 h = mtod(m, struct pfsync_header *);
1265 h->version = PFSYNC_VERSION;
1270 *sp = (void *)((char *)h + PFSYNC_HDRLEN);
1271 lwkt_reltoken(&pf_token);
1272 callout_reset(&sc->sc_tmo, hz, pfsync_timeout,
1273 LIST_FIRST(&pfsync_list));
1274 lwkt_gettoken(&pf_token);
1279 pfsync_pack_state(u_int8_t action, struct pf_state *st, int flags)
1281 struct ifnet *ifp = NULL;
1282 struct pfsync_softc *sc = pfsyncif;
1283 struct pfsync_header *h, *h_net;
1284 struct pfsync_state *sp = NULL;
1285 struct pfsync_state_upd *up = NULL;
1286 struct pfsync_state_del *dp = NULL;
1288 u_int8_t i = 255, newaction = 0;
1295 * If a packet falls in the forest and there's nobody around to
1296 * hear, does it make a sound?
1298 if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL &&
1299 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) {
1300 /* Don't leave any stale pfsync packets hanging around. */
1301 if (sc->sc_mbuf != NULL) {
1302 m_freem(sc->sc_mbuf);
1304 sc->sc_statep.s = NULL;
1309 if (action >= PFSYNC_ACT_MAX)
1313 if (sc->sc_mbuf == NULL) {
1314 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action,
1315 (void *)&sc->sc_statep.s)) == NULL) {
1319 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1321 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1322 if (h->action != action) {
1324 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action,
1325 (void *)&sc->sc_statep.s)) == NULL) {
1329 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1332 * If it's an update, look in the packet to see if
1333 * we already have an update for the state.
1335 if (action == PFSYNC_ACT_UPD && sc->sc_maxupdates) {
1336 struct pfsync_state *usp =
1337 (void *)((char *)h + PFSYNC_HDRLEN);
1339 for (i = 0; i < h->count; i++) {
1340 if (!memcmp(usp->id, &st->id,
1342 usp->creatorid == st->creatorid) {
1353 st->pfsync_time = mycpu->gd_time_seconds;
1356 /* not a "duplicate" update */
1358 sp = sc->sc_statep.s++;
1359 sc->sc_mbuf->m_pkthdr.len =
1360 sc->sc_mbuf->m_len += sizeof(struct pfsync_state);
1362 bzero(sp, sizeof(*sp));
1364 pfsync_state_export(sp, st);
1366 if (flags & PFSYNC_FLAG_STALE)
1367 sp->sync_flags |= PFSTATE_STALE;
1369 pf_state_peer_hton(&st->src, &sp->src);
1370 pf_state_peer_hton(&st->dst, &sp->dst);
1372 if (st->expire <= time_second)
1373 sp->expire = htonl(0);
1375 sp->expire = htonl(st->expire - time_second);
1378 /* do we need to build "compressed" actions for network transfer? */
1379 if (sc->sc_sync_ifp && flags & PFSYNC_FLAG_COMPRESS) {
1381 case PFSYNC_ACT_UPD:
1382 newaction = PFSYNC_ACT_UPD_C;
1384 case PFSYNC_ACT_DEL:
1385 newaction = PFSYNC_ACT_DEL_C;
1388 /* by default we just send the uncompressed states */
1394 if (sc->sc_mbuf_net == NULL) {
1395 if ((sc->sc_mbuf_net = pfsync_get_mbuf(sc, newaction,
1396 (void *)&sc->sc_statep_net.s)) == NULL) {
1401 h_net = mtod(sc->sc_mbuf_net, struct pfsync_header *);
1403 switch (newaction) {
1404 case PFSYNC_ACT_UPD_C:
1406 up = (void *)((char *)h_net +
