1 /* $OpenBSD: pf.c,v 1.1169 2023/01/06 17:44:34 sashan 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 if (sp->creatorid == 0) { 1374 DPFPRINTF(LOG_NOTICE, "%s: invalid creator id: %08x", __func__, 1375 ntohl(sp->creatorid)); 1376 return (EINVAL); 1377 } 1378 1379 if ((kif = pfi_kif_get(sp->ifname, NULL)) == NULL) { 1380 DPFPRINTF(LOG_NOTICE, "%s: unknown interface: %s", __func__, 1381 sp->ifname); 1382 if (flags & PFSYNC_SI_IOCTL) 1383 return (EINVAL); 1384 return (0); /* skip this state */ 1385 } 1386 1387 if (sp->af == 0) 1388 return (0); /* skip this state */ 1389 1390 /* 1391 * If the ruleset checksums match or the state is coming from the ioctl, 1392 * it's safe to associate the state with the rule of that number. 1393 */ 1394 if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && 1395 (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && 1396 ntohl(sp->rule) < pf_main_ruleset.rules.active.rcount) { 1397 TAILQ_FOREACH(r, pf_main_ruleset.rules.active.ptr, entries) 1398 if (ntohl(sp->rule) == n++) 1399 break; 1400 } else 1401 r = &pf_default_rule; 1402 1403 if ((r->max_states && r->states_cur >= r->max_states)) 1404 goto cleanup; 1405 1406 if (flags & PFSYNC_SI_IOCTL) 1407 pool_flags = PR_WAITOK | PR_LIMITFAIL | PR_ZERO; 1408 else 1409 pool_flags = PR_NOWAIT | PR_LIMITFAIL | PR_ZERO; 1410 1411 if ((st = pool_get(&pf_state_pl, pool_flags)) == NULL) 1412 goto cleanup; 1413 1414 if ((skw = pf_alloc_state_key(pool_flags)) == NULL) 1415 goto cleanup; 1416 1417 if ((sp->key[PF_SK_WIRE].af && 1418 (sp->key[PF_SK_WIRE].af != sp->key[PF_SK_STACK].af)) || 1419 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[0], 1420 &sp->key[PF_SK_STACK].addr[0], sp->af) || 1421 PF_ANEQ(&sp->key[PF_SK_WIRE].addr[1], 1422 &sp->key[PF_SK_STACK].addr[1], sp->af) || 1423 sp->key[PF_SK_WIRE].port[0] != sp->key[PF_SK_STACK].port[0] || 1424 sp->key[PF_SK_WIRE].port[1] != sp->key[PF_SK_STACK].port[1] || 1425 sp->key[PF_SK_WIRE].rdomain != sp->key[PF_SK_STACK].rdomain) { 1426 if ((sks = pf_alloc_state_key(pool_flags)) == NULL) 1427 goto cleanup; 1428 } else 1429 sks = pf_state_key_ref(skw); 1430 1431 /* allocate memory for scrub info */ 1432 if (pf_state_alloc_scrub_memory(&sp->src, &st->src) || 1433 pf_state_alloc_scrub_memory(&sp->dst, &st->dst)) 1434 goto cleanup; 1435 1436 /* copy to state key(s) */ 1437 skw->addr[0] = sp->key[PF_SK_WIRE].addr[0]; 1438 skw->addr[1] = sp->key[PF_SK_WIRE].addr[1]; 1439 skw->port[0] = sp->key[PF_SK_WIRE].port[0]; 1440 skw->port[1] = sp->key[PF_SK_WIRE].port[1]; 1441 skw->rdomain = ntohs(sp->key[PF_SK_WIRE].rdomain); 1442 skw->proto = sp->proto; 1443 if (!(skw->af = sp->key[PF_SK_WIRE].af)) 1444 skw->af = sp->af; 1445 skw->hash = pf_pkt_hash(skw->af, skw->proto, 1446 &skw->addr[0], &skw->addr[1], skw->port[0], skw->port[1]); 1447 1448 if (sks != skw) { 1449 sks->addr[0] = sp->key[PF_SK_STACK].addr[0]; 1450 sks->addr[1] = sp->key[PF_SK_STACK].addr[1]; 1451 sks->port[0] = sp->key[PF_SK_STACK].port[0]; 1452 sks->port[1] = sp->key[PF_SK_STACK].port[1]; 1453 sks->rdomain = ntohs(sp->key[PF_SK_STACK].rdomain); 1454 if (!(sks->af = sp->key[PF_SK_STACK].af)) 1455 sks->af = sp->af; 1456 if (sks->af != skw->af) { 1457 switch (sp->proto) { 1458 case IPPROTO_ICMP: 1459 sks->proto = IPPROTO_ICMPV6; 1460 break; 1461 case IPPROTO_ICMPV6: 1462 sks->proto = IPPROTO_ICMP; 1463 break; 1464 default: 1465 sks->proto = sp->proto; 1466 } 1467 } else 1468 sks->proto = sp->proto; 1469 1470 if (((sks->af != AF_INET) && (sks->af != AF_INET6)) || 1471 ((skw->af != AF_INET) && (skw->af != AF_INET6))) { 1472 error = EINVAL; 1473 goto cleanup; 1474 } 1475 1476 sks->hash = pf_pkt_hash(sks->af, sks->proto, 1477 &sks->addr[0], &sks->addr[1], sks->port[0], sks->port[1]); 1478 1479 } else if ((sks->af != AF_INET) && (sks->af != AF_INET6)) { 1480 error = EINVAL; 1481 goto cleanup; 1482 } 1483 st->rtableid[PF_SK_WIRE] = ntohl(sp->rtableid[PF_SK_WIRE]); 1484 st->rtableid[PF_SK_STACK] = ntohl(sp->rtableid[PF_SK_STACK]); 1485 1486 /* copy to state */ 1487 st->rt_addr = sp->rt_addr; 1488 st->rt = sp->rt; 1489 st->creation = getuptime() - ntohl(sp->creation); 1490 st->expire = getuptime(); 1491 if (ntohl(sp->expire)) { 1492 u_int32_t timeout; 1493 1494 timeout = r->timeout[sp->timeout]; 1495 if (!timeout) 1496 timeout = pf_default_rule.timeout[sp->timeout]; 1497 1498 /* sp->expire may have been adaptively scaled by export. */ 1499 st->expire -= timeout - ntohl(sp->expire); 1500 } 1501 1502 st->direction = sp->direction; 1503 st->log = sp->log; 1504 st->timeout = sp->timeout; 1505 st->state_flags = ntohs(sp->state_flags); 1506 st->max_mss = ntohs(sp->max_mss); 1507 st->min_ttl = sp->min_ttl; 1508 st->set_tos = sp->set_tos; 1509 st->set_prio[0] = sp->set_prio[0]; 1510 st->set_prio[1] = sp->set_prio[1]; 1511 1512 st->id = sp->id; 1513 st->creatorid = sp->creatorid; 1514 pf_state_peer_ntoh(&sp->src, &st->src); 1515 pf_state_peer_ntoh(&sp->dst, &st->dst); 1516 1517 st->rule.ptr = r; 1518 st->anchor.ptr = NULL; 1519 1520 st->pfsync_time = getuptime(); 1521 st->sync_state = PFSYNC_S_NONE; 1522 1523 PF_REF_INIT(st->refcnt); 1524 mtx_init(&st->mtx, IPL_NET); 1525 1526 /* XXX when we have anchors, use STATE_INC_COUNTERS */ 1527 r->states_cur++; 1528 r->states_tot++; 1529 1530 #if NPFSYNC > 0 1531 if (!ISSET(flags, PFSYNC_SI_IOCTL)) 1532 SET(st->state_flags, PFSTATE_NOSYNC); 1533 #endif 1534 1535 /* 1536 * We just set PFSTATE_NOSYNC bit, which prevents 1537 * pfsync_insert_state() to insert state to pfsync. 1538 */ 1539 if (pf_state_insert(kif, &skw, &sks, st) != 0) { 1540 /* XXX when we have anchors, use STATE_DEC_COUNTERS */ 1541 r->states_cur--; 1542 error = EEXIST; 1543 goto cleanup_state; 1544 } 1545 1546 #if NPFSYNC > 0 1547 if (!ISSET(flags, PFSYNC_SI_IOCTL)) { 1548 CLR(st->state_flags, PFSTATE_NOSYNC); 1549 if (ISSET(st->state_flags, PFSTATE_ACK)) 1550 pfsync_iack(st); 1551 } 1552 CLR(st->state_flags, PFSTATE_ACK); 1553 #endif 1554 1555 return (0); 1556 1557 cleanup: 1558 if (skw != NULL) 1559 pf_state_key_unref(skw); 1560 if (sks != NULL) 1561 pf_state_key_unref(sks); 1562 1563 cleanup_state: /* pf_state_insert frees the state keys */ 1564 if (st) { 1565 if (st->dst.scrub) 1566 pool_put(&pf_state_scrub_pl, st->dst.scrub); 1567 if (st->src.scrub) 1568 pool_put(&pf_state_scrub_pl, st->src.scrub); 1569 pool_put(&pf_state_pl, st); 1570 } 1571 return (error); 1572 } 1573 #endif /* NPFSYNC > 0 */ 1574 1575 /* END state table stuff */ 1576 1577 void 1578 pf_purge_timeout(void *unused) 1579 { 1580 /* XXX move to systqmp to avoid KERNEL_LOCK */ 1581 task_add(systq, &pf_purge_task); 1582 } 1583 1584 void 1585 pf_purge(void *xnloops) 1586 { 1587 int *nloops = xnloops; 1588 1589 /* 1590 * process a fraction of the state table every second 1591 * Note: 1592 * we no longer need PF_LOCK() here, because 1593 * pf_purge_expired_states() uses pf_state_lock to maintain 1594 * consistency. 1595 */ 1596 if (pf_default_rule.timeout[PFTM_INTERVAL] > 0) 1597 pf_purge_expired_states(1 + (pf_status.states 1598 / pf_default_rule.timeout[PFTM_INTERVAL])); 1599 1600 NET_LOCK(); 1601 1602 PF_LOCK(); 1603 /* purge other expired types every PFTM_INTERVAL seconds */ 1604 if (++(*nloops) >= pf_default_rule.timeout[PFTM_INTERVAL]) 1605 pf_purge_expired_src_nodes(); 1606 PF_UNLOCK(); 1607 1608 /* 1609 * Fragments don't require PF_LOCK(), they use their own lock. 1610 */ 1611 if ((*nloops) >= pf_default_rule.timeout[PFTM_INTERVAL]) { 1612 pf_purge_expired_fragments(); 1613 *nloops = 0; 1614 } 1615 NET_UNLOCK(); 1616 1617 timeout_add_sec(&pf_purge_to, 1); 1618 } 1619 1620 int32_t 1621 pf_state_expires(const struct pf_state *st, uint8_t stimeout) 1622 { 1623 u_int32_t timeout; 1624 u_int32_t start; 1625 u_int32_t end; 1626 u_int32_t states; 1627 1628 /* 1629 * pf_state_expires is used by the state purge task to 1630 * decide if a state is a candidate for cleanup, and by the 1631 * pfsync state export code to populate an expiry time. 1632 * 1633 * this function may be called by the state purge task while 1634 * the state is being modified. avoid inconsistent reads of 1635 * state->timeout by having the caller do the read (and any 1636 * checks it needs to do on the same variable) and then pass 1637 * their view of the timeout in here for this function to use. 1638 * the only consequence of using a stale timeout value is 1639 * that the state won't be a candidate for purging until the 1640 * next pass of the purge task. 1641 */ 1642 1643 /* handle all PFTM_* > PFTM_MAX here */ 1644 if (stimeout > PFTM_MAX) 1645 return (0); 1646 1647 KASSERT(stimeout < PFTM_MAX); 1648 1649 timeout = st->rule.ptr->timeout[stimeout]; 1650 if (!timeout) 1651 timeout = pf_default_rule.timeout[stimeout]; 1652 1653 start = st->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1654 if (start) { 1655 end = st->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1656 states = st->rule.ptr->states_cur; 1657 } else { 1658 start = pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1659 end = pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1660 states = pf_status.states; 1661 } 1662 if (end && states > start && start < end) { 1663 if (states >= end) 1664 return (0); 1665 1666 timeout = (u_int64_t)timeout * (end - states) / (end - start); 1667 } 1668 1669 return (st->expire + timeout); 1670 } 1671 1672 void 1673 pf_purge_expired_src_nodes(void) 1674 { 1675 struct pf_src_node *cur, *next; 1676 1677 PF_ASSERT_LOCKED(); 1678 1679 for (cur = RB_MIN(pf_src_tree, &tree_src_tracking); cur; cur = next) { 1680 next = RB_NEXT(pf_src_tree, &tree_src_tracking, cur); 1681 1682 if (cur->states == 0 && cur->expire <= getuptime()) { 1683 next = RB_NEXT(pf_src_tree, &tree_src_tracking, cur); 1684 pf_remove_src_node(cur); 1685 } 1686 } 1687 } 1688 1689 void 1690 pf_src_tree_remove_state(struct pf_state *st) 1691 { 1692 u_int32_t timeout; 1693 struct pf_sn_item *sni; 1694 1695 while ((sni = SLIST_FIRST(&st->src_nodes)) != NULL) { 1696 SLIST_REMOVE_HEAD(&st->src_nodes, next); 1697 if (st->src.tcp_est) 1698 --sni->sn->conn; 1699 if (--sni->sn->states == 0) { 1700 timeout = st->rule.ptr->timeout[PFTM_SRC_NODE]; 1701 if (!timeout) 1702 timeout = 1703 pf_default_rule.timeout[PFTM_SRC_NODE]; 1704 sni->sn->expire = getuptime() + timeout; 1705 } 1706 pool_put(&pf_sn_item_pl, sni); 1707 } 1708 } 1709 1710 void 1711 pf_remove_state(struct pf_state *st) 1712 { 1713 PF_ASSERT_LOCKED(); 1714 1715 /* handle load balancing related tasks */ 1716 pf_postprocess_addr(st); 1717 1718 if (st->src.state == PF_TCPS_PROXY_DST) { 1719 pf_send_tcp(st->rule.ptr, st->key[PF_SK_WIRE]->af, 1720 &st->key[PF_SK_WIRE]->addr[1], 1721 &st->key[PF_SK_WIRE]->addr[0], 1722 st->key[PF_SK_WIRE]->port[1], 1723 st->key[PF_SK_WIRE]->port[0], 1724 st->src.seqhi, st->src.seqlo + 1, 1725 TH_RST|TH_ACK, 0, 0, 0, 1, st->tag, 1726 st->key[PF_SK_WIRE]->rdomain); 1727 } 1728 if (st->key[PF_SK_STACK]->proto == IPPROTO_TCP) 1729 pf_set_protostate(st, PF_PEER_BOTH, TCPS_CLOSED); 1730 1731 RBT_REMOVE(pf_state_tree_id, &tree_id, st); 1732 #if NPFLOW > 0 1733 if (st->state_flags & PFSTATE_PFLOW) 1734 export_pflow(st); 1735 #endif /* NPFLOW > 0 */ 1736 #if NPFSYNC > 0 1737 pfsync_delete_state(st); 1738 #endif /* NPFSYNC > 0 */ 1739 st->timeout = PFTM_UNLINKED; 1740 pf_src_tree_remove_state(st); 1741 pf_detach_state(st); 1742 } 1743 1744 void 1745 pf_remove_divert_state(struct pf_state_key *sk) 1746 { 1747 struct pf_state_item *si; 1748 1749 PF_ASSERT_UNLOCKED(); 1750 1751 PF_LOCK(); 1752 PF_STATE_ENTER_WRITE(); 1753 TAILQ_FOREACH(si, &sk->sk_states, si_entry) { 1754 struct pf_state *sist = si->si_st; 1755 if (sk == sist->key[PF_SK_STACK] && sist->rule.ptr && 1756 (sist->rule.ptr->divert.type == PF_DIVERT_TO || 1757 sist->rule.ptr->divert.type == PF_DIVERT_REPLY)) { 1758 if (sist->key[PF_SK_STACK]->proto == IPPROTO_TCP && 1759 sist->key[PF_SK_WIRE] != sist->key[PF_SK_STACK]) { 1760 /* 1761 * If the local address is translated, keep 1762 * the state for "tcp.closed" seconds to 1763 * prevent its source port from being reused. 1764 */ 1765 if (sist->src.state < TCPS_FIN_WAIT_2 || 1766 sist->dst.state < TCPS_FIN_WAIT_2) { 1767 pf_set_protostate(sist, PF_PEER_BOTH, 1768 TCPS_TIME_WAIT); 1769 sist->timeout = PFTM_TCP_CLOSED; 1770 sist->expire = getuptime(); 1771 } 1772 sist->state_flags |= PFSTATE_INP_UNLINKED; 1773 } else 1774 pf_remove_state(sist); 1775 break; 1776 } 1777 } 1778 PF_STATE_EXIT_WRITE(); 1779 PF_UNLOCK(); 1780 } 1781 1782 void 1783 pf_free_state(struct pf_state *st) 1784 { 1785 struct pf_rule_item *ri; 1786 1787 PF_ASSERT_LOCKED(); 1788 1789 #if NPFSYNC > 0 1790 if (pfsync_state_in_use(st)) 1791 return; 1792 #endif /* NPFSYNC > 0 */ 1793 KASSERT(st->timeout == PFTM_UNLINKED); 1794 if (--st->rule.ptr->states_cur == 0 && 1795 st->rule.ptr->src_nodes == 0) 1796 pf_rm_rule(NULL, st->rule.ptr); 1797 if (st->anchor.ptr != NULL) 1798 if (--st->anchor.ptr->states_cur == 0) 1799 pf_rm_rule(NULL, st->anchor.ptr); 1800 while ((ri = SLIST_FIRST(&st->match_rules))) { 1801 SLIST_REMOVE_HEAD(&st->match_rules, entry); 1802 if (--ri->r->states_cur == 0 && 1803 ri->r->src_nodes == 0) 1804 pf_rm_rule(NULL, ri->r); 1805 pool_put(&pf_rule_item_pl, ri); 1806 } 1807 pf_normalize_tcp_cleanup(st); 1808 pfi_kif_unref(st->kif, PFI_KIF_REF_STATE); 1809 pf_state_list_remove(&pf_state_list, st); 1810 if (st->tag) 1811 pf_tag_unref(st->tag); 1812 pf_state_unref(st); 1813 pf_status.fcounters[FCNT_STATE_REMOVALS]++; 1814 pf_status.states--; 1815 } 1816 1817 void 1818 pf_purge_expired_states(u_int32_t maxcheck) 1819 { 1820 /* 1821 * this task/thread/context/whatever is the only thing that 1822 * removes states from the pf_state_list, so the cur reference 1823 * it holds between calls is guaranteed to still be in the 1824 * list. 1825 */ 1826 static struct pf_state *cur = NULL; 1827 1828 struct pf_state *head, *tail; 1829 struct pf_state *st; 1830 SLIST_HEAD(pf_state_gcl, pf_state) gcl = SLIST_HEAD_INITIALIZER(gcl); 1831 time_t now; 1832 1833 PF_ASSERT_UNLOCKED(); 1834 1835 rw_enter_read(&pf_state_list.pfs_rwl); 1836 1837 mtx_enter(&pf_state_list.pfs_mtx); 1838 head = TAILQ_FIRST(&pf_state_list.pfs_list); 1839 tail = TAILQ_LAST(&pf_state_list.pfs_list, pf_state_queue); 1840 mtx_leave(&pf_state_list.pfs_mtx); 1841 1842 if (head == NULL) { 1843 /* the list is empty */ 1844 rw_exit_read(&pf_state_list.pfs_rwl); 1845 return; 1846 } 1847 1848 /* (re)start at the front of the list */ 1849 if (cur == NULL) 1850 cur = head; 1851 1852 now = getuptime(); 1853 1854 do { 1855 uint8_t stimeout = cur->timeout; 1856 1857 if ((stimeout == PFTM_UNLINKED) || 1858 (pf_state_expires(cur, stimeout) <= now)) { 1859 st = pf_state_ref(cur); 1860 SLIST_INSERT_HEAD(&gcl, st, gc_list); 1861 } 1862 1863 /* don't iterate past the end of our view of the list */ 1864 if (cur == tail) { 1865 cur = NULL; 1866 break; 1867 } 1868 1869 cur = TAILQ_NEXT(cur, entry_list); 1870 } while (maxcheck--); 1871 1872 rw_exit_read(&pf_state_list.pfs_rwl); 1873 1874 if (SLIST_EMPTY(&gcl)) 1875 return; 1876 1877 NET_LOCK(); 1878 rw_enter_write(&pf_state_list.pfs_rwl); 1879 PF_LOCK(); 1880 PF_STATE_ENTER_WRITE(); 1881 SLIST_FOREACH(st, &gcl, gc_list) { 1882 if (st->timeout != PFTM_UNLINKED) 1883 pf_remove_state(st); 1884 1885 pf_free_state(st); 1886 } 1887 PF_STATE_EXIT_WRITE(); 1888 PF_UNLOCK(); 1889 rw_exit_write(&pf_state_list.pfs_rwl); 1890 NET_UNLOCK(); 1891 1892 while ((st = SLIST_FIRST(&gcl)) != NULL) { 1893 SLIST_REMOVE_HEAD(&gcl, gc_list); 1894 pf_state_unref(st); 1895 } 1896 } 1897 1898 int 1899 pf_tbladdr_setup(struct pf_ruleset *rs, struct pf_addr_wrap *aw, int wait) 1900 { 1901 if (aw->type != PF_ADDR_TABLE) 1902 return (0); 1903 if ((aw->p.tbl = pfr_attach_table(rs, aw->v.tblname, wait)) == NULL) 1904 return (1); 1905 return (0); 1906 } 1907 1908 void 1909 pf_tbladdr_remove(struct pf_addr_wrap *aw) 1910 { 1911 if (aw->type != PF_ADDR_TABLE || aw->p.tbl == NULL) 1912 return; 1913 pfr_detach_table(aw->p.tbl); 1914 aw->p.tbl = NULL; 1915 } 1916 1917 void 1918 pf_tbladdr_copyout(struct pf_addr_wrap *aw) 1919 { 1920 struct pfr_ktable *kt = aw->p.tbl; 1921 1922 if (aw->type != PF_ADDR_TABLE || kt == NULL) 1923 return; 1924 if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) 1925 kt = kt->pfrkt_root; 1926 aw->p.tbl = NULL; 1927 aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ? 1928 kt->pfrkt_cnt : -1; 1929 } 1930 1931 void 1932 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 1933 { 1934 switch (af) { 1935 case AF_INET: { 1936 u_int32_t a = ntohl(addr->addr32[0]); 1937 addlog("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 1938 (a>>8)&255, a&255); 1939 if (p) { 1940 p = ntohs(p); 1941 addlog(":%u", p); 1942 } 1943 break; 1944 } 1945 #ifdef INET6 1946 case AF_INET6: { 1947 u_int16_t b; 1948 u_int8_t i, curstart, curend, maxstart, maxend; 1949 curstart = curend = maxstart = maxend = 255; 1950 for (i = 0; i < 8; i++) { 1951 if (!addr->addr16[i]) { 1952 if (curstart == 255) 1953 curstart = i; 1954 curend = i; 1955 } else { 1956 if ((curend - curstart) > 1957 (maxend - maxstart)) { 1958 maxstart = curstart; 1959 maxend = curend; 1960 } 1961 curstart = curend = 255; 1962 } 1963 } 1964 if ((curend - curstart) > 1965 (maxend - maxstart)) { 1966 maxstart = curstart; 1967 maxend = curend; 1968 } 1969 for (i = 0; i < 8; i++) { 1970 if (i >= maxstart && i <= maxend) { 1971 if (i == 0) 1972 addlog(":"); 1973 if (i == maxend) 1974 addlog(":"); 1975 } else { 1976 b = ntohs(addr->addr16[i]); 1977 addlog("%x", b); 1978 if (i < 7) 1979 addlog(":"); 1980 } 1981 } 1982 if (p) { 1983 p = ntohs(p); 1984 addlog("[%u]", p); 1985 } 1986 break; 1987 } 1988 #endif /* INET6 */ 1989 } 1990 } 1991 1992 void 1993 pf_print_state(struct pf_state *st) 1994 { 1995 pf_print_state_parts(st, NULL, NULL); 1996 } 1997 1998 void 1999 pf_print_state_parts(struct pf_state *st, 2000 struct pf_state_key *skwp, struct pf_state_key *sksp) 2001 { 2002 struct pf_state_key *skw, *sks; 2003 u_int8_t proto, dir; 2004 2005 /* Do our best to fill these, but they're skipped if NULL */ 2006 skw = skwp ? skwp : (st ? st->key[PF_SK_WIRE] : NULL); 2007 sks = sksp ? sksp : (st ? st->key[PF_SK_STACK] : NULL); 2008 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2009 dir = st ? st->direction : 0; 2010 2011 switch (proto) { 2012 case IPPROTO_IPV4: 2013 addlog("IPv4"); 2014 break; 2015 case IPPROTO_IPV6: 2016 addlog("IPv6"); 2017 break; 2018 case IPPROTO_TCP: 2019 addlog("TCP"); 2020 break; 2021 case IPPROTO_UDP: 2022 addlog("UDP"); 2023 break; 2024 case IPPROTO_ICMP: 2025 addlog("ICMP"); 2026 break; 2027 case IPPROTO_ICMPV6: 2028 addlog("ICMPv6"); 2029 break; 2030 default: 2031 addlog("%u", proto); 2032 break; 2033 } 2034 switch (dir) { 2035 case PF_IN: 2036 addlog(" in"); 2037 break; 2038 case PF_OUT: 2039 addlog(" out"); 2040 break; 2041 } 2042 if (skw) { 2043 addlog(" wire: (%d) ", skw->rdomain); 2044 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2045 addlog(" "); 2046 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2047 } 2048 if (sks) { 2049 addlog(" stack: (%d) ", sks->rdomain); 2050 if (sks != skw) { 2051 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2052 addlog(" "); 2053 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2054 } else 2055 addlog("-"); 2056 } 2057 if (st) { 2058 if (proto == IPPROTO_TCP) { 2059 addlog(" [lo=%u high=%u win=%u modulator=%u", 2060 st->src.seqlo, st->src.seqhi, 2061 st->src.max_win, st->src.seqdiff); 2062 if (st->src.wscale && st->dst.wscale) 2063 addlog(" wscale=%u", 2064 st->src.wscale & PF_WSCALE_MASK); 2065 addlog("]"); 2066 addlog(" [lo=%u high=%u win=%u modulator=%u", 2067 st->dst.seqlo, st->dst.seqhi, 2068 st->dst.max_win, st->dst.seqdiff); 2069 if (st->src.wscale && st->dst.wscale) 2070 addlog(" wscale=%u", 2071 st->dst.wscale & PF_WSCALE_MASK); 2072 addlog("]"); 2073 } 2074 addlog(" %u:%u", st->src.state, st->dst.state); 2075 if (st->rule.ptr) 2076 addlog(" @%d", st->rule.ptr->nr); 2077 } 2078 } 2079 2080 void 2081 pf_print_flags(u_int8_t f) 2082 { 2083 if (f) 2084 addlog(" "); 2085 if (f & TH_FIN) 2086 addlog("F"); 2087 if (f & TH_SYN) 2088 addlog("S"); 2089 if (f & TH_RST) 2090 addlog("R"); 2091 if (f & TH_PUSH) 2092 addlog("P"); 2093 if (f & TH_ACK) 2094 addlog("A"); 2095 if (f & TH_URG) 2096 addlog("U"); 2097 if (f & TH_ECE) 2098 addlog("E"); 2099 if (f & TH_CWR) 2100 addlog("W"); 2101 } 2102 2103 #define PF_SET_SKIP_STEPS(i) \ 2104 do { \ 2105 while (head[i] != cur) { \ 2106 head[i]->skip[i].ptr = cur; \ 2107 head[i] = TAILQ_NEXT(head[i], entries); \ 2108 } \ 2109 } while (0) 2110 2111 void 2112 pf_calc_skip_steps(struct pf_rulequeue *rules) 2113 { 2114 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT]; 2115 int i; 2116 2117 cur = TAILQ_FIRST(rules); 2118 prev = cur; 2119 for (i = 0; i < PF_SKIP_COUNT; ++i) 2120 head[i] = cur; 2121 while (cur != NULL) { 2122 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2123 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2124 if (cur->direction != prev->direction) 2125 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2126 if (cur->onrdomain != prev->onrdomain || 2127 cur->ifnot != prev->ifnot) 2128 PF_SET_SKIP_STEPS(PF_SKIP_RDOM); 2129 if (cur->af != prev->af) 2130 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2131 if (cur->proto != prev->proto) 2132 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2133 if (cur->src.neg != prev->src.neg || 2134 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2135 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2136 if (cur->dst.neg != prev->dst.neg || 2137 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2138 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2139 if (cur->src.port[0] != prev->src.port[0] || 2140 cur->src.port[1] != prev->src.port[1] || 2141 cur->src.port_op != prev->src.port_op) 2142 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2143 if (cur->dst.port[0] != prev->dst.port[0] || 2144 cur->dst.port[1] != prev->dst.port[1] || 2145 cur->dst.port_op != prev->dst.port_op) 2146 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2147 2148 prev = cur; 2149 cur = TAILQ_NEXT(cur, entries); 2150 } 2151 for (i = 0; i < PF_SKIP_COUNT; ++i) 2152 PF_SET_SKIP_STEPS(i); 2153 } 2154 2155 int 2156 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2157 { 2158 if (aw1->type != aw2->type) 2159 return (1); 2160 switch (aw1->type) { 2161 case PF_ADDR_ADDRMASK: 2162 case PF_ADDR_RANGE: 2163 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2164 return (1); 2165 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2166 return (1); 2167 return (0); 2168 case PF_ADDR_DYNIFTL: 2169 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2170 case PF_ADDR_NONE: 2171 case PF_ADDR_NOROUTE: 2172 case PF_ADDR_URPFFAILED: 2173 return (0); 2174 case PF_ADDR_TABLE: 2175 return (aw1->p.tbl != aw2->p.tbl); 2176 case PF_ADDR_RTLABEL: 2177 return (aw1->v.rtlabel != aw2->v.rtlabel); 2178 default: 2179 addlog("invalid address type: %d\n", aw1->type); 2180 return (1); 2181 } 2182 } 2183 2184 /* This algorithm computes 'a + b - c' in ones-complement using a trick to 2185 * emulate at most one ones-complement subtraction. This thereby limits net 2186 * carries/borrows to at most one, eliminating a reduction step and saving one 2187 * each of +, >>, & and ~. 2188 * 2189 * def. x mod y = x - (x//y)*y for integer x,y 2190 * def. sum = x mod 2^16 2191 * def. accumulator = (x >> 16) mod 2^16 2192 * 2193 * The trick works as follows: subtracting exactly one u_int16_t from the 2194 * u_int32_t x incurs at most one underflow, wrapping its upper 16-bits, the 2195 * accumulator, to 2^16 - 1. Adding this to the 16-bit sum preserves the 2196 * ones-complement borrow: 2197 * 2198 * (sum + accumulator) mod 2^16 2199 * = { assume underflow: accumulator := 2^16 - 1 } 2200 * (sum + 2^16 - 1) mod 2^16 2201 * = { mod } 2202 * (sum - 1) mod 2^16 2203 * 2204 * Although this breaks for sum = 0, giving 0xffff, which is ones-complement's 2205 * other zero, not -1, that cannot occur: the 16-bit sum cannot be underflown 2206 * to zero as that requires subtraction of at least 2^16, which exceeds a 2207 * single u_int16_t's range. 2208 * 2209 * We use the following theorem to derive the implementation: 2210 * 2211 * th. (x + (y mod z)) mod z = (x + y) mod z (0) 2212 * proof. 