1 /* $NetBSD: if_pfsync.c,v 1.5 2010/01/23 01:17:23 minskim 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 1.5 2010/01/23 01:17:23 minskim 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_ops->bpf_attach(&pfsyncif->sc_if, DLT_PFSYNC, PFSYNC_HDRLEN, 195 &pfsyncif->sc_if.if_bpf); 196 197 return (0); 198 } 199 200 int 201 pfsync_clone_destroy(struct ifnet *ifp) 202 { 203 bpf_ops->bpf_detach(ifp); 204 if_detach(ifp); 205 free(pfsyncif, M_DEVBUF); 206 pfsyncif = NULL; 207 return (0); 208 } 209 210 /* 211 * Start output on the pfsync interface. 212 */ 213 void 214 pfsyncstart(struct ifnet *ifp) 215 { 216 struct mbuf *m; 217 int s; 218 219 for (;;) { 220 s = splnet(); 221 IF_DROP(&ifp->if_snd); 222 IF_DEQUEUE(&ifp->if_snd, m); 223 splx(s); 224 225 if (m == NULL) 226 return; 227 else 228 m_freem(m); 229 } 230 } 231 232 int 233 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 234 struct pf_state_peer *d) 235 { 236 if (s->scrub.scrub_flag && d->scrub == NULL) { 237 d->scrub = pool_get(&pf_state_scrub_pl, PR_NOWAIT); 238 if (d->scrub == NULL) 239 return (ENOMEM); 240 memset(d->scrub, 0, sizeof(*d->scrub)); 241 } 242 243 return (0); 244 } 245 246 int 247 pfsync_insert_net_state(struct pfsync_state *sp, u_int8_t chksum_flag) 248 { 249 struct pf_state *st = NULL; 250 struct pf_state_key *sk = NULL; 251 struct pf_rule *r = NULL; 252 struct pfi_kif *kif; 253 254 if (sp->creatorid == 0 && pf_status.debug >= PF_DEBUG_MISC) { 255 printf("pfsync_insert_net_state: invalid creator id:" 256 " %08x\n", ntohl(sp->creatorid)); 257 return (EINVAL); 258 } 259 260 kif = pfi_kif_get(sp->ifname); 261 if (kif == NULL) { 262 if (pf_status.debug >= PF_DEBUG_MISC) 263 printf("pfsync_insert_net_state: " 264 "unknown interface: %s\n", sp->ifname); 265 /* skip this state */ 266 return (0); 267 } 268 269 /* 270 * If the ruleset checksums match, it's safe to associate the state 271 * with the rule of that number. 272 */ 273 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && chksum_flag && 274 ntohl(sp->rule) < 275 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 276 r = pf_main_ruleset.rules[ 277 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 278 else 279 r = &pf_default_rule; 280 281 if (!r->max_states || r->states < r->max_states) 282 st = pool_get(&pf_state_pl, PR_NOWAIT); 283 if (st == NULL) { 284 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 285 return (ENOMEM); 286 } 287 memset(st, 0, sizeof(*st)); 288 289 if ((sk = pf_alloc_state_key(st)) == NULL) { 290 pool_put(&pf_state_pl, st); 291 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 292 return (ENOMEM); 293 } 294 295 /* allocate memory for scrub info */ 296 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 297 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) { 298 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 299 if (st->src.scrub) 300 pool_put(&pf_state_scrub_pl, st->src.scrub); 301 pool_put(&pf_state_pl, st); 302 pool_put(&pf_state_key_pl, sk); 303 return (ENOMEM); 304 } 305 306 st->rule.ptr = r; 307 /* XXX get pointers to nat_rule and anchor */ 308 309 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 310 r->states++; 311 312 /* fill in the rest of the state entry */ 313 pf_state_host_ntoh(&sp->lan, &sk->lan); 314 pf_state_host_ntoh(&sp->gwy, &sk->gwy); 315 pf_state_host_ntoh(&sp->ext, &sk->ext); 316 317 pf_state_peer_ntoh(&sp->src, &st->src); 318 pf_state_peer_ntoh(&sp->dst, &st->dst); 319 320 memcpy(&st->rt_addr, &sp->rt_addr, sizeof(st->rt_addr)); 321 st->creation = time_second - ntohl(sp->creation); 322 st->expire = ntohl(sp->expire) + time_second; 323 324 sk->af = sp->af; 325 sk->proto = sp->proto; 326 sk->direction = sp->direction; 327 st->log = sp->log; 328 st->timeout = sp->timeout; 329 st->allow_opts = sp->allow_opts; 330 331 memcpy(&st->id, sp->id, sizeof(st->id)); 332 st->creatorid = sp->creatorid; 333 st->sync_flags = PFSTATE_FROMSYNC; 334 335 if (pf_insert_state(kif, st)) { 336 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 337 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */ 338 r->states--; 339 if (st->dst.scrub) 340 pool_put(&pf_state_scrub_pl, st->dst.scrub); 341 if (st->src.scrub) 342 pool_put(&pf_state_scrub_pl, st->src.scrub); 343 pool_put(&pf_state_pl, st); 344 return (EINVAL); 345 } 346 347 return (0); 348 } 349 350 void 351 pfsync_input(struct mbuf *m, ...) 352 { 353 struct ip *ip = mtod(m, struct ip *); 354 struct pfsync_header *ph; 355 struct pfsync_softc *sc = pfsyncif; 356 struct pf_state *st; 357 struct pf_state_key *sk; 358 struct pf_state_cmp id_key; 359 struct pfsync_state *sp; 360 struct pfsync_state_upd *up; 361 struct pfsync_state_del *dp; 362 struct pfsync_state_clr *cp; 363 struct pfsync_state_upd_req *rup; 364 struct pfsync_state_bus *bus; 365 #ifdef IPSEC 366 struct pfsync_tdb *pt; 367 #endif 368 struct in_addr src; 369 struct mbuf *mp; 370 int iplen, action, error, i, s, count, offp, sfail, stale = 0; 371 u_int8_t chksum_flag = 0; 372 373 PFSYNC_STATINC(PFSYNC_STAT_IPACKETS); 374 375 /* verify that we have a sync interface configured */ 376 if (!sc || !sc->sc_sync_ifp || !pf_status.running) 377 goto done; 378 379 /* verify that the packet came in on the right interface */ 380 if (sc->sc_sync_ifp != m->m_pkthdr.rcvif) { 381 PFSYNC_STATINC(PFSYNC_STAT_BADIF); 382 goto done; 383 } 384 385 /* verify that the IP TTL is 255. */ 386 if (ip->ip_ttl != PFSYNC_DFLTTL) { 387 PFSYNC_STATINC(PFSYNC_STAT_BADTTL); 388 goto done; 389 } 390 391 iplen = ip->ip_hl << 2; 392 393 if (m->m_pkthdr.len < iplen + sizeof(*ph)) { 394 PFSYNC_STATINC(PFSYNC_STAT_HDROPS); 395 