1407 PFSYNC_HDRLEN + (i * sizeof(*up)));
1411 sc->sc_mbuf_net->m_pkthdr.len =
1412 sc->sc_mbuf_net->m_len += sizeof(*up);
1413 up = sc->sc_statep_net.u++;
1415 bzero(up, sizeof(*up));
1416 bcopy(&st->id, up->id, sizeof(up->id));
1417 up->creatorid = st->creatorid;
1419 up->timeout = st->timeout;
1420 up->expire = sp->expire;
1424 case PFSYNC_ACT_DEL_C:
1425 sc->sc_mbuf_net->m_pkthdr.len =
1426 sc->sc_mbuf_net->m_len += sizeof(*dp);
1427 dp = sc->sc_statep_net.d++;
1430 bzero(dp, sizeof(*dp));
1431 bcopy(&st->id, dp->id, sizeof(dp->id));
1432 dp->creatorid = st->creatorid;
1437 if (h->count == sc->sc_maxcount ||
1438 (sc->sc_maxupdates && (sp->updates >= sc->sc_maxupdates)))
1439 ret = pfsync_sendout(sc);
1446 pfsync_request_update(struct pfsync_state_upd *up, struct in_addr *src)
1448 struct pfsync_header *h;
1449 struct pfsync_softc *sc = pfsyncif;
1450 struct pfsync_state_upd_req *rup;
1456 if (sc->sc_mbuf == NULL) {
1457 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ,
1458 (void *)&sc->sc_statep.s)) == NULL)
1460 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1462 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1463 if (h->action != PFSYNC_ACT_UREQ) {
1465 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ,
1466 (void *)&sc->sc_statep.s)) == NULL)
1468 h = mtod(sc->sc_mbuf, struct pfsync_header *);
1473 sc->sc_sendaddr = *src;
1474 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*rup);
1476 rup = sc->sc_statep.r++;
1477 bzero(rup, sizeof(*rup));
1479 bcopy(up->id, rup->id, sizeof(rup->id));
1480 rup->creatorid = up->creatorid;
1483 if (h->count == sc->sc_maxcount)
1484 ret = pfsync_sendout(sc);
1490 pfsync_clear_states(u_int32_t creatorid, char *ifname)
1492 struct pfsync_softc *sc = pfsyncif;
1493 struct pfsync_state_clr *cp;
1500 if (sc->sc_mbuf != NULL)
1502 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_CLR,
1503 (void *)&sc->sc_statep.c)) == NULL) {
1507 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*cp);
1508 cp = sc->sc_statep.c;
1509 cp->creatorid = creatorid;
1511 strlcpy(cp->ifname, ifname, IFNAMSIZ);
1513 ret = (pfsync_sendout(sc));
1519 pfsync_timeout(void *v)
1521 struct pfsync_softc *sc = v;
1529 pfsync_send_bus(struct pfsync_softc *sc, u_int8_t status)
1531 struct pfsync_state_bus *bus;
1533 if (sc->sc_mbuf != NULL)
1536 if (pfsync_sync_ok &&
1537 (sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_BUS,
1538 (void *)&sc->sc_statep.b)) != NULL) {
1539 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*bus);
1540 bus = sc->sc_statep.b;
1541 bus->creatorid = pf_status.hostid;
1542 bus->status = status;
1543 bus->endtime = htonl(mycpu->gd_time_seconds - sc->sc_ureq_received);
1549 pfsync_bulk_update(void *v)
1551 struct pfsync_softc *sc = v;
1554 struct pf_state *state;
1556 ASSERT_LWKT_TOKEN_HELD(&pf_token);
1559 if (sc->sc_mbuf != NULL)
1563 * Grab at most PFSYNC_BULKPACKETS worth of states which have not
1564 * been sent since the latest request was made.
