1 /* $OpenBSD: pf.c,v 1.1180 2023/05/15 16:34:56 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 #else 104 struct pfsync_deferral; 105 #endif /* NPFSYNC > 0 */ 106 107 /* 108 * Global variables 109 */ 110 struct pf_state_tree pf_statetbl; 111 struct pf_queuehead pf_queues[2]; 112 struct pf_queuehead *pf_queues_active; 113 struct pf_queuehead *pf_queues_inactive; 114 115 struct pf_status pf_status; 116 117 int pf_hdr_limit = 20; /* arbitrary limit, tune in ddb */ 118 119 SHA2_CTX pf_tcp_secret_ctx; 120 u_char pf_tcp_secret[16]; 121 int pf_tcp_secret_init; 122 int pf_tcp_iss_off; 123 124 int pf_npurge; 125 struct task pf_purge_task = TASK_INITIALIZER(pf_purge, &pf_npurge); 126 struct timeout pf_purge_to = TIMEOUT_INITIALIZER(pf_purge_timeout, NULL); 127 128 enum pf_test_status { 129 PF_TEST_FAIL = -1, 130 PF_TEST_OK, 131 PF_TEST_QUICK 132 }; 133 134 struct pf_test_ctx { 135 struct pf_pdesc *pd; 136 struct pf_rule_actions act; 137 u_int8_t icmpcode; 138 u_int8_t icmptype; 139 int icmp_dir; 140 int state_icmp; 141 int tag; 142 u_short reason; 143 struct pf_rule_item *ri; 144 struct pf_src_node *sns[PF_SN_MAX]; 145 struct pf_rule_slist rules; 146 struct pf_rule *nr; 147 struct pf_rule **rm; 148 struct pf_rule *a; 149 struct pf_rule **am; 150 struct pf_ruleset **rsm; 151 struct pf_ruleset *arsm; 152 struct pf_ruleset *aruleset; 153 struct tcphdr *th; 154 }; 155 156 struct pool pf_src_tree_pl, pf_rule_pl, pf_queue_pl; 157 struct pool pf_state_pl, pf_state_key_pl, pf_state_item_pl; 158 struct pool pf_rule_item_pl, pf_sn_item_pl, pf_pktdelay_pl; 159 160 void pf_add_threshold(struct pf_threshold *); 161 int pf_check_threshold(struct pf_threshold *); 162 int pf_check_tcp_cksum(struct mbuf *, int, int, 163 sa_family_t); 164 __inline void pf_cksum_fixup(u_int16_t *, u_int16_t, u_int16_t, 165 u_int8_t); 166 void pf_cksum_fixup_a(u_int16_t *, const struct pf_addr *, 167 const struct pf_addr *, sa_family_t, u_int8_t); 168 int pf_modulate_sack(struct pf_pdesc *, 169 struct pf_state_peer *); 170 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 171 u_int16_t *, u_int16_t *); 172 int pf_change_icmp_af(struct mbuf *, int, 173 struct pf_pdesc *, struct pf_pdesc *, 174 struct pf_addr *, struct pf_addr *, sa_family_t, 175 sa_family_t); 176 int pf_translate_a(struct pf_pdesc *, struct pf_addr *, 177 struct pf_addr *); 178 void pf_translate_icmp(struct pf_pdesc *, struct pf_addr *, 179 u_int16_t *, struct pf_addr *, struct pf_addr *, 180 u_int16_t); 181 int pf_translate_icmp_af(struct pf_pdesc*, int, void *); 182 void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, int, 183 sa_family_t, struct pf_rule *, u_int); 184 void pf_detach_state(struct pf_state *); 185 struct pf_state_key *pf_state_key_attach(struct pf_state_key *, 186 struct pf_state *, int); 187 void pf_state_key_detach(struct pf_state *, int); 188 u_int32_t pf_tcp_iss(struct pf_pdesc *); 189 void pf_rule_to_actions(struct pf_rule *, 190 struct pf_rule_actions *); 191 int pf_test_rule(struct pf_pdesc *, struct pf_rule **, 192 struct pf_state **, struct pf_rule **, 193 struct pf_ruleset **, u_short *, 194 struct pfsync_deferral **); 195 static __inline int pf_create_state(struct pf_pdesc *, struct pf_rule *, 196 struct pf_rule *, struct pf_rule *, 197 struct pf_state_key **, struct pf_state_key **, 198 int *, struct pf_state **, int, 199 struct pf_rule_slist *, struct pf_rule_actions *, 200 struct pf_src_node **); 201 static __inline int pf_state_key_addr_setup(struct pf_pdesc *, void *, 202 int, struct pf_addr *, int, struct pf_addr *, 203 int, int); 204 int pf_state_key_setup(struct pf_pdesc *, struct 205 pf_state_key **, struct pf_state_key **, int); 206 int pf_tcp_track_full(struct pf_pdesc *, 207 struct pf_state **, u_short *, int *, int); 208 int pf_tcp_track_sloppy(struct pf_pdesc *, 209 struct pf_state **, u_short *); 210 static __inline int pf_synproxy(struct pf_pdesc *, struct pf_state **, 211 u_short *); 212 int pf_test_state(struct pf_pdesc *, struct pf_state **, 213 u_short *); 214 int pf_icmp_state_lookup(struct pf_pdesc *, 215 struct pf_state_key_cmp *, struct pf_state **, 216 u_int16_t, u_int16_t, int, int *, int, int); 217 int pf_test_state_icmp(struct pf_pdesc *, 218 struct pf_state **, u_short *); 219 u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, int, 220 u_int16_t); 221 static __inline int pf_set_rt_ifp(struct pf_state *, struct pf_addr *, 222 sa_family_t, struct pf_src_node **); 223 struct pf_divert *pf_get_divert(struct mbuf *); 224 int pf_walk_option(struct pf_pdesc *, struct ip *, 225 int, int, u_short *); 226 int pf_walk_header(struct pf_pdesc *, struct ip *, 227 u_short *); 228 int pf_walk_option6(struct pf_pdesc *, struct ip6_hdr *, 229 int, int, u_short *); 230 int pf_walk_header6(struct pf_pdesc *, struct ip6_hdr *, 231 u_short *); 232 void pf_print_state_parts(struct pf_state *, 233 struct pf_state_key *, struct pf_state_key *); 234 int pf_addr_wrap_neq(struct pf_addr_wrap *, 235 struct pf_addr_wrap *); 236 int pf_compare_state_keys(struct pf_state_key *, 237 struct pf_state_key *, struct pfi_kif *, u_int); 238 u_int16_t pf_pkt_hash(sa_family_t, uint8_t, 239 const struct pf_addr *, const struct pf_addr *, 240 uint16_t, uint16_t); 241 int pf_find_state(struct pf_pdesc *, 242 struct pf_state_key_cmp *, struct pf_state **); 243 int pf_src_connlimit(struct pf_state **); 244 int pf_match_rcvif(struct mbuf *, struct pf_rule *); 245 int pf_step_into_anchor(struct pf_test_ctx *, 246 struct pf_rule *); 247 int pf_match_rule(struct pf_test_ctx *, 248 struct pf_ruleset *); 249 void pf_counters_inc(int, struct pf_pdesc *, 250 struct pf_state *, struct pf_rule *, 251 struct pf_rule *); 252 253 int pf_state_key_isvalid(struct pf_state_key *); 254 struct pf_state_key *pf_state_key_ref(struct pf_state_key *); 255 void pf_state_key_unref(struct pf_state_key *); 256 void pf_state_key_link_reverse(struct pf_state_key *, 257 struct pf_state_key *); 258 void pf_state_key_unlink_reverse(struct pf_state_key *); 259 void pf_state_key_link_inpcb(struct pf_state_key *, 260 struct inpcb *); 261 void pf_state_key_unlink_inpcb(struct pf_state_key *); 262 void pf_inpcb_unlink_state_key(struct inpcb *); 263 void pf_pktenqueue_delayed(void *); 264 int32_t pf_state_expires(const struct pf_state *, uint8_t); 265 266 #if NPFLOG > 0 267 void pf_log_matches(struct pf_pdesc *, struct pf_rule *, 268 struct pf_rule *, struct pf_ruleset *, 269 struct pf_rule_slist *); 270 #endif /* NPFLOG > 0 */ 271 272 extern struct pool pfr_ktable_pl; 273 extern struct pool pfr_kentry_pl; 274 275 struct pf_pool_limit pf_pool_limits[PF_LIMIT_MAX] = { 276 { &pf_state_pl, PFSTATE_HIWAT, PFSTATE_HIWAT }, 277 { &pf_src_tree_pl, PFSNODE_HIWAT, PFSNODE_HIWAT }, 278 { &pf_frent_pl, PFFRAG_FRENT_HIWAT, PFFRAG_FRENT_HIWAT }, 279 { &pfr_ktable_pl, PFR_KTABLE_HIWAT, PFR_KTABLE_HIWAT }, 280 { &pfr_kentry_pl, PFR_KENTRY_HIWAT, PFR_KENTRY_HIWAT }, 281 { &pf_pktdelay_pl, PF_PKTDELAY_MAXPKTS, PF_PKTDELAY_MAXPKTS }, 282 { &pf_anchor_pl, PF_ANCHOR_HIWAT, PF_ANCHOR_HIWAT } 283 }; 284 285 #define BOUND_IFACE(r, k) \ 286 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : pfi_all 287 288 #define STATE_INC_COUNTERS(s) \ 289 do { \ 290 struct pf_rule_item *mrm; \ 291 s->rule.ptr->states_cur++; \ 292 s->rule.ptr->states_tot++; \ 293 if (s->anchor.ptr != NULL) { \ 294 s->anchor.ptr->states_cur++; \ 295 s->anchor.ptr->states_tot++; \ 296 } \ 297 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 298 mrm->r->states_cur++; \ 299 } while (0) 300 301 static __inline int pf_src_compare(struct pf_src_node *, struct pf_src_node *); 302 static inline int pf_state_compare_key(const struct pf_state_key *, 303 const struct pf_state_key *); 304 static inline int pf_state_compare_id(const struct pf_state *, 305 const struct pf_state *); 306 #ifdef INET6 307 static __inline void pf_cksum_uncover(u_int16_t *, u_int16_t, u_int8_t); 308 static __inline void pf_cksum_cover(u_int16_t *, u_int16_t, u_int8_t); 309 #endif /* INET6 */ 310 static __inline void pf_set_protostate(struct pf_state *, int, u_int8_t); 311 312 struct pf_src_tree tree_src_tracking; 313 314 struct pf_state_tree_id tree_id; 315 struct pf_state_list pf_state_list = PF_STATE_LIST_INITIALIZER(pf_state_list); 316 317 RB_GENERATE(pf_src_tree, pf_src_node, entry, pf_src_compare); 318 RBT_GENERATE(pf_state_tree, pf_state_key, sk_entry, pf_state_compare_key); 319 RBT_GENERATE(pf_state_tree_id, pf_state, entry_id, pf_state_compare_id); 320 321 int 322 pf_addr_compare(const struct pf_addr *a, const struct pf_addr *b, 323 sa_family_t af) 324 { 325 switch (af) { 326 case AF_INET: 327 if (a->addr32[0] > b->addr32[0]) 328 return (1); 329 if (a->addr32[0] < b->addr32[0]) 330 return (-1); 331 break; 332 #ifdef INET6 333 case AF_INET6: 334 if (a->addr32[3] > b->addr32[3]) 335 return (1); 336 if (a->addr32[3] < b->addr32[3]) 337 return (-1); 338 if (a->addr32[2] > b->addr32[2]) 339 return (1); 340 if (a->addr32[2] < b->addr32[2]) 341 return (-1); 342 if (a->addr32[1] > b->addr32[1]) 343 return (1); 344 if (a->addr32[1] < b->addr32[1]) 345 return (-1); 346 if (a->addr32[0] > b->addr32[0]) 347 return (1); 348 if (a->addr32[0] < b->addr32[0]) 349 return (-1); 350 break; 351 #endif /* INET6 */ 352 } 353 return (0); 354 } 355 356 static __inline int 357 pf_src_compare(struct pf_src_node *a, struct pf_src_node *b) 358 { 359 int diff; 360 361 if (a->rule.ptr > b->rule.ptr) 362 return (1); 363 if (a->rule.ptr < b->rule.ptr) 364 return (-1); 365 if ((diff = a->type - b->type) != 0) 366 return (diff); 367 if ((diff = a->af - b->af) != 0) 368 return (diff); 369 if ((diff = pf_addr_compare(&a->addr, &b->addr, a->af)) != 0) 370 return (diff); 371 return (0); 372 } 373 374 static __inline void 375 pf_set_protostate(struct pf_state *st, int which, u_int8_t newstate) 376 { 377 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 378 st->dst.state = newstate; 379 if (which == PF_PEER_DST) 380 return; 381 382 if (st->src.state == newstate) 383 return; 384 if (st->creatorid == pf_status.hostid && 385 st->key[PF_SK_STACK]->proto == IPPROTO_TCP && 386 !(TCPS_HAVEESTABLISHED(st->src.state) || 387 st->src.state == TCPS_CLOSED) && 388 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 389 pf_status.states_halfopen--; 390 391 st->src.state = newstate; 392 } 393 394 void 395 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 396 { 397 switch (af) { 398 case AF_INET: 399 dst->addr32[0] = src->addr32[0]; 400 break; 401 #ifdef INET6 402 case AF_INET6: 403 dst->addr32[0] = src->addr32[0]; 404 dst->addr32[1] = src->addr32[1]; 405 dst->addr32[2] = src->addr32[2]; 406 dst->addr32[3] = src->addr32[3]; 407 break; 408 #endif /* INET6 */ 409 default: 410 unhandled_af(af); 411 } 412 } 413 414 void 415 pf_init_threshold(struct pf_threshold *threshold, 416 u_int32_t limit, u_int32_t seconds) 417 { 418 threshold->limit = limit * PF_THRESHOLD_MULT; 419 threshold->seconds = seconds; 420 threshold->count = 0; 421 threshold->last = getuptime(); 422 } 423 424 void 425 pf_add_threshold(struct pf_threshold *threshold) 426 { 427 u_int32_t t = getuptime(), diff = t - threshold->last; 428 429 if (diff >= threshold->seconds) 430 threshold->count = 0; 431 else 432 threshold->count -= threshold->count * diff / 433 threshold->seconds; 434 threshold->count += PF_THRESHOLD_MULT; 435 threshold->last = t; 436 } 437 438 int 439 pf_check_threshold(struct pf_threshold *threshold) 440 { 441 return (threshold->count > threshold->limit); 442 } 443 444 void 445 pf_state_list_insert(struct pf_state_list *pfs, struct pf_state *st) 446 { 447 /* 448 * we can always put states on the end of the list. 449 * 450 * things reading the list should take a read lock, then 451 * the mutex, get the head and tail pointers, release the 452 * mutex, and then they can iterate between the head and tail. 453 */ 454 455 pf_state_ref(st); /* get a ref for the list */ 456 457 mtx_enter(&pfs->pfs_mtx); 458 TAILQ_INSERT_TAIL(&pfs->pfs_list, st, entry_list); 459 mtx_leave(&pfs->pfs_mtx); 460 } 461 462 void 463 pf_state_list_remove(struct pf_state_list *pfs, struct pf_state *st) 464 { 465 /* states can only be removed when the write lock is held */ 466 rw_assert_wrlock(&pfs->pfs_rwl); 467 468 mtx_enter(&pfs->pfs_mtx); 469 TAILQ_REMOVE(&pfs->pfs_list, st, entry_list); 470 mtx_leave(&pfs->pfs_mtx); 471 472 pf_state_unref(st); /* list no longer references the state */ 473 } 474 475 int 476 pf_src_connlimit(struct pf_state **stp) 477 { 478 int bad = 0; 479 struct pf_src_node *sn; 480 481 if ((sn = pf_get_src_node((*stp), PF_SN_NONE)) == NULL) 482 return (0); 483 484 sn->conn++; 485 (*stp)->src.tcp_est = 1; 486 pf_add_threshold(&sn->conn_rate); 487 488 if ((*stp)->rule.ptr->max_src_conn && 489 (*stp)->rule.ptr->max_src_conn < sn->conn) { 490 pf_status.lcounters[LCNT_SRCCONN]++; 491 bad++; 492 } 493 494 if ((*stp)->rule.ptr->max_src_conn_rate.limit && 495 pf_check_threshold(&sn->conn_rate)) { 496 pf_status.lcounters[LCNT_SRCCONNRATE]++; 497 bad++; 498 } 499 500 if (!bad) 501 return (0); 502 503 if ((*stp)->rule.ptr->overload_tbl) { 504 struct pfr_addr p; 505 u_int32_t killed = 0; 506 507 pf_status.lcounters[LCNT_OVERLOAD_TABLE]++; 508 if (pf_status.debug >= LOG_NOTICE) { 509 log(LOG_NOTICE, 510 "pf: pf_src_connlimit: blocking address "); 511 pf_print_host(&sn->addr, 0, 512 (*stp)->key[PF_SK_WIRE]->af); 513 } 514 515 memset(&p, 0, sizeof(p)); 516 p.pfra_af = (*stp)->key[PF_SK_WIRE]->af; 517 switch ((*stp)->key[PF_SK_WIRE]->af) { 518 case AF_INET: 519 p.pfra_net = 32; 520 p.pfra_ip4addr = sn->addr.v4; 521 break; 522 #ifdef INET6 523 case AF_INET6: 524 p.pfra_net = 128; 525 p.pfra_ip6addr = sn->addr.v6; 526 break; 527 #endif /* INET6 */ 528 } 529 530 pfr_insert_kentry((*stp)->rule.ptr->overload_tbl, 531 &p, gettime()); 532 533 /* kill existing states if that's required. */ 534 if ((*stp)->rule.ptr->flush) { 535 struct pf_state_key *sk; 536 struct pf_state *st; 537 538 pf_status.lcounters[LCNT_OVERLOAD_FLUSH]++; 539 RBT_FOREACH(st, pf_state_tree_id, &tree_id) { 540 sk = st->key[PF_SK_WIRE]; 541 /* 542 * Kill states from this source. (Only those 543 * from the same rule if PF_FLUSH_GLOBAL is not 544 * set) 545 */ 546 if (sk->af == 547 (*stp)->key[PF_SK_WIRE]->af && 548 (((*stp)->direction == PF_OUT && 549 PF_AEQ(&sn->addr, &sk->addr[1], sk->af)) || 550 ((*stp)->direction == PF_IN && 551 PF_AEQ(&sn->addr, &sk->addr[0], sk->af))) && 552 ((*stp)->rule.ptr->flush & 553 PF_FLUSH_GLOBAL || 554 (*stp)->rule.ptr == st->rule.ptr)) { 555 st->timeout = PFTM_PURGE; 556 pf_set_protostate(st, PF_PEER_BOTH, 557 TCPS_CLOSED); 558 killed++; 559 } 560 } 561 if (pf_status.debug >= LOG_NOTICE) 562 addlog(", %u states killed", killed); 563 } 564 if (pf_status.debug >= LOG_NOTICE) 565 addlog("\n"); 566 } 567 568 /* kill this state */ 569 (*stp)->timeout = PFTM_PURGE; 570 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_CLOSED); 571 return (1); 572 } 573 574 int 575 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule, 576 enum pf_sn_types type, sa_family_t af, struct pf_addr *src, 577 struct pf_addr *raddr, struct pfi_kif *kif) 578 { 579 struct pf_src_node k; 580 581 if (*sn == NULL) { 582 k.af = af; 583 k.type = type; 584 pf_addrcpy(&k.addr, src, af); 585 k.rule.ptr = rule; 586 pf_status.scounters[SCNT_SRC_NODE_SEARCH]++; 587 *sn = RB_FIND(pf_src_tree, &tree_src_tracking, &k); 588 } 589 if (*sn == NULL) { 590 if (!rule->max_src_nodes || 591 rule->src_nodes < rule->max_src_nodes) 592 (*sn) = pool_get(&pf_src_tree_pl, PR_NOWAIT | PR_ZERO); 593 else 594 pf_status.lcounters[LCNT_SRCNODES]++; 595 if ((*sn) == NULL) 596 return (-1); 597 598 pf_init_threshold(&(*sn)->conn_rate, 599 rule->max_src_conn_rate.limit, 600 rule->max_src_conn_rate.seconds); 601 602 (*sn)->type = type; 603 (*sn)->af = af; 604 (*sn)->rule.ptr = rule; 605 pf_addrcpy(&(*sn)->addr, src, af); 606 if (raddr) 607 pf_addrcpy(&(*sn)->raddr, raddr, af); 608 if (RB_INSERT(pf_src_tree, 609 &tree_src_tracking, *sn) != NULL) { 610 if (pf_status.debug >= LOG_NOTICE) { 611 log(LOG_NOTICE, 612 "pf: src_tree insert failed: "); 613 pf_print_host(&(*sn)->addr, 0, af); 614 addlog("\n"); 615 } 616 pool_put(&pf_src_tree_pl, *sn); 617 return (-1); 618 } 619 (*sn)->creation = getuptime(); 620 (*sn)->rule.ptr->src_nodes++; 621 if (kif != NULL) { 622 (*sn)->kif = kif; 623 pfi_kif_ref(kif, PFI_KIF_REF_SRCNODE); 624 } 625 pf_status.scounters[SCNT_SRC_NODE_INSERT]++; 626 pf_status.src_nodes++; 627 } else { 628 if (rule->max_src_states && 629 (*sn)->states >= rule->max_src_states) { 630 pf_status.lcounters[LCNT_SRCSTATES]++; 631 return (-1); 632 } 633 } 634 return (0); 635 } 636 637 void 638 pf_remove_src_node(struct pf_src_node *sn) 639 { 640 if (sn->states > 0 || sn->expire > getuptime()) 641 return; 642 643 sn->rule.ptr->src_nodes--; 644 if (sn->rule.ptr->states_cur == 0 && 645 sn->rule.ptr->src_nodes == 0) 646 pf_rm_rule(NULL, sn->rule.ptr); 647 RB_REMOVE(pf_src_tree, &tree_src_tracking, sn); 648 pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++; 649 pf_status.src_nodes--; 650 pfi_kif_unref(sn->kif, PFI_KIF_REF_SRCNODE); 651 pool_put(&pf_src_tree_pl, sn); 652 } 653 654 struct pf_src_node * 655 pf_get_src_node(struct pf_state *st, enum pf_sn_types type) 656 { 657 struct pf_sn_item *sni; 658 659 SLIST_FOREACH(sni, &st->src_nodes, next) 660 if (sni->sn->type == type) 661 return (sni->sn); 662 return (NULL); 663 } 664 665 void 666 pf_state_rm_src_node(struct pf_state *st, struct pf_src_node *sn) 667 { 668 struct pf_sn_item *sni, *snin, *snip = NULL; 669 670 for (sni = SLIST_FIRST(&st->src_nodes); sni; sni = snin) { 671 snin = SLIST_NEXT(sni, next); 672 if (sni->sn == sn) { 673 if (snip) 674 SLIST_REMOVE_AFTER(snip, next); 675 else 676 SLIST_REMOVE_HEAD(&st->src_nodes, next); 677 pool_put(&pf_sn_item_pl, sni); 678 sni = NULL; 679 sn->states--; 680 } 681 if (sni != NULL) 682 snip = sni; 683 } 684 } 685 686 /* state table stuff */ 687 688 static inline int 689 pf_state_compare_key(const struct pf_state_key *a, 690 const struct pf_state_key *b) 691 { 692 int diff; 693 694 if ((diff = a->hash - b->hash) != 0) 695 return (diff); 696 if ((diff = a->proto - b->proto) != 0) 697 return (diff); 698 if ((diff = a->af - b->af) != 0) 699 return (diff); 700 if ((diff = pf_addr_compare(&a->addr[0], &b->addr[0], a->af)) != 0) 701 return (diff); 702 if ((diff = pf_addr_compare(&a->addr[1], &b->addr[1], a->af)) != 0) 703 return (diff); 704 if ((diff = a->port[0] - b->port[0]) != 0) 705 return (diff); 706 if ((diff = a->port[1] - b->port[1]) != 0) 707 return (diff); 708 if ((diff = a->rdomain - b->rdomain) != 0) 709 return (diff); 710 return (0); 711 } 712 713 static inline int 714 pf_state_compare_id(const struct pf_state *a, const struct pf_state *b) 715 { 716 if (a->id > b->id) 717 return (1); 718 if (a->id < b->id) 719 return (-1); 720 if (a->creatorid > b->creatorid) 721 return (1); 722 if (a->creatorid < b->creatorid) 723 return (-1); 724 725 return (0); 726 } 727 728 /* 729 * on failure, pf_state_key_attach() releases the pf_state_key 730 * reference and returns NULL. 731 */ 732 struct pf_state_key * 733 pf_state_key_attach(struct pf_state_key *sk, struct pf_state *st, int idx) 734 { 735 struct pf_state_item *si; 736 struct pf_state_key *cur; 737 struct pf_state *oldst = NULL; 738 739 PF_ASSERT_LOCKED(); 740 741 KASSERT(st->key[idx] == NULL); 742 sk->sk_removed = 0; 743 cur = RBT_INSERT(pf_state_tree, &pf_statetbl, sk); 744 if (cur != NULL) { 745 sk->sk_removed = 1; 746 /* key exists. check for same kif, if none, add to key */ 747 TAILQ_FOREACH(si, &cur->sk_states, si_entry) { 748 struct pf_state *sist = si->si_st; 749 if (sist->kif == st->kif && 750 ((sist->key[PF_SK_WIRE]->af == sk->af && 751 sist->direction == st->direction) || 752 (sist->key[PF_SK_WIRE]->af != 753 sist->key[PF_SK_STACK]->af && 754 sk->af == sist->key[PF_SK_STACK]->af && 755 sist->direction != st->direction))) { 756 int reuse = 0; 757 758 if (sk->proto == IPPROTO_TCP && 759 sist->src.state >= TCPS_FIN_WAIT_2 && 760 sist->dst.state >= TCPS_FIN_WAIT_2) 761 reuse = 1; 762 if (pf_status.debug >= LOG_NOTICE) { 763 log(LOG_NOTICE, 764 "pf: %s key attach %s on %s: ", 765 (idx == PF_SK_WIRE) ? 766 "wire" : "stack", 767 reuse ? "reuse" : "failed", 768 st->kif->pfik_name); 769 pf_print_state_parts(st, 770 (idx == PF_SK_WIRE) ? sk : NULL, 771 (idx == PF_SK_STACK) ? sk : NULL); 772 addlog(", existing: "); 773 pf_print_state_parts(sist, 774 (idx == PF_SK_WIRE) ? sk : NULL, 775 (idx == PF_SK_STACK) ? sk : NULL); 776 addlog("\n"); 777 } 778 if (reuse) { 779 pf_set_protostate(sist, PF_PEER_BOTH, 780 TCPS_CLOSED); 781 /* remove late or sks can go away */ 782 oldst = sist; 783 } else { 784 pf_state_key_unref(sk); 785 return (NULL); /* collision! */ 786 } 787 } 788 } 789 790 /* reuse the existing state key */ 791 pf_state_key_unref(sk); 792 sk = cur; 793 } 794 795 if ((si = pool_get(&pf_state_item_pl, PR_NOWAIT)) == NULL) { 796 if (TAILQ_EMPTY(&sk->sk_states)) { 797 KASSERT(cur == NULL); 798 RBT_REMOVE(pf_state_tree, &pf_statetbl, sk); 799 sk->sk_removed = 1; 800 pf_state_key_unref(sk); 801 } 802 803 return (NULL); 804 } 805 806 st->key[idx] = pf_state_key_ref(sk); /* give a ref to state */ 807 si->si_st = pf_state_ref(st); 808 809 /* list is sorted, if-bound states before floating */ 810 if (st->kif == pfi_all) 811 TAILQ_INSERT_TAIL(&sk->sk_states, si, si_entry); 812 else 813 TAILQ_INSERT_HEAD(&sk->sk_states, si, si_entry); 814 815 if (oldst) 816 pf_remove_state(oldst); 817 818 /* caller owns the pf_state ref, which owns a pf_state_key ref now */ 819 return (sk); 820 } 821 822 void 823 pf_detach_state(struct pf_state *st) 824 { 825 KASSERT(st->key[PF_SK_WIRE] != NULL); 826 pf_state_key_detach(st, PF_SK_WIRE); 827 828 KASSERT(st->key[PF_SK_STACK] != NULL); 829 if (st->key[PF_SK_STACK] != st->key[PF_SK_WIRE]) 830 pf_state_key_detach(st, PF_SK_STACK); 831 } 832 833 void 834 pf_state_key_detach(struct pf_state *st, int idx) 835 { 836 struct pf_state_item *si; 837 struct pf_state_key *sk; 838 839 PF_ASSERT_LOCKED(); 840 841 sk = st->key[idx]; 842 if (sk == NULL) 843 return; 844 845 TAILQ_FOREACH(si, &sk->sk_states, si_entry) { 846 if (si->si_st == st) 847 break; 848 } 849 if (si == NULL) 850 return; 851 852 TAILQ_REMOVE(&sk->sk_states, si, si_entry); 853 pool_put(&pf_state_item_pl, si); 854 855 if (TAILQ_EMPTY(&sk->sk_states)) { 856 RBT_REMOVE(pf_state_tree, &pf_statetbl, sk); 857 sk->sk_removed = 1; 858 pf_state_key_unlink_reverse(sk); 859 pf_state_key_unlink_inpcb(sk); 860 pf_state_key_unref(sk); 861 } 862 863 pf_state_unref(st); 864 } 865 866 struct pf_state_key * 867 pf_alloc_state_key(int pool_flags) 868 { 869 struct pf_state_key *sk; 870 871 if ((sk = pool_get(&pf_state_key_pl, pool_flags)) == NULL) 872 return (NULL); 873 874 PF_REF_INIT(sk->sk_refcnt); 875 TAILQ_INIT(&sk->sk_states); 876 sk->sk_removed = 1; 877 878 return (sk); 879 } 880 881 static __inline int 882 pf_state_key_addr_setup(struct pf_pdesc *pd, void *arg, int sidx, 883 struct pf_addr *saddr, int didx, struct pf_addr *daddr, int af, int multi) 884 { 885 struct pf_state_key_cmp *key = arg; 886 #ifdef INET6 887 struct pf_addr *target; 888 889 if (af == AF_INET || pd->proto != IPPROTO_ICMPV6) 890 goto copy; 891 892 switch (pd->hdr.icmp6.icmp6_type) { 893 case ND_NEIGHBOR_SOLICIT: 894 if (multi) 895 return (-1); 896 target = (struct pf_addr *)&pd->hdr.nd_ns.nd_ns_target; 897 daddr = target; 898 break; 899 case ND_NEIGHBOR_ADVERT: 900 if (multi) 901 return (-1); 902 target = (struct pf_addr *)&pd->hdr.nd_ns.nd_ns_target; 903 saddr = target; 904 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 905 key->addr[didx].addr32[0] = 0; 906 key->addr[didx].addr32[1] = 0; 907 key->addr[didx].addr32[2] = 0; 908 key->addr[didx].addr32[3] = 0; 909 daddr = NULL; /* overwritten */ 910 } 911 break; 912 default: 913 if (multi) { 914 key->addr[sidx].addr32[0] = __IPV6_ADDR_INT32_MLL; 915 key->addr[sidx].addr32[1] = 0; 916 key->addr[sidx].addr32[2] = 0; 917 key->addr[sidx].addr32[3] = __IPV6_ADDR_INT32_ONE; 918 saddr = NULL; /* overwritten */ 919 } 920 } 921 copy: 922 #endif /* INET6 */ 923 if (saddr) 924 pf_addrcpy(&key->addr[sidx], saddr, af); 925 if (daddr) 926 pf_addrcpy(&key->addr[didx], daddr, af); 927 928 return (0); 929 } 930 931 int 932 pf_state_key_setup(struct pf_pdesc *pd, struct pf_state_key **skw, 933 struct pf_state_key **sks, int rtableid) 934 { 935 /* if returning error we MUST pool_put state keys ourselves */ 936 struct pf_state_key *sk1, *sk2; 937 u_int wrdom = pd->rdomain; 938 int afto = pd->af != pd->naf; 939 940 if ((sk1 = pf_alloc_state_key(PR_NOWAIT | PR_ZERO)) == NULL) 941 return (ENOMEM); 942 943 pf_state_key_addr_setup(pd, sk1, pd->sidx, pd->src, pd->didx, pd->dst, 944 pd->af, 0); 945 sk1->port[pd->sidx] = pd->osport; 946 sk1->port[pd->didx] = pd->odport; 947 sk1->proto = pd->proto; 948 sk1->af = pd->af; 949 sk1->rdomain = pd->rdomain; 950 sk1->hash = pf_pkt_hash(sk1->af, sk1->proto, 951 &sk1->addr[0], &sk1->addr[1], sk1->port[0], sk1->port[1]); 952 if (rtableid >= 0) 953 wrdom = rtable_l2(rtableid); 954 955 if (PF_ANEQ(&pd->nsaddr, pd->src, pd->af) || 956 PF_ANEQ(&pd->ndaddr, pd->dst, pd->af) || 957 pd->nsport != pd->osport || pd->ndport != pd->odport || 958 wrdom != pd->rdomain || afto) { /* NAT/NAT64 */ 959 if ((sk2 = pf_alloc_state_key(PR_NOWAIT | PR_ZERO)) == NULL) { 960 pf_state_key_unref(sk1); 961 return (ENOMEM); 962 } 963 pf_state_key_addr_setup(pd, sk2, afto ? pd->didx : pd->sidx, 964 &pd->nsaddr, afto ? pd->sidx : pd->didx, &pd->ndaddr, 965 pd->naf, 0); 966 sk2->port[afto ? pd->didx : pd->sidx] = pd->nsport; 967 sk2->port[afto ? pd->sidx : pd->didx] = pd->ndport; 968 if (afto) { 969 switch (pd->proto) { 970 case IPPROTO_ICMP: 971 sk2->proto = IPPROTO_ICMPV6; 972 break; 973 case IPPROTO_ICMPV6: 974 sk2->proto = IPPROTO_ICMP; 975 break; 976 default: 977 sk2->proto = pd->proto; 978 } 979 } else 980 sk2->proto = pd->proto; 981 sk2->af = pd->naf; 982 sk2->rdomain = wrdom; 983 sk2->hash = pf_pkt_hash(sk2->af, sk2->proto, 984 &sk2->addr[0], &sk2->addr[1], sk2->port[0], sk2->port[1]); 985 } else 986 sk2 = pf_state_key_ref(sk1); 987 988 if (pd->dir == PF_IN) { 989 *skw = sk1; 990 *sks = sk2; 991 } else { 992 *sks = sk1; 993 *skw = sk2; 994 } 995 996 if (pf_status.debug >= LOG_DEBUG) { 997 log(LOG_DEBUG, "pf: key setup: "); 998 pf_print_state_parts(NULL, *skw, *sks); 999 addlog("\n"); 1000 } 1001 1002 return (0); 1003 } 1004 1005 /* 1006 * pf_state_insert() does the following: 1007 * - links the pf_state up with pf_state_key(s). 1008 * - inserts the pf_state_keys into pf_state_tree. 1009 * - inserts the pf_state into the into pf_state_tree_id. 1010 * - tells pfsync about the state. 1011 * 1012 * pf_state_insert() owns the references to the pf_state_key structs 1013 * it is given. on failure to insert, these references are released. 