1 /* $OpenBSD: pf.c,v 1.1188 2023/10/10 16:26:06 bluhm Exp $ */ 2 3 /* 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2013 Henning Brauer <henning@openbsd.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 * 12 * - Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * - Redistributions in binary form must reproduce the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer in the documentation and/or other materials provided 17 * with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 20 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 21 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 22 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 23 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 24 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 25 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 26 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 27 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 29 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 * 32 * Effort sponsored in part by the Defense Advanced Research Projects 33 * Agency (DARPA) and Air Force Research Laboratory, Air Force 34 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 35 * 36 */ 37 38 #include "bpfilter.h" 39 #include "carp.h" 40 #include "pflog.h" 41 #include "pfsync.h" 42 #include "pflow.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/mbuf.h> 47 #include <sys/filio.h> 48 #include <sys/socket.h> 49 #include <sys/socketvar.h> 50 #include <sys/kernel.h> 51 #include <sys/time.h> 52 #include <sys/pool.h> 53 #include <sys/proc.h> 54 #include <sys/rwlock.h> 55 #include <sys/syslog.h> 56 57 #include <crypto/sha2.h> 58 59 #include <net/if.h> 60 #include <net/if_var.h> 61 #include <net/if_types.h> 62 #include <net/route.h> 63 #include <net/toeplitz.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_var.h> 67 #include <netinet/ip.h> 68 #include <netinet/in_pcb.h> 69 #include <netinet/ip_var.h> 70 #include <netinet/ip_icmp.h> 71 #include <netinet/icmp_var.h> 72 #include <netinet/tcp.h> 73 #include <netinet/tcp_seq.h> 74 #include <netinet/tcp_timer.h> 75 #include <netinet/tcp_var.h> 76 #include <netinet/tcp_fsm.h> 77 #include <netinet/udp.h> 78 #include <netinet/udp_var.h> 79 #include <netinet/ip_divert.h> 80 81 #ifdef INET6 82 #include <netinet6/in6_var.h> 83 #include <netinet/ip6.h> 84 #include <netinet6/ip6_var.h> 85 #include <netinet/icmp6.h> 86 #include <netinet6/nd6.h> 87 #include <netinet6/ip6_divert.h> 88 #endif /* INET6 */ 89 90 #include <net/pfvar.h> 91 #include <net/pfvar_priv.h> 92 93 #if NPFLOG > 0 94 #include <net/if_pflog.h> 95 #endif /* NPFLOG > 0 */ 96 97 #if NPFLOW > 0 98 #include <net/if_pflow.h> 99 #endif /* NPFLOW > 0 */ 100 101 #if NPFSYNC > 0 102 #include <net/if_pfsync.h> 103 #endif /* NPFSYNC > 0 */ 104 105 /* 106 * Global variables 107 */ 108 struct pf_state_tree pf_statetbl; 109 struct pf_queuehead pf_queues[2]; 110 struct pf_queuehead *pf_queues_active; 111 struct pf_queuehead *pf_queues_inactive; 112 113 struct pf_status pf_status; 114 115 int pf_hdr_limit = 20; /* arbitrary limit, tune in ddb */ 116 117 SHA2_CTX pf_tcp_secret_ctx; 118 u_char pf_tcp_secret[16]; 119 int pf_tcp_secret_init; 120 int pf_tcp_iss_off; 121 122 enum pf_test_status { 123 PF_TEST_FAIL = -1, 124 PF_TEST_OK, 125 PF_TEST_QUICK 126 }; 127 128 struct pf_test_ctx { 129 struct pf_pdesc *pd; 130 struct pf_rule_actions act; 131 u_int8_t icmpcode; 132 u_int8_t icmptype; 133 int icmp_dir; 134 int state_icmp; 135 int tag; 136 u_short reason; 137 struct pf_rule_item *ri; 138 struct pf_src_node *sns[PF_SN_MAX]; 139 struct pf_rule_slist rules; 140 struct pf_rule *nr; 141 struct pf_rule **rm; 142 struct pf_rule *a; 143 struct pf_rule **am; 144 struct pf_ruleset **rsm; 145 struct pf_ruleset *arsm; 146 struct pf_ruleset *aruleset; 147 struct tcphdr *th; 148 }; 149 150 struct pool pf_src_tree_pl, pf_rule_pl, pf_queue_pl; 151 struct pool pf_state_pl, pf_state_key_pl, pf_state_item_pl; 152 struct pool pf_rule_item_pl, pf_sn_item_pl, pf_pktdelay_pl; 153 154 void pf_add_threshold(struct pf_threshold *); 155 int pf_check_threshold(struct pf_threshold *); 156 int pf_check_tcp_cksum(struct mbuf *, int, int, 157 sa_family_t); 158 __inline void pf_cksum_fixup(u_int16_t *, u_int16_t, u_int16_t, 159 u_int8_t); 160 void pf_cksum_fixup_a(u_int16_t *, const struct pf_addr *, 161 const struct pf_addr *, sa_family_t, u_int8_t); 162 int pf_modulate_sack(struct pf_pdesc *, 163 struct pf_state_peer *); 164 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 165 u_int16_t *, u_int16_t *); 166 int pf_change_icmp_af(struct mbuf *, int, 167 struct pf_pdesc *, struct pf_pdesc *, 168 struct pf_addr *, struct pf_addr *, sa_family_t, 169 sa_family_t); 170 int pf_translate_a(struct pf_pdesc *, struct pf_addr *, 171 struct pf_addr *); 172 void pf_translate_icmp(struct pf_pdesc *, struct pf_addr *, 173 u_int16_t *, struct pf_addr *, struct pf_addr *, 174 u_int16_t); 175 int pf_translate_icmp_af(struct pf_pdesc*, int, void *); 176 void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, int, 177 sa_family_t, struct pf_rule *, u_int); 178 void pf_detach_state(struct pf_state *); 179 struct pf_state_key *pf_state_key_attach(struct pf_state_key *, 180 struct pf_state *, int); 181 void pf_state_key_detach(struct pf_state *, int); 182 u_int32_t pf_tcp_iss(struct pf_pdesc *); 183 void pf_rule_to_actions(struct pf_rule *, 184 struct pf_rule_actions *); 185 int pf_test_rule(struct pf_pdesc *, struct pf_rule **, 186 struct pf_state **, struct pf_rule **, 187 struct pf_ruleset **, u_short *); 188 static __inline int pf_create_state(struct pf_pdesc *, struct pf_rule *, 189 struct pf_rule *, struct pf_rule *, 190 struct pf_state_key **, struct pf_state_key **, 191 int *, struct pf_state **, int, 192 struct pf_rule_slist *, struct pf_rule_actions *, 193 struct pf_src_node **); 194 static __inline int pf_state_key_addr_setup(struct pf_pdesc *, void *, 195 int, struct pf_addr *, int, struct pf_addr *, 196 int, int); 197 int pf_state_key_setup(struct pf_pdesc *, struct 198 pf_state_key **, struct pf_state_key **, int); 199 int pf_tcp_track_full(struct pf_pdesc *, 200 struct pf_state **, u_short *, int *, int); 201 int pf_tcp_track_sloppy(struct pf_pdesc *, 202 struct pf_state **, u_short *); 203 static __inline int pf_synproxy(struct pf_pdesc *, struct pf_state **, 204 u_short *); 205 int pf_test_state(struct pf_pdesc *, struct pf_state **, 206 u_short *); 207 int pf_icmp_state_lookup(struct pf_pdesc *, 208 struct pf_state_key_cmp *, struct pf_state **, 209 u_int16_t, u_int16_t, int, int *, int, int); 210 int pf_test_state_icmp(struct pf_pdesc *, 211 struct pf_state **, u_short *); 212 u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, int, 213 u_int16_t); 214 static __inline int pf_set_rt_ifp(struct pf_state *, struct pf_addr *, 215 sa_family_t, struct pf_src_node **); 216 struct pf_divert *pf_get_divert(struct mbuf *); 217 int pf_walk_option(struct pf_pdesc *, struct ip *, 218 int, int, u_short *); 219 int pf_walk_header(struct pf_pdesc *, struct ip *, 220 u_short *); 221 int pf_walk_option6(struct pf_pdesc *, struct ip6_hdr *, 222 int, int, u_short *); 223 int pf_walk_header6(struct pf_pdesc *, struct ip6_hdr *, 224 u_short *); 225 void pf_print_state_parts(struct pf_state *, 226 struct pf_state_key *, struct pf_state_key *); 227 int pf_addr_wrap_neq(struct pf_addr_wrap *, 228 struct pf_addr_wrap *); 229 int pf_compare_state_keys(struct pf_state_key *, 230 struct pf_state_key *, struct pfi_kif *, u_int); 231 u_int16_t pf_pkt_hash(sa_family_t, uint8_t, 232 const struct pf_addr *, const struct pf_addr *, 233 uint16_t, uint16_t); 234 int pf_find_state(struct pf_pdesc *, 235 struct pf_state_key_cmp *, struct pf_state **); 236 int pf_src_connlimit(struct pf_state **); 237 int pf_match_rcvif(struct mbuf *, struct pf_rule *); 238 int pf_step_into_anchor(struct pf_test_ctx *, 239 struct pf_rule *); 240 int pf_match_rule(struct pf_test_ctx *, 241 struct pf_ruleset *); 242 void pf_counters_inc(int, struct pf_pdesc *, 243 struct pf_state *, struct pf_rule *, 244 struct pf_rule *); 245 246 int pf_state_insert(struct pfi_kif *, 247 struct pf_state_key **, struct pf_state_key **, 248 struct pf_state *); 249 250 int pf_state_key_isvalid(struct pf_state_key *); 251 struct pf_state_key *pf_state_key_ref(struct pf_state_key *); 252 void pf_state_key_unref(struct pf_state_key *); 253 void pf_state_key_link_reverse(struct pf_state_key *, 254 struct pf_state_key *); 255 void pf_state_key_unlink_reverse(struct pf_state_key *); 256 void pf_state_key_link_inpcb(struct pf_state_key *, 257 struct inpcb *); 258 void pf_state_key_unlink_inpcb(struct pf_state_key *); 259 void pf_inpcb_unlink_state_key(struct inpcb *); 260 void pf_pktenqueue_delayed(void *); 261 int32_t pf_state_expires(const struct pf_state *, uint8_t); 262 263 #if NPFLOG > 0 264 void pf_log_matches(struct pf_pdesc *, struct pf_rule *, 265 struct pf_rule *, struct pf_ruleset *, 266 struct pf_rule_slist *); 267 #endif /* NPFLOG > 0 */ 268 269 extern struct pool pfr_ktable_pl; 270 extern struct pool pfr_kentry_pl; 271 272 struct pf_pool_limit pf_pool_limits[PF_LIMIT_MAX] = { 273 { &pf_state_pl, PFSTATE_HIWAT, PFSTATE_HIWAT }, 274 { &pf_src_tree_pl, PFSNODE_HIWAT, PFSNODE_HIWAT }, 275 { &pf_frent_pl, PFFRAG_FRENT_HIWAT, PFFRAG_FRENT_HIWAT }, 276 { &pfr_ktable_pl, PFR_KTABLE_HIWAT, PFR_KTABLE_HIWAT }, 277 { &pfr_kentry_pl, PFR_KENTRY_HIWAT, PFR_KENTRY_HIWAT }, 278 { &pf_pktdelay_pl, PF_PKTDELAY_MAXPKTS, PF_PKTDELAY_MAXPKTS }, 279 { &pf_anchor_pl, PF_ANCHOR_HIWAT, PF_ANCHOR_HIWAT } 280 }; 281 282 #define BOUND_IFACE(r, k) \ 283 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : pfi_all 284 285 #define STATE_INC_COUNTERS(s) \ 286 do { \ 287 struct pf_rule_item *mrm; \ 288 s->rule.ptr->states_cur++; \ 289 s->rule.ptr->states_tot++; \ 290 if (s->anchor.ptr != NULL) { \ 291 s->anchor.ptr->states_cur++; \ 292 s->anchor.ptr->states_tot++; \ 293 } \ 294 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 295 mrm->r->states_cur++; \ 296 } while (0) 297 298 static __inline int pf_src_compare(struct pf_src_node *, struct pf_src_node *); 299 static inline int pf_state_compare_key(const struct pf_state_key *, 300 const struct pf_state_key *); 301 static inline int pf_state_compare_id(const struct pf_state *, 302 const struct pf_state *); 303 #ifdef INET6 304 static __inline void pf_cksum_uncover(u_int16_t *, u_int16_t, u_int8_t); 305 static __inline void pf_cksum_cover(u_int16_t *, u_int16_t, u_int8_t); 306 #endif /* INET6 */ 307 static __inline void pf_set_protostate(struct pf_state *, int, u_int8_t); 308 309 struct pf_src_tree tree_src_tracking; 310 311 struct pf_state_tree_id tree_id; 312 struct pf_state_list pf_state_list = PF_STATE_LIST_INITIALIZER(pf_state_list); 313 314 RB_GENERATE(pf_src_tree, pf_src_node, entry, pf_src_compare); 315 RBT_GENERATE(pf_state_tree, pf_state_key, sk_entry, pf_state_compare_key); 316 RBT_GENERATE(pf_state_tree_id, pf_state, entry_id, pf_state_compare_id); 317 318 int 319 pf_addr_compare(const struct pf_addr *a, const struct pf_addr *b, 320 sa_family_t af) 321 { 322 switch (af) { 323 case AF_INET: 324 if (a->addr32[0] > b->addr32[0]) 325 return (1); 326 if (a->addr32[0] < b->addr32[0]) 327 return (-1); 328 break; 329 #ifdef INET6 330 case AF_INET6: 331 if (a->addr32[3] > b->addr32[3]) 332 return (1); 333 if (a->addr32[3] < b->addr32[3]) 334 return (-1); 335 if (a->addr32[2] > b->addr32[2]) 336 return (1); 337 if (a->addr32[2] < b->addr32[2]) 338 return (-1); 339 if (a->addr32[1] > b->addr32[1]) 340 return (1); 341 if (a->addr32[1] < b->addr32[1]) 342 return (-1); 343 if (a->addr32[0] > b->addr32[0]) 344 return (1); 345 if (a->addr32[0] < b->addr32[0]) 346 return (-1); 347 break; 348 #endif /* INET6 */ 349 } 350 return (0); 351 } 352 353 static __inline int 354 pf_src_compare(struct pf_src_node *a, struct pf_src_node *b) 355 { 356 int diff; 357 358 if (a->rule.ptr > b->rule.ptr) 359 return (1); 360 if (a->rule.ptr < b->rule.ptr) 361 return (-1); 362 if ((diff = a->type - b->type) != 0) 363 return (diff); 364 if ((diff = a->af - b->af) != 0) 365 return (diff); 366 if ((diff = pf_addr_compare(&a->addr, &b->addr, a->af)) != 0) 367 return (diff); 368 return (0); 369 } 370 371 static __inline void 372 pf_set_protostate(struct pf_state *st, int which, u_int8_t newstate) 373 { 374 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 375 st->dst.state = newstate; 376 if (which == PF_PEER_DST) 377 return; 378 379 if (st->src.state == newstate) 380 return; 381 if (st->creatorid == pf_status.hostid && 382 st->key[PF_SK_STACK]->proto == IPPROTO_TCP && 383 !(TCPS_HAVEESTABLISHED(st->src.state) || 384 st->src.state == TCPS_CLOSED) && 385 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 386 pf_status.states_halfopen--; 387 388 st->src.state = newstate; 389 } 390 391 void 392 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 393 { 394 switch (af) { 395 case AF_INET: 396 dst->addr32[0] = src->addr32[0]; 397 break; 398 #ifdef INET6 399 case AF_INET6: 400 dst->addr32[0] = src->addr32[0]; 401 dst->addr32[1] = src->addr32[1]; 402 dst->addr32[2] = src->addr32[2]; 403 dst->addr32[3] = src->addr32[3]; 404 break; 405 #endif /* INET6 */ 406 default: 407 unhandled_af(af); 408 } 409 } 410 411 void 412 pf_init_threshold(struct pf_threshold *threshold, 413 u_int32_t limit, u_int32_t seconds) 414 { 415 threshold->limit = limit * PF_THRESHOLD_MULT; 416 threshold->seconds = seconds; 417 threshold->count = 0; 418 threshold->last = getuptime(); 419 } 420 421 void 422 pf_add_threshold(struct pf_threshold *threshold) 423 { 424 u_int32_t t = getuptime(), diff = t - threshold->last; 425 426 if (diff >= threshold->seconds) 427 threshold->count = 0; 428 else 429 threshold->count -= threshold->count * diff / 430 threshold->seconds; 431 threshold->count += PF_THRESHOLD_MULT; 432 threshold->last = t; 433 } 434 435 int 436 pf_check_threshold(struct pf_threshold *threshold) 437 { 438 return (threshold->count > threshold->limit); 439 } 440 441 void 442 pf_state_list_insert(struct pf_state_list *pfs, struct pf_state *st) 443 { 444 /* 445 * we can always put states on the end of the list. 446 * 447 * things reading the list should take a read lock, then 448 * the mutex, get the head and tail pointers, release the 449 * mutex, and then they can iterate between the head and tail. 450 */ 451 452 pf_state_ref(st); /* get a ref for the list */ 453 454 mtx_enter(&pfs->pfs_mtx); 455 TAILQ_INSERT_TAIL(&pfs->pfs_list, st, entry_list); 456 mtx_leave(&pfs->pfs_mtx); 457 } 458 459 void 460 pf_state_list_remove(struct pf_state_list *pfs, struct pf_state *st) 461 { 462 /* states can only be removed when the write lock is held */ 463 rw_assert_wrlock(&pfs->pfs_rwl); 464 465 mtx_enter(&pfs->pfs_mtx); 466 TAILQ_REMOVE(&pfs->pfs_list, st, entry_list); 467 mtx_leave(&pfs->pfs_mtx); 468 469 pf_state_unref(st); /* list no longer references the state */ 470 } 471 472 int 473 pf_src_connlimit(struct pf_state **stp) 474 { 475 int bad = 0; 476 struct pf_src_node *sn; 477 478 if ((sn = pf_get_src_node((*stp), PF_SN_NONE)) == NULL) 479 return (0); 480 481 sn->conn++; 482 (*stp)->src.tcp_est = 1; 483 pf_add_threshold(&sn->conn_rate); 484 485 if ((*stp)->rule.ptr->max_src_conn && 486 (*stp)->rule.ptr->max_src_conn < sn->conn) { 487 pf_status.lcounters[LCNT_SRCCONN]++; 488 bad++; 489 } 490 491 if ((*stp)->rule.ptr->max_src_conn_rate.limit && 492 pf_check_threshold(&sn->conn_rate)) { 493 pf_status.lcounters[LCNT_SRCCONNRATE]++; 494 bad++; 495 } 496 497 if (!bad) 498 return (0); 499 500 if ((*stp)->rule.ptr->overload_tbl) { 501 struct pfr_addr p; 502 u_int32_t killed = 0; 503 504 pf_status.lcounters[LCNT_OVERLOAD_TABLE]++; 505 if (pf_status.debug >= LOG_NOTICE) { 506 log(LOG_NOTICE, 507 "pf: pf_src_connlimit: blocking address "); 508 pf_print_host(&sn->addr, 0, 509 (*stp)->key[PF_SK_WIRE]->af); 510 } 511 512 memset(&p, 0, sizeof(p)); 513 p.pfra_af = (*stp)->key[PF_SK_WIRE]->af; 514 switch ((*stp)->key[PF_SK_WIRE]->af) { 515 case AF_INET: 516 p.pfra_net = 32; 517 p.pfra_ip4addr = sn->addr.v4; 518 break; 519 #ifdef INET6 520 case AF_INET6: 521 p.pfra_net = 128; 522 p.pfra_ip6addr = sn->addr.v6; 523 break; 524 #endif /* INET6 */ 525 } 526 527 pfr_insert_kentry((*stp)->rule.ptr->overload_tbl, 528 &p, gettime()); 529 530 /* kill existing states if that's required. */ 531 if ((*stp)->rule.ptr->flush) { 532 struct pf_state_key *sk; 533 struct pf_state *st; 534 535 pf_status.lcounters[LCNT_OVERLOAD_FLUSH]++; 536 RBT_FOREACH(st, pf_state_tree_id, &tree_id) { 537 sk = st->key[PF_SK_WIRE]; 538 /* 539 * Kill states from this source. (Only those 540 * from the same rule if PF_FLUSH_GLOBAL is not 541 * set) 542 */ 543 if (sk->af == 544 (*stp)->key[PF_SK_WIRE]->af && 545 (((*stp)->direction == PF_OUT && 546 PF_AEQ(&sn->addr, &sk->addr[1], sk->af)) || 547 ((*stp)->direction == PF_IN && 548 PF_AEQ(&sn->addr, &sk->addr[0], sk->af))) && 549 ((*stp)->rule.ptr->flush & 550 PF_FLUSH_GLOBAL || 551 (*stp)->rule.ptr == st->rule.ptr)) { 552 st->timeout = PFTM_PURGE; 553 pf_set_protostate(st, PF_PEER_BOTH, 554 TCPS_CLOSED); 555 killed++; 556 } 557 } 558 if (pf_status.debug >= LOG_NOTICE) 559 addlog(", %u states killed", killed); 560 } 561 if (pf_status.debug >= LOG_NOTICE) 562 addlog("\n"); 563 } 564 565 /* kill this state */ 566 (*stp)->timeout = PFTM_PURGE; 567 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_CLOSED); 568 return (1); 569 } 570 571 int 572 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule, 573 enum pf_sn_types type, sa_family_t af, struct pf_addr *src, 574 struct pf_addr *raddr, struct pfi_kif *kif) 575 { 576 struct pf_src_node k; 577 578 if (*sn == NULL) { 579 k.af = af; 580 k.type = type; 581 pf_addrcpy(&k.addr, src, af); 582 k.rule.ptr = rule; 583 pf_status.scounters[SCNT_SRC_NODE_SEARCH]++; 584 *sn = RB_FIND(pf_src_tree, &tree_src_tracking, &k); 585 } 586 if (*sn == NULL) { 587 if (!rule->max_src_nodes || 588 rule->src_nodes < rule->max_src_nodes) 589 (*sn) = pool_get(&pf_src_tree_pl, PR_NOWAIT | PR_ZERO); 590 else 591 pf_status.lcounters[LCNT_SRCNODES]++; 592 if ((*sn) == NULL) 593 return (-1); 594 595 pf_init_threshold(&(*sn)->conn_rate, 596 rule->max_src_conn_rate.limit, 597 rule->max_src_conn_rate.seconds); 598 599 (*sn)->type = type; 600 (*sn)->af = af; 601 (*sn)->rule.ptr = rule; 602 pf_addrcpy(&(*sn)->addr, src, af); 603 if (raddr) 604 pf_addrcpy(&(*sn)->raddr, raddr, af); 605 if (RB_INSERT(pf_src_tree, 606 &tree_src_tracking, *sn) != NULL) { 607 if (pf_status.debug >= LOG_NOTICE) { 608 log(LOG_NOTICE, 609 "pf: src_tree insert failed: "); 610 pf_print_host(&(*sn)->addr, 0, af); 611 addlog("\n"); 612 } 613 pool_put(&pf_src_tree_pl, *sn); 614 return (-1); 615 } 616 (*sn)->creation = getuptime(); 617 (*sn)->rule.ptr->src_nodes++; 618 if (kif != NULL) { 619 (*sn)->kif = kif; 620 pfi_kif_ref(kif, PFI_KIF_REF_SRCNODE); 621 } 622 pf_status.scounters[SCNT_SRC_NODE_INSERT]++; 623 pf_status.src_nodes++; 624 } else { 625 if (rule->max_src_states && 626 (*sn)->states >= rule->max_src_states) { 627 pf_status.lcounters[LCNT_SRCSTATES]++; 628 return (-1); 629 } 630 } 631 return (0); 632 } 633 634 void 635 pf_remove_src_node(struct pf_src_node *sn) 636 { 637 if (sn->states > 0 || sn->expire > getuptime()) 638 return; 639 640 sn->rule.ptr->src_nodes--; 641 if (sn->rule.ptr->states_cur == 0 && 642 sn->rule.ptr->src_nodes == 0) 643 pf_rm_rule(NULL, sn->rule.ptr); 644 RB_REMOVE(pf_src_tree, &tree_src_tracking, sn); 645 pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++; 646 pf_status.src_nodes--; 647 pfi_kif_unref(sn->kif, PFI_KIF_REF_SRCNODE); 648 pool_put(&pf_src_tree_pl, sn); 649 } 650 651 struct pf_src_node * 652 pf_get_src_node(struct pf_state *st, enum pf_sn_types type) 653 { 654 struct pf_sn_item *sni; 655 656 SLIST_FOREACH(sni, &st->src_nodes, next) 657 if (sni->sn->type == type) 658 return (sni->sn); 659 return (NULL); 660 } 661 662 void 663 pf_state_rm_src_node(struct pf_state *st, struct pf_src_node *sn) 664 { 665 struct pf_sn_item *sni, *snin, *snip = NULL; 666 667 for (sni = SLIST_FIRST(&st->src_nodes); sni; sni = snin) { 668 snin = SLIST_NEXT(sni, next); 669 if (sni->sn == sn) { 670 if (snip) 671 SLIST_REMOVE_AFTER(snip, next); 672 else 673 SLIST_REMOVE_HEAD(&st->src_nodes, next); 674 pool_put(&pf_sn_item_pl, sni); 675 sni = NULL; 676 sn->states--; 677 } 678 if (sni != NULL) 679 snip = sni; 680 } 681 } 682 683 /* state table stuff */ 684 685 static inline int 686 pf_state_compare_key(const struct pf_state_key *a, 687 const struct pf_state_key *b) 688 { 689 int diff; 690 691 if ((diff = a->hash - b->hash) != 0) 692 return (diff); 693 if ((diff = a->proto - b->proto) != 0) 694 return (diff); 695 if ((diff = a->af - b->af) != 0) 696 return (diff); 697 if ((diff = pf_addr_compare(&a->addr[0], &b->addr[0], a->af)) != 0) 698 return (diff); 699 if ((diff = pf_addr_compare(&a->addr[1], &b->addr[1], a->af)) != 0) 700 return (diff); 701 if ((diff = a->port[0] - b->port[0]) != 0) 702 return (diff); 703 if ((diff = a->port[1] - b->port[1]) != 0) 704 return (diff); 705 if ((diff = a->rdomain - b->rdomain) != 0) 706 return (diff); 707 return (0); 708 } 709 710 static inline int 711 pf_state_compare_id(const struct pf_state *a, const struct pf_state *b) 712 { 713 if (a->id > b->id) 714 return (1); 715 if (a->id < b->id) 716 return (-1); 717 if (a->creatorid > b->creatorid) 718 return (1); 719 if (a->creatorid < b->creatorid) 720 return (-1); 721 722 return (0); 723 } 724 725 /* 726 * on failure, pf_state_key_attach() releases the pf_state_key 727 * reference and returns NULL. 728 */ 729 struct pf_state_key * 730 pf_state_key_attach(struct pf_state_key *sk, struct pf_state *st, int idx) 731 { 732 struct pf_state_item *si; 733 struct pf_state_key *cur; 734 struct pf_state *oldst = NULL; 735 736 PF_ASSERT_LOCKED(); 737 738 KASSERT(st->key[idx] == NULL); 739 sk->sk_removed = 0; 740 cur = RBT_INSERT(pf_state_tree, &pf_statetbl, sk); 741 if (cur != NULL) { 742 sk->sk_removed = 1; 743 /* key exists. check for same kif, if none, add to key */ 744 TAILQ_FOREACH(si, &cur->sk_states, si_entry) { 745 struct pf_state *sist = si->si_st; 746 if (sist->kif == st->kif && 747 ((sist->key[PF_SK_WIRE]->af == sk->af && 748 sist->direction == st->direction) || 749 (sist->key[PF_SK_WIRE]->af != 750 sist->key[PF_SK_STACK]->af && 751 sk->af == sist->key[PF_SK_STACK]->af && 752 sist->direction != st->direction))) { 753 int reuse = 0; 754 755 if (sk->proto == IPPROTO_TCP && 756 sist->src.state >= TCPS_FIN_WAIT_2 && 757 sist->dst.state >= TCPS_FIN_WAIT_2) 758 reuse = 1; 759 if (pf_status.debug >= LOG_NOTICE) { 760 log(LOG_NOTICE, 761 "pf: %s key attach %s on %s: ", 762 (idx == PF_SK_WIRE) ? 763 "wire" : "stack", 764 reuse ? "reuse" : "failed", 765 st->kif->pfik_name); 766 pf_print_state_parts(st, 767 (idx == PF_SK_WIRE) ? sk : NULL, 768 (idx == PF_SK_STACK) ? sk : NULL); 769 addlog(", existing: "); 770 pf_print_state_parts(sist, 771 (idx == PF_SK_WIRE) ? sk : NULL, 772 (idx == PF_SK_STACK) ? sk : NULL); 773 addlog("\n"); 774 } 775 if (reuse) { 776 pf_set_protostate(sist, PF_PEER_BOTH, 777 TCPS_CLOSED); 778 /* remove late or sks can go away */ 779 oldst = sist; 780 } else { 781 pf_state_key_unref(sk); 782 return (NULL); /* collision! */ 783 } 784 } 785 } 786 787 /* reuse the existing state key */ 788 pf_state_key_unref(sk); 789 sk = cur; 790 } 791 792 if ((si = pool_get(&pf_state_item_pl, PR_NOWAIT)) == NULL) { 793 if (TAILQ_EMPTY(&sk->sk_states)) { 794 KASSERT(cur == NULL); 795 RBT_REMOVE(pf_state_tree, &pf_statetbl, sk); 796 sk->sk_removed = 1; 797 pf_state_key_unref(sk); 798 } 799 800 return (NULL); 801 } 802 803 st->key[idx] = pf_state_key_ref(sk); /* give a ref to state */ 804 si->si_st = pf_state_ref(st); 805 806 /* list is sorted, if-bound states before floating */ 807 if (st->kif == pfi_all) 808 TAILQ_INSERT_TAIL(&sk->sk_states, si, si_entry); 809 else 810 TAILQ_INSERT_HEAD(&sk->sk_states, si, si_entry); 811 812 if (oldst) 813 pf_remove_state(oldst); 814 815 /* caller owns the pf_state ref, which owns a pf_state_key ref now */ 816 return (sk); 817 } 818 819 void 820 pf_detach_state(struct pf_state *st) 821 { 822 KASSERT(st->key[PF_SK_WIRE] != NULL); 823 pf_state_key_detach(st, PF_SK_WIRE); 824 825 KASSERT(st->key[PF_SK_STACK] != NULL); 826 if (st->key[PF_SK_STACK] != st->key[PF_SK_WIRE]) 827 pf_state_key_detach(st, PF_SK_STACK); 828 } 829 830 void 831 pf_state_key_detach(struct pf_state *st, int idx) 832 { 833 struct pf_state_item *si; 834 struct pf_state_key *sk; 835 836 PF_ASSERT_LOCKED(); 837 838 sk = st->key[idx]; 839 if (sk == NULL) 840 return; 841 842 TAILQ_FOREACH(si, &sk->sk_states, si_entry) { 843 if (si->si_st == st) 844 break; 845 } 846 if (si == NULL) 847 return; 848 849 TAILQ_REMOVE(&sk->sk_states, si, si_entry); 850 pool_put(&pf_state_item_pl, si); 851 852 if (TAILQ_EMPTY(&sk->sk_states)) { 853 RBT_REMOVE(pf_state_tree, &pf_statetbl, sk); 854 sk->sk_removed = 1; 855 pf_state_key_unlink_reverse(sk); 856 pf_state_key_unlink_inpcb(sk); 857 pf_state_key_unref(sk); 858 } 859 860 pf_state_unref(st); 861 } 862 863 struct pf_state_key * 864 pf_alloc_state_key(int pool_flags) 865 { 866 struct pf_state_key *sk; 867 868 if ((sk = pool_get(&pf_state_key_pl, pool_flags)) == NULL) 869 return (NULL); 870 871 PF_REF_INIT(sk->sk_refcnt); 872 TAILQ_INIT(&sk->sk_states); 873 sk->sk_removed = 1; 874 875 return (sk); 876 } 877 878 static __inline int 879 pf_state_key_addr_setup(struct pf_pdesc *pd, void *arg, int sidx, 880 struct pf_addr *saddr, int didx, struct pf_addr *daddr, int af, int multi) 881 { 882 struct pf_state_key_cmp *key = arg; 883 #ifdef INET6 884 struct pf_addr *target; 885 886 if (af == AF_INET || pd->proto != IPPROTO_ICMPV6) 887 goto copy; 888 889 switch (pd->hdr.icmp6.icmp6_type) { 890 case ND_NEIGHBOR_SOLICIT: 891 if (multi) 892 return (-1); 893 target = (struct pf_addr *)&pd->hdr.nd_ns.nd_ns_target; 894 daddr = target; 895 break; 896 case ND_NEIGHBOR_ADVERT: 897 if (multi) 898 return (-1); 899 target = (struct pf_addr *)&pd->hdr.nd_ns.nd_ns_target; 900 saddr = target; 901 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 902 key->addr[didx].addr32[0] = 0; 903 key->addr[didx].addr32[1] = 0; 904 key->addr[didx].addr32[2] = 0; 905 key->addr[didx].addr32[3] = 0; 906 daddr = NULL; /* overwritten */ 907 } 908 break; 909 default: 910 if (multi) { 911 key->addr[sidx].addr32[0] = __IPV6_ADDR_INT32_MLL; 912 key->addr[sidx].addr32[1] = 0; 913 key->addr[sidx].addr32[2] = 0; 914 key->addr[sidx].addr32[3] = __IPV6_ADDR_INT32_ONE; 915 saddr = NULL; /* overwritten */ 916 } 917 } 918 copy: 919 #endif /* INET6 */ 920 if (saddr) 921 pf_addrcpy(&key->addr[sidx], saddr, af); 922 if (daddr) 923 pf_addrcpy(&key->addr[didx], daddr, af); 924 925 return (0); 926 } 927 928 int 929 pf_state_key_setup(struct pf_pdesc *pd, struct pf_state_key **skw, 930 struct pf_state_key **sks, int rtableid) 931 { 932 /* if returning error we MUST pool_put state keys ourselves */ 933 struct pf_state_key *sk1, *sk2; 934 u_int wrdom = pd->rdomain; 935 int afto = pd->af != pd->naf; 936 937 if ((sk1 = pf_alloc_state_key(PR_NOWAIT | PR_ZERO)) == NULL) 938 return (ENOMEM); 939 940 pf_state_key_addr_setup(pd, sk1, pd->sidx, pd->src, pd->didx, pd->dst, 941 pd->af, 0); 942 sk1->port[pd->sidx] = pd->osport; 943 sk1->port[pd->didx] = pd->odport; 944 sk1->proto = pd->proto; 945 sk1->af = pd->af; 946 sk1->rdomain = pd->rdomain; 947 sk1->hash = pf_pkt_hash(sk1->af, sk1->proto, 948 &sk1->addr[0], &sk1->addr[1], sk1->port[0], sk1->port[1]); 949 if (rtableid >= 0) 950 wrdom = rtable_l2(rtableid); 951 952 if (PF_ANEQ(&pd->nsaddr, pd->src, pd->af) || 953 PF_ANEQ(&pd->ndaddr, pd->dst, pd->af) || 954 pd->nsport != pd->osport || pd->ndport != pd->odport || 955 wrdom != pd->rdomain || afto) { /* NAT/NAT64 */ 956 if ((sk2 = pf_alloc_state_key(PR_NOWAIT | PR_ZERO)) == NULL) { 957 pf_state_key_unref(sk1); 958 return (ENOMEM); 959 } 960 pf_state_key_addr_setup(pd, sk2, afto ? pd->didx : pd->sidx, 961 &pd->nsaddr, afto ? pd->sidx : pd->didx, &pd->ndaddr, 962 pd->naf, 0); 963 sk2->port[afto ? pd->didx : pd->sidx] = pd->nsport; 964 sk2->port[afto ? pd->sidx : pd->didx] = pd->ndport; 965 if (afto) { 966 switch (pd->proto) { 967 case IPPROTO_ICMP: 968 sk2->proto = IPPROTO_ICMPV6; 969 break; 970 case IPPROTO_ICMPV6: 971 sk2->proto = IPPROTO_ICMP; 972 break; 973 default: 974 sk2->proto = pd->proto; 975 } 976 } else 977 sk2->proto = pd->proto; 978 sk2->af = pd->naf; 979 sk2->rdomain = wrdom; 980 sk2->hash = pf_pkt_hash(sk2->af, sk2->proto, 981 &sk2->addr[0], &sk2->addr[1], sk2->port[0], sk2->port[1]); 982 } else 983 sk2 = pf_state_key_ref(sk1); 984 985 if (pd->dir == PF_IN) { 986 *skw = sk1; 987 *sks = sk2; 988 } else { 989 *sks = sk1; 990 *skw = sk2; 991 } 992 993 if (pf_status.debug >= LOG_DEBUG) { 994 log(LOG_DEBUG, "pf: key setup: "); 995 pf_print_state_parts(NULL, *skw, *sks); 996 addlog("\n"); 997 } 998 999 return (0); 1000 } 1001 1002 /* 1003 * pf_state_insert() does the following: 1004 * - links the pf_state up with pf_state_key(s). 1005 * - inserts the pf_state_keys into pf_state_tree. 1006 * - inserts the pf_state into the into pf_state_tree_id. 1007 * - tells pfsync about the state. 1008 * 1009 * pf_state_insert() owns the references to the pf_state_key structs 1010 * it is given. on failure to insert, these references are released. 