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"
33 #include <sys/param.h>
34 #include <sys/endian.h>
37 #include <sys/systm.h>
40 #include <sys/socket.h>
41 #include <sys/kernel.h>
42 #include <sys/malloc.h>
43 #include <sys/module.h>
44 #include <sys/msgport2.h>
45 #include <sys/sockio.h>
46 #include <sys/thread2.h>
48 #include <machine/inttypes.h>
51 #include <net/if_types.h>
52 #include <net/ifq_var.h>
53 #include <net/route.h>
55 #include <net/netisr2.h>
56 #include <net/netmsg2.h>
57 #include <netinet/in.h>
58 #include <netinet/if_ether.h>
59 #include <netinet/ip_carp.h>
60 #include <netinet/tcp.h>
61 #include <netinet/tcp_seq.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip.h>
67 #include <netinet/ip_var.h>
71 #include <netinet6/nd6.h>
74 #include <net/pf/pfvar.h>
75 #include <net/pf/if_pfsync.h>
77 #define PFSYNCNAME "pfsync"
79 #define PFSYNC_MINMTU \
80 (sizeof(struct pfsync_header) + sizeof(struct pf_state))
83 #define DPRINTF(x) do { if (pfsyncdebug) kprintf x ; } while (0)
89 struct pfsync_softc *pfsyncif = NULL;
90 struct pfsyncstats pfsyncstats;
92 void pfsyncattach(int);
93 static int pfsync_clone_destroy(struct ifnet *);
94 static int pfsync_clone_create(struct if_clone *, int, caddr_t);
95 void pfsync_setmtu(struct pfsync_softc *, int);
96 int pfsync_alloc_scrub_memory(struct pfsync_state_peer *,
97 struct pf_state_peer *);
98 int pfsyncoutput(struct ifnet *, struct mbuf *, struct sockaddr *,
100 int pfsyncioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
101 void pfsyncstart(struct ifnet *, struct ifaltq_subque *);
103 struct mbuf *pfsync_get_mbuf(struct pfsync_softc *, u_int8_t, void **);
104 int pfsync_request_update(struct pfsync_state_upd *, struct in_addr *);
105 int pfsync_sendout(struct pfsync_softc *);
106 int pfsync_sendout_mbuf(struct pfsync_softc *, struct mbuf *);
107 void pfsync_timeout(void *);
108 void pfsync_send_bus(struct pfsync_softc *, u_int8_t);
109 void pfsync_bulk_update(void *);
110 void pfsync_bulkfail(void *);
112 static struct in_multi *pfsync_in_addmulti(struct ifnet *);
113 static void pfsync_in_delmulti(struct in_multi *);
115 static MALLOC_DEFINE(M_PFSYNC, PFSYNCNAME, "Packet Filter State Sync. Interface");
116 static LIST_HEAD(pfsync_list, pfsync_softc) pfsync_list;
120 struct if_clone pfsync_cloner =
121 IF_CLONE_INITIALIZER("pfsync", pfsync_clone_create, pfsync_clone_destroy, 1 ,1);
124 pfsyncattach(int npfsync)
126 if_clone_attach(&pfsync_cloner);
129 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param __unused)
131 struct pfsync_softc *sc;
134 lwkt_gettoken(&pf_token);
136 sc = kmalloc(sizeof(*sc), M_PFSYNC, M_WAITOK | M_ZERO);
139 sc->sc_mbuf_net = NULL;
140 sc->sc_mbuf_tdb = NULL;
141 sc->sc_statep.s = NULL;
142 sc->sc_statep_net.s = NULL;
143 sc->sc_statep_tdb.t = NULL;
144 sc->sc_maxupdates = 128;
145 sc->sc_sync_peer.s_addr =htonl(INADDR_PFSYNC_GROUP);
146 sc->sc_sendaddr.s_addr = htonl(INADDR_PFSYNC_GROUP);
147 sc->sc_ureq_received = 0;
148 sc->sc_ureq_sent = 0;
149 sc->sc_bulk_send_next = NULL;
150 sc->sc_bulk_terminator = NULL;
151 sc->sc_bulk_send_cpu = 0;
152 sc->sc_bulk_terminator_cpu = 0;
153 sc->sc_imo.imo_max_memberships = IP_MAX_MEMBERSHIPS;
154 lwkt_reltoken(&pf_token);
156 ksnprintf(ifp->if_xname, sizeof ifp->if_xname, "pfsync%d", unit);
157 if_initname(ifp, ifc->ifc_name, unit);
158 ifp->if_ioctl = pfsyncioctl;
159 ifp->if_output = pfsyncoutput;
160 ifp->if_start = pfsyncstart;
161 ifp->if_type = IFT_PFSYNC;
162 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
163 ifp->if_hdrlen = PFSYNC_HDRLEN;
164 ifp->if_baudrate = IF_Mbps(100);
166 pfsync_setmtu(sc, MCLBYTES);
167 callout_init(&sc->sc_tmo);
168 /* callout_init(&sc->sc_tdb_tmo); XXX we don't support tdb (yet) */
169 callout_init(&sc->sc_bulk_tmo);
170 callout_init(&sc->sc_bulkfail_tmo);
171 if_attach(ifp, NULL);
173 LIST_INSERT_HEAD(&pfsync_list, sc, sc_next);
