1 /* $NetBSD: if_pfsync.c,v 1.9 2012/03/22 20:34:38 drochner 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 1.9 2012/03/22 20:34:38 drochner 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 struct in_addr src; 365 struct mbuf *mp; 366 int iplen, action, error, i, s, count, offp, sfail, stale = 0; 367 u_int8_t chksum_flag = 0; 368 369 PFSYNC_STATINC(PFSYNC_STAT_IPACKETS); 370 371 /* verify that we have a sync interface configured */ 372 if (!sc || !sc->sc_sync_ifp || !pf_status.running) 373 goto done; 374 375 /* verify that the packet came in on the right interface */ 376 if (sc->sc_sync_ifp != m->m_pkthdr.rcvif) { 377 PFSYNC_STATINC(PFSYNC_STAT_BADIF); 378 goto done; 379 } 380 381 /* verify that the IP TTL is 255. */ 382 if (ip->ip_ttl != PFSYNC_DFLTTL) { 383 PFSYNC_STATINC(PFSYNC_STAT_BADTTL); 384 goto done; 385 } 386 387 iplen = ip->ip_hl << 2; 388 389 if (m->m_pkthdr.len < iplen + sizeof(*ph)) { 390 PFSYNC_STATINC(PFSYNC_STAT_HDROPS); 391 goto done; 392 } 393 394 if (iplen + sizeof(*ph) > m->m_len) { 395 if ((m = m_pullup(m, iplen + sizeof(*ph))) == NULL) { 396 PFSYNC_STATINC(PFSYNC_STAT_HDROPS); 397 goto done; 398 } 399 ip = mtod(m, struct ip *); 400 } 401 ph = (struct pfsync_header *)((char *)ip + iplen); 402 403 /* verify the version */ 404 if (ph->version != PFSYNC_VERSION) { 405 PFSYNC_STATINC(PFSYNC_STAT_BADVER); 406 goto done; 407 } 408 409 action = ph->action; 410 count = ph->count; 411 412 /* make sure it's a valid action code */ 413 if (action >= PFSYNC_ACT_MAX) { 414 PFSYNC_STATINC(PFSYNC_STAT_BADACT); 415 goto done; 416 } 417 418 /* Cheaper to grab this now than having to mess with mbufs later */ 419 src = ip->ip_src; 420 421 if (!bcmp(&ph->pf_chksum, &pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 422 chksum_flag++; 423 424 switch (action) { 425 case PFSYNC_ACT_CLR: { 426 struct pf_state *nexts; 427 struct pf_state_key *nextsk; 428 struct pfi_kif *kif; 429 u_int32_t creatorid; 430 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 431 sizeof(*cp), &offp)) == NULL) { 432 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 433 return; 434 } 435 cp = (struct pfsync_state_clr *)(mp->m_data + offp); 436 creatorid = cp->creatorid; 437 438 s = splsoftnet(); 439 if (cp->ifname[0] == '\0') { 440 for (st = RB_MIN(pf_state_tree_id, &tree_id); 441 st; st = nexts) { 442 nexts = RB_NEXT(pf_state_tree_id, &tree_id, st); 443 if (st->creatorid == creatorid) { 444 st->sync_flags |= PFSTATE_FROMSYNC; 445 pf_unlink_state(st); 446 } 447 } 448 } else { 449 if ((kif = pfi_kif_get(cp->ifname)) == NULL) { 450 splx(s); 451 return; 452 } 453 for (sk = RB_MIN(pf_state_tree_lan_ext, 454 &pf_statetbl_lan_ext); sk; sk = nextsk) { 455 nextsk = RB_NEXT(pf_state_tree_lan_ext, 456 &pf_statetbl_lan_ext, sk); 457 TAILQ_FOREACH(st, &sk->states, next) { 458 if (st->creatorid == creatorid) { 459 st->sync_flags |= 460 PFSTATE_FROMSYNC; 461 pf_unlink_state(st); 462 } 463 } 464 } 465 } 466 splx(s); 467 468 break; 469 } 470 case PFSYNC_ACT_INS: 471 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 472 count * sizeof(*sp), &offp)) == NULL) { 473 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 474 return; 475 } 476 477 s = splsoftnet(); 478 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 479 i < count; i++, sp++) { 480 /* check for invalid values */ 481 if (sp->timeout >= PFTM_MAX || 482 sp->src.state > PF_TCPS_PROXY_DST || 483 sp->dst.state > PF_TCPS_PROXY_DST || 484 sp->direction > PF_OUT || 485 (sp->af != AF_INET && sp->af != AF_INET6)) { 486 if (pf_status.debug >= PF_DEBUG_MISC) 487 printf("pfsync_insert: PFSYNC_ACT_INS: " 488 "invalid value\n"); 489 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 490 continue; 491 } 492 493 if ((error = pfsync_insert_net_state(sp, 494 chksum_flag))) { 495 if (error == ENOMEM) { 496 splx(s); 497 goto done; 498 } 499 continue; 500 } 501 } 502 splx(s); 503 break; 504 case PFSYNC_ACT_UPD: 505 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 506 count * sizeof(*sp), &offp)) == NULL) { 507 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 508 return; 509 } 510 511 s = splsoftnet(); 512 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 513 i < count; i++, sp++) { 514 int flags = PFSYNC_FLAG_STALE; 515 516 /* check for invalid values */ 517 if (sp->timeout >= PFTM_MAX || 518 sp->src.state > PF_TCPS_PROXY_DST || 519 sp->dst.state > PF_TCPS_PROXY_DST) { 