goto done; 396 } 397 398 if (iplen + sizeof(*ph) > m->m_len) { 399 if ((m = m_pullup(m, iplen + sizeof(*ph))) == NULL) { 400 PFSYNC_STATINC(PFSYNC_STAT_HDROPS); 401 goto done; 402 } 403 ip = mtod(m, struct ip *); 404 } 405 ph = (struct pfsync_header *)((char *)ip + iplen); 406 407 /* verify the version */ 408 if (ph->version != PFSYNC_VERSION) { 409 PFSYNC_STATINC(PFSYNC_STAT_BADVER); 410 goto done; 411 } 412 413 action = ph->action; 414 count = ph->count; 415 416 /* make sure it's a valid action code */ 417 if (action >= PFSYNC_ACT_MAX) { 418 PFSYNC_STATINC(PFSYNC_STAT_BADACT); 419 goto done; 420 } 421 422 /* Cheaper to grab this now than having to mess with mbufs later */ 423 src = ip->ip_src; 424 425 if (!bcmp(&ph->pf_chksum, &pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 426 chksum_flag++; 427 428 switch (action) { 429 case PFSYNC_ACT_CLR: { 430 struct pf_state *nexts; 431 struct pf_state_key *nextsk; 432 struct pfi_kif *kif; 433 u_int32_t creatorid; 434 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 435 sizeof(*cp), &offp)) == NULL) { 436 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 437 return; 438 } 439 cp = (struct pfsync_state_clr *)(mp->m_data + offp); 440 creatorid = cp->creatorid; 441 442 s = splsoftnet(); 443 if (cp->ifname[0] == '\0') { 444 for (st = RB_MIN(pf_state_tree_id, &tree_id); 445 st; st = nexts) { 446 nexts = RB_NEXT(pf_state_tree_id, &tree_id, st); 447 if (st->creatorid == creatorid) { 448 st->sync_flags |= PFSTATE_FROMSYNC; 449 pf_unlink_state(st); 450 } 451 } 452 } else { 453 if ((kif = pfi_kif_get(cp->ifname)) == NULL) { 454 splx(s); 455 return; 456 } 457 for (sk = RB_MIN(pf_state_tree_lan_ext, 458 &pf_statetbl_lan_ext); sk; sk = nextsk) { 459 nextsk = RB_NEXT(pf_state_tree_lan_ext, 460 &pf_statetbl_lan_ext, sk); 461 TAILQ_FOREACH(st, &sk->states, next) { 462 if (st->creatorid == creatorid) { 463 st->sync_flags |= 464 PFSTATE_FROMSYNC; 465 pf_unlink_state(st); 466 } 467 } 468 } 469 } 470 splx(s); 471 472 break; 473 } 474 case PFSYNC_ACT_INS: 475 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 476 count * sizeof(*sp), &offp)) == NULL) { 477 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 478 return; 479 } 480 481 s = splsoftnet(); 482 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 483 i < count; i++, sp++) { 484 /* check for invalid values */ 485 if (sp->timeout >= PFTM_MAX || 486 sp->src.state > PF_TCPS_PROXY_DST || 487 sp->dst.state > PF_TCPS_PROXY_DST || 488 sp->direction > PF_OUT || 489 (sp->af != AF_INET && sp->af != AF_INET6)) { 490 if (pf_status.debug >= PF_DEBUG_MISC) 491 printf("pfsync_insert: PFSYNC_ACT_INS: " 492 "invalid value\n"); 493 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 494 continue; 495 } 496 497 if ((error = pfsync_insert_net_state(sp, 498 chksum_flag))) { 499 if (error == ENOMEM) { 500 splx(s); 501 goto done; 502 } 503 continue; 504 } 505 } 506 splx(s); 507 break; 508 case PFSYNC_ACT_UPD: 509 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 510 count * sizeof(*sp), &offp)) == NULL) { 511 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 512 return; 513 } 514 515 s = splsoftnet(); 516 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 517 i < count; i++, sp++) { 518 int flags = PFSYNC_FLAG_STALE; 519 520 /* check for invalid values */ 521 if (sp->timeout >= PFTM_MAX || 522 sp->src.state > PF_TCPS_PROXY_DST || 523 sp->dst.state > PF_TCPS_PROXY_DST) { 524 if (pf_status.debug >= PF_DEBUG_MISC) 525 printf("pfsync_insert: PFSYNC_ACT_UPD: " 526 "invalid value\n"); 527 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 528 continue; 529 } 530 531 memcpy(&id_key.id, sp->id, sizeof(id_key.id)); 532 id_key.creatorid = sp->creatorid; 533 534 st = pf_find_state_byid(&id_key); 535 if (st == NULL) { 536 /* insert the update */ 537 if (pfsync_insert_net_state(sp, chksum_flag)) { 538 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 539 } 540 continue; 541 } 542 sk = st->state_key; 543 sfail = 0; 544 if (sk->proto == IPPROTO_TCP) { 545 /* 546 * The state should never go backwards except 547 * for syn-proxy states. Neither should the 548 * sequence window slide backwards. 549 */ 550 if (st->src.state > sp->src.state && 551 (st->src.state < PF_TCPS_PROXY_SRC || 552 sp->src.state >= PF_TCPS_PROXY_SRC)) 553 sfail = 1; 554 else if (SEQ_GT(st->src.seqlo, 555 ntohl(sp->src.seqlo))) 556 sfail = 3; 557 else if (st->dst.state > sp->dst.state) { 558 /* There might still be useful 559 * information about the src state here, 560 * so import that part of the update, 561 * then "fail" so we send the updated 562 * state back to the peer who is missing 563 * our what we know. */ 564 pf_state_peer_ntoh(&sp->src, &st->src); 565 /* XXX do anything with timeouts? */ 566 sfail = 7; 567 flags = 0; 568 } else if (st->dst.state >= TCPS_SYN_SENT && 569 SEQ_GT(st->dst.seqlo, ntohl(sp->dst.seqlo))) 570 sfail = 4; 571 } else { 572 /* 573 * Non-TCP protocol state machine always go 574 * forwards 575 */ 576 if (st->src.state > sp->src.state) 577 sfail = 5; 578 else if (st->dst.state > sp->dst.state) 579 sfail = 6; 580 } 581 if (sfail) { 582 if (pf_status.debug >= PF_DEBUG_MISC) 583 printf("pfsync: %s stale update " 584 "(%d) id: %016" PRIu64 "" 585 "creatorid: %08x\n", 586 (sfail < 7 ? "ignoring" 587 : "partial"), sfail, 588 be64toh(st->id), 589 ntohl(st->creatorid)); 590 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 591 592 if (!