1 /* $NetBSD: if_pfsync.c,v 1.6 2010/04/05 07:22:22 joerg Exp $ */ 2 /* $OpenBSD: if_pfsync.c,v 1.83 2007/06/26 14:44:12 mcbride Exp $ */ 3 4 /* 5 * Copyright (c) 2002 Michael Shalayeff 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, 21 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 22 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 23 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 25 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 26 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 27 * THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 __KERNEL_RCSID(0, "$NetBSD: if_pfsync.c,v 1.6 2010/04/05 07:22:22 joerg Exp $"); 32 33 #ifdef _KERNEL_OPT 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #endif 37 38 #include <sys/param.h> 39 #include <sys/proc.h> 40 #include <sys/systm.h> 41 #include <sys/time.h> 42 #include <sys/mbuf.h> 43 #include <sys/socket.h> 44 #include <sys/ioctl.h> 45 #include <sys/callout.h> 46 #include <sys/kernel.h> 47 48 #include <net/if.h> 49 #include <net/if_types.h> 50 #include <net/route.h> 51 #include <net/bpf.h> 52 #include <netinet/in.h> 53 #ifndef __NetBSD__ 54 #include <netinet/if_ether.h> 55 #else 56 #include <net/if_ether.h> 57 #endif /* __NetBSD__ */ 58 #include <netinet/tcp.h> 59 #include <netinet/tcp_seq.h> 60 61 #ifdef INET 62 #include <netinet/in_systm.h> 63 #include <netinet/in_var.h> 64 #include <netinet/ip.h> 65 #include <netinet/ip_var.h> 66 #endif 67 68 #ifdef INET6 69 #include <netinet6/nd6.h> 70 #endif /* INET6 */ 71 72 #include "carp.h" 73 #if NCARP > 0 74 extern int carp_suppress_preempt; 75 #endif 76 77 #include <net/pfvar.h> 78 #include <net/if_pfsync.h> 79 80 #ifdef __NetBSD__ 81 #include <sys/conf.h> 82 #include <sys/lwp.h> 83 #include <sys/kauth.h> 84 #include <sys/sysctl.h> 85 86 #include <net/net_stats.h> 87 88 percpu_t *pfsyncstat_percpu; 89 90 #define PFSYNC_STATINC(x) _NET_STATINC(pfsyncstat_percpu, x) 91 #endif /* __NetBSD__ */ 92 93 #include "pfsync.h" 94 95 #define PFSYNC_MINMTU \ 96 (sizeof(struct pfsync_header) + sizeof(struct pf_state)) 97 98 #ifdef PFSYNCDEBUG 99 #define DPRINTF(x) do { if (pfsyncdebug) printf x ; } while (0) 100 int pfsyncdebug; 101 #else 102 #define DPRINTF(x) 103 #endif 104 105 extern int ifqmaxlen; /* XXX */ 106 107 struct pfsync_softc *pfsyncif = NULL; 108 109 void pfsyncattach(int); 110 int pfsync_clone_create(struct if_clone *, int); 111 int pfsync_clone_destroy(struct ifnet *); 112 void pfsync_setmtu(struct pfsync_softc *, int); 113 int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, 114 struct pf_state_peer *); 115 int pfsync_insert_net_state(struct pfsync_state *, u_int8_t); 116 void pfsync_update_net_tdb(struct pfsync_tdb *); 117 int pfsyncoutput(struct ifnet *, struct mbuf *, const struct sockaddr *, 118 struct rtentry *); 119 int pfsyncioctl(struct ifnet *, u_long, void*); 120 void pfsyncstart(struct ifnet *); 121 122 struct mbuf *pfsync_get_mbuf(struct pfsync_softc *, u_int8_t, void **); 123 int pfsync_request_update(struct pfsync_state_upd *, struct in_addr *); 124 int pfsync_sendout(struct pfsync_softc *); 125 int pfsync_tdb_sendout(struct pfsync_softc *); 126 int pfsync_sendout_mbuf(struct pfsync_softc *, struct mbuf *); 127 void pfsync_timeout(void *); 128 void pfsync_tdb_timeout(void *); 129 void pfsync_send_bus(struct pfsync_softc *, u_int8_t); 130 void pfsync_bulk_update(void *); 131 void pfsync_bulkfail(void *); 132 133 int pfsync_sync_ok; 134 135 struct if_clone pfsync_cloner = 136 IF_CLONE_INITIALIZER("pfsync", pfsync_clone_create, pfsync_clone_destroy); 137 138 void 139 pfsyncattach(int npfsync) 140 { 141 if_clone_attach(&pfsync_cloner); 142 143 pfsyncstat_percpu = percpu_alloc(sizeof(uint64_t) * PFSYNC_NSTATS); 144 } 145 146 int 147 pfsync_clone_create(struct if_clone *ifc, int unit) 148 { 149 struct ifnet *ifp; 150 151 if (unit != 0) 152 return (EINVAL); 153 154 pfsync_sync_ok = 1; 155 if ((pfsyncif = malloc(sizeof(*pfsyncif), M_DEVBUF, M_NOWAIT)) == NULL) 156 return (ENOMEM); 157 memset(pfsyncif, 0, sizeof(*pfsyncif)); 158 pfsyncif->sc_mbuf = NULL; 159 pfsyncif->sc_mbuf_net = NULL; 160 pfsyncif->sc_mbuf_tdb = NULL; 161 pfsyncif->sc_statep.s = NULL; 162 pfsyncif->sc_statep_net.s = NULL; 163 pfsyncif->sc_statep_tdb.t = NULL; 164 pfsyncif->sc_maxupdates = 128; 165 pfsyncif->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP; 166 pfsyncif->sc_sendaddr.s_addr = INADDR_PFSYNC_GROUP; 167 pfsyncif->sc_ureq_received = 0; 168 pfsyncif->sc_ureq_sent = 0; 169 pfsyncif->sc_bulk_send_next = NULL; 170 pfsyncif->sc_bulk_terminator = NULL; 171 ifp = &pfsyncif->sc_if; 172 snprintf(ifp->if_xname, sizeof ifp->if_xname, "pfsync%d", unit); 173 ifp->if_softc = pfsyncif; 174 ifp->if_ioctl = pfsyncioctl; 175 ifp->if_output = pfsyncoutput; 176 ifp->if_start = pfsyncstart; 177 ifp->if_type = IFT_PFSYNC; 178 ifp->if_snd.ifq_maxlen = ifqmaxlen; 179 ifp->if_hdrlen = PFSYNC_HDRLEN; 180 pfsync_setmtu(pfsyncif, ETHERMTU); 181 182 callout_init(&pfsyncif->sc_tmo, 0); 183 callout_init(&pfsyncif->sc_tdb_tmo, 0); 184 callout_init(&pfsyncif->sc_bulk_tmo, 0); 185 callout_init(&pfsyncif->sc_bulkfail_tmo, 0); 186 callout_setfunc(&pfsyncif->sc_tmo, pfsync_timeout, pfsyncif); 187 callout_setfunc(&pfsyncif->sc_tdb_tmo, pfsync_tdb_timeout, pfsyncif); 188 callout_setfunc(&pfsyncif->sc_bulk_tmo, pfsync_bulk_update, pfsyncif); 189 callout_setfunc(&pfsyncif->sc_bulkfail_tmo, pfsync_bulkfail, pfsyncif); 190 191 if_attach(ifp); 192 if_alloc_sadl(ifp); 193 194 bpf_attach(&pfsyncif->sc_if, DLT_PFSYNC, PFSYNC_HDRLEN); 195 196 return (0); 197 } 198 199 int 200 pfsync_clone_destroy(struct ifnet *ifp) 201 { 202 bpf_detach(ifp); 203 if_detach(ifp); 204 free(pfsyncif, M_DEVBUF); 205 pfsyncif = NULL; 206 return (0); 207 } 208 209 /* 210 * Start output on the pfsync interface. 