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/sockio.h> 45 #include <sys/thread2.h> 46 47 #include <machine/inttypes.h> 48 49 #include <net/if.h> 50 #include <net/if_types.h> 51 #include <net/ifq_var.h> 52 #include <net/route.h> 53 #include <net/bpf.h> 54 #include <netinet/in.h> 55 #include <netinet/if_ether.h> 56 #include <netinet/ip_carp.h> 57 #include <netinet/tcp.h> 58 #include <netinet/tcp_seq.h> 59 60 #ifdef INET 61 #include <netinet/in_systm.h> 62 #include <netinet/in_var.h> 63 #include <netinet/ip.h> 64 #include <netinet/ip_var.h> 65 #endif 66 67 #ifdef INET6 68 #include <netinet6/nd6.h> 69 #endif /* INET6 */ 70 71 #include <net/pf/pfvar.h> 72 #include <net/pf/if_pfsync.h> 73 74 #define PFSYNCNAME "pfsync" 75 76 #define PFSYNC_MINMTU \ 77 (sizeof(struct pfsync_header) + sizeof(struct pf_state)) 78 79 #ifdef PFSYNCDEBUG 80 #define DPRINTF(x) do { if (pfsyncdebug) kprintf x ; } while (0) 81 int pfsyncdebug; 82 #else 83 #define DPRINTF(x) 84 #endif 85 86 struct pfsync_softc *pfsyncif = NULL; 87 struct pfsyncstats pfsyncstats; 88 89 void pfsyncattach(int); 90 static int pfsync_clone_destroy(struct ifnet *); 91 static int pfsync_clone_create(struct if_clone *, int, caddr_t); 92 void pfsync_setmtu(struct pfsync_softc *, int); 93 int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, 94 struct pf_state_peer *); 95 int pfsyncoutput(struct ifnet *, struct mbuf *, struct sockaddr *, 96 struct rtentry *); 97 int pfsyncioctl(struct ifnet *, u_long, caddr_t, struct ucred *); 98 void pfsyncstart(struct ifnet *, struct ifaltq_subque *); 99 100 struct mbuf *pfsync_get_mbuf(struct pfsync_softc *, u_int8_t, void **); 101 int pfsync_request_update(struct pfsync_state_upd *, struct in_addr *); 102 int pfsync_sendout(struct pfsync_softc *); 103 int pfsync_sendout_mbuf(struct pfsync_softc *, struct mbuf *); 104 void pfsync_timeout(void *); 105 void pfsync_send_bus(struct pfsync_softc *, u_int8_t); 106 void pfsync_bulk_update(void *); 107 void pfsync_bulkfail(void *); 108 109 static MALLOC_DEFINE(M_PFSYNC, PFSYNCNAME, "Packet Filter State Sync. Interface"); 110 static LIST_HEAD(pfsync_list, pfsync_softc) pfsync_list; 111 112 int pfsync_sync_ok; 113 114 struct if_clone pfsync_cloner = 115 IF_CLONE_INITIALIZER("pfsync", pfsync_clone_create, pfsync_clone_destroy, 1 ,1); 116 117 void 118 pfsyncattach(int npfsync) 119 { 120 if_clone_attach(&pfsync_cloner); 121 } 122 static int 123 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param __unused) 124 { 125 struct pfsync_softc *sc; 126 struct ifnet *ifp; 127 128 lwkt_gettoken(&pf_token); 129 130 sc = kmalloc(sizeof(*sc), M_PFSYNC, M_WAITOK | M_ZERO); 131 pfsync_sync_ok = 1; 132 sc->sc_mbuf = NULL; 133 sc->sc_mbuf_net = NULL; 134 sc->sc_mbuf_tdb = NULL; 135 sc->sc_statep.s = NULL; 136 sc->sc_statep_net.s = NULL; 137 sc->sc_statep_tdb.t = NULL; 138 sc->sc_maxupdates = 128; 139 sc->sc_sync_peer.s_addr =htonl(INADDR_PFSYNC_GROUP); 140 sc->sc_sendaddr.s_addr = htonl(INADDR_PFSYNC_GROUP); 141 sc->sc_ureq_received = 0; 142 sc->sc_ureq_sent = 0; 143 sc->sc_bulk_send_next = NULL; 144 sc->sc_bulk_terminator = NULL; 145 sc->sc_imo.imo_max_memberships = IP_MAX_MEMBERSHIPS; 146 lwkt_reltoken(&pf_token); 147 ifp = &sc->sc_if; 148 ksnprintf(ifp->if_xname, sizeof ifp->if_xname, "pfsync%d", unit); 149 if_initname(ifp, ifc->ifc_name, unit); 150 ifp->if_ioctl = pfsyncioctl; 151 ifp->if_output = pfsyncoutput; 152 ifp->if_start = pfsyncstart; 153 ifp->if_type = IFT_PFSYNC; 154 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen); 155 ifp->if_hdrlen = PFSYNC_HDRLEN; 156 ifp->if_baudrate = IF_Mbps(100); 157 ifp->if_softc = sc; 158 pfsync_setmtu(sc, MCLBYTES); 159 callout_init(&sc->sc_tmo); 160 /* callout_init(&sc->sc_tdb_tmo); XXX we don't support tdb (yet) */ 161 callout_init(&sc->sc_bulk_tmo); 162 callout_init(&sc->sc_bulkfail_tmo); 163 if_attach(ifp, NULL); 164 165 LIST_INSERT_HEAD(&pfsync_list, sc, sc_next); 166 167 168 #if NCARP > 0 169 if_addgroup(ifp, "carp"); 170 #endif 171 172 #if NBPFILTER > 0 173 bpfattach(&sc->sc_if, DLT_PFSYNC, PFSYNC_HDRLEN); 174 #endif 175 lwkt_gettoken(&pf_token); 176 177 lwkt_reltoken(&pf_token); 178 return (0); 179 } 180 181 static int 182 pfsync_clone_destroy(struct ifnet *ifp) 183 { 184 lwkt_gettoken(&pf_token); 185 lwkt_reltoken(&pf_token); 186 187 struct pfsync_softc *sc = ifp->if_softc; 188 callout_stop(&sc->sc_tmo); 189 /* callout_stop(&sc->sc_tdb_tmo); XXX we don't support tdb (yet) */ 190 callout_stop(&sc->sc_bulk_tmo); 191 callout_stop(&sc->sc_bulkfail_tmo); 192 #if NCARP > 0 193 if (!pfsync_sync_ok) 194 carp_group_demote_adj(&sc->sc_if, -1); 195 #endif 196 #if NBPFILTER > 0 197 bpfdetach(ifp); 198 #endif 199 if_detach(ifp); 200 lwkt_gettoken(&pf_token); 201 LIST_REMOVE(sc, sc_next); 202 kfree(sc, M_PFSYNC); 203 lwkt_reltoken(&pf_token); 204 205 206 return 0; 207 } 208 209 /* 210 * Start output on the pfsync interface. 211 */ 212 void 213 pfsyncstart(struct ifnet *ifp, struct ifaltq_subque *ifsq) 214 { 215 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq); 216 ifsq_purge(ifsq); 217 } 218 219 int 220 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 221 struct pf_state_peer *d) 222 { 223 if (s->scrub.scrub_flag && d->scrub == NULL) { 224 d->scrub = kmalloc(sizeof(struct pf_state_scrub), M_PFSYNC, M_NOWAIT|M_ZERO); 225 226 if (d->scrub == NULL) 227 return (ENOMEM); 228 } 229 230 return (0); 231 } 232 233 void 234 pfsync_state_export(struct pfsync_state *sp, struct pf_state *st) 235 { 236 bzero(sp, sizeof(struct pfsync_state)); 237 238 /* copy from state key */ 239 sp->key[PF_SK_WIRE].addr[0] = st->key[PF_SK_WIRE]->addr[0]; 240 sp->key[PF_SK_WIRE].addr[1] = st->key[PF_SK_WIRE]->addr[1]; 241 sp->key[PF_SK_WIRE].port[0] = st->key[PF_SK_WIRE]->port[0]; 242 sp->key[PF_SK_WIRE].port[1] = st->key[PF_SK_WIRE]->port[1]; 243 sp->key[PF_SK_STACK].addr[0] = st->key[PF_SK_STACK]->addr[0]; 244 