(sp->sync_flags & PFSTATE_STALE)) { 593 /* we have a better state, send it */ 594 if (sc->sc_mbuf != NULL && !stale) 595 pfsync_sendout(sc); 596 stale++; 597 if (!st->sync_flags) 598 pfsync_pack_state( 599 PFSYNC_ACT_UPD, st, flags); 600 } 601 continue; 602 } 603 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 604 pf_state_peer_ntoh(&sp->src, &st->src); 605 pf_state_peer_ntoh(&sp->dst, &st->dst); 606 st->expire = ntohl(sp->expire) + time_second; 607 st->timeout = sp->timeout; 608 } 609 if (stale && sc->sc_mbuf != NULL) 610 pfsync_sendout(sc); 611 splx(s); 612 break; 613 /* 614 * It's not strictly necessary for us to support the "uncompressed" 615 * delete action, but it's relatively simple and maintains consistency. 616 */ 617 case PFSYNC_ACT_DEL: 618 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 619 count * sizeof(*sp), &offp)) == NULL) { 620 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 621 return; 622 } 623 624 s = splsoftnet(); 625 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 626 i < count; i++, sp++) { 627 memcpy(&id_key.id, sp->id, sizeof(id_key.id)); 628 id_key.creatorid = sp->creatorid; 629 630 st = pf_find_state_byid(&id_key); 631 if (st == NULL) { 632 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 633 continue; 634 } 635 st->sync_flags |= PFSTATE_FROMSYNC; 636 pf_unlink_state(st); 637 } 638 splx(s); 639 break; 640 case PFSYNC_ACT_UPD_C: { 641 int update_requested = 0; 642 643 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 644 count * sizeof(*up), &offp)) == NULL) { 645 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 646 return; 647 } 648 649 s = splsoftnet(); 650 for (i = 0, up = (struct pfsync_state_upd *)(mp->m_data + offp); 651 i < count; i++, up++) { 652 /* check for invalid values */ 653 if (up->timeout >= PFTM_MAX || 654 up->src.state > PF_TCPS_PROXY_DST || 655 up->dst.state > PF_TCPS_PROXY_DST) { 656 if (pf_status.debug >= PF_DEBUG_MISC) 657 printf("pfsync_insert: " 658 "PFSYNC_ACT_UPD_C: " 659 "invalid value\n"); 660 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 661 continue; 662 } 663 664 memcpy(&id_key.id, up->id, sizeof(id_key.id)); 665 id_key.creatorid = up->creatorid; 666 667 st = pf_find_state_byid(&id_key); 668 if (st == NULL) { 669 /* We don't have this state. Ask for it. */ 670 error = pfsync_request_update(up, &src); 671 if (error == ENOMEM) { 672 splx(s); 673 goto done; 674 } 675 update_requested = 1; 676 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 677 continue; 678 } 679 sk = st->state_key; 680 sfail = 0; 681 if (sk->proto == IPPROTO_TCP) { 682 /* 683 * The state should never go backwards except 684 * for syn-proxy states. Neither should the 685 * sequence window slide backwards. 686 */ 687 if (st->src.state > up->src.state && 688 (st->src.state < PF_TCPS_PROXY_SRC || 689 up->src.state >= PF_TCPS_PROXY_SRC)) 690 sfail = 1; 691 else if (st->dst.state > up->dst.state) 692 sfail = 2; 693 else if (SEQ_GT(st->src.seqlo, 694 ntohl(up->src.seqlo))) 695 sfail = 3; 696 else if (st->dst.state >= TCPS_SYN_SENT && 697 SEQ_GT(st->dst.seqlo, ntohl(up->dst.seqlo))) 698 sfail = 4; 699 } else { 700 /* 701 * Non-TCP protocol state machine always go 702 * forwards 703 */ 704 if (st->src.state > up->src.state) 705 sfail = 5; 706 else if (st->dst.state > up->dst.state) 707 sfail = 6; 708 } 709 if (sfail) { 710 if (pf_status.debug >= PF_DEBUG_MISC) 711 printf("pfsync: ignoring stale update " 712 "(%d) id: %016" PRIu64 "" 713 "creatorid: %08x\n", sfail, 714 be64toh(st->id), 715 ntohl(st->creatorid)); 716 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 717 718 /* we have a better state, send it out */ 719 if ((!stale || update_requested) && 720 sc->sc_mbuf != NULL) { 721 pfsync_sendout(sc); 722 update_requested = 0; 723 } 724 stale++; 725 if (!st->sync_flags) 726 pfsync_pack_state(PFSYNC_ACT_UPD, st, 727 PFSYNC_FLAG_STALE); 728 continue; 729 } 730 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 731 pf_state_peer_ntoh(&up->src, &st->src); 732 pf_state_peer_ntoh(&up->dst, &st->dst); 733 st->expire = ntohl(up->expire) + time_second; 734 st->timeout = up->timeout; 735 } 736 if ((update_requested || stale) && sc->sc_mbuf) 737 pfsync_sendout(sc); 738 splx(s); 739 break; 740 } 741 case PFSYNC_ACT_DEL_C: 742 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 743 count * sizeof(*dp), &offp)) == NULL) { 744 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 745 return; 746 } 747 748 s = splsoftnet(); 749 for (i = 0, dp = (struct pfsync_state_del *)(mp->m_data + offp); 750 i < count; i++, dp++) { 751 memcpy(&id_key.id, dp->id, sizeof(id_key.id)); 752 id_key.creatorid = dp->creatorid; 753 754 st = pf_find_state_byid(&id_key); 755 if (st == NULL) { 756 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 757 continue; 758 } 759 st->sync_flags |= PFSTATE_FROMSYNC; 760 pf_unlink_state(st); 761 } 762 splx(s); 763 break; 764 case PFSYNC_ACT_INS_F: 765 case PFSYNC_ACT_DEL_F: 766 /* not implemented */ 767 break; 768 case PFSYNC_ACT_UREQ: 769 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 770 count * sizeof(*rup), &offp)) == NULL) { 771 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 772 return; 773 } 774 775 s = splsoftnet(); 776 if (sc->sc_mbuf != NULL) 777 pfsync_sendout(sc); 778 for (i = 0, 779 rup = (struct pfsync_state_upd_req *)(mp->m_data + offp); 780 i < count; i++, rup++) { 781 memcpy(&id_key.id, rup->id, sizeof(id_key.id)); 782 id_key.creatorid = rup->creatorid; 783 784 if (id_key.id == 0 && id_key.creatorid == 0) { 785 sc->sc_ureq_received = time_uptime; 786 if (sc->sc_bulk_send_next == NULL) 787 sc->sc_bulk_send_next = 788 TAILQ_FIRST(&state_list); 789 sc->sc_bulk_terminator = sc->sc_bulk_send_next; 790 if (pf_status.debug >= PF_DEBUG_MISC) 791 printf("pfsync: received " 