1 /*- 2 * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND ISC) 3 * 4 * Copyright (c) 2002 Michael Shalayeff 5 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 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 /*- 31 * Copyright (c) 2009 David Gwynne <dlg@openbsd.org> 32 * 33 * Permission to use, copy, modify, and distribute this software for any 34 * purpose with or without fee is hereby granted, provided that the above 35 * copyright notice and this permission notice appear in all copies. 36 * 37 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 38 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 39 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 40 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 41 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 42 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 43 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 44 */ 45 46 /* 47 * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $ 48 * 49 * Revisions picked from OpenBSD after revision 1.110 import: 50 * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input() 51 * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates 52 * 1.120, 1.175 - use monotonic time_uptime 53 * 1.122 - reduce number of updates for non-TCP sessions 54 * 1.125, 1.127 - rewrite merge or stale processing 55 * 1.128 - cleanups 56 * 1.146 - bzero() mbuf before sparsely filling it with data 57 * 1.170 - SIOCSIFMTU checks 58 * 1.126, 1.142 - deferred packets processing 59 * 1.173 - correct expire time processing 60 */ 61 62 #include <sys/cdefs.h> 63 __FBSDID("$FreeBSD$"); 64 65 #include "opt_inet.h" 66 #include "opt_inet6.h" 67 #include "opt_pf.h" 68 69 #include <sys/param.h> 70 #include <sys/bus.h> 71 #include <sys/endian.h> 72 #include <sys/interrupt.h> 73 #include <sys/kernel.h> 74 #include <sys/lock.h> 75 #include <sys/mbuf.h> 76 #include <sys/module.h> 77 #include <sys/mutex.h> 78 #include <sys/priv.h> 79 #include <sys/protosw.h> 80 #include <sys/smp.h> 81 #include <sys/socket.h> 82 #include <sys/sockio.h> 83 #include <sys/sysctl.h> 84 #include <sys/syslog.h> 85 86 #include <net/bpf.h> 87 #include <net/if.h> 88 #include <net/if_var.h> 89 #include <net/if_clone.h> 90 #include <net/if_types.h> 91 #include <net/vnet.h> 92 #include <net/pfvar.h> 93 #include <net/if_pfsync.h> 94 95 #include <netinet/if_ether.h> 96 #include <netinet/in.h> 97 #include <netinet/in_var.h> 98 #include <netinet/ip.h> 99 #include <netinet/ip_carp.h> 100 #include <netinet/ip_var.h> 101 #include <netinet/tcp.h> 102 #include <netinet/tcp_fsm.h> 103 #include <netinet/tcp_seq.h> 104 105 #define PFSYNC_MINPKT ( \ 106 sizeof(struct ip) + \ 107 sizeof(struct pfsync_header) + \ 108 sizeof(struct pfsync_subheader) ) 109 110 struct pfsync_bucket; 111 112 struct pfsync_pkt { 113 struct ip *ip; 114 struct in_addr src; 115 u_int8_t flags; 116 }; 117 118 static int pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *, 119 struct pfsync_state_peer *); 120 static int pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int); 121 static int pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int); 122 static int pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int); 123 static int pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int); 124 static int pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int); 125 static int pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int); 126 static int pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int); 127 static int pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int); 128 static int pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int); 129 static int pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int); 130 static int pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int); 131 static int pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int); 132 133 static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = { 134 pfsync_in_clr, /* PFSYNC_ACT_CLR */ 135 pfsync_in_ins, /* PFSYNC_ACT_INS */ 136 pfsync_in_iack, /* PFSYNC_ACT_INS_ACK */ 137 pfsync_in_upd, /* PFSYNC_ACT_UPD */ 138 pfsync_in_upd_c, /* PFSYNC_ACT_UPD_C */ 139 pfsync_in_ureq, /* PFSYNC_ACT_UPD_REQ */ 140 pfsync_in_del, /* PFSYNC_ACT_DEL */ 141 pfsync_in_del_c, /* PFSYNC_ACT_DEL_C */ 142 pfsync_in_error, /* PFSYNC_ACT_INS_F */ 143 pfsync_in_error, /* PFSYNC_ACT_DEL_F */ 144 pfsync_in_bus, /* PFSYNC_ACT_BUS */ 145 pfsync_in_tdb, /* PFSYNC_ACT_TDB */ 146 pfsync_in_eof /* PFSYNC_ACT_EOF */ 147 }; 148 149 struct pfsync_q { 150 void (*write)(struct pf_state *, void *); 151 size_t len; 152 u_int8_t action; 153 }; 154 155 /* we have one of these for every PFSYNC_S_ */ 156 static void pfsync_out_state(struct pf_state *, void *); 157 static void pfsync_out_iack(struct pf_state *, void *); 158 static void pfsync_out_upd_c(struct pf_state *, void *); 159 static void pfsync_out_del(struct pf_state *, void *); 160 161 static struct pfsync_q pfsync_qs[] = { 162 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_INS }, 163 { pfsync_out_iack, sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK }, 164 { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_UPD }, 165 { pfsync_out_upd_c, sizeof(struct pfsync_upd_c), PFSYNC_ACT_UPD_C }, 166 { pfsync_out_del, sizeof(struct pfsync_del_c), PFSYNC_ACT_DEL_C } 167 }; 168 169 static void pfsync_q_ins(struct pf_state *, int, bool); 170 static void pfsync_q_del(struct pf_state *, bool, struct pfsync_bucket *); 171 172 static void pfsync_update_state(struct pf_state *); 173 174 struct pfsync_upd_req_item { 175 TAILQ_ENTRY(pfsync_upd_req_item) ur_entry; 176 struct pfsync_upd_req ur_msg; 177 }; 178 179 struct pfsync_deferral { 180 struct pfsync_softc *pd_sc; 181 TAILQ_ENTRY(pfsync_deferral) pd_entry; 182 u_int pd_refs; 183 struct callout pd_tmo; 184 185 struct pf_state *pd_st; 186 struct mbuf *pd_m; 187 }; 188 189 struct pfsync_sofct; 190 191 struct pfsync_bucket 192 { 193 int b_id; 194 struct pfsync_softc *b_sc; 195 struct mtx b_mtx; 196 struct callout b_tmo; 197 int b_flags; 198 #define PFSYNCF_BUCKET_PUSH 0x00000001 199 200 size_t b_len; 201 TAILQ_HEAD(, pf_state) b_qs[PFSYNC_S_COUNT]; 202 TAILQ_HEAD(, pfsync_upd_req_item) b_upd_req_list; 203 TAILQ_HEAD(, pfsync_deferral) b_deferrals; 204 u_int b_deferred; 205 void *b_plus; 206 size_t b_pluslen; 207 208 struct ifaltq b_snd; 209 }; 210 211 struct pfsync_softc { 212 /* Configuration */ 213 struct ifnet *sc_ifp; 214 struct ifnet *sc_sync_if; 215 struct ip_moptions sc_imo; 216 struct in_addr sc_sync_peer; 217 uint32_t sc_flags; 218 #define PFSYNCF_OK 0x00000001 219 #define PFSYNCF_DEFER 0x00000002 220 uint8_t sc_maxupdates; 221 struct ip sc_template; 222 struct mtx sc_mtx; 223 224 /* Queued data */ 225 struct pfsync_bucket *sc_buckets; 226 227 /* Bulk update info */ 228 struct mtx sc_bulk_mtx; 229 uint32_t sc_ureq_sent; 230 int sc_bulk_tries; 231 uint32_t sc_ureq_received; 232 int sc_bulk_hashid; 233 uint64_t sc_bulk_stateid; 234 uint32_t sc_bulk_creatorid; 235 struct callout sc_bulk_tmo; 236 struct callout sc_bulkfail_tmo; 237 }; 238 239 #define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx) 240 #define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx) 241 #define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED) 242 243 #define PFSYNC_BUCKET_LOCK(b) mtx_lock(&(b)->b_mtx) 244 #define PFSYNC_BUCKET_UNLOCK(b) mtx_unlock(&(b)->b_mtx) 245 #define PFSYNC_BUCKET_LOCK_ASSERT(b) mtx_assert(&(b)->b_mtx, MA_OWNED) 246 247 #define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx) 248 #define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx) 249 #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED) 250 251 static const char pfsyncname[] = "pfsync"; 252 static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data"); 253 VNET_DEFINE_STATIC(struct pfsync_softc *, pfsyncif) = NULL; 254 #define V_pfsyncif VNET(pfsyncif) 255 VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL; 256 #define V_pfsync_swi_cookie VNET(pfsync_swi_cookie) 257 VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats); 258 #define V_pfsyncstats VNET(pfsyncstats) 259 VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW; 260 #define V_pfsync_carp_adj VNET(pfsync_carp_adj) 261 262 static void pfsync_timeout(void *); 263 static void pfsync_push(struct pfsync_bucket *); 264 static void pfsync_push_all(struct pfsync_softc *); 265 static void pfsyncintr(void *); 266 static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *, 267 void *); 268 static void pfsync_multicast_cleanup(struct pfsync_softc *); 269 static