1 /* $NetBSD: if_pfsync.c,v 1.24 2026/01/11 16:41:43 christos 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 1.24 2026/01/11 16:41:43 christos 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 #include "ioconf.h" 89 90 percpu_t *pfsyncstat_percpu; 91 92 #define PFSYNC_STATINC(x) _NET_STATINC(pfsyncstat_percpu, x) 93 #endif /* __NetBSD__ */ 94 95 #include "pfsync.h" 96 97 #define PFSYNC_MINMTU \ 98 (sizeof(struct pfsync_header) + sizeof(struct pf_state)) 99 100 #ifdef PFSYNCDEBUG 101 #define DPRINTF(x) do { if (pfsyncdebug) printf x ; } while (0) 102 int pfsyncdebug; 103 #else 104 #define DPRINTF(x) 105 #endif 106 107 extern int ifqmaxlen; /* XXX */ 108 109 struct pfsync_softc *pfsyncif = NULL; 110 111 int pfsync_clone_create(struct if_clone *, int); 112 int pfsync_clone_destroy(struct ifnet *); 113 void pfsync_setmtu(struct pfsync_softc *, int); 114 int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, 115 struct pf_state_peer *); 116 int pfsync_insert_net_state(struct pfsync_state *, u_int8_t); 117 void pfsync_update_net_tdb(struct pfsync_tdb *); 118 int pfsyncoutput(struct ifnet *, struct mbuf *, const struct sockaddr *, 119 const struct rtentry *); 120 int pfsyncioctl(struct ifnet *, u_long, void*); 121 void pfsyncstart(struct ifnet *); 122 123 struct mbuf *pfsync_get_mbuf(struct pfsync_softc *, u_int8_t, void **); 124 int pfsync_request_update(struct pfsync_state_upd *, struct in_addr *); 125 int pfsync_sendout(struct pfsync_softc *); 126 int pfsync_tdb_sendout(struct pfsync_softc *); 127 int pfsync_sendout_mbuf(struct pfsync_softc *, struct mbuf *); 128 void pfsync_timeout(void *); 129 void pfsync_tdb_timeout(void *); 130 void pfsync_send_bus(struct pfsync_softc *, u_int8_t); 131 void pfsync_bulk_update(void *); 132 void pfsync_bulkfail(void *); 133 134 int pfsync_sync_ok; 135 136 struct if_clone pfsync_cloner = 137 IF_CLONE_INITIALIZER("pfsync", pfsync_clone_create, pfsync_clone_destroy); 138 139 void 140 pfsyncattach(int npfsync) 141 { 142 if_clone_attach(&pfsync_cloner); 143 144 pfsyncstat_percpu = percpu_alloc(sizeof(uint64_t) * PFSYNC_NSTATS); 145 } 146 147 int 148 pfsync_clone_create(struct if_clone *ifc, int unit) 149 { 150 struct ifnet *ifp; 151 152 if (unit != 0) 153 return (EINVAL); 154 155 pfsync_sync_ok = 1; 156 if ((pfsyncif = malloc(sizeof(*pfsyncif), M_DEVBUF, M_NOWAIT)) == NULL) 157 return (ENOMEM); 158 memset(pfsyncif, 0, sizeof(*pfsyncif)); 159 pfsyncif->sc_mbuf = NULL; 160 pfsyncif->sc_mbuf_net = NULL; 161 pfsyncif->sc_mbuf_tdb = NULL; 162 pfsyncif->sc_statep.s = NULL; 163 pfsyncif->sc_statep_net.s = NULL; 164 pfsyncif->sc_statep_tdb.t = NULL; 165 pfsyncif->sc_maxupdates = 128; 166 pfsyncif->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP; 167 pfsyncif->sc_sendaddr.s_addr = INADDR_PFSYNC_GROUP; 168 pfsyncif->sc_ureq_received = 0; 169 pfsyncif->sc_ureq_sent = 0; 170 pfsyncif->sc_bulk_send_next = NULL; 171 pfsyncif->sc_bulk_terminator = NULL; 172 ifp = &pfsyncif->sc_if; 173 snprintf(ifp->if_xname, sizeof ifp->if_xname, "pfsync%d", unit); 174 ifp->if_softc = pfsyncif; 175 ifp->if_ioctl = pfsyncioctl; 176 ifp->if_output = pfsyncoutput; 177 ifp->if_start = pfsyncstart; 178 ifp->if_type = IFT_PFSYNC; 179 ifp->if_snd.ifq_maxlen = ifqmaxlen; 180 ifp->if_hdrlen = PFSYNC_HDRLEN; 181 pfsync_setmtu(pfsyncif, ETHERMTU); 182 183 callout_init(&pfsyncif->sc_tmo, 0); 184 callout_init(&pfsyncif->sc_tdb_tmo, 0); 185 callout_init(&pfsyncif->sc_bulk_tmo, 0); 186 callout_init(&pfsyncif->sc_bulkfail_tmo, 0); 187 callout_setfunc(&pfsyncif->sc_tmo, pfsync_timeout, pfsyncif); 188 callout_setfunc(&pfsyncif->sc_tdb_tmo, pfsync_tdb_timeout, pfsyncif); 189 