792 "bulk update request\n"); 793 pfsync_send_bus(sc, PFSYNC_BUS_START); 794 callout_schedule(&sc->sc_bulk_tmo, 1 * hz); 795 } else { 796 st = pf_find_state_byid(&id_key); 797 if (st == NULL) { 798 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 799 continue; 800 } 801 if (!st->sync_flags) 802 pfsync_pack_state(PFSYNC_ACT_UPD, 803 st, 0); 804 } 805 } 806 if (sc->sc_mbuf != NULL) 807 pfsync_sendout(sc); 808 splx(s); 809 break; 810 case PFSYNC_ACT_BUS: 811 /* If we're not waiting for a bulk update, who cares. */ 812 if (sc->sc_ureq_sent == 0) 813 break; 814 815 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 816 sizeof(*bus), &offp)) == NULL) { 817 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 818 return; 819 } 820 bus = (struct pfsync_state_bus *)(mp->m_data + offp); 821 switch (bus->status) { 822 case PFSYNC_BUS_START: 823 callout_schedule(&sc->sc_bulkfail_tmo, 824 pf_pool_limits[PF_LIMIT_STATES].limit / 825 (PFSYNC_BULKPACKETS * sc->sc_maxcount)); 826 if (pf_status.debug >= PF_DEBUG_MISC) 827 printf("pfsync: received bulk " 828 "update start\n"); 829 break; 830 case PFSYNC_BUS_END: 831 if (time_uptime - ntohl(bus->endtime) >= 832 sc->sc_ureq_sent) { 833 /* that's it, we're happy */ 834 sc->sc_ureq_sent = 0; 835 sc->sc_bulk_tries = 0; 836 callout_stop(&sc->sc_bulkfail_tmo); 837 #if NCARP > 0 838 if (!pfsync_sync_ok) 839 carp_suppress_preempt--; 840 #endif 841 pfsync_sync_ok = 1; 842 if (pf_status.debug >= PF_DEBUG_MISC) 843 printf("pfsync: received valid " 844 "bulk update end\n"); 845 } else { 846 if (pf_status.debug >= PF_DEBUG_MISC) 847 printf("pfsync: received invalid " 848 "bulk update end: bad timestamp\n"); 849 } 850 break; 851 } 852 break; 853 #ifdef IPSEC 854 case PFSYNC_ACT_TDB_UPD: 855 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 856 count * sizeof(*pt), &offp)) == NULL) { 857 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 858 return; 859 } 860 s = splsoftnet(); 861 for (i = 0, pt = (struct pfsync_tdb *)(mp->m_data + offp); 862 i < count; i++, pt++) 863 pfsync_update_net_tdb(pt); 864 splx(s); 865 break; 866 #endif 867 } 868 869 done: 870 if (m) 871 m_freem(m); 872 } 873 874 int 875 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, 876 struct rtentry *rt) 877 { 878 m_freem(m); 879 return (0); 880 } 881 882 /* ARGSUSED */ 883 int 884 pfsyncioctl(struct ifnet *ifp, u_long cmd, void* data) 885 { 886 struct lwp *l = curlwp; 887 struct pfsync_softc *sc = ifp->if_softc; 888 struct ifreq *ifr = (struct ifreq *)data; 889 struct ip_moptions *imo = &sc->sc_imo; 890 struct pfsyncreq pfsyncr; 891 struct ifnet *sifp; 892 int s, error; 893 894 switch (cmd) { 895 case SIOCSIFADDR: 896 case SIOCAIFADDR: 897 case SIOCSIFDSTADDR: 898 case SIOCSIFFLAGS: 899 if (ifp->if_flags & IFF_UP) 900 ifp->if_flags |= IFF_RUNNING; 901 else 902 ifp->if_flags &= ~IFF_RUNNING; 903 break; 904 case SIOCSIFMTU: 905 if (ifr->ifr_mtu < PFSYNC_MINMTU) 906 return (EINVAL); 907 if (ifr->ifr_mtu > MCLBYTES) 908 ifr->ifr_mtu = MCLBYTES; 909 s = splnet(); 910 if (ifr->ifr_mtu < ifp->if_mtu) 911 pfsync_sendout(sc); 912 pfsync_setmtu(sc, ifr->ifr_mtu); 913 splx(s); 914 break; 915 case SIOCGETPFSYNC: 916 if ((error = kauth_authorize_network(l->l_cred, 917 KAUTH_NETWORK_INTERFACE, 918 KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, ifp, (void *)cmd, 919 NULL)) != 0) 920 return (error); 921 memset(&pfsyncr, 0, sizeof(pfsyncr)); 922 if (sc->sc_sync_ifp) 923 strlcpy(pfsyncr.pfsyncr_syncdev, 924 sc->sc_sync_ifp->if_xname, IFNAMSIZ); 925 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 926 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 927 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)))) 928 return (error); 929 break; 930 case SIOCSETPFSYNC: 931 if ((error = kauth_authorize_network(l->l_cred, 932 KAUTH_NETWORK_INTERFACE, 933 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 934 NULL)) != 0) 935 return (error); 936 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) 937 return (error); 938 939 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 940 sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP; 941 else 942 sc->sc_sync_peer.s_addr = 943 pfsyncr.pfsyncr_syncpeer.s_addr; 944 945 if (pfsyncr.pfsyncr_maxupdates > 255) 946 return (EINVAL); 947 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 948 949 if (pfsyncr.pfsyncr_syncdev[0] == 0) { 950 sc->sc_sync_ifp = NULL; 951 if (sc->sc_mbuf_net != NULL) { 952 /* Don't keep stale pfsync packets around. */ 953 s = splnet(); 954 m_freem(sc->sc_mbuf_net); 955 sc->sc_mbuf_net = NULL; 956 sc->sc_statep_net.s = NULL; 957 splx(s); 958 } 959 if (imo->imo_num_memberships > 0) { 960 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 961 imo->imo_multicast_ifp = NULL; 962 } 963 break; 964 } 965 966 if ((sifp = ifunit(pfsyncr.pfsyncr_syncdev)) == NULL) 967 return (EINVAL); 968 969 s = splnet(); 970 if (sifp->if_mtu < sc->sc_if.if_mtu || 971 (sc->sc_sync_ifp != NULL && 972 sifp->if_mtu < sc->sc_sync_ifp->if_mtu) || 973 sifp->if_mtu < MCLBYTES - sizeof(struct ip)) 974 pfsync_sendout(sc); 975 sc->sc_sync_ifp = sifp; 976 977 pfsync_setmtu(sc, sc->sc_if.if_mtu); 978 979 if (imo->imo_num_memberships > 0) { 980 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 981 imo->imo_multicast_ifp = NULL; 982 } 983 984 if (sc->sc_sync_ifp && 985 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 986 struct in_addr addr; 987 988 if (!