1 /* $OpenBSD: if_pfsync.c,v 1.98 2008/06/29 08:42:15 mcbride Exp $ */ 2 3 /* 4 * Copyright (c) 2002 Michael Shalayeff 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, 20 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 21 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 22 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 24 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 25 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 26 * THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 #include "opt_inet.h" 30 #include "opt_inet6.h" 31 #include "opt_carp.h" 32 33 #include <sys/param.h> 34 #include <sys/endian.h> 35 #include <sys/proc.h> 36 #include <sys/priv.h> 37 #include <sys/systm.h> 38 #include <sys/time.h> 39 #include <sys/mbuf.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> 47 48 #include <machine/inttypes.h> 49 50 #include <net/if.h> 51 #include <net/if_types.h> 52 #include <net/ifq_var.h> 53 #include <net/route.h> 54 #include <net/bpf.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> 62 63 #ifdef INET 64 #include <netinet/in_systm.h> 65 #include <netinet/in_var.h> 66 #include <netinet/ip.h> 67 #include <netinet/ip_var.h> 68 #endif 69 70 #ifdef INET6 71 #include <netinet6/nd6.h> 72 #endif /* INET6 */ 73 74 #include <net/pf/pfvar.h> 75 #include <net/pf/if_pfsync.h> 76 77 #define PFSYNCNAME "pfsync" 78 79 #define PFSYNC_MINMTU \ 80 (sizeof(struct pfsync_header) + sizeof(struct pf_state)) 81 82 #ifdef PFSYNCDEBUG 83 #define DPRINTF(x) do { if (pfsyncdebug) kprintf x ; } while (0) 84 int pfsyncdebug; 85 #else 86 #define DPRINTF(x) 87 #endif 88 89 struct pfsync_softc *pfsyncif = NULL; 90 struct pfsyncstats pfsyncstats; 91 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 *, 99 struct rtentry *); 100 int pfsyncioctl(struct ifnet *, u_long, caddr_t, struct ucred *); 101 void pfsyncstart(struct ifnet *, struct ifaltq_subque *); 102 103 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 *); 111 112 static struct in_multi *pfsync_in_addmulti(struct ifnet *); 113 static void pfsync_in_delmulti(struct in_multi *); 114 115 static MALLOC_DEFINE(M_PFSYNC, PFSYNCNAME, "Packet Filter State Sync. Interface"); 116 static LIST_HEAD(pfsync_list, pfsync_softc) pfsync_list; 117 118 int pfsync_sync_ok; 119 120 struct if_clone pfsync_cloner = 121 IF_CLONE_INITIALIZER("pfsync", pfsync_clone_create, pfsync_clone_destroy, 1 ,1); 122 123 void 124 pfsyncattach(int npfsync) 125 { 126 if_clone_attach(&pfsync_cloner); 127 } 128 static int 129 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param __unused) 130 { 131 struct pfsync_softc *sc; 132 struct ifnet *ifp; 133 134 lwkt_gettoken(&pf_token); 135 136 sc = kmalloc(sizeof(*sc), M_PFSYNC, M_WAITOK | M_ZERO); 137 pfsync_sync_ok = 1; 138 sc->sc_mbuf = NULL; 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_imo.imo_max_memberships = IP_MAX_MEMBERSHIPS; 152 lwkt_reltoken(&pf_token); 153 ifp = &sc->sc_if; 154 ksnprintf(ifp->if_xname, sizeof ifp->if_xname, "pfsync%d", unit); 155 if_initname(ifp, ifc->ifc_name, unit); 156 ifp->if_ioctl = pfsyncioctl; 157 ifp->if_output = pfsyncoutput; 158 ifp->if_start = pfsyncstart; 159 ifp->if_type = IFT_PFSYNC; 160 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen); 161 ifp->if_hdrlen = PFSYNC_HDRLEN; 162 ifp->if_baudrate = IF_Mbps(100); 163 ifp->if_softc = sc; 164 pfsync_setmtu(sc, MCLBYTES); 165 callout_init(&sc->sc_tmo); 166 /* callout_init(&sc->sc_tdb_tmo); XXX we don't support tdb (yet) */ 167 callout_init(&sc->sc_bulk_tmo); 168 callout_init(&sc->sc_bulkfail_tmo); 169 if_attach(ifp, NULL); 170 171 LIST_INSERT_HEAD(&pfsync_list, sc, sc_next); 172 173 174 #if NCARP > 0 175 if_addgroup(ifp, "carp"); 176 #endif 177 178 #if NBPFILTER > 0 179 bpfattach(&sc->sc_if, DLT_PFSYNC, PFSYNC_HDRLEN); 180 #endif 181 lwkt_gettoken(&pf_token); 182 183 lwkt_reltoken(&pf_token); 184 return (0); 185 } 186 187 static int 188 pfsync_clone_destroy(struct ifnet *ifp) 189 { 190 lwkt_gettoken(&pf_token); 191 lwkt_reltoken(&pf_token); 192 193 struct pfsync_softc *sc = ifp->if_softc; 194 callout_stop(&sc->sc_tmo); 195 /* callout_stop(&sc->sc_tdb_tmo); XXX we don't support tdb (yet) */ 196 callout_stop(&sc->sc_bulk_tmo); 197 callout_stop(&sc->sc_bulkfail_tmo); 198 #if NCARP > 0 199 if (!pfsync_sync_ok) 200 carp_group_demote_adj(&sc->sc_if, -1); 201 #endif 202 #if NBPFILTER > 0 203 bpfdetach(ifp); 204 #endif 205 if_detach(ifp); 206 lwkt_gettoken(&pf_token); 207 LIST_REMOVE(sc, sc_next); 208 kfree(sc, M_PFSYNC); 209 lwkt_reltoken(&pf_token); 210 211 212 return 0; 213 } 214 215 /* 216 * Start output on the pfsync interface. 217 */ 218 void 219 pfsyncstart(struct ifnet *ifp, struct ifaltq_subque *ifsq) 220 { 221 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq); 222 ifsq_purge(ifsq); 223 } 224 225 int 226 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 227 struct pf_state_peer *d) 228 { 229 if (s->scrub.scrub_flag && d->scrub == NULL) { 230 d->scrub = kmalloc(sizeof(struct pf_state_scrub), M_PFSYNC, M_NOWAIT|M_ZERO); 231 232 if (d->scrub == NULL) 233 return (ENOMEM); 234 } 235 236 return (0); 237 } 238 239 void 240 pfsync_state_export(struct pfsync_state *sp, struct pf_state *st) 241 { 242 bzero(sp, sizeof(struct pfsync_state)); 243 244 /* copy from state key */ 245 sp->key[PF_SK_WIRE].addr[0] = st->key[PF_SK_WIRE]->addr[0]; 246 sp->key[PF_SK_WIRE].addr[1] = st->key[PF_SK_WIRE]->addr[1]; 247 sp->key[PF_SK_WIRE].port[0] = st->key[PF_SK_WIRE]->port[0]; 248 sp->key[PF_SK_WIRE].port[1] = st->key[PF_SK_WIRE]->port[1]; 249 sp->key[PF_SK_STACK].addr[0] = st->key[PF_SK_STACK]->addr[0]; 250 sp->key[PF_SK_STACK].addr[1] = st->key[PF_SK_STACK]->addr[1]; 251 sp->key[PF_SK_STACK].port[0] = st->key[PF_SK_STACK]->port[0]; 252 sp->key[PF_SK_STACK].port[1] = st->key[PF_SK_STACK]->port[1]; 253 sp->proto = st->key[PF_SK_WIRE]->proto; 254 sp->af = st->key[PF_SK_WIRE]->af; 255 256 /* copy from state */ 257 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); 258 bcopy(&st->rt_addr, &sp->rt_addr, sizeof(sp->rt_addr)); 