callout_setfunc(&pfsyncif->sc_bulk_tmo, pfsync_bulk_update, pfsyncif); 190 callout_setfunc(&pfsyncif->sc_bulkfail_tmo, pfsync_bulkfail, pfsyncif); 191 192 if_attach(ifp); 193 if_alloc_sadl(ifp); 194 195 bpf_attach(&pfsyncif->sc_if, DLT_PFSYNC, PFSYNC_HDRLEN); 196 197 return (0); 198 } 199 200 int 201 pfsync_clone_destroy(struct ifnet *ifp) 202 { 203 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 (strnlen(sp->ifname, IFNAMSIZ) == IFNAMSIZ) 255 return (EINVAL); 256 257 if (sp->creatorid == 0 && pf_status.debug >= PF_DEBUG_MISC) { 258 printf("pfsync_insert_net_state: invalid creator id:" 259 " %08x\n", ntohl(sp->creatorid)); 260 return (EINVAL); 261 } 262 263 kif = pfi_kif_get(sp->ifname); 264 if (kif == NULL) { 265 if (pf_status.debug >= PF_DEBUG_MISC) 266 printf("pfsync_insert_net_state: " 267 "unknown interface: %s\n", sp->ifname); 268 /* skip this state */ 269 return (0); 270 } 271 272 /* 273 * If the ruleset checksums match, it's safe to associate the state 274 * with the rule of that number. 275 */ 276 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && chksum_flag && 277 ntohl(sp->rule) < 278 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 279 r = pf_main_ruleset.rules[ 280 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 281 else 282 r = &pf_default_rule; 283 284 if (!r->max_states || r->states < r->max_states) 285 st = pool_get(&pf_state_pl, PR_NOWAIT); 286 if (st == NULL) { 287 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 288 return (ENOMEM); 289 } 290 memset(st, 0, sizeof(*st)); 291 292 if ((sk = pf_alloc_state_key(st)) == NULL) { 293 pool_put(&pf_state_pl, st); 294 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 295 return (ENOMEM); 296 } 297 298 /* allocate memory for scrub info */ 299 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 300 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) { 301 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 302 if (st->src.scrub) 303 pool_put(&pf_state_scrub_pl, st->src.scrub); 304 pool_put(&pf_state_pl, st); 305 pool_put(&pf_state_key_pl, sk); 306 return (ENOMEM); 307 } 308 309 st->rule.ptr = r; 310 /* XXX get pointers to nat_rule and anchor */ 311 312 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 313 r->states++; 314 315 /* fill in the rest of the state entry */ 316 pf_state_host_ntoh(&sp->lan, &sk->lan); 317 pf_state_host_ntoh(&sp->gwy, &sk->gwy); 318 pf_state_host_ntoh(&sp->ext, &sk->ext); 319 320 pf_state_peer_ntoh(&sp->src, &st->src); 321 pf_state_peer_ntoh(&sp->dst, &st->dst); 322 323 memcpy(&st->rt_addr, &sp->rt_addr, sizeof(st->rt_addr)); 324 st->creation = time_second - ntohl(sp->creation); 325 st->expire = ntohl(sp->expire) + time_second; 326 327 sk->af = sp->af; 328 sk->proto = sp->proto; 329 sk->direction = sp->direction; 330 st->log = sp->log; 331 st->timeout = sp->timeout; 332 st->allow_opts = sp->allow_opts; 333 334 memcpy(&st->id, sp->id, sizeof(st->id)); 335 st->creatorid = sp->creatorid; 336 st->sync_flags = PFSTATE_FROMSYNC; 337 338 if (pf_insert_state(kif, st)) { 339 pfi_kif_unref(kif, PFI_KIF_REF_NONE); 340 /* XXX when we have nat_rule/anchors, use STATE_DEC_COUNTERS */ 341 r->states--; 342 if (st->dst.scrub) 343 pool_put(&pf_state_scrub_pl, st->dst.scrub); 344 if (st->src.scrub) 345 pool_put(&pf_state_scrub_pl, st->src.scrub); 346 pool_put(&pf_state_pl, st); 347 return (EINVAL); 348 } 349 350 return (0); 351 } 352 353 void 354 pfsync_input(struct mbuf *m, int off, int proto) 355 { 356 struct ip *ip = mtod(m, struct ip *); 357 struct pfsync_header *ph; 358 struct pfsync_softc *sc = pfsyncif; 359 struct pf_state *st; 360 struct pf_state_key *sk; 361 