1 /* 2 * Copyright (c) 2004 The DragonFly Project. All rights reserved. 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 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 "opt_inet.h" 39 #include "opt_inet6.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/malloc.h> 44 #include <sys/mbuf.h> 45 #include <sys/filio.h> 46 #include <sys/socket.h> 47 #include <sys/socketvar.h> 48 #include <sys/kernel.h> 49 #include <sys/time.h> 50 #include <sys/sysctl.h> 51 #include <sys/endian.h> 52 #include <sys/proc.h> 53 #include <sys/kthread.h> 54 #include <sys/spinlock.h> 55 56 #include <machine/inttypes.h> 57 58 #include <sys/md5.h> 59 60 #include <net/if.h> 61 #include <net/if_types.h> 62 #include <net/bpf.h> 63 #include <net/netisr2.h> 64 #include <net/route.h> 65 66 #include <netinet/in.h> 67 #include <netinet/in_var.h> 68 #include <netinet/in_systm.h> 69 #include <netinet/ip.h> 70 #include <netinet/ip_var.h> 71 #include <netinet/tcp.h> 72 #include <netinet/tcp_seq.h> 73 #include <netinet/udp.h> 74 #include <netinet/ip_icmp.h> 75 #include <netinet/in_pcb.h> 76 #include <netinet/tcp_timer.h> 77 #include <netinet/tcp_var.h> 78 #include <netinet/udp_var.h> 79 #include <netinet/icmp_var.h> 80 #include <netinet/if_ether.h> 81 82 #include <net/pf/pfvar.h> 83 #include <net/pf/if_pflog.h> 84 85 #include <net/pf/if_pfsync.h> 86 87 #ifdef INET6 88 #include <netinet/ip6.h> 89 #include <netinet/icmp6.h> 90 #include <netinet6/nd6.h> 91 #include <netinet6/ip6_var.h> 92 #include <netinet6/in6_pcb.h> 93 #endif /* INET6 */ 94 95 #include <sys/in_cksum.h> 96 #include <sys/ucred.h> 97 #include <machine/limits.h> 98 #include <sys/msgport2.h> 99 #include <sys/spinlock2.h> 100 #include <net/netmsg2.h> 101 #include <net/toeplitz2.h> 102 103 extern int ip_optcopy(struct ip *, struct ip *); 104 extern int debug_pfugidhack; 105 106 /* 107 * pf_token - shared lock for cpu-localized operations, 108 * exclusive lock otherwise. 109 * 110 * pf_gtoken- exclusive lock used for initialization. 111 * 112 * pf_spin - only used to atomically fetch and increment stateid 113 * on 32-bit systems. 114 */ 115 struct lwkt_token pf_token = LWKT_TOKEN_INITIALIZER(pf_token); 116 struct lwkt_token pf_gtoken = LWKT_TOKEN_INITIALIZER(pf_gtoken); 117 #if __SIZEOF_LONG__ != 8 118 struct spinlock pf_spin = SPINLOCK_INITIALIZER(pf_spin, "pf_spin"); 119 #endif 120 121 #define DPFPRINTF(n, x) if (pf_status.debug >= (n)) kprintf x 122 123 #define FAIL(code) { error = (code); goto done; } 124 125 /* 126 * Global variables 127 */ 128 129 /* mask radix tree */ 130 struct radix_node_head *pf_maskhead; 131 132 /* state tables */ 133 struct pf_state_tree *pf_statetbl; /* incls one global table */ 134 struct pf_state **purge_cur; 135 struct pf_altqqueue pf_altqs[2]; 136 struct pf_palist pf_pabuf; 137 struct pf_altqqueue *pf_altqs_active; 138 struct pf_altqqueue *pf_altqs_inactive; 139 struct pf_status pf_status; 140 141 u_int32_t ticket_altqs_active; 142 u_int32_t ticket_altqs_inactive; 143 int altqs_inactive_open; 144 u_int32_t ticket_pabuf; 145 146 MD5_CTX pf_tcp_secret_ctx; 147 u_char pf_tcp_secret[16]; 148 int pf_tcp_secret_init; 149 int pf_tcp_iss_off; 150 151 struct pf_anchor_stackframe { 152 struct pf_ruleset *rs; 153 struct pf_rule *r; 154 struct pf_anchor_node *parent; 155 struct pf_anchor *child; 156 } pf_anchor_stack[64]; 157 158 struct malloc_type *pf_src_tree_pl, *pf_rule_pl, *pf_pooladdr_pl; 159 struct malloc_type *pf_state_pl, *pf_state_key_pl, *pf_state_item_pl; 160 struct malloc_type *pf_altq_pl; 161 162 void pf_print_host(struct pf_addr *, u_int16_t, u_int8_t); 163 164 void pf_init_threshold(struct pf_threshold *, u_int32_t, 165 u_int32_t); 166 void pf_add_threshold(struct pf_threshold *); 167 int pf_check_threshold(struct pf_threshold *); 168 169 void pf_change_ap(struct pf_addr *, u_int16_t *, 170 u_int16_t *, u_int16_t *, struct pf_addr *, 171 u_int16_t, u_int8_t, sa_family_t); 172 int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 173 struct tcphdr *, struct pf_state_peer *); 174 #ifdef INET6 175 void pf_change_a6(struct pf_addr *, u_int16_t *, 176 struct pf_addr *, u_int8_t); 177 #endif /* INET6 */ 178 void pf_change_icmp(struct pf_addr *, u_int16_t *, 179 struct pf_addr *, struct pf_addr *, u_int16_t, 180 u_int16_t *, u_int16_t *, u_int16_t *, 181 u_int16_t *, u_int8_t, sa_family_t); 182 void pf_send_tcp(const struct pf_rule *, sa_family_t, 183 const struct pf_addr *, const struct pf_addr *, 184 u_int16_t, u_int16_t, u_int32_t, u_int32_t, 185 u_int8_t, u_int16_t, u_int16_t, u_int8_t, int, 186 u_int16_t, struct ether_header *, struct ifnet *); 187 void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 188 sa_family_t, struct pf_rule *); 189 struct pf_rule *pf_match_translation(struct pf_pdesc *, struct mbuf *, 190 int, int, struct pfi_kif *, 191 struct pf_addr *, u_int16_t, struct pf_addr *, 192 u_int16_t, int); 193 struct pf_rule *pf_get_translation(struct pf_pdesc *, struct mbuf *, 194 int, int, struct pfi_kif *, struct pf_src_node **, 195 struct pf_state_key **, struct pf_state_key **, 196 struct pf_state_key **, struct pf_state_key **, 197 struct pf_addr *, struct pf_addr *, 198 u_int16_t, u_int16_t); 199 void pf_detach_state(struct pf_state *); 200 int pf_state_key_setup(struct pf_pdesc *, struct pf_rule *, 201 struct pf_state_key **, struct pf_state_key **, 202 struct pf_state_key **, struct pf_state_key **, 203 struct pf_addr *, struct pf_addr *, 204 u_int16_t, u_int16_t); 205 void pf_state_key_detach(struct pf_state *, int); 206 u_int32_t pf_tcp_iss(struct pf_pdesc *); 207 int pf_test_rule(struct pf_rule **, struct pf_state **, 208 int, struct pfi_kif *, struct mbuf *, int, 209 void *, struct pf_pdesc *, struct pf_rule **, 210 struct pf_ruleset **, struct ifqueue *, struct inpcb *); 211 static __inline int pf_create_state(struct pf_rule *, struct pf_rule *, 212 struct pf_rule *, struct pf_pdesc *, 213 struct pf_src_node *, struct pf_state_key *, 214 struct pf_state_key *, struct pf_state_key *, 215 struct pf_state_key *, struct mbuf *, int, 216 u_int16_t, u_int16_t, int *, struct pfi_kif *, 217 struct pf_state **, int, u_int16_t, u_int16_t, 218 int); 219 int pf_test_fragment(struct pf_rule **, int, 220 struct pfi_kif *, struct mbuf *, void *, 221 struct pf_pdesc *, struct pf_rule **, 222 struct pf_ruleset **); 223 int pf_tcp_track_full(struct pf_state_peer *, 224 struct pf_state_peer *, struct pf_state **, 225 struct pfi_kif *, struct mbuf *, int, 226 struct pf_pdesc *, u_short *, int *); 227 int pf_tcp_track_sloppy(struct pf_state_peer *, 228 struct pf_state_peer *, struct pf_state **, 229 struct pf_pdesc *, u_short *); 230 int pf_test_state_tcp(struct pf_state **, int, 231 struct pfi_kif *, struct mbuf *, int, 232 void *, struct pf_pdesc *, u_short *); 233 int pf_test_state_udp(struct pf_state **, int, 234 struct pfi_kif *, struct mbuf *, int, 235 void *, struct pf_pdesc *); 236 int pf_test_state_icmp(struct pf_state **, int, 237 struct pfi_kif *, struct mbuf *, int, 238 void *, struct pf_pdesc *, u_short *); 239 int pf_test_state_other(struct pf_state **, int, 240 struct pfi_kif *, struct mbuf *, struct pf_pdesc *); 241 void pf_step_into_anchor(int *, struct pf_ruleset **, int, 242 struct pf_rule **, struct pf_rule **, int *); 243 int pf_step_out_of_anchor(int *, struct pf_ruleset **, 244 int, struct pf_rule **, struct pf_rule **, 245 int *); 246 void pf_hash(struct pf_addr *, struct pf_addr *, 247 struct pf_poolhashkey *, sa_family_t); 248 int pf_map_addr(u_int8_t, struct pf_rule *, 249 struct pf_addr *, struct pf_addr *, 250 struct pf_addr *, struct pf_src_node **); 251 int pf_get_sport(struct pf_pdesc *, 252 sa_family_t, u_int8_t, struct pf_rule *, 253 struct pf_addr *, struct pf_addr *, 254 u_int16_t, u_int16_t, 255 struct pf_addr *, u_int16_t *, 256 u_int16_t, u_int16_t, 257 struct pf_src_node **); 258 void pf_route(struct mbuf **, struct pf_rule *, int, 259 struct ifnet *, struct pf_state *, 260 struct pf_pdesc *); 261 void pf_route6(struct mbuf **, struct pf_rule *, int, 262 struct ifnet *, struct pf_state *, 263 struct pf_pdesc *); 264 u_int8_t pf_get_wscale(struct mbuf *, int, u_int16_t, 265 sa_family_t); 266 u_int16_t pf_get_mss(struct mbuf *, int, u_int16_t, 267 sa_family_t); 268 u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 269 u_int16_t); 270 void pf_set_rt_ifp(struct pf_state *, 271 struct pf_addr *); 272 int pf_check_proto_cksum(struct mbuf *, int, int, 273 u_int8_t, sa_family_t); 274 struct pf_divert *pf_get_divert(struct mbuf *); 275 void pf_print_state_parts(struct pf_state *, 276 struct pf_state_key *, struct pf_state_key *); 277 int pf_addr_wrap_neq(struct pf_addr_wrap *, 278 struct pf_addr_wrap *); 279 struct pf_state *pf_find_state(struct pfi_kif *, 280 struct pf_state_key_cmp *, u_int, struct mbuf *); 281 int pf_src_connlimit(struct pf_state *); 282 int pf_check_congestion(struct ifqueue *); 283 284 extern int pf_end_threads; 285 286 struct pf_pool_limit pf_pool_limits[PF_LIMIT_MAX] = { 287 { &pf_state_pl, PFSTATE_HIWAT }, 288 { &pf_src_tree_pl, PFSNODE_HIWAT }, 289 { &pf_frent_pl, PFFRAG_FRENT_HIWAT }, 290 { &pfr_ktable_pl, PFR_KTABLE_HIWAT }, 291 { &pfr_kentry_pl, PFR_KENTRY_HIWAT } 292 }; 293 294 /* 295 * If route-to and direction is out we match with no further processing 296 * (rt_kif must be assigned and not equal to the out interface) 297 * If reply-to and direction is in we match with no further processing 298 * (rt_kif must be assigned and not equal to the in interface) 299 */ 300 #define STATE_LOOKUP(i, k, d, s, m) \ 301 do { \ 302 s = pf_find_state(i, k, d, m); \ 303 if (s == NULL || (s)->timeout == PFTM_PURGE) \ 304 return (PF_DROP); \ 305 if (d == PF_OUT && \ 306 (((s)->rule.ptr->rt == PF_ROUTETO && \ 307 (s)->rule.ptr->direction == PF_OUT) || \ 308 ((s)->rule.ptr->rt == PF_REPLYTO && \ 309 (s)->rule.ptr->direction == PF_IN)) && \ 310 (s)->rt_kif != NULL && \ 311 (s)->rt_kif != i) \ 312 return (PF_PASS); \ 313 } while (0) 314 315 #define BOUND_IFACE(r, k) \ 316 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : pfi_all 317 318 #define STATE_INC_COUNTERS(s) \ 319 do { \ 320 atomic_add_int(&s->rule.ptr->states_cur, 1); \ 321 s->rule.ptr->states_tot++; \ 322 if (s->anchor.ptr != NULL) { \ 323 atomic_add_int(&s->anchor.ptr->states_cur, 1); \ 324 s->anchor.ptr->states_tot++; \ 325 } \ 326 if (s->nat_rule.ptr != NULL) { \ 327 atomic_add_int(&s->nat_rule.ptr->states_cur, 1); \ 328 s->nat_rule.ptr->states_tot++; \ 329 } \ 330 } while (0) 331 332 #define STATE_DEC_COUNTERS(s) \ 333 do { \ 334 if (s->nat_rule.ptr != NULL) \ 335 atomic_add_int(&s->nat_rule.ptr->states_cur, -1); \ 336 if (s->anchor.ptr != NULL) \ 337 atomic_add_int(&s->anchor.ptr->states_cur, -1); \ 338 atomic_add_int(&s->rule.ptr->states_cur, -1); \ 339 } while (0) 340 341 static MALLOC_DEFINE(M_PFSTATEPL, "pfstatepl", "pf state pool list"); 342 static MALLOC_DEFINE(M_PFSRCTREEPL, "pfsrctpl", "pf source tree pool list"); 343 static MALLOC_DEFINE(M_PFSTATEKEYPL, "pfstatekeypl", "pf state key pool list"); 344 static