1014 * on success, the caller owns a pf_state reference that allows it 1015 * to access the state keys. 1016 */ 1017 1018 int 1019 pf_state_insert(struct pfi_kif *kif, struct pf_state_key **skwp, 1020 struct pf_state_key **sksp, struct pf_state *st) 1021 { 1022 struct pf_state_key *skw = *skwp; 1023 struct pf_state_key *sks = *sksp; 1024 int same = (skw == sks); 1025 1026 PF_ASSERT_LOCKED(); 1027 1028 st->kif = kif; 1029 PF_STATE_ENTER_WRITE(); 1030 1031 skw = pf_state_key_attach(skw, st, PF_SK_WIRE); 1032 if (skw == NULL) { 1033 pf_state_key_unref(sks); 1034 PF_STATE_EXIT_WRITE(); 1035 return (-1); 1036 } 1037 1038 if (same) { 1039 /* pf_state_key_attach might have swapped skw */ 1040 pf_state_key_unref(sks); 1041 st->key[PF_SK_STACK] = sks = pf_state_key_ref(skw); 1042 } else if (pf_state_key_attach(sks, st, PF_SK_STACK) == NULL) { 1043 pf_state_key_detach(st, PF_SK_WIRE); 1044 PF_STATE_EXIT_WRITE(); 1045 return (-1); 1046 } 1047 1048 if (st->id == 0 && st->creatorid == 0) { 1049 st->id = htobe64(pf_status.stateid++); 1050 st->creatorid = pf_status.hostid; 1051 } 1052 if (RBT_INSERT(pf_state_tree_id, &tree_id, st) != NULL) { 1053 if (pf_status.debug >= LOG_NOTICE) { 1054 log(LOG_NOTICE, "pf: state insert failed: " 1055 "id: %016llx creatorid: %08x", 1056 betoh64(st->id), ntohl(st->creatorid)); 1057 addlog("\n"); 1058 } 1059 pf_detach_state(st); 1060 PF_STATE_EXIT_WRITE(); 1061 return (-1); 1062 } 1063 pf_state_list_insert(&pf_state_list, st); 1064 pf_status.fcounters[FCNT_STATE_INSERT]++; 1065 pf_status.states++; 1066 pfi_kif_ref(kif, PFI_KIF_REF_STATE); 1067 #if NPFSYNC > 0 1068 pfsync_insert_state(st); 1069 #endif /* NPFSYNC > 0 */ 1070 PF_STATE_EXIT_WRITE(); 1071 1072 *skwp = skw; 1073 *sksp = sks; 1074 1075 return (0); 1076 } 1077 1078 struct pf_state * 1079 pf_find_state_byid(struct pf_state_cmp *key) 1080 { 1081 pf_status.fcounters[FCNT_STATE_SEARCH]++; 1082 1083 return (RBT_FIND(pf_state_tree_id, &tree_id, (struct pf_state *)key)); 1084 } 1085 1086 int 1087 pf_compare_state_keys(struct pf_state_key *a, struct pf_state_key *b, 1088 struct pfi_kif *kif, u_int dir) 1089 { 1090 /* a (from hdr) and b (new) must be exact opposites of each other */ 1091 if (a->af == b->af && a->proto == b->proto && 1092 PF_AEQ(&a->addr[0], &b->addr[1], a->af) && 1093 PF_AEQ(&a->addr[1], &b->addr[0], a->af) && 1094 a->port[0] == b->port[1] && 1095 a->port[1] == b->port[0] && a->rdomain == b->rdomain) 1096 return (0); 1097 else { 1098 /* mismatch. must not happen. */ 1099 if (pf_status.debug >= LOG_ERR) { 1100 log(LOG_ERR, 1101 "pf: state key linking mismatch! dir=%s, " 1102 "if=%s, stored af=%u, a0: ", 1103 dir == PF_OUT ? "OUT" : "IN", 1104 kif->pfik_name, a->af); 1105 pf_print_host(&a->addr[0], a->port[0], a->af); 1106 addlog(", a1: "); 1107 pf_print_host(&a->addr[1], a->port[1], a->af); 1108 addlog(", proto=%u", a->proto); 1109 addlog(", found af=%u, a0: ", b->af); 1110 pf_print_host(&b->addr[0], b->port[0], b->af); 1111 addlog(", a1: "); 1112 pf_print_host(&b->addr[1], b->port[1], b->af); 1113 addlog(", proto=%u", b->proto); 1114 addlog("\n"); 1115 } 1116 return (-1); 1117 } 1118 } 1119 1120 int 1121 pf_find_state(struct pf_pdesc *pd, struct pf_state_key_cmp *key, 1122 struct pf_state **stp) 1123 { 1124 struct pf_state_key *sk, *pkt_sk, *inp_sk; 1125 struct pf_state_item *si; 1126 struct pf_state *st = NULL; 1127 1128 pf_status.fcounters[FCNT_STATE_SEARCH]++; 1129 if (pf_status.debug >= LOG_DEBUG) { 1130 log(LOG_DEBUG, "pf: key search, %s on %s: ", 1131 pd->dir == PF_OUT ? "out" : "in", pd->kif->pfik_name); 1132 pf_print_state_parts(NULL, (struct pf_state_key *)key, NULL); 1133 addlog("\n"); 1134 } 1135 1136 inp_sk = NULL; 1137 pkt_sk = NULL; 1138 sk = NULL; 1139 if (pd->dir == PF_OUT) { 1140 /* first if block deals with outbound forwarded packet */ 1141 pkt_sk = pd->m->m_pkthdr.pf.statekey; 1142 1143 if (!pf_state_key_isvalid(pkt_sk)) { 1144 pf_mbuf_unlink_state_key(pd->m); 1145 pkt_sk = NULL; 1146 } 1147 1148 if (pkt_sk && pf_state_key_isvalid(pkt_sk->sk_reverse)) 1149 sk = pkt_sk->sk_reverse; 1150 1151 if (pkt_sk == NULL) { 1152 /* here we deal with local outbound packet */ 1153 if (pd->m->m_pkthdr.pf.inp != NULL) { 1154 inp_sk = pd->m->m_pkthdr.pf.inp->inp_pf_sk; 1155 if (pf_state_key_isvalid(inp_sk)) 1156 sk = inp_sk; 1157 else 1158 pf_inpcb_unlink_state_key( 1159 pd->m->m_pkthdr.pf.inp); 1160 } 1161 } 1162 } 1163 1164 if (sk == NULL) { 1165 if ((sk = RBT_FIND(pf_state_tree, &pf_statetbl, 1166 (struct pf_state_key *)key)) == NULL) 1167 return (PF_DROP); 1168 if (pd->dir == PF_OUT && pkt_sk && 1169 pf_compare_state_keys(pkt_sk, sk, pd->kif, pd->dir) == 0) 1170 pf_state_key_link_reverse(sk, pkt_sk); 1171 else if (pd->dir == PF_OUT && pd->m->m_pkthdr.pf.inp && 1172 !pd->m->m_pkthdr.pf.inp->inp_pf_sk && !sk->sk_inp) 1173 pf_state_key_link_inpcb(sk, pd->m->m_pkthdr.pf.inp); 1174 } 1175 1176 /* remove firewall data from outbound packet */ 1177 if (pd->dir == PF_OUT) 1178 pf_pkt_addr_changed(pd->m); 1179 1180 /* list is sorted, if-bound states before floating ones */ 1181 TAILQ_FOREACH(si, &sk->sk_states, si_entry) { 1182 struct pf_state *sist = si->si_st; 1183 if (sist->timeout != PFTM_PURGE && 1184 (sist->kif == pfi_all || sist->kif == pd->kif) && 1185 ((sist->key[PF_SK_WIRE]->af == sist->key[PF_SK_STACK]->af && 1186 sk == (pd->dir == PF_IN ? sist->key[PF_SK_WIRE] : 1187 sist->key[PF_SK_STACK])) || 1188 (sist->key[PF_SK_WIRE]->af != sist->key[PF_SK_STACK]->af 1189 && pd->dir == PF_IN && (sk == sist->key[PF_SK_STACK] || 1190 sk == sist->key[PF_SK_WIRE])))) { 1191 st = sist; 1192 break; 1193 } 1194 } 1195 1196 if (st == NULL) 1197 return (PF_DROP); 1198 if (ISSET(st->state_flags, PFSTATE_INP_UNLINKED)) 1199 return (PF_DROP); 1200 1201 if (st->rule.ptr->pktrate.limit && pd->dir == st->direction) { 1202 pf_add_threshold(&st->rule.ptr->pktrate); 1203 if (pf_check_threshold(&st->rule.ptr->pktrate)) 1204 return (PF_DROP); 1205 } 1206 1207 *stp = st; 1208 1209 return (PF_MATCH); 1210 } 1211 1212 struct pf_state * 1213 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1214 { 1215 struct pf_state_key *sk; 1216 struct pf_state_item *si, *ret = NULL; 1217 1218 pf_status.fcounters[FCNT_STATE_SEARCH]++; 1219 1220 sk = RBT_FIND(pf_state_tree, &pf_statetbl, (struct pf_state_key *)key); 1221 1222 if (sk != NULL) { 1223 TAILQ_FOREACH(si, &sk->sk_states, si_entry) { 1224 struct pf_state *sist = si->si_st; 1225 if (dir == PF_INOUT || 1226 (sk == (dir == PF_IN ? sist->key[PF_SK_WIRE] : 1227 sist->key[PF_SK_STACK]))) { 1228 if (more == NULL) 1229 return (sist); 1230 1231 if (ret) 1232 (*more)++; 1233 else 1234 ret = si; 1235 } 1236 } 1237 } 1238 return (ret ? ret->si_st : NULL); 1239 } 1240 1241 void 1242 pf_state_peer_hton(const struct pf_state_peer *s, struct pfsync_state_peer *d) 1243 { 1244 d->seqlo = htonl(s->seqlo); 1245 d->seqhi = htonl(s->seqhi); 1246 d->seqdiff = htonl(s->seqdiff); 1247 d->max_win = htons(s->max_win); 1248 d->mss = htons(s->mss); 1249 d->state = s->state; 1250 d->wscale = s->wscale; 1251 if (s->scrub) { 1252 d->scrub.pfss_flags = 1253 htons(s->scrub->pfss_flags & PFSS_TIMESTAMP); 1254 d->scrub.pfss_ttl = (s)->scrub->pfss_ttl; 1255 d->scrub.pfss_ts_mod = htonl((s)->scrub->pfss_ts_mod); 1256 d->scrub.scrub_flag = PFSYNC_SCRUB_FLAG_VALID; 1257 } 1258 } 1259 1260 void 1261 pf_state_peer_ntoh(const struct pfsync_state_peer *s, struct pf_state_peer *d) 1262 { 1263 d->seqlo = ntohl(s->seqlo); 1264 d->seqhi = ntohl(s->seqhi); 1265 d->seqdiff = ntohl(s->seqdiff); 1266 d->max_win = ntohs(s->max_win); 1267 d->mss = ntohs(s->mss); 1268 d->state = s->state; 1269 d->wscale = s->wscale; 1270 if (s->scrub.scrub_flag == PFSYNC_SCRUB_FLAG_VALID && 1271 d->scrub != NULL) { 1272 d->scrub->pfss_flags = 1273 ntohs(s->scrub.pfss_flags) & PFSS_TIMESTAMP; 1274 d->scrub->pfss_ttl = s->scrub.pfss_ttl; 1275 d->scrub->pfss_ts_mod = ntohl(s->scrub.pfss_ts_mod); 1276 } 1277 } 1278 1279 void 1280 pf_state_export(struct pfsync_state *sp, struct pf_state *st) 1281 { 1282 int32_t expire; 1283 1284 memset(sp, 0, sizeof(struct pfsync_state)); 1285 1286 /* copy from state key */ 1287 sp->key[PF_SK_WIRE].addr[0] = st->key[PF_SK_WIRE]->addr[0]; 1288 sp->key[PF_SK_WIRE].addr[1] = st->key[PF_SK_WIRE]->addr[1]; 1289 sp->key[PF_SK_WIRE].port[0] = st->key[PF_SK_WIRE]->port[0]; 1290 sp->key[PF_SK_WIRE].port[1] = st->key[PF_SK_WIRE]->port[1]; 1291 sp->key[PF_SK_WIRE].rdomain = htons(st->key[PF_SK_WIRE]->rdomain); 1292 sp->key[PF_SK_WIRE].af = st->key[PF_SK_WIRE]->af; 1293 sp->key[PF_SK_STACK].addr[0] = st->key[PF_SK_STACK]->addr[0]; 1294 sp->key[PF_SK_STACK].addr[1] = st->key[PF_SK_STACK]->addr[1]; 1295 sp->key[PF_SK_STACK].port[0] = st->key[PF_SK_STACK]->port[0]; 1296 sp->key[PF_SK_STACK].port[1] = st->key[PF_SK_STACK]->port[1]; 1297 sp->key[PF_SK_STACK].rdomain = htons(st->key[PF_SK_STACK]->rdomain); 1298 sp->key[PF_SK_STACK].af = st->key[PF_SK_STACK]->af; 1299 sp->rtableid[PF_SK_WIRE] = htonl(st->rtableid[PF_SK_WIRE]); 1300 sp->rtableid[PF_SK_STACK] = htonl(st->rtableid[PF_SK_STACK]); 1301 sp->proto = st->key[PF_SK_WIRE]->proto; 1302 sp->af = st->key[PF_SK_WIRE]->af; 1303 1304 /* copy from state */ 1305 strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); 1306 sp->rt = st->rt; 1307 sp->rt_addr = st->rt_addr; 1308 sp->creation = htonl(getuptime() - st->creation); 1309 expire = pf_state_expires(st, st->timeout); 1310 if (expire <= getuptime()) 1311 sp->expire = htonl(0); 1312 else 1313 sp->expire = htonl(expire - getuptime()); 1314 1315 sp->direction = st->direction; 1316 #if NPFLOG > 0 1317 sp->log = st->log; 1318 #endif /* NPFLOG > 0 */ 1319 sp->timeout = st->timeout; 1320 sp->state_flags = htons(st->state_flags); 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 st->pfsync_time = getuptime(); 1523 st->sync_state = PFSYNC_S_NONE; 1524 1525 PF_REF_INIT(st->refcnt); 1526 mtx_init(&st->mtx, IPL_NET); 1527 1528 /* XXX when we have anchors, use STATE_INC_COUNTERS */ 1529 r->states_cur++; 1530 r->states_tot++; 1531 1532 #if NPFSYNC > 0 1533 if (!ISSET(flags, PFSYNC_SI_IOCTL)) 1534 SET(st->state_flags, PFSTATE_NOSYNC); 1535 #endif 1536 1537 /* 1538 * We just set PFSTATE_NOSYNC bit, which prevents 1539 * pfsync_insert_state() to insert state to pfsync. 1540 */ 1541 if (pf_state_insert(kif, &skw, &sks, st) != 0) { 1542 /* XXX when we have anchors, use STATE_DEC_COUNTERS */ 1543 r->states_cur--; 1544 error = EEXIST; 1545 goto cleanup_state; 1546 } 1547 1548 #if NPFSYNC > 0 1549 if (!ISSET(flags, PFSYNC_SI_IOCTL)) { 1550 CLR(st->state_flags, PFSTATE_NOSYNC); 1551 if (ISSET(st->state_flags, PFSTATE_ACK)) 1552 pfsync_iack(st); 1553 } 1554 CLR(st->state_flags, PFSTATE_ACK); 1555 #endif 1556 1557 return (0); 1558 1559 cleanup: 1560 if (skw != NULL) 1561 pf_state_key_unref(skw); 1562 if (sks != NULL) 1563 pf_state_key_unref(sks); 1564 1565 cleanup_state: /* pf_state_insert frees the state keys */ 1566 if (st) { 1567 if (st->dst.scrub) 1568 pool_put(&pf_state_scrub_pl, st->dst.scrub); 1569 if (st->src.scrub) 1570 pool_put(&pf_state_scrub_pl, st->src.scrub); 1571 pool_put(&pf_state_pl, st); 1572 } 1573 return (error); 1574 } 1575 #endif /* NPFSYNC > 0 */ 1576 1577 /* END state table stuff */ 1578 1579 void 1580 pf_purge_timeout(void *unused) 1581 { 1582 /* XXX move to systqmp to avoid KERNEL_LOCK */ 1583 task_add(systq, &pf_purge_task); 1584 } 1585 1586 void 1587 pf_purge(void *xnloops) 1588 { 1589 int *nloops = xnloops; 1590 1591 /* 1592 * process a fraction of the state table every second 1593 * Note: 1594 * we no longer need PF_LOCK() here, because 1595 * pf_purge_expired_states() uses pf_state_lock to maintain 1596 * consistency. 1597 */ 1598 if (pf_default_rule.timeout[PFTM_INTERVAL] > 0) 1599 pf_purge_expired_states(1 + (pf_status.states 1600 / pf_default_rule.timeout[PFTM_INTERVAL])); 1601 1602 NET_LOCK(); 1603 1604 PF_LOCK(); 1605 /* purge other expired types every PFTM_INTERVAL seconds */ 1606 if (++(*nloops) >= pf_default_rule.timeout[PFTM_INTERVAL]) 1607 pf_purge_expired_src_nodes(); 1608 PF_UNLOCK(); 1609 1610 /* 1611 * Fragments don't require PF_LOCK(), they use their own lock. 1612 */ 1613 if ((*nloops) >= pf_default_rule.timeout[PFTM_INTERVAL]) { 1614 pf_purge_expired_fragments(); 1615 *nloops = 0; 1616 } 1617 NET_UNLOCK(); 1618 1619 timeout_add_sec(&pf_purge_to, 1); 1620 } 1621 1622 int32_t 1623 pf_state_expires(const struct pf_state *st, uint8_t stimeout) 1624 { 1625 u_int32_t timeout; 1626 u_int32_t start; 1627 u_int32_t end; 1628 u_int32_t states; 1629 1630 /* 1631 * pf_state_expires is used by the state purge task to 1632 * decide if a state is a candidate for cleanup, and by the 1633 * pfsync state export code to populate an expiry time. 1634 * 1635 * this function may be called by the state purge task while 1636 * the state is being modified. avoid inconsistent reads of 1637 * state->timeout by having the caller do the read (and any 1638 * checks it needs to do on the same variable) and then pass 1639 * their view of the timeout in here for this function to use. 1640 * the only consequence of using a stale timeout value is 1641 * that the state won't be a candidate for purging until the 1642 * next pass of the purge task. 1643 */ 1644 1645 /* handle all PFTM_* >= PFTM_MAX here */ 1646 if (stimeout >= PFTM_MAX) 1647 return (0); 1648 1649 KASSERT(stimeout < PFTM_MAX); 1650 1651 timeout = st->rule.ptr->timeout[stimeout]; 1652 if (!timeout) 1653 timeout = pf_default_rule.timeout[stimeout]; 1654 1655 start = st->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1656 if (start) { 1657 end = st->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1658 states = st->rule.ptr->states_cur; 1659 } else { 1660 start = pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1661 end = pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1662 states = pf_status.states; 1663 } 1664 if (end && states > start && start < end) { 1665 if (states >= end) 1666 return (0); 1667 1668 timeout = (u_int64_t)timeout * (end - states) / (end - start); 1669 } 1670 1671 return (st->expire + timeout); 1672 } 1673 1674 void 1675 pf_purge_expired_src_nodes(void) 1676 { 1677 struct pf_src_node *cur, *next; 1678 1679 PF_ASSERT_LOCKED(); 1680 1681 for (cur = RB_MIN(pf_src_tree, &tree_src_tracking); cur; cur = next) { 1682 next = RB_NEXT(pf_src_tree, &tree_src_tracking, cur); 1683 1684 if (cur->states == 0 && cur->expire <= getuptime()) { 1685 next = RB_NEXT(pf_src_tree, &tree_src_tracking, cur); 1686 pf_remove_src_node(cur); 1687 } 1688 } 1689 } 1690 1691 void 1692 pf_src_tree_remove_state(struct pf_state *st) 1693 { 1694 u_int32_t timeout; 1695 struct pf_sn_item *sni; 1696 1697 while ((sni = SLIST_FIRST(&st->src_nodes)) != NULL) { 1698 SLIST_REMOVE_HEAD(&st->src_nodes, next); 1699 if (st->src.tcp_est) 1700 --sni->sn->conn; 1701 if (--sni->sn->states == 0) { 1702 timeout = st->rule.ptr->timeout[PFTM_SRC_NODE]; 1703 if (!timeout) 1704 timeout = 1705 pf_default_rule.timeout[PFTM_SRC_NODE]; 1706 sni->sn->expire = getuptime() + timeout; 1707 } 1708 pool_put(&pf_sn_item_pl, sni); 1709 } 1710 } 1711 1712 void 1713 pf_remove_state(struct pf_state *st) 1714 { 1715 PF_ASSERT_LOCKED(); 1716 1717 /* handle load balancing related tasks */ 1718 pf_postprocess_addr(st); 1719 1720 if (st->src.state == PF_TCPS_PROXY_DST) { 1721 pf_send_tcp(st->rule.ptr, st->key[PF_SK_WIRE]->af, 1722 &st->key[PF_SK_WIRE]->addr[1], 1723 &st->key[PF_SK_WIRE]->addr[0], 1724 st->key[PF_SK_WIRE]->port[1], 1725 st->key[PF_SK_WIRE]->port[0], 1726 st->src.seqhi, st->src.seqlo + 1, 1727 TH_RST|TH_ACK, 0, 0, 0, 1, st->tag, 1728 st->key[PF_SK_WIRE]->rdomain); 1729 } 1730 if (st->key[PF_SK_STACK]->proto == IPPROTO_TCP) 1731 pf_set_protostate(st, PF_PEER_BOTH, TCPS_CLOSED); 1732 1733 RBT_REMOVE(pf_state_tree_id, &tree_id, st); 1734 #if NPFLOW > 0 1735 if (st->state_flags & PFSTATE_PFLOW) 1736 export_pflow(st); 1737 #endif /* NPFLOW > 0 */ 1738 #if NPFSYNC > 0 1739 pfsync_delete_state(st); 1740 #endif /* NPFSYNC > 0 */ 1741 st->timeout = PFTM_UNLINKED; 1742 pf_src_tree_remove_state(st); 1743 pf_detach_state(st); 1744 } 1745 1746 void 1747 pf_remove_divert_state(struct pf_state_key *sk) 1748 { 1749 struct pf_state_item *si; 1750 1751 PF_ASSERT_UNLOCKED(); 1752 1753 PF_LOCK(); 1754 PF_STATE_ENTER_WRITE(); 1755 TAILQ_FOREACH(si, &sk->sk_states, si_entry) { 1756 struct pf_state *sist = si->si_st; 1757 if (sk == sist->key[PF_SK_STACK] && sist->rule.ptr && 1758 (sist->rule.ptr->divert.type == PF_DIVERT_TO || 1759 sist->rule.ptr->divert.type == PF_DIVERT_REPLY)) { 1760 if (sist->key[PF_SK_STACK]->proto == IPPROTO_TCP && 1761 sist->key[PF_SK_WIRE] != sist->key[PF_SK_STACK]) { 1762 /* 1763 * If the local address is translated, keep 1764 * the state for "tcp.closed" seconds to 1765 * prevent its source port from being reused. 1766 */ 1767 if (sist->src.state < TCPS_FIN_WAIT_2 || 1768 sist->dst.state < TCPS_FIN_WAIT_2) { 1769 pf_set_protostate(sist, PF_PEER_BOTH, 1770 TCPS_TIME_WAIT); 1771 sist->timeout = PFTM_TCP_CLOSED; 1772 sist->expire = getuptime(); 1773 } 1774 sist->state_flags |= PFSTATE_INP_UNLINKED; 1775 } else 1776 pf_remove_state(sist); 1777 break; 1778 } 1779 } 1780 PF_STATE_EXIT_WRITE(); 1781 PF_UNLOCK(); 1782 } 1783 1784 void 1785 pf_free_state(struct pf_state *st) 1786 { 1787 struct pf_rule_item *ri; 1788 1789 PF_ASSERT_LOCKED(); 1790 1791 #if NPFSYNC > 0 1792 if (pfsync_state_in_use(st)) 1793 return; 1794 #endif /* NPFSYNC > 0 */ 1795 KASSERT(st->timeout == PFTM_UNLINKED); 1796 if (--st->rule.ptr->states_cur == 0 && 1797 st->rule.ptr->src_nodes == 0) 1798 pf_rm_rule(NULL, st->rule.ptr); 1799 if (st->anchor.ptr != NULL) 1800 if (--st->anchor.ptr->states_cur == 0) 1801 pf_rm_rule(NULL, st->anchor.ptr); 1802 while ((ri = SLIST_FIRST(&st->match_rules))) { 1803 SLIST_REMOVE_HEAD(&st->match_rules, entry); 1804 if (--ri->r->states_cur == 0 && 1805 ri->r->src_nodes == 0) 1806 pf_rm_rule(NULL, ri->r); 1807 pool_put(&pf_rule_item_pl, ri); 1808 } 1809 pf_normalize_tcp_cleanup(st); 1810 pfi_kif_unref(st->kif, PFI_KIF_REF_STATE); 1811 pf_state_list_remove(&pf_state_list, st); 1812 if (st->tag) 1813 pf_tag_unref(st->tag); 1814 pf_state_unref(st); 1815 pf_status.fcounters[FCNT_STATE_REMOVALS]++; 1816 pf_status.states--; 1817 } 1818 1819 void 1820 pf_purge_expired_states(u_int32_t maxcheck) 1821 { 1822 /* 1823 * this task/thread/context/whatever is the only thing that 1824 * removes states from the pf_state_list, so the cur reference 1825 * it holds between calls is guaranteed to still be in the 1826 * list. 1827 */ 1828 static struct pf_state *cur = NULL; 1829 1830 struct pf_state *head, *tail; 1831 struct pf_state *st; 1832 SLIST_HEAD(pf_state_gcl, pf_state) gcl = SLIST_HEAD_INITIALIZER(gcl); 1833 time_t now; 1834 1835 PF_ASSERT_UNLOCKED(); 1836 1837 rw_enter_read(&pf_state_list.pfs_rwl); 1838 1839 mtx_enter(&pf_state_list.pfs_mtx); 1840 head = TAILQ_FIRST(&pf_state_list.pfs_list); 1841 tail = TAILQ_LAST(&pf_state_list.pfs_list, pf_state_queue); 1842 mtx_leave(&pf_state_list.pfs_mtx); 1843 1844 if (head == NULL) { 1845 /* the list is empty */ 1846 rw_exit_read(&pf_state_list.pfs_rwl); 1847 return; 1848 } 1849 1850 /* (re)start at the front of the list */ 1851 if (cur == NULL) 1852 cur = head; 1853 1854 now = getuptime(); 1855 1856 do { 1857 uint8_t stimeout = cur->timeout; 1858 1859 if ((stimeout == PFTM_UNLINKED) || 1860 (pf_state_expires(cur, stimeout) <= now)) { 1861 st = pf_state_ref(cur); 1862 SLIST_INSERT_HEAD(&gcl, st, gc_list); 1863 } 1864 1865 /* don't iterate past the end of our view of the list */ 1866 if (cur == tail) { 1867 cur = NULL; 1868 break; 1869 } 1870 1871 cur = TAILQ_NEXT(cur, entry_list); 1872 } while (maxcheck--); 1873 1874 rw_exit_read(&pf_state_list.pfs_rwl); 1875 1876 if (SLIST_EMPTY(&gcl)) 1877 return; 1878 1879 NET_LOCK(); 1880 rw_enter_write(&pf_state_list.pfs_rwl); 1881 PF_LOCK(); 1882 PF_STATE_ENTER_WRITE(); 1883 SLIST_FOREACH(st, &gcl, gc_list) { 1884 if (st->timeout != PFTM_UNLINKED) 1885 pf_remove_state(st); 1886 1887 pf_free_state(st); 1888 } 1889 PF_STATE_EXIT_WRITE(); 1890 PF_UNLOCK(); 1891 rw_exit_write(&pf_state_list.pfs_rwl); 1892 NET_UNLOCK(); 1893 1894 while ((st = SLIST_FIRST(&gcl)) != NULL) { 1895 SLIST_REMOVE_HEAD(&gcl, gc_list); 1896 pf_state_unref(st); 1897 } 1898 } 1899 1900 int 1901 pf_tbladdr_setup(struct pf_ruleset *rs, struct pf_addr_wrap *aw, int wait) 1902 { 1903 if (aw->type != PF_ADDR_TABLE) 1904 return (0); 1905 if ((aw->p.tbl = pfr_attach_table(rs, aw->v.tblname, wait)) == NULL) 1906 return (1); 1907 return (0); 1908 } 1909 1910 void 1911 pf_tbladdr_remove(struct pf_addr_wrap *aw) 1912 { 1913 if (aw->type != PF_ADDR_TABLE || aw->p.tbl == NULL) 1914 return; 1915 pfr_detach_table(aw->p.tbl); 1916 aw->p.tbl = NULL; 1917 } 1918 1919 void 1920 pf_tbladdr_copyout(struct pf_addr_wrap *aw) 1921 { 1922 struct pfr_ktable *kt = aw->p.tbl; 1923 1924 if (aw->type != PF_ADDR_TABLE || kt == NULL) 1925 return; 1926 if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) 1927 kt = kt->pfrkt_root; 1928 aw->p.tbl = NULL; 1929 aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ? 1930 kt->pfrkt_cnt : -1; 1931 } 1932 1933 void 1934 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 1935 { 1936 switch (af) { 1937 case AF_INET: { 1938 u_int32_t a = ntohl(addr->addr32[0]); 1939 addlog("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 1940 (a>>8)&255, a&255); 1941 if (p) { 1942 p = ntohs(p); 1943 addlog(":%u", p); 1944 } 1945 break; 1946 } 1947 #ifdef INET6 1948 case AF_INET6: { 1949 u_int16_t b; 1950 u_int8_t i, curstart, curend, maxstart, maxend; 1951 curstart = curend = maxstart = maxend = 255; 1952 for (i = 0; i < 8; i++) { 1953 if (!addr->addr16[i]) { 1954 if (curstart == 255) 1955 curstart = i; 1956 curend = i; 1957 } else { 1958 if ((curend - curstart) > 1959 (maxend - maxstart)) { 1960 maxstart = curstart; 1961 maxend = curend; 1962 } 1963 curstart = curend = 255; 1964 } 1965 } 1966 if ((curend - curstart) > 1967 (maxend - maxstart)) { 1968 maxstart = curstart; 1969 maxend = curend; 1970 } 1971 for (i = 0; i < 8; i++) { 1972 if (i >= maxstart && i <= maxend) { 1973 if (i == 0) 1974 addlog(":"); 1975 if (i == maxend) 1976 addlog(":"); 1977 } else { 1978 b = ntohs(addr->addr16[i]); 1979 addlog("%x", b); 1980 if (i < 7) 1981 addlog(":"); 1982 } 1983 } 1984 if (p) { 1985 p = ntohs(p); 1986 addlog("[%u]", p); 1987 } 1988 break; 1989 } 1990 #endif /* INET6 */ 1991 } 1992 } 1993 1994 void 1995 pf_print_state(struct pf_state *st) 1996 { 1997 pf_print_state_parts(st, NULL, NULL); 1998 } 1999 2000 void 2001 pf_print_state_parts(struct pf_state *st, 2002 struct pf_state_key *skwp, struct pf_state_key *sksp) 2003 { 2004 struct pf_state_key *skw, *sks; 2005 u_int8_t proto, dir; 2006 2007 /* Do our best to fill these, but they're skipped if NULL */ 2008 skw = skwp ? skwp : (st ? st->key[PF_SK_WIRE] : NULL); 2009 sks = sksp ? sksp : (st ? st->key[PF_SK_STACK] : NULL); 2010 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2011 dir = st ? st->direction : 0; 2012 2013 switch (proto) { 2014 case IPPROTO_IPV4: 2015 addlog("IPv4"); 2016 break; 2017 case IPPROTO_IPV6: 2018 addlog("IPv6"); 2019 break; 2020 case IPPROTO_TCP: 2021 addlog("TCP"); 2022 break; 2023 case IPPROTO_UDP: 2024 addlog("UDP"); 2025 break; 2026 case IPPROTO_ICMP: 2027 addlog("ICMP"); 2028 break; 2029 case IPPROTO_ICMPV6: 2030 addlog("ICMPv6"); 2031 break; 2032 default: 2033 addlog("%u", proto); 2034 break; 2035 } 2036 switch (dir) { 2037 case PF_IN: 2038 addlog(" in"); 2039 break; 2040 case PF_OUT: 2041 addlog(" out"); 2042 break; 2043 } 2044 if (skw) { 2045 addlog(" wire: (%d) ", skw->rdomain); 2046 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2047 addlog(" "); 2048 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2049 } 2050 if (sks) { 2051 addlog(" stack: (%d) ", sks->rdomain); 2052 if (sks != skw) { 2053 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2054 addlog(" "); 2055 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2056 } else 2057 addlog("-"); 2058 } 2059 if (st) { 2060 if (proto == IPPROTO_TCP) { 2061 addlog(" [lo=%u high=%u win=%u modulator=%u", 2062 st->src.seqlo, st->src.seqhi, 2063 st->src.max_win, st->src.seqdiff); 2064 if (st->src.wscale && st->dst.wscale) 2065 addlog(" wscale=%u", 2066 st->src.wscale & PF_WSCALE_MASK); 2067 addlog("]"); 2068 addlog(" [lo=%u high=%u win=%u modulator=%u", 2069 st->dst.seqlo, st->dst.seqhi, 2070 st->dst.max_win, st->dst.seqdiff); 2071 if (st->src.wscale && st->dst.wscale) 2072 addlog(" wscale=%u", 2073 st->dst.wscale & PF_WSCALE_MASK); 2074 addlog("]"); 2075 } 2076 addlog(" %u:%u", st->src.state, st->dst.state); 2077 if (st->rule.ptr) 2078 addlog(" @%d", st->rule.ptr->nr); 2079 } 2080 } 2081 2082 void 2083 pf_print_flags(u_int8_t f) 2084 { 2085 if (f) 2086 addlog(" "); 2087 if (f & TH_FIN) 2088 addlog("F"); 2089 if (f & TH_SYN) 2090 addlog("S"); 2091 if (f & TH_RST) 2092 addlog("R"); 2093 if (f & TH_PUSH) 2094 addlog("P"); 2095 if (f & TH_ACK) 2096 addlog("A"); 2097 if (f & TH_URG) 2098 addlog("U"); 2099 if (f & TH_ECE) 2100 addlog("E"); 2101 if (f & TH_CWR) 2102 addlog("W"); 2103 } 2104 2105 #define PF_SET_SKIP_STEPS(i) \ 2106 do { \ 2107 while (head[i] != cur) { \ 2108 head[i]->skip[i].ptr = cur; \ 2109 head[i] = TAILQ_NEXT(head[i], entries); \ 2110 } \ 2111 } while (0) 2112 2113 void 2114 pf_calc_skip_steps(struct pf_rulequeue *rules) 2115 { 2116 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT]; 2117 int i; 2118 2119 cur = TAILQ_FIRST(rules); 2120 prev = cur; 2121 for (i = 0; i < PF_SKIP_COUNT; ++i) 2122 head[i] = cur; 2123 while (cur != NULL) { 2124 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2125 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2126 if (cur->direction != prev->direction) 2127 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2128 if (cur->onrdomain != prev->onrdomain || 2129 cur->ifnot != prev->ifnot) 2130 PF_SET_SKIP_STEPS(PF_SKIP_RDOM); 2131 if (cur->af != prev->af) 2132 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2133 if (cur->proto != prev->proto) 2134 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2135 if (cur->src.neg != prev->src.neg || 2136 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2137 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2138 if (cur->dst.neg != prev->dst.neg || 2139 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2140 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2141 if (cur->src.port[0] != prev->src.port[0] || 2142 cur->src.port[1] != prev->src.port[1] || 2143 cur->src.port_op != prev->src.port_op) 2144 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2145 if (cur->dst.port[0] != prev->dst.port[0] || 2146 cur->dst.port[1] != prev->dst.port[1] || 2147 cur->dst.port_op != prev->dst.port_op) 2148 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2149 2150 prev = cur; 2151 cur = TAILQ_NEXT(cur, entries); 2152 } 2153 for (i = 0; i < PF_SKIP_COUNT; ++i) 2154 PF_SET_SKIP_STEPS(i); 2155 } 2156 2157 int 2158 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2159 { 2160 if (aw1->type != aw2->type) 2161 return (1); 2162 switch (aw1->type) { 2163 case PF_ADDR_ADDRMASK: 2164 case PF_ADDR_RANGE: 2165 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2166 return (1); 2167 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2168 return (1); 2169 return (0); 2170 case PF_ADDR_DYNIFTL: 2171 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2172 case PF_ADDR_NONE: 2173 case PF_ADDR_NOROUTE: 2174 case PF_ADDR_URPFFAILED: 2175 return (0); 2176 case PF_ADDR_TABLE: 2177 return (aw1->p.tbl != aw2->p.tbl); 2178 case PF_ADDR_RTLABEL: 2179 return (aw1->v.rtlabel != aw2->v.rtlabel); 2180 default: 2181 addlog("invalid address type: %d\n", aw1->type); 2182 return (1); 2183 } 2184 } 2185 2186 /* This algorithm computes 'a + b - c' in ones-complement using a trick to 2187 * emulate at most one ones-complement subtraction. This thereby limits net 2188 * carries/borrows to at most one, eliminating a reduction step and saving one 2189 * each of +, >>, & and ~. 2190 * 2191 * def. x mod y = x - (x//y)*y for integer x,y 2192 * def. sum = x mod 2^16 2193 * def. accumulator = (x >> 16) mod 2^16 2194 * 2195 * The trick works as follows: subtracting exactly one u_int16_t from the 2196 * u_int32_t x incurs at most one underflow, wrapping its upper 16-bits, the 2197 * accumulator, to 2^16 - 1. Adding this to the 16-bit sum preserves the 2198 * ones-complement borrow: 2199 * 2200 * (sum + accumulator) mod 2^16 2201 * = { assume underflow: accumulator := 2^16 - 1 } 2202 * (sum + 2^16 - 1) mod 2^16 2203 * = { mod } 2204 * (sum - 1) mod 2^16 2205 * 2206 * Although this breaks for sum = 0, giving 0xffff, which is ones-complement's 2207 * other zero, not -1, that cannot occur: the 16-bit sum cannot be underflown 2208 * to zero as that requires subtraction of at least 2^16, which exceeds a 2209 * single u_int16_t's range. 2210 * 2211 * We use the following theorem to derive the implementation: 2212 * 2213 * th. (x + (y mod z)) mod z = (x + y) mod z (0) 2214 * proof. 2215 * (x + (y mod z)) mod z 2216 * = { def mod } 2217 * (x + y - (y//z)*z) mod z 2218 * = { (a + b*c) mod c = a mod c } 2219 * (x + y) mod z [end of proof] 2220 * 2221 * ... and thereby obtain: 2222 * 2223 * (sum + accumulator) mod 2^16 2224 * = { def. accumulator, def. sum } 2225 * (x mod 2^16 + (x >> 16) mod 2^16) mod 2^16 2226 * = { (0), twice } 2227 * (x + (x >> 16)) mod 2^16 2228 * = { x mod 2^n = x & (2^n - 1) } 2229 * (x + (x >> 16)) & 0xffff 2230 * 2231 * Note: this serves also as a reduction step for at most one add (as the 2232 * trailing mod 2^16 prevents further reductions by destroying carries). 