1566 state = sc->sc_bulk_send_next;
1567 cpu = sc->sc_bulk_send_cpu;
1570 /* send state update if syncable and not already sent */
1571 if (!state->sync_flags
1572 && state->timeout < PFTM_MAX
1573 && state->pfsync_time <= sc->sc_ureq_received) {
1574 pfsync_pack_state(PFSYNC_ACT_UPD, state, 0);
1578 /* figure next state to send */
1579 state = TAILQ_NEXT(state, entry_list);
1581 /* wrap to start of list if we hit the end */
1582 if (state == NULL) {
1585 state = TAILQ_FIRST(&state_list[cpu]);
1587 } while (i < sc->sc_maxcount * PFSYNC_BULKPACKETS &&
1588 cpu != sc->sc_bulk_terminator_cpu &&
1589 state != sc->sc_bulk_terminator);
1591 if (state == NULL || (cpu == sc->sc_bulk_terminator_cpu &&
1592 state == sc->sc_bulk_terminator)) {
1594 pfsync_send_bus(sc, PFSYNC_BUS_END);
1595 sc->sc_ureq_received = 0;
1596 sc->sc_bulk_send_next = NULL;
1597 sc->sc_bulk_terminator = NULL;
1598 sc->sc_bulk_send_cpu = 0;
1599 sc->sc_bulk_terminator_cpu = 0;
1600 lwkt_reltoken(&pf_token);
1601 callout_stop(&sc->sc_bulk_tmo);
1602 lwkt_gettoken(&pf_token);
1603 if (pf_status.debug >= PF_DEBUG_MISC)
1604 kprintf("pfsync: bulk update complete\n");
1606 /* look again for more in a bit */
1607 lwkt_reltoken(&pf_token);
1608 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_timeout,
1609 LIST_FIRST(&pfsync_list));
1610 lwkt_gettoken(&pf_token);
1611 sc->sc_bulk_send_next = state;
1612 sc->sc_bulk_send_cpu = cpu;
1614 if (sc->sc_mbuf != NULL)
1620 pfsync_bulkfail(void *v)
1622 struct pfsync_softc *sc = v;
1625 ASSERT_LWKT_TOKEN_HELD(&pf_token);
1627 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) {
1628 /* Try again in a bit */
1629 lwkt_reltoken(&pf_token);
1630 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulkfail,
1631 LIST_FIRST(&pfsync_list));
1632 lwkt_gettoken(&pf_token);
1634 error = pfsync_request_update(NULL, NULL);
1635 if (error == ENOMEM) {
1636 if (pf_status.debug >= PF_DEBUG_MISC)
1637 kprintf("pfsync: cannot allocate mbufs for "
1643 /* Pretend like the transfer was ok */
1644 sc->sc_ureq_sent = 0;
1645 sc->sc_bulk_tries = 0;
1647 if (!pfsync_sync_ok)
1648 carp_group_demote_adj(&sc->sc_if, -1);
1651 if (pf_status.debug >= PF_DEBUG_MISC)
1652 kprintf("pfsync: failed to receive "
1653 "bulk update status\n");
1654 lwkt_reltoken(&pf_token);
1655 callout_stop(&sc->sc_bulkfail_tmo);
1656 lwkt_gettoken(&pf_token);
1661 pfsync_sendout_handler(netmsg_t nmsg)
1663 struct netmsg_genpkt *msg = (struct netmsg_genpkt *)nmsg;
1665 pfsync_sendout_mbuf(msg->arg1, msg->m);
1669 pfsync_sendout(struct pfsync_softc *sc)
1672 struct ifnet *ifp = &sc->sc_if;
1675 struct netmsg_genpkt *msg;
1677 ASSERT_LWKT_TOKEN_HELD(&pf_token);
1679 lwkt_reltoken(&pf_token);
1680 callout_stop(&sc->sc_tmo);
1681 lwkt_gettoken(&pf_token);
1683 if (sc->sc_mbuf == NULL)
1687 sc->sc_statep.s = NULL;
1693 bpf_mtap(ifp->if_bpf, m);
1698 if (sc->sc_mbuf_net) {
1700 m = sc->sc_mbuf_net;
1701 sc->sc_mbuf_net = NULL;
1702 sc->sc_statep_net.s = NULL;