struct pf_state_cmp id_key; 362 struct pfsync_state *sp; 363 struct pfsync_state_upd *up; 364 struct pfsync_state_del *dp; 365 struct pfsync_state_clr *cp; 366 struct pfsync_state_upd_req *rup; 367 struct pfsync_state_bus *bus; 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->if_index != m->m_pkthdr.rcvif_index) { 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 } 854 855 done: 856 m_freem(m); 857 } 858 859 int 860 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, 861 const struct rtentry *rt) 862 { 863 m_freem(m); 864 return (0); 865 } 866 867 /* ARGSUSED */ 868 int 869 pfsyncioctl(struct ifnet *ifp, u_long cmd, void* data) 870 { 871 struct lwp *l = curlwp; 872 struct pfsync_softc *sc = ifp->if_softc; 873 struct ifreq *ifr = (struct ifreq *)data; 874 struct ip_moptions *imo = &sc->sc_imo; 875 struct pfsyncreq pfsyncr; 876 struct ifnet *sifp; 877 int s, error; 878 879 switch (cmd) { 880 case SIOCSIFADDR: 881 case SIOCAIFADDR: 882 case SIOCSIFDSTADDR: 883 case SIOCSIFFLAGS: 884 if (ifp->if_flags & IFF_UP) 885 ifp->if_flags |= IFF_RUNNING; 886 else 887 ifp->if_flags &= ~IFF_RUNNING; 888 break; 889 case SIOCSIFMTU: 890 if (ifr->ifr_mtu < PFSYNC_MINMTU) 891 return (EINVAL); 892 if (ifr->ifr_mtu > MCLBYTES) 893 ifr->ifr_mtu = MCLBYTES; 894 s = splnet(); 895 if (ifr->ifr_mtu < ifp->if_mtu) 896 pfsync_sendout(sc); 897 pfsync_setmtu(sc, ifr->ifr_mtu); 898 splx(s); 899 break; 900 case SIOCGETPFSYNC: 901 if ((error = kauth_authorize_network(l->l_cred, 902 KAUTH_NETWORK_INTERFACE, 903 KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, ifp, (void *)cmd, 904 NULL)) != 0) 905 return (error); 906 memset(&pfsyncr, 0, sizeof(pfsyncr)); 907 if (sc->sc_sync_ifp) 908 strlcpy(pfsyncr.pfsyncr_syncdev, 909 sc->sc_sync_ifp->if_xname, IFNAMSIZ); 910 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 911 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 912 if ((error = copyout(&pfsyncr, ifr->ifr_data, sizeof(pfsyncr)))) 913 return (error); 914 break; 915 case SIOCSETPFSYNC: 916 if ((error = kauth_authorize_network(l->l_cred, 917 KAUTH_NETWORK_INTERFACE, 918 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 919 NULL)) != 0) 920 return (error); 921 if ((error = copyin(ifr->ifr_data, &pfsyncr, sizeof(pfsyncr)))) 922 return (error); 923 924 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 925 sc->sc_sync_peer.s_addr = INADDR_PFSYNC_GROUP; 926 else 927 sc->sc_sync_peer.s_addr = 928 pfsyncr.pfsyncr_syncpeer.s_addr; 929 930 if (pfsyncr.pfsyncr_maxupdates > 255) 931 return (EINVAL); 932 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 933 934 if (pfsyncr.pfsyncr_syncdev[0] == 0) { 935 sc->sc_sync_ifp = NULL; 936 if (sc->sc_mbuf_net != NULL) { 937 /* Don't keep stale pfsync packets around. */ 938 s = splnet(); 939 m_freem(sc->sc_mbuf_net); 940 sc->sc_mbuf_net = NULL; 941 sc->sc_statep_net.s = NULL; 942 splx(s); 943 } 944 if (imo->imo_num_memberships > 0) { 945 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 946 imo->imo_multicast_if_index = 0; 947 } 948 break; 949 } 950 951 if ((sifp = ifunit(pfsyncr.pfsyncr_syncdev)) == NULL) 952 return (EINVAL); 953 954 s = splnet(); 955 if (sifp->if_mtu < sc->sc_if.if_mtu || 956 (sc->sc_sync_ifp != NULL && 957 sifp->if_mtu < sc->sc_sync_ifp->if_mtu) || 958 sifp->if_mtu < MCLBYTES - sizeof(struct ip)) 959 pfsync_sendout(sc); 960 sc->sc_sync_ifp = sifp; 961 962 pfsync_setmtu(sc, sc->sc_if.if_mtu); 963 964 if (imo->imo_num_memberships > 0) { 965 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 966 imo->imo_multicast_if_index = 0; 967 } 968 969 if (sc->sc_sync_ifp && 970 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 971 struct in_addr addr; 972 973 if (!