MALLOC_DEFINE(M_PFSTATEITEMPL, "pfstateitempl", "pf state item pool list"); 345 346 static __inline int pf_src_compare(struct pf_src_node *, struct pf_src_node *); 347 static __inline int pf_state_compare_key(struct pf_state_key *, 348 struct pf_state_key *); 349 static __inline int pf_state_compare_rkey(struct pf_state_key *, 350 struct pf_state_key *); 351 static __inline int pf_state_compare_id(struct pf_state *, 352 struct pf_state *); 353 354 struct pf_src_tree *tree_src_tracking; 355 struct pf_state_tree_id *tree_id; 356 struct pf_state_queue *state_list; 357 struct pf_counters *pf_counters; 358 359 RB_GENERATE(pf_src_tree, pf_src_node, entry, pf_src_compare); 360 RB_GENERATE(pf_state_tree, pf_state_key, entry, pf_state_compare_key); 361 RB_GENERATE(pf_state_rtree, pf_state_key, entry, pf_state_compare_rkey); 362 RB_GENERATE(pf_state_tree_id, pf_state, entry_id, pf_state_compare_id); 363 364 static __inline int 365 pf_src_compare(struct pf_src_node *a, struct pf_src_node *b) 366 { 367 int diff; 368 369 if (a->rule.ptr > b->rule.ptr) 370 return (1); 371 if (a->rule.ptr < b->rule.ptr) 372 return (-1); 373 if ((diff = a->af - b->af) != 0) 374 return (diff); 375 switch (a->af) { 376 #ifdef INET 377 case AF_INET: 378 if (a->addr.addr32[0] > b->addr.addr32[0]) 379 return (1); 380 if (a->addr.addr32[0] < b->addr.addr32[0]) 381 return (-1); 382 break; 383 #endif /* INET */ 384 #ifdef INET6 385 case AF_INET6: 386 if (a->addr.addr32[3] > b->addr.addr32[3]) 387 return (1); 388 if (a->addr.addr32[3] < b->addr.addr32[3]) 389 return (-1); 390 if (a->addr.addr32[2] > b->addr.addr32[2]) 391 return (1); 392 if (a->addr.addr32[2] < b->addr.addr32[2]) 393 return (-1); 394 if (a->addr.addr32[1] > b->addr.addr32[1]) 395 return (1); 396 if (a->addr.addr32[1] < b->addr.addr32[1]) 397 return (-1); 398 if (a->addr.addr32[0] > b->addr.addr32[0]) 399 return (1); 400 if (a->addr.addr32[0] < b->addr.addr32[0]) 401 return (-1); 402 break; 403 #endif /* INET6 */ 404 } 405 return (0); 406 } 407 408 u_int32_t 409 pf_state_hash(struct pf_state_key *sk) 410 { 411 u_int32_t hv = (u_int32_t)(((intptr_t)sk >> 6) ^ ((intptr_t)sk >> 15)); 412 if (hv == 0) /* disallow 0 */ 413 hv = 1; 414 return(hv); 415 } 416 417 #ifdef INET6 418 void 419 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 420 { 421 switch (af) { 422 #ifdef INET 423 case AF_INET: 424 dst->addr32[0] = src->addr32[0]; 425 break; 426 #endif /* INET */ 427 case AF_INET6: 428 dst->addr32[0] = src->addr32[0]; 429 dst->addr32[1] = src->addr32[1]; 430 dst->addr32[2] = src->addr32[2]; 431 dst->addr32[3] = src->addr32[3]; 432 break; 433 } 434 } 435 #endif /* INET6 */ 436 437 void 438 pf_init_threshold(struct pf_threshold *threshold, 439 u_int32_t limit, u_int32_t seconds) 440 { 441 threshold->limit = limit * PF_THRESHOLD_MULT; 442 threshold->seconds = seconds; 443 threshold->count = 0; 444 threshold->last = time_second; 445 } 446 447 void 448 pf_add_threshold(struct pf_threshold *threshold) 449 { 450 u_int32_t t = time_second, diff = t - threshold->last; 451 452 if (diff >= threshold->seconds) 453 threshold->count = 0; 454 else 455 threshold->count -= threshold->count * diff / 456 threshold->seconds; 457 threshold->count += PF_THRESHOLD_MULT; 458 threshold->last = t; 459 } 460 461 int 462 pf_check_threshold(struct pf_threshold *threshold) 463 { 464 return (threshold->count > threshold->limit); 465 } 466 467 int 468 pf_src_connlimit(struct pf_state *state) 469 { 470 int bad = 0; 471 int cpu = mycpu->gd_cpuid; 472 473 atomic_add_int(&state->src_node->conn, 1); 474 state->src.tcp_est = 1; 475 pf_add_threshold(&state->src_node->conn_rate); 476 477 if (state->rule.ptr->max_src_conn && 478 state->rule.ptr->max_src_conn < 479 state->src_node->conn) { 480 PF_INC_LCOUNTER(LCNT_SRCCONN); 481 bad++; 482 } 483 484 if (state->rule.ptr->max_src_conn_rate.limit && 485 pf_check_threshold(&state->src_node->conn_rate)) { 486 PF_INC_LCOUNTER(LCNT_SRCCONNRATE); 487 bad++; 488 } 489 490 if (!bad) 491 return 0; 492 493 if (state->rule.ptr->overload_tbl) { 494 struct pfr_addr p; 495 u_int32_t killed = 0; 496 497 PF_INC_LCOUNTER(LCNT_OVERLOAD_TABLE); 498 if (pf_status.debug >= PF_DEBUG_MISC) { 499 kprintf("pf_src_connlimit: blocking address "); 500 pf_print_host(&state->src_node->addr, 0, 501 state->key[PF_SK_WIRE]->af); 502 } 503 504 bzero(&p, sizeof(p)); 505 p.pfra_af = state->key[PF_SK_WIRE]->af; 506 switch (state->key[PF_SK_WIRE]->af) { 507 #ifdef INET 508 case AF_INET: 509 p.pfra_net = 32; 510 p.pfra_ip4addr = state->src_node->addr.v4; 511 break; 512 #endif /* INET */ 513 #ifdef INET6 514 case AF_INET6: 515 p.pfra_net = 128; 516 p.pfra_ip6addr = state->src_node->addr.v6; 517 break; 518 #endif /* INET6 */ 519 } 520 521 pfr_insert_kentry(state->rule.ptr->overload_tbl, 522 &p, time_second); 523 524 /* kill existing states if that's required. */ 525 if (state->rule.ptr->flush) { 526 struct pf_state_key *sk; 527 struct pf_state *st; 528 529 PF_INC_LCOUNTER(LCNT_OVERLOAD_FLUSH); 530 RB_FOREACH(st, pf_state_tree_id, &tree_id[cpu]) { 531 sk = st->key[PF_SK_WIRE]; 532 /* 533 * Kill states from this source. (Only those 534 * from the same rule if PF_FLUSH_GLOBAL is not 535 * set). (Only on current cpu). 536 */ 537 if (sk->af == 538 state->key[PF_SK_WIRE]->af && 539 ((state->direction == PF_OUT && 540 PF_AEQ(&state->src_node->addr, 541 &sk->addr[0], sk->af)) || 542 (state->direction == PF_IN && 543 PF_AEQ(&state->src_node->addr, 544 &sk->addr[1], sk->af))) && 545 (state->rule.ptr->flush & 546 PF_FLUSH_GLOBAL || 547 state->rule.ptr == st->rule.ptr)) { 548 st->timeout = PFTM_PURGE; 549 st->src.state = st->dst.state = 550 TCPS_CLOSED; 551 killed++; 552 } 553 } 554 if (pf_status.debug >= PF_DEBUG_MISC) 555 kprintf(", %u states killed", killed); 556 } 557 if (pf_status.debug >= PF_DEBUG_MISC) 558 kprintf("\n"); 559 } 560 561 /* kill this state */ 562 state->timeout = PFTM_PURGE; 563 state->src.state = state->dst.state = TCPS_CLOSED; 564 565 return 1; 566 } 567 568 int 569 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule, 570 struct pf_addr *src, sa_family_t af) 571 { 572 struct pf_src_node k; 573 int cpu = mycpu->gd_cpuid; 574 575 bzero(&k, sizeof(k)); /* avoid gcc warnings */ 576 if (*sn == NULL) { 577 k.af = af; 578 PF_ACPY(&k.addr, src, af); 579 if (rule->rule_flag & PFRULE_RULESRCTRACK || 580 rule->rpool.opts & PF_POOL_STICKYADDR) 581 k.rule.ptr = rule; 582 else 583 k.rule.ptr = NULL; 584 PF_INC_SCOUNTER(SCNT_SRC_NODE_SEARCH); 585 *sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k); 586 } 587 if (*sn == NULL) { 588 if (!rule->max_src_nodes || 589 rule->src_nodes < rule->max_src_nodes) 590 (*sn) = kmalloc(sizeof(struct pf_src_node), 591 M_PFSRCTREEPL, M_NOWAIT|M_ZERO); 592 else 593 PF_INC_LCOUNTER(LCNT_SRCNODES); 594 if ((*sn) == NULL) 595 return (-1); 596 597 pf_init_threshold(&(*sn)->conn_rate, 598 rule->max_src_conn_rate.limit, 599 rule->max_src_conn_rate.seconds); 600 601 (*sn)->af = af; 602 if (rule->rule_flag & PFRULE_RULESRCTRACK || 603 rule->rpool.opts & PF_POOL_STICKYADDR) 604 (*sn)->rule.ptr = rule; 605 else 606 (*sn)->rule.ptr = NULL; 607 PF_ACPY(&(*sn)->addr, src, af); 608 if (RB_INSERT(pf_src_tree, 609 &tree_src_tracking[cpu], *sn) != NULL) { 610 if (pf_status.debug >= PF_DEBUG_MISC) { 611 kprintf("pf: src_tree insert failed: "); 612 pf_print_host(&(*sn)->addr, 0, af); 613 kprintf("\n"); 614 } 615 kfree(*sn, M_PFSRCTREEPL); 616 return (-1); 617 } 618 619 /* 620 * Atomic op required to increment src_nodes in the rule 621 * because we hold a shared token here (decrements will use 622 * an exclusive token). 623 */ 624 (*sn)->creation = time_second; 625 (*sn)->ruletype = rule->action; 626 if ((*sn)->rule.ptr != NULL) 627 atomic_add_int(&(*sn)->rule.ptr->src_nodes, 1); 628 PF_INC_SCOUNTER(SCNT_SRC_NODE_INSERT); 629 atomic_add_int(&pf_status.src_nodes, 1); 630 } else { 631 if (rule->max_src_states && 632 (*sn)->states >= rule->max_src_states) { 633 PF_INC_LCOUNTER(LCNT_SRCSTATES); 634 return (-1); 635 } 636 } 637 return (0); 638 } 639 640 /* 641 * state table (indexed by the pf_state_key structure), normal RBTREE 642 * comparison. 643 */ 644 static __inline int 645 pf_state_compare_key(struct pf_state_key *a, struct pf_state_key *b) 646 { 647 int diff; 648 649 if ((diff = a->proto - b->proto) != 0) 650 return (diff); 651 if ((diff = a->af - b->af) != 0) 652 return (diff); 653 switch (a->af) { 654 #ifdef INET 655 case AF_INET: 656 if (a->addr[0].addr32[0] > b->addr[0].addr32[0]) 657 return (1); 658 if (a->addr[0].addr32[0] < b->addr[0].addr32[0]) 659 return (-1); 660 if (a->addr[1].addr32[0] > b->addr[1].addr32[0]) 661 return (1); 662 if (a->addr[1].addr32[0] < b->addr[1].addr32[0]) 663 return (-1); 664 break; 665 #endif /* INET */ 666 #ifdef INET6 667 case AF_INET6: 668 if (a->addr[0].addr32[3] > b->addr[0].addr32[3]) 669 return (1); 670 if (a->addr[0].addr32[3] < b->addr[0].addr32[3]) 671 return (-1); 672 if (a->addr[1].addr32[3] > b->addr[1].addr32[3]) 673 return (1); 674 if (a->addr[1].addr32[3] < b->addr[1].addr32[3]) 675 return (-1); 676 if (a->addr[0].addr32[2] > b->addr[0].addr32[2]) 677 return (1); 678 if (a->addr[0].addr32[2] < b->addr[0].addr32[2]) 679 return (-1); 680 if (a->addr[1].addr32[2] > b->addr[1].addr32[2]) 681 return (1); 682 if (a->addr[1].addr32[2] < b->addr[1].addr32[2]) 683 return (-1); 684 if (a->addr[0].addr32[1] > b->addr[0].addr32[1]) 685 return (1); 686 if (a->addr[0].addr32[1] < b->addr[0].addr32[1]) 687 return (-1); 688 if (a->addr[1].addr32[1] > b->addr[1].addr32[1]) 689 return (1); 690 if (a->addr[1].addr32[1] < b->addr[1].addr32[1]) 691 return (-1); 692 if (a->addr[0].addr32[0] > b->addr[0].addr32[0]) 693 return (1); 694 if (a->addr[0].addr32[0] < b->addr[0].addr32[0]) 695 return (-1); 696 if (a->addr[1].addr32[0] > b->addr[1].addr32[0]) 697 return (1); 698 if (a->addr[1].addr32[0] < b->addr[1].addr32[0]) 699 return (-1); 700 break; 701 #endif /* INET6 */ 702 } 703 704 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); 708 709 return (0); 710 } 711 712 /* 713 * Used for RB_FIND only, compare in the reverse direction. The 714 * element to be reversed is always (a), since we obviously can't 715 * reverse the state tree depicted by (b). 