2233 */ 2234 __inline void 2235 pf_cksum_fixup(u_int16_t *cksum, u_int16_t was, u_int16_t now, 2236 u_int8_t proto) 2237 { 2238 u_int32_t x; 2239 const int udp = proto == IPPROTO_UDP; 2240 2241 x = *cksum + was - now; 2242 x = (x + (x >> 16)) & 0xffff; 2243 2244 /* optimise: eliminate a branch when not udp */ 2245 if (udp && *cksum == 0x0000) 2246 return; 2247 if (udp && x == 0x0000) 2248 x = 0xffff; 2249 2250 *cksum = (u_int16_t)(x); 2251 } 2252 2253 #ifdef INET6 2254 /* pre: coverage(cksum) is superset of coverage(covered_cksum) */ 2255 static __inline void 2256 pf_cksum_uncover(u_int16_t *cksum, u_int16_t covered_cksum, u_int8_t proto) 2257 { 2258 pf_cksum_fixup(cksum, ~covered_cksum, 0x0, proto); 2259 } 2260 2261 /* pre: disjoint(coverage(cksum), coverage(uncovered_cksum)) */ 2262 static __inline void 2263 pf_cksum_cover(u_int16_t *cksum, u_int16_t uncovered_cksum, u_int8_t proto) 2264 { 2265 pf_cksum_fixup(cksum, 0x0, ~uncovered_cksum, proto); 2266 } 2267 #endif /* INET6 */ 2268 2269 /* pre: *a is 16-bit aligned within its packet 2270 * 2271 * This algorithm emulates 16-bit ones-complement sums on a twos-complement 2272 * machine by conserving ones-complement's otherwise discarded carries in the 2273 * upper bits of x. These accumulated carries when added to the lower 16-bits 2274 * over at least zero 'reduction' steps then complete the ones-complement sum. 2275 * 2276 * def. sum = x mod 2^16 2277 * def. accumulator = (x >> 16) 2278 * 2279 * At most two reduction steps 2280 * 2281 * x := sum + accumulator 2282 * = { def sum, def accumulator } 2283 * x := x mod 2^16 + (x >> 16) 2284 * = { x mod 2^n = x & (2^n - 1) } 2285 * x := (x & 0xffff) + (x >> 16) 2286 * 2287 * are necessary to incorporate the accumulated carries (at most one per add) 2288 * i.e. to reduce x < 2^16 from at most 16 carries in the upper 16 bits. 2289 * 2290 * The function is also invariant over the endian of the host. Why? 2291 * 2292 * Define the unary transpose operator ~ on a bitstring in python slice 2293 * notation as lambda m: m[P:] + m[:P] , for some constant pivot P. 2294 * 2295 * th. ~ distributes over ones-complement addition, denoted by +_1, i.e. 2296 * 2297 * ~m +_1 ~n = ~(m +_1 n) (for all bitstrings m,n of equal length) 2298 * 2299 * proof. Regard the bitstrings in m +_1 n as split at P, forming at most two 2300 * 'half-adds'. Under ones-complement addition, each half-add carries to the 2301 * other, so the sum of each half-add is unaffected by their relative 2302 * order. Therefore: 2303 * 2304 * ~m +_1 ~n 2305 * = { half-adds invariant under transposition } 2306 * ~s 2307 * = { substitute } 2308 * ~(m +_1 n) [end of proof] 2309 * 2310 * th. Summing two in-memory ones-complement 16-bit variables m,n on a machine 2311 * with the converse endian does not alter the result. 2312 * 2313 * proof. 2314 * { converse machine endian: load/store transposes, P := 8 } 2315 * ~(~m +_1 ~n) 2316 * = { ~ over +_1 } 2317 * ~~m +_1 ~~n 2318 * = { ~ is an involution } 2319 * m +_1 n [end of proof] 2320 * 2321 */ 2322 #define NEG(x) ((u_int16_t)~(x)) 2323 void 2324 pf_cksum_fixup_a(u_int16_t *cksum, const struct pf_addr *a, 2325 const struct pf_addr *an, sa_family_t af, u_int8_t proto) 2326 { 2327 u_int32_t x; 2328 const u_int16_t *n = an->addr16; 2329 const u_int16_t *o = a->addr16; 2330 const int udp = proto == IPPROTO_UDP; 2331 2332 switch (af) { 2333 case AF_INET: 2334 x = *cksum + o[0] + NEG(n[0]) + o[1] + NEG(n[1]); 2335 break; 2336 #ifdef INET6 2337 case AF_INET6: 2338 x = *cksum + o[0] + NEG(n[0]) + o[1] + NEG(n[1]) +\ 2339 o[2] + NEG(n[2]) + o[3] + NEG(n[3]) +\ 2340 o[4] + NEG(n[4]) + o[5] + NEG(n[5]) +\ 2341 o[6] + NEG(n[6]) + o[7] + NEG(n[7]); 2342 break; 2343 #endif /* INET6 */ 2344 default: 2345 unhandled_af(af); 2346 } 2347 2348 x = (x & 0xffff) + (x >> 16); 2349 x = (x & 0xffff) + (x >> 16); 2350 2351 /* optimise: eliminate a branch when not udp */ 2352 if (udp && *cksum == 0x0000) 2353 return; 2354 if (udp && x == 0x0000) 2355 x = 0xffff; 2356 2357 *cksum = (u_int16_t)(x); 2358 } 2359 2360 int 2361 pf_patch_8(struct pf_pdesc *pd, u_int8_t *f, u_int8_t v, bool hi) 2362 { 2363 int rewrite = 0; 2364 2365 if (*f != v) { 2366 u_int16_t old = htons(hi ? (*f << 8) : *f); 2367 u_int16_t new = htons(hi ? ( v << 8) : v); 2368 2369 pf_cksum_fixup(pd->pcksum, old, new, pd->proto); 2370 *f = v; 2371 rewrite = 1; 2372 } 2373 2374 return (rewrite); 2375 } 2376 2377 /* pre: *f is 16-bit aligned within its packet */ 2378 int 2379 pf_patch_16(struct pf_pdesc *pd, u_int16_t *f, u_int16_t v) 2380 { 2381 int rewrite = 0; 2382 2383 if (*f != v) { 2384 pf_cksum_fixup(pd->pcksum, *f, v, pd->proto); 2385 *f = v; 2386 rewrite = 1; 2387 } 2388 2389 return (rewrite); 2390 } 2391 2392 int 2393 pf_patch_16_unaligned(struct pf_pdesc *pd, void *f, u_int16_t v, bool hi) 2394 { 2395 int rewrite = 0; 2396 u_int8_t *fb = (u_int8_t*)f; 2397 u_int8_t *vb = (u_int8_t*)&v; 2398 2399 if (hi && ALIGNED_POINTER(f, u_int16_t)) { 2400 return (pf_patch_16(pd, f, v)); /* optimise */ 2401 } 2402 2403 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 2404 rewrite += pf_patch_8(pd, fb++, *vb++,!hi); 2405 2406 return (rewrite); 2407 } 2408 2409 /* pre: *f is 16-bit aligned within its packet */ 2410 /* pre: pd->proto != IPPROTO_UDP */ 2411 int 2412 pf_patch_32(struct pf_pdesc *pd, u_int32_t *f, u_int32_t v) 2413 { 2414 int rewrite = 0; 2415 u_int16_t *pc = pd->pcksum; 2416 u_int8_t proto = pd->proto; 2417 2418 /* optimise: inline udp fixup code is unused; let compiler scrub it */ 2419 if (proto == IPPROTO_UDP) 2420 panic("%s: udp", __func__); 2421 2422 /* optimise: skip *f != v guard; true for all use-cases */ 2423 pf_cksum_fixup(pc, *f / (1 << 16), v / (1 << 16), proto); 2424 pf_cksum_fixup(pc, *f % (1 << 16), v % (1 << 16), proto); 2425 2426 *f = v; 2427 rewrite = 1; 2428 2429 return (rewrite); 2430 } 2431 2432 int 2433 pf_patch_32_unaligned(struct pf_pdesc *pd, void *f, u_int32_t v, bool hi) 2434 { 2435 int rewrite = 0; 2436 u_int8_t *fb = (u_int8_t*)f; 2437 u_int8_t *vb = (u_int8_t*)&v; 2438 2439 if (hi && ALIGNED_POINTER(f, u_int32_t)) { 2440 return (pf_patch_32(pd, f, v)); /* optimise */ 2441 } 2442 2443 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 2444 rewrite += pf_patch_8(pd, fb++, *vb++,!hi); 2445 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 2446 rewrite += pf_patch_8(pd, fb++, *vb++,!hi); 2447 2448 return (rewrite); 2449 } 2450 2451 int 2452 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, int *icmp_dir, 2453 u_int16_t *virtual_id, u_int16_t *virtual_type) 2454 { 2455 /* 2456 * ICMP types marked with PF_OUT are typically responses to 2457 * PF_IN, and will match states in the opposite direction. 2458 * PF_IN ICMP types need to match a state with that type. 2459 */ 2460 *icmp_dir = PF_OUT; 2461 2462 /* Queries (and responses) */ 2463 switch (pd->af) { 2464 case AF_INET: 2465 switch (type) { 2466 case ICMP_ECHO: 2467 *icmp_dir = PF_IN; 2468 /* FALLTHROUGH */ 2469 case ICMP_ECHOREPLY: 2470 *virtual_type = ICMP_ECHO; 2471 *virtual_id = pd->hdr.icmp.icmp_id; 2472 break; 2473 2474 case ICMP_TSTAMP: 2475 *icmp_dir = PF_IN; 2476 /* FALLTHROUGH */ 2477 case ICMP_TSTAMPREPLY: 2478 *virtual_type = ICMP_TSTAMP; 2479 *virtual_id = pd->hdr.icmp.icmp_id; 2480 break; 2481 2482 case ICMP_IREQ: 2483 *icmp_dir = PF_IN; 2484 /* FALLTHROUGH */ 2485 case ICMP_IREQREPLY: 2486 *virtual_type = ICMP_IREQ; 2487 *virtual_id = pd->hdr.icmp.icmp_id; 2488 break; 2489 2490 case ICMP_MASKREQ: 2491 *icmp_dir = PF_IN; 2492 /* FALLTHROUGH */ 2493 case ICMP_MASKREPLY: 2494 *virtual_type = ICMP_MASKREQ; 2495 *virtual_id = pd->hdr.icmp.icmp_id; 2496 break; 2497 2498 case ICMP_IPV6_WHEREAREYOU: 2499 *icmp_dir = PF_IN; 2500 /* FALLTHROUGH */ 2501 case ICMP_IPV6_IAMHERE: 2502 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2503 *virtual_id = 0; /* Nothing sane to match on! */ 2504 break; 2505 2506 case ICMP_MOBILE_REGREQUEST: 2507 *icmp_dir = PF_IN; 2508 /* FALLTHROUGH */ 2509 case ICMP_MOBILE_REGREPLY: 2510 *virtual_type = ICMP_MOBILE_REGREQUEST; 2511 *virtual_id = 0; /* Nothing sane to match on! */ 2512 break; 2513 2514 case ICMP_ROUTERSOLICIT: 2515 *icmp_dir = PF_IN; 2516 /* FALLTHROUGH */ 2517 case ICMP_ROUTERADVERT: 2518 *virtual_type = ICMP_ROUTERSOLICIT; 2519 *virtual_id = 0; /* Nothing sane to match on! */ 2520 break; 2521 2522 /* These ICMP types map to other connections */ 2523 case ICMP_UNREACH: 2524 case ICMP_SOURCEQUENCH: 2525 case ICMP_REDIRECT: 2526 case ICMP_TIMXCEED: 2527 case ICMP_PARAMPROB: 2528 /* These will not be used, but set them anyway */ 2529 *icmp_dir = PF_IN; 2530 *virtual_type = htons(type); 2531 *virtual_id = 0; 2532 return (1); /* These types match to another state */ 2533 2534 /* 2535 * All remaining ICMP types get their own states, 2536 * and will only match in one direction. 2537 */ 2538 default: 2539 *icmp_dir = PF_IN; 2540 *virtual_type = type; 2541 *virtual_id = 0; 2542 break; 2543 } 2544 break; 2545 #ifdef INET6 2546 case AF_INET6: 2547 switch (type) { 2548 case ICMP6_ECHO_REQUEST: 2549 *icmp_dir = PF_IN; 2550 /* FALLTHROUGH */ 2551 case ICMP6_ECHO_REPLY: 2552 *virtual_type = ICMP6_ECHO_REQUEST; 2553 *virtual_id = pd->hdr.icmp6.icmp6_id; 2554 break; 2555 2556 case MLD_LISTENER_QUERY: 2557 case MLD_LISTENER_REPORT: { 2558 struct mld_hdr *mld = &pd->hdr.mld; 2559 u_int32_t h; 2560 2561 /* 2562 * Listener Report can be sent by clients 2563 * without an associated Listener Query. 2564 * In addition to that, when Report is sent as a 2565 * reply to a Query its source and destination 2566 * address are different. 2567 */ 2568 *icmp_dir = PF_IN; 2569 *virtual_type = MLD_LISTENER_QUERY; 2570 /* generate fake id for these messages */ 2571 h = mld->mld_addr.s6_addr32[0] ^ 2572 mld->mld_addr.s6_addr32[1] ^ 2573 mld->mld_addr.s6_addr32[2] ^ 2574 mld->mld_addr.s6_addr32[3]; 2575 *virtual_id = (h >> 16) ^ (h & 0xffff); 2576 break; 2577 } 2578 2579 /* 2580 * ICMP6_FQDN and ICMP6_NI query/reply are the same type as 2581 * ICMP6_WRU 2582 */ 2583 case ICMP6_WRUREQUEST: 2584 *icmp_dir = PF_IN; 2585 /* FALLTHROUGH */ 2586 case ICMP6_WRUREPLY: 2587 *virtual_type = ICMP6_WRUREQUEST; 2588 *virtual_id = 0; /* Nothing sane to match on! */ 2589 break; 2590 2591 case MLD_MTRACE: 2592 *icmp_dir = PF_IN; 2593 /* FALLTHROUGH */ 2594 case MLD_MTRACE_RESP: 2595 *virtual_type = MLD_MTRACE; 2596 *virtual_id = 0; /* Nothing sane to match on! */ 2597 break; 2598 2599 case ND_NEIGHBOR_SOLICIT: 2600 *icmp_dir = PF_IN; 2601 /* FALLTHROUGH */ 2602 case ND_NEIGHBOR_ADVERT: { 2603 struct nd_neighbor_solicit *nd = &pd->hdr.nd_ns; 2604 u_int32_t h; 2605 2606 *virtual_type = ND_NEIGHBOR_SOLICIT; 2607 /* generate fake id for these messages */ 2608 h = nd->nd_ns_target.s6_addr32[0] ^ 2609 nd->nd_ns_target.s6_addr32[1] ^ 2610 nd->nd_ns_target.s6_addr32[2] ^ 2611 nd->nd_ns_target.s6_addr32[3]; 2612 *virtual_id = (h >> 16) ^ (h & 0xffff); 2613 /* 2614 * the extra work here deals with 'keep state' option 2615 * at pass rule for unsolicited advertisement. By 2616 * returning 1 (state_icmp = 1) we override 'keep 2617 * state' to 'no state' so we don't create state for 2618 * unsolicited advertisements. No one expects answer to 2619 * unsolicited advertisements so we should be good. 2620 */ 2621 if (type == ND_NEIGHBOR_ADVERT) { 2622 *virtual_type = htons(*virtual_type); 2623 return (1); 2624 } 2625 break; 2626 } 2627 2628 /* 2629 * These ICMP types map to other connections. 2630 * ND_REDIRECT can't be in this list because the triggering 2631 * packet header is optional. 2632 */ 2633 case ICMP6_DST_UNREACH: 2634 case ICMP6_PACKET_TOO_BIG: 2635 case ICMP6_TIME_EXCEEDED: 2636 case ICMP6_PARAM_PROB: 2637 /* These will not be used, but set them anyway */ 2638 *icmp_dir = PF_IN; 2639 *virtual_type = htons(type); 2640 *virtual_id = 0; 2641 return (1); /* These types match to another state */ 2642 /* 2643 * All remaining ICMP6 types get their own states, 2644 * and will only match in one direction. 2645 */ 2646 default: 2647 *icmp_dir = PF_IN; 2648 *virtual_type = type; 2649 *virtual_id = 0; 2650 break; 2651 } 2652 break; 2653 #endif /* INET6 */ 2654 } 2655 *virtual_type = htons(*virtual_type); 2656 return (0); /* These types match to their own state */ 2657 } 2658 2659 void 2660 pf_translate_icmp(struct pf_pdesc *pd, struct pf_addr *qa, u_int16_t *qp, 2661 struct pf_addr *oa, struct pf_addr *na, u_int16_t np) 2662 { 2663 /* note: doesn't trouble to fixup quoted checksums, if any */ 2664 2665 /* change quoted protocol port */ 2666 if (qp != NULL) 2667 pf_patch_16(pd, qp, np); 2668 2669 /* change quoted ip address */ 2670 pf_cksum_fixup_a(pd->pcksum, qa, na, pd->af, pd->proto); 2671 pf_addrcpy(qa, na, pd->af); 2672 2673 /* change network-header's ip address */ 2674 if (oa) 2675 pf_translate_a(pd, oa, na); 2676 } 2677 2678 /* pre: *a is 16-bit aligned within its packet */ 2679 /* *a is a network header src/dst address */ 2680 int 2681 pf_translate_a(struct pf_pdesc *pd, struct pf_addr *a, struct pf_addr *an) 2682 { 2683 int rewrite = 0; 2684 2685 /* warning: !PF_ANEQ != PF_AEQ */ 2686 if (!PF_ANEQ(a, an, pd->af)) 2687 return (0); 2688 2689 /* fixup transport pseudo-header, if any */ 2690 switch (pd->proto) { 2691 case IPPROTO_TCP: /* FALLTHROUGH */ 2692 case IPPROTO_UDP: /* FALLTHROUGH */ 2693 case IPPROTO_ICMPV6: 2694 pf_cksum_fixup_a(pd->pcksum, a, an, pd->af, pd->proto); 2695 break; 2696 default: 2697 break; /* assume no pseudo-header */ 2698 } 2699 2700 pf_addrcpy(a, an, pd->af); 2701 rewrite = 1; 2702 2703 return (rewrite); 2704 } 2705 2706 #ifdef INET6 2707 /* pf_translate_af() may change pd->m, adjust local copies after calling */ 2708 int 2709 pf_translate_af(struct pf_pdesc *pd) 2710 { 2711 static const struct pf_addr zero; 2712 struct ip *ip4; 2713 struct ip6_hdr *ip6; 2714 int copyback = 0; 2715 u_int hlen, ohlen, dlen; 2716 u_int16_t *pc; 2717 u_int8_t af_proto, naf_proto; 2718 2719 hlen = (pd->naf == AF_INET) ? sizeof(*ip4) : sizeof(*ip6); 2720 ohlen = pd->off; 2721 dlen = pd->tot_len - pd->off; 2722 pc = pd->pcksum; 2723 2724 af_proto = naf_proto = pd->proto; 2725 if (naf_proto == IPPROTO_ICMP) 2726 af_proto = IPPROTO_ICMPV6; 2727 if (naf_proto == IPPROTO_ICMPV6) 2728 af_proto = IPPROTO_ICMP; 2729 2730 /* uncover stale pseudo-header */ 2731 switch (af_proto) { 2732 case IPPROTO_ICMPV6: 2733 /* optimise: unchanged for TCP/UDP */ 2734 pf_cksum_fixup(pc, htons(af_proto), 0x0, af_proto); 2735 pf_cksum_fixup(pc, htons(dlen), 0x0, af_proto); 2736 /* FALLTHROUGH */ 2737 case IPPROTO_UDP: /* FALLTHROUGH */ 2738 case IPPROTO_TCP: 2739 pf_cksum_fixup_a(pc, pd->src, &zero, pd->af, af_proto); 2740 pf_cksum_fixup_a(pc, pd->dst, &zero, pd->af, af_proto); 2741 copyback = 1; 2742 break; 2743 default: 2744 break; /* assume no pseudo-header */ 2745 } 2746 2747 /* replace the network header */ 2748 m_adj(pd->m, pd->off); 2749 pd->src = NULL; 2750 pd->dst = NULL; 2751 2752 if ((M_PREPEND(pd->m, hlen, M_DONTWAIT)) == NULL) { 2753 pd->m = NULL; 2754 return (-1); 2755 } 2756 2757 pd->off = hlen; 2758 pd->tot_len += hlen - ohlen; 2759 2760 switch (pd->naf) { 2761 case AF_INET: 2762 ip4 = mtod(pd->m, struct ip *); 2763 memset(ip4, 0, hlen); 2764 ip4->ip_v = IPVERSION; 2765 ip4->ip_hl = hlen >> 2; 2766 ip4->ip_tos = pd->tos; 2767 ip4->ip_len = htons(hlen + dlen); 2768 ip4->ip_id = htons(ip_randomid()); 2769 ip4->ip_off = htons(IP_DF); 2770 ip4->ip_ttl = pd->ttl; 2771 ip4->ip_p = pd->proto; 2772 ip4->ip_src = pd->nsaddr.v4; 2773 ip4->ip_dst = pd->ndaddr.v4; 2774 break; 2775 case AF_INET6: 2776 ip6 = mtod(pd->m, struct ip6_hdr *); 2777 memset(ip6, 0, hlen); 2778 ip6->ip6_vfc = IPV6_VERSION; 2779 ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20); 2780 ip6->ip6_plen = htons(dlen); 2781 ip6->ip6_nxt = pd->proto; 2782 if (!pd->ttl || pd->ttl > IPV6_DEFHLIM) 2783 ip6->ip6_hlim = IPV6_DEFHLIM; 2784 else 2785 ip6->ip6_hlim = pd->ttl; 2786 ip6->ip6_src = pd->nsaddr.v6; 2787 ip6->ip6_dst = pd->ndaddr.v6; 2788 break; 2789 default: 2790 unhandled_af(pd->naf); 2791 } 2792 2793 /* UDP over IPv6 must be checksummed per rfc2460 p27 */ 2794 if (naf_proto == IPPROTO_UDP && *pc == 0x0000 && 2795 pd->naf == AF_INET6) { 2796 pd->m->m_pkthdr.csum_flags |= M_UDP_CSUM_OUT; 2797 } 2798 2799 /* cover fresh pseudo-header */ 2800 switch (naf_proto) { 2801 case IPPROTO_ICMPV6: 2802 /* optimise: unchanged for TCP/UDP */ 2803 pf_cksum_fixup(pc, 0x0, htons(naf_proto), naf_proto); 2804 pf_cksum_fixup(pc, 0x0, htons(dlen), naf_proto); 2805 /* FALLTHROUGH */ 2806 case IPPROTO_UDP: /* FALLTHROUGH */ 2807 case IPPROTO_TCP: 2808 pf_cksum_fixup_a(pc, &zero, &pd->nsaddr, pd->naf, naf_proto); 2809 pf_cksum_fixup_a(pc, &zero, &pd->ndaddr, pd->naf, naf_proto); 2810 copyback = 1; 2811 break; 2812 default: 2813 break; /* assume no pseudo-header */ 2814 } 2815 2816 /* flush pd->pcksum */ 2817 if (copyback) 2818 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 2819 2820 return (0); 2821 } 2822 2823 int 2824 pf_change_icmp_af(struct mbuf *m, int ipoff2, struct pf_pdesc *pd, 2825 struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst, 2826 sa_family_t af, sa_family_t naf) 2827 { 2828 struct mbuf *n = NULL; 2829 struct ip *ip4; 2830 struct ip6_hdr *ip6; 2831 u_int hlen, ohlen, dlen; 2832 int d; 2833 2834 if (af == naf || (af != AF_INET && af != AF_INET6) || 2835 (naf != AF_INET && naf != AF_INET6)) 2836 return (-1); 2837 2838 /* split the mbuf chain on the quoted ip/ip6 header boundary */ 2839 if ((n = m_split(m, ipoff2, M_DONTWAIT)) == NULL) 2840 return (-1); 2841 2842 /* new quoted header */ 2843 hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 2844 /* old quoted header */ 2845 ohlen = pd2->off - ipoff2; 2846 2847 /* trim old quoted header */ 2848 pf_cksum_uncover(pd->pcksum, in_cksum(n, ohlen), pd->proto); 2849 m_adj(n, ohlen); 2850 2851 /* prepend a new, translated, quoted header */ 2852 if ((M_PREPEND(n, hlen, M_DONTWAIT)) == NULL) 2853 return (-1); 2854 2855 switch (naf) { 2856 case AF_INET: 2857 ip4 = mtod(n, struct ip *); 2858 memset(ip4, 0, sizeof(*ip4)); 2859 ip4->ip_v = IPVERSION; 2860 ip4->ip_hl = sizeof(*ip4) >> 2; 2861 ip4->ip_len = htons(sizeof(*ip4) + pd2->tot_len - ohlen); 2862 ip4->ip_id = htons(ip_randomid()); 2863 ip4->ip_off = htons(IP_DF); 2864 ip4->ip_ttl = pd2->ttl; 2865 if (pd2->proto == IPPROTO_ICMPV6) 2866 ip4->ip_p = IPPROTO_ICMP; 2867 else 2868 ip4->ip_p = pd2->proto; 2869 ip4->ip_src = src->v4; 2870 ip4->ip_dst = dst->v4; 2871 in_hdr_cksum_out(n, NULL); 2872 break; 2873 case AF_INET6: 2874 ip6 = mtod(n, struct ip6_hdr *); 2875 memset(ip6, 0, sizeof(*ip6)); 2876 ip6->ip6_vfc = IPV6_VERSION; 2877 ip6->ip6_plen = htons(pd2->tot_len - ohlen); 2878 if (pd2->proto == IPPROTO_ICMP) 2879 ip6->ip6_nxt = IPPROTO_ICMPV6; 2880 else 2881 ip6->ip6_nxt = pd2->proto; 2882 if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM) 2883 ip6->ip6_hlim = IPV6_DEFHLIM; 2884 else 2885 ip6->ip6_hlim = pd2->ttl; 2886 ip6->ip6_src = src->v6; 2887 ip6->ip6_dst = dst->v6; 2888 break; 2889 } 2890 2891 /* cover new quoted header */ 2892 /* optimise: any new AF_INET header of ours sums to zero */ 2893 if (naf != AF_INET) { 2894 pf_cksum_cover(pd->pcksum, in_cksum(n, hlen), pd->proto); 2895 } 2896 2897 /* reattach modified quoted packet to outer header */ 2898 { 2899 int nlen = n->m_pkthdr.len; 2900 m_cat(m, n); 2901 m->m_pkthdr.len += nlen; 2902 } 2903 2904 /* account for altered length */ 2905 d = hlen - ohlen; 2906 2907 if (pd->proto == IPPROTO_ICMPV6) { 2908 /* fixup pseudo-header */ 2909 dlen = pd->tot_len - pd->off; 2910 pf_cksum_fixup(pd->pcksum, 2911 htons(dlen), htons(dlen + d), pd->proto); 2912 } 2913 2914 pd->tot_len += d; 2915 pd2->tot_len += d; 2916 pd2->off += d; 2917 2918 /* note: not bothering to update network headers as 2919 these due for rewrite by pf_translate_af() */ 2920 2921 return (0); 2922 } 2923 2924 2925 #define PTR_IP(field) (offsetof(struct ip, field)) 2926 #define PTR_IP6(field) (offsetof(struct ip6_hdr, field)) 2927 2928 int 2929 pf_translate_icmp_af(struct pf_pdesc *pd, int af, void *arg) 2930 { 2931 struct icmp *icmp4; 2932 struct icmp6_hdr *icmp6; 2933 u_int32_t mtu; 2934 int32_t ptr = -1; 2935 u_int8_t type; 2936 u_int8_t code; 2937 2938 switch (af) { 2939 case AF_INET: 2940 icmp6 = arg; 2941 type = icmp6->icmp6_type; 2942 code = icmp6->icmp6_code; 2943 mtu = ntohl(icmp6->icmp6_mtu); 2944 2945 switch (type) { 2946 case ICMP6_ECHO_REQUEST: 2947 type = ICMP_ECHO; 2948 break; 2949 case ICMP6_ECHO_REPLY: 2950 type = ICMP_ECHOREPLY; 2951 break; 2952 case ICMP6_DST_UNREACH: 2953 type = ICMP_UNREACH; 2954 switch (code) { 2955 case ICMP6_DST_UNREACH_NOROUTE: 2956 case ICMP6_DST_UNREACH_BEYONDSCOPE: 2957 case ICMP6_DST_UNREACH_ADDR: 2958 code = ICMP_UNREACH_HOST; 2959 break; 2960 case ICMP6_DST_UNREACH_ADMIN: 2961 code = ICMP_UNREACH_HOST_PROHIB; 2962 break; 2963 case ICMP6_DST_UNREACH_NOPORT: 2964 code = ICMP_UNREACH_PORT; 2965 break; 2966 default: 2967 return (-1); 2968 } 2969 break; 2970 case ICMP6_PACKET_TOO_BIG: 2971 type = ICMP_UNREACH; 2972 code = ICMP_UNREACH_NEEDFRAG; 2973 mtu -= 20; 2974 break; 2975 case ICMP6_TIME_EXCEEDED: 2976 type = ICMP_TIMXCEED; 2977 break; 2978 case ICMP6_PARAM_PROB: 2979 switch (code) { 2980 case ICMP6_PARAMPROB_HEADER: 2981 type = ICMP_PARAMPROB; 2982 code = ICMP_PARAMPROB_ERRATPTR; 2983 ptr = ntohl(icmp6->icmp6_pptr); 2984 2985 if (ptr == PTR_IP6(ip6_vfc)) 2986 ; /* preserve */ 2987 else if (ptr == PTR_IP6(ip6_vfc) + 1) 2988 ptr = PTR_IP(ip_tos); 2989 else if (ptr == PTR_IP6(ip6_plen) || 2990 ptr == PTR_IP6(ip6_plen) + 1) 2991 ptr = PTR_IP(ip_len); 2992 else if (ptr == PTR_IP6(ip6_nxt)) 2993 ptr = PTR_IP(ip_p); 2994 else if (ptr == PTR_IP6(ip6_hlim)) 2995 ptr = PTR_IP(ip_ttl); 2996 else if (ptr >= PTR_IP6(ip6_src) && 2997 ptr < PTR_IP6(ip6_dst)) 2998 ptr = PTR_IP(ip_src); 2999 else if (ptr >= PTR_IP6(ip6_dst) && 3000 ptr < sizeof(struct ip6_hdr)) 3001 ptr = PTR_IP(ip_dst); 3002 else { 3003 return (-1); 3004 } 3005 break; 3006 case ICMP6_PARAMPROB_NEXTHEADER: 3007 type = ICMP_UNREACH; 3008 code = ICMP_UNREACH_PROTOCOL; 3009 break; 3010 default: 3011 return (-1); 3012 } 3013 break; 3014 default: 3015 return (-1); 3016 } 3017 3018 pf_patch_8(pd, &icmp6->icmp6_type, type, PF_HI); 3019 pf_patch_8(pd, &icmp6->icmp6_code, code, PF_LO); 3020 3021 /* aligns well with a icmpv4 nextmtu */ 3022 pf_patch_32(pd, &icmp6->icmp6_mtu, htonl(mtu)); 3023 3024 /* icmpv4 pptr is a one most significant byte */ 3025 if (ptr >= 0) 3026 pf_patch_32(pd, &icmp6->icmp6_pptr, htonl(ptr << 24)); 3027 break; 3028 case AF_INET6: 3029 icmp4 = arg; 3030 type = icmp4->icmp_type; 3031 code = icmp4->icmp_code; 3032 mtu = ntohs(icmp4->icmp_nextmtu); 3033 3034 switch (type) { 3035 case ICMP_ECHO: 3036 type = ICMP6_ECHO_REQUEST; 3037 break; 3038 case ICMP_ECHOREPLY: 3039 type = ICMP6_ECHO_REPLY; 3040 break; 3041 case ICMP_UNREACH: 3042 type = ICMP6_DST_UNREACH; 3043 switch (code) { 3044 case ICMP_UNREACH_NET: 3045 case ICMP_UNREACH_HOST: 3046 case ICMP_UNREACH_NET_UNKNOWN: 3047 case ICMP_UNREACH_HOST_UNKNOWN: 3048 case ICMP_UNREACH_ISOLATED: 3049 case ICMP_UNREACH_TOSNET: 3050 case ICMP_UNREACH_TOSHOST: 3051 code = ICMP6_DST_UNREACH_NOROUTE; 3052 break; 3053 case ICMP_UNREACH_PORT: 3054 code = ICMP6_DST_UNREACH_NOPORT; 3055 break; 3056 case ICMP_UNREACH_NET_PROHIB: 3057 case ICMP_UNREACH_HOST_PROHIB: 3058 case ICMP_UNREACH_FILTER_PROHIB: 3059 case ICMP_UNREACH_PRECEDENCE_CUTOFF: 3060 code = ICMP6_DST_UNREACH_ADMIN; 3061 break; 3062 case ICMP_UNREACH_PROTOCOL: 3063 type = ICMP6_PARAM_PROB; 3064 code = ICMP6_PARAMPROB_NEXTHEADER; 3065 ptr = offsetof(struct ip6_hdr, ip6_nxt); 3066 break; 3067 case ICMP_UNREACH_NEEDFRAG: 3068 type = ICMP6_PACKET_TOO_BIG; 3069 code = 0; 3070 mtu += 20; 3071 break; 3072 default: 3073 return (-1); 3074 } 3075 break; 3076 case ICMP_TIMXCEED: 3077 type = ICMP6_TIME_EXCEEDED; 3078 break; 3079 case ICMP_PARAMPROB: 3080 type = ICMP6_PARAM_PROB; 3081 switch (code) { 3082 case ICMP_PARAMPROB_ERRATPTR: 3083 code = ICMP6_PARAMPROB_HEADER; 3084 break; 3085 case ICMP_PARAMPROB_LENGTH: 3086 code = ICMP6_PARAMPROB_HEADER; 3087 break; 3088 default: 3089 return (-1); 3090 } 3091 3092 ptr = icmp4->icmp_pptr; 3093 if (ptr == 0 || ptr == PTR_IP(ip_tos)) 3094 ; /* preserve */ 3095 else if (ptr == PTR_IP(ip_len) || 3096 ptr == PTR_IP(ip_len) + 1) 3097 ptr = PTR_IP6(ip6_plen); 3098 else if (ptr == PTR_IP(ip_ttl)) 3099 ptr = PTR_IP6(ip6_hlim); 3100 else if (ptr == PTR_IP(ip_p)) 3101 ptr = PTR_IP6(ip6_nxt); 3102 else if (ptr >= PTR_IP(ip_src) && 3103 ptr < PTR_IP(ip_dst)) 3104 ptr = PTR_IP6(ip6_src); 3105 else if (ptr >= PTR_IP(ip_dst) && 3106 ptr < sizeof(struct ip)) 3107 ptr = PTR_IP6(ip6_dst); 3108 else { 3109 return (-1); 3110 } 3111 break; 3112 default: 3113 return (-1); 3114 } 3115 3116 pf_patch_8(pd, &icmp4->icmp_type, type, PF_HI); 3117 pf_patch_8(pd, &icmp4->icmp_code, code, PF_LO); 3118 pf_patch_16(pd, &icmp4->icmp_nextmtu, htons(mtu)); 3119 if (ptr >= 0) 3120 pf_patch_32(pd, &icmp4->icmp_void, htonl(ptr)); 3121 break; 3122 } 3123 3124 return (0); 3125 } 3126 #endif /* INET6 */ 3127 3128 /* 3129 * Need to modulate the sequence numbers in the TCP SACK option 3130 * (credits to Krzysztof Pfaff for report and patch) 3131 */ 3132 int 3133 pf_modulate_sack(struct pf_pdesc *pd, struct pf_state_peer *dst) 3134 { 3135 struct sackblk sack; 3136 int copyback = 0, i; 3137 int olen, optsoff; 3138 u_int8_t opts[MAX_TCPOPTLEN], *opt, *eoh; 3139 3140 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 3141 optsoff = pd->off + sizeof(struct tcphdr); 3142 #define TCPOLEN_MINSACK (TCPOLEN_SACK + 2) 3143 if (olen < TCPOLEN_MINSACK || 3144 !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, NULL, pd->af)) 3145 return (0); 3146 3147 eoh = opts + olen; 3148 opt = opts; 3149 while ((opt = pf_find_tcpopt(opt, opts, olen, 3150 TCPOPT_SACK, TCPOLEN_MINSACK)) != NULL) 3151 { 3152 size_t safelen = MIN(opt[1], (eoh - opt)); 3153 for (i = 2; i + TCPOLEN_SACK <= safelen; i += TCPOLEN_SACK) { 3154 size_t startoff = (opt + i) - opts; 3155 memcpy(&sack, &opt[i], sizeof(sack)); 3156 pf_patch_32_unaligned(pd, &sack.start, 3157 htonl(ntohl(sack.start) - dst->seqdiff), 3158 PF_ALGNMNT(startoff)); 3159 pf_patch_32_unaligned(pd, &sack.end, 3160 htonl(ntohl(sack.end) - dst->seqdiff), 3161 PF_ALGNMNT(startoff + sizeof(sack.start))); 3162 memcpy(&opt[i], &sack, sizeof(sack)); 3163 } 3164 copyback = 1; 3165 opt += opt[1]; 3166 } 3167 3168 if (copyback) 3169 m_copyback(pd->m, optsoff, olen, opts, M_NOWAIT); 3170 return (copyback); 3171 } 3172 3173 struct mbuf * 3174 pf_build_tcp(const struct pf_rule *r, sa_family_t af, 3175 const struct pf_addr *saddr, const struct pf_addr *daddr, 3176 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3177 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 3178 u_int16_t rtag, u_int sack, u_int rdom) 3179 { 3180 struct mbuf *m; 3181 int len, tlen; 3182 struct ip *h; 3183 #ifdef INET6 3184 struct ip6_hdr *h6; 3185 #endif /* INET6 */ 3186 struct tcphdr *th; 3187 char *opt; 3188 3189 /* maximum segment size tcp option */ 3190 tlen = sizeof(struct tcphdr); 3191 if (mss) 3192 tlen += 4; 3193 if (sack) 3194 tlen += 2; 3195 3196 switch (af) { 3197 case AF_INET: 3198 len = sizeof(struct ip) + tlen; 3199 break; 3200 #ifdef INET6 3201 case AF_INET6: 3202 len = sizeof(struct ip6_hdr) + tlen; 3203 break; 3204 #endif /* INET6 */ 3205 default: 3206 unhandled_af(af); 3207 } 3208 3209 /* create outgoing mbuf */ 3210 m = m_gethdr(M_DONTWAIT, MT_HEADER); 3211 if (m == NULL) 3212 return (NULL); 3213 if (tag) 3214 m->m_pkthdr.pf.flags |= PF_TAG_GENERATED; 3215 m->m_pkthdr.pf.tag = rtag; 3216 m->m_pkthdr.ph_rtableid = rdom; 3217 if (r && (r->scrub_flags & PFSTATE_SETPRIO)) 3218 m->m_pkthdr.pf.prio = r->set_prio[0]; 3219 if (r && r->qid) 3220 m->m_pkthdr.pf.qid = r->qid; 3221 m->m_data += max_linkhdr; 3222 m->m_pkthdr.len = m->m_len = len; 3223 m->m_pkthdr.ph_ifidx = 0; 3224 m->m_pkthdr.csum_flags |= M_TCP_CSUM_OUT; 3225 memset(m->m_data, 0, len); 3226 switch (af) { 3227 case AF_INET: 3228 h = mtod(m, struct ip *); 3229 h->ip_p = IPPROTO_TCP; 3230 h->ip_len = htons(tlen); 3231 h->ip_v = 4; 3232 h->ip_hl = sizeof(*h) >> 2; 3233 h->ip_tos = IPTOS_LOWDELAY; 3234 h->ip_len = htons(len); 3235 h->ip_off = htons(ip_mtudisc ? IP_DF : 0); 3236 h->ip_ttl = ttl ? ttl : ip_defttl; 3237 h->ip_sum = 0; 3238 h->ip_src.s_addr = saddr->v4.s_addr; 3239 h->ip_dst.s_addr = daddr->v4.s_addr; 3240 3241 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3242 break; 3243 #ifdef INET6 3244 case AF_INET6: 3245 h6 = mtod(m, struct ip6_hdr *); 3246 h6->ip6_nxt = IPPROTO_TCP; 3247 h6->ip6_plen = htons(tlen); 3248 h6->ip6_vfc |= IPV6_VERSION; 3249 h6->ip6_hlim = IPV6_DEFHLIM; 3250 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3251 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3252 3253 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3254 break; 3255 #endif /* INET6 */ 3256 default: 3257 unhandled_af(af); 3258 } 3259 3260 /* TCP header */ 3261 th->th_sport = sport; 3262 th->th_dport = dport; 3263 th->th_seq = htonl(seq); 3264 th->th_ack = htonl(ack); 3265 th->th_off = tlen >> 2; 3266 th->th_flags = flags; 3267 th->th_win = htons(win); 3268 3269 opt = (char *)(th + 1); 3270 if (mss) { 3271 opt[0] = TCPOPT_MAXSEG; 3272 opt[1] = 4; 3273 mss = htons(mss); 3274 memcpy((opt + 2), &mss, 2); 3275 opt += 4; 3276 } 3277 if (sack) { 3278 opt[0] = TCPOPT_SACK_PERMITTED; 3279 opt[1] = 2; 3280 opt += 2; 3281 } 3282 3283 return (m); 3284 } 3285 3286 void 3287 pf_send_tcp(const struct pf_rule *r, sa_family_t af, 3288 const struct pf_addr *saddr, const struct pf_addr *daddr, 3289 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3290 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 3291 u_int16_t rtag, u_int rdom) 3292 { 3293 struct mbuf *m; 3294 3295 if ((m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, 3296 flags, win, mss, ttl, tag, rtag, 0, rdom)) == NULL) 3297 return; 3298 3299 switch (af) { 3300 case AF_INET: 3301 ip_send(m); 3302 break; 3303 #ifdef INET6 3304 case AF_INET6: 3305 ip6_send(m); 3306 break; 3307 #endif /* INET6 */ 3308 } 3309 } 3310 3311 static void 3312 pf_send_challenge_ack(struct pf_pdesc *pd, struct pf_state *st, 3313 struct pf_state_peer *src, struct pf_state_peer *dst) 3314 { 3315 /* 3316 * We are sending challenge ACK as a response to SYN packet, which 3317 * matches existing state (modulo TCP window check). Therefore packet 3318 * must be sent on behalf of destination. 