sp->key[PF_SK_STACK].addr[1] = st->key[PF_SK_STACK]->addr[1]; 245 sp->key[PF_SK_STACK].port[0] = st->key[PF_SK_STACK]->port[0]; 246 sp->key[PF_SK_STACK].port[1] = st->key[PF_SK_STACK]->port[1]; 247 sp->proto = st->key[PF_SK_WIRE]->proto; 248 sp->af = st->key[PF_SK_WIRE]->af; 249 250 /* copy from state */ 251 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); 252 bcopy(&st->rt_addr, &sp->rt_addr, sizeof(sp->rt_addr)); 253 sp->creation = htonl(time_second - st->creation); 254 sp->expire = pf_state_expires(st); 255 if (sp->expire <= time_second) 256 sp->expire = htonl(0); 257 else 258 sp->expire = htonl(sp->expire - time_second); 259 260 sp->direction = st->direction; 261 sp->log = st->log; 262 sp->timeout = st->timeout; 263 sp->state_flags = st->state_flags; 264 if (st->src_node) 265 sp->sync_flags |= PFSYNC_FLAG_SRCNODE; 266 if (st->nat_src_node) 267 sp->sync_flags |= PFSYNC_FLAG_NATSRCNODE; 268 269 bcopy(&st->id, &sp->id, sizeof(sp->id)); 270 sp->creatorid = st->creatorid; 271 pf_state_peer_hton(&st->src, &sp->src); 272 pf_state_peer_hton(&st->dst, &sp->dst); 273 274 if (st->rule.ptr == NULL) 275 sp->rule = htonl(-1); 276 else 277 sp->rule = htonl(st->rule.ptr->nr); 278 if (st->anchor.ptr == NULL) 279 sp->anchor = htonl(-1); 280 else 281 sp->anchor = htonl(st->anchor.ptr->nr); 282 if (st->nat_rule.ptr == NULL) 283 sp->nat_rule = htonl(-1); 284 else 285 sp->nat_rule = htonl(st->nat_rule.ptr->nr); 286 287 pf_state_counter_hton(st->packets[0], sp->packets[0]); 288 pf_state_counter_hton(st->packets[1], sp->packets[1]); 289 pf_state_counter_hton(st->bytes[0], sp->bytes[0]); 290 pf_state_counter_hton(st->bytes[1], sp->bytes[1]); 291 292 } 293 294 int 295 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags) 296 { 297 struct pf_state *st = NULL; 298 struct pf_state_key *skw = NULL, *sks = NULL; 299 struct pf_rule *r = NULL; 300 struct pfi_kif *kif; 301 int pool_flags; 302 int error; 303 304 if (sp->creatorid == 0 && pf_status.debug >= PF_DEBUG_MISC) { 305 kprintf("pfsync_insert_net_state: invalid creator id:" 306 " %08x\n", ntohl(sp->creatorid)); 307 return (EINVAL); 308 } 309 310 if ((kif = pfi_kif_get(sp->ifname)) == NULL) { 311 if (pf_status.debug >= PF_DEBUG_MISC) 312 kprintf("pfsync_insert_net_state: " 313 "unknown interface: %s\n", sp->ifname); 314 if (flags & PFSYNC_SI_IOCTL) 315 return (EINVAL); 316 return (0); /* skip this state */ 317 } 318 319 /* 320 * If the ruleset checksums match or the state is coming from the ioctl, 321 * it's safe to associate the state with the rule of that number. 322 */ 323 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && 324 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) < 325 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 326 r = pf_main_ruleset.rules[ 327 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 328 else 329 r = &pf_default_rule; 330 331 if ((r->max_states && r->states_cur >= r->max_states)) 332 goto cleanup; 333 334 if (flags & PFSYNC_SI_IOCTL) 335 pool_flags = M_WAITOK | M_NULLOK | M_ZERO; 336 else 337 pool_flags = M_WAITOK | M_ZERO; 338 339 if ((st = kmalloc(sizeof(struct pf_state), M_PFSYNC, pool_flags)) == NULL) 340 goto cleanup; 341 342 if ((skw = pf_alloc_state_key(pool_flags)) == NULL) 343 goto cleanup; 344 345 if (PF_ANEQ(&sp->key[PF_SK_WIRE].addr[0], 346 &sp->key[PF_SK_STACK].addr[0], sp->af) || 347 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[1], 348 &sp->key[PF_SK_STACK].addr[1], sp->af) || 349 sp->key[PF_SK_WIRE].port[0] != sp->key[PF_SK_STACK].port[0] || 350 sp->key[PF_SK_WIRE].port[1] != sp->key[PF_SK_STACK].port[1]) { 351 if ((sks = pf_alloc_state_key(pool_flags)) == NULL) 352 goto cleanup; 353 } else 354 sks = skw; 355 356 /* allocate memory for scrub info */ 357 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 358 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) 359 goto cleanup; 360 361 /* copy to state key(s) */ 362 skw->addr[0] = sp->key[PF_SK_WIRE].addr[0]; 363 skw->addr[1] = sp->key[PF_SK_WIRE].addr[1]; 364 skw->port[0] = sp->key[PF_SK_WIRE].port[0]; 365 skw->port[1] = sp->key[PF_SK_WIRE].port[1]; 366 skw->proto = sp->proto; 367 skw->af = sp->af; 368 if (sks != skw) { 369 sks->addr[0] = sp->key[PF_SK_STACK].addr[0]; 370 sks->addr[1] = sp->key[PF_SK_STACK].addr[1]; 371 sks->port[0] = sp->key[PF_SK_STACK].port[0]; 372 sks->port[1] = sp->key[PF_SK_STACK].port[1]; 373 sks->proto = sp->proto; 374 sks->af = sp->af; 375 } 376 377 /* copy to state */ 378 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr)); 379 st->creation = time_second - ntohl(sp->creation); 380 st->expire = time_second; 381 if (sp->expire) { 382 /* XXX No adaptive scaling. */ 383 st->expire -= r->timeout[sp->timeout] - ntohl(sp->expire); 384 } 385 386 st->expire = ntohl(sp->expire) + time_second; 387 st->direction = sp->direction; 388 st->log = sp->log; 389 st->timeout = sp->timeout; 390 st->state_flags = sp->state_flags; 391 if (!(flags & PFSYNC_SI_IOCTL)) 392 st->sync_flags = PFSTATE_FROMSYNC; 393 394 bcopy(sp->id, &st->id, sizeof(st->id)); 395 st->creatorid = sp->creatorid; 396 pf_state_peer_ntoh(&sp->src, &st->src); 397 pf_state_peer_ntoh(&sp->dst, &st->dst); 398 399 st->rule.ptr = r; 400 st->nat_rule.ptr = NULL; 401 st->anchor.ptr = NULL; 402 st->rt_kif = NULL; 403 404 st->pfsync_time = 0; 405 406 407 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 408 r->states_cur++; 409 r->states_tot++; 410 411 if ((error = pf_state_insert(kif, skw, sks, st)) != 0) { 412 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */ 413 r->states_cur--; 414 goto cleanup_state; 415 } 416 417 return (0); 418 419 cleanup: 420 error = ENOMEM; 421 if (skw == sks) 422 sks = NULL; 423 if (skw != NULL) 424 kfree(skw, M_PFSYNC); 425 if (sks != NULL) 426 kfree(sks, M_PFSYNC); 427 428 cleanup_state: /* pf_state_insert frees the state keys */ 429 if (st) { 430 if (st->dst.scrub) 431 kfree(st->dst.scrub, M_PFSYNC); 432 if (st->src.scrub) 433 kfree(st->src.scrub, M_PFSYNC); 434 kfree(st, M_PFSYNC); 435 } 436 return (error); 437 } 438 439 void 440 pfsync_input(struct mbuf *m, ...) 