716 */ 717 static __inline int 718 pf_state_compare_rkey(struct pf_state_key *a, struct pf_state_key *b) 719 { 720 int diff; 721 722 if ((diff = a->proto - b->proto) != 0) 723 return (diff); 724 if ((diff = a->af - b->af) != 0) 725 return (diff); 726 switch (a->af) { 727 #ifdef INET 728 case AF_INET: 729 if (a->addr[1].addr32[0] > b->addr[0].addr32[0]) 730 return (1); 731 if (a->addr[1].addr32[0] < b->addr[0].addr32[0]) 732 return (-1); 733 if (a->addr[0].addr32[0] > b->addr[1].addr32[0]) 734 return (1); 735 if (a->addr[0].addr32[0] < b->addr[1].addr32[0]) 736 return (-1); 737 break; 738 #endif /* INET */ 739 #ifdef INET6 740 case AF_INET6: 741 if (a->addr[1].addr32[3] > b->addr[0].addr32[3]) 742 return (1); 743 if (a->addr[1].addr32[3] < b->addr[0].addr32[3]) 744 return (-1); 745 if (a->addr[0].addr32[3] > b->addr[1].addr32[3]) 746 return (1); 747 if (a->addr[0].addr32[3] < b->addr[1].addr32[3]) 748 return (-1); 749 if (a->addr[1].addr32[2] > b->addr[0].addr32[2]) 750 return (1); 751 if (a->addr[1].addr32[2] < b->addr[0].addr32[2]) 752 return (-1); 753 if (a->addr[0].addr32[2] > b->addr[1].addr32[2]) 754 return (1); 755 if (a->addr[0].addr32[2] < b->addr[1].addr32[2]) 756 return (-1); 757 if (a->addr[1].addr32[1] > b->addr[0].addr32[1]) 758 return (1); 759 if (a->addr[1].addr32[1] < b->addr[0].addr32[1]) 760 return (-1); 761 if (a->addr[0].addr32[1] > b->addr[1].addr32[1]) 762 return (1); 763 if (a->addr[0].addr32[1] < b->addr[1].addr32[1]) 764 return (-1); 765 if (a->addr[1].addr32[0] > b->addr[0].addr32[0]) 766 return (1); 767 if (a->addr[1].addr32[0] < b->addr[0].addr32[0]) 768 return (-1); 769 if (a->addr[0].addr32[0] > b->addr[1].addr32[0]) 770 return (1); 771 if (a->addr[0].addr32[0] < b->addr[1].addr32[0]) 772 return (-1); 773 break; 774 #endif /* INET6 */ 775 } 776 777 if ((diff = a->port[1] - b->port[0]) != 0) 778 return (diff); 779 if ((diff = a->port[0] - b->port[1]) != 0) 780 return (diff); 781 782 return (0); 783 } 784 785 static __inline int 786 pf_state_compare_id(struct pf_state *a, struct pf_state *b) 787 { 788 if (a->id > b->id) 789 return (1); 790 if (a->id < b->id) 791 return (-1); 792 if (a->creatorid > b->creatorid) 793 return (1); 794 if (a->creatorid < b->creatorid) 795 return (-1); 796 797 return (0); 798 } 799 800 int 801 pf_state_key_attach(struct pf_state_key *sk, struct pf_state *s, int idx) 802 { 803 struct pf_state_item *si; 804 struct pf_state_key *cur; 805 int cpu; 806 int error; 807 808 /* 809 * PFSTATE_STACK_GLOBAL is set when the state might not hash to the 810 * current cpu. The keys are managed on the global statetbl tree 811 * for this case. Only translations (RDR, NAT) can cause this. 812 * 813 * When this flag is not set we must still check the global statetbl 814 * for a collision, and if we find one we set the HALF_DUPLEX flag 815 * in the state. 816 */ 817 if (s->state_flags & PFSTATE_STACK_GLOBAL) { 818 cpu = ncpus; 819 lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE); 820 } else { 821 cpu = mycpu->gd_cpuid; 822 lockmgr(&pf_global_statetbl_lock, LK_SHARED); 823 } 824 KKASSERT(s->key[idx] == NULL); /* XXX handle this? */ 825 826 if (pf_status.debug >= PF_DEBUG_MISC) { 827 kprintf("state_key attach cpu %d (%08x:%d) %s (%08x:%d)\n", 828 cpu, 829 ntohl(sk->addr[0].addr32[0]), ntohs(sk->port[0]), 830 (idx == PF_SK_WIRE ? "->" : "<-"), 831 ntohl(sk->addr[1].addr32[0]), ntohs(sk->port[1])); 832 } 833 834 /* 835 * Check whether (e.g.) a PASS rule being put on a per-cpu tree 836 * collides with a translation rule on the global tree. This is 837 * NOT an error. We *WANT* to establish state for this case so the 838 * packet path is short-cutted and doesn't need to scan the ruleset 839 * on every packet. But the established state will only see one 840 * side of a two-way packet conversation. To prevent this from 841 * causing problems (e.g. generating a RST), we force PFSTATE_SLOPPY 842 * to be set on the established state. 843 * 844 * A collision against RDR state can only occur with a PASS IN in the 845 * opposite direction or a PASS OUT in the forwards direction. This 846 * is because RDRs are processed on the input side. 847 * 848 * A collision against NAT state can only occur with a PASS IN in the 849 * forwards direction or a PASS OUT in the opposite direction. This 850 * is because NATs are processed on the output side. 851 * 852 * In both situations we need to do a reverse addr/port test because 853 * the PASS IN or PASS OUT only establishes if it doesn't match the 854 * established RDR state in the forwards direction. The direction 855 * flag has to be ignored (it will be one way for a PASS IN and the 856 * other way for a PASS OUT). 857 * 858 * pf_global_statetbl_lock will be locked shared when testing and 859 * not entering into the global state table. 860 */ 861 if (cpu != ncpus && 862 (cur = RB_FIND(pf_state_rtree, 863 (struct pf_state_rtree *)&pf_statetbl[ncpus], 864 sk)) != NULL) { 865 TAILQ_FOREACH(si, &cur->states, entry) { 866 /* 867 * NOTE: We must ignore direction mismatches. 868 */ 869 if (si->s->kif == s->kif) { 870 s->state_flags |= PFSTATE_HALF_DUPLEX | 871 PFSTATE_SLOPPY; 872 if (pf_status.debug >= PF_DEBUG_MISC) { 873 kprintf( 874 "pf: %s key attach collision " 875 "on %s: ", 876 (idx == PF_SK_WIRE) ? 877 "wire" : "stack", 878 s->kif->pfik_name); 879 pf_print_state_parts(s, 880 (idx == PF_SK_WIRE) ? sk : NULL, 881 (idx == PF_SK_STACK) ? sk : NULL); 882 kprintf("\n"); 883 } 884 break; 885 } 886 } 887 } 888 889 /* 890 * Enter into either the per-cpu or the global state table. 891 * 892 * pf_global_statetbl_lock will be locked exclusively when entering 893 * into the global state table. 894 */ 895 if ((cur = RB_INSERT(pf_state_tree, &pf_statetbl[cpu], sk)) != NULL) { 896 /* key exists. check for same kif, if none, add to key */ 897 TAILQ_FOREACH(si, &cur->states, entry) { 898 if (si->s->kif == s->kif && 899 si->s->direction == s->direction) { 900 if (pf_status.debug >= PF_DEBUG_MISC) { 901 kprintf( 902 "pf: %s key attach failed on %s: ", 903 (idx == PF_SK_WIRE) ? 904 "wire" : "stack", 905 s->kif->pfik_name); 906 pf_print_state_parts(s, 907 (idx == PF_SK_WIRE) ? sk : NULL, 908 (idx == PF_SK_STACK) ? sk : NULL); 909 kprintf("\n"); 910 } 911 kfree(sk, M_PFSTATEKEYPL); 912 error = -1; 913 goto failed; /* collision! */ 914 } 915 } 916 kfree(sk, M_PFSTATEKEYPL); 917 918 s->key[idx] = cur; 919 } else { 920 s->key[idx] = sk; 921 } 922 923 if ((si = kmalloc(sizeof(struct pf_state_item), 924 M_PFSTATEITEMPL, M_NOWAIT)) == NULL) { 925 pf_state_key_detach(s, idx); 926 error = -1; 927 goto failed; /* collision! */ 928 } 929 si->s = s; 930 931 /* list is sorted, if-bound states before floating */ 932 if (s->kif == pfi_all) 933 TAILQ_INSERT_TAIL(&s->key[idx]->states, si, entry); 934 else 935 TAILQ_INSERT_HEAD(&s->key[idx]->states, si, entry); 936 937 error = 0; 938 failed: 939 lockmgr(&pf_global_statetbl_lock, LK_RELEASE); 940 return error; 941 } 942 943 /* 944 * NOTE: Can only be called indirectly via the purge thread with pf_token 945 * exclusively locked. 946 */ 947 void 948 pf_detach_state(struct pf_state *s) 949 { 950 if (s->key[PF_SK_WIRE] == s->key[PF_SK_STACK]) 951 s->key[PF_SK_WIRE] = NULL; 952 953 if (s->key[PF_SK_STACK] != NULL) 954 pf_state_key_detach(s, PF_SK_STACK); 955 956 if (s->key[PF_SK_WIRE] != NULL) 957 pf_state_key_detach(s, PF_SK_WIRE); 958 } 959 960 /* 961 * NOTE: Can only be called indirectly via the purge thread with pf_token 962 * exclusively locked. 963 */ 964 void 965 pf_state_key_detach(struct pf_state *s, int idx) 966 { 967 struct pf_state_item *si; 968 int cpu; 969 970 /* 971 * PFSTATE_STACK_GLOBAL is set for translations when the translated 972 * address/port is not localized to the same cpu that the untranslated 973 * address/port is on. The wire pf_state_key is managed on the global 974 * statetbl tree for this case. 975 */ 976 if (s->state_flags & PFSTATE_STACK_GLOBAL) { 977 cpu = ncpus; 978 lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE); 979 } else { 980 cpu = mycpu->gd_cpuid; 981 } 982 983 si = TAILQ_FIRST(&s->key[idx]->states); 984 while (si && si->s != s) 985 si = TAILQ_NEXT(si, entry); 986 987 if (si) { 988 TAILQ_REMOVE(&s->key[idx]->states, si, entry); 989 kfree(si, M_PFSTATEITEMPL); 990 } 991 992 if (TAILQ_EMPTY(&s->key[idx]->states)) { 993 RB_REMOVE(pf_state_tree, &pf_statetbl[cpu], s->key[idx]); 994 if (s->key[idx]->reverse) 995 s->key[idx]->reverse->reverse = NULL; 996 if (s->key[idx]->inp) 997 s->key[idx]->inp->inp_pf_sk = NULL; 998 kfree(s->key[idx], M_PFSTATEKEYPL); 999 } 1000 s->key[idx] = NULL; 1001 1002 if (s->state_flags & PFSTATE_STACK_GLOBAL) 1003 lockmgr(&pf_global_statetbl_lock, LK_RELEASE); 1004 } 1005 1006 struct pf_state_key * 1007 pf_alloc_state_key(int pool_flags) 1008 { 1009 struct pf_state_key *sk; 1010 1011 sk = kmalloc(sizeof(struct pf_state_key), M_PFSTATEKEYPL, pool_flags); 1012 if (sk) { 1013 TAILQ_INIT(&sk->states); 1014 } 1015 return (sk); 1016 } 1017 1018 int 1019 pf_state_key_setup(struct pf_pdesc *pd, struct pf_rule *nr, 1020 struct pf_state_key **skw, struct pf_state_key **sks, 1021 struct pf_state_key **skp, struct pf_state_key **nkp, 1022 struct pf_addr *saddr, struct pf_addr *daddr, 1023 u_int16_t sport, u_int16_t dport) 1024 { 1025 KKASSERT((*skp == NULL && *nkp == NULL)); 1026 1027 if ((*skp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL) 1028 return (ENOMEM); 1029 1030 PF_ACPY(&(*skp)->addr[pd->sidx], saddr, pd->af); 1031 PF_ACPY(&(*skp)->addr[pd->didx], daddr, pd->af); 1032 (*skp)->port[pd->sidx] = sport; 1033 (*skp)->port[pd->didx] = dport; 1034 (*skp)->proto = pd->proto; 1035 (*skp)->af = pd->af; 1036 1037 if (nr != NULL) { 1038 if ((*nkp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL) 1039 return (ENOMEM); /* caller must handle cleanup */ 1040 1041 /* XXX maybe just bcopy and TAILQ_INIT(&(*nkp)->states) */ 1042 PF_ACPY(&(*nkp)->addr[0], &(*skp)->addr[0], pd->af); 1043 PF_ACPY(&(*nkp)->addr[1], &(*skp)->addr[1], pd->af); 1044 (*nkp)->port[0] = (*skp)->port[0]; 1045 (*nkp)->port[1] = (*skp)->port[1]; 1046 (*nkp)->proto = pd->proto; 1047 (*nkp)->af = pd->af; 1048 } else { 1049 *nkp = *skp; 1050 } 1051 1052 if (pd->dir == PF_IN) { 1053 *skw = *skp; 1054 *sks = *nkp; 1055 } else { 1056 *sks = *skp; 1057 *skw = *nkp; 1058 } 1059 return (0); 1060 } 1061 1062 /* 1063 * Insert pf_state with one or two state keys (allowing a reverse path lookup 1064 * which is used by NAT). In the NAT case skw is the initiator (?) and 1065 * sks is the target. 1066 */ 1067 int 1068 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw, 1069 struct pf_state_key *sks, struct pf_state *s) 1070 { 1071 int cpu = mycpu->gd_cpuid; 1072 1073 s->kif = kif; 1074 s->cpuid = cpu; 1075 1076 if (skw == sks) { 1077 if (pf_state_key_attach(skw, s, PF_SK_WIRE)) 1078 return (-1); 1079 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1080 } else { 1081 /* 1082 skw->reverse = sks; 1083 sks->reverse = skw; 1084 */ 1085 if (pf_state_key_attach(skw, s, PF_SK_WIRE)) { 1086 kfree(sks, M_PFSTATEKEYPL); 1087 return (-1); 1088 } 1089 if (pf_state_key_attach(sks, s, PF_SK_STACK)) { 1090 pf_state_key_detach(s, PF_SK_WIRE); 1091 return (-1); 1092 } 1093 } 1094 1095 if (s->id == 0 && s->creatorid == 0) { 1096 u_int64_t sid; 1097 1098 #if __SIZEOF_LONG__ == 8 1099 sid = atomic_fetchadd_long(&pf_status.stateid, 1); 1100 #else 1101 spin_lock(&pf_spin); 1102 sid = pf_status.stateid++; 1103 spin_unlock(&pf_spin); 1104 #endif 1105 s->id = htobe64(sid); 1106 s->creatorid = pf_status.hostid; 1107 } 1108 1109 /* 1110 * Calculate hash code for altq 1111 */ 1112 s->hash = crc32(s->key[PF_SK_WIRE], PF_STATE_KEY_HASH_LENGTH); 1113 1114 if (RB_INSERT(pf_state_tree_id, &tree_id[cpu], s) != NULL) { 1115 if (pf_status.debug >= PF_DEBUG_MISC) { 1116 kprintf("pf: state insert failed: " 1117 "id: %016jx creatorid: %08x", 1118 (uintmax_t)be64toh(s->id), ntohl(s->creatorid)); 1119 if (s->sync_flags & PFSTATE_FROMSYNC) 1120 kprintf(" (from sync)"); 1121 kprintf("\n"); 1122 } 1123 pf_detach_state(s); 1124 return (-1); 1125 } 1126 TAILQ_INSERT_TAIL(&state_list[cpu], s, entry_list); 1127 PF_INC_FCOUNTER(FCNT_STATE_INSERT); 1128 atomic_add_int(&pf_status.states, 1); 1129 pfi_kif_ref(kif, PFI_KIF_REF_STATE); 1130 pfsync_insert_state(s); 1131 return (0); 1132 } 1133 1134 struct pf_state * 1135 pf_find_state_byid(struct pf_state_cmp *key) 1136 { 1137 int cpu = mycpu->gd_cpuid; 1138 1139 PF_INC_FCOUNTER(FCNT_STATE_SEARCH); 1140 1141 return (RB_FIND(pf_state_tree_id, &tree_id[cpu], 1142 (struct pf_state *)key)); 1143 } 1144 1145 /* 1146 * WARNING! May return a state structure that was localized to another cpu, 1147 * destruction is typically protected by the callers pf_token. 