1011 * on success, the caller owns a pf_state reference that allows it 1012 * to access the state keys. 1013 */ 1014 1015 int 1016 pf_state_insert(struct pfi_kif *kif, struct pf_state_key **skwp, 1017 struct pf_state_key **sksp, struct pf_state *st) 1018 { 1019 struct pf_state_key *skw = *skwp; 1020 struct pf_state_key *sks = *sksp; 1021 int same = (skw == sks); 1022 1023 PF_ASSERT_LOCKED(); 1024 1025 st->kif = kif; 1026 PF_STATE_ENTER_WRITE(); 1027 1028 skw = pf_state_key_attach(skw, st, PF_SK_WIRE); 1029 if (skw == NULL) { 1030 pf_state_key_unref(sks); 1031 PF_STATE_EXIT_WRITE(); 1032 return (-1); 1033 } 1034 1035 if (same) { 1036 /* pf_state_key_attach might have swapped skw */ 1037 pf_state_key_unref(sks); 1038 st->key[PF_SK_STACK] = sks = pf_state_key_ref(skw); 1039 } else if (pf_state_key_attach(sks, st, PF_SK_STACK) == NULL) { 1040 pf_state_key_detach(st, PF_SK_WIRE); 1041 PF_STATE_EXIT_WRITE(); 1042 return (-1); 1043 } 1044 1045 if (st->id == 0 && st->creatorid == 0) { 1046 st->id = htobe64(pf_status.stateid++); 1047 st->creatorid = pf_status.hostid; 1048 } 1049 if (RBT_INSERT(pf_state_tree_id, &tree_id, st) != NULL) { 1050 if (pf_status.debug >= LOG_NOTICE) { 1051 log(LOG_NOTICE, "pf: state insert failed: " 1052 "id: %016llx creatorid: %08x", 1053 betoh64(st->id), ntohl(st->creatorid)); 1054 addlog("\n"); 1055 } 1056 pf_detach_state(st); 1057 PF_STATE_EXIT_WRITE(); 1058 return (-1); 1059 } 1060 pf_state_list_insert(&pf_state_list, st); 1061 pf_status.fcounters[FCNT_STATE_INSERT]++; 1062 pf_status.states++; 1063 pfi_kif_ref(kif, PFI_KIF_REF_STATE); 1064 PF_STATE_EXIT_WRITE(); 1065 1066 #if NPFSYNC > 0 1067 pfsync_insert_state(st); 1068 #endif /* NPFSYNC > 0 */ 1069 1070 *skwp = skw; 1071 *sksp = sks; 1072 1073 return (0); 1074 } 1075 1076 struct pf_state * 1077 pf_find_state_byid(struct pf_state_cmp *key) 1078 { 1079 pf_status.fcounters[FCNT_STATE_SEARCH]++; 1080 1081 return (RBT_FIND(pf_state_tree_id, &tree_id, (struct pf_state *)key)); 1082 } 1083 1084 int 1085 pf_compare_state_keys(struct pf_state_key *a, struct pf_state_key *b, 1086 struct pfi_kif *kif, u_int dir) 1087 { 1088 /* a (from hdr) and b (new) must be exact opposites of each other */ 1089 if (a->af == b->af && a->proto == b->proto && 1090 PF_AEQ(&a->addr[0], &b->addr[1], a->af) && 1091 PF_AEQ(&a->addr[1], &b->addr[0], a->af) && 1092 a->port[0] == b->port[1] && 1093 a->port[1] == b->port[0] && a->rdomain == b->rdomain) 1094 return (0); 1095 else { 1096 /* mismatch. must not happen. */ 1097 if (pf_status.debug >= LOG_ERR) { 1098 log(LOG_ERR, 1099 "pf: state key linking mismatch! dir=%s, " 1100 "if=%s, stored af=%u, a0: ", 1101 dir == PF_OUT ? "OUT" : "IN", 1102 kif->pfik_name, a->af); 1103 pf_print_host(&a->addr[0], a->port[0], a->af); 1104 addlog(", a1: "); 1105 pf_print_host(&a->addr[1], a->port[1], a->af); 1106 addlog(", proto=%u", a->proto); 1107 addlog(", found af=%u, a0: ", b->af); 1108 pf_print_host(&b->addr[0], b->port[0], b->af); 1109 addlog(", a1: "); 1110 pf_print_host(&b->addr[1], b->port[1], b->af); 1111 addlog(", proto=%u", b->proto); 1112 addlog("\n"); 1113 } 1114 return (-1); 1115 } 1116 } 1117 1118 int 1119 pf_find_state(struct pf_pdesc *pd, struct pf_state_key_cmp *key, 1120 struct pf_state **stp) 1121 { 1122 struct pf_state_key *sk, *pkt_sk, *inp_sk; 1123 struct pf_state_item *si; 1124 struct pf_state *st = NULL; 1125 1126 pf_status.fcounters[FCNT_STATE_SEARCH]++; 1127 if (pf_status.debug >= LOG_DEBUG) { 1128 log(LOG_DEBUG, "pf: key search, %s on %s: ", 1129 pd->dir == PF_OUT ? "out" : "in", pd->kif->pfik_name); 1130 pf_print_state_parts(NULL, (struct pf_state_key *)key, NULL); 1131 addlog("\n"); 1132 } 1133 1134 inp_sk = NULL; 1135 pkt_sk = NULL; 1136 sk = NULL; 1137 if (pd->dir == PF_OUT) { 1138 /* first if block deals with outbound forwarded packet */ 1139 pkt_sk = pd->m->m_pkthdr.pf.statekey; 1140 1141 if (!pf_state_key_isvalid(pkt_sk)) { 1142 pf_mbuf_unlink_state_key(pd->m); 1143 pkt_sk = NULL; 1144 } 1145 1146 if (pkt_sk && pf_state_key_isvalid(pkt_sk->sk_reverse)) 1147 sk = pkt_sk->sk_reverse; 1148 1149 if (pkt_sk == NULL) { 1150 /* here we deal with local outbound packet */ 1151 if (pd->m->m_pkthdr.pf.inp != NULL) { 1152 inp_sk = pd->m->m_pkthdr.pf.inp->inp_pf_sk; 1153 if (pf_state_key_isvalid(inp_sk)) 1154 sk = inp_sk; 1155 else 1156 pf_inpcb_unlink_state_key( 1157 pd->m->m_pkthdr.pf.inp); 1158 } 1159 } 1160 } 1161 1162 if (sk == NULL) { 1163 if ((sk = RBT_FIND(pf_state_tree, &pf_statetbl, 1164 (struct pf_state_key *)key)) == NULL) 1165 return (PF_DROP); 1166 if (pd->dir == PF_OUT && pkt_sk && 1167 pf_compare_state_keys(pkt_sk, sk, pd->kif, pd->dir) == 0) 1168 pf_state_key_link_reverse(sk, pkt_sk); 1169 else if (pd->dir == PF_OUT && pd->m->m_pkthdr.pf.inp && 1170 !pd->m->m_pkthdr.pf.inp->inp_pf_sk && !sk->sk_inp) 1171 pf_state_key_link_inpcb(sk, pd->m->m_pkthdr.pf.inp); 1172 } 1173 1174 /* remove firewall data from outbound packet */ 1175 if (pd->dir == PF_OUT) 1176 pf_pkt_addr_changed(pd->m); 1177 1178 /* list is sorted, if-bound states before floating ones */ 1179 TAILQ_FOREACH(si, &sk->sk_states, si_entry) { 1180 struct pf_state *sist = si->si_st; 1181 if (sist->timeout != PFTM_PURGE && 1182 (sist->kif == pfi_all || sist->kif == pd->kif) && 1183 ((sist->key[PF_SK_WIRE]->af == sist->key[PF_SK_STACK]->af && 1184 sk == (pd->dir == PF_IN ? sist->key[PF_SK_WIRE] : 1185 sist->key[PF_SK_STACK])) || 1186 (sist->key[PF_SK_WIRE]->af != sist->key[PF_SK_STACK]->af 1187 && pd->dir == PF_IN && (sk == sist->key[PF_SK_STACK] || 1188 sk == sist->key[PF_SK_WIRE])))) { 1189 st = sist; 1190 break; 1191 } 1192 } 1193 1194 if (st == NULL) 1195 return (PF_DROP); 1196 if (ISSET(st->state_flags, PFSTATE_INP_UNLINKED)) 1197 return (PF_DROP); 1198 1199 if (st->rule.ptr->pktrate.limit && pd->dir == st->direction) { 1200 pf_add_threshold(&st->rule.ptr->pktrate); 1201 if (pf_check_threshold(&st->rule.ptr->pktrate)) 1202 return (PF_DROP); 1203 } 1204 1205 *stp = st; 1206 1207 return (PF_MATCH); 1208 } 1209 1210 struct pf_state * 1211 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1212 { 1213 struct pf_state_key *sk; 1214 struct pf_state_item *si, *ret = NULL; 1215 1216 pf_status.fcounters[FCNT_STATE_SEARCH]++; 1217 1218 sk = RBT_FIND(pf_state_tree, &pf_statetbl, (struct pf_state_key *)key); 1219 1220 if (sk != NULL) { 1221 TAILQ_FOREACH(si, &sk->sk_states, si_entry) { 1222 struct pf_state *sist = si->si_st; 1223 if (dir == PF_INOUT || 1224 (sk == (dir == PF_IN ? sist->key[PF_SK_WIRE] : 1225 sist->key[PF_SK_STACK]))) { 1226 if (more == NULL) 1227 return (sist); 1228 1229 if (ret) 1230 (*more)++; 1231 else 1232 ret = si; 1233 } 1234 } 1235 } 1236 return (ret ? ret->si_st : NULL); 1237 } 1238 1239 void 1240 pf_state_peer_hton(const struct pf_state_peer *s, struct pfsync_state_peer *d) 1241 { 1242 d->seqlo = htonl(s->seqlo); 1243 d->seqhi = htonl(s->seqhi); 1244 d->seqdiff = htonl(s->seqdiff); 1245 d->max_win = htons(s->max_win); 1246 d->mss = htons(s->mss); 1247 d->state = s->state; 1248 d->wscale = s->wscale; 1249 if (s->scrub) { 1250 d->scrub.pfss_flags = 1251 htons(s->scrub->pfss_flags & PFSS_TIMESTAMP); 1252 d->scrub.pfss_ttl = (s)->scrub->pfss_ttl; 1253 d->scrub.pfss_ts_mod = htonl((s)->scrub->pfss_ts_mod); 1254 d->scrub.scrub_flag = PFSYNC_SCRUB_FLAG_VALID; 1255 } 1256 } 1257 1258 void 1259 pf_state_peer_ntoh(const struct pfsync_state_peer *s, struct pf_state_peer *d) 1260 { 1261 d->seqlo = ntohl(s->seqlo); 1262 d->seqhi = ntohl(s->seqhi); 1263 d->seqdiff = ntohl(s->seqdiff); 1264 d->max_win = ntohs(s->max_win); 1265 d->mss = ntohs(s->mss); 1266 d->state = s->state; 1267 d->wscale = s->wscale; 1268 if (s->scrub.scrub_flag == PFSYNC_SCRUB_FLAG_VALID && 1269 d->scrub != NULL) { 1270 d->scrub->pfss_flags = 1271 ntohs(s->scrub.pfss_flags) & PFSS_TIMESTAMP; 1272 d->scrub->pfss_ttl = s->scrub.pfss_ttl; 1273 d->scrub->pfss_ts_mod = ntohl(s->scrub.pfss_ts_mod); 1274 } 1275 } 1276 1277 void 1278 pf_state_export(struct pfsync_state *sp, struct pf_state *st) 1279 { 1280 int32_t expire; 1281 1282 memset(sp, 0, sizeof(struct pfsync_state)); 1283 1284 /* copy from state key */ 1285 sp->key[PF_SK_WIRE].addr[0] = st->key[PF_SK_WIRE]->addr[0]; 1286 sp->key[PF_SK_WIRE].addr[1] = st->key[PF_SK_WIRE]->addr[1]; 1287 sp->key[PF_SK_WIRE].port[0] = st->key[PF_SK_WIRE]->port[0]; 1288 sp->key[PF_SK_WIRE].port[1] = st->key[PF_SK_WIRE]->port[1]; 1289 sp->key[PF_SK_WIRE].rdomain = htons(st->key[PF_SK_WIRE]->rdomain); 1290 sp->key[PF_SK_WIRE].af = st->key[PF_SK_WIRE]->af; 1291 sp->key[PF_SK_STACK].addr[0] = st->key[PF_SK_STACK]->addr[0]; 1292 sp->key[PF_SK_STACK].addr[1] = st->key[PF_SK_STACK]->addr[1]; 1293 sp->key[PF_SK_STACK].port[0] = st->key[PF_SK_STACK]->port[0]; 1294 sp->key[PF_SK_STACK].port[1] = st->key[PF_SK_STACK]->port[1]; 1295 sp->key[PF_SK_STACK].rdomain = htons(st->key[PF_SK_STACK]->rdomain); 1296 sp->key[PF_SK_STACK].af = st->key[PF_SK_STACK]->af; 1297 sp->rtableid[PF_SK_WIRE] = htonl(st->rtableid[PF_SK_WIRE]); 1298 sp->rtableid[PF_SK_STACK] = htonl(st->rtableid[PF_SK_STACK]); 1299 sp->proto = st->key[PF_SK_WIRE]->proto; 1300 sp->af = st->key[PF_SK_WIRE]->af; 1301 1302 /* copy from state */ 1303 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); 1304 sp->rt = st->rt; 1305 sp->rt_addr = st->rt_addr; 1306 sp->creation = htonl(getuptime() - st->creation); 1307 expire = pf_state_expires(st, st->timeout); 1308 if (expire <= getuptime()) 1309 sp->expire = htonl(0); 1310 else 1311 sp->expire = htonl(expire - getuptime()); 1312 1313 sp->direction = st->direction; 1314 #if NPFLOG > 0 1315 sp->log = st->log; 1316 #endif /* NPFLOG > 0 */ 1317 sp->timeout = st->timeout; 1318 sp->state_flags = htons(st->state_flags); 1319 if (READ_ONCE(st->sync_defer) != NULL) 1320 sp->state_flags |= htons(PFSTATE_ACK); 1321 if (!SLIST_EMPTY(&st->src_nodes)) 1322 sp->sync_flags |= PFSYNC_FLAG_SRCNODE; 1323 1324 sp->id = st->id; 1325 sp->creatorid = st->creatorid; 1326 pf_state_peer_hton(&st->src, &sp->src); 1327 pf_state_peer_hton(&st->dst, &sp->dst); 1328 1329 if (st->rule.ptr == NULL) 1330 sp->rule = htonl(-1); 1331 else 1332 sp->rule = htonl(st->rule.ptr->nr); 1333 if (st->anchor.ptr == NULL) 1334 sp->anchor = htonl(-1); 1335 else 1336 sp->anchor = htonl(st->anchor.ptr->nr); 1337 sp->nat_rule = htonl(-1); /* left for compat, nat_rule is gone */ 1338 1339 pf_state_counter_hton(st->packets[0], sp->packets[0]); 1340 pf_state_counter_hton(st->packets[1], sp->packets[1]); 1341 pf_state_counter_hton(st->bytes[0], sp->bytes[0]); 1342 pf_state_counter_hton(st->bytes[1], sp->bytes[1]); 1343 1344 sp->max_mss = htons(st->max_mss); 1345 sp->min_ttl = st->min_ttl; 1346 sp->set_tos = st->set_tos; 1347 sp->set_prio[0] = st->set_prio[0]; 1348 sp->set_prio[1] = st->set_prio[1]; 1349 } 1350 1351 int 1352 pf_state_alloc_scrub_memory(const struct pfsync_state_peer *s, 1353 struct pf_state_peer *d) 1354 { 1355 if (s->scrub.scrub_flag && d->scrub == NULL) 1356 return (pf_normalize_tcp_alloc(d)); 1357 1358 return (0); 1359 } 1360 1361 #if NPFSYNC > 0 1362 int 1363 pf_state_import(const struct pfsync_state *sp, int flags) 1364 { 1365 struct pf_state *st = NULL; 1366 struct pf_state_key *skw = NULL, *sks = NULL; 1367 struct pf_rule *r = NULL; 1368 struct pfi_kif *kif; 1369 int pool_flags; 1370 int error = ENOMEM; 1371 int n = 0; 1372 1373 PF_ASSERT_LOCKED(); 1374 1375 if (sp->creatorid == 0) { 1376 DPFPRINTF(LOG_NOTICE, "%s: invalid creator id: %08x", __func__, 1377 ntohl(sp->creatorid)); 1378 return (EINVAL); 1379 } 1380 1381 if ((kif = pfi_kif_get(sp->ifname, NULL)) == NULL) { 1382 DPFPRINTF(LOG_NOTICE, "%s: unknown interface: %s", __func__, 1383 sp->ifname); 1384 if (flags & PFSYNC_SI_IOCTL) 1385 return (EINVAL); 1386 return (0); /* skip this state */ 1387 } 1388 1389 if (sp->af == 0) 1390 return (0); /* skip this state */ 1391 1392 /* 1393 * If the ruleset checksums match or the state is coming from the ioctl, 1394 * it's safe to associate the state with the rule of that number. 1395 */ 1396 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && 1397 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && 1398 ntohl(sp->rule) < pf_main_ruleset.rules.active.rcount) { 1399 TAILQ_FOREACH(r, pf_main_ruleset.rules.active.ptr, entries) 1400 if (ntohl(sp->rule) == n++) 1401 break; 1402 } else 1403 r = &pf_default_rule; 1404 1405 if ((r->max_states && r->states_cur >= r->max_states)) 1406 goto cleanup; 1407 1408 if (flags & PFSYNC_SI_IOCTL) 1409 pool_flags = PR_WAITOK | PR_LIMITFAIL | PR_ZERO; 1410 else 1411 pool_flags = PR_NOWAIT | PR_LIMITFAIL | PR_ZERO; 1412 1413 if ((st = pool_get(&pf_state_pl, pool_flags)) == NULL) 1414 goto cleanup; 1415 1416 if ((skw = pf_alloc_state_key(pool_flags)) == NULL) 1417 goto cleanup; 1418 1419 if ((sp->key[PF_SK_WIRE].af && 1420 (sp->key[PF_SK_WIRE].af != sp->key[PF_SK_STACK].af)) || 1421 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[0], 1422 &sp->key[PF_SK_STACK].addr[0], sp->af) || 1423 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[1], 1424 &sp->key[PF_SK_STACK].addr[1], sp->af) || 1425 sp->key[PF_SK_WIRE].port[0] != sp->key[PF_SK_STACK].port[0] || 1426 sp->key[PF_SK_WIRE].port[1] != sp->key[PF_SK_STACK].port[1] || 1427 sp->key[PF_SK_WIRE].rdomain != sp->key[PF_SK_STACK].rdomain) { 1428 if ((sks = pf_alloc_state_key(pool_flags)) == NULL) 1429 goto cleanup; 1430 } else 1431 sks = pf_state_key_ref(skw); 1432 1433 /* allocate memory for scrub info */ 1434 if (pf_state_alloc_scrub_memory(&sp->src, &st->src) || 1435 pf_state_alloc_scrub_memory(&sp->dst, &st->dst)) 1436 goto cleanup; 1437 1438 /* copy to state key(s) */ 1439 skw->addr[0] = sp->key[PF_SK_WIRE].addr[0]; 1440 skw->addr[1] = sp->key[PF_SK_WIRE].addr[1]; 1441 skw->port[0] = sp->key[PF_SK_WIRE].port[0]; 1442 skw->port[1] = sp->key[PF_SK_WIRE].port[1]; 1443 skw->rdomain = ntohs(sp->key[PF_SK_WIRE].rdomain); 1444 skw->proto = sp->proto; 1445 if (!(skw->af = sp->key[PF_SK_WIRE].af)) 1446 skw->af = sp->af; 1447 skw->hash = pf_pkt_hash(skw->af, skw->proto, 1448 &skw->addr[0], &skw->addr[1], skw->port[0], skw->port[1]); 1449 1450 if (sks != skw) { 1451 sks->addr[0] = sp->key[PF_SK_STACK].addr[0]; 1452 sks->addr[1] = sp->key[PF_SK_STACK].addr[1]; 1453 sks->port[0] = sp->key[PF_SK_STACK].port[0]; 1454 sks->port[1] = sp->key[PF_SK_STACK].port[1]; 1455 sks->rdomain = ntohs(sp->key[PF_SK_STACK].rdomain); 1456 if (!(sks->af = sp->key[PF_SK_STACK].af)) 1457 sks->af = sp->af; 1458 if (sks->af != skw->af) { 1459 switch (sp->proto) { 1460 case IPPROTO_ICMP: 1461 sks->proto = IPPROTO_ICMPV6; 1462 break; 1463 case IPPROTO_ICMPV6: 1464 sks->proto = IPPROTO_ICMP; 1465 break; 1466 default: 1467 sks->proto = sp->proto; 1468 } 1469 } else 1470 sks->proto = sp->proto; 1471 1472 if (((sks->af != AF_INET) && (sks->af != AF_INET6)) || 1473 ((skw->af != AF_INET) && (skw->af != AF_INET6))) { 1474 error = EINVAL; 1475 goto cleanup; 1476 } 1477 1478 sks->hash = pf_pkt_hash(sks->af, sks->proto, 1479 &sks->addr[0], &sks->addr[1], sks->port[0], sks->port[1]); 1480 1481 } else if ((sks->af != AF_INET) && (sks->af != AF_INET6)) { 1482 error = EINVAL; 1483 goto cleanup; 1484 } 1485 st->rtableid[PF_SK_WIRE] = ntohl(sp->rtableid[PF_SK_WIRE]); 1486 st->rtableid[PF_SK_STACK] = ntohl(sp->rtableid[PF_SK_STACK]); 1487 1488 /* copy to state */ 1489 st->rt_addr = sp->rt_addr; 1490 st->rt = sp->rt; 1491 st->creation = getuptime() - ntohl(sp->creation); 1492 st->expire = getuptime(); 1493 if (ntohl(sp->expire)) { 1494 u_int32_t timeout; 1495 1496 timeout = r->timeout[sp->timeout]; 1497 if (!timeout) 1498 timeout = pf_default_rule.timeout[sp->timeout]; 1499 1500 /* sp->expire may have been adaptively scaled by export. */ 1501 st->expire -= timeout - ntohl(sp->expire); 1502 } 1503 1504 st->direction = sp->direction; 1505 st->log = sp->log; 1506 st->timeout = sp->timeout; 1507 st->state_flags = ntohs(sp->state_flags); 1508 st->max_mss = ntohs(sp->max_mss); 1509 st->min_ttl = sp->min_ttl; 1510 st->set_tos = sp->set_tos; 1511 st->set_prio[0] = sp->set_prio[0]; 1512 st->set_prio[1] = sp->set_prio[1]; 1513 1514 st->id = sp->id; 1515 st->creatorid = sp->creatorid; 1516 pf_state_peer_ntoh(&sp->src, &st->src); 1517 pf_state_peer_ntoh(&sp->dst, &st->dst); 1518 1519 st->rule.ptr = r; 1520 st->anchor.ptr = NULL; 1521 1522 PF_REF_INIT(st->refcnt); 1523 mtx_init(&st->mtx, IPL_NET); 1524 1525 /* XXX when we have anchors, use STATE_INC_COUNTERS */ 1526 r->states_cur++; 1527 r->states_tot++; 1528 1529 st->sync_state = PFSYNC_S_NONE; 1530 st->pfsync_time = getuptime(); 1531 #if NPFSYNC > 0 1532 pfsync_init_state(st, skw, sks, flags); 1533 #endif 1534 1535 if (pf_state_insert(kif, &skw, &sks, st) != 0) { 1536 /* XXX when we have anchors, use STATE_DEC_COUNTERS */ 1537 r->states_cur--; 1538 error = EEXIST; 1539 goto cleanup_state; 1540 } 1541 1542 return (0); 1543 1544 cleanup: 1545 if (skw != NULL) 1546 pf_state_key_unref(skw); 1547 if (sks != NULL) 1548 pf_state_key_unref(sks); 1549 1550 cleanup_state: /* pf_state_insert frees the state keys */ 1551 if (st) { 1552 if (st->dst.scrub) 1553 pool_put(&pf_state_scrub_pl, st->dst.scrub); 1554 if (st->src.scrub) 1555 pool_put(&pf_state_scrub_pl, st->src.scrub); 1556 pool_put(&pf_state_pl, st); 1557 } 1558 return (error); 1559 } 1560 #endif /* NPFSYNC > 0 */ 1561 1562 /* END state table stuff */ 1563 1564 void pf_purge_states(void *); 1565 struct task pf_purge_states_task = 1566 TASK_INITIALIZER(pf_purge_states, NULL); 1567 1568 void pf_purge_states_tick(void *); 1569 struct timeout pf_purge_states_to = 1570 TIMEOUT_INITIALIZER(pf_purge_states_tick, NULL); 1571 1572 unsigned int pf_purge_expired_states(unsigned int, unsigned int); 1573 1574 /* 1575 * how many states to scan this interval. 1576 * 1577 * this is set when the timeout fires, and reduced by the task. the 1578 * task will reschedule itself until the limit is reduced to zero, 1579 * and then it adds the timeout again. 1580 */ 1581 unsigned int pf_purge_states_limit; 1582 1583 /* 1584 * limit how many states are processed with locks held per run of 1585 * the state purge task. 1586 */ 1587 unsigned int pf_purge_states_collect = 64; 1588 1589 void 1590 pf_purge_states_tick(void *null) 1591 { 1592 unsigned int limit = pf_status.states; 1593 unsigned int interval = pf_default_rule.timeout[PFTM_INTERVAL]; 1594 1595 if (limit == 0) { 1596 timeout_add_sec(&pf_purge_states_to, 1); 1597 return; 1598 } 1599 1600 /* 1601 * process a fraction of the state table every second 1602 */ 1603 1604 if (interval > 1) 1605 limit /= interval; 1606 1607 pf_purge_states_limit = limit; 1608 task_add(systqmp, &pf_purge_states_task); 1609 } 1610 1611 void 1612 pf_purge_states(void *null) 1613 { 1614 unsigned int limit; 1615 unsigned int scanned; 1616 1617 limit = pf_purge_states_limit; 1618 if (limit < pf_purge_states_collect) 1619 limit = pf_purge_states_collect; 1620 1621 scanned = pf_purge_expired_states(limit, pf_purge_states_collect); 1622 if (scanned >= pf_purge_states_limit) { 1623 /* we've run out of states to scan this "interval" */ 1624 timeout_add_sec(&pf_purge_states_to, 1); 1625 return; 1626 } 1627 1628 pf_purge_states_limit -= scanned; 1629 task_add(systqmp, &pf_purge_states_task); 1630 } 1631 1632 void pf_purge_tick(void *); 1633 struct timeout pf_purge_to = 1634 TIMEOUT_INITIALIZER(pf_purge_tick, NULL); 1635 1636 void pf_purge(void *); 1637 struct task pf_purge_task = 1638 TASK_INITIALIZER(pf_purge, NULL); 1639 1640 void 1641 pf_purge_tick(void *null) 1642 { 1643 task_add(systqmp, &pf_purge_task); 1644 } 1645 1646 void 1647 pf_purge(void *null) 1648 { 1649 unsigned int interval = max(1, pf_default_rule.timeout[PFTM_INTERVAL]); 1650 1651 PF_LOCK(); 1652 1653 pf_purge_expired_src_nodes(); 1654 1655 PF_UNLOCK(); 1656 1657 /* 1658 * Fragments don't require PF_LOCK(), they use their own lock. 1659 */ 1660 pf_purge_expired_fragments(); 1661 1662 /* interpret the interval as idle time between runs */ 1663 timeout_add_sec(&pf_purge_to, interval); 1664 } 1665 1666 int32_t 1667 pf_state_expires(const struct pf_state *st, uint8_t stimeout) 1668 { 1669 u_int32_t timeout; 1670 u_int32_t start; 1671 u_int32_t end; 1672 u_int32_t states; 1673 1674 /* 1675 * pf_state_expires is used by the state purge task to 1676 * decide if a state is a candidate for cleanup, and by the 1677 * pfsync state export code to populate an expiry time. 1678 * 1679 * this function may be called by the state purge task while 1680 * the state is being modified. avoid inconsistent reads of 1681 * state->timeout by having the caller do the read (and any 1682 * checks it needs to do on the same variable) and then pass 1683 * their view of the timeout in here for this function to use. 1684 * the only consequence of using a stale timeout value is 1685 * that the state won't be a candidate for purging until the 1686 * next pass of the purge task. 1687 */ 1688 1689 /* handle all PFTM_* >= PFTM_MAX here */ 1690 if (stimeout >= PFTM_MAX) 1691 return (0); 1692 1693 KASSERT(stimeout < PFTM_MAX); 1694 1695 timeout = st->rule.ptr->timeout[stimeout]; 1696 if (!timeout) 1697 timeout = pf_default_rule.timeout[stimeout]; 1698 1699 start = st->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1700 if (start) { 1701 end = st->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1702 states = st->rule.ptr->states_cur; 1703 } else { 1704 start = pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1705 end = pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1706 states = pf_status.states; 1707 } 1708 if (end && states > start && start < end) { 1709 if (states >= end) 1710 return (0); 1711 1712 timeout = (u_int64_t)timeout * (end - states) / (end - start); 1713 } 1714 1715 return (st->expire + timeout); 1716 } 1717 1718 void 1719 pf_purge_expired_src_nodes(void) 1720 { 1721 struct pf_src_node *cur, *next; 1722 1723 PF_ASSERT_LOCKED(); 1724 1725 for (cur = RB_MIN(pf_src_tree, &tree_src_tracking); cur; cur = next) { 1726 next = RB_NEXT(pf_src_tree, &tree_src_tracking, cur); 1727 1728 if (cur->states == 0 && cur->expire <= getuptime()) { 1729 next = RB_NEXT(pf_src_tree, &tree_src_tracking, cur); 1730 pf_remove_src_node(cur); 1731 } 1732 } 1733 } 1734 1735 void 1736 pf_src_tree_remove_state(struct pf_state *st) 1737 { 1738 u_int32_t timeout; 1739 struct pf_sn_item *sni; 1740 1741 while ((sni = SLIST_FIRST(&st->src_nodes)) != NULL) { 1742 SLIST_REMOVE_HEAD(&st->src_nodes, next); 1743 if (st->src.tcp_est) 1744 --sni->sn->conn; 1745 if (--sni->sn->states == 0) { 1746 timeout = st->rule.ptr->timeout[PFTM_SRC_NODE]; 1747 if (!timeout) 1748 timeout = 1749 pf_default_rule.timeout[PFTM_SRC_NODE]; 1750 sni->sn->expire = getuptime() + timeout; 1751 } 1752 pool_put(&pf_sn_item_pl, sni); 1753 } 1754 } 1755 1756 void 1757 pf_remove_state(struct pf_state *st) 1758 { 1759 PF_ASSERT_LOCKED(); 1760 1761 if (st->timeout == PFTM_UNLINKED) 1762 return; 1763 1764 st->timeout = PFTM_UNLINKED; 1765 1766 /* handle load balancing related tasks */ 1767 pf_postprocess_addr(st); 1768 1769 if (st->src.state == PF_TCPS_PROXY_DST) { 1770 pf_send_tcp(st->rule.ptr, st->key[PF_SK_WIRE]->af, 1771 &st->key[PF_SK_WIRE]->addr[1], 1772 &st->key[PF_SK_WIRE]->addr[0], 1773 st->key[PF_SK_WIRE]->port[1], 1774 st->key[PF_SK_WIRE]->port[0], 1775 st->src.seqhi, st->src.seqlo + 1, 1776 TH_RST|TH_ACK, 0, 0, 0, 1, st->tag, 1777 st->key[PF_SK_WIRE]->rdomain); 1778 } 1779 if (st->key[PF_SK_STACK]->proto == IPPROTO_TCP) 1780 pf_set_protostate(st, PF_PEER_BOTH, TCPS_CLOSED); 1781 1782 RBT_REMOVE(pf_state_tree_id, &tree_id, st); 1783 #if NPFLOW > 0 1784 if (st->state_flags & PFSTATE_PFLOW) 1785 export_pflow(st); 1786 #endif /* NPFLOW > 0 */ 1787 #if NPFSYNC > 0 1788 pfsync_delete_state(st); 1789 #endif /* NPFSYNC > 0 */ 1790 pf_src_tree_remove_state(st); 1791 pf_detach_state(st); 1792 } 1793 1794 void 1795 pf_remove_divert_state(struct pf_state_key *sk) 1796 { 1797 struct pf_state_item *si; 1798 1799 PF_ASSERT_UNLOCKED(); 1800 1801 PF_LOCK(); 1802 PF_STATE_ENTER_WRITE(); 1803 TAILQ_FOREACH(si, &sk->sk_states, si_entry) { 1804 struct pf_state *sist = si->si_st; 1805 if (sk == sist->key[PF_SK_STACK] && sist->rule.ptr && 1806 (sist->rule.ptr->divert.type == PF_DIVERT_TO || 1807 sist->rule.ptr->divert.type == PF_DIVERT_REPLY)) { 1808 if (sist->key[PF_SK_STACK]->proto == IPPROTO_TCP && 1809 sist->key[PF_SK_WIRE] != sist->key[PF_SK_STACK]) { 1810 /* 1811 * If the local address is translated, keep 1812 * the state for "tcp.closed" seconds to 1813 * prevent its source port from being reused. 1814 */ 1815 if (sist->src.state < TCPS_FIN_WAIT_2 || 1816 sist->dst.state < TCPS_FIN_WAIT_2) { 1817 pf_set_protostate(sist, PF_PEER_BOTH, 1818 TCPS_TIME_WAIT); 1819 sist->timeout = PFTM_TCP_CLOSED; 1820 sist->expire = getuptime(); 1821 } 1822 sist->state_flags |= PFSTATE_INP_UNLINKED; 1823 } else 1824 pf_remove_state(sist); 1825 break; 1826 } 1827 } 1828 PF_STATE_EXIT_WRITE(); 1829 PF_UNLOCK(); 1830 } 1831 1832 void 1833 pf_free_state(struct pf_state *st) 1834 { 1835 struct pf_rule_item *ri; 1836 1837 PF_ASSERT_LOCKED(); 1838 1839 #if NPFSYNC > 0 1840 if (pfsync_state_in_use(st)) 1841 return; 1842 #endif /* NPFSYNC > 0 */ 1843 1844 KASSERT(st->timeout == PFTM_UNLINKED); 1845 if (--st->rule.ptr->states_cur == 0 && 1846 st->rule.ptr->src_nodes == 0) 1847 pf_rm_rule(NULL, st->rule.ptr); 1848 if (st->anchor.ptr != NULL) 1849 if (--st->anchor.ptr->states_cur == 0) 1850 pf_rm_rule(NULL, st->anchor.ptr); 1851 while ((ri = SLIST_FIRST(&st->match_rules))) { 1852 SLIST_REMOVE_HEAD(&st->match_rules, entry); 1853 if (--ri->r->states_cur == 0 && 1854 ri->r->src_nodes == 0) 1855 pf_rm_rule(NULL, ri->r); 1856 pool_put(&pf_rule_item_pl, ri); 1857 } 1858 pf_normalize_tcp_cleanup(st); 1859 pfi_kif_unref(st->kif, PFI_KIF_REF_STATE); 1860 pf_state_list_remove(&pf_state_list, st); 1861 if (st->tag) 1862 pf_tag_unref(st->tag); 1863 pf_state_unref(st); 1864 pf_status.fcounters[FCNT_STATE_REMOVALS]++; 1865 pf_status.states--; 1866 } 1867 1868 unsigned int 1869 pf_purge_expired_states(const unsigned int limit, const unsigned int collect) 1870 { 1871 /* 1872 * this task/thread/context/whatever is the only thing that 1873 * removes states from the pf_state_list, so the cur reference 1874 * it holds between calls is guaranteed to still be in the 1875 * list. 1876 */ 1877 static struct pf_state *cur = NULL; 1878 1879 struct pf_state *head, *tail; 1880 struct pf_state *st; 1881 SLIST_HEAD(pf_state_gcl, pf_state) gcl = SLIST_HEAD_INITIALIZER(gcl); 1882 time_t now; 1883 unsigned int scanned; 1884 unsigned int collected = 0; 1885 1886 PF_ASSERT_UNLOCKED(); 1887 1888 rw_enter_read(&pf_state_list.pfs_rwl); 1889 1890 mtx_enter(&pf_state_list.pfs_mtx); 1891 head = TAILQ_FIRST(&pf_state_list.pfs_list); 1892 tail = TAILQ_LAST(&pf_state_list.pfs_list, pf_state_queue); 1893 mtx_leave(&pf_state_list.pfs_mtx); 1894 1895 if (head == NULL) { 1896 /* the list is empty */ 1897 rw_exit_read(&pf_state_list.pfs_rwl); 1898 return (limit); 1899 } 1900 1901 /* (re)start at the front of the list */ 1902 if (cur == NULL) 1903 cur = head; 1904 1905 now = getuptime(); 1906 1907 for (scanned = 0; scanned < limit; scanned++) { 1908 uint8_t stimeout = cur->timeout; 1909 unsigned int limited = 0; 1910 1911 if ((stimeout == PFTM_UNLINKED) || 1912 (pf_state_expires(cur, stimeout) <= now)) { 1913 st = pf_state_ref(cur); 1914 SLIST_INSERT_HEAD(&gcl, st, gc_list); 1915 1916 if (++collected >= collect) 1917 limited = 1; 1918 } 1919 1920 /* don't iterate past the end of our view of the list */ 1921 if (cur == tail) { 1922 cur = NULL; 1923 break; 1924 } 1925 1926 cur = TAILQ_NEXT(cur, entry_list); 1927 1928 /* don't spend too much time here. */ 1929 if (ISSET(READ_ONCE(curcpu()->ci_schedstate.spc_schedflags), 1930 SPCF_SHOULDYIELD) || limited) 1931 break; 1932 } 1933 1934 rw_exit_read(&pf_state_list.pfs_rwl); 1935 1936 if (SLIST_EMPTY(&gcl)) 1937 return (scanned); 1938 1939 rw_enter_write(&pf_state_list.pfs_rwl); 1940 PF_LOCK(); 1941 PF_STATE_ENTER_WRITE(); 1942 SLIST_FOREACH(st, &gcl, gc_list) { 1943 if (st->timeout != PFTM_UNLINKED) 1944 pf_remove_state(st); 1945 1946 pf_free_state(st); 1947 } 1948 PF_STATE_EXIT_WRITE(); 1949 PF_UNLOCK(); 1950 rw_exit_write(&pf_state_list.pfs_rwl); 1951 1952 while ((st = SLIST_FIRST(&gcl)) != NULL) { 1953 SLIST_REMOVE_HEAD(&gcl, gc_list); 1954 pf_state_unref(st); 1955 } 1956 1957 return (scanned); 1958 } 1959 1960 int 1961 pf_tbladdr_setup(struct pf_ruleset *rs, struct pf_addr_wrap *aw, int wait) 1962 { 1963 if (aw->type != PF_ADDR_TABLE) 1964 return (0); 1965 if ((aw->p.tbl = pfr_attach_table(rs, aw->v.tblname, wait)) == NULL) 1966 return (1); 1967 return (0); 1968 } 1969 1970 void 1971 pf_tbladdr_remove(struct pf_addr_wrap *aw) 1972 { 1973 if (aw->type != PF_ADDR_TABLE || aw->p.tbl == NULL) 1974 return; 1975 pfr_detach_table(aw->p.tbl); 1976 aw->p.tbl = NULL; 1977 } 1978 1979 void 1980 pf_tbladdr_copyout(struct pf_addr_wrap *aw) 1981 { 1982 struct pfr_ktable *kt = aw->p.tbl; 1983 1984 if (aw->type != PF_ADDR_TABLE || kt == NULL) 1985 return; 1986 if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) 1987 kt = kt->pfrkt_root; 1988 aw->p.tbl = NULL; 1989 aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ? 