2213 * (x + (y mod z)) mod z 2214 * = { def mod } 2215 * (x + y - (y//z)*z) mod z 2216 * = { (a + b*c) mod c = a mod c } 2217 * (x + y) mod z [end of proof] 2218 * 2219 * ... and thereby obtain: 2220 * 2221 * (sum + accumulator) mod 2^16 2222 * = { def. accumulator, def. sum } 2223 * (x mod 2^16 + (x >> 16) mod 2^16) mod 2^16 2224 * = { (0), twice } 2225 * (x + (x >> 16)) mod 2^16 2226 * = { x mod 2^n = x & (2^n - 1) } 2227 * (x + (x >> 16)) & 0xffff 2228 * 2229 * Note: this serves also as a reduction step for at most one add (as the 2230 * trailing mod 2^16 prevents further reductions by destroying carries). 2231 */ 2232 __inline void 2233 pf_cksum_fixup(u_int16_t *cksum, u_int16_t was, u_int16_t now, 2234 u_int8_t proto) 2235 { 2236 u_int32_t x; 2237 const int udp = proto == IPPROTO_UDP; 2238 2239 x = *cksum + was - now; 2240 x = (x + (x >> 16)) & 0xffff; 2241 2242 /* optimise: eliminate a branch when not udp */ 2243 if (udp && *cksum == 0x0000) 2244 return; 2245 if (udp && x == 0x0000) 2246 x = 0xffff; 2247 2248 *cksum = (u_int16_t)(x); 2249 } 2250 2251 #ifdef INET6 2252 /* pre: coverage(cksum) is superset of coverage(covered_cksum) */ 2253 static __inline void 2254 pf_cksum_uncover(u_int16_t *cksum, u_int16_t covered_cksum, u_int8_t proto) 2255 { 2256 pf_cksum_fixup(cksum, ~covered_cksum, 0x0, proto); 2257 } 2258 2259 /* pre: disjoint(coverage(cksum), coverage(uncovered_cksum)) */ 2260 static __inline void 2261 pf_cksum_cover(u_int16_t *cksum, u_int16_t uncovered_cksum, u_int8_t proto) 2262 { 2263 pf_cksum_fixup(cksum, 0x0, ~uncovered_cksum, proto); 2264 } 2265 #endif /* INET6 */ 2266 2267 /* pre: *a is 16-bit aligned within its packet 2268 * 2269 * This algorithm emulates 16-bit ones-complement sums on a twos-complement 2270 * machine by conserving ones-complement's otherwise discarded carries in the 2271 * upper bits of x. These accumulated carries when added to the lower 16-bits 2272 * over at least zero 'reduction' steps then complete the ones-complement sum. 2273 * 2274 * def. sum = x mod 2^16 2275 * def. accumulator = (x >> 16) 2276 * 2277 * At most two reduction steps 2278 * 2279 * x := sum + accumulator 2280 * = { def sum, def accumulator } 2281 * x := x mod 2^16 + (x >> 16) 2282 * = { x mod 2^n = x & (2^n - 1) } 2283 * x := (x & 0xffff) + (x >> 16) 2284 * 2285 * are necessary to incorporate the accumulated carries (at most one per add) 2286 * i.e. to reduce x < 2^16 from at most 16 carries in the upper 16 bits. 2287 * 2288 * The function is also invariant over the endian of the host. Why? 2289 * 2290 * Define the unary transpose operator ~ on a bitstring in python slice 2291 * notation as lambda m: m[P:] + m[:P] , for some constant pivot P. 2292 * 2293 * th. ~ distributes over ones-complement addition, denoted by +_1, i.e. 2294 * 2295 * ~m +_1 ~n = ~(m +_1 n) (for all bitstrings m,n of equal length) 2296 * 2297 * proof. Regard the bitstrings in m +_1 n as split at P, forming at most two 2298 * 'half-adds'. Under ones-complement addition, each half-add carries to the 2299 * other, so the sum of each half-add is unaffected by their relative 2300 * order. Therefore: 2301 * 2302 * ~m +_1 ~n 2303 * = { half-adds invariant under transposition } 2304 * ~s 2305 * = { substitute } 2306 * ~(m +_1 n) [end of proof] 2307 * 2308 * th. Summing two in-memory ones-complement 16-bit variables m,n on a machine 2309 * with the converse endian does not alter the result. 2310 * 2311 * proof. 2312 * { converse machine endian: load/store transposes, P := 8 } 2313 * ~(~m +_1 ~n) 2314 * = { ~ over +_1 } 2315 * ~~m +_1 ~~n 2316 * = { ~ is an involution } 2317 * m +_1 n [end of proof] 2318 * 2319 */ 2320 #define NEG(x) ((u_int16_t)~(x)) 2321 void 2322 pf_cksum_fixup_a(u_int16_t *cksum, const struct pf_addr *a, 2323 const struct pf_addr *an, sa_family_t af, u_int8_t proto) 2324 { 2325 u_int32_t x; 2326 const u_int16_t *n = an->addr16; 2327 const u_int16_t *o = a->addr16; 2328 const int udp = proto == IPPROTO_UDP; 2329 2330 switch (af) { 2331 case AF_INET: 2332 x = *cksum + o[0] + NEG(n[0]) + o[1] + NEG(n[1]); 2333 break; 2334 #ifdef INET6 2335 case AF_INET6: 2336 x = *cksum + o[0] + NEG(n[0]) + o[1] + NEG(n[1]) +\ 2337 o[2] + NEG(n[2]) + o[3] + NEG(n[3]) +\ 2338 o[4] + NEG(n[4]) + o[5] + NEG(n[5]) +\ 2339 o[6] + NEG(n[6]) + o[7] + NEG(n[7]); 2340 break; 2341 #endif /* INET6 */ 2342 default: 2343 unhandled_af(af); 2344 } 2345 2346 x = (x & 0xffff) + (x >> 16); 2347 x = (x & 0xffff) + (x >> 16); 2348 2349 /* optimise: eliminate a branch when not udp */ 2350 if (udp && *cksum == 0x0000) 2351 return; 2352 if (udp && x == 0x0000) 2353 x = 0xffff; 2354 2355 *cksum = (u_int16_t)(x); 2356 } 2357 2358 int 2359 pf_patch_8(struct pf_pdesc *pd, u_int8_t *f, u_int8_t v, bool hi) 2360 { 2361 int rewrite = 0; 2362 2363 if (*f != v) { 2364 u_int16_t old = htons(hi ? (*f << 8) : *f); 2365 u_int16_t new = htons(hi ? ( v << 8) : v); 2366 2367 pf_cksum_fixup(pd->pcksum, old, new, pd->proto); 2368 *f = v; 2369 rewrite = 1; 2370 } 2371 2372 return (rewrite); 2373 } 2374 2375 /* pre: *f is 16-bit aligned within its packet */ 2376 int 2377 pf_patch_16(struct pf_pdesc *pd, u_int16_t *f, u_int16_t v) 2378 { 2379 int rewrite = 0; 2380 2381 if (*f != v) { 2382 pf_cksum_fixup(pd->pcksum, *f, v, pd->proto); 2383 *f = v; 2384 rewrite = 1; 2385 } 2386 2387 return (rewrite); 2388 } 2389 2390 int 2391 pf_patch_16_unaligned(struct pf_pdesc *pd, void *f, u_int16_t v, bool hi) 2392 { 2393 int rewrite = 0; 2394 u_int8_t *fb = (u_int8_t*)f; 2395 u_int8_t *vb = (u_int8_t*)&v; 2396 2397 if (hi && ALIGNED_POINTER(f, u_int16_t)) { 2398 return (pf_patch_16(pd, f, v)); /* optimise */ 2399 } 2400 2401 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 2402 rewrite += pf_patch_8(pd, fb++, *vb++,!hi); 2403 2404 return (rewrite); 2405 } 2406 2407 /* pre: *f is 16-bit aligned within its packet */ 2408 /* pre: pd->proto != IPPROTO_UDP */ 2409 int 2410 pf_patch_32(struct pf_pdesc *pd, u_int32_t *f, u_int32_t v) 2411 { 2412 int rewrite = 0; 2413 u_int16_t *pc = pd->pcksum; 2414 u_int8_t proto = pd->proto; 2415 2416 /* optimise: inline udp fixup code is unused; let compiler scrub it */ 2417 if (proto == IPPROTO_UDP) 2418 panic("%s: udp", __func__); 2419 2420 /* optimise: skip *f != v guard; true for all use-cases */ 2421 pf_cksum_fixup(pc, *f / (1 << 16), v / (1 << 16), proto); 2422 pf_cksum_fixup(pc, *f % (1 << 16), v % (1 << 16), proto); 2423 2424 *f = v; 2425 rewrite = 1; 2426 2427 return (rewrite); 2428 } 2429 2430 int 2431 pf_patch_32_unaligned(struct pf_pdesc *pd, void *f, u_int32_t v, bool hi) 2432 { 2433 int rewrite = 0; 2434 u_int8_t *fb = (u_int8_t*)f; 2435 u_int8_t *vb = (u_int8_t*)&v; 2436 2437 if (hi && ALIGNED_POINTER(f, u_int32_t)) { 2438 return (pf_patch_32(pd, f, v)); /* optimise */ 2439 } 2440 2441 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 2442 rewrite += pf_patch_8(pd, fb++, *vb++,!hi); 2443 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 2444 rewrite += pf_patch_8(pd, fb++, *vb++,!hi); 2445 2446 return (rewrite); 2447 } 2448 2449 int 2450 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, int *icmp_dir, 2451 u_int16_t *virtual_id, u_int16_t *virtual_type) 2452 { 2453 /* 2454 * ICMP types marked with PF_OUT are typically responses to 2455 * PF_IN, and will match states in the opposite direction. 2456 * PF_IN ICMP types need to match a state with that type. 2457 */ 2458 *icmp_dir = PF_OUT; 2459 2460 /* Queries (and responses) */ 2461 switch (pd->af) { 2462 case AF_INET: 2463 switch (type) { 2464 case ICMP_ECHO: 2465 *icmp_dir = PF_IN; 2466 /* FALLTHROUGH */ 2467 case ICMP_ECHOREPLY: 2468 *virtual_type = ICMP_ECHO; 2469 *virtual_id = pd->hdr.icmp.icmp_id; 2470 break; 2471 2472 case ICMP_TSTAMP: 2473 *icmp_dir = PF_IN; 2474 /* FALLTHROUGH */ 2475 case ICMP_TSTAMPREPLY: 2476 *virtual_type = ICMP_TSTAMP; 2477 *virtual_id = pd->hdr.icmp.icmp_id; 2478 break; 2479 2480 case ICMP_IREQ: 2481 *icmp_dir = PF_IN; 2482 /* FALLTHROUGH */ 2483 case ICMP_IREQREPLY: 2484 *virtual_type = ICMP_IREQ; 2485 *virtual_id = pd->hdr.icmp.icmp_id; 2486 break; 2487 2488 case ICMP_MASKREQ: 2489 *icmp_dir = PF_IN; 2490 /* FALLTHROUGH */ 2491 case ICMP_MASKREPLY: 2492 *virtual_type = ICMP_MASKREQ; 2493 *virtual_id = pd->hdr.icmp.icmp_id; 2494 break; 2495 2496 case ICMP_IPV6_WHEREAREYOU: 2497 *icmp_dir = PF_IN; 2498 /* FALLTHROUGH */ 2499 case ICMP_IPV6_IAMHERE: 2500 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2501 *virtual_id = 0; /* Nothing sane to match on! */ 2502 break; 2503 2504 case ICMP_MOBILE_REGREQUEST: 2505 *icmp_dir = PF_IN; 2506 /* FALLTHROUGH */ 2507 case ICMP_MOBILE_REGREPLY: 2508 *virtual_type = ICMP_MOBILE_REGREQUEST; 2509 *virtual_id = 0; /* Nothing sane to match on! */ 2510 break; 2511 2512 case ICMP_ROUTERSOLICIT: 2513 *icmp_dir = PF_IN; 2514 /* FALLTHROUGH */ 2515 case ICMP_ROUTERADVERT: 2516 *virtual_type = ICMP_ROUTERSOLICIT; 2517 *virtual_id = 0; /* Nothing sane to match on! */ 2518 break; 2519 2520 /* These ICMP types map to other connections */ 2521 case ICMP_UNREACH: 2522 case ICMP_SOURCEQUENCH: 2523 case ICMP_REDIRECT: 2524 case ICMP_TIMXCEED: 2525 case ICMP_PARAMPROB: 2526 /* These will not be used, but set them anyway */ 2527 *icmp_dir = PF_IN; 2528 *virtual_type = htons(type); 2529 *virtual_id = 0; 2530 return (1); /* These types match to another state */ 2531 2532 /* 2533 * All remaining ICMP types get their own states, 2534 * and will only match in one direction. 2535 */ 2536 default: 2537 *icmp_dir = PF_IN; 2538 *virtual_type = type; 2539 *virtual_id = 0; 2540 break; 2541 } 2542 break; 2543 #ifdef INET6 2544 case AF_INET6: 2545 switch (type) { 2546 case ICMP6_ECHO_REQUEST: 2547 *icmp_dir = PF_IN; 2548 /* FALLTHROUGH */ 2549 case ICMP6_ECHO_REPLY: 2550 *virtual_type = ICMP6_ECHO_REQUEST; 2551 *virtual_id = pd->hdr.icmp6.icmp6_id; 2552 break; 2553 2554 case MLD_LISTENER_QUERY: 2555 case MLD_LISTENER_REPORT: { 2556 struct mld_hdr *mld = &pd->hdr.mld; 2557 u_int32_t h; 2558 2559 /* 2560 * Listener Report can be sent by clients 2561 * without an associated Listener Query. 2562 * In addition to that, when Report is sent as a 2563 * reply to a Query its source and destination 2564 * address are different. 2565 */ 2566 *icmp_dir = PF_IN; 2567 *virtual_type = MLD_LISTENER_QUERY; 2568 /* generate fake id for these messages */ 2569 h = mld->mld_addr.s6_addr32[0] ^ 2570 mld->mld_addr.s6_addr32[1] ^ 2571 mld->mld_addr.s6_addr32[2] ^ 2572 mld->mld_addr.s6_addr32[3]; 2573 *virtual_id = (h >> 16) ^ (h & 0xffff); 2574 break; 2575 } 2576 2577 /* 2578 * ICMP6_FQDN and ICMP6_NI query/reply are the same type as 2579 * ICMP6_WRU 2580 */ 2581 case ICMP6_WRUREQUEST: 2582 *icmp_dir = PF_IN; 2583 /* FALLTHROUGH */ 2584 case ICMP6_WRUREPLY: 2585 *virtual_type = ICMP6_WRUREQUEST; 2586 *virtual_id = 0; /* Nothing sane to match on! */ 2587 break; 2588 2589 case MLD_MTRACE: 2590 *icmp_dir = PF_IN; 2591 /* FALLTHROUGH */ 2592 case MLD_MTRACE_RESP: 2593 *virtual_type = MLD_MTRACE; 2594 *virtual_id = 0; /* Nothing sane to match on! */ 2595 break; 2596 2597 case ND_NEIGHBOR_SOLICIT: 2598 *icmp_dir = PF_IN; 2599 /* FALLTHROUGH */ 2600 case ND_NEIGHBOR_ADVERT: { 2601 struct nd_neighbor_solicit *nd = &pd->hdr.nd_ns; 2602 u_int32_t h; 2603 2604 *virtual_type = ND_NEIGHBOR_SOLICIT; 2605 /* generate fake id for these messages */ 2606 h = nd->nd_ns_target.s6_addr32[0] ^ 2607 nd->nd_ns_target.s6_addr32[1] ^ 2608 nd->nd_ns_target.s6_addr32[2] ^ 2609 nd->nd_ns_target.s6_addr32[3]; 2610 *virtual_id = (h >> 16) ^ (h & 0xffff); 2611 break; 2612 } 2613 2614 /* 2615 * These ICMP types map to other connections. 2616 * ND_REDIRECT can't be in this list because the triggering 2617 * packet header is optional. 2618 */ 2619 case ICMP6_DST_UNREACH: 2620 case ICMP6_PACKET_TOO_BIG: 2621 case ICMP6_TIME_EXCEEDED: 2622 case ICMP6_PARAM_PROB: 2623 /* These will not be used, but set them anyway */ 2624 *icmp_dir = PF_IN; 2625 *virtual_type = htons(type); 2626 *virtual_id = 0; 2627 return (1); /* These types match to another state */ 2628 /* 2629 * All remaining ICMP6 types get their own states, 2630 * and will only match in one direction. 2631 */ 2632 default: 2633 *icmp_dir = PF_IN; 2634 *virtual_type = type; 2635 *virtual_id = 0; 2636 break; 2637 } 2638 break; 2639 #endif /* INET6 */ 2640 } 2641 *virtual_type = htons(*virtual_type); 2642 return (0); /* These types match to their own state */ 2643 } 2644 2645 void 2646 pf_translate_icmp(struct pf_pdesc *pd, struct pf_addr *qa, u_int16_t *qp, 2647 struct pf_addr *oa, struct pf_addr *na, u_int16_t np) 2648 { 2649 /* note: doesn't trouble to fixup quoted checksums, if any */ 2650 2651 /* change quoted protocol port */ 2652 if (qp != NULL) 2653 pf_patch_16(pd, qp, np); 2654 2655 /* change quoted ip address */ 2656 pf_cksum_fixup_a(pd->pcksum, qa, na, pd->af, pd->proto); 2657 pf_addrcpy(qa, na, pd->af); 2658 2659 /* change network-header's ip address */ 2660 if (oa) 2661 pf_translate_a(pd, oa, na); 2662 } 2663 2664 /* pre: *a is 16-bit aligned within its packet */ 2665 /* *a is a network header src/dst address */ 2666 int 2667 pf_translate_a(struct pf_pdesc *pd, struct pf_addr *a, struct pf_addr *an) 2668 { 2669 int rewrite = 0; 2670 2671 /* warning: !PF_ANEQ != PF_AEQ */ 2672 if (!PF_ANEQ(a, an, pd->af)) 2673 return (0); 2674 2675 /* fixup transport pseudo-header, if any */ 2676 switch (pd->proto) { 2677 case IPPROTO_TCP: /* FALLTHROUGH */ 2678 case IPPROTO_UDP: /* FALLTHROUGH */ 2679 case IPPROTO_ICMPV6: 2680 pf_cksum_fixup_a(pd->pcksum, a, an, pd->af, pd->proto); 2681 break; 2682 default: 2683 break; /* assume no pseudo-header */ 2684 } 2685 2686 pf_addrcpy(a, an, pd->af); 2687 rewrite = 1; 2688 2689 return (rewrite); 2690 } 2691 2692 #ifdef INET6 2693 /* pf_translate_af() may change pd->m, adjust local copies after calling */ 2694 int 2695 pf_translate_af(struct pf_pdesc *pd) 2696 { 2697 static const struct pf_addr zero; 2698 struct ip *ip4; 2699 struct ip6_hdr *ip6; 2700 int copyback = 0; 2701 u_int hlen, ohlen, dlen; 2702 u_int16_t *pc; 2703 u_int8_t af_proto, naf_proto; 2704 2705 hlen = (pd->naf == AF_INET) ? sizeof(*ip4) : sizeof(*ip6); 2706 ohlen = pd->off; 2707 dlen = pd->tot_len - pd->off; 2708 pc = pd->pcksum; 2709 2710 af_proto = naf_proto = pd->proto; 2711 if (naf_proto == IPPROTO_ICMP) 2712 af_proto = IPPROTO_ICMPV6; 2713 if (naf_proto == IPPROTO_ICMPV6) 2714 af_proto = IPPROTO_ICMP; 2715 2716 /* uncover stale pseudo-header */ 2717 switch (af_proto) { 2718 case IPPROTO_ICMPV6: 2719 /* optimise: unchanged for TCP/UDP */ 2720 pf_cksum_fixup(pc, htons(af_proto), 0x0, af_proto); 2721 pf_cksum_fixup(pc, htons(dlen), 0x0, af_proto); 2722 /* FALLTHROUGH */ 2723 case IPPROTO_UDP: /* FALLTHROUGH */ 2724 case IPPROTO_TCP: 2725 pf_cksum_fixup_a(pc, pd->src, &zero, pd->af, af_proto); 2726 pf_cksum_fixup_a(pc, pd->dst, &zero, pd->af, af_proto); 2727 copyback = 1; 2728 break; 2729 default: 2730 break; /* assume no pseudo-header */ 2731 } 2732 2733 /* replace the network header */ 2734 m_adj(pd->m, pd->off); 2735 pd->src = NULL; 2736 pd->dst = NULL; 2737 2738 if ((M_PREPEND(pd->m, hlen, M_DONTWAIT)) == NULL) { 2739 pd->m = NULL; 2740 return (-1); 2741 } 2742 2743 pd->off = hlen; 2744 pd->tot_len += hlen - ohlen; 2745 2746 switch (pd->naf) { 2747 case AF_INET: 2748 ip4 = mtod(pd->m, struct ip *); 2749 memset(ip4, 0, hlen); 2750 ip4->ip_v = IPVERSION; 2751 ip4->ip_hl = hlen >> 2; 2752 ip4->ip_tos = pd->tos; 2753 ip4->ip_len = htons(hlen + dlen); 2754 ip4->ip_id = htons(ip_randomid()); 2755 ip4->ip_off = htons(IP_DF); 2756 ip4->ip_ttl = pd->ttl; 2757 ip4->ip_p = pd->proto; 2758 ip4->ip_src = pd->nsaddr.v4; 2759 ip4->ip_dst = pd->ndaddr.v4; 2760 break; 2761 case AF_INET6: 2762 ip6 = mtod(pd->m, struct ip6_hdr *); 2763 memset(ip6, 0, hlen); 2764 ip6->ip6_vfc = IPV6_VERSION; 2765 ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20); 2766 ip6->ip6_plen = htons(dlen); 2767 ip6->ip6_nxt = pd->proto; 2768 if (!pd->ttl || pd->ttl > IPV6_DEFHLIM) 2769 ip6->ip6_hlim = IPV6_DEFHLIM; 2770 else 2771 ip6->ip6_hlim = pd->ttl; 2772 ip6->ip6_src = pd->nsaddr.v6; 2773 ip6->ip6_dst = pd->ndaddr.v6; 2774 break; 2775 default: 2776 unhandled_af(pd->naf); 2777 } 2778 2779 /* UDP over IPv6 must be checksummed per rfc2460 p27 */ 2780 if (naf_proto == IPPROTO_UDP && *pc == 0x0000 && 2781 pd->naf == AF_INET6) { 2782 pd->m->m_pkthdr.csum_flags |= M_UDP_CSUM_OUT; 2783 } 2784 2785 /* cover fresh pseudo-header */ 2786 switch (naf_proto) { 2787 case IPPROTO_ICMPV6: 2788 /* optimise: unchanged for TCP/UDP */ 2789 pf_cksum_fixup(pc, 0x0, htons(naf_proto), naf_proto); 2790 pf_cksum_fixup(pc, 0x0, htons(dlen), naf_proto); 2791 /* FALLTHROUGH */ 2792 case IPPROTO_UDP: /* FALLTHROUGH */ 2793 case IPPROTO_TCP: 2794 pf_cksum_fixup_a(pc, &zero, &pd->nsaddr, pd->naf, naf_proto); 2795 pf_cksum_fixup_a(pc, &zero, &pd->ndaddr, pd->naf, naf_proto); 2796 copyback = 1; 2797 break; 2798 default: 2799 break; /* assume no pseudo-header */ 2800 } 2801 2802 /* flush pd->pcksum */ 2803 if (copyback) 2804 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 2805 2806 return (0); 2807 } 2808 2809 int 2810 pf_change_icmp_af(struct mbuf *m, int ipoff2, struct pf_pdesc *pd, 2811 struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst, 2812 sa_family_t af, sa_family_t naf) 2813 { 2814 struct mbuf *n = NULL; 2815 struct ip *ip4; 2816 struct ip6_hdr *ip6; 2817 u_int hlen, ohlen, dlen; 2818 int d; 2819 2820 if (af == naf || (af != AF_INET && af != AF_INET6) || 2821 (naf != AF_INET && naf != AF_INET6)) 2822 return (-1); 2823 2824 /* split the mbuf chain on the quoted ip/ip6 header boundary */ 2825 if ((n = m_split(m, ipoff2, M_DONTWAIT)) == NULL) 2826 return (-1); 2827 2828 /* new quoted header */ 2829 hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 2830 /* old quoted header */ 2831 ohlen = pd2->off - ipoff2; 2832 2833 /* trim old quoted header */ 2834 pf_cksum_uncover(pd->pcksum, in_cksum(n, ohlen), pd->proto); 2835 m_adj(n, ohlen); 2836 2837 /* prepend a new, translated, quoted header */ 2838 if ((M_PREPEND(n, hlen, M_DONTWAIT)) == NULL) 2839 return (-1); 2840 2841 switch (naf) { 2842 case AF_INET: 2843 ip4 = mtod(n, struct ip *); 2844 memset(ip4, 0, sizeof(*ip4)); 2845 ip4->ip_v = IPVERSION; 2846 ip4->ip_hl = sizeof(*ip4) >> 2; 2847 ip4->ip_len = htons(sizeof(*ip4) + pd2->tot_len - ohlen); 2848 ip4->ip_id = htons(ip_randomid()); 2849 ip4->ip_off = htons(IP_DF); 2850 ip4->ip_ttl = pd2->ttl; 2851 if (pd2->proto == IPPROTO_ICMPV6) 2852 ip4->ip_p = IPPROTO_ICMP; 2853 else 2854 ip4->ip_p = pd2->proto; 2855 ip4->ip_src = src->v4; 2856 ip4->ip_dst = dst->v4; 2857 ip4->ip_sum = in_cksum(n, ip4->ip_hl << 2); 2858 break; 2859 case AF_INET6: 2860 ip6 = mtod(n, struct ip6_hdr *); 2861 memset(ip6, 0, sizeof(*ip6)); 2862 ip6->ip6_vfc = IPV6_VERSION; 2863 ip6->ip6_plen = htons(pd2->tot_len - ohlen); 2864 if (pd2->proto == IPPROTO_ICMP) 2865 ip6->ip6_nxt = IPPROTO_ICMPV6; 2866 else 2867 ip6->ip6_nxt = pd2->proto; 2868 if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM) 2869 ip6->ip6_hlim = IPV6_DEFHLIM; 2870 else 2871 ip6->ip6_hlim = pd2->ttl; 2872 ip6->ip6_src = src->v6; 2873 ip6->ip6_dst = dst->v6; 2874 break; 2875 } 2876 2877 /* cover new quoted header */ 2878 /* optimise: any new AF_INET header of ours sums to zero */ 2879 if (naf != AF_INET) { 2880 pf_cksum_cover(pd->pcksum, in_cksum(n, hlen), pd->proto); 2881 } 2882 2883 /* reattach modified quoted packet to outer header */ 2884 { 2885 int nlen = n->m_pkthdr.len; 2886 m_cat(m, n); 2887 m->m_pkthdr.len += nlen; 2888 } 2889 2890 /* account for altered length */ 2891 d = hlen - ohlen; 2892 2893 if (pd->proto == IPPROTO_ICMPV6) { 2894 /* fixup pseudo-header */ 2895 dlen = pd->tot_len - pd->off; 2896 pf_cksum_fixup(pd->pcksum, 2897 htons(dlen), htons(dlen + d), pd->proto); 2898 } 2899 2900 pd->tot_len += d; 2901 pd2->tot_len += d; 2902 pd2->off += d; 2903 2904 /* note: not bothering to update network headers as 2905 these due for rewrite by pf_translate_af() */ 2906 2907 return (0); 2908 } 2909 2910 2911 #define PTR_IP(field) (offsetof(struct ip, field)) 2912 #define PTR_IP6(field) (offsetof(struct ip6_hdr, field)) 2913 2914 int 2915 pf_translate_icmp_af(struct pf_pdesc *pd, int af, void *arg) 2916 { 2917 struct icmp *icmp4; 2918 struct icmp6_hdr *icmp6; 2919 u_int32_t mtu; 2920 int32_t ptr = -1; 2921 u_int8_t type; 2922 u_int8_t code; 2923 2924 switch (af) { 2925 case AF_INET: 2926 icmp6 = arg; 2927 type = icmp6->icmp6_type; 2928 code = icmp6->icmp6_code; 2929 mtu = ntohl(icmp6->icmp6_mtu); 2930 2931 switch (type) { 2932 case ICMP6_ECHO_REQUEST: 2933 type = ICMP_ECHO; 2934 break; 2935 case ICMP6_ECHO_REPLY: 2936 type = ICMP_ECHOREPLY; 2937 break; 2938 case ICMP6_DST_UNREACH: 2939 type = ICMP_UNREACH; 2940 switch (code) { 2941 case ICMP6_DST_UNREACH_NOROUTE: 2942 case ICMP6_DST_UNREACH_BEYONDSCOPE: 2943 case ICMP6_DST_UNREACH_ADDR: 2944 code = ICMP_UNREACH_HOST; 2945 break; 2946 case ICMP6_DST_UNREACH_ADMIN: 2947 code = ICMP_UNREACH_HOST_PROHIB; 2948 break; 2949 case ICMP6_DST_UNREACH_NOPORT: 2950 code = ICMP_UNREACH_PORT; 2951 break; 2952 default: 2953 return (-1); 2954 } 2955 break; 2956 case ICMP6_PACKET_TOO_BIG: 2957 type = ICMP_UNREACH; 2958 code = ICMP_UNREACH_NEEDFRAG; 2959 mtu -= 20; 2960 break; 2961 case ICMP6_TIME_EXCEEDED: 2962 type = ICMP_TIMXCEED; 2963 break; 2964 case ICMP6_PARAM_PROB: 2965 switch (code) { 2966 case ICMP6_PARAMPROB_HEADER: 2967 type = ICMP_PARAMPROB; 2968 code = ICMP_PARAMPROB_ERRATPTR; 2969 ptr = ntohl(icmp6->icmp6_pptr); 2970 2971 if (ptr == PTR_IP6(ip6_vfc)) 2972 ; /* preserve */ 2973 else if (ptr == PTR_IP6(ip6_vfc) + 1) 2974 ptr = PTR_IP(ip_tos); 2975 else if (ptr == PTR_IP6(ip6_plen) || 2976 ptr == PTR_IP6(ip6_plen) + 1) 2977 ptr = PTR_IP(ip_len); 2978 else if (ptr == PTR_IP6(ip6_nxt)) 2979 ptr = PTR_IP(ip_p); 2980 else if (ptr == PTR_IP6(ip6_hlim)) 2981 ptr = PTR_IP(ip_ttl); 2982 else if (ptr >= PTR_IP6(ip6_src) && 2983 ptr < PTR_IP6(ip6_dst)) 2984 ptr = PTR_IP(ip_src); 2985 else if (ptr >= PTR_IP6(ip6_dst) && 2986 ptr < sizeof(struct ip6_hdr)) 2987 ptr = PTR_IP(ip_dst); 2988 else { 2989 return (-1); 2990 } 2991 break; 2992 case ICMP6_PARAMPROB_NEXTHEADER: 2993 type = ICMP_UNREACH; 2994 code = ICMP_UNREACH_PROTOCOL; 2995 break; 2996 default: 2997 return (-1); 2998 } 2999 break; 3000 default: 3001 return (-1); 3002 } 3003 3004 pf_patch_8(pd, &icmp6->icmp6_type, type, PF_HI); 3005 pf_patch_8(pd, &icmp6->icmp6_code, code, PF_LO); 3006 3007 /* aligns well with a icmpv4 nextmtu */ 3008 pf_patch_32(pd, &icmp6->icmp6_mtu, htonl(mtu)); 3009 3010 /* icmpv4 pptr is a one most significant byte */ 3011 if (ptr >= 0) 3012 pf_patch_32(pd, &icmp6->icmp6_pptr, htonl(ptr << 24)); 3013 break; 3014 case AF_INET6: 3015 icmp4 = arg; 3016 type = icmp4->icmp_type; 3017 code = icmp4->icmp_code; 3018 mtu = ntohs(icmp4->icmp_nextmtu); 3019 3020 switch (type) { 3021 case ICMP_ECHO: 3022 type = ICMP6_ECHO_REQUEST; 3023 break; 3024 case ICMP_ECHOREPLY: 3025 type = ICMP6_ECHO_REPLY; 3026 break; 3027 case ICMP_UNREACH: 3028 type = ICMP6_DST_UNREACH; 3029 switch (code) { 3030 case ICMP_UNREACH_NET: 3031 case ICMP_UNREACH_HOST: 3032 case ICMP_UNREACH_NET_UNKNOWN: 3033 case ICMP_UNREACH_HOST_UNKNOWN: 3034 case ICMP_UNREACH_ISOLATED: 3035 case ICMP_UNREACH_TOSNET: 3036 case ICMP_UNREACH_TOSHOST: 3037 code = ICMP6_DST_UNREACH_NOROUTE; 3038 break; 3039 case ICMP_UNREACH_PORT: 3040 code = ICMP6_DST_UNREACH_NOPORT; 3041 break; 3042 case ICMP_UNREACH_NET_PROHIB: 3043 case ICMP_UNREACH_HOST_PROHIB: 3044 case ICMP_UNREACH_FILTER_PROHIB: 3045 case ICMP_UNREACH_PRECEDENCE_CUTOFF: 3046 code = ICMP6_DST_UNREACH_ADMIN; 3047 break; 3048 case ICMP_UNREACH_PROTOCOL: 3049 type = ICMP6_PARAM_PROB; 3050 code = ICMP6_PARAMPROB_NEXTHEADER; 3051 ptr = offsetof(struct ip6_hdr, ip6_nxt); 3052 break; 3053 case ICMP_UNREACH_NEEDFRAG: 3054 type = ICMP6_PACKET_TOO_BIG; 3055 code = 0; 3056 mtu += 20; 3057 break; 3058 default: 3059 return (-1); 3060 } 3061 break; 3062 case ICMP_TIMXCEED: 3063 type = ICMP6_TIME_EXCEEDED; 3064 break; 3065 case ICMP_PARAMPROB: 3066 type = ICMP6_PARAM_PROB; 3067 switch (code) { 3068 case ICMP_PARAMPROB_ERRATPTR: 3069 code = ICMP6_PARAMPROB_HEADER; 3070 break; 3071 case ICMP_PARAMPROB_LENGTH: 3072 code = ICMP6_PARAMPROB_HEADER; 3073 break; 3074 default: 3075 return (-1); 3076 } 3077 3078 ptr = icmp4->icmp_pptr; 3079 if (ptr == 0 || ptr == PTR_IP(ip_tos)) 3080 ; /* preserve */ 3081 else if (ptr == PTR_IP(ip_len) || 3082 ptr == PTR_IP(ip_len) + 1) 3083 ptr = PTR_IP6(ip6_plen); 3084 else if (ptr == PTR_IP(ip_ttl)) 3085 ptr = PTR_IP6(ip6_hlim); 3086 else if (ptr == PTR_IP(ip_p)) 3087 ptr = PTR_IP6(ip6_nxt); 3088 else if (ptr >= PTR_IP(ip_src) && 3089 ptr < PTR_IP(ip_dst)) 3090 ptr = PTR_IP6(ip6_src); 3091 else if (ptr >= PTR_IP(ip_dst) && 3092 ptr < sizeof(struct ip)) 3093 ptr = PTR_IP6(ip6_dst); 3094 else { 3095 return (-1); 3096 } 3097 break; 3098 default: 3099 return (-1); 3100 } 3101 3102 pf_patch_8(pd, &icmp4->icmp_type, type, PF_HI); 3103 pf_patch_8(pd, &icmp4->icmp_code, code, PF_LO); 3104 pf_patch_16(pd, &icmp4->icmp_nextmtu, htons(mtu)); 3105 if (ptr >= 0) 3106 pf_patch_32(pd, &icmp4->icmp_void, htonl(ptr)); 3107 break; 3108 } 3109 3110 return (0); 3111 } 3112 #endif /* INET6 */ 3113 3114 /* 3115 * Need to modulate the sequence numbers in the TCP SACK option 3116 * (credits to Krzysztof Pfaff for report and patch) 3117 */ 3118 int 3119 pf_modulate_sack(struct pf_pdesc *pd, struct pf_state_peer *dst) 3120 { 3121 struct sackblk sack; 3122 int copyback = 0, i; 3123 int olen, optsoff; 3124 u_int8_t opts[MAX_TCPOPTLEN], *opt, *eoh; 3125 3126 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 3127 optsoff = pd->off + sizeof(struct tcphdr); 3128 #define TCPOLEN_MINSACK (TCPOLEN_SACK + 2) 3129 if (olen < TCPOLEN_MINSACK || 3130 !