211 */ 212 void 213 pfsyncstart(struct ifnet *ifp) 214 { 215 struct mbuf *m; 216 int s; 217 218 for (;;) { 219 s = splnet(); 220 IF_DROP(&ifp->if_snd); 221 IF_DEQUEUE(&ifp->if_snd, m); 222 splx(s); 223 224 if (m == NULL) 225 return; 226 else 227 m_freem(m); 228 } 229 } 230 231 int 232 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 233 struct pf_state_peer *d) 234 { 235 if (s->scrub.scrub_flag && d->scrub == NULL) { 236 d->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT); 237 if (d->scrub == NULL) 238 return (ENOMEM); 239 memset(d->scrub, 0, sizeof(*d->scrub)); 240 } 241 242 return (0); 243 } 244 245 int 246 pfsync_insert_net_state(struct pfsync_state *sp, u_int8_t chksum_flag) 247 { 248 struct pf_state *st = NULL; 249 struct pf_state_key *sk = NULL; 250 struct pf_rule *r = NULL; 251 struct pfi_kif *kif; 252 253 if (sp->creatorid == 0 && pf_status.debug >= PF_DEBUG_MISC) { 254 printf("pfsync_insert_net_state: invalid creator id:" 255 " %08x\n", ntohl(sp->creatorid)); 256 return (EINVAL); 257 } 258 259 kif = pfi_kif_get(sp->ifname); 260 if (kif == NULL) { 261 if (pf_status.debug >= PF_DEBUG_MISC) 262 printf("pfsync_insert_net_state: " 263 "unknown interface: %s\n", sp->ifname); 264 /* skip this state */ 265 return (0); 266 } 267 268 /* 269 * If the ruleset checksums match, it's safe to associate the state 270 * with the rule of that number. 271 */ 272 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && chksum_flag && 273 ntohl(sp->rule) < 274 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 275 r = pf_main_ruleset.rules[ 276 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 277 else 278 r = &pf_default_rule; 279 280 if (!r->max_states || r->states < r->max_states) 281 st = pool_get(&pf_state_pl, PR_NOWAIT); 282 if (st == NULL) { 283 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 284 return (ENOMEM); 285 } 286 memset(st, 0, sizeof(*st)); 287 288 if ((sk = pf_alloc_state_key(st)) == NULL) { 289 pool_put(&pf_state_pl, st); 290 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 291 return (ENOMEM); 292 } 293 294 /* allocate memory for scrub info */ 295 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 296 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) { 297 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 298 if (st->src.scrub) 299 pool_put(&pf_state_scrub_pl, st->src.scrub); 300 pool_put(&pf_state_pl, st); 301 pool_put(&pf_state_key_pl, sk); 302 return (ENOMEM); 303 } 304 305 st->rule.ptr = r; 306 /* XXX get pointers to nat_rule and anchor */ 307 308 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 309 r->states++; 310 311 /* fill in the rest of the state entry */ 312 pf_state_host_ntoh(&sp->lan, &sk->lan); 313 pf_state_host_ntoh(&sp->gwy, &sk->gwy); 314 pf_state_host_ntoh(&sp->ext, &sk->ext); 315 316 pf_state_peer_ntoh(&sp->src, &st->src); 317 pf_state_peer_ntoh(&sp->dst, &st->dst); 318 319 memcpy(&st->rt_addr, &sp->rt_addr, sizeof(st->rt_addr)); 320 st->creation = time_second - ntohl(sp->creation); 321 st->expire = ntohl(sp->expire) + time_second; 322 323 sk->af = sp->af; 324 sk->proto = sp->proto; 325 sk->direction = sp->direction; 326 st->log = sp->log; 327 st->timeout = sp->timeout; 328 st->allow_opts = sp->allow_opts; 329 330 memcpy(&st->id, sp->id, sizeof(st->id)); 331 st->creatorid = sp->creatorid; 332 st->sync_flags = PFSTATE_FROMSYNC; 333 334 if (pf_insert_state(kif, st)) { 335 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 336 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */ 337 r->states--; 338 if (st->dst.scrub) 339 pool_put(&pf_state_scrub_pl, st->dst.scrub); 340 if (st->src.scrub) 341 pool_put(&pf_state_scrub_pl, st->src.scrub); 342 pool_put(&pf_state_pl, st); 343 return (EINVAL); 344 } 345 346 return (0); 347 } 348 349 void 350 pfsync_input(struct mbuf *m, ...) 351 { 352 struct ip *ip = mtod(m, struct ip *); 353 struct pfsync_header *ph; 354 struct pfsync_softc *sc = pfsyncif; 355 struct pf_state *st; 356 struct pf_state_key *sk; 357 struct pf_state_cmp id_key; 358 struct pfsync_state *sp; 359 struct pfsync_state_upd *up; 360 struct pfsync_state_del *dp; 361 struct pfsync_state_clr *cp; 362 struct pfsync_state_upd_req *rup; 363 struct pfsync_state_bus *bus; 364 #ifdef IPSEC 365 struct pfsync_tdb *pt; 366 #endif 367 struct in_addr src; 368 struct mbuf *mp; 369 int iplen, action, error, i, s, count, offp, sfail, stale = 0; 370 u_int8_t chksum_flag = 0; 371 372 PFSYNC_STATINC(PFSYNC_STAT_IPACKETS); 373 374 /* verify that we have a sync interface configured */ 375 if (!sc || !sc->sc_sync_ifp || !pf_status.running) 376 goto done; 377 378 /* verify that the packet came in on the right interface */ 379 if (sc->sc_sync_ifp != m->m_pkthdr.rcvif) { 380 PFSYNC_STATINC(PFSYNC_STAT_BADIF); 381 goto done; 382 } 383 384 /* verify that the IP TTL is 255. */ 385 if (ip->ip_ttl != PFSYNC_DFLTTL) { 386 PFSYNC_STATINC(PFSYNC_STAT_BADTTL); 387 goto done; 388 } 389 390 iplen = ip->ip_hl << 2; 391 392 if (m->m_pkthdr.len < iplen + sizeof(*ph)) { 393 PFSYNC_STATINC(PFSYNC_STAT_HDROPS); 394 goto done; 395 } 396 397 if (iplen + sizeof(*ph) > m->m_len) { 398 if ((m = m_pullup(m, iplen + sizeof(*ph))) == NULL) { 399 PFSYNC_STATINC(PFSYNC_STAT_HDROPS); 400 goto done; 401 } 402 ip = mtod(m, struct ip *); 403 } 404 ph = (struct pfsync_header *)((char *)ip + iplen); 405 406 /* verify the version */ 407 if (ph->version != PFSYNC_VERSION) { 408 PFSYNC_STATINC(PFSYNC_STAT_BADVER); 409 goto done; 410 } 411 412 action = ph->action; 413 count = ph->count; 414 415 /* make sure it's a valid action code */ 416 if (action >= PFSYNC_ACT_MAX) { 417 PFSYNC_STATINC(PFSYNC_STAT_BADACT); 418 goto done; 419 } 420 421 /* Cheaper to grab this now than having to mess with mbufs later */ 422 src = ip->ip_src; 423 424 if (!bcmp(&ph->pf_chksum, &pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 425 chksum_flag++; 426 427 switch (action) { 428 case PFSYNC_ACT_CLR: { 429 struct pf_state *nexts; 430 struct pf_state_key *nextsk; 431 struct pfi_kif *kif; 432 u_int32_t creatorid; 433 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 434 sizeof(*cp), &offp)) == NULL) { 435 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 436 return; 437 } 438 cp = (struct pfsync_state_clr *)(mp->m_data + offp); 439 creatorid = cp->creatorid; 440 441 s = splsoftnet(); 442 if (cp->ifname[0] == '\0') { 443 for (st = RB_MIN(pf_state_tree_id, &tree_id); 444 st; st = nexts) { 445 nexts = RB_NEXT(pf_state_tree_id, &tree_id, st); 446 if (st->creatorid == creatorid) { 447 st->sync_flags |= PFSTATE_FROMSYNC; 448 pf_unlink_state(st); 449 } 450 } 451 } else { 452 if ((kif = pfi_kif_get(cp->ifname)) == NULL) { 453 splx(s); 454 return; 455 } 456 for (sk = RB_MIN(pf_state_tree_lan_ext, 457 &pf_statetbl_lan_ext); sk; sk = nextsk) { 458 nextsk = RB_NEXT(pf_state_tree_lan_ext, 459 &pf_statetbl_lan_ext, sk); 460 TAILQ_FOREACH(st, &sk->states, next) { 461 if (st->creatorid == creatorid) { 462 st->sync_flags |= 463 PFSTATE_FROMSYNC; 464 pf_unlink_state(st); 465 } 466 } 467 } 468 } 469 splx(s); 470 471 break; 472 } 473 case PFSYNC_ACT_INS: 474 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 475 count * sizeof(*sp), &offp)) == NULL) { 476 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 477 return; 478 } 479 480 s = splsoftnet(); 481 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 482 i < count; i++, sp++) { 483 /* check for invalid values */ 484 if (sp->timeout >= PFTM_MAX || 485 sp->src.state > PF_TCPS_PROXY_DST || 486 sp->dst.state > PF_TCPS_PROXY_DST || 487 sp->direction > PF_OUT || 488 (sp->af != AF_INET && sp->af != AF_INET6)) { 489 if (pf_status.debug >= PF_DEBUG_MISC) 490 printf("pfsync_insert: PFSYNC_ACT_INS: " 491 "invalid value\n"); 492 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 493 continue; 494 } 495 496 if ((error = pfsync_insert_net_state(sp, 497 chksum_flag))) { 498 if (error == ENOMEM) { 499 splx(s); 500 goto done; 501 } 502 continue; 503 } 504 } 505 splx(s); 506 break; 507 case PFSYNC_ACT_UPD: 508 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 509 count * sizeof(*sp), &offp)) == NULL) { 510 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 511 return; 512 } 513 514 s = splsoftnet(); 515 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 516 i < count; i++, sp++) { 517 int flags = PFSYNC_FLAG_STALE; 518 519 /* check for invalid values */ 520 if (sp->timeout >= PFTM_MAX || 521 sp->src.state > PF_TCPS_PROXY_DST || 522 sp->dst.state > PF_TCPS_PROXY_DST) { 523 if (pf_status.debug >= PF_DEBUG_MISC) 524 printf("pfsync_insert: PFSYNC_ACT_UPD: " 525 "invalid value\n"); 526 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 527 continue; 528 } 529 530 memcpy(&id_key.id, sp->id, sizeof(id_key.id)); 531 id_key.creatorid = sp->creatorid; 532 533 st = pf_find_state_byid(&id_key); 534 if (st == NULL) { 535 /* insert the update */ 536 if (pfsync_insert_net_state(sp, chksum_flag)) { 537 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 538 } 539 continue; 540 } 541 sk = st->state_key; 542 sfail = 0; 543 if (sk->proto == IPPROTO_TCP) { 544 /* 545 * The state should never go backwards except 546 * for syn-proxy states. Neither should the 547 * sequence window slide backwards. 548 */ 549 if (st->src.state > sp->src.state && 550 (st->src.state < PF_TCPS_PROXY_SRC || 551 sp->src.state >= PF_TCPS_PROXY_SRC)) 552 sfail = 1; 553 else if (SEQ_GT(st->src.seqlo, 554 ntohl(sp->src.seqlo))) 555 sfail = 3; 556 else if (st->dst.state > sp->dst.state) { 557 /* There might still be useful 558 * information about the src state here, 559 * so import that part of the update, 560 * then "fail" so we send the updated 561 * state back to the peer who is missing 562 * our what we know. */ 563 pf_state_peer_ntoh(&sp->src, &st->src); 564 /* XXX do anything with timeouts? */ 565 sfail = 7; 566 flags = 0; 567 } else if (st->dst.state >= TCPS_SYN_SENT && 568 SEQ_GT(st->dst.seqlo, ntohl(sp->dst.seqlo))) 569 sfail = 4; 570 } else { 571 /* 572 * Non-TCP protocol state machine always go 573 * forwards 574 */ 575 if (st->src.state > sp->src.state) 576 sfail = 5; 577 else if (st->dst.state > sp->dst.state) 578 sfail = 6; 579 } 580 if (sfail) { 581 if (pf_status.debug >= PF_DEBUG_MISC) 582 printf("pfsync: %s stale update " 583 "(%d) id: %016" PRIu64 "" 584 "creatorid: %08x\n", 585 (sfail < 7 ? "ignoring" 586 : "partial"), sfail, 587 be64toh(st->id), 588 ntohl(st->creatorid)); 589 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 590 591 if (!(sp->sync_flags & PFSTATE_STALE)) { 592 /* we have a better state, send it */ 593 if (sc->sc_mbuf != NULL && !stale) 594 pfsync_sendout(sc); 595 stale++; 596 if (!st->sync_flags) 597 pfsync_pack_state( 598 PFSYNC_ACT_UPD, st, flags); 599 } 600 continue; 601 } 602 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 603 pf_state_peer_ntoh(&sp->src, &st->src); 604 pf_state_peer_ntoh(&sp->dst, &st->dst); 605 st->expire = ntohl(sp->expire) + time_second; 606 st->timeout = sp->timeout; 607 } 608 if (stale && sc->sc_mbuf != NULL) 609 pfsync_sendout(sc); 610 splx(s); 611 break; 612 /* 613 * It's not strictly necessary for us to support the "uncompressed" 614 * delete action, but it's relatively simple and maintains consistency. 615 */ 616 case PFSYNC_ACT_DEL: 617 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 618 count * sizeof(*sp), &offp)) == NULL) { 619 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 620 return; 621 } 622 623 s = splsoftnet(); 624 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 625 i < count; i++, sp++) { 626 memcpy(&id_key.id, sp->id, sizeof(id_key.id)); 627 id_key.creatorid = sp->creatorid; 628 629 st = pf_find_state_byid(&id_key); 630 if (st == NULL) { 631 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 632 continue; 633 } 634 st->sync_flags |= PFSTATE_FROMSYNC; 635 pf_unlink_state(st); 636 } 637 splx(s); 638 break; 639 case PFSYNC_ACT_UPD_C: { 640 int update_requested = 0; 641 642 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 643 count * sizeof(*up), &offp)) == NULL) { 644 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 645 return; 646 } 647 648 s = splsoftnet(); 649 for (i = 0, up = (struct pfsync_state_upd *)(mp->m_data + offp); 650 i < count; i++, up++) { 651 /* check for invalid values */ 652 if (up->timeout >= PFTM_MAX || 653 up->src.state > PF_TCPS_PROXY_DST || 654 up->dst.state > PF_TCPS_PROXY_DST) { 655 if (pf_status.debug >= PF_DEBUG_MISC) 656 printf("pfsync_insert: " 657 "PFSYNC_ACT_UPD_C: " 658 "invalid value\n"); 659 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 660 continue; 661 } 662 663 memcpy(&id_key.id, up->id, sizeof(id_key.id)); 664 id_key.creatorid = up->creatorid; 665 666 st = pf_find_state_byid(&id_key); 667 if (st == NULL) { 668 /* We don't have this state. Ask for it. */ 669 error = pfsync_request_update(up, &src); 670 if (error == ENOMEM) { 671 splx(s); 672 goto done; 673 } 674 update_requested = 1; 675 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 676 continue; 677 } 678 sk = st->state_key; 679 sfail = 0; 680 if (sk->proto == IPPROTO_TCP) { 681 /* 682 * The state should never go backwards except 683 * for syn-proxy states. Neither should the 684 * sequence window slide backwards. 