1990 kt->pfrkt_cnt : -1; 1991 } 1992 1993 void 1994 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 1995 { 1996 switch (af) { 1997 case AF_INET: { 1998 u_int32_t a = ntohl(addr->addr32[0]); 1999 addlog("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2000 (a>>8)&255, a&255); 2001 if (p) { 2002 p = ntohs(p); 2003 addlog(":%u", p); 2004 } 2005 break; 2006 } 2007 #ifdef INET6 2008 case AF_INET6: { 2009 u_int16_t b; 2010 u_int8_t i, curstart, curend, maxstart, maxend; 2011 curstart = curend = maxstart = maxend = 255; 2012 for (i = 0; i < 8; i++) { 2013 if (!addr->addr16[i]) { 2014 if (curstart == 255) 2015 curstart = i; 2016 curend = i; 2017 } else { 2018 if ((curend - curstart) > 2019 (maxend - maxstart)) { 2020 maxstart = curstart; 2021 maxend = curend; 2022 } 2023 curstart = curend = 255; 2024 } 2025 } 2026 if ((curend - curstart) > 2027 (maxend - maxstart)) { 2028 maxstart = curstart; 2029 maxend = curend; 2030 } 2031 for (i = 0; i < 8; i++) { 2032 if (i >= maxstart && i <= maxend) { 2033 if (i == 0) 2034 addlog(":"); 2035 if (i == maxend) 2036 addlog(":"); 2037 } else { 2038 b = ntohs(addr->addr16[i]); 2039 addlog("%x", b); 2040 if (i < 7) 2041 addlog(":"); 2042 } 2043 } 2044 if (p) { 2045 p = ntohs(p); 2046 addlog("[%u]", p); 2047 } 2048 break; 2049 } 2050 #endif /* INET6 */ 2051 } 2052 } 2053 2054 void 2055 pf_print_state(struct pf_state *st) 2056 { 2057 pf_print_state_parts(st, NULL, NULL); 2058 } 2059 2060 void 2061 pf_print_state_parts(struct pf_state *st, 2062 struct pf_state_key *skwp, struct pf_state_key *sksp) 2063 { 2064 struct pf_state_key *skw, *sks; 2065 u_int8_t proto, dir; 2066 2067 /* Do our best to fill these, but they're skipped if NULL */ 2068 skw = skwp ? skwp : (st ? st->key[PF_SK_WIRE] : NULL); 2069 sks = sksp ? sksp : (st ? st->key[PF_SK_STACK] : NULL); 2070 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2071 dir = st ? st->direction : 0; 2072 2073 switch (proto) { 2074 case IPPROTO_IPV4: 2075 addlog("IPv4"); 2076 break; 2077 case IPPROTO_IPV6: 2078 addlog("IPv6"); 2079 break; 2080 case IPPROTO_TCP: 2081 addlog("TCP"); 2082 break; 2083 case IPPROTO_UDP: 2084 addlog("UDP"); 2085 break; 2086 case IPPROTO_ICMP: 2087 addlog("ICMP"); 2088 break; 2089 case IPPROTO_ICMPV6: 2090 addlog("ICMPv6"); 2091 break; 2092 default: 2093 addlog("%u", proto); 2094 break; 2095 } 2096 switch (dir) { 2097 case PF_IN: 2098 addlog(" in"); 2099 break; 2100 case PF_OUT: 2101 addlog(" out"); 2102 break; 2103 } 2104 if (skw) { 2105 addlog(" wire: (%d) ", skw->rdomain); 2106 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2107 addlog(" "); 2108 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2109 } 2110 if (sks) { 2111 addlog(" stack: (%d) ", sks->rdomain); 2112 if (sks != skw) { 2113 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2114 addlog(" "); 2115 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2116 } else 2117 addlog("-"); 2118 } 2119 if (st) { 2120 if (proto == IPPROTO_TCP) { 2121 addlog(" [lo=%u high=%u win=%u modulator=%u", 2122 st->src.seqlo, st->src.seqhi, 2123 st->src.max_win, st->src.seqdiff); 2124 if (st->src.wscale && st->dst.wscale) 2125 addlog(" wscale=%u", 2126 st->src.wscale & PF_WSCALE_MASK); 2127 addlog("]"); 2128 addlog(" [lo=%u high=%u win=%u modulator=%u", 2129 st->dst.seqlo, st->dst.seqhi, 2130 st->dst.max_win, st->dst.seqdiff); 2131 if (st->src.wscale && st->dst.wscale) 2132 addlog(" wscale=%u", 2133 st->dst.wscale & PF_WSCALE_MASK); 2134 addlog("]"); 2135 } 2136 addlog(" %u:%u", st->src.state, st->dst.state); 2137 if (st->rule.ptr) 2138 addlog(" @%d", st->rule.ptr->nr); 2139 } 2140 } 2141 2142 void 2143 pf_print_flags(u_int8_t f) 2144 { 2145 if (f) 2146 addlog(" "); 2147 if (f & TH_FIN) 2148 addlog("F"); 2149 if (f & TH_SYN) 2150 addlog("S"); 2151 if (f & TH_RST) 2152 addlog("R"); 2153 if (f & TH_PUSH) 2154 addlog("P"); 2155 if (f & TH_ACK) 2156 addlog("A"); 2157 if (f & TH_URG) 2158 addlog("U"); 2159 if (f & TH_ECE) 2160 addlog("E"); 2161 if (f & TH_CWR) 2162 addlog("W"); 2163 } 2164 2165 #define PF_SET_SKIP_STEPS(i) \ 2166 do { \ 2167 while (head[i] != cur) { \ 2168 head[i]->skip[i].ptr = cur; \ 2169 head[i] = TAILQ_NEXT(head[i], entries); \ 2170 } \ 2171 } while (0) 2172 2173 void 2174 pf_calc_skip_steps(struct pf_rulequeue *rules) 2175 { 2176 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT]; 2177 int i; 2178 2179 cur = TAILQ_FIRST(rules); 2180 prev = cur; 2181 for (i = 0; i < PF_SKIP_COUNT; ++i) 2182 head[i] = cur; 2183 while (cur != NULL) { 2184 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2185 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2186 if (cur->direction != prev->direction) 2187 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2188 if (cur->onrdomain != prev->onrdomain || 2189 cur->ifnot != prev->ifnot) 2190 PF_SET_SKIP_STEPS(PF_SKIP_RDOM); 2191 if (cur->af != prev->af) 2192 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2193 if (cur->proto != prev->proto) 2194 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2195 if (cur->src.neg != prev->src.neg || 2196 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2197 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2198 if (cur->dst.neg != prev->dst.neg || 2199 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2200 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2201 if (cur->src.port[0] != prev->src.port[0] || 2202 cur->src.port[1] != prev->src.port[1] || 2203 cur->src.port_op != prev->src.port_op) 2204 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2205 if (cur->dst.port[0] != prev->dst.port[0] || 2206 cur->dst.port[1] != prev->dst.port[1] || 2207 cur->dst.port_op != prev->dst.port_op) 2208 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2209 2210 prev = cur; 2211 cur = TAILQ_NEXT(cur, entries); 2212 } 2213 for (i = 0; i < PF_SKIP_COUNT; ++i) 2214 PF_SET_SKIP_STEPS(i); 2215 } 2216 2217 int 2218 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2219 { 2220 if (aw1->type != aw2->type) 2221 return (1); 2222 switch (aw1->type) { 2223 case PF_ADDR_ADDRMASK: 2224 case PF_ADDR_RANGE: 2225 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2226 return (1); 2227 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2228 return (1); 2229 return (0); 2230 case PF_ADDR_DYNIFTL: 2231 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2232 case PF_ADDR_NONE: 2233 case PF_ADDR_NOROUTE: 2234 case PF_ADDR_URPFFAILED: 2235 return (0); 2236 case PF_ADDR_TABLE: 2237 return (aw1->p.tbl != aw2->p.tbl); 2238 case PF_ADDR_RTLABEL: 2239 return (aw1->v.rtlabel != aw2->v.rtlabel); 2240 default: 2241 addlog("invalid address type: %d\n", aw1->type); 2242 return (1); 2243 } 2244 } 2245 2246 /* This algorithm computes 'a + b - c' in ones-complement using a trick to 2247 * emulate at most one ones-complement subtraction. This thereby limits net 2248 * carries/borrows to at most one, eliminating a reduction step and saving one 2249 * each of +, >>, & and ~. 2250 * 2251 * def. x mod y = x - (x//y)*y for integer x,y 2252 * def. sum = x mod 2^16 2253 * def. accumulator = (x >> 16) mod 2^16 2254 * 2255 * The trick works as follows: subtracting exactly one u_int16_t from the 2256 * u_int32_t x incurs at most one underflow, wrapping its upper 16-bits, the 2257 * accumulator, to 2^16 - 1. Adding this to the 16-bit sum preserves the 2258 * ones-complement borrow: 2259 * 2260 * (sum + accumulator) mod 2^16 2261 * = { assume underflow: accumulator := 2^16 - 1 } 2262 * (sum + 2^16 - 1) mod 2^16 2263 * = { mod } 2264 * (sum - 1) mod 2^16 2265 * 2266 * Although this breaks for sum = 0, giving 0xffff, which is ones-complement's 2267 * other zero, not -1, that cannot occur: the 16-bit sum cannot be underflown 2268 * to zero as that requires subtraction of at least 2^16, which exceeds a 2269 * single u_int16_t's range. 2270 * 2271 * We use the following theorem to derive the implementation: 2272 * 2273 * th. (x + (y mod z)) mod z = (x + y) mod z (0) 2274 * proof. 2275 * (x + (y mod z)) mod z 2276 * = { def mod } 2277 * (x + y - (y//z)*z) mod z 2278 * = { (a + b*c) mod c = a mod c } 2279 * (x + y) mod z [end of proof] 2280 * 2281 * ... and thereby obtain: 2282 * 2283 * (sum + accumulator) mod 2^16 2284 * = { def. accumulator, def. sum } 2285 * (x mod 2^16 + (x >> 16) mod 2^16) mod 2^16 2286 * = { (0), twice } 2287 * (x + (x >> 16)) mod 2^16 2288 * = { x mod 2^n = x & (2^n - 1) } 2289 * (x + (x >> 16)) & 0xffff 2290 * 2291 * Note: this serves also as a reduction step for at most one add (as the 2292 * trailing mod 2^16 prevents further reductions by destroying carries). 2293 */ 2294 __inline void 2295 pf_cksum_fixup(u_int16_t *cksum, u_int16_t was, u_int16_t now, 2296 u_int8_t proto) 2297 { 2298 u_int32_t x; 2299 const int udp = proto == IPPROTO_UDP; 2300 2301 x = *cksum + was - now; 2302 x = (x + (x >> 16)) & 0xffff; 2303 2304 /* optimise: eliminate a branch when not udp */ 2305 if (udp && *cksum == 0x0000) 2306 return; 2307 if (udp && x == 0x0000) 2308 x = 0xffff; 2309 2310 *cksum = (u_int16_t)(x); 2311 } 2312 2313 #ifdef INET6 2314 /* pre: coverage(cksum) is superset of coverage(covered_cksum) */ 2315 static __inline void 2316 pf_cksum_uncover(u_int16_t *cksum, u_int16_t covered_cksum, u_int8_t proto) 2317 { 2318 pf_cksum_fixup(cksum, ~covered_cksum, 0x0, proto); 2319 } 2320 2321 /* pre: disjoint(coverage(cksum), coverage(uncovered_cksum)) */ 2322 static __inline void 2323 pf_cksum_cover(u_int16_t *cksum, u_int16_t uncovered_cksum, u_int8_t proto) 2324 { 2325 pf_cksum_fixup(cksum, 0x0, ~uncovered_cksum, proto); 2326 } 2327 #endif /* INET6 */ 2328 2329 /* pre: *a is 16-bit aligned within its packet 2330 * 2331 * This algorithm emulates 16-bit ones-complement sums on a twos-complement 2332 * machine by conserving ones-complement's otherwise discarded carries in the 2333 * upper bits of x. These accumulated carries when added to the lower 16-bits 2334 * over at least zero 'reduction' steps then complete the ones-complement sum. 2335 * 2336 * def. sum = x mod 2^16 2337 * def. accumulator = (x >> 16) 2338 * 2339 * At most two reduction steps 2340 * 2341 * x := sum + accumulator 2342 * = { def sum, def accumulator } 2343 * x := x mod 2^16 + (x >> 16) 2344 * = { x mod 2^n = x & (2^n - 1) } 2345 * x := (x & 0xffff) + (x >> 16) 2346 * 2347 * are necessary to incorporate the accumulated carries (at most one per add) 2348 * i.e. to reduce x < 2^16 from at most 16 carries in the upper 16 bits. 2349 * 2350 * The function is also invariant over the endian of the host. Why? 2351 * 2352 * Define the unary transpose operator ~ on a bitstring in python slice 2353 * notation as lambda m: m[P:] + m[:P] , for some constant pivot P. 2354 * 2355 * th. ~ distributes over ones-complement addition, denoted by +_1, i.e. 2356 * 2357 * ~m +_1 ~n = ~(m +_1 n) (for all bitstrings m,n of equal length) 2358 * 2359 * proof. Regard the bitstrings in m +_1 n as split at P, forming at most two 2360 * 'half-adds'. Under ones-complement addition, each half-add carries to the 2361 * other, so the sum of each half-add is unaffected by their relative 2362 * order. Therefore: 2363 * 2364 * ~m +_1 ~n 2365 * = { half-adds invariant under transposition } 2366 * ~s 2367 * = { substitute } 2368 * ~(m +_1 n) [end of proof] 2369 * 2370 * th. Summing two in-memory ones-complement 16-bit variables m,n on a machine 2371 * with the converse endian does not alter the result. 2372 * 2373 * proof. 2374 * { converse machine endian: load/store transposes, P := 8 } 2375 * ~(~m +_1 ~n) 2376 * = { ~ over +_1 } 2377 * ~~m +_1 ~~n 2378 * = { ~ is an involution } 2379 * m +_1 n [end of proof] 2380 * 2381 */ 2382 #define NEG(x) ((u_int16_t)~(x)) 2383 void 2384 pf_cksum_fixup_a(u_int16_t *cksum, const struct pf_addr *a, 2385 const struct pf_addr *an, sa_family_t af, u_int8_t proto) 2386 { 2387 u_int32_t x; 2388 const u_int16_t *n = an->addr16; 2389 const u_int16_t *o = a->addr16; 2390 const int udp = proto == IPPROTO_UDP; 2391 2392 switch (af) { 2393 case AF_INET: 2394 x = *cksum + o[0] + NEG(n[0]) + o[1] + NEG(n[1]); 2395 break; 2396 #ifdef INET6 2397 case AF_INET6: 2398 x = *cksum + o[0] + NEG(n[0]) + o[1] + NEG(n[1]) +\ 2399 o[2] + NEG(n[2]) + o[3] + NEG(n[3]) +\ 2400 o[4] + NEG(n[4]) + o[5] + NEG(n[5]) +\ 2401 o[6] + NEG(n[6]) + o[7] + NEG(n[7]); 2402 break; 2403 #endif /* INET6 */ 2404 default: 2405 unhandled_af(af); 2406 } 2407 2408 x = (x & 0xffff) + (x >> 16); 2409 x = (x & 0xffff) + (x >> 16); 2410 2411 /* optimise: eliminate a branch when not udp */ 2412 if (udp && *cksum == 0x0000) 2413 return; 2414 if (udp && x == 0x0000) 2415 x = 0xffff; 2416 2417 *cksum = (u_int16_t)(x); 2418 } 2419 2420 int 2421 pf_patch_8(struct pf_pdesc *pd, u_int8_t *f, u_int8_t v, bool hi) 2422 { 2423 int rewrite = 0; 2424 2425 if (*f != v) { 2426 u_int16_t old = htons(hi ? (*f << 8) : *f); 2427 u_int16_t new = htons(hi ? ( v << 8) : v); 2428 2429 pf_cksum_fixup(pd->pcksum, old, new, pd->proto); 2430 *f = v; 2431 rewrite = 1; 2432 } 2433 2434 return (rewrite); 2435 } 2436 2437 /* pre: *f is 16-bit aligned within its packet */ 2438 int 2439 pf_patch_16(struct pf_pdesc *pd, u_int16_t *f, u_int16_t v) 2440 { 2441 int rewrite = 0; 2442 2443 if (*f != v) { 2444 pf_cksum_fixup(pd->pcksum, *f, v, pd->proto); 2445 *f = v; 2446 rewrite = 1; 2447 } 2448 2449 return (rewrite); 2450 } 2451 2452 int 2453 pf_patch_16_unaligned(struct pf_pdesc *pd, void *f, u_int16_t v, bool hi) 2454 { 2455 int rewrite = 0; 2456 u_int8_t *fb = (u_int8_t*)f; 2457 u_int8_t *vb = (u_int8_t*)&v; 2458 2459 if (hi && ALIGNED_POINTER(f, u_int16_t)) { 2460 return (pf_patch_16(pd, f, v)); /* optimise */ 2461 } 2462 2463 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 2464 rewrite += pf_patch_8(pd, fb++, *vb++,!hi); 2465 2466 return (rewrite); 2467 } 2468 2469 /* pre: *f is 16-bit aligned within its packet */ 2470 /* pre: pd->proto != IPPROTO_UDP */ 2471 int 2472 pf_patch_32(struct pf_pdesc *pd, u_int32_t *f, u_int32_t v) 2473 { 2474 int rewrite = 0; 2475 u_int16_t *pc = pd->pcksum; 2476 u_int8_t proto = pd->proto; 2477 2478 /* optimise: inline udp fixup code is unused; let compiler scrub it */ 2479 if (proto == IPPROTO_UDP) 2480 panic("%s: udp", __func__); 2481 2482 /* optimise: skip *f != v guard; true for all use-cases */ 2483 pf_cksum_fixup(pc, *f / (1 << 16), v / (1 << 16), proto); 2484 pf_cksum_fixup(pc, *f % (1 << 16), v % (1 << 16), proto); 2485 2486 *f = v; 2487 rewrite = 1; 2488 2489 return (rewrite); 2490 } 2491 2492 int 2493 pf_patch_32_unaligned(struct pf_pdesc *pd, void *f, u_int32_t v, bool hi) 2494 { 2495 int rewrite = 0; 2496 u_int8_t *fb = (u_int8_t*)f; 2497 u_int8_t *vb = (u_int8_t*)&v; 2498 2499 if (hi && ALIGNED_POINTER(f, u_int32_t)) { 2500 return (pf_patch_32(pd, f, v)); /* optimise */ 2501 } 2502 2503 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 2504 rewrite += pf_patch_8(pd, fb++, *vb++,!hi); 2505 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 2506 rewrite += pf_patch_8(pd, fb++, *vb++,!hi); 2507 2508 return (rewrite); 2509 } 2510 2511 int 2512 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, int *icmp_dir, 2513 u_int16_t *virtual_id, u_int16_t *virtual_type) 2514 { 2515 /* 2516 * ICMP types marked with PF_OUT are typically responses to 2517 * PF_IN, and will match states in the opposite direction. 2518 * PF_IN ICMP types need to match a state with that type. 2519 */ 2520 *icmp_dir = PF_OUT; 2521 2522 /* Queries (and responses) */ 2523 switch (pd->af) { 2524 case AF_INET: 2525 switch (type) { 2526 case ICMP_ECHO: 2527 *icmp_dir = PF_IN; 2528 /* FALLTHROUGH */ 2529 case ICMP_ECHOREPLY: 2530 *virtual_type = ICMP_ECHO; 2531 *virtual_id = pd->hdr.icmp.icmp_id; 2532 break; 2533 2534 case ICMP_TSTAMP: 2535 *icmp_dir = PF_IN; 2536 /* FALLTHROUGH */ 2537 case ICMP_TSTAMPREPLY: 2538 *virtual_type = ICMP_TSTAMP; 2539 *virtual_id = pd->hdr.icmp.icmp_id; 2540 break; 2541 2542 case ICMP_IREQ: 2543 *icmp_dir = PF_IN; 2544 /* FALLTHROUGH */ 2545 case ICMP_IREQREPLY: 2546 *virtual_type = ICMP_IREQ; 2547 *virtual_id = pd->hdr.icmp.icmp_id; 2548 break; 2549 2550 case ICMP_MASKREQ: 2551 *icmp_dir = PF_IN; 2552 /* FALLTHROUGH */ 2553 case ICMP_MASKREPLY: 2554 *virtual_type = ICMP_MASKREQ; 2555 *virtual_id = pd->hdr.icmp.icmp_id; 2556 break; 2557 2558 case ICMP_IPV6_WHEREAREYOU: 2559 *icmp_dir = PF_IN; 2560 /* FALLTHROUGH */ 2561 case ICMP_IPV6_IAMHERE: 2562 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2563 *virtual_id = 0; /* Nothing sane to match on! */ 2564 break; 2565 2566 case ICMP_MOBILE_REGREQUEST: 2567 *icmp_dir = PF_IN; 2568 /* FALLTHROUGH */ 2569 case ICMP_MOBILE_REGREPLY: 2570 *virtual_type = ICMP_MOBILE_REGREQUEST; 2571 *virtual_id = 0; /* Nothing sane to match on! */ 2572 break; 2573 2574 case ICMP_ROUTERSOLICIT: 2575 *icmp_dir = PF_IN; 2576 /* FALLTHROUGH */ 2577 case ICMP_ROUTERADVERT: 2578 *virtual_type = ICMP_ROUTERSOLICIT; 2579 *virtual_id = 0; /* Nothing sane to match on! */ 2580 break; 2581 2582 /* These ICMP types map to other connections */ 2583 case ICMP_UNREACH: 2584 case ICMP_SOURCEQUENCH: 2585 case ICMP_REDIRECT: 2586 case ICMP_TIMXCEED: 2587 case ICMP_PARAMPROB: 2588 /* These will not be used, but set them anyway */ 2589 *icmp_dir = PF_IN; 2590 *virtual_type = htons(type); 2591 *virtual_id = 0; 2592 return (1); /* These types match to another state */ 2593 2594 /* 2595 * All remaining ICMP types get their own states, 2596 * and will only match in one direction. 2597 */ 2598 default: 2599 *icmp_dir = PF_IN; 2600 *virtual_type = type; 2601 *virtual_id = 0; 2602 break; 2603 } 2604 break; 2605 #ifdef INET6 2606 case AF_INET6: 2607 switch (type) { 2608 case ICMP6_ECHO_REQUEST: 2609 *icmp_dir = PF_IN; 2610 /* FALLTHROUGH */ 2611 case ICMP6_ECHO_REPLY: 2612 *virtual_type = ICMP6_ECHO_REQUEST; 2613 *virtual_id = pd->hdr.icmp6.icmp6_id; 2614 break; 2615 2616 case MLD_LISTENER_QUERY: 2617 case MLD_LISTENER_REPORT: { 2618 struct mld_hdr *mld = &pd->hdr.mld; 2619 u_int32_t h; 2620 2621 /* 2622 * Listener Report can be sent by clients 2623 * without an associated Listener Query. 2624 * In addition to that, when Report is sent as a 2625 * reply to a Query its source and destination 2626 * address are different. 2627 */ 2628 *icmp_dir = PF_IN; 2629 *virtual_type = MLD_LISTENER_QUERY; 2630 /* generate fake id for these messages */ 2631 h = mld->mld_addr.s6_addr32[0] ^ 2632 mld->mld_addr.s6_addr32[1] ^ 2633 mld->mld_addr.s6_addr32[2] ^ 2634 mld->mld_addr.s6_addr32[3]; 2635 *virtual_id = (h >> 16) ^ (h & 0xffff); 2636 break; 2637 } 2638 2639 /* 2640 * ICMP6_FQDN and ICMP6_NI query/reply are the same type as 2641 * ICMP6_WRU 2642 */ 2643 case ICMP6_WRUREQUEST: 2644 *icmp_dir = PF_IN; 2645 /* FALLTHROUGH */ 2646 case ICMP6_WRUREPLY: 2647 *virtual_type = ICMP6_WRUREQUEST; 2648 *virtual_id = 0; /* Nothing sane to match on! */ 2649 break; 2650 2651 case MLD_MTRACE: 2652 *icmp_dir = PF_IN; 2653 /* FALLTHROUGH */ 2654 case MLD_MTRACE_RESP: 2655 *virtual_type = MLD_MTRACE; 2656 *virtual_id = 0; /* Nothing sane to match on! */ 2657 break; 2658 2659 case ND_NEIGHBOR_SOLICIT: 2660 *icmp_dir = PF_IN; 2661 /* FALLTHROUGH */ 2662 case ND_NEIGHBOR_ADVERT: { 2663 struct nd_neighbor_solicit *nd = &pd->hdr.nd_ns; 2664 u_int32_t h; 2665 2666 *virtual_type = ND_NEIGHBOR_SOLICIT; 2667 /* generate fake id for these messages */ 2668 h = nd->nd_ns_target.s6_addr32[0] ^ 2669 nd->nd_ns_target.s6_addr32[1] ^ 2670 nd->nd_ns_target.s6_addr32[2] ^ 2671 nd->nd_ns_target.s6_addr32[3]; 2672 *virtual_id = (h >> 16) ^ (h & 0xffff); 2673 /* 2674 * the extra work here deals with 'keep state' option 2675 * at pass rule for unsolicited advertisement. By 2676 * returning 1 (state_icmp = 1) we override 'keep 2677 * state' to 'no state' so we don't create state for 2678 * unsolicited advertisements. No one expects answer to 2679 * unsolicited advertisements so we should be good. 2680 */ 2681 if (type == ND_NEIGHBOR_ADVERT) { 2682 *virtual_type = htons(*virtual_type); 2683 return (1); 2684 } 2685 break; 2686 } 2687 2688 /* 2689 * These ICMP types map to other connections. 2690 * ND_REDIRECT can't be in this list because the triggering 2691 * packet header is optional. 2692 */ 2693 case ICMP6_DST_UNREACH: 2694 case ICMP6_PACKET_TOO_BIG: 2695 case ICMP6_TIME_EXCEEDED: 2696 case ICMP6_PARAM_PROB: 2697 /* These will not be used, but set them anyway */ 2698 *icmp_dir = PF_IN; 2699 *virtual_type = htons(type); 2700 *virtual_id = 0; 2701 return (1); /* These types match to another state */ 2702 /* 2703 * All remaining ICMP6 types get their own states, 2704 * and will only match in one direction. 2705 */ 2706 default: 2707 *icmp_dir = PF_IN; 2708 *virtual_type = type; 2709 *virtual_id = 0; 2710 break; 2711 } 2712 break; 2713 #endif /* INET6 */ 2714 } 2715 *virtual_type = htons(*virtual_type); 2716 return (0); /* These types match to their own state */ 2717 } 2718 2719 void 2720 pf_translate_icmp(struct pf_pdesc *pd, struct pf_addr *qa, u_int16_t *qp, 2721 struct pf_addr *oa, struct pf_addr *na, u_int16_t np) 2722 { 2723 /* note: doesn't trouble to fixup quoted checksums, if any */ 2724 2725 /* change quoted protocol port */ 2726 if (qp != NULL) 2727 pf_patch_16(pd, qp, np); 2728 2729 /* change quoted ip address */ 2730 pf_cksum_fixup_a(pd->pcksum, qa, na, pd->af, pd->proto); 2731 pf_addrcpy(qa, na, pd->af); 2732 2733 /* change network-header's ip address */ 2734 if (oa) 2735 pf_translate_a(pd, oa, na); 2736 } 2737 2738 /* pre: *a is 16-bit aligned within its packet */ 2739 /* *a is a network header src/dst address */ 2740 int 2741 pf_translate_a(struct pf_pdesc *pd, struct pf_addr *a, struct pf_addr *an) 2742 { 2743 int rewrite = 0; 2744 2745 /* warning: !PF_ANEQ != PF_AEQ */ 2746 if (!PF_ANEQ(a, an, pd->af)) 2747 return (0); 2748 2749 /* fixup transport pseudo-header, if any */ 2750 switch (pd->proto) { 2751 case IPPROTO_TCP: /* FALLTHROUGH */ 2752 case IPPROTO_UDP: /* FALLTHROUGH */ 2753 case IPPROTO_ICMPV6: 2754 pf_cksum_fixup_a(pd->pcksum, a, an, pd->af, pd->proto); 2755 break; 2756 default: 2757 break; /* assume no pseudo-header */ 2758 } 2759 2760 pf_addrcpy(a, an, pd->af); 2761 rewrite = 1; 2762 2763 return (rewrite); 2764 } 2765 2766 #ifdef INET6 2767 /* pf_translate_af() may change pd->m, adjust local copies after calling */ 2768 int 2769 pf_translate_af(struct pf_pdesc *pd) 2770 { 2771 static const struct pf_addr zero; 2772 struct ip *ip4; 2773 struct ip6_hdr *ip6; 2774 int copyback = 0; 2775 u_int hlen, ohlen, dlen; 2776 u_int16_t *pc; 2777 u_int8_t af_proto, naf_proto; 2778 2779 hlen = (pd->naf == AF_INET) ? sizeof(*ip4) : sizeof(*ip6); 2780 ohlen = pd->off; 2781 dlen = pd->tot_len - pd->off; 2782 pc = pd->pcksum; 2783 2784 af_proto = naf_proto = pd->proto; 2785 if (naf_proto == IPPROTO_ICMP) 2786 af_proto = IPPROTO_ICMPV6; 2787 if (naf_proto == IPPROTO_ICMPV6) 2788 af_proto = IPPROTO_ICMP; 2789 2790 /* uncover stale pseudo-header */ 2791 switch (af_proto) { 2792 case IPPROTO_ICMPV6: 2793 /* optimise: unchanged for TCP/UDP */ 2794 pf_cksum_fixup(pc, htons(af_proto), 0x0, af_proto); 2795 pf_cksum_fixup(pc, htons(dlen), 0x0, af_proto); 2796 /* FALLTHROUGH */ 2797 case IPPROTO_UDP: /* FALLTHROUGH */ 2798 case IPPROTO_TCP: 2799 pf_cksum_fixup_a(pc, pd->src, &zero, pd->af, af_proto); 2800 pf_cksum_fixup_a(pc, pd->dst, &zero, pd->af, af_proto); 2801 copyback = 1; 2802 break; 2803 default: 2804 break; /* assume no pseudo-header */ 2805 } 2806 2807 /* replace the network header */ 2808 m_adj(pd->m, pd->off); 2809 pd->src = NULL; 2810 pd->dst = NULL; 2811 2812 if ((M_PREPEND(pd->m, hlen, M_DONTWAIT)) == NULL) { 2813 pd->m = NULL; 2814 return (-1); 2815 } 2816 2817 pd->off = hlen; 2818 pd->tot_len += hlen - ohlen; 2819 2820 switch (pd->naf) { 2821 case AF_INET: 2822 ip4 = mtod(pd->m, struct ip *); 2823 memset(ip4, 0, hlen); 2824 ip4->ip_v = IPVERSION; 2825 ip4->ip_hl = hlen >> 2; 2826 ip4->ip_tos = pd->tos; 2827 ip4->ip_len = htons(hlen + dlen); 2828 ip4->ip_id = htons(ip_randomid()); 2829 ip4->ip_off = htons(IP_DF); 2830 ip4->ip_ttl = pd->ttl; 2831 ip4->ip_p = pd->proto; 2832 ip4->ip_src = pd->nsaddr.v4; 2833 ip4->ip_dst = pd->ndaddr.v4; 2834 break; 2835 case AF_INET6: 2836 ip6 = mtod(pd->m, struct ip6_hdr *); 2837 memset(ip6, 0, hlen); 2838 ip6->ip6_vfc = IPV6_VERSION; 2839 ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20); 2840 ip6->ip6_plen = htons(dlen); 2841 ip6->ip6_nxt = pd->proto; 2842 if (!pd->ttl || pd->ttl > IPV6_DEFHLIM) 2843 ip6->ip6_hlim = IPV6_DEFHLIM; 2844 else 2845 ip6->ip6_hlim = pd->ttl; 2846 ip6->ip6_src = pd->nsaddr.v6; 2847 ip6->ip6_dst = pd->ndaddr.v6; 2848 break; 2849 default: 2850 unhandled_af(pd->naf); 2851 } 2852 2853 /* UDP over IPv6 must be checksummed per rfc2460 p27 */ 2854 if (naf_proto == IPPROTO_UDP && *pc == 0x0000 && 2855 pd->naf == AF_INET6) { 2856 pd->m->m_pkthdr.csum_flags |= M_UDP_CSUM_OUT; 2857 } 2858 2859 /* cover fresh pseudo-header */ 2860 switch (naf_proto) { 2861 case IPPROTO_ICMPV6: 2862 /* optimise: unchanged for TCP/UDP */ 2863 pf_cksum_fixup(pc, 0x0, htons(naf_proto), naf_proto); 2864 pf_cksum_fixup(pc, 0x0, htons(dlen), naf_proto); 2865 /* FALLTHROUGH */ 2866 case IPPROTO_UDP: /* FALLTHROUGH */ 2867 case IPPROTO_TCP: 2868 pf_cksum_fixup_a(pc, &zero, &pd->nsaddr, pd->naf, naf_proto); 2869 pf_cksum_fixup_a(pc, &zero, &pd->ndaddr, pd->naf, naf_proto); 2870 copyback = 1; 2871 break; 2872 default: 2873 break; /* assume no pseudo-header */ 2874 } 2875 2876 /* flush pd->pcksum */ 2877 if (copyback) 2878 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 2879 2880 return (0); 2881 } 2882 2883 int 2884 pf_change_icmp_af(struct mbuf *m, int ipoff2, struct pf_pdesc *pd, 2885 struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst, 2886 sa_family_t af, sa_family_t naf) 2887 { 2888 struct mbuf *n = NULL; 2889 struct ip *ip4; 2890 struct ip6_hdr *ip6; 2891 u_int hlen, ohlen, dlen; 2892 int d; 2893 2894 if (af == naf || (af != AF_INET && af != AF_INET6) || 2895 (naf != AF_INET && naf != AF_INET6)) 2896 return (-1); 2897 2898 /* split the mbuf chain on the quoted ip/ip6 header boundary */ 2899 if ((n = m_split(m, ipoff2, M_DONTWAIT)) == NULL) 2900 return (-1); 2901 2902 /* new quoted header */ 2903 hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 2904 /* old quoted header */ 2905 ohlen = pd2->off - ipoff2; 2906 2907 /* trim old quoted header */ 2908 pf_cksum_uncover(pd->pcksum, in_cksum(n, ohlen), pd->proto); 2909 m_adj(n, ohlen); 2910 2911 /* prepend a new, translated, quoted header */ 2912 if ((M_PREPEND(n, hlen, M_DONTWAIT)) == NULL) 2913 return (-1); 2914 2915 switch (naf) { 2916 case AF_INET: 2917 ip4 = mtod(n, struct ip *); 2918 memset(ip4, 0, sizeof(*ip4)); 2919 ip4->ip_v = IPVERSION; 2920 ip4->ip_hl = sizeof(*ip4) >> 2; 2921 ip4->ip_len = htons(sizeof(*ip4) + pd2->tot_len - ohlen); 2922 ip4->ip_id = htons(ip_randomid()); 2923 ip4->ip_off = htons(IP_DF); 2924 ip4->ip_ttl = pd2->ttl; 2925 if (pd2->proto == IPPROTO_ICMPV6) 2926 ip4->ip_p = IPPROTO_ICMP; 2927 else 2928 ip4->ip_p = pd2->proto; 2929 ip4->ip_src = src->v4; 2930 ip4->ip_dst = dst->v4; 2931 in_hdr_cksum_out(n, NULL); 2932 break; 2933 case AF_INET6: 2934 ip6 = mtod(n, struct ip6_hdr *); 2935 memset(ip6, 0, sizeof(*ip6)); 2936 ip6->ip6_vfc = IPV6_VERSION; 2937 ip6->ip6_plen = htons(pd2->tot_len - ohlen); 2938 if (pd2->proto == IPPROTO_ICMP) 2939 ip6->ip6_nxt = IPPROTO_ICMPV6; 2940 else 2941 ip6->ip6_nxt = pd2->proto; 2942 if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM) 2943 ip6->ip6_hlim = IPV6_DEFHLIM; 2944 else 2945 ip6->ip6_hlim = pd2->ttl; 2946 ip6->ip6_src = src->v6; 2947 ip6->ip6_dst = dst->v6; 2948 break; 2949 } 2950 2951 /* cover new quoted header */ 2952 /* optimise: any new AF_INET header of ours sums to zero */ 2953 if (naf != AF_INET) { 2954 pf_cksum_cover(pd->pcksum, in_cksum(n, hlen), pd->proto); 2955 } 2956 2957 /* reattach modified quoted packet to outer header */ 2958 { 2959 int nlen = n->m_pkthdr.len; 2960 m_cat(m, n); 2961 m->m_pkthdr.len += nlen; 2962 } 2963 2964 /* account for altered length */ 2965 d = hlen - ohlen; 2966 2967 if (pd->proto == IPPROTO_ICMPV6) { 2968 /* fixup pseudo-header */ 2969 dlen = pd->tot_len - pd->off; 2970 pf_cksum_fixup(pd->pcksum, 2971 htons(dlen), htons(dlen + d), pd->proto); 2972 } 2973 2974 pd->tot_len += d; 2975 pd2->tot_len += d; 2976 pd2->off += d; 2977 2978 /* note: not bothering to update network headers as 2979 these due for rewrite by pf_translate_af() */ 2980 2981 return (0); 2982 } 2983 2984 2985 #define PTR_IP(field) (offsetof(struct ip, field)) 2986 #define PTR_IP6(field) (offsetof(struct ip6_hdr, field)) 2987 2988 int 2989 pf_translate_icmp_af(struct pf_pdesc *pd, int af, void *arg) 2990 { 2991 struct icmp *icmp4; 2992 struct icmp6_hdr *icmp6; 2993 u_int32_t mtu; 2994 int32_t ptr = -1; 2995 u_int8_t type; 2996 u_int8_t code; 2997 2998 switch (af) { 2999 case AF_INET: 3000 icmp6 = arg; 3001 type = icmp6->icmp6_type; 3002 code = icmp6->icmp6_code; 3003 mtu = ntohl(icmp6->icmp6_mtu); 3004 3005 switch (type) { 3006 case ICMP6_ECHO_REQUEST: 3007 type = ICMP_ECHO; 3008 break; 3009 case ICMP6_ECHO_REPLY: 3010 type = ICMP_ECHOREPLY; 3011 break; 3012 case ICMP6_DST_UNREACH: 3013 type = ICMP_UNREACH; 3014 switch (code) { 3015 case ICMP6_DST_UNREACH_NOROUTE: 3016 case ICMP6_DST_UNREACH_BEYONDSCOPE: 3017 case ICMP6_DST_UNREACH_ADDR: 3018 code = ICMP_UNREACH_HOST; 3019 break; 3020 case ICMP6_DST_UNREACH_ADMIN: 3021 code = ICMP_UNREACH_HOST_PROHIB; 3022 break; 3023 case ICMP6_DST_UNREACH_NOPORT: 3024 code = ICMP_UNREACH_PORT; 3025 break; 3026 default: 3027 return (-1); 3028 } 3029 break; 3030 case ICMP6_PACKET_TOO_BIG: 3031 type = ICMP_UNREACH; 3032 code = ICMP_UNREACH_NEEDFRAG; 3033 mtu -= 20; 3034 break; 3035 case ICMP6_TIME_EXCEEDED: 3036 type = ICMP_TIMXCEED; 3037 break; 3038 case ICMP6_PARAM_PROB: 3039 switch (code) { 3040 case ICMP6_PARAMPROB_HEADER: 3041 type = ICMP_PARAMPROB; 3042 code = ICMP_PARAMPROB_ERRATPTR; 3043 ptr = ntohl(icmp6->icmp6_pptr); 3044 3045 if (ptr == PTR_IP6(ip6_vfc)) 3046 ; /* preserve */ 3047 else if (ptr == PTR_IP6(ip6_vfc) + 1) 3048 ptr = PTR_IP(ip_tos); 3049 else if (ptr == PTR_IP6(ip6_plen) || 3050 ptr == PTR_IP6(ip6_plen) + 1) 3051 ptr = PTR_IP(ip_len); 3052 else if (ptr == PTR_IP6(ip6_nxt)) 3053 ptr = PTR_IP(ip_p); 3054 else if (ptr == PTR_IP6(ip6_hlim)) 3055 ptr = PTR_IP(ip_ttl); 3056 else if (ptr >= PTR_IP6(ip6_src) && 3057 ptr < PTR_IP6(ip6_dst)) 3058 ptr = PTR_IP(ip_src); 3059 else if (ptr >= PTR_IP6(ip6_dst) && 3060 ptr < sizeof(struct ip6_hdr)) 3061 ptr = PTR_IP(ip_dst); 3062 else { 3063 return (-1); 3064 } 3065 break; 3066 case ICMP6_PARAMPROB_NEXTHEADER: 3067 type = ICMP_UNREACH; 3068 code = ICMP_UNREACH_PROTOCOL; 3069 break; 3070 default: 3071 return (-1); 3072 } 3073 break; 3074 default: 3075 return (-1); 3076 } 3077 3078 pf_patch_8(pd, &icmp6->icmp6_type, type, PF_HI); 3079 pf_patch_8(pd, &icmp6->icmp6_code, code, PF_LO); 3080 3081 /* aligns well with a icmpv4 nextmtu */ 3082 pf_patch_32(pd, &icmp6->icmp6_mtu, htonl(mtu)); 3083 3084 /* icmpv4 pptr is a one most significant byte */ 3085 if (ptr >= 0) 3086 pf_patch_32(pd, &icmp6->icmp6_pptr, htonl(ptr << 24)); 3087 break; 3088 case AF_INET6: 3089 icmp4 = arg; 3090 type = icmp4->icmp_type; 3091 code = icmp4->icmp_code; 3092 mtu = ntohs(icmp4->icmp_nextmtu); 3093 3094 switch (type) { 3095 case ICMP_ECHO: 3096 type = ICMP6_ECHO_REQUEST; 3097 break; 3098 case ICMP_ECHOREPLY: 3099 type = ICMP6_ECHO_REPLY; 3100 break; 3101 case ICMP_UNREACH: 3102 type = ICMP6_DST_UNREACH; 3103 switch (code) { 3104 case ICMP_UNREACH_NET: 3105 case ICMP_UNREACH_HOST: 3106 case ICMP_UNREACH_NET_UNKNOWN: 3107 case ICMP_UNREACH_HOST_UNKNOWN: 3108 case ICMP_UNREACH_ISOLATED: 3109 case ICMP_UNREACH_TOSNET: 3110 case ICMP_UNREACH_TOSHOST: 3111 code = ICMP6_DST_UNREACH_NOROUTE; 3112 break; 3113 case ICMP_UNREACH_PORT: 3114 code = ICMP6_DST_UNREACH_NOPORT; 3115 break; 3116 case ICMP_UNREACH_NET_PROHIB: 3117 case ICMP_UNREACH_HOST_PROHIB: 3118 case ICMP_UNREACH_FILTER_PROHIB: 3119 case ICMP_UNREACH_PRECEDENCE_CUTOFF: 3120 code = ICMP6_DST_UNREACH_ADMIN; 3121 break; 3122 case ICMP_UNREACH_PROTOCOL: 3123 type = ICMP6_PARAM_PROB; 3124 code = ICMP6_PARAMPROB_NEXTHEADER; 3125 ptr = offsetof(struct ip6_hdr, ip6_nxt); 3126 break; 3127 case ICMP_UNREACH_NEEDFRAG: 3128 type = ICMP6_PACKET_TOO_BIG; 3129 code = 0; 3130 mtu += 20; 3131 break; 3132 default: 3133 return (-1); 3134 } 3135 break; 3136 case ICMP_TIMXCEED: 3137 type = ICMP6_TIME_EXCEEDED; 3138 break; 3139 case ICMP_PARAMPROB: 3140 type = ICMP6_PARAM_PROB; 3141 switch (code) { 3142 case ICMP_PARAMPROB_ERRATPTR: 3143 code = ICMP6_PARAMPROB_HEADER; 3144 break; 3145 case ICMP_PARAMPROB_LENGTH: 3146 code = ICMP6_PARAMPROB_HEADER; 3147 break; 3148 default: 3149 return (-1); 3150 } 3151 3152 ptr = icmp4->icmp_pptr; 3153 if (ptr == 0 || ptr == PTR_IP(ip_tos)) 3154 ; /* preserve */ 3155 else if (ptr == PTR_IP(ip_len) || 3156 ptr == PTR_IP(ip_len) + 1) 3157 ptr = PTR_IP6(ip6_plen); 3158 else if (ptr == PTR_IP(ip_ttl)) 3159 ptr = PTR_IP6(ip6_hlim); 3160 else if (ptr == PTR_IP(ip_p)) 3161 ptr = PTR_IP6(ip6_nxt); 3162 else if (ptr >= PTR_IP(ip_src) && 3163 ptr < PTR_IP(ip_dst)) 3164 ptr = PTR_IP6(ip6_src); 3165 else if (ptr >= PTR_IP(ip_dst) && 3166 ptr < sizeof(struct ip)) 3167 ptr = PTR_IP6(ip6_dst); 3168 else { 3169 return (-1); 3170 } 3171 break; 3172 default: 3173 return (-1); 3174 } 3175 3176 pf_patch_8(pd, &icmp4->icmp_type, type, PF_HI); 3177 pf_patch_8(pd, &icmp4->icmp_code, code, PF_LO); 3178 pf_patch_16(pd, &icmp4->icmp_nextmtu, htons(mtu)); 3179 if (ptr >= 0) 3180 pf_patch_32(pd, &icmp4->icmp_void, htonl(ptr)); 3181 break; 3182 } 3183 3184 return (0); 3185 } 3186 #endif /* INET6 */ 3187 3188 /* 3189 * Need to modulate the sequence numbers in the TCP SACK option 3190 * (credits to Krzysztof Pfaff for report and patch) 3191 */ 3192 int 3193 pf_modulate_sack(struct pf_pdesc *pd, struct pf_state_peer *dst) 3194 { 3195 struct sackblk sack; 3196 int copyback = 0, i; 3197 int olen, optsoff; 3198 u_int8_t opts[MAX_TCPOPTLEN], *opt, *eoh; 3199 3200 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 3201 optsoff = pd->off + sizeof(struct tcphdr); 3202 #define TCPOLEN_MINSACK (TCPOLEN_SACK + 2) 3203 if (olen < TCPOLEN_MINSACK || 3204 !