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, NULL, pd->af)) 3131 return (0); 3132 3133 eoh = opts + olen; 3134 opt = opts; 3135 while ((opt = pf_find_tcpopt(opt, opts, olen, 3136 TCPOPT_SACK, TCPOLEN_MINSACK)) != NULL) 3137 { 3138 size_t safelen = MIN(opt[1], (eoh - opt)); 3139 for (i = 2; i + TCPOLEN_SACK <= safelen; i += TCPOLEN_SACK) { 3140 size_t startoff = (opt + i) - opts; 3141 memcpy(&sack, &opt[i], sizeof(sack)); 3142 pf_patch_32_unaligned(pd, &sack.start, 3143 htonl(ntohl(sack.start) - dst->seqdiff), 3144 PF_ALGNMNT(startoff)); 3145 pf_patch_32_unaligned(pd, &sack.end, 3146 htonl(ntohl(sack.end) - dst->seqdiff), 3147 PF_ALGNMNT(startoff + sizeof(sack.start))); 3148 memcpy(&opt[i], &sack, sizeof(sack)); 3149 } 3150 copyback = 1; 3151 opt += opt[1]; 3152 } 3153 3154 if (copyback) 3155 m_copyback(pd->m, optsoff, olen, opts, M_NOWAIT); 3156 return (copyback); 3157 } 3158 3159 struct mbuf * 3160 pf_build_tcp(const struct pf_rule *r, sa_family_t af, 3161 const struct pf_addr *saddr, const struct pf_addr *daddr, 3162 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3163 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 3164 u_int16_t rtag, u_int sack, u_int rdom) 3165 { 3166 struct mbuf *m; 3167 int len, tlen; 3168 struct ip *h; 3169 #ifdef INET6 3170 struct ip6_hdr *h6; 3171 #endif /* INET6 */ 3172 struct tcphdr *th; 3173 char *opt; 3174 3175 /* maximum segment size tcp option */ 3176 tlen = sizeof(struct tcphdr); 3177 if (mss) 3178 tlen += 4; 3179 if (sack) 3180 tlen += 2; 3181 3182 switch (af) { 3183 case AF_INET: 3184 len = sizeof(struct ip) + tlen; 3185 break; 3186 #ifdef INET6 3187 case AF_INET6: 3188 len = sizeof(struct ip6_hdr) + tlen; 3189 break; 3190 #endif /* INET6 */ 3191 default: 3192 unhandled_af(af); 3193 } 3194 3195 /* create outgoing mbuf */ 3196 m = m_gethdr(M_DONTWAIT, MT_HEADER); 3197 if (m == NULL) 3198 return (NULL); 3199 if (tag) 3200 m->m_pkthdr.pf.flags |= PF_TAG_GENERATED; 3201 m->m_pkthdr.pf.tag = rtag; 3202 m->m_pkthdr.ph_rtableid = rdom; 3203 if (r && (r->scrub_flags & PFSTATE_SETPRIO)) 3204 m->m_pkthdr.pf.prio = r->set_prio[0]; 3205 if (r && r->qid) 3206 m->m_pkthdr.pf.qid = r->qid; 3207 m->m_data += max_linkhdr; 3208 m->m_pkthdr.len = m->m_len = len; 3209 m->m_pkthdr.ph_ifidx = 0; 3210 m->m_pkthdr.csum_flags |= M_TCP_CSUM_OUT; 3211 memset(m->m_data, 0, len); 3212 switch (af) { 3213 case AF_INET: 3214 h = mtod(m, struct ip *); 3215 h->ip_p = IPPROTO_TCP; 3216 h->ip_len = htons(tlen); 3217 h->ip_v = 4; 3218 h->ip_hl = sizeof(*h) >> 2; 3219 h->ip_tos = IPTOS_LOWDELAY; 3220 h->ip_len = htons(len); 3221 h->ip_off = htons(ip_mtudisc ? IP_DF : 0); 3222 h->ip_ttl = ttl ? ttl : ip_defttl; 3223 h->ip_sum = 0; 3224 h->ip_src.s_addr = saddr->v4.s_addr; 3225 h->ip_dst.s_addr = daddr->v4.s_addr; 3226 3227 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3228 break; 3229 #ifdef INET6 3230 case AF_INET6: 3231 h6 = mtod(m, struct ip6_hdr *); 3232 h6->ip6_nxt = IPPROTO_TCP; 3233 h6->ip6_plen = htons(tlen); 3234 h6->ip6_vfc |= IPV6_VERSION; 3235 h6->ip6_hlim = IPV6_DEFHLIM; 3236 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3237 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3238 3239 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3240 break; 3241 #endif /* INET6 */ 3242 default: 3243 unhandled_af(af); 3244 } 3245 3246 /* TCP header */ 3247 th->th_sport = sport; 3248 th->th_dport = dport; 3249 th->th_seq = htonl(seq); 3250 th->th_ack = htonl(ack); 3251 th->th_off = tlen >> 2; 3252 th->th_flags = flags; 3253 th->th_win = htons(win); 3254 3255 opt = (char *)(th + 1); 3256 if (mss) { 3257 opt[0] = TCPOPT_MAXSEG; 3258 opt[1] = 4; 3259 mss = htons(mss); 3260 memcpy((opt + 2), &mss, 2); 3261 opt += 4; 3262 } 3263 if (sack) { 3264 opt[0] = TCPOPT_SACK_PERMITTED; 3265 opt[1] = 2; 3266 opt += 2; 3267 } 3268 3269 return (m); 3270 } 3271 3272 void 3273 pf_send_tcp(const struct pf_rule *r, sa_family_t af, 3274 const struct pf_addr *saddr, const struct pf_addr *daddr, 3275 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3276 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 3277 u_int16_t rtag, u_int rdom) 3278 { 3279 struct mbuf *m; 3280 3281 if ((m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, 3282 flags, win, mss, ttl, tag, rtag, 0, rdom)) == NULL) 3283 return; 3284 3285 switch (af) { 3286 case AF_INET: 3287 ip_send(m); 3288 break; 3289 #ifdef INET6 3290 case AF_INET6: 3291 ip6_send(m); 3292 break; 3293 #endif /* INET6 */ 3294 } 3295 } 3296 3297 static void 3298 pf_send_challenge_ack(struct pf_pdesc *pd, struct pf_state *st, 3299 struct pf_state_peer *src, struct pf_state_peer *dst) 3300 { 3301 /* 3302 * We are sending challenge ACK as a response to SYN packet, which 3303 * matches existing state (modulo TCP window check). Therefore packet 3304 * must be sent on behalf of destination. 3305 * 3306 * We expect sender to remain either silent, or send RST packet 3307 * so both, firewall and remote peer, can purge dead state from 3308 * memory. 3309 */ 3310 pf_send_tcp(st->rule.ptr, pd->af, pd->dst, pd->src, 3311 pd->hdr.tcp.th_dport, pd->hdr.tcp.th_sport, dst->seqlo, 3312 src->seqlo, TH_ACK, 0, 0, st->rule.ptr->return_ttl, 1, 0, 3313 pd->rdomain); 3314 } 3315 3316 void 3317 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, int param, 3318 sa_family_t af, struct pf_rule *r, u_int rdomain) 3319 { 3320 struct mbuf *m0; 3321 3322 if ((m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT)) == NULL) 3323 return; 3324 3325 m0->m_pkthdr.pf.flags |= PF_TAG_GENERATED; 3326 m0->m_pkthdr.ph_rtableid = rdomain; 3327 if (r && (r->scrub_flags & PFSTATE_SETPRIO)) 3328 m0->m_pkthdr.pf.prio = r->set_prio[0]; 3329 if (r && r->qid) 3330 m0->m_pkthdr.pf.qid = r->qid; 3331 3332 switch (af) { 3333 case AF_INET: 3334 icmp_error(m0, type, code, 0, param); 3335 break; 3336 #ifdef INET6 3337 case AF_INET6: 3338 icmp6_error(m0, type, code, param); 3339 break; 3340 #endif /* INET6 */ 3341 } 3342 } 3343 3344 /* 3345 * Return ((n = 0) == (a = b [with mask m])) 3346 * Note: n != 0 => returns (a != b [with mask m]) 3347 */ 3348 int 3349 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3350 struct pf_addr *b, sa_family_t af) 3351 { 3352 switch (af) { 3353 case AF_INET: 3354 if ((a->addr32[0] & m->addr32[0]) == 3355 (b->addr32[0] & m->addr32[0])) 3356 return (n == 0); 3357 break; 3358 #ifdef INET6 3359 case AF_INET6: 3360 if (((a->addr32[0] & m->addr32[0]) == 3361 (b->addr32[0] & m->addr32[0])) && 3362 ((a->addr32[1] & m->addr32[1]) == 3363 (b->addr32[1] & m->addr32[1])) && 3364 ((a->addr32[2] & m->addr32[2]) == 3365 (b->addr32[2] & m->addr32[2])) && 3366 ((a->addr32[3] & m->addr32[3]) == 3367 (b->addr32[3] & m->addr32[3]))) 3368 return (n == 0); 3369 break; 3370 #endif /* INET6 */ 3371 } 3372 3373 return (n != 0); 3374 } 3375 3376 /* 3377 * Return 1 if b <= a <= e, otherwise return 0. 3378 */ 3379 int 3380 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3381 struct pf_addr *a, sa_family_t af) 3382 { 3383 switch (af) { 3384 case AF_INET: 3385 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3386 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3387 return (0); 3388 break; 3389 #ifdef INET6 3390 case AF_INET6: { 3391 int i; 3392 3393 /* check a >= b */ 3394 for (i = 0; i < 4; ++i) 3395 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3396 break; 3397 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3398 return (0); 3399 /* check a <= e */ 3400 for (i = 0; i < 4; ++i) 3401 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3402 break; 3403 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3404 return (0); 3405 break; 3406 } 3407 #endif /* INET6 */ 3408 } 3409 return (1); 3410 } 3411 3412 int 3413 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3414 { 3415 switch (op) { 3416 case PF_OP_IRG: 3417 return ((p > a1) && (p < a2)); 3418 case PF_OP_XRG: 3419 return ((p < a1) || (p > a2)); 3420 case PF_OP_RRG: 3421 return ((p >= a1) && (p <= a2)); 3422 case PF_OP_EQ: 3423 return (p == a1); 3424 case PF_OP_NE: 3425 return (p != a1); 3426 case PF_OP_LT: 3427 return (p < a1); 3428 case PF_OP_LE: 3429 return (p <= a1); 3430 case PF_OP_GT: 3431 return (p > a1); 3432 case PF_OP_GE: 3433 return (p >= a1); 3434 } 3435 return (0); /* never reached */ 3436 } 3437 3438 int 3439 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3440 { 3441 return (pf_match(op, ntohs(a1), ntohs(a2), ntohs(p))); 3442 } 3443 3444 int 3445 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3446 { 3447 if (u == -1 && op != PF_OP_EQ && op != PF_OP_NE) 3448 return (0); 3449 return (pf_match(op, a1, a2, u)); 3450 } 3451 3452 int 3453 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3454 { 3455 if (g == -1 && op != PF_OP_EQ && op != PF_OP_NE) 3456 return (0); 3457 return (pf_match(op, a1, a2, g)); 3458 } 3459 3460 int 3461 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag) 3462 { 3463 if (*tag == -1) 3464 *tag = m->m_pkthdr.pf.tag; 3465 3466 return ((!r->match_tag_not && r->match_tag == *tag) || 3467 (r->match_tag_not && r->match_tag != *tag)); 3468 } 3469 3470 int 3471 pf_match_rcvif(struct mbuf *m, struct pf_rule *r) 3472 { 3473 struct ifnet *ifp; 3474 #if NCARP > 0 3475 struct ifnet *ifp0; 3476 #endif 3477 struct pfi_kif *kif; 3478 3479 ifp = if_get(m->m_pkthdr.ph_ifidx); 3480 if (ifp == NULL) 3481 return (0); 3482 3483 #if NCARP > 0 3484 if (ifp->if_type == IFT_CARP && 3485 (ifp0 = if_get(ifp->if_carpdevidx)) != NULL) { 3486 kif = (struct pfi_kif *)ifp0->if_pf_kif; 3487 if_put(ifp0); 3488 } else 3489 #endif /* NCARP */ 3490 kif = (struct pfi_kif *)ifp->if_pf_kif; 3491 3492 if_put(ifp); 3493 3494 if (kif == NULL) { 3495 DPFPRINTF(LOG_ERR, 3496 "%s: kif == NULL, @%d via %s", __func__, 3497 r->nr, r->rcv_ifname); 3498 return (0); 3499 } 3500 3501 return (pfi_kif_match(r->rcv_kif, kif)); 3502 } 3503 3504 void 3505 pf_tag_packet(struct mbuf *m, int tag, int rtableid) 3506 { 3507 if (tag > 0) 3508 m->m_pkthdr.pf.tag = tag; 3509 if (rtableid >= 0) 3510 m->m_pkthdr.ph_rtableid = (u_int)rtableid; 3511 } 3512 3513 void 3514 pf_anchor_stack_init(void) 3515 { 3516 struct pf_anchor_stackframe *stack; 3517 3518 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3519 stack[PF_ANCHOR_STACK_MAX].sf_stack_top = &stack[0]; 3520 cpumem_leave(pf_anchor_stack, stack); 3521 } 3522 3523 int 3524 pf_anchor_stack_is_full(struct pf_anchor_stackframe *sf) 3525 { 3526 struct pf_anchor_stackframe *stack; 3527 int rv; 3528 3529 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3530 rv = (sf == &stack[PF_ANCHOR_STACK_MAX]); 3531 cpumem_leave(pf_anchor_stack, stack); 3532 3533 return (rv); 3534 } 3535 3536 int 3537 pf_anchor_stack_is_empty(struct pf_anchor_stackframe *sf) 3538 { 3539 struct pf_anchor_stackframe *stack; 3540 int rv; 3541 3542 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3543 rv = (sf == &stack[0]); 3544 cpumem_leave(pf_anchor_stack, stack); 3545 3546 return (rv); 3547 } 3548 3549 struct pf_anchor_stackframe * 3550 pf_anchor_stack_top(void) 3551 { 3552 struct pf_anchor_stackframe *stack; 3553 struct pf_anchor_stackframe *top_sf; 3554 3555 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3556 top_sf = stack[PF_ANCHOR_STACK_MAX].sf_stack_top; 3557 cpumem_leave(pf_anchor_stack, stack); 3558 3559 return (top_sf); 3560 } 3561 3562 int 3563 pf_anchor_stack_push(struct pf_ruleset *rs, struct pf_rule *r, 3564 struct pf_anchor *child, int jump_target) 3565 { 3566 struct pf_anchor_stackframe *stack; 3567 struct pf_anchor_stackframe *top_sf = pf_anchor_stack_top(); 3568 3569 top_sf++; 3570 if (pf_anchor_stack_is_full(top_sf)) 3571 return (-1); 3572 3573 top_sf->sf_rs = rs; 3574 top_sf->sf_r = r; 3575 top_sf->sf_child = child; 3576 top_sf->sf_jump_target = jump_target; 3577 3578 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3579 3580 if ((top_sf <= &stack[0]) || (top_sf >= &stack[PF_ANCHOR_STACK_MAX])) 3581 panic("%s: top frame outside of anchor stack range", __func__); 3582 3583 stack[PF_ANCHOR_STACK_MAX].sf_stack_top = top_sf; 3584 cpumem_leave(pf_anchor_stack, stack); 3585 3586 return (0); 3587 } 3588 3589 int 3590 pf_anchor_stack_pop(struct pf_ruleset **rs, struct pf_rule **r, 3591 struct pf_anchor **child, int *jump_target) 3592 { 3593 struct pf_anchor_stackframe *top_sf = pf_anchor_stack_top(); 3594 struct pf_anchor_stackframe *stack; 3595 int on_top; 3596 3597 stack = (struct pf_anchor_stackframe *)cpumem_enter(pf_anchor_stack); 3598 if (pf_anchor_stack_is_empty(top_sf)) { 3599 on_top = -1; 3600 } else { 3601 if ((top_sf <= &stack[0]) || 3602 (top_sf >= &stack[PF_ANCHOR_STACK_MAX])) 3603 panic("%s: top frame outside of anchor stack range", 3604 __func__); 3605 3606 *rs = top_sf->sf_rs; 3607 *r = top_sf->sf_r; 3608 *child = top_sf->sf_child; 3609 *jump_target = top_sf->sf_jump_target; 3610 top_sf--; 3611 stack[PF_ANCHOR_STACK_MAX].sf_stack_top = top_sf; 3612 on_top = 0; 3613 } 3614 cpumem_leave(pf_anchor_stack, stack); 3615 3616 return (on_top); 3617 } 3618 3619 void 3620 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3621 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3622 { 3623 switch (af) { 3624 case AF_INET: 3625 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3626 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3627 break; 3628 #ifdef INET6 3629 case AF_INET6: 3630 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3631 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3632 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3633 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3634 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3635 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3636 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3637 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3638 break; 3639 #endif /* INET6 */ 3640 default: 3641 unhandled_af(af); 3642 } 3643 } 3644 3645 void 3646 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3647 { 3648 switch (af) { 3649 case AF_INET: 3650 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3651 break; 3652 #ifdef INET6 3653 case AF_INET6: 3654 if (addr->addr32[3] == 0xffffffff) { 3655 addr->addr32[3] = 0; 3656 if (addr->addr32[2] == 0xffffffff) { 3657 addr->addr32[2] = 0; 3658 if (addr->addr32[1] == 0xffffffff) { 3659 addr->addr32[1] = 0; 3660 addr->addr32[0] = 3661 htonl(ntohl(addr->addr32[0]) + 1); 3662 } else 3663 addr->addr32[1] = 3664 htonl(ntohl(addr->addr32[1]) + 1); 3665 } else 3666 addr->addr32[2] = 3667 htonl(ntohl(addr->addr32[2]) + 1); 3668 } else 3669 addr->addr32[3] = 3670 htonl(ntohl(addr->addr32[3]) + 1); 3671 break; 3672 #endif /* INET6 */ 3673 default: 3674 unhandled_af(af); 3675 } 3676 } 3677 3678 int 3679 pf_socket_lookup(struct pf_pdesc *pd) 3680 { 3681 struct pf_addr *saddr, *daddr; 3682 u_int16_t sport, dport; 3683 struct inpcbtable *tb; 3684 struct inpcb *inp; 3685 3686 pd->lookup.uid = -1; 3687 pd->lookup.gid = -1; 3688 pd->lookup.pid = NO_PID; 3689 switch (pd->virtual_proto) { 3690 case IPPROTO_TCP: 3691 sport = pd->hdr.tcp.th_sport; 3692 dport = pd->hdr.tcp.th_dport; 3693 PF_ASSERT_LOCKED(); 3694 NET_ASSERT_LOCKED(); 3695 tb = &tcbtable; 3696 break; 3697 case IPPROTO_UDP: 3698 sport = pd->hdr.udp.uh_sport; 3699 dport = pd->hdr.udp.uh_dport; 3700 PF_ASSERT_LOCKED(); 3701 NET_ASSERT_LOCKED(); 3702 tb = &udbtable; 3703 break; 3704 default: 3705 return (-1); 3706 } 3707 if (pd->dir == PF_IN) { 3708 saddr = pd->src; 3709 daddr = pd->dst; 3710 } else { 3711 u_int16_t p; 3712 3713 p = sport; 3714 sport = dport; 3715 dport = p; 3716 saddr = pd->dst; 3717 daddr = pd->src; 3718 } 3719 switch (pd->af) { 3720 case AF_INET: 3721 /* 3722 * Fails when rtable is changed while evaluating the ruleset 3723 * The socket looked up will not match the one hit in the end. 3724 */ 3725 inp = in_pcblookup(tb, saddr->v4, sport, daddr->v4, dport, 3726 pd->rdomain); 3727 if (inp == NULL) { 3728 inp = in_pcblookup_listen(tb, daddr->v4, dport, 3729 NULL, pd->rdomain); 3730 if (inp == NULL) 3731 return (-1); 3732 } 3733 break; 3734 #ifdef INET6 3735 case AF_INET6: 3736 inp = in6_pcblookup(tb, &saddr->v6, sport, &daddr->v6, 3737 dport, pd->rdomain); 3738 if (inp == NULL) { 3739 inp = in6_pcblookup_listen(tb, &daddr->v6, dport, 3740 NULL, pd->rdomain); 3741 if (inp == NULL) 3742 return (-1); 3743 } 3744 break; 3745 #endif /* INET6 */ 3746 default: 3747 unhandled_af(pd->af); 3748 } 3749 pd->lookup.uid = inp->inp_socket->so_euid; 3750 pd->lookup.gid = inp->inp_socket->so_egid; 3751 pd->lookup.pid = inp->inp_socket->so_cpid; 3752 in_pcbunref(inp); 3753 return (1); 3754 } 3755 3756 /* post: r => (r[0] == type /\ r[1] >= min_typelen >= 2 "validity" 3757 * /\ (eoh - r) >= min_typelen >= 2 "safety" ) 3758 * 3759 * warning: r + r[1] may exceed opts bounds for r[1] > min_typelen 3760 */ 3761 u_int8_t* 3762 pf_find_tcpopt(u_int8_t *opt, u_int8_t *opts, size_t hlen, u_int8_t type, 3763 u_int8_t min_typelen) 3764 { 3765 u_int8_t *eoh = opts + hlen; 3766 3767 if (min_typelen < 2) 3768 return (NULL); 3769 3770 while ((eoh - opt) >= min_typelen) { 3771 switch (*opt) { 3772 case TCPOPT_EOL: 3773 /* FALLTHROUGH - Workaround the failure of some 3774 systems to NOP-pad their bzero'd option buffers, 3775 producing spurious EOLs */ 3776 case TCPOPT_NOP: 3777 opt++; 3778 continue; 3779 default: 3780 if (opt[0] == type && 3781 opt[1] >= min_typelen) 3782 return (opt); 3783 } 3784 3785 opt += MAX(opt[1], 2); /* evade infinite loops */ 3786 } 3787 3788 return (NULL); 3789 } 3790 3791 u_int8_t 3792 pf_get_wscale(struct pf_pdesc *pd) 3793 { 3794 int olen; 3795 u_int8_t opts[MAX_TCPOPTLEN], *opt; 3796 u_int8_t wscale = 0; 3797 3798 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 3799 if (olen < TCPOLEN_WINDOW || !pf_pull_hdr(pd->m, 3800 pd->off + sizeof(struct tcphdr), opts, olen, NULL, NULL, pd->af)) 3801 return (0); 3802 3803 opt = opts; 3804 while ((opt = pf_find_tcpopt(opt, opts, olen, 3805 TCPOPT_WINDOW, TCPOLEN_WINDOW)) != NULL) { 3806 wscale = opt[2]; 3807 wscale = MIN(wscale, TCP_MAX_WINSHIFT); 3808 wscale |= PF_WSCALE_FLAG; 3809 3810 opt += opt[1]; 3811 } 3812 3813 return (wscale); 3814 } 3815 3816 u_int16_t 3817 pf_get_mss(struct pf_pdesc *pd) 3818 { 3819 int olen; 3820 u_int8_t opts[MAX_TCPOPTLEN], *opt; 3821 u_int16_t mss = tcp_mssdflt; 3822 3823 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 3824 if (olen < TCPOLEN_MAXSEG || !pf_pull_hdr(pd->m, 3825 pd->off + sizeof(struct tcphdr), opts, olen, NULL, NULL, pd->af)) 3826 return (0); 3827 3828 opt = opts; 3829 while ((opt = pf_find_tcpopt(opt, opts, olen, 3830 TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) { 3831 memcpy(&mss, (opt + 2), 2); 3832 mss = ntohs(mss); 3833 3834 opt += opt[1]; 3835 } 3836 return (mss); 3837 } 3838 3839 u_int16_t 3840 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 3841 { 3842 struct ifnet *ifp; 3843 struct sockaddr_in *dst; 3844 #ifdef INET6 3845 struct sockaddr_in6 *dst6; 3846 #endif /* INET6 */ 3847 struct rtentry *rt = NULL; 3848 struct sockaddr_storage ss; 3849 int hlen; 3850 u_int16_t mss = tcp_mssdflt; 3851 3852 memset(&ss, 0, sizeof(ss)); 3853 3854 switch (af) { 3855 case AF_INET: 3856 hlen = sizeof(struct ip); 3857 dst = (struct sockaddr_in *)&ss; 3858 dst->sin_family = AF_INET; 3859 dst->sin_len = sizeof(*dst); 3860 dst->sin_addr = addr->v4; 3861 rt = rtalloc(sintosa(dst), 0, rtableid); 3862 break; 3863 #ifdef INET6 3864 case AF_INET6: 3865 hlen = sizeof(struct ip6_hdr); 3866 dst6 = (struct sockaddr_in6 *)&ss; 3867 dst6->sin6_family = AF_INET6; 3868 dst6->sin6_len = sizeof(*dst6); 3869 dst6->sin6_addr = addr->v6; 3870 rt = rtalloc(sin6tosa(dst6), 0, rtableid); 3871 break; 3872 #endif /* INET6 */ 3873 } 3874 3875 if (rt != NULL && (ifp = if_get(rt->rt_ifidx)) != NULL) { 3876 mss = ifp->if_mtu - hlen - sizeof(struct tcphdr); 3877 mss = max(tcp_mssdflt, mss); 3878 if_put(ifp); 3879 } 3880 rtfree(rt); 3881 mss = min(mss, offer); 3882 mss = max(mss, 64); /* sanity - at least max opt space */ 3883 return (mss); 3884 } 3885 3886 static __inline int 3887 pf_set_rt_ifp(struct pf_state *st, struct pf_addr *saddr, sa_family_t af, 3888 struct pf_src_node **sns) 3889 { 3890 struct pf_rule *r = st->rule.ptr; 3891 int rv; 3892 3893 if (!r->rt) 3894 return (0); 3895 3896 rv = pf_map_addr(af, r, saddr, &st->rt_addr, NULL, sns, 3897 &r->route, PF_SN_ROUTE); 3898 if (rv == 0) 3899 st->rt = r->rt; 3900 3901 return (rv); 3902 } 3903 3904 u_int32_t 3905 pf_tcp_iss(struct pf_pdesc *pd) 3906 { 3907 SHA2_CTX ctx; 3908 union { 3909 uint8_t bytes[SHA512_DIGEST_LENGTH]; 3910 uint32_t words[1]; 3911 } digest; 3912 3913 if (pf_tcp_secret_init == 0) { 3914 arc4random_buf(pf_tcp_secret, sizeof(pf_tcp_secret)); 3915 SHA512Init(&pf_tcp_secret_ctx); 3916 SHA512Update(&pf_tcp_secret_ctx, pf_tcp_secret, 3917 sizeof(pf_tcp_secret)); 3918 pf_tcp_secret_init = 1; 3919 } 3920 ctx = pf_tcp_secret_ctx; 3921 3922 SHA512Update(&ctx, &pd->rdomain, sizeof(pd->rdomain)); 3923 SHA512Update(&ctx, &pd->hdr.tcp.th_sport, sizeof(u_short)); 3924 SHA512Update(&ctx, &pd->hdr.tcp.th_dport, sizeof(u_short)); 3925 switch (pd->af) { 3926 case AF_INET: 3927 SHA512Update(&ctx, &pd->src->v4, sizeof(struct in_addr)); 3928 SHA512Update(&ctx, &pd->dst->v4, sizeof(struct in_addr)); 3929 break; 3930 #ifdef INET6 3931 case AF_INET6: 3932 SHA512Update(&ctx, &pd->src->v6, sizeof(struct in6_addr)); 3933 SHA512Update(&ctx, &pd->dst->v6, sizeof(struct in6_addr)); 3934 break; 3935 #endif /* INET6 */ 3936 } 3937 SHA512Final(digest.bytes, &ctx); 3938 pf_tcp_iss_off += 4096; 3939 return (digest.words[0] + READ_ONCE(tcp_iss) + pf_tcp_iss_off); 3940 } 3941 3942 void 3943 pf_rule_to_actions(struct pf_rule *r, struct pf_rule_actions *a) 3944 { 3945 if (r->qid) 3946 a->qid = r->qid; 3947 if (r->pqid) 3948 a->pqid = r->pqid; 3949 if (r->rtableid >= 0) 3950 a->rtableid = r->rtableid; 3951 #if NPFLOG > 0 3952 a->log |= r->log; 3953 #endif /* NPFLOG > 0 */ 3954 if (r->scrub_flags & PFSTATE_SETTOS) 3955 a->set_tos = r->set_tos; 3956 if (r->min_ttl) 3957 a->min_ttl = r->min_ttl; 3958 if (r->max_mss) 3959 a->max_mss = r->max_mss; 3960 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 3961 PFSTATE_SETTOS|PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 3962 if (r->scrub_flags & PFSTATE_SETPRIO) { 3963 a->set_prio[0] = r->set_prio[0]; 3964 a->set_prio[1] = r->set_prio[1]; 3965 } 3966 if (r->rule_flag & PFRULE_SETDELAY) 3967 a->delay = r->delay; 3968 } 3969 3970 #define PF_TEST_ATTRIB(t, a) \ 3971 if (t) { \ 3972 r = a; \ 3973 continue; \ 3974 } else do { \ 3975 } while (0) 3976 3977 enum pf_test_status 3978 pf_match_rule(struct pf_test_ctx *ctx, struct pf_ruleset *ruleset) 3979 { 3980 struct pf_rule *r; 3981 struct pf_anchor *child = NULL; 3982 int target; 3983 3984 pf_anchor_stack_init(); 3985 enter_ruleset: 3986 r = TAILQ_FIRST(ruleset->rules.active.ptr); 3987 while (r != NULL) { 3988 PF_TEST_ATTRIB(r->rule_flag & PFRULE_EXPIRED, 3989 TAILQ_NEXT(r, entries)); 3990 r->evaluations++; 3991 PF_TEST_ATTRIB( 3992 (pfi_kif_match(r->kif, ctx->pd->kif) == r->ifnot), 3993 r->skip[PF_SKIP_IFP].ptr); 3994 PF_TEST_ATTRIB((r->direction && r->direction != ctx->pd->dir), 3995 r->skip[PF_SKIP_DIR].ptr); 3996 PF_TEST_ATTRIB((r->onrdomain >= 0 && 3997 (r->onrdomain == ctx->pd->rdomain) == r->ifnot), 3998 r->skip[PF_SKIP_RDOM].ptr); 3999 PF_TEST_ATTRIB((r->af && r->af != ctx->pd->af), 4000 r->skip[PF_SKIP_AF].ptr); 4001 PF_TEST_ATTRIB((r->proto && r->proto != ctx->pd->proto), 4002 r->skip[PF_SKIP_PROTO].ptr); 4003 PF_TEST_ATTRIB((PF_MISMATCHAW(&r->src.addr, &ctx->pd->nsaddr, 4004 ctx->pd->naf, r->src.neg, ctx->pd->kif, 4005 ctx->act.rtableid)), 4006 r->skip[PF_SKIP_SRC_ADDR].ptr); 4007 PF_TEST_ATTRIB((PF_MISMATCHAW(&r->dst.addr, &ctx->pd->ndaddr, 4008 ctx->pd->af, r->dst.neg, NULL, ctx->act.rtableid)), 4009 r->skip[PF_SKIP_DST_ADDR].ptr); 4010 4011 switch (ctx->pd->virtual_proto) { 4012 case PF_VPROTO_FRAGMENT: 4013 /* tcp/udp only. port_op always 0 in other cases */ 4014 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 4015 TAILQ_NEXT(r, entries)); 4016 PF_TEST_ATTRIB((ctx->pd->proto == IPPROTO_TCP && 4017 r->flagset), 4018 TAILQ_NEXT(r, entries)); 4019 /* icmp only. type/code always 0 in other cases */ 4020 PF_TEST_ATTRIB((r->type || r->code), 4021 TAILQ_NEXT(r, entries)); 4022 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 4023 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 4024 TAILQ_NEXT(r, entries)); 4025 break; 4026 4027 case IPPROTO_TCP: 4028 PF_TEST_ATTRIB(((r->flagset & ctx->th->th_flags) != 4029 r->flags), 4030 TAILQ_NEXT(r, entries)); 4031 PF_TEST_ATTRIB((r->os_fingerprint != PF_OSFP_ANY && 4032 !pf_osfp_match(pf_osfp_fingerprint(ctx->pd), 4033 r->os_fingerprint)), 4034 TAILQ_NEXT(r, entries)); 4035 /* FALLTHROUGH */ 4036 4037 case IPPROTO_UDP: 4038 /* tcp/udp only. port_op always 0 in other cases */ 4039 PF_TEST_ATTRIB((r->src.port_op && 4040 !pf_match_port(r->src.port_op, r->src.port[0], 4041 r->src.port[1], ctx->pd->nsport)), 4042 r->skip[PF_SKIP_SRC_PORT].ptr); 4043 PF_TEST_ATTRIB((r->dst.port_op && 4044 !pf_match_port(r->dst.port_op, r->dst.port[0], 4045 r->dst.port[1], ctx->pd->ndport)), 4046 r->skip[PF_SKIP_DST_PORT].ptr); 4047 /* tcp/udp only. uid.op always 0 in other cases */ 4048 PF_TEST_ATTRIB((r->uid.op && (ctx->pd->lookup.done || 4049 (ctx->pd->lookup.done = 4050 pf_socket_lookup(ctx->pd), 1)) && 4051 !pf_match_uid(r->uid.op, r->uid.uid[0], 4052 r->uid.uid[1], ctx->pd->lookup.uid)), 4053 TAILQ_NEXT(r, entries)); 4054 /* tcp/udp only. gid.op always 0 in other cases */ 4055 PF_TEST_ATTRIB((r->gid.op && (ctx->pd->lookup.done || 4056 (ctx->pd->lookup.done = 4057 pf_socket_lookup(ctx->pd), 1)) && 4058 !pf_match_gid(r->gid.op, r->gid.gid[0], 4059 r->gid.gid[1], ctx->pd->lookup.gid)), 4060 TAILQ_NEXT(r, entries)); 4061 break; 4062 4063 case IPPROTO_ICMP: 4064 case IPPROTO_ICMPV6: 4065 /* icmp only. type always 0 in other cases */ 4066 PF_TEST_ATTRIB((r->type && 4067 r->type != ctx->icmptype + 1), 4068 TAILQ_NEXT(r, entries)); 4069 /* icmp only. type always 0 in other cases */ 4070 PF_TEST_ATTRIB((r->code && 4071 r->code != ctx->icmpcode + 1), 4072 TAILQ_NEXT(r, entries)); 4073 /* icmp only. don't create states on replies */ 4074 PF_TEST_ATTRIB((r->keep_state && !ctx->state_icmp && 4075 (r->rule_flag & PFRULE_STATESLOPPY) == 0 && 4076 ctx->icmp_dir != PF_IN), 4077 TAILQ_NEXT(r, entries)); 4078 break; 4079 4080 default: 4081 break; 4082 } 4083 4084 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 4085 ctx->pd->virtual_proto != PF_VPROTO_FRAGMENT), 4086 TAILQ_NEXT(r, entries)); 4087 PF_TEST_ATTRIB((r->tos && !