3319 * 3320 * We expect sender to remain either silent, or send RST packet 3321 * so both, firewall and remote peer, can purge dead state from 3322 * memory. 3323 */ 3324 pf_send_tcp(st->rule.ptr, pd->af, pd->dst, pd->src, 3325 pd->hdr.tcp.th_dport, pd->hdr.tcp.th_sport, dst->seqlo, 3326 src->seqlo, TH_ACK, 0, 0, st->rule.ptr->return_ttl, 1, 0, 3327 pd->rdomain); 3328 } 3329 3330 void 3331 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, int param, 3332 sa_family_t af, struct pf_rule *r, u_int rdomain) 3333 { 3334 struct mbuf *m0; 3335 3336 if ((m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT)) == NULL) 3337 return; 3338 3339 m0->m_pkthdr.pf.flags |= PF_TAG_GENERATED; 3340 m0->m_pkthdr.ph_rtableid = rdomain; 3341 if (r && (r->scrub_flags & PFSTATE_SETPRIO)) 3342 m0->m_pkthdr.pf.prio = r->set_prio[0]; 3343 if (r && r->qid) 3344 m0->m_pkthdr.pf.qid = r->qid; 3345 3346 switch (af) { 3347 case AF_INET: 3348 icmp_error(m0, type, code, 0, param); 3349 break; 3350 #ifdef INET6 3351 case AF_INET6: 3352 icmp6_error(m0, type, code, param); 3353 break; 3354 #endif /* INET6 */ 3355 } 3356 } 3357 3358 /* 3359 * Return ((n = 0) == (a = b [with mask m])) 3360 * Note: n != 0 => returns (a != b [with mask m]) 3361 */ 3362 int 3363 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3364 struct pf_addr *b, sa_family_t af) 3365 { 3366 switch (af) { 3367 case AF_INET: 3368 if ((a->addr32[0] & m->addr32[0]) == 3369 (b->addr32[0] & m->addr32[0])) 3370 return (n == 0); 3371 break; 3372 #ifdef INET6 3373 case AF_INET6: 3374 if (((a->addr32[0] & m->addr32[0]) == 3375 (b->addr32[0] & m->addr32[0])) && 3376 ((a->addr32[1] & m->addr32[1]) == 3377 (b->addr32[1] & m->addr32[1])) && 3378 ((a->addr32[2] & m->addr32[2]) == 3379 (b->addr32[2] & m->addr32[2])) && 3380 ((a->addr32[3] & m->addr32[3]) == 3381 (b->addr32[3] & m->addr32[3]))) 3382 return (n == 0); 3383 break; 3384 #endif /* INET6 */ 3385 } 3386 3387 return (n != 0); 3388 } 3389 3390 /* 3391 * Return 1 if b <= a <= e, otherwise return 0. 3392 */ 3393 int 3394 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3395 struct pf_addr *a, sa_family_t af) 3396 { 3397 switch (af) { 3398 case AF_INET: 3399 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3400 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3401 return (0); 3402 break; 3403 #ifdef INET6 3404 case AF_INET6: { 3405 int i; 3406 3407 /* check a >= b */ 3408 for (i = 0; i < 4; ++i) 3409 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3410 break; 3411 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3412 return (0); 3413 /* check a <= e */ 3414 for (i = 0; i < 4; ++i) 3415 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3416 break; 3417 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3418 return (0); 3419 break; 3420 } 3421 #endif /* INET6 */ 3422 } 3423 return (1); 3424 } 3425 3426 int 3427 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3428 { 3429 switch (op) { 3430 case PF_OP_IRG: 3431 return ((p > a1) && (p < a2)); 3432 case PF_OP_XRG: 3433 return ((p < a1) || (p > a2)); 3434 case PF_OP_RRG: 3435 return ((p >= a1) && (p <= a2)); 3436 case PF_OP_EQ: 3437 return (p == a1); 3438 case PF_OP_NE: 3439 return (p != a1); 3440 case PF_OP_LT: 3441 return (p < a1); 3442 case PF_OP_LE: 3443 return (p <= a1); 3444 case PF_OP_GT: 3445 return (p > a1); 3446 case PF_OP_GE: 3447 return (p >= a1); 3448 } 3449 return (0); /* never reached */ 3450 } 3451 3452 int 3453 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3454 { 3455 return (pf_match(op, ntohs(a1), ntohs(a2), ntohs(p))); 3456 } 3457 3458 int 3459 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3460 { 3461 if (u == -1 && op != PF_OP_EQ && op != PF_OP_NE) 3462 return (0); 3463 return (pf_match(op, a1, a2, u)); 3464 } 3465 3466 int 3467 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3468 { 3469 if (g == -1 && op != PF_OP_EQ && op != PF_OP_NE) 3470 return (0); 3471 return (pf_match(op, a1, a2, g)); 3472 } 3473 3474 int 3475 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag) 3476 { 3477 if (*tag == -1) 3478 *tag = m->m_pkthdr.pf.tag; 3479 3480 return ((!r->match_tag_not && r->match_tag == *tag) || 3481 (r->match_tag_not && r->match_tag != *tag)); 3482 } 3483 3484 int 3485 pf_match_rcvif(struct mbuf *m, struct pf_rule *r) 3486 { 3487 struct ifnet *ifp; 3488 #if NCARP > 0 3489 struct ifnet *ifp0; 3490 #endif 3491 struct pfi_kif *kif; 3492 3493 ifp = if_get(m->m_pkthdr.ph_ifidx); 3494 if (ifp == NULL) 3495 return (0); 3496 3497 #if NCARP > 0 3498 if (ifp->if_type == IFT_CARP && 3499 (ifp0 = if_get(ifp->if_carpdevidx)) != NULL) { 3500 kif = (struct pfi_kif *)ifp0->if_pf_kif; 3501 if_put(ifp0); 3502 } else 3503 #endif /* NCARP */ 3504 kif = (struct pfi_kif *)ifp->if_pf_kif; 3505 3506 if_put(ifp); 3507 3508 if (kif == NULL) { 3509 DPFPRINTF(LOG_ERR, 3510 "%s: kif == NULL, @%d via %s", __func__, 3511 r->nr, r->rcv_ifname); 3512 return (0); 3513 } 3514 3515 return (pfi_kif_match(r->rcv_kif, kif)); 3516 } 3517 3518 void 3519 pf_tag_packet(struct mbuf *m, int tag, int rtableid) 3520 { 3521 if (tag > 0) 3522 m->m_pkthdr.pf.tag = tag; 3523 if (rtableid >= 0) 3524 m->m_pkthdr.ph_rtableid = (u_int)rtableid; 3525 } 3526 3527 void 3528 pf_anchor_stack_init(void) 3529 { 3530 struct pf_anchor_stackframe *stack; 3531 3532 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3533 stack[PF_ANCHOR_STACK_MAX].sf_stack_top = &stack[0]; 3534 cpumem_leave(pf_anchor_stack, stack); 3535 } 3536 3537 int 3538 pf_anchor_stack_is_full(struct pf_anchor_stackframe *sf) 3539 { 3540 struct pf_anchor_stackframe *stack; 3541 int rv; 3542 3543 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3544 rv = (sf == &stack[PF_ANCHOR_STACK_MAX]); 3545 cpumem_leave(pf_anchor_stack, stack); 3546 3547 return (rv); 3548 } 3549 3550 int 3551 pf_anchor_stack_is_empty(struct pf_anchor_stackframe *sf) 3552 { 3553 struct pf_anchor_stackframe *stack; 3554 int rv; 3555 3556 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3557 rv = (sf == &stack[0]); 3558 cpumem_leave(pf_anchor_stack, stack); 3559 3560 return (rv); 3561 } 3562 3563 struct pf_anchor_stackframe * 3564 pf_anchor_stack_top(void) 3565 { 3566 struct pf_anchor_stackframe *stack; 3567 struct pf_anchor_stackframe *top_sf; 3568 3569 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3570 top_sf = stack[PF_ANCHOR_STACK_MAX].sf_stack_top; 3571 cpumem_leave(pf_anchor_stack, stack); 3572 3573 return (top_sf); 3574 } 3575 3576 int 3577 pf_anchor_stack_push(struct pf_ruleset *rs, struct pf_rule *r, 3578 struct pf_anchor *child, int jump_target) 3579 { 3580 struct pf_anchor_stackframe *stack; 3581 struct pf_anchor_stackframe *top_sf = pf_anchor_stack_top(); 3582 3583 top_sf++; 3584 if (pf_anchor_stack_is_full(top_sf)) 3585 return (-1); 3586 3587 top_sf->sf_rs = rs; 3588 top_sf->sf_r = r; 3589 top_sf->sf_child = child; 3590 top_sf->sf_jump_target = jump_target; 3591 3592 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3593 3594 if ((top_sf <= &stack[0]) || (top_sf >= &stack[PF_ANCHOR_STACK_MAX])) 3595 panic("%s: top frame outside of anchor stack range", __func__); 3596 3597 stack[PF_ANCHOR_STACK_MAX].sf_stack_top = top_sf; 3598 cpumem_leave(pf_anchor_stack, stack); 3599 3600 return (0); 3601 } 3602 3603 int 3604 pf_anchor_stack_pop(struct pf_ruleset **rs, struct pf_rule **r, 3605 struct pf_anchor **child, int *jump_target) 3606 { 3607 struct pf_anchor_stackframe *top_sf = pf_anchor_stack_top(); 3608 struct pf_anchor_stackframe *stack; 3609 int on_top; 3610 3611 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3612 if (pf_anchor_stack_is_empty(top_sf)) { 3613 on_top = -1; 3614 } else { 3615 if ((top_sf <= &stack[0]) || 3616 (top_sf >= &stack[PF_ANCHOR_STACK_MAX])) 3617 panic("%s: top frame outside of anchor stack range", 3618 __func__); 3619 3620 *rs = top_sf->sf_rs; 3621 *r = top_sf->sf_r; 3622 *child = top_sf->sf_child; 3623 *jump_target = top_sf->sf_jump_target; 3624 top_sf--; 3625 stack[PF_ANCHOR_STACK_MAX].sf_stack_top = top_sf; 3626 on_top = 0; 3627 } 3628 cpumem_leave(pf_anchor_stack, stack); 3629 3630 return (on_top); 3631 } 3632 3633 void 3634 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3635 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3636 { 3637 switch (af) { 3638 case AF_INET: 3639 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3640 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3641 break; 3642 #ifdef INET6 3643 case AF_INET6: 3644 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3645 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3646 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3647 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3648 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3649 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3650 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3651 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3652 break; 3653 #endif /* INET6 */ 3654 default: 3655 unhandled_af(af); 3656 } 3657 } 3658 3659 void 3660 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3661 { 3662 switch (af) { 3663 case AF_INET: 3664 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3665 break; 3666 #ifdef INET6 3667 case AF_INET6: 3668 if (addr->addr32[3] == 0xffffffff) { 3669 addr->addr32[3] = 0; 3670 if (addr->addr32[2] == 0xffffffff) { 3671 addr->addr32[2] = 0; 3672 if (addr->addr32[1] == 0xffffffff) { 3673 addr->addr32[1] = 0; 3674 addr->addr32[0] = 3675 htonl(ntohl(addr->addr32[0]) + 1); 3676 } else 3677 addr->addr32[1] = 3678 htonl(ntohl(addr->addr32[1]) + 1); 3679 } else 3680 addr->addr32[2] = 3681 htonl(ntohl(addr->addr32[2]) + 1); 3682 } else 3683 addr->addr32[3] = 3684 htonl(ntohl(addr->addr32[3]) + 1); 3685 break; 3686 #endif /* INET6 */ 3687 default: 3688 unhandled_af(af); 3689 } 3690 } 3691 3692 int 3693 pf_socket_lookup(struct pf_pdesc *pd) 3694 { 3695 struct pf_addr *saddr, *daddr; 3696 u_int16_t sport, dport; 3697 struct inpcbtable *tb; 3698 struct inpcb *inp; 3699 3700 pd->lookup.uid = -1; 3701 pd->lookup.gid = -1; 3702 pd->lookup.pid = NO_PID; 3703 switch (pd->virtual_proto) { 3704 case IPPROTO_TCP: 3705 sport = pd->hdr.tcp.th_sport; 3706 dport = pd->hdr.tcp.th_dport; 3707 PF_ASSERT_LOCKED(); 3708 NET_ASSERT_LOCKED(); 3709 tb = &tcbtable; 3710 break; 3711 case IPPROTO_UDP: 3712 sport = pd->hdr.udp.uh_sport; 3713 dport = pd->hdr.udp.uh_dport; 3714 PF_ASSERT_LOCKED(); 3715 NET_ASSERT_LOCKED(); 3716 tb = &udbtable; 3717 break; 3718 default: 3719 return (-1); 3720 } 3721 if (pd->dir == PF_IN) { 3722 saddr = pd->src; 3723 daddr = pd->dst; 3724 } else { 3725 u_int16_t p; 3726 3727 p = sport; 3728 sport = dport; 3729 dport = p; 3730 saddr = pd->dst; 3731 daddr = pd->src; 3732 } 3733 switch (pd->af) { 3734 case AF_INET: 3735 /* 3736 * Fails when rtable is changed while evaluating the ruleset 3737 * The socket looked up will not match the one hit in the end. 3738 */ 3739 inp = in_pcblookup(tb, saddr->v4, sport, daddr->v4, dport, 3740 pd->rdomain); 3741 if (inp == NULL) { 3742 inp = in_pcblookup_listen(tb, daddr->v4, dport, 3743 NULL, pd->rdomain); 3744 if (inp == NULL) 3745 return (-1); 3746 } 3747 break; 3748 #ifdef INET6 3749 case AF_INET6: 3750 inp = in6_pcblookup(tb, &saddr->v6, sport, &daddr->v6, 3751 dport, pd->rdomain); 3752 if (inp == NULL) { 3753 inp = in6_pcblookup_listen(tb, &daddr->v6, dport, 3754 NULL, pd->rdomain); 3755 if (inp == NULL) 3756 return (-1); 3757 } 3758 break; 3759 #endif /* INET6 */ 3760 default: 3761 unhandled_af(pd->af); 3762 } 3763 pd->lookup.uid = inp->inp_socket->so_euid; 3764 pd->lookup.gid = inp->inp_socket->so_egid; 3765 pd->lookup.pid = inp->inp_socket->so_cpid; 3766 in_pcbunref(inp); 3767 return (1); 3768 } 3769 3770 /* post: r => (r[0] == type /\ r[1] >= min_typelen >= 2 "validity" 3771 * /\ (eoh - r) >= min_typelen >= 2 "safety" ) 3772 * 3773 * warning: r + r[1] may exceed opts bounds for r[1] > min_typelen 3774 */ 3775 u_int8_t* 3776 pf_find_tcpopt(u_int8_t *opt, u_int8_t *opts, size_t hlen, u_int8_t type, 3777 u_int8_t min_typelen) 3778 { 3779 u_int8_t *eoh = opts + hlen; 3780 3781 if (min_typelen < 2) 3782 return (NULL); 3783 3784 while ((eoh - opt) >= min_typelen) { 3785 switch (*opt) { 3786 case TCPOPT_EOL: 3787 /* FALLTHROUGH - Workaround the failure of some 3788 systems to NOP-pad their bzero'd option buffers, 3789 producing spurious EOLs */ 3790 case TCPOPT_NOP: 3791 opt++; 3792 continue; 3793 default: 3794 if (opt[0] == type && 3795 opt[1] >= min_typelen) 3796 return (opt); 3797 } 3798 3799 opt += MAX(opt[1], 2); /* evade infinite loops */ 3800 } 3801 3802 return (NULL); 3803 } 3804 3805 u_int8_t 3806 pf_get_wscale(struct pf_pdesc *pd) 3807 { 3808 int olen; 3809 u_int8_t opts[MAX_TCPOPTLEN], *opt; 3810 u_int8_t wscale = 0; 3811 3812 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 3813 if (olen < TCPOLEN_WINDOW || !pf_pull_hdr(pd->m, 3814 pd->off + sizeof(struct tcphdr), opts, olen, NULL, NULL, pd->af)) 3815 return (0); 3816 3817 opt = opts; 3818 while ((opt = pf_find_tcpopt(opt, opts, olen, 3819 TCPOPT_WINDOW, TCPOLEN_WINDOW)) != NULL) { 3820 wscale = opt[2]; 3821 wscale = MIN(wscale, TCP_MAX_WINSHIFT); 3822 wscale |= PF_WSCALE_FLAG; 3823 3824 opt += opt[1]; 3825 } 3826 3827 return (wscale); 3828 } 3829 3830 u_int16_t 3831 pf_get_mss(struct pf_pdesc *pd) 3832 { 3833 int olen; 3834 u_int8_t opts[MAX_TCPOPTLEN], *opt; 3835 u_int16_t mss = tcp_mssdflt; 3836 3837 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 3838 if (olen < TCPOLEN_MAXSEG || !pf_pull_hdr(pd->m, 3839 pd->off + sizeof(struct tcphdr), opts, olen, NULL, NULL, pd->af)) 3840 return (0); 3841 3842 opt = opts; 3843 while ((opt = pf_find_tcpopt(opt, opts, olen, 3844 TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) { 3845 memcpy(&mss, (opt + 2), 2); 3846 mss = ntohs(mss); 3847 3848 opt += opt[1]; 3849 } 3850 return (mss); 3851 } 3852 3853 u_int16_t 3854 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 3855 { 3856 struct ifnet *ifp; 3857 struct sockaddr_in *dst; 3858 #ifdef INET6 3859 struct sockaddr_in6 *dst6; 3860 #endif /* INET6 */ 3861 struct rtentry *rt = NULL; 3862 struct sockaddr_storage ss; 3863 int hlen; 3864 u_int16_t mss = tcp_mssdflt; 3865 3866 memset(&ss, 0, sizeof(ss)); 3867 3868 switch (af) { 3869 case AF_INET: 3870 hlen = sizeof(struct ip); 3871 dst = (struct sockaddr_in *)&ss; 3872 dst->sin_family = AF_INET; 3873 dst->sin_len = sizeof(*dst); 3874 dst->sin_addr = addr->v4; 3875 rt = rtalloc(sintosa(dst), 0, rtableid); 3876 break; 3877 #ifdef INET6 3878 case AF_INET6: 3879 hlen = sizeof(struct ip6_hdr); 3880 dst6 = (struct sockaddr_in6 *)&ss; 3881 dst6->sin6_family = AF_INET6; 3882 dst6->sin6_len = sizeof(*dst6); 3883 dst6->sin6_addr = addr->v6; 3884 rt = rtalloc(sin6tosa(dst6), 0, rtableid); 3885 break; 3886 #endif /* INET6 */ 3887 } 3888 3889 if (rt != NULL && (ifp = if_get(rt->rt_ifidx)) != NULL) { 3890 mss = ifp->if_mtu - hlen - sizeof(struct tcphdr); 3891 mss = max(tcp_mssdflt, mss); 3892 if_put(ifp); 3893 } 3894 rtfree(rt); 3895 mss = min(mss, offer); 3896 mss = max(mss, 64); /* sanity - at least max opt space */ 3897 return (mss); 3898 } 3899 3900 static __inline int 3901 pf_set_rt_ifp(struct pf_state *st, struct pf_addr *saddr, sa_family_t af, 3902 struct pf_src_node **sns) 3903 { 3904 struct pf_rule *r = st->rule.ptr; 3905 int rv; 3906 3907 if (!r->rt) 3908 return (0); 3909 3910 rv = pf_map_addr(af, r, saddr, &st->rt_addr, NULL, sns, 3911 &r->route, PF_SN_ROUTE); 3912 if (rv == 0) 3913 st->rt = r->rt; 3914 3915 return (rv); 3916 } 3917 3918 u_int32_t 3919 pf_tcp_iss(struct pf_pdesc *pd) 3920 { 3921 SHA2_CTX ctx; 3922 union { 3923 uint8_t bytes[SHA512_DIGEST_LENGTH]; 3924 uint32_t words[1]; 3925 } digest; 3926 3927 if (pf_tcp_secret_init == 0) { 3928 arc4random_buf(pf_tcp_secret, sizeof(pf_tcp_secret)); 3929 SHA512Init(&pf_tcp_secret_ctx); 3930 SHA512Update(&pf_tcp_secret_ctx, pf_tcp_secret, 3931 sizeof(pf_tcp_secret)); 3932 pf_tcp_secret_init = 1; 3933 } 3934 ctx = pf_tcp_secret_ctx; 3935 3936 SHA512Update(&ctx, &pd->rdomain, sizeof(pd->rdomain)); 3937 SHA512Update(&ctx, &pd->hdr.tcp.th_sport, sizeof(u_short)); 3938 SHA512Update(&ctx, &pd->hdr.tcp.th_dport, sizeof(u_short)); 3939 switch (pd->af) { 3940 case AF_INET: 3941 SHA512Update(&ctx, &pd->src->v4, sizeof(struct in_addr)); 3942 SHA512Update(&ctx, &pd->dst->v4, sizeof(struct in_addr)); 3943 break; 3944 #ifdef INET6 3945 case AF_INET6: 3946 SHA512Update(&ctx, &pd->src->v6, sizeof(struct in6_addr)); 3947 SHA512Update(&ctx, &pd->dst->v6, sizeof(struct in6_addr)); 3948 break; 3949 #endif /* INET6 */ 3950 } 3951 SHA512Final(digest.bytes, &ctx); 3952 pf_tcp_iss_off += 4096; 3953 return (digest.words[0] + READ_ONCE(tcp_iss) + pf_tcp_iss_off); 3954 } 3955 3956 void 3957 pf_rule_to_actions(struct pf_rule *r, struct pf_rule_actions *a) 3958 { 3959 if (r->qid) 3960 a->qid = r->qid; 3961 if (r->pqid) 3962 a->pqid = r->pqid; 3963 if (r->rtableid >= 0) 3964 a->rtableid = r->rtableid; 3965 #if NPFLOG > 0 3966 a->log |= r->log; 3967 #endif /* NPFLOG > 0 */ 3968 if (r->scrub_flags & PFSTATE_SETTOS) 3969 a->set_tos = r->set_tos; 3970 if (r->min_ttl) 3971 a->min_ttl = r->min_ttl; 3972 if (r->max_mss) 3973 a->max_mss = r->max_mss; 3974 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 3975 PFSTATE_SETTOS|PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 3976 if (r->scrub_flags & PFSTATE_SETPRIO) { 3977 a->set_prio[0] = r->set_prio[0]; 3978 a->set_prio[1] = r->set_prio[1]; 3979 } 3980 if (r->rule_flag & PFRULE_SETDELAY) 3981 a->delay = r->delay; 3982 } 3983 3984 #define PF_TEST_ATTRIB(t, a) \ 3985 if (t) { \ 3986 r = a; \ 3987 continue; \ 3988 } else do { \ 3989 } while (0) 3990 3991 enum pf_test_status 3992 pf_match_rule(struct pf_test_ctx *ctx, struct pf_ruleset *ruleset) 3993 { 3994 struct pf_rule *r; 3995 struct pf_anchor *child = NULL; 3996 int target; 3997 3998 pf_anchor_stack_init(); 3999 enter_ruleset: 4000 r = TAILQ_FIRST(ruleset->rules.active.ptr); 4001 while (r != NULL) { 4002 PF_TEST_ATTRIB(r->rule_flag & PFRULE_EXPIRED, 4003 TAILQ_NEXT(r, entries)); 4004 r->evaluations++; 4005 PF_TEST_ATTRIB( 4006 (pfi_kif_match(r->kif, ctx->pd->kif) == r->ifnot), 4007 r->skip[PF_SKIP_IFP].ptr); 4008 PF_TEST_ATTRIB((r->direction && r->direction != ctx->pd->dir), 4009 r->skip[PF_SKIP_DIR].ptr); 4010 PF_TEST_ATTRIB((r->onrdomain >= 0 && 4011 (r->onrdomain == ctx->pd->rdomain) == r->ifnot), 4012 r->skip[PF_SKIP_RDOM].ptr); 4013 PF_TEST_ATTRIB((r->af && r->af != ctx->pd->af), 4014 r->skip[PF_SKIP_AF].ptr); 4015 PF_TEST_ATTRIB((r->proto && r->proto != ctx->pd->proto), 4016 r->skip[PF_SKIP_PROTO].ptr); 4017 PF_TEST_ATTRIB((PF_MISMATCHAW(&r->src.addr, &ctx->pd->nsaddr, 4018 ctx->pd->naf, r->src.neg, ctx->pd->kif, 4019 ctx->act.rtableid)), 4020 r->skip[PF_SKIP_SRC_ADDR].ptr); 4021 PF_TEST_ATTRIB((PF_MISMATCHAW(&r->dst.addr, &ctx->pd->ndaddr, 4022 ctx->pd->af, r->dst.neg, NULL, ctx->act.rtableid)), 4023 r->skip[PF_SKIP_DST_ADDR].ptr); 4024 4025 switch (ctx->pd->virtual_proto) { 4026 case PF_VPROTO_FRAGMENT: 4027 /* tcp/udp only. port_op always 0 in other cases */ 4028 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 4029 TAILQ_NEXT(r, entries)); 4030 PF_TEST_ATTRIB((ctx->pd->proto == IPPROTO_TCP && 4031 r->flagset), 4032 TAILQ_NEXT(r, entries)); 4033 /* icmp only. type/code always 0 in other cases */ 4034 PF_TEST_ATTRIB((r->type || r->code), 4035 TAILQ_NEXT(r, entries)); 4036 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 4037 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 4038 TAILQ_NEXT(r, entries)); 4039 break; 4040 4041 case IPPROTO_TCP: 4042 PF_TEST_ATTRIB(((r->flagset & ctx->th->th_flags) != 4043 r->flags), 4044 TAILQ_NEXT(r, entries)); 4045 PF_TEST_ATTRIB((r->os_fingerprint != PF_OSFP_ANY && 4046 !pf_osfp_match(pf_osfp_fingerprint(ctx->pd), 4047 r->os_fingerprint)), 4048 TAILQ_NEXT(r, entries)); 4049 /* FALLTHROUGH */ 4050 4051 case IPPROTO_UDP: 4052 /* tcp/udp only. port_op always 0 in other cases */ 4053 PF_TEST_ATTRIB((r->src.port_op && 4054 !pf_match_port(r->src.port_op, r->src.port[0], 4055 r->src.port[1], ctx->pd->nsport)), 4056 r->skip[PF_SKIP_SRC_PORT].ptr); 4057 PF_TEST_ATTRIB((r->dst.port_op && 4058 !pf_match_port(r->dst.port_op, r->dst.port[0], 4059 r->dst.port[1], ctx->pd->ndport)), 4060 r->skip[PF_SKIP_DST_PORT].ptr); 4061 /* tcp/udp only. uid.op always 0 in other cases */ 4062 PF_TEST_ATTRIB((r->uid.op && (ctx->pd->lookup.done || 4063 (ctx->pd->lookup.done = 4064 pf_socket_lookup(ctx->pd), 1)) && 4065 !pf_match_uid(r->uid.op, r->uid.uid[0], 4066 r->uid.uid[1], ctx->pd->lookup.uid)), 4067 TAILQ_NEXT(r, entries)); 4068 /* tcp/udp only. gid.op always 0 in other cases */ 4069 PF_TEST_ATTRIB((r->gid.op && (ctx->pd->lookup.done || 4070 (ctx->pd->lookup.done = 4071 pf_socket_lookup(ctx->pd), 1)) && 4072 !pf_match_gid(r->gid.op, r->gid.gid[0], 4073 r->gid.gid[1], ctx->pd->lookup.gid)), 4074 TAILQ_NEXT(r, entries)); 4075 break; 4076 4077 case IPPROTO_ICMP: 4078 /* icmp only. type always 0 in other cases */ 4079 PF_TEST_ATTRIB((r->type && 4080 r->type != ctx->icmptype + 1), 4081 TAILQ_NEXT(r, entries)); 4082 /* icmp only. type always 0 in other cases */ 4083 PF_TEST_ATTRIB((r->code && 4084 r->code != ctx->icmpcode + 1), 4085 TAILQ_NEXT(r, entries)); 4086 /* icmp only. don't create states on replies */ 4087 PF_TEST_ATTRIB((r->keep_state && !ctx->state_icmp && 4088 (r->rule_flag & PFRULE_STATESLOPPY) == 0 && 4089 ctx->icmp_dir != PF_IN), 4090 TAILQ_NEXT(r, entries)); 4091 break; 4092 4093 case IPPROTO_ICMPV6: 4094 /* icmp only. type always 0 in other cases */ 4095 PF_TEST_ATTRIB((r->type && 4096 r->type != ctx->icmptype + 1), 4097 TAILQ_NEXT(r, entries)); 4098 /* icmp only. type always 0 in other cases */ 4099 PF_TEST_ATTRIB((r->code && 4100 r->code != ctx->icmpcode + 1), 4101 TAILQ_NEXT(r, entries)); 4102 /* icmp only. don't create states on replies */ 4103 PF_TEST_ATTRIB((r->keep_state && !ctx->state_icmp && 4104 (r->rule_flag & PFRULE_STATESLOPPY) == 0 && 4105 ctx->icmp_dir != PF_IN && 4106 ctx->icmptype != ND_NEIGHBOR_ADVERT), 4107 TAILQ_NEXT(r, entries)); 4108 break; 4109 4110 default: 4111 break; 4112 } 4113 4114 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 4115 ctx->pd->virtual_proto != PF_VPROTO_FRAGMENT), 4116 TAILQ_NEXT(r, entries)); 4117 PF_TEST_ATTRIB((r->tos && !(r->tos == ctx->pd->tos)), 4118 TAILQ_NEXT(r, entries)); 4119 PF_TEST_ATTRIB((r->prob && 4120 r->prob <= arc4random_uniform(UINT_MAX - 1) + 1), 4121 TAILQ_NEXT(r, entries)); 4122 PF_TEST_ATTRIB((r->match_tag && 4123 !pf_match_tag(ctx->pd->m, r, &ctx->tag)), 4124 TAILQ_NEXT(r, entries)); 4125 PF_TEST_ATTRIB((r->rcv_kif && pf_match_rcvif(ctx->pd->m, r) == 4126 r->rcvifnot), 4127 TAILQ_NEXT(r, entries)); 4128 PF_TEST_ATTRIB((r->prio && 4129 (r->prio == PF_PRIO_ZERO ? 0 : r->prio) != 4130 ctx->pd->m->m_pkthdr.pf.prio), 4131 TAILQ_NEXT(r, entries)); 4132 4133 /* must be last! */ 4134 if (r->pktrate.limit) { 4135 pf_add_threshold(&r->pktrate); 4136 PF_TEST_ATTRIB((pf_check_threshold(&r->pktrate)), 4137 TAILQ_NEXT(r, entries)); 4138 } 4139 4140 /* FALLTHROUGH */ 4141 if (r->tag) 4142 ctx->tag = r->tag; 4143 if (r->anchor == NULL) { 4144 4145 if (r->rule_flag & PFRULE_ONCE) { 4146 u_int32_t rule_flag; 4147 4148 rule_flag = r->rule_flag; 4149 if (((rule_flag & PFRULE_EXPIRED) == 0) && 4150 atomic_cas_uint(&r->rule_flag, rule_flag, 4151 rule_flag | PFRULE_EXPIRED) == rule_flag) { 4152 r->exptime = gettime(); 4153 } else { 4154 r = TAILQ_NEXT(r, entries); 4155 continue; 4156 } 4157 } 4158 4159 if (r->action == PF_MATCH) { 4160 if ((ctx->ri = pool_get(&pf_rule_item_pl, 4161 PR_NOWAIT)) == NULL) { 4162 REASON_SET(&ctx->reason, PFRES_MEMORY); 4163 return (PF_TEST_FAIL); 4164 } 4165 ctx->ri->r = r; 4166 /* order is irrelevant */ 4167 SLIST_INSERT_HEAD(&ctx->rules, ctx->ri, entry); 4168 ctx->ri = NULL; 4169 pf_rule_to_actions(r, &ctx->act); 4170 if (r->rule_flag & PFRULE_AFTO) 4171 ctx->pd->naf = r->naf; 4172 if (pf_get_transaddr(r, ctx->pd, ctx->sns, 4173 &ctx->nr) == -1) { 4174 REASON_SET(&ctx->reason, 4175 PFRES_TRANSLATE); 4176 return (PF_TEST_FAIL); 4177 } 4178 #if NPFLOG > 0 4179 if (r->log) { 4180 REASON_SET(&ctx->reason, PFRES_MATCH); 4181 pflog_packet(ctx->pd, ctx->reason, r, 4182 ctx->a, ruleset, NULL); 4183 } 4184 #endif /* NPFLOG > 0 */ 4185 } else { 4186 /* 4187 * found matching r 4188 */ 4189 *ctx->rm = r; 4190 /* 4191 * anchor, with ruleset, where r belongs to 4192 */ 4193 *ctx->am = ctx->a; 4194 /* 4195 * ruleset where r belongs to 4196 */ 4197 *ctx->rsm = ruleset; 4198 /* 4199 * ruleset, where anchor belongs to. 4200 */ 4201 ctx->arsm = ctx->aruleset; 4202 } 4203 4204 #if NPFLOG > 0 4205 if (ctx->act.log & PF_LOG_MATCHES) 4206 pf_log_matches(ctx->pd, r, ctx->a, ruleset, 4207 &ctx->rules); 4208 #endif /* NPFLOG > 0 */ 4209 4210 if (r->quick) 4211 return (PF_TEST_QUICK); 4212 } else { 4213 ctx->a = r; 4214 ctx->aruleset = &r->anchor->ruleset; 4215 if (r->anchor_wildcard) { 4216 RB_FOREACH(child, pf_anchor_node, 4217 &r->anchor->children) { 4218 if (pf_anchor_stack_push(ruleset, r, child, 4219 PF_NEXT_CHILD) != 0) 4220 return (PF_TEST_FAIL); 4221 4222 ruleset = &child->ruleset; 4223 goto enter_ruleset; 4224 next_child: 4225 continue; /* with RB_FOREACH() */ 4226 } 4227 } else { 4228 if (pf_anchor_stack_push(ruleset, r, child, 4229 PF_NEXT_RULE) != 0) 4230 return (PF_TEST_FAIL); 4231 4232 ruleset = &r->anchor->ruleset; 4233 child = NULL; 4234 goto enter_ruleset; 4235 next_rule: 4236 ; 4237 } 4238 } 4239 r = TAILQ_NEXT(r, entries); 4240 } 4241 4242 if (pf_anchor_stack_pop(&ruleset, &r, &child, &target) == 0) { 4243 /* stop if any rule matched within quick anchors. */ 4244 if (r->quick == PF_TEST_QUICK && *ctx->am == r) 4245 return (PF_TEST_QUICK); 4246 4247 switch (target) { 4248 case PF_NEXT_CHILD: 4249 goto next_child; 4250 case PF_NEXT_RULE: 4251 goto next_rule; 4252 default: 4253 panic("%s: unknown jump target", __func__); 4254 } 4255 } 4256 4257 return (PF_TEST_OK); 4258 } 4259 4260 int 4261 pf_test_rule(struct pf_pdesc *pd, struct pf_rule **rm, struct pf_state **sm, 4262 struct pf_rule **am, struct pf_ruleset **rsm, u_short *reason, 4263 struct pfsync_deferral **pdeferral) 4264 { 4265 struct pf_rule *r = NULL; 4266 struct pf_rule *a = NULL; 4267 struct pf_ruleset *ruleset = NULL; 4268 struct pf_state_key *skw = NULL, *sks = NULL; 4269 int rewrite = 0; 4270 u_int16_t virtual_type, virtual_id; 4271 int action = PF_DROP; 4272 struct pf_test_ctx ctx; 4273 int rv; 4274 4275 PF_ASSERT_LOCKED(); 4276 4277 memset(&ctx, 0, sizeof(ctx)); 4278 ctx.pd = pd; 4279 ctx.rm = rm; 4280 ctx.am = am; 4281 ctx.rsm = rsm; 4282 ctx.th = &pd->hdr.tcp; 4283 ctx.act.rtableid = pd->rdomain; 4284 ctx.tag = -1; 4285 SLIST_INIT(&ctx.rules); 4286 4287 if (pd->dir == PF_IN && if_congested()) { 4288 REASON_SET(&ctx.reason, PFRES_CONGEST); 4289 return (PF_DROP); 4290 } 4291 4292 switch (pd->virtual_proto) { 4293 case IPPROTO_ICMP: 4294 ctx.icmptype = pd->hdr.icmp.icmp_type; 4295 ctx.icmpcode = pd->hdr.icmp.icmp_code; 4296 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 4297 &ctx.icmp_dir, &virtual_id, &virtual_type); 4298 if (ctx.icmp_dir == PF_IN) { 4299 pd->osport = pd->nsport = virtual_id; 4300 pd->odport = pd->ndport = virtual_type; 4301 } else { 4302 pd->osport = pd->nsport = virtual_type; 4303 pd->odport = pd->ndport = virtual_id; 4304 } 4305 break; 4306 #ifdef INET6 4307 case IPPROTO_ICMPV6: 4308 ctx.icmptype = pd->hdr.icmp6.icmp6_type; 4309 ctx.icmpcode = pd->hdr.icmp6.icmp6_code; 4310 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 4311 &ctx.icmp_dir, &virtual_id, &virtual_type); 4312 if (ctx.icmp_dir == PF_IN) { 4313 pd->osport = pd->nsport = virtual_id; 4314 pd->odport = pd->ndport = virtual_type; 4315 } else { 4316 pd->osport = pd->nsport = virtual_type; 4317 pd->odport = pd->ndport = virtual_id; 4318 } 4319 break; 4320 #endif /* INET6 */ 4321 } 4322 4323 ruleset = &pf_main_ruleset; 4324 rv = pf_match_rule(&ctx, ruleset); 4325 if (rv == PF_TEST_FAIL) { 4326 /* 4327 * Reason has been set in pf_match_rule() already. 