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, pd->af)) 3205 return (0); 3206 3207 eoh = opts + olen; 3208 opt = opts; 3209 while ((opt = pf_find_tcpopt(opt, opts, olen, 3210 TCPOPT_SACK, TCPOLEN_MINSACK)) != NULL) 3211 { 3212 size_t safelen = MIN(opt[1], (eoh - opt)); 3213 for (i = 2; i + TCPOLEN_SACK <= safelen; i += TCPOLEN_SACK) { 3214 size_t startoff = (opt + i) - opts; 3215 memcpy(&sack, &opt[i], sizeof(sack)); 3216 pf_patch_32_unaligned(pd, &sack.start, 3217 htonl(ntohl(sack.start) - dst->seqdiff), 3218 PF_ALGNMNT(startoff)); 3219 pf_patch_32_unaligned(pd, &sack.end, 3220 htonl(ntohl(sack.end) - dst->seqdiff), 3221 PF_ALGNMNT(startoff + sizeof(sack.start))); 3222 memcpy(&opt[i], &sack, sizeof(sack)); 3223 } 3224 copyback = 1; 3225 opt += opt[1]; 3226 } 3227 3228 if (copyback) 3229 m_copyback(pd->m, optsoff, olen, opts, M_NOWAIT); 3230 return (copyback); 3231 } 3232 3233 struct mbuf * 3234 pf_build_tcp(const struct pf_rule *r, sa_family_t af, 3235 const struct pf_addr *saddr, const struct pf_addr *daddr, 3236 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3237 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 3238 u_int16_t rtag, u_int sack, u_int rdom) 3239 { 3240 struct mbuf *m; 3241 int len, tlen; 3242 struct ip *h; 3243 #ifdef INET6 3244 struct ip6_hdr *h6; 3245 #endif /* INET6 */ 3246 struct tcphdr *th; 3247 char *opt; 3248 3249 /* maximum segment size tcp option */ 3250 tlen = sizeof(struct tcphdr); 3251 if (mss) 3252 tlen += 4; 3253 if (sack) 3254 tlen += 2; 3255 3256 switch (af) { 3257 case AF_INET: 3258 len = sizeof(struct ip) + tlen; 3259 break; 3260 #ifdef INET6 3261 case AF_INET6: 3262 len = sizeof(struct ip6_hdr) + tlen; 3263 break; 3264 #endif /* INET6 */ 3265 default: 3266 unhandled_af(af); 3267 } 3268 3269 /* create outgoing mbuf */ 3270 m = m_gethdr(M_DONTWAIT, MT_HEADER); 3271 if (m == NULL) 3272 return (NULL); 3273 if (tag) 3274 m->m_pkthdr.pf.flags |= PF_TAG_GENERATED; 3275 m->m_pkthdr.pf.tag = rtag; 3276 m->m_pkthdr.ph_rtableid = rdom; 3277 if (r && (r->scrub_flags & PFSTATE_SETPRIO)) 3278 m->m_pkthdr.pf.prio = r->set_prio[0]; 3279 if (r && r->qid) 3280 m->m_pkthdr.pf.qid = r->qid; 3281 m->m_data += max_linkhdr; 3282 m->m_pkthdr.len = m->m_len = len; 3283 m->m_pkthdr.ph_ifidx = 0; 3284 m->m_pkthdr.csum_flags |= M_TCP_CSUM_OUT; 3285 memset(m->m_data, 0, len); 3286 switch (af) { 3287 case AF_INET: 3288 h = mtod(m, struct ip *); 3289 h->ip_p = IPPROTO_TCP; 3290 h->ip_len = htons(tlen); 3291 h->ip_v = 4; 3292 h->ip_hl = sizeof(*h) >> 2; 3293 h->ip_tos = IPTOS_LOWDELAY; 3294 h->ip_len = htons(len); 3295 h->ip_off = htons(ip_mtudisc ? IP_DF : 0); 3296 h->ip_ttl = ttl ? ttl : ip_defttl; 3297 h->ip_sum = 0; 3298 h->ip_src.s_addr = saddr->v4.s_addr; 3299 h->ip_dst.s_addr = daddr->v4.s_addr; 3300 3301 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3302 break; 3303 #ifdef INET6 3304 case AF_INET6: 3305 h6 = mtod(m, struct ip6_hdr *); 3306 h6->ip6_nxt = IPPROTO_TCP; 3307 h6->ip6_plen = htons(tlen); 3308 h6->ip6_vfc |= IPV6_VERSION; 3309 h6->ip6_hlim = IPV6_DEFHLIM; 3310 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3311 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3312 3313 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3314 break; 3315 #endif /* INET6 */ 3316 default: 3317 unhandled_af(af); 3318 } 3319 3320 /* TCP header */ 3321 th->th_sport = sport; 3322 th->th_dport = dport; 3323 th->th_seq = htonl(seq); 3324 th->th_ack = htonl(ack); 3325 th->th_off = tlen >> 2; 3326 th->th_flags = flags; 3327 th->th_win = htons(win); 3328 3329 opt = (char *)(th + 1); 3330 if (mss) { 3331 opt[0] = TCPOPT_MAXSEG; 3332 opt[1] = 4; 3333 mss = htons(mss); 3334 memcpy((opt + 2), &mss, 2); 3335 opt += 4; 3336 } 3337 if (sack) { 3338 opt[0] = TCPOPT_SACK_PERMITTED; 3339 opt[1] = 2; 3340 opt += 2; 3341 } 3342 3343 return (m); 3344 } 3345 3346 void 3347 pf_send_tcp(const struct pf_rule *r, sa_family_t af, 3348 const struct pf_addr *saddr, const struct pf_addr *daddr, 3349 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3350 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 3351 u_int16_t rtag, u_int rdom) 3352 { 3353 struct mbuf *m; 3354 3355 if ((m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, 3356 flags, win, mss, ttl, tag, rtag, 0, rdom)) == NULL) 3357 return; 3358 3359 switch (af) { 3360 case AF_INET: 3361 ip_send(m); 3362 break; 3363 #ifdef INET6 3364 case AF_INET6: 3365 ip6_send(m); 3366 break; 3367 #endif /* INET6 */ 3368 } 3369 } 3370 3371 static void 3372 pf_send_challenge_ack(struct pf_pdesc *pd, struct pf_state *st, 3373 struct pf_state_peer *src, struct pf_state_peer *dst) 3374 { 3375 /* 3376 * We are sending challenge ACK as a response to SYN packet, which 3377 * matches existing state (modulo TCP window check). Therefore packet 3378 * must be sent on behalf of destination. 3379 * 3380 * We expect sender to remain either silent, or send RST packet 3381 * so both, firewall and remote peer, can purge dead state from 3382 * memory. 3383 */ 3384 pf_send_tcp(st->rule.ptr, pd->af, pd->dst, pd->src, 3385 pd->hdr.tcp.th_dport, pd->hdr.tcp.th_sport, dst->seqlo, 3386 src->seqlo, TH_ACK, 0, 0, st->rule.ptr->return_ttl, 1, 0, 3387 pd->rdomain); 3388 } 3389 3390 void 3391 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, int param, 3392 sa_family_t af, struct pf_rule *r, u_int rdomain) 3393 { 3394 struct mbuf *m0; 3395 3396 if ((m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT)) == NULL) 3397 return; 3398 3399 m0->m_pkthdr.pf.flags |= PF_TAG_GENERATED; 3400 m0->m_pkthdr.ph_rtableid = rdomain; 3401 if (r && (r->scrub_flags & PFSTATE_SETPRIO)) 3402 m0->m_pkthdr.pf.prio = r->set_prio[0]; 3403 if (r && r->qid) 3404 m0->m_pkthdr.pf.qid = r->qid; 3405 3406 switch (af) { 3407 case AF_INET: 3408 icmp_error(m0, type, code, 0, param); 3409 break; 3410 #ifdef INET6 3411 case AF_INET6: 3412 icmp6_error(m0, type, code, param); 3413 break; 3414 #endif /* INET6 */ 3415 } 3416 } 3417 3418 /* 3419 * Return ((n = 0) == (a = b [with mask m])) 3420 * Note: n != 0 => returns (a != b [with mask m]) 3421 */ 3422 int 3423 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3424 struct pf_addr *b, sa_family_t af) 3425 { 3426 switch (af) { 3427 case AF_INET: 3428 if ((a->addr32[0] & m->addr32[0]) == 3429 (b->addr32[0] & m->addr32[0])) 3430 return (n == 0); 3431 break; 3432 #ifdef INET6 3433 case AF_INET6: 3434 if (((a->addr32[0] & m->addr32[0]) == 3435 (b->addr32[0] & m->addr32[0])) && 3436 ((a->addr32[1] & m->addr32[1]) == 3437 (b->addr32[1] & m->addr32[1])) && 3438 ((a->addr32[2] & m->addr32[2]) == 3439 (b->addr32[2] & m->addr32[2])) && 3440 ((a->addr32[3] & m->addr32[3]) == 3441 (b->addr32[3] & m->addr32[3]))) 3442 return (n == 0); 3443 break; 3444 #endif /* INET6 */ 3445 } 3446 3447 return (n != 0); 3448 } 3449 3450 /* 3451 * Return 1 if b <= a <= e, otherwise return 0. 3452 */ 3453 int 3454 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3455 struct pf_addr *a, sa_family_t af) 3456 { 3457 switch (af) { 3458 case AF_INET: 3459 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3460 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3461 return (0); 3462 break; 3463 #ifdef INET6 3464 case AF_INET6: { 3465 int i; 3466 3467 /* check a >= b */ 3468 for (i = 0; i < 4; ++i) 3469 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3470 break; 3471 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3472 return (0); 3473 /* check a <= e */ 3474 for (i = 0; i < 4; ++i) 3475 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3476 break; 3477 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3478 return (0); 3479 break; 3480 } 3481 #endif /* INET6 */ 3482 } 3483 return (1); 3484 } 3485 3486 int 3487 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3488 { 3489 switch (op) { 3490 case PF_OP_IRG: 3491 return ((p > a1) && (p < a2)); 3492 case PF_OP_XRG: 3493 return ((p < a1) || (p > a2)); 3494 case PF_OP_RRG: 3495 return ((p >= a1) && (p <= a2)); 3496 case PF_OP_EQ: 3497 return (p == a1); 3498 case PF_OP_NE: 3499 return (p != a1); 3500 case PF_OP_LT: 3501 return (p < a1); 3502 case PF_OP_LE: 3503 return (p <= a1); 3504 case PF_OP_GT: 3505 return (p > a1); 3506 case PF_OP_GE: 3507 return (p >= a1); 3508 } 3509 return (0); /* never reached */ 3510 } 3511 3512 int 3513 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3514 { 3515 return (pf_match(op, ntohs(a1), ntohs(a2), ntohs(p))); 3516 } 3517 3518 int 3519 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3520 { 3521 if (u == -1 && op != PF_OP_EQ && op != PF_OP_NE) 3522 return (0); 3523 return (pf_match(op, a1, a2, u)); 3524 } 3525 3526 int 3527 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3528 { 3529 if (g == -1 && op != PF_OP_EQ && op != PF_OP_NE) 3530 return (0); 3531 return (pf_match(op, a1, a2, g)); 3532 } 3533 3534 int 3535 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag) 3536 { 3537 if (*tag == -1) 3538 *tag = m->m_pkthdr.pf.tag; 3539 3540 return ((!r->match_tag_not && r->match_tag == *tag) || 3541 (r->match_tag_not && r->match_tag != *tag)); 3542 } 3543 3544 int 3545 pf_match_rcvif(struct mbuf *m, struct pf_rule *r) 3546 { 3547 struct ifnet *ifp; 3548 #if NCARP > 0 3549 struct ifnet *ifp0; 3550 #endif 3551 struct pfi_kif *kif; 3552 3553 ifp = if_get(m->m_pkthdr.ph_ifidx); 3554 if (ifp == NULL) 3555 return (0); 3556 3557 #if NCARP > 0 3558 if (ifp->if_type == IFT_CARP && 3559 (ifp0 = if_get(ifp->if_carpdevidx)) != NULL) { 3560 kif = (struct pfi_kif *)ifp0->if_pf_kif; 3561 if_put(ifp0); 3562 } else 3563 #endif /* NCARP */ 3564 kif = (struct pfi_kif *)ifp->if_pf_kif; 3565 3566 if_put(ifp); 3567 3568 if (kif == NULL) { 3569 DPFPRINTF(LOG_ERR, 3570 "%s: kif == NULL, @%d via %s", __func__, 3571 r->nr, r->rcv_ifname); 3572 return (0); 3573 } 3574 3575 return (pfi_kif_match(r->rcv_kif, kif)); 3576 } 3577 3578 void 3579 pf_tag_packet(struct mbuf *m, int tag, int rtableid) 3580 { 3581 if (tag > 0) 3582 m->m_pkthdr.pf.tag = tag; 3583 if (rtableid >= 0) 3584 m->m_pkthdr.ph_rtableid = (u_int)rtableid; 3585 } 3586 3587 void 3588 pf_anchor_stack_init(void) 3589 { 3590 struct pf_anchor_stackframe *stack; 3591 3592 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3593 stack[PF_ANCHOR_STACK_MAX].sf_stack_top = &stack[0]; 3594 cpumem_leave(pf_anchor_stack, stack); 3595 } 3596 3597 int 3598 pf_anchor_stack_is_full(struct pf_anchor_stackframe *sf) 3599 { 3600 struct pf_anchor_stackframe *stack; 3601 int rv; 3602 3603 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3604 rv = (sf == &stack[PF_ANCHOR_STACK_MAX]); 3605 cpumem_leave(pf_anchor_stack, stack); 3606 3607 return (rv); 3608 } 3609 3610 int 3611 pf_anchor_stack_is_empty(struct pf_anchor_stackframe *sf) 3612 { 3613 struct pf_anchor_stackframe *stack; 3614 int rv; 3615 3616 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3617 rv = (sf == &stack[0]); 3618 cpumem_leave(pf_anchor_stack, stack); 3619 3620 return (rv); 3621 } 3622 3623 struct pf_anchor_stackframe * 3624 pf_anchor_stack_top(void) 3625 { 3626 struct pf_anchor_stackframe *stack; 3627 struct pf_anchor_stackframe *top_sf; 3628 3629 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3630 top_sf = stack[PF_ANCHOR_STACK_MAX].sf_stack_top; 3631 cpumem_leave(pf_anchor_stack, stack); 3632 3633 return (top_sf); 3634 } 3635 3636 int 3637 pf_anchor_stack_push(struct pf_ruleset *rs, struct pf_rule *r, 3638 struct pf_anchor *child, int jump_target) 3639 { 3640 struct pf_anchor_stackframe *stack; 3641 struct pf_anchor_stackframe *top_sf = pf_anchor_stack_top(); 3642 3643 top_sf++; 3644 if (pf_anchor_stack_is_full(top_sf)) 3645 return (-1); 3646 3647 top_sf->sf_rs = rs; 3648 top_sf->sf_r = r; 3649 top_sf->sf_child = child; 3650 top_sf->sf_jump_target = jump_target; 3651 3652 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3653 3654 if ((top_sf <= &stack[0]) || (top_sf >= &stack[PF_ANCHOR_STACK_MAX])) 3655 panic("%s: top frame outside of anchor stack range", __func__); 3656 3657 stack[PF_ANCHOR_STACK_MAX].sf_stack_top = top_sf; 3658 cpumem_leave(pf_anchor_stack, stack); 3659 3660 return (0); 3661 } 3662 3663 int 3664 pf_anchor_stack_pop(struct pf_ruleset **rs, struct pf_rule **r, 3665 struct pf_anchor **child, int *jump_target) 3666 { 3667 struct pf_anchor_stackframe *top_sf = pf_anchor_stack_top(); 3668 struct pf_anchor_stackframe *stack; 3669 int on_top; 3670 3671 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3672 if (pf_anchor_stack_is_empty(top_sf)) { 3673 on_top = -1; 3674 } else { 3675 if ((top_sf <= &stack[0]) || 3676 (top_sf >= &stack[PF_ANCHOR_STACK_MAX])) 3677 panic("%s: top frame outside of anchor stack range", 3678 __func__); 3679 3680 *rs = top_sf->sf_rs; 3681 *r = top_sf->sf_r; 3682 *child = top_sf->sf_child; 3683 *jump_target = top_sf->sf_jump_target; 3684 top_sf--; 3685 stack[PF_ANCHOR_STACK_MAX].sf_stack_top = top_sf; 3686 on_top = 0; 3687 } 3688 cpumem_leave(pf_anchor_stack, stack); 3689 3690 return (on_top); 3691 } 3692 3693 void 3694 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3695 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3696 { 3697 switch (af) { 3698 case AF_INET: 3699 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3700 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3701 break; 3702 #ifdef INET6 3703 case AF_INET6: 3704 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3705 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3706 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3707 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3708 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3709 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3710 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3711 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3712 break; 3713 #endif /* INET6 */ 3714 default: 3715 unhandled_af(af); 3716 } 3717 } 3718 3719 void 3720 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3721 { 3722 switch (af) { 3723 case AF_INET: 3724 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3725 break; 3726 #ifdef INET6 3727 case AF_INET6: 3728 if (addr->addr32[3] == 0xffffffff) { 3729 addr->addr32[3] = 0; 3730 if (addr->addr32[2] == 0xffffffff) { 3731 addr->addr32[2] = 0; 3732 if (addr->addr32[1] == 0xffffffff) { 3733 addr->addr32[1] = 0; 3734 addr->addr32[0] = 3735 htonl(ntohl(addr->addr32[0]) + 1); 3736 } else 3737 addr->addr32[1] = 3738 htonl(ntohl(addr->addr32[1]) + 1); 3739 } else 3740 addr->addr32[2] = 3741 htonl(ntohl(addr->addr32[2]) + 1); 3742 } else 3743 addr->addr32[3] = 3744 htonl(ntohl(addr->addr32[3]) + 1); 3745 break; 3746 #endif /* INET6 */ 3747 default: 3748 unhandled_af(af); 3749 } 3750 } 3751 3752 int 3753 pf_socket_lookup(struct pf_pdesc *pd) 3754 { 3755 struct pf_addr *saddr, *daddr; 3756 u_int16_t sport, dport; 3757 struct inpcbtable *tb; 3758 struct inpcb *inp; 3759 3760 pd->lookup.uid = -1; 3761 pd->lookup.gid = -1; 3762 pd->lookup.pid = NO_PID; 3763 switch (pd->virtual_proto) { 3764 case IPPROTO_TCP: 3765 sport = pd->hdr.tcp.th_sport; 3766 dport = pd->hdr.tcp.th_dport; 3767 PF_ASSERT_LOCKED(); 3768 NET_ASSERT_LOCKED(); 3769 tb = &tcbtable; 3770 break; 3771 case IPPROTO_UDP: 3772 sport = pd->hdr.udp.uh_sport; 3773 dport = pd->hdr.udp.uh_dport; 3774 PF_ASSERT_LOCKED(); 3775 NET_ASSERT_LOCKED(); 3776 tb = &udbtable; 3777 break; 3778 default: 3779 return (-1); 3780 } 3781 if (pd->dir == PF_IN) { 3782 saddr = pd->src; 3783 daddr = pd->dst; 3784 } else { 3785 u_int16_t p; 3786 3787 p = sport; 3788 sport = dport; 3789 dport = p; 3790 saddr = pd->dst; 3791 daddr = pd->src; 3792 } 3793 switch (pd->af) { 3794 case AF_INET: 3795 /* 3796 * Fails when rtable is changed while evaluating the ruleset 3797 * The socket looked up will not match the one hit in the end. 3798 */ 3799 inp = in_pcblookup(tb, saddr->v4, sport, daddr->v4, dport, 3800 pd->rdomain); 3801 if (inp == NULL) { 3802 inp = in_pcblookup_listen(tb, daddr->v4, dport, 3803 NULL, pd->rdomain); 3804 if (inp == NULL) 3805 return (-1); 3806 } 3807 break; 3808 #ifdef INET6 3809 case AF_INET6: 3810 inp = in6_pcblookup(tb, &saddr->v6, sport, &daddr->v6, 3811 dport, pd->rdomain); 3812 if (inp == NULL) { 3813 inp = in6_pcblookup_listen(tb, &daddr->v6, dport, 3814 NULL, pd->rdomain); 3815 if (inp == NULL) 3816 return (-1); 3817 } 3818 break; 3819 #endif /* INET6 */ 3820 default: 3821 unhandled_af(pd->af); 3822 } 3823 pd->lookup.uid = inp->inp_socket->so_euid; 3824 pd->lookup.gid = inp->inp_socket->so_egid; 3825 pd->lookup.pid = inp->inp_socket->so_cpid; 3826 in_pcbunref(inp); 3827 return (1); 3828 } 3829 3830 /* post: r => (r[0] == type /\ r[1] >= min_typelen >= 2 "validity" 3831 * /\ (eoh - r) >= min_typelen >= 2 "safety" ) 3832 * 3833 * warning: r + r[1] may exceed opts bounds for r[1] > min_typelen 3834 */ 3835 u_int8_t* 3836 pf_find_tcpopt(u_int8_t *opt, u_int8_t *opts, size_t hlen, u_int8_t type, 3837 u_int8_t min_typelen) 3838 { 3839 u_int8_t *eoh = opts + hlen; 3840 3841 if (min_typelen < 2) 3842 return (NULL); 3843 3844 while ((eoh - opt) >= min_typelen) { 3845 switch (*opt) { 3846 case TCPOPT_EOL: 3847 /* FALLTHROUGH - Workaround the failure of some 3848 systems to NOP-pad their bzero'd option buffers, 3849 producing spurious EOLs */ 3850 case TCPOPT_NOP: 3851 opt++; 3852 continue; 3853 default: 3854 if (opt[0] == type && 3855 opt[1] >= min_typelen) 3856 return (opt); 3857 } 3858 3859 opt += MAX(opt[1], 2); /* evade infinite loops */ 3860 } 3861 3862 return (NULL); 3863 } 3864 3865 u_int8_t 3866 pf_get_wscale(struct pf_pdesc *pd) 3867 { 3868 int olen; 3869 u_int8_t opts[MAX_TCPOPTLEN], *opt; 3870 u_int8_t wscale = 0; 3871 3872 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 3873 if (olen < TCPOLEN_WINDOW || !pf_pull_hdr(pd->m, 3874 pd->off + sizeof(struct tcphdr), opts, olen, NULL, pd->af)) 3875 return (0); 3876 3877 opt = opts; 3878 while ((opt = pf_find_tcpopt(opt, opts, olen, 3879 TCPOPT_WINDOW, TCPOLEN_WINDOW)) != NULL) { 3880 wscale = opt[2]; 3881 wscale = MIN(wscale, TCP_MAX_WINSHIFT); 3882 wscale |= PF_WSCALE_FLAG; 3883 3884 opt += opt[1]; 3885 } 3886 3887 return (wscale); 3888 } 3889 3890 u_int16_t 3891 pf_get_mss(struct pf_pdesc *pd) 3892 { 3893 int olen; 3894 u_int8_t opts[MAX_TCPOPTLEN], *opt; 3895 u_int16_t mss = tcp_mssdflt; 3896 3897 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 3898 if (olen < TCPOLEN_MAXSEG || !pf_pull_hdr(pd->m, 3899 pd->off + sizeof(struct tcphdr), opts, olen, NULL, pd->af)) 3900 return (0); 3901 3902 opt = opts; 3903 while ((opt = pf_find_tcpopt(opt, opts, olen, 3904 TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) { 3905 memcpy(&mss, (opt + 2), 2); 3906 mss = ntohs(mss); 3907 3908 opt += opt[1]; 3909 } 3910 return (mss); 3911 } 3912 3913 u_int16_t 3914 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 3915 { 3916 struct ifnet *ifp; 3917 struct sockaddr_in *dst; 3918 #ifdef INET6 3919 struct sockaddr_in6 *dst6; 3920 #endif /* INET6 */ 3921 struct rtentry *rt = NULL; 3922 struct sockaddr_storage ss; 3923 int hlen; 3924 u_int16_t mss = tcp_mssdflt; 3925 3926 memset(&ss, 0, sizeof(ss)); 3927 3928 switch (af) { 3929 case AF_INET: 3930 hlen = sizeof(struct ip); 3931 dst = (struct sockaddr_in *)&ss; 3932 dst->sin_family = AF_INET; 3933 dst->sin_len = sizeof(*dst); 3934 dst->sin_addr = addr->v4; 3935 rt = rtalloc(sintosa(dst), 0, rtableid); 3936 break; 3937 #ifdef INET6 3938 case AF_INET6: 3939 hlen = sizeof(struct ip6_hdr); 3940 dst6 = (struct sockaddr_in6 *)&ss; 3941 dst6->sin6_family = AF_INET6; 3942 dst6->sin6_len = sizeof(*dst6); 3943 dst6->sin6_addr = addr->v6; 3944 rt = rtalloc(sin6tosa(dst6), 0, rtableid); 3945 break; 3946 #endif /* INET6 */ 3947 } 3948 3949 if (rt != NULL && (ifp = if_get(rt->rt_ifidx)) != NULL) { 3950 mss = ifp->if_mtu - hlen - sizeof(struct tcphdr); 3951 mss = max(tcp_mssdflt, mss); 3952 if_put(ifp); 3953 } 3954 rtfree(rt); 3955 mss = min(mss, offer); 3956 mss = max(mss, 64); /* sanity - at least max opt space */ 3957 return (mss); 3958 } 3959 3960 static __inline int 3961 pf_set_rt_ifp(struct pf_state *st, struct pf_addr *saddr, sa_family_t af, 3962 struct pf_src_node **sns) 3963 { 3964 struct pf_rule *r = st->rule.ptr; 3965 int rv; 3966 3967 if (!r->rt) 3968 return (0); 3969 3970 rv = pf_map_addr(af, r, saddr, &st->rt_addr, NULL, sns, 3971 &r->route, PF_SN_ROUTE); 3972 if (rv == 0) 3973 st->rt = r->rt; 3974 3975 return (rv); 3976 } 3977 3978 u_int32_t 3979 pf_tcp_iss(struct pf_pdesc *pd) 3980 { 3981 SHA2_CTX ctx; 3982 union { 3983 uint8_t bytes[SHA512_DIGEST_LENGTH]; 3984 uint32_t words[1]; 3985 } digest; 3986 3987 if (pf_tcp_secret_init == 0) { 3988 arc4random_buf(pf_tcp_secret, sizeof(pf_tcp_secret)); 3989 SHA512Init(&pf_tcp_secret_ctx); 3990 SHA512Update(&pf_tcp_secret_ctx, pf_tcp_secret, 3991 sizeof(pf_tcp_secret)); 3992 pf_tcp_secret_init = 1; 3993 } 3994 ctx = pf_tcp_secret_ctx; 3995 3996 SHA512Update(&ctx, &pd->rdomain, sizeof(pd->rdomain)); 3997 SHA512Update(&ctx, &pd->hdr.tcp.th_sport, sizeof(u_short)); 3998 SHA512Update(&ctx, &pd->hdr.tcp.th_dport, sizeof(u_short)); 3999 switch (pd->af) { 4000 case AF_INET: 4001 SHA512Update(&ctx, &pd->src->v4, sizeof(struct in_addr)); 4002 SHA512Update(&ctx, &pd->dst->v4, sizeof(struct in_addr)); 4003 break; 4004 #ifdef INET6 4005 case AF_INET6: 4006 SHA512Update(&ctx, &pd->src->v6, sizeof(struct in6_addr)); 4007 SHA512Update(&ctx, &pd->dst->v6, sizeof(struct in6_addr)); 4008 break; 4009 #endif /* INET6 */ 4010 } 4011 SHA512Final(digest.bytes, &ctx); 4012 pf_tcp_iss_off += 4096; 4013 return (digest.words[0] + READ_ONCE(tcp_iss) + pf_tcp_iss_off); 4014 } 4015 4016 void 4017 pf_rule_to_actions(struct pf_rule *r, struct pf_rule_actions *a) 4018 { 4019 if (r->qid) 4020 a->qid = r->qid; 4021 if (r->pqid) 4022 a->pqid = r->pqid; 4023 if (r->rtableid >= 0) 4024 a->rtableid = r->rtableid; 4025 #if NPFLOG > 0 4026 a->log |= r->log; 4027 #endif /* NPFLOG > 0 */ 4028 if (r->scrub_flags & PFSTATE_SETTOS) 4029 a->set_tos = r->set_tos; 4030 if (r->min_ttl) 4031 a->min_ttl = r->min_ttl; 4032 if (r->max_mss) 4033 a->max_mss = r->max_mss; 4034 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4035 PFSTATE_SETTOS|PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4036 if (r->scrub_flags & PFSTATE_SETPRIO) { 4037 a->set_prio[0] = r->set_prio[0]; 4038 a->set_prio[1] = r->set_prio[1]; 4039 } 4040 if (r->rule_flag & PFRULE_SETDELAY) 4041 a->delay = r->delay; 4042 } 4043 4044 #define PF_TEST_ATTRIB(t, a) \ 4045 if (t) { \ 4046 r = a; \ 4047 continue; \ 4048 } else do { \ 4049 } while (0) 4050 4051 enum pf_test_status 4052 pf_match_rule(struct pf_test_ctx *ctx, struct pf_ruleset *ruleset) 4053 { 4054 struct pf_rule *r; 4055 struct pf_anchor *child = NULL; 4056 int target; 4057 4058 pf_anchor_stack_init(); 4059 enter_ruleset: 4060 r = TAILQ_FIRST(ruleset->rules.active.ptr); 4061 while (r != NULL) { 4062 PF_TEST_ATTRIB(r->rule_flag & PFRULE_EXPIRED, 4063 TAILQ_NEXT(r, entries)); 4064 r->evaluations++; 4065 PF_TEST_ATTRIB( 4066 (pfi_kif_match(r->kif, ctx->pd->kif) == r->ifnot), 4067 r->skip[PF_SKIP_IFP].ptr); 4068 PF_TEST_ATTRIB((r->direction && r->direction != ctx->pd->dir), 4069 r->skip[PF_SKIP_DIR].ptr); 4070 PF_TEST_ATTRIB((r->onrdomain >= 0 && 4071 (r->onrdomain == ctx->pd->rdomain) == r->ifnot), 4072 r->skip[PF_SKIP_RDOM].ptr); 4073 PF_TEST_ATTRIB((r->af && r->af != ctx->pd->af), 4074 r->skip[PF_SKIP_AF].ptr); 4075 PF_TEST_ATTRIB((r->proto && r->proto != ctx->pd->proto), 4076 r->skip[PF_SKIP_PROTO].ptr); 4077 PF_TEST_ATTRIB((PF_MISMATCHAW(&r->src.addr, &ctx->pd->nsaddr, 4078 ctx->pd->naf, r->src.neg, ctx->pd->kif, 4079 ctx->act.rtableid)), 4080 r->skip[PF_SKIP_SRC_ADDR].ptr); 4081 PF_TEST_ATTRIB((PF_MISMATCHAW(&r->dst.addr, &ctx->pd->ndaddr, 4082 ctx->pd->af, r->dst.neg, NULL, ctx->act.rtableid)), 4083 r->skip[PF_SKIP_DST_ADDR].ptr); 4084 4085 switch (ctx->pd->virtual_proto) { 4086 case PF_VPROTO_FRAGMENT: 4087 /* tcp/udp only. port_op always 0 in other cases */ 4088 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 4089 TAILQ_NEXT(r, entries)); 4090 PF_TEST_ATTRIB((ctx->pd->proto == IPPROTO_TCP && 4091 r->flagset), 4092 TAILQ_NEXT(r, entries)); 4093 /* icmp only. type/code always 0 in other cases */ 4094 PF_TEST_ATTRIB((r->type || r->code), 4095 TAILQ_NEXT(r, entries)); 4096 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 4097 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 4098 TAILQ_NEXT(r, entries)); 4099 break; 4100 4101 case IPPROTO_TCP: 4102 PF_TEST_ATTRIB(((r->flagset & ctx->th->th_flags) != 4103 r->flags), 4104 TAILQ_NEXT(r, entries)); 4105 PF_TEST_ATTRIB((r->os_fingerprint != PF_OSFP_ANY && 4106 !pf_osfp_match(pf_osfp_fingerprint(ctx->pd), 4107 r->os_fingerprint)), 4108 TAILQ_NEXT(r, entries)); 4109 /* FALLTHROUGH */ 4110 4111 case IPPROTO_UDP: 4112 /* tcp/udp only. port_op always 0 in other cases */ 4113 PF_TEST_ATTRIB((r->src.port_op && 4114 !pf_match_port(r->src.port_op, r->src.port[0], 4115 r->src.port[1], ctx->pd->nsport)), 4116 r->skip[PF_SKIP_SRC_PORT].ptr); 4117 PF_TEST_ATTRIB((r->dst.port_op && 4118 !pf_match_port(r->dst.port_op, r->dst.port[0], 4119 r->dst.port[1], ctx->pd->ndport)), 4120 r->skip[PF_SKIP_DST_PORT].ptr); 4121 /* tcp/udp only. uid.op always 0 in other cases */ 4122 PF_TEST_ATTRIB((r->uid.op && (ctx->pd->lookup.done || 4123 (ctx->pd->lookup.done = 4124 pf_socket_lookup(ctx->pd), 1)) && 4125 !pf_match_uid(r->uid.op, r->uid.uid[0], 4126 r->uid.uid[1], ctx->pd->lookup.uid)), 4127 TAILQ_NEXT(r, entries)); 4128 /* tcp/udp only. gid.op always 0 in other cases */ 4129 PF_TEST_ATTRIB((r->gid.op && (ctx->pd->lookup.done || 4130 (ctx->pd->lookup.done = 4131 pf_socket_lookup(ctx->pd), 1)) && 4132 !pf_match_gid(r->gid.op, r->gid.gid[0], 4133 r->gid.gid[1], ctx->pd->lookup.gid)), 4134 TAILQ_NEXT(r, entries)); 4135 break; 4136 4137 case IPPROTO_ICMP: 4138 /* icmp only. type always 0 in other cases */ 4139 PF_TEST_ATTRIB((r->type && 4140 r->type != ctx->icmptype + 1), 4141 TAILQ_NEXT(r, entries)); 4142 /* icmp only. type always 0 in other cases */ 4143 PF_TEST_ATTRIB((r->code && 4144 r->code != ctx->icmpcode + 1), 4145 TAILQ_NEXT(r, entries)); 4146 /* icmp only. don't create states on replies */ 4147 PF_TEST_ATTRIB((r->keep_state && !ctx->state_icmp && 4148 (r->rule_flag & PFRULE_STATESLOPPY) == 0 && 4149 ctx->icmp_dir != PF_IN), 4150 TAILQ_NEXT(r, entries)); 4151 break; 4152 4153 case IPPROTO_ICMPV6: 4154 /* icmp only. type always 0 in other cases */ 4155 PF_TEST_ATTRIB((r->type && 4156 r->type != ctx->icmptype + 1), 4157 TAILQ_NEXT(r, entries)); 4158 /* icmp only. type always 0 in other cases */ 4159 PF_TEST_ATTRIB((r->code && 4160 r->code != ctx->icmpcode + 1), 4161 TAILQ_NEXT(r, entries)); 4162 /* icmp only. don't create states on replies */ 4163 PF_TEST_ATTRIB((r->keep_state && !ctx->state_icmp && 4164 (r->rule_flag & PFRULE_STATESLOPPY) == 0 && 4165 ctx->icmp_dir != PF_IN && 4166 ctx->icmptype != ND_NEIGHBOR_ADVERT), 4167 TAILQ_NEXT(r, entries)); 4168 break; 4169 4170 default: 4171 break; 4172 } 4173 4174 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 4175 ctx->pd->virtual_proto != PF_VPROTO_FRAGMENT), 4176 TAILQ_NEXT(r, entries)); 4177 PF_TEST_ATTRIB((r->tos && !