(r->tos == ctx->pd->tos)), 4088 TAILQ_NEXT(r, entries)); 4089 PF_TEST_ATTRIB((r->prob && 4090 r->prob <= arc4random_uniform(UINT_MAX - 1) + 1), 4091 TAILQ_NEXT(r, entries)); 4092 PF_TEST_ATTRIB((r->match_tag && 4093 !pf_match_tag(ctx->pd->m, r, &ctx->tag)), 4094 TAILQ_NEXT(r, entries)); 4095 PF_TEST_ATTRIB((r->rcv_kif && pf_match_rcvif(ctx->pd->m, r) == 4096 r->rcvifnot), 4097 TAILQ_NEXT(r, entries)); 4098 PF_TEST_ATTRIB((r->prio && 4099 (r->prio == PF_PRIO_ZERO ? 0 : r->prio) != 4100 ctx->pd->m->m_pkthdr.pf.prio), 4101 TAILQ_NEXT(r, entries)); 4102 4103 /* must be last! */ 4104 if (r->pktrate.limit) { 4105 pf_add_threshold(&r->pktrate); 4106 PF_TEST_ATTRIB((pf_check_threshold(&r->pktrate)), 4107 TAILQ_NEXT(r, entries)); 4108 } 4109 4110 /* FALLTHROUGH */ 4111 if (r->tag) 4112 ctx->tag = r->tag; 4113 if (r->anchor == NULL) { 4114 4115 if (r->rule_flag & PFRULE_ONCE) { 4116 u_int32_t rule_flag; 4117 4118 rule_flag = r->rule_flag; 4119 if (((rule_flag & PFRULE_EXPIRED) == 0) && 4120 atomic_cas_uint(&r->rule_flag, rule_flag, 4121 rule_flag | PFRULE_EXPIRED) == rule_flag) { 4122 r->exptime = gettime(); 4123 } else { 4124 r = TAILQ_NEXT(r, entries); 4125 continue; 4126 } 4127 } 4128 4129 if (r->action == PF_MATCH) { 4130 if ((ctx->ri = pool_get(&pf_rule_item_pl, 4131 PR_NOWAIT)) == NULL) { 4132 REASON_SET(&ctx->reason, PFRES_MEMORY); 4133 return (PF_TEST_FAIL); 4134 } 4135 ctx->ri->r = r; 4136 /* order is irrelevant */ 4137 SLIST_INSERT_HEAD(&ctx->rules, ctx->ri, entry); 4138 ctx->ri = NULL; 4139 pf_rule_to_actions(r, &ctx->act); 4140 if (r->rule_flag & PFRULE_AFTO) 4141 ctx->pd->naf = r->naf; 4142 if (pf_get_transaddr(r, ctx->pd, ctx->sns, 4143 &ctx->nr) == -1) { 4144 REASON_SET(&ctx->reason, 4145 PFRES_TRANSLATE); 4146 return (PF_TEST_FAIL); 4147 } 4148 #if NPFLOG > 0 4149 if (r->log) { 4150 REASON_SET(&ctx->reason, PFRES_MATCH); 4151 pflog_packet(ctx->pd, ctx->reason, r, 4152 ctx->a, ruleset, NULL); 4153 } 4154 #endif /* NPFLOG > 0 */ 4155 } else { 4156 /* 4157 * found matching r 4158 */ 4159 *ctx->rm = r; 4160 /* 4161 * anchor, with ruleset, where r belongs to 4162 */ 4163 *ctx->am = ctx->a; 4164 /* 4165 * ruleset where r belongs to 4166 */ 4167 *ctx->rsm = ruleset; 4168 /* 4169 * ruleset, where anchor belongs to. 4170 */ 4171 ctx->arsm = ctx->aruleset; 4172 } 4173 4174 #if NPFLOG > 0 4175 if (ctx->act.log & PF_LOG_MATCHES) 4176 pf_log_matches(ctx->pd, r, ctx->a, ruleset, 4177 &ctx->rules); 4178 #endif /* NPFLOG > 0 */ 4179 4180 if (r->quick) 4181 return (PF_TEST_QUICK); 4182 } else { 4183 ctx->a = r; 4184 ctx->aruleset = &r->anchor->ruleset; 4185 if (r->anchor_wildcard) { 4186 RB_FOREACH(child, pf_anchor_node, 4187 &r->anchor->children) { 4188 if (pf_anchor_stack_push(ruleset, r, child, 4189 PF_NEXT_CHILD) != 0) 4190 return (PF_TEST_FAIL); 4191 4192 ruleset = &child->ruleset; 4193 goto enter_ruleset; 4194 next_child: 4195 continue; /* with RB_FOREACH() */ 4196 } 4197 } else { 4198 if (pf_anchor_stack_push(ruleset, r, child, 4199 PF_NEXT_RULE) != 0) 4200 return (PF_TEST_FAIL); 4201 4202 ruleset = &r->anchor->ruleset; 4203 child = NULL; 4204 goto enter_ruleset; 4205 next_rule: 4206 ; 4207 } 4208 } 4209 r = TAILQ_NEXT(r, entries); 4210 } 4211 4212 if (pf_anchor_stack_pop(&ruleset, &r, &child, &target) == 0) { 4213 switch (target) { 4214 case PF_NEXT_CHILD: 4215 goto next_child; 4216 case PF_NEXT_RULE: 4217 goto next_rule; 4218 default: 4219 panic("%s: unknown jump target", __func__); 4220 } 4221 } 4222 4223 return (PF_TEST_OK); 4224 } 4225 4226 int 4227 pf_test_rule(struct pf_pdesc *pd, struct pf_rule **rm, struct pf_state **sm, 4228 struct pf_rule **am, struct pf_ruleset **rsm, u_short *reason, 4229 struct pfsync_deferral **pdeferral) 4230 { 4231 struct pf_rule *r = NULL; 4232 struct pf_rule *a = NULL; 4233 struct pf_ruleset *ruleset = NULL; 4234 struct pf_state_key *skw = NULL, *sks = NULL; 4235 int rewrite = 0; 4236 u_int16_t virtual_type, virtual_id; 4237 int action = PF_DROP; 4238 struct pf_test_ctx ctx; 4239 int rv; 4240 4241 memset(&ctx, 0, sizeof(ctx)); 4242 ctx.pd = pd; 4243 ctx.rm = rm; 4244 ctx.am = am; 4245 ctx.rsm = rsm; 4246 ctx.th = &pd->hdr.tcp; 4247 ctx.act.rtableid = pd->rdomain; 4248 ctx.tag = -1; 4249 SLIST_INIT(&ctx.rules); 4250 4251 if (pd->dir == PF_IN && if_congested()) { 4252 REASON_SET(&ctx.reason, PFRES_CONGEST); 4253 return (PF_DROP); 4254 } 4255 4256 switch (pd->virtual_proto) { 4257 case IPPROTO_ICMP: 4258 ctx.icmptype = pd->hdr.icmp.icmp_type; 4259 ctx.icmpcode = pd->hdr.icmp.icmp_code; 4260 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 4261 &ctx.icmp_dir, &virtual_id, &virtual_type); 4262 if (ctx.icmp_dir == PF_IN) { 4263 pd->osport = pd->nsport = virtual_id; 4264 pd->odport = pd->ndport = virtual_type; 4265 } else { 4266 pd->osport = pd->nsport = virtual_type; 4267 pd->odport = pd->ndport = virtual_id; 4268 } 4269 break; 4270 #ifdef INET6 4271 case IPPROTO_ICMPV6: 4272 ctx.icmptype = pd->hdr.icmp6.icmp6_type; 4273 ctx.icmpcode = pd->hdr.icmp6.icmp6_code; 4274 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 4275 &ctx.icmp_dir, &virtual_id, &virtual_type); 4276 if (ctx.icmp_dir == PF_IN) { 4277 pd->osport = pd->nsport = virtual_id; 4278 pd->odport = pd->ndport = virtual_type; 4279 } else { 4280 pd->osport = pd->nsport = virtual_type; 4281 pd->odport = pd->ndport = virtual_id; 4282 } 4283 break; 4284 #endif /* INET6 */ 4285 } 4286 4287 ruleset = &pf_main_ruleset; 4288 rv = pf_match_rule(&ctx, ruleset); 4289 if (rv == PF_TEST_FAIL) { 4290 /* 4291 * Reason has been set in pf_match_rule() already. 4292 */ 4293 goto cleanup; 4294 } 4295 4296 r = *ctx.rm; /* matching rule */ 4297 a = *ctx.am; /* rule that defines an anchor containing 'r' */ 4298 ruleset = *ctx.rsm;/* ruleset of the anchor defined by the rule 'a' */ 4299 ctx.aruleset = ctx.arsm;/* ruleset of the 'a' rule itself */ 4300 4301 /* apply actions for last matching pass/block rule */ 4302 pf_rule_to_actions(r, &ctx.act); 4303 if (r->rule_flag & PFRULE_AFTO) 4304 pd->naf = r->naf; 4305 if (pf_get_transaddr(r, pd, ctx.sns, &ctx.nr) == -1) { 4306 REASON_SET(&ctx.reason, PFRES_TRANSLATE); 4307 goto cleanup; 4308 } 4309 REASON_SET(&ctx.reason, PFRES_MATCH); 4310 4311 #if NPFLOG > 0 4312 if (r->log) 4313 pflog_packet(pd, ctx.reason, r, a, ruleset, NULL); 4314 if (ctx.act.log & PF_LOG_MATCHES) 4315 pf_log_matches(pd, r, a, ruleset, &ctx.rules); 4316 #endif /* NPFLOG > 0 */ 4317 4318 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 4319 (r->action == PF_DROP) && 4320 ((r->rule_flag & PFRULE_RETURNRST) || 4321 (r->rule_flag & PFRULE_RETURNICMP) || 4322 (r->rule_flag & PFRULE_RETURN))) { 4323 if (pd->proto == IPPROTO_TCP && 4324 ((r->rule_flag & PFRULE_RETURNRST) || 4325 (r->rule_flag & PFRULE_RETURN)) && 4326 !(ctx.th->th_flags & TH_RST)) { 4327 u_int32_t ack = 4328 ntohl(ctx.th->th_seq) + pd->p_len; 4329 4330 if (pf_check_tcp_cksum(pd->m, pd->off, 4331 pd->tot_len - pd->off, pd->af)) 4332 REASON_SET(&ctx.reason, PFRES_PROTCKSUM); 4333 else { 4334 if (ctx.th->th_flags & TH_SYN) 4335 ack++; 4336 if (ctx.th->th_flags & TH_FIN) 4337 ack++; 4338 pf_send_tcp(r, pd->af, pd->dst, 4339 pd->src, ctx.th->th_dport, 4340 ctx.th->th_sport, ntohl(ctx.th->th_ack), 4341 ack, TH_RST|TH_ACK, 0, 0, r->return_ttl, 4342 1, 0, pd->rdomain); 4343 } 4344 } else if ((pd->proto != IPPROTO_ICMP || 4345 ICMP_INFOTYPE(ctx.icmptype)) && pd->af == AF_INET && 4346 r->return_icmp) 4347 pf_send_icmp(pd->m, r->return_icmp >> 8, 4348 r->return_icmp & 255, 0, pd->af, r, pd->rdomain); 4349 else if ((pd->proto != IPPROTO_ICMPV6 || 4350 (ctx.icmptype >= ICMP6_ECHO_REQUEST && 4351 ctx.icmptype != ND_REDIRECT)) && pd->af == AF_INET6 && 4352 r->return_icmp6) 4353 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 4354 r->return_icmp6 & 255, 0, pd->af, r, pd->rdomain); 4355 } 4356 4357 if (r->action == PF_DROP) 4358 goto cleanup; 4359 4360 pf_tag_packet(pd->m, ctx.tag, ctx.act.rtableid); 4361 if (ctx.act.rtableid >= 0 && 4362 rtable_l2(ctx.act.rtableid) != pd->rdomain) 4363 pd->destchg = 1; 4364 4365 if (r->action == PF_PASS && pd->badopts != 0 && ! r->allow_opts) { 4366 REASON_SET(&ctx.reason, PFRES_IPOPTIONS); 4367 #if NPFLOG > 0 4368 pd->pflog |= PF_LOG_FORCE; 4369 #endif /* NPFLOG > 0 */ 4370 DPFPRINTF(LOG_NOTICE, "dropping packet with " 4371 "ip/ipv6 options in pf_test_rule()"); 4372 goto cleanup; 4373 } 4374 4375 action = PF_PASS; 4376 4377 if (pd->virtual_proto != PF_VPROTO_FRAGMENT 4378 && !ctx.state_icmp && r->keep_state) { 4379 4380 if (r->rule_flag & PFRULE_SRCTRACK && 4381 pf_insert_src_node(&ctx.sns[PF_SN_NONE], r, PF_SN_NONE, 4382 pd->af, pd->src, NULL, NULL) != 0) { 4383 REASON_SET(&ctx.reason, PFRES_SRCLIMIT); 4384 goto cleanup; 4385 } 4386 4387 if (r->max_states && (r->states_cur >= r->max_states)) { 4388 pf_status.lcounters[LCNT_STATES]++; 4389 REASON_SET(&ctx.reason, PFRES_MAXSTATES); 4390 goto cleanup; 4391 } 4392 4393 action = pf_create_state(pd, r, a, ctx.nr, &skw, &sks, 4394 &rewrite, sm, ctx.tag, &ctx.rules, &ctx.act, ctx.sns); 4395 4396 if (action != PF_PASS) 4397 goto cleanup; 4398 if (sks != skw) { 4399 struct pf_state_key *sk; 4400 4401 if (pd->dir == PF_IN) 4402 sk = sks; 4403 else 4404 sk = skw; 4405 rewrite += pf_translate(pd, 4406 &sk->addr[pd->af == pd->naf ? pd->sidx : pd->didx], 4407 sk->port[pd->af == pd->naf ? pd->sidx : pd->didx], 4408 &sk->addr[pd->af == pd->naf ? pd->didx : pd->sidx], 4409 sk->port[pd->af == pd->naf ? pd->didx : pd->sidx], 4410 virtual_type, ctx.icmp_dir); 4411 } 4412 4413 #ifdef INET6 4414 if (rewrite && skw->af != sks->af) 4415 action = PF_AFRT; 4416 #endif /* INET6 */ 4417 4418 } else { 4419 while ((ctx.ri = SLIST_FIRST(&ctx.rules))) { 4420 SLIST_REMOVE_HEAD(&ctx.rules, entry); 4421 pool_put(&pf_rule_item_pl, ctx.ri); 4422 } 4423 } 4424 4425 /* copy back packet headers if needed */ 4426 if (rewrite && pd->hdrlen) { 4427 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 4428 } 4429 4430 #if NPFSYNC > 0 4431 if (*sm != NULL && !ISSET((*sm)->state_flags, PFSTATE_NOSYNC) && 4432 pd->dir == PF_OUT && pfsync_is_up()) { 4433 /* 4434 * We want the state created, but we dont 4435 * want to send this in case a partner 4436 * firewall has to know about it to allow 4437 * replies through it. 4438 */ 4439 if (pfsync_defer(*sm, pd->m, pdeferral)) 4440 return (PF_DEFER); 4441 } 4442 #endif /* NPFSYNC > 0 */ 4443 4444 return (action); 4445 4446 cleanup: 4447 while ((ctx.ri = SLIST_FIRST(&ctx.rules))) { 4448 SLIST_REMOVE_HEAD(&ctx.rules, entry); 4449 pool_put(&pf_rule_item_pl, ctx.ri); 4450 } 4451 4452 return (action); 4453 } 4454 4455 static __inline int 4456 pf_create_state(struct pf_pdesc *pd, struct pf_rule *r, struct pf_rule *a, 4457 struct pf_rule *nr, struct pf_state_key **skw, struct pf_state_key **sks, 4458 int *rewrite, struct pf_state **sm, int tag, struct pf_rule_slist *rules, 4459 struct pf_rule_actions *act, struct pf_src_node *sns[PF_SN_MAX]) 4460 { 4461 struct pf_state *st = NULL; 4462 struct tcphdr *th = &pd->hdr.tcp; 4463 u_int16_t mss = tcp_mssdflt; 4464 u_short reason; 4465 u_int i; 4466 4467 st = pool_get(&pf_state_pl, PR_NOWAIT | PR_ZERO); 4468 if (st == NULL) { 4469 REASON_SET(&reason, PFRES_MEMORY); 4470 goto csfailed; 4471 } 4472 st->rule.ptr = r; 4473 st->anchor.ptr = a; 4474 st->natrule.ptr = nr; 4475 if (r->allow_opts) 4476 st->state_flags |= PFSTATE_ALLOWOPTS; 4477 if (r->rule_flag & PFRULE_STATESLOPPY) 4478 st->state_flags |= PFSTATE_SLOPPY; 4479 if (r->rule_flag & PFRULE_PFLOW) 4480 st->state_flags |= PFSTATE_PFLOW; 4481 #if NPFLOG > 0 4482 st->log = act->log & PF_LOG_ALL; 4483 #endif /* NPFLOG > 0 */ 4484 st->qid = act->qid; 4485 st->pqid = act->pqid; 4486 st->rtableid[pd->didx] = act->rtableid; 4487 st->rtableid[pd->sidx] = -1; /* return traffic is routed normally */ 4488 st->min_ttl = act->min_ttl; 4489 st->set_tos = act->set_tos; 4490 st->max_mss = act->max_mss; 4491 st->state_flags |= act->flags; 4492 #if NPFSYNC > 0 4493 st->sync_state = PFSYNC_S_NONE; 4494 #endif /* NPFSYNC > 0 */ 4495 st->set_prio[0] = act->set_prio[0]; 4496 st->set_prio[1] = act->set_prio[1]; 4497 st->delay = act->delay; 4498 SLIST_INIT(&st->src_nodes); 4499 /* 4500 * must initialize refcnt, before pf_state_insert() gets called. 4501 * pf_state_inserts() grabs reference for pfsync! 4502 */ 4503 PF_REF_INIT(st->refcnt); 4504 mtx_init(&st->mtx, IPL_NET); 4505 4506 switch (pd->proto) { 4507 case IPPROTO_TCP: 4508 st->src.seqlo = ntohl(th->th_seq); 4509 st->src.seqhi = st->src.seqlo + pd->p_len + 1; 4510 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4511 r->keep_state == PF_STATE_MODULATE) { 4512 /* Generate sequence number modulator */ 4513 st->src.seqdiff = pf_tcp_iss(pd) - st->src.seqlo; 4514 if (st->src.seqdiff == 0) 4515 st->src.seqdiff = 1; 4516 pf_patch_32(pd, &th->th_seq, 4517 htonl(st->src.seqlo + st->src.seqdiff)); 4518 *rewrite = 1; 4519 } else 4520 st->src.seqdiff = 0; 4521 if (th->th_flags & TH_SYN) { 4522 st->src.seqhi++; 4523 st->src.wscale = pf_get_wscale(pd); 4524 } 4525 st->src.max_win = MAX(ntohs(th->th_win), 1); 4526 if (st->src.wscale & PF_WSCALE_MASK) { 4527 /* Remove scale factor from initial window */ 4528 int win = st->src.max_win; 4529 win += 1 << (st->src.wscale & PF_WSCALE_MASK); 4530 st->src.max_win = (win - 1) >> 4531 (st->src.wscale & PF_WSCALE_MASK); 4532 } 4533 if (th->th_flags & TH_FIN) 4534 st->src.seqhi++; 4535 st->dst.seqhi = 1; 4536 st->dst.max_win = 1; 4537 pf_set_protostate(st, PF_PEER_SRC, TCPS_SYN_SENT); 4538 pf_set_protostate(st, PF_PEER_DST, TCPS_CLOSED); 4539 st->timeout = PFTM_TCP_FIRST_PACKET; 4540 pf_status.states_halfopen++; 4541 break; 4542 case IPPROTO_UDP: 4543 pf_set_protostate(st, PF_PEER_SRC, PFUDPS_SINGLE); 4544 pf_set_protostate(st, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 4545 st->timeout = PFTM_UDP_FIRST_PACKET; 4546 break; 4547 case IPPROTO_ICMP: 4548 #ifdef INET6 4549 case IPPROTO_ICMPV6: 4550 #endif /* INET6 */ 4551 st->timeout = PFTM_ICMP_FIRST_PACKET; 4552 break; 4553 default: 4554 pf_set_protostate(st, PF_PEER_SRC, PFOTHERS_SINGLE); 4555 pf_set_protostate(st, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 4556 st->timeout = PFTM_OTHER_FIRST_PACKET; 4557 } 4558 4559 st->creation = getuptime(); 4560 st->expire = getuptime(); 4561 4562 if (pd->proto == IPPROTO_TCP) { 4563 if (st->state_flags & PFSTATE_SCRUB_TCP && 4564 pf_normalize_tcp_init(pd, &st->src)) { 4565 REASON_SET(&reason, PFRES_MEMORY); 4566 goto csfailed; 4567 } 4568 if (st->state_flags & PFSTATE_SCRUB_TCP && st->src.scrub && 4569 pf_normalize_tcp_stateful(pd, &reason, st, 4570 &st->src, &st->dst, rewrite)) { 4571 /* This really shouldn't happen!!! */ 4572 DPFPRINTF(LOG_ERR, 4573 "%s: tcp normalize failed on first pkt", __func__); 4574 goto csfailed; 4575 } 4576 } 4577 st->direction = pd->dir; 4578 4579 if (pf_state_key_setup(pd, skw, sks, act->rtableid)) { 4580 REASON_SET(&reason, PFRES_MEMORY); 4581 goto csfailed; 4582 } 4583 4584 if (pf_set_rt_ifp(st, pd->src, (*skw)->af, sns) != 0) { 4585 REASON_SET(&reason, PFRES_NOROUTE); 4586 goto csfailed; 4587 } 4588 4589 for (i = 0; i < PF_SN_MAX; i++) 4590 if (sns[i] != NULL) { 4591 struct pf_sn_item *sni; 4592 4593 sni = pool_get(&pf_sn_item_pl, PR_NOWAIT); 4594 if (sni == NULL) { 4595 REASON_SET(&reason, PFRES_MEMORY); 4596 goto csfailed; 4597 } 4598 sni->sn = sns[i]; 4599 SLIST_INSERT_HEAD(&st->src_nodes, sni, next); 4600 sni->sn->states++; 4601 } 4602 4603 if (pf_state_insert(BOUND_IFACE(r, pd->kif), skw, sks, st)) { 4604 *sks = *skw = NULL; 4605 REASON_SET(&reason, PFRES_STATEINS); 4606 goto csfailed; 4607 } else 4608 *sm = st; 4609 4610 /* 4611 * Make state responsible for rules it binds here. 4612 */ 4613 memcpy(&st->match_rules, rules, sizeof(st->match_rules)); 4614 memset(rules, 0, sizeof(*rules)); 4615 STATE_INC_COUNTERS(st); 4616 4617 if (tag > 0) { 4618 pf_tag_ref(tag); 4619 st->tag = tag; 4620 } 4621 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 4622 TH_SYN && r->keep_state == PF_STATE_SYNPROXY && pd->dir == PF_IN) { 4623 int rtid = pd->rdomain; 4624 if (act->rtableid >= 0) 4625 rtid = act->rtableid; 4626 pf_set_protostate(st, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 4627 st->src.seqhi = arc4random(); 4628 /* Find mss option */ 4629 mss = pf_get_mss(pd); 4630 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 4631 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 4632 st->src.mss = mss; 4633 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 4634 th->th_sport, st->src.seqhi, ntohl(th->th_seq) + 1, 4635 TH_SYN|TH_ACK, 0, st->src.mss, 0, 1, 0, pd->rdomain); 4636 REASON_SET(&reason, PFRES_SYNPROXY); 4637 return (PF_SYNPROXY_DROP); 4638 } 4639 4640 return (PF_PASS); 4641 4642 csfailed: 4643 if (st) { 4644 pf_normalize_tcp_cleanup(st); /* safe even w/o init */ 4645 pf_src_tree_remove_state(st); 4646 pool_put(&pf_state_pl, st); 4647 } 4648 4649 for (i = 0; i < PF_SN_MAX; i++) 4650 if (sns[i] != NULL) 4651 pf_remove_src_node(sns[i]); 4652 4653 return (PF_DROP); 4654 } 4655 4656 int 4657 pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport, 4658 struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type, 4659 int icmp_dir) 4660 { 4661 int rewrite = 0; 4662 int afto = pd->af != pd->naf; 4663 4664 if (afto || PF_ANEQ(daddr, pd->dst, pd->af)) 4665 pd->destchg = 1; 4666 4667 switch (pd->proto) { 4668 case IPPROTO_TCP: /* FALLTHROUGH */ 4669 case IPPROTO_UDP: 4670 rewrite += pf_patch_16(pd, pd->sport, sport); 4671 rewrite += pf_patch_16(pd, pd->dport, dport); 4672 break; 4673 4674 case IPPROTO_ICMP: 4675 if (pd->af != AF_INET) 4676 return (0); 4677 4678 #ifdef INET6 4679 if (afto) { 4680 if (pf_translate_icmp_af(pd, AF_INET6, &pd->hdr.icmp)) 4681 return (0); 4682 pd->proto = IPPROTO_ICMPV6; 4683 rewrite = 1; 4684 } 4685 #endif /* INET6 */ 4686 if (virtual_type == htons(ICMP_ECHO)) { 4687 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 4688 rewrite += pf_patch_16(pd, 4689 &pd->hdr.icmp.icmp_id, icmpid); 4690 } 4691 break; 4692 4693 #ifdef INET6 4694 case IPPROTO_ICMPV6: 4695 if (pd->af != AF_INET6) 4696 return (0); 4697 4698 if (afto) { 4699 if (pf_translate_icmp_af(pd, AF_INET, &pd->hdr.icmp6)) 4700 return (0); 4701 pd->proto = IPPROTO_ICMP; 4702 rewrite = 1; 4703 } 4704 if (virtual_type == htons(ICMP6_ECHO_REQUEST)) { 4705 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 4706 rewrite += pf_patch_16(pd, 4707 &pd->hdr.icmp6.icmp6_id, icmpid); 4708 } 4709 break; 4710 #endif /* INET6 */ 4711 } 4712 4713 if (!afto) { 4714 rewrite += pf_translate_a(pd, pd->src, saddr); 4715 rewrite += pf_translate_a(pd, pd->dst, daddr); 4716 } 4717 4718 return (rewrite); 4719 } 4720 4721 int 4722 pf_tcp_track_full(struct pf_pdesc *pd, struct pf_state **stp, u_short *reason, 4723 int *copyback, int reverse) 4724 { 4725 struct tcphdr *th = &pd->hdr.tcp; 4726 struct pf_state_peer *src, *dst; 4727 u_int16_t win = ntohs(th->th_win); 4728 u_int32_t ack, end, data_end, seq, orig_seq; 4729 u_int8_t sws, dws, psrc, pdst; 4730 int ackskew; 4731 4732 if ((pd->dir == (*stp)->direction && !reverse) || 4733 (pd->dir != (*stp)->direction && reverse)) { 4734 src = &(*stp)->src; 4735 dst = &(*stp)->dst; 4736 psrc = PF_PEER_SRC; 4737 pdst = PF_PEER_DST; 4738 } else { 4739 src = &(*stp)->dst; 4740 dst = &(*stp)->src; 4741 psrc = PF_PEER_DST; 4742 pdst = PF_PEER_SRC; 4743 } 4744 4745 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 4746 sws = src->wscale & PF_WSCALE_MASK; 4747 dws = dst->wscale & PF_WSCALE_MASK; 4748 } else 4749 sws = dws = 0; 4750 4751 /* 4752 * Sequence tracking algorithm from Guido van Rooij's paper: 4753 * http://www.madison-gurkha.com/publications/tcp_filtering/ 4754 * tcp_filtering.ps 4755 */ 4756 4757 orig_seq = seq = ntohl(th->th_seq); 4758 if (src->seqlo == 0) { 4759 /* First packet from this end. Set its state */ 4760 4761 if (((*stp)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 4762 src->scrub == NULL) { 4763 if (pf_normalize_tcp_init(pd, src)) { 4764 REASON_SET(reason, PFRES_MEMORY); 4765 return (PF_DROP); 4766 } 4767 } 4768 4769 /* Deferred generation of sequence number modulator */ 4770 if (dst->seqdiff && !src->seqdiff) { 4771 /* use random iss for the TCP server */ 4772 while ((src->seqdiff = arc4random() - seq) == 0) 4773 continue; 4774 ack = ntohl(th->th_ack) - dst->seqdiff; 4775 pf_patch_32(pd, &th->th_seq, htonl(seq + src->seqdiff)); 4776 pf_patch_32(pd, &th->th_ack, htonl(ack)); 4777 *copyback = 1; 4778 } else { 4779 ack = ntohl(th->th_ack); 4780 } 4781 4782 end = seq + pd->p_len; 4783 if (th->th_flags & TH_SYN) { 4784 end++; 4785 if (dst->wscale & PF_WSCALE_FLAG) { 4786 src->wscale = pf_get_wscale(pd); 4787 if (src->wscale & PF_WSCALE_FLAG) { 4788 /* Remove scale factor from initial 4789 * window */ 4790 sws = src->wscale & PF_WSCALE_MASK; 4791 win = ((u_int32_t)win + (1 << sws) - 1) 4792 >> sws; 4793 dws = dst->wscale & PF_WSCALE_MASK; 4794 } else { 4795 /* fixup other window */ 4796 dst->max_win = MIN(TCP_MAXWIN, 4797 (u_int32_t)dst->max_win << 4798 (dst->wscale & PF_WSCALE_MASK)); 4799 /* in case of a retrans SYN|ACK */ 4800 dst->wscale = 0; 4801 } 4802 } 4803 } 4804 data_end = end; 4805 if (th->th_flags & TH_FIN) 4806 end++; 4807 4808 src->seqlo = seq; 4809 if (src->state < TCPS_SYN_SENT) 4810 pf_set_protostate(*stp, psrc, TCPS_SYN_SENT); 4811 4812 /* 4813 * May need to slide the window (seqhi may have been set by 4814 * the crappy stack check or if we picked up the connection 4815 * after establishment) 4816 */ 4817 if (src->seqhi == 1 || 4818 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 4819 src->seqhi = end + MAX(1, dst->max_win << dws); 4820 if (win > src->max_win) 4821 src->max_win = win; 4822 4823 } else { 4824 ack = ntohl(th->th_ack) - dst->seqdiff; 4825 if (src->seqdiff) { 4826 /* Modulate sequence numbers */ 4827 pf_patch_32(pd, &th->th_seq, htonl(seq + src->seqdiff)); 4828 pf_patch_32(pd, &th->th_ack, htonl(ack)); 4829 *copyback = 1; 4830 } 4831 end = seq + pd->p_len; 4832 if (th->th_flags & TH_SYN) 4833 end++; 4834 data_end = end; 4835 if (th->th_flags & TH_FIN) 4836 end++; 4837 } 4838 4839 if ((th->th_flags & TH_ACK) == 0) { 4840 /* Let it pass through the ack skew check */ 4841 ack = dst->seqlo; 4842 } else if ((ack == 0 && 4843 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 4844 /* broken tcp stacks do not set ack */ 4845 (dst->state < TCPS_SYN_SENT)) { 4846 /* 4847 * Many stacks (ours included) will set the ACK number in an 4848 * FIN|ACK if the SYN times out -- no sequence to ACK. 4849 */ 4850 ack = dst->seqlo; 4851 } 4852 4853 if (seq == end) { 4854 /* Ease sequencing restrictions on no data packets */ 4855 seq = src->seqlo; 4856 data_end = end = seq; 4857 } 4858 4859 ackskew = dst->seqlo - ack; 4860 4861 4862 /* 4863 * Need to demodulate the sequence numbers in any TCP SACK options 4864 * (Selective ACK). We could optionally validate the SACK values 4865 * against the current ACK window, either forwards or backwards, but 4866 * I'm not confident that SACK has been implemented properly 4867 * everywhere. It wouldn't surprise me if several stacks accidently 4868 * SACK too far backwards of previously ACKed data. There really aren't 4869 * any security implications of bad SACKing unless the target stack 4870 * doesn't validate the option length correctly. Someone trying to 4871 * spoof into a TCP connection won't bother blindly sending SACK 4872 * options anyway. 