685 */ 686 if (st->src.state > up->src.state && 687 (st->src.state < PF_TCPS_PROXY_SRC || 688 up->src.state >= PF_TCPS_PROXY_SRC)) 689 sfail = 1; 690 else if (st->dst.state > up->dst.state) 691 sfail = 2; 692 else if (SEQ_GT(st->src.seqlo, 693 ntohl(up->src.seqlo))) 694 sfail = 3; 695 else if (st->dst.state >= TCPS_SYN_SENT && 696 SEQ_GT(st->dst.seqlo, ntohl(up->dst.seqlo))) 697 sfail = 4; 698 } else { 699 /* 700 * Non-TCP protocol state machine always go 701 * forwards 702 */ 703 if (st->src.state > up->src.state) 704 sfail = 5; 705 else if (st->dst.state > up->dst.state) 706 sfail = 6; 707 } 708 if (sfail) { 709 if (pf_status.debug >= PF_DEBUG_MISC) 710 printf("pfsync: ignoring stale update " 711 "(%d) id: %016" PRIu64 "" 712 "creatorid: %08x\n", sfail, 713 be64toh(st->id), 714 ntohl(st->creatorid)); 715 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 716 717 /* we have a better state, send it out */ 718 if ((!stale || update_requested) && 719 sc->sc_mbuf != NULL) { 720 pfsync_sendout(sc); 721 update_requested = 0; 722 } 723 stale++; 724 if (!st->sync_flags) 725 pfsync_pack_state(PFSYNC_ACT_UPD, st, 726 PFSYNC_FLAG_STALE); 727 continue; 728 } 729 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 730 pf_state_peer_ntoh(&up->src, &st->src); 731 pf_state_peer_ntoh(&up->dst, &st->dst); 732 st->expire = ntohl(up->expire) + time_second; 733 st->timeout = up->timeout; 734 } 735 if ((update_requested || stale) && sc->sc_mbuf) 736 pfsync_sendout(sc); 737 splx(s); 738 break; 739 } 740 case PFSYNC_ACT_DEL_C: 741 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 742 count * sizeof(*dp), &offp)) == NULL) { 743 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 744 return; 745 } 746 747 s = splsoftnet(); 748 for (i = 0, dp = (struct pfsync_state_del *)(mp->m_data + offp); 749 i < count; i++, dp++) { 750 memcpy(&id_key.id, dp->id, sizeof(id_key.id)); 751 id_key.creatorid = dp->creatorid; 752 753 st = pf_find_state_byid(&id_key); 754 if (st == NULL) { 755 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 756 continue; 757 } 758 st->sync_flags |= PFSTATE_FROMSYNC; 759 pf_unlink_state(st); 760 } 761 splx(s); 762 break; 763 case PFSYNC_ACT_INS_F: 764 case PFSYNC_ACT_DEL_F: 765 /* not implemented */ 766 break; 767 case PFSYNC_ACT_UREQ: 768 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 769 count * sizeof(*rup), &offp)) == NULL) { 770 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 771 return; 772 } 773 774 s = splsoftnet(); 775 if (sc->sc_mbuf != NULL) 776 pfsync_sendout(sc); 777 for (i = 0, 778 rup = (struct pfsync_state_upd_req *)(mp->m_data + offp); 779 i < count; i++, rup++) { 780 memcpy(&id_key.id, rup->id, sizeof(id_key.id)); 781 id_key.creatorid = rup->creatorid; 782 783 if (id_key.id == 0 && id_key.creatorid == 0) { 784 sc->sc_ureq_received = time_uptime; 785 if (sc->sc_bulk_send_next == NULL) 786 sc->sc_bulk_send_next = 787 TAILQ_FIRST(&state_list); 788 sc->sc_bulk_terminator = sc->sc_bulk_send_next; 789 if (pf_status.debug >= PF_DEBUG_MISC) 790 printf("pfsync: received " 791 "bulk update request\n"); 792 pfsync_send_bus(sc, PFSYNC_BUS_START); 793 callout_schedule(&sc->sc_bulk_tmo, 1 * hz); 794 } else { 795 st = pf_find_state_byid(&id_key); 796 if (st == NULL) { 797 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 798 continue; 799 } 800 if (!st->sync_flags) 801 pfsync_pack_state(PFSYNC_ACT_UPD, 802 st, 0); 803 } 804 } 805 if (sc->sc_mbuf != NULL) 806 pfsync_sendout(sc); 807 splx(s); 808 break; 809 case PFSYNC_ACT_BUS: 810 /* If we're not waiting for a bulk update, who cares. */ 811 if (sc->sc_ureq_sent == 0) 812 break; 813 814 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 815 sizeof(*bus), &offp)) == NULL) { 816 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 817 return; 818 } 819 bus = (struct pfsync_state_bus *)(mp->m_data + offp); 820 switch (bus->status) { 821 case PFSYNC_BUS_START: 822 callout_schedule(&sc->sc_bulkfail_tmo, 823 pf_pool_limits[PF_LIMIT_STATES].limit / 824 (PFSYNC_BULKPACKETS * sc->sc_maxcount)); 825 if (pf_status.debug >= PF_DEBUG_MISC) 826 printf("pfsync: received bulk " 827 "update start\n"); 828 break; 829 case PFSYNC_BUS_END: 830 if (time_uptime - ntohl(bus->endtime) >= 831 sc->sc_ureq_sent) { 832 /* that's it, we're happy */ 833 sc->sc_ureq_sent = 0; 834 sc->sc_bulk_tries = 0; 835 callout_stop(&sc->sc_bulkfail_tmo); 836 #if NCARP > 0 837 if (!pfsync_sync_ok) 838 carp_suppress_preempt--; 839 #endif 840 pfsync_sync_ok = 1; 841 if (pf_status.debug >= PF_DEBUG_MISC) 842 printf("pfsync: received valid " 843 "bulk update end\n"); 844 } else { 845 if (pf_status.debug >= PF_DEBUG_MISC) 846 printf("pfsync: received invalid " 847 "bulk update end: bad timestamp\n"); 848 } 849 break; 850 } 851 break; 852 #ifdef IPSEC 853 case PFSYNC_ACT_TDB_UPD: 854 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 855 count * sizeof(*pt), &offp)) == NULL) { 856 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 857 return; 858 } 859 s = splsoftnet(); 860 for (i = 0, pt = (struct pfsync_tdb *)(mp->m_data + offp); 861 i < count; i++, pt++) 862 pfsync_update_net_tdb(pt); 863 splx(s); 864 break; 865 #endif 866 } 867 868 done: 869 if (m) 870 m_freem(m); 871 } 872 873 int 874 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, 875 struct rtentry *rt) 876 { 877 m_freem(m); 878 return (0); 879 } 880 881 /* ARGSUSED */ 882 int 883 pfsyncioctl(struct ifnet *ifp, u_long cmd, void* data) 884 { 885 struct lwp *l = curlwp; 886 struct pfsync_softc *sc = ifp->if_softc; 887 struct ifreq *ifr = (struct ifreq *)data; 888 struct ip_moptions *imo = &sc->sc_imo; 889 struct pfsyncreq pfsyncr; 890 struct ifnet *sifp; 891 int s, error; 892 893 switch (cmd) { 894 case SIOCSIFADDR: 895 case SIOCAIFADDR: 896 case SIOCSIFDSTADDR: 897 case SIOCSIFFLAGS: 898 if (ifp->if_flags & IFF_UP) 899 ifp->if_flags |= IFF_RUNNING; 900 else 901 ifp->if_flags &= ~IFF_RUNNING; 902 break; 903 case SIOCSIFMTU: 904 if (ifr->ifr_mtu < PFSYNC_MINMTU) 905 return (EINVAL); 906 