441 { 442 struct ip *ip = mtod(m, struct ip *); 443 struct pfsync_header *ph; 444 struct pfsync_softc *sc = pfsyncif; 445 struct pf_state *st; 446 struct pf_state_key *sk; 447 struct pf_state_item *si; 448 struct pf_state_cmp id_key; 449 struct pfsync_state *sp; 450 struct pfsync_state_upd *up; 451 struct pfsync_state_del *dp; 452 struct pfsync_state_clr *cp; 453 struct pfsync_state_upd_req *rup; 454 struct pfsync_state_bus *bus; 455 #ifdef IPSEC 456 struct pfsync_tdb *pt; 457 #endif 458 struct in_addr src; 459 struct mbuf *mp; 460 int iplen, action, error, i, count, offp, sfail, stale = 0; 461 u_int8_t flags = 0; 462 463 /* This function is not yet called from anywhere */ 464 /* Still we assume for safety that pf_token must be held */ 465 ASSERT_LWKT_TOKEN_HELD(&pf_token); 466 467 pfsyncstats.pfsyncs_ipackets++; 468 469 /* verify that we have a sync interface configured */ 470 if (!sc || !sc->sc_sync_ifp || !pf_status.running) 471 goto done; 472 473 /* verify that the packet came in on the right interface */ 474 if (sc->sc_sync_ifp != m->m_pkthdr.rcvif) { 475 pfsyncstats.pfsyncs_badif++; 476 goto done; 477 } 478 479 /* verify that the IP TTL is 255. */ 480 if (ip->ip_ttl != PFSYNC_DFLTTL) { 481 pfsyncstats.pfsyncs_badttl++; 482 goto done; 483 } 484 485 iplen = ip->ip_hl << 2; 486 487 if (m->m_pkthdr.len < iplen + sizeof(*ph)) { 488 pfsyncstats.pfsyncs_hdrops++; 489 goto done; 490 } 491 492 if (iplen + sizeof(*ph) > m->m_len) { 493 if ((m = m_pullup(m, iplen + sizeof(*ph))) == NULL) { 494 pfsyncstats.pfsyncs_hdrops++; 495 goto done; 496 } 497 ip = mtod(m, struct ip *); 498 } 499 ph = (struct pfsync_header *)((char *)ip + iplen); 500 501 /* verify the version */ 502 if (ph->version != PFSYNC_VERSION) { 503 pfsyncstats.pfsyncs_badver++; 504 goto done; 505 } 506 507 action = ph->action; 508 count = ph->count; 509 510 /* make sure it's a valid action code */ 511 if (action >= PFSYNC_ACT_MAX) { 512 pfsyncstats.pfsyncs_badact++; 513 goto done; 514 } 515 516 /* Cheaper to grab this now than having to mess with mbufs later */ 517 src = ip->ip_src; 518 519 if (!bcmp(&ph->pf_chksum, &pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 520 flags |= PFSYNC_SI_CKSUM; 521 522 switch (action) { 523 case PFSYNC_ACT_CLR: { 524 struct pf_state *nexts; 525 struct pf_state_key *nextsk; 526 struct pfi_kif *kif; 527 u_int32_t creatorid; 528 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 529 sizeof(*cp), &offp)) == NULL) { 530 pfsyncstats.pfsyncs_badlen++; 531 return; 532 } 533 cp = (struct pfsync_state_clr *)(mp->m_data + offp); 534 creatorid = cp->creatorid; 535 536 crit_enter(); 537 if (cp->ifname[0] == '\0') { 538 for (st = RB_MIN(pf_state_tree_id, &tree_id); 539 st; st = nexts) { 540 nexts = RB_NEXT(pf_state_tree_id, &tree_id, st); 541 if (st->creatorid == creatorid) { 542 st->sync_flags |= PFSTATE_FROMSYNC; 543 pf_unlink_state(st); 544 } 545 } 546 } else { 547 if ((kif = pfi_kif_get(cp->ifname)) == NULL) { 548 crit_exit(); 549 return; 550 } 551 /* XXX correct? */ 552 for (sk = RB_MIN(pf_state_tree, 553 &pf_statetbl); sk; sk = nextsk) { 554 nextsk = RB_NEXT(pf_state_tree, 555 &pf_statetbl, sk); 556 TAILQ_FOREACH(si, &sk->states, entry) { 557 if (si->s->creatorid == creatorid) { 558 si->s->sync_flags |= 559 PFSTATE_FROMSYNC; 560 pf_unlink_state(si->s); 561 } 562 } 563 } 564 } 565 crit_exit(); 566 567 break; 568 } 569 case PFSYNC_ACT_INS: 570 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 571 count * sizeof(*sp), &offp)) == NULL) { 572 pfsyncstats.pfsyncs_badlen++; 573 return; 574 } 575 576 crit_enter(); 577 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 578 i < count; i++, sp++) { 579 /* check for invalid values */ 580 if (sp->timeout >= PFTM_MAX || 581 sp->src.state > PF_TCPS_PROXY_DST || 582 sp->dst.state > PF_TCPS_PROXY_DST || 583 sp->direction > PF_OUT || 584 (sp->af != AF_INET && sp->af != AF_INET6)) { 585 if (pf_status.debug >= PF_DEBUG_MISC) 586 kprintf("pfsync_insert: PFSYNC_ACT_INS: " 587 "invalid value\n"); 588 pfsyncstats.pfsyncs_badval++; 589 continue; 590 } 591 592 if ((error = pfsync_state_import(sp, flags))) { 593 if (error == ENOMEM) { 594 crit_exit(); 595 goto done; 596 } 597 continue; 598 } 599 } 600 crit_exit(); 601 break; 602 case PFSYNC_ACT_UPD: 603 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 604 count * sizeof(*sp), &offp)) == NULL) { 605 pfsyncstats.pfsyncs_badlen++; 606 return; 607 } 608 609 crit_enter(); 610 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 611 i < count; i++, sp++) { 612 int flags = PFSYNC_FLAG_STALE; 613 614 /* check for invalid values */ 615 if (sp->timeout >= PFTM_MAX || 616 sp->src.state > PF_TCPS_PROXY_DST || 617 sp->dst.state > PF_TCPS_PROXY_DST) { 618 if (pf_status.debug >= PF_DEBUG_MISC) 619 kprintf("pfsync_insert: PFSYNC_ACT_UPD: " 620 "invalid value\n"); 621 pfsyncstats.pfsyncs_badval++; 622 continue; 623 } 624 625 bcopy(sp->id, &id_key.id, sizeof(id_key.id)); 626 id_key.creatorid = sp->creatorid; 627 628 st = pf_find_state_byid(&id_key); 629 if (st == NULL) { 630 /* insert the update */ 631 if (pfsync_state_import(sp, flags)) 632 pfsyncstats.pfsyncs_badstate++; 633 continue; 634 } 635 sk = st->key[PF_SK_WIRE]; /* XXX right one? */ 636 sfail = 0; 637 if (sk->proto == IPPROTO_TCP) { 638 /* 639 * The state should never go backwards except 640 * for syn-proxy states. Neither should the 641 * sequence window slide backwards. 