177 if_addgroup(ifp, "carp");
181 bpfattach(&sc->sc_if, DLT_PFSYNC, PFSYNC_HDRLEN);
183 lwkt_gettoken(&pf_token);
185 lwkt_reltoken(&pf_token);
190 pfsync_clone_destroy(struct ifnet *ifp)
192 lwkt_gettoken(&pf_token);
193 lwkt_reltoken(&pf_token);
195 struct pfsync_softc *sc = ifp->if_softc;
196 callout_stop(&sc->sc_tmo);
197 /* callout_stop(&sc->sc_tdb_tmo); XXX we don't support tdb (yet) */
198 callout_stop(&sc->sc_bulk_tmo);
199 callout_stop(&sc->sc_bulkfail_tmo);
202 carp_group_demote_adj(&sc->sc_if, -1);
208 lwkt_gettoken(&pf_token);
209 LIST_REMOVE(sc, sc_next);
211 lwkt_reltoken(&pf_token);
218 * Start output on the pfsync interface.
221 pfsyncstart(struct ifnet *ifp, struct ifaltq_subque *ifsq)
223 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
228 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s,
229 struct pf_state_peer *d)
231 if (s->scrub.scrub_flag && d->scrub == NULL) {
232 d->scrub = kmalloc(sizeof(struct pf_state_scrub), M_PFSYNC, M_NOWAIT|M_ZERO);
234 if (d->scrub == NULL)
242 pfsync_state_export(struct pfsync_state *sp, struct pf_state *st)
244 bzero(sp, sizeof(struct pfsync_state));
246 /* copy from state key */
247 sp->key[PF_SK_WIRE].addr[0] = st->key[PF_SK_WIRE]->addr[0];
248 sp->key[PF_SK_WIRE].addr[1] = st->key[PF_SK_WIRE]->addr[1];
249 sp->key[PF_SK_WIRE].port[0] = st->key[PF_SK_WIRE]->port[0];
250 sp->key[PF_SK_WIRE].port[1] = st->key[PF_SK_WIRE]->port[1];
251 sp->key[PF_SK_STACK].addr[0] = st->key[PF_SK_STACK]->addr[0];
252 sp->key[PF_SK_STACK].addr[1] = st->key[PF_SK_STACK]->addr[1];
253 sp->key[PF_SK_STACK].port[0] = st->key[PF_SK_STACK]->port[0];
254 sp->key[PF_SK_STACK].port[1] = st->key[PF_SK_STACK]->port[1];
255 sp->proto = st->key[PF_SK_WIRE]->proto;
256 sp->af = st->key[PF_SK_WIRE]->af;
258 /* copy from state */
259 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname));
260 bcopy(&st->rt_addr, &sp->rt_addr, sizeof(sp->rt_addr));
261 sp->creation = htonl(time_second - st->creation);
262 sp->expire = pf_state_expires(st);
263 if (sp->expire <= time_second)
264 sp->expire = htonl(0);
266 sp->expire = htonl(sp->expire - time_second);
268 sp->direction = st->direction;
270 sp->cpuid = st->cpuid;
271 sp->pickup_mode = st->pickup_mode;
272 sp->timeout = st->timeout;
273 sp->state_flags = st->state_flags;
275 sp->sync_flags |= PFSYNC_FLAG_SRCNODE;
276 if (st->nat_src_node)
277 sp->sync_flags |= PFSYNC_FLAG_NATSRCNODE;
279 bcopy(&st->id, &sp->id, sizeof(sp->id));
280 sp->creatorid = st->creatorid;
281 pf_state_peer_hton(&st->src, &sp->src);
282 pf_state_peer_hton(&st->dst, &sp->dst);
284 if (st->rule.ptr == NULL)
285 sp->rule = htonl(-1);
287 sp->rule = htonl(st->rule.ptr->nr);
288 if (st->anchor.ptr == NULL)
289 sp->anchor = htonl(-1);
291 sp->anchor = htonl(st->anchor.ptr->nr);
292 if (st->nat_rule.ptr == NULL)
293 sp->nat_rule = htonl(-1);
295 sp->nat_rule = htonl(st->nat_rule.ptr->nr);
297 pf_state_counter_hton(st->packets[0], sp->packets[0]);
298 pf_state_counter_hton(st->packets[1], sp->packets[1]);
299 pf_state_counter_hton(st->bytes[0], sp->bytes[0]);
300 pf_state_counter_hton(st->bytes[1], sp->bytes[1]);
305 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags)
307 struct pf_state *st = NULL;
308 struct pf_state_key *skw = NULL, *sks = NULL;
309 struct pf_rule *r = NULL;
314 if (sp->creatorid == 0 && pf_status.debug >= PF_DEBUG_MISC) {
315 kprintf("pfsync_insert_net_state: invalid creator id:"
316 " %08x\n", ntohl(sp->creatorid));
320 if ((kif = pfi_kif_get(sp->ifname)) == NULL) {
321 if (pf_status.debug >= PF_DEBUG_MISC)
322 kprintf("pfsync_insert_net_state: "
323 "unknown interface: %s\n", sp->ifname);
324 if (flags & PFSYNC_SI_IOCTL)
326 return (0); /* skip this state */
330 * If the ruleset checksums match or the state is coming from the ioctl,
331 * it's safe to associate the state with the rule of that number.