4328 */ 4329 goto cleanup; 4330 } 4331 4332 r = *ctx.rm; /* matching rule */ 4333 a = *ctx.am; /* rule that defines an anchor containing 'r' */ 4334 ruleset = *ctx.rsm;/* ruleset of the anchor defined by the rule 'a' */ 4335 ctx.aruleset = ctx.arsm;/* ruleset of the 'a' rule itself */ 4336 4337 /* apply actions for last matching pass/block rule */ 4338 pf_rule_to_actions(r, &ctx.act); 4339 if (r->rule_flag & PFRULE_AFTO) 4340 pd->naf = r->naf; 4341 if (pf_get_transaddr(r, pd, ctx.sns, &ctx.nr) == -1) { 4342 REASON_SET(&ctx.reason, PFRES_TRANSLATE); 4343 goto cleanup; 4344 } 4345 REASON_SET(&ctx.reason, PFRES_MATCH); 4346 4347 #if NPFLOG > 0 4348 if (r->log) 4349 pflog_packet(pd, ctx.reason, r, a, ruleset, NULL); 4350 if (ctx.act.log & PF_LOG_MATCHES) 4351 pf_log_matches(pd, r, a, ruleset, &ctx.rules); 4352 #endif /* NPFLOG > 0 */ 4353 4354 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 4355 (r->action == PF_DROP) && 4356 ((r->rule_flag & PFRULE_RETURNRST) || 4357 (r->rule_flag & PFRULE_RETURNICMP) || 4358 (r->rule_flag & PFRULE_RETURN))) { 4359 if (pd->proto == IPPROTO_TCP && 4360 ((r->rule_flag & PFRULE_RETURNRST) || 4361 (r->rule_flag & PFRULE_RETURN)) && 4362 !(ctx.th->th_flags & TH_RST)) { 4363 u_int32_t ack = 4364 ntohl(ctx.th->th_seq) + pd->p_len; 4365 4366 if (pf_check_tcp_cksum(pd->m, pd->off, 4367 pd->tot_len - pd->off, pd->af)) 4368 REASON_SET(&ctx.reason, PFRES_PROTCKSUM); 4369 else { 4370 if (ctx.th->th_flags & TH_SYN) 4371 ack++; 4372 if (ctx.th->th_flags & TH_FIN) 4373 ack++; 4374 pf_send_tcp(r, pd->af, pd->dst, 4375 pd->src, ctx.th->th_dport, 4376 ctx.th->th_sport, ntohl(ctx.th->th_ack), 4377 ack, TH_RST|TH_ACK, 0, 0, r->return_ttl, 4378 1, 0, pd->rdomain); 4379 } 4380 } else if ((pd->proto != IPPROTO_ICMP || 4381 ICMP_INFOTYPE(ctx.icmptype)) && pd->af == AF_INET && 4382 r->return_icmp) 4383 pf_send_icmp(pd->m, r->return_icmp >> 8, 4384 r->return_icmp & 255, 0, pd->af, r, pd->rdomain); 4385 else if ((pd->proto != IPPROTO_ICMPV6 || 4386 (ctx.icmptype >= ICMP6_ECHO_REQUEST && 4387 ctx.icmptype != ND_REDIRECT)) && pd->af == AF_INET6 && 4388 r->return_icmp6) 4389 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 4390 r->return_icmp6 & 255, 0, pd->af, r, pd->rdomain); 4391 } 4392 4393 if (r->action == PF_DROP) 4394 goto cleanup; 4395 4396 pf_tag_packet(pd->m, ctx.tag, ctx.act.rtableid); 4397 if (ctx.act.rtableid >= 0 && 4398 rtable_l2(ctx.act.rtableid) != pd->rdomain) 4399 pd->destchg = 1; 4400 4401 if (r->action == PF_PASS && pd->badopts != 0 && ! r->allow_opts) { 4402 REASON_SET(&ctx.reason, PFRES_IPOPTIONS); 4403 #if NPFLOG > 0 4404 pd->pflog |= PF_LOG_FORCE; 4405 #endif /* NPFLOG > 0 */ 4406 DPFPRINTF(LOG_NOTICE, "dropping packet with " 4407 "ip/ipv6 options in pf_test_rule()"); 4408 goto cleanup; 4409 } 4410 4411 action = PF_PASS; 4412 4413 if (pd->virtual_proto != PF_VPROTO_FRAGMENT 4414 && !ctx.state_icmp && r->keep_state) { 4415 4416 if (r->rule_flag & PFRULE_SRCTRACK && 4417 pf_insert_src_node(&ctx.sns[PF_SN_NONE], r, PF_SN_NONE, 4418 pd->af, pd->src, NULL, NULL) != 0) { 4419 REASON_SET(&ctx.reason, PFRES_SRCLIMIT); 4420 goto cleanup; 4421 } 4422 4423 if (r->max_states && (r->states_cur >= r->max_states)) { 4424 pf_status.lcounters[LCNT_STATES]++; 4425 REASON_SET(&ctx.reason, PFRES_MAXSTATES); 4426 goto cleanup; 4427 } 4428 4429 action = pf_create_state(pd, r, a, ctx.nr, &skw, &sks, 4430 &rewrite, sm, ctx.tag, &ctx.rules, &ctx.act, ctx.sns); 4431 4432 if (action != PF_PASS) 4433 goto cleanup; 4434 if (sks != skw) { 4435 struct pf_state_key *sk; 4436 4437 if (pd->dir == PF_IN) 4438 sk = sks; 4439 else 4440 sk = skw; 4441 rewrite += pf_translate(pd, 4442 &sk->addr[pd->af == pd->naf ? pd->sidx : pd->didx], 4443 sk->port[pd->af == pd->naf ? pd->sidx : pd->didx], 4444 &sk->addr[pd->af == pd->naf ? pd->didx : pd->sidx], 4445 sk->port[pd->af == pd->naf ? pd->didx : pd->sidx], 4446 virtual_type, ctx.icmp_dir); 4447 } 4448 4449 #ifdef INET6 4450 if (rewrite && skw->af != sks->af) 4451 action = PF_AFRT; 4452 #endif /* INET6 */ 4453 4454 } else { 4455 while ((ctx.ri = SLIST_FIRST(&ctx.rules))) { 4456 SLIST_REMOVE_HEAD(&ctx.rules, entry); 4457 pool_put(&pf_rule_item_pl, ctx.ri); 4458 } 4459 } 4460 4461 /* copy back packet headers if needed */ 4462 if (rewrite && pd->hdrlen) { 4463 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 4464 } 4465 4466 #if NPFSYNC > 0 4467 if (*sm != NULL && !ISSET((*sm)->state_flags, PFSTATE_NOSYNC) && 4468 pd->dir == PF_OUT && pfsync_is_up()) { 4469 /* 4470 * We want the state created, but we dont 4471 * want to send this in case a partner 4472 * firewall has to know about it to allow 4473 * replies through it. 4474 */ 4475 if (pfsync_defer(*sm, pd->m, pdeferral)) 4476 return (PF_DEFER); 4477 } 4478 #endif /* NPFSYNC > 0 */ 4479 4480 return (action); 4481 4482 cleanup: 4483 while ((ctx.ri = SLIST_FIRST(&ctx.rules))) { 4484 SLIST_REMOVE_HEAD(&ctx.rules, entry); 4485 pool_put(&pf_rule_item_pl, ctx.ri); 4486 } 4487 4488 return (action); 4489 } 4490 4491 static __inline int 4492 pf_create_state(struct pf_pdesc *pd, struct pf_rule *r, struct pf_rule *a, 4493 struct pf_rule *nr, struct pf_state_key **skw, struct pf_state_key **sks, 4494 int *rewrite, struct pf_state **sm, int tag, struct pf_rule_slist *rules, 4495 struct pf_rule_actions *act, struct pf_src_node *sns[PF_SN_MAX]) 4496 { 4497 struct pf_state *st = NULL; 4498 struct tcphdr *th = &pd->hdr.tcp; 4499 u_int16_t mss = tcp_mssdflt; 4500 u_short reason; 4501 u_int i; 4502 4503 st = pool_get(&pf_state_pl, PR_NOWAIT | PR_ZERO); 4504 if (st == NULL) { 4505 REASON_SET(&reason, PFRES_MEMORY); 4506 goto csfailed; 4507 } 4508 st->rule.ptr = r; 4509 st->anchor.ptr = a; 4510 st->natrule.ptr = nr; 4511 if (r->allow_opts) 4512 st->state_flags |= PFSTATE_ALLOWOPTS; 4513 if (r->rule_flag & PFRULE_STATESLOPPY) 4514 st->state_flags |= PFSTATE_SLOPPY; 4515 if (r->rule_flag & PFRULE_PFLOW) 4516 st->state_flags |= PFSTATE_PFLOW; 4517 #if NPFLOG > 0 4518 st->log = act->log & PF_LOG_ALL; 4519 #endif /* NPFLOG > 0 */ 4520 st->qid = act->qid; 4521 st->pqid = act->pqid; 4522 st->rtableid[pd->didx] = act->rtableid; 4523 st->rtableid[pd->sidx] = -1; /* return traffic is routed normally */ 4524 st->min_ttl = act->min_ttl; 4525 st->set_tos = act->set_tos; 4526 st->max_mss = act->max_mss; 4527 st->state_flags |= act->flags; 4528 #if NPFSYNC > 0 4529 st->sync_state = PFSYNC_S_NONE; 4530 #endif /* NPFSYNC > 0 */ 4531 st->set_prio[0] = act->set_prio[0]; 4532 st->set_prio[1] = act->set_prio[1]; 4533 st->delay = act->delay; 4534 SLIST_INIT(&st->src_nodes); 4535 /* 4536 * must initialize refcnt, before pf_state_insert() gets called. 4537 * pf_state_inserts() grabs reference for pfsync! 4538 */ 4539 PF_REF_INIT(st->refcnt); 4540 mtx_init(&st->mtx, IPL_NET); 4541 4542 switch (pd->proto) { 4543 case IPPROTO_TCP: 4544 st->src.seqlo = ntohl(th->th_seq); 4545 st->src.seqhi = st->src.seqlo + pd->p_len + 1; 4546 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4547 r->keep_state == PF_STATE_MODULATE) { 4548 /* Generate sequence number modulator */ 4549 st->src.seqdiff = pf_tcp_iss(pd) - st->src.seqlo; 4550 if (st->src.seqdiff == 0) 4551 st->src.seqdiff = 1; 4552 pf_patch_32(pd, &th->th_seq, 4553 htonl(st->src.seqlo + st->src.seqdiff)); 4554 *rewrite = 1; 4555 } else 4556 st->src.seqdiff = 0; 4557 if (th->th_flags & TH_SYN) { 4558 st->src.seqhi++; 4559 st->src.wscale = pf_get_wscale(pd); 4560 } 4561 st->src.max_win = MAX(ntohs(th->th_win), 1); 4562 if (st->src.wscale & PF_WSCALE_MASK) { 4563 /* Remove scale factor from initial window */ 4564 int win = st->src.max_win; 4565 win += 1 << (st->src.wscale & PF_WSCALE_MASK); 4566 st->src.max_win = (win - 1) >> 4567 (st->src.wscale & PF_WSCALE_MASK); 4568 } 4569 if (th->th_flags & TH_FIN) 4570 st->src.seqhi++; 4571 st->dst.seqhi = 1; 4572 st->dst.max_win = 1; 4573 pf_set_protostate(st, PF_PEER_SRC, TCPS_SYN_SENT); 4574 pf_set_protostate(st, PF_PEER_DST, TCPS_CLOSED); 4575 st->timeout = PFTM_TCP_FIRST_PACKET; 4576 pf_status.states_halfopen++; 4577 break; 4578 case IPPROTO_UDP: 4579 pf_set_protostate(st, PF_PEER_SRC, PFUDPS_SINGLE); 4580 pf_set_protostate(st, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 4581 st->timeout = PFTM_UDP_FIRST_PACKET; 4582 break; 4583 case IPPROTO_ICMP: 4584 #ifdef INET6 4585 case IPPROTO_ICMPV6: 4586 #endif /* INET6 */ 4587 st->timeout = PFTM_ICMP_FIRST_PACKET; 4588 break; 4589 default: 4590 pf_set_protostate(st, PF_PEER_SRC, PFOTHERS_SINGLE); 4591 pf_set_protostate(st, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 4592 st->timeout = PFTM_OTHER_FIRST_PACKET; 4593 } 4594 4595 st->creation = getuptime(); 4596 st->expire = getuptime(); 4597 4598 if (pd->proto == IPPROTO_TCP) { 4599 if (st->state_flags & PFSTATE_SCRUB_TCP && 4600 pf_normalize_tcp_init(pd, &st->src)) { 4601 REASON_SET(&reason, PFRES_MEMORY); 4602 goto csfailed; 4603 } 4604 if (st->state_flags & PFSTATE_SCRUB_TCP && st->src.scrub && 4605 pf_normalize_tcp_stateful(pd, &reason, st, 4606 &st->src, &st->dst, rewrite)) { 4607 /* This really shouldn't happen!!! */ 4608 DPFPRINTF(LOG_ERR, 4609 "%s: tcp normalize failed on first pkt", __func__); 4610 goto csfailed; 4611 } 4612 } 4613 st->direction = pd->dir; 4614 4615 if (pf_state_key_setup(pd, skw, sks, act->rtableid)) { 4616 REASON_SET(&reason, PFRES_MEMORY); 4617 goto csfailed; 4618 } 4619 4620 if (pf_set_rt_ifp(st, pd->src, (*skw)->af, sns) != 0) { 4621 REASON_SET(&reason, PFRES_NOROUTE); 4622 goto csfailed; 4623 } 4624 4625 for (i = 0; i < PF_SN_MAX; i++) 4626 if (sns[i] != NULL) { 4627 struct pf_sn_item *sni; 4628 4629 sni = pool_get(&pf_sn_item_pl, PR_NOWAIT); 4630 if (sni == NULL) { 4631 REASON_SET(&reason, PFRES_MEMORY); 4632 goto csfailed; 4633 } 4634 sni->sn = sns[i]; 4635 SLIST_INSERT_HEAD(&st->src_nodes, sni, next); 4636 sni->sn->states++; 4637 } 4638 4639 if (pf_state_insert(BOUND_IFACE(r, pd->kif), skw, sks, st)) { 4640 *sks = *skw = NULL; 4641 REASON_SET(&reason, PFRES_STATEINS); 4642 goto csfailed; 4643 } else 4644 *sm = st; 4645 4646 /* 4647 * Make state responsible for rules it binds here. 4648 */ 4649 memcpy(&st->match_rules, rules, sizeof(st->match_rules)); 4650 memset(rules, 0, sizeof(*rules)); 4651 STATE_INC_COUNTERS(st); 4652 4653 if (tag > 0) { 4654 pf_tag_ref(tag); 4655 st->tag = tag; 4656 } 4657 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 4658 TH_SYN && r->keep_state == PF_STATE_SYNPROXY && pd->dir == PF_IN) { 4659 int rtid = pd->rdomain; 4660 if (act->rtableid >= 0) 4661 rtid = act->rtableid; 4662 pf_set_protostate(st, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 4663 st->src.seqhi = arc4random(); 4664 /* Find mss option */ 4665 mss = pf_get_mss(pd); 4666 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 4667 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 4668 st->src.mss = mss; 4669 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 4670 th->th_sport, st->src.seqhi, ntohl(th->th_seq) + 1, 4671 TH_SYN|TH_ACK, 0, st->src.mss, 0, 1, 0, pd->rdomain); 4672 REASON_SET(&reason, PFRES_SYNPROXY); 4673 return (PF_SYNPROXY_DROP); 4674 } 4675 4676 return (PF_PASS); 4677 4678 csfailed: 4679 if (st) { 4680 pf_normalize_tcp_cleanup(st); /* safe even w/o init */ 4681 pf_src_tree_remove_state(st); 4682 pool_put(&pf_state_pl, st); 4683 } 4684 4685 for (i = 0; i < PF_SN_MAX; i++) 4686 if (sns[i] != NULL) 4687 pf_remove_src_node(sns[i]); 4688 4689 return (PF_DROP); 4690 } 4691 4692 int 4693 pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport, 4694 struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type, 4695 int icmp_dir) 4696 { 4697 int rewrite = 0; 4698 int afto = pd->af != pd->naf; 4699 4700 if (afto || PF_ANEQ(daddr, pd->dst, pd->af)) 4701 pd->destchg = 1; 4702 4703 switch (pd->proto) { 4704 case IPPROTO_TCP: /* FALLTHROUGH */ 4705 case IPPROTO_UDP: 4706 rewrite += pf_patch_16(pd, pd->sport, sport); 4707 rewrite += pf_patch_16(pd, pd->dport, dport); 4708 break; 4709 4710 case IPPROTO_ICMP: 4711 if (pd->af != AF_INET) 4712 return (0); 4713 4714 #ifdef INET6 4715 if (afto) { 4716 if (pf_translate_icmp_af(pd, AF_INET6, &pd->hdr.icmp)) 4717 return (0); 4718 pd->proto = IPPROTO_ICMPV6; 4719 rewrite = 1; 4720 } 4721 #endif /* INET6 */ 4722 if (virtual_type == htons(ICMP_ECHO)) { 4723 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 4724 rewrite += pf_patch_16(pd, 4725 &pd->hdr.icmp.icmp_id, icmpid); 4726 } 4727 break; 4728 4729 #ifdef INET6 4730 case IPPROTO_ICMPV6: 4731 if (pd->af != AF_INET6) 4732 return (0); 4733 4734 if (afto) { 4735 if (pf_translate_icmp_af(pd, AF_INET, &pd->hdr.icmp6)) 4736 return (0); 4737 pd->proto = IPPROTO_ICMP; 4738 rewrite = 1; 4739 } 4740 if (virtual_type == htons(ICMP6_ECHO_REQUEST)) { 4741 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 4742 rewrite += pf_patch_16(pd, 4743 &pd->hdr.icmp6.icmp6_id, icmpid); 4744 } 4745 break; 4746 #endif /* INET6 */ 4747 } 4748 4749 if (!afto) { 4750 rewrite += pf_translate_a(pd, pd->src, saddr); 4751 rewrite += pf_translate_a(pd, pd->dst, daddr); 4752 } 4753 4754 return (rewrite); 4755 } 4756 4757 int 4758 pf_tcp_track_full(struct pf_pdesc *pd, struct pf_state **stp, u_short *reason, 4759 int *copyback, int reverse) 4760 { 4761 struct tcphdr *th = &pd->hdr.tcp; 4762 struct pf_state_peer *src, *dst; 4763 u_int16_t win = ntohs(th->th_win); 4764 u_int32_t ack, end, data_end, seq, orig_seq; 4765 u_int8_t sws, dws, psrc, pdst; 4766 int ackskew; 4767 4768 if ((pd->dir == (*stp)->direction && !reverse) || 4769 (pd->dir != (*stp)->direction && reverse)) { 4770 src = &(*stp)->src; 4771 dst = &(*stp)->dst; 4772 psrc = PF_PEER_SRC; 4773 pdst = PF_PEER_DST; 4774 } else { 4775 src = &(*stp)->dst; 4776 dst = &(*stp)->src; 4777 psrc = PF_PEER_DST; 4778 pdst = PF_PEER_SRC; 4779 } 4780 4781 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 4782 sws = src->wscale & PF_WSCALE_MASK; 4783 dws = dst->wscale & PF_WSCALE_MASK; 4784 } else 4785 sws = dws = 0; 4786 4787 /* 4788 * Sequence tracking algorithm from Guido van Rooij's paper: 4789 * http://www.madison-gurkha.com/publications/tcp_filtering/ 4790 * tcp_filtering.ps 4791 */ 4792 4793 orig_seq = seq = ntohl(th->th_seq); 4794 if (src->seqlo == 0) { 4795 /* First packet from this end. Set its state */ 4796 4797 if (((*stp)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 4798 src->scrub == NULL) { 4799 if (pf_normalize_tcp_init(pd, src)) { 4800 REASON_SET(reason, PFRES_MEMORY); 4801 return (PF_DROP); 4802 } 4803 } 4804 4805 /* Deferred generation of sequence number modulator */ 4806 if (dst->seqdiff && !src->seqdiff) { 4807 /* use random iss for the TCP server */ 4808 while ((src->seqdiff = arc4random() - seq) == 0) 4809 continue; 4810 ack = ntohl(th->th_ack) - dst->seqdiff; 4811 pf_patch_32(pd, &th->th_seq, htonl(seq + src->seqdiff)); 4812 pf_patch_32(pd, &th->th_ack, htonl(ack)); 4813 *copyback = 1; 4814 } else { 4815 ack = ntohl(th->th_ack); 4816 } 4817 4818 end = seq + pd->p_len; 4819 if (th->th_flags & TH_SYN) { 4820 end++; 4821 if (dst->wscale & PF_WSCALE_FLAG) { 4822 src->wscale = pf_get_wscale(pd); 4823 if (src->wscale & PF_WSCALE_FLAG) { 4824 /* Remove scale factor from initial 4825 * window */ 4826 sws = src->wscale & PF_WSCALE_MASK; 4827 win = ((u_int32_t)win + (1 << sws) - 1) 4828 >> sws; 4829 dws = dst->wscale & PF_WSCALE_MASK; 4830 } else { 4831 /* fixup other window */ 4832 dst->max_win = MIN(TCP_MAXWIN, 4833 (u_int32_t)dst->max_win << 4834 (dst->wscale & PF_WSCALE_MASK)); 4835 /* in case of a retrans SYN|ACK */ 4836 dst->wscale = 0; 4837 } 4838 } 4839 } 4840 data_end = end; 4841 if (th->th_flags & TH_FIN) 4842 end++; 4843 4844 src->seqlo = seq; 4845 if (src->state < TCPS_SYN_SENT) 4846 pf_set_protostate(*stp, psrc, TCPS_SYN_SENT); 4847 4848 /* 4849 * May need to slide the window (seqhi may have been set by 4850 * the crappy stack check or if we picked up the connection 4851 * after establishment) 4852 */ 4853 if (src->seqhi == 1 || 4854 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 4855 src->seqhi = end + MAX(1, dst->max_win << dws); 4856 if (win > src->max_win) 4857 src->max_win = win; 4858 4859 } else { 4860 ack = ntohl(th->th_ack) - dst->seqdiff; 4861 if (src->seqdiff) { 4862 /* Modulate sequence numbers */ 4863 pf_patch_32(pd, &th->th_seq, htonl(seq + src->seqdiff)); 4864 pf_patch_32(pd, &th->th_ack, htonl(ack)); 4865 *copyback = 1; 4866 } 4867 end = seq + pd->p_len; 4868 if (th->th_flags & TH_SYN) 4869 end++; 4870 data_end = end; 4871 if (th->th_flags & TH_FIN) 4872 end++; 4873 } 4874 4875 if ((th->th_flags & TH_ACK) == 0) { 4876 /* Let it pass through the ack skew check */ 4877 ack = dst->seqlo; 4878 } else if ((ack == 0 && 4879 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 4880 /* broken tcp stacks do not set ack */ 4881 (dst->state < TCPS_SYN_SENT)) { 4882 /* 4883 * Many stacks (ours included) will set the ACK number in an 4884 * FIN|ACK if the SYN times out -- no sequence to ACK. 4885 */ 4886 ack = dst->seqlo; 4887 } 4888 4889 if (seq == end) { 4890 /* Ease sequencing restrictions on no data packets */ 4891 seq = src->seqlo; 4892 data_end = end = seq; 4893 } 4894 4895 ackskew = dst->seqlo - ack; 4896 4897 4898 /* 4899 * Need to demodulate the sequence numbers in any TCP SACK options 4900 * (Selective ACK). We could optionally validate the SACK values 4901 * against the current ACK window, either forwards or backwards, but 4902 * I'm not confident that SACK has been implemented properly 4903 * everywhere. It wouldn't surprise me if several stacks accidently 4904 * SACK too far backwards of previously ACKed data. There really aren't 4905 * any security implications of bad SACKing unless the target stack 4906 * doesn't validate the option length correctly. Someone trying to 4907 * spoof into a TCP connection won't bother blindly sending SACK 4908 * options anyway. 4909 */ 4910 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 4911 if (pf_modulate_sack(pd, dst)) 4912 *copyback = 1; 4913 } 4914 4915 4916 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 4917 if (SEQ_GEQ(src->seqhi, data_end) && 4918 /* Last octet inside other's window space */ 4919 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 4920 /* Retrans: not more than one window back */ 4921 (ackskew >= -MAXACKWINDOW) && 4922 /* Acking not more than one reassembled fragment backwards */ 4923 (ackskew <= (MAXACKWINDOW << sws)) && 4924 /* Acking not more than one window forward */ 4925 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 4926 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 4927 /* Require an exact/+1 sequence match on resets when possible */ 4928 4929 if (dst->scrub || src->scrub) { 4930 if (pf_normalize_tcp_stateful(pd, reason, *stp, src, 4931 dst, copyback)) 4932 return (PF_DROP); 4933 } 4934 4935 /* update max window */ 4936 if (src->max_win < win) 4937 src->max_win = win; 4938 /* synchronize sequencing */ 4939 if (SEQ_GT(end, src->seqlo)) 4940 src->seqlo = end; 4941 /* slide the window of what the other end can send */ 4942 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4943 dst->seqhi = ack + MAX((win << sws), 1); 4944 4945 /* update states */ 4946 if (th->th_flags & TH_SYN) 4947 if (src->state < TCPS_SYN_SENT) 4948 pf_set_protostate(*stp, psrc, TCPS_SYN_SENT); 4949 if (th->th_flags & TH_FIN) 4950 if (src->state < TCPS_CLOSING) 4951 pf_set_protostate(*stp, psrc, TCPS_CLOSING); 4952 if (th->th_flags & TH_ACK) { 4953 if (dst->state == TCPS_SYN_SENT) { 4954 pf_set_protostate(*stp, pdst, 4955 TCPS_ESTABLISHED); 4956 if (src->state == TCPS_ESTABLISHED && 4957 !SLIST_EMPTY(&(*stp)->src_nodes) && 4958 pf_src_connlimit(stp)) { 4959 REASON_SET(reason, PFRES_SRCLIMIT); 4960 return (PF_DROP); 4961 } 4962 } else if (dst->state == TCPS_CLOSING) 4963 pf_set_protostate(*stp, pdst, 4964 TCPS_FIN_WAIT_2); 4965 } 4966 if (th->th_flags & TH_RST) 4967 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_TIME_WAIT); 4968 4969 /* update expire time */ 4970 (*stp)->expire = getuptime(); 4971 if (src->state >= TCPS_FIN_WAIT_2 && 4972 dst->state >= TCPS_FIN_WAIT_2) 4973 (*stp)->timeout = PFTM_TCP_CLOSED; 4974 else if (src->state >= TCPS_CLOSING && 4975 dst->state >= TCPS_CLOSING) 4976 (*stp)->timeout = PFTM_TCP_FIN_WAIT; 4977 else if (src->state < TCPS_ESTABLISHED || 4978 dst->state < TCPS_ESTABLISHED) 4979 (*stp)->timeout = PFTM_TCP_OPENING; 4980 else if (src->state >= TCPS_CLOSING || 4981 dst->state >= TCPS_CLOSING) 4982 (*stp)->timeout = PFTM_TCP_CLOSING; 4983 else 4984 (*stp)->timeout = PFTM_TCP_ESTABLISHED; 4985 4986 /* Fall through to PASS packet */ 4987 } else if ((dst->state < TCPS_SYN_SENT || 4988 dst->state >= TCPS_FIN_WAIT_2 || 4989 src->state >= TCPS_FIN_WAIT_2) && 4990 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 4991 /* Within a window forward of the originating packet */ 4992 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 4993 /* Within a window backward of the originating packet */ 4994 4995 /* 4996 * This currently handles three situations: 4997 * 1) Stupid stacks will shotgun SYNs before their peer 4998 * replies. 4999 * 2) When PF catches an already established stream (the 5000 * firewall rebooted, the state table was flushed, routes 5001 * changed...) 5002 * 3) Packets get funky immediately after the connection 5003 * closes (this should catch Solaris spurious ACK|FINs 5004 * that web servers like to spew after a close) 5005 * 5006 * This must be a little more careful than the above code 5007 * since packet floods will also be caught here. We don't 5008 * update the TTL here to mitigate the damage of a packet 5009 * flood and so the same code can handle awkward establishment 5010 * and a loosened connection close. 5011 * In the establishment case, a correct peer response will 5012 * validate the connection, go through the normal state code 5013 * and keep updating the state TTL. 5014 */ 5015 5016 if (pf_status.debug >= LOG_NOTICE) { 5017 log(LOG_NOTICE, "pf: loose state match: "); 5018 pf_print_state(*stp); 5019 pf_print_flags(th->th_flags); 5020 addlog(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5021 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5022 pd->p_len, ackskew, (*stp)->packets[0], 5023 (*stp)->packets[1], 5024 pd->dir == PF_IN ? "in" : "out", 5025 pd->dir == (*stp)->direction ? "fwd" : "rev"); 5026 } 5027 5028 if (dst->scrub || src->scrub) { 5029 if (pf_normalize_tcp_stateful(pd, reason, *stp, src, 5030 dst, copyback)) 5031 return (PF_DROP); 5032 } 5033 5034 /* update max window */ 5035 if (src->max_win < win) 5036 src->max_win = win; 5037 /* synchronize sequencing */ 5038 if (SEQ_GT(end, src->seqlo)) 5039 src->seqlo = end; 5040 /* slide the window of what the other end can send */ 5041 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5042 dst->seqhi = ack + MAX((win << sws), 1); 5043 5044 /* 5045 * Cannot set dst->seqhi here since this could be a shotgunned 5046 * SYN and not an already established connection. 5047 */ 5048 if (th->th_flags & TH_FIN) 5049 if (src->state < TCPS_CLOSING) 5050 pf_set_protostate(*stp, psrc, TCPS_CLOSING); 5051 if (th->th_flags & TH_RST) 5052 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_TIME_WAIT); 5053 5054 /* Fall through to PASS packet */ 5055 } else { 5056 if ((*stp)->dst.state == TCPS_SYN_SENT && 5057 (*stp)->src.state == TCPS_SYN_SENT) { 5058 /* Send RST for state mismatches during handshake */ 5059 if (!(th->th_flags & TH_RST)) 5060 pf_send_tcp((*stp)->rule.ptr, pd->af, 5061 pd->dst, pd->src, th->th_dport, 5062 th->th_sport, ntohl(th->th_ack), 0, 5063 TH_RST, 0, 0, 5064 (*stp)->rule.ptr->return_ttl, 1, 0, 5065 pd->rdomain); 5066 src->seqlo = 0; 5067 src->seqhi = 1; 5068 src->max_win = 1; 5069 } else if (pf_status.debug >= LOG_NOTICE) { 5070 log(LOG_NOTICE, "pf: BAD state: "); 5071 pf_print_state(*stp); 5072 pf_print_flags(th->th_flags); 5073 addlog(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5074 "pkts=%llu:%llu dir=%s,%s\n", 5075 seq, orig_seq, ack, pd->p_len, ackskew, 5076 (*stp)->packets[0], (*stp)->packets[1], 5077 pd->dir == PF_IN ? "in" : "out", 5078 pd->dir == (*stp)->direction ? "fwd" : "rev"); 5079 addlog("pf: State failure on: %c %c %c %c | %c %c\n", 5080 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 5081 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5082 ' ': '2', 5083 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5084 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5085 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ? 5086 ' ' :'5', 5087 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5088 } 5089 REASON_SET(reason, PFRES_BADSTATE); 5090 return (PF_DROP); 5091 } 5092 5093 return (PF_PASS); 5094 } 5095 5096 int 5097 pf_tcp_track_sloppy(struct pf_pdesc *pd, struct pf_state **stp, 5098 u_short *reason) 5099 { 5100 struct tcphdr *th = &pd->hdr.tcp; 5101 struct pf_state_peer *src, *dst; 5102 u_int8_t psrc, pdst; 5103 5104 if (pd->dir == (*stp)->direction) { 5105 src = &(*stp)->src; 5106 dst = &(*stp)->dst; 5107 psrc = PF_PEER_SRC; 5108 pdst = PF_PEER_DST; 5109 } else { 5110 src = &(*stp)->dst; 5111 dst = &(*stp)->src; 5112 psrc = PF_PEER_DST; 5113 pdst = PF_PEER_SRC; 5114 } 5115 5116 if (th->th_flags & TH_SYN) 5117 if (src->state < TCPS_SYN_SENT) 5118 pf_set_protostate(*stp, psrc, TCPS_SYN_SENT); 5119 if (th->th_flags & TH_FIN) 5120 if (src->state < TCPS_CLOSING) 5121 pf_set_protostate(*stp, psrc, TCPS_CLOSING); 5122 if (th->th_flags & TH_ACK) { 5123 if (dst->state == TCPS_SYN_SENT) { 5124 pf_set_protostate(*stp, pdst, TCPS_ESTABLISHED); 5125 if (src->state == TCPS_ESTABLISHED && 5126 !SLIST_EMPTY(&(*stp)->src_nodes) && 5127 pf_src_connlimit(stp)) { 5128 REASON_SET(reason, PFRES_SRCLIMIT); 5129 return (PF_DROP); 5130 } 5131 } else if (dst->state == TCPS_CLOSING) { 5132 pf_set_protostate(*stp, pdst, TCPS_FIN_WAIT_2); 5133 } else if (src->state == TCPS_SYN_SENT && 5134 dst->state < TCPS_SYN_SENT) { 5135 /* 5136 * Handle a special sloppy case where we only see one 5137 * half of the connection. If there is a ACK after 5138 * the initial SYN without ever seeing a packet from 5139 * the destination, set the connection to established. 5140 */ 5141 pf_set_protostate(*stp, PF_PEER_BOTH, 5142 TCPS_ESTABLISHED); 5143 if (!SLIST_EMPTY(&(*stp)->src_nodes) && 5144 pf_src_connlimit(stp)) { 5145 REASON_SET(reason, PFRES_SRCLIMIT); 5146 return (PF_DROP); 5147 } 5148 } else if (src->state == TCPS_CLOSING && 5149 dst->state == TCPS_ESTABLISHED && 5150 dst->seqlo == 0) { 5151 /* 5152 * Handle the closing of half connections where we 5153 * don't see the full bidirectional FIN/ACK+ACK 5154 * handshake. 