520 if (pf_status.debug >= PF_DEBUG_MISC) 521 printf("pfsync_insert: PFSYNC_ACT_UPD: " 522 "invalid value\n"); 523 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 524 continue; 525 } 526 527 memcpy(&id_key.id, sp->id, sizeof(id_key.id)); 528 id_key.creatorid = sp->creatorid; 529 530 st = pf_find_state_byid(&id_key); 531 if (st == NULL) { 532 /* insert the update */ 533 if (pfsync_insert_net_state(sp, chksum_flag)) { 534 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 535 } 536 continue; 537 } 538 sk = st->state_key; 539 sfail = 0; 540 if (sk->proto == IPPROTO_TCP) { 541 /* 542 * The state should never go backwards except 543 * for syn-proxy states. Neither should the 544 * sequence window slide backwards. 545 */ 546 if (st->src.state > sp->src.state && 547 (st->src.state < PF_TCPS_PROXY_SRC || 548 sp->src.state >= PF_TCPS_PROXY_SRC)) 549 sfail = 1; 550 else if (SEQ_GT(st->src.seqlo, 551 ntohl(sp->src.seqlo))) 552 sfail = 3; 553 else if (st->dst.state > sp->dst.state) { 554 /* There might still be useful 555 * information about the src state here, 556 * so import that part of the update, 557 * then "fail" so we send the updated 558 * state back to the peer who is missing 559 * our what we know. */ 560 pf_state_peer_ntoh(&sp->src, &st->src); 561 /* XXX do anything with timeouts? */ 562 sfail = 7; 563 flags = 0; 564 } else if (st->dst.state >= TCPS_SYN_SENT && 565 SEQ_GT(st->dst.seqlo, ntohl(sp->dst.seqlo))) 566 sfail = 4; 567 } else { 568 /* 569 * Non-TCP protocol state machine always go 570 * forwards 571 */ 572 if (st->src.state > sp->src.state) 573 sfail = 5; 574 else if (st->dst.state > sp->dst.state) 575 sfail = 6; 576 } 577 if (sfail) { 578 if (pf_status.debug >= PF_DEBUG_MISC) 579 printf("pfsync: %s stale update " 580 "(%d) id: %016" PRIu64 "" 581 "creatorid: %08x\n", 582 (sfail < 7 ? "ignoring" 583 : "partial"), sfail, 584 be64toh(st->id), 585 ntohl(st->creatorid)); 586 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 587 588 if (!(sp->sync_flags & PFSTATE_STALE)) { 589 /* we have a better state, send it */ 590 if (sc->sc_mbuf != NULL && !stale) 591 pfsync_sendout(sc); 592 stale++; 593 if (!st->sync_flags) 594 pfsync_pack_state( 595 PFSYNC_ACT_UPD, st, flags); 596 } 597 continue; 598 } 599 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 600 pf_state_peer_ntoh(&sp->src, &st->src); 601 pf_state_peer_ntoh(&sp->dst, &st->dst); 602 st->expire = ntohl(sp->expire) + time_second; 603 st->timeout = sp->timeout; 604 } 605 if (stale && sc->sc_mbuf != NULL) 606 pfsync_sendout(sc); 607 splx(s); 608 break; 609 /* 610 * It's not strictly necessary for us to support the "uncompressed" 611 * delete action, but it's relatively simple and maintains consistency. 612 */ 613 case PFSYNC_ACT_DEL: 614 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 615 count * sizeof(*sp), &offp)) == NULL) { 616 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 617 return; 618 } 619 620 s = splsoftnet(); 621 for (i = 0, sp = (struct pfsync_state *)(mp->m_data + offp); 622 i < count; i++, sp++) { 623 memcpy(&id_key.id, sp->id, sizeof(id_key.id)); 624 id_key.creatorid = sp->creatorid; 625 626 st = pf_find_state_byid(&id_key); 627 if (st == NULL) { 628 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 629 continue; 630 } 631 st->sync_flags |= PFSTATE_FROMSYNC; 632 pf_unlink_state(st); 633 } 634 splx(s); 635 break; 636 case PFSYNC_ACT_UPD_C: { 637 int update_requested = 0; 638 639 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 640 count * sizeof(*up), &offp)) == NULL) { 641 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 642 return; 643 } 644 645 s = splsoftnet(); 646 for (i = 0, up = (struct pfsync_state_upd *)(mp->m_data + offp); 647 i < count; i++, up++) { 648 /* check for invalid values */ 649 if (up->timeout >= PFTM_MAX || 650 up->src.state > PF_TCPS_PROXY_DST || 651 up->dst.state > PF_TCPS_PROXY_DST) { 652 if (pf_status.debug >= PF_DEBUG_MISC) 653 printf("pfsync_insert: " 654 "PFSYNC_ACT_UPD_C: " 655 "invalid value\n"); 656 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 657 continue; 658 } 659 660 memcpy(&id_key.id, up->id, sizeof(id_key.id)); 661 id_key.creatorid = up->creatorid; 662 663 st = pf_find_state_byid(&id_key); 664 if (st == NULL) { 665 /* We don't have this state. Ask for it. */ 666 error = pfsync_request_update(up, &src); 667 if (error == ENOMEM) { 668 splx(s); 669 goto done; 670 } 671 update_requested = 1; 672 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 673 continue; 674 } 675 sk = st->state_key; 676 sfail = 0; 677 if (sk->proto == IPPROTO_TCP) { 678 /* 679 * The state should never go backwards except 680 * for syn-proxy states. Neither should the 681 * sequence window slide backwards. 