void pfsync_pointers_init(void); 270 static void pfsync_pointers_uninit(void); 271 static int pfsync_init(void); 272 static void pfsync_uninit(void); 273 274 static unsigned long pfsync_buckets; 275 276 SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC"); 277 SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW, 278 &VNET_NAME(pfsyncstats), pfsyncstats, 279 "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)"); 280 SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW, 281 &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment"); 282 SYSCTL_ULONG(_net_pfsync, OID_AUTO, pfsync_buckets, CTLFLAG_RDTUN, 283 &pfsync_buckets, 0, "Number of pfsync hash buckets"); 284 285 static int pfsync_clone_create(struct if_clone *, int, caddr_t); 286 static void pfsync_clone_destroy(struct ifnet *); 287 static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, 288 struct pf_state_peer *); 289 static int pfsyncoutput(struct ifnet *, struct mbuf *, 290 const struct sockaddr *, struct route *); 291 static int pfsyncioctl(struct ifnet *, u_long, caddr_t); 292 293 static int pfsync_defer(struct pf_state *, struct mbuf *); 294 static void pfsync_undefer(struct pfsync_deferral *, int); 295 static void pfsync_undefer_state(struct pf_state *, int); 296 static void pfsync_defer_tmo(void *); 297 298 static void pfsync_request_update(u_int32_t, u_int64_t); 299 static bool pfsync_update_state_req(struct pf_state *); 300 301 static void pfsync_drop(struct pfsync_softc *); 302 static void pfsync_sendout(int, int); 303 static void pfsync_send_plus(void *, size_t); 304 305 static void pfsync_bulk_start(void); 306 static void pfsync_bulk_status(u_int8_t); 307 static void pfsync_bulk_update(void *); 308 static void pfsync_bulk_fail(void *); 309 310 static void pfsync_detach_ifnet(struct ifnet *); 311 #ifdef IPSEC 312 static void pfsync_update_net_tdb(struct pfsync_tdb *); 313 #endif 314 static struct pfsync_bucket *pfsync_get_bucket(struct pfsync_softc *, 315 struct pf_state *); 316 317 318 #define PFSYNC_MAX_BULKTRIES 12 319 320 VNET_DEFINE(struct if_clone *, pfsync_cloner); 321 #define V_pfsync_cloner VNET(pfsync_cloner) 322 323 static int 324 pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param) 325 { 326 struct pfsync_softc *sc; 327 struct ifnet *ifp; 328 struct pfsync_bucket *b; 329 int c, q; 330 331 if (unit != 0) 332 return (EINVAL); 333 334 if (! pfsync_buckets) 335 pfsync_buckets = mp_ncpus * 2; 336 337 sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO); 338 sc->sc_flags |= PFSYNCF_OK; 339 sc->sc_maxupdates = 128; 340 341 ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC); 342 if (ifp == NULL) { 343 free(sc, M_PFSYNC); 344 return (ENOSPC); 345 } 346 if_initname(ifp, pfsyncname, unit); 347 ifp->if_softc = sc; 348 ifp->if_ioctl = pfsyncioctl; 349 ifp->if_output = pfsyncoutput; 350 ifp->if_type = IFT_PFSYNC; 351 ifp->if_hdrlen = sizeof(struct pfsync_header); 352 ifp->if_mtu = ETHERMTU; 353 mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF); 354 mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF); 355 callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0); 356 callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0); 357 358 if_attach(ifp); 359 360 bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN); 361 362 sc->sc_buckets = mallocarray(pfsync_buckets, sizeof(*sc->sc_buckets), 363 M_PFSYNC, M_ZERO | M_WAITOK); 364 for (c = 0; c < pfsync_buckets; c++) { 365 b = &sc->sc_buckets[c]; 366 mtx_init(&b->b_mtx, "pfsync bucket", NULL, MTX_DEF); 367 368 b->b_id = c; 369 b->b_sc = sc; 370 b->b_len = PFSYNC_MINPKT; 371 372 for (q = 0; q < PFSYNC_S_COUNT; q++) 373 TAILQ_INIT(&b->b_qs[q]); 374 375 TAILQ_INIT(&b->b_upd_req_list); 376 TAILQ_INIT(&b->b_deferrals); 377 378 callout_init(&b->b_tmo, 1); 379 380 b->b_snd.ifq_maxlen = ifqmaxlen; 381 } 382 383 V_pfsyncif = sc; 384 385 return (0); 386 } 387 388 static void 389 pfsync_clone_destroy(struct ifnet *ifp) 390 { 391 struct pfsync_softc *sc = ifp->if_softc; 392 struct pfsync_bucket *b; 393 int c; 394 395 for (c = 0; c < pfsync_buckets; c++) { 396 b = &sc->sc_buckets[c]; 397 /* 398 * At this stage, everything should have already been 399 * cleared by pfsync_uninit(), and we have only to 400 * drain callouts. 401 */ 402 while (b->b_deferred > 0) { 403 struct pfsync_deferral *pd = 404 TAILQ_FIRST(&b->b_deferrals); 405 406 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry); 407 b->b_deferred--; 408 if (callout_stop(&pd->pd_tmo) > 0) { 409 pf_release_state(pd->pd_st); 410 m_freem(pd->pd_m); 411 free(pd, M_PFSYNC); 412 } else { 413 pd->pd_refs++; 414 callout_drain(&pd->pd_tmo); 415 free(pd, M_PFSYNC); 416 } 417 } 418 419 callout_drain(&b->b_tmo); 420 } 421 422 callout_drain(&sc->sc_bulkfail_tmo); 423 callout_drain(&sc->sc_bulk_tmo); 424 425 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 426 (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy"); 427 bpfdetach(ifp); 428 if_detach(ifp); 429 430 pfsync_drop(sc); 431 432 if_free(ifp); 433 if (sc->sc_imo.imo_membership) 434 pfsync_multicast_cleanup(sc); 435 mtx_destroy(&sc->sc_mtx); 436 mtx_destroy(&sc->sc_bulk_mtx); 437 438 free(sc->sc_buckets, M_PFSYNC); 439 free(sc, M_PFSYNC); 440 441 V_pfsyncif = NULL; 442 } 443 444 static int 445 pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, 446 struct pf_state_peer *d) 447 { 448 if (s->scrub.scrub_flag && d->scrub == NULL) { 449 d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO); 450 if (d->scrub == NULL) 451 return (ENOMEM); 452 } 453 454 return (0); 455 } 456 457 458 static int 459 pfsync_state_import(struct pfsync_state *sp, u_int8_t flags) 460 { 461 struct pfsync_softc *sc = V_pfsyncif; 462 #ifndef __NO_STRICT_ALIGNMENT 463 struct pfsync_state_key key[2]; 464 #endif 465 struct pfsync_state_key *kw, *ks; 466 struct pf_state *st = NULL; 467 struct pf_state_key *skw = NULL, *sks = NULL; 468 struct pf_rule *r = NULL; 469 struct pfi_kif *kif; 470 int error; 471 472 PF_RULES_RASSERT(); 473 474 if (sp->creatorid == 0) { 475 if (V_pf_status.debug >= PF_DEBUG_MISC) 476 printf("%s: invalid creator id: %08x\n", __func__, 477 ntohl(sp->creatorid)); 478 return (EINVAL); 479 } 480 481 if ((kif = pfi_kif_find(sp->ifname)) == NULL) { 482 if (V_pf_status.debug >= PF_DEBUG_MISC) 483 printf("%s: unknown interface: %s\n", __func__, 484 sp->ifname); 485 if (flags & PFSYNC_SI_IOCTL) 486 return (EINVAL); 487 return (0); /* skip this state */ 488 } 489 490 /* 491 * If the ruleset checksums match or the state is coming from the ioctl, 492 * it's safe to associate the state with the rule of that number. 493 */ 494 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && 495 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) < 496 pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) 497 r = pf_main_ruleset.rules[ 498 PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; 499 else 500 r = &V_pf_default_rule; 501 502 if ((r->max_states && 503 counter_u64_fetch(r->states_cur) >= r->max_states)) 504 goto cleanup; 505 506 /* 507 * XXXGL: consider M_WAITOK in ioctl path after. 508 */ 509 if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL) 510 goto cleanup; 511 512 if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL) 513 goto cleanup; 514 515 #ifndef __NO_STRICT_ALIGNMENT 516 bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2); 517 kw = &key[PF_SK_WIRE]; 518 ks = &key[PF_SK_STACK]; 519 #else 520 kw = &sp->key[PF_SK_WIRE]; 521 ks = &sp->key[PF_SK_STACK]; 522 #endif 523 524 if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) || 525 PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) || 526 kw->port[0] != ks->port[0] || 527 kw->port[1] != ks->port[1]) { 528 sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 529 if (sks == NULL) 530 goto cleanup; 531 } else 532 sks = skw; 533 534 /* allocate memory for scrub info */ 535 if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || 536 pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) 537 goto cleanup; 538 539 /* Copy to state key(s). */ 540 skw->addr[0] = kw->addr[0]; 541 skw->addr[1] = kw->addr[1]; 542 skw->port[0] = kw->port[0]; 543 skw->port[1] = kw->port[1]; 544 skw->proto = sp->proto; 545 skw->af = sp->af; 546 if (sks != skw) { 547 sks->addr[0] = ks->addr[0]; 548 sks->addr[1] = ks->addr[1]; 549 sks->port[0] = ks->port[0]; 550 sks->port[1] = ks->port[1]; 551 sks->proto = sp->proto; 552 sks->af = sp->af; 553 } 554 555 /* copy to state */ 556 bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr)); 557 st->creation = time_uptime - ntohl(sp->creation); 558 st->expire = time_uptime; 559 if (sp->expire) { 560 uint32_t timeout; 561 562 timeout = r->timeout[sp->timeout]; 563 if (!timeout) 564 timeout = V_pf_default_rule.timeout[sp->timeout]; 565 566 /* sp->expire may have been adaptively scaled by export. */ 567 st->expire -= timeout - ntohl(sp->expire); 568 } 569 570 st->direction = sp->direction; 571 st->log = sp->log; 572 st->timeout = sp->timeout; 573 st->state_flags = sp->state_flags; 574 575 st->id = sp->id; 576 st->creatorid = sp->creatorid; 577 pf_state_peer_ntoh(&sp->src, &st->src); 578 pf_state_peer_ntoh(&sp->dst, &st->dst); 579 580 st->rule.ptr = r; 581 st->nat_rule.ptr = NULL; 582 st->anchor.ptr = NULL; 583 st->rt_kif = NULL; 584 585 st->pfsync_time = time_uptime; 586 st->sync_state = PFSYNC_S_NONE; 587 588 if (!