(sc->sc_sync_ifp->if_flags & IFF_MULTICAST)) { 989 sc->sc_sync_ifp = NULL; 990 splx(s); 991 return (EADDRNOTAVAIL); 992 } 993 994 addr.s_addr = INADDR_PFSYNC_GROUP; 995 996 if ((imo->imo_membership[0] = 997 in_addmulti(&addr, sc->sc_sync_ifp)) == NULL) { 998 sc->sc_sync_ifp = NULL; 999 splx(s); 1000 return (ENOBUFS); 1001 } 1002 imo->imo_num_memberships++; 1003 imo->imo_multicast_ifp = sc->sc_sync_ifp; 1004 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 1005 imo->imo_multicast_loop = 0; 1006 } 1007 1008 if (sc->sc_sync_ifp || 1009 sc->sc_sendaddr.s_addr != INADDR_PFSYNC_GROUP) { 1010 /* Request a full state table update. */ 1011 sc->sc_ureq_sent = time_uptime; 1012 #if NCARP > 0 1013 if (pfsync_sync_ok) 1014 carp_suppress_preempt ++; 1015 #endif 1016 pfsync_sync_ok = 0; 1017 if (pf_status.debug >= PF_DEBUG_MISC) 1018 printf("pfsync: requesting bulk update\n"); 1019 callout_schedule(&sc->sc_bulkfail_tmo, 5 * hz); 1020 error = pfsync_request_update(NULL, NULL); 1021 if (error == ENOMEM) { 1022 splx(s); 1023 return (ENOMEM); 1024 } 1025 pfsync_sendout(sc); 1026 } 1027 splx(s); 1028 1029 break; 1030 1031 default: 1032 return ifioctl_common(ifp, cmd, data); 1033 } 1034 1035 return (0); 1036 } 1037 1038 void 1039 pfsync_setmtu(struct pfsync_softc *sc, int mtu_req) 1040 { 1041 int mtu; 1042 1043 if (sc->sc_sync_ifp && sc->sc_sync_ifp->if_mtu < mtu_req) 1044 mtu = sc->sc_sync_ifp->if_mtu; 1045 else 1046 mtu = mtu_req; 1047 1048 sc->sc_maxcount = (mtu - sizeof(struct pfsync_header)) / 1049 sizeof(struct pfsync_state); 1050 if (sc->sc_maxcount > 254) 1051 sc->sc_maxcount = 254; 1052 sc->sc_if.if_mtu = sizeof(struct pfsync_header) + 1053 sc->sc_maxcount * sizeof(struct pfsync_state); 1054 } 1055 1056 struct mbuf * 1057 pfsync_get_mbuf(struct pfsync_softc *sc, u_int8_t action, void **sp) 1058 { 1059 struct pfsync_header *h; 1060 struct mbuf *m; 1061 int len; 1062 1063 MGETHDR(m, M_DONTWAIT, MT_DATA); 1064 if (m == NULL) { 1065 sc->sc_if.if_oerrors++; 1066 return (NULL); 1067 } 1068 1069 switch (action) { 1070 case PFSYNC_ACT_CLR: 1071 len = sizeof(struct pfsync_header) + 1072 sizeof(struct pfsync_state_clr); 1073 break; 1074 case PFSYNC_ACT_UPD_C: 1075 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd)) + 1076 sizeof(struct pfsync_header); 1077 break; 1078 case PFSYNC_ACT_DEL_C: 1079 len = (sc->sc_maxcount * sizeof(struct pfsync_state_del)) + 1080 sizeof(struct pfsync_header); 1081 break; 1082 case PFSYNC_ACT_UREQ: 1083 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd_req)) + 1084 sizeof(struct pfsync_header); 1085 break; 1086 case PFSYNC_ACT_BUS: 1087 len = sizeof(struct pfsync_header) + 1088 sizeof(struct pfsync_state_bus); 1089 break; 1090 case PFSYNC_ACT_TDB_UPD: 1091 len = (sc->sc_maxcount * sizeof(struct pfsync_tdb)) + 1092 sizeof(struct pfsync_header); 1093 break; 1094 default: 1095 len = (sc->sc_maxcount * sizeof(struct pfsync_state)) + 1096 sizeof(struct pfsync_header); 1097 break; 1098 } 1099 1100 if (len > MHLEN) { 1101 MCLGET(m, M_DONTWAIT); 1102 if ((m->m_flags & M_EXT) == 0) { 1103 m_free(m); 1104 sc->sc_if.if_oerrors++; 1105 return (NULL); 1106 } 1107 m->m_data += (MCLBYTES - len) &~ (sizeof(long) - 1); 1108 } else 1109 MH_ALIGN(m, len); 1110 1111 m->m_pkthdr.rcvif = NULL; 1112 m->m_pkthdr.len = m->m_len = sizeof(struct pfsync_header); 1113 h = mtod(m, struct pfsync_header *); 1114 h->version = PFSYNC_VERSION; 1115 h->af = 0; 1116 h->count = 0; 1117 h->action = action; 1118 if (action != PFSYNC_ACT_TDB_UPD) 1119 memcpy(&h->pf_chksum, &pf_status.pf_chksum, 1120 PF_MD5_DIGEST_LENGTH); 1121 1122 *sp = (void *)((char *)h + PFSYNC_HDRLEN); 1123 if (action == PFSYNC_ACT_TDB_UPD) 1124 callout_schedule(&sc->sc_tdb_tmo, hz); 1125 else 1126 callout_schedule(&sc->sc_tmo, hz); 1127 return (m); 1128 } 1129 1130 int 1131 pfsync_pack_state(u_int8_t action, struct pf_state *st, int flags) 1132 { 1133 struct ifnet *ifp = NULL; 1134 struct pfsync_softc *sc = pfsyncif; 1135 struct pfsync_header *h, *h_net; 1136 struct pfsync_state *sp = NULL; 1137 struct pfsync_state_upd *up = NULL; 1138 struct pfsync_state_del *dp = NULL; 1139 struct pf_state_key *sk = st->state_key; 1140 struct pf_rule *r; 1141 u_long secs; 1142 int s, ret = 0; 1143 u_int8_t i = 255, newaction = 0; 1144 1145 if (sc == NULL) 1146 return (0); 1147 ifp = &sc->sc_if; 1148 1149 /* 1150 * If a packet falls in the forest and there's nobody around to 1151 * hear, does it make a sound? 1152 */ 1153 if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL && 1154 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1155 /* Don't leave any stale pfsync packets hanging around. */ 1156 if (sc->sc_mbuf != NULL) { 1157 m_freem(sc->sc_mbuf); 1158 sc->sc_mbuf = NULL; 1159 sc->sc_statep.s = NULL; 1160 } 1161 return (0); 1162 } 1163 1164 if (action >= PFSYNC_ACT_MAX) 1165 return (EINVAL); 1166 1167 s = splnet(); 1168 if (sc->sc_mbuf == NULL) { 1169 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1170 (void *)&sc->sc_statep.s)) == NULL) { 1171 splx(s); 1172 return (ENOMEM); 1173 } 1174 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1175 } else { 1176 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1177 if (h->action != action) { 1178 pfsync_sendout(sc); 1179 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1180 (void *)&sc->sc_statep.s)) == NULL) { 1181 splx(s); 1182 return (ENOMEM); 1183 } 1184 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1185 } else { 1186 /* 1187 * If it's an update, look in the packet to see if 1188 * we already have an update for the state. 