(sc->sc_sync_ifp->if_flags & IFF_MULTICAST)) { 974 sc->sc_sync_ifp = NULL; 975 splx(s); 976 return (EADDRNOTAVAIL); 977 } 978 979 addr.s_addr = INADDR_PFSYNC_GROUP; 980 981 if ((imo->imo_membership[0] = 982 in_addmulti(&addr, sc->sc_sync_ifp)) == NULL) { 983 sc->sc_sync_ifp = NULL; 984 splx(s); 985 return (ENOBUFS); 986 } 987 imo->imo_num_memberships++; 988 imo->imo_multicast_if_index = if_get_index(sc->sc_sync_ifp); 989 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 990 imo->imo_multicast_loop = 0; 991 } 992 993 if (sc->sc_sync_ifp || 994 sc->sc_sendaddr.s_addr != INADDR_PFSYNC_GROUP) { 995 /* Request a full state table update. */ 996 sc->sc_ureq_sent = time_uptime; 997 #if NCARP > 0 998 if (pfsync_sync_ok) 999 carp_suppress_preempt ++; 1000 #endif 1001 pfsync_sync_ok = 0; 1002 if (pf_status.debug >= PF_DEBUG_MISC) 1003 printf("pfsync: requesting bulk update\n"); 1004 callout_schedule(&sc->sc_bulkfail_tmo, 5 * hz); 1005 error = pfsync_request_update(NULL, NULL); 1006 if (error == ENOMEM) { 1007 splx(s); 1008 return (ENOMEM); 1009 } 1010 pfsync_sendout(sc); 1011 } 1012 splx(s); 1013 1014 break; 1015 1016 default: 1017 return ifioctl_common(ifp, cmd, data); 1018 } 1019 1020 return (0); 1021 } 1022 1023 void 1024 pfsync_setmtu(struct pfsync_softc *sc, int mtu_req) 1025 { 1026 int mtu; 1027 1028 if (sc->sc_sync_ifp && sc->sc_sync_ifp->if_mtu < mtu_req) 1029 mtu = sc->sc_sync_ifp->if_mtu; 1030 else 1031 mtu = mtu_req; 1032 1033 sc->sc_maxcount = (mtu - sizeof(struct pfsync_header)) / 1034 sizeof(struct pfsync_state); 1035 if (sc->sc_maxcount > 254) 1036 sc->sc_maxcount = 254; 1037 sc->sc_if.if_mtu = sizeof(struct pfsync_header) + 1038 sc->sc_maxcount * sizeof(struct pfsync_state); 1039 } 1040 1041 struct mbuf * 1042 pfsync_get_mbuf(struct pfsync_softc *sc, u_int8_t action, void **sp) 1043 { 1044 struct pfsync_header *h; 1045 struct mbuf *m; 1046 int len; 1047 1048 MGETHDR(m, M_DONTWAIT, MT_DATA); 1049 if (m == NULL) { 1050 #ifdef __NetBSD__ 1051 if_statinc(&sc->sc_if, if_oerrors); 1052 #else 1053 sc->sc_if.if_oerrors++; 1054 #endif /* __NetBSD__ */ 1055 return (NULL); 1056 } 1057 1058 switch (action) { 1059 case PFSYNC_ACT_CLR: 1060 len = sizeof(struct pfsync_header) + 1061 sizeof(struct pfsync_state_clr); 1062 break; 1063 case PFSYNC_ACT_UPD_C: 1064 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd)) + 1065 sizeof(struct pfsync_header); 1066 break; 1067 case PFSYNC_ACT_DEL_C: 1068 len = (sc->sc_maxcount * sizeof(struct pfsync_state_del)) + 1069 sizeof(struct pfsync_header); 1070 break; 1071 case PFSYNC_ACT_UREQ: 1072 len = (sc->sc_maxcount * sizeof(struct pfsync_state_upd_req)) + 1073 sizeof(struct pfsync_header); 1074 break; 1075 case PFSYNC_ACT_BUS: 1076 len = sizeof(struct pfsync_header) + 1077 sizeof(struct pfsync_state_bus); 1078 break; 1079 case PFSYNC_ACT_TDB_UPD: 1080 len = (sc->sc_maxcount * sizeof(struct pfsync_tdb)) + 1081 sizeof(struct pfsync_header); 1082 break; 1083 default: 1084 len = (sc->sc_maxcount * sizeof(struct pfsync_state)) + 1085 sizeof(struct pfsync_header); 1086 break; 1087 } 1088 1089 if (len > MHLEN) { 1090 MCLGET(m, M_DONTWAIT); 1091 if ((m->m_flags & M_EXT) == 0) { 1092 m_free(m); 1093 #ifdef __NetBSD__ 1094 if_statinc(&sc->sc_if, if_oerrors); 1095 #else 1096 sc->sc_if.if_oerrors++; 1097 #endif /* __NetBSD__ */ 1098 return (NULL); 1099 } 1100 m->m_data += (MCLBYTES - len) &~ (sizeof(long) - 1); 1101 } else 1102 m_align(m, len); 1103 1104 m_reset_rcvif(m); 1105 m->m_pkthdr.len = m->m_len = sizeof(struct pfsync_header); 1106 h = mtod(m, struct pfsync_header *); 1107 h->version = PFSYNC_VERSION; 1108 