4873 */ 4874 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 4875 if (pf_modulate_sack(pd, dst)) 4876 *copyback = 1; 4877 } 4878 4879 4880 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 4881 if (SEQ_GEQ(src->seqhi, data_end) && 4882 /* Last octet inside other's window space */ 4883 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 4884 /* Retrans: not more than one window back */ 4885 (ackskew >= -MAXACKWINDOW) && 4886 /* Acking not more than one reassembled fragment backwards */ 4887 (ackskew <= (MAXACKWINDOW << sws)) && 4888 /* Acking not more than one window forward */ 4889 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 4890 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 4891 /* Require an exact/+1 sequence match on resets when possible */ 4892 4893 if (dst->scrub || src->scrub) { 4894 if (pf_normalize_tcp_stateful(pd, reason, *stp, src, 4895 dst, copyback)) 4896 return (PF_DROP); 4897 } 4898 4899 /* update max window */ 4900 if (src->max_win < win) 4901 src->max_win = win; 4902 /* synchronize sequencing */ 4903 if (SEQ_GT(end, src->seqlo)) 4904 src->seqlo = end; 4905 /* slide the window of what the other end can send */ 4906 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4907 dst->seqhi = ack + MAX((win << sws), 1); 4908 4909 /* update states */ 4910 if (th->th_flags & TH_SYN) 4911 if (src->state < TCPS_SYN_SENT) 4912 pf_set_protostate(*stp, psrc, TCPS_SYN_SENT); 4913 if (th->th_flags & TH_FIN) 4914 if (src->state < TCPS_CLOSING) 4915 pf_set_protostate(*stp, psrc, TCPS_CLOSING); 4916 if (th->th_flags & TH_ACK) { 4917 if (dst->state == TCPS_SYN_SENT) { 4918 pf_set_protostate(*stp, pdst, 4919 TCPS_ESTABLISHED); 4920 if (src->state == TCPS_ESTABLISHED && 4921 !SLIST_EMPTY(&(*stp)->src_nodes) && 4922 pf_src_connlimit(stp)) { 4923 REASON_SET(reason, PFRES_SRCLIMIT); 4924 return (PF_DROP); 4925 } 4926 } else if (dst->state == TCPS_CLOSING) 4927 pf_set_protostate(*stp, pdst, 4928 TCPS_FIN_WAIT_2); 4929 } 4930 if (th->th_flags & TH_RST) 4931 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_TIME_WAIT); 4932 4933 /* update expire time */ 4934 (*stp)->expire = getuptime(); 4935 if (src->state >= TCPS_FIN_WAIT_2 && 4936 dst->state >= TCPS_FIN_WAIT_2) 4937 (*stp)->timeout = PFTM_TCP_CLOSED; 4938 else if (src->state >= TCPS_CLOSING && 4939 dst->state >= TCPS_CLOSING) 4940 (*stp)->timeout = PFTM_TCP_FIN_WAIT; 4941 else if (src->state < TCPS_ESTABLISHED || 4942 dst->state < TCPS_ESTABLISHED) 4943 (*stp)->timeout = PFTM_TCP_OPENING; 4944 else if (src->state >= TCPS_CLOSING || 4945 dst->state >= TCPS_CLOSING) 4946 (*stp)->timeout = PFTM_TCP_CLOSING; 4947 else 4948 (*stp)->timeout = PFTM_TCP_ESTABLISHED; 4949 4950 /* Fall through to PASS packet */ 4951 } else if ((dst->state < TCPS_SYN_SENT || 4952 dst->state >= TCPS_FIN_WAIT_2 || 4953 src->state >= TCPS_FIN_WAIT_2) && 4954 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 4955 /* Within a window forward of the originating packet */ 4956 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 4957 /* Within a window backward of the originating packet */ 4958 4959 /* 4960 * This currently handles three situations: 4961 * 1) Stupid stacks will shotgun SYNs before their peer 4962 * replies. 4963 * 2) When PF catches an already established stream (the 4964 * firewall rebooted, the state table was flushed, routes 4965 * changed...) 4966 * 3) Packets get funky immediately after the connection 4967 * closes (this should catch Solaris spurious ACK|FINs 4968 * that web servers like to spew after a close) 4969 * 4970 * This must be a little more careful than the above code 4971 * since packet floods will also be caught here. We don't 4972 * update the TTL here to mitigate the damage of a packet 4973 * flood and so the same code can handle awkward establishment 4974 * and a loosened connection close. 4975 * In the establishment case, a correct peer response will 4976 * validate the connection, go through the normal state code 4977 * and keep updating the state TTL. 4978 */ 4979 4980 if (pf_status.debug >= LOG_NOTICE) { 4981 log(LOG_NOTICE, "pf: loose state match: "); 4982 pf_print_state(*stp); 4983 pf_print_flags(th->th_flags); 4984 addlog(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4985 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 4986 pd->p_len, ackskew, (*stp)->packets[0], 4987 (*stp)->packets[1], 4988 pd->dir == PF_IN ? "in" : "out", 4989 pd->dir == (*stp)->direction ? "fwd" : "rev"); 4990 } 4991 4992 if (dst->scrub || src->scrub) { 4993 if (pf_normalize_tcp_stateful(pd, reason, *stp, src, 4994 dst, copyback)) 4995 return (PF_DROP); 4996 } 4997 4998 /* update max window */ 4999 if (src->max_win < win) 5000 src->max_win = win; 5001 /* synchronize sequencing */ 5002 if (SEQ_GT(end, src->seqlo)) 5003 src->seqlo = end; 5004 /* slide the window of what the other end can send */ 5005 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5006 dst->seqhi = ack + MAX((win << sws), 1); 5007 5008 /* 5009 * Cannot set dst->seqhi here since this could be a shotgunned 5010 * SYN and not an already established connection. 5011 */ 5012 if (th->th_flags & TH_FIN) 5013 if (src->state < TCPS_CLOSING) 5014 pf_set_protostate(*stp, psrc, TCPS_CLOSING); 5015 if (th->th_flags & TH_RST) 5016 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_TIME_WAIT); 5017 5018 /* Fall through to PASS packet */ 5019 } else { 5020 if ((*stp)->dst.state == TCPS_SYN_SENT && 5021 (*stp)->src.state == TCPS_SYN_SENT) { 5022 /* Send RST for state mismatches during handshake */ 5023 if (!(th->th_flags & TH_RST)) 5024 pf_send_tcp((*stp)->rule.ptr, pd->af, 5025 pd->dst, pd->src, th->th_dport, 5026 th->th_sport, ntohl(th->th_ack), 0, 5027 TH_RST, 0, 0, 5028 (*stp)->rule.ptr->return_ttl, 1, 0, 5029 pd->rdomain); 5030 src->seqlo = 0; 5031 src->seqhi = 1; 5032 src->max_win = 1; 5033 } else if (pf_status.debug >= LOG_NOTICE) { 5034 log(LOG_NOTICE, "pf: BAD state: "); 5035 pf_print_state(*stp); 5036 pf_print_flags(th->th_flags); 5037 addlog(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5038 "pkts=%llu:%llu dir=%s,%s\n", 5039 seq, orig_seq, ack, pd->p_len, ackskew, 5040 (*stp)->packets[0], (*stp)->packets[1], 5041 pd->dir == PF_IN ? "in" : "out", 5042 pd->dir == (*stp)->direction ? "fwd" : "rev"); 5043 addlog("pf: State failure on: %c %c %c %c | %c %c\n", 5044 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 5045 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5046 ' ': '2', 5047 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5048 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5049 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ? 5050 ' ' :'5', 5051 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5052 } 5053 REASON_SET(reason, PFRES_BADSTATE); 5054 return (PF_DROP); 5055 } 5056 5057 return (PF_PASS); 5058 } 5059 5060 int 5061 pf_tcp_track_sloppy(struct pf_pdesc *pd, struct pf_state **stp, 5062 u_short *reason) 5063 { 5064 struct tcphdr *th = &pd->hdr.tcp; 5065 struct pf_state_peer *src, *dst; 5066 u_int8_t psrc, pdst; 5067 5068 if (pd->dir == (*stp)->direction) { 5069 src = &(*stp)->src; 5070 dst = &(*stp)->dst; 5071 psrc = PF_PEER_SRC; 5072 pdst = PF_PEER_DST; 5073 } else { 5074 src = &(*stp)->dst; 5075 dst = &(*stp)->src; 5076 psrc = PF_PEER_DST; 5077 pdst = PF_PEER_SRC; 5078 } 5079 5080 if (th->th_flags & TH_SYN) 5081 if (src->state < TCPS_SYN_SENT) 5082 pf_set_protostate(*stp, psrc, TCPS_SYN_SENT); 5083 if (th->th_flags & TH_FIN) 5084 if (src->state < TCPS_CLOSING) 5085 pf_set_protostate(*stp, psrc, TCPS_CLOSING); 5086 if (th->th_flags & TH_ACK) { 5087 if (dst->state == TCPS_SYN_SENT) { 5088 pf_set_protostate(*stp, pdst, TCPS_ESTABLISHED); 5089 if (src->state == TCPS_ESTABLISHED && 5090 !SLIST_EMPTY(&(*stp)->src_nodes) && 5091 pf_src_connlimit(stp)) { 5092 REASON_SET(reason, PFRES_SRCLIMIT); 5093 return (PF_DROP); 5094 } 5095 } else if (dst->state == TCPS_CLOSING) { 5096 pf_set_protostate(*stp, pdst, TCPS_FIN_WAIT_2); 5097 } else if (src->state == TCPS_SYN_SENT && 5098 dst->state < TCPS_SYN_SENT) { 5099 /* 5100 * Handle a special sloppy case where we only see one 5101 * half of the connection. If there is a ACK after 5102 * the initial SYN without ever seeing a packet from 5103 * the destination, set the connection to established. 5104 */ 5105 pf_set_protostate(*stp, PF_PEER_BOTH, 5106 TCPS_ESTABLISHED); 5107 if (!SLIST_EMPTY(&(*stp)->src_nodes) && 5108 pf_src_connlimit(stp)) { 5109 REASON_SET(reason, PFRES_SRCLIMIT); 5110 return (PF_DROP); 5111 } 5112 } else if (src->state == TCPS_CLOSING && 5113 dst->state == TCPS_ESTABLISHED && 5114 dst->seqlo == 0) { 5115 /* 5116 * Handle the closing of half connections where we 5117 * don't see the full bidirectional FIN/ACK+ACK 5118 * handshake. 5119 */ 5120 pf_set_protostate(*stp, pdst, TCPS_CLOSING); 5121 } 5122 } 5123 if (th->th_flags & TH_RST) 5124 pf_set_protostate(*stp, PF_PEER_BOTH, TCPS_TIME_WAIT); 5125 5126 /* update expire time */ 5127 (*stp)->expire = getuptime(); 5128 if (src->state >= TCPS_FIN_WAIT_2 && 5129 dst->state >= TCPS_FIN_WAIT_2) 5130 (*stp)->timeout = PFTM_TCP_CLOSED; 5131 else if (src->state >= TCPS_CLOSING && 5132 dst->state >= TCPS_CLOSING) 5133 (*stp)->timeout = PFTM_TCP_FIN_WAIT; 5134 else if (src->state < TCPS_ESTABLISHED || 5135 dst->state < TCPS_ESTABLISHED) 5136 (*stp)->timeout = PFTM_TCP_OPENING; 5137 else if (src->state >= TCPS_CLOSING || 5138 dst->state >= TCPS_CLOSING) 5139 (*stp)->timeout = PFTM_TCP_CLOSING; 5140 else 5141 (*stp)->timeout = PFTM_TCP_ESTABLISHED; 5142 5143 return (PF_PASS); 5144 } 5145 5146 static __inline int 5147 pf_synproxy(struct pf_pdesc *pd, struct pf_state **stp, u_short *reason) 5148 { 5149 struct pf_state_key *sk = (*stp)->key[pd->didx]; 5150 5151 if ((*stp)->src.state == PF_TCPS_PROXY_SRC) { 5152 struct tcphdr *th = &pd->hdr.tcp; 5153 5154 if (pd->dir != (*stp)->direction) { 5155 REASON_SET(reason, PFRES_SYNPROXY); 5156 return (PF_SYNPROXY_DROP); 5157 } 5158 if (th->th_flags & TH_SYN) { 5159 if (ntohl(th->th_seq) != (*stp)->src.seqlo) { 5160 REASON_SET(reason, PFRES_SYNPROXY); 5161 return (PF_DROP); 5162 } 5163 pf_send_tcp((*stp)->rule.ptr, pd->af, pd->dst, 5164 pd->src, th->th_dport, th->th_sport, 5165 (*stp)->src.seqhi, ntohl(th->th_seq) + 1, 5166 TH_SYN|TH_ACK, 0, (*stp)->src.mss, 0, 1, 5167 0, pd->rdomain); 5168 REASON_SET(reason, PFRES_SYNPROXY); 5169 return (PF_SYNPROXY_DROP); 5170 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 5171 (ntohl(th->th_ack) != (*stp)->src.seqhi + 1) || 5172 (ntohl(th->th_seq) != (*stp)->src.seqlo + 1)) { 5173 REASON_SET(reason, PFRES_SYNPROXY); 5174 return (PF_DROP); 5175 } else if (!SLIST_EMPTY(&(*stp)->src_nodes) && 5176 pf_src_connlimit(stp)) { 5177 REASON_SET(reason, PFRES_SRCLIMIT); 5178 return (PF_DROP); 5179 } else 5180 pf_set_protostate(*stp, PF_PEER_SRC, 5181 PF_TCPS_PROXY_DST); 5182 } 5183 if ((*stp)->src.state == PF_TCPS_PROXY_DST) { 5184 struct tcphdr *th = &pd->hdr.tcp; 5185 5186 if (pd->dir == (*stp)->direction) { 5187 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 5188 (ntohl(th->th_ack) != (*stp)->src.seqhi + 1) || 5189 (ntohl(th->th_seq) != (*stp)->src.seqlo + 1)) { 5190 REASON_SET(reason, PFRES_SYNPROXY); 5191 return (PF_DROP); 5192 } 5193 (*stp)->src.max_win = MAX(ntohs(th->th_win), 1); 5194 if ((*stp)->dst.seqhi == 1) 5195 (*stp)->dst.seqhi = arc4random(); 5196 pf_send_tcp((*stp)->rule.ptr, pd->af, 5197 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5198 sk->port[pd->sidx], sk->port[pd->didx], 5199 (*stp)->dst.seqhi, 0, TH_SYN, 0, 5200 (*stp)->src.mss, 0, 0, (*stp)->tag, 5201 sk->rdomain); 5202 REASON_SET(reason, PFRES_SYNPROXY); 5203 return (PF_SYNPROXY_DROP); 5204 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 5205 (TH_SYN|TH_ACK)) || 5206 (ntohl(th->th_ack) != (*stp)->dst.seqhi + 1)) { 5207 REASON_SET(reason, PFRES_SYNPROXY); 5208 return (PF_DROP); 5209 } else { 5210 (*stp)->dst.max_win = MAX(ntohs(th->th_win), 1); 5211 (*stp)->dst.seqlo = ntohl(th->th_seq); 5212 pf_send_tcp((*stp)->rule.ptr, pd->af, pd->dst, 5213 pd->src, th->th_dport, th->th_sport, 5214 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 5215 TH_ACK, (*stp)->src.max_win, 0, 0, 0, 5216 (*stp)->tag, pd->rdomain); 5217 pf_send_tcp((*stp)->rule.ptr, pd->af, 5218 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5219 sk->port[pd->sidx], sk->port[pd->didx], 5220 (*stp)->src.seqhi + 1, (*stp)->src.seqlo + 1, 5221 TH_ACK, (*stp)->dst.max_win, 0, 0, 1, 5222 0, sk->rdomain); 5223 (*stp)->src.seqdiff = (*stp)->dst.seqhi - 5224 (*stp)->src.seqlo; 5225 (*stp)->dst.seqdiff = (*stp)->src.seqhi - 5226 (*stp)->dst.seqlo; 5227 (*stp)->src.seqhi = (*stp)->src.seqlo + 5228 (*stp)->dst.max_win; 5229 (*stp)->dst.seqhi = (*stp)->dst.seqlo + 5230 (*stp)->src.max_win; 5231 (*stp)->src.wscale = (*stp)->dst.wscale = 0; 5232 pf_set_protostate(*stp, PF_PEER_BOTH, 5233 TCPS_ESTABLISHED); 5234 REASON_SET(reason, PFRES_SYNPROXY); 5235 return (PF_SYNPROXY_DROP); 5236 } 5237 } 5238 return (PF_PASS); 5239 } 5240 5241 int 5242 pf_test_state(struct pf_pdesc *pd, struct pf_state **stp, u_short *reason) 5243 { 5244 int copyback = 0; 5245 struct pf_state_peer *src, *dst; 5246 int action; 5247 struct inpcb *inp = pd->m->m_pkthdr.pf.inp; 5248 u_int8_t psrc, pdst; 5249 5250 action = PF_PASS; 5251 if (pd->dir == (*stp)->direction) { 5252 src = &(*stp)->src; 5253 dst = &(*stp)->dst; 5254 psrc = PF_PEER_SRC; 5255 pdst = PF_PEER_DST; 5256 } else { 5257 src = &(*stp)->dst; 5258 dst = &(*stp)->src; 5259 psrc = PF_PEER_DST; 5260 pdst = PF_PEER_SRC; 5261 } 5262 5263 switch (pd->virtual_proto) { 5264 case IPPROTO_TCP: 5265 if ((action = pf_synproxy(pd, stp, reason)) != PF_PASS) 5266 return (action); 5267 if ((pd->hdr.tcp.th_flags & (TH_SYN|TH_ACK)) == TH_SYN) { 5268 5269 if (dst->state >= TCPS_FIN_WAIT_2 && 5270 src->state >= TCPS_FIN_WAIT_2) { 5271 if (pf_status.debug >= LOG_NOTICE) { 5272 log(LOG_NOTICE, "pf: state reuse "); 5273 pf_print_state(*stp); 5274 pf_print_flags(pd->hdr.tcp.th_flags); 5275 addlog("\n"); 5276 } 5277 /* XXX make sure it's the same direction ?? */ 5278 (*stp)->timeout = PFTM_PURGE; 5279 pf_state_unref(*stp); 5280 *stp = NULL; 5281 pf_mbuf_link_inpcb(pd->m, inp); 5282 return (PF_DROP); 5283 } else if (dst->state >= TCPS_ESTABLISHED && 5284 src->state >= TCPS_ESTABLISHED) { 5285 /* 5286 * SYN matches existing state??? 5287 * Typically happens when sender boots up after 5288 * sudden panic. Certain protocols (NFSv3) are 5289 * always using same port numbers. Challenge 5290 * ACK enables all parties (firewall and peers) 5291 * to get in sync again. 5292 */ 5293 pf_send_challenge_ack(pd, *stp, src, dst); 5294 return (PF_DROP); 5295 } 5296 } 5297 5298 if ((*stp)->state_flags & PFSTATE_SLOPPY) { 5299 if (pf_tcp_track_sloppy(pd, stp, reason) == PF_DROP) 5300 return (PF_DROP); 5301 } else { 5302 if (pf_tcp_track_full(pd, stp, reason, ©back, 5303 PF_REVERSED_KEY((*stp)->key, pd->af)) == PF_DROP) 5304 return (PF_DROP); 5305 } 5306 break; 5307 case IPPROTO_UDP: 5308 /* update states */ 5309 if (src->state < PFUDPS_SINGLE) 5310 pf_set_protostate(*stp, psrc, PFUDPS_SINGLE); 5311 if (dst->state == PFUDPS_SINGLE) 5312 pf_set_protostate(*stp, pdst, PFUDPS_MULTIPLE); 5313 5314 /* update expire time */ 5315 (*stp)->expire = getuptime(); 5316 if (src->state == PFUDPS_MULTIPLE && 5317 dst->state == PFUDPS_MULTIPLE) 5318 (*stp)->timeout = PFTM_UDP_MULTIPLE; 5319 else 5320 (*stp)->timeout = PFTM_UDP_SINGLE; 5321 break; 5322 default: 5323 /* update states */ 5324 if (src->state < PFOTHERS_SINGLE) 5325 pf_set_protostate(*stp, psrc, PFOTHERS_SINGLE); 5326 if (dst->state == PFOTHERS_SINGLE) 5327 pf_set_protostate(*stp, pdst, PFOTHERS_MULTIPLE); 5328 5329 /* update expire time */ 5330 (*stp)->expire = getuptime(); 5331 if (src->state == PFOTHERS_MULTIPLE && 5332 dst->state == PFOTHERS_MULTIPLE) 5333 (*stp)->timeout = PFTM_OTHER_MULTIPLE; 5334 else 5335 (*stp)->timeout = PFTM_OTHER_SINGLE; 5336 break; 5337 } 5338 5339 /* translate source/destination address, if necessary */ 5340 if ((*stp)->key[PF_SK_WIRE] != (*stp)->key[PF_SK_STACK]) { 5341 struct pf_state_key *nk; 5342 int afto, sidx, didx; 5343 5344 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5345 nk = (*stp)->key[pd->sidx]; 5346 else 5347 nk = (*stp)->key[pd->didx]; 5348 5349 afto = pd->af != nk->af; 5350 sidx = afto ? pd->didx : pd->sidx; 5351 didx = afto ? pd->sidx : pd->didx; 5352 5353 #ifdef INET6 5354 if (afto) { 5355 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], nk->af); 5356 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], nk->af); 5357 pd->naf = nk->af; 5358 action = PF_AFRT; 5359 } 5360 #endif /* INET6 */ 5361 5362 if (!afto) 5363 pf_translate_a(pd, pd->src, &nk->addr[sidx]); 5364 5365 if (pd->sport != NULL) 5366 pf_patch_16(pd, pd->sport, nk->port[sidx]); 5367 5368 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 5369 pd->rdomain != nk->rdomain) 5370 pd->destchg = 1; 5371 5372 if (!afto) 5373 pf_translate_a(pd, pd->dst, &nk->addr[didx]); 5374 5375 if (pd->dport != NULL) 5376 pf_patch_16(pd, pd->dport, nk->port[didx]); 5377 5378 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5379 copyback = 1; 5380 } 5381 5382 if (copyback && pd->hdrlen > 0) { 5383 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 5384 } 5385 5386 return (action); 5387 } 5388 5389 int 5390 pf_icmp_state_lookup(struct pf_pdesc *pd, struct pf_state_key_cmp *key, 5391 struct pf_state **stp, u_int16_t icmpid, u_int16_t type, 5392 int icmp_dir, int *iidx, int multi, int inner) 5393 { 5394 int direction, action; 5395 5396 key->af = pd->af; 5397 key->proto = pd->proto; 5398 key->rdomain = pd->rdomain; 5399 if (icmp_dir == PF_IN) { 5400 *iidx = pd->sidx; 5401 key->port[pd->sidx] = icmpid; 5402 key->port[pd->didx] = type; 5403 } else { 5404 *iidx = pd->didx; 5405 key->port[pd->sidx] = type; 5406 key->port[pd->didx] = icmpid; 5407 } 5408 5409 if (pf_state_key_addr_setup(pd, key, pd->sidx, pd->src, pd->didx, 5410 pd->dst, pd->af, multi)) 5411 return (PF_DROP); 5412 5413 key->hash = pf_pkt_hash(key->af, key->proto, 5414 &key->addr[0], &key->addr[1], 0, 0); 5415 5416 action = pf_find_state(pd, key, stp); 5417 if (action != PF_MATCH) 5418 return (action); 5419 5420 if ((*stp)->state_flags & PFSTATE_SLOPPY) 5421 return (-1); 5422 5423 /* Is this ICMP message flowing in right direction? */ 5424 if ((*stp)->key[PF_SK_WIRE]->af != (*stp)->key[PF_SK_STACK]->af) 5425 direction = (pd->af == (*stp)->key[PF_SK_WIRE]->af) ? 5426 PF_IN : PF_OUT; 5427 else 5428 direction = (*stp)->direction; 5429 if ((((!inner && direction == pd->dir) || 5430 (inner && direction != pd->dir)) ? 5431 PF_IN : PF_OUT) != icmp_dir) { 5432 if (pf_status.debug >= LOG_NOTICE) { 5433 log(LOG_NOTICE, 5434 "pf: icmp type %d in wrong direction (%d): ", 5435 ntohs(type), icmp_dir); 5436 pf_print_state(*stp); 5437 addlog("\n"); 5438 } 5439 return (PF_DROP); 5440 } 5441 return (-1); 5442 } 5443 5444 int 5445 pf_test_state_icmp(struct pf_pdesc *pd, struct pf_state **stp, 5446 u_short *reason) 5447 { 5448 u_int16_t virtual_id, virtual_type; 5449 u_int8_t icmptype, icmpcode; 5450 int icmp_dir, iidx, ret, copyback = 0; 5451 5452 struct pf_state_key_cmp key; 5453 5454 switch (pd->proto) { 5455 case IPPROTO_ICMP: 5456 icmptype = pd->hdr.icmp.icmp_type; 5457 icmpcode = pd->hdr.icmp.icmp_code; 5458 break; 5459 #ifdef INET6 5460 case IPPROTO_ICMPV6: 5461 icmptype = pd->hdr.icmp6.icmp6_type; 5462 icmpcode = pd->hdr.icmp6.icmp6_code; 5463 break; 5464 #endif /* INET6 */ 5465 default: 5466 panic("unhandled proto %d", pd->proto); 5467 } 5468 5469 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &virtual_id, 5470 &virtual_type) == 0) { 5471 /* 5472 * ICMP query/reply message not related to a TCP/UDP packet. 5473 * Search for an ICMP state. 5474 */ 5475 ret = pf_icmp_state_lookup(pd, &key, stp, 5476 virtual_id, virtual_type, icmp_dir, &iidx, 5477 0, 0); 5478 /* IPv6? try matching a multicast address */ 5479 if (ret == PF_DROP && pd->af == AF_INET6 && icmp_dir == PF_OUT) 5480 ret = pf_icmp_state_lookup(pd, &key, stp, virtual_id, 5481 virtual_type, icmp_dir, &iidx, 1, 0); 5482 if (ret >= 0) 5483 return (ret); 5484 5485 (*stp)->expire = getuptime(); 5486 (*stp)->timeout = PFTM_ICMP_ERROR_REPLY; 5487 5488 /* translate source/destination address, if necessary */ 5489 if ((*stp)->key[PF_SK_WIRE] != (*stp)->key[PF_SK_STACK]) { 5490 struct pf_state_key *nk; 5491 int afto, sidx, didx; 5492 5493 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5494 nk = (*stp)->key[pd->sidx]; 5495 else 5496 nk = (*stp)->key[pd->didx]; 5497 5498 afto = pd->af != nk->af; 5499 sidx = afto ? pd->didx : pd->sidx; 5500 didx = afto ? pd->sidx : pd->didx; 5501 iidx = afto ? !iidx : iidx; 5502 #ifdef INET6 5503 if (afto) { 5504 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], 5505 nk->af); 5506 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], 5507 nk->af); 5508 pd->naf = nk->af; 5509 } 5510 #endif /* INET6 */ 5511 if (!afto) { 5512 pf_translate_a(pd, pd->src, &nk->addr[sidx]); 5513 pf_translate_a(pd, pd->dst, &nk->addr[didx]); 5514 } 5515 5516 if (pd->rdomain != nk->rdomain) 5517 pd->destchg = 1; 5518 if (!afto && PF_ANEQ(pd->dst, 5519 &nk->addr[didx], pd->af)) 5520 pd->destchg = 1; 5521 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5522 5523 switch (pd->af) { 5524 case AF_INET: 5525 #ifdef INET6 5526 if (afto) { 5527 if (pf_translate_icmp_af(pd, AF_INET6, 5528 &pd->hdr.icmp)) 5529 return (PF_DROP); 5530 pd->proto = IPPROTO_ICMPV6; 5531 } 5532 #endif /* INET6 */ 5533 pf_patch_16(pd, 5534 &pd->hdr.icmp.icmp_id, nk->port[iidx]); 5535 5536 m_copyback(pd->m, pd->off, ICMP_MINLEN, 5537 &pd->hdr.icmp, M_NOWAIT); 5538 copyback = 1; 5539 break; 5540 #ifdef INET6 5541 case AF_INET6: 5542 if (afto) { 5543 if (pf_translate_icmp_af(pd, AF_INET, 5544 &pd->hdr.icmp6)) 5545 return (PF_DROP); 5546 pd->proto = IPPROTO_ICMP; 5547 } 5548 5549 pf_patch_16(pd, 5550 &pd->hdr.icmp6.icmp6_id, nk->port[iidx]); 5551 5552 m_copyback(pd->m, pd->off, 5553 sizeof(struct icmp6_hdr), &pd->hdr.icmp6, 5554 M_NOWAIT); 5555 copyback = 1; 5556 break; 5557 #endif /* INET6 */ 5558 } 5559 #ifdef INET6 5560 if (afto) 5561 return (PF_AFRT); 5562 #endif /* INET6 */ 5563 } 5564 } else { 5565 /* 5566 * ICMP error message in response to a TCP/UDP packet. 