1705 msg = &m->m_hdr.mh_genmsg;
1706 netmsg_init(&msg->base, NULL, &netisr_apanic_rport, 0,
1707 pfsync_sendout_handler);
1710 netisr_sendmsg(&msg->base, 0);
1716 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m)
1721 if (sc->sc_sync_ifp ||
1722 sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) {
1723 M_PREPEND(m, sizeof(struct ip), M_WAITOK);
1725 pfsyncstats.pfsyncs_onomem++;
1728 ip = mtod(m, struct ip *);
1729 ip->ip_v = IPVERSION;
1730 ip->ip_hl = sizeof(*ip) >> 2;
1731 ip->ip_tos = IPTOS_LOWDELAY;
1732 ip->ip_len = htons(m->m_pkthdr.len);
1733 ip->ip_id = htons(ip_randomid());
1734 ip->ip_off = htons(IP_DF);
1735 ip->ip_ttl = PFSYNC_DFLTTL;
1736 ip->ip_p = IPPROTO_PFSYNC;
1739 bzero(&sa, sizeof(sa));
1740 ip->ip_src.s_addr = INADDR_ANY;
1742 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP)
1743 m->m_flags |= M_MCAST;
1744 ip->ip_dst = sc->sc_sendaddr;
1745 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr;
1747 pfsyncstats.pfsyncs_opackets++;
1749 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL))
1750 pfsyncstats.pfsyncs_oerrors++;
1758 pfsync_modevent(module_t mod, int type, void *data)
1762 struct pfsync_softc *pfs_if, *tmp;
1764 lwkt_gettoken(&pf_token);
1768 LIST_INIT(&pfsync_list);
1769 lwkt_reltoken(&pf_token);
1770 if_clone_attach(&pfsync_cloner);
1771 lwkt_gettoken(&pf_token);
1772 /* Override the function pointer for pf_ioctl.c */
1776 lwkt_reltoken(&pf_token);
1777 if_clone_detach(&pfsync_cloner);
1778 lwkt_gettoken(&pf_token);
1779 LIST_FOREACH_MUTABLE(pfs_if, &pfsync_list, sc_next, tmp) {
1780 pfsync_clone_destroy(&pfs_if->sc_if);
1789 lwkt_reltoken(&pf_token);
1793 static moduledata_t pfsync_mod = {
1799 #define PFSYNC_MODVER 44
1801 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
1802 MODULE_VERSION(pfsync, PFSYNC_MODVER);
1803 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER);
1806 pfsync_in_addmulti_dispatch(netmsg_t nmsg)
1808 struct lwkt_msg *lmsg = &nmsg->lmsg;
1809 struct ifnet *ifp = lmsg->u.ms_resultp;
1810 struct in_addr addr;
1812 addr.s_addr = INADDR_PFSYNC_GROUP;
1813 lmsg->u.ms_resultp = in_addmulti(&addr, ifp);
1815 lwkt_replymsg(lmsg, 0);
1818 static struct in_multi *
1819 pfsync_in_addmulti(struct ifnet *ifp)
1821 struct netmsg_base nmsg;
1822 struct lwkt_msg *lmsg = &nmsg.lmsg;
1824 netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
1825 pfsync_in_addmulti_dispatch);
1826 lmsg->u.ms_resultp = ifp;
1828 lwkt_domsg(netisr_cpuport(0), lmsg, 0);
1829 return lmsg->u.ms_resultp;
1833 pfsync_in_delmulti_dispatch(netmsg_t nmsg)
1835 struct lwkt_msg *lmsg = &nmsg->lmsg;
1837 in_delmulti(lmsg->u.ms_resultp);
1838 lwkt_replymsg(lmsg, 0);
1842 pfsync_in_delmulti(struct in_multi *inm)
1844 struct netmsg_base nmsg;
1845 struct lwkt_msg *lmsg = &nmsg.lmsg;
1847 netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
1848 pfsync_in_delmulti_dispatch);
1849 lmsg->u.ms_resultp = inm;
1851 lwkt_domsg(netisr_cpuport(0), lmsg, 0);