(flags & PFSYNC_SI_IOCTL)) 589 st->state_flags |= PFSTATE_NOSYNC; 590 591 if ((error = pf_state_insert(kif, skw, sks, st)) != 0) 592 goto cleanup_state; 593 594 /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ 595 counter_u64_add(r->states_cur, 1); 596 counter_u64_add(r->states_tot, 1); 597 598 if (!(flags & PFSYNC_SI_IOCTL)) { 599 st->state_flags &= ~PFSTATE_NOSYNC; 600 if (st->state_flags & PFSTATE_ACK) { 601 pfsync_q_ins(st, PFSYNC_S_IACK, true); 602 pfsync_push_all(sc); 603 } 604 } 605 st->state_flags &= ~PFSTATE_ACK; 606 PF_STATE_UNLOCK(st); 607 608 return (0); 609 610 cleanup: 611 error = ENOMEM; 612 if (skw == sks) 613 sks = NULL; 614 if (skw != NULL) 615 uma_zfree(V_pf_state_key_z, skw); 616 if (sks != NULL) 617 uma_zfree(V_pf_state_key_z, sks); 618 619 cleanup_state: /* pf_state_insert() frees the state keys. */ 620 if (st) { 621 if (st->dst.scrub) 622 uma_zfree(V_pf_state_scrub_z, st->dst.scrub); 623 if (st->src.scrub) 624 uma_zfree(V_pf_state_scrub_z, st->src.scrub); 625 uma_zfree(V_pf_state_z, st); 626 } 627 return (error); 628 } 629 630 static int 631 pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused) 632 { 633 struct pfsync_softc *sc = V_pfsyncif; 634 struct pfsync_pkt pkt; 635 struct mbuf *m = *mp; 636 struct ip *ip = mtod(m, struct ip *); 637 struct pfsync_header *ph; 638 struct pfsync_subheader subh; 639 640 int offset, len; 641 int rv; 642 uint16_t count; 643 644 PF_RULES_RLOCK_TRACKER; 645 646 *mp = NULL; 647 V_pfsyncstats.pfsyncs_ipackets++; 648 649 /* Verify that we have a sync interface configured. */ 650 if (!sc || !sc->sc_sync_if || !V_pf_status.running || 651 (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 652 goto done; 653 654 /* verify that the packet came in on the right interface */ 655 if (sc->sc_sync_if != m->m_pkthdr.rcvif) { 656 V_pfsyncstats.pfsyncs_badif++; 657 goto done; 658 } 659 660 if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1); 661 if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); 662 /* verify that the IP TTL is 255. */ 663 if (ip->ip_ttl != PFSYNC_DFLTTL) { 664 V_pfsyncstats.pfsyncs_badttl++; 665 goto done; 666 } 667 668 offset = ip->ip_hl << 2; 669 if (m->m_pkthdr.len < offset + sizeof(*ph)) { 670 V_pfsyncstats.pfsyncs_hdrops++; 671 goto done; 672 } 673 674 if (offset + sizeof(*ph) > m->m_len) { 675 if (m_pullup(m, offset + sizeof(*ph)) == NULL) { 676 V_pfsyncstats.pfsyncs_hdrops++; 677 return (IPPROTO_DONE); 678 } 679 ip = mtod(m, struct ip *); 680 } 681 ph = (struct pfsync_header *)((char *)ip + offset); 682 683 /* verify the version */ 684 if (ph->version != PFSYNC_VERSION) { 685 V_pfsyncstats.pfsyncs_badver++; 686 goto done; 687 } 688 689 len = ntohs(ph->len) + offset; 690 if (m->m_pkthdr.len < len) { 691 V_pfsyncstats.pfsyncs_badlen++; 692 goto done; 693 } 694 695 /* Cheaper to grab this now than having to mess with mbufs later */ 696 pkt.ip = ip; 697 pkt.src = ip->ip_src; 698 pkt.flags = 0; 699 700 /* 701 * Trusting pf_chksum during packet processing, as well as seeking 702 * in interface name tree, require holding PF_RULES_RLOCK(). 703 */ 704 PF_RULES_RLOCK(); 705 if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) 706 pkt.flags |= PFSYNC_SI_CKSUM; 707 708 offset += sizeof(*ph); 709 while (offset <= len - sizeof(subh)) { 710 m_copydata(m, offset, sizeof(subh), (caddr_t)&subh); 711 offset += sizeof(subh); 712 713 if (subh.action >= PFSYNC_ACT_MAX) { 714 V_pfsyncstats.pfsyncs_badact++; 715 PF_RULES_RUNLOCK(); 716 goto done; 717 } 718 719 count = ntohs(subh.count); 720 V_pfsyncstats.pfsyncs_iacts[subh.action] += count; 721 rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count); 722 if (rv == -1) { 723 PF_RULES_RUNLOCK(); 724 return (IPPROTO_DONE); 725 } 726 727 offset += rv; 728 } 729 PF_RULES_RUNLOCK(); 730 731 done: 732 m_freem(m); 733 return (IPPROTO_DONE); 734 } 735 736 static int 737 pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 738 { 739 struct pfsync_clr *clr; 740 struct mbuf *mp; 741 int len = sizeof(*clr) * count; 742 int i, offp; 743 u_int32_t creatorid; 744 745 mp = m_pulldown(m, offset, len, &offp); 746 if (mp == NULL) { 747 V_pfsyncstats.pfsyncs_badlen++; 748 return (-1); 749 } 750 clr = (struct pfsync_clr *)(mp->m_data + offp); 751 752 for (i = 0; i < count; i++) { 753 creatorid = clr[i].creatorid; 754 755 if (clr[i].ifname[0] != '\0' && 756 pfi_kif_find(clr[i].ifname) == NULL) 757 continue; 758 759 for (int i = 0; i <= pf_hashmask; i++) { 760 struct pf_idhash *ih = &V_pf_idhash[i]; 761 struct pf_state *s; 762 relock: 763 PF_HASHROW_LOCK(ih); 764 LIST_FOREACH(s, &ih->states, entry) { 765 if (s->creatorid == creatorid) { 766 s->state_flags |= PFSTATE_NOSYNC; 767 pf_unlink_state(s, PF_ENTER_LOCKED); 768 goto relock; 769 } 770 } 771 PF_HASHROW_UNLOCK(ih); 772 } 773 } 774 775 return (len); 776 } 777 778 static int 779 pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 780 { 781 struct mbuf *mp; 782 struct pfsync_state *sa, *sp; 783 int len = sizeof(*sp) * count; 784 int i, offp; 785 786 mp = m_pulldown(m, offset, len, &offp); 787 if (mp == NULL) { 788 V_pfsyncstats.pfsyncs_badlen++; 789 return (-1); 790 } 791 sa = (struct pfsync_state *)(mp->m_data + offp); 792 793 for (i = 0; i < count; i++) { 794 sp = &sa[i]; 795 796 /* Check for invalid values. */ 797 if (sp->timeout >= PFTM_MAX || 798 sp->src.state > PF_TCPS_PROXY_DST || 799 sp->dst.state > PF_TCPS_PROXY_DST || 800 sp->direction > PF_OUT || 801 (sp->af != AF_INET && sp->af != AF_INET6)) { 802 if (V_pf_status.debug >= PF_DEBUG_MISC) 803 printf("%s: invalid value\n", __func__); 804 V_pfsyncstats.pfsyncs_badval++; 805 continue; 806 } 807 808 if (pfsync_state_import(sp, pkt->flags) == ENOMEM) 809 /* Drop out, but process the rest of the actions. */ 810 break; 811 } 812 813 return (len); 814 } 815 816 static int 817 pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 818 { 819 struct pfsync_ins_ack *ia, *iaa; 820 struct pf_state *st; 821 822 struct mbuf *mp; 823 int len = count * sizeof(*ia); 824 int offp, i; 825 826 mp = m_pulldown(m, offset, len, &offp); 827 if (mp == NULL) { 828 V_pfsyncstats.pfsyncs_badlen++; 829 return (-1); 830 } 831 iaa = (struct pfsync_ins_ack *)(mp->m_data + offp); 832 833 for (i = 0; i < count; i++) { 834 ia = &iaa[i]; 835 836 st = pf_find_state_byid(ia->id, ia->creatorid); 837 if (st == NULL) 838 continue; 839 840 if (st->state_flags & PFSTATE_ACK) { 841 pfsync_undefer_state(st, 0); 842 } 843 PF_STATE_UNLOCK(st); 844 } 845 /* 846 * XXX this is not yet implemented, but we know the size of the 847 * message so we can skip it. 848 */ 849 850 return (count * sizeof(struct pfsync_ins_ack)); 851 } 852 853 static int 854 pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src, 855 struct pfsync_state_peer *dst) 856 { 857 int sync = 0; 858 859 PF_STATE_LOCK_ASSERT(st); 860 861 /* 862 * The state should never go backwards except 863 * for syn-proxy states. Neither should the 864 * sequence window slide backwards. 