642 */ 643 if (st->src.state > sp->src.state && 644 (st->src.state < PF_TCPS_PROXY_SRC || 645 sp->src.state >= PF_TCPS_PROXY_SRC)) 646 sfail = 1; 647 else if (SEQ_GT(st->src.seqlo, 648 ntohl(sp->src.seqlo))) 649 sfail = 3; 650 else if (st->dst.state > sp->dst.state) { 651 /* There might still be useful 652 * information about the src state here, 653 * so import that part of the update, 654 * then "fail" so we send the updated 655 * state back to the peer who is missing 656 * our what we know. */ 657 pf_state_peer_ntoh(&sp->src, &st->src); 658 /* XXX do anything with timeouts? */ 659 sfail = 7; 660 flags = 0; 661 } else if (st->dst.state >= TCPS_SYN_SENT && 662 SEQ_GT(st->dst.seqlo, ntohl(sp->dst.seqlo))) 663 sfail = 4; 664 } else { 665 /* 666 * Non-TCP protocol state machine always go 667 * forwards 668 */ 669 if (st->src.state > sp->src.state) 670 sfail = 5; 671 else if (st->dst.state > sp->dst.state) 672 sfail = 6; 673 } 674 if (sfail) { 675 if (pf_status.debug >= PF_DEBUG_MISC) 676 kprintf("pfsync: %s stale update " 677 "(%d) id: %016jx " 678 "creatorid: %08x\n", 679 (sfail < 7 ? "ignoring" 680 : "partial"), sfail, 681 (uintmax_t)be64toh(st->id), 682 ntohl(st->creatorid)); 683 pfsyncstats.pfsyncs_stale++; 684 685 if (!(sp->sync_flags & PFSTATE_STALE)) { 686 /* we have a better state, send it */ 687 if (sc->sc_mbuf != NULL && !stale) 688 pfsync_sendout(sc); 689 stale++; 690 if (!st->sync_flags) 691 pfsync_pack_state( 692 PFSYNC_ACT_UPD, st, flags); 693 } 694 continue; 695 } 696 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 697 pf_state_peer_ntoh(&sp->src, &st->src); 698 pf_state_peer_ntoh(&sp->dst, &st->dst); 699 st->expire = ntohl(sp->expire) + time_second; 700 st->timeout = sp->timeout; 701 } 702 if (stale && sc->sc_mbuf != NULL) 703 pfsync_sendout(sc); 704 crit_exit(); 705 break; 706 /* 707 * It's not strictly necessary for us to support the "uncompressed" 708 * delete action, but it's relatively simple and maintains consistency. 709 */ 710 case PFSYNC_ACT_DEL: 711 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 712 count * sizeof(*sp), &offp)) == NULL) { 713 pfsyncstats.pfsyncs_badlen++; 714 return; 715 } 716 717 crit_enter(); 718 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 719 i < count; i++, sp++) { 720 bcopy(sp->id, &id_key.id, sizeof(id_key.id)); 721 id_key.creatorid = sp->creatorid; 722 723 st = pf_find_state_byid(&id_key); 724 if (st == NULL) { 725 pfsyncstats.pfsyncs_badstate++; 726 continue; 727 } 728 st->sync_flags |= PFSTATE_FROMSYNC; 729 pf_unlink_state(st); 730 } 731 crit_exit(); 732 break; 733 case PFSYNC_ACT_UPD_C: { 734 int update_requested = 0; 735 736 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 737 count * sizeof(*up), &offp)) == NULL) { 738 pfsyncstats.pfsyncs_badlen++; 739 return; 740 } 741 742 crit_enter(); 743 for (i = 0, up = (struct pfsync_state_upd *)(mp->m_data + offp); 744 i < count; i++, up++) { 745 /* check for invalid values */ 746 if (up->timeout >= PFTM_MAX || 747 up->src.state > PF_TCPS_PROXY_DST || 748 up->dst.state > PF_TCPS_PROXY_DST) { 749 if (pf_status.debug >= PF_DEBUG_MISC) 750 kprintf("pfsync_insert: " 751 "PFSYNC_ACT_UPD_C: " 752 "invalid value\n"); 753 pfsyncstats.pfsyncs_badval++; 754 continue; 755 } 756 757 bcopy(up->id, &id_key.id, sizeof(id_key.id)); 758 id_key.creatorid = up->creatorid; 759 760 st = pf_find_state_byid(&id_key); 761 if (st == NULL) { 762 /* We don't have this state. Ask for it. */ 763 error = pfsync_request_update(up, &src); 764 if (error == ENOMEM) { 765 crit_exit(); 766 goto done; 767 } 768 update_requested = 1; 769 pfsyncstats.pfsyncs_badstate++; 770 continue; 771 } 772 sk = st->key[PF_SK_WIRE]; /* XXX right one? */ 773 sfail = 0; 774 if (sk->proto == IPPROTO_TCP) { 775 /* 776 * The state should never go backwards except 777 * for syn-proxy states. Neither should the 778 * sequence window slide backwards. 779 */ 780 if (st->src.state > up->src.state && 781 (st->src.state < PF_TCPS_PROXY_SRC || 782 up->src.state >= PF_TCPS_PROXY_SRC)) 783 sfail = 1; 784 else if (st->dst.state > up->dst.state) 785 sfail = 2; 786 else if (SEQ_GT(st->src.seqlo, 787 ntohl(up->src.seqlo))) 788 sfail = 3; 789 else if (st->dst.state >= TCPS_SYN_SENT && 790 SEQ_GT(st->dst.seqlo, ntohl(up->dst.seqlo))) 791 sfail = 4; 792 } else { 793 /* 794 * Non-TCP protocol state machine always go 795 * forwards 796 */ 797 if (st->src.state > up->src.state) 798 sfail = 5; 799 else if (st->dst.state > up->dst.state) 800 sfail = 6; 801 } 802 if (sfail) { 803 if (pf_status.debug >= PF_DEBUG_MISC) 804 kprintf("pfsync: ignoring stale update " 805 "(%d) id: %016" PRIx64 " " 806 "creatorid: %08x\n", sfail, 807 be64toh(st->id), 808 ntohl(st->creatorid)); 809 pfsyncstats.pfsyncs_stale++; 810 811 /* we have a better state, send it out */ 812 if ((!stale || update_requested) && 813 sc->sc_mbuf != NULL) { 814 pfsync_sendout(sc); 815 update_requested = 0; 816 } 817 stale++; 818 if (!st->sync_flags) 819 pfsync_pack_state(PFSYNC_ACT_UPD, st, 820 PFSYNC_FLAG_STALE); 821 continue; 822 } 823 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 824 pf_state_peer_ntoh(&up->src, &st->src); 825 pf_state_peer_ntoh(&up->dst, &st->dst); 826 st->expire = ntohl(up->expire) + time_second; 827 st->timeout = up->timeout; 828 } 829 if ((update_requested || stale) && sc->sc_mbuf) 830 pfsync_sendout(sc); 831 crit_exit(); 832 break; 833 } 834 case PFSYNC_ACT_DEL_C: 835 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 836 count * sizeof(*dp), &offp)) == NULL) { 837 pfsyncstats.pfsyncs_badlen++; 838 return; 839 } 840 841 crit_enter(); 842 for (i = 0, dp = (struct pfsync_state_del *)(mp->m_data + offp); 843 i < count; i++, dp++) { 844 bcopy(dp->id, &id_key.id, sizeof(id_key.id)); 845 id_key.creatorid = dp->creatorid; 846 847 st = pf_find_state_byid(&id_key); 848 if (st == NULL) { 849 pfsyncstats.pfsyncs_badstate++; 850 continue; 851 } 852 st->sync_flags |= PFSTATE_FROMSYNC; 853 pf_unlink_state(st); 854 } 855 crit_exit(); 856 break; 857 case PFSYNC_ACT_INS_F: 858 case PFSYNC_ACT_DEL_F: 859 /* not implemented */ 860 break; 861 case PFSYNC_ACT_UREQ: 862 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 863 count * sizeof(*rup), &offp)) == NULL) { 864 pfsyncstats.pfsyncs_badlen++; 865 return; 