1148 * The element can only be destroyed 1149 */ 1150 struct pf_state * 1151 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir, 1152 struct mbuf *m) 1153 { 1154 struct pf_state_key *skey = (void *)key; 1155 struct pf_state_key *sk; 1156 struct pf_state_item *si; 1157 struct pf_state *s; 1158 int cpu = mycpu->gd_cpuid; 1159 int globalstl = 0; 1160 1161 PF_INC_FCOUNTER(FCNT_STATE_SEARCH); 1162 1163 if (dir == PF_OUT && m->m_pkthdr.pf.statekey && 1164 ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->reverse) { 1165 sk = ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->reverse; 1166 } else { 1167 sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey); 1168 if (sk == NULL) { 1169 lockmgr(&pf_global_statetbl_lock, LK_SHARED); 1170 sk = RB_FIND(pf_state_tree, &pf_statetbl[ncpus], skey); 1171 if (sk == NULL) { 1172 lockmgr(&pf_global_statetbl_lock, LK_RELEASE); 1173 return (NULL); 1174 } 1175 globalstl = 1; 1176 } 1177 if (dir == PF_OUT && m->m_pkthdr.pf.statekey) { 1178 ((struct pf_state_key *) 1179 m->m_pkthdr.pf.statekey)->reverse = sk; 1180 sk->reverse = m->m_pkthdr.pf.statekey; 1181 } 1182 } 1183 if (dir == PF_OUT) 1184 m->m_pkthdr.pf.statekey = NULL; 1185 1186 /* list is sorted, if-bound states before floating ones */ 1187 TAILQ_FOREACH(si, &sk->states, entry) { 1188 if ((si->s->kif == pfi_all || si->s->kif == kif) && 1189 sk == (dir == PF_IN ? si->s->key[PF_SK_WIRE] : 1190 si->s->key[PF_SK_STACK])) { 1191 break; 1192 } 1193 } 1194 1195 /* 1196 * Extract state before potentially releasing the global statetbl 1197 * lock. Ignore the state if the create is still in-progress as 1198 * it can be deleted out from under us by the owning localized cpu. 1199 * However, if CREATEINPROG is not set, state can only be deleted 1200 * by the purge thread which we are protected from via our shared 1201 * pf_token. 1202 */ 1203 if (si) { 1204 s = si->s; 1205 if (s && (s->state_flags & PFSTATE_CREATEINPROG)) 1206 s = NULL; 1207 } else { 1208 s = NULL; 1209 } 1210 if (globalstl) 1211 lockmgr(&pf_global_statetbl_lock, LK_RELEASE); 1212 return s; 1213 } 1214 1215 /* 1216 * WARNING! May return a state structure that was localized to another cpu, 1217 * destruction is typically protected by the callers pf_token. 1218 */ 1219 struct pf_state * 1220 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1221 { 1222 struct pf_state_key *skey = (void *)key; 1223 struct pf_state_key *sk; 1224 struct pf_state_item *si, *ret = NULL; 1225 struct pf_state *s; 1226 int cpu = mycpu->gd_cpuid; 1227 int globalstl = 0; 1228 1229 PF_INC_FCOUNTER(FCNT_STATE_SEARCH); 1230 1231 sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey); 1232 if (sk == NULL) { 1233 lockmgr(&pf_global_statetbl_lock, LK_SHARED); 1234 sk = RB_FIND(pf_state_tree, &pf_statetbl[ncpus], skey); 1235 globalstl = 1; 1236 } 1237 if (sk != NULL) { 1238 TAILQ_FOREACH(si, &sk->states, entry) 1239 if (dir == PF_INOUT || 1240 (sk == (dir == PF_IN ? si->s->key[PF_SK_WIRE] : 1241 si->s->key[PF_SK_STACK]))) { 1242 if (more == NULL) { 1243 ret = si; 1244 break; 1245 } 1246 if (ret) 1247 (*more)++; 1248 else 1249 ret = si; 1250 } 1251 } 1252 1253 /* 1254 * Extract state before potentially releasing the global statetbl 1255 * lock. Ignore the state if the create is still in-progress as 1256 * it can be deleted out from under us by the owning localized cpu. 1257 * However, if CREATEINPROG is not set, state can only be deleted 1258 * by the purge thread which we are protected from via our shared 1259 * pf_token. 1260 */ 1261 if (ret) { 1262 s = ret->s; 1263 if (s && (s->state_flags & PFSTATE_CREATEINPROG)) 1264 s = NULL; 1265 } else { 1266 s = NULL; 1267 } 1268 if (globalstl) 1269 lockmgr(&pf_global_statetbl_lock, LK_RELEASE); 1270 return s; 1271 } 1272 1273 /* END state table stuff */ 1274 1275 void 1276 pf_purge_thread(void *v) 1277 { 1278 globaldata_t save_gd = mycpu; 1279 int nloops = 0; 1280 int locked = 0; 1281 int nn; 1282 int endingit; 1283 1284 for (;;) { 1285 tsleep(pf_purge_thread, PWAIT, "pftm", 1 * hz); 1286 1287 endingit = pf_end_threads; 1288 1289 for (nn = 0; nn < ncpus; ++nn) { 1290 lwkt_setcpu_self(globaldata_find(nn)); 1291 1292 lwkt_gettoken(&pf_token); 1293 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE); 1294 crit_enter(); 1295 1296 /* 1297 * process a fraction of the state table every second 1298 */ 1299 if(!pf_purge_expired_states( 1300 1 + (pf_status.states / 1301 pf_default_rule.timeout[ 1302 PFTM_INTERVAL]), 0)) { 1303 pf_purge_expired_states( 1304 1 + (pf_status.states / 1305 pf_default_rule.timeout[ 1306 PFTM_INTERVAL]), 1); 1307 } 1308 1309 /* 1310 * purge other expired types every PFTM_INTERVAL 1311 * seconds 1312 */ 1313 if (++nloops >= 1314 pf_default_rule.timeout[PFTM_INTERVAL]) { 1315 pf_purge_expired_fragments(); 1316 if (!pf_purge_expired_src_nodes(locked)) { 1317 pf_purge_expired_src_nodes(1); 1318 } 1319 nloops = 0; 1320 } 1321 1322 /* 1323 * If terminating the thread, clean everything out 1324 * (on all cpus). 1325 */ 1326 if (endingit) { 1327 pf_purge_expired_states(pf_status.states, 0); 1328 pf_purge_expired_fragments(); 1329 pf_purge_expired_src_nodes(1); 1330 } 1331 1332 crit_exit(); 1333 lockmgr(&pf_consistency_lock, LK_RELEASE); 1334 lwkt_reltoken(&pf_token); 1335 } 1336 lwkt_setcpu_self(save_gd); 1337 if (endingit) 1338 break; 1339 } 1340 1341 /* 1342 * Thread termination 1343 */ 1344 pf_end_threads++; 1345 wakeup(pf_purge_thread); 1346 kthread_exit(); 1347 } 1348 1349 u_int32_t 1350 pf_state_expires(const struct pf_state *state) 1351 { 1352 u_int32_t timeout; 1353 u_int32_t start; 1354 u_int32_t end; 1355 u_int32_t states; 1356 1357 /* handle all PFTM_* > PFTM_MAX here */ 1358 if (state->timeout == PFTM_PURGE) 1359 return (time_second); 1360 if (state->timeout == PFTM_UNTIL_PACKET) 1361 return (0); 1362 KKASSERT(state->timeout != PFTM_UNLINKED); 1363 KKASSERT(state->timeout < PFTM_MAX); 1364 timeout = state->rule.ptr->timeout[state->timeout]; 1365 if (!timeout) 1366 timeout = pf_default_rule.timeout[state->timeout]; 1367 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1368 if (start) { 1369 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1370 states = state->rule.ptr->states_cur; 1371 } else { 1372 start = pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1373 end = pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1374 states = pf_status.states; 1375 } 1376 if (end && states > start && start < end) { 1377 if (states < end) 1378 return (state->expire + timeout * (end - states) / 1379 (end - start)); 1380 else 1381 return (time_second); 1382 } 1383 return (state->expire + timeout); 1384 } 1385 1386 /* 1387 * (called with exclusive pf_token) 1388 */ 1389 int 1390 pf_purge_expired_src_nodes(int waslocked) 1391 { 1392 struct pf_src_node *cur, *next; 1393 int locked = waslocked; 1394 int cpu = mycpu->gd_cpuid; 1395 1396 for (cur = RB_MIN(pf_src_tree, &tree_src_tracking[cpu]); 1397 cur; 1398 cur = next) { 1399 next = RB_NEXT(pf_src_tree, &tree_src_tracking[cpu], cur); 1400 1401 if (cur->states <= 0 && cur->expire <= time_second) { 1402 if (!locked) { 1403 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE); 1404 next = RB_NEXT(pf_src_tree, 1405 &tree_src_tracking[cpu], cur); 1406 locked = 1; 1407 } 1408 if (cur->rule.ptr != NULL) { 1409 /* 1410 * decrements in rule should be ok, token is 1411 * held exclusively in this code path. 1412 */ 1413 atomic_add_int(&cur->rule.ptr->src_nodes, -1); 1414 if (cur->rule.ptr->states_cur <= 0 && 1415 cur->rule.ptr->max_src_nodes <= 0) 1416 pf_rm_rule(NULL, cur->rule.ptr); 1417 } 1418 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], cur); 1419 PF_INC_SCOUNTER(SCNT_SRC_NODE_REMOVALS); 1420 atomic_add_int(&pf_status.src_nodes, -1); 1421 kfree(cur, M_PFSRCTREEPL); 1422 } 1423 } 1424 if (locked && !waslocked) 1425 lockmgr(&pf_consistency_lock, LK_RELEASE); 1426 return(1); 1427 } 1428 1429 void 1430 pf_src_tree_remove_state(struct pf_state *s) 1431 { 1432 u_int32_t timeout; 1433 1434 if (s->src_node != NULL) { 1435 if (s->src.tcp_est) 1436 atomic_add_int(&s->src_node->conn, -1); 1437 if (--s->src_node->states <= 0) { 1438 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE]; 1439 if (!timeout) { 1440 timeout = 1441 pf_default_rule.timeout[PFTM_SRC_NODE]; 1442 } 1443 s->src_node->expire = time_second + timeout; 1444 } 1445 } 1446 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 1447 if (--s->nat_src_node->states <= 0) { 1448 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE]; 1449 if (!timeout) 1450 timeout = 1451 pf_default_rule.timeout[PFTM_SRC_NODE]; 1452 s->nat_src_node->expire = time_second + timeout; 1453 } 1454 } 1455 s->src_node = s->nat_src_node = NULL; 1456 } 1457 1458 /* callers should be at crit_enter() */ 1459 void 1460 pf_unlink_state(struct pf_state *cur) 1461 { 1462 int cpu = mycpu->gd_cpuid; 1463 1464 if (cur->src.state == PF_TCPS_PROXY_DST) { 1465 /* XXX wire key the right one? */ 1466 pf_send_tcp(cur->rule.ptr, cur->key[PF_SK_WIRE]->af, 1467 &cur->key[PF_SK_WIRE]->addr[1], 1468 &cur->key[PF_SK_WIRE]->addr[0], 1469 cur->key[PF_SK_WIRE]->port[1], 1470 cur->key[PF_SK_WIRE]->port[0], 1471 cur->src.seqhi, cur->src.seqlo + 1, 1472 TH_RST|TH_ACK, 0, 0, 0, 1, cur->tag, NULL, NULL); 1473 } 1474 RB_REMOVE(pf_state_tree_id, &tree_id[cpu], cur); 1475 if (cur->creatorid == pf_status.hostid) 1476 pfsync_delete_state(cur); 1477 cur->timeout = PFTM_UNLINKED; 1478 pf_src_tree_remove_state(cur); 1479 pf_detach_state(cur); 1480 } 1481 1482 /* 1483 * callers should be at crit_enter() and hold pf_consistency_lock exclusively. 1484 * pf_token must also be held exclusively. 1485 */ 1486 void 1487 pf_free_state(struct pf_state *cur) 1488 { 1489 int cpu = mycpu->gd_cpuid; 1490 1491 KKASSERT(cur->cpuid == cpu); 1492 1493 if (pfsyncif != NULL && 1494 (pfsyncif->sc_bulk_send_next == cur || 1495 pfsyncif->sc_bulk_terminator == cur)) 1496 return; 1497 KKASSERT(cur->timeout == PFTM_UNLINKED); 1498 /* 1499 * decrements in rule should be ok, token is 1500 * held exclusively in this code path. 1501 */ 1502 if (--cur->rule.ptr->states_cur <= 0 && 1503 cur->rule.ptr->src_nodes <= 0) 1504 pf_rm_rule(NULL, cur->rule.ptr); 1505 if (cur->nat_rule.ptr != NULL) { 1506 if (--cur->nat_rule.ptr->states_cur <= 0 && 1507 cur->nat_rule.ptr->src_nodes <= 0) { 1508 pf_rm_rule(NULL, cur->nat_rule.ptr); 1509 } 1510 } 1511 if (cur->anchor.ptr != NULL) { 1512 if (--cur->anchor.ptr->states_cur <= 0) 1513 pf_rm_rule(NULL, cur->anchor.ptr); 1514 } 1515 pf_normalize_tcp_cleanup(cur); 1516 pfi_kif_unref(cur->kif, PFI_KIF_REF_STATE); 1517 1518 /* 1519 * We may be freeing pf_purge_expired_states()'s saved scan entry, 1520 * adjust it if necessary. 