865 */ 866 if ((st->src.state > src->state && 867 (st->src.state < PF_TCPS_PROXY_SRC || 868 src->state >= PF_TCPS_PROXY_SRC)) || 869 870 (st->src.state == src->state && 871 SEQ_GT(st->src.seqlo, ntohl(src->seqlo)))) 872 sync++; 873 else 874 pf_state_peer_ntoh(src, &st->src); 875 876 if ((st->dst.state > dst->state) || 877 878 (st->dst.state >= TCPS_SYN_SENT && 879 SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo)))) 880 sync++; 881 else 882 pf_state_peer_ntoh(dst, &st->dst); 883 884 return (sync); 885 } 886 887 static int 888 pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 889 { 890 struct pfsync_softc *sc = V_pfsyncif; 891 struct pfsync_state *sa, *sp; 892 struct pf_state *st; 893 int sync; 894 895 struct mbuf *mp; 896 int len = count * sizeof(*sp); 897 int offp, i; 898 899 mp = m_pulldown(m, offset, len, &offp); 900 if (mp == NULL) { 901 V_pfsyncstats.pfsyncs_badlen++; 902 return (-1); 903 } 904 sa = (struct pfsync_state *)(mp->m_data + offp); 905 906 for (i = 0; i < count; i++) { 907 sp = &sa[i]; 908 909 /* check for invalid values */ 910 if (sp->timeout >= PFTM_MAX || 911 sp->src.state > PF_TCPS_PROXY_DST || 912 sp->dst.state > PF_TCPS_PROXY_DST) { 913 if (V_pf_status.debug >= PF_DEBUG_MISC) { 914 printf("pfsync_input: PFSYNC_ACT_UPD: " 915 "invalid value\n"); 916 } 917 V_pfsyncstats.pfsyncs_badval++; 918 continue; 919 } 920 921 st = pf_find_state_byid(sp->id, sp->creatorid); 922 if (st == NULL) { 923 /* insert the update */ 924 if (pfsync_state_import(sp, pkt->flags)) 925 V_pfsyncstats.pfsyncs_badstate++; 926 continue; 927 } 928 929 if (st->state_flags & PFSTATE_ACK) { 930 pfsync_undefer_state(st, 1); 931 } 932 933 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) 934 sync = pfsync_upd_tcp(st, &sp->src, &sp->dst); 935 else { 936 sync = 0; 937 938 /* 939 * Non-TCP protocol state machine always go 940 * forwards 941 */ 942 if (st->src.state > sp->src.state) 943 sync++; 944 else 945 pf_state_peer_ntoh(&sp->src, &st->src); 946 if (st->dst.state > sp->dst.state) 947 sync++; 948 else 949 pf_state_peer_ntoh(&sp->dst, &st->dst); 950 } 951 if (sync < 2) { 952 pfsync_alloc_scrub_memory(&sp->dst, &st->dst); 953 pf_state_peer_ntoh(&sp->dst, &st->dst); 954 st->expire = time_uptime; 955 st->timeout = sp->timeout; 956 } 957 st->pfsync_time = time_uptime; 958 959 if (sync) { 960 V_pfsyncstats.pfsyncs_stale++; 961 962 pfsync_update_state(st); 963 PF_STATE_UNLOCK(st); 964 pfsync_push_all(sc); 965 continue; 966 } 967 PF_STATE_UNLOCK(st); 968 } 969 970 return (len); 971 } 972 973 static int 974 pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 975 { 976 struct pfsync_softc *sc = V_pfsyncif; 977 struct pfsync_upd_c *ua, *up; 978 struct pf_state *st; 979 int len = count * sizeof(*up); 980 int sync; 981 struct mbuf *mp; 982 int offp, i; 983 984 mp = m_pulldown(m, offset, len, &offp); 985 if (mp == NULL) { 986 V_pfsyncstats.pfsyncs_badlen++; 987 return (-1); 988 } 989 ua = (struct pfsync_upd_c *)(mp->m_data + offp); 990 991 for (i = 0; i < count; i++) { 992 up = &ua[i]; 993 994 /* check for invalid values */ 995 if (up->timeout >= PFTM_MAX || 996 up->src.state > PF_TCPS_PROXY_DST || 997 up->dst.state > PF_TCPS_PROXY_DST) { 998 if (V_pf_status.debug >= PF_DEBUG_MISC) { 999 printf("pfsync_input: " 1000 "PFSYNC_ACT_UPD_C: " 1001 "invalid value\n"); 1002 } 1003 V_pfsyncstats.pfsyncs_badval++; 1004 continue; 1005 } 1006 1007 st = pf_find_state_byid(up->id, up->creatorid); 1008 if (st == NULL) { 1009 /* We don't have this state. Ask for it. */ 1010 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]); 1011 pfsync_request_update(up->creatorid, up->id); 1012 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]); 1013 continue; 1014 } 1015 1016 if (st->state_flags & PFSTATE_ACK) { 1017 pfsync_undefer_state(st, 1); 1018 } 1019 1020 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) 1021 sync = pfsync_upd_tcp(st, &up->src, &up->dst); 1022 else { 1023 sync = 0; 1024 1025 /* 1026 * Non-TCP protocol state machine always go 1027 * forwards 1028 */ 1029 if (st->src.state > up->src.state) 1030 sync++; 1031 else 1032 pf_state_peer_ntoh(&up->src, &st->src); 1033 if (st->dst.state > up->dst.state) 1034 sync++; 1035 else 1036 pf_state_peer_ntoh(&up->dst, &st->dst); 1037 } 1038 if (sync < 2) { 1039 pfsync_alloc_scrub_memory(&up->dst, &st->dst); 1040 pf_state_peer_ntoh(&up->dst, &st->dst); 1041 st->expire = time_uptime; 1042 st->timeout = up->timeout; 1043 } 1044 st->pfsync_time = time_uptime; 1045 1046 if (sync) { 1047 V_pfsyncstats.pfsyncs_stale++; 1048 1049 pfsync_update_state(st); 1050 PF_STATE_UNLOCK(st); 1051 pfsync_push_all(sc); 1052 continue; 1053 } 1054 PF_STATE_UNLOCK(st); 1055 } 1056 1057 return (len); 1058 } 1059 1060 static int 1061 pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1062 { 1063 struct pfsync_upd_req *ur, *ura; 1064 struct mbuf *mp; 1065 int len = count * sizeof(*ur); 1066 int i, offp; 1067 1068 struct pf_state *st; 1069 1070 mp = m_pulldown(m, offset, len, &offp); 1071 if (mp == NULL) { 1072 V_pfsyncstats.pfsyncs_badlen++; 1073 return (-1); 1074 } 1075 ura = (struct pfsync_upd_req *)(mp->m_data + offp); 1076 1077 for (i = 0; i < count; i++) { 1078 ur = &ura[i]; 1079 1080 if (ur->id == 0 && ur->creatorid == 0) 1081 pfsync_bulk_start(); 1082 else { 1083 st = pf_find_state_byid(ur->id, ur->creatorid); 1084 if (st == NULL) { 1085 V_pfsyncstats.pfsyncs_badstate++; 1086 continue; 1087 } 1088 if (st->state_flags & PFSTATE_NOSYNC) { 1089 PF_STATE_UNLOCK(st); 1090 continue; 1091 } 1092 1093 pfsync_update_state_req(st); 1094 PF_STATE_UNLOCK(st); 1095 } 1096 } 1097 1098 return (len); 1099 } 1100 1101 static int 1102 pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1103 { 1104 struct mbuf *mp; 1105 struct pfsync_state *sa, *sp; 1106 struct pf_state *st; 1107 int len = count * sizeof(*sp); 1108 int offp, i; 1109 1110 mp = m_pulldown(m, offset, len, &offp); 1111 if (mp == NULL) { 1112 V_pfsyncstats.pfsyncs_badlen++; 1113 return (-1); 1114 } 1115 sa = (struct pfsync_state *)(mp->m_data + offp); 1116 1117 for (i = 0; i < count; i++) { 1118 sp = &sa[i]; 1119 1120 st = pf_find_state_byid(sp->id, sp->creatorid); 1121 if (st == NULL) { 1122 V_pfsyncstats.pfsyncs_badstate++; 1123 continue; 1124 } 1125 st->state_flags |= PFSTATE_NOSYNC; 1126 pf_unlink_state(st, PF_ENTER_LOCKED); 1127 } 1128 1129 return (len); 1130 } 1131 1132 static int 1133 pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1134 { 1135 struct mbuf *mp; 1136 struct pfsync_del_c *sa, *sp; 1137 struct pf_state *st; 1138 int len = count * sizeof(*sp); 1139 int offp, i; 1140 1141 mp = m_pulldown(m, offset, len, &offp); 1142 if (mp == NULL) { 1143 V_pfsyncstats.pfsyncs_badlen++; 1144 return (-1); 1145 } 1146 sa = (struct pfsync_del_c *)(mp->m_data + offp); 1147 1148 for (i = 0; i < count; i++) { 1149 sp = &sa[i]; 1150 1151 st = pf_find_state_byid(sp->id, sp->creatorid); 1152 if (st == NULL) { 1153 V_pfsyncstats.pfsyncs_badstate++; 1154 continue; 1155 } 1156 1157 st->state_flags |= PFSTATE_NOSYNC; 1158 pf_unlink_state(st, PF_ENTER_LOCKED); 1159 } 1160 1161 return (len); 1162 } 1163 1164 static int 1165 pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1166 { 1167 struct pfsync_softc *sc = V_pfsyncif; 1168 struct pfsync_bus *bus; 1169 struct mbuf *mp; 1170 int len = count * sizeof(*bus); 1171 int offp; 1172 1173 PFSYNC_BLOCK(sc); 1174 1175 /* If we're not waiting for a bulk update, who cares. */ 1176 if (sc->sc_ureq_sent == 0) { 1177 PFSYNC_BUNLOCK(sc); 1178 return (len); 1179 } 1180 1181 mp = m_pulldown(m, offset, len, &offp); 1182 if (mp == NULL) { 1183 PFSYNC_BUNLOCK(sc); 1184 V_pfsyncstats.pfsyncs_badlen++; 1185 return (-1); 1186 } 1187 bus = (struct pfsync_bus *)(mp->m_data + offp); 1188 1189 switch (bus->status) { 1190 case PFSYNC_BUS_START: 1191 callout_reset(&sc->sc_bulkfail_tmo, 4 * hz + 1192 V_pf_limits[PF_LIMIT_STATES].limit / 1193 ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) / 1194 sizeof(struct pfsync_state)), 1195 pfsync_bulk_fail, sc); 1196 if (V_pf_status.debug >= PF_DEBUG_MISC) 1197 printf("pfsync: received bulk update start\n"); 1198 break; 1199 1200 case PFSYNC_BUS_END: 1201 if (time_uptime - ntohl(bus->endtime) >= 1202 sc->sc_ureq_sent) { 1203 /* that's it, we're happy */ 1204 sc->sc_ureq_sent = 0; 1205 sc->sc_bulk_tries = 0; 1206 callout_stop(&sc->sc_bulkfail_tmo); 1207 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1208 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 1209 "pfsync bulk done"); 1210 sc->sc_flags |= PFSYNCF_OK; 1211 if (V_pf_status.debug >= PF_DEBUG_MISC) 1212 printf("pfsync: received valid " 1213 "bulk update end\n"); 1214 } else { 1215 if (V_pf_status.debug >= PF_DEBUG_MISC) 1216 printf("pfsync: received invalid " 1217 "bulk update end: bad timestamp\n"); 1218 } 1219 break; 1220 } 1221 PFSYNC_BUNLOCK(sc); 1222 1223 return (len); 1224 } 1225 1226 static int 1227 pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1228 { 1229 int len = count * sizeof(struct pfsync_tdb); 1230 1231 #if defined(IPSEC) 1232 struct pfsync_tdb *tp; 1233 struct mbuf *mp; 1234 int offp; 1235 int i; 1236 int s; 1237 1238 mp = m_pulldown(m, offset, len, &offp); 1239 if (mp == NULL) { 1240 V_pfsyncstats.pfsyncs_badlen++; 1241 return (-1); 1242 } 1243 tp = (struct pfsync_tdb *)(mp->m_data + offp); 1244 1245 for (i = 0; i < count; i++) 1246 pfsync_update_net_tdb(&tp[i]); 1247 #endif 1248 1249 return (len); 1250 } 1251 1252 #if defined(IPSEC) 1253 /* Update an in-kernel