1189 */ 1190 if (action == PFSYNC_ACT_UPD && sc->sc_maxupdates) { 1191 struct pfsync_state *usp = 1192 (void *)((char *)h + PFSYNC_HDRLEN); 1193 1194 for (i = 0; i < h->count; i++) { 1195 if (!memcmp(usp->id, &st->id, 1196 PFSYNC_ID_LEN) && 1197 usp->creatorid == st->creatorid) { 1198 sp = usp; 1199 sp->updates++; 1200 break; 1201 } 1202 usp++; 1203 } 1204 } 1205 } 1206 } 1207 1208 secs = time_second; 1209 1210 st->pfsync_time = time_uptime; 1211 1212 if (sp == NULL) { 1213 /* not a "duplicate" update */ 1214 i = 255; 1215 sp = sc->sc_statep.s++; 1216 sc->sc_mbuf->m_pkthdr.len = 1217 sc->sc_mbuf->m_len += sizeof(struct pfsync_state); 1218 h->count++; 1219 memset(sp, 0, sizeof(*sp)); 1220 1221 memcpy(sp->id, &st->id, sizeof(sp->id)); 1222 sp->creatorid = st->creatorid; 1223 1224 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); 1225 pf_state_host_hton(&sk->lan, &sp->lan); 1226 pf_state_host_hton(&sk->gwy, &sp->gwy); 1227 pf_state_host_hton(&sk->ext, &sp->ext); 1228 1229 memcpy(&sp->rt_addr, &st->rt_addr, sizeof(sp->rt_addr)); 1230 1231 sp->creation = htonl(secs - st->creation); 1232 pf_state_counter_hton(st->packets[0], sp->packets[0]); 1233 pf_state_counter_hton(st->packets[1], sp->packets[1]); 1234 pf_state_counter_hton(st->bytes[0], sp->bytes[0]); 1235 pf_state_counter_hton(st->bytes[1], sp->bytes[1]); 1236 if ((r = st->rule.ptr) == NULL) 1237 sp->rule = htonl(-1); 1238 else 1239 sp->rule = htonl(r->nr); 1240 if ((r = st->anchor.ptr) == NULL) 1241 sp->anchor = htonl(-1); 1242 else 1243 sp->anchor = htonl(r->nr); 1244 sp->af = sk->af; 1245 sp->proto = sk->proto; 1246 sp->direction = sk->direction; 1247 sp->log = st->log; 1248 sp->allow_opts = st->allow_opts; 1249 sp->timeout = st->timeout; 1250 1251 if (flags & PFSYNC_FLAG_STALE) 1252 sp->sync_flags |= PFSTATE_STALE; 1253 } 1254 1255 pf_state_peer_hton(&st->src, &sp->src); 1256 pf_state_peer_hton(&st->dst, &sp->dst); 1257 1258 if (st->expire <= secs) 1259 sp->expire = htonl(0); 1260 else 1261 sp->expire = htonl(st->expire - secs); 1262 1263 /* do we need to build "compressed" actions for network transfer? */ 1264 if (sc->sc_sync_ifp && flags & PFSYNC_FLAG_COMPRESS) { 1265 switch (action) { 1266 case PFSYNC_ACT_UPD: 1267 newaction = PFSYNC_ACT_UPD_C; 1268 break; 1269 case PFSYNC_ACT_DEL: 1270 newaction = PFSYNC_ACT_DEL_C; 1271 break; 1272 default: 1273 /* by default we just send the uncompressed states */ 1274 break; 1275 } 1276 } 1277 1278 if (newaction) { 1279 if (sc->sc_mbuf_net == NULL) { 1280 if ((sc->sc_mbuf_net = pfsync_get_mbuf(sc, newaction, 1281 (void *)&sc->sc_statep_net.s)) == NULL) { 1282 splx(s); 1283 return (ENOMEM); 1284 } 1285 } 1286 h_net = mtod(sc->sc_mbuf_net, struct pfsync_header *); 1287 1288 switch (newaction) { 1289 case PFSYNC_ACT_UPD_C: 1290 if (i != 255) { 1291 up = (void *)((char *)h_net + 1292 PFSYNC_HDRLEN + (i * sizeof(*up))); 1293 up->updates++; 1294 } else { 1295 h_net->count++; 1296 sc->sc_mbuf_net->m_pkthdr.len = 1297 sc->sc_mbuf_net->m_len += sizeof(*up); 1298 up = sc->sc_statep_net.u++; 1299 1300 memset(up, 0, sizeof(*up)); 1301 memcpy(up->id, &st->id, sizeof(up->id)); 1302 up->creatorid = st->creatorid; 1303 } 1304 up->timeout = st->timeout; 1305 up->expire = sp->expire; 1306 up->src = sp->src; 1307 up->dst = sp->dst; 1308 break; 1309 case PFSYNC_ACT_DEL_C: 1310 sc->sc_mbuf_net->m_pkthdr.len = 1311 sc->sc_mbuf_net->m_len += sizeof(*dp); 1312 dp = sc->sc_statep_net.d++; 1313 h_net->count++; 1314 1315 memset(dp, 0, sizeof(*dp)); 1316 memcpy(dp->id, &st->id, sizeof(dp->id)); 1317 dp->creatorid = st->creatorid; 1318 break; 1319 } 1320 } 1321 1322 if (h->count == sc->sc_maxcount || 1323 (sc->sc_maxupdates && (sp->updates >= sc->sc_maxupdates))) 1324 ret = pfsync_sendout(sc); 1325 1326 splx(s); 1327 return (ret); 1328 } 1329 1330 /* This must be called in splnet() */ 1331 int 1332 pfsync_request_update(struct pfsync_state_upd *up, struct in_addr *src) 1333 { 1334 struct ifnet *ifp = NULL; 1335 struct pfsync_header *h; 1336 struct pfsync_softc *sc = pfsyncif; 1337 struct pfsync_state_upd_req *rup; 1338 int ret = 0; 1339 1340 if (sc == NULL) 1341 return (0); 1342 1343 ifp = &sc->sc_if; 1344 if (sc->sc_mbuf == NULL) { 1345 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1346 (void *)&sc->sc_statep.s)) == NULL) 1347 return (ENOMEM); 1348 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1349 } else { 1350 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1351 if (h->action != PFSYNC_ACT_UREQ) { 1352 pfsync_sendout(sc); 1353 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1354 (void *)&sc->sc_statep.s)) == NULL) 1355 return (ENOMEM); 1356 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1357 } 1358 } 1359 1360 if (src != NULL) 1361 sc->sc_sendaddr = *src; 1362 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*rup); 1363 h->count++; 1364 rup = sc->sc_statep.r++; 1365 memset(rup, 0, sizeof(*rup)); 1366 if (up != NULL) { 1367 memcpy(rup->id, up->id, sizeof(rup->id)); 1368 rup->creatorid = up->creatorid; 1369 } 1370 1371 if (h->count == sc->sc_maxcount) 1372 ret = pfsync_sendout(sc); 1373 1374 return (ret); 1375 } 1376 1377 int 1378 pfsync_clear_states(u_int32_t creatorid, char *ifname) 1379 { 1380 struct ifnet *ifp = NULL; 1381 struct pfsync_softc *sc = pfsyncif; 1382 struct pfsync_state_clr *cp; 1383 int s, ret; 1384 1385 if (sc == NULL) 1386 return (0); 1387 1388 ifp = &sc->sc_if; 1389 s = splnet(); 1390 if (sc->sc_mbuf != NULL) 1391 pfsync_sendout(sc); 1392 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_CLR, 1393 (void *)&sc->sc_statep.c)) == NULL) { 1394 splx(s); 1395 return (ENOMEM); 1396 } 1397 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*cp); 1398 cp = sc->sc_statep.c; 1399 cp->creatorid = creatorid; 1400 if (ifname != NULL) 1401 strlcpy(cp->ifname, ifname, IFNAMSIZ); 1402 1403 ret = (pfsync_sendout(sc)); 