(r->tos == ctx->pd->tos)), 4178 TAILQ_NEXT(r, entries)); 4179 PF_TEST_ATTRIB((r->prob && 4180 r->prob <= arc4random_uniform(UINT_MAX - 1) + 1), 4181 TAILQ_NEXT(r, entries)); 4182 PF_TEST_ATTRIB((r->match_tag && 4183 !pf_match_tag(ctx->pd->m, r, &ctx->tag)), 4184 TAILQ_NEXT(r, entries)); 4185 PF_TEST_ATTRIB((r->rcv_kif && pf_match_rcvif(ctx->pd->m, r) == 4186 r->rcvifnot), 4187 TAILQ_NEXT(r, entries)); 4188 PF_TEST_ATTRIB((r->prio && 4189 (r->prio == PF_PRIO_ZERO ? 0 : r->prio) != 4190 ctx->pd->m->m_pkthdr.pf.prio), 4191 TAILQ_NEXT(r, entries)); 4192 4193 /* must be last! */ 4194 if (r->pktrate.limit) { 4195 pf_add_threshold(&r->pktrate); 4196 PF_TEST_ATTRIB((pf_check_threshold(&r->pktrate)), 4197 TAILQ_NEXT(r, entries)); 4198 } 4199 4200 /* FALLTHROUGH */ 4201 if (r->tag) 4202 ctx->tag = r->tag; 4203 if (r->anchor == NULL) { 4204 4205 if (r->rule_flag & PFRULE_ONCE) { 4206 u_int32_t rule_flag; 4207 4208 rule_flag = r->rule_flag; 4209 if (((rule_flag & PFRULE_EXPIRED) == 0) && 4210 atomic_cas_uint(&r->rule_flag, rule_flag, 4211 rule_flag | PFRULE_EXPIRED) == rule_flag) { 4212 r->exptime = gettime(); 4213 } else { 4214 r = TAILQ_NEXT(r, entries); 4215 continue; 4216 } 4217 } 4218 4219 if (r->action == PF_MATCH) { 4220 if ((ctx->ri = pool_get(&pf_rule_item_pl, 4221 PR_NOWAIT)) == NULL) { 4222 REASON_SET(&ctx->reason, PFRES_MEMORY); 4223 return (PF_TEST_FAIL); 4224 } 4225 ctx->ri->r = r; 4226 /* order is irrelevant */ 4227 SLIST_INSERT_HEAD(&ctx->rules, ctx->ri, entry); 4228 ctx->ri = NULL; 4229 pf_rule_to_actions(r, &ctx->act); 4230 if (r->rule_flag & PFRULE_AFTO) 4231 ctx->pd->naf = r->naf; 4232 if (pf_get_transaddr(r, ctx->pd, ctx->sns, 4233 &ctx->nr) == -1) { 4234 REASON_SET(&ctx->reason, 4235 PFRES_TRANSLATE); 4236 return (PF_TEST_FAIL); 4237 } 4238 #if NPFLOG > 0 4239 if (r->log) { 4240 REASON_SET(&ctx->reason, PFRES_MATCH); 4241 pflog_packet(ctx->pd, ctx->reason, r, 4242 ctx->a, ruleset, NULL); 4243 } 4244 #endif /* NPFLOG > 0 */ 4245 } else { 4246 /* 4247 * found matching r 4248 */ 4249 *ctx->rm = r; 4250 /* 4251 * anchor, with ruleset, where r belongs to 4252 */ 4253 *ctx->am = ctx->a; 4254 /* 4255 * ruleset where r belongs to 4256 */ 4257 *ctx->rsm = ruleset; 4258 /* 4259 * ruleset, where anchor belongs to. 4260 */ 4261 ctx->arsm = ctx->aruleset; 4262 } 4263 4264 #if NPFLOG > 0 4265 if (ctx->act.log & PF_LOG_MATCHES) 4266 pf_log_matches(ctx->pd, r, ctx->a, ruleset, 4267 &ctx->rules); 4268 #endif /* NPFLOG > 0 */ 4269 4270 if (r->quick) 4271 return (PF_TEST_QUICK); 4272 } else { 4273 ctx->a = r; 4274 ctx->aruleset = &r->anchor->ruleset; 4275 if (r->anchor_wildcard) { 4276 RB_FOREACH(child, pf_anchor_node, 4277 &r->anchor->children) { 4278 if (pf_anchor_stack_push(ruleset, r, child, 4279 PF_NEXT_CHILD) != 0) 4280 return (PF_TEST_FAIL); 4281 4282 ruleset = &child->ruleset; 4283 goto enter_ruleset; 4284 next_child: 4285 continue; /* with RB_FOREACH() */ 4286 } 4287 } else { 4288 if (pf_anchor_stack_push(ruleset, r, child, 4289 PF_NEXT_RULE) != 0) 4290 return (PF_TEST_FAIL); 4291 4292 ruleset = &r->anchor->ruleset; 4293 child = NULL; 4294 goto enter_ruleset; 4295 next_rule: 4296 ; 4297 } 4298 } 4299 r = TAILQ_NEXT(r, entries); 4300 } 4301 4302 if (pf_anchor_stack_pop(&ruleset, &r, &child, &target) == 0) { 4303 /* stop if any rule matched within quick anchors. */ 4304 if (r->quick == PF_TEST_QUICK && *ctx->am == r) 4305 return (PF_TEST_QUICK); 4306 4307 switch (target) { 4308 case PF_NEXT_CHILD: 4309 goto next_child; 4310 case PF_NEXT_RULE: 4311 goto next_rule; 4312 default: 4313 panic("%s: unknown jump target", __func__); 4314 } 4315 } 4316 4317 return (PF_TEST_OK); 4318 } 4319 4320 int 4321 pf_test_rule(struct pf_pdesc *pd, struct pf_rule **rm, struct pf_state **sm, 4322 struct pf_rule **am, struct pf_ruleset **rsm, u_short *reason) 4323 { 4324 struct pf_rule *r = NULL; 4325 struct pf_rule *a = NULL; 4326 struct pf_ruleset *ruleset = NULL; 4327 struct pf_state_key *skw = NULL, *sks = NULL; 4328 int rewrite = 0; 4329 u_int16_t virtual_type, virtual_id; 4330 int action = PF_DROP; 4331 struct pf_test_ctx ctx; 4332 int rv; 4333 4334 PF_ASSERT_LOCKED(); 4335 4336 memset(&ctx, 0, sizeof(ctx)); 4337 ctx.pd = pd; 4338 ctx.rm = rm; 4339 ctx.am = am; 4340 ctx.rsm = rsm; 4341 ctx.th = &pd->hdr.tcp; 4342 ctx.act.rtableid = pd->rdomain; 4343 ctx.tag = -1; 4344 SLIST_INIT(&ctx.rules); 4345 4346 if (pd->dir == PF_IN && if_congested()) { 4347 REASON_SET(&ctx.reason, PFRES_CONGEST); 4348 return (PF_DROP); 4349 } 4350 4351 switch (pd->virtual_proto) { 4352 case IPPROTO_ICMP: 4353 ctx.icmptype = pd->hdr.icmp.icmp_type; 4354 ctx.icmpcode = pd->hdr.icmp.icmp_code; 4355 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 4356 &ctx.icmp_dir, &virtual_id, &virtual_type); 4357 if (ctx.icmp_dir == PF_IN) { 4358 pd->osport = pd->nsport = virtual_id; 4359 pd->odport = pd->ndport = virtual_type; 4360 } else { 4361 pd->osport = pd->nsport = virtual_type; 4362 pd->odport = pd->ndport = virtual_id; 4363 } 4364 break; 4365 #ifdef INET6 4366 case IPPROTO_ICMPV6: 4367 ctx.icmptype = pd->hdr.icmp6.icmp6_type; 4368 ctx.icmpcode = pd->hdr.icmp6.icmp6_code; 4369 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 4370 &ctx.icmp_dir, &virtual_id, &virtual_type); 4371 if (ctx.icmp_dir == PF_IN) { 4372 pd->osport = pd->nsport = virtual_id; 4373 pd->odport = pd->ndport = virtual_type; 4374 } else { 4375 pd->osport = pd->nsport = virtual_type; 4376 pd->odport = pd->ndport = virtual_id; 4377 } 4378 break; 4379 #endif /* INET6 */ 4380 } 4381 4382 ruleset = &pf_main_ruleset; 4383 rv = pf_match_rule(&ctx, ruleset); 4384 if (rv == PF_TEST_FAIL) { 4385 /* 4386 * Reason has been set in pf_match_rule() already. 4387 */ 4388 goto cleanup; 4389 } 4390 4391 r = *ctx.rm; /* matching rule */ 4392 a = *ctx.am; /* rule that defines an anchor containing 'r' */ 4393 ruleset = *ctx.rsm;/* ruleset of the anchor defined by the rule 'a' */ 4394 ctx.aruleset = ctx.arsm;/* ruleset of the 'a' rule itself */ 4395 4396 /* apply actions for last matching pass/block rule */ 4397 pf_rule_to_actions(r, &ctx.act); 4398 if (r->rule_flag & PFRULE_AFTO) 4399 pd->naf = r->naf; 4400 if (pf_get_transaddr(r, pd, ctx.sns, &ctx.nr) == -1) { 4401 REASON_SET(&ctx.reason, PFRES_TRANSLATE); 4402 goto cleanup; 4403 } 4404 REASON_SET(&ctx.reason, PFRES_MATCH); 4405 4406 #if NPFLOG > 0 4407 if (r->log) 4408 pflog_packet(pd, ctx.reason, r, a, ruleset, NULL); 4409 if (ctx.act.log & PF_LOG_MATCHES) 4410 pf_log_matches(pd, r, a, ruleset, &ctx.rules); 4411 #endif /* NPFLOG > 0 */ 4412 4413 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 4414 (r->action == PF_DROP) && 4415 ((r->rule_flag & PFRULE_RETURNRST) || 4416 (r->rule_flag & PFRULE_RETURNICMP) || 4417 (r->rule_flag & PFRULE_RETURN))) { 4418 if (pd->proto == IPPROTO_TCP && 4419 ((r->rule_flag & PFRULE_RETURNRST) || 4420 (r->rule_flag & PFRULE_RETURN)) && 4421 !(ctx.th->th_flags & TH_RST)) { 4422 u_int32_t ack = 4423 ntohl(ctx.th->th_seq) + pd->p_len; 4424 4425 if (pf_check_tcp_cksum(pd->m, pd->off, 4426 pd->tot_len - pd->off, pd->af)) 4427 REASON_SET(&ctx.reason, PFRES_PROTCKSUM); 4428 else { 4429 if (ctx.th->th_flags & TH_SYN) 4430 ack++; 4431 if (ctx.th->th_flags & TH_FIN) 4432 ack++; 4433 pf_send_tcp(r, pd->af, pd->dst, 4434 pd->src, ctx.th->th_dport, 4435 ctx.th->th_sport, ntohl(ctx.th->th_ack), 4436 ack, TH_RST|TH_ACK, 0, 0, r->return_ttl, 4437 1, 0, pd->rdomain); 4438 } 4439 } else if ((pd->proto != IPPROTO_ICMP || 4440 ICMP_INFOTYPE(ctx.icmptype)) && pd->af == AF_INET && 4441 r->return_icmp) 4442 pf_send_icmp(pd->m, r->return_icmp >> 8, 4443 r->return_icmp & 255, 0, pd->af, r, pd->rdomain); 4444 else if ((pd->proto != IPPROTO_ICMPV6 || 4445 (ctx.icmptype >= ICMP6_ECHO_REQUEST && 4446 ctx.icmptype != ND_REDIRECT)) && pd->af == AF_INET6 && 4447 r->return_icmp6) 4448 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 4449 r->return_icmp6 & 255, 0, pd->af, r, pd->rdomain); 4450 } 4451 4452 if (r->action == PF_DROP) 4453 goto cleanup; 4454 4455 pf_tag_packet(pd->m, ctx.tag, ctx.act.rtableid); 4456 if (ctx.act.rtableid >= 0 && 4457 rtable_l2(ctx.act.rtableid) != pd->rdomain) 4458 pd->destchg = 1; 4459 4460 if (r->action == PF_PASS && pd->badopts != 0 && ! r->allow_opts) { 4461 REASON_SET(&ctx.reason, PFRES_IPOPTIONS); 4462 #if NPFLOG > 0 4463 pd->pflog |= PF_LOG_FORCE; 4464 #endif /* NPFLOG > 0 */ 4465 DPFPRINTF(LOG_NOTICE, "dropping packet with " 4466 "ip/ipv6 options in pf_test_rule()"); 4467 goto cleanup; 4468 } 4469 4470 if (pd->virtual_proto != PF_VPROTO_FRAGMENT 4471 && !ctx.state_icmp && r->keep_state) { 4472 4473 if (r->rule_flag & PFRULE_SRCTRACK && 4474 pf_insert_src_node(&ctx.sns[PF_SN_NONE], r, PF_SN_NONE, 4475 pd->af, pd->src, NULL, NULL) != 0) { 4476 REASON_SET(&ctx.reason, PFRES_SRCLIMIT); 4477 goto cleanup; 4478 } 4479 4480 if (r->max_states && (r->states_cur >= r->max_states)) { 4481 pf_status.lcounters[LCNT_STATES]++; 4482 REASON_SET(&ctx.reason, PFRES_MAXSTATES); 4483 goto cleanup; 4484 } 4485 4486 action = pf_create_state(pd, r, a, ctx.nr, &skw, &sks, 4487 &rewrite, sm, ctx.tag, &ctx.rules, &ctx.act, ctx.sns); 4488 4489 if (action != PF_PASS) 4490 goto cleanup; 4491 if (sks != skw) { 4492 struct pf_state_key *sk; 4493 4494 if (pd->dir == PF_IN) 4495 sk = sks; 4496 else 4497 sk = skw; 4498 rewrite += pf_translate(pd, 4499 &sk->addr[pd->af == pd->naf ? pd->sidx : pd->didx], 4500 sk->port[pd->af == pd->naf ? pd->sidx : pd->didx], 4501 &sk->addr[pd->af == pd->naf ? pd->didx : pd->sidx], 4502 sk->port[pd->af == pd->naf ? pd->didx : pd->sidx], 4503 virtual_type, ctx.icmp_dir); 4504 } 4505 4506 #ifdef INET6 4507 if (rewrite && skw->af != sks->af) 4508 action = PF_AFRT; 4509 #endif /* INET6 */ 4510 4511 } else { 4512 action = PF_PASS; 4513 4514 while ((ctx.ri = SLIST_FIRST(&ctx.rules))) { 4515 SLIST_REMOVE_HEAD(&ctx.rules, entry); 4516 pool_put(&pf_rule_item_pl, ctx.ri); 4517 } 4518 } 4519 4520 /* copy back packet headers if needed */ 4521 if (rewrite && pd->hdrlen) { 4522 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 4523 } 4524 4525 #if NPFSYNC > 0 4526 if (*sm != NULL && !ISSET((*sm)->state_flags, PFSTATE_NOSYNC) && 4527 pd->dir == PF_OUT && pfsync_is_up()) { 4528 /* 4529 * We want the state created, but we dont 4530 * want to send this in case a partner 4531 * firewall has to know about it to allow 4532 * replies through it. 4533 */ 4534 if (pfsync_defer(*sm, pd->m)) 4535 return (PF_DEFER); 4536 } 4537 #endif /* NPFSYNC > 0 */ 4538 4539 return (action); 4540 4541 cleanup: 4542 while ((ctx.ri = SLIST_FIRST(&ctx.rules))) { 4543 SLIST_REMOVE_HEAD(&ctx.rules, entry); 4544 pool_put(&pf_rule_item_pl, ctx.ri); 4545 } 4546 4547 return (action); 4548 } 4549 4550 static __inline int 4551 pf_create_state(struct pf_pdesc *pd, struct pf_rule *r, struct pf_rule *a, 4552 struct pf_rule *nr, struct pf_state_key **skw, struct pf_state_key **sks, 4553 int *rewrite, struct pf_state **sm, int tag, struct pf_rule_slist *rules, 4554 struct pf_rule_actions *act, struct pf_src_node *sns[PF_SN_MAX]) 4555 { 4556 struct pf_state *st = NULL; 4557 struct tcphdr *th = &pd->hdr.tcp; 4558 u_int16_t mss = tcp_mssdflt; 4559 u_short reason; 4560 u_int i; 4561 4562 st = pool_get(&pf_state_pl, PR_NOWAIT | PR_ZERO); 4563 if (st == NULL) { 4564 REASON_SET(&reason, PFRES_MEMORY); 4565 goto csfailed; 4566 } 4567 st->rule.ptr = r; 4568 st->anchor.ptr = a; 4569 st->natrule.ptr = nr; 4570 if (r->allow_opts) 4571 st->state_flags |= PFSTATE_ALLOWOPTS; 4572 if (r->rule_flag & PFRULE_STATESLOPPY) 4573 st->state_flags |= PFSTATE_SLOPPY; 4574 if (r->rule_flag & PFRULE_PFLOW) 4575 st->state_flags |= PFSTATE_PFLOW; 4576 if (r->rule_flag & PFRULE_NOSYNC) 4577 st->state_flags |= PFSTATE_NOSYNC; 4578 #if NPFLOG > 0 4579 st->log = act->log & PF_LOG_ALL; 4580 #endif /* NPFLOG > 0 */ 4581 st->qid = act->qid; 4582 st->pqid = act->pqid; 4583 st->rtableid[pd->didx] = act->rtableid; 4584 st->rtableid[pd->sidx] = -1; /* return traffic is routed normally */ 4585 st->min_ttl = act->min_ttl; 4586 st->set_tos = act->set_tos; 4587 st->max_mss = act->max_mss; 4588 st->state_flags |= act->flags; 4589 #if NPFSYNC > 0 4590 st->sync_state = PFSYNC_S_NONE; 4591 #endif /* NPFSYNC > 0 */ 4592 st->set_prio[0] = act->set_prio[0]; 4593 st->set_prio[1] = act->set_prio[1]; 4594 st->delay = act->delay; 4595 SLIST_INIT(&st->src_nodes); 4596 4597 /* 4598 * must initialize refcnt, before pf_state_insert() gets called. 4599 * pf_state_inserts() grabs reference for pfsync! 4600 */ 4601 PF_REF_INIT(st->refcnt); 4602 mtx_init(&st->mtx, IPL_NET); 4603 4604 switch (pd->proto) { 4605 case IPPROTO_TCP: 4606 st->src.seqlo = ntohl(th->th_seq); 4607 st->src.seqhi = st->src.seqlo + pd->p_len + 1; 4608 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4609 r->keep_state == PF_STATE_MODULATE) { 4610 /* Generate sequence number modulator */ 4611 st->src.seqdiff = pf_tcp_iss(pd) - st->src.seqlo; 4612 if (st->src.seqdiff == 0) 4613 st->src.seqdiff = 1; 4614 pf_patch_32(pd, &th->th_seq, 4615 htonl(st->src.seqlo + st->src.seqdiff)); 4616 *rewrite = 1; 4617 } else 4618 st->src.seqdiff = 0; 4619 if (th->th_flags & TH_SYN) { 4620 st->src.seqhi++; 4621 st->src.wscale = pf_get_wscale(pd); 4622 } 4623 st->src.max_win = MAX(ntohs(th->th_win), 1); 4624 if (st->src.wscale & PF_WSCALE_MASK) { 4625 /* Remove scale factor from initial window */ 4626 int win = st->src.max_win; 4627 win += 1 << (st->src.wscale & PF_WSCALE_MASK); 4628 st->src.max_win = (win - 1) >> 4629 (st->src.wscale & PF_WSCALE_MASK); 4630 } 4631 if (th->th_flags & TH_FIN) 4632 st->src.seqhi++; 4633 st->dst.seqhi = 1; 4634 st->dst.max_win = 1; 4635 pf_set_protostate(st, PF_PEER_SRC, TCPS_SYN_SENT); 4636 pf_set_protostate(st, PF_PEER_DST, TCPS_CLOSED); 4637 st->timeout = PFTM_TCP_FIRST_PACKET; 4638 pf_status.states_halfopen++; 4639 break; 4640 case IPPROTO_UDP: 4641 pf_set_protostate(st, PF_PEER_SRC, PFUDPS_SINGLE); 4642 pf_set_protostate(st, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 4643 st->timeout = PFTM_UDP_FIRST_PACKET; 4644 break; 4645 case IPPROTO_ICMP: 4646 #ifdef INET6 4647 case IPPROTO_ICMPV6: 4648 #endif /* INET6 */ 4649 st->timeout = PFTM_ICMP_FIRST_PACKET; 4650 break; 4651 default: 4652 pf_set_protostate(st, PF_PEER_SRC, PFOTHERS_SINGLE); 4653 pf_set_protostate(st, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 4654 st->timeout = PFTM_OTHER_FIRST_PACKET; 4655 } 4656 4657 st->creation = getuptime(); 4658 st->expire = getuptime(); 4659 4660 if (pd->proto == IPPROTO_TCP) { 4661 if (st->state_flags & PFSTATE_SCRUB_TCP && 4662 pf_normalize_tcp_init(pd, &st->src)) { 4663 REASON_SET(&reason, PFRES_MEMORY); 4664 goto csfailed; 4665 } 4666 if (st->state_flags & PFSTATE_SCRUB_TCP && st->src.scrub && 4667 pf_normalize_tcp_stateful(pd, &reason, st, 4668 &st->src, &st->dst, rewrite)) { 4669 /* This really shouldn't happen!!! */ 4670 DPFPRINTF(LOG_ERR, 4671 "%s: tcp normalize failed on first pkt", __func__); 4672 goto csfailed; 4673 } 4674 } 4675 st->direction = pd->dir; 4676 4677 if (pf_state_key_setup(pd, skw, sks, act->rtableid)) { 4678 REASON_SET(&reason, PFRES_MEMORY); 4679 goto csfailed; 4680 } 4681 4682 if (pf_set_rt_ifp(st, pd->src, (*skw)->af, sns) != 0) { 4683 REASON_SET(&reason, PFRES_NOROUTE); 4684 goto csfailed; 4685 } 4686 4687 for (i = 0; i < PF_SN_MAX; i++) 4688 if (sns[i] != NULL) { 4689 struct pf_sn_item *sni; 4690 4691 sni = pool_get(&pf_sn_item_pl, PR_NOWAIT); 4692 if (sni == NULL) { 4693 REASON_SET(&reason, PFRES_MEMORY); 4694 goto csfailed; 4695 } 4696 sni->sn = sns[i]; 4697 SLIST_INSERT_HEAD(&st->src_nodes, sni, next); 4698 sni->sn->states++; 4699 } 4700 4701 #if NPFSYNC > 0 4702 pfsync_init_state(st, *skw, *sks, 0); 4703 #endif 4704 4705 if (pf_state_insert(BOUND_IFACE(r, pd->kif), skw, sks, st)) { 4706 *sks = *skw = NULL; 4707 REASON_SET(&reason, PFRES_STATEINS); 4708 goto csfailed; 4709 } else 4710 *sm = st; 4711 4712 /* 4713 * Make state responsible for rules it binds here. 4714 */ 4715 memcpy(&st->match_rules, rules, sizeof(st->match_rules)); 4716 memset(rules, 0, sizeof(*rules)); 4717 STATE_INC_COUNTERS(st); 4718 4719 if (tag > 0) { 4720 pf_tag_ref(tag); 4721 st->tag = tag; 4722 } 4723 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 4724 TH_SYN && r->keep_state == PF_STATE_SYNPROXY && pd->dir == PF_IN) { 4725 int rtid = pd->rdomain; 4726 if (act->rtableid >= 0) 4727 rtid = act->rtableid; 4728 pf_set_protostate(st, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 4729 st->src.seqhi = arc4random(); 4730 /* Find mss option */ 4731 mss = pf_get_mss(pd); 4732 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 4733 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 4734 st->src.mss = mss; 4735 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 4736 th->th_sport, st->src.seqhi, ntohl(th->th_seq) + 1, 4737 TH_SYN|TH_ACK, 0, st->src.mss, 0, 1, 0, pd->rdomain); 4738 REASON_SET(&reason, PFRES_SYNPROXY); 4739 return (PF_SYNPROXY_DROP); 4740 } 4741 4742 return (PF_PASS); 4743 4744 csfailed: 4745 if (st) { 4746 pf_normalize_tcp_cleanup(st); /* safe even w/o init */ 4747 pf_src_tree_remove_state(st); 4748 pool_put(&pf_state_pl, st); 4749 } 4750 4751 for (i = 0; i < PF_SN_MAX; i++) 4752 if (sns[i] != NULL) 4753 pf_remove_src_node(sns[i]); 4754 4755 return (PF_DROP); 4756 } 4757 4758 int 4759 pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport, 4760 struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type, 4761 int icmp_dir) 4762 { 4763 int rewrite = 0; 4764 int afto = pd->af != pd->naf; 4765 4766 if (afto || PF_ANEQ(daddr, pd->dst, pd->af)) 4767 pd->destchg = 1; 4768 4769 switch (pd->proto) { 4770 case IPPROTO_TCP: /* FALLTHROUGH */ 4771 case IPPROTO_UDP: 4772 rewrite += pf_patch_16(pd, pd->sport, sport); 4773 rewrite += pf_patch_16(pd, pd->dport, dport); 4774 break; 4775 4776 case IPPROTO_ICMP: 4777 if (pd->af != AF_INET) 4778 return (0); 4779 4780 #ifdef INET6 4781 if (afto) { 4782 if (pf_translate_icmp_af(pd, AF_INET6, &pd->hdr.icmp)) 4783 return (0); 4784 pd->proto = IPPROTO_ICMPV6; 4785 rewrite = 1; 4786 } 4787 #endif /* INET6 */ 4788 if (virtual_type == htons(ICMP_ECHO)) { 4789 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 4790 rewrite += pf_patch_16(pd, 4791 &pd->hdr.icmp.icmp_id, icmpid); 4792 } 4793 break; 4794 4795 #ifdef INET6 4796 case IPPROTO_ICMPV6: 4797 if (pd->af != AF_INET6) 4798 return (0); 4799 4800 if (afto) { 4801 if (pf_translate_icmp_af(pd, AF_INET, &pd->hdr.icmp6)) 4802 return (0); 4803 pd->proto = IPPROTO_ICMP; 4804 rewrite = 1; 4805 } 4806 if (virtual_type == htons(ICMP6_ECHO_REQUEST)) { 4807 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 4808 rewrite += pf_patch_16(pd, 4809 &pd->hdr.icmp6.icmp6_id, icmpid); 4810 } 4811 break; 4812 #endif /* INET6 */ 4813 } 4814 4815 if (!afto) { 4816 rewrite += pf_translate_a(pd, pd->src, saddr); 4817 rewrite += pf_translate_a(pd, pd->dst, daddr); 4818 } 4819 4820 return (rewrite); 4821 } 4822 4823 int 4824 pf_tcp_track_full(struct pf_pdesc *pd, struct pf_state **stp, u_short *reason, 4825 int *copyback, int reverse) 4826 { 4827 struct tcphdr *th = &pd->hdr.tcp; 4828 struct pf_state_peer *src, *dst; 4829 u_int16_t win = ntohs(th->th_win); 4830 u_int32_t ack, end, data_end, seq, orig_seq; 4831 u_int8_t sws, dws, psrc, pdst; 4832 int ackskew; 4833 4834 if ((pd->dir == (*stp)->direction && !reverse) || 4835 (pd->dir != (*stp)->direction && reverse)) { 4836 src = &(*stp)->src; 4837 dst = &(*stp)->dst; 4838 psrc = PF_PEER_SRC; 4839 pdst = PF_PEER_DST; 4840 } else { 4841 src = &(*stp)->dst; 4842 dst = &(*stp)->src; 4843 psrc = PF_PEER_DST; 4844 pdst = PF_PEER_SRC; 4845 } 4846 4847 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 4848 sws = src->wscale & PF_WSCALE_MASK; 4849 dws = dst->wscale & PF_WSCALE_MASK; 4850 } else 4851 sws = dws = 0; 4852 4853 /* 4854 * Sequence tracking algorithm from Guido van Rooij's paper: 4855 * http://www.madison-gurkha.com/publications/tcp_filtering/ 4856 * tcp_filtering.ps 4857 */ 4858 4859 orig_seq = seq = ntohl(th->th_seq); 4860 if (src->seqlo == 0) { 4861 /* First packet from this end. Set its state */ 4862 4863 if (((*stp)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 4864 src->scrub == NULL) { 4865 if (pf_normalize_tcp_init(pd, src)) { 4866 REASON_SET(reason, PFRES_MEMORY); 4867 return (PF_DROP); 4868 } 4869 } 4870 4871 /* Deferred generation of sequence number modulator */ 4872 if (dst->seqdiff && !src->seqdiff) { 4873 /* use random iss for the TCP server */ 4874 while ((src->seqdiff = arc4random() - seq) == 0) 4875 continue; 4876 ack = ntohl(th->th_ack) - dst->seqdiff; 4877 pf_patch_32(pd, &th->th_seq, htonl(seq + src->seqdiff)); 4878 pf_patch_32(pd, &th->th_ack, htonl(ack)); 4879 *copyback = 1; 4880 } else { 4881 ack = ntohl(th->th_ack); 4882 } 4883 4884 end = seq + pd->p_len; 4885 if (th->th_flags & TH_SYN) { 4886 end++; 4887 if (dst->wscale & PF_WSCALE_FLAG) { 4888 src->wscale = pf_get_wscale(pd); 4889 if (src->wscale & PF_WSCALE_FLAG) { 4890 /* Remove scale factor from initial 4891 * window */ 4892 sws = src->wscale & PF_WSCALE_MASK; 4893 win = ((u_int32_t)win + (1 << sws) - 1) 4894 >> sws; 4895 dws = dst->wscale & PF_WSCALE_MASK; 4896 } else { 4897 /* fixup other window */ 4898 dst->max_win = MIN(TCP_MAXWIN, 4899 (u_int32_t)dst->max_win << 4900 (dst->wscale & PF_WSCALE_MASK)); 4901 /* in case of a retrans SYN|ACK */ 4902 dst->wscale = 0; 4903 } 4904 } 4905 } 4906 data_end = end; 4907 if (th->th_flags & TH_FIN) 4908 end++; 4909 4910 src->seqlo = seq; 4911 if (src->state < TCPS_SYN_SENT) 4912 pf_set_protostate(*stp, psrc, TCPS_SYN_SENT); 4913 4914 /* 4915 * May need to slide the window (seqhi may have been set by 4916 * the crappy stack check or if we picked up the connection 4917 * after establishment) 4918 */ 4919 if (src->seqhi == 1 || 4920 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 4921 src->seqhi = end + MAX(1, dst->max_win << dws); 4922 if (win > src->max_win) 4923 src->max_win = win; 4924 4925 } else { 4926 ack = ntohl(th->th_ack) - dst->seqdiff; 4927 if (src->seqdiff) { 4928 /* Modulate sequence numbers */ 4929 pf_patch_32(pd, &th->th_seq, htonl(seq + src->seqdiff)); 4930 pf_patch_32(pd, &th->th_ack, htonl(ack)); 4931 *copyback = 1; 4932 } 4933 end = seq + pd->p_len; 4934 if (th->th_flags & TH_SYN) 4935 end++; 4936 data_end = end; 4937 if (th->th_flags & TH_FIN) 4938 end++; 4939 } 4940 4941 if ((th->th_flags & TH_ACK) == 0) { 4942 /* Let it pass through the ack skew check */ 4943 ack = dst->seqlo; 4944 } else if ((ack == 0 && 4945 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 4946 /* broken tcp stacks do not set ack */ 4947 (dst->state < TCPS_SYN_SENT)) { 4948 /* 4949 * Many stacks (ours included) will set the ACK number in an 4950 * FIN|ACK if the SYN times out -- no sequence to ACK. 4951 */ 4952 ack = dst->seqlo; 4953 } 4954 4955 if (seq == end) { 4956 /* Ease sequencing restrictions on no data packets */ 4957 seq = src->seqlo; 4958 data_end = end = seq; 4959 } 4960 4961 ackskew = dst->seqlo - ack; 4962 4963 4964 /* 4965 * Need to demodulate the sequence numbers in any TCP SACK options 4966 * (Selective ACK). We could optionally validate the SACK values 4967 * against the current ACK window, either forwards or backwards, but 4968 * I'm not confident that SACK has been implemented properly 4969 * everywhere. It wouldn't surprise me if several stacks accidently 4970 * SACK too far backwards of previously ACKed data. There really aren't 4971 * any security implications of bad SACKing unless the target stack 4972 * doesn't validate the option length correctly. Someone trying to 4973 * spoof into a TCP connection won't bother blindly sending SACK 4974 * options anyway. 4975 */ 4976 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 4977 if (pf_modulate_sack(pd, dst)) 4978 *copyback = 1; 4979 } 4980 4981 4982 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 4983 if (SEQ_GEQ(src->seqhi, data_end) && 4984 /* Last octet inside other's window space */ 4985 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 4986 /* Retrans: not more than one window back */ 4987 (ackskew >= -MAXACKWINDOW) && 4988 /* Acking not more than one reassembled fragment backwards */ 4989 (ackskew <= (MAXACKWINDOW << sws)) && 4990 /* Acking not more than one window forward */ 4991 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 4992 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 4993 /* Require an exact/+1 sequence match on resets when possible */ 4994 4995 if (dst->scrub || src->scrub) { 4996 if (pf_normalize_tcp_stateful(pd, reason, *stp, src, 4997 dst, copyback)) 4998 return (PF_DROP); 4999 } 5000 5001 /* update max window */ 5002 if (src->max_win < win) 5003 src->max_win = win; 5004 /* synchronize sequencing */ 5005 if (SEQ_GT(end, src->seqlo)) 5006 src->seqlo = end; 5007 /* slide the window of what the other end can send */ 5008 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5009 dst->seqhi = ack + MAX((win << sws), 1); 5010 5011 /* update states */ 5012 if (th->th_flags & TH_SYN) 5013 if (src->state < TCPS_SYN_SENT) 5014 pf_set_protostate(*stp, psrc, TCPS_SYN_SENT); 5015 if (th->th_flags & TH_FIN) 5016 if (src->state < TCPS_CLOSING) 5017 pf_set_protostate(*stp, psrc, TCPS_CLOSING); 5018 if (th->th_flags & TH_ACK) { 5019 if (dst->state == TCPS_SYN_SENT) { 5020 pf_set_protostate(*stp, pdst, 5021 TCPS_ESTABLISHED); 5022 if (src->state == TCPS_ESTABLISHED && 5023 !SLIST_EMPTY(&(*stp)->src_nodes) && 5024 pf_src_connlimit(stp)) { 5025 REASON_SET(reason, PFRES_SRCLIMIT); 5026 return (PF_DROP); 5027 } 5028 } else if (dst->state == TCPS_CLOSING) 5029 pf_set_protostate(*stp, pdst, 5030 TCPS_FIN_WAIT_2); 5031 } 5032 if (th->th_flags & TH_RST) 5033 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_TIME_WAIT); 5034 5035 /* update expire time */ 5036 (*stp)->expire = getuptime(); 5037 if (src->state >= TCPS_FIN_WAIT_2 && 5038 dst->state >= TCPS_FIN_WAIT_2) 5039 (*stp)->timeout = PFTM_TCP_CLOSED; 5040 else if (src->state >= TCPS_CLOSING && 5041 dst->state >= TCPS_CLOSING) 5042 (*stp)->timeout = PFTM_TCP_FIN_WAIT; 5043 else if (src->state < TCPS_ESTABLISHED || 5044 dst->state < TCPS_ESTABLISHED) 5045 (*stp)->timeout = PFTM_TCP_OPENING; 5046 else if (src->state >= TCPS_CLOSING || 5047 dst->state >= TCPS_CLOSING) 5048 (*stp)->timeout = PFTM_TCP_CLOSING; 5049 else 5050 (*stp)->timeout = PFTM_TCP_ESTABLISHED; 5051 5052 /* Fall through to PASS packet */ 5053 } else if ((dst->state < TCPS_SYN_SENT || 5054 dst->state >= TCPS_FIN_WAIT_2 || 5055 src->state >= TCPS_FIN_WAIT_2) && 5056 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 5057 /* Within a window forward of the originating packet */ 5058 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5059 /* Within a window backward of the originating packet */ 5060 5061 /* 5062 * This currently handles three situations: 5063 * 1) Stupid stacks will shotgun SYNs before their peer 5064 * replies. 5065 * 2) When PF catches an already established stream (the 5066 * firewall rebooted, the state table was flushed, routes 5067 * changed...) 5068 * 3) Packets get funky immediately after the connection 5069 * closes (this should catch Solaris spurious ACK|FINs 5070 * that web servers like to spew after a close) 5071 * 5072 * This must be a little more careful than the above code 5073 * since packet floods will also be caught here. We don't 5074 * update the TTL here to mitigate the damage of a packet 5075 * flood and so the same code can handle awkward establishment 5076 * and a loosened connection close. 5077 * In the establishment case, a correct peer response will 5078 * validate the connection, go through the normal state code 5079 * and keep updating the state TTL. 5080 */ 5081 5082 if (pf_status.debug >= LOG_NOTICE) { 5083 log(LOG_NOTICE, "pf: loose state match: "); 5084 pf_print_state(*stp); 5085 pf_print_flags(th->th_flags); 5086 addlog(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5087 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5088 pd->p_len, ackskew, (*stp)->packets[0], 5089 (*stp)->packets[1], 5090 pd->dir == PF_IN ? "in" : "out", 5091 pd->dir == (*stp)->direction ? "fwd" : "rev"); 5092 } 5093 5094 if (dst->scrub || src->scrub) { 5095 if (pf_normalize_tcp_stateful(pd, reason, *stp, src, 5096 dst, copyback)) 5097 return (PF_DROP); 5098 } 5099 5100 /* update max window */ 5101 if (src->max_win < win) 5102 src->max_win = win; 5103 /* synchronize sequencing */ 5104 if (SEQ_GT(end, src->seqlo)) 5105 src->seqlo = end; 5106 /* slide the window of what the other end can send */ 5107 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5108 dst->seqhi = ack + MAX((win << sws), 1); 5109 5110 /* 5111 * Cannot set dst->seqhi here since this could be a shotgunned 5112 * SYN and not an already established connection. 