h->af = 0; 1109 h->count = 0; 1110 h->action = action; 1111 if (action != PFSYNC_ACT_TDB_UPD) 1112 memcpy(&h->pf_chksum, &pf_status.pf_chksum, 1113 PF_MD5_DIGEST_LENGTH); 1114 1115 *sp = (void *)((char *)h + PFSYNC_HDRLEN); 1116 if (action == PFSYNC_ACT_TDB_UPD) 1117 callout_schedule(&sc->sc_tdb_tmo, hz); 1118 else 1119 callout_schedule(&sc->sc_tmo, hz); 1120 return (m); 1121 } 1122 1123 int 1124 pfsync_pack_state(u_int8_t action, struct pf_state *st, int flags) 1125 { 1126 struct ifnet *ifp = NULL; 1127 struct pfsync_softc *sc = pfsyncif; 1128 struct pfsync_header *h, *h_net; 1129 struct pfsync_state *sp = NULL; 1130 struct pfsync_state_upd *up = NULL; 1131 struct pfsync_state_del *dp = NULL; 1132 struct pf_state_key *sk = st->state_key; 1133 struct pf_rule *r; 1134 u_long secs; 1135 int s, ret = 0; 1136 u_int8_t i = 255, newaction = 0; 1137 1138 if (sc == NULL) 1139 return (0); 1140 ifp = &sc->sc_if; 1141 1142 /* 1143 * If a packet falls in the forest and there's nobody around to 1144 * hear, does it make a sound? 1145 */ 1146 if (ifp->if_bpf == NULL && sc->sc_sync_ifp == NULL && 1147 sc->sc_sync_peer.s_addr == INADDR_PFSYNC_GROUP) { 1148 /* Don't leave any stale pfsync packets hanging around. */ 1149 if (sc->sc_mbuf != NULL) { 1150 m_freem(sc->sc_mbuf); 1151 sc->sc_mbuf = NULL; 1152 sc->sc_statep.s = NULL; 1153 } 1154 return (0); 1155 } 1156 1157 if (action >= PFSYNC_ACT_MAX) 1158 return (EINVAL); 1159 1160 s = splnet(); 1161 if (sc->sc_mbuf == NULL) { 1162 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1163 (void *)&sc->sc_statep.s)) == NULL) { 1164 splx(s); 1165 return (ENOMEM); 1166 } 1167 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1168 } else { 1169 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1170 if (h->action != action) { 1171 pfsync_sendout(sc); 1172 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, action, 1173 (void *)&sc->sc_statep.s)) == NULL) { 1174 splx(s); 1175 return (ENOMEM); 1176 } 1177 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1178 } else { 1179 /* 1180 * If it's an update, look in the packet to see if 1181 * we already have an update for the state. 1182 */ 1183 if (action == PFSYNC_ACT_UPD && sc->sc_maxupdates) { 1184 struct pfsync_state *usp = 1185 (void *)((char *)h + PFSYNC_HDRLEN); 1186 1187 for (i = 0; i < h->count; i++) { 1188 if (!memcmp(usp->id, &st->id, 1189 PFSYNC_ID_LEN) && 1190 usp->creatorid == st->creatorid) { 1191 sp = usp; 1192 sp->updates++; 1193 break; 1194 } 1195 usp++; 1196 } 1197 } 1198 } 1199 } 1200 1201 secs = time_second; 1202 1203 st->pfsync_time = time_uptime; 1204 1205 if (sp == NULL) { 1206 /* not a "duplicate" update */ 1207 i = 255; 1208 sp = sc->sc_statep.s++; 1209 sc->sc_mbuf->m_pkthdr.len = 1210 sc->sc_mbuf->m_len += sizeof(struct pfsync_state); 1211 h->count++; 1212 memset(sp, 0, sizeof(*sp)); 1213 1214 memcpy(sp->id, &st->id, sizeof(sp->id)); 1215 sp->creatorid = st->creatorid; 1216 1217 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); 1218 pf_state_host_hton(&sk->lan, &sp->lan); 1219 pf_state_host_hton(&sk->gwy, &sp->gwy); 1220 pf_state_host_hton(&sk->ext, &sp->ext); 1221 1222 memcpy(&sp->rt_addr, &st->rt_addr, sizeof(sp->rt_addr)); 1223 1224 sp->creation = htonl(secs - st->creation); 1225 pf_state_counter_hton(st->packets[0], sp->packets[0]); 1226 pf_state_counter_hton(st->packets[1], sp->packets[1]); 1227 pf_state_counter_hton(st->bytes[0], sp->bytes[0]); 1228 pf_state_counter_hton(st->bytes[1], sp->bytes[1]); 1229 if ((r = st->rule.ptr) == NULL) 1230 sp->rule = htonl(-1); 1231 else 1232 sp->rule = htonl(r->nr); 1233 if ((r = st->anchor.ptr) == NULL) 1234 sp->anchor = htonl(-1); 1235 else 1236 sp->anchor = htonl(r->nr); 1237 sp->af = sk->af; 1238 sp->proto = sk->proto; 1239 sp->direction = sk->direction; 1240 sp->log = st->log; 1241 sp->allow_opts = st->allow_opts; 1242 sp->timeout = st->timeout; 1243 1244 if (flags & PFSYNC_FLAG_STALE) 1245 sp->sync_flags |= PFSTATE_STALE; 1246 } 1247 1248 pf_state_peer_hton(&st->src, &sp->src); 1249 pf_state_peer_hton(&st->dst, &sp->dst); 1250 1251 if (st->expire <= secs) 1252 sp->expire = htonl(0); 1253 else 1254 sp->expire = htonl(st->expire - secs); 1255 1256 /* do we need to build "compressed" actions for network transfer? */ 1257 if (sc->sc_sync_ifp && flags & PFSYNC_FLAG_COMPRESS) { 1258 switch (action) { 1259 case PFSYNC_ACT_UPD: 1260 newaction = PFSYNC_ACT_UPD_C; 1261 break; 1262 case PFSYNC_ACT_DEL: 1263 newaction = PFSYNC_ACT_DEL_C; 1264 break; 1265 default: 1266 /* by default we just send the uncompressed states */ 1267 break; 1268 } 1269 } 1270 1271 if (newaction) { 1272 if (sc->sc_mbuf_net == NULL) { 1273 if ((sc->sc_mbuf_net = pfsync_get_mbuf(sc, newaction, 1274 (void *)&sc->sc_statep_net.s)) == NULL) { 1275 splx(s); 1276 return (ENOMEM); 1277 } 1278 } 1279 h_net = mtod(sc->sc_mbuf_net, struct pfsync_header *); 1280 1281 switch (newaction) { 1282 case PFSYNC_ACT_UPD_C: 1283 if (i != 255) { 1284 up = (void *)((char *)h_net + 1285 PFSYNC_HDRLEN + (i * sizeof(*up))); 1286 up->updates++; 1287 } else { 1288 h_net->count++; 1289 sc->sc_mbuf_net->m_pkthdr.len = 1290 sc->sc_mbuf_net->m_len += sizeof(*up); 1291 up = sc->sc_statep_net.u++; 1292 1293 memset(up, 0, sizeof(*up)); 1294 memcpy(up->id, &st->id, sizeof(up->id)); 1295 up->creatorid = st->creatorid; 1296 } 1297 up->timeout = st->timeout; 1298 up->expire = sp->expire; 1299 up->src = sp->src; 1300 up->dst = sp->dst; 1301 break; 1302 case PFSYNC_ACT_DEL_C: 1303 sc->sc_mbuf_net->m_pkthdr.len = 1304 sc->sc_mbuf_net->m_len += sizeof(*dp); 1305 dp = sc->sc_statep_net.d++; 1306 h_net->count++; 1307 1308 memset(dp, 0, sizeof(*dp)); 1309 memcpy(dp->id, &st->id, sizeof(dp->id)); 1310 dp->creatorid = st->creatorid; 1311 break; 1312 } 1313 } 1314 1315 if (h->count == sc->sc_maxcount || 1316 (sc->sc_maxupdates && (sp->updates >= sc->sc_maxupdates))) 1317 ret = pfsync_sendout(sc); 1318 1319 splx(s); 1320 return (ret); 1321 } 1322 1323 /* This must be called in splnet() */ 1324 int 1325 pfsync_request_update(struct pfsync_state_upd *up, struct in_addr *src) 1326 { 1327 struct pfsync_header *h; 1328 struct pfsync_softc *sc = pfsyncif; 1329 struct pfsync_state_upd_req *rup; 1330 int ret = 0; 1331 1332 if (sc == NULL) 1333 return (0); 1334 1335 if (sc->sc_mbuf == NULL) { 1336 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1337 (void *)&sc->sc_statep.s)) == NULL) 1338 return (ENOMEM); 1339 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1340 } else { 1341 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1342 if (h->action != PFSYNC_ACT_UREQ) { 1343 pfsync_sendout(sc); 1344 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_UREQ, 1345 (void *)&sc->sc_statep.s)) == NULL) 1346 return (ENOMEM); 1347 h = mtod(sc->sc_mbuf, struct pfsync_header *); 1348 } 1349 } 1350 1351 if (src != NULL) 1352 sc->sc_sendaddr = *src; 1353 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*rup); 1354 h->count++; 1355 rup = sc->sc_statep.r++; 1356 memset(rup, 0, sizeof(*rup)); 1357 if (up != NULL) { 1358 memcpy(rup->id, up->id, sizeof(rup->id)); 1359 rup->creatorid = up->creatorid; 1360 } 1361 1362 if (h->count == sc->sc_maxcount) 1363 ret = pfsync_sendout(sc); 1364 1365 return (ret); 1366 } 1367 1368 int 1369 pfsync_clear_states(u_int32_t creatorid, char *ifname) 