259 sp->creation = htonl(time_second - st->creation); 260 sp->expire = pf_state_expires(st); 261 if (sp->expire <= time_second) 262 sp->expire = htonl(0); 263 else 264 sp->expire = htonl(sp->expire - time_second); 265 266 sp->direction = st->direction; 267 sp->log = st->log; 268 sp->timeout = st->timeout; 269 sp->state_flags = st->state_flags; 270 if (st->src_node) 271 sp->sync_flags |= PFSYNC_FLAG_SRCNODE; 272 if (st->nat_src_node) 273 sp->sync_flags |= PFSYNC_FLAG_NATSRCNODE; 274 275 bcopy(&st->id, &sp->id, sizeof(sp->id)); 276 sp->creatorid = st->creatorid; 277 pf_state_peer_hton(&st->src, &sp->src); 278 pf_state_peer_hton(&st->dst, &sp->dst); 279 280 if (st->rule.ptr == NULL) 281 sp->rule = htonl(-1); 282 else 283 sp->rule = htonl(st->rule.ptr->nr); 284 if (st->anchor.ptr == NULL) 285 sp->anchor = htonl(-1); 286 else 287 sp->anchor = htonl(st->anchor.ptr->nr); 288 if (st->nat_rule.ptr == NULL) 289 sp->nat_rule = htonl(-1); 290 else 291 sp->nat_rule = htonl(st->nat_rule.ptr->nr); 292 293 pf_state_counter_hton(st->packets[0], sp->packets[0]); 294 pf_state_counter_hton(st->packets[1], sp->packets[1]); 295 pf_state_counter_hton(st->bytes[0], sp->bytes[0]); 296 pf_state_counter_hton(st->bytes[1], sp->bytes[1]); 297 298 } 299 300 int 301 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags) 302 { 303 struct pf_state *st = NULL; 304 struct pf_state_key *skw = NULL, *sks = NULL; 305 struct pf_rule *r = NULL; 306 struct pfi_kif *kif; 307 int pool_flags; 308 int error; 309 310 if (sp->creatorid == 0 && pf_status.debug >= PF_DEBUG_MISC) { 311 kprintf("pfsync_insert_net_state: invalid creator id:" 312 " %08x\n", ntohl(sp->creatorid)); 313 return (EINVAL); 314 } 315 316 if ((kif = pfi_kif_get(sp->ifname)) == NULL) { 317 if (pf_status.debug >= PF_DEBUG_MISC) 318 kprintf("pfsync_insert_net_state: " 319 "unknown interface: %s\n", sp->ifname); 320 if (flags & PFSYNC_SI_IOCTL) 321 return (EINVAL); 322 return (0); /* skip this state */ 323 } 324 325 /* 326 * If the ruleset checksums match or the state is coming from the ioctl, 327 * it's safe to associate the state with the rule of that number. 328 */ 329 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && 330 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) < 331 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 332 r = pf_main_ruleset.rules[ 333 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 334 else 335 r = &pf_default_rule; 336 337 if ((r->max_states && r->states_cur >= r->max_states)) 338 goto cleanup; 339 340 if (flags & PFSYNC_SI_IOCTL) 341 pool_flags = M_WAITOK | M_NULLOK | M_ZERO; 342 else 343 pool_flags = M_WAITOK | M_ZERO; 344 345 if ((st = kmalloc(sizeof(struct pf_state), M_PFSYNC, pool_flags)) == NULL) 346 goto cleanup; 347 348 if ((skw = pf_alloc_state_key(pool_flags)) == NULL) 349 goto cleanup; 350 351 if (PF_ANEQ(&sp->key[PF_SK_WIRE].addr[0], 352 &sp->key[PF_SK_STACK].addr[0], sp->af) || 353 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[1], 354 &sp->key[PF_SK_STACK].addr[1], sp->af) || 355 sp->key[PF_SK_WIRE].port[0] != sp->key[PF_SK_STACK].port[0] || 356 sp->key[PF_SK_WIRE].port[1] != sp->key[PF_SK_STACK].port[1]) { 357 if ((sks = pf_alloc_state_key(pool_flags)) == NULL) 358 goto cleanup; 359 } else 360 sks = skw; 361 362 /* allocate memory for scrub info */ 363 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 364 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) 365 goto cleanup; 366 367 /* copy to state key(s) */ 368 skw->addr[0] = sp->key[PF_SK_WIRE].addr[0]; 369 skw->addr[1] = sp->key[PF_SK_WIRE].addr[1]; 370 skw->port[0] = sp->key[PF_SK_WIRE].port[0]; 371 skw->port[1] = sp->key[PF_SK_WIRE].port[1]; 372 skw->proto = sp->proto; 373 skw->af = sp->af; 374 if (sks != skw) { 375 sks->addr[0] = sp->key[PF_SK_STACK].addr[0]; 376 sks->addr[1] = sp->key[PF_SK_STACK].addr[1]; 377 sks->port[0] = sp->key[PF_SK_STACK].port[0]; 378 sks->port[1] = sp->key[PF_SK_STACK].port[1]; 379 sks->proto = sp->proto; 380 sks->af = sp->af; 381 } 382 383 /* copy to state */ 384 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr)); 385 st->creation = time_second - ntohl(sp->creation); 386 st->expire = time_second; 387 if (sp->expire) { 388 /* XXX No adaptive scaling. */ 389 st->expire -= r->timeout[sp->timeout] - ntohl(sp->expire); 390 } 391 392 st->expire = ntohl(sp->expire) + time_second; 393 st->direction = sp->direction; 394 st->log = sp->log; 395 st->timeout = sp->timeout; 396 st->state_flags = sp->state_flags; 397 if (!(flags & PFSYNC_SI_IOCTL)) 398 st->sync_flags = PFSTATE_FROMSYNC; 399 400 bcopy(sp->id, &st->id, sizeof(st->id)); 401 st->creatorid = sp->creatorid; 402 pf_state_peer_ntoh(&sp->src, &st->src); 403 pf_state_peer_ntoh(&sp->dst, &st->dst); 404 405 st->rule.ptr = r; 406 st->nat_rule.ptr = NULL; 407 st->anchor.ptr = NULL; 408 st->rt_kif = NULL; 409 410 st->pfsync_time = 0; 411 412 413 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 414 r->states_cur++; 415 r->states_tot++; 416 417 if ((error = pf_state_insert(kif, skw, sks, st)) != 0) { 418 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */ 419 r->states_cur--; 420 goto cleanup_state; 421 } 422 423 return (0); 424 425 cleanup: 426 error = ENOMEM; 427 if (skw == sks) 428 sks = NULL; 429 if (skw != NULL) 430 kfree(skw, M_PFSYNC); 431 if (sks != NULL) 432 kfree(sks, M_PFSYNC); 433 434 cleanup_state: /* pf_state_insert frees the state keys */ 435 if (st) { 436 if (st->dst.scrub) 437 kfree(st->dst.scrub, M_PFSYNC); 438 if (st->src.scrub) 439 kfree(st->src.scrub, M_PFSYNC); 440 kfree(st, M_PFSYNC); 441 } 442 return (error); 443 } 444 445 void 446 pfsync_input(struct mbuf *m, ...) 