5113 */ 5114 if (th->th_flags & TH_FIN) 5115 if (src->state < TCPS_CLOSING) 5116 pf_set_protostate(*stp, psrc, TCPS_CLOSING); 5117 if (th->th_flags & TH_RST) 5118 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_TIME_WAIT); 5119 5120 /* Fall through to PASS packet */ 5121 } else { 5122 if ((*stp)->dst.state == TCPS_SYN_SENT && 5123 (*stp)->src.state == TCPS_SYN_SENT) { 5124 /* Send RST for state mismatches during handshake */ 5125 if (!(th->th_flags & TH_RST)) 5126 pf_send_tcp((*stp)->rule.ptr, pd->af, 5127 pd->dst, pd->src, th->th_dport, 5128 th->th_sport, ntohl(th->th_ack), 0, 5129 TH_RST, 0, 0, 5130 (*stp)->rule.ptr->return_ttl, 1, 0, 5131 pd->rdomain); 5132 src->seqlo = 0; 5133 src->seqhi = 1; 5134 src->max_win = 1; 5135 } else if (pf_status.debug >= LOG_NOTICE) { 5136 log(LOG_NOTICE, "pf: BAD state: "); 5137 pf_print_state(*stp); 5138 pf_print_flags(th->th_flags); 5139 addlog(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5140 "pkts=%llu:%llu dir=%s,%s\n", 5141 seq, orig_seq, ack, pd->p_len, ackskew, 5142 (*stp)->packets[0], (*stp)->packets[1], 5143 pd->dir == PF_IN ? "in" : "out", 5144 pd->dir == (*stp)->direction ? "fwd" : "rev"); 5145 addlog("pf: State failure on: %c %c %c %c | %c %c\n", 5146 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 5147 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5148 ' ': '2', 5149 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5150 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5151 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ? 5152 ' ' :'5', 5153 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5154 } 5155 REASON_SET(reason, PFRES_BADSTATE); 5156 return (PF_DROP); 5157 } 5158 5159 return (PF_PASS); 5160 } 5161 5162 int 5163 pf_tcp_track_sloppy(struct pf_pdesc *pd, struct pf_state **stp, 5164 u_short *reason) 5165 { 5166 struct tcphdr *th = &pd->hdr.tcp; 5167 struct pf_state_peer *src, *dst; 5168 u_int8_t psrc, pdst; 5169 5170 if (pd->dir == (*stp)->direction) { 5171 src = &(*stp)->src; 5172 dst = &(*stp)->dst; 5173 psrc = PF_PEER_SRC; 5174 pdst = PF_PEER_DST; 5175 } else { 5176 src = &(*stp)->dst; 5177 dst = &(*stp)->src; 5178 psrc = PF_PEER_DST; 5179 pdst = PF_PEER_SRC; 5180 } 5181 5182 if (th->th_flags & TH_SYN) 5183 if (src->state < TCPS_SYN_SENT) 5184 pf_set_protostate(*stp, psrc, TCPS_SYN_SENT); 5185 if (th->th_flags & TH_FIN) 5186 if (src->state < TCPS_CLOSING) 5187 pf_set_protostate(*stp, psrc, TCPS_CLOSING); 5188 if (th->th_flags & TH_ACK) { 5189 if (dst->state == TCPS_SYN_SENT) { 5190 pf_set_protostate(*stp, pdst, TCPS_ESTABLISHED); 5191 if (src->state == TCPS_ESTABLISHED && 5192 !SLIST_EMPTY(&(*stp)->src_nodes) && 5193 pf_src_connlimit(stp)) { 5194 REASON_SET(reason, PFRES_SRCLIMIT); 5195 return (PF_DROP); 5196 } 5197 } else if (dst->state == TCPS_CLOSING) { 5198 pf_set_protostate(*stp, pdst, TCPS_FIN_WAIT_2); 5199 } else if (src->state == TCPS_SYN_SENT && 5200 dst->state < TCPS_SYN_SENT) { 5201 /* 5202 * Handle a special sloppy case where we only see one 5203 * half of the connection. If there is a ACK after 5204 * the initial SYN without ever seeing a packet from 5205 * the destination, set the connection to established. 5206 */ 5207 pf_set_protostate(*stp, PF_PEER_BOTH, 5208 TCPS_ESTABLISHED); 5209 if (!SLIST_EMPTY(&(*stp)->src_nodes) && 5210 pf_src_connlimit(stp)) { 5211 REASON_SET(reason, PFRES_SRCLIMIT); 5212 return (PF_DROP); 5213 } 5214 } else if (src->state == TCPS_CLOSING && 5215 dst->state == TCPS_ESTABLISHED && 5216 dst->seqlo == 0) { 5217 /* 5218 * Handle the closing of half connections where we 5219 * don't see the full bidirectional FIN/ACK+ACK 5220 * handshake. 5221 */ 5222 pf_set_protostate(*stp, pdst, TCPS_CLOSING); 5223 } 5224 } 5225 if (th->th_flags & TH_RST) 5226 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_TIME_WAIT); 5227 5228 /* update expire time */ 5229 (*stp)->expire = getuptime(); 5230 if (src->state >= TCPS_FIN_WAIT_2 && 5231 dst->state >= TCPS_FIN_WAIT_2) 5232 (*stp)->timeout = PFTM_TCP_CLOSED; 5233 else if (src->state >= TCPS_CLOSING && 5234 dst->state >= TCPS_CLOSING) 5235 (*stp)->timeout = PFTM_TCP_FIN_WAIT; 5236 else if (src->state < TCPS_ESTABLISHED || 5237 dst->state < TCPS_ESTABLISHED) 5238 (*stp)->timeout = PFTM_TCP_OPENING; 5239 else if (src->state >= TCPS_CLOSING || 5240 dst->state >= TCPS_CLOSING) 5241 (*stp)->timeout = PFTM_TCP_CLOSING; 5242 else 5243 (*stp)->timeout = PFTM_TCP_ESTABLISHED; 5244 5245 return (PF_PASS); 5246 } 5247 5248 static __inline int 5249 pf_synproxy(struct pf_pdesc *pd, struct pf_state **stp, u_short *reason) 5250 { 5251 struct pf_state_key *sk = (*stp)->key[pd->didx]; 5252 5253 if ((*stp)->src.state == PF_TCPS_PROXY_SRC) { 5254 struct tcphdr *th = &pd->hdr.tcp; 5255 5256 if (pd->dir != (*stp)->direction) { 5257 REASON_SET(reason, PFRES_SYNPROXY); 5258 return (PF_SYNPROXY_DROP); 5259 } 5260 if (th->th_flags & TH_SYN) { 5261 if (ntohl(th->th_seq) != (*stp)->src.seqlo) { 5262 REASON_SET(reason, PFRES_SYNPROXY); 5263 return (PF_DROP); 5264 } 5265 pf_send_tcp((*stp)->rule.ptr, pd->af, pd->dst, 5266 pd->src, th->th_dport, th->th_sport, 5267 (*stp)->src.seqhi, ntohl(th->th_seq) + 1, 5268 TH_SYN|TH_ACK, 0, (*stp)->src.mss, 0, 1, 5269 0, pd->rdomain); 5270 REASON_SET(reason, PFRES_SYNPROXY); 5271 return (PF_SYNPROXY_DROP); 5272 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 5273 (ntohl(th->th_ack) != (*stp)->src.seqhi + 1) || 5274 (ntohl(th->th_seq) != (*stp)->src.seqlo + 1)) { 5275 REASON_SET(reason, PFRES_SYNPROXY); 5276 return (PF_DROP); 5277 } else if (!SLIST_EMPTY(&(*stp)->src_nodes) && 5278 pf_src_connlimit(stp)) { 5279 REASON_SET(reason, PFRES_SRCLIMIT); 5280 return (PF_DROP); 5281 } else 5282 pf_set_protostate(*stp, PF_PEER_SRC, 5283 PF_TCPS_PROXY_DST); 5284 } 5285 if ((*stp)->src.state == PF_TCPS_PROXY_DST) { 5286 struct tcphdr *th = &pd->hdr.tcp; 5287 5288 if (pd->dir == (*stp)->direction) { 5289 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 5290 (ntohl(th->th_ack) != (*stp)->src.seqhi + 1) || 5291 (ntohl(th->th_seq) != (*stp)->src.seqlo + 1)) { 5292 REASON_SET(reason, PFRES_SYNPROXY); 5293 return (PF_DROP); 5294 } 5295 (*stp)->src.max_win = MAX(ntohs(th->th_win), 1); 5296 if ((*stp)->dst.seqhi == 1) 5297 (*stp)->dst.seqhi = arc4random(); 5298 pf_send_tcp((*stp)->rule.ptr, pd->af, 5299 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5300 sk->port[pd->sidx], sk->port[pd->didx], 5301 (*stp)->dst.seqhi, 0, TH_SYN, 0, 5302 (*stp)->src.mss, 0, 0, (*stp)->tag, 5303 sk->rdomain); 5304 REASON_SET(reason, PFRES_SYNPROXY); 5305 return (PF_SYNPROXY_DROP); 5306 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 5307 (TH_SYN|TH_ACK)) || 5308 (ntohl(th->th_ack) != (*stp)->dst.seqhi + 1)) { 5309 REASON_SET(reason, PFRES_SYNPROXY); 5310 return (PF_DROP); 5311 } else { 5312 (*stp)->dst.max_win = MAX(ntohs(th->th_win), 1); 5313 (*stp)->dst.seqlo = ntohl(th->th_seq); 5314 pf_send_tcp((*stp)->rule.ptr, pd->af, pd->dst, 5315 pd->src, th->th_dport, th->th_sport, 5316 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 5317 TH_ACK, (*stp)->src.max_win, 0, 0, 0, 5318 (*stp)->tag, pd->rdomain); 5319 pf_send_tcp((*stp)->rule.ptr, pd->af, 5320 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5321 sk->port[pd->sidx], sk->port[pd->didx], 5322 (*stp)->src.seqhi + 1, (*stp)->src.seqlo + 1, 5323 TH_ACK, (*stp)->dst.max_win, 0, 0, 1, 5324 0, sk->rdomain); 5325 (*stp)->src.seqdiff = (*stp)->dst.seqhi - 5326 (*stp)->src.seqlo; 5327 (*stp)->dst.seqdiff = (*stp)->src.seqhi - 5328 (*stp)->dst.seqlo; 5329 (*stp)->src.seqhi = (*stp)->src.seqlo + 5330 (*stp)->dst.max_win; 5331 (*stp)->dst.seqhi = (*stp)->dst.seqlo + 5332 (*stp)->src.max_win; 5333 (*stp)->src.wscale = (*stp)->dst.wscale = 0; 5334 pf_set_protostate(*stp, PF_PEER_BOTH, 5335 TCPS_ESTABLISHED); 5336 REASON_SET(reason, PFRES_SYNPROXY); 5337 return (PF_SYNPROXY_DROP); 5338 } 5339 } 5340 return (PF_PASS); 5341 } 5342 5343 int 5344 pf_test_state(struct pf_pdesc *pd, struct pf_state **stp, u_short *reason) 5345 { 5346 int copyback = 0; 5347 struct pf_state_peer *src, *dst; 5348 int action; 5349 struct inpcb *inp = pd->m->m_pkthdr.pf.inp; 5350 u_int8_t psrc, pdst; 5351 5352 action = PF_PASS; 5353 if (pd->dir == (*stp)->direction) { 5354 src = &(*stp)->src; 5355 dst = &(*stp)->dst; 5356 psrc = PF_PEER_SRC; 5357 pdst = PF_PEER_DST; 5358 } else { 5359 src = &(*stp)->dst; 5360 dst = &(*stp)->src; 5361 psrc = PF_PEER_DST; 5362 pdst = PF_PEER_SRC; 5363 } 5364 5365 switch (pd->virtual_proto) { 5366 case IPPROTO_TCP: 5367 if ((action = pf_synproxy(pd, stp, reason)) != PF_PASS) 5368 return (action); 5369 if ((pd->hdr.tcp.th_flags & (TH_SYN|TH_ACK)) == TH_SYN) { 5370 5371 if (dst->state >= TCPS_FIN_WAIT_2 && 5372 src->state >= TCPS_FIN_WAIT_2) { 5373 if (pf_status.debug >= LOG_NOTICE) { 5374 log(LOG_NOTICE, "pf: state reuse "); 5375 pf_print_state(*stp); 5376 pf_print_flags(pd->hdr.tcp.th_flags); 5377 addlog("\n"); 5378 } 5379 /* XXX make sure it's the same direction ?? */ 5380 (*stp)->timeout = PFTM_PURGE; 5381 pf_state_unref(*stp); 5382 *stp = NULL; 5383 pf_mbuf_link_inpcb(pd->m, inp); 5384 return (PF_DROP); 5385 } else if (dst->state >= TCPS_ESTABLISHED && 5386 src->state >= TCPS_ESTABLISHED) { 5387 /* 5388 * SYN matches existing state??? 5389 * Typically happens when sender boots up after 5390 * sudden panic. Certain protocols (NFSv3) are 5391 * always using same port numbers. Challenge 5392 * ACK enables all parties (firewall and peers) 5393 * to get in sync again. 5394 */ 5395 pf_send_challenge_ack(pd, *stp, src, dst); 5396 return (PF_DROP); 5397 } 5398 } 5399 5400 if ((*stp)->state_flags & PFSTATE_SLOPPY) { 5401 if (pf_tcp_track_sloppy(pd, stp, reason) == PF_DROP) 5402 return (PF_DROP); 5403 } else { 5404 if (pf_tcp_track_full(pd, stp, reason, ©back, 5405 PF_REVERSED_KEY((*stp)->key, pd->af)) == PF_DROP) 5406 return (PF_DROP); 5407 } 5408 break; 5409 case IPPROTO_UDP: 5410 /* update states */ 5411 if (src->state < PFUDPS_SINGLE) 5412 pf_set_protostate(*stp, psrc, PFUDPS_SINGLE); 5413 if (dst->state == PFUDPS_SINGLE) 5414 pf_set_protostate(*stp, pdst, PFUDPS_MULTIPLE); 5415 5416 /* update expire time */ 5417 (*stp)->expire = getuptime(); 5418 if (src->state == PFUDPS_MULTIPLE && 5419 dst->state == PFUDPS_MULTIPLE) 5420 (*stp)->timeout = PFTM_UDP_MULTIPLE; 5421 else 5422 (*stp)->timeout = PFTM_UDP_SINGLE; 5423 break; 5424 default: 5425 /* update states */ 5426 if (src->state < PFOTHERS_SINGLE) 5427 pf_set_protostate(*stp, psrc, PFOTHERS_SINGLE); 5428 if (dst->state == PFOTHERS_SINGLE) 5429 pf_set_protostate(*stp, pdst, PFOTHERS_MULTIPLE); 5430 5431 /* update expire time */ 5432 (*stp)->expire = getuptime(); 5433 if (src->state == PFOTHERS_MULTIPLE && 5434 dst->state == PFOTHERS_MULTIPLE) 5435 (*stp)->timeout = PFTM_OTHER_MULTIPLE; 5436 else 5437 (*stp)->timeout = PFTM_OTHER_SINGLE; 5438 break; 5439 } 5440 5441 /* translate source/destination address, if necessary */ 5442 if ((*stp)->key[PF_SK_WIRE] != (*stp)->key[PF_SK_STACK]) { 5443 struct pf_state_key *nk; 5444 int afto, sidx, didx; 5445 5446 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5447 nk = (*stp)->key[pd->sidx]; 5448 else 5449 nk = (*stp)->key[pd->didx]; 5450 5451 afto = pd->af != nk->af; 5452 sidx = afto ? pd->didx : pd->sidx; 5453 didx = afto ? pd->sidx : pd->didx; 5454 5455 #ifdef INET6 5456 if (afto) { 5457 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], nk->af); 5458 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], nk->af); 5459 pd->naf = nk->af; 5460 action = PF_AFRT; 5461 } 5462 #endif /* INET6 */ 5463 5464 if (!afto) 5465 pf_translate_a(pd, pd->src, &nk->addr[sidx]); 5466 5467 if (pd->sport != NULL) 5468 pf_patch_16(pd, pd->sport, nk->port[sidx]); 5469 5470 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 5471 pd->rdomain != nk->rdomain) 5472 pd->destchg = 1; 5473 5474 if (!afto) 5475 pf_translate_a(pd, pd->dst, &nk->addr[didx]); 5476 5477 if (pd->dport != NULL) 5478 pf_patch_16(pd, pd->dport, nk->port[didx]); 5479 5480 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5481 copyback = 1; 5482 } 5483 5484 if (copyback && pd->hdrlen > 0) { 5485 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 5486 } 5487 5488 return (action); 5489 } 5490 5491 int 5492 pf_icmp_state_lookup(struct pf_pdesc *pd, struct pf_state_key_cmp *key, 5493 struct pf_state **stp, u_int16_t icmpid, u_int16_t type, 5494 int icmp_dir, int *iidx, int multi, int inner) 5495 { 5496 int direction, action; 5497 5498 key->af = pd->af; 5499 key->proto = pd->proto; 5500 key->rdomain = pd->rdomain; 5501 if (icmp_dir == PF_IN) { 5502 *iidx = pd->sidx; 5503 key->port[pd->sidx] = icmpid; 5504 key->port[pd->didx] = type; 5505 } else { 5506 *iidx = pd->didx; 5507 key->port[pd->sidx] = type; 5508 key->port[pd->didx] = icmpid; 5509 } 5510 5511 if (pf_state_key_addr_setup(pd, key, pd->sidx, pd->src, pd->didx, 5512 pd->dst, pd->af, multi)) 5513 return (PF_DROP); 5514 5515 key->hash = pf_pkt_hash(key->af, key->proto, 5516 &key->addr[0], &key->addr[1], 0, 0); 5517 5518 action = pf_find_state(pd, key, stp); 5519 if (action != PF_MATCH) 5520 return (action); 5521 5522 if ((*stp)->state_flags & PFSTATE_SLOPPY) 5523 return (-1); 5524 5525 /* Is this ICMP message flowing in right direction? */ 5526 if ((*stp)->key[PF_SK_WIRE]->af != (*stp)->key[PF_SK_STACK]->af) 5527 direction = (pd->af == (*stp)->key[PF_SK_WIRE]->af) ? 5528 PF_IN : PF_OUT; 5529 else 5530 direction = (*stp)->direction; 5531 if ((((!inner && direction == pd->dir) || 5532 (inner && direction != pd->dir)) ? 5533 PF_IN : PF_OUT) != icmp_dir) { 5534 if (pf_status.debug >= LOG_NOTICE) { 5535 log(LOG_NOTICE, 5536 "pf: icmp type %d in wrong direction (%d): ", 5537 ntohs(type), icmp_dir); 5538 pf_print_state(*stp); 5539 addlog("\n"); 5540 } 5541 return (PF_DROP); 5542 } 5543 return (-1); 5544 } 5545 5546 int 5547 pf_test_state_icmp(struct pf_pdesc *pd, struct pf_state **stp, 5548 u_short *reason) 5549 { 5550 u_int16_t virtual_id, virtual_type; 5551 u_int8_t icmptype, icmpcode; 5552 int icmp_dir, iidx, ret, copyback = 0; 5553 5554 struct pf_state_key_cmp key; 5555 5556 switch (pd->proto) { 5557 case IPPROTO_ICMP: 5558 icmptype = pd->hdr.icmp.icmp_type; 5559 icmpcode = pd->hdr.icmp.icmp_code; 5560 break; 5561 #ifdef INET6 5562 case IPPROTO_ICMPV6: 5563 icmptype = pd->hdr.icmp6.icmp6_type; 5564 icmpcode = pd->hdr.icmp6.icmp6_code; 5565 break; 5566 #endif /* INET6 */ 5567 default: 5568 panic("unhandled proto %d", pd->proto); 5569 } 5570 5571 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &virtual_id, 5572 &virtual_type) == 0) { 5573 /* 5574 * ICMP query/reply message not related to a TCP/UDP packet. 5575 * Search for an ICMP state. 5576 */ 5577 ret = pf_icmp_state_lookup(pd, &key, stp, 5578 virtual_id, virtual_type, icmp_dir, &iidx, 5579 0, 0); 5580 /* IPv6? try matching a multicast address */ 5581 if (ret == PF_DROP && pd->af == AF_INET6 && icmp_dir == PF_OUT) 5582 ret = pf_icmp_state_lookup(pd, &key, stp, virtual_id, 5583 virtual_type, icmp_dir, &iidx, 1, 0); 5584 if (ret >= 0) 5585 return (ret); 5586 5587 (*stp)->expire = getuptime(); 5588 (*stp)->timeout = PFTM_ICMP_ERROR_REPLY; 5589 5590 /* translate source/destination address, if necessary */ 5591 if ((*stp)->key[PF_SK_WIRE] != (*stp)->key[PF_SK_STACK]) { 5592 struct pf_state_key *nk; 5593 int afto, sidx, didx; 5594 5595 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5596 nk = (*stp)->key[pd->sidx]; 5597 else 5598 nk = (*stp)->key[pd->didx]; 5599 5600 afto = pd->af != nk->af; 5601 sidx = afto ? pd->didx : pd->sidx; 5602 didx = afto ? pd->sidx : pd->didx; 5603 iidx = afto ? !iidx : iidx; 5604 #ifdef INET6 5605 if (afto) { 5606 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], 5607 nk->af); 5608 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], 5609 nk->af); 5610 pd->naf = nk->af; 5611 } 5612 #endif /* INET6 */ 5613 if (!afto) { 5614 pf_translate_a(pd, pd->src, &nk->addr[sidx]); 5615 pf_translate_a(pd, pd->dst, &nk->addr[didx]); 5616 } 5617 5618 if (pd->rdomain != nk->rdomain) 5619 pd->destchg = 1; 5620 if (!afto && PF_ANEQ(pd->dst, 5621 &nk->addr[didx], pd->af)) 5622 pd->destchg = 1; 5623 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5624 5625 switch (pd->af) { 5626 case AF_INET: 5627 #ifdef INET6 5628 if (afto) { 5629 if (pf_translate_icmp_af(pd, AF_INET6, 5630 &pd->hdr.icmp)) 5631 return (PF_DROP); 5632 pd->proto = IPPROTO_ICMPV6; 5633 } 5634 #endif /* INET6 */ 5635 pf_patch_16(pd, 5636 &pd->hdr.icmp.icmp_id, nk->port[iidx]); 5637 5638 m_copyback(pd->m, pd->off, ICMP_MINLEN, 5639 &pd->hdr.icmp, M_NOWAIT); 5640 copyback = 1; 5641 break; 5642 #ifdef INET6 5643 case AF_INET6: 5644 if (afto) { 5645 if (pf_translate_icmp_af(pd, AF_INET, 5646 &pd->hdr.icmp6)) 5647 return (PF_DROP); 5648 pd->proto = IPPROTO_ICMP; 5649 } 5650 5651 pf_patch_16(pd, 5652 &pd->hdr.icmp6.icmp6_id, nk->port[iidx]); 5653 5654 m_copyback(pd->m, pd->off, 5655 sizeof(struct icmp6_hdr), &pd->hdr.icmp6, 5656 M_NOWAIT); 5657 copyback = 1; 5658 break; 5659 #endif /* INET6 */ 5660 } 5661 #ifdef INET6 5662 if (afto) 5663 return (PF_AFRT); 5664 #endif /* INET6 */ 5665 } 5666 } else { 5667 /* 5668 * ICMP error message in response to a TCP/UDP packet. 5669 * Extract the inner TCP/UDP header and search for that state. 5670 */ 5671 struct pf_pdesc pd2; 5672 struct ip h2; 5673 #ifdef INET6 5674 struct ip6_hdr h2_6; 5675 #endif /* INET6 */ 5676 int ipoff2; 5677 5678 /* Initialize pd2 fields valid for both packets with pd. */ 5679 memset(&pd2, 0, sizeof(pd2)); 5680 pd2.af = pd->af; 5681 pd2.dir = pd->dir; 5682 pd2.kif = pd->kif; 5683 pd2.m = pd->m; 5684 pd2.rdomain = pd->rdomain; 5685 /* Payload packet is from the opposite direction. */ 5686 pd2.sidx = (pd2.dir == PF_IN) ? 1 : 0; 5687 pd2.didx = (pd2.dir == PF_IN) ? 0 : 1; 5688 switch (pd->af) { 5689 case AF_INET: 5690 /* offset of h2 in mbuf chain */ 5691 ipoff2 = pd->off + ICMP_MINLEN; 5692 5693 if (!pf_pull_hdr(pd2.m, ipoff2, &h2, sizeof(h2), 5694 reason, pd2.af)) { 5695 DPFPRINTF(LOG_NOTICE, 5696 "ICMP error message too short (ip)"); 5697 return (PF_DROP); 5698 } 5699 /* 5700 * ICMP error messages don't refer to non-first 5701 * fragments 5702 */ 5703 if (h2.ip_off & htons(IP_OFFMASK)) { 5704 REASON_SET(reason, PFRES_FRAG); 5705 return (PF_DROP); 5706 } 5707 5708 /* offset of protocol header that follows h2 */ 5709 pd2.off = ipoff2; 5710 if (pf_walk_header(&pd2, &h2, reason) != PF_PASS) 5711 return (PF_DROP); 5712 5713 pd2.tot_len = ntohs(h2.ip_len); 5714 pd2.src = (struct pf_addr *)&h2.ip_src; 5715 pd2.dst = (struct pf_addr *)&h2.ip_dst; 5716 break; 5717 #ifdef INET6 5718 case AF_INET6: 5719 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 5720 5721 if (!pf_pull_hdr(pd2.m, ipoff2, &h2_6, sizeof(h2_6), 5722 reason, pd2.af)) { 5723 DPFPRINTF(LOG_NOTICE, 5724 "ICMP error message too short (ip6)"); 5725 return (PF_DROP); 5726 } 5727 5728 pd2.off = ipoff2; 5729 if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS) 5730 return (PF_DROP); 5731 5732 pd2.tot_len = ntohs(h2_6.ip6_plen) + 5733 sizeof(struct ip6_hdr); 5734 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 5735 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 5736 break; 5737 #endif /* INET6 */ 5738 default: 5739 unhandled_af(pd->af); 5740 } 5741 5742 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 5743 if (pf_status.debug >= LOG_NOTICE) { 5744 log(LOG_NOTICE, 5745 "pf: BAD ICMP %d:%d outer dst: ", 5746 icmptype, icmpcode); 5747 pf_print_host(pd->src, 0, pd->af); 5748 addlog(" -> "); 5749 pf_print_host(pd->dst, 0, pd->af); 5750 addlog(" inner src: "); 5751 pf_print_host(pd2.src, 0, pd2.af); 5752 addlog(" -> "); 5753 pf_print_host(pd2.dst, 0, pd2.af); 5754 addlog("\n"); 5755 } 5756 REASON_SET(reason, PFRES_BADSTATE); 5757 return (PF_DROP); 5758 } 5759 5760 switch (pd2.proto) { 5761 case IPPROTO_TCP: { 5762 struct tcphdr *th = &pd2.hdr.tcp; 5763 u_int32_t seq; 5764 struct pf_state_peer *src, *dst; 5765 u_int8_t dws; 5766 int action; 5767 5768 /* 5769 * Only the first 8 bytes of the TCP header can be 5770 * expected. Don't access any TCP header fields after 5771 * th_seq, an ackskew test is not possible. 5772 */ 5773 if (!pf_pull_hdr(pd2.m, pd2.off, th, 8, reason, 5774 pd2.af)) { 5775 DPFPRINTF(LOG_NOTICE, 5776 "ICMP error message too short (tcp)"); 5777 return (PF_DROP); 5778 } 5779 5780 key.af = pd2.af; 5781 key.proto = IPPROTO_TCP; 5782 key.rdomain = pd2.rdomain; 5783 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 5784 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 5785 key.port[pd2.sidx] = th->th_sport; 5786 key.port[pd2.didx] = th->th_dport; 5787 key.hash = pf_pkt_hash(pd2.af, pd2.proto, 5788 pd2.src, pd2.dst, th->th_sport, th->th_dport); 5789 5790 action = pf_find_state(&pd2, &key, stp); 5791 if (action != PF_MATCH) 5792 return (action); 5793 5794 if (pd2.dir == (*stp)->direction) { 5795 if (PF_REVERSED_KEY((*stp)->key, pd->af)) { 5796 src = &(*stp)->src; 5797 dst = &(*stp)->dst; 5798 } else { 5799 src = &(*stp)->dst; 5800 dst = &(*stp)->src; 5801 } 5802 } else { 5803 if (PF_REVERSED_KEY((*stp)->key, pd->af)) { 5804 src = &(*stp)->dst; 5805 dst = &(*stp)->src; 5806 } else { 5807 src = &(*stp)->src; 5808 dst = &(*stp)->dst; 5809 } 5810 } 5811 5812 if (src->wscale && dst->wscale) 5813 dws = dst->wscale & PF_WSCALE_MASK; 5814 else 5815 dws = 0; 5816 5817 /* Demodulate sequence number */ 5818 seq = ntohl(th->th_seq) - src->seqdiff; 5819 if (src->seqdiff) { 5820 pf_patch_32(pd, &th->th_seq, htonl(seq)); 5821 copyback = 1; 5822 } 5823 5824 if (!((*stp)->state_flags & PFSTATE_SLOPPY) && 5825 (!SEQ_GEQ(src->seqhi, seq) || !SEQ_GEQ(seq, 5826 src->seqlo - (dst->max_win << dws)))) { 5827 if (pf_status.debug >= LOG_NOTICE) { 5828 log(LOG_NOTICE, 5829 "pf: BAD ICMP %d:%d ", 5830 icmptype, icmpcode); 5831 pf_print_host(pd->src, 0, pd->af); 5832 addlog(" -> "); 5833 pf_print_host(pd->dst, 0, pd->af); 5834 addlog(" state: "); 5835 pf_print_state(*stp); 5836 addlog(" seq=%u\n", seq); 5837 } 5838 REASON_SET(reason, PFRES_BADSTATE); 5839 return (PF_DROP); 5840 } else { 5841 if (pf_status.debug >= LOG_DEBUG) { 5842 log(LOG_DEBUG, 5843 "pf: OK ICMP %d:%d ", 5844 icmptype, icmpcode); 5845 pf_print_host(pd->src, 0, pd->af); 5846 addlog(" -> "); 5847 pf_print_host(pd->dst, 0, pd->af); 5848 addlog(" state: "); 5849 pf_print_state(*stp); 5850 addlog(" seq=%u\n", seq); 5851 } 5852 } 5853 5854 /* translate source/destination address, if necessary */ 5855 if ((*stp)->key[PF_SK_WIRE] != 5856 (*stp)->key[PF_SK_STACK]) { 5857 struct pf_state_key *nk; 5858 int afto, sidx, didx; 5859 5860 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5861 nk = (*stp)->key[pd->sidx]; 5862 else 5863 nk = (*stp)->key[pd->didx]; 5864 5865 afto = pd->af != nk->af; 5866 sidx = afto ? pd2.didx : pd2.sidx; 5867 didx = afto ? pd2.sidx : pd2.didx; 5868 5869 #ifdef INET6 5870 if (afto) { 5871 if (pf_translate_icmp_af(pd, nk->af, 5872 &pd->hdr.icmp)) 5873 return (PF_DROP); 5874 m_copyback(pd->m, pd->off, 5875 sizeof(struct icmp6_hdr), 5876 &pd->hdr.icmp6, M_NOWAIT); 5877 if (pf_change_icmp_af(pd->m, ipoff2, 5878 pd, &pd2, &nk->addr[sidx], 5879 &nk->addr[didx], pd->af, nk->af)) 5880 return (PF_DROP); 5881 if (nk->af == AF_INET) 5882 pd->proto = IPPROTO_ICMP; 5883 else 5884 pd->proto = IPPROTO_ICMPV6; 5885 pd->m->m_pkthdr.ph_rtableid = 5886 nk->rdomain; 5887 pd->destchg = 1; 5888 pf_addrcpy(&pd->nsaddr, 5889 &nk->addr[pd2.sidx], nk->af); 5890 pf_addrcpy(&pd->ndaddr, 5891 &nk->addr[pd2.didx], nk->af); 5892 pd->naf = nk->af; 5893 5894 pf_patch_16(pd, 5895 &th->th_sport, nk->port[sidx]); 5896 pf_patch_16(pd, 5897 &th->th_dport, nk->port[didx]); 5898 5899 m_copyback(pd2.m, pd2.off, 8, th, 5900 M_NOWAIT); 5901 return (PF_AFRT); 5902 } 5903 #endif /* INET6 */ 5904 if (PF_ANEQ(pd2.src, 5905 &nk->addr[pd2.sidx], pd2.af) || 5906 nk->port[pd2.sidx] != th->th_sport) 5907 pf_translate_icmp(pd, pd2.src, 5908 &th->th_sport, pd->dst, 5909 &nk->addr[pd2.sidx], 5910 nk->port[pd2.sidx]); 5911 5912 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 5913 pd2.af) || pd2.rdomain != nk->rdomain) 5914 pd->destchg = 1; 5915 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5916 5917 if (PF_ANEQ(pd2.dst, 5918 &nk->addr[pd2.didx], pd2.af) || 5919 nk->port[pd2.didx] != th->th_dport) 5920 pf_translate_icmp(pd, pd2.dst, 5921 &th->th_dport, pd->src, 5922 &nk->addr[pd2.didx], 5923 nk->port[pd2.didx]); 5924 copyback = 1; 5925 } 5926 5927 if (copyback) { 5928 switch (pd2.af) { 5929 case AF_INET: 5930 m_copyback(pd->m, pd->off, ICMP_MINLEN, 5931 &pd->hdr.icmp, M_NOWAIT); 5932 m_copyback(pd2.m, ipoff2, sizeof(h2), 5933 &h2, M_NOWAIT); 5934 break; 5935 #ifdef INET6 5936 case AF_INET6: 5937 m_copyback(pd->m, pd->off, 5938 sizeof(struct icmp6_hdr), 5939 &pd->hdr.icmp6, M_NOWAIT); 5940 m_copyback(pd2.m, ipoff2, sizeof(h2_6), 5941 &h2_6, M_NOWAIT); 5942 break; 5943 #endif /* INET6 */ 5944 } 5945 m_copyback(pd2.m, pd2.off, 8, th, M_NOWAIT); 5946 } 5947 break; 5948 } 5949 case IPPROTO_UDP: { 5950 struct udphdr *uh = &pd2.hdr.udp; 5951 int action; 5952 5953 if (!pf_pull_hdr(pd2.m, pd2.off, uh, sizeof(*uh), 5954 reason, pd2.af)) { 5955 DPFPRINTF(LOG_NOTICE, 5956 "ICMP error message too short (udp)"); 5957 return (PF_DROP); 5958 } 5959 5960 key.af = pd2.af; 5961 key.proto = IPPROTO_UDP; 5962 key.rdomain = pd2.rdomain; 5963 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 5964 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 5965 key.port[pd2.sidx] = uh->uh_sport; 5966 key.port[pd2.didx] = uh->uh_dport; 5967 key.hash = pf_pkt_hash(pd2.af, pd2.proto, 5968 pd2.src, pd2.dst, uh->uh_sport, uh->uh_dport); 5969 5970 action = pf_find_state(&pd2, &key, stp); 5971 if (action != PF_MATCH) 5972 return (action); 5973 5974 /* translate source/destination address, if necessary */ 5975 if ((*stp)->key[PF_SK_WIRE] != 5976 (*stp)->key[PF_SK_STACK]) { 5977 struct pf_state_key *nk; 5978 int afto, sidx, didx; 5979 5980 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5981 nk = (*stp)->key[pd->sidx]; 5982 else 5983 nk = (*stp)->key[pd->didx]; 5984 5985 afto = pd->af != nk->af; 5986 sidx = afto ? pd2.didx : pd2.sidx; 5987 didx = afto ? pd2.sidx : pd2.didx; 5988 5989 #ifdef INET6 5990 if (afto) { 5991 if (pf_translate_icmp_af(pd, nk->af, 5992 &pd->hdr.icmp)) 5993 return (PF_DROP); 5994 m_copyback(pd->m, pd->off, 5995 sizeof(struct icmp6_hdr), 5996 &pd->hdr.icmp6, M_NOWAIT); 5997 if (pf_change_icmp_af(pd->m, ipoff2, 5998 pd, &pd2, &nk->addr[sidx], 5999 &nk->addr[didx], pd->af, nk->af)) 6000 return (PF_DROP); 6001 if (nk->af == AF_INET) 6002 pd->proto = IPPROTO_ICMP; 6003 else 6004 pd->proto = IPPROTO_ICMPV6; 6005 pd->m->m_pkthdr.ph_rtableid = 6006 nk->rdomain; 6007 pd->destchg = 1; 6008 pf_addrcpy(&pd->nsaddr, 6009 &nk->addr[pd2.sidx], nk->af); 6010 pf_addrcpy(&pd->ndaddr, 6011 &nk->addr[pd2.didx], nk->af); 6012 pd->naf = nk->af; 6013 6014 pf_patch_16(pd, 6015 &uh->uh_sport, nk->port[sidx]); 6016 pf_patch_16(pd, 6017 &uh->uh_dport, nk->port[didx]); 6018 6019 m_copyback(pd2.m, pd2.off, sizeof(*uh), 6020 uh, M_NOWAIT); 6021 return (PF_AFRT); 6022 } 6023 #endif /* INET6 */ 6024 6025 if (PF_ANEQ(pd2.src, 6026 &nk->addr[pd2.sidx], pd2.af) || 6027 nk->port[pd2.sidx] != uh->uh_sport) 6028 pf_translate_icmp(pd, pd2.src, 6029 &uh->uh_sport, pd->dst, 6030 &nk->addr[pd2.sidx], 6031 nk->port[pd2.sidx]); 6032 6033 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 6034 pd2.af) || pd2.rdomain != nk->rdomain) 6035 pd->destchg = 1; 6036 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 6037 6038 if (PF_ANEQ(pd2.dst, 6039 &nk->addr[pd2.didx], pd2.af) || 6040 nk->port[pd2.didx] != uh->uh_dport) 6041 pf_translate_icmp(pd, pd2.dst, 6042 &uh->uh_dport, pd->src, 6043 &nk->addr[pd2.didx], 6044 nk->port[pd2.didx]); 6045 6046 switch (pd2.af) { 6047 case AF_INET: 6048 m_copyback(pd->m, pd->off, ICMP_MINLEN, 6049 &pd->hdr.icmp, M_NOWAIT); 6050 m_copyback(pd2.m, ipoff2, sizeof(h2), 6051 &h2, M_NOWAIT); 6052 break; 6053 #ifdef INET6 6054 case AF_INET6: 6055 m_copyback(pd->m, pd->off, 6056 sizeof(struct icmp6_hdr), 6057 &pd->hdr.icmp6, M_NOWAIT); 6058 m_copyback(pd2.m, ipoff2, sizeof(h2_6), 6059 &h2_6, M_NOWAIT); 6060 break; 6061 #endif /* INET6 */ 6062 } 6063 /* Avoid recomputing quoted UDP checksum. 6064 * note: udp6 0 csum invalid per rfc2460 p27. 6065 * but presumed nothing cares in this context */ 6066 pf_patch_16(pd, &uh->uh_sum, 0); 6067 m_copyback(pd2.m, pd2.off, sizeof(*uh), uh, 6068 M_NOWAIT); 6069 copyback = 1; 6070 } 6071 break; 6072 } 6073 case IPPROTO_ICMP: { 6074 struct icmp *iih = &pd2.hdr.icmp; 6075 6076 if (pd2.af != AF_INET) { 6077 REASON_SET(reason, PFRES_NORM); 6078 return (PF_DROP); 6079 } 6080 6081 if (!pf_pull_hdr(pd2.m, pd2.off, iih, ICMP_MINLEN, 6082 reason, pd2.af)) { 6083 DPFPRINTF(LOG_NOTICE, 6084 "ICMP error message too short (icmp)"); 6085 return (PF_DROP); 6086 } 6087 6088 pf_icmp_mapping(&pd2, iih->icmp_type, 6089 &icmp_dir, &virtual_id, &virtual_type); 6090 6091 ret = pf_icmp_state_lookup(&pd2, &key, stp, 6092 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 6093 if (ret >= 0) 6094 return (ret); 6095 6096 /* translate source/destination address, if necessary */ 6097 if ((*stp)->key[PF_SK_WIRE] != 6098 (*stp)->key[PF_SK_STACK]) { 6099 struct pf_state_key *nk; 6100 int afto, sidx, didx; 6101 6102 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 6103 nk = (*stp)->key[pd->sidx]; 6104 else 6105 nk = (*stp)->key[pd->didx]; 6106 6107 afto = pd->af != nk->af; 6108 sidx = afto ? pd2.didx : pd2.sidx; 6109 didx = afto ? pd2.sidx : pd2.didx; 6110 iidx = afto ? !iidx : iidx; 6111 6112 #ifdef INET6 6113 if (afto) { 6114 if (nk->af != AF_INET6) 6115 return (PF_DROP); 6116 if (pf_translate_icmp_af(pd, nk->af, 6117 &pd->hdr.icmp)) 6118 return (PF_DROP); 6119 m_copyback(pd->m, pd->off, 6120 sizeof(struct icmp6_hdr), 6121 &pd->hdr.icmp6, M_NOWAIT); 6122 if (pf_change_icmp_af(pd->m, ipoff2, 6123 pd, &pd2, &nk->addr[sidx], 6124 &nk->addr[didx], pd->af, nk->af)) 6125 return (PF_DROP); 6126 pd->proto = IPPROTO_ICMPV6; 6127 if (pf_translate_icmp_af(pd, 6128 nk->af, iih)) 6129 return (PF_DROP); 6130 if (virtual_type == htons(ICMP_ECHO)) 6131 pf_patch_16(pd, &iih->icmp_id, 6132 nk->port[iidx]); 6133 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, 6134 iih, M_NOWAIT); 6135 pd->m->m_pkthdr.ph_rtableid = 6136 nk->rdomain; 6137 pd->destchg = 1; 6138 pf_addrcpy(&pd->nsaddr, 6139 &nk->addr[pd2.sidx], nk->af); 6140 pf_addrcpy(&pd->ndaddr, 6141 &nk->addr[pd2.didx], nk->af); 6142 pd->naf = nk->af; 6143 return (PF_AFRT); 6144 } 6145 #endif /* INET6 */ 6146 6147 if (PF_ANEQ(pd2.src, 6148 &nk->addr[pd2.sidx], pd2.af) || 6149 (virtual_type == htons(ICMP_ECHO) && 6150 nk->port[iidx] != iih->icmp_id)) 6151 pf_translate_icmp(pd, pd2.src, 6152 (virtual_type == htons(ICMP_ECHO)) ? 