1521 */ 1522 if (purge_cur[cpu] == cur) { 1523 kprintf("PURGE CONFLICT\n"); 1524 purge_cur[cpu] = TAILQ_NEXT(purge_cur[cpu], entry_list); 1525 } 1526 TAILQ_REMOVE(&state_list[cpu], cur, entry_list); 1527 if (cur->tag) 1528 pf_tag_unref(cur->tag); 1529 kfree(cur, M_PFSTATEPL); 1530 PF_INC_FCOUNTER(FCNT_STATE_REMOVALS); 1531 atomic_add_int(&pf_status.states, -1); 1532 } 1533 1534 int 1535 pf_purge_expired_states(u_int32_t maxcheck, int waslocked) 1536 { 1537 struct pf_state *cur; 1538 int locked = waslocked; 1539 int cpu = mycpu->gd_cpuid; 1540 1541 while (maxcheck--) { 1542 /* 1543 * Wrap to start of list when we hit the end 1544 */ 1545 cur = purge_cur[cpu]; 1546 if (cur == NULL) { 1547 cur = TAILQ_FIRST(&state_list[cpu]); 1548 if (cur == NULL) 1549 break; /* list empty */ 1550 } 1551 1552 /* 1553 * Setup next (purge_cur) while we process this one. If 1554 * we block and something else deletes purge_cur, 1555 * pf_free_state() will adjust it further ahead. 1556 */ 1557 purge_cur[cpu] = TAILQ_NEXT(cur, entry_list); 1558 1559 if (cur->timeout == PFTM_UNLINKED) { 1560 /* free unlinked state */ 1561 if (! locked) { 1562 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE); 1563 locked = 1; 1564 } 1565 pf_free_state(cur); 1566 } else if (pf_state_expires(cur) <= time_second) { 1567 /* unlink and free expired state */ 1568 pf_unlink_state(cur); 1569 if (! locked) { 1570 if (!lockmgr(&pf_consistency_lock, LK_EXCLUSIVE)) 1571 return (0); 1572 locked = 1; 1573 } 1574 pf_free_state(cur); 1575 } 1576 } 1577 1578 if (locked) 1579 lockmgr(&pf_consistency_lock, LK_RELEASE); 1580 return (1); 1581 } 1582 1583 int 1584 pf_tbladdr_setup(struct pf_ruleset *rs, struct pf_addr_wrap *aw) 1585 { 1586 if (aw->type != PF_ADDR_TABLE) 1587 return (0); 1588 if ((aw->p.tbl = pfr_attach_table(rs, aw->v.tblname)) == NULL) 1589 return (1); 1590 return (0); 1591 } 1592 1593 void 1594 pf_tbladdr_remove(struct pf_addr_wrap *aw) 1595 { 1596 if (aw->type != PF_ADDR_TABLE || aw->p.tbl == NULL) 1597 return; 1598 pfr_detach_table(aw->p.tbl); 1599 aw->p.tbl = NULL; 1600 } 1601 1602 void 1603 pf_tbladdr_copyout(struct pf_addr_wrap *aw) 1604 { 1605 struct pfr_ktable *kt = aw->p.tbl; 1606 1607 if (aw->type != PF_ADDR_TABLE || kt == NULL) 1608 return; 1609 if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) 1610 kt = kt->pfrkt_root; 1611 aw->p.tbl = NULL; 1612 aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ? 1613 kt->pfrkt_cnt : -1; 1614 } 1615 1616 void 1617 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 1618 { 1619 switch (af) { 1620 #ifdef INET 1621 case AF_INET: { 1622 u_int32_t a = ntohl(addr->addr32[0]); 1623 kprintf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 1624 (a>>8)&255, a&255); 1625 if (p) { 1626 p = ntohs(p); 1627 kprintf(":%u", p); 1628 } 1629 break; 1630 } 1631 #endif /* INET */ 1632 #ifdef INET6 1633 case AF_INET6: { 1634 u_int16_t b; 1635 u_int8_t i, curstart, curend, maxstart, maxend; 1636 curstart = curend = maxstart = maxend = 255; 1637 for (i = 0; i < 8; i++) { 1638 if (!addr->addr16[i]) { 1639 if (curstart == 255) 1640 curstart = i; 1641 curend = i; 1642 } else { 1643 if ((curend - curstart) > 1644 (maxend - maxstart)) { 1645 maxstart = curstart; 1646 maxend = curend; 1647 } 1648 curstart = curend = 255; 1649 } 1650 } 1651 if ((curend - curstart) > 1652 (maxend - maxstart)) { 1653 maxstart = curstart; 1654 maxend = curend; 1655 } 1656 for (i = 0; i < 8; i++) { 1657 if (i >= maxstart && i <= maxend) { 1658 if (i == 0) 1659 kprintf(":"); 1660 if (i == maxend) 1661 kprintf(":"); 1662 } else { 1663 b = ntohs(addr->addr16[i]); 1664 kprintf("%x", b); 1665 if (i < 7) 1666 kprintf(":"); 1667 } 1668 } 1669 if (p) { 1670 p = ntohs(p); 1671 kprintf("[%u]", p); 1672 } 1673 break; 1674 } 1675 #endif /* INET6 */ 1676 } 1677 } 1678 1679 void 1680 pf_print_state(struct pf_state *s) 1681 { 1682 pf_print_state_parts(s, NULL, NULL); 1683 } 1684 1685 void 1686 pf_print_state_parts(struct pf_state *s, 1687 struct pf_state_key *skwp, struct pf_state_key *sksp) 1688 { 1689 struct pf_state_key *skw, *sks; 1690 u_int8_t proto, dir; 1691 1692 /* Do our best to fill these, but they're skipped if NULL */ 1693 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 1694 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 1695 proto = skw ? skw->proto : (sks ? sks->proto : 0); 1696 dir = s ? s->direction : 0; 1697 1698 switch (proto) { 1699 case IPPROTO_TCP: 1700 kprintf("TCP "); 1701 break; 1702 case IPPROTO_UDP: 1703 kprintf("UDP "); 1704 break; 1705 case IPPROTO_ICMP: 1706 kprintf("ICMP "); 1707 break; 1708 case IPPROTO_ICMPV6: 1709 kprintf("ICMPV6 "); 1710 break; 1711 default: 1712 kprintf("%u ", skw->proto); 1713 break; 1714 } 1715 switch (dir) { 1716 case PF_IN: 1717 kprintf(" in"); 1718 break; 1719 case PF_OUT: 1720 kprintf(" out"); 1721 break; 1722 } 1723 if (skw) { 1724 kprintf(" wire: "); 1725 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 1726 kprintf(" "); 1727 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 1728 } 1729 if (sks) { 1730 kprintf(" stack: "); 1731 if (sks != skw) { 1732 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 1733 kprintf(" "); 1734 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 1735 } else 1736 kprintf("-"); 1737 } 1738 if (s) { 1739 if (proto == IPPROTO_TCP) { 1740 kprintf(" [lo=%u high=%u win=%u modulator=%u", 1741 s->src.seqlo, s->src.seqhi, 1742 s->src.max_win, s->src.seqdiff); 1743 if (s->src.wscale && s->dst.wscale) 1744 kprintf(" wscale=%u", 1745 s->src.wscale & PF_WSCALE_MASK); 1746 kprintf("]"); 1747 kprintf(" [lo=%u high=%u win=%u modulator=%u", 1748 s->dst.seqlo, s->dst.seqhi, 1749 s->dst.max_win, s->dst.seqdiff); 1750 if (s->src.wscale && s->dst.wscale) 1751 kprintf(" wscale=%u", 1752 s->dst.wscale & PF_WSCALE_MASK); 1753 kprintf("]"); 1754 } 1755 kprintf(" %u:%u", s->src.state, s->dst.state); 1756 } 1757 } 1758 1759 void 1760 pf_print_flags(u_int8_t f) 1761 { 1762 if (f) 1763 kprintf(" "); 1764 if (f & TH_FIN) 1765 kprintf("F"); 1766 if (f & TH_SYN) 1767 kprintf("S"); 1768 if (f & TH_RST) 1769 kprintf("R"); 1770 if (f & TH_PUSH) 1771 kprintf("P"); 1772 if (f & TH_ACK) 1773 kprintf("A"); 1774 if (f & TH_URG) 1775 kprintf("U"); 1776 if (f & TH_ECE) 1777 kprintf("E"); 1778 if (f & TH_CWR) 1779 kprintf("W"); 1780 } 1781 1782 #define PF_SET_SKIP_STEPS(i) \ 1783 do { \ 1784 while (head[i] != cur) { \ 1785 head[i]->skip[i].ptr = cur; \ 1786 head[i] = TAILQ_NEXT(head[i], entries); \ 1787 } \ 1788 } while (0) 1789 1790 void 1791 pf_calc_skip_steps(struct pf_rulequeue *rules) 1792 { 1793 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT]; 1794 int i; 1795 1796 cur = TAILQ_FIRST(rules); 1797 prev = cur; 1798 for (i = 0; i < PF_SKIP_COUNT; ++i) 1799 head[i] = cur; 1800 while (cur != NULL) { 1801 1802 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 1803 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 1804 if (cur->direction != prev->direction) 1805 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 1806 if (cur->af != prev->af) 1807 PF_SET_SKIP_STEPS(PF_SKIP_AF); 1808 if (cur->proto != prev->proto) 1809 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 1810 if (cur->src.neg != prev->src.neg || 1811 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 1812 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 1813 if (cur->src.port[0] != prev->src.port[0] || 1814 cur->src.port[1] != prev->src.port[1] || 1815 cur->src.port_op != prev->src.port_op) 1816 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 1817 if (cur->dst.neg != prev->dst.neg || 1818 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 1819 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 1820 if (cur->dst.port[0] != prev->dst.port[0] || 1821 cur->dst.port[1] != prev->dst.port[1] || 1822 cur->dst.port_op != prev->dst.port_op) 1823 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 1824 1825 prev = cur; 1826 cur = TAILQ_NEXT(cur, entries); 1827 } 1828 for (i = 0; i < PF_SKIP_COUNT; ++i) 1829 PF_SET_SKIP_STEPS(i); 1830 } 1831 1832 int 1833 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 1834 { 1835 if (aw1->type != aw2->type) 1836 return (1); 1837 switch (aw1->type) { 1838 case PF_ADDR_ADDRMASK: 1839 case PF_ADDR_RANGE: 1840 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0)) 1841 return (1); 1842 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0)) 1843 return (1); 1844 return (0); 1845 case PF_ADDR_DYNIFTL: 1846 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 1847 case PF_ADDR_NOROUTE: 1848 case PF_ADDR_URPFFAILED: 1849 return (0); 1850 case PF_ADDR_TABLE: 1851 return (aw1->p.tbl != aw2->p.tbl); 1852 case PF_ADDR_RTLABEL: 1853 return (aw1->v.rtlabel != aw2->v.rtlabel); 1854 default: 1855 kprintf("invalid address type: %d\n", aw1->type); 1856 return (1); 1857 } 1858 } 1859 1860 u_int16_t 1861 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 1862 { 1863 u_int32_t l; 1864 1865 if (udp && !cksum) 1866 return (0x0000); 1867 l = cksum + old - new; 1868 l = (l >> 16) + (l & 65535); 1869 l = l & 65535; 1870 if (udp && !l) 1871 return (0xFFFF); 1872 return (l); 1873 } 1874 1875 void 1876 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc, 1877 struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af) 1878 { 1879 struct pf_addr ao; 1880 u_int16_t po = *p; 1881 1882 PF_ACPY(&ao, a, af); 1883 PF_ACPY(a, an, af); 1884 1885 *p = pn; 1886 1887 switch (af) { 1888 #ifdef INET 1889 case AF_INET: 1890 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 1891 ao.addr16[0], an->addr16[0], 0), 1892 ao.addr16[1], an->addr16[1], 0); 1893 *p = pn; 1894 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 1895 ao.addr16[0], an->addr16[0], u), 1896 ao.addr16[1], an->addr16[1], u), 1897 po, pn, u); 1898 break; 1899 #endif /* INET */ 1900 #ifdef INET6 1901 case AF_INET6: 1902 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 1903 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 1904 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 1905 ao.addr16[0], an->addr16[0], u), 1906 ao.addr16[1], an->addr16[1], u), 1907 ao.addr16[2], an->addr16[2], u), 1908 ao.addr16[3], an->addr16[3], u), 1909 ao.addr16[4], an->addr16[4], u), 1910 ao.addr16[5], an->addr16[5], u), 1911 ao.addr16[6], an->addr16[6], u), 1912 ao.addr16[7], an->addr16[7], u), 1913 po, pn, u); 1914 break; 1915 #endif /* INET6 */ 1916 } 1917 } 1918 1919 1920 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 1921 void 1922 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 1923 { 1924 u_int32_t ao; 1925 1926 memcpy(&ao, a, sizeof(ao)); 1927 memcpy(a, &an, sizeof(u_int32_t)); 1928 