866 } 867 868 crit_enter(); 869 if (sc->sc_mbuf != NULL) 870 pfsync_sendout(sc); 871 for (i = 0, 872 rup = (struct pfsync_state_upd_req *)(mp->m_data + offp); 873 i < count; i++, rup++) { 874 bcopy(rup->id, &id_key.id, sizeof(id_key.id)); 875 id_key.creatorid = rup->creatorid; 876 877 if (id_key.id == 0 && id_key.creatorid == 0) { 878 sc->sc_ureq_received = mycpu->gd_time_seconds; 879 if (sc->sc_bulk_send_next == NULL) 880 sc->sc_bulk_send_next = 881 TAILQ_FIRST(&state_list); 882 sc->sc_bulk_terminator = sc->sc_bulk_send_next; 883 if (pf_status.debug >= PF_DEBUG_MISC) 884 kprintf("pfsync: received " 885 "bulk update request\n"); 886 pfsync_send_bus(sc, PFSYNC_BUS_START); 887 lwkt_reltoken(&pf_token); 888 callout_init(&sc->sc_bulk_tmo); 889 lwkt_gettoken(&pf_token); 890 } else { 891 st = pf_find_state_byid(&id_key); 892 if (st == NULL) { 893 pfsyncstats.pfsyncs_badstate++; 894 continue; 895 } 896 if (!st->sync_flags) 897 pfsync_pack_state(PFSYNC_ACT_UPD, 898 st, 0); 899 } 900 } 901 if (sc->sc_mbuf != NULL) 902 pfsync_sendout(sc); 903 crit_exit(); 904 break; 905 case PFSYNC_ACT_BUS: 906 /* If we're not waiting for a bulk update, who cares. */ 907 if (sc->sc_ureq_sent == 0) 908 break; 909 910 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 911 sizeof(*bus), &offp)) == NULL) { 912 pfsyncstats.pfsyncs_badlen++; 913 return; 914 } 915 bus = (struct pfsync_state_bus *)(mp->m_data + offp); 916 switch (bus->status) { 917 case PFSYNC_BUS_START: 918 lwkt_reltoken(&pf_token); 919 callout_reset(&sc->sc_bulkfail_tmo, 920 pf_pool_limits[PF_LIMIT_STATES].limit / 921 (PFSYNC_BULKPACKETS * sc->sc_maxcount), 922 pfsync_bulkfail, LIST_FIRST(&pfsync_list)); 923 lwkt_gettoken(&pf_token); 924 if (pf_status.debug >= PF_DEBUG_MISC) 925 kprintf("pfsync: received bulk " 926 "update start\n"); 927 break; 928 case PFSYNC_BUS_END: 929 if (mycpu->gd_time_seconds - ntohl(bus->endtime) >= 930 sc->sc_ureq_sent) { 931 /* that's it, we're happy */ 932 sc->sc_ureq_sent = 0; 933 sc->sc_bulk_tries = 0; 934 lwkt_reltoken(&pf_token); 935 callout_stop(&sc->sc_bulkfail_tmo); 936 lwkt_gettoken(&pf_token); 937 #if NCARP > 0 938 if (!pfsync_sync_ok) { 939 lwkt_reltoken(&pf_token); 940 carp_group_demote_adj(&sc->sc_if, -1); 941 lwkt_gettoken(&pf_token); 942 } 943 #endif 944 pfsync_sync_ok = 1; 945 if (pf_status.debug >= PF_DEBUG_MISC) 946 kprintf("pfsync: received valid " 947 "bulk update end\n"); 948 } else { 949 if (pf_status.debug >= PF_DEBUG_MISC) 950 kprintf("pfsync: received invalid " 951 "bulk update end: bad timestamp\n"); 952 } 953 break; 954 } 955 break; 956 #ifdef IPSEC 957 case PFSYNC_ACT_TDB_UPD: 958 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 959 count * sizeof(*pt), &offp)) == NULL) { 960 pfsyncstats.pfsyncs_badlen++; 961 return; 962 } 963 crit_enter(); 964 for (i = 0, pt = (struct pfsync_tdb *)(mp->m_data + offp); 965 i < count; i++, pt++) 966 pfsync_update_net_tdb(pt); 967 crit_exit(); 968 break; 969 #endif 970 } 971 972 done: 973 if (m) 974 m_freem(m); 975 } 976 977 int 978 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst, 979 struct rtentry *rt) 980 { 981 m_freem(m); 982 return (0); 983 } 984 985 /* ARGSUSED */ 986 int 987 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr) 988 { 989 struct pfsync_softc *sc = ifp->if_softc; 990 struct ifreq *ifr = (struct ifreq *)data; 991 struct ip_moptions *imo = &sc->sc_imo; 992 struct pfsyncreq pfsyncr; 993 struct ifnet *sifp; 994 int error; 995 996 lwkt_gettoken(&pf_token); 997 998 switch (cmd) { 999 case SIOCSIFADDR: 1000 case SIOCAIFADDR: 1001 case SIOCSIFDSTADDR: 1002 case SIOCSIFFLAGS: 1003 if (ifp->if_flags & IFF_UP) 1004 ifp->if_flags |= IFF_RUNNING; 1005 else 1006 ifp->if_flags &= ~IFF_RUNNING; 1007 break; 1008 case SIOCSIFMTU: 1009 if (ifr->ifr_mtu < PFSYNC_MINMTU) { 1010 lwkt_reltoken(&pf_token); 1011 return (EINVAL); 1012 } 1013 if (ifr->ifr_mtu > MCLBYTES) 1014 ifr->ifr_mtu = MCLBYTES; 1015 crit_enter(); 1016 if (ifr->ifr_mtu < ifp->if_mtu) 1017 pfsync_sendout(sc); 1018 pfsync_setmtu(sc, ifr->ifr_mtu); 1019 crit_exit(); 1020 break; 1021 case SIOCGETPFSYNC: 1022 bzero(&pfsyncr, sizeof(pfsyncr)); 1023 if (sc->sc_sync_ifp) 1024 strlcpy(pfsyncr.pfsyncr_syncdev, 1025 sc->sc_sync_ifp->if_xname, IFNAMSIZ); 1026 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 1027 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 1028 lwkt_reltoken(&pf_token); 1029 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)))) 1030 return (error); 1031 lwkt_gettoken(&pf_token); 1032 break; 1033 case SIOCSETPFSYNC: 1034 if ((error = priv_check_cred(cr, PRIV_ROOT, NULL_CRED_OKAY)) != 0) { 1035 lwkt_reltoken(&pf_token); 1036 return (error); 1037 } 1038 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) { 1039 lwkt_reltoken(&pf_token); 1040 return (error); 1041 } 1042 1043 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 1044 sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP; 1045 else 1046 sc->sc_sync_peer.s_addr = 1047 pfsyncr.pfsyncr_syncpeer.s_addr; 1048 1049 if (pfsyncr.pfsyncr_maxupdates > 255) { 1050 lwkt_reltoken(&pf_token); 1051 return (EINVAL); 1052 } 1053 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 1054 1055 if (pfsyncr.pfsyncr_syncdev[0] == 0) { 1056 sc->sc_sync_ifp = NULL; 1057 if (sc->sc_mbuf_net != NULL) { 1058 /* Don't keep stale pfsync packets around. */ 1059 crit_enter(); 1060 m_freem(sc->sc_mbuf_net); 1061 sc->sc_mbuf_net = NULL; 1062 sc->sc_statep_net.s = NULL; 1063 crit_exit(); 1064 } 1065 if (imo->imo_num_memberships > 0) { 1066 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 1067 imo->imo_multicast_ifp = NULL; 1068 } 1069 break; 1070 } 1071 1072 if ((sifp = ifunit(pfsyncr.pfsyncr_syncdev)) == NULL) { 1073 lwkt_reltoken(&pf_token); 1074 return (EINVAL); 1075 } 1076 1077 crit_enter(); 1078 if (sifp->if_mtu < sc->sc_if.if_mtu || 1079 (sc->sc_sync_ifp != NULL && 1080 sifp->if_mtu < sc->sc_sync_ifp->if_mtu) || 1081 sifp->if_mtu < MCLBYTES - sizeof(struct ip)) 1082 pfsync_sendout(sc); 1083 sc->sc_sync_ifp = sifp; 1084 1085 pfsync_setmtu(sc, sc->sc_if.if_mtu); 1086 1087 if (imo->imo_num_memberships > 0) { 1088 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 1089 imo->imo_multicast_ifp = NULL; 1090 } 1091 1092 if (sc->sc_sync_ifp && 1093 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1094 struct in_addr addr; 1095 1096 if (!