tdb. Silently fail if no tdb is found. */ 1254 static void 1255 pfsync_update_net_tdb(struct pfsync_tdb *pt) 1256 { 1257 struct tdb *tdb; 1258 int s; 1259 1260 /* check for invalid values */ 1261 if (ntohl(pt->spi) <= SPI_RESERVED_MAX || 1262 (pt->dst.sa.sa_family != AF_INET && 1263 pt->dst.sa.sa_family != AF_INET6)) 1264 goto bad; 1265 1266 tdb = gettdb(pt->spi, &pt->dst, pt->sproto); 1267 if (tdb) { 1268 pt->rpl = ntohl(pt->rpl); 1269 pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes); 1270 1271 /* Neither replay nor byte counter should ever decrease. */ 1272 if (pt->rpl < tdb->tdb_rpl || 1273 pt->cur_bytes < tdb->tdb_cur_bytes) { 1274 goto bad; 1275 } 1276 1277 tdb->tdb_rpl = pt->rpl; 1278 tdb->tdb_cur_bytes = pt->cur_bytes; 1279 } 1280 return; 1281 1282 bad: 1283 if (V_pf_status.debug >= PF_DEBUG_MISC) 1284 printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: " 1285 "invalid value\n"); 1286 V_pfsyncstats.pfsyncs_badstate++; 1287 return; 1288 } 1289 #endif 1290 1291 1292 static int 1293 pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1294 { 1295 /* check if we are at the right place in the packet */ 1296 if (offset != m->m_pkthdr.len) 1297 V_pfsyncstats.pfsyncs_badlen++; 1298 1299 /* we're done. free and let the caller return */ 1300 m_freem(m); 1301 return (-1); 1302 } 1303 1304 static int 1305 pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) 1306 { 1307 V_pfsyncstats.pfsyncs_badact++; 1308 1309 m_freem(m); 1310 return (-1); 1311 } 1312 1313 static int 1314 pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, 1315 struct route *rt) 1316 { 1317 m_freem(m); 1318 return (0); 1319 } 1320 1321 /* ARGSUSED */ 1322 static int 1323 pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1324 { 1325 struct pfsync_softc *sc = ifp->if_softc; 1326 struct ifreq *ifr = (struct ifreq *)data; 1327 struct pfsyncreq pfsyncr; 1328 int error; 1329 int c; 1330 1331 switch (cmd) { 1332 case SIOCSIFFLAGS: 1333 PFSYNC_LOCK(sc); 1334 if (ifp->if_flags & IFF_UP) { 1335 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1336 PFSYNC_UNLOCK(sc); 1337 pfsync_pointers_init(); 1338 } else { 1339 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1340 PFSYNC_UNLOCK(sc); 1341 pfsync_pointers_uninit(); 1342 } 1343 break; 1344 case SIOCSIFMTU: 1345 if (!sc->sc_sync_if || 1346 ifr->ifr_mtu <= PFSYNC_MINPKT || 1347 ifr->ifr_mtu > sc->sc_sync_if->if_mtu) 1348 return (EINVAL); 1349 if (ifr->ifr_mtu < ifp->if_mtu) { 1350 for (c = 0; c < pfsync_buckets; c++) { 1351 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]); 1352 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT) 1353 pfsync_sendout(1, c); 1354 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]); 1355 } 1356 } 1357 ifp->if_mtu = ifr->ifr_mtu; 1358 break; 1359 case SIOCGETPFSYNC: 1360 bzero(&pfsyncr, sizeof(pfsyncr)); 1361 PFSYNC_LOCK(sc); 1362 if (sc->sc_sync_if) { 1363 strlcpy(pfsyncr.pfsyncr_syncdev, 1364 sc->sc_sync_if->if_xname, IFNAMSIZ); 1365 } 1366 pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; 1367 pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; 1368 pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER == 1369 (sc->sc_flags & PFSYNCF_DEFER)); 1370 PFSYNC_UNLOCK(sc); 1371 return (copyout(&pfsyncr, ifr_data_get_ptr(ifr), 1372 sizeof(pfsyncr))); 1373 1374 case SIOCSETPFSYNC: 1375 { 1376 struct ip_moptions *imo = &sc->sc_imo; 1377 struct ifnet *sifp; 1378 struct ip *ip; 1379 void *mship = NULL; 1380 1381 if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0) 1382 return (error); 1383 if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr, 1384 sizeof(pfsyncr)))) 1385 return (error); 1386 1387 if (pfsyncr.pfsyncr_maxupdates > 255) 1388 return (EINVAL); 1389 1390 if (pfsyncr.pfsyncr_syncdev[0] == 0) 1391 sifp = NULL; 1392 else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL) 1393 return (EINVAL); 1394 1395 if (sifp != NULL && ( 1396 pfsyncr.pfsyncr_syncpeer.s_addr == 0 || 1397 pfsyncr.pfsyncr_syncpeer.s_addr == 1398 htonl(INADDR_PFSYNC_GROUP))) 1399 mship = malloc((sizeof(struct in_multi *) * 1400 IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO); 1401 1402 PFSYNC_LOCK(sc); 1403 if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) 1404 sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP); 1405 else 1406 sc->sc_sync_peer.s_addr = 1407 pfsyncr.pfsyncr_syncpeer.s_addr; 1408 1409 sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; 1410 if (pfsyncr.pfsyncr_defer) { 1411 sc->sc_flags |= PFSYNCF_DEFER; 1412 V_pfsync_defer_ptr = pfsync_defer; 1413 } else { 1414 sc->sc_flags &= ~PFSYNCF_DEFER; 1415 V_pfsync_defer_ptr = NULL; 1416 } 1417 1418 if (sifp == NULL) { 1419 if (sc->sc_sync_if) 1420 if_rele(sc->sc_sync_if); 1421 sc->sc_sync_if = NULL; 1422 if (imo->imo_membership) 1423 pfsync_multicast_cleanup(sc); 1424 PFSYNC_UNLOCK(sc); 1425 break; 1426 } 1427 1428 for (c = 0; c < pfsync_buckets; c++) { 1429 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[c]); 1430 if (sc->sc_buckets[c].b_len > PFSYNC_MINPKT && 1431 (sifp->if_mtu < sc->sc_ifp->if_mtu || 1432 (sc->sc_sync_if != NULL && 1433 sifp->if_mtu < sc->sc_sync_if->if_mtu) || 1434 sifp->if_mtu < MCLBYTES - sizeof(struct ip))) 1435 pfsync_sendout(1, c); 1436 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[c]); 1437 } 1438 1439 if (imo->imo_membership) 1440 pfsync_multicast_cleanup(sc); 1441 1442 if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) { 1443 error = pfsync_multicast_setup(sc, sifp, mship); 1444 if (error) { 1445 if_rele(sifp); 1446 free(mship, M_PFSYNC); 1447 PFSYNC_UNLOCK(sc); 1448 return (error); 1449 } 1450 } 1451 if (sc->sc_sync_if) 1452 if_rele(sc->sc_sync_if); 1453 sc->sc_sync_if = sifp; 1454 1455 ip = &sc->sc_template; 1456 bzero(ip, sizeof(*ip)); 1457 ip->ip_v = IPVERSION; 1458 ip->ip_hl = sizeof(sc->sc_template) >> 2; 1459 ip->ip_tos = IPTOS_LOWDELAY; 1460 /* len and id are set later. */ 1461 ip->ip_off = htons(IP_DF); 1462 ip->ip_ttl = PFSYNC_DFLTTL; 1463 ip->ip_p = IPPROTO_PFSYNC; 1464 ip->ip_src.s_addr = INADDR_ANY; 1465 ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr; 1466 1467 /* Request a full state table update. */ 1468 if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 1469 (*carp_demote_adj_p)(V_pfsync_carp_adj, 1470 "pfsync bulk start"); 1471 sc->sc_flags &= ~PFSYNCF_OK; 1472 if (V_pf_status.debug >= PF_DEBUG_MISC) 1473 printf("pfsync: requesting bulk update\n"); 1474 PFSYNC_UNLOCK(sc); 1475 PFSYNC_BUCKET_LOCK(&sc->sc_buckets[0]); 1476 pfsync_request_update(0, 0); 1477 PFSYNC_BUCKET_UNLOCK(&sc->sc_buckets[0]); 1478 PFSYNC_BLOCK(sc); 1479 sc->sc_ureq_sent = time_uptime; 1480 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, 1481 sc); 1482 PFSYNC_BUNLOCK(sc); 1483 1484 break; 1485 } 1486 default: 1487 return (ENOTTY); 1488 } 1489 1490 return (0); 1491 } 1492 1493 static void 1494 pfsync_out_state(struct pf_state *st, void *buf) 1495 { 1496 struct pfsync_state *sp = buf; 1497 1498 pfsync_state_export(sp, st); 1499 } 1500 1501 static void 1502 pfsync_out_iack(struct pf_state *st, void *buf) 1503 { 1504 struct pfsync_ins_ack *iack = buf; 1505 1506 iack->id = st->id; 1507 iack->creatorid = st->creatorid; 1508 } 1509 1510 static void 1511 pfsync_out_upd_c(struct pf_state *st, void *buf) 1512 { 1513 struct pfsync_upd_c *up = buf; 1514 1515 bzero(up, sizeof(*up)); 1516 up->id = st->id; 1517 pf_state_peer_hton(&st->src, &up->src); 1518 pf_state_peer_hton(&st->dst, &up->dst); 1519 up->creatorid = st->creatorid; 1520 up->timeout = st->timeout; 1521 } 1522 1523 static void 1524 pfsync_out_del(struct pf_state *st, void *buf) 1525 { 1526 struct pfsync_del_c *dp = buf; 1527 1528 dp->id = st->id; 1529 dp->creatorid = st->creatorid; 1530 st->state_flags |= PFSTATE_NOSYNC; 1531 } 1532 1533 static void 1534 pfsync_drop(struct pfsync_softc *sc) 1535 { 1536 struct pf_state *st, *next; 1537 struct pfsync_upd_req_item *ur; 1538 struct pfsync_bucket *b; 1539 int c, q; 1540 1541 for (c = 0; c < pfsync_buckets; c++) { 1542 b = &sc->sc_buckets[c]; 1543 for (q = 0; q < PFSYNC_S_COUNT; q++) { 1544 if (TAILQ_EMPTY(&b->b_qs[q])) 1545 continue; 1546 1547 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, next) { 1548 KASSERT(st->sync_state == q, 1549 ("%s: st->sync_state == q", 1550 __func__)); 1551 st->sync_state = PFSYNC_S_NONE; 1552 pf_release_state(st); 1553 } 1554 TAILQ_INIT(&b->b_qs[q]); 1555 } 1556 1557 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) { 1558 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry); 1559 free(ur, M_PFSYNC); 1560 } 1561 1562 b->b_len = PFSYNC_MINPKT; 1563 b->b_plus = NULL; 