if (ifr->ifr_mtu > MCLBYTES) 907 ifr->ifr_mtu = MCLBYTES; 908 s = splnet(); 909 if (ifr->ifr_mtu < ifp->if_mtu) 910 pfsync_sendout(sc); 911 pfsync_setmtu(sc, ifr->ifr_mtu); 912 splx(s); 913 break; 914 case SIOCGETPFSYNC: 915 if ((error = kauth_authorize_network(l->l_cred, 916 KAUTH_NETWORK_INTERFACE, 917 KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, ifp, (void *)cmd, 918 NULL)) != 0) 919 return (error); 920 memset(&pfsyncr, 0, sizeof(pfsyncr)); 921 if (sc->sc_sync_ifp) 922 strlcpy(pfsyncr.pfsyncr_syncdev, 923 sc->sc_sync_ifp->if_xname, IFNAMSIZ); 924 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 925 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 926 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)))) 927 return (error); 928 break; 929 case SIOCSETPFSYNC: 930 if ((error = kauth_authorize_network(l->l_cred, 931 KAUTH_NETWORK_INTERFACE, 932 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 933 NULL)) != 0) 934 return (error); 935 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) 936 return (error); 937 938 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 939 sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP; 940 else 941 sc->sc_sync_peer.s_addr = 942 pfsyncr.pfsyncr_syncpeer.s_addr; 943 944 if (pfsyncr.pfsyncr_maxupdates > 255) 945 return (EINVAL); 946 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 947 948 if (pfsyncr.pfsyncr_syncdev[0] == 0) { 949 sc->sc_sync_ifp = NULL; 950 if (sc->sc_mbuf_net != NULL) { 951 /* Don't keep stale pfsync packets around. */ 952 s = splnet(); 953 m_freem(sc->sc_mbuf_net); 954 sc->sc_mbuf_net = NULL; 955 sc->sc_statep_net.s = NULL; 956 splx(s); 957 } 958 if (imo->imo_num_memberships > 0) { 959 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 960 imo->imo_multicast_ifp = NULL; 961 } 962 break; 963 } 964 965 if ((sifp = ifunit(pfsyncr.pfsyncr_syncdev)) == NULL) 966 return (EINVAL); 967 968 s = splnet(); 969 if (sifp->if_mtu < sc->sc_if.if_mtu || 970 (sc->sc_sync_ifp != NULL && 971 sifp->if_mtu < sc->sc_sync_ifp->if_mtu) || 972 sifp->if_mtu < MCLBYTES - sizeof(struct ip)) 973 pfsync_sendout(sc); 974 sc->sc_sync_ifp = sifp; 975 976 pfsync_setmtu(sc, sc->sc_if.if_mtu); 977 978 if (imo->imo_num_memberships > 0) { 979 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 980 imo->imo_multicast_ifp = NULL; 981 } 982 983 if (sc->sc_sync_ifp && 984 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 985 struct in_addr addr; 986 987 if (!(sc->sc_sync_ifp->if_flags & IFF_MULTICAST)) { 988 sc->sc_sync_ifp = NULL; 989 splx(s); 990 return (EADDRNOTAVAIL); 991 } 992 993 addr.s_addr = INADDR_PFSYNC_GROUP; 994 995 if ((imo->imo_membership[0] = 996 in_addmulti(&addr, sc->sc_sync_ifp)) == NULL) { 997 sc->sc_sync_ifp = NULL; 998 splx(s); 999 return (ENOBUFS); 1000 } 1001 imo->imo_num_memberships++; 1002 imo->imo_multicast_ifp = sc->sc_sync_ifp; 1003 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 1004 imo->imo_multicast_loop = 0; 1005 } 1006 1007 if (sc->sc_sync_ifp || 1008 sc->sc_sendaddr.s_addr != INADDR_PFSYNC_GROUP) { 1009 /* Request a full state table update. */ 1010 sc->sc_ureq_sent = time_uptime; 1011 #if NCARP > 0 1012 if (pfsync_sync_ok) 1013 carp_suppress_preempt ++; 1014 #endif 1015 pfsync_sync_ok = 0; 1016 if (pf_status.debug >= PF_DEBUG_MISC) 1017 printf("pfsync: requesting bulk update\n"); 1018 callout_schedule(&sc->sc_bulkfail_tmo, 5 * hz); 1019 error = pfsync_request_update(NULL, NULL); 1020 if (error == ENOMEM) { 1021 splx(s); 1022 return (ENOMEM); 1023 } 1024 pfsync_sendout(sc); 1025 } 1026 splx(s); 1027 1028 break; 1029 1030 default: 1031 return ifioctl_common(ifp, cmd, data); 1032 } 1033 1034 return (0); 1035 } 1036 1037 void 1038 pfsync_setmtu(struct pfsync_softc *sc, int mtu_req) 1039 { 1040 int mtu; 1041 1042 if (sc->sc_sync_ifp && sc->sc_sync_ifp->if_mtu < mtu_req) 1043 mtu = sc->sc_sync_ifp->if_mtu; 1044 else 1045 mtu = mtu_req; 1046 1047 sc->sc_maxcount = (mtu - sizeof(struct pfsync_header)) / 1048 sizeof(struct pfsync_state); 1049 if (sc->sc_maxcount > 254) 1050 sc->sc_maxcount = 254; 1051 sc->sc_if.if_mtu = sizeof(struct pfsync_header) + 1052 sc->sc_maxcount * sizeof(struct pfsync_state); 1053 } 1054 1055 struct mbuf * 1056 pfsync_get_mbuf(struct pfsync_softc *sc, u_int8_t action, void **sp) 1057 { 1058 struct pfsync_header *h; 1059 struct mbuf *m; 1060 int len; 1061 1062 MGETHDR(m, M_DONTWAIT, MT_DATA); 1063 if (m == NULL) { 1064 sc->sc_if.if_oerrors++; 1065 return (NULL); 1066 } 1067 1068 switch (action) { 1069 case PFSYNC_ACT_CLR: 1070 len = sizeof(struct pfsync_header) + 1071 sizeof(struct pfsync_state_clr); 1072 break; 1073 case PFSYNC_ACT_UPD_C: 1074 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd)) + 1075 sizeof(struct pfsync_header); 1076 break; 1077 case PFSYNC_ACT_DEL_C: 1078 len = (sc->sc_maxcount * sizeof(struct pfsync_state_del)) + 1079 sizeof(struct pfsync_header); 1080 break; 1081 case PFSYNC_ACT_UREQ: 1082 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd_req)) + 1083 sizeof(struct pfsync_header); 1084 break; 1085 case PFSYNC_ACT_BUS: 1086 len = sizeof(struct pfsync_header) + 1087 sizeof(struct pfsync_state_bus); 1088 break; 1089 case PFSYNC_ACT_TDB_UPD: 1090 len = (sc->sc_maxcount * sizeof(struct pfsync_tdb)) + 1091 sizeof(struct pfsync_header); 1092 break; 1093 default: 1094 len = (sc->sc_maxcount * sizeof(struct pfsync_state)) + 1095 sizeof(struct pfsync_header); 1096 break; 1097 } 1098 1099 if (len > MHLEN) { 1100 MCLGET(m, M_DONTWAIT); 1101 if ((m->m_flags & M_EXT) == 0) { 1102 m_free(m); 1103 sc->sc_if.if_oerrors++; 1104 return (NULL); 1105 } 1106 m->m_data += (MCLBYTES - len) &~ (sizeof(long) - 1); 1107 } else 1108 MH_ALIGN(m, len); 1109 1110 m->m_pkthdr.rcvif = NULL; 1111 m->m_pkthdr.len = m->m_len = sizeof(struct pfsync_header); 1112 h = mtod(m, struct pfsync_header *); 1113 h->version = PFSYNC_VERSION; 1114 h->af = 0; 1115 h->count = 0; 1116 h->action = action; 1117 if (action != PFSYNC_ACT_TDB_UPD) 1118 memcpy(&h->pf_chksum, &pf_status.pf_chksum, 1119 PF_MD5_DIGEST_LENGTH); 1120 1121 *sp = (void *)((char *)h + PFSYNC_HDRLEN); 1122 if (action == PFSYNC_ACT_TDB_UPD) 1123 callout_schedule(&sc->sc_tdb_tmo, hz); 1124 else 1125 callout_schedule(&sc->sc_tmo, hz); 1126 return (m); 1127 } 1128 1129 int 1130 pfsync_pack_state(u_int8_t action, struct pf_state *st, int flags) 1131 { 1132 struct ifnet *ifp = NULL; 1133 struct pfsync_softc *sc = pfsyncif; 1134 struct pfsync_header *h, *h_net; 1135 struct pfsync_state *sp = NULL; 1136 struct pfsync_state_upd *up = NULL; 1137 struct pfsync_state_del *dp = NULL; 1138 struct pf_state_key *sk = st->state_key; 1139 struct pf_rule *r; 1140 u_long secs; 1141 int s, ret = 0; 1142 u_int8_t i = 255, newaction = 0; 1143 1144 if (sc == NULL) 1145 return (0); 1146 ifp = &sc->sc_if; 1147 1148 /* 1149 * If a packet falls in the forest and there's nobody around to 1150 * hear, does it make a sound? 