447 { 448 struct ip *ip = mtod(m, struct ip *); 449 struct pfsync_header *ph; 450 struct pfsync_softc *sc = pfsyncif; 451 struct pf_state *st; 452 struct pf_state_key *sk; 453 struct pf_state_item *si; 454 struct pf_state_cmp id_key; 455 struct pfsync_state *sp; 456 struct pfsync_state_upd *up; 457 struct pfsync_state_del *dp; 458 struct pfsync_state_clr *cp; 459 struct pfsync_state_upd_req *rup; 460 struct pfsync_state_bus *bus; 461 #ifdef IPSEC 462 struct pfsync_tdb *pt; 463 #endif 464 struct in_addr src; 465 struct mbuf *mp; 466 int iplen, action, error, i, count, offp, sfail, stale = 0; 467 u_int8_t flags = 0; 468 469 /* This function is not yet called from anywhere */ 470 /* Still we assume for safety that pf_token must be held */ 471 ASSERT_LWKT_TOKEN_HELD(&pf_token); 472 473 pfsyncstats.pfsyncs_ipackets++; 474 475 /* verify that we have a sync interface configured */ 476 if (!sc || !sc->sc_sync_ifp || !pf_status.running) 477 goto done; 478 479 /* verify that the packet came in on the right interface */ 480 if (sc->sc_sync_ifp != m->m_pkthdr.rcvif) { 481 pfsyncstats.pfsyncs_badif++; 482 goto done; 483 } 484 485 /* verify that the IP TTL is 255. */ 486 if (ip->ip_ttl != PFSYNC_DFLTTL) { 487 pfsyncstats.pfsyncs_badttl++; 488 goto done; 489 } 490 491 iplen = ip->ip_hl << 2; 492 493 if (m->m_pkthdr.len < iplen + sizeof(*ph)) { 494 pfsyncstats.pfsyncs_hdrops++; 495 goto done; 496 } 497 498 if (iplen + sizeof(*ph) > m->m_len) { 499 if ((m = m_pullup(m, iplen + sizeof(*ph))) == NULL) { 500 pfsyncstats.pfsyncs_hdrops++; 501 goto done; 502 } 503 ip = mtod(m, struct ip *); 504 } 505 ph = (struct pfsync_header *)((char *)ip + iplen); 506 507 /* verify the version */ 508 if (ph->version != PFSYNC_VERSION) { 509 pfsyncstats.pfsyncs_badver++; 510 goto done; 511 } 512 513 action = ph->action; 514 count = ph->count; 515 516 /* make sure it's a valid action code */ 517 if (action >= PFSYNC_ACT_MAX) { 518 pfsyncstats.pfsyncs_badact++; 519 goto done; 520 } 521 522 /* Cheaper to grab this now than having to mess with mbufs later */ 523 src = ip->ip_src; 524 525 if (!bcmp(&ph->pf_chksum, &pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 526 flags |= PFSYNC_SI_CKSUM; 527 528 switch (action) { 529 case PFSYNC_ACT_CLR: { 530 struct pf_state *nexts; 531 struct pf_state_key *nextsk; 532 struct pfi_kif *kif; 533 u_int32_t creatorid; 534 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 535 sizeof(*cp), &offp)) == NULL) { 536 pfsyncstats.pfsyncs_badlen++; 537 return; 538 } 539 cp = (struct pfsync_state_clr *)(mp->m_data + offp); 540 creatorid = cp->creatorid; 541 542 crit_enter(); 543 if (cp->ifname[0] == '\0') { 544 for (st = RB_MIN(pf_state_tree_id, &tree_id); 545 st; st = nexts) { 546 nexts = RB_NEXT(pf_state_tree_id, &tree_id, st); 547 if (st->creatorid == creatorid) { 548 st->sync_flags |= PFSTATE_FROMSYNC; 549 pf_unlink_state(st); 550 } 551 } 552 } else { 553 if ((kif = pfi_kif_get(cp->ifname)) == NULL) { 554 crit_exit(); 555 return; 556 } 557 /* XXX correct? */ 558 for (sk = RB_MIN(pf_state_tree, 559 &pf_statetbl); sk; sk = nextsk) { 560 nextsk = RB_NEXT(pf_state_tree, 561 &pf_statetbl, sk); 562 TAILQ_FOREACH(si, &sk->states, entry) { 563 if (si->s->creatorid == creatorid) { 564 si->s->sync_flags |= 565 PFSTATE_FROMSYNC; 566 pf_unlink_state(si->s); 567 } 568 } 569 } 570 } 571 crit_exit(); 572 573 break; 574 } 575 case PFSYNC_ACT_INS: 576 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 577 count * sizeof(*sp), &offp)) == NULL) { 578 pfsyncstats.pfsyncs_badlen++; 579 return; 580 } 581 582 crit_enter(); 583 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 584 i < count; i++, sp++) { 585 /* check for invalid values */ 586 if (sp->timeout >= PFTM_MAX || 587 sp->src.state > PF_TCPS_PROXY_DST || 588 sp->dst.state > PF_TCPS_PROXY_DST || 589 sp->direction > PF_OUT || 590 (sp->af != AF_INET && sp->af != AF_INET6)) { 591 if (pf_status.debug >= PF_DEBUG_MISC) 592 kprintf("pfsync_insert: PFSYNC_ACT_INS: " 593 "invalid value\n"); 594 pfsyncstats.pfsyncs_badval++; 595 continue; 596 } 597 598 if ((error = pfsync_state_import(sp, flags))) { 599 if (error == ENOMEM) { 600 crit_exit(); 601 goto done; 602 } 603 } 604 } 605 crit_exit(); 606 break; 607 case PFSYNC_ACT_UPD: 608 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 609 count * sizeof(*sp), &offp)) == NULL) { 610 pfsyncstats.pfsyncs_badlen++; 611 return; 612 } 613 614 crit_enter(); 615 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 616 i < count; i++, sp++) { 617 int flags = PFSYNC_FLAG_STALE; 618 619 /* check for invalid values */ 620 if (sp->timeout >= PFTM_MAX || 621 sp->src.state > PF_TCPS_PROXY_DST || 622 sp->dst.state > PF_TCPS_PROXY_DST) { 623 if (pf_status.debug >= PF_DEBUG_MISC) 624 kprintf("pfsync_insert: PFSYNC_ACT_UPD: " 625 "invalid value\n"); 626 pfsyncstats.pfsyncs_badval++; 627 continue; 628 } 629 630 bcopy(sp->id, &id_key.id, sizeof(id_key.id)); 631 id_key.creatorid = sp->creatorid; 632 633 st = pf_find_state_byid(&id_key); 634 if (st == NULL) { 635 /* insert the update */ 636 if (pfsync_state_import(sp, flags)) 637 pfsyncstats.pfsyncs_badstate++; 638 continue; 639 } 640 sk = st->key[PF_SK_WIRE]; /* XXX right one? */ 641 sfail = 0; 642 if (sk->proto == IPPROTO_TCP) { 643 /* 644 * The state should never go backwards except 645 * for syn-proxy states. Neither should the 646 * sequence window slide backwards. 647 */ 648 if (st->src.state > sp->src.state && 649 (st->src.state < PF_TCPS_PROXY_SRC || 650 sp->src.state >= PF_TCPS_PROXY_SRC)) 651 sfail = 1; 652 else if (SEQ_GT(st->src.seqlo, 653 ntohl(sp->src.seqlo))) 654 sfail = 3; 655 else if (st->dst.state > sp->dst.state) { 656 /* There might still be useful 657 * information about the src state here, 658 * so import that part of the update, 659 * then "fail" so we send the updated 660 * state back to the peer who is missing 661 * our what we know. */ 662 pf_state_peer_ntoh(&sp->src, &st->src); 663 /* XXX do anything with timeouts? */ 664 sfail = 7; 665 flags = 0; 666 } else if (st->dst.state >= TCPS_SYN_SENT && 667 SEQ_GT(st->dst.seqlo, ntohl(sp->dst.seqlo))) 668 sfail = 4; 669 } else { 670 /* 671 * Non-TCP protocol state machine always go 672 * forwards 673 */ 674 if (st->src.state > sp->src.state) 675 sfail = 5; 676 else if (st->dst.state > sp->dst.state) 677 sfail = 6; 678 } 679 if (sfail) { 680 if (pf_status.debug >= PF_DEBUG_MISC) 681 kprintf("pfsync: %s stale update " 682 "(%d) id: %016jx " 683 "creatorid: %08x\n", 684 (sfail < 7 ? "ignoring" 685 : "partial"), sfail, 686 (uintmax_t)be64toh(st->id), 687 ntohl(st->creatorid)); 688 pfsyncstats.pfsyncs_stale++; 689 690 if (!