(sc->sc_sync_ifp->if_flags & IFF_MULTICAST)) { 1097 sc->sc_sync_ifp = NULL; 1098 lwkt_reltoken(&pf_token); 1099 crit_exit(); 1100 return (EADDRNOTAVAIL); 1101 } 1102 1103 addr.s_addr = INADDR_PFSYNC_GROUP; 1104 1105 if ((imo->imo_membership[0] = 1106 in_addmulti(&addr, sc->sc_sync_ifp)) == NULL) { 1107 sc->sc_sync_ifp = NULL; 1108 lwkt_reltoken(&pf_token); 1109 crit_exit(); 1110 return (ENOBUFS); 1111 } 1112 imo->imo_num_memberships++; 1113 imo->imo_multicast_ifp = sc->sc_sync_ifp; 1114 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 1115 imo->imo_multicast_loop = 0; 1116 } 1117 1118 if (sc->sc_sync_ifp || 1119 sc->sc_sendaddr.s_addr != INADDR_PFSYNC_GROUP) { 1120 /* Request a full state table update. */ 1121 sc->sc_ureq_sent = mycpu->gd_time_seconds; 1122 #if NCARP > 0 1123 if (pfsync_sync_ok) 1124 carp_group_demote_adj(&sc->sc_if, 1); 1125 #endif 1126 pfsync_sync_ok = 0; 1127 if (pf_status.debug >= PF_DEBUG_MISC) 1128 kprintf("pfsync: requesting bulk update\n"); 1129 lwkt_reltoken(&pf_token); 1130 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, 1131 pfsync_bulkfail, LIST_FIRST(&pfsync_list)); 1132 lwkt_gettoken(&pf_token); 1133 error = pfsync_request_update(NULL, NULL); 1134 if (error == ENOMEM) { 1135 lwkt_reltoken(&pf_token); 1136 crit_exit(); 1137 return (ENOMEM); 1138 } 1139 pfsync_sendout(sc); 1140 } 1141 crit_exit(); 1142 1143 break; 1144 1145 default: 1146 lwkt_reltoken(&pf_token); 1147 return (ENOTTY); 1148 } 1149 1150 lwkt_reltoken(&pf_token); 1151 return (0); 1152 } 1153 1154 void 1155 pfsync_setmtu(struct pfsync_softc *sc, int mtu_req) 1156 { 1157 int mtu; 1158 1159 if (sc->sc_sync_ifp && sc->sc_sync_ifp->if_mtu < mtu_req) 1160 mtu = sc->sc_sync_ifp->if_mtu; 1161 else 1162 mtu = mtu_req; 1163 1164 sc->sc_maxcount = (mtu - sizeof(struct pfsync_header)) / 1165 sizeof(struct pfsync_state); 1166 if (sc->sc_maxcount > 254) 1167 sc->sc_maxcount = 254; 1168 sc->sc_if.if_mtu = sizeof(struct pfsync_header) + 1169 sc->sc_maxcount * sizeof(struct pfsync_state); 1170 } 1171 1172 struct mbuf * 1173 pfsync_get_mbuf(struct pfsync_softc *sc, u_int8_t action, void **sp) 1174 { 1175 struct pfsync_header *h; 1176 struct mbuf *m; 1177 int len; 1178 1179 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1180 1181 MGETHDR(m, M_WAITOK, MT_DATA); 1182 if (m == NULL) { 1183 IFNET_STAT_INC(&sc->sc_if, oerrors, 1); 1184 return (NULL); 1185 } 1186 1187 switch (action) { 1188 case PFSYNC_ACT_CLR: 1189 len = sizeof(struct pfsync_header) + 1190 sizeof(struct pfsync_state_clr); 1191 break; 1192 case PFSYNC_ACT_UPD_C: 1193 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd)) + 1194 sizeof(struct pfsync_header); 1195 break; 1196 case PFSYNC_ACT_DEL_C: 1197 len = (sc->sc_maxcount * sizeof(struct pfsync_state_del)) + 1198 sizeof(struct pfsync_header); 1199 break; 1200 case PFSYNC_ACT_UREQ: 1201 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd_req)) + 1202 sizeof(struct pfsync_header); 1203 break; 1204 case PFSYNC_ACT_BUS: 1205 len = sizeof(struct pfsync_header) + 1206 sizeof(struct pfsync_state_bus); 1207 break; 1208 case PFSYNC_ACT_TDB_UPD: 1209 len = (sc->sc_maxcount * sizeof(struct pfsync_tdb)) + 1210 sizeof(struct pfsync_header); 1211 break; 1212 default: 1213 len = (sc->sc_maxcount * sizeof(struct pfsync_state)) + 1214 sizeof(struct pfsync_header); 1215 break; 1216 } 1217 1218 if (len > MHLEN) { 1219 MCLGET(m, M_WAITOK); 1220 if ((m->m_flags & M_EXT) == 0) { 1221 m_free(m); 1222 IFNET_STAT_INC(&sc->sc_if, oerrors, 1); 1223 return (NULL); 1224 } 1225 m->m_data += (MCLBYTES - len) &~ (sizeof(long) - 1); 1226 } else 1227 MH_ALIGN(m, len); 1228 1229 m->m_pkthdr.rcvif = NULL; 1230 m->m_pkthdr.len = m->m_len = sizeof(struct pfsync_header); 1231 h = mtod(m, struct pfsync_header *); 1232 h->version = PFSYNC_VERSION; 1233 h->af = 0; 1234 h->count = 0; 1235 h->action = action; 1236 1237 *sp = (void *)((char *)h + PFSYNC_HDRLEN); 1238 lwkt_reltoken(&pf_token); 1239 callout_reset(&sc->sc_tmo, hz, pfsync_timeout, 1240 LIST_FIRST(&pfsync_list)); 1241 lwkt_gettoken(&pf_token); 1242 return (m); 1243 } 1244 1245 int 1246 pfsync_pack_state(u_int8_t action, struct pf_state *st, int flags) 1247 { 1248 struct ifnet *ifp = NULL; 1249 struct pfsync_softc *sc = pfsyncif; 1250 struct pfsync_header *h, *h_net; 1251 struct pfsync_state *sp = NULL; 1252 struct pfsync_state_upd *up = NULL; 1253 struct pfsync_state_del *dp = NULL; 1254 int ret = 0; 1255 u_int8_t i = 255, newaction = 0; 1256 1257 if (sc == NULL) 1258 return (0); 1259 ifp = &sc->sc_if; 1260 1261 /* 1262 * If a packet falls in the forest and there's nobody around to 1263 * hear, does it make a sound? 1264 */ 1265 if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL && 1266 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1267 /* Don't leave any stale pfsync packets hanging around. */ 1268 if (sc->sc_mbuf != NULL) { 1269 m_freem(sc->sc_mbuf); 1270 sc->sc_mbuf = NULL; 1271 sc->sc_statep.s = NULL; 1272 } 1273 return (0); 1274 } 1275 1276 if (action >= PFSYNC_ACT_MAX) 1277 return (EINVAL); 1278 1279 crit_enter(); 1280 if (sc->sc_mbuf == NULL) { 1281 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1282 (void *)&sc->sc_statep.s)) == NULL) { 1283 crit_exit(); 1284 return (ENOMEM); 1285 } 1286 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1287 } else { 1288 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1289 if (h->action != action) { 1290 pfsync_sendout(sc); 1291 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1292 (void *)&sc->sc_statep.s)) == NULL) { 1293 crit_exit(); 1294 return (ENOMEM); 1295 } 1296 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1297 } else { 1298 /* 1299 * If it's an update, look in the packet to see if 1300 * we already have an update for the state. 