1151 */ 1152 if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL && 1153 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1154 /* Don't leave any stale pfsync packets hanging around. */ 1155 if (sc->sc_mbuf != NULL) { 1156 m_freem(sc->sc_mbuf); 1157 sc->sc_mbuf = NULL; 1158 sc->sc_statep.s = NULL; 1159 } 1160 return (0); 1161 } 1162 1163 if (action >= PFSYNC_ACT_MAX) 1164 return (EINVAL); 1165 1166 s = splnet(); 1167 if (sc->sc_mbuf == NULL) { 1168 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1169 (void *)&sc->sc_statep.s)) == NULL) { 1170 splx(s); 1171 return (ENOMEM); 1172 } 1173 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1174 } else { 1175 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1176 if (h->action != action) { 1177 pfsync_sendout(sc); 1178 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1179 (void *)&sc->sc_statep.s)) == NULL) { 1180 splx(s); 1181 return (ENOMEM); 1182 } 1183 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1184 } else { 1185 /* 1186 * If it's an update, look in the packet to see if 1187 * we already have an update for the state. 1188 */ 1189 if (action == PFSYNC_ACT_UPD && sc->sc_maxupdates) { 1190 struct pfsync_state *usp = 1191 (void *)((char *)h + PFSYNC_HDRLEN); 1192 1193 for (i = 0; i < h->count; i++) { 1194 if (!memcmp(usp->id, &st->id, 1195 PFSYNC_ID_LEN) && 1196 usp->creatorid == st->creatorid) { 1197 sp = usp; 1198 sp->updates++; 1199 break; 1200 } 1201 usp++; 1202 } 1203 } 1204 } 1205 } 1206 1207 secs = time_second; 1208 1209 st->pfsync_time = time_uptime; 1210 1211 if (sp == NULL) { 1212 /* not a "duplicate" update */ 1213 i = 255; 1214 sp = sc->sc_statep.s++; 1215 sc->sc_mbuf->m_pkthdr.len = 1216 sc->sc_mbuf->m_len += sizeof(struct pfsync_state); 1217 h->count++; 1218 memset(sp, 0, sizeof(*sp)); 1219 1220 memcpy(sp->id, &st->id, sizeof(sp->id)); 1221 sp->creatorid = st->creatorid; 1222 1223 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); 1224 pf_state_host_hton(&sk->lan, &sp->lan); 1225 pf_state_host_hton(&sk->gwy, &sp->gwy); 1226 pf_state_host_hton(&sk->ext, &sp->ext); 1227 1228 memcpy(&sp->rt_addr, &st->rt_addr, sizeof(sp->rt_addr)); 1229 1230 sp->creation = htonl(secs - st->creation); 1231 pf_state_counter_hton(st->packets[0], sp->packets[0]); 1232 pf_state_counter_hton(st->packets[1], sp->packets[1]); 1233 pf_state_counter_hton(st->bytes[0], sp->bytes[0]); 1234 pf_state_counter_hton(st->bytes[1], sp->bytes[1]); 1235 if ((r = st->rule.ptr) == NULL) 1236 sp->rule = htonl(-1); 1237 else 1238 sp->rule = htonl(r->nr); 1239 if ((r = st->anchor.ptr) == NULL) 1240 sp->anchor = htonl(-1); 1241 else 1242 sp->anchor = htonl(r->nr); 1243 sp->af = sk->af; 1244 sp->proto = sk->proto; 1245 sp->direction = sk->direction; 1246 sp->log = st->log; 1247 sp->allow_opts = st->allow_opts; 1248 sp->timeout = st->timeout; 1249 1250 if (flags & PFSYNC_FLAG_STALE) 1251 sp->sync_flags |= PFSTATE_STALE; 1252 } 1253 1254 pf_state_peer_hton(&st->src, &sp->src); 1255 pf_state_peer_hton(&st->dst, &sp->dst); 1256 1257 if (st->expire <= secs) 1258 sp->expire = htonl(0); 1259 else 1260 sp->expire = htonl(st->expire - secs); 1261 1262 /* do we need to build "compressed" actions for network transfer? */ 1263 if (sc->sc_sync_ifp && flags & PFSYNC_FLAG_COMPRESS) { 1264 switch (action) { 1265 case PFSYNC_ACT_UPD: 1266 newaction = PFSYNC_ACT_UPD_C; 1267 break; 1268 case PFSYNC_ACT_DEL: 1269 newaction = PFSYNC_ACT_DEL_C; 1270 break; 1271 default: 1272 /* by default we just send the uncompressed states */ 1273 break; 1274 } 1275 } 1276 1277 if (newaction) { 1278 if (sc->sc_mbuf_net == NULL) { 1279 if ((sc->sc_mbuf_net = pfsync_get_mbuf(sc, newaction, 1280 (void *)&sc->sc_statep_net.s)) == NULL) { 1281 splx(s); 1282 return (ENOMEM); 1283 } 1284 } 1285 h_net = mtod(sc->sc_mbuf_net, struct pfsync_header *); 1286 1287 switch (newaction) { 1288 case PFSYNC_ACT_UPD_C: 1289 if (i != 255) { 1290 up = (void *)((char *)h_net + 1291 PFSYNC_HDRLEN + (i * sizeof(*up))); 1292 up->updates++; 1293 } else { 1294 h_net->count++; 1295 sc->sc_mbuf_net->m_pkthdr.len = 1296 sc->sc_mbuf_net->m_len += sizeof(*up); 1297 up = sc->sc_statep_net.u++; 1298 1299 memset(up, 0, sizeof(*up)); 1300 memcpy(up->id, &st->id, sizeof(up->id)); 1301 up->creatorid = st->creatorid; 1302 } 1303 up->timeout = st->timeout; 1304 up->expire = sp->expire; 1305 up->src = sp->src; 1306 up->dst = sp->dst; 1307 break; 1308 case PFSYNC_ACT_DEL_C: 1309 sc->sc_mbuf_net->m_pkthdr.len = 1310 sc->sc_mbuf_net->m_len += sizeof(*dp); 1311 dp = sc->sc_statep_net.d++; 1312 h_net->count++; 1313 1314 memset(dp, 0, sizeof(*dp)); 1315 memcpy(dp->id, &st->id, sizeof(dp->id)); 1316 dp->creatorid = st->creatorid; 1317 break; 1318 } 1319 } 1320 1321 if (h->count == sc->sc_maxcount || 1322 (sc->sc_maxupdates && (sp->updates >= sc->sc_maxupdates))) 1323 ret = pfsync_sendout(sc); 1324 1325 splx(s); 1326 return (ret); 1327 } 1328 1329 /* This must be called in splnet() */ 1330 int 1331 pfsync_request_update(struct pfsync_state_upd *up, struct in_addr *src) 1332 { 1333 struct ifnet *ifp = NULL; 1334 struct pfsync_header *h; 1335 struct pfsync_softc *sc = pfsyncif; 1336 struct pfsync_state_upd_req *rup; 1337 int ret = 0; 1338 1339 if (sc == NULL) 1340 return (0); 1341 1342 ifp = &sc->sc_if; 1343 if (sc->sc_mbuf == NULL) { 1344 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1345 (void *)&sc->sc_statep.s)) == NULL) 1346 return (ENOMEM); 1347 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1348 } else { 1349 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1350 if (h->action != PFSYNC_ACT_UREQ) { 