(sp->sync_flags & PFSTATE_STALE)) { 691 /* we have a better state, send it */ 692 if (sc->sc_mbuf != NULL && !stale) 693 pfsync_sendout(sc); 694 stale++; 695 if (!st->sync_flags) 696 pfsync_pack_state( 697 PFSYNC_ACT_UPD, st, flags); 698 } 699 continue; 700 } 701 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 702 pf_state_peer_ntoh(&sp->src, &st->src); 703 pf_state_peer_ntoh(&sp->dst, &st->dst); 704 st->expire = ntohl(sp->expire) + time_second; 705 st->timeout = sp->timeout; 706 } 707 if (stale && sc->sc_mbuf != NULL) 708 pfsync_sendout(sc); 709 crit_exit(); 710 break; 711 /* 712 * It's not strictly necessary for us to support the "uncompressed" 713 * delete action, but it's relatively simple and maintains consistency. 714 */ 715 case PFSYNC_ACT_DEL: 716 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 717 count * sizeof(*sp), &offp)) == NULL) { 718 pfsyncstats.pfsyncs_badlen++; 719 return; 720 } 721 722 crit_enter(); 723 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 724 i < count; i++, sp++) { 725 bcopy(sp->id, &id_key.id, sizeof(id_key.id)); 726 id_key.creatorid = sp->creatorid; 727 728 st = pf_find_state_byid(&id_key); 729 if (st == NULL) { 730 pfsyncstats.pfsyncs_badstate++; 731 continue; 732 } 733 st->sync_flags |= PFSTATE_FROMSYNC; 734 pf_unlink_state(st); 735 } 736 crit_exit(); 737 break; 738 case PFSYNC_ACT_UPD_C: { 739 int update_requested = 0; 740 741 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 742 count * sizeof(*up), &offp)) == NULL) { 743 pfsyncstats.pfsyncs_badlen++; 744 return; 745 } 746 747 crit_enter(); 748 for (i = 0, up = (struct pfsync_state_upd *)(mp->m_data + offp); 749 i < count; i++, up++) { 750 /* check for invalid values */ 751 if (up->timeout >= PFTM_MAX || 752 up->src.state > PF_TCPS_PROXY_DST || 753 up->dst.state > PF_TCPS_PROXY_DST) { 754 if (pf_status.debug >= PF_DEBUG_MISC) 755 kprintf("pfsync_insert: " 756 "PFSYNC_ACT_UPD_C: " 757 "invalid value\n"); 758 pfsyncstats.pfsyncs_badval++; 759 continue; 760 } 761 762 bcopy(up->id, &id_key.id, sizeof(id_key.id)); 763 id_key.creatorid = up->creatorid; 764 765 st = pf_find_state_byid(&id_key); 766 if (st == NULL) { 767 /* We don't have this state. Ask for it. */ 768 error = pfsync_request_update(up, &src); 769 if (error == ENOMEM) { 770 crit_exit(); 771 goto done; 772 } 773 update_requested = 1; 774 pfsyncstats.pfsyncs_badstate++; 775 continue; 776 } 777 sk = st->key[PF_SK_WIRE]; /* XXX right one? */ 778 sfail = 0; 779 if (sk->proto == IPPROTO_TCP) { 780 /* 781 * The state should never go backwards except 782 * for syn-proxy states. Neither should the 783 * sequence window slide backwards. 784 */ 785 if (st->src.state > up->src.state && 786 (st->src.state < PF_TCPS_PROXY_SRC || 787 up->src.state >= PF_TCPS_PROXY_SRC)) 788 sfail = 1; 789 else if (st->dst.state > up->dst.state) 790 sfail = 2; 791 else if (SEQ_GT(st->src.seqlo, 792 ntohl(up->src.seqlo))) 793 sfail = 3; 794 else if (st->dst.state >= TCPS_SYN_SENT && 795 SEQ_GT(st->dst.seqlo, ntohl(up->dst.seqlo))) 796 sfail = 4; 797 } else { 798 /* 799 * Non-TCP protocol state machine always go 800 * forwards 801 */ 802 if (st->src.state > up->src.state) 803 sfail = 5; 804 else if (st->dst.state > up->dst.state) 805 sfail = 6; 806 } 807 if (sfail) { 808 if (pf_status.debug >= PF_DEBUG_MISC) 809 kprintf("pfsync: ignoring stale update " 810 "(%d) id: %016" PRIx64 " " 811 "creatorid: %08x\n", sfail, 812 be64toh(st->id), 813 ntohl(st->creatorid)); 814 pfsyncstats.pfsyncs_stale++; 815 816 /* we have a better state, send it out */ 817 if ((!stale || update_requested) && 818 sc->sc_mbuf != NULL) { 819 pfsync_sendout(sc); 820 update_requested = 0; 821 } 822 stale++; 823 if (!st->sync_flags) 824 pfsync_pack_state(PFSYNC_ACT_UPD, st, 825 PFSYNC_FLAG_STALE); 826 continue; 827 } 828 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 829 pf_state_peer_ntoh(&up->src, &st->src); 830 pf_state_peer_ntoh(&up->dst, &st->dst); 831 st->expire = ntohl(up->expire) + time_second; 832 st->timeout = up->timeout; 833 } 834 if ((update_requested || stale) && sc->sc_mbuf) 835 pfsync_sendout(sc); 836 crit_exit(); 837 break; 838 } 839 case PFSYNC_ACT_DEL_C: 840 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 841 count * sizeof(*dp), &offp)) == NULL) { 842 pfsyncstats.pfsyncs_badlen++; 843 return; 844 } 845 846 crit_enter(); 847 for (i = 0, dp = (struct pfsync_state_del *)(mp->m_data + offp); 848 i < count; i++, dp++) { 849 bcopy(dp->id, &id_key.id, sizeof(id_key.id)); 850 id_key.creatorid = dp->creatorid; 851 852 st = pf_find_state_byid(&id_key); 853 if (st == NULL) { 854 pfsyncstats.pfsyncs_badstate++; 855 continue; 856 } 857 st->sync_flags |= PFSTATE_FROMSYNC; 858 pf_unlink_state(st); 859 } 860 crit_exit(); 861 break; 862 case PFSYNC_ACT_INS_F: 863 case PFSYNC_ACT_DEL_F: 864 /* not implemented */ 865 break; 866 case PFSYNC_ACT_UREQ: 867 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 868 count * sizeof(*rup), &offp)) == NULL) { 869 pfsyncstats.pfsyncs_badlen++; 870 return; 871 } 872 873 crit_enter(); 874 if (sc->sc_mbuf != NULL) 875 pfsync_sendout(sc); 876 for (i = 0, 877 rup = (struct pfsync_state_upd_req *)(mp->m_data + offp); 878 i < count; i++, rup++) { 879 bcopy(rup->id, &id_key.id, sizeof(id_key.id)); 880 id_key.creatorid = rup->creatorid; 881 882 if (id_key.id == 0 && id_key.creatorid == 0) { 883 sc->sc_ureq_received = mycpu->gd_time_seconds; 884 if (sc->sc_bulk_send_next == NULL) 885 sc->sc_bulk_send_next = 886 TAILQ_FIRST(&state_list); 887 sc->sc_bulk_terminator = sc->sc_bulk_send_next; 888 if (pf_status.debug >= PF_DEBUG_MISC) 889 kprintf("pfsync: received " 890 "bulk update request\n"); 891 pfsync_send_bus(sc, PFSYNC_BUS_START); 892 lwkt_reltoken(&pf_token); 893 callout_init(&sc->sc_bulk_tmo); 