333 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) &&
334 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) <
335 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount)
336 r = pf_main_ruleset.rules[
337 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)];
339 r = &pf_default_rule;
341 if ((r->max_states && r->states_cur >= r->max_states))
344 if (flags & PFSYNC_SI_IOCTL)
345 pool_flags = M_WAITOK | M_NULLOK | M_ZERO;
347 pool_flags = M_WAITOK | M_ZERO;
349 if ((st = kmalloc(sizeof(struct pf_state), M_PFSYNC, pool_flags)) == NULL)
351 lockinit(&st->lk, "pfstlk", 0, 0);
353 if ((skw = pf_alloc_state_key(pool_flags)) == NULL)
356 if (PF_ANEQ(&sp->key[PF_SK_WIRE].addr[0],
357 &sp->key[PF_SK_STACK].addr[0], sp->af) ||
358 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[1],
359 &sp->key[PF_SK_STACK].addr[1], sp->af) ||
360 sp->key[PF_SK_WIRE].port[0] != sp->key[PF_SK_STACK].port[0] ||
361 sp->key[PF_SK_WIRE].port[1] != sp->key[PF_SK_STACK].port[1]) {
362 if ((sks = pf_alloc_state_key(pool_flags)) == NULL)
367 /* allocate memory for scrub info */
368 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) ||
369 pfsync_alloc_scrub_memory(&sp->dst, &st->dst))
372 /* copy to state key(s) */
373 skw->addr[0] = sp->key[PF_SK_WIRE].addr[0];
374 skw->addr[1] = sp->key[PF_SK_WIRE].addr[1];
375 skw->port[0] = sp->key[PF_SK_WIRE].port[0];
376 skw->port[1] = sp->key[PF_SK_WIRE].port[1];
377 skw->proto = sp->proto;
380 sks->addr[0] = sp->key[PF_SK_STACK].addr[0];
381 sks->addr[1] = sp->key[PF_SK_STACK].addr[1];
382 sks->port[0] = sp->key[PF_SK_STACK].port[0];
383 sks->port[1] = sp->key[PF_SK_STACK].port[1];
384 sks->proto = sp->proto;
389 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr));
390 st->creation = time_second - ntohl(sp->creation);
391 st->expire = time_second;
393 /* XXX No adaptive scaling. */
394 st->expire -= r->timeout[sp->timeout] - ntohl(sp->expire);
397 st->expire = ntohl(sp->expire) + time_second;
398 st->direction = sp->direction;
400 st->timeout = sp->timeout;
401 st->state_flags = sp->state_flags;
402 if (!(flags & PFSYNC_SI_IOCTL))
403 st->sync_flags = PFSTATE_FROMSYNC;
405 bcopy(sp->id, &st->id, sizeof(st->id));
406 st->creatorid = sp->creatorid;
407 pf_state_peer_ntoh(&sp->src, &st->src);
408 pf_state_peer_ntoh(&sp->dst, &st->dst);
411 st->nat_rule.ptr = NULL;
412 st->anchor.ptr = NULL;
418 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */
422 if ((error = pf_state_insert(kif, skw, sks, st)) != 0) {
423 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */
435 kfree(skw, M_PFSYNC);
437 kfree(sks, M_PFSYNC);
439 cleanup_state: /* pf_state_insert frees the state keys */
442 kfree(st->dst.scrub, M_PFSYNC);
444 kfree(st->src.scrub, M_PFSYNC);
451 pfsync_input(struct mbuf *m, ...)