1351 pfsync_sendout(sc); 1352 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1353 (void *)&sc->sc_statep.s)) == NULL) 1354 return (ENOMEM); 1355 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1356 } 1357 } 1358 1359 if (src != NULL) 1360 sc->sc_sendaddr = *src; 1361 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*rup); 1362 h->count++; 1363 rup = sc->sc_statep.r++; 1364 memset(rup, 0, sizeof(*rup)); 1365 if (up != NULL) { 1366 memcpy(rup->id, up->id, sizeof(rup->id)); 1367 rup->creatorid = up->creatorid; 1368 } 1369 1370 if (h->count == sc->sc_maxcount) 1371 ret = pfsync_sendout(sc); 1372 1373 return (ret); 1374 } 1375 1376 int 1377 pfsync_clear_states(u_int32_t creatorid, char *ifname) 1378 { 1379 struct ifnet *ifp = NULL; 1380 struct pfsync_softc *sc = pfsyncif; 1381 struct pfsync_state_clr *cp; 1382 int s, ret; 1383 1384 if (sc == NULL) 1385 return (0); 1386 1387 ifp = &sc->sc_if; 1388 s = splnet(); 1389 if (sc->sc_mbuf != NULL) 1390 pfsync_sendout(sc); 1391 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_CLR, 1392 (void *)&sc->sc_statep.c)) == NULL) { 1393 splx(s); 1394 return (ENOMEM); 1395 } 1396 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*cp); 1397 cp = sc->sc_statep.c; 1398 cp->creatorid = creatorid; 1399 if (ifname != NULL) 1400 strlcpy(cp->ifname, ifname, IFNAMSIZ); 1401 1402 ret = (pfsync_sendout(sc)); 1403 splx(s); 1404 return (ret); 1405 } 1406 1407 void 1408 pfsync_timeout(void *v) 1409 { 1410 struct pfsync_softc *sc = v; 1411 int s; 1412 1413 s = splnet(); 1414 pfsync_sendout(sc); 1415 splx(s); 1416 } 1417 1418 void 1419 pfsync_tdb_timeout(void *v) 1420 { 1421 struct pfsync_softc *sc = v; 1422 int s; 1423 1424 s = splnet(); 1425 pfsync_tdb_sendout(sc); 1426 splx(s); 1427 } 1428 1429 /* This must be called in splnet() */ 1430 void 1431 pfsync_send_bus(struct pfsync_softc *sc, u_int8_t status) 1432 { 1433 struct pfsync_state_bus *bus; 1434 1435 if (sc->sc_mbuf != NULL) 1436 pfsync_sendout(sc); 1437 1438 if (pfsync_sync_ok && 1439 (sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_BUS, 1440 (void *)&sc->sc_statep.b)) != NULL) { 1441 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*bus); 1442 bus = sc->sc_statep.b; 1443 bus->creatorid = pf_status.hostid; 1444 bus->status = status; 1445 bus->endtime = htonl(time_uptime - sc->sc_ureq_received); 1446 pfsync_sendout(sc); 1447 } 1448 } 1449 1450 void 1451 pfsync_bulk_update(void *v) 1452 { 1453 struct pfsync_softc *sc = v; 1454 int s, i = 0; 1455 struct pf_state *state; 1456 1457 s = splnet(); 1458 if (sc->sc_mbuf != NULL) 1459 pfsync_sendout(sc); 1460 1461 /* 1462 * Grab at most PFSYNC_BULKPACKETS worth of states which have not 1463 * been sent since the latest request was made. 1464 */ 1465 state = sc->sc_bulk_send_next; 1466 if (state) 1467 do { 1468 /* send state update if syncable and not already sent */ 1469 if (!state->sync_flags 1470 && state->timeout < PFTM_MAX 1471 && state->pfsync_time <= sc->sc_ureq_received) { 1472 pfsync_pack_state(PFSYNC_ACT_UPD, state, 0); 1473 i++; 1474 } 1475 1476 /* figure next state to send */ 1477 state = TAILQ_NEXT(state, entry_list); 1478 1479 /* wrap to start of list if we hit the end */ 1480 if (!state) 1481 state = TAILQ_FIRST(&state_list); 1482 } while (i < sc->sc_maxcount * PFSYNC_BULKPACKETS && 1483 state != sc->sc_bulk_terminator); 1484 1485 if (!state || state == sc->sc_bulk_terminator) { 1486 /* we're done */ 1487 pfsync_send_bus(sc, PFSYNC_BUS_END); 1488 sc->sc_ureq_received = 0; 1489 sc->sc_bulk_send_next = NULL; 1490 sc->sc_bulk_terminator = NULL; 1491 callout_stop(&sc->sc_bulk_tmo); 1492 if (pf_status.debug >= PF_DEBUG_MISC) 1493 printf("pfsync: bulk update complete\n"); 1494 } else { 1495 /* look again for more in a bit */ 1496 callout_schedule(&sc->sc_bulk_tmo, 1); 1497 sc->sc_bulk_send_next = state; 1498 } 1499 if (sc->sc_mbuf != NULL) 1500 pfsync_sendout(sc); 1501 splx(s); 1502 } 1503 1504 void 1505 pfsync_bulkfail(void *v) 1506 { 1507 struct pfsync_softc *sc = v; 1508 int s, error; 1509 1510 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 1511 /* Try again in a bit */ 1512 callout_schedule(&sc->sc_bulkfail_tmo, 5 * hz); 1513 s = splnet(); 1514 error = pfsync_request_update(NULL, NULL); 1515 if (error == ENOMEM) { 1516 if (pf_status.debug >= PF_DEBUG_MISC) 1517 printf("pfsync: cannot allocate mbufs for " 1518 "bulk update\n"); 1519 } else 1520 pfsync_sendout(sc); 1521 splx(s); 1522 } else { 1523 /* Pretend like the transfer was ok */ 1524 sc->sc_ureq_sent = 0; 1525 sc->sc_bulk_tries = 0; 1526 #if NCARP > 0 1527 if (!pfsync_sync_ok) 1528 carp_suppress_preempt --; 1529 #endif 1530 pfsync_sync_ok = 1; 1531 if (pf_status.debug >= PF_DEBUG_MISC) 1532 printf("pfsync: failed to receive " 1533 "bulk update status\n"); 1534 callout_stop(&sc->sc_bulkfail_tmo); 1535 } 1536 } 1537 1538 /* This must be called in splnet() */ 1539 int 1540 pfsync_sendout(struct pfsync_softc *sc) 1541 { 1542 struct ifnet *ifp = &sc->sc_if; 1543 struct mbuf *m; 1544 1545 callout_stop(&sc->sc_tmo); 1546 1547 if (sc->sc_mbuf == NULL) 1548 return (0); 1549 m = sc->sc_mbuf; 1550 sc->sc_mbuf = NULL; 1551 sc->sc_statep.s = NULL; 1552 1553 bpf_mtap(ifp, m); 1554 1555 if (sc->sc_mbuf_net) { 1556 m_freem(m); 1557 m = sc->sc_mbuf_net; 1558 sc->sc_mbuf_net = NULL; 1559 sc->sc_statep_net.s = NULL; 1560 } 1561 1562 return pfsync_sendout_mbuf(sc, m); 1563 } 1564 1565 int 1566 pfsync_tdb_sendout(struct pfsync_softc *sc) 1567 { 1568 struct ifnet *ifp = &sc->sc_if; 1569 struct mbuf *m; 1570 1571 callout_stop(&sc->sc_tdb_tmo); 1572 1573 if (sc->sc_mbuf_tdb == NULL) 1574 return (0); 1575 m = sc->sc_mbuf_tdb; 1576 sc->sc_mbuf_tdb = NULL; 1577 sc->sc_statep_tdb.t = NULL; 1578 1579 bpf_mtap(ifp, m); 1580 1581 return pfsync_sendout_mbuf(sc, m); 1582 } 1583 1584 int 1585 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m) 1586 { 1587 struct sockaddr sa; 1588 struct ip *ip; 1589 1590 if (sc->sc_sync_ifp || 1591 sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) { 1592 M_PREPEND(m, sizeof(struct ip), M_DONTWAIT); 1593 if (m == NULL) { 1594 PFSYNC_STATINC(PFSYNC_STAT_ONOMEM); 1595 return (0); 1596 } 1597 ip = mtod(m, struct ip *); 1598 ip->ip_v = IPVERSION; 1599 ip->ip_hl = sizeof(*ip) >> 2; 1600 ip->ip_tos = IPTOS_LOWDELAY; 1601 ip->ip_len = htons(m->m_pkthdr.len); 1602 ip->ip_id = htons(ip_randomid(0)); 1603 ip->ip_off = htons(IP_DF); 1604 ip->ip_ttl = PFSYNC_DFLTTL; 1605 ip->ip_p = IPPROTO_PFSYNC; 1606 ip->ip_sum = 0; 1607 1608 memset(&sa, 0, sizeof(sa)); 1609 ip->ip_src.s_addr = INADDR_ANY; 1610 1611 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP) 1612 m->m_flags |= M_MCAST; 1613 ip->ip_dst = sc->sc_sendaddr; 1614 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr; 1615 1616 PFSYNC_STATINC(PFSYNC_STAT_OPACKETS); 1617 1618 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL)) { 1619 PFSYNC_STATINC(PFSYNC_STAT_OERRORS); 1620 } 1621 } else 1622 m_freem(m); 1623 1624 return (0); 1625 } 1626 1627 #ifdef IPSEC 1628 /* Update an in-kernel tdb. Silently fail if no tdb is found. */ 1629 void 1630 pfsync_update_net_tdb(struct pfsync_tdb *pt) 1631 { 1632 struct tdb *tdb; 1633 int s; 1634 1635 /* check for invalid values */ 1636 if (ntohl(pt->spi) <= SPI_RESERVED_MAX || 1637 (pt->dst.sa.sa_family != AF_INET && 1638 pt->dst.sa.sa_family != AF_INET6)) 1639 goto bad; 1640 1641 s = spltdb(); 1642 tdb = gettdb(pt->spi, &pt->dst, pt->sproto); 1643 if (tdb) { 1644 pt->rpl = ntohl(pt->rpl); 1645 pt->cur_bytes = betoh64(pt->cur_bytes); 1646 1647 /* Neither replay nor byte counter should ever decrease. */ 1648 if (pt->rpl < tdb->tdb_rpl || 1649 pt->cur_bytes < tdb->tdb_cur_bytes) { 1650 splx(s); 1651 goto bad; 1652 } 1653 1654 tdb->tdb_rpl = pt->rpl; 1655 tdb->tdb_cur_bytes = pt->cur_bytes; 1656 } 1657 splx(s); 1658 return; 1659 1660 bad: 1661 if (pf_status.debug >= PF_DEBUG_MISC) 1662 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: " 1663 "invalid value\n"); 1664 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 1665 return; 1666 } 1667 1668 /* One of our local tdbs have been updated, need to sync rpl with others */ 1669 int 1670 pfsync_update_tdb(struct tdb *tdb, int output) 1671 { 1672 struct ifnet *ifp = NULL; 1673 struct pfsync_softc *sc = pfsyncif; 1674 struct pfsync_header *h; 1675 struct pfsync_tdb *pt = NULL; 1676 int s, i, ret; 1677 1678 if (sc == NULL) 1679 return (0); 1680 1681 ifp = &sc->sc_if; 1682 if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL && 1683 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1684 /* Don't leave any stale pfsync packets hanging around. */ 1685 if (sc->sc_mbuf_tdb != NULL) { 1686 m_freem(sc->sc_mbuf_tdb); 1687 sc->sc_mbuf_tdb = NULL; 1688 sc->sc_statep_tdb.t = NULL; 1689 } 1690 return (0); 1691 } 1692 1693 s = splnet(); 1694 if (sc->sc_mbuf_tdb == NULL) { 1695 if ((sc->sc_mbuf_tdb = pfsync_get_mbuf(sc, PFSYNC_ACT_TDB_UPD, 1696 (void *)&sc->sc_statep_tdb.t)) == NULL) { 1697 splx(s); 1698 return (ENOMEM); 1699 } 1700 h = mtod(sc->sc_mbuf_tdb, struct pfsync_header *); 1701 } else { 1702 h = mtod(sc->sc_mbuf_tdb, struct pfsync_header *); 1703 if (h->action != PFSYNC_ACT_TDB_UPD) { 1704 /* 1705 * XXX will never happen as long as there's 1706 * only one "TDB action". 1707 */ 1708 pfsync_tdb_sendout(sc); 1709 sc->sc_mbuf_tdb = pfsync_get_mbuf(sc, 1710 PFSYNC_ACT_TDB_UPD, (void *)&sc->sc_statep_tdb.t); 1711 if (sc->sc_mbuf_tdb == NULL) { 1712 splx(s); 1713 return (ENOMEM); 1714 } 1715 h = mtod(sc->sc_mbuf_tdb, struct pfsync_header *); 1716 } else if (sc->sc_maxupdates) { 1717 /* 1718 * If it's an update, look in the packet to see if 1719 * we already have an update for the state. 1720 */ 1721 struct pfsync_tdb *u = 1722 (void *)((char *)h + PFSYNC_HDRLEN); 1723 1724 for (i = 0; !pt && i < h->count; i++) { 1725 if (tdb->tdb_spi == u->spi && 1726 tdb->tdb_sproto == u->sproto && 1727 !bcmp(&tdb->tdb_dst, &u->dst, 1728 SA_LEN(&u->dst.sa))) { 1729 pt = u; 1730 pt->updates++; 1731 } 1732 u++; 1733 } 1734 } 1735 } 1736 1737 if (pt == NULL) { 1738 /* not a "duplicate" update */ 1739 pt = sc->sc_statep_tdb.t++; 1740 sc->sc_mbuf_tdb->m_pkthdr.len = 1741 sc->sc_mbuf_tdb->m_len += sizeof(struct pfsync_tdb); 1742 h->count++; 1743 memset(pt, 0, sizeof(*pt)); 1744 1745 pt->spi = tdb->tdb_spi; 1746 memcpy(&pt->dst, &tdb->tdb_dst, sizeof pt->dst); 1747 pt->sproto = tdb->tdb_sproto; 1748 } 1749 1750 /* 1751 * When a failover happens, the master's rpl is probably above 1752 * what we see here (we may be up to a second late), so 1753 * increase it a bit for outbound tdbs to manage most such 1754 * situations. 1755 * 1756 * For now, just add an offset that is likely to be larger 1757 * than the number of packets we can see in one second. The RFC 1758 * just says the next packet must have a higher seq value. 1759 * 1760 * XXX What is a good algorithm for this? We could use 1761 * a rate-determined increase, but to know it, we would have 1762 * to extend struct tdb. 1763 * XXX pt->rpl can wrap over MAXINT, but if so the real tdb 1764 * will soon be replaced anyway. For now, just don't handle 1765 * this edge case. 1766 */ 1767 #define RPL_INCR 16384 1768 pt->rpl = htonl(tdb->tdb_rpl + (output ? RPL_INCR : 0)); 1769 pt->cur_bytes = htobe64(tdb->tdb_cur_bytes); 1770 1771 if (h->count == sc->sc_maxcount || 1772 (sc->sc_maxupdates && (pt->updates >= sc->sc_maxupdates))) 1773 ret = pfsync_tdb_sendout(sc); 1774 1775 splx(s); 1776 return (ret); 1777 } 1778 #endif 1779 1780 static int 1781 sysctl_net_inet_pfsync_stats(SYSCTLFN_ARGS) 1782 { 1783 1784 return (NETSTAT_SYSCTL(pfsyncstat_percpu, PFSYNC_NSTATS)); 1785 } 1786 1787 SYSCTL_SETUP(sysctl_net_inet_pfsync_setup, "sysctl net.inet.pfsync subtree setup") 1788 { 1789 1790 sysctl_createv(clog, 0, NULL, NULL, 1791 CTLFLAG_PERMANENT, 1792 CTLTYPE_NODE, "net", NULL, 1793 NULL, 0, NULL, 0, 1794 CTL_NET, CTL_EOL); 1795 sysctl_createv(clog, 0, NULL, NULL, 1796 CTLFLAG_PERMANENT, 1797 CTLTYPE_NODE, "inet", NULL, 1798 NULL, 0, NULL, 0, 1799 CTL_NET, PF_INET, CTL_EOL); 1800 sysctl_createv(clog, 0, NULL, NULL, 1801 CTLFLAG_PERMANENT, 1802 CTLTYPE_NODE, "pfsync", 1803 SYSCTL_DESCR("pfsync related settings"), 1804 NULL, 0, NULL, 0, 1805 CTL_NET, PF_INET, IPPROTO_PFSYNC, CTL_EOL); 1806 sysctl_createv(clog, 0, NULL, NULL, 1807 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1808 CTLTYPE_STRUCT, "stats", 1809 SYSCTL_DESCR("pfsync statistics"), 1810 sysctl_net_inet_pfsync_stats, 0, NULL, 0, 1811 CTL_NET, PF_INET, IPPROTO_PFSYNC, 1812 CTL_CREATE, CTL_EOL); 1813 } 1814