1301 */ 1302 if (action == PFSYNC_ACT_UPD && sc->sc_maxupdates) { 1303 struct pfsync_state *usp = 1304 (void *)((char *)h + PFSYNC_HDRLEN); 1305 1306 for (i = 0; i < h->count; i++) { 1307 if (!memcmp(usp->id, &st->id, 1308 PFSYNC_ID_LEN) && 1309 usp->creatorid == st->creatorid) { 1310 sp = usp; 1311 sp->updates++; 1312 break; 1313 } 1314 usp++; 1315 } 1316 } 1317 } 1318 } 1319 1320 st->pfsync_time = mycpu->gd_time_seconds; 1321 1322 if (sp == NULL) { 1323 /* not a "duplicate" update */ 1324 i = 255; 1325 sp = sc->sc_statep.s++; 1326 sc->sc_mbuf->m_pkthdr.len = 1327 sc->sc_mbuf->m_len += sizeof(struct pfsync_state); 1328 h->count++; 1329 bzero(sp, sizeof(*sp)); 1330 1331 pfsync_state_export(sp, st); 1332 1333 if (flags & PFSYNC_FLAG_STALE) 1334 sp->sync_flags |= PFSTATE_STALE; 1335 } else { 1336 pf_state_peer_hton(&st->src, &sp->src); 1337 pf_state_peer_hton(&st->dst, &sp->dst); 1338 1339 if (st->expire <= time_second) 1340 sp->expire = htonl(0); 1341 else 1342 sp->expire = htonl(st->expire - time_second); 1343 } 1344 1345 /* do we need to build "compressed" actions for network transfer? */ 1346 if (sc->sc_sync_ifp && flags & PFSYNC_FLAG_COMPRESS) { 1347 switch (action) { 1348 case PFSYNC_ACT_UPD: 1349 newaction = PFSYNC_ACT_UPD_C; 1350 break; 1351 case PFSYNC_ACT_DEL: 1352 newaction = PFSYNC_ACT_DEL_C; 1353 break; 1354 default: 1355 /* by default we just send the uncompressed states */ 1356 break; 1357 } 1358 } 1359 1360 if (newaction) { 1361 if (sc->sc_mbuf_net == NULL) { 1362 if ((sc->sc_mbuf_net = pfsync_get_mbuf(sc, newaction, 1363 (void *)&sc->sc_statep_net.s)) == NULL) { 1364 crit_exit(); 1365 return (ENOMEM); 1366 } 1367 } 1368 h_net = mtod(sc->sc_mbuf_net, struct pfsync_header *); 1369 1370 switch (newaction) { 1371 case PFSYNC_ACT_UPD_C: 1372 if (i != 255) { 1373 up = (void *)((char *)h_net + 1374 PFSYNC_HDRLEN + (i * sizeof(*up))); 1375 up->updates++; 1376 } else { 1377 h_net->count++; 1378 sc->sc_mbuf_net->m_pkthdr.len = 1379 sc->sc_mbuf_net->m_len += sizeof(*up); 1380 up = sc->sc_statep_net.u++; 1381 1382 bzero(up, sizeof(*up)); 1383 bcopy(&st->id, up->id, sizeof(up->id)); 1384 up->creatorid = st->creatorid; 1385 } 1386 up->timeout = st->timeout; 1387 up->expire = sp->expire; 1388 up->src = sp->src; 1389 up->dst = sp->dst; 1390 break; 1391 case PFSYNC_ACT_DEL_C: 1392 sc->sc_mbuf_net->m_pkthdr.len = 1393 sc->sc_mbuf_net->m_len += sizeof(*dp); 1394 dp = sc->sc_statep_net.d++; 1395 h_net->count++; 1396 1397 bzero(dp, sizeof(*dp)); 1398 bcopy(&st->id, dp->id, sizeof(dp->id)); 1399 dp->creatorid = st->creatorid; 1400 break; 1401 } 1402 } 1403 1404 if (h->count == sc->sc_maxcount || 1405 (sc->sc_maxupdates && (sp->updates >= sc->sc_maxupdates))) 1406 ret = pfsync_sendout(sc); 1407 1408 crit_exit(); 1409 return (ret); 1410 } 1411 1412 int 1413 pfsync_request_update(struct pfsync_state_upd *up, struct in_addr *src) 1414 { 1415 struct pfsync_header *h; 1416 struct pfsync_softc *sc = pfsyncif; 1417 struct pfsync_state_upd_req *rup; 1418 int ret = 0; 1419 1420 if (sc == NULL) 1421 return (0); 1422 1423 if (sc->sc_mbuf == NULL) { 1424 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1425 (void *)&sc->sc_statep.s)) == NULL) 1426 return (ENOMEM); 1427 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1428 } else { 1429 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1430 if (h->action != PFSYNC_ACT_UREQ) { 1431 pfsync_sendout(sc); 1432 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1433 (void *)&sc->sc_statep.s)) == NULL) 1434 return (ENOMEM); 1435 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1436 } 1437 } 1438 1439 if (src != NULL) 1440 sc->sc_sendaddr = *src; 1441 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*rup); 1442 h->count++; 1443 rup = sc->sc_statep.r++; 1444 bzero(rup, sizeof(*rup)); 1445 if (up != NULL) { 1446 bcopy(up->id, rup->id, sizeof(rup->id)); 1447 rup->creatorid = up->creatorid; 1448 } 1449 1450 if (h->count == sc->sc_maxcount) 1451 ret = pfsync_sendout(sc); 1452 1453 return (ret); 1454 } 1455 1456 int 1457 pfsync_clear_states(u_int32_t creatorid, char *ifname) 1458 { 1459 struct pfsync_softc *sc = pfsyncif; 1460 struct pfsync_state_clr *cp; 1461 int ret; 1462 1463 if (sc == NULL) 1464 return (0); 1465 1466 crit_enter(); 1467 if (sc->sc_mbuf != NULL) 1468 pfsync_sendout(sc); 1469 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_CLR, 1470 (void *)&sc->sc_statep.c)) == NULL) { 1471 crit_exit(); 1472 return (ENOMEM); 1473 } 1474 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*cp); 1475 cp = sc->sc_statep.c; 1476 cp->creatorid = creatorid; 1477 if (ifname != NULL) 1478 strlcpy(cp->ifname, ifname, IFNAMSIZ); 1479 1480 ret = (pfsync_sendout(sc)); 1481 crit_exit(); 1482 return (ret); 1483 } 1484 1485 void 1486 pfsync_timeout(void *v) 1487 { 1488 struct pfsync_softc *sc = v; 1489 1490 crit_enter(); 1491 pfsync_sendout(sc); 1492 crit_exit(); 1493 } 1494 1495 void 1496 pfsync_send_bus(struct pfsync_softc *sc, u_int8_t status) 1497 { 1498 struct pfsync_state_bus *bus; 1499 1500 if (sc->sc_mbuf != NULL) 1501 pfsync_sendout(sc); 1502 1503 if (pfsync_sync_ok && 1504 (sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_BUS, 1505 (void *)&sc->sc_statep.b)) != NULL) { 1506 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*bus); 1507 bus = sc->sc_statep.b; 1508 bus->creatorid = pf_status.hostid; 1509 bus->status = status; 1510 bus->endtime = htonl(mycpu->gd_time_seconds - sc->sc_ureq_received); 1511 pfsync_sendout(sc); 1512 } 1513 } 1514 1515 void 1516 pfsync_bulk_update(void *v) 1517 { 1518 struct pfsync_softc *sc = v; 1519 int i = 0; 