682 */ 683 if (st->src.state > up->src.state && 684 (st->src.state < PF_TCPS_PROXY_SRC || 685 up->src.state >= PF_TCPS_PROXY_SRC)) 686 sfail = 1; 687 else if (st->dst.state > up->dst.state) 688 sfail = 2; 689 else if (SEQ_GT(st->src.seqlo, 690 ntohl(up->src.seqlo))) 691 sfail = 3; 692 else if (st->dst.state >= TCPS_SYN_SENT && 693 SEQ_GT(st->dst.seqlo, ntohl(up->dst.seqlo))) 694 sfail = 4; 695 } else { 696 /* 697 * Non-TCP protocol state machine always go 698 * forwards 699 */ 700 if (st->src.state > up->src.state) 701 sfail = 5; 702 else if (st->dst.state > up->dst.state) 703 sfail = 6; 704 } 705 if (sfail) { 706 if (pf_status.debug >= PF_DEBUG_MISC) 707 printf("pfsync: ignoring stale update " 708 "(%d) id: %016" PRIu64 "" 709 "creatorid: %08x\n", sfail, 710 be64toh(st->id), 711 ntohl(st->creatorid)); 712 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 713 714 /* we have a better state, send it out */ 715 if ((!stale || update_requested) && 716 sc->sc_mbuf != NULL) { 717 pfsync_sendout(sc); 718 update_requested = 0; 719 } 720 stale++; 721 if (!st->sync_flags) 722 pfsync_pack_state(PFSYNC_ACT_UPD, st, 723 PFSYNC_FLAG_STALE); 724 continue; 725 } 726 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 727 pf_state_peer_ntoh(&up->src, &st->src); 728 pf_state_peer_ntoh(&up->dst, &st->dst); 729 st->expire = ntohl(up->expire) + time_second; 730 st->timeout = up->timeout; 731 } 732 if ((update_requested || stale) && sc->sc_mbuf) 733 pfsync_sendout(sc); 734 splx(s); 735 break; 736 } 737 case PFSYNC_ACT_DEL_C: 738 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 739 count * sizeof(*dp), &offp)) == NULL) { 740 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 741 return; 742 } 743 744 s = splsoftnet(); 745 for (i = 0, dp = (struct pfsync_state_del *)(mp->m_data + offp); 746 i < count; i++, dp++) { 747 memcpy(&id_key.id, dp->id, sizeof(id_key.id)); 748 id_key.creatorid = dp->creatorid; 749 750 st = pf_find_state_byid(&id_key); 751 if (st == NULL) { 752 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 753 continue; 754 } 755 st->sync_flags |= PFSTATE_FROMSYNC; 756 pf_unlink_state(st); 757 } 758 splx(s); 759 break; 760 case PFSYNC_ACT_INS_F: 761 case PFSYNC_ACT_DEL_F: 762 /* not implemented */ 763 break; 764 case PFSYNC_ACT_UREQ: 765 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 766 count * sizeof(*rup), &offp)) == NULL) { 767 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 768 return; 769 } 770 771 s = splsoftnet(); 772 if (sc->sc_mbuf != NULL) 773 pfsync_sendout(sc); 774 for (i = 0, 775 rup = (struct pfsync_state_upd_req *)(mp->m_data + offp); 776 i < count; i++, rup++) { 777 memcpy(&id_key.id, rup->id, sizeof(id_key.id)); 778 id_key.creatorid = rup->creatorid; 779 780 if (id_key.id == 0 && id_key.creatorid == 0) { 781 sc->sc_ureq_received = time_uptime; 782 if (sc->sc_bulk_send_next == NULL) 783 sc->sc_bulk_send_next = 784 TAILQ_FIRST(&state_list); 785 sc->sc_bulk_terminator = sc->sc_bulk_send_next; 786 if (pf_status.debug >= PF_DEBUG_MISC) 787 printf("pfsync: received " 788 "bulk update request\n"); 789 pfsync_send_bus(sc, PFSYNC_BUS_START); 790 callout_schedule(&sc->sc_bulk_tmo, 1 * hz); 791 } else { 792 st = pf_find_state_byid(&id_key); 793 if (st == NULL) { 794 PFSYNC_STATINC(PFSYNC_STAT_BADSTATE); 795 continue; 796 } 797 if (!st->sync_flags) 798 pfsync_pack_state(PFSYNC_ACT_UPD, 799 st, 0); 800 } 801 } 802 if (sc->sc_mbuf != NULL) 803 pfsync_sendout(sc); 804 splx(s); 805 break; 806 case PFSYNC_ACT_BUS: 807 /* If we're not waiting for a bulk update, who cares. */ 808 if (sc->sc_ureq_sent == 0) 809 break; 810 811 if ((mp = m_pulldown(m, iplen + sizeof(*ph), 812 sizeof(*bus), &offp)) == NULL) { 813 PFSYNC_STATINC(PFSYNC_STAT_BADLEN); 814 return; 815 } 816 bus = (struct pfsync_state_bus *)(mp->m_data + offp); 817 switch (bus->status) { 818 case PFSYNC_BUS_START: 819 callout_schedule(&sc->sc_bulkfail_tmo, 820 pf_pool_limits[PF_LIMIT_STATES].limit / 821 (PFSYNC_BULKPACKETS * sc->sc_maxcount)); 822 if (pf_status.debug >= PF_DEBUG_MISC) 823 printf("pfsync: received