1564 } 1565 } 1566 1567 static void 1568 pfsync_sendout(int schedswi, int c) 1569 { 1570 struct pfsync_softc *sc = V_pfsyncif; 1571 struct ifnet *ifp = sc->sc_ifp; 1572 struct mbuf *m; 1573 struct ip *ip; 1574 struct pfsync_header *ph; 1575 struct pfsync_subheader *subh; 1576 struct pf_state *st, *st_next; 1577 struct pfsync_upd_req_item *ur; 1578 struct pfsync_bucket *b = &sc->sc_buckets[c]; 1579 int offset; 1580 int q, count = 0; 1581 1582 KASSERT(sc != NULL, ("%s: null sc", __func__)); 1583 KASSERT(b->b_len > PFSYNC_MINPKT, 1584 ("%s: sc_len %zu", __func__, b->b_len)); 1585 PFSYNC_BUCKET_LOCK_ASSERT(b); 1586 1587 if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) { 1588 pfsync_drop(sc); 1589 return; 1590 } 1591 1592 m = m_get2(max_linkhdr + b->b_len, M_NOWAIT, MT_DATA, M_PKTHDR); 1593 if (m == NULL) { 1594 if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1); 1595 V_pfsyncstats.pfsyncs_onomem++; 1596 return; 1597 } 1598 m->m_data += max_linkhdr; 1599 m->m_len = m->m_pkthdr.len = b->b_len; 1600 1601 /* build the ip header */ 1602 ip = (struct ip *)m->m_data; 1603 bcopy(&sc->sc_template, ip, sizeof(*ip)); 1604 offset = sizeof(*ip); 1605 1606 ip->ip_len = htons(m->m_pkthdr.len); 1607 ip_fillid(ip); 1608 1609 /* build the pfsync header */ 1610 ph = (struct pfsync_header *)(m->m_data + offset); 1611 bzero(ph, sizeof(*ph)); 1612 offset += sizeof(*ph); 1613 1614 ph->version = PFSYNC_VERSION; 1615 ph->len = htons(b->b_len - sizeof(*ip)); 1616 bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH); 1617 1618 /* walk the queues */ 1619 for (q = 0; q < PFSYNC_S_COUNT; q++) { 1620 if (TAILQ_EMPTY(&b->b_qs[q])) 1621 continue; 1622 1623 subh = (struct pfsync_subheader *)(m->m_data + offset); 1624 offset += sizeof(*subh); 1625 1626 count = 0; 1627 TAILQ_FOREACH_SAFE(st, &b->b_qs[q], sync_list, st_next) { 1628 KASSERT(st->sync_state == q, 1629 ("%s: st->sync_state == q", 1630 __func__)); 1631 /* 1632 * XXXGL: some of write methods do unlocked reads 1633 * of state data :( 1634 */ 1635 pfsync_qs[q].write(st, m->m_data + offset); 1636 offset += pfsync_qs[q].len; 1637 st->sync_state = PFSYNC_S_NONE; 1638 pf_release_state(st); 1639 count++; 1640 } 1641 TAILQ_INIT(&b->b_qs[q]); 1642 1643 bzero(subh, sizeof(*subh)); 1644 subh->action = pfsync_qs[q].action; 1645 subh->count = htons(count); 1646 V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count; 1647 } 1648 1649 if (!TAILQ_EMPTY(&b->b_upd_req_list)) { 1650 subh = (struct pfsync_subheader *)(m->m_data + offset); 1651 offset += sizeof(*subh); 1652 1653 count = 0; 1654 while ((ur = TAILQ_FIRST(&b->b_upd_req_list)) != NULL) { 1655 TAILQ_REMOVE(&b->b_upd_req_list, ur, ur_entry); 1656 1657 bcopy(&ur->ur_msg, m->m_data + offset, 1658 sizeof(ur->ur_msg)); 1659 offset += sizeof(ur->ur_msg); 1660 free(ur, M_PFSYNC); 1661 count++; 1662 } 1663 1664 bzero(subh, sizeof(*subh)); 1665 subh->action = PFSYNC_ACT_UPD_REQ; 1666 subh->count = htons(count); 1667 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count; 1668 } 1669 1670 /* has someone built a custom region for us to add? */ 1671 if (b->b_plus != NULL) { 1672 bcopy(b->b_plus, m->m_data + offset, b->b_pluslen); 1673 offset += b->b_pluslen; 1674 1675 b->b_plus = NULL; 1676 } 1677 1678 subh = (struct pfsync_subheader *)(m->m_data + offset); 1679 offset += sizeof(*subh); 1680 1681 bzero(subh, sizeof(*subh)); 1682 subh->action = PFSYNC_ACT_EOF; 1683 subh->count = htons(1); 1684 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++; 1685 1686 /* we're done, let's put it on the wire */ 1687 if (ifp->if_bpf) { 1688 m->m_data += sizeof(*ip); 1689 m->m_len = m->m_pkthdr.len = b->b_len - sizeof(*ip); 1690 BPF_MTAP(ifp, m); 1691 m->m_data -= sizeof(*ip); 1692 m->m_len = m->m_pkthdr.len = b->b_len; 1693 } 1694 1695 if (sc->sc_sync_if == NULL) { 1696 b->b_len = PFSYNC_MINPKT; 1697 m_freem(m); 1698 return; 1699 } 1700 1701 if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1); 1702 if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len); 1703 b->b_len = PFSYNC_MINPKT; 1704 1705 if (!_IF_QFULL(&b->b_snd)) 1706 _IF_ENQUEUE(&b->b_snd, m); 1707 else { 1708 m_freem(m); 1709 if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1); 1710 } 1711 if (schedswi) 1712 swi_sched(V_pfsync_swi_cookie, 0); 1713 } 1714 1715 static void 1716 pfsync_insert_state(struct pf_state *st) 1717 { 1718 struct pfsync_softc *sc = V_pfsyncif; 1719 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1720 1721 if (st->state_flags & PFSTATE_NOSYNC) 1722 return; 1723 1724 if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) || 1725 st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) { 1726 st->state_flags |= PFSTATE_NOSYNC; 1727 return; 1728 } 1729 1730 KASSERT(st->sync_state == PFSYNC_S_NONE, 1731 ("%s: st->sync_state %u", __func__, st->sync_state)); 1732 1733 PFSYNC_BUCKET_LOCK(b); 1734 if (b->b_len == PFSYNC_MINPKT) 1735 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b); 1736 1737 pfsync_q_ins(st, PFSYNC_S_INS, true); 1738 PFSYNC_BUCKET_UNLOCK(b); 1739 1740 st->sync_updates = 0; 1741 } 1742 1743 static int 1744 pfsync_defer(struct pf_state *st, struct mbuf *m) 1745 { 1746 struct pfsync_softc *sc = V_pfsyncif; 1747 struct pfsync_deferral *pd; 1748 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1749 1750 if (m->m_flags & (M_BCAST|M_MCAST)) 1751 return (0); 1752 1753 PFSYNC_LOCK(sc); 1754 1755 if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) || 1756 !(sc->sc_flags & PFSYNCF_DEFER)) { 1757 PFSYNC_UNLOCK(sc); 1758 return (0); 1759 } 1760 1761 if (b->b_deferred >= 128) 1762 pfsync_undefer(TAILQ_FIRST(&b->b_deferrals), 0); 1763 1764 pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT); 1765 if (pd == NULL) 1766 return (0); 1767 b->b_deferred++; 1768 1769 m->m_flags |= M_SKIP_FIREWALL; 1770 st->state_flags |= PFSTATE_ACK; 1771 1772 pd->pd_sc = sc; 1773 pd->pd_refs = 0; 1774 pd->pd_st = st; 1775 pf_ref_state(st); 1776 pd->pd_m = m; 1777 1778 TAILQ_INSERT_TAIL(&b->b_deferrals, pd, pd_entry); 1779 callout_init_mtx(&pd->pd_tmo, &b->b_mtx, CALLOUT_RETURNUNLOCKED); 1780 callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd); 1781 1782 pfsync_push(b); 1783 1784 return (1); 1785 } 1786 1787 static void 1788 pfsync_undefer(struct pfsync_deferral *pd, int drop) 1789 { 1790 struct pfsync_softc *sc = pd->pd_sc; 1791 struct mbuf *m = pd->pd_m; 1792 struct pf_state *st = pd->pd_st; 1793 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1794 1795 PFSYNC_BUCKET_LOCK_ASSERT(b); 1796 1797 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry); 1798 b->b_deferred--; 1799 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1800 free(pd, M_PFSYNC); 1801 pf_release_state(st); 1802 1803 if (drop) 1804 m_freem(m); 1805 else { 1806 _IF_ENQUEUE(&b->b_snd, m); 1807 pfsync_push(b); 1808 } 1809 } 1810 1811 static void 1812 pfsync_defer_tmo(void *arg) 1813 { 1814 struct pfsync_deferral *pd = arg; 1815 struct pfsync_softc *sc = pd->pd_sc; 1816 struct mbuf *m = pd->pd_m; 1817 struct pf_state *st = pd->pd_st; 1818 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1819 1820 PFSYNC_BUCKET_LOCK_ASSERT(b); 1821 1822 CURVNET_SET(m->m_pkthdr.rcvif->if_vnet); 1823 1824 TAILQ_REMOVE(&b->b_deferrals, pd, pd_entry); 1825 b->b_deferred--; 1826 pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ 1827 if (pd->pd_refs == 0) 1828 free(pd, M_PFSYNC); 1829 PFSYNC_UNLOCK(sc); 1830 1831 ip_output(m, NULL, NULL, 0, NULL, NULL); 1832 1833 pf_release_state(st); 1834 1835 CURVNET_RESTORE(); 1836 } 1837 1838 static void 1839 pfsync_undefer_state(struct pf_state *st, int drop) 1840 { 1841 struct pfsync_softc *sc = V_pfsyncif; 1842 struct pfsync_deferral *pd; 1843 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1844 1845 PFSYNC_BUCKET_LOCK(b); 1846 1847 TAILQ_FOREACH(pd, &b->b_deferrals, pd_entry) { 1848 if (pd->pd_st == st) { 1849 if (callout_stop(&pd->pd_tmo) > 0) 1850 pfsync_undefer(pd, drop); 1851 1852 PFSYNC_BUCKET_UNLOCK(b); 1853 return; 1854 } 1855 } 1856 PFSYNC_BUCKET_UNLOCK(b); 1857 1858 panic("%s: unable to find deferred state", __func__); 1859 } 1860 1861 static struct pfsync_bucket* 1862 pfsync_get_bucket(struct pfsync_softc *sc, struct pf_state *st) 1863 { 1864 int c = PF_IDHASH(st) % pfsync_buckets; 1865 return &sc->sc_buckets[c]; 1866 } 1867 1868 static void 1869 pfsync_update_state(struct pf_state *st) 1870 { 1871 struct pfsync_softc *sc = V_pfsyncif; 1872 bool sync = false, ref = true; 1873 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1874 1875 PF_STATE_LOCK_ASSERT(st); 1876 PFSYNC_BUCKET_LOCK(b); 1877 1878 if (st->state_flags & PFSTATE_ACK) 