1370 { 1371 struct pfsync_softc *sc = pfsyncif; 1372 struct pfsync_state_clr *cp; 1373 int s, ret; 1374 1375 if (sc == NULL) 1376 return (0); 1377 1378 s = splnet(); 1379 if (sc->sc_mbuf != NULL) 1380 pfsync_sendout(sc); 1381 if ((sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_CLR, 1382 (void *)&sc->sc_statep.c)) == NULL) { 1383 splx(s); 1384 return (ENOMEM); 1385 } 1386 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*cp); 1387 cp = sc->sc_statep.c; 1388 cp->creatorid = creatorid; 1389 if (ifname != NULL) 1390 strlcpy(cp->ifname, ifname, IFNAMSIZ); 1391 1392 ret = (pfsync_sendout(sc)); 1393 splx(s); 1394 return (ret); 1395 } 1396 1397 void 1398 pfsync_timeout(void *v) 1399 { 1400 struct pfsync_softc *sc = v; 1401 int s; 1402 1403 s = splnet(); 1404 pfsync_sendout(sc); 1405 splx(s); 1406 } 1407 1408 void 1409 pfsync_tdb_timeout(void *v) 1410 { 1411 struct pfsync_softc *sc = v; 1412 int s; 1413 1414 s = splnet(); 1415 pfsync_tdb_sendout(sc); 1416 splx(s); 1417 } 1418 1419 /* This must be called in splnet() */ 1420 void 1421 pfsync_send_bus(struct pfsync_softc *sc, u_int8_t status) 1422 { 1423 struct pfsync_state_bus *bus; 1424 1425 if (sc->sc_mbuf != NULL) 1426 pfsync_sendout(sc); 1427 1428 if (pfsync_sync_ok && 1429 (sc->sc_mbuf = pfsync_get_mbuf(sc, PFSYNC_ACT_BUS, 1430 (void *)&sc->sc_statep.b)) != NULL) { 1431 sc->sc_mbuf->m_pkthdr.len = sc->sc_mbuf->m_len += sizeof(*bus); 1432 bus = sc->sc_statep.b; 1433 bus->creatorid = pf_status.hostid; 1434 bus->status = status; 1435 bus->endtime = htonl(time_uptime - sc->sc_ureq_received); 1436 pfsync_sendout(sc); 1437 } 1438 } 1439 1440 void 1441 pfsync_bulk_update(void *v) 1442 { 1443 struct pfsync_softc *sc = v; 1444 int s, i = 0; 1445 struct pf_state *state; 1446 1447 s = splnet(); 1448 if (sc->sc_mbuf != NULL) 1449 pfsync_sendout(sc); 1450 1451 /* 1452 * Grab at most PFSYNC_BULKPACKETS worth of states which have not 1453 * been sent since the latest request was made. 1454 */ 1455 state = sc->sc_bulk_send_next; 1456 if (state) 1457 do { 1458 /* send state update if syncable and not already sent */ 1459 if (!state->sync_flags 1460 && state->timeout < PFTM_MAX 1461 && state->pfsync_time <= sc->sc_ureq_received) { 1462 pfsync_pack_state(PFSYNC_ACT_UPD, state, 0); 1463 i++; 1464 } 1465 1466 /* figure next state to send */ 1467 state = TAILQ_NEXT(state, entry_list); 1468 1469 /* wrap to start of list if we hit the end */ 1470 if (!state) 1471 state = TAILQ_FIRST(&state_list); 1472 } while (i < sc->sc_maxcount * PFSYNC_BULKPACKETS && 1473 state != sc->sc_bulk_terminator); 1474 1475 if (!state || state == sc->sc_bulk_terminator) { 1476 /* we're done */ 1477 pfsync_send_bus(sc, PFSYNC_BUS_END); 1478 sc->sc_ureq_received = 0; 1479 sc->sc_bulk_send_next = NULL; 1480 sc->sc_bulk_terminator = NULL; 1481 callout_stop(&sc->sc_bulk_tmo); 1482 if (pf_status.debug >= PF_DEBUG_MISC) 1483 printf("pfsync: bulk update complete\n"); 1484 } else { 1485 /* look again for more in a bit */ 1486 callout_schedule(&sc->sc_bulk_tmo, 1); 1487 sc->sc_bulk_send_next = state; 1488 } 1489 if (sc->sc_mbuf != NULL) 1490 pfsync_sendout(sc); 1491 splx(s); 1492 } 1493 1494 void 1495 pfsync_bulkfail(void *v) 1496 { 1497 struct pfsync_softc *sc = v; 1498 int s, error; 1499 1500 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 1501 /* Try again in a bit */ 1502 callout_schedule(&sc->sc_bulkfail_tmo, 5 * hz); 1503 s = splnet(); 1504 error = pfsync_request_update(NULL, NULL); 1505 if (error == ENOMEM) { 1506 if (pf_status.debug >= PF_DEBUG_MISC) 1507 printf("pfsync: cannot allocate mbufs for " 1508 "bulk update\n"); 1509 } else 1510 pfsync_sendout(sc); 