453 struct ip *ip = mtod(m, struct ip *);
454 struct pfsync_header *ph;
455 struct pfsync_softc *sc = pfsyncif;
457 struct pf_state_key *sk;
458 struct pf_state_item *si;
459 struct pf_state_cmp id_key;
460 struct pfsync_state *sp;
461 struct pfsync_state_upd *up;
462 struct pfsync_state_del *dp;
463 struct pfsync_state_clr *cp;
464 struct pfsync_state_upd_req *rup;
465 struct pfsync_state_bus *bus;
467 struct pfsync_tdb *pt;
471 int iplen, action, error, i, count, offp, sfail, stale = 0;
474 /* This function is not yet called from anywhere */
475 /* Still we assume for safety that pf_token must be held */
476 ASSERT_LWKT_TOKEN_HELD(&pf_token);
478 pfsyncstats.pfsyncs_ipackets++;
480 /* verify that we have a sync interface configured */
481 if (!sc || !sc->sc_sync_ifp || !pf_status.running)
484 /* verify that the packet came in on the right interface */
485 if (sc->sc_sync_ifp != m->m_pkthdr.rcvif) {
486 pfsyncstats.pfsyncs_badif++;
490 /* verify that the IP TTL is 255. */
491 if (ip->ip_ttl != PFSYNC_DFLTTL) {
492 pfsyncstats.pfsyncs_badttl++;
496 iplen = ip->ip_hl << 2;
498 if (m->m_pkthdr.len < iplen + sizeof(*ph)) {
499 pfsyncstats.pfsyncs_hdrops++;
503 if (iplen + sizeof(*ph) > m->m_len) {
504 if ((m = m_pullup(m, iplen + sizeof(*ph))) == NULL) {
505 pfsyncstats.pfsyncs_hdrops++;
508 ip = mtod(m, struct ip *);
510 ph = (struct pfsync_header *)((char *)ip + iplen);
512 /* verify the version */
513 if (ph->version != PFSYNC_VERSION) {
514 pfsyncstats.pfsyncs_badver++;
521 /* make sure it's a valid action code */
522 if (action >= PFSYNC_ACT_MAX) {
523 pfsyncstats.pfsyncs_badact++;
527 /* Cheaper to grab this now than having to mess with mbufs later */
530 if (!bcmp(&ph->pf_chksum, &pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH))
531 flags |= PFSYNC_SI_CKSUM;
534 case PFSYNC_ACT_CLR: {
535 struct pf_state *nexts;
536 struct pf_state_key *nextsk;
538 globaldata_t save_gd = mycpu;
542 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
543 sizeof(*cp), &offp)) == NULL) {
544 pfsyncstats.pfsyncs_badlen++;
547 cp = (struct pfsync_state_clr *)(mp->m_data + offp);
548 creatorid = cp->creatorid;
551 if (cp->ifname[0] == '\0') {
552 lwkt_gettoken(&pf_token);
553 for (nn = 0; nn < ncpus; ++nn) {
554 lwkt_setcpu_self(globaldata_find(nn));
555 for (st = RB_MIN(pf_state_tree_id,
558 nexts = RB_NEXT(pf_state_tree_id,
560 if (st->creatorid == creatorid) {
567 lwkt_setcpu_self(save_gd);
568 lwkt_reltoken(&pf_token);
570 if ((kif = pfi_kif_get(cp->ifname)) == NULL) {
575 lwkt_gettoken(&pf_token);
576 for (nn = 0; nn < ncpus; ++nn) {
577 lwkt_setcpu_self(globaldata_find(nn));
578 for (sk = RB_MIN(pf_state_tree,
582 nextsk = RB_NEXT(pf_state_tree,
583 &pf_statetbl[n], sk);
584 TAILQ_FOREACH(si, &sk->states, entry) {
585 if (si->s->creatorid ==
589 pf_unlink_state(si->s);
594 lwkt_setcpu_self(save_gd);
595 lwkt_reltoken(&pf_token);
602 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
603 count * sizeof(*sp), &offp)) == NULL) {
604 pfsyncstats.pfsyncs_badlen++;
609 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
610 i < count; i++, sp++) {
611 /* check for invalid values */
612 if (sp->timeout >= PFTM_MAX ||
613 sp->src.state > PF_TCPS_PROXY_DST ||
614 sp->dst.state > PF_TCPS_PROXY_DST ||
615 sp->direction > PF_OUT ||
616 (sp->af != AF_INET && sp->af != AF_INET6)) {
617 if (pf_status.debug >= PF_DEBUG_MISC)
618 kprintf("pfsync_insert: PFSYNC_ACT_INS: "
620 pfsyncstats.pfsyncs_badval++;
624 if ((error = pfsync_state_import(sp, flags))) {
625 if (error == ENOMEM) {
634 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
635 count * sizeof(*sp), &offp)) == NULL) {
636 pfsyncstats.pfsyncs_badlen++;
641 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
642 i < count; i++, sp++) {
643 int flags = PFSYNC_FLAG_STALE;
645 /* check for invalid values */
646 if (sp->timeout >= PFTM_MAX ||
647 sp->src.state > PF_TCPS_PROXY_DST ||
648 sp->dst.state > PF_TCPS_PROXY_DST) {
649 if (pf_status.debug >= PF_DEBUG_MISC)
650 kprintf("pfsync_insert: PFSYNC_ACT_UPD: "
652 pfsyncstats.pfsyncs_badval++;
656 bcopy(sp->id, &id_key.id, sizeof(id_key.id));
657 id_key.creatorid = sp->creatorid;
659 st = pf_find_state_byid(&id_key);
661 /* insert the update */
662 if (pfsync_state_import(sp, flags))
663 pfsyncstats.pfsyncs_badstate++;
666 sk = st->key[PF_SK_WIRE]; /* XXX right one? */
668 if (sk->proto == IPPROTO_TCP) {
670 * The state should never go backwards except
671 * for syn-proxy states. Neither should the
672 * sequence window slide backwards.