894 lwkt_gettoken(&pf_token); 895 } else { 896 st = pf_find_state_byid(&id_key); 897 if (st == NULL) { 898 pfsyncstats.pfsyncs_badstate++; 899 continue; 900 } 901 if (!st->sync_flags) 902 pfsync_pack_state(PFSYNC_ACT_UPD, 903 st, 0); 904 } 905 } 906 if (sc->sc_mbuf != NULL) 907 pfsync_sendout(sc); 908 crit_exit(); 909 break; 910 case PFSYNC_ACT_BUS: 911 /* If we're not waiting for a bulk update, who cares. */ 912 if (sc->sc_ureq_sent == 0) 913 break; 914 915 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 916 sizeof(*bus), &offp)) == NULL) { 917 pfsyncstats.pfsyncs_badlen++; 918 return; 919 } 920 bus = (struct pfsync_state_bus *)(mp->m_data + offp); 921 switch (bus->status) { 922 case PFSYNC_BUS_START: 923 lwkt_reltoken(&pf_token); 924 callout_reset(&sc->sc_bulkfail_tmo, 925 pf_pool_limits[PF_LIMIT_STATES].limit / 926 (PFSYNC_BULKPACKETS * sc->sc_maxcount), 927 pfsync_bulkfail, LIST_FIRST(&pfsync_list)); 928 lwkt_gettoken(&pf_token); 929 if (pf_status.debug >= PF_DEBUG_MISC) 930 kprintf("pfsync: received bulk " 931 "update start\n"); 932 break; 933 case PFSYNC_BUS_END: 934 if (mycpu->gd_time_seconds - ntohl(bus->endtime) >= 935 sc->sc_ureq_sent) { 936 /* that's it, we're happy */ 937 sc->sc_ureq_sent = 0; 938 sc->sc_bulk_tries = 0; 939 lwkt_reltoken(&pf_token); 940 callout_stop(&sc->sc_bulkfail_tmo); 941 lwkt_gettoken(&pf_token); 942 #if NCARP > 0 943 if (!pfsync_sync_ok) { 944 lwkt_reltoken(&pf_token); 945 carp_group_demote_adj(&sc->sc_if, -1); 946 lwkt_gettoken(&pf_token); 947 } 948 #endif 949 pfsync_sync_ok = 1; 950 if (pf_status.debug >= PF_DEBUG_MISC) 951 kprintf("pfsync: received valid " 952 "bulk update end\n"); 953 } else { 954 if (pf_status.debug >= PF_DEBUG_MISC) 955 kprintf("pfsync: received invalid " 956 "bulk update end: bad timestamp\n"); 957 } 958 break; 959 } 960 break; 961 #ifdef IPSEC 962 case PFSYNC_ACT_TDB_UPD: 963 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 964 count * sizeof(*pt), &offp)) == NULL) { 965 pfsyncstats.pfsyncs_badlen++; 966 return; 967 } 968 crit_enter(); 969 for (i = 0, pt = (struct pfsync_tdb *)(mp->m_data + offp); 970 i < count; i++, pt++) 971 pfsync_update_net_tdb(pt); 972 crit_exit(); 973 break; 974 #endif 975 } 976 977 done: 978 if (m) 979 m_freem(m); 980 } 981 982 int 983 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst, 984 struct rtentry *rt) 985 { 986 m_freem(m); 987 return (0); 988 } 989 990 /* ARGSUSED */ 991 int 992 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr) 993 { 994 struct pfsync_softc *sc = ifp->if_softc; 995 struct ifreq *ifr = (struct ifreq *)data; 996 struct ip_moptions *imo = &sc->sc_imo; 997 struct pfsyncreq pfsyncr; 998 struct ifnet *sifp; 999 int error; 1000 1001 lwkt_gettoken(&pf_token); 1002 1003 switch (cmd) { 1004 case SIOCSIFADDR: 1005 case SIOCAIFADDR: 1006 case SIOCSIFDSTADDR: 1007 case SIOCSIFFLAGS: 1008 if (ifp->if_flags & IFF_UP) 1009 ifp->if_flags |= IFF_RUNNING; 1010 else 1011 ifp->if_flags &= ~IFF_RUNNING; 1012 break; 1013 case SIOCSIFMTU: 1014 if (ifr->ifr_mtu < PFSYNC_MINMTU) { 1015 lwkt_reltoken(&pf_token); 1016 return (EINVAL); 1017 } 1018 if (ifr->ifr_mtu > MCLBYTES) 1019 ifr->ifr_mtu = MCLBYTES; 1020 crit_enter(); 1021 if (ifr->ifr_mtu < ifp->if_mtu) 1022 pfsync_sendout(sc); 1023 pfsync_setmtu(sc, ifr->ifr_mtu); 1024 crit_exit(); 1025 break; 1026 case SIOCGETPFSYNC: 1027 bzero(&pfsyncr, sizeof(pfsyncr)); 1028 if (sc->sc_sync_ifp) 1029 strlcpy(pfsyncr.pfsyncr_syncdev, 1030 sc->sc_sync_ifp->if_xname, IFNAMSIZ); 1031 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 1032 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 1033 lwkt_reltoken(&pf_token); 1034 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)))) 1035 return (error); 1036 lwkt_gettoken(&pf_token); 1037 break; 1038 case SIOCSETPFSYNC: 1039 if ((error = priv_check_cred(cr, PRIV_ROOT, NULL_CRED_OKAY)) != 0) { 1040 lwkt_reltoken(&pf_token); 1041 return (error); 1042 } 1043 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) { 1044 lwkt_reltoken(&pf_token); 1045 return (error); 1046 } 1047 1048 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 1049 sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP; 1050 else 1051 sc->sc_sync_peer.s_addr = 1052 pfsyncr.pfsyncr_syncpeer.s_addr; 1053 1054 if (pfsyncr.pfsyncr_maxupdates > 255) { 1055 lwkt_reltoken(&pf_token); 1056 return (EINVAL); 1057 } 1058 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 1059 1060 if (pfsyncr.pfsyncr_syncdev[0] == 0) { 1061 sc->sc_sync_ifp = NULL; 1062 if (sc->sc_mbuf_net != NULL) { 1063 /* Don't keep stale pfsync packets around. */ 1064 crit_enter(); 1065 m_freem(sc->sc_mbuf_net); 1066 sc->sc_mbuf_net = NULL; 1067 sc->sc_statep_net.s = NULL; 1068 crit_exit(); 1069 } 1070 if (imo->imo_num_memberships > 0) { 1071 pfsync_in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 1072 imo->imo_multicast_ifp = NULL; 1073 } 1074 break; 1075 } 1076 1077 if ((sifp = ifunit(pfsyncr.pfsyncr_syncdev)) == NULL) { 1078 lwkt_reltoken(&pf_token); 1079 return (EINVAL); 1080 } 1081 1082 crit_enter(); 1083 if (sifp->if_mtu < sc->sc_if.if_mtu || 1084 (sc->sc_sync_ifp != NULL && 1085 sifp->if_mtu < sc->sc_sync_ifp->if_mtu) || 1086 sifp->if_mtu < MCLBYTES - sizeof(struct ip)) 1087 pfsync_sendout(sc); 1088 sc->sc_sync_ifp = sifp; 1089 1090 pfsync_setmtu(sc, sc->sc_if.if_mtu); 1091 1092 if (imo->imo_num_memberships > 0) { 1093 pfsync_in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 1094 imo->imo_multicast_ifp = NULL; 1095 } 1096 1097 if (sc->sc_sync_ifp && 1098 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1099 if (!