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 1929 ao % 65536, an % 65536, u); 1930 } 1931 1932 #ifdef INET6 1933 void 1934 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 1935 { 1936 struct pf_addr ao; 1937 1938 PF_ACPY(&ao, a, AF_INET6); 1939 PF_ACPY(a, an, AF_INET6); 1940 1941 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 1942 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 1943 pf_cksum_fixup(pf_cksum_fixup(*c, 1944 ao.addr16[0], an->addr16[0], u), 1945 ao.addr16[1], an->addr16[1], u), 1946 ao.addr16[2], an->addr16[2], u), 1947 ao.addr16[3], an->addr16[3], u), 1948 ao.addr16[4], an->addr16[4], u), 1949 ao.addr16[5], an->addr16[5], u), 1950 ao.addr16[6], an->addr16[6], u), 1951 ao.addr16[7], an->addr16[7], u); 1952 } 1953 #endif /* INET6 */ 1954 1955 void 1956 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 1957 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 1958 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 1959 { 1960 struct pf_addr oia, ooa; 1961 1962 PF_ACPY(&oia, ia, af); 1963 if (oa) 1964 PF_ACPY(&ooa, oa, af); 1965 1966 /* Change inner protocol port, fix inner protocol checksum. */ 1967 if (ip != NULL) { 1968 u_int16_t oip = *ip; 1969 u_int32_t opc = 0; 1970 1971 if (pc != NULL) 1972 opc = *pc; 1973 *ip = np; 1974 if (pc != NULL) 1975 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 1976 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 1977 if (pc != NULL) 1978 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 1979 } 1980 /* Change inner ip address, fix inner ip and icmp checksums. */ 1981 PF_ACPY(ia, na, af); 1982 switch (af) { 1983 #ifdef INET 1984 case AF_INET: { 1985 u_int32_t oh2c = *h2c; 1986 1987 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 1988 oia.addr16[0], ia->addr16[0], 0), 1989 oia.addr16[1], ia->addr16[1], 0); 1990 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 1991 oia.addr16[0], ia->addr16[0], 0), 1992 oia.addr16[1], ia->addr16[1], 0); 1993 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 1994 break; 1995 } 1996 #endif /* INET */ 1997 #ifdef INET6 1998 case AF_INET6: 1999 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2000 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2001 pf_cksum_fixup(pf_cksum_fixup(*ic, 2002 oia.addr16[0], ia->addr16[0], u), 2003 oia.addr16[1], ia->addr16[1], u), 2004 oia.addr16[2], ia->addr16[2], u), 2005 oia.addr16[3], ia->addr16[3], u), 2006 oia.addr16[4], ia->addr16[4], u), 2007 oia.addr16[5], ia->addr16[5], u), 2008 oia.addr16[6], ia->addr16[6], u), 2009 oia.addr16[7], ia->addr16[7], u); 2010 break; 2011 #endif /* INET6 */ 2012 } 2013 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2014 if (oa) { 2015 PF_ACPY(oa, na, af); 2016 switch (af) { 2017 #ifdef INET 2018 case AF_INET: 2019 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2020 ooa.addr16[0], oa->addr16[0], 0), 2021 ooa.addr16[1], oa->addr16[1], 0); 2022 break; 2023 #endif /* INET */ 2024 #ifdef INET6 2025 case AF_INET6: 2026 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2027 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2028 pf_cksum_fixup(pf_cksum_fixup(*ic, 2029 ooa.addr16[0], oa->addr16[0], u), 2030 ooa.addr16[1], oa->addr16[1], u), 2031 ooa.addr16[2], oa->addr16[2], u), 2032 ooa.addr16[3], oa->addr16[3], u), 2033 ooa.addr16[4], oa->addr16[4], u), 2034 ooa.addr16[5], oa->addr16[5], u), 2035 ooa.addr16[6], oa->addr16[6], u), 2036 ooa.addr16[7], oa->addr16[7], u); 2037 break; 2038 #endif /* INET6 */ 2039 } 2040 } 2041 } 2042 2043 2044 /* 2045 * Need to modulate the sequence numbers in the TCP SACK option 2046 * (credits to Krzysztof Pfaff for report and patch) 2047 */ 2048 int 2049 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2050 struct tcphdr *th, struct pf_state_peer *dst) 2051 { 2052 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2053 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2054 int copyback = 0, i, olen; 2055 struct raw_sackblock sack; 2056 2057 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2058 if (hlen < TCPOLEN_SACKLEN || 2059 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2060 return 0; 2061 2062 while (hlen >= TCPOLEN_SACKLEN) { 2063 olen = opt[1]; 2064 switch (*opt) { 2065 case TCPOPT_EOL: /* FALLTHROUGH */ 2066 case TCPOPT_NOP: 2067 opt++; 2068 hlen--; 2069 break; 2070 case TCPOPT_SACK: 2071 if (olen > hlen) 2072 olen = hlen; 2073 if (olen >= TCPOLEN_SACKLEN) { 2074 for (i = 2; i + TCPOLEN_SACK <= olen; 2075 i += TCPOLEN_SACK) { 2076 memcpy(&sack, &opt[i], sizeof(sack)); 2077 pf_change_a(&sack.rblk_start, &th->th_sum, 2078 htonl(ntohl(sack.rblk_start) - 2079 dst->seqdiff), 0); 2080 pf_change_a(&sack.rblk_end, &th->th_sum, 2081 htonl(ntohl(sack.rblk_end) - 2082 dst->seqdiff), 0); 2083 memcpy(&opt[i], &sack, sizeof(sack)); 2084 } 2085 copyback = 1; 2086 } 2087 /* FALLTHROUGH */ 2088 default: 2089 if (olen < 2) 2090 olen = 2; 2091 hlen -= olen; 2092 opt += olen; 2093 } 2094 } 2095 2096 if (copyback) 2097 m_copyback(m, off + sizeof(*th), thoptlen, opts); 2098 return (copyback); 2099 } 2100 2101 void 2102 pf_send_tcp(const struct pf_rule *r, sa_family_t af, 2103 const struct pf_addr *saddr, const struct pf_addr *daddr, 2104 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2105 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2106 u_int16_t rtag, struct ether_header *eh, struct ifnet *ifp) 2107 { 2108 struct mbuf *m; 2109 int len = 0, tlen; 2110 #ifdef INET 2111 struct ip *h = NULL; 2112 #endif /* INET */ 2113 #ifdef INET6 2114 struct ip6_hdr *h6 = NULL; 2115 #endif /* INET6 */ 2116 struct tcphdr *th = NULL; 2117 char *opt; 2118 2119 ASSERT_LWKT_TOKEN_HELD(&pf_token); 2120 2121 /* maximum segment size tcp option */ 2122 tlen = sizeof(struct tcphdr); 2123 if (mss) 2124 tlen += 4; 2125 2126 switch (af) { 2127 #ifdef INET 2128 case AF_INET: 2129 len = sizeof(struct ip) + tlen; 2130 break; 2131 #endif /* INET */ 2132 #ifdef INET6 2133 case AF_INET6: 2134 len = sizeof(struct ip6_hdr) + tlen; 2135 break; 2136 #endif /* INET6 */ 2137 } 2138 2139 /* 2140 * Create outgoing mbuf. 2141 * 2142 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags, 2143 * so make sure pf.flags is clear. 2144 */ 2145 m = m_gethdr(M_NOWAIT, MT_HEADER); 2146 if (m == NULL) { 2147 return; 2148 } 2149 if (tag) 2150 m->m_pkthdr.fw_flags |= PF_MBUF_TAGGED; 2151 m->m_pkthdr.pf.flags = 0; 2152 m->m_pkthdr.pf.tag = rtag; 2153 /* XXX Recheck when upgrading to > 4.4 */ 2154 m->m_pkthdr.pf.statekey = NULL; 2155 if (r != NULL && r->rtableid >= 0) 2156 m->m_pkthdr.pf.rtableid = r->rtableid; 2157 2158 #ifdef ALTQ 2159 if (r != NULL && r->qid) { 2160 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE; 2161 m->m_pkthdr.pf.qid = r->qid; 2162 m->m_pkthdr.pf.ecn_af = af; 2163 m->m_pkthdr.pf.hdr = mtod(m, struct ip *); 2164 } 2165 #endif /* ALTQ */ 2166 m->m_data += max_linkhdr; 2167 m->m_pkthdr.len = m->m_len = len; 2168 m->m_pkthdr.rcvif = NULL; 2169 bzero(m->m_data, len); 2170 switch (af) { 2171 #ifdef INET 2172 case AF_INET: 2173 h = mtod(m, struct ip *); 2174 2175 /* IP header fields included in the TCP checksum */ 2176 h->ip_p = IPPROTO_TCP; 2177 h->ip_len = tlen; 2178 h->ip_src.s_addr = saddr->v4.s_addr; 2179 h->ip_dst.s_addr = daddr->v4.s_addr; 2180 2181 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2182 break; 2183 #endif /* INET */ 2184 #ifdef INET6 2185 case AF_INET6: 2186 h6 = mtod(m, struct ip6_hdr *); 2187 2188 /* IP header fields included in the TCP checksum */ 2189 h6->ip6_nxt = IPPROTO_TCP; 2190 h6->ip6_plen = htons(tlen); 2191 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2192 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2193 2194 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2195 break; 2196 #endif /* INET6 */ 2197 } 2198 2199 /* TCP header */ 2200 th->th_sport = sport; 2201 th->th_dport = dport; 2202 th->th_seq = htonl(seq); 2203 th->th_ack = htonl(ack); 2204 th->th_off = tlen >> 2; 2205 th->th_flags = flags; 2206 th->th_win = htons(win); 2207 2208 if (mss) { 2209 opt = (char *)(th + 1); 2210 opt[0] = TCPOPT_MAXSEG; 2211 opt[1] = 4; 2212 mss = htons(mss); 2213 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2214 } 2215 2216 switch (af) { 2217 #ifdef INET 2218 case AF_INET: 2219 /* TCP checksum */ 2220 th->th_sum = in_cksum(m, len); 2221 2222 /* Finish the IP header */ 2223 h->ip_v = 4; 2224 h->ip_hl = sizeof(*h) >> 2; 2225 h->ip_tos = IPTOS_LOWDELAY; 2226 h->ip_len = len; 2227 h->ip_off = path_mtu_discovery ? IP_DF : 0; 2228 h->ip_ttl = ttl ? ttl : ip_defttl; 2229 h->ip_sum = 0; 2230 if (eh == NULL) { 2231 lwkt_reltoken(&pf_token); 2232 ip_output(m, NULL, NULL, 0, NULL, NULL); 2233 lwkt_gettoken(&pf_token); 2234 } else { 2235 struct route ro; 2236 struct rtentry rt; 2237 struct ether_header *e = (void *)ro.ro_dst.sa_data; 2238 2239 if (ifp == NULL) { 2240 m_freem(m); 2241 return; 2242 } 2243 rt.rt_ifp = ifp; 2244 ro.ro_rt = &rt; 2245 ro.ro_dst.sa_len = sizeof(ro.ro_dst); 2246 ro.ro_dst.sa_family = pseudo_AF_HDRCMPLT; 2247 bcopy(eh->ether_dhost, e->ether_shost, ETHER_ADDR_LEN); 2248 bcopy(eh->ether_shost, e->ether_dhost, ETHER_ADDR_LEN); 2249 e->ether_type = eh->ether_type; 2250 /* XXX_IMPORT: later */ 2251 lwkt_reltoken(&pf_token); 2252 ip_output(m, NULL, &ro, 0, NULL, NULL); 2253 lwkt_gettoken(&pf_token); 2254 } 2255 break; 2256 #endif /* INET */ 2257 #ifdef INET6 2258 case AF_INET6: 2259 /* TCP checksum */ 2260 th->th_sum = in6_cksum(m, IPPROTO_TCP, 2261 sizeof(struct ip6_hdr), tlen); 2262 2263 h6->ip6_vfc |= IPV6_VERSION; 2264 h6->ip6_hlim = IPV6_DEFHLIM; 2265 2266 lwkt_reltoken(&pf_token); 2267 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 2268 lwkt_gettoken(&pf_token); 2269 break; 2270 #endif /* INET6 */ 2271 } 2272 } 2273 2274 void 2275 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 2276 struct pf_rule *r) 2277 { 2278 struct mbuf *m0; 2279 2280 /* 2281 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags, 2282 * so make sure pf.flags is clear. 2283 */ 2284 if ((m0 = m_copy(m, 0, M_COPYALL)) == NULL) 2285 return; 2286 2287 m0->m_pkthdr.fw_flags |= PF_MBUF_TAGGED; 2288 m0->m_pkthdr.pf.flags = 0; 2289 /* XXX Re-Check when Upgrading to > 4.4 */ 2290 m0->m_pkthdr.pf.statekey = NULL; 2291 2292 if (r->rtableid >= 0) 2293 m0->m_pkthdr.pf.rtableid = r->rtableid; 2294 2295 #ifdef ALTQ 2296 if (r->qid) { 2297 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE; 2298 m0->m_pkthdr.pf.qid = r->qid; 2299 m0->m_pkthdr.pf.ecn_af = af; 2300 m0->m_pkthdr.pf.hdr = mtod(m0, struct ip *); 2301 } 2302 #endif /* ALTQ */ 2303 2304 switch (af) { 2305 #ifdef INET 2306 case AF_INET: 2307 icmp_error(m0, type, code, 0, 0); 2308 break; 2309 #endif /* INET */ 2310 #ifdef INET6 2311 case AF_INET6: 2312 icmp6_error(m0, type, code, 0); 2313 break; 2314 #endif /* INET6 */ 2315 } 2316 } 2317 2318 /* 2319 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 2320 * If n is 0, they match if they are equal. If n is != 0, they match if they 2321 * are different. 