1511 splx(s); 1512 } else { 1513 /* Pretend like the transfer was ok */ 1514 sc->sc_ureq_sent = 0; 1515 sc->sc_bulk_tries = 0; 1516 #if NCARP > 0 1517 if (!pfsync_sync_ok) 1518 carp_suppress_preempt --; 1519 #endif 1520 pfsync_sync_ok = 1; 1521 if (pf_status.debug >= PF_DEBUG_MISC) 1522 printf("pfsync: failed to receive " 1523 "bulk update status\n"); 1524 callout_stop(&sc->sc_bulkfail_tmo); 1525 } 1526 } 1527 1528 /* This must be called in splnet() */ 1529 int 1530 pfsync_sendout(struct pfsync_softc *sc) 1531 { 1532 struct ifnet *ifp = &sc->sc_if; 1533 struct mbuf *m; 1534 1535 callout_stop(&sc->sc_tmo); 1536 1537 if (sc->sc_mbuf == NULL) 1538 return (0); 1539 m = sc->sc_mbuf; 1540 sc->sc_mbuf = NULL; 1541 sc->sc_statep.s = NULL; 1542 1543 bpf_mtap(ifp, m, BPF_D_OUT); 1544 1545 if (sc->sc_mbuf_net) { 1546 m_freem(m); 1547 m = sc->sc_mbuf_net; 1548 sc->sc_mbuf_net = NULL; 1549 sc->sc_statep_net.s = NULL; 1550 } 1551 1552 return pfsync_sendout_mbuf(sc, m); 1553 } 1554 1555 int 1556 pfsync_tdb_sendout(struct pfsync_softc *sc) 1557 { 1558 struct ifnet *ifp = &sc->sc_if; 1559 struct mbuf *m; 1560 1561 callout_stop(&sc->sc_tdb_tmo); 1562 1563 if (sc->sc_mbuf_tdb == NULL) 1564 return (0); 1565 m = sc->sc_mbuf_tdb; 1566 sc->sc_mbuf_tdb = NULL; 1567 sc->sc_statep_tdb.t = NULL; 1568 1569 bpf_mtap(ifp, m, BPF_D_OUT); 1570 1571 return pfsync_sendout_mbuf(sc, m); 1572 } 1573 1574 int 1575 pfsync_sendout_mbuf(struct pfsync_softc *sc, struct mbuf *m) 1576 { 1577 struct sockaddr sa; 1578 struct ip *ip; 1579 1580 if (sc->sc_sync_ifp || 1581 sc->sc_sync_peer.s_addr != INADDR_PFSYNC_GROUP) { 1582 M_PREPEND(m, sizeof(struct ip), M_DONTWAIT); 1583 if (m == NULL) { 1584 PFSYNC_STATINC(PFSYNC_STAT_ONOMEM); 1585 return (0); 1586 } 1587 ip = mtod(m, struct ip *); 1588 ip->ip_v = IPVERSION; 1589 ip->ip_hl = sizeof(*ip) >> 2; 1590 ip->ip_tos = IPTOS_LOWDELAY; 1591 ip->ip_len = htons(m->m_pkthdr.len); 1592 ip->ip_id = ip_randomid(); 1593 ip->ip_off = htons(IP_DF); 1594 ip->ip_ttl = PFSYNC_DFLTTL; 1595 ip->ip_p = IPPROTO_PFSYNC; 1596 ip->ip_sum = 0; 1597 1598 memset(&sa, 0, sizeof(sa)); 1599 ip->ip_src.s_addr = INADDR_ANY; 1600 1601 if (sc->sc_sendaddr.s_addr == INADDR_PFSYNC_GROUP) 1602 m->m_flags |= M_MCAST; 1603 ip->ip_dst = sc->sc_sendaddr; 1604 sc->sc_sendaddr.s_addr = sc->sc_sync_peer.s_addr; 1605 1606 PFSYNC_STATINC(PFSYNC_STAT_OPACKETS); 1607 1608 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL)) { 1609 PFSYNC_STATINC(PFSYNC_STAT_OERRORS); 1610 } 1611 } else 1612 m_freem(m); 1613 1614 return (0); 1615 } 1616 1617 static int 1618 sysctl_net_inet_pfsync_stats(SYSCTLFN_ARGS) 1619 { 1620 1621 return (NETSTAT_SYSCTL(pfsyncstat_percpu, PFSYNC_NSTATS)); 1622 } 1623 1624 SYSCTL_SETUP(sysctl_net_inet_pfsync_setup, "sysctl net.inet.pfsync subtree setup") 1625 { 1626 1627 sysctl_createv(clog, 0, NULL, NULL, 1628 CTLFLAG_PERMANENT, 1629 CTLTYPE_NODE, "net", NULL, 1630 NULL, 0, NULL, 0, 1631 CTL_NET, CTL_EOL); 1632 sysctl_createv(clog, 0, NULL, NULL, 1633 CTLFLAG_PERMANENT, 1634 CTLTYPE_NODE, "inet", NULL, 1635 NULL, 0, NULL, 0, 1636 CTL_NET, PF_INET, CTL_EOL); 1637 sysctl_createv(clog, 0, NULL, NULL, 1638 CTLFLAG_PERMANENT, 1639 CTLTYPE_NODE, "pfsync", 1640 SYSCTL_DESCR("pfsync related settings"), 1641 NULL, 0, NULL, 0, 1642 CTL_NET, PF_INET, IPPROTO_PFSYNC, CTL_EOL); 1643 sysctl_createv(clog, 0, NULL, NULL, 1644 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1645 CTLTYPE_STRUCT, "stats", 1646 SYSCTL_DESCR("pfsync statistics"), 1647 sysctl_net_inet_pfsync_stats, 0, NULL, 0, 1648 CTL_NET, PF_INET, IPPROTO_PFSYNC, 1649 CTL_CREATE, CTL_EOL); 1650 } 1651
Indexes created Sat Feb 21 16:20:20 UTC 2026