(sc->sc_sync_ifp->if_flags & IFF_MULTICAST)) { 1100 sc->sc_sync_ifp = NULL; 1101 lwkt_reltoken(&pf_token); 1102 crit_exit(); 1103 return (EADDRNOTAVAIL); 1104 } 1105 1106 if ((imo->imo_membership[0] = 1107 pfsync_in_addmulti(sc->sc_sync_ifp)) == NULL) { 1108 sc->sc_sync_ifp = NULL; 1109 lwkt_reltoken(&pf_token); 1110 crit_exit(); 1111 return (ENOBUFS); 1112 } 1113 imo->imo_num_memberships++; 1114 imo->imo_multicast_ifp = sc->sc_sync_ifp; 1115 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 1116 imo->imo_multicast_loop = 0; 1117 } 1118 1119 if (sc->sc_sync_ifp || 1120 sc->sc_sendaddr.s_addr != INADDR_PFSYNC_GROUP) { 1121 /* Request a full state table update. */ 1122 sc->sc_ureq_sent = mycpu->gd_time_seconds; 1123 #if NCARP > 0 1124 if (pfsync_sync_ok) 1125 carp_group_demote_adj(&sc->sc_if, 1); 1126 #endif 1127 pfsync_sync_ok = 0; 1128 if (pf_status.debug >= PF_DEBUG_MISC) 1129 kprintf("pfsync: requesting bulk update\n"); 1130 lwkt_reltoken(&pf_token); 1131 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, 1132 pfsync_bulkfail, LIST_FIRST(&pfsync_list)); 1133 lwkt_gettoken(&pf_token); 1134 error = pfsync_request_update(NULL, NULL); 1135 if (error == ENOMEM) { 1136 lwkt_reltoken(&pf_token); 1137 crit_exit(); 1138 return (ENOMEM); 1139 } 1140 pfsync_sendout(sc); 1141 } 1142 crit_exit(); 1143 1144 break; 1145 1146 default: 1147 lwkt_reltoken(&pf_token); 1148 return (ENOTTY); 1149 } 1150 1151 lwkt_reltoken(&pf_token); 1152 return (0); 1153 } 1154 1155 void 1156 pfsync_setmtu(struct pfsync_softc *sc, int mtu_req) 1157 { 1158 int mtu; 1159 1160 if (sc->sc_sync_ifp && sc->sc_sync_ifp->if_mtu < mtu_req) 1161 mtu = sc->sc_sync_ifp->if_mtu; 1162 else 1163 mtu = mtu_req; 1164 1165 sc->sc_maxcount = (mtu - sizeof(struct pfsync_header)) / 1166 sizeof(struct pfsync_state); 1167 if (sc->sc_maxcount > 254) 1168 sc->sc_maxcount = 254; 1169 sc->sc_if.if_mtu = sizeof(struct pfsync_header) + 1170 sc->sc_maxcount * sizeof(struct pfsync_state); 1171 } 1172 1173 struct mbuf * 1174 pfsync_get_mbuf(struct pfsync_softc *sc, u_int8_t action, void **sp) 1175 { 1176 struct pfsync_header *h; 1177 struct mbuf *m; 1178 int len; 1179 1180 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1181 1182 MGETHDR(m, M_WAITOK, MT_DATA); 1183 if (m == NULL) { 1184 IFNET_STAT_INC(&sc->sc_if, oerrors, 1); 1185 return (NULL); 1186 } 1187 1188 switch (action) { 1189 case PFSYNC_ACT_CLR: 1190 len = sizeof(struct pfsync_header) + 1191 sizeof(struct pfsync_state_clr); 1192 break; 1193 case PFSYNC_ACT_UPD_C: 1194 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd)) + 1195 sizeof(struct pfsync_header); 1196 break; 1197 case PFSYNC_ACT_DEL_C: 1198 len = (sc->sc_maxcount * sizeof(struct pfsync_state_del)) + 1199 sizeof(struct pfsync_header); 1200 break; 1201 case PFSYNC_ACT_UREQ: 1202 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd_req)) + 1203 sizeof(struct pfsync_header); 1204 break; 1205 case PFSYNC_ACT_BUS: 1206 len = sizeof(struct pfsync_header) + 1207 sizeof(struct pfsync_state_bus); 1208 break; 1209 case PFSYNC_ACT_TDB_UPD: 1210 len = (sc->sc_maxcount * sizeof(struct pfsync_tdb)) + 1211 sizeof(struct pfsync_header); 1212 break; 1213 default: 1214 len = (sc->sc_maxcount * sizeof(struct pfsync_state)) + 1215 sizeof(struct pfsync_header); 1216 break; 1217 } 1218 1219 if (len > MHLEN) { 1220 MCLGET(m, M_WAITOK); 1221 if ((m->m_flags & M_EXT) == 0) { 1222 m_free(m); 1223 IFNET_STAT_INC(&sc->sc_if, oerrors, 1); 1224 return (NULL); 1225 } 1226 m->m_data += (MCLBYTES - len) &~ (sizeof(long) - 1); 1227 } else 1228 MH_ALIGN(m, len); 1229 1230 m->m_pkthdr.rcvif = NULL; 1231 m->m_pkthdr.len = m->m_len = sizeof(struct pfsync_header); 1232 h = mtod(m, struct pfsync_header *); 1233 h->version = PFSYNC_VERSION; 1234 h->af = 0; 1235 h->count = 0; 1236 h->action = action; 1237 1238 *sp = (void *)((char *)h + PFSYNC_HDRLEN); 1239 lwkt_reltoken(&pf_token); 1240 callout_reset(&sc->sc_tmo, hz, pfsync_timeout, 1241 LIST_FIRST(&pfsync_list)); 1242 lwkt_gettoken(&pf_token); 1243 return (m); 1244 } 1245 1246 int 1247 pfsync_pack_state(u_int8_t action, struct pf_state *st, int flags) 1248 { 1249 struct ifnet *ifp = NULL; 1250 struct pfsync_softc *sc = pfsyncif; 1251 struct pfsync_header *h, *h_net; 1252 struct pfsync_state *sp = NULL; 1253 struct pfsync_state_upd *up = NULL; 1254 struct pfsync_state_del *dp = NULL; 1255 int ret = 0; 1256 u_int8_t i = 255, newaction = 0; 1257 1258 if (sc == NULL) 1259 return (0); 1260 ifp = &sc->sc_if; 1261 1262 /* 1263 * If a packet falls in the forest and there's nobody around to 1264 * hear, does it make a sound? 1265 */ 1266 if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL && 1267 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1268 /* Don't leave any stale pfsync packets hanging around. */ 1269 if (sc->sc_mbuf != NULL) { 1270 m_freem(sc->sc_mbuf); 1271 sc->sc_mbuf = NULL; 1272 sc->sc_statep.s = NULL; 1273 } 1274 return (0); 1275 } 1276 1277 if (action >= PFSYNC_ACT_MAX) 1278 return (EINVAL); 1279 1280 crit_enter(); 1281 if (sc->sc_mbuf == NULL) { 1282 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1283 (void *)&sc->sc_statep.s)) == NULL) { 1284 crit_exit(); 1285 return (ENOMEM); 1286 } 1287 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1288 } else { 1289 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1290 if (h->action != action) { 1291 pfsync_sendout(sc); 1292 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1293 (void *)&sc->sc_statep.s)) == NULL) { 1294 crit_exit(); 1295 return (ENOMEM); 1296 } 1297 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1298 } else { 1299 /* 1300 * If it's an update, look in the packet to see if 1301 * we already have an update for the state. 