5155 */ 5156 pf_set_protostate(*stp, pdst, TCPS_CLOSING); 5157 } 5158 } 5159 if (th->th_flags & TH_RST) 5160 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_TIME_WAIT); 5161 5162 /* update expire time */ 5163 (*stp)->expire = getuptime(); 5164 if (src->state >= TCPS_FIN_WAIT_2 && 5165 dst->state >= TCPS_FIN_WAIT_2) 5166 (*stp)->timeout = PFTM_TCP_CLOSED; 5167 else if (src->state >= TCPS_CLOSING && 5168 dst->state >= TCPS_CLOSING) 5169 (*stp)->timeout = PFTM_TCP_FIN_WAIT; 5170 else if (src->state < TCPS_ESTABLISHED || 5171 dst->state < TCPS_ESTABLISHED) 5172 (*stp)->timeout = PFTM_TCP_OPENING; 5173 else if (src->state >= TCPS_CLOSING || 5174 dst->state >= TCPS_CLOSING) 5175 (*stp)->timeout = PFTM_TCP_CLOSING; 5176 else 5177 (*stp)->timeout = PFTM_TCP_ESTABLISHED; 5178 5179 return (PF_PASS); 5180 } 5181 5182 static __inline int 5183 pf_synproxy(struct pf_pdesc *pd, struct pf_state **stp, u_short *reason) 5184 { 5185 struct pf_state_key *sk = (*stp)->key[pd->didx]; 5186 5187 if ((*stp)->src.state == PF_TCPS_PROXY_SRC) { 5188 struct tcphdr *th = &pd->hdr.tcp; 5189 5190 if (pd->dir != (*stp)->direction) { 5191 REASON_SET(reason, PFRES_SYNPROXY); 5192 return (PF_SYNPROXY_DROP); 5193 } 5194 if (th->th_flags & TH_SYN) { 5195 if (ntohl(th->th_seq) != (*stp)->src.seqlo) { 5196 REASON_SET(reason, PFRES_SYNPROXY); 5197 return (PF_DROP); 5198 } 5199 pf_send_tcp((*stp)->rule.ptr, pd->af, pd->dst, 5200 pd->src, th->th_dport, th->th_sport, 5201 (*stp)->src.seqhi, ntohl(th->th_seq) + 1, 5202 TH_SYN|TH_ACK, 0, (*stp)->src.mss, 0, 1, 5203 0, pd->rdomain); 5204 REASON_SET(reason, PFRES_SYNPROXY); 5205 return (PF_SYNPROXY_DROP); 5206 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 5207 (ntohl(th->th_ack) != (*stp)->src.seqhi + 1) || 5208 (ntohl(th->th_seq) != (*stp)->src.seqlo + 1)) { 5209 REASON_SET(reason, PFRES_SYNPROXY); 5210 return (PF_DROP); 5211 } else if (!SLIST_EMPTY(&(*stp)->src_nodes) && 5212 pf_src_connlimit(stp)) { 5213 REASON_SET(reason, PFRES_SRCLIMIT); 5214 return (PF_DROP); 5215 } else 5216 pf_set_protostate(*stp, PF_PEER_SRC, 5217 PF_TCPS_PROXY_DST); 5218 } 5219 if ((*stp)->src.state == PF_TCPS_PROXY_DST) { 5220 struct tcphdr *th = &pd->hdr.tcp; 5221 5222 if (pd->dir == (*stp)->direction) { 5223 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 5224 (ntohl(th->th_ack) != (*stp)->src.seqhi + 1) || 5225 (ntohl(th->th_seq) != (*stp)->src.seqlo + 1)) { 5226 REASON_SET(reason, PFRES_SYNPROXY); 5227 return (PF_DROP); 5228 } 5229 (*stp)->src.max_win = MAX(ntohs(th->th_win), 1); 5230 if ((*stp)->dst.seqhi == 1) 5231 (*stp)->dst.seqhi = arc4random(); 5232 pf_send_tcp((*stp)->rule.ptr, pd->af, 5233 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5234 sk->port[pd->sidx], sk->port[pd->didx], 5235 (*stp)->dst.seqhi, 0, TH_SYN, 0, 5236 (*stp)->src.mss, 0, 0, (*stp)->tag, 5237 sk->rdomain); 5238 REASON_SET(reason, PFRES_SYNPROXY); 5239 return (PF_SYNPROXY_DROP); 5240 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 5241 (TH_SYN|TH_ACK)) || 5242 (ntohl(th->th_ack) != (*stp)->dst.seqhi + 1)) { 5243 REASON_SET(reason, PFRES_SYNPROXY); 5244 return (PF_DROP); 5245 } else { 5246 (*stp)->dst.max_win = MAX(ntohs(th->th_win), 1); 5247 (*stp)->dst.seqlo = ntohl(th->th_seq); 5248 pf_send_tcp((*stp)->rule.ptr, pd->af, pd->dst, 5249 pd->src, th->th_dport, th->th_sport, 5250 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 5251 TH_ACK, (*stp)->src.max_win, 0, 0, 0, 5252 (*stp)->tag, pd->rdomain); 5253 pf_send_tcp((*stp)->rule.ptr, pd->af, 5254 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5255 sk->port[pd->sidx], sk->port[pd->didx], 5256 (*stp)->src.seqhi + 1, (*stp)->src.seqlo + 1, 5257 TH_ACK, (*stp)->dst.max_win, 0, 0, 1, 5258 0, sk->rdomain); 5259 (*stp)->src.seqdiff = (*stp)->dst.seqhi - 5260 (*stp)->src.seqlo; 5261 (*stp)->dst.seqdiff = (*stp)->src.seqhi - 5262 (*stp)->dst.seqlo; 5263 (*stp)->src.seqhi = (*stp)->src.seqlo + 5264 (*stp)->dst.max_win; 5265 (*stp)->dst.seqhi = (*stp)->dst.seqlo + 5266 (*stp)->src.max_win; 5267 (*stp)->src.wscale = (*stp)->dst.wscale = 0; 5268 pf_set_protostate(*stp, PF_PEER_BOTH, 5269 TCPS_ESTABLISHED); 5270 REASON_SET(reason, PFRES_SYNPROXY); 5271 return (PF_SYNPROXY_DROP); 5272 } 5273 } 5274 return (PF_PASS); 5275 } 5276 5277 int 5278 pf_test_state(struct pf_pdesc *pd, struct pf_state **stp, u_short *reason) 5279 { 5280 int copyback = 0; 5281 struct pf_state_peer *src, *dst; 5282 int action; 5283 struct inpcb *inp = pd->m->m_pkthdr.pf.inp; 5284 u_int8_t psrc, pdst; 5285 5286 action = PF_PASS; 5287 if (pd->dir == (*stp)->direction) { 5288 src = &(*stp)->src; 5289 dst = &(*stp)->dst; 5290 psrc = PF_PEER_SRC; 5291 pdst = PF_PEER_DST; 5292 } else { 5293 src = &(*stp)->dst; 5294 dst = &(*stp)->src; 5295 psrc = PF_PEER_DST; 5296 pdst = PF_PEER_SRC; 5297 } 5298 5299 switch (pd->virtual_proto) { 5300 case IPPROTO_TCP: 5301 if ((action = pf_synproxy(pd, stp, reason)) != PF_PASS) 5302 return (action); 5303 if ((pd->hdr.tcp.th_flags & (TH_SYN|TH_ACK)) == TH_SYN) { 5304 5305 if (dst->state >= TCPS_FIN_WAIT_2 && 5306 src->state >= TCPS_FIN_WAIT_2) { 5307 if (pf_status.debug >= LOG_NOTICE) { 5308 log(LOG_NOTICE, "pf: state reuse "); 5309 pf_print_state(*stp); 5310 pf_print_flags(pd->hdr.tcp.th_flags); 5311 addlog("\n"); 5312 } 5313 /* XXX make sure it's the same direction ?? */ 5314 (*stp)->timeout = PFTM_PURGE; 5315 pf_state_unref(*stp); 5316 *stp = NULL; 5317 pf_mbuf_link_inpcb(pd->m, inp); 5318 return (PF_DROP); 5319 } else if (dst->state >= TCPS_ESTABLISHED && 5320 src->state >= TCPS_ESTABLISHED) { 5321 /* 5322 * SYN matches existing state??? 5323 * Typically happens when sender boots up after 5324 * sudden panic. Certain protocols (NFSv3) are 5325 * always using same port numbers. Challenge 5326 * ACK enables all parties (firewall and peers) 5327 * to get in sync again. 5328 */ 5329 pf_send_challenge_ack(pd, *stp, src, dst); 5330 return (PF_DROP); 5331 } 5332 } 5333 5334 if ((*stp)->state_flags & PFSTATE_SLOPPY) { 5335 if (pf_tcp_track_sloppy(pd, stp, reason) == PF_DROP) 5336 return (PF_DROP); 5337 } else { 5338 if (pf_tcp_track_full(pd, stp, reason, ©back, 5339 PF_REVERSED_KEY((*stp)->key, pd->af)) == PF_DROP) 5340 return (PF_DROP); 5341 } 5342 break; 5343 case IPPROTO_UDP: 5344 /* update states */ 5345 if (src->state < PFUDPS_SINGLE) 5346 pf_set_protostate(*stp, psrc, PFUDPS_SINGLE); 5347 if (dst->state == PFUDPS_SINGLE) 5348 pf_set_protostate(*stp, pdst, PFUDPS_MULTIPLE); 5349 5350 /* update expire time */ 5351 (*stp)->expire = getuptime(); 5352 if (src->state == PFUDPS_MULTIPLE && 5353 dst->state == PFUDPS_MULTIPLE) 5354 (*stp)->timeout = PFTM_UDP_MULTIPLE; 5355 else 5356 (*stp)->timeout = PFTM_UDP_SINGLE; 5357 break; 5358 default: 5359 /* update states */ 5360 if (src->state < PFOTHERS_SINGLE) 5361 pf_set_protostate(*stp, psrc, PFOTHERS_SINGLE); 5362 if (dst->state == PFOTHERS_SINGLE) 5363 pf_set_protostate(*stp, pdst, PFOTHERS_MULTIPLE); 5364 5365 /* update expire time */ 5366 (*stp)->expire = getuptime(); 5367 if (src->state == PFOTHERS_MULTIPLE && 5368 dst->state == PFOTHERS_MULTIPLE) 5369 (*stp)->timeout = PFTM_OTHER_MULTIPLE; 5370 else 5371 (*stp)->timeout = PFTM_OTHER_SINGLE; 5372 break; 5373 } 5374 5375 /* translate source/destination address, if necessary */ 5376 if ((*stp)->key[PF_SK_WIRE] != (*stp)->key[PF_SK_STACK]) { 5377 struct pf_state_key *nk; 5378 int afto, sidx, didx; 5379 5380 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5381 nk = (*stp)->key[pd->sidx]; 5382 else 5383 nk = (*stp)->key[pd->didx]; 5384 5385 afto = pd->af != nk->af; 5386 sidx = afto ? pd->didx : pd->sidx; 5387 didx = afto ? pd->sidx : pd->didx; 5388 5389 #ifdef INET6 5390 if (afto) { 5391 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], nk->af); 5392 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], nk->af); 5393 pd->naf = nk->af; 5394 action = PF_AFRT; 5395 } 5396 #endif /* INET6 */ 5397 5398 if (!afto) 5399 pf_translate_a(pd, pd->src, &nk->addr[sidx]); 5400 5401 if (pd->sport != NULL) 5402 pf_patch_16(pd, pd->sport, nk->port[sidx]); 5403 5404 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 5405 pd->rdomain != nk->rdomain) 5406 pd->destchg = 1; 5407 5408 if (!afto) 5409 pf_translate_a(pd, pd->dst, &nk->addr[didx]); 5410 5411 if (pd->dport != NULL) 5412 pf_patch_16(pd, pd->dport, nk->port[didx]); 5413 5414 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5415 copyback = 1; 5416 } 5417 5418 if (copyback && pd->hdrlen > 0) { 5419 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 5420 } 5421 5422 return (action); 5423 } 5424 5425 int 5426 pf_icmp_state_lookup(struct pf_pdesc *pd, struct pf_state_key_cmp *key, 5427 struct pf_state **stp, u_int16_t icmpid, u_int16_t type, 5428 int icmp_dir, int *iidx, int multi, int inner) 5429 { 5430 int direction, action; 5431 5432 key->af = pd->af; 5433 key->proto = pd->proto; 5434 key->rdomain = pd->rdomain; 5435 if (icmp_dir == PF_IN) { 5436 *iidx = pd->sidx; 5437 key->port[pd->sidx] = icmpid; 5438 key->port[pd->didx] = type; 5439 } else { 5440 *iidx = pd->didx; 5441 key->port[pd->sidx] = type; 5442 key->port[pd->didx] = icmpid; 5443 } 5444 5445 if (pf_state_key_addr_setup(pd, key, pd->sidx, pd->src, pd->didx, 5446 pd->dst, pd->af, multi)) 5447 return (PF_DROP); 5448 5449 key->hash = pf_pkt_hash(key->af, key->proto, 5450 &key->addr[0], &key->addr[1], 0, 0); 5451 5452 action = pf_find_state(pd, key, stp); 5453 if (action != PF_MATCH) 5454 return (action); 5455 5456 if ((*stp)->state_flags & PFSTATE_SLOPPY) 5457 return (-1); 5458 5459 /* Is this ICMP message flowing in right direction? */ 5460 if ((*stp)->key[PF_SK_WIRE]->af != (*stp)->key[PF_SK_STACK]->af) 5461 direction = (pd->af == (*stp)->key[PF_SK_WIRE]->af) ? 5462 PF_IN : PF_OUT; 5463 else 5464 direction = (*stp)->direction; 5465 if ((((!inner && direction == pd->dir) || 5466 (inner && direction != pd->dir)) ? 5467 PF_IN : PF_OUT) != icmp_dir) { 5468 if (pf_status.debug >= LOG_NOTICE) { 5469 log(LOG_NOTICE, 5470 "pf: icmp type %d in wrong direction (%d): ", 5471 ntohs(type), icmp_dir); 5472 pf_print_state(*stp); 5473 addlog("\n"); 5474 } 5475 return (PF_DROP); 5476 } 5477 return (-1); 5478 } 5479 5480 int 5481 pf_test_state_icmp(struct pf_pdesc *pd, struct pf_state **stp, 5482 u_short *reason) 5483 { 5484 u_int16_t virtual_id, virtual_type; 5485 u_int8_t icmptype, icmpcode; 5486 int icmp_dir, iidx, ret, copyback = 0; 5487 5488 struct pf_state_key_cmp key; 5489 5490 switch (pd->proto) { 5491 case IPPROTO_ICMP: 5492 icmptype = pd->hdr.icmp.icmp_type; 5493 icmpcode = pd->hdr.icmp.icmp_code; 5494 break; 5495 #ifdef INET6 5496 case IPPROTO_ICMPV6: 5497 icmptype = pd->hdr.icmp6.icmp6_type; 5498 icmpcode = pd->hdr.icmp6.icmp6_code; 5499 break; 5500 #endif /* INET6 */ 5501 default: 5502 panic("unhandled proto %d", pd->proto); 5503 } 5504 5505 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &virtual_id, 5506 &virtual_type) == 0) { 5507 /* 5508 * ICMP query/reply message not related to a TCP/UDP packet. 5509 * Search for an ICMP state. 5510 */ 5511 ret = pf_icmp_state_lookup(pd, &key, stp, 5512 virtual_id, virtual_type, icmp_dir, &iidx, 5513 0, 0); 5514 /* IPv6? try matching a multicast address */ 5515 if (ret == PF_DROP && pd->af == AF_INET6 && icmp_dir == PF_OUT) 5516 ret = pf_icmp_state_lookup(pd, &key, stp, virtual_id, 5517 virtual_type, icmp_dir, &iidx, 1, 0); 5518 if (ret >= 0) 5519 return (ret); 5520 5521 (*stp)->expire = getuptime(); 5522 (*stp)->timeout = PFTM_ICMP_ERROR_REPLY; 5523 5524 /* translate source/destination address, if necessary */ 5525 if ((*stp)->key[PF_SK_WIRE] != (*stp)->key[PF_SK_STACK]) { 5526 struct pf_state_key *nk; 5527 int afto, sidx, didx; 5528 5529 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5530 nk = (*stp)->key[pd->sidx]; 5531 else 5532 nk = (*stp)->key[pd->didx]; 5533 5534 afto = pd->af != nk->af; 5535 sidx = afto ? pd->didx : pd->sidx; 5536 didx = afto ? pd->sidx : pd->didx; 5537 iidx = afto ? !iidx : iidx; 5538 #ifdef INET6 5539 if (afto) { 5540 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], 5541 nk->af); 5542 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], 5543 nk->af); 5544 pd->naf = nk->af; 5545 } 5546 #endif /* INET6 */ 5547 if (!afto) { 5548 pf_translate_a(pd, pd->src, &nk->addr[sidx]); 5549 pf_translate_a(pd, pd->dst, &nk->addr[didx]); 5550 } 5551 5552 if (pd->rdomain != nk->rdomain) 5553 pd->destchg = 1; 5554 if (!afto && PF_ANEQ(pd->dst, 5555 &nk->addr[didx], pd->af)) 5556 pd->destchg = 1; 5557 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5558 5559 switch (pd->af) { 5560 case AF_INET: 5561 #ifdef INET6 5562 if (afto) { 5563 if (pf_translate_icmp_af(pd, AF_INET6, 5564 &pd->hdr.icmp)) 5565 return (PF_DROP); 5566 pd->proto = IPPROTO_ICMPV6; 5567 } 5568 #endif /* INET6 */ 5569 pf_patch_16(pd, 5570 &pd->hdr.icmp.icmp_id, nk->port[iidx]); 5571 5572 m_copyback(pd->m, pd->off, ICMP_MINLEN, 5573 &pd->hdr.icmp, M_NOWAIT); 5574 copyback = 1; 5575 break; 5576 #ifdef INET6 5577 case AF_INET6: 5578 if (afto) { 5579 if (pf_translate_icmp_af(pd, AF_INET, 5580 &pd->hdr.icmp6)) 5581 return (PF_DROP); 5582 pd->proto = IPPROTO_ICMP; 5583 } 5584 5585 pf_patch_16(pd, 5586 &pd->hdr.icmp6.icmp6_id, nk->port[iidx]); 5587 5588 m_copyback(pd->m, pd->off, 5589 sizeof(struct icmp6_hdr), &pd->hdr.icmp6, 5590 M_NOWAIT); 5591 copyback = 1; 5592 break; 5593 #endif /* INET6 */ 5594 } 5595 #ifdef INET6 5596 if (afto) 5597 return (PF_AFRT); 5598 #endif /* INET6 */ 5599 } 5600 } else { 5601 /* 5602 * ICMP error message in response to a TCP/UDP packet. 5603 * Extract the inner TCP/UDP header and search for that state. 5604 */ 5605 struct pf_pdesc pd2; 5606 struct ip h2; 5607 #ifdef INET6 5608 struct ip6_hdr h2_6; 5609 #endif /* INET6 */ 5610 int ipoff2; 5611 5612 /* Initialize pd2 fields valid for both packets with pd. */ 5613 memset(&pd2, 0, sizeof(pd2)); 5614 pd2.af = pd->af; 5615 pd2.dir = pd->dir; 5616 pd2.kif = pd->kif; 5617 pd2.m = pd->m; 5618 pd2.rdomain = pd->rdomain; 5619 /* Payload packet is from the opposite direction. */ 5620 pd2.sidx = (pd2.dir == PF_IN) ? 1 : 0; 5621 pd2.didx = (pd2.dir == PF_IN) ? 0 : 1; 5622 switch (pd->af) { 5623 case AF_INET: 5624 /* offset of h2 in mbuf chain */ 5625 ipoff2 = pd->off + ICMP_MINLEN; 5626 5627 if (!pf_pull_hdr(pd2.m, ipoff2, &h2, sizeof(h2), 5628 NULL, reason, pd2.af)) { 5629 DPFPRINTF(LOG_NOTICE, 5630 "ICMP error message too short (ip)"); 5631 return (PF_DROP); 5632 } 5633 /* 5634 * ICMP error messages don't refer to non-first 5635 * fragments 5636 */ 5637 if (h2.ip_off & htons(IP_OFFMASK)) { 5638 REASON_SET(reason, PFRES_FRAG); 5639 return (PF_DROP); 5640 } 5641 5642 /* offset of protocol header that follows h2 */ 5643 pd2.off = ipoff2; 5644 if (pf_walk_header(&pd2, &h2, reason) != PF_PASS) 5645 return (PF_DROP); 5646 5647 pd2.tot_len = ntohs(h2.ip_len); 5648 pd2.src = (struct pf_addr *)&h2.ip_src; 5649 pd2.dst = (struct pf_addr *)&h2.ip_dst; 5650 break; 5651 #ifdef INET6 5652 case AF_INET6: 5653 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 5654 5655 if (!pf_pull_hdr(pd2.m, ipoff2, &h2_6, sizeof(h2_6), 5656 NULL, reason, pd2.af)) { 5657 DPFPRINTF(LOG_NOTICE, 5658 "ICMP error message too short (ip6)"); 5659 return (PF_DROP); 5660 } 5661 5662 pd2.off = ipoff2; 5663 if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS) 5664 return (PF_DROP); 5665 5666 pd2.tot_len = ntohs(h2_6.ip6_plen) + 5667 sizeof(struct ip6_hdr); 5668 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 5669 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 5670 break; 5671 #endif /* INET6 */ 5672 default: 5673 unhandled_af(pd->af); 5674 } 5675 5676 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 5677 if (pf_status.debug >= LOG_NOTICE) { 5678 log(LOG_NOTICE, 5679 "pf: BAD ICMP %d:%d outer dst: ", 5680 icmptype, icmpcode); 5681 pf_print_host(pd->src, 0, pd->af); 5682 addlog(" -> "); 5683 pf_print_host(pd->dst, 0, pd->af); 5684 addlog(" inner src: "); 5685 pf_print_host(pd2.src, 0, pd2.af); 5686 addlog(" -> "); 5687 pf_print_host(pd2.dst, 0, pd2.af); 5688 addlog("\n"); 5689 } 5690 REASON_SET(reason, PFRES_BADSTATE); 5691 return (PF_DROP); 5692 } 5693 5694 switch (pd2.proto) { 5695 case IPPROTO_TCP: { 5696 struct tcphdr *th = &pd2.hdr.tcp; 5697 u_int32_t seq; 5698 struct pf_state_peer *src, *dst; 5699 u_int8_t dws; 5700 int action; 5701 5702 /* 5703 * Only the first 8 bytes of the TCP header can be 5704 * expected. Don't access any TCP header fields after 5705 * th_seq, an ackskew test is not possible. 5706 */ 5707 if (!pf_pull_hdr(pd2.m, pd2.off, th, 8, NULL, reason, 5708 pd2.af)) { 5709 DPFPRINTF(LOG_NOTICE, 5710 "ICMP error message too short (tcp)"); 5711 return (PF_DROP); 5712 } 5713 5714 key.af = pd2.af; 5715 key.proto = IPPROTO_TCP; 5716 key.rdomain = pd2.rdomain; 5717 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 5718 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 5719 key.port[pd2.sidx] = th->th_sport; 5720 key.port[pd2.didx] = th->th_dport; 5721 key.hash = pf_pkt_hash(pd2.af, pd2.proto, 5722 pd2.src, pd2.dst, th->th_sport, th->th_dport); 5723 5724 action = pf_find_state(&pd2, &key, stp); 5725 if (action != PF_MATCH) 5726 return (action); 5727 5728 if (pd2.dir == (*stp)->direction) { 5729 if (PF_REVERSED_KEY((*stp)->key, pd->af)) { 5730 src = &(*stp)->src; 5731 dst = &(*stp)->dst; 5732 } else { 5733 src = &(*stp)->dst; 5734 dst = &(*stp)->src; 5735 } 5736 } else { 5737 if (PF_REVERSED_KEY((*stp)->key, pd->af)) { 5738 src = &(*stp)->dst; 5739 dst = &(*stp)->src; 5740 } else { 5741 src = &(*stp)->src; 5742 dst = &(*stp)->dst; 5743 } 5744 } 5745 5746 if (src->wscale && dst->wscale) 5747 dws = dst->wscale & PF_WSCALE_MASK; 5748 else 5749 dws = 0; 5750 5751 /* Demodulate sequence number */ 5752 seq = ntohl(th->th_seq) - src->seqdiff; 5753 if (src->seqdiff) { 5754 pf_patch_32(pd, &th->th_seq, htonl(seq)); 5755 copyback = 1; 5756 } 5757 5758 if (!((*stp)->state_flags & PFSTATE_SLOPPY) && 5759 (!SEQ_GEQ(src->seqhi, seq) || !SEQ_GEQ(seq, 5760 src->seqlo - (dst->max_win << dws)))) { 5761 if (pf_status.debug >= LOG_NOTICE) { 5762 log(LOG_NOTICE, 5763 "pf: BAD ICMP %d:%d ", 5764 icmptype, icmpcode); 5765 pf_print_host(pd->src, 0, pd->af); 5766 addlog(" -> "); 5767 pf_print_host(pd->dst, 0, pd->af); 5768 addlog(" state: "); 5769 pf_print_state(*stp); 5770 addlog(" seq=%u\n", seq); 5771 } 5772 REASON_SET(reason, PFRES_BADSTATE); 5773 return (PF_DROP); 5774 } else { 5775 if (pf_status.debug >= LOG_DEBUG) { 5776 log(LOG_DEBUG, 5777 "pf: OK ICMP %d:%d ", 5778 icmptype, icmpcode); 5779 pf_print_host(pd->src, 0, pd->af); 5780 addlog(" -> "); 5781 pf_print_host(pd->dst, 0, pd->af); 5782 addlog(" state: "); 5783 pf_print_state(*stp); 5784 addlog(" seq=%u\n", seq); 5785 } 5786 } 5787 5788 /* translate source/destination address, if necessary */ 5789 if ((*stp)->key[PF_SK_WIRE] != 5790 (*stp)->key[PF_SK_STACK]) { 5791 struct pf_state_key *nk; 5792 int afto, sidx, didx; 5793 5794 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5795 nk = (*stp)->key[pd->sidx]; 5796 else 5797 nk = (*stp)->key[pd->didx]; 5798 5799 afto = pd->af != nk->af; 5800 sidx = afto ? pd2.didx : pd2.sidx; 5801 didx = afto ? pd2.sidx : pd2.didx; 5802 5803 #ifdef INET6 5804 if (afto) { 5805 if (pf_translate_icmp_af(pd, nk->af, 5806 &pd->hdr.icmp)) 5807 return (PF_DROP); 5808 m_copyback(pd->m, pd->off, 5809 sizeof(struct icmp6_hdr), 5810 &pd->hdr.icmp6, M_NOWAIT); 5811 if (pf_change_icmp_af(pd->m, ipoff2, 5812 pd, &pd2, &nk->addr[sidx], 5813 &nk->addr[didx], pd->af, nk->af)) 5814 return (PF_DROP); 5815 if (nk->af == AF_INET) 5816 pd->proto = IPPROTO_ICMP; 5817 else 5818 pd->proto = IPPROTO_ICMPV6; 5819 pd->m->m_pkthdr.ph_rtableid = 5820 nk->rdomain; 5821 pd->destchg = 1; 5822 pf_addrcpy(&pd->nsaddr, 5823 &nk->addr[pd2.sidx], nk->af); 5824 pf_addrcpy(&pd->ndaddr, 5825 &nk->addr[pd2.didx], nk->af); 5826 pd->naf = nk->af; 5827 5828 pf_patch_16(pd, 5829 &th->th_sport, nk->port[sidx]); 5830 pf_patch_16(pd, 5831 &th->th_dport, nk->port[didx]); 5832 5833 m_copyback(pd2.m, pd2.off, 8, th, 5834 M_NOWAIT); 5835 return (PF_AFRT); 5836 } 5837 #endif /* INET6 */ 5838 if (PF_ANEQ(pd2.src, 5839 &nk->addr[pd2.sidx], pd2.af) || 5840 nk->port[pd2.sidx] != th->th_sport) 5841 pf_translate_icmp(pd, pd2.src, 5842 &th->th_sport, pd->dst, 5843 &nk->addr[pd2.sidx], 5844 nk->port[pd2.sidx]); 5845 5846 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 5847 pd2.af) || pd2.rdomain != nk->rdomain) 5848 pd->destchg = 1; 5849 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5850 5851 if (PF_ANEQ(pd2.dst, 5852 &nk->addr[pd2.didx], pd2.af) || 5853 nk->port[pd2.didx] != th->th_dport) 5854 pf_translate_icmp(pd, pd2.dst, 5855 &th->th_dport, pd->src, 5856 &nk->addr[pd2.didx], 5857 nk->port[pd2.didx]); 5858 copyback = 1; 5859 } 5860 5861 if (copyback) { 5862 switch (pd2.af) { 5863 case AF_INET: 5864 m_copyback(pd->m, pd->off, ICMP_MINLEN, 5865 &pd->hdr.icmp, M_NOWAIT); 5866 m_copyback(pd2.m, ipoff2, sizeof(h2), 5867 &h2, M_NOWAIT); 5868 break; 5869 #ifdef INET6 5870 case AF_INET6: 5871 m_copyback(pd->m, pd->off, 5872 sizeof(struct icmp6_hdr), 5873 &pd->hdr.icmp6, M_NOWAIT); 5874 m_copyback(pd2.m, ipoff2, sizeof(h2_6), 5875 &h2_6, M_NOWAIT); 5876 break; 5877 #endif /* INET6 */ 5878 } 5879 m_copyback(pd2.m, pd2.off, 8, th, M_NOWAIT); 5880 } 5881 break; 5882 } 5883 case IPPROTO_UDP: { 5884 struct udphdr *uh = &pd2.hdr.udp; 5885 int action; 5886 5887 if (!pf_pull_hdr(pd2.m, pd2.off, uh, sizeof(*uh), 5888 NULL, reason, pd2.af)) { 5889 DPFPRINTF(LOG_NOTICE, 5890 "ICMP error message too short (udp)"); 5891 return (PF_DROP); 5892 } 5893 5894 key.af = pd2.af; 5895 key.proto = IPPROTO_UDP; 5896 key.rdomain = pd2.rdomain; 5897 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 5898 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 5899 key.port[pd2.sidx] = uh->uh_sport; 5900 key.port[pd2.didx] = uh->uh_dport; 5901 key.hash = pf_pkt_hash(pd2.af, pd2.proto, 5902 pd2.src, pd2.dst, uh->uh_sport, uh->uh_dport); 5903 5904 action = pf_find_state(&pd2, &key, stp); 5905 if (action != PF_MATCH) 5906 return (action); 5907 5908 /* translate source/destination address, if necessary */ 5909 if ((*stp)->key[PF_SK_WIRE] != 5910 (*stp)->key[PF_SK_STACK]) { 5911 struct pf_state_key *nk; 5912 int afto, sidx, didx; 5913 5914 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5915 nk = (*stp)->key[pd->sidx]; 5916 else 5917 nk = (*stp)->key[pd->didx]; 5918 5919 afto = pd->af != nk->af; 5920 sidx = afto ? pd2.didx : pd2.sidx; 5921 didx = afto ? pd2.sidx : pd2.didx; 5922 5923 #ifdef INET6 5924 if (afto) { 5925 if (pf_translate_icmp_af(pd, nk->af, 5926 &pd->hdr.icmp)) 5927 return (PF_DROP); 5928 m_copyback(pd->m, pd->off, 5929 sizeof(struct icmp6_hdr), 5930 &pd->hdr.icmp6, M_NOWAIT); 5931 if (pf_change_icmp_af(pd->m, ipoff2, 5932 pd, &pd2, &nk->addr[sidx], 5933 &nk->addr[didx], pd->af, nk->af)) 5934 return (PF_DROP); 5935 if (nk->af == AF_INET) 5936 pd->proto = IPPROTO_ICMP; 5937 else 5938 pd->proto = IPPROTO_ICMPV6; 5939 pd->m->m_pkthdr.ph_rtableid = 5940 nk->rdomain; 5941 pd->destchg = 1; 5942 pf_addrcpy(&pd->nsaddr, 5943 &nk->addr[pd2.sidx], nk->af); 5944 pf_addrcpy(&pd->ndaddr, 5945 &nk->addr[pd2.didx], nk->af); 5946 pd->naf = nk->af; 5947 5948 pf_patch_16(pd, 5949 &uh->uh_sport, nk->port[sidx]); 5950 pf_patch_16(pd, 5951 &uh->uh_dport, nk->port[didx]); 5952 5953 m_copyback(pd2.m, pd2.off, sizeof(*uh), 5954 uh, M_NOWAIT); 5955 return (PF_AFRT); 5956 } 5957 #endif /* INET6 */ 5958 5959 if (PF_ANEQ(pd2.src, 5960 &nk->addr[pd2.sidx], pd2.af) || 5961 nk->port[pd2.sidx] != uh->uh_sport) 5962 pf_translate_icmp(pd, pd2.src, 5963 &uh->uh_sport, pd->dst, 5964 &nk->addr[pd2.sidx], 5965 nk->port[pd2.sidx]); 5966 5967 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 5968 pd2.af) || pd2.rdomain != nk->rdomain) 5969 pd->destchg = 1; 5970 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5971 5972 if (PF_ANEQ(pd2.dst, 5973 &nk->addr[pd2.didx], pd2.af) || 5974 nk->port[pd2.didx] != uh->uh_dport) 5975 pf_translate_icmp(pd, pd2.dst, 5976 &uh->uh_dport, pd->src, 5977 &nk->addr[pd2.didx], 5978 nk->port[pd2.didx]); 5979 5980 switch (pd2.af) { 5981 case AF_INET: 5982 m_copyback(pd->m, pd->off, ICMP_MINLEN, 5983 &pd->hdr.icmp, M_NOWAIT); 5984 m_copyback(pd2.m, ipoff2, sizeof(h2), 5985 &h2, M_NOWAIT); 5986 break; 5987 #ifdef INET6 5988 case AF_INET6: 5989 m_copyback(pd->m, pd->off, 5990 sizeof(struct icmp6_hdr), 5991 &pd->hdr.icmp6, M_NOWAIT); 5992 m_copyback(pd2.m, ipoff2, sizeof(h2_6), 5993 &h2_6, M_NOWAIT); 5994 break; 5995 #endif /* INET6 */ 5996 } 5997 /* Avoid recomputing quoted UDP checksum. 5998 * note: udp6 0 csum invalid per rfc2460 p27. 5999 * but presumed nothing cares in this context */ 6000 pf_patch_16(pd, &uh->uh_sum, 0); 6001 m_copyback(pd2.m, pd2.off, sizeof(*uh), uh, 6002 M_NOWAIT); 6003 copyback = 1; 6004 } 6005 break; 6006 } 6007 case IPPROTO_ICMP: { 6008 struct icmp *iih = &pd2.hdr.icmp; 6009 6010 if (pd2.af != AF_INET) { 6011 REASON_SET(reason, PFRES_NORM); 6012 return (PF_DROP); 6013 } 6014 6015 if (!pf_pull_hdr(pd2.m, pd2.off, iih, ICMP_MINLEN, 6016 NULL, reason, pd2.af)) { 6017 DPFPRINTF(LOG_NOTICE, 6018 "ICMP error message too short (icmp)"); 6019 return (PF_DROP); 6020 } 6021 6022 pf_icmp_mapping(&pd2, iih->icmp_type, 6023 &icmp_dir, &virtual_id, &virtual_type); 6024 6025 ret = pf_icmp_state_lookup(&pd2, &key, stp, 6026 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 6027 if (ret >= 0) 6028 return (ret); 6029 6030 /* translate source/destination address, if necessary */ 6031 if ((*stp)->key[PF_SK_WIRE] != 6032 (*stp)->key[PF_SK_STACK]) { 6033 struct pf_state_key *nk; 6034 int afto, sidx, didx; 6035 6036 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 6037 nk = (*stp)->key[pd->sidx]; 6038 else 6039 nk = (*stp)->key[pd->didx]; 6040 6041 afto = pd->af != nk->af; 6042 sidx = afto ? pd2.didx : pd2.sidx; 6043 didx = afto ? pd2.sidx : pd2.didx; 6044 iidx = afto ? !iidx : iidx; 6045 6046 #ifdef INET6 6047 if (afto) { 6048 if (nk->af != AF_INET6) 6049 return (PF_DROP); 6050 if (pf_translate_icmp_af(pd, nk->af, 6051 &pd->hdr.icmp)) 6052 return (PF_DROP); 6053 m_copyback(pd->m, pd->off, 6054 sizeof(struct icmp6_hdr), 6055 &pd->hdr.icmp6, M_NOWAIT); 6056 if (pf_change_icmp_af(pd->m, ipoff2, 6057 pd, &pd2, &nk->addr[sidx], 6058 &nk->addr[didx], pd->af, nk->af)) 6059 return (PF_DROP); 6060 pd->proto = IPPROTO_ICMPV6; 6061 if (pf_translate_icmp_af(pd, 6062 nk->af, iih)) 6063 return (PF_DROP); 6064 if (virtual_type == htons(ICMP_ECHO)) 6065 pf_patch_16(pd, &iih->icmp_id, 6066 nk->port[iidx]); 6067 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, 6068 iih, M_NOWAIT); 6069 pd->m->m_pkthdr.ph_rtableid = 6070 nk->rdomain; 6071 pd->destchg = 1; 6072 pf_addrcpy(&pd->nsaddr, 6073 &nk->addr[pd2.sidx], nk->af); 6074 pf_addrcpy(&pd->ndaddr, 6075 &nk->addr[pd2.didx], nk->af); 6076 pd->naf = nk->af; 6077 return (PF_AFRT); 6078 } 6079 #endif /* INET6 */ 6080 6081 if (PF_ANEQ(pd2.src, 6082 &nk->addr[pd2.sidx], pd2.af) || 6083 (virtual_type == htons(ICMP_ECHO) && 6084 nk->port[iidx] != iih->icmp_id)) 6085 pf_translate_icmp(pd, pd2.src, 6086 (virtual_type == htons(ICMP_ECHO)) ? 6087 &iih->icmp_id : NULL, 6088 pd->dst, &nk->addr[pd2.sidx], 6089 (virtual_type == htons(ICMP_ECHO)) ? 