1404 splx(s); 1405 return (ret); 1406 } 1407 1408 void 1409 pfsync_timeout(void *v) 1410 { 1411 struct pfsync_softc *sc = v; 1412 int s; 1413 1414 s = splnet(); 1415 pfsync_sendout(sc); 1416 splx(s); 1417 } 1418 1419 void 1420 pfsync_tdb_timeout(void *v) 1421 { 1422 struct pfsync_softc *sc = v; 1423 int s; 1424 1425 s = splnet(); 1426 pfsync_tdb_sendout(sc); 1427 splx(s); 1428 } 1429 1430 /* This must be called in splnet() */ 1431 void 1432 pfsync_send_bus(struct pfsync_softc *sc, u_int8_t status) 1433 { 1434 struct pfsync_state_bus *bus; 1435 1436 if (sc->sc_mbuf != NULL) 1437 pfsync_sendout(sc); 1438 1439 if (pfsync_sync_ok && 1440 (sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_BUS, 1441 (void *)&sc->sc_statep.b)) != NULL) { 1442 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*bus); 1443 bus = sc->sc_statep.b; 1444 bus->creatorid = pf_status.hostid; 1445 bus->status = status; 1446 bus->endtime = htonl(time_uptime - sc->sc_ureq_received); 1447 pfsync_sendout(sc); 1448 } 1449 } 1450 1451 void 1452 pfsync_bulk_update(void *v) 1453 { 1454 struct pfsync_softc *sc = v; 1455 int s, i = 0; 1456 struct pf_state *state; 1457 1458 s = splnet(); 1459 if (sc->sc_mbuf != NULL) 1460 pfsync_sendout(sc); 1461 1462 /* 1463 * Grab at most PFSYNC_BULKPACKETS worth of states which have not 1464 * been sent since the latest request was made. 1465 */ 1466 state = sc->sc_bulk_send_next; 1467 if (state) 1468 do { 1469 /* send state update if syncable and not already sent */ 1470 if (!state->sync_flags 1471 && state->timeout < PFTM_MAX 1472 && state->pfsync_time <= sc->sc_ureq_received) { 1473 pfsync_pack_state(PFSYNC_ACT_UPD, state, 0); 1474 i++; 1475 } 1476 1477 /* figure next state to send */ 1478 state = TAILQ_NEXT(state, entry_list); 1479 1480 /* wrap to start of list if we hit the end */ 1481 if (!state) 1482 state = TAILQ_FIRST(&state_list); 1483 } while (i < sc->sc_maxcount * PFSYNC_BULKPACKETS && 1484 state != sc->sc_bulk_terminator); 1485 1486 if (!state || state == sc->sc_bulk_terminator) { 1487 /* we're done */ 1488 pfsync_send_bus(sc, PFSYNC_BUS_END); 1489 sc->sc_ureq_received = 0; 1490 sc->sc_bulk_send_next = NULL; 1491 sc->sc_bulk_terminator = NULL; 1492 callout_stop(&sc->sc_bulk_tmo); 1493 if (pf_status.debug >= PF_DEBUG_MISC) 1494 printf("pfsync: bulk update complete\n"); 1495 } else { 1496 /* look again for more in a bit */ 1497 callout_schedule(&sc->sc_bulk_tmo, 1); 1498 sc->sc_bulk_send_next = state; 1499 } 1500 if (sc->sc_mbuf != NULL) 1501 pfsync_sendout(sc); 1502 splx(s); 1503 } 1504 1505 void 1506 pfsync_bulkfail(void *v) 1507 { 1508 struct pfsync_softc *sc = v; 1509 int s, error; 1510 1511 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 1512 /* Try again in a bit */ 1513 callout_schedule(&sc->sc_bulkfail_tmo, 5 * hz); 1514 s = splnet(); 1515 error = pfsync_request_update(NULL, NULL); 1516 if (error == ENOMEM) { 1517 if (pf_status.debug >= PF_DEBUG_MISC) 1518 printf("pfsync: cannot allocate mbufs for " 1519 "bulk update\n"); 1520 } else 1521 pfsync_sendout(sc); 1522 splx(s); 1523 } else { 1524 /* Pretend like the transfer was ok */ 1525 sc->sc_ureq_sent = 0; 1526 sc->sc_bulk_tries = 0; 1527 #if NCARP > 0 1528 if (!pfsync_sync_ok) 1529 carp_suppress_preempt --; 1530 #endif 1531 pfsync_sync_ok = 1; 1532 if (pf_status.debug >= PF_DEBUG_MISC) 1533 printf("pfsync: failed to receive " 1534 "bulk update status\n"); 1535 callout_stop(&sc->sc_bulkfail_tmo); 1536 } 1537 } 1538 1539 /* This must be called in splnet() */ 1540 int 1541 pfsync_sendout(struct pfsync_softc *sc) 1542 { 1543 struct ifnet *ifp = &sc->sc_if; 1544 struct mbuf *m; 1545 1546 callout_stop(&sc->sc_tmo); 1547 1548 if (sc->sc_mbuf == NULL) 1549 return (0); 1550 m = sc->sc_mbuf; 1551 sc->sc_mbuf = NULL; 1552 sc->sc_statep.s = NULL; 1553 1554 if (ifp->if_bpf) 1555 bpf_ops->bpf_mtap(ifp->if_bpf, m); 1556 1557 if (sc->sc_mbuf_net) { 1558 m_freem(m); 1559 m = sc->sc_mbuf_net; 1560 sc->sc_mbuf_net = NULL; 1561 sc->sc_statep_net.s = NULL; 1562 } 1563 1564 return pfsync_sendout_mbuf(sc, m); 1565 } 1566 1567 int 1568 pfsync_tdb_sendout(struct pfsync_softc *sc) 1569 { 1570 struct ifnet *ifp = &sc->sc_if; 1571 struct mbuf *m; 1572 1573 callout_stop(&sc->sc_tdb_tmo); 1574 1575 if (sc->sc_mbuf_tdb == NULL) 1576 return (0); 1577 m = sc->sc_mbuf_tdb; 1578 sc->sc_mbuf_tdb = NULL; 1579 sc->sc_statep_tdb.t = NULL; 1580 1581 if (ifp->if_bpf) 1582 bpf_ops->bpf_mtap(ifp->if_bpf, m); 1583 1584 return pfsync_sendout_mbuf(sc, m); 1585 } 1586 1587 int 1588 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m) 1589 { 1590 struct sockaddr sa; 1591 struct ip *ip; 1592 1593 if (sc->sc_sync_ifp || 1594 sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) { 1595 M_PREPEND(m, sizeof(struct ip), M_DONTWAIT); 1596 if (m == NULL) { 1597 PFSYNC_STATINC(PFSYNC_STAT_ONOMEM); 1598 return (0); 1599 } 1600 ip = mtod(m, struct ip *); 1601 ip->ip_v = IPVERSION; 1602 ip->ip_hl = sizeof(*ip) >> 2; 1603 ip->ip_tos = IPTOS_LOWDELAY; 1604 ip->ip_len = htons(m->m_pkthdr.len); 1605 ip->ip_id = htons(ip_randomid(0)); 1606 ip->ip_off = htons(IP_DF); 1607 ip->ip_ttl = PFSYNC_DFLTTL; 1608 ip->ip_p = IPPROTO_PFSYNC; 1609 ip->ip_sum = 0; 1610 1611 memset(&sa, 0, sizeof(sa)); 1612 ip->ip_src.s_addr = INADDR_ANY; 1613 1614 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP) 1615 m->m_flags |= M_MCAST; 1616 ip->ip_dst = sc->sc_sendaddr; 1617 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr; 1618 1619 PFSYNC_STATINC(PFSYNC_STAT_OPACKETS); 1620 1621 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL)) { 1622 PFSYNC_STATINC(PFSYNC_STAT_OERRORS); 1623 } 1624 } else 1625 m_freem(m); 1626 1627 return (0); 1628 } 1629 1630 #ifdef IPSEC 1631 /* Update an in-kernel tdb. Silently fail if no tdb is found. */ 1632 void 1633 pfsync_update_net_tdb(struct pfsync_tdb *pt) 1634 { 1635 struct tdb *tdb; 1636 int s; 1637 1638 /* check for invalid values */ 1639 if (ntohl(pt->spi) <= SPI_RESERVED_MAX || 1640 (pt->dst.sa.sa_family != AF_INET && 1641 pt->dst.sa.sa_family != AF_INET6)) 1642 goto bad; 1643 1644 s = spltdb(); 1645 tdb = gettdb(pt->spi, &pt->dst, pt->sproto); 1646 if (tdb) { 1647 pt->rpl = ntohl(pt->rpl); 1648 pt->cur_bytes = betoh64(pt->cur_bytes); 1649 1650 /* Neither replay nor byte counter should ever decrease. */ 1651 if (pt->rpl < tdb->tdb_rpl || 1652 pt->cur_bytes < tdb->tdb_cur_bytes) { 1653 splx(s); 1654 goto bad; 1655 } 1656 1657 tdb->tdb_rpl = pt->rpl; 1658 tdb->tdb_cur_bytes = pt->cur_bytes; 1659 } 1660 splx(s); 1661 return; 1662 1663 bad: 1664 if (pf_status.debug >= PF_DEBUG_MISC) 1665 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: " 1666 "invalid value\n"); 1667 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 1668 return; 1669 } 1670 1671 /* One of our local tdbs have been updated, need to sync rpl with others */ 1672 int 1673 pfsync_update_tdb(struct tdb *tdb, int output) 1674 { 1675 struct ifnet *ifp = NULL; 1676 struct pfsync_softc *sc = pfsyncif; 1677 struct pfsync_header *h; 1678 struct pfsync_tdb *pt = NULL; 1679 int s, i, ret; 1680 1681 if (sc == NULL) 1682 return (0); 1683 1684 ifp = &sc->sc_if; 1685 if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL && 1686 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1687 /* Don't leave any stale pfsync packets hanging around. */ 1688 if (sc->sc_mbuf_tdb != NULL) { 1689 m_freem(sc->sc_mbuf_tdb); 1690 sc->sc_mbuf_tdb = NULL; 1691 sc->sc_statep_tdb.t = NULL; 1692 } 1693 return (0); 1694 } 1695 1696 s = splnet(); 1697 if (sc->sc_mbuf_tdb == NULL) { 1698 if ((sc->sc_mbuf_tdb = pfsync_get_mbuf(sc, PFSYNC_ACT_TDB_UPD, 1699 (void *)&sc->sc_statep_tdb.t)) == NULL) { 1700 splx(s); 1701 return (ENOMEM); 1702 } 1703 h = mtod(sc->sc_mbuf_tdb, struct pfsync_header *); 1704 } else { 1705 h = mtod(sc->sc_mbuf_tdb, struct pfsync_header *); 1706 if (h->action != PFSYNC_ACT_TDB_UPD) { 1707 /* 1708 * XXX will never happen as long as there's 1709 * only one "TDB action". 1710 */ 1711 pfsync_tdb_sendout(sc); 1712 sc->sc_mbuf_tdb = pfsync_get_mbuf(sc, 1713 PFSYNC_ACT_TDB_UPD, (void *)&sc->sc_statep_tdb.t); 1714 if (sc->sc_mbuf_tdb == NULL) { 1715 splx(s); 1716 return (ENOMEM); 1717 } 1718 h = mtod(sc->sc_mbuf_tdb, struct pfsync_header *); 1719 } else if (sc->sc_maxupdates) { 1720 /* 1721 * If it's an update, look in the packet to see if 1722 * we already have an update for the state. 1723 */ 1724 struct pfsync_tdb *u = 1725 (void *)((char *)h + PFSYNC_HDRLEN); 1726 1727 for (i = 0; !pt && i < h->count; i++) { 1728 if (tdb->tdb_spi == u->spi && 1729 tdb->tdb_sproto == u->sproto && 1730 !bcmp(&tdb->tdb_dst, &u->dst, 1731 SA_LEN(&u->dst.sa))) { 1732 pt = u; 1733 pt->updates++; 1734 } 1735 u++; 1736 } 1737 } 1738 } 1739 1740 if (pt == NULL) { 1741 /* not a "duplicate" update */ 1742 pt = sc->sc_statep_tdb.t++; 1743 sc->sc_mbuf_tdb->m_pkthdr.len = 1744 sc->sc_mbuf_tdb->m_len += sizeof(struct pfsync_tdb); 1745 h->count++; 1746 memset(pt, 0, sizeof(*pt)); 1747 1748 pt->spi = tdb->tdb_spi; 1749 memcpy(&pt->dst, &tdb->tdb_dst, sizeof pt->dst); 1750 pt->sproto = tdb->tdb_sproto; 1751 } 1752 1753 /* 1754 * When a failover happens, the master's rpl is probably above 1755 * what we see here (we may be up to a second late), so 1756 * increase it a bit for outbound tdbs to manage most such 1757 * situations. 1758 * 1759 * For now, just add an offset that is likely to be larger 1760 * than the number of packets we can see in one second. The RFC 1761 * just says the next packet must have a higher seq value. 1762 * 1763 * XXX What is a good algorithm for this? We could use 1764 * a rate-determined increase, but to know it, we would have 1765 * to extend struct tdb. 1766 * XXX pt->rpl can wrap over MAXINT, but if so the real tdb 1767 * will soon be replaced anyway. For now, just don't handle 1768 * this edge case. 1769 */ 1770 #define RPL_INCR 16384 1771 pt->rpl = htonl(tdb->tdb_rpl + (output ? RPL_INCR : 0)); 1772 pt->cur_bytes = htobe64(tdb->tdb_cur_bytes); 1773 1774 if (h->count == sc->sc_maxcount || 1775 (sc->sc_maxupdates && (pt->updates >= sc->sc_maxupdates))) 1776 ret = pfsync_tdb_sendout(sc); 1777 1778 splx(s); 1779 return (ret); 1780 } 1781 #endif 1782 1783 static int 1784 sysctl_net_inet_pfsync_stats(SYSCTLFN_ARGS) 1785 { 1786 1787 return (NETSTAT_SYSCTL(pfsyncstat_percpu, PFSYNC_NSTATS)); 1788 } 1789 1790 SYSCTL_SETUP(sysctl_net_inet_pfsync_setup, "sysctl net.inet.pfsync subtree setup") 1791 { 1792 1793 sysctl_createv(clog, 0, NULL, NULL, 1794 CTLFLAG_PERMANENT, 1795 CTLTYPE_NODE, "net", NULL, 1796 NULL, 0, NULL, 0, 1797 CTL_NET, CTL_EOL); 1798 sysctl_createv(clog, 0, NULL, NULL, 1799 CTLFLAG_PERMANENT, 1800 CTLTYPE_NODE, "inet", NULL, 1801 NULL, 0, NULL, 0, 1802 CTL_NET, PF_INET, CTL_EOL); 1803 sysctl_createv(clog, 0, NULL, NULL, 1804 CTLFLAG_PERMANENT, 1805 CTLTYPE_NODE, "pfsync", 1806 SYSCTL_DESCR("pfsync related settings"), 1807 NULL, 0, NULL, 0, 1808 CTL_NET, PF_INET, IPPROTO_PFSYNC, CTL_EOL); 1809 sysctl_createv(clog, 0, NULL, NULL, 1810 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1811 CTLTYPE_STRUCT, "stats", 1812 SYSCTL_DESCR("pfsync statistics"), 1813 sysctl_net_inet_pfsync_stats, 0, NULL, 0, 1814 CTL_NET, PF_INET, IPPROTO_PFSYNC, 1815 CTL_CREATE, CTL_EOL); 1816 } 1817