1879 pfsync_undefer_state(st, 0); 1880 if (st->state_flags & PFSTATE_NOSYNC) { 1881 if (st->sync_state != PFSYNC_S_NONE) 1882 pfsync_q_del(st, true, b); 1883 PFSYNC_BUCKET_UNLOCK(b); 1884 return; 1885 } 1886 1887 if (b->b_len == PFSYNC_MINPKT) 1888 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b); 1889 1890 switch (st->sync_state) { 1891 case PFSYNC_S_UPD_C: 1892 case PFSYNC_S_UPD: 1893 case PFSYNC_S_INS: 1894 /* we're already handling it */ 1895 1896 if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) { 1897 st->sync_updates++; 1898 if (st->sync_updates >= sc->sc_maxupdates) 1899 sync = true; 1900 } 1901 break; 1902 1903 case PFSYNC_S_IACK: 1904 pfsync_q_del(st, false, b); 1905 ref = false; 1906 /* FALLTHROUGH */ 1907 1908 case PFSYNC_S_NONE: 1909 pfsync_q_ins(st, PFSYNC_S_UPD_C, ref); 1910 st->sync_updates = 0; 1911 break; 1912 1913 default: 1914 panic("%s: unexpected sync state %d", __func__, st->sync_state); 1915 } 1916 1917 if (sync || (time_uptime - st->pfsync_time) < 2) 1918 pfsync_push(b); 1919 1920 PFSYNC_BUCKET_UNLOCK(b); 1921 } 1922 1923 static void 1924 pfsync_request_update(u_int32_t creatorid, u_int64_t id) 1925 { 1926 struct pfsync_softc *sc = V_pfsyncif; 1927 struct pfsync_bucket *b = &sc->sc_buckets[0]; 1928 struct pfsync_upd_req_item *item; 1929 size_t nlen = sizeof(struct pfsync_upd_req); 1930 1931 PFSYNC_BUCKET_LOCK_ASSERT(b); 1932 1933 /* 1934 * This code does a bit to prevent multiple update requests for the 1935 * same state being generated. It searches current subheader queue, 1936 * but it doesn't lookup into queue of already packed datagrams. 1937 */ 1938 TAILQ_FOREACH(item, &b->b_upd_req_list, ur_entry) 1939 if (item->ur_msg.id == id && 1940 item->ur_msg.creatorid == creatorid) 1941 return; 1942 1943 item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT); 1944 if (item == NULL) 1945 return; /* XXX stats */ 1946 1947 item->ur_msg.id = id; 1948 item->ur_msg.creatorid = creatorid; 1949 1950 if (TAILQ_EMPTY(&b->b_upd_req_list)) 1951 nlen += sizeof(struct pfsync_subheader); 1952 1953 if (b->b_len + nlen > sc->sc_ifp->if_mtu) { 1954 pfsync_sendout(1, 0); 1955 1956 nlen = sizeof(struct pfsync_subheader) + 1957 sizeof(struct pfsync_upd_req); 1958 } 1959 1960 TAILQ_INSERT_TAIL(&b->b_upd_req_list, item, ur_entry); 1961 b->b_len += nlen; 1962 } 1963 1964 static bool 1965 pfsync_update_state_req(struct pf_state *st) 1966 { 1967 struct pfsync_softc *sc = V_pfsyncif; 1968 bool ref = true, full = false; 1969 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 1970 1971 PF_STATE_LOCK_ASSERT(st); 1972 PFSYNC_BUCKET_LOCK(b); 1973 1974 if (st->state_flags & PFSTATE_NOSYNC) { 1975 if (st->sync_state != PFSYNC_S_NONE) 1976 pfsync_q_del(st, true, b); 1977 PFSYNC_BUCKET_UNLOCK(b); 1978 return (full); 1979 } 1980 1981 switch (st->sync_state) { 1982 case PFSYNC_S_UPD_C: 1983 case PFSYNC_S_IACK: 1984 pfsync_q_del(st, false, b); 1985 ref = false; 1986 /* FALLTHROUGH */ 1987 1988 case PFSYNC_S_NONE: 1989 pfsync_q_ins(st, PFSYNC_S_UPD, ref); 1990 pfsync_push(b); 1991 break; 1992 1993 case PFSYNC_S_INS: 1994 case PFSYNC_S_UPD: 1995 case PFSYNC_S_DEL: 1996 /* we're already handling it */ 1997 break; 1998 1999 default: 2000 panic("%s: unexpected sync state %d", __func__, st->sync_state); 2001 } 2002 2003 if ((sc->sc_ifp->if_mtu - b->b_len) < sizeof(struct pfsync_state)) 2004 full = true; 2005 2006 PFSYNC_BUCKET_UNLOCK(b); 2007 2008 return (full); 2009 } 2010 2011 static void 2012 pfsync_delete_state(struct pf_state *st) 2013 { 2014 struct pfsync_softc *sc = V_pfsyncif; 2015 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 2016 bool ref = true; 2017 2018 PFSYNC_BUCKET_LOCK(b); 2019 if (st->state_flags & PFSTATE_ACK) 2020 pfsync_undefer_state(st, 1); 2021 if (st->state_flags & PFSTATE_NOSYNC) { 2022 if (st->sync_state != PFSYNC_S_NONE) 2023 pfsync_q_del(st, true, b); 2024 PFSYNC_BUCKET_UNLOCK(b); 2025 return; 2026 } 2027 2028 if (b->b_len == PFSYNC_MINPKT) 2029 callout_reset(&b->b_tmo, 1 * hz, pfsync_timeout, b); 2030 2031 switch (st->sync_state) { 2032 case PFSYNC_S_INS: 2033 /* We never got to tell the world so just forget about it. */ 2034 pfsync_q_del(st, true, b); 2035 break; 2036 2037 case PFSYNC_S_UPD_C: 2038 case PFSYNC_S_UPD: 2039 case PFSYNC_S_IACK: 2040 pfsync_q_del(st, false, b); 2041 ref = false; 2042 /* FALLTHROUGH */ 2043 2044 case PFSYNC_S_NONE: 2045 pfsync_q_ins(st, PFSYNC_S_DEL, ref); 2046 break; 2047 2048 default: 2049 panic("%s: unexpected sync state %d", __func__, st->sync_state); 2050 } 2051 2052 PFSYNC_BUCKET_UNLOCK(b); 2053 } 2054 2055 static void 2056 pfsync_clear_states(u_int32_t creatorid, const char *ifname) 2057 { 2058 struct { 2059 struct pfsync_subheader subh; 2060 struct pfsync_clr clr; 2061 } __packed r; 2062 2063 bzero(&r, sizeof(r)); 2064 2065 r.subh.action = PFSYNC_ACT_CLR; 2066 r.subh.count = htons(1); 2067 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++; 2068 2069 strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname)); 2070 r.clr.creatorid = creatorid; 2071 2072 pfsync_send_plus(&r, sizeof(r)); 2073 } 2074 2075 static void 2076 pfsync_q_ins(struct pf_state *st, int q, bool ref) 2077 { 2078 struct pfsync_softc *sc = V_pfsyncif; 2079 size_t nlen = pfsync_qs[q].len; 2080 struct pfsync_bucket *b = pfsync_get_bucket(sc, st); 2081 2082 PFSYNC_BUCKET_LOCK_ASSERT(b); 2083 2084 KASSERT(st->sync_state == PFSYNC_S_NONE, 2085 ("%s: st->sync_state %u", __func__, st->sync_state)); 2086 KASSERT(b->b_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu", 2087 b->b_len)); 2088 2089 if (TAILQ_EMPTY(&b->b_qs[q])) 2090 nlen += sizeof(struct pfsync_subheader); 2091 2092 if (b->b_len + nlen > sc->sc_ifp->if_mtu) { 2093 pfsync_sendout(1, b->b_id); 2094 2095 nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len; 2096 } 2097 2098 b->b_len += nlen; 2099 TAILQ_INSERT_TAIL(&b->b_qs[q], st, sync_list); 2100 st->sync_state = q; 2101 if (ref) 2102 pf_ref_state(st); 2103 } 2104 2105 static void 2106 pfsync_q_del(struct pf_state *st, bool unref, struct pfsync_bucket *b) 2107 { 2108 int q = st->sync_state; 2109 2110 PFSYNC_BUCKET_LOCK_ASSERT(b); 2111 KASSERT(st->sync_state != PFSYNC_S_NONE, 2112 ("%s: st->sync_state != PFSYNC_S_NONE", __func__)); 2113 2114 b->b_len -= pfsync_qs[q].len; 2115 TAILQ_REMOVE(&b->b_qs[q], st, sync_list); 2116 st->sync_state = PFSYNC_S_NONE; 2117 if (unref) 2118 pf_release_state(st); 2119 2120 if (TAILQ_EMPTY(&b->b_qs[q])) 2121 b->b_len -= sizeof(struct pfsync_subheader); 2122 } 2123 2124 static void 2125 pfsync_bulk_start(void) 2126 { 2127 struct pfsync_softc *sc = V_pfsyncif; 2128 2129 if (V_pf_status.debug >= PF_DEBUG_MISC) 2130 printf("pfsync: received bulk update request\n"); 2131 2132 PFSYNC_BLOCK(sc); 2133 2134 sc->sc_ureq_received = time_uptime; 2135 sc->sc_bulk_hashid = 0; 2136 sc->sc_bulk_stateid = 0; 2137 pfsync_bulk_status(PFSYNC_BUS_START); 2138 callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc); 2139 PFSYNC_BUNLOCK(sc); 2140 } 2141 2142 static void 2143 pfsync_bulk_update(void *arg) 2144 { 2145 struct pfsync_softc *sc = arg; 2146 struct pf_state *s; 2147 int i, sent = 0; 2148 2149 PFSYNC_BLOCK_ASSERT(sc); 2150 CURVNET_SET(sc->sc_ifp->if_vnet); 2151 2152 /* 2153 * Start with last state from previous invocation. 2154 * It may had gone, in this case start from the 2155 * hash slot. 2156 */ 2157 s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid); 2158 2159 if (s != NULL) 2160 i = PF_IDHASH(s); 2161 else 2162 i = sc->sc_bulk_hashid; 2163 2164 for (; i <= pf_hashmask; i++) { 2165 struct pf_idhash *ih = &V_pf_idhash[i]; 2166 2167 if (s != NULL) 2168 PF_HASHROW_ASSERT(ih); 2169 else { 2170 PF_HASHROW_LOCK(ih); 2171 s = LIST_FIRST(&ih->states); 2172 } 2173 2174 for (; s; s = LIST_NEXT(s, entry)) { 2175 if (s->sync_state == PFSYNC_S_NONE && 2176 s->timeout < PFTM_MAX && 2177 s->pfsync_time <= sc->sc_ureq_received) { 2178 if (pfsync_update_state_req(s)) { 2179 /* We've filled a packet. */ 2180 sc->sc_bulk_hashid = i; 2181 sc->sc_bulk_stateid = s->id; 2182 sc->sc_bulk_creatorid = s->creatorid; 2183 PF_HASHROW_UNLOCK(ih); 2184 callout_reset(&sc->sc_bulk_tmo, 1, 2185 pfsync_bulk_update, sc); 2186 goto full; 2187 } 2188 sent++; 2189 } 2190 } 2191 PF_HASHROW_UNLOCK(ih); 2192 } 2193 2194 /* We're done. */ 2195 pfsync_bulk_status(PFSYNC_BUS_END); 2196 full: 2197 CURVNET_RESTORE(); 2198 } 2199 2200 static void 2201 pfsync_bulk_status(u_int8_t status) 2202 { 2203 struct { 2204 struct pfsync_subheader subh; 2205 struct pfsync_bus bus; 2206 } __packed r; 2207 2208 struct pfsync_softc *sc = V_pfsyncif; 2209 2210 bzero(&r, sizeof(r)); 