1302 */ 1303 if (action == PFSYNC_ACT_UPD && sc->sc_maxupdates) { 1304 struct pfsync_state *usp = 1305 (void *)((char *)h + PFSYNC_HDRLEN); 1306 1307 for (i = 0; i < h->count; i++) { 1308 if (!memcmp(usp->id, &st->id, 1309 PFSYNC_ID_LEN) && 1310 usp->creatorid == st->creatorid) { 1311 sp = usp; 1312 sp->updates++; 1313 break; 1314 } 1315 usp++; 1316 } 1317 } 1318 } 1319 } 1320 1321 st->pfsync_time = mycpu->gd_time_seconds; 1322 1323 if (sp == NULL) { 1324 /* not a "duplicate" update */ 1325 i = 255; 1326 sp = sc->sc_statep.s++; 1327 sc->sc_mbuf->m_pkthdr.len = 1328 sc->sc_mbuf->m_len += sizeof(struct pfsync_state); 1329 h->count++; 1330 bzero(sp, sizeof(*sp)); 1331 1332 pfsync_state_export(sp, st); 1333 1334 if (flags & PFSYNC_FLAG_STALE) 1335 sp->sync_flags |= PFSTATE_STALE; 1336 } else { 1337 pf_state_peer_hton(&st->src, &sp->src); 1338 pf_state_peer_hton(&st->dst, &sp->dst); 1339 1340 if (st->expire <= time_second) 1341 sp->expire = htonl(0); 1342 else 1343 sp->expire = htonl(st->expire - time_second); 1344 } 1345 1346 /* do we need to build "compressed" actions for network transfer? */ 1347 if (sc->sc_sync_ifp && flags & PFSYNC_FLAG_COMPRESS) { 1348 switch (action) { 1349 case PFSYNC_ACT_UPD: 1350 newaction = PFSYNC_ACT_UPD_C; 1351 break; 1352 case PFSYNC_ACT_DEL: 1353 newaction = PFSYNC_ACT_DEL_C; 1354 break; 1355 default: 1356 /* by default we just send the uncompressed states */ 1357 break; 1358 } 1359 } 1360 1361 if (newaction) { 1362 if (sc->sc_mbuf_net == NULL) { 1363 if ((sc->sc_mbuf_net = pfsync_get_mbuf(sc, newaction, 1364 (void *)&sc->sc_statep_net.s)) == NULL) { 1365 crit_exit(); 1366 return (ENOMEM); 1367 } 1368 } 1369 h_net = mtod(sc->sc_mbuf_net, struct pfsync_header *); 1370 1371 switch (newaction) { 1372 case PFSYNC_ACT_UPD_C: 1373 if (i != 255) { 1374 up = (void *)((char *)h_net + 1375 PFSYNC_HDRLEN + (i * sizeof(*up))); 1376 up->updates++; 1377 } else { 1378 h_net->count++; 1379 sc->sc_mbuf_net->m_pkthdr.len = 1380 sc->sc_mbuf_net->m_len += sizeof(*up); 1381 up = sc->sc_statep_net.u++; 1382 1383 bzero(up, sizeof(*up)); 1384 bcopy(&st->id, up->id, sizeof(up->id)); 1385 up->creatorid = st->creatorid; 1386 } 1387 up->timeout = st->timeout; 1388 up->expire = sp->expire; 1389 up->src = sp->src; 1390 up->dst = sp->dst; 1391 break; 1392 case PFSYNC_ACT_DEL_C: 1393 sc->sc_mbuf_net->m_pkthdr.len = 1394 sc->sc_mbuf_net->m_len += sizeof(*dp); 1395 dp = sc->sc_statep_net.d++; 1396 h_net->count++; 1397 1398 bzero(dp, sizeof(*dp)); 1399 bcopy(&st->id, dp->id, sizeof(dp->id)); 1400 dp->creatorid = st->creatorid; 1401 break; 1402 } 1403 } 1404 1405 if (h->count == sc->sc_maxcount || 1406 (sc->sc_maxupdates && (sp->updates >= sc->sc_maxupdates))) 1407 ret = pfsync_sendout(sc); 1408 1409 crit_exit(); 1410 return (ret); 1411 } 1412 1413 int 1414 pfsync_request_update(struct pfsync_state_upd *up, struct in_addr *src) 1415 { 1416 struct pfsync_header *h; 1417 struct pfsync_softc *sc = pfsyncif; 1418 struct pfsync_state_upd_req *rup; 1419 int ret = 0; 1420 1421 if (sc == NULL) 1422 return (0); 1423 1424 if (sc->sc_mbuf == NULL) { 1425 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1426 (void *)&sc->sc_statep.s)) == NULL) 1427 return (ENOMEM); 1428 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1429 } else { 1430 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1431 if (h->action != PFSYNC_ACT_UREQ) { 1432 pfsync_sendout(sc); 1433 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1434 (void *)&sc->sc_statep.s)) == NULL) 1435 return (ENOMEM); 1436 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1437 } 1438 } 1439 1440 if (src != NULL) 1441 sc->sc_sendaddr = *src; 1442 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*rup); 1443 h->count++; 1444 rup = sc->sc_statep.r++; 1445 bzero(rup, sizeof(*rup)); 1446 if (up != NULL) { 1447 bcopy(up->id, rup->id, sizeof(rup->id)); 1448 rup->creatorid = up->creatorid; 1449 } 1450 1451 if (h->count == sc->sc_maxcount) 1452 ret = pfsync_sendout(sc); 1453 1454 return (ret); 1455 } 1456 1457 int 1458 pfsync_clear_states(u_int32_t creatorid, char *ifname) 1459 { 1460 struct pfsync_softc *sc = pfsyncif; 1461 struct pfsync_state_clr *cp; 1462 int ret; 1463 1464 if (sc == NULL) 1465 return (0); 1466 1467 crit_enter(); 1468 if (sc->sc_mbuf != NULL) 1469 pfsync_sendout(sc); 1470 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_CLR, 1471 (void *)&sc->sc_statep.c)) == NULL) { 1472 crit_exit(); 1473 return (ENOMEM); 1474 } 1475 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*cp); 1476 cp = sc->sc_statep.c; 1477 cp->creatorid = creatorid; 1478 if (ifname != NULL) 1479 strlcpy(cp->ifname, ifname, IFNAMSIZ); 1480 1481 ret = (pfsync_sendout(sc)); 1482 crit_exit(); 1483 return (ret); 1484 } 1485 1486 void 1487 pfsync_timeout(void *v) 1488 { 1489 struct pfsync_softc *sc = v; 1490 1491 crit_enter(); 1492 pfsync_sendout(sc); 1493 crit_exit(); 1494 } 1495 1496 void 1497 pfsync_send_bus(struct pfsync_softc *sc, u_int8_t status) 1498 { 1499 struct pfsync_state_bus *bus; 1500 1501 if (sc->sc_mbuf != NULL) 1502 pfsync_sendout(sc); 1503 1504 if (pfsync_sync_ok && 1505 (sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_BUS, 1506 (void *)&sc->sc_statep.b)) != NULL) { 1507 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*bus); 1508 bus = sc->sc_statep.b; 1509 bus->creatorid = pf_status.hostid; 1510 bus->status = status; 1511 bus->endtime = htonl(mycpu->gd_time_seconds - sc->sc_ureq_received); 1512 pfsync_sendout(sc); 1513 } 1514 } 1515 1516 void 1517 pfsync_bulk_update(void *v) 1518 { 1519 struct pfsync_softc *sc = v; 1520 int i = 0; 1521 struct pf_state *state; 1522 1523 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1524 1525 crit_enter(); 1526 if (sc->sc_mbuf != NULL) 1527 pfsync_sendout(sc); 1528 1529 /* 1530 * Grab at most PFSYNC_BULKPACKETS worth of states which have not 1531 * been sent since the latest request was made. 1532 */ 1533 state = sc->sc_bulk_send_next; 1534 if (state) 1535 do { 1536 /* send state update if syncable and not already sent */ 1537 if (!state->sync_flags 1538 && state->timeout < PFTM_MAX 1539 && state->pfsync_time <= sc->sc_ureq_received) { 1540 pfsync_pack_state(PFSYNC_ACT_UPD, state, 0); 1541 i++; 1542 } 1543 1544 /* figure next state to send */ 1545 state = TAILQ_NEXT(state, entry_list); 1546 1547 /* wrap to start of list if we hit the end */ 1548 if (!state) 1549 state = TAILQ_FIRST(&state_list); 1550 } while (i < sc->sc_maxcount * PFSYNC_BULKPACKETS && 1551 state != sc->sc_bulk_terminator); 1552 1553 if (!state || state == sc->sc_bulk_terminator) { 1554 /* we're done */ 1555 pfsync_send_bus(sc, PFSYNC_BUS_END); 1556 sc->sc_ureq_received = 0; 1557 sc->sc_bulk_send_next = NULL; 1558 sc->sc_bulk_terminator = NULL; 1559 lwkt_reltoken(&pf_token); 1560 callout_stop(&sc->sc_bulk_tmo); 1561 lwkt_gettoken(&pf_token); 1562 if (pf_status.debug >= PF_DEBUG_MISC) 1563 kprintf("pfsync: bulk update complete\n"); 1564 } else { 1565 /* look again for more in a bit */ 1566 lwkt_reltoken(&pf_token); 1567 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_timeout, 1568 LIST_FIRST(&pfsync_list)); 1569 lwkt_gettoken(&pf_token); 1570 sc->sc_bulk_send_next = state; 1571 } 1572 if (sc->sc_mbuf != NULL) 1573 pfsync_sendout(sc); 1574 crit_exit(); 1575 } 1576 1577 void 1578 pfsync_bulkfail(void *v) 1579 { 1580 struct pfsync_softc *sc = v; 1581 int error; 1582 1583 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1584 1585 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 1586 /* Try again in a bit */ 1587 lwkt_reltoken(&pf_token); 1588 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulkfail, 1589 LIST_FIRST(&pfsync_list)); 1590 lwkt_gettoken(&pf_token); 1591 crit_enter(); 1592 error = pfsync_request_update(NULL, NULL); 1593 if (error == ENOMEM) { 1594 if (pf_status.debug >= PF_DEBUG_MISC) 1595 kprintf("pfsync: cannot allocate mbufs for " 1596 "bulk update\n"); 1597 } else 1598 pfsync_sendout(sc); 1599 crit_exit(); 1600 } else { 1601 /* Pretend like the transfer was ok */ 1602 sc->sc_ureq_sent = 0; 1603 sc->sc_bulk_tries = 0; 1604 #if NCARP > 0 1605 if (!pfsync_sync_ok) 1606 carp_group_demote_adj(&sc->sc_if, -1); 1607 #endif 1608 pfsync_sync_ok = 1; 1609 if (pf_status.debug >= PF_DEBUG_MISC) 1610 kprintf("pfsync: failed to receive " 1611 "bulk update status\n"); 1612 lwkt_reltoken(&pf_token); 1613 callout_stop(&sc->sc_bulkfail_tmo); 1614 lwkt_gettoken(&pf_token); 1615 } 1616 } 1617 1618 /* This must be called in splnet() */ 1619 int 1620 pfsync_sendout(struct pfsync_softc *sc) 1621 { 1622 #if NBPFILTER > 0 1623 struct ifnet *ifp = &sc->sc_if; 1624 #endif 1625 struct mbuf *m; 1626 1627 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1628 1629 lwkt_reltoken(&pf_token); 1630 callout_stop(&sc->sc_tmo); 1631 lwkt_gettoken(&pf_token); 1632 1633 if (sc->sc_mbuf == NULL) 1634 return (0); 1635 m = sc->sc_mbuf; 1636 sc->sc_mbuf = NULL; 1637 sc->sc_statep.s = NULL; 1638 1639 #if NBPFILTER > 0 1640 if (ifp->if_bpf) { 1641 bpf_gettoken(); 1642 if (ifp->if_bpf) 1643 bpf_mtap(ifp->if_bpf, m, BPF_DIRECTION_OUT); 1644 bpf_reltoken(); 1645 } 1646 #endif 1647 1648 if (sc->sc_mbuf_net) { 1649 m_freem(m); 1650 m = sc->sc_mbuf_net; 1651 sc->sc_mbuf_net = NULL; 1652 sc->sc_statep_net.s = NULL; 1653 } 1654 1655 return pfsync_sendout_mbuf(sc, m); 1656 } 1657 1658 int 1659 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m) 1660 { 1661 struct sockaddr sa; 1662 struct ip *ip; 1663 1664 if (sc->sc_sync_ifp || 1665 sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) { 1666 M_PREPEND(m, sizeof(struct ip), M_WAITOK); 1667 if (m == NULL) { 1668 pfsyncstats.pfsyncs_onomem++; 1669 return (0); 1670 } 1671 ip = mtod(m, struct ip *); 1672 ip->ip_v = IPVERSION; 1673 ip->ip_hl = sizeof(*ip) >> 2; 1674 ip->ip_tos = IPTOS_LOWDELAY; 1675 ip->ip_len = htons(m->m_pkthdr.len); 1676 ip->ip_id = htons(ip_randomid()); 1677 ip->ip_off = htons(IP_DF); 1678 ip->ip_ttl = PFSYNC_DFLTTL; 1679 ip->ip_p = IPPROTO_PFSYNC; 1680 ip->ip_sum = 0; 1681 1682 bzero(&sa, sizeof(sa)); 1683 ip->ip_src.s_addr = INADDR_ANY; 1684 1685 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP) 1686 m->m_flags |= M_MCAST; 1687 ip->ip_dst = sc->sc_sendaddr; 1688 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr; 1689 1690 pfsyncstats.pfsyncs_opackets++; 1691 1692 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL)) 1693 pfsyncstats.pfsyncs_oerrors++; 1694 } else 1695 m_freem(m); 1696 1697 return (0); 1698 } 1699 1700 static int 1701 pfsync_modevent(module_t mod, int type, void *data) 1702 { 1703 int error = 0; 1704 1705 struct pfsync_softc *pfs_if, *tmp; 1706 1707 lwkt_gettoken(&pf_token); 1708 1709 switch (type) { 1710 case MOD_LOAD: 1711 LIST_INIT(&pfsync_list); 1712 lwkt_reltoken(&pf_token); 1713 if_clone_attach(&pfsync_cloner); 1714 lwkt_gettoken(&pf_token); 1715 /* Override the function pointer for pf_ioctl.c */ 1716 break; 1717 1718 case MOD_UNLOAD: 1719 lwkt_reltoken(&pf_token); 1720 if_clone_detach(&pfsync_cloner); 1721 lwkt_gettoken(&pf_token); 1722 LIST_FOREACH_MUTABLE(pfs_if, &pfsync_list, sc_next, tmp) { 1723 pfsync_clone_destroy(&pfs_if->sc_if); 1724 } 1725 break; 1726 1727 default: 1728 error = EINVAL; 1729 break; 1730 } 1731 1732 lwkt_reltoken(&pf_token); 1733 return error; 1734 } 1735 1736 static moduledata_t pfsync_mod = { 1737 "pfsync", 1738 pfsync_modevent, 1739 0 1740 }; 1741 1742 #define PFSYNC_MODVER 44 1743 1744 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 1745 MODULE_VERSION(pfsync, PFSYNC_MODVER); 1746 1747 static void 1748 pfsync_in_addmulti_dispatch(netmsg_t nmsg) 1749 { 1750 struct lwkt_msg *lmsg = &nmsg->lmsg; 1751 struct ifnet *ifp = lmsg->u.ms_resultp; 1752 struct in_addr addr; 1753 1754 addr.s_addr = INADDR_PFSYNC_GROUP; 1755 lmsg->u.ms_resultp = in_addmulti(&addr, ifp); 1756 1757 lwkt_replymsg(lmsg, 0); 1758 } 1759 1760 static struct in_multi * 1761 pfsync_in_addmulti(struct ifnet *ifp) 1762 { 1763 struct netmsg_base nmsg; 1764 struct lwkt_msg *lmsg = &nmsg.lmsg; 1765 1766 netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0, 1767 pfsync_in_addmulti_dispatch); 1768 lmsg->u.ms_resultp = ifp; 1769 1770 lwkt_domsg(netisr_cpuport(0), lmsg, 0); 1771 return lmsg->u.ms_resultp; 1772 } 1773 1774 static void 1775 pfsync_in_delmulti_dispatch(netmsg_t nmsg) 1776 { 1777 struct lwkt_msg *lmsg = &nmsg->lmsg; 1778 1779 in_delmulti(lmsg->u.ms_resultp); 1780 lwkt_replymsg(lmsg, 0); 1781 } 1782 1783 static void 1784 pfsync_in_delmulti(struct in_multi *inm) 1785 { 1786 struct netmsg_base nmsg; 1787 struct lwkt_msg *lmsg = &nmsg.lmsg; 1788 1789 netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0, 1790 pfsync_in_delmulti_dispatch); 1791 lmsg->u.ms_resultp = inm; 1792 1793 lwkt_domsg(netisr_cpuport(0), lmsg, 0); 1794 } 1795