bulk " 824 "update start\n"); 825 break; 826 case PFSYNC_BUS_END: 827 if (time_uptime - ntohl(bus->endtime) >= 828 sc->sc_ureq_sent) { 829 /* that's it, we're happy */ 830 sc->sc_ureq_sent = 0; 831 sc->sc_bulk_tries = 0; 832 callout_stop(&sc->sc_bulkfail_tmo); 833 #if NCARP > 0 834 if (!pfsync_sync_ok) 835 carp_suppress_preempt--; 836 #endif 837 pfsync_sync_ok = 1; 838 if (pf_status.debug >= PF_DEBUG_MISC) 839 printf("pfsync: received valid " 840 "bulk update end\n"); 841 } else { 842 if (pf_status.debug >= PF_DEBUG_MISC) 843 printf("pfsync: received invalid " 844 "bulk update end: bad timestamp\n"); 845 } 846 break; 847 } 848 break; 849 } 850 851 done: 852 if (m) 853 m_freem(m); 854 } 855 856 int 857 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, 858 struct rtentry *rt) 859 { 860 m_freem(m); 861 return (0); 862 } 863 864 /* ARGSUSED */ 865 int 866 pfsyncioctl(struct ifnet *ifp, u_long cmd, void* data) 867 { 868 struct lwp *l = curlwp; 869 struct pfsync_softc *sc = ifp->if_softc; 870 struct ifreq *ifr = (struct ifreq *)data; 871 struct ip_moptions *imo = &sc->sc_imo; 872 struct pfsyncreq pfsyncr; 873 struct ifnet *sifp; 874 int s, error; 875 876 switch (cmd) { 877 case SIOCSIFADDR: 878 case SIOCAIFADDR: 879 case SIOCSIFDSTADDR: 880 case SIOCSIFFLAGS: 881 if (ifp->if_flags & IFF_UP) 882 ifp->if_flags |= IFF_RUNNING; 883 else 884 ifp->if_flags &= ~IFF_RUNNING; 885 break; 886 case SIOCSIFMTU: 887 if (ifr->ifr_mtu < PFSYNC_MINMTU) 888 return (EINVAL); 889 if (ifr->ifr_mtu > MCLBYTES) 890 ifr->ifr_mtu = MCLBYTES; 891 s = splnet(); 892 if (ifr->ifr_mtu < ifp->if_mtu) 893 pfsync_sendout(sc); 894 pfsync_setmtu(sc, ifr->ifr_mtu); 895 splx(s); 896 break; 897 case SIOCGETPFSYNC: 898 if ((error = kauth_authorize_network(l->l_cred, 899 KAUTH_NETWORK_INTERFACE, 900 KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, ifp, (void *)cmd, 901 NULL)) != 0) 902 return (error); 903 memset(&pfsyncr, 0, sizeof(pfsyncr)); 904 if (sc->sc_sync_ifp) 905 strlcpy(pfsyncr.pfsyncr_syncdev, 906 sc->sc_sync_ifp->if_xname, IFNAMSIZ); 907 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 908 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 909 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)))) 910 return (error); 911 break; 912 case SIOCSETPFSYNC: 913 if ((error = kauth_authorize_network(l->l_cred, 914 KAUTH_NETWORK_INTERFACE, 915 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 916 NULL)) != 0) 917 return (error); 918 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) 919 return (error); 920 921 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 922 sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP; 923 else 924 sc->sc_sync_peer.s_addr = 925 pfsyncr.pfsyncr_syncpeer.s_addr; 926 927 if (pfsyncr.pfsyncr_maxupdates > 255) 928 return (EINVAL); 929 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 930 931 if (pfsyncr.pfsyncr_syncdev[0] == 0) { 932 sc->sc_sync_ifp = NULL; 933 if (sc->sc_mbuf_net != NULL) { 934 /* Don't keep stale pfsync packets around. */ 935 s = splnet(); 936 m_freem(sc->sc_mbuf_net); 937 sc->sc_mbuf_net = NULL; 938 sc->sc_statep_net.s = NULL; 939 splx(s); 940 } 941 if (imo->imo_num_memberships > 0) { 942 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 943 imo->imo_multicast_ifp = NULL; 944 } 945 break; 946 } 947 948 if ((sifp = ifunit(pfsyncr.pfsyncr_syncdev)) == NULL) 949 return (EINVAL); 950 951 s = splnet(); 952 if (sifp->if_mtu < sc->sc_if.if_mtu || 953 (sc->sc_sync_ifp != NULL && 954 sifp->if_mtu < sc->sc_sync_ifp->if_mtu) || 955 sifp->if_mtu < MCLBYTES - sizeof(struct ip)) 956 pfsync_sendout(sc); 957 sc->sc_sync_ifp = sifp; 958 959 pfsync_setmtu(sc, sc->sc_if.if_mtu); 960 961 if (imo->imo_num_memberships > 0) { 962 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 963 imo->imo_multicast_ifp = NULL; 964 } 965 966 if (sc->sc_sync_ifp && 967 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 968 struct in_addr addr; 969 970 if (!