6153 &iih->icmp_id : NULL, 6154 pd->dst, &nk->addr[pd2.sidx], 6155 (virtual_type == htons(ICMP_ECHO)) ? 6156 nk->port[iidx] : 0); 6157 6158 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 6159 pd2.af) || pd2.rdomain != nk->rdomain) 6160 pd->destchg = 1; 6161 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 6162 6163 if (PF_ANEQ(pd2.dst, 6164 &nk->addr[pd2.didx], pd2.af)) 6165 pf_translate_icmp(pd, pd2.dst, NULL, 6166 pd->src, &nk->addr[pd2.didx], 0); 6167 6168 m_copyback(pd->m, pd->off, ICMP_MINLEN, 6169 &pd->hdr.icmp, M_NOWAIT); 6170 m_copyback(pd2.m, ipoff2, sizeof(h2), &h2, 6171 M_NOWAIT); 6172 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, iih, 6173 M_NOWAIT); 6174 copyback = 1; 6175 } 6176 break; 6177 } 6178 #ifdef INET6 6179 case IPPROTO_ICMPV6: { 6180 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 6181 6182 if (pd2.af != AF_INET6) { 6183 REASON_SET(reason, PFRES_NORM); 6184 return (PF_DROP); 6185 } 6186 6187 if (!pf_pull_hdr(pd2.m, pd2.off, iih, 6188 sizeof(struct icmp6_hdr), reason, pd2.af)) { 6189 DPFPRINTF(LOG_NOTICE, 6190 "ICMP error message too short (icmp6)"); 6191 return (PF_DROP); 6192 } 6193 6194 pf_icmp_mapping(&pd2, iih->icmp6_type, 6195 &icmp_dir, &virtual_id, &virtual_type); 6196 ret = pf_icmp_state_lookup(&pd2, &key, stp, 6197 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 6198 /* IPv6? try matching a multicast address */ 6199 if (ret == PF_DROP && pd2.af == AF_INET6 && 6200 icmp_dir == PF_OUT) 6201 ret = pf_icmp_state_lookup(&pd2, &key, stp, 6202 virtual_id, virtual_type, icmp_dir, &iidx, 6203 1, 1); 6204 if (ret >= 0) 6205 return (ret); 6206 6207 /* translate source/destination address, if necessary */ 6208 if ((*stp)->key[PF_SK_WIRE] != 6209 (*stp)->key[PF_SK_STACK]) { 6210 struct pf_state_key *nk; 6211 int afto, sidx, didx; 6212 6213 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 6214 nk = (*stp)->key[pd->sidx]; 6215 else 6216 nk = (*stp)->key[pd->didx]; 6217 6218 afto = pd->af != nk->af; 6219 sidx = afto ? pd2.didx : pd2.sidx; 6220 didx = afto ? pd2.sidx : pd2.didx; 6221 iidx = afto ? !iidx : iidx; 6222 6223 if (afto) { 6224 if (nk->af != AF_INET) 6225 return (PF_DROP); 6226 if (pf_translate_icmp_af(pd, nk->af, 6227 &pd->hdr.icmp)) 6228 return (PF_DROP); 6229 m_copyback(pd->m, pd->off, 6230 sizeof(struct icmp6_hdr), 6231 &pd->hdr.icmp6, M_NOWAIT); 6232 if (pf_change_icmp_af(pd->m, ipoff2, 6233 pd, &pd2, &nk->addr[sidx], 6234 &nk->addr[didx], pd->af, nk->af)) 6235 return (PF_DROP); 6236 pd->proto = IPPROTO_ICMP; 6237 if (pf_translate_icmp_af(pd, 6238 nk->af, iih)) 6239 return (PF_DROP); 6240 if (virtual_type == 6241 htons(ICMP6_ECHO_REQUEST)) 6242 pf_patch_16(pd, &iih->icmp6_id, 6243 nk->port[iidx]); 6244 m_copyback(pd2.m, pd2.off, 6245 sizeof(struct icmp6_hdr), iih, 6246 M_NOWAIT); 6247 pd->m->m_pkthdr.ph_rtableid = 6248 nk->rdomain; 6249 pd->destchg = 1; 6250 pf_addrcpy(&pd->nsaddr, 6251 &nk->addr[pd2.sidx], nk->af); 6252 pf_addrcpy(&pd->ndaddr, 6253 &nk->addr[pd2.didx], nk->af); 6254 pd->naf = nk->af; 6255 return (PF_AFRT); 6256 } 6257 6258 if (PF_ANEQ(pd2.src, 6259 &nk->addr[pd2.sidx], pd2.af) || 6260 ((virtual_type == 6261 htons(ICMP6_ECHO_REQUEST)) && 6262 nk->port[pd2.sidx] != iih->icmp6_id)) 6263 pf_translate_icmp(pd, pd2.src, 6264 (virtual_type == 6265 htons(ICMP6_ECHO_REQUEST)) 6266 ? &iih->icmp6_id : NULL, 6267 pd->dst, &nk->addr[pd2.sidx], 6268 (virtual_type == 6269 htons(ICMP6_ECHO_REQUEST)) 6270 ? nk->port[iidx] : 0); 6271 6272 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 6273 pd2.af) || pd2.rdomain != nk->rdomain) 6274 pd->destchg = 1; 6275 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 6276 6277 if (PF_ANEQ(pd2.dst, 6278 &nk->addr[pd2.didx], pd2.af)) 6279 pf_translate_icmp(pd, pd2.dst, NULL, 6280 pd->src, &nk->addr[pd2.didx], 0); 6281 6282 m_copyback(pd->m, pd->off, 6283 sizeof(struct icmp6_hdr), &pd->hdr.icmp6, 6284 M_NOWAIT); 6285 m_copyback(pd2.m, ipoff2, sizeof(h2_6), &h2_6, 6286 M_NOWAIT); 6287 m_copyback(pd2.m, pd2.off, 6288 sizeof(struct icmp6_hdr), iih, M_NOWAIT); 6289 copyback = 1; 6290 } 6291 break; 6292 } 6293 #endif /* INET6 */ 6294 default: { 6295 int action; 6296 6297 key.af = pd2.af; 6298 key.proto = pd2.proto; 6299 key.rdomain = pd2.rdomain; 6300 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 6301 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 6302 key.port[0] = key.port[1] = 0; 6303 key.hash = pf_pkt_hash(pd2.af, pd2.proto, 6304 pd2.src, pd2.dst, 0, 0); 6305 6306 action = pf_find_state(&pd2, &key, stp); 6307 if (action != PF_MATCH) 6308 return (action); 6309 6310 /* translate source/destination address, if necessary */ 6311 if ((*stp)->key[PF_SK_WIRE] != 6312 (*stp)->key[PF_SK_STACK]) { 6313 struct pf_state_key *nk = 6314 (*stp)->key[pd->didx]; 6315 6316 if (PF_ANEQ(pd2.src, 6317 &nk->addr[pd2.sidx], pd2.af)) 6318 pf_translate_icmp(pd, pd2.src, NULL, 6319 pd->dst, &nk->addr[pd2.sidx], 0); 6320 6321 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 6322 pd2.af) || pd2.rdomain != nk->rdomain) 6323 pd->destchg = 1; 6324 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 6325 6326 if (PF_ANEQ(pd2.dst, 6327 &nk->addr[pd2.didx], pd2.af)) 6328 pf_translate_icmp(pd, pd2.dst, NULL, 6329 pd->src, &nk->addr[pd2.didx], 0); 6330 6331 switch (pd2.af) { 6332 case AF_INET: 6333 m_copyback(pd->m, pd->off, ICMP_MINLEN, 6334 &pd->hdr.icmp, M_NOWAIT); 6335 m_copyback(pd2.m, ipoff2, sizeof(h2), 6336 &h2, M_NOWAIT); 6337 break; 6338 #ifdef INET6 6339 case AF_INET6: 6340 m_copyback(pd->m, pd->off, 6341 sizeof(struct icmp6_hdr), 6342 &pd->hdr.icmp6, M_NOWAIT); 6343 m_copyback(pd2.m, ipoff2, sizeof(h2_6), 6344 &h2_6, M_NOWAIT); 6345 break; 6346 #endif /* INET6 */ 6347 } 6348 copyback = 1; 6349 } 6350 break; 6351 } 6352 } 6353 } 6354 if (copyback) { 6355 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 6356 } 6357 6358 return (PF_PASS); 6359 } 6360 6361 /* 6362 * ipoff and off are measured from the start of the mbuf chain. 6363 * h must be at "ipoff" on the mbuf chain. 6364 */ 6365 void * 6366 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 6367 u_short *reasonp, sa_family_t af) 6368 { 6369 int iplen = 0; 6370 6371 switch (af) { 6372 case AF_INET: { 6373 struct ip *h = mtod(m, struct ip *); 6374 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 6375 6376 if (fragoff) { 6377 REASON_SET(reasonp, PFRES_FRAG); 6378 return (NULL); 6379 } 6380 iplen = ntohs(h->ip_len); 6381 break; 6382 } 6383 #ifdef INET6 6384 case AF_INET6: { 6385 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 6386 6387 iplen = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 6388 break; 6389 } 6390 #endif /* INET6 */ 6391 } 6392 if (m->m_pkthdr.len < off + len || iplen < off + len) { 6393 REASON_SET(reasonp, PFRES_SHORT); 6394 return (NULL); 6395 } 6396 m_copydata(m, off, len, p); 6397 return (p); 6398 } 6399 6400 int 6401 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif, 6402 int rtableid) 6403 { 6404 struct sockaddr_storage ss; 6405 struct sockaddr_in *dst; 6406 int ret = 1; 6407 int check_mpath; 6408 #ifdef INET6 6409 struct sockaddr_in6 *dst6; 6410 #endif /* INET6 */ 6411 struct rtentry *rt = NULL; 6412 6413 check_mpath = 0; 6414 memset(&ss, 0, sizeof(ss)); 6415 switch (af) { 6416 case AF_INET: 6417 dst = (struct sockaddr_in *)&ss; 6418 dst->sin_family = AF_INET; 6419 dst->sin_len = sizeof(*dst); 6420 dst->sin_addr = addr->v4; 6421 if (ipmultipath) 6422 check_mpath = 1; 6423 break; 6424 #ifdef INET6 6425 case AF_INET6: 6426 /* 6427 * Skip check for addresses with embedded interface scope, 6428 * as they would always match anyway. 6429 */ 6430 if (IN6_IS_SCOPE_EMBED(&addr->v6)) 6431 goto out; 6432 dst6 = (struct sockaddr_in6 *)&ss; 6433 dst6->sin6_family = AF_INET6; 6434 dst6->sin6_len = sizeof(*dst6); 6435 dst6->sin6_addr = addr->v6; 6436 if (ip6_multipath) 6437 check_mpath = 1; 6438 break; 6439 #endif /* INET6 */ 6440 } 6441 6442 /* Skip checks for ipsec interfaces */ 6443 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 6444 goto out; 6445 6446 rt = rtalloc(sstosa(&ss), 0, rtableid); 6447 if (rt != NULL) { 6448 /* No interface given, this is a no-route check */ 6449 if (kif == NULL) 6450 goto out; 6451 6452 if (kif->pfik_ifp == NULL) { 6453 ret = 0; 6454 goto out; 6455 } 6456 6457 /* Perform uRPF check if passed input interface */ 6458 ret = 0; 6459 do { 6460 if (rt->rt_ifidx == kif->pfik_ifp->if_index) { 6461 ret = 1; 6462 #if NCARP > 0 6463 } else { 6464 struct ifnet *ifp; 6465 6466 ifp = if_get(rt->rt_ifidx); 6467 if (ifp != NULL && ifp->if_type == IFT_CARP && 6468 ifp->if_carpdevidx == 6469 kif->pfik_ifp->if_index) 6470 ret = 1; 6471 if_put(ifp); 6472 #endif /* NCARP */ 6473 } 6474 6475 rt = rtable_iterate(rt); 6476 } while (check_mpath == 1 && rt != NULL && ret == 0); 6477 } else 6478 ret = 0; 6479 out: 6480 rtfree(rt); 6481 return (ret); 6482 } 6483 6484 int 6485 pf_rtlabel_match(struct pf_addr *addr, sa_family_t af, struct pf_addr_wrap *aw, 6486 int rtableid) 6487 { 6488 struct sockaddr_storage ss; 6489 struct sockaddr_in *dst; 6490 #ifdef INET6 6491 struct sockaddr_in6 *dst6; 6492 #endif /* INET6 */ 6493 struct rtentry *rt; 6494 int ret = 0; 6495 6496 memset(&ss, 0, sizeof(ss)); 6497 switch (af) { 6498 case AF_INET: 6499 dst = (struct sockaddr_in *)&ss; 6500 dst->sin_family = AF_INET; 6501 dst->sin_len = sizeof(*dst); 6502 dst->sin_addr = addr->v4; 6503 break; 6504 #ifdef INET6 6505 case AF_INET6: 6506 dst6 = (struct sockaddr_in6 *)&ss; 6507 dst6->sin6_family = AF_INET6; 6508 dst6->sin6_len = sizeof(*dst6); 6509 dst6->sin6_addr = addr->v6; 6510 break; 6511 #endif /* INET6 */ 6512 } 6513 6514 rt = rtalloc(sstosa(&ss), RT_RESOLVE, rtableid); 6515 if (rt != NULL) { 6516 if (rt->rt_labelid == aw->v.rtlabel) 6517 ret = 1; 6518 rtfree(rt); 6519 } 6520 6521 return (ret); 6522 } 6523 6524 /* pf_route() may change pd->m, adjust local copies after calling */ 6525 void 6526 pf_route(struct pf_pdesc *pd, struct pf_state *st) 6527 { 6528 struct mbuf *m0; 6529 struct mbuf_list ml; 6530 struct sockaddr_in *dst, sin; 6531 struct rtentry *rt = NULL; 6532 struct ip *ip; 6533 struct ifnet *ifp = NULL; 6534 unsigned int rtableid; 6535 6536 if (pd->m->m_pkthdr.pf.routed++ > 3) { 6537 m_freem(pd->m); 6538 pd->m = NULL; 6539 return; 6540 } 6541 6542 if (st->rt == PF_DUPTO) { 6543 if ((m0 = m_dup_pkt(pd->m, max_linkhdr, M_NOWAIT)) == NULL) 6544 return; 6545 } else { 6546 if ((st->rt == PF_REPLYTO) == (st->direction == pd->dir)) 6547 return; 6548 m0 = pd->m; 6549 pd->m = NULL; 6550 } 6551 6552 if (m0->m_len < sizeof(struct ip)) { 6553 DPFPRINTF(LOG_ERR, 6554 "%s: m0->m_len < sizeof(struct ip)", __func__); 6555 goto bad; 6556 } 6557 6558 ip = mtod(m0, struct ip *); 6559 6560 if (pd->dir == PF_IN) { 6561 if (ip->ip_ttl <= IPTTLDEC) { 6562 if (st->rt != PF_DUPTO) { 6563 pf_send_icmp(m0, ICMP_TIMXCEED, 6564 ICMP_TIMXCEED_INTRANS, 0, 6565 pd->af, st->rule.ptr, pd->rdomain); 6566 } 6567 goto bad; 6568 } 6569 ip->ip_ttl -= IPTTLDEC; 6570 } 6571 6572 memset(&sin, 0, sizeof(sin)); 6573 dst = &sin; 6574 dst->sin_family = AF_INET; 6575 dst->sin_len = sizeof(*dst); 6576 dst->sin_addr = st->rt_addr.v4; 6577 rtableid = m0->m_pkthdr.ph_rtableid; 6578 6579 rt = rtalloc_mpath(sintosa(dst), &ip->ip_src.s_addr, rtableid); 6580 if (!rtisvalid(rt)) { 6581 if (st->rt != PF_DUPTO) { 6582 pf_send_icmp(m0, ICMP_UNREACH, ICMP_UNREACH_HOST, 6583 0, pd->af, st->rule.ptr, pd->rdomain); 6584 } 6585 ipstat_inc(ips_noroute); 6586 goto bad; 6587 } 6588 6589 ifp = if_get(rt->rt_ifidx); 6590 if (ifp == NULL) 6591 goto bad; 6592 6593 /* A locally generated packet may have invalid source address. */ 6594 if ((ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET && 6595 (ifp->if_flags & IFF_LOOPBACK) == 0) 6596 ip->ip_src = ifatoia(rt->rt_ifa)->ia_addr.sin_addr; 6597 6598 if (st->rt != PF_DUPTO && pd->dir == PF_IN) { 6599 if (pf_test(AF_INET, PF_OUT, ifp, &m0) != PF_PASS) 6600 goto bad; 6601 else if (m0 == NULL) 6602 goto done; 6603 if (m0->m_len < sizeof(struct ip)) { 6604 DPFPRINTF(LOG_ERR, 6605 "%s: m0->m_len < sizeof(struct ip)", __func__); 6606 goto bad; 6607 } 6608 ip = mtod(m0, struct ip *); 6609 } 6610 6611 if (if_output_tso(ifp, &m0, sintosa(dst), rt, ifp->if_mtu) || 6612 m0 == NULL) 6613 goto done; 6614 6615 /* 6616 * Too large for interface; fragment if possible. 6617 * Must be able to put at least 8 bytes per fragment. 6618 */ 6619 if (ip->ip_off & htons(IP_DF)) { 6620 ipstat_inc(ips_cantfrag); 6621 if (st->rt != PF_DUPTO) 6622 pf_send_icmp(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 6623 ifp->if_mtu, pd->af, st->rule.ptr, pd->rdomain); 6624 goto bad; 6625 } 6626 6627 if (ip_fragment(m0, &ml, ifp, ifp->if_mtu) || 6628 if_output_ml(ifp, &ml, sintosa(dst), rt)) 6629 goto done; 6630 ipstat_inc(ips_fragmented); 6631 6632 done: 6633 if_put(ifp); 6634 rtfree(rt); 6635 return; 6636 6637 bad: 6638 m_freem(m0); 6639 goto done; 6640 } 6641 6642 #ifdef INET6 6643 /* pf_route6() may change pd->m, adjust local copies after calling */ 6644 void 6645 pf_route6(struct pf_pdesc *pd, struct pf_state *st) 6646 { 6647 struct mbuf *m0; 6648 struct sockaddr_in6 *dst, sin6; 6649 struct rtentry *rt = NULL; 6650 struct ip6_hdr *ip6; 6651 struct ifnet *ifp = NULL; 6652 struct m_tag *mtag; 6653 unsigned int rtableid; 6654 6655 if (pd->m->m_pkthdr.pf.routed++ > 3) { 6656 m_freem(pd->m); 6657 pd->m = NULL; 6658 return; 6659 } 6660 6661 if (st->rt == PF_DUPTO) { 6662 if ((m0 = m_dup_pkt(pd->m, max_linkhdr, M_NOWAIT)) == NULL) 6663 return; 6664 } else { 6665 if ((st->rt == PF_REPLYTO) == (st->direction == pd->dir)) 6666 return; 6667 m0 = pd->m; 6668 pd->m = NULL; 6669 } 6670 6671 if (m0->m_len < sizeof(struct ip6_hdr)) { 6672 DPFPRINTF(LOG_ERR, 6673 "%s: m0->m_len < sizeof(struct ip6_hdr)", __func__); 6674 goto bad; 6675 } 6676 ip6 = mtod(m0, struct ip6_hdr *); 6677 6678 if (pd->dir == PF_IN) { 6679 if (ip6->ip6_hlim <= IPV6_HLIMDEC) { 6680 if (st->rt != PF_DUPTO) { 6681 pf_send_icmp(m0, ICMP6_TIME_EXCEEDED, 6682 ICMP6_TIME_EXCEED_TRANSIT, 0, 6683 pd->af, st->rule.ptr, pd->rdomain); 6684 } 6685 goto bad; 6686 } 6687 ip6->ip6_hlim -= IPV6_HLIMDEC; 6688 } 6689 6690 memset(&sin6, 0, sizeof(sin6)); 6691 dst = &sin6; 6692 dst->sin6_family = AF_INET6; 6693 dst->sin6_len = sizeof(*dst); 6694 dst->sin6_addr = st->rt_addr.v6; 6695 rtableid = m0->m_pkthdr.ph_rtableid; 6696 6697 rt = rtalloc_mpath(sin6tosa(dst), &ip6->ip6_src.s6_addr32[0], 6698 rtableid); 6699 if (!rtisvalid(rt)) { 6700 if (st->rt != PF_DUPTO) { 6701 pf_send_icmp(m0, ICMP6_DST_UNREACH, 6702 ICMP6_DST_UNREACH_NOROUTE, 0, 6703 pd->af, st->rule.ptr, pd->rdomain); 6704 } 6705 ip6stat_inc(ip6s_noroute); 6706 goto bad; 6707 } 6708 6709 ifp = if_get(rt->rt_ifidx); 6710 if (ifp == NULL) 6711 goto bad; 6712 6713 /* A locally generated packet may have invalid source address. */ 6714 if (IN6_IS_ADDR_LOOPBACK(&ip6->ip6_src) && 6715 (ifp->if_flags & IFF_LOOPBACK) == 0) 6716 ip6->ip6_src = ifatoia6(rt->rt_ifa)->ia_addr.sin6_addr; 6717 6718 if (st->rt != PF_DUPTO && pd->dir == PF_IN) { 6719 if (pf_test(AF_INET6, PF_OUT, ifp, &m0) != PF_PASS) 6720 goto bad; 6721 else if (m0 == NULL) 6722 goto done; 6723 if (m0->m_len < sizeof(struct ip6_hdr)) { 6724 DPFPRINTF(LOG_ERR, 6725 "%s: m0->m_len < sizeof(struct ip6_hdr)", __func__); 6726 goto bad; 6727 } 6728 } 6729 6730 /* 6731 * If packet has been reassembled by PF earlier, we have to 6732 * use pf_refragment6() here to turn it back to fragments. 6733 */ 6734 if ((mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL))) { 6735 (void) pf_refragment6(&m0, mtag, dst, ifp, rt); 6736 goto done; 6737 } 6738 6739 if (if_output_tso(ifp, &m0, sin6tosa(dst), rt, ifp->if_mtu) || 6740 m0 == NULL) 6741 goto done; 6742 6743 ip6stat_inc(ip6s_cantfrag); 6744 if (st->rt != PF_DUPTO) 6745 pf_send_icmp(m0, ICMP6_PACKET_TOO_BIG, 0, 6746 ifp->if_mtu, pd->af, st->rule.ptr, pd->rdomain); 6747 goto bad; 6748 6749 done: 6750 if_put(ifp); 6751 rtfree(rt); 6752 return; 6753 6754 bad: 6755 m_freem(m0); 6756 goto done; 6757 } 6758 #endif /* INET6 */ 6759 6760 /* 6761 * check TCP checksum and set mbuf flag 6762 * off is the offset where the protocol header starts 6763 * len is the total length of protocol header plus payload 6764 * returns 0 when the checksum is valid, otherwise returns 1. 6765 * if the _OUT flag is set the checksum isn't done yet, consider these ok 6766 */ 6767 int 6768 pf_check_tcp_cksum(struct mbuf *m, int off, int len, sa_family_t af) 6769 { 6770 u_int16_t sum; 6771 6772 if (m->m_pkthdr.csum_flags & 6773 (M_TCP_CSUM_IN_OK | M_TCP_CSUM_OUT)) { 6774 return (0); 6775 } 6776 if (m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_BAD || 6777 off < sizeof(struct ip) || 6778 m->m_pkthdr.len < off + len) { 6779 return (1); 6780 } 6781 6782 /* need to do it in software */ 6783 tcpstat_inc(tcps_inswcsum); 6784 6785 switch (af) { 6786 case AF_INET: 6787 if (m->m_len < sizeof(struct ip)) 6788 return (1); 6789 6790 sum = in4_cksum(m, IPPROTO_TCP, off, len); 6791 break; 6792 #ifdef INET6 6793 case AF_INET6: 6794 if (m->m_len < sizeof(struct ip6_hdr)) 6795 return (1); 6796 6797 sum = in6_cksum(m, IPPROTO_TCP, off, len); 6798 break; 6799 #endif /* INET6 */ 6800 default: 6801 unhandled_af(af); 6802 } 6803 if (sum) { 6804 tcpstat_inc(tcps_rcvbadsum); 6805 m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_BAD; 6806 return (1); 6807 } 6808 6809 m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK; 6810 return (0); 6811 } 6812 6813 struct pf_divert * 6814 pf_find_divert(struct mbuf *m) 6815 { 6816 struct m_tag *mtag; 6817 6818 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) 6819 return (NULL); 6820 6821 return ((struct pf_divert *)(mtag + 1)); 6822 } 6823 6824 struct pf_divert * 6825 pf_get_divert(struct mbuf *m) 6826 { 6827 struct m_tag *mtag; 6828 6829 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) { 6830 mtag = m_tag_get(PACKET_TAG_PF_DIVERT, sizeof(struct pf_divert), 6831 M_NOWAIT); 6832 if (mtag == NULL) 6833 return (NULL); 6834 memset(mtag + 1, 0, sizeof(struct pf_divert)); 6835 m_tag_prepend(m, mtag); 6836 } 6837 6838 return ((struct pf_divert *)(mtag + 1)); 6839 } 6840 6841 int 6842 pf_walk_option(struct pf_pdesc *pd, struct ip *h, int off, int end, 6843 u_short *reason) 6844 { 6845 uint8_t type, length, opts[15 * 4 - sizeof(struct ip)]; 6846 6847 /* IP header in payload of ICMP packet may be too short */ 6848 if (pd->m->m_pkthdr.len < end) { 6849 DPFPRINTF(LOG_NOTICE, "IP option too short"); 6850 REASON_SET(reason, PFRES_SHORT); 6851 return (PF_DROP); 6852 } 6853 6854 KASSERT(end - off <= sizeof(opts)); 6855 m_copydata(pd->m, off, end - off, opts); 6856 end -= off; 6857 off = 0; 6858 6859 while (off < end) { 6860 type = opts[off]; 6861 if (type == IPOPT_EOL) 6862 break; 6863 if (type == IPOPT_NOP) { 6864 off++; 6865 continue; 6866 } 6867 if (off + 2 > end) { 6868 DPFPRINTF(LOG_NOTICE, "IP length opt"); 6869 REASON_SET(reason, PFRES_IPOPTIONS); 6870 return (PF_DROP); 6871 } 6872 length = opts[off + 1]; 6873 if (length < 2) { 6874 DPFPRINTF(LOG_NOTICE, "IP short opt"); 6875 REASON_SET(reason, PFRES_IPOPTIONS); 6876 return (PF_DROP); 6877 } 6878 if (off + length > end) { 6879 DPFPRINTF(LOG_NOTICE, "IP long opt"); 6880 REASON_SET(reason, PFRES_IPOPTIONS); 6881 return (PF_DROP); 6882 } 6883 switch (type) { 6884 case IPOPT_RA: 6885 SET(pd->badopts, PF_OPT_ROUTER_ALERT); 6886 break; 6887 default: 6888 SET(pd->badopts, PF_OPT_OTHER); 6889 break; 6890 } 6891 off += length; 6892 } 6893 6894 return (PF_PASS); 6895 } 6896 6897 int 6898 pf_walk_header(struct pf_pdesc *pd, struct ip *h, u_short *reason) 6899 { 6900 struct ip6_ext ext; 6901 u_int32_t hlen, end; 6902 int hdr_cnt; 6903 6904 hlen = h->ip_hl << 2; 6905 if (hlen < sizeof(struct ip) || hlen > ntohs(h->ip_len)) { 6906 REASON_SET(reason, PFRES_SHORT); 6907 return (PF_DROP); 6908 } 6909 if (hlen != sizeof(struct ip)) { 6910 if (pf_walk_option(pd, h, pd->off + sizeof(struct ip), 6911 pd->off + hlen, reason) != PF_PASS) 6912 return (PF_DROP); 6913 /* header options which contain only padding is fishy */ 6914 if (pd->badopts == 0) 6915 SET(pd->badopts, PF_OPT_OTHER); 6916 } 6917 end = pd->off + ntohs(h->ip_len); 6918 pd->off += hlen; 6919 pd->proto = h->ip_p; 6920 /* IGMP packets have router alert options, allow them */ 6921 if (pd->proto == IPPROTO_IGMP) { 6922 /* 6923 * According to RFC 1112 ttl must be set to 1 in all IGMP 6924 * packets sent to 224.0.0.1 6925 */ 6926 if ((h->ip_ttl != 1) && 6927 (h->ip_dst.s_addr == INADDR_ALLHOSTS_GROUP)) { 6928 DPFPRINTF(LOG_NOTICE, "Invalid IGMP"); 6929 REASON_SET(reason, PFRES_IPOPTIONS); 6930 return (PF_DROP); 6931 } 6932 CLR(pd->badopts, PF_OPT_ROUTER_ALERT); 6933 } 6934 /* stop walking over non initial fragments */ 6935 if ((h->ip_off & htons(IP_OFFMASK)) != 0) 6936 return (PF_PASS); 6937 6938 for (hdr_cnt = 0; hdr_cnt < pf_hdr_limit; hdr_cnt++) { 6939 switch (pd->proto) { 6940 case IPPROTO_AH: 6941 /* fragments may be short */ 6942 if ((h->ip_off & htons(IP_MF | IP_OFFMASK)) != 0 && 6943 end < pd->off + sizeof(ext)) 6944 return (PF_PASS); 6945 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 6946 reason, AF_INET)) { 6947 DPFPRINTF(LOG_NOTICE, "IP short exthdr"); 6948 return (PF_DROP); 6949 } 6950 pd->off += (ext.ip6e_len + 2) * 4; 6951 pd->proto = ext.ip6e_nxt; 6952 break; 6953 default: 6954 return (PF_PASS); 6955 } 6956 } 6957 DPFPRINTF(LOG_NOTICE, "IPv4 nested authentication header limit"); 6958 REASON_SET(reason, PFRES_IPOPTIONS); 6959 return (PF_DROP); 6960 } 6961 6962 #ifdef INET6 6963 int 6964 pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end, 6965 u_short *reason) 6966 { 6967 struct ip6_opt opt; 6968 struct ip6_opt_jumbo jumbo; 6969 6970 while (off < end) { 6971 if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type, 6972 sizeof(opt.ip6o_type), reason, AF_INET6)) { 6973 DPFPRINTF(LOG_NOTICE, "IPv6 short opt type"); 6974 return (PF_DROP); 6975 } 6976 if (opt.ip6o_type == IP6OPT_PAD1) { 6977 off++; 6978 continue; 6979 } 6980 if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), 6981 reason, AF_INET6)) { 6982 DPFPRINTF(LOG_NOTICE, "IPv6 short opt"); 6983 return (PF_DROP); 6984 } 6985 if (off + sizeof(opt) + opt.ip6o_len > end) { 6986 DPFPRINTF(LOG_NOTICE, "IPv6 long opt"); 6987 REASON_SET(reason, PFRES_IPOPTIONS); 6988 return (PF_DROP); 6989 } 6990 switch (opt.ip6o_type) { 6991 case IP6OPT_PADN: 6992 break; 6993 case IP6OPT_JUMBO: 6994 SET(pd->badopts, PF_OPT_JUMBO); 6995 if (pd->jumbolen != 0) { 6996 DPFPRINTF(LOG_NOTICE, "IPv6 multiple jumbo"); 6997 REASON_SET(reason, PFRES_IPOPTIONS); 6998 return (PF_DROP); 6999 } 7000 if (ntohs(h->ip6_plen) != 0) { 7001 DPFPRINTF(LOG_NOTICE, "IPv6 bad jumbo plen"); 7002 REASON_SET(reason, PFRES_IPOPTIONS); 7003 return (PF_DROP); 7004 } 7005 if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), 7006 reason, AF_INET6)) { 7007 DPFPRINTF(LOG_NOTICE, "IPv6 short jumbo"); 7008 return (PF_DROP); 7009 } 7010 memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len, 7011 sizeof(pd->jumbolen)); 7012 pd->jumbolen = ntohl(pd->jumbolen); 7013 if (pd->jumbolen < IPV6_MAXPACKET) { 7014 DPFPRINTF(LOG_NOTICE, "IPv6 short jumbolen"); 7015 REASON_SET(reason, PFRES_IPOPTIONS); 7016 return (PF_DROP); 7017 } 7018 break; 7019 case IP6OPT_ROUTER_ALERT: 7020 SET(pd->badopts, PF_OPT_ROUTER_ALERT); 7021 break; 7022 default: 7023 SET(pd->badopts, PF_OPT_OTHER); 7024 break; 7025 } 7026 off += sizeof(opt) + opt.ip6o_len; 7027 } 7028 7029 return (PF_PASS); 7030 } 7031 7032 int 7033 pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason) 7034 { 7035 struct ip6_frag frag; 7036 struct ip6_ext ext; 7037 struct icmp6_hdr icmp6; 7038 struct ip6_rthdr rthdr; 7039 u_int32_t end; 7040 int hdr_cnt, fraghdr_cnt = 0, rthdr_cnt = 0; 7041 7042 pd->off += sizeof(struct ip6_hdr); 7043 end = pd->off + ntohs(h->ip6_plen); 7044 pd->fragoff = pd->extoff = pd->jumbolen = 0; 7045 pd->proto = h->ip6_nxt; 7046 7047 for (hdr_cnt = 0; hdr_cnt < pf_hdr_limit; hdr_cnt++) { 7048 switch (pd->proto) { 7049 case IPPROTO_ROUTING: 7050 case IPPROTO_DSTOPTS: 7051 SET(pd->badopts, PF_OPT_OTHER); 7052 break; 7053 case IPPROTO_HOPOPTS: 7054 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 7055 reason, AF_INET6)) { 7056 DPFPRINTF(LOG_NOTICE, "IPv6 short exthdr"); 7057 return (PF_DROP); 7058 } 7059 if (pf_walk_option6(pd, h, pd->off + sizeof(ext), 7060 pd->off + (ext.ip6e_len + 1) * 8, reason) 7061 != PF_PASS) 7062 return (PF_DROP); 7063 /* option header which contains only padding is fishy */ 7064 if (pd->badopts == 0) 7065 SET(pd->badopts, PF_OPT_OTHER); 7066 break; 7067 } 7068 switch (pd->proto) { 7069 case IPPROTO_FRAGMENT: 7070 if (fraghdr_cnt++) { 7071 DPFPRINTF(LOG_NOTICE, "IPv6 multiple fragment"); 7072 REASON_SET(reason, PFRES_FRAG); 7073 return (PF_DROP); 7074 } 7075 /* jumbo payload packets cannot be fragmented */ 7076 if (pd->jumbolen != 0) { 7077 DPFPRINTF(LOG_NOTICE, "IPv6 fragmented jumbo"); 7078 REASON_SET(reason, PFRES_FRAG); 7079 return (PF_DROP); 7080 } 7081 if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag), 7082 reason, AF_INET6)) { 7083 DPFPRINTF(LOG_NOTICE, "IPv6 short fragment"); 7084 return (PF_DROP); 7085 } 7086 /* stop walking over non initial fragments */ 7087 if (ntohs((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) { 7088 pd->fragoff = pd->off; 7089 return (PF_PASS); 7090 } 7091 /* RFC6946: reassemble only non atomic fragments */ 7092 if (frag.ip6f_offlg & IP6F_MORE_FRAG) 7093 pd->fragoff = pd->off; 7094 pd->off += sizeof(frag); 7095 pd->proto = frag.ip6f_nxt; 7096 break; 7097 case IPPROTO_ROUTING: 7098 if (rthdr_cnt++) { 7099 DPFPRINTF(LOG_NOTICE, "IPv6 multiple rthdr"); 7100 REASON_SET(reason, PFRES_IPOPTIONS); 7101 return (PF_DROP); 7102 } 7103 /* fragments may be short */ 7104 if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) { 7105 pd->off = pd->fragoff; 7106 pd->proto = IPPROTO_FRAGMENT; 7107 return (PF_PASS); 7108 } 7109 if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr), 7110 reason, AF_INET6)) { 7111 DPFPRINTF(LOG_NOTICE, "IPv6 short rthdr"); 7112 return (PF_DROP); 7113 } 7114 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 7115 DPFPRINTF(LOG_NOTICE, "IPv6 rthdr0"); 7116 REASON_SET(reason, PFRES_IPOPTIONS); 7117 return (PF_DROP); 7118 } 7119 /* FALLTHROUGH */ 7120 case IPPROTO_HOPOPTS: 7121 /* RFC2460 4.1: Hop-by-Hop only after IPv6 header */ 7122 if (pd->proto == IPPROTO_HOPOPTS && hdr_cnt > 0) { 7123 DPFPRINTF(LOG_NOTICE, "IPv6 hopopts not first"); 7124 REASON_SET(reason, PFRES_IPOPTIONS); 7125 return (PF_DROP); 7126 } 7127 /* FALLTHROUGH */ 7128 case IPPROTO_AH: 7129 case IPPROTO_DSTOPTS: 7130 /* fragments may be short */ 7131 if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) { 7132 pd->off = pd->fragoff; 7133 pd->proto = IPPROTO_FRAGMENT; 7134 return (PF_PASS); 7135 } 7136 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 7137 reason, AF_INET6)) { 7138 DPFPRINTF(LOG_NOTICE, "IPv6 short exthdr"); 7139 return (PF_DROP); 7140 } 7141 /* reassembly needs the ext header before the frag */ 7142 if (pd->fragoff == 0) 7143 pd->extoff = pd->off; 7144 if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0 && 7145 ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) { 7146 DPFPRINTF(LOG_NOTICE, "IPv6 missing jumbo"); 7147 REASON_SET(reason, PFRES_IPOPTIONS); 7148 return (PF_DROP); 7149 } 7150 if (pd->proto == IPPROTO_AH) 7151 pd->off += (ext.ip6e_len + 2) * 4; 7152 else 7153 pd->off += (ext.ip6e_len + 1) * 8; 7154 pd->proto = ext.ip6e_nxt; 7155 break; 7156 case IPPROTO_ICMPV6: 7157 /* fragments may be short, ignore inner header then */ 7158 if (pd->fragoff != 0 && end < pd->off + sizeof(icmp6)) { 7159 pd->off = pd->fragoff; 7160 pd->proto = IPPROTO_FRAGMENT; 7161 return (PF_PASS); 7162 } 7163 if (!pf_pull_hdr(pd->m, pd->off, &icmp6, sizeof(icmp6), 7164 reason, AF_INET6)) { 7165 DPFPRINTF(LOG_NOTICE, "IPv6 short icmp6hdr"); 7166 return (PF_DROP); 7167 } 7168 /* ICMP multicast packets have router alert options */ 7169 switch (icmp6.icmp6_type) { 7170 case MLD_LISTENER_QUERY: 7171 case MLD_LISTENER_REPORT: 7172 case MLD_LISTENER_DONE: 7173 case MLDV2_LISTENER_REPORT: 7174 /* 7175 * According to RFC 2710 all MLD messages are 7176 * sent with hop-limit (ttl) set to 1, and link 7177 * local source address. If either one is 7178 * missing then MLD message is invalid and 7179 * should be discarded. 