674 if (st->src.state > sp->src.state &&
675 (st->src.state < PF_TCPS_PROXY_SRC ||
676 sp->src.state >= PF_TCPS_PROXY_SRC))
678 else if (SEQ_GT(st->src.seqlo,
679 ntohl(sp->src.seqlo)))
681 else if (st->dst.state > sp->dst.state) {
682 /* There might still be useful
683 * information about the src state here,
684 * so import that part of the update,
685 * then "fail" so we send the updated
686 * state back to the peer who is missing
687 * our what we know. */
688 pf_state_peer_ntoh(&sp->src, &st->src);
689 /* XXX do anything with timeouts? */
692 } else if (st->dst.state >= TCPS_SYN_SENT &&
693 SEQ_GT(st->dst.seqlo, ntohl(sp->dst.seqlo)))
697 * Non-TCP protocol state machine always go
700 if (st->src.state > sp->src.state)
702 else if (st->dst.state > sp->dst.state)
706 if (pf_status.debug >= PF_DEBUG_MISC)
707 kprintf("pfsync: %s stale update "
710 (sfail < 7 ? "ignoring"
712 (uintmax_t)be64toh(st->id),
713 ntohl(st->creatorid));
714 pfsyncstats.pfsyncs_stale++;
716 if (!(sp->sync_flags & PFSTATE_STALE)) {
717 /* we have a better state, send it */
718 if (sc->sc_mbuf != NULL && !stale)
723 PFSYNC_ACT_UPD, st, flags);
727 pfsync_alloc_scrub_memory(&sp->dst, &st->dst);
728 pf_state_peer_ntoh(&sp->src, &st->src);
729 pf_state_peer_ntoh(&sp->dst, &st->dst);
730 st->expire = ntohl(sp->expire) + time_second;
731 st->timeout = sp->timeout;
733 if (stale && sc->sc_mbuf != NULL)
738 * It's not strictly necessary for us to support the "uncompressed"
739 * delete action, but it's relatively simple and maintains consistency.
742 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
743 count * sizeof(*sp), &offp)) == NULL) {
744 pfsyncstats.pfsyncs_badlen++;
749 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp);
750 i < count; i++, sp++) {
751 bcopy(sp->id, &id_key.id, sizeof(id_key.id));
752 id_key.creatorid = sp->creatorid;
754 st = pf_find_state_byid(&id_key);
756 pfsyncstats.pfsyncs_badstate++;
759 st->sync_flags |= PFSTATE_FROMSYNC;
764 case PFSYNC_ACT_UPD_C: {
765 int update_requested = 0;
767 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
768 count * sizeof(*up), &offp)) == NULL) {
769 pfsyncstats.pfsyncs_badlen++;
774 for (i = 0, up = (struct pfsync_state_upd *)(mp->m_data + offp);
775 i < count; i++, up++) {
776 /* check for invalid values */
777 if (up->timeout >= PFTM_MAX ||
778 up->src.state > PF_TCPS_PROXY_DST ||
779 up->dst.state > PF_TCPS_PROXY_DST) {
780 if (pf_status.debug >= PF_DEBUG_MISC)
781 kprintf("pfsync_insert: "
784 pfsyncstats.pfsyncs_badval++;
788 bcopy(up->id, &id_key.id, sizeof(id_key.id));
789 id_key.creatorid = up->creatorid;
791 st = pf_find_state_byid(&id_key);
793 /* We don't have this state. Ask for it. */
794 error = pfsync_request_update(up, &src);
795 if (error == ENOMEM) {
799 update_requested = 1;
800 pfsyncstats.pfsyncs_badstate++;
803 sk = st->key[PF_SK_WIRE]; /* XXX right one? */
805 if (sk->proto == IPPROTO_TCP) {
807 * The state should never go backwards except
808 * for syn-proxy states. Neither should the
809 * sequence window slide backwards.