5567 * Extract the inner TCP/UDP header and search for that state. 5568 */ 5569 struct pf_pdesc pd2; 5570 struct ip h2; 5571 #ifdef INET6 5572 struct ip6_hdr h2_6; 5573 #endif /* INET6 */ 5574 int ipoff2; 5575 5576 /* Initialize pd2 fields valid for both packets with pd. */ 5577 memset(&pd2, 0, sizeof(pd2)); 5578 pd2.af = pd->af; 5579 pd2.dir = pd->dir; 5580 pd2.kif = pd->kif; 5581 pd2.m = pd->m; 5582 pd2.rdomain = pd->rdomain; 5583 /* Payload packet is from the opposite direction. */ 5584 pd2.sidx = (pd2.dir == PF_IN) ? 1 : 0; 5585 pd2.didx = (pd2.dir == PF_IN) ? 0 : 1; 5586 switch (pd->af) { 5587 case AF_INET: 5588 /* offset of h2 in mbuf chain */ 5589 ipoff2 = pd->off + ICMP_MINLEN; 5590 5591 if (!pf_pull_hdr(pd2.m, ipoff2, &h2, sizeof(h2), 5592 NULL, reason, pd2.af)) { 5593 DPFPRINTF(LOG_NOTICE, 5594 "ICMP error message too short (ip)"); 5595 return (PF_DROP); 5596 } 5597 /* 5598 * ICMP error messages don't refer to non-first 5599 * fragments 5600 */ 5601 if (h2.ip_off & htons(IP_OFFMASK)) { 5602 REASON_SET(reason, PFRES_FRAG); 5603 return (PF_DROP); 5604 } 5605 5606 /* offset of protocol header that follows h2 */ 5607 pd2.off = ipoff2; 5608 if (pf_walk_header(&pd2, &h2, reason) != PF_PASS) 5609 return (PF_DROP); 5610 5611 pd2.tot_len = ntohs(h2.ip_len); 5612 pd2.src = (struct pf_addr *)&h2.ip_src; 5613 pd2.dst = (struct pf_addr *)&h2.ip_dst; 5614 break; 5615 #ifdef INET6 5616 case AF_INET6: 5617 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 5618 5619 if (!pf_pull_hdr(pd2.m, ipoff2, &h2_6, sizeof(h2_6), 5620 NULL, reason, pd2.af)) { 5621 DPFPRINTF(LOG_NOTICE, 5622 "ICMP error message too short (ip6)"); 5623 return (PF_DROP); 5624 } 5625 5626 pd2.off = ipoff2; 5627 if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS) 5628 return (PF_DROP); 5629 5630 pd2.tot_len = ntohs(h2_6.ip6_plen) + 5631 sizeof(struct ip6_hdr); 5632 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 5633 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 5634 break; 5635 #endif /* INET6 */ 5636 default: 5637 unhandled_af(pd->af); 5638 } 5639 5640 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 5641 if (pf_status.debug >= LOG_NOTICE) { 5642 log(LOG_NOTICE, 5643 "pf: BAD ICMP %d:%d outer dst: ", 5644 icmptype, icmpcode); 5645 pf_print_host(pd->src, 0, pd->af); 5646 addlog(" -> "); 5647 pf_print_host(pd->dst, 0, pd->af); 5648 addlog(" inner src: "); 5649 pf_print_host(pd2.src, 0, pd2.af); 5650 addlog(" -> "); 5651 pf_print_host(pd2.dst, 0, pd2.af); 5652 addlog("\n"); 5653 } 5654 REASON_SET(reason, PFRES_BADSTATE); 5655 return (PF_DROP); 5656 } 5657 5658 switch (pd2.proto) { 5659 case IPPROTO_TCP: { 5660 struct tcphdr *th = &pd2.hdr.tcp; 5661 u_int32_t seq; 5662 struct pf_state_peer *src, *dst; 5663 u_int8_t dws; 5664 int action; 5665 5666 /* 5667 * Only the first 8 bytes of the TCP header can be 5668 * expected. Don't access any TCP header fields after 5669 * th_seq, an ackskew test is not possible. 5670 */ 5671 if (!pf_pull_hdr(pd2.m, pd2.off, th, 8, NULL, reason, 5672 pd2.af)) { 5673 DPFPRINTF(LOG_NOTICE, 5674 "ICMP error message too short (tcp)"); 5675 return (PF_DROP); 5676 } 5677 5678 key.af = pd2.af; 5679 key.proto = IPPROTO_TCP; 5680 key.rdomain = pd2.rdomain; 5681 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 5682 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 5683 key.port[pd2.sidx] = th->th_sport; 5684 key.port[pd2.didx] = th->th_dport; 5685 key.hash = pf_pkt_hash(pd2.af, pd2.proto, 5686 pd2.src, pd2.dst, th->th_sport, th->th_dport); 5687 5688 action = pf_find_state(&pd2, &key, stp); 5689 if (action != PF_MATCH) 5690 return (action); 5691 5692 if (pd2.dir == (*stp)->direction) { 5693 if (PF_REVERSED_KEY((*stp)->key, pd->af)) { 5694 src = &(*stp)->src; 5695 dst = &(*stp)->dst; 5696 } else { 5697 src = &(*stp)->dst; 5698 dst = &(*stp)->src; 5699 } 5700 } else { 5701 if (PF_REVERSED_KEY((*stp)->key, pd->af)) { 5702 src = &(*stp)->dst; 5703 dst = &(*stp)->src; 5704 } else { 5705 src = &(*stp)->src; 5706 dst = &(*stp)->dst; 5707 } 5708 } 5709 5710 if (src->wscale && dst->wscale) 5711 dws = dst->wscale & PF_WSCALE_MASK; 5712 else 5713 dws = 0; 5714 5715 /* Demodulate sequence number */ 5716 seq = ntohl(th->th_seq) - src->seqdiff; 5717 if (src->seqdiff) { 5718 pf_patch_32(pd, &th->th_seq, htonl(seq)); 5719 copyback = 1; 5720 } 5721 5722 if (!((*stp)->state_flags & PFSTATE_SLOPPY) && 5723 (!SEQ_GEQ(src->seqhi, seq) || !SEQ_GEQ(seq, 5724 src->seqlo - (dst->max_win << dws)))) { 5725 if (pf_status.debug >= LOG_NOTICE) { 5726 log(LOG_NOTICE, 5727 "pf: BAD ICMP %d:%d ", 5728 icmptype, icmpcode); 5729 pf_print_host(pd->src, 0, pd->af); 5730 addlog(" -> "); 5731 pf_print_host(pd->dst, 0, pd->af); 5732 addlog(" state: "); 5733 pf_print_state(*stp); 5734 addlog(" seq=%u\n", seq); 5735 } 5736 REASON_SET(reason, PFRES_BADSTATE); 5737 return (PF_DROP); 5738 } else { 5739 if (pf_status.debug >= LOG_DEBUG) { 5740 log(LOG_DEBUG, 5741 "pf: OK ICMP %d:%d ", 5742 icmptype, icmpcode); 5743 pf_print_host(pd->src, 0, pd->af); 5744 addlog(" -> "); 5745 pf_print_host(pd->dst, 0, pd->af); 5746 addlog(" state: "); 5747 pf_print_state(*stp); 5748 addlog(" seq=%u\n", seq); 5749 } 5750 } 5751 5752 /* translate source/destination address, if necessary */ 5753 if ((*stp)->key[PF_SK_WIRE] != 5754 (*stp)->key[PF_SK_STACK]) { 5755 struct pf_state_key *nk; 5756 int afto, sidx, didx; 5757 5758 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5759 nk = (*stp)->key[pd->sidx]; 5760 else 5761 nk = (*stp)->key[pd->didx]; 5762 5763 afto = pd->af != nk->af; 5764 sidx = afto ? pd2.didx : pd2.sidx; 5765 didx = afto ? pd2.sidx : pd2.didx; 5766 5767 #ifdef INET6 5768 if (afto) { 5769 if (pf_translate_icmp_af(pd, nk->af, 5770 &pd->hdr.icmp)) 5771 return (PF_DROP); 5772 m_copyback(pd->m, pd->off, 5773 sizeof(struct icmp6_hdr), 5774 &pd->hdr.icmp6, M_NOWAIT); 5775 if (pf_change_icmp_af(pd->m, ipoff2, 5776 pd, &pd2, &nk->addr[sidx], 5777 &nk->addr[didx], pd->af, nk->af)) 5778 return (PF_DROP); 5779 if (nk->af == AF_INET) 5780 pd->proto = IPPROTO_ICMP; 5781 else 5782 pd->proto = IPPROTO_ICMPV6; 5783 pd->m->m_pkthdr.ph_rtableid = 5784 nk->rdomain; 5785 pd->destchg = 1; 5786 pf_addrcpy(&pd->nsaddr, 5787 &nk->addr[pd2.sidx], nk->af); 5788 pf_addrcpy(&pd->ndaddr, 5789 &nk->addr[pd2.didx], nk->af); 5790 pd->naf = nk->af; 5791 5792 pf_patch_16(pd, 5793 &th->th_sport, nk->port[sidx]); 5794 pf_patch_16(pd, 5795 &th->th_dport, nk->port[didx]); 5796 5797 m_copyback(pd2.m, pd2.off, 8, th, 5798 M_NOWAIT); 5799 return (PF_AFRT); 5800 } 5801 #endif /* INET6 */ 5802 if (PF_ANEQ(pd2.src, 5803 &nk->addr[pd2.sidx], pd2.af) || 5804 nk->port[pd2.sidx] != th->th_sport) 5805 pf_translate_icmp(pd, pd2.src, 5806 &th->th_sport, pd->dst, 5807 &nk->addr[pd2.sidx], 5808 nk->port[pd2.sidx]); 5809 5810 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 5811 pd2.af) || pd2.rdomain != nk->rdomain) 5812 pd->destchg = 1; 5813 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5814 5815 if (PF_ANEQ(pd2.dst, 5816 &nk->addr[pd2.didx], pd2.af) || 5817 nk->port[pd2.didx] != th->th_dport) 5818 pf_translate_icmp(pd, pd2.dst, 5819 &th->th_dport, pd->src, 5820 &nk->addr[pd2.didx], 5821 nk->port[pd2.didx]); 5822 copyback = 1; 5823 } 5824 5825 if (copyback) { 5826 switch (pd2.af) { 5827 case AF_INET: 5828 m_copyback(pd->m, pd->off, ICMP_MINLEN, 5829 &pd->hdr.icmp, M_NOWAIT); 5830 m_copyback(pd2.m, ipoff2, sizeof(h2), 5831 &h2, M_NOWAIT); 5832 break; 5833 #ifdef INET6 5834 case AF_INET6: 5835 m_copyback(pd->m, pd->off, 5836 sizeof(struct icmp6_hdr), 5837 &pd->hdr.icmp6, M_NOWAIT); 5838 m_copyback(pd2.m, ipoff2, sizeof(h2_6), 5839 &h2_6, M_NOWAIT); 5840 break; 5841 #endif /* INET6 */ 5842 } 5843 m_copyback(pd2.m, pd2.off, 8, th, M_NOWAIT); 5844 } 5845 break; 5846 } 5847 case IPPROTO_UDP: { 5848 struct udphdr *uh = &pd2.hdr.udp; 5849 int action; 5850 5851 if (!pf_pull_hdr(pd2.m, pd2.off, uh, sizeof(*uh), 5852 NULL, reason, pd2.af)) { 5853 DPFPRINTF(LOG_NOTICE, 5854 "ICMP error message too short (udp)"); 5855 return (PF_DROP); 5856 } 5857 5858 key.af = pd2.af; 5859 key.proto = IPPROTO_UDP; 5860 key.rdomain = pd2.rdomain; 5861 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 5862 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 5863 key.port[pd2.sidx] = uh->uh_sport; 5864 key.port[pd2.didx] = uh->uh_dport; 5865 key.hash = pf_pkt_hash(pd2.af, pd2.proto, 5866 pd2.src, pd2.dst, uh->uh_sport, uh->uh_dport); 5867 5868 action = pf_find_state(&pd2, &key, stp); 5869 if (action != PF_MATCH) 5870 return (action); 5871 5872 /* translate source/destination address, if necessary */ 5873 if ((*stp)->key[PF_SK_WIRE] != 5874 (*stp)->key[PF_SK_STACK]) { 5875 struct pf_state_key *nk; 5876 int afto, sidx, didx; 5877 5878 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 5879 nk = (*stp)->key[pd->sidx]; 5880 else 5881 nk = (*stp)->key[pd->didx]; 5882 5883 afto = pd->af != nk->af; 5884 sidx = afto ? pd2.didx : pd2.sidx; 5885 didx = afto ? pd2.sidx : pd2.didx; 5886 5887 #ifdef INET6 5888 if (afto) { 5889 if (pf_translate_icmp_af(pd, nk->af, 5890 &pd->hdr.icmp)) 5891 return (PF_DROP); 5892 m_copyback(pd->m, pd->off, 5893 sizeof(struct icmp6_hdr), 5894 &pd->hdr.icmp6, M_NOWAIT); 5895 if (pf_change_icmp_af(pd->m, ipoff2, 5896 pd, &pd2, &nk->addr[sidx], 5897 &nk->addr[didx], pd->af, nk->af)) 5898 return (PF_DROP); 5899 if (nk->af == AF_INET) 5900 pd->proto = IPPROTO_ICMP; 5901 else 5902 pd->proto = IPPROTO_ICMPV6; 5903 pd->m->m_pkthdr.ph_rtableid = 5904 nk->rdomain; 5905 pd->destchg = 1; 5906 pf_addrcpy(&pd->nsaddr, 5907 &nk->addr[pd2.sidx], nk->af); 5908 pf_addrcpy(&pd->ndaddr, 5909 &nk->addr[pd2.didx], nk->af); 5910 pd->naf = nk->af; 5911 5912 pf_patch_16(pd, 5913 &uh->uh_sport, nk->port[sidx]); 5914 pf_patch_16(pd, 5915 &uh->uh_dport, nk->port[didx]); 5916 5917 m_copyback(pd2.m, pd2.off, sizeof(*uh), 5918 uh, M_NOWAIT); 5919 return (PF_AFRT); 5920 } 5921 #endif /* INET6 */ 5922 5923 if (PF_ANEQ(pd2.src, 5924 &nk->addr[pd2.sidx], pd2.af) || 5925 nk->port[pd2.sidx] != uh->uh_sport) 5926 pf_translate_icmp(pd, pd2.src, 5927 &uh->uh_sport, pd->dst, 5928 &nk->addr[pd2.sidx], 5929 nk->port[pd2.sidx]); 5930 5931 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 5932 pd2.af) || pd2.rdomain != nk->rdomain) 5933 pd->destchg = 1; 5934 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 5935 5936 if (PF_ANEQ(pd2.dst, 5937 &nk->addr[pd2.didx], pd2.af) || 5938 nk->port[pd2.didx] != uh->uh_dport) 5939 pf_translate_icmp(pd, pd2.dst, 5940 &uh->uh_dport, pd->src, 5941 &nk->addr[pd2.didx], 5942 nk->port[pd2.didx]); 5943 5944 switch (pd2.af) { 5945 case AF_INET: 5946 m_copyback(pd->m, pd->off, ICMP_MINLEN, 5947 &pd->hdr.icmp, M_NOWAIT); 5948 m_copyback(pd2.m, ipoff2, sizeof(h2), 5949 &h2, M_NOWAIT); 5950 break; 5951 #ifdef INET6 5952 case AF_INET6: 5953 m_copyback(pd->m, pd->off, 5954 sizeof(struct icmp6_hdr), 5955 &pd->hdr.icmp6, M_NOWAIT); 5956 m_copyback(pd2.m, ipoff2, sizeof(h2_6), 5957 &h2_6, M_NOWAIT); 5958 break; 5959 #endif /* INET6 */ 5960 } 5961 /* Avoid recomputing quoted UDP checksum. 5962 * note: udp6 0 csum invalid per rfc2460 p27. 5963 * but presumed nothing cares in this context */ 5964 pf_patch_16(pd, &uh->uh_sum, 0); 5965 m_copyback(pd2.m, pd2.off, sizeof(*uh), uh, 5966 M_NOWAIT); 5967 copyback = 1; 5968 } 5969 break; 5970 } 5971 case IPPROTO_ICMP: { 5972 struct icmp *iih = &pd2.hdr.icmp; 5973 5974 if (pd2.af != AF_INET) { 5975 REASON_SET(reason, PFRES_NORM); 5976 return (PF_DROP); 5977 } 5978 5979 if (!pf_pull_hdr(pd2.m, pd2.off, iih, ICMP_MINLEN, 5980 NULL, reason, pd2.af)) { 5981 DPFPRINTF(LOG_NOTICE, 5982 "ICMP error message too short (icmp)"); 5983 return (PF_DROP); 5984 } 5985 5986 pf_icmp_mapping(&pd2, iih->icmp_type, 5987 &icmp_dir, &virtual_id, &virtual_type); 5988 5989 ret = pf_icmp_state_lookup(&pd2, &key, stp, 5990 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 5991 if (ret >= 0) 5992 return (ret); 5993 5994 /* translate source/destination address, if necessary */ 5995 if ((*stp)->key[PF_SK_WIRE] != 5996 (*stp)->key[PF_SK_STACK]) { 5997 struct pf_state_key *nk; 5998 int afto, sidx, didx; 5999 6000 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 6001 nk = (*stp)->key[pd->sidx]; 6002 else 6003 nk = (*stp)->key[pd->didx]; 6004 6005 afto = pd->af != nk->af; 6006 sidx = afto ? pd2.didx : pd2.sidx; 6007 didx = afto ? pd2.sidx : pd2.didx; 6008 iidx = afto ? !iidx : iidx; 6009 6010 #ifdef INET6 6011 if (afto) { 6012 if (nk->af != AF_INET6) 6013 return (PF_DROP); 6014 if (pf_translate_icmp_af(pd, nk->af, 6015 &pd->hdr.icmp)) 6016 return (PF_DROP); 6017 m_copyback(pd->m, pd->off, 6018 sizeof(struct icmp6_hdr), 6019 &pd->hdr.icmp6, M_NOWAIT); 6020 if (pf_change_icmp_af(pd->m, ipoff2, 6021 pd, &pd2, &nk->addr[sidx], 6022 &nk->addr[didx], pd->af, nk->af)) 6023 return (PF_DROP); 6024 pd->proto = IPPROTO_ICMPV6; 6025 if (pf_translate_icmp_af(pd, 6026 nk->af, iih)) 6027 return (PF_DROP); 6028 if (virtual_type == htons(ICMP_ECHO)) 6029 pf_patch_16(pd, &iih->icmp_id, 6030 nk->port[iidx]); 6031 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, 6032 iih, M_NOWAIT); 6033 pd->m->m_pkthdr.ph_rtableid = 6034 nk->rdomain; 6035 pd->destchg = 1; 6036 pf_addrcpy(&pd->nsaddr, 6037 &nk->addr[pd2.sidx], nk->af); 6038 pf_addrcpy(&pd->ndaddr, 6039 &nk->addr[pd2.didx], nk->af); 6040 pd->naf = nk->af; 6041 return (PF_AFRT); 6042 } 6043 #endif /* INET6 */ 6044 6045 if (PF_ANEQ(pd2.src, 6046 &nk->addr[pd2.sidx], pd2.af) || 6047 (virtual_type == htons(ICMP_ECHO) && 6048 nk->port[iidx] != iih->icmp_id)) 6049 pf_translate_icmp(pd, pd2.src, 6050 (virtual_type == htons(ICMP_ECHO)) ? 6051 &iih->icmp_id : NULL, 6052 pd->dst, &nk->addr[pd2.sidx], 6053 (virtual_type == htons(ICMP_ECHO)) ? 6054 nk->port[iidx] : 0); 6055 6056 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 6057 pd2.af) || pd2.rdomain != nk->rdomain) 6058 pd->destchg = 1; 6059 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 6060 6061 if (PF_ANEQ(pd2.dst, 6062 &nk->addr[pd2.didx], pd2.af)) 6063 pf_translate_icmp(pd, pd2.dst, NULL, 6064 pd->src, &nk->addr[pd2.didx], 0); 6065 6066 m_copyback(pd->m, pd->off, ICMP_MINLEN, 6067 &pd->hdr.icmp, M_NOWAIT); 6068 m_copyback(pd2.m, ipoff2, sizeof(h2), &h2, 6069 M_NOWAIT); 6070 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, iih, 6071 M_NOWAIT); 6072 copyback = 1; 6073 } 6074 break; 6075 } 6076 #ifdef INET6 6077 case IPPROTO_ICMPV6: { 6078 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 6079 6080 if (pd2.af != AF_INET6) { 6081 REASON_SET(reason, PFRES_NORM); 6082 return (PF_DROP); 6083 } 6084 6085 if (!pf_pull_hdr(pd2.m, pd2.off, iih, 6086 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 6087 DPFPRINTF(LOG_NOTICE, 6088 "ICMP error message too short (icmp6)"); 6089 return (PF_DROP); 6090 } 6091 6092 pf_icmp_mapping(&pd2, iih->icmp6_type, 6093 &icmp_dir, &virtual_id, &virtual_type); 6094 ret = pf_icmp_state_lookup(&pd2, &key, stp, 6095 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 6096 /* IPv6? try matching a multicast address */ 6097 if (ret == PF_DROP && pd2.af == AF_INET6 && 6098 icmp_dir == PF_OUT) 6099 ret = pf_icmp_state_lookup(&pd2, &key, stp, 6100 virtual_id, virtual_type, icmp_dir, &iidx, 6101 1, 1); 6102 if (ret >= 0) 6103 return (ret); 6104 6105 /* translate source/destination address, if necessary */ 6106 if ((*stp)->key[PF_SK_WIRE] != 6107 (*stp)->key[PF_SK_STACK]) { 6108 struct pf_state_key *nk; 6109 int afto, sidx, didx; 6110 6111 if (PF_REVERSED_KEY((*stp)->key, pd->af)) 6112 nk = (*stp)->key[pd->sidx]; 6113 else 6114 nk = (*stp)->key[pd->didx]; 6115 6116 afto = pd->af != nk->af; 6117 sidx = afto ? pd2.didx : pd2.sidx; 6118 didx = afto ? pd2.sidx : pd2.didx; 6119 iidx = afto ? !iidx : iidx; 6120 6121 if (afto) { 6122 if (nk->af != AF_INET) 6123 return (PF_DROP); 6124 if (pf_translate_icmp_af(pd, nk->af, 6125 &pd->hdr.icmp)) 6126 return (PF_DROP); 6127 m_copyback(pd->m, pd->off, 6128 sizeof(struct icmp6_hdr), 6129 &pd->hdr.icmp6, M_NOWAIT); 6130 if (pf_change_icmp_af(pd->m, ipoff2, 6131 pd, &pd2, &nk->addr[sidx], 6132 &nk->addr[didx], pd->af, nk->af)) 6133 return (PF_DROP); 6134 pd->proto = IPPROTO_ICMP; 6135 if (pf_translate_icmp_af(pd, 6136 nk->af, iih)) 6137 return (PF_DROP); 6138 if (virtual_type == 6139 htons(ICMP6_ECHO_REQUEST)) 6140 pf_patch_16(pd, &iih->icmp6_id, 6141 nk->port[iidx]); 6142 m_copyback(pd2.m, pd2.off, 6143 sizeof(struct icmp6_hdr), iih, 6144 M_NOWAIT); 6145 pd->m->m_pkthdr.ph_rtableid = 6146 nk->rdomain; 6147 pd->destchg = 1; 6148 pf_addrcpy(&pd->nsaddr, 6149 &nk->addr[pd2.sidx], nk->af); 6150 pf_addrcpy(&pd->ndaddr, 6151 &nk->addr[pd2.didx], nk->af); 6152 pd->naf = nk->af; 6153 return (PF_AFRT); 6154 } 6155 6156 if (PF_ANEQ(pd2.src, 6157 &nk->addr[pd2.sidx], pd2.af) || 6158 ((virtual_type == 6159 htons(ICMP6_ECHO_REQUEST)) && 6160 nk->port[pd2.sidx] != iih->icmp6_id)) 6161 pf_translate_icmp(pd, pd2.src, 6162 (virtual_type == 6163 htons(ICMP6_ECHO_REQUEST)) 6164 ? &iih->icmp6_id : NULL, 6165 pd->dst, &nk->addr[pd2.sidx], 6166 (virtual_type == 6167 htons(ICMP6_ECHO_REQUEST)) 6168 ? nk->port[iidx] : 0); 6169 6170 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 6171 pd2.af) || pd2.rdomain != nk->rdomain) 6172 pd->destchg = 1; 6173 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 6174 6175 if (PF_ANEQ(pd2.dst, 6176 &nk->addr[pd2.didx], pd2.af)) 6177 pf_translate_icmp(pd, pd2.dst, NULL, 6178 pd->src, &nk->addr[pd2.didx], 0); 6179 6180 m_copyback(pd->m, pd->off, 6181 sizeof(struct icmp6_hdr), &pd->hdr.icmp6, 6182 M_NOWAIT); 6183 m_copyback(pd2.m, ipoff2, sizeof(h2_6), &h2_6, 6184 M_NOWAIT); 6185 m_copyback(pd2.m, pd2.off, 6186 sizeof(struct icmp6_hdr), iih, M_NOWAIT); 6187 copyback = 1; 6188 } 6189 break; 6190 } 6191 #endif /* INET6 */ 6192 default: { 6193 int action; 6194 6195 key.af = pd2.af; 6196 key.proto = pd2.proto; 6197 key.rdomain = pd2.rdomain; 6198 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 6199 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 6200 key.port[0] = key.port[1] = 0; 6201 key.hash = pf_pkt_hash(pd2.af, pd2.proto, 6202 pd2.src, pd2.dst, 0, 0); 6203 6204 action = pf_find_state(&pd2, &key, stp); 6205 if (action != PF_MATCH) 6206 return (action); 6207 6208 /* translate source/destination address, if necessary */ 6209 if ((*stp)->key[PF_SK_WIRE] != 6210 (*stp)->key[PF_SK_STACK]) { 6211 struct pf_state_key *nk = 6212 (*stp)->key[pd->didx]; 6213 6214 if (PF_ANEQ(pd2.src, 6215 &nk->addr[pd2.sidx], pd2.af)) 6216 pf_translate_icmp(pd, pd2.src, NULL, 6217 pd->dst, &nk->addr[pd2.sidx], 0); 6218 6219 if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], 6220 pd2.af) || pd2.rdomain != nk->rdomain) 6221 pd->destchg = 1; 6222 pd->m->m_pkthdr.ph_rtableid = nk->rdomain; 6223 6224 if (PF_ANEQ(pd2.dst, 6225 &nk->addr[pd2.didx], pd2.af)) 6226 pf_translate_icmp(pd, pd2.dst, NULL, 6227 pd->src, &nk->addr[pd2.didx], 0); 6228 6229 switch (pd2.af) { 6230 case AF_INET: 6231 m_copyback(pd->m, pd->off, ICMP_MINLEN, 6232 &pd->hdr.icmp, M_NOWAIT); 6233 m_copyback(pd2.m, ipoff2, sizeof(h2), 6234 &h2, M_NOWAIT); 6235 break; 6236 #ifdef INET6 6237 case AF_INET6: 6238 m_copyback(pd->m, pd->off, 6239 sizeof(struct icmp6_hdr), 6240 &pd->hdr.icmp6, M_NOWAIT); 6241 m_copyback(pd2.m, ipoff2, sizeof(h2_6), 6242 &h2_6, M_NOWAIT); 6243 break; 6244 #endif /* INET6 */ 6245 } 6246 copyback = 1; 6247 } 6248 break; 6249 } 6250 } 6251 } 6252 if (copyback) { 6253 m_copyback(pd->m, pd->off, pd->hdrlen, &pd->hdr, M_NOWAIT); 6254 } 6255 6256 return (PF_PASS); 6257 } 6258 6259 /* 6260 * ipoff and off are measured from the start of the mbuf chain. 6261 * h must be at "ipoff" on the mbuf chain. 6262 */ 6263 void * 6264 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 6265 u_short *actionp, u_short *reasonp, sa_family_t af) 6266 { 6267 int iplen = 0; 6268 6269 switch (af) { 6270 case AF_INET: { 6271 struct ip *h = mtod(m, struct ip *); 6272 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 6273 6274 if (fragoff) { 6275 if (fragoff >= len) 6276 ACTION_SET(actionp, PF_PASS); 6277 else { 6278 ACTION_SET(actionp, PF_DROP); 6279 REASON_SET(reasonp, PFRES_FRAG); 6280 } 6281 return (NULL); 6282 } 6283 iplen = ntohs(h->ip_len); 6284 break; 6285 } 6286 #ifdef INET6 6287 case AF_INET6: { 6288 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 6289 6290 iplen = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 6291 break; 6292 } 6293 #endif /* INET6 */ 6294 } 6295 if (m->m_pkthdr.len < off + len || iplen < off + len) { 6296 ACTION_SET(actionp, PF_DROP); 6297 REASON_SET(reasonp, PFRES_SHORT); 6298 return (NULL); 6299 } 6300 m_copydata(m, off, len, p); 6301 return (p); 6302 } 6303 6304 int 6305 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif, 6306 int rtableid) 6307 { 6308 struct sockaddr_storage ss; 6309 struct sockaddr_in *dst; 6310 int ret = 1; 6311 int check_mpath; 6312 #ifdef INET6 6313 struct sockaddr_in6 *dst6; 6314 #endif /* INET6 */ 6315 struct rtentry *rt = NULL; 6316 6317 check_mpath = 0; 6318 memset(&ss, 0, sizeof(ss)); 6319 switch (af) { 6320 case AF_INET: 6321 dst = (struct sockaddr_in *)&ss; 6322 dst->sin_family = AF_INET; 6323 dst->sin_len = sizeof(*dst); 6324 dst->sin_addr = addr->v4; 6325 if (ipmultipath) 6326 check_mpath = 1; 6327 break; 6328 #ifdef INET6 6329 case AF_INET6: 6330 /* 6331 * Skip check for addresses with embedded interface scope, 6332 * as they would always match anyway. 6333 */ 6334 if (IN6_IS_SCOPE_EMBED(&addr->v6)) 6335 goto out; 6336 dst6 = (struct sockaddr_in6 *)&ss; 6337 dst6->sin6_family = AF_INET6; 6338 dst6->sin6_len = sizeof(*dst6); 6339 dst6->sin6_addr = addr->v6; 6340 if (ip6_multipath) 6341 check_mpath = 1; 6342 break; 6343 #endif /* INET6 */ 6344 } 6345 6346 /* Skip checks for ipsec interfaces */ 6347 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 6348 goto out; 6349 6350 rt = rtalloc(sstosa(&ss), 0, rtableid); 6351 if (rt != NULL) { 6352 /* No interface given, this is a no-route check */ 6353 if (kif == NULL) 6354 goto out; 6355 6356 if (kif->pfik_ifp == NULL) { 6357 ret = 0; 6358 goto out; 6359 } 6360 6361 /* Perform uRPF check if passed input interface */ 6362 ret = 0; 6363 do { 6364 if (rt->rt_ifidx == kif->pfik_ifp->if_index) { 6365 ret = 1; 6366 #if NCARP > 0 6367 } else { 6368 struct ifnet *ifp; 6369 6370 ifp = if_get(rt->rt_ifidx); 6371 if (ifp != NULL && ifp->if_type == IFT_CARP && 6372 ifp->if_carpdevidx == 6373 kif->pfik_ifp->if_index) 6374 ret = 1; 6375 if_put(ifp); 6376 #endif /* NCARP */ 6377 } 6378 6379 rt = rtable_iterate(rt); 6380 } while (check_mpath == 1 && rt != NULL && ret == 0); 6381 } else 6382 ret = 0; 6383 out: 6384 rtfree(rt); 6385 return (ret); 6386 } 6387 6388 int 6389 pf_rtlabel_match(struct pf_addr *addr, sa_family_t af, struct pf_addr_wrap *aw, 6390 int rtableid) 6391 { 6392 struct sockaddr_storage ss; 6393 struct sockaddr_in *dst; 6394 #ifdef INET6 6395 struct sockaddr_in6 *dst6; 6396 #endif /* INET6 */ 6397 struct rtentry *rt; 6398 int ret = 0; 6399 6400 memset(&ss, 0, sizeof(ss)); 6401 switch (af) { 6402 case AF_INET: 6403 dst = (struct sockaddr_in *)&ss; 6404 dst->sin_family = AF_INET; 6405 dst->sin_len = sizeof(*dst); 6406 dst->sin_addr = addr->v4; 6407 break; 6408 #ifdef INET6 6409 case AF_INET6: 6410 dst6 = (struct sockaddr_in6 *)&ss; 6411 dst6->sin6_family = AF_INET6; 6412 dst6->sin6_len = sizeof(*dst6); 6413 dst6->sin6_addr = addr->v6; 6414 break; 6415 #endif /* INET6 */ 6416 } 6417 6418 rt = rtalloc(sstosa(&ss), RT_RESOLVE, rtableid); 6419 if (rt != NULL) { 6420 if (rt->rt_labelid == aw->v.rtlabel) 6421 ret = 1; 6422 rtfree(rt); 6423 } 6424 6425 return (ret); 6426 } 6427 6428 /* pf_route() may change pd->m, adjust local copies after calling */ 6429 void 6430 pf_route(struct pf_pdesc *pd, struct pf_state *st) 6431 { 6432 struct mbuf *m0; 6433 struct mbuf_list fml; 6434 struct sockaddr_in *dst, sin; 6435 struct rtentry *rt = NULL; 6436 struct ip *ip; 6437 struct ifnet *ifp = NULL; 6438 int error = 0; 6439 unsigned int rtableid; 6440 6441 if (pd->m->m_pkthdr.pf.routed++ > 3) { 6442 m_freem(pd->m); 6443 pd->m = NULL; 6444 return; 6445 } 6446 6447 if (st->rt == PF_DUPTO) { 6448 if ((m0 = m_dup_pkt(pd->m, max_linkhdr, M_NOWAIT)) == NULL) 6449 return; 6450 } else { 6451 if ((st->rt == PF_REPLYTO) == (st->direction == pd->dir)) 6452 return; 6453 m0 = pd->m; 6454 pd->m = NULL; 6455 } 6456 6457 if (m0->m_len < sizeof(struct ip)) { 6458 DPFPRINTF(LOG_ERR, 6459 "%s: m0->m_len < sizeof(struct ip)", __func__); 6460 goto bad; 6461 } 6462 6463 ip = mtod(m0, struct ip *); 6464 6465 if (pd->dir == PF_IN) { 6466 if (ip->ip_ttl <= IPTTLDEC) { 6467 if (st->rt != PF_DUPTO) { 6468 pf_send_icmp(m0, ICMP_TIMXCEED, 6469 ICMP_TIMXCEED_INTRANS, 0, 6470 pd->af, st->rule.ptr, pd->rdomain); 6471 } 6472 goto bad; 6473 } 6474 ip->ip_ttl -= IPTTLDEC; 6475 } 6476 6477 memset(&sin, 0, sizeof(sin)); 6478 dst = &sin; 6479 dst->sin_family = AF_INET; 6480 dst->sin_len = sizeof(*dst); 6481 dst->sin_addr = st->rt_addr.v4; 6482 rtableid = m0->m_pkthdr.ph_rtableid; 6483 6484 rt = rtalloc_mpath(sintosa(dst), &ip->ip_src.s_addr, rtableid); 6485 if (!rtisvalid(rt)) { 6486 if (st->rt != PF_DUPTO) { 6487 pf_send_icmp(m0, ICMP_UNREACH, ICMP_UNREACH_HOST, 6488 0, pd->af, st->rule.ptr, pd->rdomain); 6489 } 6490 ipstat_inc(ips_noroute); 6491 goto bad; 6492 } 6493 6494 ifp = if_get(rt->rt_ifidx); 6495 if (ifp == NULL) 6496 goto bad; 6497 6498 /* A locally generated packet may have invalid source address. */ 6499 if ((ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET && 6500 (ifp->if_flags & IFF_LOOPBACK) == 0) 6501 ip->ip_src = ifatoia(rt->rt_ifa)->ia_addr.sin_addr; 6502 6503 if (st->rt != PF_DUPTO && pd->dir == PF_IN) { 6504 if (pf_test(AF_INET, PF_OUT, ifp, &m0) != PF_PASS) 6505 goto bad; 6506 else if (m0 == NULL) 6507 goto done; 6508 if (m0->m_len < sizeof(struct ip)) { 6509 DPFPRINTF(LOG_ERR, 6510 "%s: m0->m_len < sizeof(struct ip)", __func__); 6511 goto bad; 6512 } 6513 ip = mtod(m0, struct ip *); 6514 } 6515 6516 in_proto_cksum_out(m0, ifp); 6517 6518 if (ntohs(ip->ip_len) <= ifp->if_mtu) { 6519 ip->ip_sum = 0; 6520 if (ifp->if_capabilities & IFCAP_CSUM_IPv4) 6521 m0->m_pkthdr.csum_flags |= M_IPV4_CSUM_OUT; 6522 else { 6523 ipstat_inc(ips_outswcsum); 6524 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 6525 } 6526 error = ifp->if_output(ifp, m0, sintosa(dst), rt); 6527 goto done; 6528 } 6529 6530 /* 6531 * Too large for interface; fragment if possible. 