7180 */ 7181 if ((h->ip6_hlim != 1) || 7182 !IN6_IS_ADDR_LINKLOCAL(&h->ip6_src)) { 7183 DPFPRINTF(LOG_NOTICE, "Invalid MLD"); 7184 REASON_SET(reason, PFRES_IPOPTIONS); 7185 return (PF_DROP); 7186 } 7187 CLR(pd->badopts, PF_OPT_ROUTER_ALERT); 7188 break; 7189 } 7190 return (PF_PASS); 7191 case IPPROTO_TCP: 7192 case IPPROTO_UDP: 7193 /* fragments may be short, ignore inner header then */ 7194 if (pd->fragoff != 0 && end < pd->off + 7195 (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) : 7196 pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) : 7197 sizeof(struct icmp6_hdr))) { 7198 pd->off = pd->fragoff; 7199 pd->proto = IPPROTO_FRAGMENT; 7200 } 7201 /* FALLTHROUGH */ 7202 default: 7203 return (PF_PASS); 7204 } 7205 } 7206 DPFPRINTF(LOG_NOTICE, "IPv6 nested extension header limit"); 7207 REASON_SET(reason, PFRES_IPOPTIONS); 7208 return (PF_DROP); 7209 } 7210 #endif /* INET6 */ 7211 7212 u_int16_t 7213 pf_pkt_hash(sa_family_t af, uint8_t proto, 7214 const struct pf_addr *src, const struct pf_addr *dst, 7215 uint16_t sport, uint16_t dport) 7216 { 7217 uint32_t hash; 7218 7219 hash = src->addr32[0] ^ dst->addr32[0]; 7220 #ifdef INET6 7221 if (af == AF_INET6) { 7222 hash ^= src->addr32[1] ^ dst->addr32[1]; 7223 hash ^= src->addr32[2] ^ dst->addr32[2]; 7224 hash ^= src->addr32[3] ^ dst->addr32[3]; 7225 } 7226 #endif 7227 7228 switch (proto) { 7229 case IPPROTO_TCP: 7230 case IPPROTO_UDP: 7231 hash ^= sport ^ dport; 7232 break; 7233 } 7234 7235 return stoeplitz_n32(hash); 7236 } 7237 7238 int 7239 pf_setup_pdesc(struct pf_pdesc *pd, sa_family_t af, int dir, 7240 struct pfi_kif *kif, struct mbuf *m, u_short *reason) 7241 { 7242 memset(pd, 0, sizeof(*pd)); 7243 pd->dir = dir; 7244 pd->kif = kif; /* kif is NULL when called by pflog */ 7245 pd->m = m; 7246 pd->sidx = (dir == PF_IN) ? 0 : 1; 7247 pd->didx = (dir == PF_IN) ? 1 : 0; 7248 pd->af = pd->naf = af; 7249 pd->rdomain = rtable_l2(pd->m->m_pkthdr.ph_rtableid); 7250 7251 switch (pd->af) { 7252 case AF_INET: { 7253 struct ip *h; 7254 7255 /* Check for illegal packets */ 7256 if (pd->m->m_pkthdr.len < (int)sizeof(struct ip)) { 7257 REASON_SET(reason, PFRES_SHORT); 7258 return (PF_DROP); 7259 } 7260 7261 h = mtod(pd->m, struct ip *); 7262 if (pd->m->m_pkthdr.len < ntohs(h->ip_len)) { 7263 REASON_SET(reason, PFRES_SHORT); 7264 return (PF_DROP); 7265 } 7266 7267 if (pf_walk_header(pd, h, reason) != PF_PASS) 7268 return (PF_DROP); 7269 7270 pd->src = (struct pf_addr *)&h->ip_src; 7271 pd->dst = (struct pf_addr *)&h->ip_dst; 7272 pd->tot_len = ntohs(h->ip_len); 7273 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK; 7274 pd->ttl = h->ip_ttl; 7275 pd->virtual_proto = (h->ip_off & htons(IP_MF | IP_OFFMASK)) ? 7276 PF_VPROTO_FRAGMENT : pd->proto; 7277 7278 break; 7279 } 7280 #ifdef INET6 7281 case AF_INET6: { 7282 struct ip6_hdr *h; 7283 7284 /* Check for illegal packets */ 7285 if (pd->m->m_pkthdr.len < (int)sizeof(struct ip6_hdr)) { 7286 REASON_SET(reason, PFRES_SHORT); 7287 return (PF_DROP); 7288 } 7289 7290 h = mtod(pd->m, struct ip6_hdr *); 7291 if (pd->m->m_pkthdr.len < 7292 sizeof(struct ip6_hdr) + ntohs(h->ip6_plen)) { 7293 REASON_SET(reason, PFRES_SHORT); 7294 return (PF_DROP); 7295 } 7296 7297 if (pf_walk_header6(pd, h, reason) != PF_PASS) 7298 return (PF_DROP); 7299 7300 #if 1 7301 /* 7302 * we do not support jumbogram yet. if we keep going, zero 7303 * ip6_plen will do something bad, so drop the packet for now. 7304 */ 7305 if (pd->jumbolen != 0) { 7306 REASON_SET(reason, PFRES_NORM); 7307 return (PF_DROP); 7308 } 7309 #endif /* 1 */ 7310 7311 pd->src = (struct pf_addr *)&h->ip6_src; 7312 pd->dst = (struct pf_addr *)&h->ip6_dst; 7313 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 7314 pd->tos = (ntohl(h->ip6_flow) & 0x0fc00000) >> 20; 7315 pd->ttl = h->ip6_hlim; 7316 pd->virtual_proto = (pd->fragoff != 0) ? 7317 PF_VPROTO_FRAGMENT : pd->proto; 7318 7319 break; 7320 } 7321 #endif /* INET6 */ 7322 default: 7323 panic("pf_setup_pdesc called with illegal af %u", pd->af); 7324 7325 } 7326 7327 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 7328 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 7329 7330 switch (pd->virtual_proto) { 7331 case IPPROTO_TCP: { 7332 struct tcphdr *th = &pd->hdr.tcp; 7333 7334 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), 7335 reason, pd->af)) 7336 return (PF_DROP); 7337 pd->hdrlen = sizeof(*th); 7338 if (th->th_dport == 0 || 7339 pd->off + (th->th_off << 2) > pd->tot_len || 7340 (th->th_off << 2) < sizeof(struct tcphdr)) { 7341 REASON_SET(reason, PFRES_SHORT); 7342 return (PF_DROP); 7343 } 7344 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 7345 pd->sport = &th->th_sport; 7346 pd->dport = &th->th_dport; 7347 pd->pcksum = &th->th_sum; 7348 break; 7349 } 7350 case IPPROTO_UDP: { 7351 struct udphdr *uh = &pd->hdr.udp; 7352 7353 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), 7354 reason, pd->af)) 7355 return (PF_DROP); 7356 pd->hdrlen = sizeof(*uh); 7357 if (uh->uh_dport == 0 || 7358 pd->off + ntohs(uh->uh_ulen) > pd->tot_len || 7359 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 7360 REASON_SET(reason, PFRES_SHORT); 7361 return (PF_DROP); 7362 } 7363 pd->sport = &uh->uh_sport; 7364 pd->dport = &uh->uh_dport; 7365 pd->pcksum = &uh->uh_sum; 7366 break; 7367 } 7368 case IPPROTO_ICMP: { 7369 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 7370 reason, pd->af)) 7371 return (PF_DROP); 7372 pd->hdrlen = ICMP_MINLEN; 7373 if (pd->off + pd->hdrlen > pd->tot_len) { 7374 REASON_SET(reason, PFRES_SHORT); 7375 return (PF_DROP); 7376 } 7377 pd->pcksum = &pd->hdr.icmp.icmp_cksum; 7378 break; 7379 } 7380 #ifdef INET6 7381 case IPPROTO_ICMPV6: { 7382 size_t icmp_hlen = sizeof(struct icmp6_hdr); 7383 7384 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 7385 reason, pd->af)) 7386 return (PF_DROP); 7387 /* ICMP headers we look further into to match state */ 7388 switch (pd->hdr.icmp6.icmp6_type) { 7389 case MLD_LISTENER_QUERY: 7390 case MLD_LISTENER_REPORT: 7391 icmp_hlen = sizeof(struct mld_hdr); 7392 break; 7393 case ND_NEIGHBOR_SOLICIT: 7394 case ND_NEIGHBOR_ADVERT: 7395 icmp_hlen = sizeof(struct nd_neighbor_solicit); 7396 /* FALLTHROUGH */ 7397 case ND_ROUTER_SOLICIT: 7398 case ND_ROUTER_ADVERT: 7399 case ND_REDIRECT: 7400 if (pd->ttl != 255) { 7401 REASON_SET(reason, PFRES_NORM); 7402 return (PF_DROP); 7403 } 7404 break; 7405 } 7406 if (icmp_hlen > sizeof(struct icmp6_hdr) && 7407 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 7408 reason, pd->af)) 7409 return (PF_DROP); 7410 pd->hdrlen = icmp_hlen; 7411 if (pd->off + pd->hdrlen > pd->tot_len) { 7412 REASON_SET(reason, PFRES_SHORT); 7413 return (PF_DROP); 7414 } 7415 pd->pcksum = &pd->hdr.icmp6.icmp6_cksum; 7416 break; 7417 } 7418 #endif /* INET6 */ 7419 } 7420 7421 if (pd->sport) 7422 pd->osport = pd->nsport = *pd->sport; 7423 if (pd->dport) 7424 pd->odport = pd->ndport = *pd->dport; 7425 7426 pd->hash = pf_pkt_hash(pd->af, pd->proto, 7427 pd->src, pd->dst, pd->osport, pd->odport); 7428 7429 return (PF_PASS); 7430 } 7431 7432 void 7433 pf_counters_inc(int action, struct pf_pdesc *pd, struct pf_state *st, 7434 struct pf_rule *r, struct pf_rule *a) 7435 { 7436 int dirndx; 7437 pd->kif->pfik_bytes[pd->af == AF_INET6][pd->dir == PF_OUT] 7438 [action != PF_PASS] += pd->tot_len; 7439 pd->kif->pfik_packets[pd->af == AF_INET6][pd->dir == PF_OUT] 7440 [action != PF_PASS]++; 7441 7442 if (action == PF_PASS || action == PF_AFRT || r->action == PF_DROP) { 7443 dirndx = (pd->dir == PF_OUT); 7444 r->packets[dirndx]++; 7445 r->bytes[dirndx] += pd->tot_len; 7446 if (a != NULL) { 7447 a->packets[dirndx]++; 7448 a->bytes[dirndx] += pd->tot_len; 7449 } 7450 if (st != NULL) { 7451 struct pf_rule_item *ri; 7452 struct pf_sn_item *sni; 7453 7454 SLIST_FOREACH(sni, &st->src_nodes, next) { 7455 sni->sn->packets[dirndx]++; 7456 sni->sn->bytes[dirndx] += pd->tot_len; 7457 } 7458 dirndx = (pd->dir == st->direction) ? 0 : 1; 7459 st->packets[dirndx]++; 7460 st->bytes[dirndx] += pd->tot_len; 7461 7462 SLIST_FOREACH(ri, &st->match_rules, entry) { 7463 ri->r->packets[dirndx]++; 7464 ri->r->bytes[dirndx] += pd->tot_len; 7465 7466 if (ri->r->src.addr.type == PF_ADDR_TABLE) 7467 pfr_update_stats(ri->r->src.addr.p.tbl, 7468 &st->key[(st->direction == PF_IN)]-> 7469 addr[(st->direction == PF_OUT)], 7470 pd, ri->r->action, ri->r->src.neg); 7471 if (ri->r->dst.addr.type == PF_ADDR_TABLE) 7472 pfr_update_stats(ri->r->dst.addr.p.tbl, 7473 &st->key[(st->direction == PF_IN)]-> 7474 addr[(st->direction == PF_IN)], 7475 pd, ri->r->action, ri->r->dst.neg); 7476 } 7477 } 7478 if (r->src.addr.type == PF_ADDR_TABLE) 7479 pfr_update_stats(r->src.addr.p.tbl, 7480 (st == NULL) ? pd->src : 7481 &st->key[(st->direction == PF_IN)]-> 7482 addr[(st->direction == PF_OUT)], 7483 pd, r->action, r->src.neg); 7484 if (r->dst.addr.type == PF_ADDR_TABLE) 7485 pfr_update_stats(r->dst.addr.p.tbl, 7486 (st == NULL) ? pd->dst : 7487 &st->key[(st->direction == PF_IN)]-> 7488 addr[(st->direction == PF_IN)], 7489 pd, r->action, r->dst.neg); 7490 } 7491 } 7492 7493 int 7494 pf_test(sa_family_t af, int fwdir, struct ifnet *ifp, struct mbuf **m0) 7495 { 7496 #if NCARP > 0 7497 struct ifnet *ifp0; 7498 #endif 7499 struct pfi_kif *kif; 7500 u_short action, reason = 0; 7501 struct pf_rule *a = NULL, *r = &pf_default_rule; 7502 struct pf_state *st = NULL; 7503 struct pf_state_key_cmp key; 7504 struct pf_ruleset *ruleset = NULL; 7505 struct pf_pdesc pd; 7506 int dir = (fwdir == PF_FWD) ? PF_OUT : fwdir; 7507 u_int32_t qid, pqid = 0; 7508 int have_pf_lock = 0; 7509 7510 if (!pf_status.running) 7511 return (PF_PASS); 7512 7513 #if NCARP > 0 7514 if (ifp->if_type == IFT_CARP && 7515 (ifp0 = if_get(ifp->if_carpdevidx)) != NULL) { 7516 kif = (struct pfi_kif *)ifp0->if_pf_kif; 7517 if_put(ifp0); 7518 } else 7519 #endif /* NCARP */ 7520 kif = (struct pfi_kif *)ifp->if_pf_kif; 7521 7522 if (kif == NULL) { 7523 DPFPRINTF(LOG_ERR, 7524 "%s: kif == NULL, if_xname %s", __func__, ifp->if_xname); 7525 return (PF_DROP); 7526 } 7527 if (kif->pfik_flags & PFI_IFLAG_SKIP) 7528 return (PF_PASS); 7529 7530 #ifdef DIAGNOSTIC 7531 if (((*m0)->m_flags & M_PKTHDR) == 0) 7532 panic("non-M_PKTHDR is passed to pf_test"); 7533 #endif /* DIAGNOSTIC */ 7534 7535 if ((*m0)->m_pkthdr.pf.flags & PF_TAG_GENERATED) 7536 return (PF_PASS); 7537 7538 if ((*m0)->m_pkthdr.pf.flags & PF_TAG_DIVERTED_PACKET) { 7539 (*m0)->m_pkthdr.pf.flags &= ~PF_TAG_DIVERTED_PACKET; 7540 return (PF_PASS); 7541 } 7542 7543 if ((*m0)->m_pkthdr.pf.flags & PF_TAG_REFRAGMENTED) { 7544 (*m0)->m_pkthdr.pf.flags &= ~PF_TAG_REFRAGMENTED; 7545 return (PF_PASS); 7546 } 7547 7548 action = pf_setup_pdesc(&pd, af, dir, kif, *m0, &reason); 7549 if (action != PF_PASS) { 7550 #if NPFLOG > 0 7551 pd.pflog |= PF_LOG_FORCE; 7552 #endif /* NPFLOG > 0 */ 7553 goto done; 7554 } 7555 7556 /* packet normalization and reassembly */ 7557 switch (pd.af) { 7558 case AF_INET: 7559 action = pf_normalize_ip(&pd, &reason); 7560 break; 7561 #ifdef INET6 7562 case AF_INET6: 7563 action = pf_normalize_ip6(&pd, &reason); 7564 break; 7565 #endif /* INET6 */ 7566 } 7567 *m0 = pd.m; 7568 /* if packet sits in reassembly queue, return without error */ 7569 if (pd.m == NULL) 7570 return PF_PASS; 7571 7572 if (action != PF_PASS) { 7573 #if NPFLOG > 0 7574 pd.pflog |= PF_LOG_FORCE; 7575 #endif /* NPFLOG > 0 */ 7576 goto done; 7577 } 7578 7579 /* if packet has been reassembled, update packet description */ 7580 if (pf_status.reass && pd.virtual_proto == PF_VPROTO_FRAGMENT) { 7581 action = pf_setup_pdesc(&pd, af, dir, kif, pd.m, &reason); 7582 if (action != PF_PASS) { 7583 #if NPFLOG > 0 7584 pd.pflog |= PF_LOG_FORCE; 7585 #endif /* NPFLOG > 0 */ 7586 goto done; 7587 } 7588 } 7589 pd.m->m_pkthdr.pf.flags |= PF_TAG_PROCESSED; 7590 7591 /* 7592 * Avoid pcb-lookups from the forwarding path. They should never 7593 * match and would cause MP locking problems. 7594 */ 7595 if (fwdir == PF_FWD) { 7596 pd.lookup.done = -1; 7597 pd.lookup.uid = -1; 7598 pd.lookup.gid = -1; 7599 pd.lookup.pid = NO_PID; 7600 } 7601 7602 switch (pd.virtual_proto) { 7603 7604 case PF_VPROTO_FRAGMENT: { 7605 /* 7606 * handle fragments that aren't reassembled by 7607 * normalization 7608 */ 7609 PF_LOCK(); 7610 have_pf_lock = 1; 7611 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, &reason); 7612 st = pf_state_ref(st); 7613 if (action != PF_PASS) 7614 REASON_SET(&reason, PFRES_FRAG); 7615 break; 7616 } 7617 7618 case IPPROTO_ICMP: { 7619 if (pd.af != AF_INET) { 7620 action = PF_DROP; 7621 REASON_SET(&reason, PFRES_NORM); 7622 DPFPRINTF(LOG_NOTICE, 7623 "dropping IPv6 packet with ICMPv4 payload"); 7624 break; 7625 } 7626 PF_STATE_ENTER_READ(); 7627 action = pf_test_state_icmp(&pd, &st, &reason); 7628 st = pf_state_ref(st); 7629 PF_STATE_EXIT_READ(); 7630 if (action == PF_PASS || action == PF_AFRT) { 7631 #if NPFSYNC > 0 7632 pfsync_update_state(st); 7633 #endif /* NPFSYNC > 0 */ 7634 r = st->rule.ptr; 7635 a = st->anchor.ptr; 7636 #if NPFLOG > 0 7637 pd.pflog |= st->log; 7638 #endif /* NPFLOG > 0 */ 7639 } else if (st == NULL) { 7640 PF_LOCK(); 7641 have_pf_lock = 1; 7642 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, 7643 &reason); 7644 st = pf_state_ref(st); 7645 } 7646 break; 7647 } 7648 7649 #ifdef INET6 7650 case IPPROTO_ICMPV6: { 7651 if (pd.af != AF_INET6) { 7652 action = PF_DROP; 7653 REASON_SET(&reason, PFRES_NORM); 7654 DPFPRINTF(LOG_NOTICE, 7655 "dropping IPv4 packet with ICMPv6 payload"); 7656 break; 7657 } 7658 PF_STATE_ENTER_READ(); 7659 action = pf_test_state_icmp(&pd, &st, &reason); 7660 st = pf_state_ref(st); 7661 PF_STATE_EXIT_READ(); 7662 if (action == PF_PASS || action == PF_AFRT) { 7663 #if NPFSYNC > 0 7664 pfsync_update_state(st); 7665 #endif /* NPFSYNC > 0 */ 7666 r = st->rule.ptr; 7667 a = st->anchor.ptr; 7668 #if NPFLOG > 0 7669 pd.pflog |= st->log; 7670 #endif /* NPFLOG > 0 */ 7671 } else if (st == NULL) { 7672 PF_LOCK(); 7673 have_pf_lock = 1; 7674 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, 7675 &reason); 7676 st = pf_state_ref(st); 7677 } 7678 break; 7679 } 7680 #endif /* INET6 */ 7681 7682 default: 7683 if (pd.virtual_proto == IPPROTO_TCP) { 7684 if (pd.dir == PF_IN && (pd.hdr.tcp.th_flags & 7685 (TH_SYN|TH_ACK)) == TH_SYN && 7686 pf_synflood_check(&pd)) { 7687 PF_LOCK(); 7688 have_pf_lock = 1; 7689 pf_syncookie_send(&pd); 7690 action = PF_DROP; 7691 break; 7692 } 7693 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 7694 pqid = 1; 7695 action = pf_normalize_tcp(&pd); 7696 if (action == PF_DROP) 7697 break; 7698 } 7699 7700 key.af = pd.af; 7701 key.proto = pd.virtual_proto; 7702 key.rdomain = pd.rdomain; 7703 pf_addrcpy(&key.addr[pd.sidx], pd.src, key.af); 7704 pf_addrcpy(&key.addr[pd.didx], pd.dst, key.af); 7705 key.port[pd.sidx] = pd.osport; 7706 key.port[pd.didx] = pd.odport; 7707 key.hash = pd.hash; 7708 7709 PF_STATE_ENTER_READ(); 7710 action = pf_find_state(&pd, &key, &st); 7711 st = pf_state_ref(st); 7712 PF_STATE_EXIT_READ(); 7713 7714 /* check for syncookies if tcp ack and no active state */ 7715 if (pd.dir == PF_IN && pd.virtual_proto == IPPROTO_TCP && 7716 (st == NULL || (st->src.state >= TCPS_FIN_WAIT_2 && 7717 st->dst.state >= TCPS_FIN_WAIT_2)) && 7718 (pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 7719 pf_syncookie_validate(&pd)) { 7720 struct mbuf *msyn = pf_syncookie_recreate_syn(&pd); 7721 if (msyn) { 7722 action = pf_test(af, fwdir, ifp, &msyn); 7723 m_freem(msyn); 7724 if (action == PF_PASS || action == PF_AFRT) { 7725 PF_STATE_ENTER_READ(); 7726 pf_state_unref(st); 7727 action = pf_find_state(&pd, &key, &st); 7728 st = pf_state_ref(st); 7729 PF_STATE_EXIT_READ(); 7730 if (st == NULL) 7731 return (PF_DROP); 7732 st->src.seqhi = st->dst.seqhi = 7733 ntohl(pd.hdr.tcp.th_ack) - 1; 7734 st->src.seqlo = 7735 ntohl(pd.hdr.tcp.th_seq) - 1; 7736 pf_set_protostate(st, PF_PEER_SRC, 7737 PF_TCPS_PROXY_DST); 7738 } 7739 } else 7740 action = PF_DROP; 7741 } 7742 7743 if (action == PF_MATCH) 7744 action = pf_test_state(&pd, &st, &reason); 7745 7746 if (action == PF_PASS || action == PF_AFRT) { 7747 #if NPFSYNC > 0 7748 pfsync_update_state(st); 7749 #endif /* NPFSYNC > 0 */ 7750 r = st->rule.ptr; 7751 a = st->anchor.ptr; 7752 #if NPFLOG > 0 7753 pd.pflog |= st->log; 7754 #endif /* NPFLOG > 0 */ 7755 } else if (st == NULL) { 7756 PF_LOCK(); 7757 have_pf_lock = 1; 7758 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, 7759 &reason); 7760 st = pf_state_ref(st); 7761 } 7762 7763 if (pd.virtual_proto == IPPROTO_TCP) { 7764 if (st) { 7765 if (st->max_mss) 7766 pf_normalize_mss(&pd, st->max_mss); 7767 } else if (r->max_mss) 7768 pf_normalize_mss(&pd, r->max_mss); 7769 } 7770 7771 break; 7772 } 7773 7774 if (have_pf_lock != 0) 7775 PF_UNLOCK(); 7776 7777 /* 7778 * At the moment, we rely on NET_LOCK() to prevent removal of items 7779 * we've collected above ('r', 'anchor' and 'ruleset'). They'll have 7780 * to be refcounted when NET_LOCK() is gone. 7781 */ 7782 7783 done: 7784 if (action != PF_DROP) { 7785 if (st) { 7786 /* The non-state case is handled in pf_test_rule() */ 7787 if (action == PF_PASS && pd.badopts != 0 && 7788 !(st->state_flags & PFSTATE_ALLOWOPTS)) { 7789 action = PF_DROP; 7790 REASON_SET(&reason, PFRES_IPOPTIONS); 7791 #if NPFLOG > 0 7792 pd.pflog |= PF_LOG_FORCE; 7793 #endif /* NPFLOG > 0 */ 7794 DPFPRINTF(LOG_NOTICE, "dropping packet with " 7795 "ip/ipv6 options in pf_test()"); 7796 } 7797 7798 pf_scrub(pd.m, st->state_flags, pd.af, st->min_ttl, 7799 st->set_tos); 7800 pf_tag_packet(pd.m, st->tag, st->rtableid[pd.didx]); 7801 if (pqid || (pd.tos & IPTOS_LOWDELAY)) { 7802 qid = st->pqid; 7803 if (st->state_flags & PFSTATE_SETPRIO) { 7804 pd.m->m_pkthdr.pf.prio = 7805 st->set_prio[1]; 7806 } 7807 } else { 7808 qid = st->qid; 7809 if (st->state_flags & PFSTATE_SETPRIO) { 7810 pd.m->m_pkthdr.pf.prio = 7811 st->set_prio[0]; 7812 } 7813 } 7814 pd.m->m_pkthdr.pf.delay = st->delay; 7815 } else { 7816 pf_scrub(pd.m, r->scrub_flags, pd.af, r->min_ttl, 7817 r->set_tos); 7818 if (pqid || (pd.tos & IPTOS_LOWDELAY)) { 7819 qid = r->pqid; 7820 if (r->scrub_flags & PFSTATE_SETPRIO) 7821 pd.m->m_pkthdr.pf.prio = r->set_prio[1]; 7822 } else { 7823 qid = r->qid; 7824 if (r->scrub_flags & PFSTATE_SETPRIO) 7825 pd.m->m_pkthdr.pf.prio = r->set_prio[0]; 7826 } 7827 pd.m->m_pkthdr.pf.delay = r->delay; 7828 } 7829 } 7830 7831 if (action == PF_PASS && qid) 7832 pd.m->m_pkthdr.pf.qid = qid; 7833 if (pd.dir == PF_IN && st && st->key[PF_SK_STACK]) 7834 pf_mbuf_link_state_key(pd.m, st->key[PF_SK_STACK]); 7835 if (pd.dir == PF_OUT && 7836 pd.m->m_pkthdr.pf.inp && !pd.m->m_pkthdr.pf.inp->inp_pf_sk && 7837 st && st->key[PF_SK_STACK] && !st->key[PF_SK_STACK]->sk_inp) 7838 pf_state_key_link_inpcb(st->key[PF_SK_STACK], 7839 pd.m->m_pkthdr.pf.inp); 7840 7841 if (st != NULL && !ISSET(pd.m->m_pkthdr.csum_flags, M_FLOWID)) { 7842 pd.m->m_pkthdr.ph_flowid = st->key[PF_SK_WIRE]->hash; 7843 SET(pd.m->m_pkthdr.csum_flags, M_FLOWID); 7844 } 7845 7846 /* 7847 * connections redirected to loopback should not match sockets 7848 * bound specifically to loopback due to security implications, 7849 * see in_pcblookup_listen(). 7850 */ 7851 if (pd.destchg) 7852 if ((pd.af == AF_INET && (ntohl(pd.dst->v4.s_addr) >> 7853 IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) || 7854 (pd.af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))) 7855 pd.m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST; 7856 /* We need to redo the route lookup on outgoing routes. */ 7857 if (pd.destchg && pd.dir == PF_OUT) 7858 pd.m->m_pkthdr.pf.flags |= PF_TAG_REROUTE; 7859 7860 if (pd.dir == PF_IN && action == PF_PASS && 7861 (r->divert.type == PF_DIVERT_TO || 7862 r->divert.type == PF_DIVERT_REPLY)) { 7863 struct pf_divert *divert; 7864 7865 if ((divert = pf_get_divert(pd.m))) { 7866 pd.m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED; 7867 divert->addr = r->divert.addr; 7868 divert->port = r->divert.port; 7869 divert->rdomain = pd.rdomain; 7870 divert->type = r->divert.type; 7871 } 7872 } 7873 7874 if (action == PF_PASS && r->divert.type == PF_DIVERT_PACKET) 7875 action = PF_DIVERT; 7876 7877 #if NPFLOG > 0 7878 if (pd.pflog) { 7879 struct pf_rule_item *ri; 7880 7881 if (pd.pflog & PF_LOG_FORCE || r->log & PF_LOG_ALL) 7882 pflog_packet(&pd, reason, r, a, ruleset, NULL); 7883 if (st) { 7884 SLIST_FOREACH(ri, &st->match_rules, entry) 7885 if (ri->r->log & PF_LOG_ALL) 7886 pflog_packet(&pd, reason, ri->r, a, 7887 ruleset, NULL); 7888 } 7889 } 7890 #endif /* NPFLOG > 0 */ 7891 7892 pf_counters_inc(action, &pd, st, r, a); 7893 7894 switch (action) { 7895 case PF_SYNPROXY_DROP: 7896 m_freem(pd.m); 7897 /* FALLTHROUGH */ 7898 case PF_DEFER: 7899 pd.m = NULL; 7900 action = PF_PASS; 7901 break; 7902 case PF_DIVERT: 7903 switch (pd.af) { 7904 case AF_INET: 7905 divert_packet(pd.m, pd.dir, r->divert.port); 7906 pd.m = NULL; 7907 break; 7908 #ifdef INET6 7909 case AF_INET6: 7910 divert6_packet(pd.m, pd.dir, r->divert.port); 7911 pd.m = NULL; 7912 break; 7913 #endif /* INET6 */ 7914 } 7915 action = PF_PASS; 7916 break; 7917 #ifdef INET6 7918 case PF_AFRT: 7919 if (pf_translate_af(&pd)) { 7920 action = PF_DROP; 7921 break; 7922 } 7923 pd.m->m_pkthdr.pf.flags |= PF_TAG_GENERATED; 7924 switch (pd.naf) { 7925 case AF_INET: 7926 if (pd.dir == PF_IN) { 7927 if (ipforwarding == 0) { 7928 ipstat_inc(ips_cantforward); 7929 action = PF_DROP; 7930 break; 7931 } 7932 ip_forward(pd.m, ifp, NULL, 1); 7933 } else 7934 ip_output(pd.m, NULL, NULL, 0, NULL, NULL, 0); 7935 break; 7936 case AF_INET6: 7937 if (pd.dir == PF_IN) { 7938 if (ip6_forwarding == 0) { 7939 ip6stat_inc(ip6s_cantforward); 7940 action = PF_DROP; 7941 break; 7942 } 7943 ip6_forward(pd.m, NULL, 1); 7944 } else 7945 ip6_output(pd.m, NULL, NULL, 0, NULL, NULL); 7946 break; 7947 } 7948 if (action != PF_DROP) { 7949 pd.m = NULL; 7950 action = PF_PASS; 7951 } 7952 break; 7953 #endif /* INET6 */ 7954 case PF_DROP: 7955 m_freem(pd.m); 7956 pd.m = NULL; 7957 break; 7958 default: 7959 if (st && st->rt) { 7960 switch (pd.af) { 7961 case AF_INET: 7962 pf_route(&pd, st); 7963 break; 7964 #ifdef INET6 7965 case AF_INET6: 7966 pf_route6(&pd, st); 7967 break; 7968 #endif /* INET6 */ 7969 } 7970 } 7971 break; 7972 } 7973 7974 #ifdef INET6 7975 /* if reassembled packet passed, create new fragments */ 7976 if (pf_status.reass && action == PF_PASS && pd.m && fwdir == PF_FWD && 7977 pd.af == AF_INET6) { 7978 struct m_tag *mtag; 7979 7980 if ((mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL))) 7981 action = pf_refragment6(&pd.m, mtag, NULL, NULL, NULL); 7982 } 7983 #endif /* INET6 */ 7984 if (st && action != PF_DROP) { 7985 if (!st->if_index_in && dir == PF_IN) 7986 st->if_index_in = ifp->if_index; 7987 else if (!st->if_index_out && dir == PF_OUT) 7988 st->if_index_out = ifp->if_index; 7989 } 7990 7991 *m0 = pd.m; 7992 7993 pf_state_unref(st); 7994 7995 return (action); 7996 } 7997 7998 int 7999 pf_ouraddr(struct mbuf *m) 8000 { 8001 struct pf_state_key *sk; 8002 8003 if (m->m_pkthdr.pf.flags & PF_TAG_DIVERTED) 8004 return (1); 8005 8006 sk = m->m_pkthdr.pf.statekey; 8007 if (sk != NULL) { 8008 if (sk->sk_inp != NULL) 8009 return (1); 8010 } 8011 8012 return (-1); 8013 } 8014 8015 /* 8016 * must be called whenever any addressing information such as 8017 * address, port, protocol has changed 8018 */ 8019 void 8020 pf_pkt_addr_changed(struct mbuf *m) 8021 { 8022 pf_mbuf_unlink_state_key(m); 8023 pf_mbuf_unlink_inpcb(m); 8024 } 8025 8026 struct inpcb * 8027 pf_inp_lookup(struct mbuf *m) 8028 { 8029 struct inpcb *inp = NULL; 8030 struct pf_state_key *sk = m->m_pkthdr.pf.statekey; 8031 8032 if (!pf_state_key_isvalid(sk)) 8033 pf_mbuf_unlink_state_key(m); 8034 else 8035 inp = m->m_pkthdr.pf.statekey->sk_inp; 8036 8037 if (inp && inp->inp_pf_sk) 8038 KASSERT(m->m_pkthdr.pf.statekey == inp->inp_pf_sk); 8039 8040 in_pcbref(inp); 8041 return (inp); 8042 } 8043 8044 void 8045 pf_inp_link(struct mbuf *m, struct inpcb *inp) 8046 { 8047 struct pf_state_key *sk = m->m_pkthdr.pf.statekey; 8048 8049 if (!pf_state_key_isvalid(sk)) { 8050 pf_mbuf_unlink_state_key(m); 8051 return; 8052 } 8053 8054 /* 8055 * we don't need to grab PF-lock here. At worst case we link inp to 8056 * state, which might be just being marked as deleted by another 8057 * thread. 8058 */ 8059 if (inp && !sk->sk_inp && !inp->inp_pf_sk) 8060 pf_state_key_link_inpcb(sk, inp); 8061 8062 /* The statekey has finished finding the inp, it is no longer needed. */ 8063 pf_mbuf_unlink_state_key(m); 8064 } 8065 8066 void 8067 pf_inp_unlink(struct inpcb *inp) 8068 { 8069 pf_inpcb_unlink_state_key(inp); 8070 } 8071 8072 void 8073 pf_state_key_link_reverse(struct pf_state_key *sk, struct pf_state_key *skrev) 8074 { 8075 struct pf_state_key *old_reverse; 8076 8077 old_reverse = atomic_cas_ptr(&sk->sk_reverse, NULL, skrev); 8078 if (old_reverse != NULL) 8079 KASSERT(old_reverse == skrev); 8080 else { 8081 pf_state_key_ref(skrev); 8082 8083 /* 8084 * NOTE: if sk == skrev, then KASSERT() below holds true, we 8085 * still want to grab a reference in such case, because 8086 * pf_state_key_unlink_reverse() does not check whether keys 8087 * are identical or not. 8088 */ 8089 old_reverse = atomic_cas_ptr(&skrev->sk_reverse, NULL, sk); 8090 if (old_reverse != NULL) 8091 KASSERT(old_reverse == sk); 8092 8093 pf_state_key_ref(sk); 8094 } 8095 } 8096 8097 #if NPFLOG > 0 8098 void 8099 pf_log_matches(struct pf_pdesc *pd, struct pf_rule *rm, struct pf_rule *am, 8100 struct pf_ruleset *ruleset, struct pf_rule_slist *matchrules) 8101 { 8102 struct pf_rule_item *ri; 8103 8104 /* if this is the log(matches) rule, packet has been logged already */ 8105 if (rm->log & PF_LOG_MATCHES) 8106 return; 8107 8108 SLIST_FOREACH(ri, matchrules, entry) 8109 if (ri->r->log & PF_LOG_MATCHES) 8110 pflog_packet(pd, PFRES_MATCH, rm, am, ruleset, ri->r); 8111 } 8112 #endif /* NPFLOG > 0 */ 8113 8114 struct pf_state_key * 8115 pf_state_key_ref(struct pf_state_key *sk) 8116 { 8117 if (sk != NULL) 8118 PF_REF_TAKE(sk->sk_refcnt); 8119 8120 return (sk); 8121 } 8122 8123 void 8124 pf_state_key_unref(struct pf_state_key *sk) 8125 { 8126 if (PF_REF_RELE(sk->sk_refcnt)) { 8127 /* state key must be removed from tree */ 8128 KASSERT(!pf_state_key_isvalid(sk)); 8129 /* state key must be unlinked from reverse key */ 8130 KASSERT(sk->sk_reverse == NULL); 8131 /* state key must be unlinked from socket */ 8132 KASSERT(sk->sk_inp == NULL); 8133 pool_put(&pf_state_key_pl, sk); 8134 } 8135 } 8136 8137 int 8138 pf_state_key_isvalid(struct pf_state_key *sk) 8139 { 8140 return ((sk != NULL) && (sk->sk_removed == 0)); 8141 } 8142 8143 void 8144 pf_mbuf_link_state_key(struct mbuf *m, struct pf_state_key *sk) 8145 { 8146 KASSERT(m->m_pkthdr.pf.statekey == NULL); 8147 m->m_pkthdr.pf.statekey = pf_state_key_ref(sk); 8148 } 8149 8150 void 8151 pf_mbuf_unlink_state_key(struct mbuf *m) 8152 { 8153 struct pf_state_key *sk = m->m_pkthdr.pf.statekey; 8154 8155 if (sk != NULL) { 8156 m->m_pkthdr.pf.statekey = NULL; 8157 pf_state_key_unref(sk); 8158 } 8159 } 8160 8161 void 8162 pf_mbuf_link_inpcb(struct mbuf *m, struct inpcb *inp) 8163 { 8164 KASSERT(m->m_pkthdr.pf.inp == NULL); 8165 m->m_pkthdr.pf.inp = in_pcbref(inp); 8166 } 8167 8168 void 8169 pf_mbuf_unlink_inpcb(struct mbuf *m) 8170 { 8171 struct inpcb *inp = m->m_pkthdr.pf.inp; 8172 8173 if (inp != NULL) { 8174 m->m_pkthdr.pf.inp = NULL; 8175 in_pcbunref(inp); 8176 } 8177 } 8178 8179 void 8180 pf_state_key_link_inpcb(struct pf_state_key *sk, struct inpcb *inp) 8181 { 8182 KASSERT(sk->sk_inp == NULL); 8183 sk->sk_inp = in_pcbref(inp); 8184 KASSERT(inp->inp_pf_sk == NULL); 8185 inp->inp_pf_sk = pf_state_key_ref(sk); 8186 } 8187 8188 void 8189 pf_inpcb_unlink_state_key(struct inpcb *inp) 8190 { 8191 struct pf_state_key *sk = inp->inp_pf_sk; 8192 8193 if (sk != NULL) { 8194 KASSERT(sk->sk_inp == inp); 8195 sk->sk_inp = NULL; 8196 inp->inp_pf_sk = NULL; 8197 pf_state_key_unref(sk); 8198 in_pcbunref(inp); 8199 } 8200 } 8201 8202 void 8203 pf_state_key_unlink_inpcb(struct pf_state_key *sk) 8204 { 8205 struct inpcb *inp = sk->sk_inp; 8206 8207 if (inp != NULL) { 8208 KASSERT(inp->inp_pf_sk == sk); 8209 sk->sk_inp = NULL; 8210 inp->inp_pf_sk = NULL; 8211 pf_state_key_unref(sk); 8212 in_pcbunref(inp); 8213 } 8214 } 8215 8216 void 8217 pf_state_key_unlink_reverse(struct pf_state_key *sk) 8218 { 8219 struct pf_state_key *skrev = sk->sk_reverse; 8220 8221 /* Note that sk and skrev may be equal, then we unref twice. */ 8222 if (skrev != NULL) { 8223 KASSERT(skrev->sk_reverse == sk); 8224 sk->sk_reverse = NULL; 8225 skrev->sk_reverse = NULL; 8226 pf_state_key_unref(skrev); 8227 pf_state_key_unref(sk); 8228 } 8229 } 8230 8231 struct pf_state * 8232 pf_state_ref(struct pf_state *st) 8233 { 8234 if (st != NULL) 8235 PF_REF_TAKE(st->refcnt); 8236 return (st); 8237 } 8238 8239 void 8240 pf_state_unref(struct pf_state *st) 8241 { 8242 if ((st != NULL) && PF_REF_RELE(st->refcnt)) { 8243 /* never inserted or removed */ 8244 #if NPFSYNC > 0 8245 KASSERT((TAILQ_NEXT(st, sync_list) == NULL) || 8246 ((TAILQ_NEXT(st, sync_list) == _Q_INVALID) && 8247 (st->sync_state >= PFSYNC_S_NONE))); 8248 #endif /* NPFSYNC */ 8249 KASSERT((TAILQ_NEXT(st, entry_list) == NULL) || 8250 (TAILQ_NEXT(st, entry_list) == _Q_INVALID)); 8251 8252 pf_state_key_unref(st->key[PF_SK_WIRE]); 8253 pf_state_key_unref(st->key[PF_SK_STACK]); 8254 8255 pool_put(&pf_state_pl, st); 8256 } 8257 } 8258 8259 int 8260 pf_delay_pkt(struct mbuf *m, u_int ifidx) 8261 { 8262 struct pf_pktdelay *pdy; 8263 8264 if ((pdy = pool_get(&pf_pktdelay_pl, PR_NOWAIT)) == NULL) { 8265 m_freem(m); 8266 return (ENOBUFS); 8267 } 8268 pdy->ifidx = ifidx; 8269 pdy->m = m; 8270 timeout_set(&pdy->to, pf_pktenqueue_delayed, pdy); 8271 timeout_add_msec(&pdy->to, m->m_pkthdr.pf.delay); 8272 m->m_pkthdr.pf.delay = 0; 8273 return (0); 8274 } 8275 8276 void 8277 pf_pktenqueue_delayed(void *arg) 8278 { 8279 struct pf_pktdelay *pdy = arg; 8280 struct ifnet *ifp; 8281 8282 ifp = if_get(pdy->ifidx); 8283 if (ifp != NULL) { 8284 if_enqueue(ifp, pdy->m); 8285 if_put(ifp); 8286 } else 8287 m_freem(pdy->m); 8288 8289 pool_put(&pf_pktdelay_pl, pdy); 8290 } 8291