6090 nk->port[iidx] : 0); 6091 6092 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 6093 pd2.af) || pd2.rdomain != nk->rdomain) 6094 pd->destchg = 1; 6095 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 6096 6097 if (PF_ANEQ(pd2.dst, 6098 &nk->addr[pd2.didx], pd2.af)) 6099 pf_translate_icmp(pd, pd2.dst, NULL, 6100 pd->src, &nk->addr[pd2.didx], 0); 6101 6102 m_copyback(pd->m, pd->off, ICMP_MINLEN, 6103 &pd->hdr.icmp, M_NOWAIT); 6104 m_copyback(pd2.m, ipoff2, sizeof(h2), &h2, 6105 M_NOWAIT); 6106 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, iih, 6107 M_NOWAIT); 6108 copyback = 1; 6109 } 6110 break; 6111 } 6112 #ifdef INET6 6113 case IPPROTO_ICMPV6: { 6114 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 6115 6116 if (pd2.af != AF_INET6) { 6117 REASON_SET(reason, PFRES_NORM); 6118 return (PF_DROP); 6119 } 6120 6121 if (!pf_pull_hdr(pd2.m, pd2.off, iih, 6122 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 6123 DPFPRINTF(LOG_NOTICE, 6124 "ICMP error message too short (icmp6)"); 6125 return (PF_DROP); 6126 } 6127 6128 pf_icmp_mapping(&pd2, iih->icmp6_type, 6129 &icmp_dir, &virtual_id, &virtual_type); 6130 ret = pf_icmp_state_lookup(&pd2, &key, stp, 6131 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 6132 /* IPv6? try matching a multicast address */ 6133 if (ret == PF_DROP && pd2.af == AF_INET6 && 6134 icmp_dir == PF_OUT) 6135 ret = pf_icmp_state_lookup(&pd2, &key, stp, 6136 virtual_id, virtual_type, icmp_dir, &iidx, 6137 1, 1); 6138 if (ret >= 0) 6139 return (ret); 6140 6141 /* translate source/destination address, if necessary */ 6142 if ((*stp)->key[PF_SK_WIRE] != 6143 (*stp)->key[PF_SK_STACK]) { 6144 struct pf_state_key *nk; 6145 int afto, sidx, didx; 6146 6147 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 6148 nk = (*stp)->key[pd->sidx]; 6149 else 6150 nk = (*stp)->key[pd->didx]; 6151 6152 afto = pd->af != nk->af; 6153 sidx = afto ? pd2.didx : pd2.sidx; 6154 didx = afto ? pd2.sidx : pd2.didx; 6155 iidx = afto ? !iidx : iidx; 6156 6157 if (afto) { 6158 if (nk->af != AF_INET) 6159 return (PF_DROP); 6160 if (pf_translate_icmp_af(pd, nk->af, 6161 &pd->hdr.icmp)) 6162 return (PF_DROP); 6163 m_copyback(pd->m, pd->off, 6164 sizeof(struct icmp6_hdr), 6165 &pd->hdr.icmp6, M_NOWAIT); 6166 if (pf_change_icmp_af(pd->m, ipoff2, 6167 pd, &pd2, &nk->addr[sidx], 6168 &nk->addr[didx], pd->af, nk->af)) 6169 return (PF_DROP); 6170 pd->proto = IPPROTO_ICMP; 6171 if (pf_translate_icmp_af(pd, 6172 nk->af, iih)) 6173 return (PF_DROP); 6174 if (virtual_type == 6175 htons(ICMP6_ECHO_REQUEST)) 6176 pf_patch_16(pd, &iih->icmp6_id, 6177 nk->port[iidx]); 6178 m_copyback(pd2.m, pd2.off, 6179 sizeof(struct icmp6_hdr), iih, 6180 M_NOWAIT); 6181 pd->m->m_pkthdr.ph_rtableid = 6182 nk->rdomain; 6183 pd->destchg = 1; 6184 pf_addrcpy(&pd->nsaddr, 6185 &nk->addr[pd2.sidx], nk->af); 6186 pf_addrcpy(&pd->ndaddr, 6187 &nk->addr[pd2.didx], nk->af); 6188 pd->naf = nk->af; 6189 return (PF_AFRT); 6190 } 6191 6192 if (PF_ANEQ(pd2.src, 6193 &nk->addr[pd2.sidx], pd2.af) || 6194 ((virtual_type == 6195 htons(ICMP6_ECHO_REQUEST)) && 6196 nk->port[pd2.sidx] != iih->icmp6_id)) 6197 pf_translate_icmp(pd, pd2.src, 6198 (virtual_type == 6199 htons(ICMP6_ECHO_REQUEST)) 6200 ? &iih->icmp6_id : NULL, 6201 pd->dst, &nk->addr[pd2.sidx], 6202 (virtual_type == 6203 htons(ICMP6_ECHO_REQUEST)) 6204 ? nk->port[iidx] : 0); 6205 6206 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 6207 pd2.af) || pd2.rdomain != nk->rdomain) 6208 pd->destchg = 1; 6209 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 6210 6211 if (PF_ANEQ(pd2.dst, 6212 &nk->addr[pd2.didx], pd2.af)) 6213 pf_translate_icmp(pd, pd2.dst, NULL, 6214 pd->src, &nk->addr[pd2.didx], 0); 6215 6216 m_copyback(pd->m, pd->off, 6217 sizeof(struct icmp6_hdr), &pd->hdr.icmp6, 6218 M_NOWAIT); 6219 m_copyback(pd2.m, ipoff2, sizeof(h2_6), &h2_6, 6220 M_NOWAIT); 6221 m_copyback(pd2.m, pd2.off, 6222 sizeof(struct icmp6_hdr), iih, M_NOWAIT); 6223 copyback = 1; 6224 } 6225 break; 6226 } 6227 #endif /* INET6 */ 6228 default: { 6229 int action; 6230 6231 key.af = pd2.af; 6232 key.proto = pd2.proto; 6233 key.rdomain = pd2.rdomain; 6234 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 6235 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 6236 key.port[0] = key.port[1] = 0; 6237 key.hash = pf_pkt_hash(pd2.af, pd2.proto, 6238 pd2.src, pd2.dst, 0, 0); 6239 6240 action = pf_find_state(&pd2, &key, stp); 6241 if (action != PF_MATCH) 6242 return (action); 6243 6244 /* translate source/destination address, if necessary */ 6245 if ((*stp)->key[PF_SK_WIRE] != 6246 (*stp)->key[PF_SK_STACK]) { 6247 struct pf_state_key *nk = 6248 (*stp)->key[pd->didx]; 6249 6250 if (PF_ANEQ(pd2.src, 6251 &nk->addr[pd2.sidx], pd2.af)) 6252 pf_translate_icmp(pd, pd2.src, NULL, 6253 pd->dst, &nk->addr[pd2.sidx], 0); 6254 6255 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 6256 pd2.af) || pd2.rdomain != nk->rdomain) 6257 pd->destchg = 1; 6258 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 6259 6260 if (PF_ANEQ(pd2.dst, 6261 &nk->addr[pd2.didx], pd2.af)) 6262 pf_translate_icmp(pd, pd2.dst, NULL, 6263 pd->src, &nk->addr[pd2.didx], 0); 6264 6265 switch (pd2.af) { 6266 case AF_INET: 6267 m_copyback(pd->m, pd->off, ICMP_MINLEN, 6268 &pd->hdr.icmp, M_NOWAIT); 6269 m_copyback(pd2.m, ipoff2, sizeof(h2), 6270 &h2, M_NOWAIT); 6271 break; 6272 #ifdef INET6 6273 case AF_INET6: 6274 m_copyback(pd->m, pd->off, 6275 sizeof(struct icmp6_hdr), 6276 &pd->hdr.icmp6, M_NOWAIT); 6277 m_copyback(pd2.m, ipoff2, sizeof(h2_6), 6278 &h2_6, M_NOWAIT); 6279 break; 6280 #endif /* INET6 */ 6281 } 6282 copyback = 1; 6283 } 6284 break; 6285 } 6286 } 6287 } 6288 if (copyback) { 6289 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 6290 } 6291 6292 return (PF_PASS); 6293 } 6294 6295 /* 6296 * ipoff and off are measured from the start of the mbuf chain. 6297 * h must be at "ipoff" on the mbuf chain. 6298 */ 6299 void * 6300 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 6301 u_short *actionp, u_short *reasonp, sa_family_t af) 6302 { 6303 int iplen = 0; 6304 6305 switch (af) { 6306 case AF_INET: { 6307 struct ip *h = mtod(m, struct ip *); 6308 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 6309 6310 if (fragoff) { 6311 if (fragoff >= len) 6312 ACTION_SET(actionp, PF_PASS); 6313 else { 6314 ACTION_SET(actionp, PF_DROP); 6315 REASON_SET(reasonp, PFRES_FRAG); 6316 } 6317 return (NULL); 6318 } 6319 iplen = ntohs(h->ip_len); 6320 break; 6321 } 6322 #ifdef INET6 6323 case AF_INET6: { 6324 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 6325 6326 iplen = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 6327 break; 6328 } 6329 #endif /* INET6 */ 6330 } 6331 if (m->m_pkthdr.len < off + len || iplen < off + len) { 6332 ACTION_SET(actionp, PF_DROP); 6333 REASON_SET(reasonp, PFRES_SHORT); 6334 return (NULL); 6335 } 6336 m_copydata(m, off, len, p); 6337 return (p); 6338 } 6339 6340 int 6341 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif, 6342 int rtableid) 6343 { 6344 struct sockaddr_storage ss; 6345 struct sockaddr_in *dst; 6346 int ret = 1; 6347 int check_mpath; 6348 #ifdef INET6 6349 struct sockaddr_in6 *dst6; 6350 #endif /* INET6 */ 6351 struct rtentry *rt = NULL; 6352 6353 check_mpath = 0; 6354 memset(&ss, 0, sizeof(ss)); 6355 switch (af) { 6356 case AF_INET: 6357 dst = (struct sockaddr_in *)&ss; 6358 dst->sin_family = AF_INET; 6359 dst->sin_len = sizeof(*dst); 6360 dst->sin_addr = addr->v4; 6361 if (ipmultipath) 6362 check_mpath = 1; 6363 break; 6364 #ifdef INET6 6365 case AF_INET6: 6366 /* 6367 * Skip check for addresses with embedded interface scope, 6368 * as they would always match anyway. 6369 */ 6370 if (IN6_IS_SCOPE_EMBED(&addr->v6)) 6371 goto out; 6372 dst6 = (struct sockaddr_in6 *)&ss; 6373 dst6->sin6_family = AF_INET6; 6374 dst6->sin6_len = sizeof(*dst6); 6375 dst6->sin6_addr = addr->v6; 6376 if (ip6_multipath) 6377 check_mpath = 1; 6378 break; 6379 #endif /* INET6 */ 6380 } 6381 6382 /* Skip checks for ipsec interfaces */ 6383 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 6384 goto out; 6385 6386 rt = rtalloc(sstosa(&ss), 0, rtableid); 6387 if (rt != NULL) { 6388 /* No interface given, this is a no-route check */ 6389 if (kif == NULL) 6390 goto out; 6391 6392 if (kif->pfik_ifp == NULL) { 6393 ret = 0; 6394 goto out; 6395 } 6396 6397 /* Perform uRPF check if passed input interface */ 6398 ret = 0; 6399 do { 6400 if (rt->rt_ifidx == kif->pfik_ifp->if_index) { 6401 ret = 1; 6402 #if NCARP > 0 6403 } else { 6404 struct ifnet *ifp; 6405 6406 ifp = if_get(rt->rt_ifidx); 6407 if (ifp != NULL && ifp->if_type == IFT_CARP && 6408 ifp->if_carpdevidx == 6409 kif->pfik_ifp->if_index) 6410 ret = 1; 6411 if_put(ifp); 6412 #endif /* NCARP */ 6413 } 6414 6415 rt = rtable_iterate(rt); 6416 } while (check_mpath == 1 && rt != NULL && ret == 0); 6417 } else 6418 ret = 0; 6419 out: 6420 rtfree(rt); 6421 return (ret); 6422 } 6423 6424 int 6425 pf_rtlabel_match(struct pf_addr *addr, sa_family_t af, struct pf_addr_wrap *aw, 6426 int rtableid) 6427 { 6428 struct sockaddr_storage ss; 6429 struct sockaddr_in *dst; 6430 #ifdef INET6 6431 struct sockaddr_in6 *dst6; 6432 #endif /* INET6 */ 6433 struct rtentry *rt; 6434 int ret = 0; 6435 6436 memset(&ss, 0, sizeof(ss)); 6437 switch (af) { 6438 case AF_INET: 6439 dst = (struct sockaddr_in *)&ss; 6440 dst->sin_family = AF_INET; 6441 dst->sin_len = sizeof(*dst); 6442 dst->sin_addr = addr->v4; 6443 break; 6444 #ifdef INET6 6445 case AF_INET6: 6446 dst6 = (struct sockaddr_in6 *)&ss; 6447 dst6->sin6_family = AF_INET6; 6448 dst6->sin6_len = sizeof(*dst6); 6449 dst6->sin6_addr = addr->v6; 6450 break; 6451 #endif /* INET6 */ 6452 } 6453 6454 rt = rtalloc(sstosa(&ss), RT_RESOLVE, rtableid); 6455 if (rt != NULL) { 6456 if (rt->rt_labelid == aw->v.rtlabel) 6457 ret = 1; 6458 rtfree(rt); 6459 } 6460 6461 return (ret); 6462 } 6463 6464 /* pf_route() may change pd->m, adjust local copies after calling */ 6465 void 6466 pf_route(struct pf_pdesc *pd, struct pf_state *st) 6467 { 6468 struct mbuf *m0; 6469 struct mbuf_list ml; 6470 struct sockaddr_in *dst, sin; 6471 struct rtentry *rt = NULL; 6472 struct ip *ip; 6473 struct ifnet *ifp = NULL; 6474 unsigned int rtableid; 6475 6476 if (pd->m->m_pkthdr.pf.routed++ > 3) { 6477 m_freem(pd->m); 6478 pd->m = NULL; 6479 return; 6480 } 6481 6482 if (st->rt == PF_DUPTO) { 6483 if ((m0 = m_dup_pkt(pd->m, max_linkhdr, M_NOWAIT)) == NULL) 6484 return; 6485 } else { 6486 if ((st->rt == PF_REPLYTO) == (st->direction == pd->dir)) 6487 return; 6488 m0 = pd->m; 6489 pd->m = NULL; 6490 } 6491 6492 if (m0->m_len < sizeof(struct ip)) { 6493 DPFPRINTF(LOG_ERR, 6494 "%s: m0->m_len < sizeof(struct ip)", __func__); 6495 goto bad; 6496 } 6497 6498 ip = mtod(m0, struct ip *); 6499 6500 if (pd->dir == PF_IN) { 6501 if (ip->ip_ttl <= IPTTLDEC) { 6502 if (st->rt != PF_DUPTO) { 6503 pf_send_icmp(m0, ICMP_TIMXCEED, 6504 ICMP_TIMXCEED_INTRANS, 0, 6505 pd->af, st->rule.ptr, pd->rdomain); 6506 } 6507 goto bad; 6508 } 6509 ip->ip_ttl -= IPTTLDEC; 6510 } 6511 6512 memset(&sin, 0, sizeof(sin)); 6513 dst = &sin; 6514 dst->sin_family = AF_INET; 6515 dst->sin_len = sizeof(*dst); 6516 dst->sin_addr = st->rt_addr.v4; 6517 rtableid = m0->m_pkthdr.ph_rtableid; 6518 6519 rt = rtalloc_mpath(sintosa(dst), &ip->ip_src.s_addr, rtableid); 6520 if (!rtisvalid(rt)) { 6521 if (st->rt != PF_DUPTO) { 6522 pf_send_icmp(m0, ICMP_UNREACH, ICMP_UNREACH_HOST, 6523 0, pd->af, st->rule.ptr, pd->rdomain); 6524 } 6525 ipstat_inc(ips_noroute); 6526 goto bad; 6527 } 6528 6529 ifp = if_get(rt->rt_ifidx); 6530 if (ifp == NULL) 6531 goto bad; 6532 6533 /* A locally generated packet may have invalid source address. */ 6534 if ((ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET && 6535 (ifp->if_flags & IFF_LOOPBACK) == 0) 6536 ip->ip_src = ifatoia(rt->rt_ifa)->ia_addr.sin_addr; 6537 6538 if (st->rt != PF_DUPTO && pd->dir == PF_IN) { 6539 if (pf_test(AF_INET, PF_OUT, ifp, &m0) != PF_PASS) 6540 goto bad; 6541 else if (m0 == NULL) 6542 goto done; 6543 if (m0->m_len < sizeof(struct ip)) { 6544 DPFPRINTF(LOG_ERR, 6545 "%s: m0->m_len < sizeof(struct ip)", __func__); 6546 goto bad; 6547 } 6548 ip = mtod(m0, struct ip *); 6549 } 6550 6551 if (ntohs(ip->ip_len) <= ifp->if_mtu) { 6552 in_hdr_cksum_out(m0, ifp); 6553 in_proto_cksum_out(m0, ifp); 6554 ifp->if_output(ifp, m0, sintosa(dst), rt); 6555 goto done; 6556 } 6557 6558 if (tcp_if_output_tso(ifp, &m0, sintosa(dst), rt, 6559 IFCAP_TSOv4, ifp->if_mtu) || m0 == NULL) 6560 goto done; 6561 6562 /* 6563 * Too large for interface; fragment if possible. 6564 * Must be able to put at least 8 bytes per fragment. 6565 */ 6566 if (ip->ip_off & htons(IP_DF)) { 6567 ipstat_inc(ips_cantfrag); 6568 if (st->rt != PF_DUPTO) 6569 pf_send_icmp(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 6570 ifp->if_mtu, pd->af, st->rule.ptr, pd->rdomain); 6571 goto bad; 6572 } 6573 6574 if (ip_fragment(m0, &ml, ifp, ifp->if_mtu) || 6575 if_output_ml(ifp, &ml, sintosa(dst), rt)) 6576 goto done; 6577 ipstat_inc(ips_fragmented); 6578 6579 done: 6580 if_put(ifp); 6581 rtfree(rt); 6582 return; 6583 6584 bad: 6585 m_freem(m0); 6586 goto done; 6587 } 6588 6589 #ifdef INET6 6590 /* pf_route6() may change pd->m, adjust local copies after calling */ 6591 void 6592 pf_route6(struct pf_pdesc *pd, struct pf_state *st) 6593 { 6594 struct mbuf *m0; 6595 struct sockaddr_in6 *dst, sin6; 6596 struct rtentry *rt = NULL; 6597 struct ip6_hdr *ip6; 6598 struct ifnet *ifp = NULL; 6599 struct m_tag *mtag; 6600 unsigned int rtableid; 6601 6602 if (pd->m->m_pkthdr.pf.routed++ > 3) { 6603 m_freem(pd->m); 6604 pd->m = NULL; 6605 return; 6606 } 6607 6608 if (st->rt == PF_DUPTO) { 6609 if ((m0 = m_dup_pkt(pd->m, max_linkhdr, M_NOWAIT)) == NULL) 6610 return; 6611 } else { 6612 if ((st->rt == PF_REPLYTO) == (st->direction == pd->dir)) 6613 return; 6614 m0 = pd->m; 6615 pd->m = NULL; 6616 } 6617 6618 if (m0->m_len < sizeof(struct ip6_hdr)) { 6619 DPFPRINTF(LOG_ERR, 6620 "%s: m0->m_len < sizeof(struct ip6_hdr)", __func__); 6621 goto bad; 6622 } 6623 ip6 = mtod(m0, struct ip6_hdr *); 6624 6625 if (pd->dir == PF_IN) { 6626 if (ip6->ip6_hlim <= IPV6_HLIMDEC) { 6627 if (st->rt != PF_DUPTO) { 6628 pf_send_icmp(m0, ICMP6_TIME_EXCEEDED, 6629 ICMP6_TIME_EXCEED_TRANSIT, 0, 6630 pd->af, st->rule.ptr, pd->rdomain); 6631 } 6632 goto bad; 6633 } 6634 ip6->ip6_hlim -= IPV6_HLIMDEC; 6635 } 6636 6637 memset(&sin6, 0, sizeof(sin6)); 6638 dst = &sin6; 6639 dst->sin6_family = AF_INET6; 6640 dst->sin6_len = sizeof(*dst); 6641 dst->sin6_addr = st->rt_addr.v6; 6642 rtableid = m0->m_pkthdr.ph_rtableid; 6643 6644 rt = rtalloc_mpath(sin6tosa(dst), &ip6->ip6_src.s6_addr32[0], 6645 rtableid); 6646 if (!rtisvalid(rt)) { 6647 if (st->rt != PF_DUPTO) { 6648 pf_send_icmp(m0, ICMP6_DST_UNREACH, 6649 ICMP6_DST_UNREACH_NOROUTE, 0, 6650 pd->af, st->rule.ptr, pd->rdomain); 6651 } 6652 ip6stat_inc(ip6s_noroute); 6653 goto bad; 6654 } 6655 6656 ifp = if_get(rt->rt_ifidx); 6657 if (ifp == NULL) 6658 goto bad; 6659 6660 /* A locally generated packet may have invalid source address. */ 6661 if (IN6_IS_ADDR_LOOPBACK(&ip6->ip6_src) && 6662 (ifp->if_flags & IFF_LOOPBACK) == 0) 6663 ip6->ip6_src = ifatoia6(rt->rt_ifa)->ia_addr.sin6_addr; 6664 6665 if (st->rt != PF_DUPTO && pd->dir == PF_IN) { 6666 if (pf_test(AF_INET6, PF_OUT, ifp, &m0) != PF_PASS) 6667 goto bad; 6668 else if (m0 == NULL) 6669 goto done; 6670 if (m0->m_len < sizeof(struct ip6_hdr)) { 6671 DPFPRINTF(LOG_ERR, 6672 "%s: m0->m_len < sizeof(struct ip6_hdr)", __func__); 6673 goto bad; 6674 } 6675 } 6676 6677 /* 6678 * If packet has been reassembled by PF earlier, we have to 6679 * use pf_refragment6() here to turn it back to fragments. 6680 */ 6681 if ((mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL))) { 6682 (void) pf_refragment6(&m0, mtag, dst, ifp, rt); 6683 goto done; 6684 } 6685 6686 if (m0->m_pkthdr.len <= ifp->if_mtu) { 6687 in6_proto_cksum_out(m0, ifp); 6688 ifp->if_output(ifp, m0, sin6tosa(dst), rt); 6689 goto done; 6690 } 6691 6692 if (tcp_if_output_tso(ifp, &m0, sin6tosa(dst), rt, 6693 IFCAP_TSOv6, ifp->if_mtu) || m0 == NULL) 6694 goto done; 6695 6696 ip6stat_inc(ip6s_cantfrag); 6697 if (st->rt != PF_DUPTO) 6698 pf_send_icmp(m0, ICMP6_PACKET_TOO_BIG, 0, 6699 ifp->if_mtu, pd->af, st->rule.ptr, pd->rdomain); 6700 goto bad; 6701 6702 done: 6703 if_put(ifp); 6704 rtfree(rt); 6705 return; 6706 6707 bad: 6708 m_freem(m0); 6709 goto done; 6710 } 6711 #endif /* INET6 */ 6712 6713 /* 6714 * check TCP checksum and set mbuf flag 6715 * off is the offset where the protocol header starts 6716 * len is the total length of protocol header plus payload 6717 * returns 0 when the checksum is valid, otherwise returns 1. 6718 * if the _OUT flag is set the checksum isn't done yet, consider these ok 6719 */ 6720 int 6721 pf_check_tcp_cksum(struct mbuf *m, int off, int len, sa_family_t af) 6722 { 6723 u_int16_t sum; 6724 6725 if (m->m_pkthdr.csum_flags & 6726 (M_TCP_CSUM_IN_OK | M_TCP_CSUM_OUT)) { 6727 return (0); 6728 } 6729 if (m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_BAD || 6730 off < sizeof(struct ip) || 6731 m->m_pkthdr.len < off + len) { 6732 return (1); 6733 } 6734 6735 /* need to do it in software */ 6736 tcpstat_inc(tcps_inswcsum); 6737 6738 switch (af) { 6739 case AF_INET: 6740 if (m->m_len < sizeof(struct ip)) 6741 return (1); 6742 6743 sum = in4_cksum(m, IPPROTO_TCP, off, len); 6744 break; 6745 #ifdef INET6 6746 case AF_INET6: 6747 if (m->m_len < sizeof(struct ip6_hdr)) 6748 return (1); 6749 6750 sum = in6_cksum(m, IPPROTO_TCP, off, len); 6751 break; 6752 #endif /* INET6 */ 6753 default: 6754 unhandled_af(af); 6755 } 6756 if (sum) { 6757 tcpstat_inc(tcps_rcvbadsum); 6758 m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_BAD; 6759 return (1); 6760 } 6761 6762 m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK; 6763 return (0); 6764 } 6765 6766 struct pf_divert * 6767 pf_find_divert(struct mbuf *m) 6768 { 6769 struct m_tag *mtag; 6770 6771 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) 6772 return (NULL); 6773 6774 return ((struct pf_divert *)(mtag + 1)); 6775 } 6776 6777 struct pf_divert * 6778 pf_get_divert(struct mbuf *m) 6779 { 6780 struct m_tag *mtag; 6781 6782 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) { 6783 mtag = m_tag_get(PACKET_TAG_PF_DIVERT, sizeof(struct pf_divert), 6784 M_NOWAIT); 6785 if (mtag == NULL) 6786 return (NULL); 6787 memset(mtag + 1, 0, sizeof(struct pf_divert)); 6788 m_tag_prepend(m, mtag); 6789 } 6790 6791 return ((struct pf_divert *)(mtag + 1)); 6792 } 6793 6794 int 6795 pf_walk_option(struct pf_pdesc *pd, struct ip *h, int off, int end, 6796 u_short *reason) 6797 { 6798 uint8_t type, length, opts[15 * 4 - sizeof(struct ip)]; 6799 6800 /* IP header in payload of ICMP packet may be too short */ 6801 if (pd->m->m_pkthdr.len < end) { 6802 DPFPRINTF(LOG_NOTICE, "IP option too short"); 6803 REASON_SET(reason, PFRES_SHORT); 6804 return (PF_DROP); 6805 } 6806 6807 KASSERT(end - off <= sizeof(opts)); 6808 m_copydata(pd->m, off, end - off, opts); 6809 end -= off; 6810 off = 0; 6811 6812 while (off < end) { 6813 type = opts[off]; 6814 if (type == IPOPT_EOL) 6815 break; 6816 if (type == IPOPT_NOP) { 6817 off++; 6818 continue; 6819 } 6820 if (off + 2 > end) { 6821 DPFPRINTF(LOG_NOTICE, "IP length opt"); 6822 REASON_SET(reason, PFRES_IPOPTIONS); 6823 return (PF_DROP); 6824 } 6825 length = opts[off + 1]; 6826 if (length < 2) { 6827 DPFPRINTF(LOG_NOTICE, "IP short opt"); 6828 REASON_SET(reason, PFRES_IPOPTIONS); 6829 return (PF_DROP); 6830 } 6831 if (off + length > end) { 6832 DPFPRINTF(LOG_NOTICE, "IP long opt"); 6833 REASON_SET(reason, PFRES_IPOPTIONS); 6834 return (PF_DROP); 6835 } 6836 switch (type) { 6837 case IPOPT_RA: 6838 SET(pd->badopts, PF_OPT_ROUTER_ALERT); 6839 break; 6840 default: 6841 SET(pd->badopts, PF_OPT_OTHER); 6842 break; 6843 } 6844 off += length; 6845 } 6846 6847 return (PF_PASS); 6848 } 6849 6850 int 6851 pf_walk_header(struct pf_pdesc *pd, struct ip *h, u_short *reason) 6852 { 6853 struct ip6_ext ext; 6854 u_int32_t hlen, end; 6855 int hdr_cnt; 6856 6857 hlen = h->ip_hl << 2; 6858 if (hlen < sizeof(struct ip) || hlen > ntohs(h->ip_len)) { 6859 REASON_SET(reason, PFRES_SHORT); 6860 return (PF_DROP); 6861 } 6862 if (hlen != sizeof(struct ip)) { 6863 if (pf_walk_option(pd, h, pd->off + sizeof(struct ip), 6864 pd->off + hlen, reason) != PF_PASS) 6865 return (PF_DROP); 6866 /* header options which contain only padding is fishy */ 6867 if (pd->badopts == 0) 6868 SET(pd->badopts, PF_OPT_OTHER); 6869 } 6870 end = pd->off + ntohs(h->ip_len); 6871 pd->off += hlen; 6872 pd->proto = h->ip_p; 6873 /* IGMP packets have router alert options, allow them */ 6874 if (pd->proto == IPPROTO_IGMP) { 6875 /* 6876 * According to RFC 1112 ttl must be set to 1 in all IGMP 6877 * packets sent to 224.0.0.1 6878 */ 6879 if ((h->ip_ttl != 1) && 6880 (h->ip_dst.s_addr == INADDR_ALLHOSTS_GROUP)) { 6881 DPFPRINTF(LOG_NOTICE, "Invalid IGMP"); 6882 REASON_SET(reason, PFRES_IPOPTIONS); 6883 return (PF_DROP); 6884 } 6885 CLR(pd->badopts, PF_OPT_ROUTER_ALERT); 6886 } 6887 /* stop walking over non initial fragments */ 6888 if ((h->ip_off & htons(IP_OFFMASK)) != 0) 6889 return (PF_PASS); 6890 6891 for (hdr_cnt = 0; hdr_cnt < pf_hdr_limit; hdr_cnt++) { 6892 switch (pd->proto) { 6893 case IPPROTO_AH: 6894 /* fragments may be short */ 6895 if ((h->ip_off & htons(IP_MF | IP_OFFMASK)) != 0 && 6896 end < pd->off + sizeof(ext)) 6897 return (PF_PASS); 6898 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 6899 NULL, reason, AF_INET)) { 6900 DPFPRINTF(LOG_NOTICE, "IP short exthdr"); 6901 return (PF_DROP); 6902 } 6903 pd->off += (ext.ip6e_len + 2) * 4; 6904 pd->proto = ext.ip6e_nxt; 6905 break; 6906 default: 6907 return (PF_PASS); 6908 } 6909 } 6910 DPFPRINTF(LOG_NOTICE, "IPv4 nested authentication header limit"); 6911 REASON_SET(reason, PFRES_IPOPTIONS); 6912 return (PF_DROP); 6913 } 6914 6915 #ifdef INET6 6916 int 6917 pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end, 6918 u_short *reason) 6919 { 6920 struct ip6_opt opt; 6921 struct ip6_opt_jumbo jumbo; 6922 6923 while (off < end) { 6924 if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type, 6925 sizeof(opt.ip6o_type), NULL, reason, AF_INET6)) { 6926 DPFPRINTF(LOG_NOTICE, "IPv6 short opt type"); 6927 return (PF_DROP); 6928 } 6929 if (opt.ip6o_type == IP6OPT_PAD1) { 6930 off++; 6931 continue; 6932 } 6933 if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), 6934 NULL, reason, AF_INET6)) { 6935 DPFPRINTF(LOG_NOTICE, "IPv6 short opt"); 6936 return (PF_DROP); 6937 } 6938 if (off + sizeof(opt) + opt.ip6o_len > end) { 6939 DPFPRINTF(LOG_NOTICE, "IPv6 long opt"); 6940 REASON_SET(reason, PFRES_IPOPTIONS); 6941 return (PF_DROP); 6942 } 6943 switch (opt.ip6o_type) { 6944 case IP6OPT_PADN: 6945 break; 6946 case IP6OPT_JUMBO: 6947 SET(pd->badopts, PF_OPT_JUMBO); 6948 if (pd->jumbolen != 0) { 6949 DPFPRINTF(LOG_NOTICE, "IPv6 multiple jumbo"); 6950 REASON_SET(reason, PFRES_IPOPTIONS); 6951 return (PF_DROP); 6952 } 6953 if (ntohs(h->ip6_plen) != 0) { 6954 DPFPRINTF(LOG_NOTICE, "IPv6 bad jumbo plen"); 6955 REASON_SET(reason, PFRES_IPOPTIONS); 6956 return (PF_DROP); 6957 } 6958 if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), 6959 NULL, reason, AF_INET6)) { 6960 DPFPRINTF(LOG_NOTICE, "IPv6 short jumbo"); 6961 return (PF_DROP); 6962 } 6963 memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len, 6964 sizeof(pd->jumbolen)); 6965 pd->jumbolen = ntohl(pd->jumbolen); 6966 if (pd->jumbolen < IPV6_MAXPACKET) { 6967 DPFPRINTF(LOG_NOTICE, "IPv6 short jumbolen"); 6968 REASON_SET(reason, PFRES_IPOPTIONS); 6969 return (PF_DROP); 6970 } 6971 break; 6972 case IP6OPT_ROUTER_ALERT: 6973 SET(pd->badopts, PF_OPT_ROUTER_ALERT); 6974 break; 6975 default: 6976 SET(pd->badopts, PF_OPT_OTHER); 6977 break; 6978 } 6979 off += sizeof(opt) + opt.ip6o_len; 6980 } 6981 6982 return (PF_PASS); 6983 } 6984 6985 int 6986 pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason) 6987 { 6988 struct ip6_frag frag; 6989 struct ip6_ext ext; 6990 struct icmp6_hdr icmp6; 6991 struct ip6_rthdr rthdr; 6992 u_int32_t end; 6993 int hdr_cnt, fraghdr_cnt = 0, rthdr_cnt = 0; 6994 6995 pd->off += sizeof(struct ip6_hdr); 6996 end = pd->off + ntohs(h->ip6_plen); 6997 pd->fragoff = pd->extoff = pd->jumbolen = 0; 6998 pd->proto = h->ip6_nxt; 6999 7000 for (hdr_cnt = 0; hdr_cnt < pf_hdr_limit; hdr_cnt++) { 7001 switch (pd->proto) { 7002 case IPPROTO_ROUTING: 7003 case IPPROTO_DSTOPTS: 7004 SET(pd->badopts, PF_OPT_OTHER); 7005 break; 7006 case IPPROTO_HOPOPTS: 7007 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 7008 NULL, reason, AF_INET6)) { 7009 DPFPRINTF(LOG_NOTICE, "IPv6 short exthdr"); 7010 return (PF_DROP); 7011 } 7012 if (pf_walk_option6(pd, h, pd->off + sizeof(ext), 7013 pd->off + (ext.ip6e_len + 1) * 8, reason) 7014 != PF_PASS) 7015 return (PF_DROP); 7016 /* option header which contains only padding is fishy */ 7017 if (pd->badopts == 0) 7018 SET(pd->badopts, PF_OPT_OTHER); 7019 break; 7020 } 7021 switch (pd->proto) { 7022 case IPPROTO_FRAGMENT: 7023 if (fraghdr_cnt++) { 7024 DPFPRINTF(LOG_NOTICE, "IPv6 multiple fragment"); 7025 REASON_SET(reason, PFRES_FRAG); 7026 return (PF_DROP); 7027 } 7028 /* jumbo payload packets cannot be fragmented */ 7029 if (pd->jumbolen != 0) { 7030 DPFPRINTF(LOG_NOTICE, "IPv6 fragmented jumbo"); 7031 REASON_SET(reason, PFRES_FRAG); 7032 return (PF_DROP); 7033 } 7034 if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag), 7035 NULL, reason, AF_INET6)) { 7036 DPFPRINTF(LOG_NOTICE, "IPv6 short fragment"); 7037 return (PF_DROP); 7038 } 7039 /* stop walking over non initial fragments */ 7040 if (ntohs((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) { 7041 pd->fragoff = pd->off; 7042 return (PF_PASS); 7043 } 7044 /* RFC6946: reassemble only non atomic fragments */ 7045 if (frag.ip6f_offlg & IP6F_MORE_FRAG) 7046 pd->fragoff = pd->off; 7047 pd->off += sizeof(frag); 7048 pd->proto = frag.ip6f_nxt; 7049 break; 7050 case IPPROTO_ROUTING: 7051 if (rthdr_cnt++) { 7052 DPFPRINTF(LOG_NOTICE, "IPv6 multiple rthdr"); 7053 REASON_SET(reason, PFRES_IPOPTIONS); 7054 return (PF_DROP); 7055 } 7056 /* fragments may be short */ 7057 if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) { 7058 pd->off = pd->fragoff; 7059 pd->proto = IPPROTO_FRAGMENT; 7060 return (PF_PASS); 7061 } 7062 if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr), 7063 NULL, reason, AF_INET6)) { 7064 DPFPRINTF(LOG_NOTICE, "IPv6 short rthdr"); 7065 return (PF_DROP); 7066 } 7067 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 7068 DPFPRINTF(LOG_NOTICE, "IPv6 rthdr0"); 7069 REASON_SET(reason, PFRES_IPOPTIONS); 7070 return (PF_DROP); 7071 } 7072 /* FALLTHROUGH */ 7073 case IPPROTO_HOPOPTS: 7074 /* RFC2460 4.1: Hop-by-Hop only after IPv6 header */ 7075 if (pd->proto == IPPROTO_HOPOPTS && hdr_cnt > 0) { 7076 DPFPRINTF(LOG_NOTICE, "IPv6 hopopts not first"); 7077 REASON_SET(reason, PFRES_IPOPTIONS); 7078 return (PF_DROP); 7079 } 7080 /* FALLTHROUGH */ 7081 case IPPROTO_AH: 7082 case IPPROTO_DSTOPTS: 7083 /* fragments may be short */ 7084 if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) { 7085 pd->off = pd->fragoff; 7086 pd->proto = IPPROTO_FRAGMENT; 7087 return (PF_PASS); 7088 } 7089 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 7090 NULL, reason, AF_INET6)) { 7091 DPFPRINTF(LOG_NOTICE, "IPv6 short exthdr"); 7092 return (PF_DROP); 7093 } 7094 /* reassembly needs the ext header before the frag */ 7095 if (pd->fragoff == 0) 7096 pd->extoff = pd->off; 7097 if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0 && 7098 ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) { 7099 DPFPRINTF(LOG_NOTICE, "IPv6 missing jumbo"); 7100 REASON_SET(reason, PFRES_IPOPTIONS); 7101 return (PF_DROP); 7102 } 7103 if (pd->proto == IPPROTO_AH) 7104 pd->off += (ext.ip6e_len + 2) * 4; 7105 else 7106 pd->off += (ext.ip6e_len + 1) * 8; 7107 pd->proto = ext.ip6e_nxt; 7108 break; 7109 case IPPROTO_ICMPV6: 7110 /* fragments may be short, ignore inner header then */ 7111 if (pd->fragoff != 0 && end < pd->off + sizeof(icmp6)) { 7112 pd->off = pd->fragoff; 7113 pd->proto = IPPROTO_FRAGMENT; 7114 return (PF_PASS); 7115 } 7116 if (!pf_pull_hdr(pd->m, pd->off, &icmp6, sizeof(icmp6), 7117 NULL, reason, AF_INET6)) { 7118 DPFPRINTF(LOG_NOTICE, "IPv6 short icmp6hdr"); 7119 return (PF_DROP); 7120 } 7121 /* ICMP multicast packets have router alert options */ 7122 switch (icmp6.icmp6_type) { 7123 case MLD_LISTENER_QUERY: 7124 case MLD_LISTENER_REPORT: 7125 case MLD_LISTENER_DONE: 7126 case MLDV2_LISTENER_REPORT: 7127 /* 7128 * According to RFC 2710 all MLD messages are 7129 * sent with hop-limit (ttl) set to 1, and link 7130 * local source address. If either one is 7131 * missing then MLD message is invalid and 7132 * should be discarded. 7133 */ 7134 if ((h->ip6_hlim != 1) || 7135 !IN6_IS_ADDR_LINKLOCAL(&h->ip6_src)) { 7136 DPFPRINTF(LOG_NOTICE, "Invalid MLD"); 7137 REASON_SET(reason, PFRES_IPOPTIONS); 7138 return (PF_DROP); 7139 } 7140 CLR(pd->badopts, PF_OPT_ROUTER_ALERT); 7141 break; 7142 } 7143 return (PF_PASS); 7144 case IPPROTO_TCP: 7145 case IPPROTO_UDP: 7146 /* fragments may be short, ignore inner header then */ 7147 if (pd->fragoff != 0 && end < pd->off + 7148 (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) : 7149 pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) : 7150 sizeof(struct icmp6_hdr))) { 7151 pd->off = pd->fragoff; 7152 pd->proto = IPPROTO_FRAGMENT; 7153 } 7154 /* FALLTHROUGH */ 7155 default: 7156 return (PF_PASS); 7157 } 7158 } 7159 DPFPRINTF(LOG_NOTICE, "IPv6 nested extension header limit"); 7160 REASON_SET(reason, PFRES_IPOPTIONS); 7161 return (PF_DROP); 7162 } 7163 #endif /* INET6 */ 7164 7165 u_int16_t 7166 pf_pkt_hash(sa_family_t af, uint8_t proto, 7167 const struct pf_addr *src, const struct pf_addr *dst, 7168 uint16_t sport, uint16_t dport) 7169 { 7170 uint32_t hash; 7171 7172 hash = src->addr32[0] ^ dst->addr32[0]; 7173 #ifdef INET6 7174 if (af == AF_INET6) { 7175 hash ^= src->addr32[1] ^ dst->addr32[1]; 7176 hash ^= src->addr32[2] ^ dst->addr32[2]; 7177 hash ^= src->addr32[3] ^ dst->addr32[3]; 7178 } 7179 #endif 7180 7181 switch (proto) { 7182 case IPPROTO_TCP: 7183 case IPPROTO_UDP: 7184 hash ^= sport ^ dport; 7185 break; 7186 } 7187 7188 return stoeplitz_n32(hash); 7189 } 7190 7191 int 7192 pf_setup_pdesc(struct pf_pdesc *pd, sa_family_t af, int dir, 7193 struct pfi_kif *kif, struct mbuf *m, u_short *reason) 7194 { 7195 memset(pd, 0, sizeof(*pd)); 7196 pd->dir = dir; 7197 pd->kif = kif; /* kif is NULL when called by pflog */ 7198 pd->m = m; 7199 pd->sidx = (dir == PF_IN) ? 0 : 1; 7200 pd->didx = (dir == PF_IN) ? 