(sc->sc_sync_ifp->if_flags & IFF_MULTICAST)) { 971 sc->sc_sync_ifp = NULL; 972 splx(s); 973 return (EADDRNOTAVAIL); 974 } 975 976 addr.s_addr = INADDR_PFSYNC_GROUP; 977 978 if ((imo->imo_membership[0] = 979 in_addmulti(&addr, sc->sc_sync_ifp)) == NULL) { 980 sc->sc_sync_ifp = NULL; 981 splx(s); 982 return (ENOBUFS); 983 } 984 imo->imo_num_memberships++; 985 imo->imo_multicast_ifp = sc->sc_sync_ifp; 986 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 987 imo->imo_multicast_loop = 0; 988 } 989 990 if (sc->sc_sync_ifp || 991 sc->sc_sendaddr.s_addr != INADDR_PFSYNC_GROUP) { 992 /* Request a full state table update. */ 993 sc->sc_ureq_sent = time_uptime; 994 #if NCARP > 0 995 if (pfsync_sync_ok) 996 carp_suppress_preempt ++; 997 #endif 998 pfsync_sync_ok = 0; 999 if (pf_status.debug >= PF_DEBUG_MISC) 1000 printf("pfsync: requesting bulk update\n"); 1001 callout_schedule(&sc->sc_bulkfail_tmo, 5 * hz); 1002 error = pfsync_request_update(NULL, NULL); 1003 if (error == ENOMEM) { 1004 splx(s); 1005 return (ENOMEM); 1006 } 1007 pfsync_sendout(sc); 1008 } 1009 splx(s); 1010 1011 break; 1012 1013 default: 1014 return ifioctl_common(ifp, cmd, data); 1015 } 1016 1017 return (0); 1018 } 1019 1020 void 1021 pfsync_setmtu(struct pfsync_softc *sc, int mtu_req) 1022 { 1023 int mtu; 1024 1025 if (sc->sc_sync_ifp && sc->sc_sync_ifp->if_mtu < mtu_req) 1026 mtu = sc->sc_sync_ifp->if_mtu; 1027 else 1028 mtu = mtu_req; 1029 1030 sc->sc_maxcount = (mtu - sizeof(struct pfsync_header)) / 1031 sizeof(struct pfsync_state); 1032 if (sc->sc_maxcount > 254) 1033 sc->sc_maxcount = 254; 1034 sc->sc_if.if_mtu = sizeof(struct pfsync_header) + 1035 sc->sc_maxcount * sizeof(struct pfsync_state); 1036 } 1037 1038 struct mbuf * 1039 pfsync_get_mbuf(struct pfsync_softc *sc, u_int8_t action, void **sp) 1040 { 1041 struct pfsync_header *h; 1042 struct mbuf *m; 1043 int len; 1044 1045 MGETHDR(m, M_DONTWAIT, MT_DATA); 1046 if (m == NULL) { 1047 sc->sc_if.if_oerrors++; 1048 return (NULL); 1049 } 1050 1051 switch (action) { 1052 case PFSYNC_ACT_CLR: 1053 len = sizeof(struct pfsync_header) + 1054 sizeof(struct pfsync_state_clr); 1055 break; 1056 case PFSYNC_ACT_UPD_C: 1057 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd)) + 1058 sizeof(struct pfsync_header); 1059 break; 1060 case PFSYNC_ACT_DEL_C: 1061 len = (sc->sc_maxcount * sizeof(struct pfsync_state_del)) + 1062 sizeof(struct pfsync_header); 1063 break; 1064 case PFSYNC_ACT_UREQ: 1065 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd_req)) + 1066 sizeof(struct pfsync_header); 1067 break; 1068 case PFSYNC_ACT_BUS: 1069 len = sizeof(struct pfsync_header) + 1070 sizeof(struct pfsync_state_bus); 1071 break; 1072 case PFSYNC_ACT_TDB_UPD: 1073 len = (sc->sc_maxcount * sizeof(struct pfsync_tdb)) + 1074 sizeof(struct pfsync_header); 1075 break; 1076 default: 1077 len = (sc->sc_maxcount * sizeof(struct pfsync_state)) + 1078 sizeof(struct pfsync_header); 1079 break; 1080 } 1081 1082 if (len > MHLEN) { 1083 MCLGET(m, M_DONTWAIT); 1084 if ((m->m_flags & M_EXT) == 0) { 1085 m_free(m); 1086 sc->sc_if.if_oerrors++; 1087 return (NULL); 1088 } 1089 m->m_data += (MCLBYTES - len) &~ (sizeof(long) - 1); 1090 } else 1091 MH_ALIGN(m, len); 1092 1093 m->m_pkthdr.rcvif = NULL; 1094 m->m_pkthdr.len = m->m_len = sizeof(struct pfsync_header); 1095 h = mtod(m, struct pfsync_header *); 1096 h->version = PFSYNC_VERSION; 1097 h->af = 0; 1098 h->count = 0; 1099 h->action = action; 1100 if (action != PFSYNC_ACT_TDB_UPD) 1101 memcpy(&h->pf_chksum, &pf_status.pf_chksum, 1102 PF_MD5_DIGEST_LENGTH); 1103 1104 *sp = (void *)((char *)h + PFSYNC_HDRLEN); 1105 if (action == PFSYNC_ACT_TDB_UPD) 1106 callout_schedule(&sc->sc_tdb_tmo, hz); 1107 else 1108 callout_schedule(&sc->sc_tmo, hz); 1109 return (m); 1110 } 1111 1112 int 1113 pfsync_pack_state(u_int8_t action, struct pf_state *st, int flags) 1114 { 1115 struct ifnet *ifp = NULL; 1116 struct pfsync_softc *sc = pfsyncif; 1117 struct pfsync_header *h, *h_net; 1118 struct pfsync_state *sp = NULL; 1119 struct pfsync_state_upd *up = NULL; 1120 struct pfsync_state_del *dp = NULL; 1121 struct pf_state_key *sk = st->state_key; 1122 struct pf_rule *r; 1123 u_long secs; 1124 int s, ret = 0; 1125 u_int8_t i = 255, newaction = 0; 1126 1127 if (sc == NULL) 1128 return (0); 1129 ifp = &sc->sc_if; 1130 1131 /* 1132 * If a packet falls in the forest and there's nobody around to 1133 * hear, does it make a sound? 