1520 struct pf_state *state; 1521 1522 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1523 1524 crit_enter(); 1525 if (sc->sc_mbuf != NULL) 1526 pfsync_sendout(sc); 1527 1528 /* 1529 * Grab at most PFSYNC_BULKPACKETS worth of states which have not 1530 * been sent since the latest request was made. 1531 */ 1532 state = sc->sc_bulk_send_next; 1533 if (state) 1534 do { 1535 /* send state update if syncable and not already sent */ 1536 if (!state->sync_flags 1537 && state->timeout < PFTM_MAX 1538 && state->pfsync_time <= sc->sc_ureq_received) { 1539 pfsync_pack_state(PFSYNC_ACT_UPD, state, 0); 1540 i++; 1541 } 1542 1543 /* figure next state to send */ 1544 state = TAILQ_NEXT(state, entry_list); 1545 1546 /* wrap to start of list if we hit the end */ 1547 if (!state) 1548 state = TAILQ_FIRST(&state_list); 1549 } while (i < sc->sc_maxcount * PFSYNC_BULKPACKETS && 1550 state != sc->sc_bulk_terminator); 1551 1552 if (!state || state == sc->sc_bulk_terminator) { 1553 /* we're done */ 1554 pfsync_send_bus(sc, PFSYNC_BUS_END); 1555 sc->sc_ureq_received = 0; 1556 sc->sc_bulk_send_next = NULL; 1557 sc->sc_bulk_terminator = NULL; 1558 lwkt_reltoken(&pf_token); 1559 callout_stop(&sc->sc_bulk_tmo); 1560 lwkt_gettoken(&pf_token); 1561 if (pf_status.debug >= PF_DEBUG_MISC) 1562 kprintf("pfsync: bulk update complete\n"); 1563 } else { 1564 /* look again for more in a bit */ 1565 lwkt_reltoken(&pf_token); 1566 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_timeout, 1567 LIST_FIRST(&pfsync_list)); 1568 lwkt_gettoken(&pf_token); 1569 sc->sc_bulk_send_next = state; 1570 } 1571 if (sc->sc_mbuf != NULL) 1572 pfsync_sendout(sc); 1573 crit_exit(); 1574 } 1575 1576 void 1577 pfsync_bulkfail(void *v) 1578 { 1579 struct pfsync_softc *sc = v; 1580 int error; 1581 1582 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1583 1584 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 1585 /* Try again in a bit */ 1586 lwkt_reltoken(&pf_token); 1587 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulkfail, 1588 LIST_FIRST(&pfsync_list)); 1589 lwkt_gettoken(&pf_token); 1590 crit_enter(); 1591 error = pfsync_request_update(NULL, NULL); 1592 if (error == ENOMEM) { 1593 if (pf_status.debug >= PF_DEBUG_MISC) 1594 kprintf("pfsync: cannot allocate mbufs for " 1595 "bulk update\n"); 1596 } else 1597 pfsync_sendout(sc); 1598 crit_exit(); 1599 } else { 1600 /* Pretend like the transfer was ok */ 1601 sc->sc_ureq_sent = 0; 1602 sc->sc_bulk_tries = 0; 1603 #if NCARP > 0 1604 if (!pfsync_sync_ok) 1605 carp_group_demote_adj(&sc->sc_if, -1); 1606 #endif 1607 pfsync_sync_ok = 1; 1608 if (pf_status.debug >= PF_DEBUG_MISC) 1609 kprintf("pfsync: failed to receive " 1610 "bulk update status\n"); 1611 lwkt_reltoken(&pf_token); 1612 callout_stop(&sc->sc_bulkfail_tmo); 1613 lwkt_gettoken(&pf_token); 1614 } 1615 } 1616 1617 /* This must be called in splnet() */ 1618 int 1619 pfsync_sendout(struct pfsync_softc *sc) 1620 { 1621 #if NBPFILTER > 0 1622 struct ifnet *ifp = &sc->sc_if; 1623 #endif 1624 struct mbuf *m; 1625 1626 ASSERT_LWKT_TOKEN_HELD(&pf_token); 1627 1628 lwkt_reltoken(&pf_token); 1629 callout_stop(&sc->sc_tmo); 1630 lwkt_gettoken(&pf_token); 1631 1632 if (sc->sc_mbuf == NULL) 1633 return (0); 1634 m = sc->sc_mbuf; 1635 sc->sc_mbuf = NULL; 1636 sc->sc_statep.s = NULL; 1637 1638 #if NBPFILTER > 0 1639 if (ifp->if_bpf) { 1640 bpf_gettoken(); 1641 if (ifp->if_bpf) 1642 bpf_mtap(ifp->if_bpf, m, BPF_DIRECTION_OUT); 1643 bpf_reltoken(); 1644 } 1645 #endif 1646 1647 if (sc->sc_mbuf_net) { 1648 m_freem(m); 1649 m = sc->sc_mbuf_net; 1650 sc->sc_mbuf_net = NULL; 1651 sc->sc_statep_net.s = NULL; 1652 } 1653 1654 return pfsync_sendout_mbuf(sc, m); 1655 } 1656 1657 int 1658 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m) 1659 { 1660 struct sockaddr sa; 1661 struct ip *ip; 1662 1663 if (sc->sc_sync_ifp || 1664 sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) { 1665 M_PREPEND(m, sizeof(struct ip), M_WAITOK); 1666 if (m == NULL) { 1667 pfsyncstats.pfsyncs_onomem++; 1668 return (0); 1669 } 1670 ip = mtod(m, struct ip *); 1671 ip->ip_v = IPVERSION; 1672 ip->ip_hl = sizeof(*ip) >> 2; 1673 ip->ip_tos = IPTOS_LOWDELAY; 1674 ip->ip_len = htons(m->m_pkthdr.len); 1675 ip->ip_id = htons(ip_randomid()); 1676 ip->ip_off = htons(IP_DF); 1677 ip->ip_ttl = PFSYNC_DFLTTL; 1678 ip->ip_p = IPPROTO_PFSYNC; 1679 ip->ip_sum = 0; 1680 1681 bzero(&sa, sizeof(sa)); 1682 ip->ip_src.s_addr = INADDR_ANY; 1683 1684 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP) 1685 m->m_flags |= M_MCAST; 1686 ip->ip_dst = sc->sc_sendaddr; 1687 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr; 1688 1689 pfsyncstats.pfsyncs_opackets++; 1690 1691 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL)) 1692 pfsyncstats.pfsyncs_oerrors++; 1693 } else 1694 m_freem(m); 1695 1696 return (0); 1697 } 1698 1699 static int 1700 pfsync_modevent(module_t mod, int type, void *data) 1701 { 1702 int error = 0; 1703 1704 struct pfsync_softc *pfs_if, *tmp; 1705 1706 lwkt_gettoken(&pf_token); 1707 1708 switch (type) { 1709 case MOD_LOAD: 1710 LIST_INIT(&pfsync_list); 1711 lwkt_reltoken(&pf_token); 1712 if_clone_attach(&pfsync_cloner); 1713 lwkt_gettoken(&pf_token); 1714 /* Override the function pointer for pf_ioctl.c */ 1715 break; 1716 1717 case MOD_UNLOAD: 1718 lwkt_reltoken(&pf_token); 1719 if_clone_detach(&pfsync_cloner); 1720 lwkt_gettoken(&pf_token); 1721 LIST_FOREACH_MUTABLE(pfs_if, &pfsync_list, sc_next, tmp) { 1722 pfsync_clone_destroy(&pfs_if->sc_if); 1723 } 1724 break; 1725 1726 default: 1727 error = EINVAL; 1728 break; 1729 } 1730 1731 lwkt_reltoken(&pf_token); 1732 return error; 1733 } 1734 1735 static moduledata_t pfsync_mod = { 1736 "pfsync", 1737 pfsync_modevent, 1738 0 1739 }; 1740 1741 #define PFSYNC_MODVER 44 1742 1743 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 1744 MODULE_VERSION(pfsync, PFSYNC_MODVER); 1745 1746 1747 1748