811 if (st->src.state > up->src.state &&
812 (st->src.state < PF_TCPS_PROXY_SRC ||
813 up->src.state >= PF_TCPS_PROXY_SRC))
815 else if (st->dst.state > up->dst.state)
817 else if (SEQ_GT(st->src.seqlo,
818 ntohl(up->src.seqlo)))
820 else if (st->dst.state >= TCPS_SYN_SENT &&
821 SEQ_GT(st->dst.seqlo, ntohl(up->dst.seqlo)))
825 * Non-TCP protocol state machine always go
828 if (st->src.state > up->src.state)
830 else if (st->dst.state > up->dst.state)
834 if (pf_status.debug >= PF_DEBUG_MISC)
835 kprintf("pfsync: ignoring stale update "
836 "(%d) id: %016" PRIx64 " "
837 "creatorid: %08x\n", sfail,
839 ntohl(st->creatorid));
840 pfsyncstats.pfsyncs_stale++;
842 /* we have a better state, send it out */
843 if ((!stale || update_requested) &&
844 sc->sc_mbuf != NULL) {
846 update_requested = 0;
850 pfsync_pack_state(PFSYNC_ACT_UPD, st,
854 pfsync_alloc_scrub_memory(&up->dst, &st->dst);
855 pf_state_peer_ntoh(&up->src, &st->src);
856 pf_state_peer_ntoh(&up->dst, &st->dst);
857 st->expire = ntohl(up->expire) + time_second;
858 st->timeout = up->timeout;
860 if ((update_requested || stale) && sc->sc_mbuf)
865 case PFSYNC_ACT_DEL_C:
866 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
867 count * sizeof(*dp), &offp)) == NULL) {
868 pfsyncstats.pfsyncs_badlen++;
873 for (i = 0, dp = (struct pfsync_state_del *)(mp->m_data + offp);
874 i < count; i++, dp++) {
875 bcopy(dp->id, &id_key.id, sizeof(id_key.id));
876 id_key.creatorid = dp->creatorid;
878 st = pf_find_state_byid(&id_key);
880 pfsyncstats.pfsyncs_badstate++;
883 st->sync_flags |= PFSTATE_FROMSYNC;
888 case PFSYNC_ACT_INS_F:
889 case PFSYNC_ACT_DEL_F:
890 /* not implemented */
892 case PFSYNC_ACT_UREQ:
893 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
894 count * sizeof(*rup), &offp)) == NULL) {
895 pfsyncstats.pfsyncs_badlen++;
900 if (sc->sc_mbuf != NULL)
903 rup = (struct pfsync_state_upd_req *)(mp->m_data + offp);
904 i < count; i++, rup++) {
905 bcopy(rup->id, &id_key.id, sizeof(id_key.id));
906 id_key.creatorid = rup->creatorid;
908 if (id_key.id == 0 && id_key.creatorid == 0) {
909 sc->sc_ureq_received = mycpu->gd_time_seconds;
910 if (sc->sc_bulk_send_next == NULL) {
911 if (++sc->sc_bulk_send_cpu >= ncpus)
912 sc->sc_bulk_send_cpu = 0;
913 sc->sc_bulk_send_next =
914 TAILQ_FIRST(&state_list[sc->sc_bulk_send_cpu]);
916 sc->sc_bulk_terminator =
917 sc->sc_bulk_send_next;
918 sc->sc_bulk_terminator_cpu =
919 sc->sc_bulk_send_cpu;
920 if (pf_status.debug >= PF_DEBUG_MISC)
921 kprintf("pfsync: received "
922 "bulk update request\n");
923 pfsync_send_bus(sc, PFSYNC_BUS_START);
924 lwkt_reltoken(&pf_token);
925 callout_init(&sc->sc_bulk_tmo);
926 lwkt_gettoken(&pf_token);
928 st = pf_find_state_byid(&id_key);
930 pfsyncstats.pfsyncs_badstate++;
934 pfsync_pack_state(PFSYNC_ACT_UPD,
938 if (sc->sc_mbuf != NULL)
943 /* If we're not waiting for a bulk update, who cares. */
944 if (sc->sc_ureq_sent == 0)
947 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
948 sizeof(*bus), &offp)) == NULL) {
949 pfsyncstats.pfsyncs_badlen++;
952 bus = (struct pfsync_state_bus *)(mp->m_data + offp);
953 switch (bus->status) {
954 case PFSYNC_BUS_START:
955 lwkt_reltoken(&pf_token);
956 callout_reset(&sc->sc_bulkfail_tmo,