6532 * Must be able to put at least 8 bytes per fragment. 6533 */ 6534 if (ip->ip_off & htons(IP_DF)) { 6535 ipstat_inc(ips_cantfrag); 6536 if (st->rt != PF_DUPTO) 6537 pf_send_icmp(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 6538 ifp->if_mtu, pd->af, st->rule.ptr, pd->rdomain); 6539 goto bad; 6540 } 6541 6542 error = ip_fragment(m0, &fml, ifp, ifp->if_mtu); 6543 if (error) 6544 goto done; 6545 6546 while ((m0 = ml_dequeue(&fml)) != NULL) { 6547 error = ifp->if_output(ifp, m0, sintosa(dst), rt); 6548 if (error) 6549 break; 6550 } 6551 if (error) 6552 ml_purge(&fml); 6553 else 6554 ipstat_inc(ips_fragmented); 6555 6556 done: 6557 if_put(ifp); 6558 rtfree(rt); 6559 return; 6560 6561 bad: 6562 m_freem(m0); 6563 goto done; 6564 } 6565 6566 #ifdef INET6 6567 /* pf_route6() may change pd->m, adjust local copies after calling */ 6568 void 6569 pf_route6(struct pf_pdesc *pd, struct pf_state *st) 6570 { 6571 struct mbuf *m0; 6572 struct sockaddr_in6 *dst, sin6; 6573 struct rtentry *rt = NULL; 6574 struct ip6_hdr *ip6; 6575 struct ifnet *ifp = NULL; 6576 struct m_tag *mtag; 6577 unsigned int rtableid; 6578 6579 if (pd->m->m_pkthdr.pf.routed++ > 3) { 6580 m_freem(pd->m); 6581 pd->m = NULL; 6582 return; 6583 } 6584 6585 if (st->rt == PF_DUPTO) { 6586 if ((m0 = m_dup_pkt(pd->m, max_linkhdr, M_NOWAIT)) == NULL) 6587 return; 6588 } else { 6589 if ((st->rt == PF_REPLYTO) == (st->direction == pd->dir)) 6590 return; 6591 m0 = pd->m; 6592 pd->m = NULL; 6593 } 6594 6595 if (m0->m_len < sizeof(struct ip6_hdr)) { 6596 DPFPRINTF(LOG_ERR, 6597 "%s: m0->m_len < sizeof(struct ip6_hdr)", __func__); 6598 goto bad; 6599 } 6600 ip6 = mtod(m0, struct ip6_hdr *); 6601 6602 if (pd->dir == PF_IN) { 6603 if (ip6->ip6_hlim <= IPV6_HLIMDEC) { 6604 if (st->rt != PF_DUPTO) { 6605 pf_send_icmp(m0, ICMP6_TIME_EXCEEDED, 6606 ICMP6_TIME_EXCEED_TRANSIT, 0, 6607 pd->af, st->rule.ptr, pd->rdomain); 6608 } 6609 goto bad; 6610 } 6611 ip6->ip6_hlim -= IPV6_HLIMDEC; 6612 } 6613 6614 memset(&sin6, 0, sizeof(sin6)); 6615 dst = &sin6; 6616 dst->sin6_family = AF_INET6; 6617 dst->sin6_len = sizeof(*dst); 6618 dst->sin6_addr = st->rt_addr.v6; 6619 rtableid = m0->m_pkthdr.ph_rtableid; 6620 6621 rt = rtalloc_mpath(sin6tosa(dst), &ip6->ip6_src.s6_addr32[0], 6622 rtableid); 6623 if (!rtisvalid(rt)) { 6624 if (st->rt != PF_DUPTO) { 6625 pf_send_icmp(m0, ICMP6_DST_UNREACH, 6626 ICMP6_DST_UNREACH_NOROUTE, 0, 6627 pd->af, st->rule.ptr, pd->rdomain); 6628 } 6629 ip6stat_inc(ip6s_noroute); 6630 goto bad; 6631 } 6632 6633 ifp = if_get(rt->rt_ifidx); 6634 if (ifp == NULL) 6635 goto bad; 6636 6637 /* A locally generated packet may have invalid source address. */ 6638 if (IN6_IS_ADDR_LOOPBACK(&ip6->ip6_src) && 6639 (ifp->if_flags & IFF_LOOPBACK) == 0) 6640 ip6->ip6_src = ifatoia6(rt->rt_ifa)->ia_addr.sin6_addr; 6641 6642 if (st->rt != PF_DUPTO && pd->dir == PF_IN) { 6643 if (pf_test(AF_INET6, PF_OUT, ifp, &m0) != PF_PASS) 6644 goto bad; 6645 else if (m0 == NULL) 6646 goto done; 6647 if (m0->m_len < sizeof(struct ip6_hdr)) { 6648 DPFPRINTF(LOG_ERR, 6649 "%s: m0->m_len < sizeof(struct ip6_hdr)", __func__); 6650 goto bad; 6651 } 6652 } 6653 6654 in6_proto_cksum_out(m0, ifp); 6655 6656 /* 6657 * If packet has been reassembled by PF earlier, we have to 6658 * use pf_refragment6() here to turn it back to fragments. 6659 */ 6660 if ((mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL))) { 6661 (void) pf_refragment6(&m0, mtag, dst, ifp, rt); 6662 } else if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 6663 ifp->if_output(ifp, m0, sin6tosa(dst), rt); 6664 } else { 6665 ip6stat_inc(ip6s_cantfrag); 6666 if (st->rt != PF_DUPTO) 6667 pf_send_icmp(m0, ICMP6_PACKET_TOO_BIG, 0, 6668 ifp->if_mtu, pd->af, st->rule.ptr, pd->rdomain); 6669 goto bad; 6670 } 6671 6672 done: 6673 if_put(ifp); 6674 rtfree(rt); 6675 return; 6676 6677 bad: 6678 m_freem(m0); 6679 goto done; 6680 } 6681 #endif /* INET6 */ 6682 6683 /* 6684 * check TCP checksum and set mbuf flag 6685 * off is the offset where the protocol header starts 6686 * len is the total length of protocol header plus payload 6687 * returns 0 when the checksum is valid, otherwise returns 1. 6688 * if the _OUT flag is set the checksum isn't done yet, consider these ok 6689 */ 6690 int 6691 pf_check_tcp_cksum(struct mbuf *m, int off, int len, sa_family_t af) 6692 { 6693 u_int16_t sum; 6694 6695 if (m->m_pkthdr.csum_flags & 6696 (M_TCP_CSUM_IN_OK | M_TCP_CSUM_OUT)) { 6697 return (0); 6698 } 6699 if (m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_BAD || 6700 off < sizeof(struct ip) || 6701 m->m_pkthdr.len < off + len) { 6702 return (1); 6703 } 6704 6705 /* need to do it in software */ 6706 tcpstat_inc(tcps_inswcsum); 6707 6708 switch (af) { 6709 case AF_INET: 6710 if (m->m_len < sizeof(struct ip)) 6711 return (1); 6712 6713 sum = in4_cksum(m, IPPROTO_TCP, off, len); 6714 break; 6715 #ifdef INET6 6716 case AF_INET6: 6717 if (m->m_len < sizeof(struct ip6_hdr)) 6718 return (1); 6719 6720 sum = in6_cksum(m, IPPROTO_TCP, off, len); 6721 break; 6722 #endif /* INET6 */ 6723 default: 6724 unhandled_af(af); 6725 } 6726 if (sum) { 6727 tcpstat_inc(tcps_rcvbadsum); 6728 m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_BAD; 6729 return (1); 6730 } 6731 6732 m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK; 6733 return (0); 6734 } 6735 6736 struct pf_divert * 6737 pf_find_divert(struct mbuf *m) 6738 { 6739 struct m_tag *mtag; 6740 6741 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) 6742 return (NULL); 6743 6744 return ((struct pf_divert *)(mtag + 1)); 6745 } 6746 6747 struct pf_divert * 6748 pf_get_divert(struct mbuf *m) 6749 { 6750 struct m_tag *mtag; 6751 6752 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) { 6753 mtag = m_tag_get(PACKET_TAG_PF_DIVERT, sizeof(struct pf_divert), 6754 M_NOWAIT); 6755 if (mtag == NULL) 6756 return (NULL); 6757 memset(mtag + 1, 0, sizeof(struct pf_divert)); 6758 m_tag_prepend(m, mtag); 6759 } 6760 6761 return ((struct pf_divert *)(mtag + 1)); 6762 } 6763 6764 int 6765 pf_walk_option(struct pf_pdesc *pd, struct ip *h, int off, int end, 6766 u_short *reason) 6767 { 6768 uint8_t type, length, opts[15 * 4 - sizeof(struct ip)]; 6769 6770 /* IP header in payload of ICMP packet may be too short */ 6771 if (pd->m->m_pkthdr.len < end) { 6772 DPFPRINTF(LOG_NOTICE, "IP option too short"); 6773 REASON_SET(reason, PFRES_SHORT); 6774 return (PF_DROP); 6775 } 6776 6777 KASSERT(end - off <= sizeof(opts)); 6778 m_copydata(pd->m, off, end - off, opts); 6779 end -= off; 6780 off = 0; 6781 6782 while (off < end) { 6783 type = opts[off]; 6784 if (type == IPOPT_EOL) 6785 break; 6786 if (type == IPOPT_NOP) { 6787 off++; 6788 continue; 6789 } 6790 if (off + 2 > end) { 6791 DPFPRINTF(LOG_NOTICE, "IP length opt"); 6792 REASON_SET(reason, PFRES_IPOPTIONS); 6793 return (PF_DROP); 6794 } 6795 length = opts[off + 1]; 6796 if (length < 2) { 6797 DPFPRINTF(LOG_NOTICE, "IP short opt"); 6798 REASON_SET(reason, PFRES_IPOPTIONS); 6799 return (PF_DROP); 6800 } 6801 if (off + length > end) { 6802 DPFPRINTF(LOG_NOTICE, "IP long opt"); 6803 REASON_SET(reason, PFRES_IPOPTIONS); 6804 return (PF_DROP); 6805 } 6806 switch (type) { 6807 case IPOPT_RA: 6808 SET(pd->badopts, PF_OPT_ROUTER_ALERT); 6809 break; 6810 default: 6811 SET(pd->badopts, PF_OPT_OTHER); 6812 break; 6813 } 6814 off += length; 6815 } 6816 6817 return (PF_PASS); 6818 } 6819 6820 int 6821 pf_walk_header(struct pf_pdesc *pd, struct ip *h, u_short *reason) 6822 { 6823 struct ip6_ext ext; 6824 u_int32_t hlen, end; 6825 int hdr_cnt; 6826 6827 hlen = h->ip_hl << 2; 6828 if (hlen < sizeof(struct ip) || hlen > ntohs(h->ip_len)) { 6829 REASON_SET(reason, PFRES_SHORT); 6830 return (PF_DROP); 6831 } 6832 if (hlen != sizeof(struct ip)) { 6833 if (pf_walk_option(pd, h, pd->off + sizeof(struct ip), 6834 pd->off + hlen, reason) != PF_PASS) 6835 return (PF_DROP); 6836 /* header options which contain only padding is fishy */ 6837 if (pd->badopts == 0) 6838 SET(pd->badopts, PF_OPT_OTHER); 6839 } 6840 end = pd->off + ntohs(h->ip_len); 6841 pd->off += hlen; 6842 pd->proto = h->ip_p; 6843 /* IGMP packets have router alert options, allow them */ 6844 if (pd->proto == IPPROTO_IGMP) { 6845 /* According to RFC 1112 ttl must be set to 1. */ 6846 if ((h->ip_ttl != 1) || !IN_MULTICAST(h->ip_dst.s_addr)) { 6847 DPFPRINTF(LOG_NOTICE, "Invalid IGMP"); 6848 REASON_SET(reason, PFRES_IPOPTIONS); 6849 return (PF_DROP); 6850 } 6851 CLR(pd->badopts, PF_OPT_ROUTER_ALERT); 6852 } 6853 /* stop walking over non initial fragments */ 6854 if ((h->ip_off & htons(IP_OFFMASK)) != 0) 6855 return (PF_PASS); 6856 6857 for (hdr_cnt = 0; hdr_cnt < pf_hdr_limit; hdr_cnt++) { 6858 switch (pd->proto) { 6859 case IPPROTO_AH: 6860 /* fragments may be short */ 6861 if ((h->ip_off & htons(IP_MF | IP_OFFMASK)) != 0 && 6862 end < pd->off + sizeof(ext)) 6863 return (PF_PASS); 6864 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 6865 NULL, reason, AF_INET)) { 6866 DPFPRINTF(LOG_NOTICE, "IP short exthdr"); 6867 return (PF_DROP); 6868 } 6869 pd->off += (ext.ip6e_len + 2) * 4; 6870 pd->proto = ext.ip6e_nxt; 6871 break; 6872 default: 6873 return (PF_PASS); 6874 } 6875 } 6876 DPFPRINTF(LOG_NOTICE, "IPv4 nested authentication header limit"); 6877 REASON_SET(reason, PFRES_IPOPTIONS); 6878 return (PF_DROP); 6879 } 6880 6881 #ifdef INET6 6882 int 6883 pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end, 6884 u_short *reason) 6885 { 6886 struct ip6_opt opt; 6887 struct ip6_opt_jumbo jumbo; 6888 6889 while (off < end) { 6890 if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type, 6891 sizeof(opt.ip6o_type), NULL, reason, AF_INET6)) { 6892 DPFPRINTF(LOG_NOTICE, "IPv6 short opt type"); 6893 return (PF_DROP); 6894 } 6895 if (opt.ip6o_type == IP6OPT_PAD1) { 6896 off++; 6897 continue; 6898 } 6899 if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), 6900 NULL, reason, AF_INET6)) { 6901 DPFPRINTF(LOG_NOTICE, "IPv6 short opt"); 6902 return (PF_DROP); 6903 } 6904 if (off + sizeof(opt) + opt.ip6o_len > end) { 6905 DPFPRINTF(LOG_NOTICE, "IPv6 long opt"); 6906 REASON_SET(reason, PFRES_IPOPTIONS); 6907 return (PF_DROP); 6908 } 6909 switch (opt.ip6o_type) { 6910 case IP6OPT_PADN: 6911 break; 6912 case IP6OPT_JUMBO: 6913 SET(pd->badopts, PF_OPT_JUMBO); 6914 if (pd->jumbolen != 0) { 6915 DPFPRINTF(LOG_NOTICE, "IPv6 multiple jumbo"); 6916 REASON_SET(reason, PFRES_IPOPTIONS); 6917 return (PF_DROP); 6918 } 6919 if (ntohs(h->ip6_plen) != 0) { 6920 DPFPRINTF(LOG_NOTICE, "IPv6 bad jumbo plen"); 6921 REASON_SET(reason, PFRES_IPOPTIONS); 6922 return (PF_DROP); 6923 } 6924 if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), 6925 NULL, reason, AF_INET6)) { 6926 DPFPRINTF(LOG_NOTICE, "IPv6 short jumbo"); 6927 return (PF_DROP); 6928 } 6929 memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len, 6930 sizeof(pd->jumbolen)); 6931 pd->jumbolen = ntohl(pd->jumbolen); 6932 if (pd->jumbolen < IPV6_MAXPACKET) { 6933 DPFPRINTF(LOG_NOTICE, "IPv6 short jumbolen"); 6934 REASON_SET(reason, PFRES_IPOPTIONS); 6935 return (PF_DROP); 6936 } 6937 break; 6938 case IP6OPT_ROUTER_ALERT: 6939 SET(pd->badopts, PF_OPT_ROUTER_ALERT); 6940 break; 6941 default: 6942 SET(pd->badopts, PF_OPT_OTHER); 6943 break; 6944 } 6945 off += sizeof(opt) + opt.ip6o_len; 6946 } 6947 6948 return (PF_PASS); 6949 } 6950 6951 int 6952 pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason) 6953 { 6954 struct ip6_frag frag; 6955 struct ip6_ext ext; 6956 struct icmp6_hdr icmp6; 6957 struct ip6_rthdr rthdr; 6958 u_int32_t end; 6959 int hdr_cnt, fraghdr_cnt = 0, rthdr_cnt = 0; 6960 6961 pd->off += sizeof(struct ip6_hdr); 6962 end = pd->off + ntohs(h->ip6_plen); 6963 pd->fragoff = pd->extoff = pd->jumbolen = 0; 6964 pd->proto = h->ip6_nxt; 6965 6966 for (hdr_cnt = 0; hdr_cnt < pf_hdr_limit; hdr_cnt++) { 6967 switch (pd->proto) { 6968 case IPPROTO_ROUTING: 6969 case IPPROTO_DSTOPTS: 6970 SET(pd->badopts, PF_OPT_OTHER); 6971 break; 6972 case IPPROTO_HOPOPTS: 6973 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 6974 NULL, reason, AF_INET6)) { 6975 DPFPRINTF(LOG_NOTICE, "IPv6 short exthdr"); 6976 return (PF_DROP); 6977 } 6978 if (pf_walk_option6(pd, h, pd->off + sizeof(ext), 6979 pd->off + (ext.ip6e_len + 1) * 8, reason) 6980 != PF_PASS) 6981 return (PF_DROP); 6982 /* option header which contains only padding is fishy */ 6983 if (pd->badopts == 0) 6984 SET(pd->badopts, PF_OPT_OTHER); 6985 break; 6986 } 6987 switch (pd->proto) { 6988 case IPPROTO_FRAGMENT: 6989 if (fraghdr_cnt++) { 6990 DPFPRINTF(LOG_NOTICE, "IPv6 multiple fragment"); 6991 REASON_SET(reason, PFRES_FRAG); 6992 return (PF_DROP); 6993 } 6994 /* jumbo payload packets cannot be fragmented */ 6995 if (pd->jumbolen != 0) { 6996 DPFPRINTF(LOG_NOTICE, "IPv6 fragmented jumbo"); 6997 REASON_SET(reason, PFRES_FRAG); 6998 return (PF_DROP); 6999 } 7000 if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag), 7001 NULL, reason, AF_INET6)) { 7002 DPFPRINTF(LOG_NOTICE, "IPv6 short fragment"); 7003 return (PF_DROP); 7004 } 7005 /* stop walking over non initial fragments */ 7006 if (ntohs((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) { 7007 pd->fragoff = pd->off; 7008 return (PF_PASS); 7009 } 7010 /* RFC6946: reassemble only non atomic fragments */ 7011 if (frag.ip6f_offlg & IP6F_MORE_FRAG) 7012 pd->fragoff = pd->off; 7013 pd->off += sizeof(frag); 7014 pd->proto = frag.ip6f_nxt; 7015 break; 7016 case IPPROTO_ROUTING: 7017 if (rthdr_cnt++) { 7018 DPFPRINTF(LOG_NOTICE, "IPv6 multiple rthdr"); 7019 REASON_SET(reason, PFRES_IPOPTIONS); 7020 return (PF_DROP); 7021 } 7022 /* fragments may be short */ 7023 if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) { 7024 pd->off = pd->fragoff; 7025 pd->proto = IPPROTO_FRAGMENT; 7026 return (PF_PASS); 7027 } 7028 if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr), 7029 NULL, reason, AF_INET6)) { 7030 DPFPRINTF(LOG_NOTICE, "IPv6 short rthdr"); 7031 return (PF_DROP); 7032 } 7033 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 7034 DPFPRINTF(LOG_NOTICE, "IPv6 rthdr0"); 7035 REASON_SET(reason, PFRES_IPOPTIONS); 7036 return (PF_DROP); 7037 } 7038 /* FALLTHROUGH */ 7039 case IPPROTO_HOPOPTS: 7040 /* RFC2460 4.1: Hop-by-Hop only after IPv6 header */ 7041 if (pd->proto == IPPROTO_HOPOPTS && hdr_cnt > 0) { 7042 DPFPRINTF(LOG_NOTICE, "IPv6 hopopts not first"); 7043 REASON_SET(reason, PFRES_IPOPTIONS); 7044 return (PF_DROP); 7045 } 7046 /* FALLTHROUGH */ 7047 case IPPROTO_AH: 7048 case IPPROTO_DSTOPTS: 7049 /* fragments may be short */ 7050 if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) { 7051 pd->off = pd->fragoff; 7052 pd->proto = IPPROTO_FRAGMENT; 7053 return (PF_PASS); 7054 } 7055 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 7056 NULL, reason, AF_INET6)) { 7057 DPFPRINTF(LOG_NOTICE, "IPv6 short exthdr"); 7058 return (PF_DROP); 7059 } 7060 /* reassembly needs the ext header before the frag */ 7061 if (pd->fragoff == 0) 7062 pd->extoff = pd->off; 7063 if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0 && 7064 ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) { 7065 DPFPRINTF(LOG_NOTICE, "IPv6 missing jumbo"); 7066 REASON_SET(reason, PFRES_IPOPTIONS); 7067 return (PF_DROP); 7068 } 7069 if (pd->proto == IPPROTO_AH) 7070 pd->off += (ext.ip6e_len + 2) * 4; 7071 else 7072 pd->off += (ext.ip6e_len + 1) * 8; 7073 pd->proto = ext.ip6e_nxt; 7074 break; 7075 case IPPROTO_ICMPV6: 7076 /* fragments may be short, ignore inner header then */ 7077 if (pd->fragoff != 0 && end < pd->off + sizeof(icmp6)) { 7078 pd->off = pd->fragoff; 7079 pd->proto = IPPROTO_FRAGMENT; 7080 return (PF_PASS); 7081 } 7082 if (!pf_pull_hdr(pd->m, pd->off, &icmp6, sizeof(icmp6), 7083 NULL, reason, AF_INET6)) { 7084 DPFPRINTF(LOG_NOTICE, "IPv6 short icmp6hdr"); 7085 return (PF_DROP); 7086 } 7087 /* ICMP multicast packets have router alert options */ 7088 switch (icmp6.icmp6_type) { 7089 case MLD_LISTENER_QUERY: 7090 case MLD_LISTENER_REPORT: 7091 case MLD_LISTENER_DONE: 7092 case MLDV2_LISTENER_REPORT: 7093 /* 7094 * According to RFC 2710 all MLD messages are 7095 * sent with hop-limit (ttl) set to 1, and link 7096 * local source address. If either one is 7097 * missing then MLD message is invalid and 7098 * should be discarded. 7099 */ 7100 if ((h->ip6_hlim != 1) || 7101 !IN6_IS_ADDR_LINKLOCAL(&h->ip6_src)) { 7102 DPFPRINTF(LOG_NOTICE, "Invalid MLD"); 7103 REASON_SET(reason, PFRES_IPOPTIONS); 7104 return (PF_DROP); 7105 } 7106 CLR(pd->badopts, PF_OPT_ROUTER_ALERT); 7107 break; 7108 } 7109 return (PF_PASS); 7110 case IPPROTO_TCP: 7111 case IPPROTO_UDP: 7112 /* fragments may be short, ignore inner header then */ 7113 if (pd->fragoff != 0 && end < pd->off + 7114 (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) : 7115 pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) : 7116 sizeof(struct icmp6_hdr))) { 7117 pd->off = pd->fragoff; 7118 pd->proto = IPPROTO_FRAGMENT; 7119 } 7120 /* FALLTHROUGH */ 7121 default: 7122 return (PF_PASS); 7123 } 7124 } 7125 DPFPRINTF(LOG_NOTICE, "IPv6 nested extension header limit"); 7126 REASON_SET(reason, PFRES_IPOPTIONS); 7127 return (PF_DROP); 7128 } 7129 #endif /* INET6 */ 7130 7131 u_int16_t 7132 pf_pkt_hash(sa_family_t af, uint8_t proto, 7133 const struct pf_addr *src, const struct pf_addr *dst, 7134 uint16_t sport, uint16_t dport) 7135 { 7136 uint32_t hash; 7137 7138 hash = src->addr32[0] ^ dst->addr32[0]; 7139 #ifdef INET6 7140 if (af == AF_INET6) { 7141 hash ^= src->addr32[1] ^ dst->addr32[1]; 7142 hash ^= src->addr32[2] ^ dst->addr32[2]; 7143 hash ^= src->addr32[3] ^ dst->addr32[3]; 7144 } 7145 #endif 7146 7147 switch (proto) { 7148 case IPPROTO_TCP: 7149 case IPPROTO_UDP: 7150 hash ^= sport ^ dport; 7151 break; 7152 } 7153 7154 return stoeplitz_n32(hash); 7155 } 7156 7157 int 7158 pf_setup_pdesc(struct pf_pdesc *pd, sa_family_t af, int dir, 7159 struct pfi_kif *kif, struct mbuf *m, u_short *reason) 7160 { 7161 memset(pd, 0, sizeof(*pd)); 7162 pd->dir = dir; 7163 pd->kif = kif; /* kif is NULL when called by pflog */ 7164 pd->m = m; 7165 pd->sidx = (dir == PF_IN) ? 0 : 1; 7166 pd->didx = (dir == PF_IN) ? 1 : 0; 7167 pd->af = pd->naf = af; 7168 pd->rdomain = rtable_l2(pd->m->m_pkthdr.ph_rtableid); 7169 7170 switch (pd->af) { 7171 case AF_INET: { 7172 struct ip *h; 7173 7174 /* Check for illegal packets */ 7175 if (pd->m->m_pkthdr.len < (int)sizeof(struct ip)) { 7176 REASON_SET(reason, PFRES_SHORT); 7177 return (PF_DROP); 7178 } 7179 7180 h = mtod(pd->m, struct ip *); 7181 if (pd->m->m_pkthdr.len < ntohs(h->ip_len)) { 7182 REASON_SET(reason, PFRES_SHORT); 7183 return (PF_DROP); 7184 } 7185 7186 if (pf_walk_header(pd, h, reason) != PF_PASS) 7187 return (PF_DROP); 7188 7189 pd->src = (struct pf_addr *)&h->ip_src; 7190 pd->dst = (struct pf_addr *)&h->ip_dst; 7191 pd->tot_len = ntohs(h->ip_len); 7192 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK; 7193 pd->ttl = h->ip_ttl; 7194 pd->virtual_proto = (h->ip_off & htons(IP_MF | IP_OFFMASK)) ? 7195 PF_VPROTO_FRAGMENT : pd->proto; 7196 7197 break; 7198 } 7199 #ifdef INET6 7200 case AF_INET6: { 7201 struct ip6_hdr *h; 7202 7203 /* Check for illegal packets */ 7204 if (pd->m->m_pkthdr.len < (int)sizeof(struct ip6_hdr)) { 7205 REASON_SET(reason, PFRES_SHORT); 7206 return (PF_DROP); 7207 } 7208 7209 h = mtod(pd->m, struct ip6_hdr *); 7210 if (pd->m->m_pkthdr.len < 7211 sizeof(struct ip6_hdr) + ntohs(h->ip6_plen)) { 7212 REASON_SET(reason, PFRES_SHORT); 7213 return (PF_DROP); 7214 } 7215 7216 if (pf_walk_header6(pd, h, reason) != PF_PASS) 7217 return (PF_DROP); 7218 7219 #if 1 7220 /* 7221 * we do not support jumbogram yet. if we keep going, zero 7222 * ip6_plen will do something bad, so drop the packet for now. 7223 */ 7224 if (pd->jumbolen != 0) { 7225 REASON_SET(reason, PFRES_NORM); 7226 return (PF_DROP); 7227 } 7228 #endif /* 1 */ 7229 7230 pd->src = (struct pf_addr *)&h->ip6_src; 7231 pd->dst = (struct pf_addr *)&h->ip6_dst; 7232 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 7233 pd->tos = (ntohl(h->ip6_flow) & 0x0fc00000) >> 20; 7234 pd->ttl = h->ip6_hlim; 7235 pd->virtual_proto = (pd->fragoff != 0) ? 7236 PF_VPROTO_FRAGMENT : pd->proto; 7237 7238 break; 7239 } 7240 #endif /* INET6 */ 7241 default: 7242 panic("pf_setup_pdesc called with illegal af %u", pd->af); 7243 7244 } 7245 7246 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 7247 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 7248 7249 switch (pd->virtual_proto) { 7250 case IPPROTO_TCP: { 7251 struct tcphdr *th = &pd->hdr.tcp; 7252 7253 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), 7254 NULL, reason, pd->af)) 7255 return (PF_DROP); 7256 pd->hdrlen = sizeof(*th); 7257 if (pd->off + (th->th_off << 2) > pd->tot_len || 7258 (th->th_off << 2) < sizeof(struct tcphdr)) { 7259 REASON_SET(reason, PFRES_SHORT); 7260 return (PF_DROP); 7261 } 7262 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 7263 pd->sport = &th->th_sport; 7264 pd->dport = &th->th_dport; 7265 pd->pcksum = &th->th_sum; 7266 break; 7267 } 7268 case IPPROTO_UDP: { 7269 struct udphdr *uh = &pd->hdr.udp; 7270 7271 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), 7272 NULL, reason, pd->af)) 7273 return (PF_DROP); 7274 pd->hdrlen = sizeof(*uh); 7275 if (uh->uh_dport == 0 || 7276 pd->off + ntohs(uh->uh_ulen) > pd->tot_len || 7277 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 7278 REASON_SET(reason, PFRES_SHORT); 7279 return (PF_DROP); 7280 } 7281 pd->sport = &uh->uh_sport; 7282 pd->dport = &uh->uh_dport; 7283 pd->pcksum = &uh->uh_sum; 7284 break; 7285 } 7286 case IPPROTO_ICMP: { 7287 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 7288 NULL, reason, pd->af)) 7289 return (PF_DROP); 7290 pd->hdrlen = ICMP_MINLEN; 7291 if (pd->off + pd->hdrlen > pd->tot_len) { 7292 REASON_SET(reason, PFRES_SHORT); 7293 return (PF_DROP); 7294 } 7295 pd->pcksum = &pd->hdr.icmp.icmp_cksum; 7296 break; 7297 } 7298 #ifdef INET6 7299 case IPPROTO_ICMPV6: { 7300 size_t icmp_hlen = sizeof(struct icmp6_hdr); 7301 7302 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 7303 NULL, reason, pd->af)) 7304 return (PF_DROP); 7305 /* ICMP headers we look further into to match state */ 7306 switch (pd->hdr.icmp6.icmp6_type) { 7307 case MLD_LISTENER_QUERY: 7308 case MLD_LISTENER_REPORT: 7309 icmp_hlen = sizeof(struct mld_hdr); 7310 break; 7311 case ND_NEIGHBOR_SOLICIT: 7312 case ND_NEIGHBOR_ADVERT: 7313 icmp_hlen = sizeof(struct nd_neighbor_solicit); 7314 /* FALLTHROUGH */ 7315 case ND_ROUTER_SOLICIT: 7316 case ND_ROUTER_ADVERT: 7317 case ND_REDIRECT: 7318 if (pd->ttl != 255) { 7319 REASON_SET(reason, PFRES_NORM); 7320 return (PF_DROP); 7321 } 7322 break; 7323 } 7324 if (icmp_hlen > sizeof(struct icmp6_hdr) && 7325 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 7326 NULL, reason, pd->af)) 7327 return (PF_DROP); 7328 pd->hdrlen = icmp_hlen; 7329 if (pd->off + pd->hdrlen > pd->tot_len) { 7330 REASON_SET(reason, PFRES_SHORT); 7331 return (PF_DROP); 7332 } 7333 pd->pcksum = &pd->hdr.icmp6.icmp6_cksum; 7334 break; 7335 } 7336 #endif /* INET6 */ 7337 } 7338 7339 if (pd->sport) 7340 pd->osport = pd->nsport = *pd->sport; 7341 if (pd->dport) 7342 pd->odport = pd->ndport = *pd->dport; 7343 7344 pd->hash = pf_pkt_hash(pd->af, pd->proto, 7345 pd->src, pd->dst, pd->osport, pd->odport); 7346 7347 return (PF_PASS); 7348 } 7349 7350 void 7351 pf_counters_inc(int action, struct pf_pdesc *pd, struct pf_state *st, 7352 struct pf_rule *r, struct pf_rule *a) 7353 { 7354 int dirndx; 7355 pd->kif->pfik_bytes[pd->af == AF_INET6][pd->dir == PF_OUT] 7356 [action != PF_PASS] += pd->tot_len; 7357 pd->kif->pfik_packets[pd->af == AF_INET6][pd->dir == PF_OUT] 7358 [action != PF_PASS]++; 7359 7360 if (action == PF_PASS || action == PF_AFRT || r->action == PF_DROP) { 7361 dirndx = (pd->dir == PF_OUT); 7362 r->packets[dirndx]++; 7363 r->bytes[dirndx] += pd->tot_len; 7364 if (a != NULL) { 7365 a->packets[dirndx]++; 7366 a->bytes[dirndx] += pd->tot_len; 7367 } 7368 if (st != NULL) { 7369 struct pf_rule_item *ri; 7370 struct pf_sn_item *sni; 7371 7372 SLIST_FOREACH(sni, &st->src_nodes, next) { 7373 sni->sn->packets[dirndx]++; 7374 sni->sn->bytes[dirndx] += pd->tot_len; 7375 } 7376 dirndx = (pd->dir == st->direction) ? 