1 : 0; 7201 pd->af = pd->naf = af; 7202 pd->rdomain = rtable_l2(pd->m->m_pkthdr.ph_rtableid); 7203 7204 switch (pd->af) { 7205 case AF_INET: { 7206 struct ip *h; 7207 7208 /* Check for illegal packets */ 7209 if (pd->m->m_pkthdr.len < (int)sizeof(struct ip)) { 7210 REASON_SET(reason, PFRES_SHORT); 7211 return (PF_DROP); 7212 } 7213 7214 h = mtod(pd->m, struct ip *); 7215 if (pd->m->m_pkthdr.len < ntohs(h->ip_len)) { 7216 REASON_SET(reason, PFRES_SHORT); 7217 return (PF_DROP); 7218 } 7219 7220 if (pf_walk_header(pd, h, reason) != PF_PASS) 7221 return (PF_DROP); 7222 7223 pd->src = (struct pf_addr *)&h->ip_src; 7224 pd->dst = (struct pf_addr *)&h->ip_dst; 7225 pd->tot_len = ntohs(h->ip_len); 7226 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK; 7227 pd->ttl = h->ip_ttl; 7228 pd->virtual_proto = (h->ip_off & htons(IP_MF | IP_OFFMASK)) ? 7229 PF_VPROTO_FRAGMENT : pd->proto; 7230 7231 break; 7232 } 7233 #ifdef INET6 7234 case AF_INET6: { 7235 struct ip6_hdr *h; 7236 7237 /* Check for illegal packets */ 7238 if (pd->m->m_pkthdr.len < (int)sizeof(struct ip6_hdr)) { 7239 REASON_SET(reason, PFRES_SHORT); 7240 return (PF_DROP); 7241 } 7242 7243 h = mtod(pd->m, struct ip6_hdr *); 7244 if (pd->m->m_pkthdr.len < 7245 sizeof(struct ip6_hdr) + ntohs(h->ip6_plen)) { 7246 REASON_SET(reason, PFRES_SHORT); 7247 return (PF_DROP); 7248 } 7249 7250 if (pf_walk_header6(pd, h, reason) != PF_PASS) 7251 return (PF_DROP); 7252 7253 #if 1 7254 /* 7255 * we do not support jumbogram yet. if we keep going, zero 7256 * ip6_plen will do something bad, so drop the packet for now. 7257 */ 7258 if (pd->jumbolen != 0) { 7259 REASON_SET(reason, PFRES_NORM); 7260 return (PF_DROP); 7261 } 7262 #endif /* 1 */ 7263 7264 pd->src = (struct pf_addr *)&h->ip6_src; 7265 pd->dst = (struct pf_addr *)&h->ip6_dst; 7266 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 7267 pd->tos = (ntohl(h->ip6_flow) & 0x0fc00000) >> 20; 7268 pd->ttl = h->ip6_hlim; 7269 pd->virtual_proto = (pd->fragoff != 0) ? 7270 PF_VPROTO_FRAGMENT : pd->proto; 7271 7272 break; 7273 } 7274 #endif /* INET6 */ 7275 default: 7276 panic("pf_setup_pdesc called with illegal af %u", pd->af); 7277 7278 } 7279 7280 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 7281 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 7282 7283 switch (pd->virtual_proto) { 7284 case IPPROTO_TCP: { 7285 struct tcphdr *th = &pd->hdr.tcp; 7286 7287 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), 7288 NULL, reason, pd->af)) 7289 return (PF_DROP); 7290 pd->hdrlen = sizeof(*th); 7291 if (th->th_dport == 0 || 7292 pd->off + (th->th_off << 2) > pd->tot_len || 7293 (th->th_off << 2) < sizeof(struct tcphdr)) { 7294 REASON_SET(reason, PFRES_SHORT); 7295 return (PF_DROP); 7296 } 7297 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 7298 pd->sport = &th->th_sport; 7299 pd->dport = &th->th_dport; 7300 pd->pcksum = &th->th_sum; 7301 break; 7302 } 7303 case IPPROTO_UDP: { 7304 struct udphdr *uh = &pd->hdr.udp; 7305 7306 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), 7307 NULL, reason, pd->af)) 7308 return (PF_DROP); 7309 pd->hdrlen = sizeof(*uh); 7310 if (uh->uh_dport == 0 || 7311 pd->off + ntohs(uh->uh_ulen) > pd->tot_len || 7312 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 7313 REASON_SET(reason, PFRES_SHORT); 7314 return (PF_DROP); 7315 } 7316 pd->sport = &uh->uh_sport; 7317 pd->dport = &uh->uh_dport; 7318 pd->pcksum = &uh->uh_sum; 7319 break; 7320 } 7321 case IPPROTO_ICMP: { 7322 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 7323 NULL, reason, pd->af)) 7324 return (PF_DROP); 7325 pd->hdrlen = ICMP_MINLEN; 7326 if (pd->off + pd->hdrlen > pd->tot_len) { 7327 REASON_SET(reason, PFRES_SHORT); 7328 return (PF_DROP); 7329 } 7330 pd->pcksum = &pd->hdr.icmp.icmp_cksum; 7331 break; 7332 } 7333 #ifdef INET6 7334 case IPPROTO_ICMPV6: { 7335 size_t icmp_hlen = sizeof(struct icmp6_hdr); 7336 7337 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 7338 NULL, reason, pd->af)) 7339 return (PF_DROP); 7340 /* ICMP headers we look further into to match state */ 7341 switch (pd->hdr.icmp6.icmp6_type) { 7342 case MLD_LISTENER_QUERY: 7343 case MLD_LISTENER_REPORT: 7344 icmp_hlen = sizeof(struct mld_hdr); 7345 break; 7346 case ND_NEIGHBOR_SOLICIT: 7347 case ND_NEIGHBOR_ADVERT: 7348 icmp_hlen = sizeof(struct nd_neighbor_solicit); 7349 /* FALLTHROUGH */ 7350 case ND_ROUTER_SOLICIT: 7351 case ND_ROUTER_ADVERT: 7352 case ND_REDIRECT: 7353 if (pd->ttl != 255) { 7354 REASON_SET(reason, PFRES_NORM); 7355 return (PF_DROP); 7356 } 7357 break; 7358 } 7359 if (icmp_hlen > sizeof(struct icmp6_hdr) && 7360 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 7361 NULL, reason, pd->af)) 7362 return (PF_DROP); 7363 pd->hdrlen = icmp_hlen; 7364 if (pd->off + pd->hdrlen > pd->tot_len) { 7365 REASON_SET(reason, PFRES_SHORT); 7366 return (PF_DROP); 7367 } 7368 pd->pcksum = &pd->hdr.icmp6.icmp6_cksum; 7369 break; 7370 } 7371 #endif /* INET6 */ 7372 } 7373 7374 if (pd->sport) 7375 pd->osport = pd->nsport = *pd->sport; 7376 if (pd->dport) 7377 pd->odport = pd->ndport = *pd->dport; 7378 7379 pd->hash = pf_pkt_hash(pd->af, pd->proto, 7380 pd->src, pd->dst, pd->osport, pd->odport); 7381 7382 return (PF_PASS); 7383 } 7384 7385 void 7386 pf_counters_inc(int action, struct pf_pdesc *pd, struct pf_state *st, 7387 struct pf_rule *r, struct pf_rule *a) 7388 { 7389 int dirndx; 7390 pd->kif->pfik_bytes[pd->af == AF_INET6][pd->dir == PF_OUT] 7391 [action != PF_PASS] += pd->tot_len; 7392 pd->kif->pfik_packets[pd->af == AF_INET6][pd->dir == PF_OUT] 7393 [action != PF_PASS]++; 7394 7395 if (action == PF_PASS || action == PF_AFRT || r->action == PF_DROP) { 7396 dirndx = (pd->dir == PF_OUT); 7397 r->packets[dirndx]++; 7398 r->bytes[dirndx] += pd->tot_len; 7399 if (a != NULL) { 7400 a->packets[dirndx]++; 7401 a->bytes[dirndx] += pd->tot_len; 7402 } 7403 if (st != NULL) { 7404 struct pf_rule_item *ri; 7405 struct pf_sn_item *sni; 7406 7407 SLIST_FOREACH(sni, &st->src_nodes, next) { 7408 sni->sn->packets[dirndx]++; 7409 sni->sn->bytes[dirndx] += pd->tot_len; 7410 } 7411 dirndx = (pd->dir == st->direction) ? 0 : 1; 7412 st->packets[dirndx]++; 7413 st->bytes[dirndx] += pd->tot_len; 7414 7415 SLIST_FOREACH(ri, &st->match_rules, entry) { 7416 ri->r->packets[dirndx]++; 7417 ri->r->bytes[dirndx] += pd->tot_len; 7418 7419 if (ri->r->src.addr.type == PF_ADDR_TABLE) 7420 pfr_update_stats(ri->r->src.addr.p.tbl, 7421 &st->key[(st->direction == PF_IN)]-> 7422 addr[(st->direction == PF_OUT)], 7423 pd, ri->r->action, ri->r->src.neg); 7424 if (ri->r->dst.addr.type == PF_ADDR_TABLE) 7425 pfr_update_stats(ri->r->dst.addr.p.tbl, 7426 &st->key[(st->direction == PF_IN)]-> 7427 addr[(st->direction == PF_IN)], 7428 pd, ri->r->action, ri->r->dst.neg); 7429 } 7430 } 7431 if (r->src.addr.type == PF_ADDR_TABLE) 7432 pfr_update_stats(r->src.addr.p.tbl, 7433 (st == NULL) ? pd->src : 7434 &st->key[(st->direction == PF_IN)]-> 7435 addr[(st->direction == PF_OUT)], 7436 pd, r->action, r->src.neg); 7437 if (r->dst.addr.type == PF_ADDR_TABLE) 7438 pfr_update_stats(r->dst.addr.p.tbl, 7439 (st == NULL) ? pd->dst : 7440 &st->key[(st->direction == PF_IN)]-> 7441 addr[(st->direction == PF_IN)], 7442 pd, r->action, r->dst.neg); 7443 } 7444 } 7445 7446 int 7447 pf_test(sa_family_t af, int fwdir, struct ifnet *ifp, struct mbuf **m0) 7448 { 7449 #if NCARP > 0 7450 struct ifnet *ifp0; 7451 #endif 7452 struct pfi_kif *kif; 7453 u_short action, reason = 0; 7454 struct pf_rule *a = NULL, *r = &pf_default_rule; 7455 struct pf_state *st = NULL; 7456 struct pf_state_key_cmp key; 7457 struct pf_ruleset *ruleset = NULL; 7458 struct pf_pdesc pd; 7459 int dir = (fwdir == PF_FWD) ? PF_OUT : fwdir; 7460 u_int32_t qid, pqid = 0; 7461 int have_pf_lock = 0; 7462 struct pfsync_deferral *deferral = NULL; 7463 7464 if (!pf_status.running) 7465 return (PF_PASS); 7466 7467 #if NCARP > 0 7468 if (ifp->if_type == IFT_CARP && 7469 (ifp0 = if_get(ifp->if_carpdevidx)) != NULL) { 7470 kif = (struct pfi_kif *)ifp0->if_pf_kif; 7471 if_put(ifp0); 7472 } else 7473 #endif /* NCARP */ 7474 kif = (struct pfi_kif *)ifp->if_pf_kif; 7475 7476 if (kif == NULL) { 7477 DPFPRINTF(LOG_ERR, 7478 "%s: kif == NULL, if_xname %s", __func__, ifp->if_xname); 7479 return (PF_DROP); 7480 } 7481 if (kif->pfik_flags & PFI_IFLAG_SKIP) 7482 return (PF_PASS); 7483 7484 #ifdef DIAGNOSTIC 7485 if (((*m0)->m_flags & M_PKTHDR) == 0) 7486 panic("non-M_PKTHDR is passed to pf_test"); 7487 #endif /* DIAGNOSTIC */ 7488 7489 if ((*m0)->m_pkthdr.pf.flags & PF_TAG_GENERATED) 7490 return (PF_PASS); 7491 7492 if ((*m0)->m_pkthdr.pf.flags & PF_TAG_DIVERTED_PACKET) { 7493 (*m0)->m_pkthdr.pf.flags &= ~PF_TAG_DIVERTED_PACKET; 7494 return (PF_PASS); 7495 } 7496 7497 if ((*m0)->m_pkthdr.pf.flags & PF_TAG_REFRAGMENTED) { 7498 (*m0)->m_pkthdr.pf.flags &= ~PF_TAG_REFRAGMENTED; 7499 return (PF_PASS); 7500 } 7501 7502 action = pf_setup_pdesc(&pd, af, dir, kif, *m0, &reason); 7503 if (action != PF_PASS) { 7504 #if NPFLOG > 0 7505 pd.pflog |= PF_LOG_FORCE; 7506 #endif /* NPFLOG > 0 */ 7507 goto done; 7508 } 7509 7510 /* packet normalization and reassembly */ 7511 switch (pd.af) { 7512 case AF_INET: 7513 action = pf_normalize_ip(&pd, &reason); 7514 break; 7515 #ifdef INET6 7516 case AF_INET6: 7517 action = pf_normalize_ip6(&pd, &reason); 7518 break; 7519 #endif /* INET6 */ 7520 } 7521 *m0 = pd.m; 7522 /* if packet sits in reassembly queue, return without error */ 7523 if (pd.m == NULL) 7524 return PF_PASS; 7525 7526 if (action != PF_PASS) { 7527 #if NPFLOG > 0 7528 pd.pflog |= PF_LOG_FORCE; 7529 #endif /* NPFLOG > 0 */ 7530 goto done; 7531 } 7532 7533 /* if packet has been reassembled, update packet description */ 7534 if (pf_status.reass && pd.virtual_proto == PF_VPROTO_FRAGMENT) { 7535 action = pf_setup_pdesc(&pd, af, dir, kif, pd.m, &reason); 7536 if (action != PF_PASS) { 7537 #if NPFLOG > 0 7538 pd.pflog |= PF_LOG_FORCE; 7539 #endif /* NPFLOG > 0 */ 7540 goto done; 7541 } 7542 } 7543 pd.m->m_pkthdr.pf.flags |= PF_TAG_PROCESSED; 7544 7545 /* 7546 * Avoid pcb-lookups from the forwarding path. They should never 7547 * match and would cause MP locking problems. 7548 */ 7549 if (fwdir == PF_FWD) { 7550 pd.lookup.done = -1; 7551 pd.lookup.uid = -1; 7552 pd.lookup.gid = -1; 7553 pd.lookup.pid = NO_PID; 7554 } 7555 7556 switch (pd.virtual_proto) { 7557 7558 case PF_VPROTO_FRAGMENT: { 7559 /* 7560 * handle fragments that aren't reassembled by 7561 * normalization 7562 */ 7563 PF_LOCK(); 7564 have_pf_lock = 1; 7565 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, &reason, 7566 &deferral); 7567 st = pf_state_ref(st); 7568 if (action != PF_PASS) 7569 REASON_SET(&reason, PFRES_FRAG); 7570 break; 7571 } 7572 7573 case IPPROTO_ICMP: { 7574 if (pd.af != AF_INET) { 7575 action = PF_DROP; 7576 REASON_SET(&reason, PFRES_NORM); 7577 DPFPRINTF(LOG_NOTICE, 7578 "dropping IPv6 packet with ICMPv4 payload"); 7579 break; 7580 } 7581 PF_STATE_ENTER_READ(); 7582 action = pf_test_state_icmp(&pd, &st, &reason); 7583 st = pf_state_ref(st); 7584 PF_STATE_EXIT_READ(); 7585 if (action == PF_PASS || action == PF_AFRT) { 7586 #if NPFSYNC > 0 7587 pfsync_update_state(st); 7588 #endif /* NPFSYNC > 0 */ 7589 r = st->rule.ptr; 7590 a = st->anchor.ptr; 7591 #if NPFLOG > 0 7592 pd.pflog |= st->log; 7593 #endif /* NPFLOG > 0 */ 7594 } else if (st == NULL) { 7595 PF_LOCK(); 7596 have_pf_lock = 1; 7597 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, 7598 &reason, &deferral); 7599 st = pf_state_ref(st); 7600 } 7601 break; 7602 } 7603 7604 #ifdef INET6 7605 case IPPROTO_ICMPV6: { 7606 if (pd.af != AF_INET6) { 7607 action = PF_DROP; 7608 REASON_SET(&reason, PFRES_NORM); 7609 DPFPRINTF(LOG_NOTICE, 7610 "dropping IPv4 packet with ICMPv6 payload"); 7611 break; 7612 } 7613 PF_STATE_ENTER_READ(); 7614 action = pf_test_state_icmp(&pd, &st, &reason); 7615 st = pf_state_ref(st); 7616 PF_STATE_EXIT_READ(); 7617 if (action == PF_PASS || action == PF_AFRT) { 7618 #if NPFSYNC > 0 7619 pfsync_update_state(st); 7620 #endif /* NPFSYNC > 0 */ 7621 r = st->rule.ptr; 7622 a = st->anchor.ptr; 7623 #if NPFLOG > 0 7624 pd.pflog |= st->log; 7625 #endif /* NPFLOG > 0 */ 7626 } else if (st == NULL) { 7627 PF_LOCK(); 7628 have_pf_lock = 1; 7629 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, 7630 &reason, &deferral); 7631 st = pf_state_ref(st); 7632 } 7633 break; 7634 } 7635 #endif /* INET6 */ 7636 7637 default: 7638 if (pd.virtual_proto == IPPROTO_TCP) { 7639 if (pd.dir == PF_IN && (pd.hdr.tcp.th_flags & 7640 (TH_SYN|TH_ACK)) == TH_SYN && 7641 pf_synflood_check(&pd)) { 7642 PF_LOCK(); 7643 have_pf_lock = 1; 7644 pf_syncookie_send(&pd); 7645 action = PF_DROP; 7646 break; 7647 } 7648 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 7649 pqid = 1; 7650 action = pf_normalize_tcp(&pd); 7651 if (action == PF_DROP) 7652 break; 7653 } 7654 7655 key.af = pd.af; 7656 key.proto = pd.virtual_proto; 7657 key.rdomain = pd.rdomain; 7658 pf_addrcpy(&key.addr[pd.sidx], pd.src, key.af); 7659 pf_addrcpy(&key.addr[pd.didx], pd.dst, key.af); 7660 key.port[pd.sidx] = pd.osport; 7661 key.port[pd.didx] = pd.odport; 7662 key.hash = pd.hash; 7663 7664 PF_STATE_ENTER_READ(); 7665 action = pf_find_state(&pd, &key, &st); 7666 st = pf_state_ref(st); 7667 PF_STATE_EXIT_READ(); 7668 7669 /* check for syncookies if tcp ack and no active state */ 7670 if (pd.dir == PF_IN && pd.virtual_proto == IPPROTO_TCP && 7671 (st == NULL || (st->src.state >= TCPS_FIN_WAIT_2 && 7672 st->dst.state >= TCPS_FIN_WAIT_2)) && 7673 (pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 7674 pf_syncookie_validate(&pd)) { 7675 struct mbuf *msyn = pf_syncookie_recreate_syn(&pd); 7676 if (msyn) { 7677 action = pf_test(af, fwdir, ifp, &msyn); 7678 m_freem(msyn); 7679 if (action == PF_PASS || action == PF_AFRT) { 7680 PF_STATE_ENTER_READ(); 7681 pf_state_unref(st); 7682 action = pf_find_state(&pd, &key, &st); 7683 st = pf_state_ref(st); 7684 PF_STATE_EXIT_READ(); 7685 if (st == NULL) 7686 return (PF_DROP); 7687 st->src.seqhi = st->dst.seqhi = 7688 ntohl(pd.hdr.tcp.th_ack) - 1; 7689 st->src.seqlo = 7690 ntohl(pd.hdr.tcp.th_seq) - 1; 7691 pf_set_protostate(st, PF_PEER_SRC, 7692 PF_TCPS_PROXY_DST); 7693 } 7694 } else 7695 action = PF_DROP; 7696 } 7697 7698 if (action == PF_MATCH) 7699 action = pf_test_state(&pd, &st, &reason); 7700 7701 if (action == PF_PASS || action == PF_AFRT) { 7702 #if NPFSYNC > 0 7703 pfsync_update_state(st); 7704 #endif /* NPFSYNC > 0 */ 7705 r = st->rule.ptr; 7706 a = st->anchor.ptr; 7707 #if NPFLOG > 0 7708 pd.pflog |= st->log; 7709 #endif /* NPFLOG > 0 */ 7710 } else if (st == NULL) { 7711 PF_LOCK(); 7712 have_pf_lock = 1; 7713 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, 7714 &reason, &deferral); 7715 st = pf_state_ref(st); 7716 } 7717 7718 if (pd.virtual_proto == IPPROTO_TCP) { 7719 if (st) { 7720 if (st->max_mss) 7721 pf_normalize_mss(&pd, st->max_mss); 7722 } else if (r->max_mss) 7723 pf_normalize_mss(&pd, r->max_mss); 7724 } 7725 7726 break; 7727 } 7728 7729 if (have_pf_lock != 0) 7730 PF_UNLOCK(); 7731 7732 /* 7733 * At the moment, we rely on NET_LOCK() to prevent removal of items 7734 * we've collected above ('r', 'anchor' and 'ruleset'). They'll have 7735 * to be refcounted when NET_LOCK() is gone. 7736 */ 7737 7738 done: 7739 if (action != PF_DROP) { 7740 if (st) { 7741 /* The non-state case is handled in pf_test_rule() */ 7742 if (action == PF_PASS && pd.badopts != 0 && 7743 !(st->state_flags & PFSTATE_ALLOWOPTS)) { 7744 action = PF_DROP; 7745 REASON_SET(&reason, PFRES_IPOPTIONS); 7746 #if NPFLOG > 0 7747 pd.pflog |= PF_LOG_FORCE; 7748 #endif /* NPFLOG > 0 */ 7749 DPFPRINTF(LOG_NOTICE, "dropping packet with " 7750 "ip/ipv6 options in pf_test()"); 7751 } 7752 7753 pf_scrub(pd.m, st->state_flags, pd.af, st->min_ttl, 7754 st->set_tos); 7755 pf_tag_packet(pd.m, st->tag, st->rtableid[pd.didx]); 7756 if (pqid || (pd.tos & IPTOS_LOWDELAY)) { 7757 qid = st->pqid; 7758 if (st->state_flags & PFSTATE_SETPRIO) { 7759 pd.m->m_pkthdr.pf.prio = 7760 st->set_prio[1]; 7761 } 7762 } else { 7763 qid = st->qid; 7764 if (st->state_flags & PFSTATE_SETPRIO) { 7765 pd.m->m_pkthdr.pf.prio = 7766 st->set_prio[0]; 7767 } 7768 } 7769 pd.m->m_pkthdr.pf.delay = st->delay; 7770 } else { 7771 pf_scrub(pd.m, r->scrub_flags, pd.af, r->min_ttl, 7772 r->set_tos); 7773 if (pqid || (pd.tos & IPTOS_LOWDELAY)) { 7774 qid = r->pqid; 7775 if (r->scrub_flags & PFSTATE_SETPRIO) 7776 pd.m->m_pkthdr.pf.prio = r->set_prio[1]; 7777 } else { 7778 qid = r->qid; 7779 if (r->scrub_flags & PFSTATE_SETPRIO) 7780 pd.m->m_pkthdr.pf.prio = r->set_prio[0]; 7781 } 7782 pd.m->m_pkthdr.pf.delay = r->delay; 7783 } 7784 } 7785 7786 if (action == PF_PASS && qid) 7787 pd.m->m_pkthdr.pf.qid = qid; 7788 if (pd.dir == PF_IN && st && st->key[PF_SK_STACK]) 7789 pf_mbuf_link_state_key(pd.m, st->key[PF_SK_STACK]); 7790 if (pd.dir == PF_OUT && 7791 pd.m->m_pkthdr.pf.inp && !pd.m->m_pkthdr.pf.inp->inp_pf_sk && 7792 st && st->key[PF_SK_STACK] && !st->key[PF_SK_STACK]->sk_inp) 7793 pf_state_key_link_inpcb(st->key[PF_SK_STACK], 7794 pd.m->m_pkthdr.pf.inp); 7795 7796 if (st != NULL && !ISSET(pd.m->m_pkthdr.csum_flags, M_FLOWID)) { 7797 pd.m->m_pkthdr.ph_flowid = st->key[PF_SK_WIRE]->hash; 7798 SET(pd.m->m_pkthdr.csum_flags, M_FLOWID); 7799 } 7800 7801 /* 7802 * connections redirected to loopback should not match sockets 7803 * bound specifically to loopback due to security implications, 7804 * see in_pcblookup_listen(). 7805 */ 7806 if (pd.destchg) 7807 if ((pd.af == AF_INET && (ntohl(pd.dst->v4.s_addr) >> 7808 IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) || 7809 (pd.af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))) 7810 pd.m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST; 7811 /* We need to redo the route lookup on outgoing routes. */ 7812 if (pd.destchg && pd.dir == PF_OUT) 7813 pd.m->m_pkthdr.pf.flags |= PF_TAG_REROUTE; 7814 7815 if (pd.dir == PF_IN && action == PF_PASS && 7816 (r->divert.type == PF_DIVERT_TO || 7817 r->divert.type == PF_DIVERT_REPLY)) { 7818 struct pf_divert *divert; 7819 7820 if ((divert = pf_get_divert(pd.m))) { 7821 pd.m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED; 7822 divert->addr = r->divert.addr; 7823 divert->port = r->divert.port; 7824 divert->rdomain = pd.rdomain; 7825 divert->type = r->divert.type; 7826 } 7827 } 7828 7829 if (action == PF_PASS && r->divert.type == PF_DIVERT_PACKET) 7830 action = PF_DIVERT; 7831 7832 #if NPFLOG > 0 7833 if (pd.pflog) { 7834 struct pf_rule_item *ri; 7835 7836 if (pd.pflog & PF_LOG_FORCE || r->log & PF_LOG_ALL) 7837 pflog_packet(&pd, reason, r, a, ruleset, NULL); 7838 if (st) { 7839 SLIST_FOREACH(ri, &st->match_rules, entry) 7840 if (ri->r->log & PF_LOG_ALL) 7841 pflog_packet(&pd, reason, ri->r, a, 7842 ruleset, NULL); 7843 } 7844 } 7845 #endif /* NPFLOG > 0 */ 7846 7847 pf_counters_inc(action, &pd, st, r, a); 7848 7849 switch (action) { 7850 case PF_SYNPROXY_DROP: 7851 m_freem(pd.m); 7852 /* FALLTHROUGH */ 7853 case PF_DEFER: 7854 #if NPFSYNC > 0 7855 /* 7856 * We no longer hold PF_LOCK() here, so we can dispatch 7857 * deferral if we are asked to do so. 7858 */ 7859 if (deferral != NULL) 7860 pfsync_undefer(deferral, 0); 7861 #endif /* NPFSYNC > 0 */ 7862 pd.m = NULL; 7863 action = PF_PASS; 7864 break; 7865 case PF_DIVERT: 7866 switch (pd.af) { 7867 case AF_INET: 7868 divert_packet(pd.m, pd.dir, r->divert.port); 7869 pd.m = NULL; 7870 break; 7871 #ifdef INET6 7872 case AF_INET6: 7873 divert6_packet(pd.m, pd.dir, r->divert.port); 7874 pd.m = NULL; 7875 break; 7876 #endif /* INET6 */ 7877 } 7878 action = PF_PASS; 7879 break; 7880 #ifdef INET6 7881 case PF_AFRT: 7882 if (pf_translate_af(&pd)) { 7883 action = PF_DROP; 7884 break; 7885 } 7886 pd.m->m_pkthdr.pf.flags |= PF_TAG_GENERATED; 7887 switch (pd.naf) { 7888 case AF_INET: 7889 if (pd.dir == PF_IN) { 7890 if (ipforwarding == 0) { 7891 ipstat_inc(ips_cantforward); 7892 action = PF_DROP; 7893 break; 7894 } 7895 ip_forward(pd.m, ifp, NULL, 1); 7896 } else 7897 ip_output(pd.m, NULL, NULL, 0, NULL, NULL, 0); 7898 break; 7899 case AF_INET6: 7900 if (pd.dir == PF_IN) { 7901 if (ip6_forwarding == 0) { 7902 ip6stat_inc(ip6s_cantforward); 7903 action = PF_DROP; 7904 break; 7905 } 7906 ip6_forward(pd.m, NULL, 1); 7907 } else 7908 ip6_output(pd.m, NULL, NULL, 0, NULL, NULL); 7909 break; 7910 } 7911 if (action != PF_DROP) { 7912 pd.m = NULL; 7913 action = PF_PASS; 7914 } 7915 break; 7916 #endif /* INET6 */ 7917 case PF_DROP: 7918 m_freem(pd.m); 7919 pd.m = NULL; 7920 break; 7921 default: 7922 if (st && st->rt) { 7923 switch (pd.af) { 7924 case AF_INET: 7925 pf_route(&pd, st); 7926 break; 7927 #ifdef INET6 7928 case AF_INET6: 7929 pf_route6(&pd, st); 7930 break; 7931 #endif /* INET6 */ 7932 } 7933 } 7934 break; 7935 } 7936 7937 #ifdef INET6 7938 /* if reassembled packet passed, create new fragments */ 7939 if (pf_status.reass && action == PF_PASS && pd.m && fwdir == PF_FWD && 7940 pd.af == AF_INET6) { 7941 struct m_tag *mtag; 7942 7943 if ((mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL))) 7944 action = pf_refragment6(&pd.m, mtag, NULL, NULL, NULL); 7945 } 7946 #endif /* INET6 */ 7947 if (st && action != PF_DROP) { 7948 if (!st->if_index_in && dir == PF_IN) 7949 st->if_index_in = ifp->if_index; 7950 else if (!st->if_index_out && dir == PF_OUT) 7951 st->if_index_out = ifp->if_index; 7952 } 7953 7954 *m0 = pd.m; 7955 7956 pf_state_unref(st); 7957 7958 return (action); 7959 } 7960 7961 int 7962 pf_ouraddr(struct mbuf *m) 7963 { 7964 struct pf_state_key *sk; 7965 7966 if (m->m_pkthdr.pf.flags & PF_TAG_DIVERTED) 7967 return (1); 7968 7969 sk = m->m_pkthdr.pf.statekey; 7970 if (sk != NULL) { 7971 if (sk->sk_inp != NULL) 7972 return (1); 7973 } 7974 7975 return (-1); 7976 } 7977 7978 /* 7979 * must be called whenever any addressing information such as 7980 * address, port, protocol has changed 7981 */ 7982 void 7983 pf_pkt_addr_changed(struct mbuf *m) 7984 { 7985 pf_mbuf_unlink_state_key(m); 7986 pf_mbuf_unlink_inpcb(m); 7987 } 7988 7989 struct inpcb * 7990 pf_inp_lookup(struct mbuf *m) 7991 { 7992 struct inpcb *inp = NULL; 7993 struct pf_state_key *sk = m->m_pkthdr.pf.statekey; 7994 7995 if (!pf_state_key_isvalid(sk)) 7996 pf_mbuf_unlink_state_key(m); 7997 else 7998 inp = m->m_pkthdr.pf.statekey->sk_inp; 7999 8000 if (inp && inp->inp_pf_sk) 8001 KASSERT(m->m_pkthdr.pf.statekey == inp->inp_pf_sk); 8002 8003 in_pcbref(inp); 8004 return (inp); 8005 } 8006 8007 void 8008 pf_inp_link(struct mbuf *m, struct inpcb *inp) 8009 { 8010 struct pf_state_key *sk = m->m_pkthdr.pf.statekey; 8011 8012 if (!pf_state_key_isvalid(sk)) { 8013 pf_mbuf_unlink_state_key(m); 8014 return; 8015 } 8016 8017 /* 8018 * we don't need to grab PF-lock here. At worst case we link inp to 8019 * state, which might be just being marked as deleted by another 8020 * thread. 8021 */ 8022 if (inp && !sk->sk_inp && !inp->inp_pf_sk) 8023 pf_state_key_link_inpcb(sk, inp); 8024 8025 /* The statekey has finished finding the inp, it is no longer needed. */ 8026 pf_mbuf_unlink_state_key(m); 8027 } 8028 8029 void 8030 pf_inp_unlink(struct inpcb *inp) 8031 { 8032 pf_inpcb_unlink_state_key(inp); 8033 } 8034 8035 void 8036 pf_state_key_link_reverse(struct pf_state_key *sk, struct pf_state_key *skrev) 8037 { 8038 struct pf_state_key *old_reverse; 8039 8040 old_reverse = atomic_cas_ptr(&sk->sk_reverse, NULL, skrev); 8041 if (old_reverse != NULL) 8042 KASSERT(old_reverse == skrev); 8043 else { 8044 pf_state_key_ref(skrev); 8045 8046 /* 8047 * NOTE: if sk == skrev, then KASSERT() below holds true, we 8048 * still want to grab a reference in such case, because 8049 * pf_state_key_unlink_reverse() does not check whether keys 8050 * are identical or not. 8051 */ 8052 old_reverse = atomic_cas_ptr(&skrev->sk_reverse, NULL, sk); 8053 if (old_reverse != NULL) 8054 KASSERT(old_reverse == sk); 8055 8056 pf_state_key_ref(sk); 8057 } 8058 } 8059 8060 #if NPFLOG > 0 8061 void 8062 pf_log_matches(struct pf_pdesc *pd, struct pf_rule *rm, struct pf_rule *am, 8063 struct pf_ruleset *ruleset, struct pf_rule_slist *matchrules) 8064 { 8065 struct pf_rule_item *ri; 8066 8067 /* if this is the log(matches) rule, packet has been logged already */ 8068 if (rm->log & PF_LOG_MATCHES) 8069 return; 8070 8071 SLIST_FOREACH(ri, matchrules, entry) 8072 if (ri->r->log & PF_LOG_MATCHES) 8073 pflog_packet(pd, PFRES_MATCH, rm, am, ruleset, ri->r); 8074 } 8075 #endif /* NPFLOG > 0 */ 8076 8077 struct pf_state_key * 8078 pf_state_key_ref(struct pf_state_key *sk) 8079 { 8080 if (sk != NULL) 8081 PF_REF_TAKE(sk->sk_refcnt); 8082 8083 return (sk); 8084 } 8085 8086 void 8087 pf_state_key_unref(struct pf_state_key *sk) 8088 { 8089 if (PF_REF_RELE(sk->sk_refcnt)) { 8090 /* state key must be removed from tree */ 8091 KASSERT(!pf_state_key_isvalid(sk)); 8092 /* state key must be unlinked from reverse key */ 8093 KASSERT(sk->sk_reverse == NULL); 8094 /* state key must be unlinked from socket */ 8095 KASSERT(sk->sk_inp == NULL); 8096 pool_put(&pf_state_key_pl, sk); 8097 } 8098 } 8099 8100 int 8101 pf_state_key_isvalid(struct pf_state_key *sk) 8102 { 8103 return ((sk != NULL) && (sk->sk_removed == 0)); 8104 } 8105 8106 void 8107 pf_mbuf_link_state_key(struct mbuf *m, struct pf_state_key *sk) 8108 { 8109 KASSERT(m->m_pkthdr.pf.statekey == NULL); 8110 m->m_pkthdr.pf.statekey = pf_state_key_ref(sk); 8111 } 8112 8113 void 8114 pf_mbuf_unlink_state_key(struct mbuf *m) 8115 { 8116 struct pf_state_key *sk = m->m_pkthdr.pf.statekey; 8117 8118 if (sk != NULL) { 8119 m->m_pkthdr.pf.statekey = NULL; 8120 pf_state_key_unref(sk); 8121 } 8122 } 8123 8124 void 8125 pf_mbuf_link_inpcb(struct mbuf *m, struct inpcb *inp) 8126 { 8127 KASSERT(m->m_pkthdr.pf.inp == NULL); 8128 m->m_pkthdr.pf.inp = in_pcbref(inp); 8129 } 8130 8131 void 8132 pf_mbuf_unlink_inpcb(struct mbuf *m) 8133 { 8134 struct inpcb *inp = m->m_pkthdr.pf.inp; 8135 8136 if (inp != NULL) { 8137 m->m_pkthdr.pf.inp = NULL; 8138 in_pcbunref(inp); 8139 } 8140 } 8141 8142 void 8143 pf_state_key_link_inpcb(struct pf_state_key *sk, struct inpcb *inp) 8144 { 8145 KASSERT(sk->sk_inp == NULL); 8146 sk->sk_inp = in_pcbref(inp); 8147 KASSERT(inp->inp_pf_sk == NULL); 8148 inp->inp_pf_sk = pf_state_key_ref(sk); 8149 } 8150 8151 void 8152 pf_inpcb_unlink_state_key(struct inpcb *inp) 8153 { 8154 struct pf_state_key *sk = inp->inp_pf_sk; 8155 8156 if (sk != NULL) { 8157 KASSERT(sk->sk_inp == inp); 8158 sk->sk_inp = NULL; 8159 inp->inp_pf_sk = NULL; 8160 pf_state_key_unref(sk); 8161 in_pcbunref(inp); 8162 } 8163 } 8164 8165 void 8166 pf_state_key_unlink_inpcb(struct pf_state_key *sk) 8167 { 8168 struct inpcb *inp = sk->sk_inp; 8169 8170 if (inp != NULL) { 8171 KASSERT(inp->inp_pf_sk == sk); 8172 sk->sk_inp = NULL; 8173 inp->inp_pf_sk = NULL; 8174 pf_state_key_unref(sk); 8175 in_pcbunref(inp); 8176 } 8177 } 8178 8179 void 8180 pf_state_key_unlink_reverse(struct pf_state_key *sk) 8181 { 8182 struct pf_state_key *skrev = sk->sk_reverse; 8183 8184 /* Note that sk and skrev may be equal, then we unref twice. */ 8185 if (skrev != NULL) { 8186 KASSERT(skrev->sk_reverse == sk); 8187 sk->sk_reverse = NULL; 8188 skrev->sk_reverse = NULL; 8189 pf_state_key_unref(skrev); 8190 pf_state_key_unref(sk); 8191 } 8192 } 8193 8194 struct pf_state * 8195 pf_state_ref(struct pf_state *st) 8196 { 8197 if (st != NULL) 8198 PF_REF_TAKE(st->refcnt); 8199 return (st); 8200 } 8201 8202 void 8203 pf_state_unref(struct pf_state *st) 8204 { 8205 if ((st != NULL) && PF_REF_RELE(st->refcnt)) { 8206 /* never inserted or removed */ 8207 #if NPFSYNC > 0 8208 KASSERT((TAILQ_NEXT(st, sync_list) == NULL) || 8209 ((TAILQ_NEXT(st, sync_list) == _Q_INVALID) && 8210 (st->sync_state == PFSYNC_S_NONE))); 8211 #endif /* NPFSYNC */ 8212 KASSERT((TAILQ_NEXT(st, entry_list) == NULL) || 8213 (TAILQ_NEXT(st, entry_list) == _Q_INVALID)); 8214 8215 pf_state_key_unref(st->key[PF_SK_WIRE]); 8216 pf_state_key_unref(st->key[PF_SK_STACK]); 8217 8218 pool_put(&pf_state_pl, st); 8219 } 8220 } 8221 8222 int 8223 pf_delay_pkt(struct mbuf *m, u_int ifidx) 8224 { 8225 struct pf_pktdelay *pdy; 8226 8227 if ((pdy = pool_get(&pf_pktdelay_pl, PR_NOWAIT)) == NULL) { 8228 m_freem(m); 8229 return (ENOBUFS); 8230 } 8231 pdy->ifidx = ifidx; 8232 pdy->m = m; 8233 timeout_set(&pdy->to, pf_pktenqueue_delayed, pdy); 8234 timeout_add_msec(&pdy->to, m->m_pkthdr.pf.delay); 8235 m->m_pkthdr.pf.delay = 0; 8236 return (0); 8237 } 8238 8239 void 8240 pf_pktenqueue_delayed(void *arg) 8241 { 8242 struct pf_pktdelay *pdy = arg; 8243 struct ifnet *ifp; 8244 8245 ifp = if_get(pdy->ifidx); 8246 if (ifp != NULL) { 8247 if_enqueue(ifp, pdy->m); 8248 if_put(ifp); 8249 } else 8250 m_freem(pdy->m); 8251 8252 pool_put(&pf_pktdelay_pl, pdy); 8253 } 8254