957 pf_pool_limits[PF_LIMIT_STATES].limit /
958 (PFSYNC_BULKPACKETS * sc->sc_maxcount),
959 pfsync_bulkfail, LIST_FIRST(&pfsync_list));
960 lwkt_gettoken(&pf_token);
961 if (pf_status.debug >= PF_DEBUG_MISC)
962 kprintf("pfsync: received bulk "
966 if (mycpu->gd_time_seconds - ntohl(bus->endtime) >=
968 /* that's it, we're happy */
969 sc->sc_ureq_sent = 0;
970 sc->sc_bulk_tries = 0;
971 lwkt_reltoken(&pf_token);
972 callout_stop(&sc->sc_bulkfail_tmo);
973 lwkt_gettoken(&pf_token);
975 if (!pfsync_sync_ok) {
976 lwkt_reltoken(&pf_token);
977 carp_group_demote_adj(&sc->sc_if, -1);
978 lwkt_gettoken(&pf_token);
982 if (pf_status.debug >= PF_DEBUG_MISC)
983 kprintf("pfsync: received valid "
984 "bulk update end\n");
986 if (pf_status.debug >= PF_DEBUG_MISC)
987 kprintf("pfsync: received invalid "
988 "bulk update end: bad timestamp\n");
994 case PFSYNC_ACT_TDB_UPD:
995 if ((mp = m_pulldown(m, iplen + sizeof(*ph),
996 count * sizeof(*pt), &offp)) == NULL) {
997 pfsyncstats.pfsyncs_badlen++;
1001 for (i = 0, pt = (struct pfsync_tdb *)(mp->m_data + offp);
1002 i < count; i++, pt++)
1003 pfsync_update_net_tdb(pt);
1015 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
1024 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
1026 struct pfsync_softc *sc = ifp->if_softc;
1027 struct ifreq *ifr = (struct ifreq *)data;
1028 struct ip_moptions *imo = &sc->sc_imo;
1029 struct pfsyncreq pfsyncr;
1033 lwkt_gettoken(&pf_token);
1038 case SIOCSIFDSTADDR:
1040 if (ifp->if_flags & IFF_UP)
1041 ifp->if_flags |= IFF_RUNNING;
1043 ifp->if_flags &= ~IFF_RUNNING;
1046 if (ifr->ifr_mtu < PFSYNC_MINMTU) {
1047 lwkt_reltoken(&pf_token);
1050 if (ifr->ifr_mtu > MCLBYTES)
1051 ifr->ifr_mtu = MCLBYTES;
1053 if (ifr->ifr_mtu < ifp->if_mtu)
1055 pfsync_setmtu(sc, ifr->ifr_mtu);
1059 bzero(&pfsyncr, sizeof(pfsyncr));
1060 if (sc->sc_sync_ifp)
1061 strlcpy(pfsyncr.pfsyncr_syncdev,
1062 sc->sc_sync_ifp->if_xname, IFNAMSIZ);
1063 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer;
1064 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates;
1065 lwkt_reltoken(&pf_token);
1066 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr))))
1068 lwkt_gettoken(&pf_token);
1071 if ((error = priv_check_cred(cr, PRIV_ROOT, NULL_CRED_OKAY)) != 0) {
1072 lwkt_reltoken(&pf_token);
1075 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) {
1076 lwkt_reltoken(&pf_token);
1080 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0)
1081 sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP;
1083 sc->sc_sync_peer.s_addr =
1084 pfsyncr.pfsyncr_syncpeer.s_addr;
1086 if (pfsyncr.pfsyncr_maxupdates > 255) {
1087 lwkt_reltoken(&pf_token);
1090 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates;
1092 if (pfsyncr.pfsyncr_syncdev[0] == 0) {
1093 sc->sc_sync_ifp = NULL;
1094 if (sc->sc_mbuf_net != NULL) {
1095 /* Don't keep stale pfsync packets around. */
1097 m_freem(sc->sc_mbuf_net);
1098 sc->sc_mbuf_net = NULL;
1099 sc->sc_statep_net.s = NULL;
1102 if (imo->imo_num_memberships > 0) {
1103 pfsync_in_delmulti(imo->imo_membership[--imo->imo_num_memberships]);
1104 imo->imo_multicast_ifp = NULL;
1109 if ((sifp = ifunit(pfsyncr.pfsyncr_syncdev)) == NULL) {
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);