0 : 1; 7377 st->packets[dirndx]++; 7378 st->bytes[dirndx] += pd->tot_len; 7379 7380 SLIST_FOREACH(ri, &st->match_rules, entry) { 7381 ri->r->packets[dirndx]++; 7382 ri->r->bytes[dirndx] += pd->tot_len; 7383 7384 if (ri->r->src.addr.type == PF_ADDR_TABLE) 7385 pfr_update_stats(ri->r->src.addr.p.tbl, 7386 &st->key[(st->direction == PF_IN)]-> 7387 addr[(st->direction == PF_OUT)], 7388 pd, ri->r->action, ri->r->src.neg); 7389 if (ri->r->dst.addr.type == PF_ADDR_TABLE) 7390 pfr_update_stats(ri->r->dst.addr.p.tbl, 7391 &st->key[(st->direction == PF_IN)]-> 7392 addr[(st->direction == PF_IN)], 7393 pd, ri->r->action, ri->r->dst.neg); 7394 } 7395 } 7396 if (r->src.addr.type == PF_ADDR_TABLE) 7397 pfr_update_stats(r->src.addr.p.tbl, 7398 (st == NULL) ? pd->src : 7399 &st->key[(st->direction == PF_IN)]-> 7400 addr[(st->direction == PF_OUT)], 7401 pd, r->action, r->src.neg); 7402 if (r->dst.addr.type == PF_ADDR_TABLE) 7403 pfr_update_stats(r->dst.addr.p.tbl, 7404 (st == NULL) ? pd->dst : 7405 &st->key[(st->direction == PF_IN)]-> 7406 addr[(st->direction == PF_IN)], 7407 pd, r->action, r->dst.neg); 7408 } 7409 } 7410 7411 int 7412 pf_test(sa_family_t af, int fwdir, struct ifnet *ifp, struct mbuf **m0) 7413 { 7414 #if NCARP > 0 7415 struct ifnet *ifp0; 7416 #endif 7417 struct pfi_kif *kif; 7418 u_short action, reason = 0; 7419 struct pf_rule *a = NULL, *r = &pf_default_rule; 7420 struct pf_state *st = NULL; 7421 struct pf_state_key_cmp key; 7422 struct pf_ruleset *ruleset = NULL; 7423 struct pf_pdesc pd; 7424 int dir = (fwdir == PF_FWD) ? PF_OUT : fwdir; 7425 u_int32_t qid, pqid = 0; 7426 int have_pf_lock = 0; 7427 struct pfsync_deferral *deferral = NULL; 7428 7429 if (!pf_status.running) 7430 return (PF_PASS); 7431 7432 #if NCARP > 0 7433 if (ifp->if_type == IFT_CARP && 7434 (ifp0 = if_get(ifp->if_carpdevidx)) != NULL) { 7435 kif = (struct pfi_kif *)ifp0->if_pf_kif; 7436 if_put(ifp0); 7437 } else 7438 #endif /* NCARP */ 7439 kif = (struct pfi_kif *)ifp->if_pf_kif; 7440 7441 if (kif == NULL) { 7442 DPFPRINTF(LOG_ERR, 7443 "%s: kif == NULL, if_xname %s", __func__, ifp->if_xname); 7444 return (PF_DROP); 7445 } 7446 if (kif->pfik_flags & PFI_IFLAG_SKIP) 7447 return (PF_PASS); 7448 7449 #ifdef DIAGNOSTIC 7450 if (((*m0)->m_flags & M_PKTHDR) == 0) 7451 panic("non-M_PKTHDR is passed to pf_test"); 7452 #endif /* DIAGNOSTIC */ 7453 7454 if ((*m0)->m_pkthdr.pf.flags & PF_TAG_GENERATED) 7455 return (PF_PASS); 7456 7457 if ((*m0)->m_pkthdr.pf.flags & PF_TAG_DIVERTED_PACKET) { 7458 (*m0)->m_pkthdr.pf.flags &= ~PF_TAG_DIVERTED_PACKET; 7459 return (PF_PASS); 7460 } 7461 7462 if ((*m0)->m_pkthdr.pf.flags & PF_TAG_REFRAGMENTED) { 7463 (*m0)->m_pkthdr.pf.flags &= ~PF_TAG_REFRAGMENTED; 7464 return (PF_PASS); 7465 } 7466 7467 action = pf_setup_pdesc(&pd, af, dir, kif, *m0, &reason); 7468 if (action != PF_PASS) { 7469 #if NPFLOG > 0 7470 pd.pflog |= PF_LOG_FORCE; 7471 #endif /* NPFLOG > 0 */ 7472 goto done; 7473 } 7474 7475 /* packet normalization and reassembly */ 7476 switch (pd.af) { 7477 case AF_INET: 7478 action = pf_normalize_ip(&pd, &reason); 7479 break; 7480 #ifdef INET6 7481 case AF_INET6: 7482 action = pf_normalize_ip6(&pd, &reason); 7483 break; 7484 #endif /* INET6 */ 7485 } 7486 *m0 = pd.m; 7487 /* if packet sits in reassembly queue, return without error */ 7488 if (pd.m == NULL) 7489 return PF_PASS; 7490 7491 if (action != PF_PASS) { 7492 #if NPFLOG > 0 7493 pd.pflog |= PF_LOG_FORCE; 7494 #endif /* NPFLOG > 0 */ 7495 goto done; 7496 } 7497 7498 /* if packet has been reassembled, update packet description */ 7499 if (pf_status.reass && pd.virtual_proto == PF_VPROTO_FRAGMENT) { 7500 action = pf_setup_pdesc(&pd, af, dir, kif, pd.m, &reason); 7501 if (action != PF_PASS) { 7502 #if NPFLOG > 0 7503 pd.pflog |= PF_LOG_FORCE; 7504 #endif /* NPFLOG > 0 */ 7505 goto done; 7506 } 7507 } 7508 pd.m->m_pkthdr.pf.flags |= PF_TAG_PROCESSED; 7509 7510 /* 7511 * Avoid pcb-lookups from the forwarding path. They should never 7512 * match and would cause MP locking problems. 7513 */ 7514 if (fwdir == PF_FWD) { 7515 pd.lookup.done = -1; 7516 pd.lookup.uid = -1; 7517 pd.lookup.gid = -1; 7518 pd.lookup.pid = NO_PID; 7519 } 7520 7521 switch (pd.virtual_proto) { 7522 7523 case PF_VPROTO_FRAGMENT: { 7524 /* 7525 * handle fragments that aren't reassembled by 7526 * normalization 7527 */ 7528 PF_LOCK(); 7529 have_pf_lock = 1; 7530 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, &reason, 7531 &deferral); 7532 st = pf_state_ref(st); 7533 if (action != PF_PASS) 7534 REASON_SET(&reason, PFRES_FRAG); 7535 break; 7536 } 7537 7538 case IPPROTO_ICMP: { 7539 if (pd.af != AF_INET) { 7540 action = PF_DROP; 7541 REASON_SET(&reason, PFRES_NORM); 7542 DPFPRINTF(LOG_NOTICE, 7543 "dropping IPv6 packet with ICMPv4 payload"); 7544 break; 7545 } 7546 PF_STATE_ENTER_READ(); 7547 action = pf_test_state_icmp(&pd, &st, &reason); 7548 st = pf_state_ref(st); 7549 PF_STATE_EXIT_READ(); 7550 if (action == PF_PASS || action == PF_AFRT) { 7551 #if NPFSYNC > 0 7552 pfsync_update_state(st); 7553 #endif /* NPFSYNC > 0 */ 7554 r = st->rule.ptr; 7555 a = st->anchor.ptr; 7556 #if NPFLOG > 0 7557 pd.pflog |= st->log; 7558 #endif /* NPFLOG > 0 */ 7559 } else if (st == NULL) { 7560 PF_LOCK(); 7561 have_pf_lock = 1; 7562 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, 7563 &reason, &deferral); 7564 st = pf_state_ref(st); 7565 } 7566 break; 7567 } 7568 7569 #ifdef INET6 7570 case IPPROTO_ICMPV6: { 7571 if (pd.af != AF_INET6) { 7572 action = PF_DROP; 7573 REASON_SET(&reason, PFRES_NORM); 7574 DPFPRINTF(LOG_NOTICE, 7575 "dropping IPv4 packet with ICMPv6 payload"); 7576 break; 7577 } 7578 PF_STATE_ENTER_READ(); 7579 action = pf_test_state_icmp(&pd, &st, &reason); 7580 st = pf_state_ref(st); 7581 PF_STATE_EXIT_READ(); 7582 if (action == PF_PASS || action == PF_AFRT) { 7583 #if NPFSYNC > 0 7584 pfsync_update_state(st); 7585 #endif /* NPFSYNC > 0 */ 7586 r = st->rule.ptr; 7587 a = st->anchor.ptr; 7588 #if NPFLOG > 0 7589 pd.pflog |= st->log; 7590 #endif /* NPFLOG > 0 */ 7591 } else if (st == NULL) { 7592 PF_LOCK(); 7593 have_pf_lock = 1; 7594 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, 7595 &reason, &deferral); 7596 st = pf_state_ref(st); 7597 } 7598 break; 7599 } 7600 #endif /* INET6 */ 7601 7602 default: 7603 if (pd.virtual_proto == IPPROTO_TCP) { 7604 if (pd.dir == PF_IN && (pd.hdr.tcp.th_flags & 7605 (TH_SYN|TH_ACK)) == TH_SYN && 7606 pf_synflood_check(&pd)) { 7607 PF_LOCK(); 7608 have_pf_lock = 1; 7609 pf_syncookie_send(&pd); 7610 action = PF_DROP; 7611 break; 7612 } 7613 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 7614 pqid = 1; 7615 action = pf_normalize_tcp(&pd); 7616 if (action == PF_DROP) 7617 break; 7618 } 7619 7620 key.af = pd.af; 7621 key.proto = pd.virtual_proto; 7622 key.rdomain = pd.rdomain; 7623 pf_addrcpy(&key.addr[pd.sidx], pd.src, key.af); 7624 pf_addrcpy(&key.addr[pd.didx], pd.dst, key.af); 7625 key.port[pd.sidx] = pd.osport; 7626 key.port[pd.didx] = pd.odport; 7627 key.hash = pd.hash; 7628 7629 PF_STATE_ENTER_READ(); 7630 action = pf_find_state(&pd, &key, &st); 7631 st = pf_state_ref(st); 7632 PF_STATE_EXIT_READ(); 7633 7634 /* check for syncookies if tcp ack and no active state */ 7635 if (pd.dir == PF_IN && pd.virtual_proto == IPPROTO_TCP && 7636 (st == NULL || (st->src.state >= TCPS_FIN_WAIT_2 && 7637 st->dst.state >= TCPS_FIN_WAIT_2)) && 7638 (pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 7639 pf_syncookie_validate(&pd)) { 7640 struct mbuf *msyn = pf_syncookie_recreate_syn(&pd); 7641 if (msyn) { 7642 action = pf_test(af, fwdir, ifp, &msyn); 7643 m_freem(msyn); 7644 if (action == PF_PASS || action == PF_AFRT) { 7645 PF_STATE_ENTER_READ(); 7646 pf_state_unref(st); 7647 action = pf_find_state(&pd, &key, &st); 7648 st = pf_state_ref(st); 7649 PF_STATE_EXIT_READ(); 7650 if (st == NULL) 7651 return (PF_DROP); 7652 st->src.seqhi = st->dst.seqhi = 7653 ntohl(pd.hdr.tcp.th_ack) - 1; 7654 st->src.seqlo = 7655 ntohl(pd.hdr.tcp.th_seq) - 1; 7656 pf_set_protostate(st, PF_PEER_SRC, 7657 PF_TCPS_PROXY_DST); 7658 } 7659 } else 7660 action = PF_DROP; 7661 } 7662 7663 if (action == PF_MATCH) 7664 action = pf_test_state(&pd, &st, &reason); 7665 7666 if (action == PF_PASS || action == PF_AFRT) { 7667 #if NPFSYNC > 0 7668 pfsync_update_state(st); 7669 #endif /* NPFSYNC > 0 */ 7670 r = st->rule.ptr; 7671 a = st->anchor.ptr; 7672 #if NPFLOG > 0 7673 pd.pflog |= st->log; 7674 #endif /* NPFLOG > 0 */ 7675 } else if (st == NULL) { 7676 PF_LOCK(); 7677 have_pf_lock = 1; 7678 action = pf_test_rule(&pd, &r, &st, &a, &ruleset, 7679 &reason, &deferral); 7680 st = pf_state_ref(st); 7681 } 7682 7683 if (pd.virtual_proto == IPPROTO_TCP) { 7684 if (st) { 7685 if (st->max_mss) 7686 pf_normalize_mss(&pd, st->max_mss); 7687 } else if (r->max_mss) 7688 pf_normalize_mss(&pd, r->max_mss); 7689 } 7690 7691 break; 7692 } 7693 7694 if (have_pf_lock != 0) 7695 PF_UNLOCK(); 7696 7697 /* 7698 * At the moment, we rely on NET_LOCK() to prevent removal of items 7699 * we've collected above ('r', 'anchor' and 'ruleset'). They'll have 7700 * to be refcounted when NET_LOCK() is gone. 7701 */ 7702 7703 done: 7704 if (action != PF_DROP) { 7705 if (st) { 7706 /* The non-state case is handled in pf_test_rule() */ 7707 if (action == PF_PASS && pd.badopts != 0 && 7708 !(st->state_flags & PFSTATE_ALLOWOPTS)) { 7709 action = PF_DROP; 7710 REASON_SET(&reason, PFRES_IPOPTIONS); 7711 #if NPFLOG > 0 7712 pd.pflog |= PF_LOG_FORCE; 7713 #endif /* NPFLOG > 0 */ 7714 DPFPRINTF(LOG_NOTICE, "dropping packet with " 7715 "ip/ipv6 options in pf_test()"); 7716 } 7717 7718 pf_scrub(pd.m, st->state_flags, pd.af, st->min_ttl, 7719 st->set_tos); 7720 pf_tag_packet(pd.m, st->tag, st->rtableid[pd.didx]); 7721 if (pqid || (pd.tos & IPTOS_LOWDELAY)) { 7722 qid = st->pqid; 7723 if (st->state_flags & PFSTATE_SETPRIO) { 7724 pd.m->m_pkthdr.pf.prio = 7725 st->set_prio[1]; 7726 } 7727 } else { 7728 qid = st->qid; 7729 if (st->state_flags & PFSTATE_SETPRIO) { 7730 pd.m->m_pkthdr.pf.prio = 7731 st->set_prio[0]; 7732 } 7733 } 7734 pd.m->m_pkthdr.pf.delay = st->delay; 7735 } else { 7736 pf_scrub(pd.m, r->scrub_flags, pd.af, r->min_ttl, 7737 r->set_tos); 7738 if (pqid || (pd.tos & IPTOS_LOWDELAY)) { 7739 qid = r->pqid; 7740 if (r->scrub_flags & PFSTATE_SETPRIO) 7741 pd.m->m_pkthdr.pf.prio = r->set_prio[1]; 7742 } else { 7743 qid = r->qid; 7744 if (r->scrub_flags & PFSTATE_SETPRIO) 7745 pd.m->m_pkthdr.pf.prio = r->set_prio[0]; 7746 } 7747 pd.m->m_pkthdr.pf.delay = r->delay; 7748 } 7749 } 7750 7751 if (action == PF_PASS && qid) 7752 pd.m->m_pkthdr.pf.qid = qid; 7753 if (pd.dir == PF_IN && st && st->key[PF_SK_STACK]) 7754 pf_mbuf_link_state_key(pd.m, st->key[PF_SK_STACK]); 7755 if (pd.dir == PF_OUT && 7756 pd.m->m_pkthdr.pf.inp && !pd.m->m_pkthdr.pf.inp->inp_pf_sk && 7757 st && st->key[PF_SK_STACK] && !st->key[PF_SK_STACK]->sk_inp) 7758 pf_state_key_link_inpcb(st->key[PF_SK_STACK], 7759 pd.m->m_pkthdr.pf.inp); 7760 7761 if (st != NULL && !ISSET(pd.m->m_pkthdr.csum_flags, M_FLOWID)) { 7762 pd.m->m_pkthdr.ph_flowid = st->key[PF_SK_WIRE]->hash; 7763 SET(pd.m->m_pkthdr.csum_flags, M_FLOWID); 7764 } 7765 7766 /* 7767 * connections redirected to loopback should not match sockets 7768 * bound specifically to loopback due to security implications, 7769 * see in_pcblookup_listen(). 7770 */ 7771 if (pd.destchg) 7772 if ((pd.af == AF_INET && (ntohl(pd.dst->v4.s_addr) >> 7773 IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) || 7774 (pd.af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))) 7775 pd.m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST; 7776 /* We need to redo the route lookup on outgoing routes. */ 7777 if (pd.destchg && pd.dir == PF_OUT) 7778 pd.m->m_pkthdr.pf.flags |= PF_TAG_REROUTE; 7779 7780 if (pd.dir == PF_IN && action == PF_PASS && 7781 (r->divert.type == PF_DIVERT_TO || 7782 r->divert.type == PF_DIVERT_REPLY)) { 7783 struct pf_divert *divert; 7784 7785 if ((divert = pf_get_divert(pd.m))) { 7786 pd.m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED; 7787 divert->addr = r->divert.addr; 7788 divert->port = r->divert.port; 7789 divert->rdomain = pd.rdomain; 7790 divert->type = r->divert.type; 7791 } 7792 } 7793 7794 if (action == PF_PASS && r->divert.type == PF_DIVERT_PACKET) 7795 action = PF_DIVERT; 7796 7797 #if NPFLOG > 0 7798 if (pd.pflog) { 7799 struct pf_rule_item *ri; 7800 7801 if (pd.pflog & PF_LOG_FORCE || r->log & PF_LOG_ALL) 7802 pflog_packet(&pd, reason, r, a, ruleset, NULL); 7803 if (st) { 7804 SLIST_FOREACH(ri, &st->match_rules, entry) 7805 if (ri->r->log & PF_LOG_ALL) 7806 pflog_packet(&pd, reason, ri->r, a, 7807 ruleset, NULL); 7808 } 7809 } 7810 #endif /* NPFLOG > 0 */ 7811 7812 pf_counters_inc(action, &pd, st, r, a); 7813 7814 switch (action) { 7815 case PF_SYNPROXY_DROP: 7816 m_freem(pd.m); 7817 /* FALLTHROUGH */ 7818 case PF_DEFER: 7819 #if NPFSYNC > 0 7820 /* 7821 * We no longer hold PF_LOCK() here, so we can dispatch 7822 * deferral if we are asked to do so. 7823 */ 7824 if (deferral != NULL) 7825 pfsync_undefer(deferral, 0); 7826 #endif /* NPFSYNC > 0 */ 7827 pd.m = NULL; 7828 action = PF_PASS; 7829 break; 7830 case PF_DIVERT: 7831 switch (pd.af) { 7832 case AF_INET: 7833 divert_packet(pd.m, pd.dir, r->divert.port); 7834 pd.m = NULL; 7835 break; 7836 #ifdef INET6 7837 case AF_INET6: 7838 divert6_packet(pd.m, pd.dir, r->divert.port); 7839 pd.m = NULL; 7840 break; 7841 #endif /* INET6 */ 7842 } 7843 action = PF_PASS; 7844 break; 7845 #ifdef INET6 7846 case PF_AFRT: 7847 if (pf_translate_af(&pd)) { 7848 action = PF_DROP; 7849 break; 7850 } 7851 pd.m->m_pkthdr.pf.flags |= PF_TAG_GENERATED; 7852 switch (pd.naf) { 7853 case AF_INET: 7854 if (pd.dir == PF_IN) { 7855 if (ipforwarding == 0) { 7856 ipstat_inc(ips_cantforward); 7857 action = PF_DROP; 7858 break; 7859 } 7860 ip_forward(pd.m, ifp, NULL, 1); 7861 } else 7862 ip_output(pd.m, NULL, NULL, 0, NULL, NULL, 0); 7863 break; 7864 case AF_INET6: 7865 if (pd.dir == PF_IN) { 7866 if (ip6_forwarding == 0) { 7867 ip6stat_inc(ip6s_cantforward); 7868 action = PF_DROP; 7869 break; 7870 } 7871 ip6_forward(pd.m, NULL, 1); 7872 } else 7873 ip6_output(pd.m, NULL, NULL, 0, NULL, NULL); 7874 break; 7875 } 7876 if (action != PF_DROP) { 7877 pd.m = NULL; 7878 action = PF_PASS; 7879 } 7880 break; 7881 #endif /* INET6 */ 7882 case PF_DROP: 7883 m_freem(pd.m); 7884 pd.m = NULL; 7885 break; 7886 default: 7887 if (st && st->rt) { 7888 switch (pd.af) { 7889 case AF_INET: 7890 pf_route(&pd, st); 7891 break; 7892 #ifdef INET6 7893 case AF_INET6: 7894 pf_route6(&pd, st); 7895 break; 7896 #endif /* INET6 */ 7897 } 7898 } 7899 break; 7900 } 7901 7902 #ifdef INET6 7903 /* if reassembled packet passed, create new fragments */ 7904 if (pf_status.reass && action == PF_PASS && pd.m && fwdir == PF_FWD && 7905 pd.af == AF_INET6) { 7906 struct m_tag *mtag; 7907 7908 if ((mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL))) 7909 action = pf_refragment6(&pd.m, mtag, NULL, NULL, NULL); 7910 } 7911 #endif /* INET6 */ 7912 if (st && action != PF_DROP) { 7913 if (!st->if_index_in && dir == PF_IN) 7914 st->if_index_in = ifp->if_index; 7915 else if (!st->if_index_out && dir == PF_OUT) 7916 st->if_index_out = ifp->if_index; 7917 } 7918 7919 *m0 = pd.m; 7920 7921 pf_state_unref(st); 7922 7923 return (action); 7924 } 7925 7926 int 7927 pf_ouraddr(struct mbuf *m) 7928 { 7929 struct pf_state_key *sk; 7930 7931 if (m->m_pkthdr.pf.flags & PF_TAG_DIVERTED) 7932 return (1); 7933 7934 sk = m->m_pkthdr.pf.statekey; 7935 if (sk != NULL) { 7936 if (sk->sk_inp != NULL) 7937 return (1); 7938 } 7939 7940 return (-1); 7941 } 7942 7943 /* 7944 * must be called whenever any addressing information such as 7945 * address, port, protocol has changed 7946 */ 7947 void 7948 pf_pkt_addr_changed(struct mbuf *m) 7949 { 7950 pf_mbuf_unlink_state_key(m); 7951 pf_mbuf_unlink_inpcb(m); 7952 } 7953 7954 struct inpcb * 7955 pf_inp_lookup(struct mbuf *m) 7956 { 7957 struct inpcb *inp = NULL; 7958 struct pf_state_key *sk = m->m_pkthdr.pf.statekey; 7959 7960 if (!pf_state_key_isvalid(sk)) 7961 pf_mbuf_unlink_state_key(m); 7962 else 7963 inp = m->m_pkthdr.pf.statekey->sk_inp; 7964 7965 if (inp && inp->inp_pf_sk) 7966 KASSERT(m->m_pkthdr.pf.statekey == inp->inp_pf_sk); 7967 7968 in_pcbref(inp); 7969 return (inp); 7970 } 7971 7972 void 7973 pf_inp_link(struct mbuf *m, struct inpcb *inp) 7974 { 7975 struct pf_state_key *sk = m->m_pkthdr.pf.statekey; 7976 7977 if (!pf_state_key_isvalid(sk)) { 7978 pf_mbuf_unlink_state_key(m); 7979 return; 7980 } 7981 7982 /* 7983 * we don't need to grab PF-lock here. At worst case we link inp to 7984 * state, which might be just being marked as deleted by another 7985 * thread. 7986 */ 7987 if (inp && !sk->sk_inp && !inp->inp_pf_sk) 7988 pf_state_key_link_inpcb(sk, inp); 7989 7990 /* The statekey has finished finding the inp, it is no longer needed. */ 7991 pf_mbuf_unlink_state_key(m); 7992 } 7993 7994 void 7995 pf_inp_unlink(struct inpcb *inp) 7996 { 7997 pf_inpcb_unlink_state_key(inp); 7998 } 7999 8000 void 8001 pf_state_key_link_reverse(struct pf_state_key *sk, struct pf_state_key *skrev) 8002 { 8003 struct pf_state_key *old_reverse; 8004 8005 old_reverse = atomic_cas_ptr(&sk->sk_reverse, NULL, skrev); 8006 if (old_reverse != NULL) 8007 KASSERT(old_reverse == skrev); 8008 else { 8009 pf_state_key_ref(skrev); 8010 8011 /* 8012 * NOTE: if sk == skrev, then KASSERT() below holds true, we 8013 * still want to grab a reference in such case, because 8014 * pf_state_key_unlink_reverse() does not check whether keys 8015 * are identical or not. 8016 */ 8017 old_reverse = atomic_cas_ptr(&skrev->sk_reverse, NULL, sk); 8018 if (old_reverse != NULL) 8019 KASSERT(old_reverse == sk); 8020 8021 pf_state_key_ref(sk); 8022 } 8023 } 8024 8025 #if NPFLOG > 0 8026 void 8027 pf_log_matches(struct pf_pdesc *pd, struct pf_rule *rm, struct pf_rule *am, 8028 struct pf_ruleset *ruleset, struct pf_rule_slist *matchrules) 8029 { 8030 struct pf_rule_item *ri; 8031 8032 /* if this is the log(matches) rule, packet has been logged already */ 8033 if (rm->log & PF_LOG_MATCHES) 8034 return; 8035 8036 SLIST_FOREACH(ri, matchrules, entry) 8037 if (ri->r->log & PF_LOG_MATCHES) 8038 pflog_packet(pd, PFRES_MATCH, rm, am, ruleset, ri->r); 8039 } 8040 #endif /* NPFLOG > 0 */ 8041 8042 struct pf_state_key * 8043 pf_state_key_ref(struct pf_state_key *sk) 8044 { 8045 if (sk != NULL) 8046 PF_REF_TAKE(sk->sk_refcnt); 8047 8048 return (sk); 8049 } 8050 8051 void 8052 pf_state_key_unref(struct pf_state_key *sk) 8053 { 8054 if (PF_REF_RELE(sk->sk_refcnt)) { 8055 /* state key must be removed from tree */ 8056 KASSERT(!pf_state_key_isvalid(sk)); 8057 /* state key must be unlinked from reverse key */ 8058 KASSERT(sk->sk_reverse == NULL); 8059 /* state key must be unlinked from socket */ 8060 KASSERT(sk->sk_inp == NULL); 8061 pool_put(&pf_state_key_pl, sk); 8062 } 8063 } 8064 8065 int 8066 pf_state_key_isvalid(struct pf_state_key *sk) 8067 { 8068 return ((sk != NULL) && (sk->sk_removed == 0)); 8069 } 8070 8071 void 8072 pf_mbuf_link_state_key(struct mbuf *m, struct pf_state_key *sk) 8073 { 8074 KASSERT(m->m_pkthdr.pf.statekey == NULL); 8075 m->m_pkthdr.pf.statekey = pf_state_key_ref(sk); 8076 } 8077 8078 void 8079 pf_mbuf_unlink_state_key(struct mbuf *m) 8080 { 8081 struct pf_state_key *sk = m->m_pkthdr.pf.statekey; 8082 8083 if (sk != NULL) { 8084 m->m_pkthdr.pf.statekey = NULL; 8085 pf_state_key_unref(sk); 8086 } 8087 } 8088 8089 void 8090 pf_mbuf_link_inpcb(struct mbuf *m, struct inpcb *inp) 8091 { 8092 KASSERT(m->m_pkthdr.pf.inp == NULL); 8093 m->m_pkthdr.pf.inp = in_pcbref(inp); 8094 } 8095 8096 void 8097 pf_mbuf_unlink_inpcb(struct mbuf *m) 8098 { 8099 struct inpcb *inp = m->m_pkthdr.pf.inp; 8100 8101 if (inp != NULL) { 8102 m->m_pkthdr.pf.inp = NULL; 8103 in_pcbunref(inp); 8104 } 8105 } 8106 8107 void 8108 pf_state_key_link_inpcb(struct pf_state_key *sk, struct inpcb *inp) 8109 { 8110 KASSERT(sk->sk_inp == NULL); 8111 sk->sk_inp = in_pcbref(inp); 8112 KASSERT(inp->inp_pf_sk == NULL); 8113 inp->inp_pf_sk = pf_state_key_ref(sk); 8114 } 8115 8116 void 8117 pf_inpcb_unlink_state_key(struct inpcb *inp) 8118 { 8119 struct pf_state_key *sk = inp->inp_pf_sk; 8120 8121 if (sk != NULL) { 8122 KASSERT(sk->sk_inp == inp); 8123 sk->sk_inp = NULL; 8124 inp->inp_pf_sk = NULL; 8125 pf_state_key_unref(sk); 8126 in_pcbunref(inp); 8127 } 8128 } 8129 8130 void 8131 pf_state_key_unlink_inpcb(struct pf_state_key *sk) 8132 { 8133 struct inpcb *inp = sk->sk_inp; 8134 8135 if (inp != NULL) { 8136 KASSERT(inp->inp_pf_sk == sk); 8137 sk->sk_inp = NULL; 8138 inp->inp_pf_sk = NULL; 8139 pf_state_key_unref(sk); 8140 in_pcbunref(inp); 8141 } 8142 } 8143 8144 void 8145 pf_state_key_unlink_reverse(struct pf_state_key *sk) 8146 { 8147 struct pf_state_key *skrev = sk->sk_reverse; 8148 8149 /* Note that sk and skrev may be equal, then we unref twice. */ 8150 if (skrev != NULL) { 8151 KASSERT(skrev->sk_reverse == sk); 8152 sk->sk_reverse = NULL; 8153 skrev->sk_reverse = NULL; 8154 pf_state_key_unref(skrev); 8155 pf_state_key_unref(sk); 8156 } 8157 } 8158 8159 struct pf_state * 8160 pf_state_ref(struct pf_state *st) 8161 { 8162 if (st != NULL) 8163 PF_REF_TAKE(st->refcnt); 8164 return (st); 8165 } 8166 8167 void 8168 pf_state_unref(struct pf_state *st) 8169 { 8170 if ((st != NULL) && PF_REF_RELE(st->refcnt)) { 8171 /* never inserted or removed */ 8172 #if NPFSYNC > 0 8173 KASSERT((TAILQ_NEXT(st, sync_list) == NULL) || 8174 ((TAILQ_NEXT(st, sync_list) == _Q_INVALID) && 8175 (st->sync_state == PFSYNC_S_NONE))); 8176 #endif /* NPFSYNC */ 8177 KASSERT((TAILQ_NEXT(st, entry_list) == NULL) || 8178 (TAILQ_NEXT(st, entry_list) == _Q_INVALID)); 8179 8180 pf_state_key_unref(st->key[PF_SK_WIRE]); 8181 pf_state_key_unref(st->key[PF_SK_STACK]); 8182 8183 pool_put(&pf_state_pl, st); 8184 } 8185 } 8186 8187 int 8188 pf_delay_pkt(struct mbuf *m, u_int ifidx) 8189 { 8190 struct pf_pktdelay *pdy; 8191 8192 if ((pdy = pool_get(&pf_pktdelay_pl, PR_NOWAIT)) == NULL) { 8193 m_freem(m); 8194 return (ENOBUFS); 8195 } 8196 pdy->ifidx = ifidx; 8197 pdy->m = m; 8198 timeout_set(&pdy->to, pf_pktenqueue_delayed, pdy); 8199 timeout_add_msec(&pdy->to, m->m_pkthdr.pf.delay); 8200 m->m_pkthdr.pf.delay = 0; 8201 return (0); 8202 } 8203 8204 void 8205 pf_pktenqueue_delayed(void *arg) 8206 { 8207 struct pf_pktdelay *pdy = arg; 8208 struct ifnet *ifp; 8209 8210 ifp = if_get(pdy->ifidx); 8211 if (ifp != NULL) { 8212 if_enqueue(ifp, pdy->m); 8213 if_put(ifp); 8214 } else 8215 m_freem(pdy->m); 8216 8217 pool_put(&pf_pktdelay_pl, pdy); 8218 } 8219