1134 */ 1135 if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL && 1136 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1137 /* Don't leave any stale pfsync packets hanging around. */ 1138 if (sc->sc_mbuf != NULL) { 1139 m_freem(sc->sc_mbuf); 1140 sc->sc_mbuf = NULL; 1141 sc->sc_statep.s = NULL; 1142 } 1143 return (0); 1144 } 1145 1146 if (action >= PFSYNC_ACT_MAX) 1147 return (EINVAL); 1148 1149 s = splnet(); 1150 if (sc->sc_mbuf == NULL) { 1151 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1152 (void *)&sc->sc_statep.s)) == NULL) { 1153 splx(s); 1154 return (ENOMEM); 1155 } 1156 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1157 } else { 1158 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1159 if (h->action != action) { 1160 pfsync_sendout(sc); 1161 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1162 (void *)&sc->sc_statep.s)) == NULL) { 1163 splx(s); 1164 return (ENOMEM); 1165 } 1166 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1167 } else { 1168 /* 1169 * If it's an update, look in the packet to see if 1170 * we already have an update for the state. 1171 */ 1172 if (action == PFSYNC_ACT_UPD && sc->sc_maxupdates) { 1173 struct pfsync_state *usp = 1174 (void *)((char *)h + PFSYNC_HDRLEN); 1175 1176 for (i = 0; i < h->count; i++) { 1177 if (!memcmp(usp->id, &st->id, 1178 PFSYNC_ID_LEN) && 1179 usp->creatorid == st->creatorid) { 1180 sp = usp; 1181 sp->updates++; 1182 break; 1183 } 1184 usp++; 1185 } 1186 } 1187 } 1188 } 1189 1190 secs = time_second; 1191 1192 st->pfsync_time = time_uptime; 1193 1194 if (sp == NULL) { 1195 /* not a "duplicate" update */ 1196 i = 255; 1197 sp = sc->sc_statep.s++; 1198 sc->sc_mbuf->m_pkthdr.len = 1199 sc->sc_mbuf->m_len += sizeof(struct pfsync_state); 1200 h->count++; 1201 memset(sp, 0, sizeof(*sp)); 1202 1203 memcpy(sp->id, &st->id, sizeof(sp->id)); 1204 sp->creatorid = st->creatorid; 1205 1206 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); 1207 pf_state_host_hton(&sk->lan, &sp->lan); 1208 pf_state_host_hton(&sk->gwy, &sp->gwy); 1209 pf_state_host_hton(&sk->ext, &sp->ext); 1210 1211 memcpy(&sp->rt_addr, &st->rt_addr, sizeof(sp->rt_addr)); 1212 1213 sp->creation = htonl(secs - st->creation); 1214 pf_state_counter_hton(st->packets[0], sp->packets[0]); 1215 pf_state_counter_hton(st->packets[1], sp->packets[1]); 1216 pf_state_counter_hton(st->bytes[0], sp->bytes[0]); 1217 pf_state_counter_hton(st->bytes[1], sp->bytes[1]); 1218 if ((r = st->rule.ptr) == NULL) 1219 sp->rule = htonl(-1); 1220 else 1221 sp->rule = htonl(r->nr); 1222 if ((r = st->anchor.ptr) == NULL) 1223 sp->anchor = htonl(-1); 1224 else 1225 sp->anchor = htonl(r->nr); 1226 sp->af = sk->af; 1227 sp->proto = sk->proto; 1228 sp->direction = sk->direction; 1229 sp->log = st->log; 1230 sp->allow_opts = st->allow_opts; 1231 sp->timeout = st->timeout; 1232 1233 if (flags & PFSYNC_FLAG_STALE) 1234 sp->sync_flags |= PFSTATE_STALE; 1235 } 1236 1237 pf_state_peer_hton(&st->src, &sp->src); 1238 pf_state_peer_hton(&st->dst, &sp->dst); 1239 1240 if (st->expire <= secs) 1241 sp->expire = htonl(0); 1242 else 1243 sp->expire = htonl(st->expire - secs); 1244 1245 /* do we need to build "compressed" actions for network transfer? */ 1246 if (sc->sc_sync_ifp && flags & PFSYNC_FLAG_COMPRESS) { 1247 switch (action) { 1248 case PFSYNC_ACT_UPD: 1249 newaction = PFSYNC_ACT_UPD_C; 1250 break; 1251 case PFSYNC_ACT_DEL: 1252 newaction = PFSYNC_ACT_DEL_C; 1253 break; 1254 default: 1255 /* by default we just send the uncompressed states */ 1256 break; 1257 } 1258 } 1259 1260 if (newaction) { 1261 if (sc->sc_mbuf_net == NULL) { 1262 if ((sc->sc_mbuf_net = pfsync_get_mbuf(sc, newaction, 1263 (void *)&sc->sc_statep_net.s)) == NULL) { 1264 splx(s); 1265 return (ENOMEM); 1266 } 1267 } 1268 h_net = mtod(sc->sc_mbuf_net, struct pfsync_header *); 1269 1270 switch (newaction) { 1271 case PFSYNC_ACT_UPD_C: 1272 if (i != 255) { 1273 up = (void *)((char *)h_net + 1274 PFSYNC_HDRLEN + (i * sizeof(*up))); 1275 up->updates++; 1276 } else { 1277 h_net->count++; 1278 sc->sc_mbuf_net->m_pkthdr.len = 1279 sc->sc_mbuf_net->m_len += sizeof(*up); 1280 up = sc->sc_statep_net.u++; 1281 1282 memset(up, 0, sizeof(*up)); 1283 memcpy(up->id, &st->id, sizeof(up->id)); 1284 up->creatorid = st->creatorid; 1285 } 1286 up->timeout = st->timeout; 1287 up->expire = sp->expire; 1288 up->src = sp->src; 1289 up->dst = sp->dst; 1290 break; 1291 case PFSYNC_ACT_DEL_C: 1292 sc->sc_mbuf_net->m_pkthdr.len = 1293 sc->sc_mbuf_net->m_len += sizeof(*dp); 1294 dp = sc->sc_statep_net.d++; 1295 h_net->count++; 1296 1297 memset(dp, 0, sizeof(*dp)); 1298 memcpy(dp->id, &st->id, sizeof(dp->id)); 1299 dp->creatorid = st->creatorid; 1300 break; 1301 } 1302 } 1303 1304 if (h->count == sc->sc_maxcount || 1305 (sc->sc_maxupdates && (sp->updates >= sc->sc_maxupdates))) 