1660 /* This must be called in splnet() */
1662 pfsync_sendout(struct pfsync_softc *sc)
1665 struct ifnet *ifp = &sc->sc_if;
1669 ASSERT_LWKT_TOKEN_HELD(&pf_token);
1671 lwkt_reltoken(&pf_token);
1672 callout_stop(&sc->sc_tmo);
1673 lwkt_gettoken(&pf_token);
1675 if (sc->sc_mbuf == NULL)
1679 sc->sc_statep.s = NULL;
1685 bpf_mtap(ifp->if_bpf, m, BPF_DIRECTION_OUT);
1690 if (sc->sc_mbuf_net) {
1692 m = sc->sc_mbuf_net;
1693 sc->sc_mbuf_net = NULL;
1694 sc->sc_statep_net.s = NULL;
1697 return pfsync_sendout_mbuf(sc, m);
1701 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m)
1706 if (sc->sc_sync_ifp ||
1707 sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) {
1708 M_PREPEND(m, sizeof(struct ip), M_WAITOK);
1710 pfsyncstats.pfsyncs_onomem++;
1713 ip = mtod(m, struct ip *);
1714 ip->ip_v = IPVERSION;
1715 ip->ip_hl = sizeof(*ip) >> 2;
1716 ip->ip_tos = IPTOS_LOWDELAY;
1717 ip->ip_len = htons(m->m_pkthdr.len);
1718 ip->ip_id = htons(ip_randomid());
1719 ip->ip_off = htons(IP_DF);
1720 ip->ip_ttl = PFSYNC_DFLTTL;
1721 ip->ip_p = IPPROTO_PFSYNC;
1724 bzero(&sa, sizeof(sa));
1725 ip->ip_src.s_addr = INADDR_ANY;
1727 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP)
1728 m->m_flags |= M_MCAST;
1729 ip->ip_dst = sc->sc_sendaddr;
1730 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr;
1732 pfsyncstats.pfsyncs_opackets++;
1734 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL))
1735 pfsyncstats.pfsyncs_oerrors++;
1743 pfsync_modevent(module_t mod, int type, void *data)
1747 struct pfsync_softc *pfs_if, *tmp;
1749 lwkt_gettoken(&pf_token);
1753 LIST_INIT(&pfsync_list);
1754 lwkt_reltoken(&pf_token);
1755 if_clone_attach(&pfsync_cloner);
1756 lwkt_gettoken(&pf_token);
1757 /* Override the function pointer for pf_ioctl.c */
1761 lwkt_reltoken(&pf_token);
1762 if_clone_detach(&pfsync_cloner);
1763 lwkt_gettoken(&pf_token);
1764 LIST_FOREACH_MUTABLE(pfs_if, &pfsync_list, sc_next, tmp) {
1765 pfsync_clone_destroy(&pfs_if->sc_if);
1774 lwkt_reltoken(&pf_token);
1778 static moduledata_t pfsync_mod = {
1784 #define PFSYNC_MODVER 44
1786 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
1787 MODULE_VERSION(pfsync, PFSYNC_MODVER);
1790 pfsync_in_addmulti_dispatch(netmsg_t nmsg)
1792 struct lwkt_msg *lmsg = &nmsg->lmsg;
1793 struct ifnet *ifp = lmsg->u.ms_resultp;
1794 struct in_addr addr;
1796 addr.s_addr = INADDR_PFSYNC_GROUP;
1797 lmsg->u.ms_resultp = in_addmulti(&addr, ifp);
1799 lwkt_replymsg(lmsg, 0);
1802 static struct in_multi *
1803 pfsync_in_addmulti(struct ifnet *ifp)
1805 struct netmsg_base nmsg;
1806 struct lwkt_msg *lmsg = &nmsg.lmsg;
1808 netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
1809 pfsync_in_addmulti_dispatch);
1810 lmsg->u.ms_resultp = ifp;
1812 lwkt_domsg(netisr_cpuport(0), lmsg, 0);
1813 return lmsg->u.ms_resultp;
1817 pfsync_in_delmulti_dispatch(netmsg_t nmsg)
1819 struct lwkt_msg *lmsg = &nmsg->lmsg;
1821 in_delmulti(lmsg->u.ms_resultp);
1822 lwkt_replymsg(lmsg, 0);
1826 pfsync_in_delmulti(struct in_multi *inm)
1828 struct netmsg_base nmsg;
1829 struct lwkt_msg *lmsg = &nmsg.lmsg;
1831 netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
1832 pfsync_in_delmulti_dispatch);
1833 lmsg->u.ms_resultp = inm;
1835 lwkt_domsg(netisr_cpuport(0), lmsg, 0);