2211 2212 r.subh.action = PFSYNC_ACT_BUS; 2213 r.subh.count = htons(1); 2214 V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++; 2215 2216 r.bus.creatorid = V_pf_status.hostid; 2217 r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received); 2218 r.bus.status = status; 2219 2220 pfsync_send_plus(&r, sizeof(r)); 2221 } 2222 2223 static void 2224 pfsync_bulk_fail(void *arg) 2225 { 2226 struct pfsync_softc *sc = arg; 2227 struct pfsync_bucket *b = &sc->sc_buckets[0]; 2228 2229 CURVNET_SET(sc->sc_ifp->if_vnet); 2230 2231 PFSYNC_BLOCK_ASSERT(sc); 2232 2233 if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { 2234 /* Try again */ 2235 callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, 2236 pfsync_bulk_fail, V_pfsyncif); 2237 PFSYNC_BUCKET_LOCK(b); 2238 pfsync_request_update(0, 0); 2239 PFSYNC_BUCKET_UNLOCK(b); 2240 } else { 2241 /* Pretend like the transfer was ok. */ 2242 sc->sc_ureq_sent = 0; 2243 sc->sc_bulk_tries = 0; 2244 PFSYNC_LOCK(sc); 2245 if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) 2246 (*carp_demote_adj_p)(-V_pfsync_carp_adj, 2247 "pfsync bulk fail"); 2248 sc->sc_flags |= PFSYNCF_OK; 2249 PFSYNC_UNLOCK(sc); 2250 if (V_pf_status.debug >= PF_DEBUG_MISC) 2251 printf("pfsync: failed to receive bulk update\n"); 2252 } 2253 2254 CURVNET_RESTORE(); 2255 } 2256 2257 static void 2258 pfsync_send_plus(void *plus, size_t pluslen) 2259 { 2260 struct pfsync_softc *sc = V_pfsyncif; 2261 struct pfsync_bucket *b = &sc->sc_buckets[0]; 2262 2263 PFSYNC_BUCKET_LOCK(b); 2264 2265 if (b->b_len + pluslen > sc->sc_ifp->if_mtu) 2266 pfsync_sendout(1, b->b_id); 2267 2268 b->b_plus = plus; 2269 b->b_len += (b->b_pluslen = pluslen); 2270 2271 pfsync_sendout(1, b->b_id); 2272 PFSYNC_BUCKET_UNLOCK(b); 2273 } 2274 2275 static void 2276 pfsync_timeout(void *arg) 2277 { 2278 struct pfsync_bucket *b = arg; 2279 2280 CURVNET_SET(b->b_sc->sc_ifp->if_vnet); 2281 PFSYNC_BUCKET_LOCK(b); 2282 pfsync_push(b); 2283 PFSYNC_BUCKET_UNLOCK(b); 2284 CURVNET_RESTORE(); 2285 } 2286 2287 static void 2288 pfsync_push(struct pfsync_bucket *b) 2289 { 2290 2291 PFSYNC_BUCKET_LOCK_ASSERT(b); 2292 2293 b->b_flags |= PFSYNCF_BUCKET_PUSH; 2294 swi_sched(V_pfsync_swi_cookie, 0); 2295 } 2296 2297 static void 2298 pfsync_push_all(struct pfsync_softc *sc) 2299 { 2300 int c; 2301 struct pfsync_bucket *b; 2302 2303 for (c = 0; c < pfsync_buckets; c++) { 2304 b = &sc->sc_buckets[c]; 2305 2306 PFSYNC_BUCKET_LOCK(b); 2307 pfsync_push(b); 2308 PFSYNC_BUCKET_UNLOCK(b); 2309 } 2310 } 2311 2312 static void 2313 pfsyncintr(void *arg) 2314 { 2315 struct pfsync_softc *sc = arg; 2316 struct pfsync_bucket *b; 2317 struct mbuf *m, *n; 2318 int c; 2319 2320 CURVNET_SET(sc->sc_ifp->if_vnet); 2321 2322 for (c = 0; c < pfsync_buckets; c++) { 2323 b = &sc->sc_buckets[c]; 2324 2325 PFSYNC_BUCKET_LOCK(b); 2326 if ((b->b_flags & PFSYNCF_BUCKET_PUSH) && b->b_len > PFSYNC_MINPKT) { 2327 pfsync_sendout(0, b->b_id); 2328 b->b_flags &= ~PFSYNCF_BUCKET_PUSH; 2329 } 2330 _IF_DEQUEUE_ALL(&b->b_snd, m); 2331 PFSYNC_BUCKET_UNLOCK(b); 2332 2333 for (; m != NULL; m = n) { 2334 2335 n = m->m_nextpkt; 2336 m->m_nextpkt = NULL; 2337 2338 /* 2339 * We distinguish between a deferral packet and our 2340 * own pfsync packet based on M_SKIP_FIREWALL 2341 * flag. This is XXX. 2342 */ 2343 if (m->m_flags & M_SKIP_FIREWALL) 2344 ip_output(m, NULL, NULL, 0, NULL, NULL); 2345 else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 2346 NULL) == 0) 2347 V_pfsyncstats.pfsyncs_opackets++; 2348 else 2349 V_pfsyncstats.pfsyncs_oerrors++; 2350 } 2351 } 2352 CURVNET_RESTORE(); 2353 } 2354 2355 static int 2356 pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship) 2357 { 2358 struct ip_moptions *imo = &sc->sc_imo; 2359 int error; 2360 2361 if (!(ifp->if_flags & IFF_MULTICAST)) 2362 return (EADDRNOTAVAIL); 2363 2364 imo->imo_membership = (struct in_multi **)mship; 2365 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS; 2366 imo->imo_multicast_vif = -1; 2367 2368 if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL, 2369 &imo->imo_membership[0])) != 0) { 2370 imo->imo_membership = NULL; 2371 return (error); 2372 } 2373 imo->imo_num_memberships++; 2374 imo->imo_multicast_ifp = ifp; 2375 imo->imo_multicast_ttl = PFSYNC_DFLTTL; 2376 imo->imo_multicast_loop = 0; 2377 2378 return (0); 2379 } 2380 2381 static void 2382 pfsync_multicast_cleanup(struct pfsync_softc *sc) 2383 { 2384 struct ip_moptions *imo = &sc->sc_imo; 2385 2386 in_leavegroup(imo->imo_membership[0], NULL); 2387 free(imo->imo_membership, M_PFSYNC); 2388 imo->imo_membership = NULL; 2389 imo->imo_multicast_ifp = NULL; 2390 } 2391 2392 void 2393 pfsync_detach_ifnet(struct ifnet *ifp) 2394 { 2395 struct pfsync_softc *sc = V_pfsyncif; 2396 2397 if (sc == NULL) 2398 return; 2399 2400 PFSYNC_LOCK(sc); 2401 2402 if (sc->sc_sync_if == ifp) { 2403 /* We don't need mutlicast cleanup here, because the interface 2404 * is going away. We do need to ensure we don't try to do 2405 * cleanup later. 2406 */ 2407 sc->sc_imo.imo_membership = NULL; 2408 sc->sc_imo.imo_multicast_ifp = NULL; 2409 sc->sc_sync_if = NULL; 2410 } 2411 2412 PFSYNC_UNLOCK(sc); 2413 } 2414 2415 #ifdef INET 2416 extern struct domain inetdomain; 2417 static struct protosw in_pfsync_protosw = { 2418 .pr_type = SOCK_RAW, 2419 .pr_domain = &inetdomain, 2420 .pr_protocol = IPPROTO_PFSYNC, 2421 .pr_flags = PR_ATOMIC|PR_ADDR, 2422 .pr_input = pfsync_input, 2423 .pr_output = rip_output, 2424 .pr_ctloutput = rip_ctloutput, 2425 .pr_usrreqs = &rip_usrreqs 2426 }; 2427 #endif 2428 2429 static void 2430 pfsync_pointers_init() 2431 { 2432 2433 PF_RULES_WLOCK(); 2434 V_pfsync_state_import_ptr = pfsync_state_import; 2435 V_pfsync_insert_state_ptr = pfsync_insert_state; 2436 V_pfsync_update_state_ptr = pfsync_update_state; 2437 V_pfsync_delete_state_ptr = pfsync_delete_state; 2438 V_pfsync_clear_states_ptr = pfsync_clear_states; 2439 V_pfsync_defer_ptr = pfsync_defer; 2440 PF_RULES_WUNLOCK(); 2441 } 2442 2443 static void 2444 pfsync_pointers_uninit() 2445 { 2446 2447 PF_RULES_WLOCK(); 2448 V_pfsync_state_import_ptr = NULL; 2449 V_pfsync_insert_state_ptr = NULL; 2450 V_pfsync_update_state_ptr = NULL; 2451 V_pfsync_delete_state_ptr = NULL; 2452 V_pfsync_clear_states_ptr = NULL; 2453 V_pfsync_defer_ptr = NULL; 2454 PF_RULES_WUNLOCK(); 2455 } 2456 2457 static void 2458 vnet_pfsync_init(const void *unused __unused) 2459 { 2460 int error; 2461 2462 V_pfsync_cloner = if_clone_simple(pfsyncname, 2463 pfsync_clone_create, pfsync_clone_destroy, 1); 2464 error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif, 2465 SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie); 2466 if (error) { 2467 if_clone_detach(V_pfsync_cloner); 2468 log(LOG_INFO, "swi_add() failed in %s\n", __func__); 2469 } 2470 2471 pfsync_pointers_init(); 2472 } 2473 VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY, 2474 vnet_pfsync_init, NULL); 2475 2476 static void 2477 vnet_pfsync_uninit(const void *unused __unused) 2478 { 2479 2480 pfsync_pointers_uninit(); 2481 2482 if_clone_detach(V_pfsync_cloner); 2483 swi_remove(V_pfsync_swi_cookie); 2484 } 2485 2486 VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_FOURTH, 2487 vnet_pfsync_uninit, NULL); 2488 2489 static int 2490 pfsync_init() 2491 { 2492 #ifdef INET 2493 int error; 2494 2495 pfsync_detach_ifnet_ptr = pfsync_detach_ifnet; 2496 2497 error = pf_proto_register(PF_INET, &in_pfsync_protosw); 2498 if (error) 2499 return (error); 2500 error = ipproto_register(IPPROTO_PFSYNC); 2501 if (error) { 2502 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); 2503 return (error); 2504 } 2505 #endif 2506 2507 return (0); 2508 } 2509 2510 static void 2511 pfsync_uninit() 2512 { 2513 pfsync_detach_ifnet_ptr = NULL; 2514 2515 #ifdef INET 2516 ipproto_unregister(IPPROTO_PFSYNC); 2517 pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); 2518 #endif 2519 } 2520 2521 static int 2522 pfsync_modevent(module_t mod, int type, void *data) 2523 { 2524 int error = 0; 2525 2526 switch (type) { 2527 case MOD_LOAD: 2528 error = pfsync_init(); 2529 break; 2530 case MOD_UNLOAD: 2531 pfsync_uninit(); 2532 break; 2533 default: 2534 error = EINVAL; 2535 break; 2536 } 2537 2538 return (error); 2539 } 2540 2541 static moduledata_t pfsync_mod = { 2542 pfsyncname, 2543 pfsync_modevent, 2544 0 2545 }; 2546 2547 #define PFSYNC_MODVER 1 2548 2549 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */ 2550 DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY); 2551 MODULE_VERSION(pfsync, PFSYNC_MODVER); 2552 MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER); 2553