1306 ret = pfsync_sendout(sc); 1307 1308 splx(s); 1309 return (ret); 1310 } 1311 1312 /* This must be called in splnet() */ 1313 int 1314 pfsync_request_update(struct pfsync_state_upd *up, struct in_addr *src) 1315 { 1316 struct ifnet *ifp = NULL; 1317 struct pfsync_header *h; 1318 struct pfsync_softc *sc = pfsyncif; 1319 struct pfsync_state_upd_req *rup; 1320 int ret = 0; 1321 1322 if (sc == NULL) 1323 return (0); 1324 1325 ifp = &sc->sc_if; 1326 if (sc->sc_mbuf == NULL) { 1327 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1328 (void *)&sc->sc_statep.s)) == NULL) 1329 return (ENOMEM); 1330 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1331 } else { 1332 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1333 if (h->action != PFSYNC_ACT_UREQ) { 1334 pfsync_sendout(sc); 1335 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1336 (void *)&sc->sc_statep.s)) == NULL) 1337 return (ENOMEM); 1338 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1339 } 1340 } 1341 1342 if (src != NULL) 1343 sc->sc_sendaddr = *src; 1344 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*rup); 1345 h->count++; 1346 rup = sc->sc_statep.r++; 1347 memset(rup, 0, sizeof(*rup)); 1348 if (up != NULL) { 1349 memcpy(rup->id, up->id, sizeof(rup->id)); 1350 rup->creatorid = up->creatorid; 1351 } 1352 1353 if (h->count == sc->sc_maxcount) 1354 ret = pfsync_sendout(sc); 1355 1356 return (ret); 1357 } 1358 1359 int 1360 pfsync_clear_states(u_int32_t creatorid, char *ifname) 1361 { 1362 struct ifnet *ifp = NULL; 1363 struct pfsync_softc *sc = pfsyncif; 1364 struct pfsync_state_clr *cp; 1365 int s, ret; 1366 1367 if (sc == NULL) 1368 return (0); 1369 1370 ifp = &sc->sc_if; 1371 s = splnet(); 1372 if (sc->sc_mbuf != NULL) 1373 pfsync_sendout(sc); 1374 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_CLR, 1375 (void *)&sc->sc_statep.c)) == NULL) { 1376 splx(s); 1377 return (ENOMEM); 1378 } 1379 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*cp); 1380 cp = sc->sc_statep.c; 1381 cp->creatorid = creatorid; 1382 if (ifname != NULL) 1383 strlcpy(cp->ifname, ifname, IFNAMSIZ); 1384 1385 ret = (pfsync_sendout(sc)); 1386 splx(s); 1387 return (ret); 1388 } 1389 1390 void 1391 pfsync_timeout(void *v) 1392 { 1393 struct pfsync_softc *sc = v; 1394 int s; 1395 1396 s = splnet(); 1397 pfsync_sendout(sc); 1398 splx(s); 1399 } 1400 1401 void 1402 pfsync_tdb_timeout(void *v) 1403 { 1404 struct pfsync_softc *sc = v; 1405 int s; 1406 1407 s = splnet(); 1408 pfsync_tdb_sendout(sc); 1409 splx(s); 1410 } 1411 1412 /* This must be called in splnet() */ 1413 void 1414 pfsync_send_bus(struct pfsync_softc *sc, u_int8_t status) 1415 { 1416 struct pfsync_state_bus *bus; 1417 1418 if (sc->sc_mbuf != NULL) 1419 pfsync_sendout(sc); 1420 1421 if (pfsync_sync_ok && 1422 (sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_BUS, 1423 (void *)&sc->sc_statep.b)) != NULL) { 1424 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*bus); 1425 bus = sc->sc_statep.b; 1426 bus->creatorid = pf_status.hostid; 1427 bus->status = status; 1428 bus->endtime = htonl(time_uptime - sc->sc_ureq_received); 1429 pfsync_sendout(sc); 1430 } 1431 } 1432 1433 void 1434 pfsync_bulk_update(void *v) 1435 { 1436 struct pfsync_softc *sc = v; 1437 int s, i = 0; 1438 struct pf_state *state; 1439 1440 s = splnet(); 1441 if (sc->sc_mbuf != NULL) 1442 pfsync_sendout(sc); 1443 1444 /* 1445 * Grab at most PFSYNC_BULKPACKETS worth of states which have not 1446 * been sent since the latest request was made. 1447 */ 1448 state = sc->sc_bulk_send_next; 1449 if (state) 1450 do { 1451 /* send state update if syncable and not already sent */ 1452 if (!state->sync_flags 1453 && state->timeout < PFTM_MAX 1454 && state->pfsync_time <= sc->sc_ureq_received) { 1455 pfsync_pack_state(PFSYNC_ACT_UPD, state, 0); 1456 i++; 1457 } 1458 1459 /* figure next state to send */ 1460 state = TAILQ_NEXT(state, entry_list); 1461 1462 /* wrap to start of list if we hit the end */ 1463 if (!state) 1464 state = TAILQ_FIRST(&state_list); 1465 } while (i < sc->sc_maxcount * PFSYNC_BULKPACKETS && 1466 state != sc->sc_bulk_terminator); 1467 1468 if (!state || state == sc->sc_bulk_terminator) { 1469 /* we're done */ 1470 pfsync_send_bus(sc, PFSYNC_BUS_END); 1471 sc->sc_ureq_received = 0; 1472 sc->sc_bulk_send_next = NULL; 1473 sc->sc_bulk_terminator = NULL; 1474 callout_stop(&sc->sc_bulk_tmo); 1475 if (pf_status.debug >= PF_DEBUG_MISC) 1476 printf("pfsync: bulk update complete\n"); 1477 } else { 1478 /* look again for more in a bit */ 1479 callout_schedule(&sc->sc_bulk_tmo, 1); 1480 sc->sc_bulk_send_next = state; 1481 } 1482 if (sc->sc_mbuf != NULL) 1483 pfsync_sendout(sc); 1484 splx(s); 1485 } 1486 1487 void 1488 pfsync_bulkfail(void *v) 1489 { 1490 struct pfsync_softc *sc = v; 1491 int s, error; 1492 1493 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 1494 /* Try again in a bit */ 1495 callout_schedule(&sc->sc_bulkfail_tmo, 5 * hz); 1496 s = splnet(); 1497 error = pfsync_request_update(NULL, NULL); 1498 if (error == ENOMEM) { 1499 if (pf_status.debug >= PF_DEBUG_MISC) 1500 printf("pfsync: cannot allocate mbufs for " 1501 "bulk update\n"); 1502 } else 1503 pfsync_sendout(sc); 1504 splx(s); 1505 } else { 1506 /* Pretend like the transfer was ok */ 1507 sc->sc_ureq_sent = 0; 1508 sc->sc_bulk_tries = 0; 1509 #if NCARP > 0 1510 if (!pfsync_sync_ok) 1511 carp_suppress_preempt --; 1512 #endif 1513 pfsync_sync_ok = 1; 1514 if (pf_status.debug >= PF_DEBUG_MISC) 1515 printf("pfsync: failed to receive " 1516 "bulk update status\n"); 1517 callout_stop(&sc->sc_bulkfail_tmo); 1518 } 1519 } 1520 1521 /* This must be called in splnet() */ 1522 int 1523 pfsync_sendout(struct pfsync_softc *sc) 1524 { 1525 struct ifnet *ifp = &sc->sc_if; 1526 struct mbuf *m; 1527 1528 callout_stop(&sc->sc_tmo); 1529 1530 if (sc->sc_mbuf == NULL) 1531 return (0); 1532 m = sc->sc_mbuf; 1533 sc->sc_mbuf = NULL; 1534 sc->sc_statep.s = NULL; 1535 1536 bpf_mtap(ifp, m); 1537 1538 if (sc->sc_mbuf_net) { 1539 m_freem(m); 1540 m = sc->sc_mbuf_net; 1541 sc->sc_mbuf_net = NULL; 1542 sc->sc_statep_net.s = NULL; 1543 } 1544 1545 return pfsync_sendout_mbuf(sc, m); 1546 } 1547 1548 int 1549 pfsync_tdb_sendout(struct pfsync_softc *sc) 1550 { 1551 struct ifnet *ifp = &sc->sc_if; 1552 struct mbuf *m; 1553 1554 callout_stop(&sc->sc_tdb_tmo); 1555 1556 if (sc->sc_mbuf_tdb == NULL) 1557 return (0); 1558 m = sc->sc_mbuf_tdb; 1559 sc->sc_mbuf_tdb = NULL; 1560 sc->sc_statep_tdb.t = NULL; 1561 1562 bpf_mtap(ifp, m); 1563 1564 return pfsync_sendout_mbuf(sc, m); 1565 } 1566 1567 int 1568 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m) 1569 { 1570 struct sockaddr sa; 1571 struct ip *ip; 1572 1573 if (sc->sc_sync_ifp || 1574 sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) { 1575 M_PREPEND(m, sizeof(struct ip), M_DONTWAIT); 1576 if (m == NULL) { 1577 PFSYNC_STATINC(PFSYNC_STAT_ONOMEM); 1578 return (0); 1579 } 1580 ip = mtod(m, struct ip *); 1581 ip->ip_v = IPVERSION; 1582 ip->ip_hl = sizeof(*ip) >> 2; 1583 ip->ip_tos = IPTOS_LOWDELAY; 1584 ip->ip_len = htons(m->m_pkthdr.len); 1585 ip->ip_id = htons(ip_randomid(ip_ids, 0)); 1586 ip->ip_off = htons(IP_DF); 1587 ip->ip_ttl = PFSYNC_DFLTTL; 1588 ip->ip_p = IPPROTO_PFSYNC; 1589 ip->ip_sum = 0; 1590 1591 memset(&sa, 0, sizeof(sa)); 1592 ip->ip_src.s_addr = INADDR_ANY; 1593 1594 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP) 1595 m->m_flags |= M_MCAST; 1596 ip->ip_dst = sc->sc_sendaddr; 1597 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr; 1598 1599 PFSYNC_STATINC(PFSYNC_STAT_OPACKETS); 1600 1601 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL)) { 1602 PFSYNC_STATINC(PFSYNC_STAT_OERRORS); 1603 } 1604 } else 1605 m_freem(m); 1606 1607 return (0); 1608 } 1609 1610 static int 1611 sysctl_net_inet_pfsync_stats(SYSCTLFN_ARGS) 1612 { 1613 1614 return (NETSTAT_SYSCTL(pfsyncstat_percpu, PFSYNC_NSTATS)); 1615 } 1616 1617 SYSCTL_SETUP(sysctl_net_inet_pfsync_setup, "sysctl net.inet.pfsync subtree setup") 1618 { 1619 1620 sysctl_createv(clog, 0, NULL, NULL, 1621 CTLFLAG_PERMANENT, 1622 CTLTYPE_NODE, "net", NULL, 1623 NULL, 0, NULL, 0, 1624 CTL_NET, CTL_EOL); 1625 sysctl_createv(clog, 0, NULL, NULL, 1626 CTLFLAG_PERMANENT, 1627 CTLTYPE_NODE, "inet", NULL, 1628 NULL, 0, NULL, 0, 1629 CTL_NET, PF_INET, CTL_EOL); 1630 sysctl_createv(clog, 0, NULL, NULL, 1631 CTLFLAG_PERMANENT, 1632 CTLTYPE_NODE, "pfsync", 1633 SYSCTL_DESCR("pfsync related settings"), 1634 NULL, 0, NULL, 0, 1635 CTL_NET, PF_INET, IPPROTO_PFSYNC, CTL_EOL); 1636 sysctl_createv(clog, 0, NULL, NULL, 1637 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1638 CTLTYPE_STRUCT, "stats", 1639 SYSCTL_DESCR("pfsync statistics"), 1640 sysctl_net_inet_pfsync_stats, 0, NULL, 0, 1641 CTL_NET, PF_INET, IPPROTO_PFSYNC, 1642 CTL_CREATE, CTL_EOL); 1643 } 1644