2322 */ 2323 int 2324 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 2325 struct pf_addr *b, sa_family_t af) 2326 { 2327 int match = 0; 2328 2329 switch (af) { 2330 #ifdef INET 2331 case AF_INET: 2332 if ((a->addr32[0] & m->addr32[0]) == 2333 (b->addr32[0] & m->addr32[0])) 2334 match++; 2335 break; 2336 #endif /* INET */ 2337 #ifdef INET6 2338 case AF_INET6: 2339 if (((a->addr32[0] & m->addr32[0]) == 2340 (b->addr32[0] & m->addr32[0])) && 2341 ((a->addr32[1] & m->addr32[1]) == 2342 (b->addr32[1] & m->addr32[1])) && 2343 ((a->addr32[2] & m->addr32[2]) == 2344 (b->addr32[2] & m->addr32[2])) && 2345 ((a->addr32[3] & m->addr32[3]) == 2346 (b->addr32[3] & m->addr32[3]))) 2347 match++; 2348 break; 2349 #endif /* INET6 */ 2350 } 2351 if (match) { 2352 if (n) 2353 return (0); 2354 else 2355 return (1); 2356 } else { 2357 if (n) 2358 return (1); 2359 else 2360 return (0); 2361 } 2362 } 2363 2364 /* 2365 * Return 1 if b <= a <= e, otherwise return 0. 2366 */ 2367 int 2368 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 2369 struct pf_addr *a, sa_family_t af) 2370 { 2371 switch (af) { 2372 #ifdef INET 2373 case AF_INET: 2374 if ((a->addr32[0] < b->addr32[0]) || 2375 (a->addr32[0] > e->addr32[0])) 2376 return (0); 2377 break; 2378 #endif /* INET */ 2379 #ifdef INET6 2380 case AF_INET6: { 2381 int i; 2382 2383 /* check a >= b */ 2384 for (i = 0; i < 4; ++i) 2385 if (a->addr32[i] > b->addr32[i]) 2386 break; 2387 else if (a->addr32[i] < b->addr32[i]) 2388 return (0); 2389 /* check a <= e */ 2390 for (i = 0; i < 4; ++i) 2391 if (a->addr32[i] < e->addr32[i]) 2392 break; 2393 else if (a->addr32[i] > e->addr32[i]) 2394 return (0); 2395 break; 2396 } 2397 #endif /* INET6 */ 2398 } 2399 return (1); 2400 } 2401 2402 int 2403 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 2404 { 2405 switch (op) { 2406 case PF_OP_IRG: 2407 return ((p > a1) && (p < a2)); 2408 case PF_OP_XRG: 2409 return ((p < a1) || (p > a2)); 2410 case PF_OP_RRG: 2411 return ((p >= a1) && (p <= a2)); 2412 case PF_OP_EQ: 2413 return (p == a1); 2414 case PF_OP_NE: 2415 return (p != a1); 2416 case PF_OP_LT: 2417 return (p < a1); 2418 case PF_OP_LE: 2419 return (p <= a1); 2420 case PF_OP_GT: 2421 return (p > a1); 2422 case PF_OP_GE: 2423 return (p >= a1); 2424 } 2425 return (0); /* never reached */ 2426 } 2427 2428 int 2429 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 2430 { 2431 a1 = ntohs(a1); 2432 a2 = ntohs(a2); 2433 p = ntohs(p); 2434 return (pf_match(op, a1, a2, p)); 2435 } 2436 2437 int 2438 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 2439 { 2440 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2441 return (0); 2442 return (pf_match(op, a1, a2, u)); 2443 } 2444 2445 int 2446 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 2447 { 2448 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2449 return (0); 2450 return (pf_match(op, a1, a2, g)); 2451 } 2452 2453 int 2454 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag) 2455 { 2456 if (*tag == -1) 2457 *tag = m->m_pkthdr.pf.tag; 2458 2459 return ((!r->match_tag_not && r->match_tag == *tag) || 2460 (r->match_tag_not && r->match_tag != *tag)); 2461 } 2462 2463 int 2464 pf_tag_packet(struct mbuf *m, int tag, int rtableid) 2465 { 2466 if (tag <= 0 && rtableid < 0) 2467 return (0); 2468 2469 if (tag > 0) 2470 m->m_pkthdr.pf.tag = tag; 2471 if (rtableid >= 0) 2472 m->m_pkthdr.pf.rtableid = rtableid; 2473 2474 return (0); 2475 } 2476 2477 void 2478 pf_step_into_anchor(int *depth, struct pf_ruleset **rs, int n, 2479 struct pf_rule **r, struct pf_rule **a, int *match) 2480 { 2481 struct pf_anchor_stackframe *f; 2482 2483 (*r)->anchor->match = 0; 2484 if (match) 2485 *match = 0; 2486 if (*depth >= NELEM(pf_anchor_stack)) { 2487 kprintf("pf_step_into_anchor: stack overflow\n"); 2488 *r = TAILQ_NEXT(*r, entries); 2489 return; 2490 } else if (*depth == 0 && a != NULL) 2491 *a = *r; 2492 f = pf_anchor_stack + (*depth)++; 2493 f->rs = *rs; 2494 f->r = *r; 2495 if ((*r)->anchor_wildcard) { 2496 f->parent = &(*r)->anchor->children; 2497 if ((f->child = RB_MIN(pf_anchor_node, f->parent)) == 2498 NULL) { 2499 *r = NULL; 2500 return; 2501 } 2502 *rs = &f->child->ruleset; 2503 } else { 2504 f->parent = NULL; 2505 f->child = NULL; 2506 *rs = &(*r)->anchor->ruleset; 2507 } 2508 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2509 } 2510 2511 int 2512 pf_step_out_of_anchor(int *depth, struct pf_ruleset **rs, int n, 2513 struct pf_rule **r, struct pf_rule **a, int *match) 2514 { 2515 struct pf_anchor_stackframe *f; 2516 int quick = 0; 2517 2518 do { 2519 if (*depth <= 0) 2520 break; 2521 f = pf_anchor_stack + *depth - 1; 2522 if (f->parent != NULL && f->child != NULL) { 2523 if (f->child->match || 2524 (match != NULL && *match)) { 2525 f->r->anchor->match = 1; 2526 *match = 0; 2527 } 2528 f->child = RB_NEXT(pf_anchor_node, f->parent, f->child); 2529 if (f->child != NULL) { 2530 *rs = &f->child->ruleset; 2531 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2532 if (*r == NULL) 2533 continue; 2534 else 2535 break; 2536 } 2537 } 2538 (*depth)--; 2539 if (*depth == 0 && a != NULL) 2540 *a = NULL; 2541 *rs = f->rs; 2542 if (f->r->anchor->match || (match != NULL && *match)) 2543 quick = f->r->quick; 2544 *r = TAILQ_NEXT(f->r, entries); 2545 } while (*r == NULL); 2546 2547 return (quick); 2548 } 2549 2550 #ifdef INET6 2551 void 2552 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 2553 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 2554 { 2555 switch (af) { 2556 #ifdef INET 2557 case AF_INET: 2558 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 2559 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 2560 break; 2561 #endif /* INET */ 2562 case AF_INET6: 2563 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 2564 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 2565 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 2566 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 2567 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 2568 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 2569 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 2570 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 2571 break; 2572 } 2573 } 2574 2575 void 2576 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 2577 { 2578 switch (af) { 2579 #ifdef INET 2580 case AF_INET: 2581 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 2582 break; 2583 #endif /* INET */ 2584 case AF_INET6: 2585 if (addr->addr32[3] == 0xffffffff) { 2586 addr->addr32[3] = 0; 2587 if (addr->addr32[2] == 0xffffffff) { 2588 addr->addr32[2] = 0; 2589 if (addr->addr32[1] == 0xffffffff) { 2590 addr->addr32[1] = 0; 2591 addr->addr32[0] = 2592 htonl(ntohl(addr->addr32[0]) + 1); 2593 } else 2594 addr->addr32[1] = 2595 htonl(ntohl(addr->addr32[1]) + 1); 2596 } else 2597 addr->addr32[2] = 2598 htonl(ntohl(addr->addr32[2]) + 1); 2599 } else 2600 addr->addr32[3] = 2601 htonl(ntohl(addr->addr32[3]) + 1); 2602 break; 2603 } 2604 } 2605 #endif /* INET6 */ 2606 2607 #define mix(a,b,c) \ 2608 do { \ 2609 a -= b; a -= c; a ^= (c >> 13); \ 2610 b -= c; b -= a; b ^= (a << 8); \ 2611 c -= a; c -= b; c ^= (b >> 13); \ 2612 a -= b; a -= c; a ^= (c >> 12); \ 2613 b -= c; b -= a; b ^= (a << 16); \ 2614 c -= a; c -= b; c ^= (b >> 5); \ 2615 a -= b; a -= c; a ^= (c >> 3); \ 2616 b -= c; b -= a; b ^= (a << 10); \ 2617 c -= a; c -= b; c ^= (b >> 15); \ 2618 } while (0) 2619 2620 /* 2621 * hash function based on bridge_hash in if_bridge.c 2622 */ 2623 void 2624 pf_hash(struct pf_addr *inaddr, struct pf_addr *hash, 2625 struct pf_poolhashkey *key, sa_family_t af) 2626 { 2627 u_int32_t a = 0x9e3779b9, b = 0x9e3779b9, c = key->key32[0]; 2628 2629 switch (af) { 2630 #ifdef INET 2631 case AF_INET: 2632 a += inaddr->addr32[0]; 2633 b += key->key32[1]; 2634 mix(a, b, c); 2635 hash->addr32[0] = c + key->key32[2]; 2636 break; 2637 #endif /* INET */ 2638 #ifdef INET6 2639 case AF_INET6: 2640 a += inaddr->addr32[0]; 2641 b += inaddr->addr32[2]; 2642 mix(a, b, c); 2643 hash->addr32[0] = c; 2644 a += inaddr->addr32[1]; 2645 b += inaddr->addr32[3]; 2646 c += key->key32[1]; 2647 mix(a, b, c); 2648 hash->addr32[1] = c; 2649 a += inaddr->addr32[2]; 2650 b += inaddr->addr32[1]; 2651 c += key->key32[2]; 2652 mix(a, b, c); 2653 hash->addr32[2] = c; 2654 a += inaddr->addr32[3]; 2655 b += inaddr->addr32[0]; 2656 c += key->key32[3]; 2657 mix(a, b, c); 2658 hash->addr32[3] = c; 2659 break; 2660 #endif /* INET6 */ 2661 } 2662 } 2663 2664 int 2665 pf_map_addr(sa_family_t af, struct pf_rule *r, struct pf_addr *saddr, 2666 struct pf_addr *naddr, struct pf_addr *init_addr, struct pf_src_node **sn) 2667 { 2668 unsigned char hash[16]; 2669 struct pf_pool *rpool = &r->rpool; 2670 struct pf_pooladdr *acur = rpool->cur; 2671 struct pf_pooladdr *cur; 2672 struct pf_addr *raddr; 2673 struct pf_addr *rmask; 2674 struct pf_addr counter; 2675 struct pf_src_node k; 2676 int cpu = mycpu->gd_cpuid; 2677 int tblidx; 2678 2679 bzero(hash, sizeof(hash)); /* avoid gcc warnings */ 2680 2681 /* 2682 * NOTE! rpool->cur and rpool->tblidx can be iterators and thus 2683 * may represent a SMP race due to the shared nature of the 2684 * rpool structure. We allow the race and ensure that updates 2685 * do not create a fatal condition. 2686 */ 2687 cpu_ccfence(); 2688 cur = acur; 2689 raddr = &cur->addr.v.a.addr; 2690 rmask = &cur->addr.v.a.mask; 2691 2692 if (*sn == NULL && r->rpool.opts & PF_POOL_STICKYADDR && 2693 (r->rpool.opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) { 2694 k.af = af; 2695 PF_ACPY(&k.addr, saddr, af); 2696 if (r->rule_flag & PFRULE_RULESRCTRACK || 2697 r->rpool.opts & PF_POOL_STICKYADDR) 2698 k.rule.ptr = r; 2699 else 2700 k.rule.ptr = NULL; 2701 PF_INC_SCOUNTER(SCNT_SRC_NODE_SEARCH); 2702 *sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k); 2703 if (*sn != NULL && !PF_AZERO(&(*sn)->raddr, af)) { 2704 PF_ACPY(naddr, &(*sn)->raddr, af); 2705 if (pf_status.debug >= PF_DEBUG_MISC) { 2706 kprintf("pf_map_addr: src tracking maps "); 2707 pf_print_host(&k.addr, 0, af); 2708 kprintf(" to "); 2709 pf_print_host(naddr, 0, af); 2710 kprintf("\n"); 2711 } 2712 return (0); 2713 } 2714 } 2715 2716 if (cur->addr.type == PF_ADDR_NOROUTE) 2717 return (1); 2718 if (cur->addr.type == PF_ADDR_DYNIFTL) { 2719 switch (af) { 2720 #ifdef INET 2721 case AF_INET: 2722 if (cur->addr.p.dyn->pfid_acnt4 < 1 && 2723 (rpool->opts & PF_POOL_TYPEMASK) != 2724 PF_POOL_ROUNDROBIN) 2725 return (1); 2726 raddr = &cur->addr.p.dyn->pfid_addr4; 2727 rmask = &cur->addr.p.dyn->pfid_mask4; 2728 break; 2729 #endif /* INET */ 2730 #ifdef INET6 2731 case AF_INET6: 2732 if (cur->addr.p.dyn->pfid_acnt6 < 1 && 2733 (rpool->opts & PF_POOL_TYPEMASK) != 2734 PF_POOL_ROUNDROBIN) 2735 return (1); 2736 raddr = &cur->addr.p.dyn->pfid_addr6; 2737 rmask = &cur->addr.p.dyn->pfid_mask6; 2738 break; 2739 #endif /* INET6 */ 2740 } 2741 } else if (cur->addr.type == PF_ADDR_TABLE) { 2742 if ((rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_ROUNDROBIN) 2743 return (1); /* unsupported */ 2744 } else { 2745 raddr = &cur->addr.v.a.addr; 2746 rmask = &cur->addr.v.a.mask; 2747 } 2748 2749 switch (rpool->opts & PF_POOL_TYPEMASK) { 2750 case PF_POOL_NONE: 2751 PF_ACPY(naddr, raddr, af); 2752 break; 2753 case PF_POOL_BITMASK: 2754 PF_POOLMASK(naddr, raddr, rmask, saddr, af); 2755 break; 2756 case PF_POOL_RANDOM: 2757 if (init_addr != NULL && PF_AZERO(init_addr, af)) { 2758 switch (af) { 2759 #ifdef INET 2760 case AF_INET: 2761 counter.addr32[0] = htonl(karc4random()); 2762 break; 2763 #endif /* INET */ 2764 #ifdef INET6 2765 case AF_INET6: 2766 if (rmask->addr32[3] != 0xffffffff) 2767 counter.addr32[3] = 2768 htonl(karc4random()); 2769 else 2770 break; 2771 if (rmask->addr32[2] != 0xffffffff) 2772 counter.addr32[2] = 2773 htonl(karc4random()); 2774 else 2775 break; 2776 if (rmask->addr32[1] != 0xffffffff) 2777 counter.addr32[1] = 2778 htonl(karc4random()); 2779 else 2780 break; 2781 if (rmask->addr32[0] != 0xffffffff) 2782 counter.addr32[0] = 2783 htonl(karc4random()); 2784 break; 2785 #endif /* INET6 */ 2786 } 2787 PF_POOLMASK(naddr, raddr, rmask, &counter, af); 2788 PF_ACPY(init_addr, naddr, af); 2789 2790 } else { 2791 counter = rpool->counter; 2792 cpu_ccfence(); 2793 PF_AINC(&counter, af); 2794 PF_POOLMASK(naddr, raddr, rmask, &counter, af); 2795 rpool->counter = counter; 2796 } 2797 break; 2798 case PF_POOL_SRCHASH: 2799 pf_hash(saddr, (struct pf_addr *)&hash, &rpool->key, af); 2800 PF_POOLMASK(naddr, raddr, rmask, (struct pf_addr *)&hash, af); 2801 break; 2802 case PF_POOL_ROUNDROBIN: 2803 tblidx = rpool->tblidx; 2804 counter = rpool->counter; 2805 if (cur->addr.type == PF_ADDR_TABLE) { 2806 if (!pfr_pool_get(cur->addr.p.tbl, 2807 &tblidx, &counter, 2808 &raddr, &rmask, af)) { 2809 goto get_addr; 2810 } 2811 } else if (cur->addr.type == PF_ADDR_DYNIFTL) { 2812 if (!pfr_pool_get(cur->addr.p.dyn->pfid_kt, 2813 &tblidx, &counter, 2814 &raddr, &rmask, af)) { 2815 goto get_addr; 2816 } 2817 } else if (pf_match_addr(0, raddr, rmask, 2818 &counter, af)) { 2819 goto get_addr; 2820 } 2821 2822 try_next: 2823 if ((cur = TAILQ_NEXT(cur, entries)) == NULL) 2824 cur = TAILQ_FIRST(&rpool->list); 2825 if (cur->addr.type == PF_ADDR_TABLE) { 2826 tblidx = -1; 2827 if (pfr_pool_get(cur->addr.p.tbl, 2828 &tblidx, &counter, 2829 &raddr, &rmask, af)) { 2830 /* table contains no address of type 'af' */ 2831 if (cur != acur) 2832 goto try_next; 2833 return (1); 2834 } 2835 } else if (cur->addr.type == PF_ADDR_DYNIFTL) { 2836 tblidx = -1; 2837 if (pfr_pool_get(cur->addr.p.dyn->pfid_kt, 2838 &tblidx, &counter, 2839 &raddr, &rmask, af)) { 2840 /* table contains no address of type 'af' */ 2841 if (cur != acur) 2842 goto try_next; 2843 return (1); 2844 } 2845 } else { 2846 raddr = &cur->addr.v.a.addr; 2847 rmask = &cur->addr.v.a.mask; 2848 PF_ACPY(&counter, raddr, af); 2849 } 2850 2851 get_addr: 2852 rpool->cur = cur; 2853 rpool->tblidx = tblidx; 2854 PF_ACPY(naddr, &counter, af); 2855 if (init_addr != NULL && PF_AZERO(init_addr, af)) 2856 PF_ACPY(init_addr, naddr, af); 2857 PF_AINC(&counter, af); 2858 rpool->counter = counter; 2859 break; 2860 } 2861 if (*sn != NULL) 2862 PF_ACPY(&(*sn)->raddr, naddr, af); 2863 2864 if (pf_status.debug >= PF_DEBUG_MISC && 2865 (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) { 2866 kprintf("pf_map_addr: selected address "); 2867 pf_print_host(naddr, 0, af); 2868 kprintf("\n"); 2869 } 2870 2871 return (0); 2872 } 2873 2874 int 2875 pf_get_sport(struct pf_pdesc *pd, sa_family_t af, 2876 u_int8_t proto, struct pf_rule *r, 2877 struct pf_addr *saddr, struct pf_addr *daddr, 2878 u_int16_t sport, u_int16_t dport, 2879 struct pf_addr *naddr, u_int16_t *nport, 2880 u_int16_t low, u_int16_t high, struct pf_src_node **sn) 2881 { 2882 struct pf_state_key_cmp key; 2883 struct pf_addr init_addr; 2884 u_int16_t cut; 2885 u_int32_t hash_base = 0; 2886 int do_hash = 0; 2887 2888 bzero(&init_addr, sizeof(init_addr)); 2889 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn)) 2890 return (1); 2891 2892 if (proto == IPPROTO_ICMP) { 2893 low = 1; 2894 high = 65535; 2895 } 2896 2897 bzero(&key, sizeof(key)); 2898 key.af = af; 2899 key.proto = proto; 2900 key.port[0] = dport; 2901 PF_ACPY(&key.addr[0], daddr, key.af); 2902 2903 do { 2904 PF_ACPY(&key.addr[1], naddr, key.af); 2905 2906 /* 2907 * We want to select a port that calculates to a toeplitz hash 2908 * that masks to the same cpu, otherwise the response may 2909 * not see the new state. 2910 * 2911 * We can still do this even if the kernel is disregarding 2912 * the hash and vectoring the packets to a specific cpu, 2913 * but it will reduce the number of ports we can use. 2914 */ 2915 switch(af) { 2916 case AF_INET: 2917 if (proto == IPPROTO_TCP) { 2918 do_hash = 1; 2919 hash_base = toeplitz_piecemeal_port(dport) ^ 2920 toeplitz_piecemeal_addr(daddr->v4.s_addr) ^ 2921 toeplitz_piecemeal_addr(naddr->v4.s_addr); 2922 } 2923 break; 2924 case AF_INET6: 2925 /* XXX TODO XXX */ 2926 default: 2927 /* XXX TODO XXX */ 2928 break; 2929 } 2930 2931 /* 2932 * port search; start random, step; 2933 * similar 2 portloop in in_pcbbind 2934 * 2935 * WARNING! We try to match such that the kernel will 2936 * dispatch the translated host/port to the same 2937 * cpu, but this might not be possible. 2938 * 2939 * In the case where the port is fixed, or for the 2940 * UDP case (whos toeplitz does not incorporate the 2941 * port), we set not_cpu_localized which ultimately 2942 * causes the pf_state_tree element 2943 * 2944 * XXX fixed ports present a problem for cpu localization. 2945 */ 2946 if (!(proto == IPPROTO_TCP || 2947 proto == IPPROTO_UDP || 2948 proto == IPPROTO_ICMP)) { 2949 /* 2950 * non-specific protocol, leave port intact. 2951 */ 2952 key.port[1] = sport; 2953 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) { 2954 *nport = sport; 2955 pd->not_cpu_localized = 1; 2956 return (0); 2957 } 2958 } else if (low == 0 && high == 0) { 2959 /* 2960 * static-port same as originator. 2961 */ 2962 key.port[1] = sport; 2963 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) { 2964 *nport = sport; 2965 pd->not_cpu_localized = 1; 2966 return (0); 2967 } 2968 } else if (low == high) { 2969 /* 2970 * specific port as specified. 2971 */ 2972 key.port[1] = htons(low); 2973 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) { 2974 *nport = htons(low); 2975 pd->not_cpu_localized = 1; 2976 return (0); 2977 } 2978 } else { 2979 /* 2980 * normal dynamic port 2981 */ 2982 u_int16_t tmp; 2983 2984 if (low > high) { 2985 tmp = low; 2986 low = high; 2987 high = tmp; 2988 } 2989 /* low < high */ 2990 cut = htonl(karc4random()) % (1 + high - low) + low; 2991 /* low <= cut <= high */ 2992 for (tmp = cut; tmp <= high; ++(tmp)) { 2993 key.port[1] = htons(tmp); 2994 if (do_hash) { 2995 uint32_t hash; 2996 2997 hash = hash_base ^ 2998 toeplitz_piecemeal_port(key.port[1]); 2999 if (netisr_hashcpu(hash) != mycpuid) 3000 continue; 3001 } 3002 if (pf_find_state_all(&key, PF_IN, NULL) == 3003 NULL && !in_baddynamic(tmp, proto)) { 3004 if (proto == IPPROTO_UDP) 3005 pd->not_cpu_localized = 1; 3006 *nport = htons(tmp); 3007 return (0); 3008 } 3009 } 3010 for (tmp = cut - 1; tmp >= low; --(tmp)) { 3011 key.port[1] = htons(tmp); 3012 if (do_hash) { 3013 uint32_t hash; 3014 3015 hash = hash_base ^ 3016 toeplitz_piecemeal_port(key.port[1]); 3017 if (netisr_hashcpu(hash) != mycpuid) 3018 continue; 3019 } 3020 if (pf_find_state_all(&key, PF_IN, NULL) == 3021 NULL && !in_baddynamic(tmp, proto)) { 3022 if (proto == IPPROTO_UDP) 3023 pd->not_cpu_localized = 1; 3024 *nport = htons(tmp); 3025 return (0); 3026 } 3027 } 3028 } 3029 3030 /* 3031 * Next address 3032 */ 3033 switch (r->rpool.opts & PF_POOL_TYPEMASK) { 3034 case PF_POOL_RANDOM: 3035 case PF_POOL_ROUNDROBIN: 3036 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn)) 3037 return (1); 3038 break; 3039 case PF_POOL_NONE: 3040 case PF_POOL_SRCHASH: 3041 case PF_POOL_BITMASK: 3042 default: 3043 return (1); 3044 } 3045 } while (! PF_AEQ(&init_addr, naddr, af) ); 3046 return (1); /* none available */ 3047 } 3048 3049 struct pf_rule * 3050 pf_match_translation(struct pf_pdesc *pd, struct mbuf *m, int off, 3051 int direction, struct pfi_kif *kif, struct pf_addr *saddr, u_int16_t sport, 3052 struct pf_addr *daddr, u_int16_t dport, int rs_num) 3053 { 3054 struct pf_rule *r, *rm = NULL; 3055 struct pf_ruleset *ruleset = NULL; 3056 int tag = -1; 3057 int rtableid = -1; 3058 int asd = 0; 3059 3060 r = TAILQ_FIRST(pf_main_ruleset.rules[rs_num].active.ptr); 3061 while (r && rm == NULL) { 3062 struct pf_rule_addr *src = NULL, *dst = NULL; 3063 struct pf_addr_wrap *xdst = NULL; 3064 struct pf_pooladdr *cur; 3065 3066 if (r->action == PF_BINAT && direction == PF_IN) { 3067 src = &r->dst; 3068 cur = r->rpool.cur; /* SMP race possible */ 3069 cpu_ccfence(); 3070 if (cur) 3071 xdst = &cur->addr; 3072 } else { 3073 src = &r->src; 3074 dst = &r->dst; 3075 } 3076 3077 r->evaluations++; 3078 if (pfi_kif_match(r->kif, kif) == r->ifnot) 3079 r = r->skip[PF_SKIP_IFP].ptr; 3080 else if (r->direction && r->direction != direction) 3081 r = r->skip[PF_SKIP_DIR].ptr; 3082 else if (r->af && r->af != pd->af) 3083 r = r->skip[PF_SKIP_AF].ptr; 3084 else if (r->proto && r->proto != pd->proto) 3085 r = r->skip[PF_SKIP_PROTO].ptr; 3086 else if (PF_MISMATCHAW(&src->addr, saddr, pd->af, 3087 src->neg, kif)) 3088 r = r->skip[src == &r->src ? PF_SKIP_SRC_ADDR : 3089 PF_SKIP_DST_ADDR].ptr; 3090 else if (src->port_op && !pf_match_port(src->port_op, 3091 src->port[0], src->port[1], sport)) 3092 r = r->skip[src == &r->src ? PF_SKIP_SRC_PORT : 3093 PF_SKIP_DST_PORT].ptr; 3094 else if (dst != NULL && 3095 PF_MISMATCHAW(&dst->addr, daddr, pd->af, dst->neg, NULL)) 3096 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3097 else if (xdst != NULL && PF_MISMATCHAW(xdst, daddr, pd->af, 3098 0, NULL)) 3099 r = TAILQ_NEXT(r, entries); 3100 else if (dst != NULL && dst->port_op && 3101 !pf_match_port(dst->port_op, dst->port[0], 3102 dst->port[1], dport)) 3103 r = r->skip[PF_SKIP_DST_PORT].ptr; 3104 else if (r->match_tag && !pf_match_tag(m, r, &tag)) 3105 r = TAILQ_NEXT(r, entries); 3106 else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto != 3107 IPPROTO_TCP || !pf_osfp_match(pf_osfp_fingerprint(pd, m, 3108 off, pd->hdr.tcp), r->os_fingerprint))) 3109 r = TAILQ_NEXT(r, entries); 3110 else { 3111 if (r->tag) 3112 tag = r->tag; 3113 if (r->rtableid >= 0) 3114 rtableid = r->rtableid; 3115 if (r->anchor == NULL) { 3116 rm = r; 3117 } else 3118 pf_step_into_anchor(&asd, &ruleset, rs_num, 3119 &r, NULL, NULL); 3120 } 3121 if (r == NULL) 3122 pf_step_out_of_anchor(&asd, &ruleset, rs_num, &r, 3123 NULL, NULL); 3124 } 3125 if (pf_tag_packet(m, tag, rtableid)) 3126 return (NULL); 3127 if (rm != NULL && (rm->action == PF_NONAT || 3128 rm->action == PF_NORDR || rm->action == PF_NOBINAT)) 3129 return (NULL); 3130 return (rm); 3131 } 3132 3133 struct pf_rule * 3134 pf_get_translation(struct pf_pdesc *pd, struct mbuf *m, int off, int direction, 3135 struct pfi_kif *kif, struct pf_src_node **sn, 3136 struct pf_state_key **skw, struct pf_state_key **sks, 3137 struct pf_state_key **skp, struct pf_state_key **nkp, 3138 struct pf_addr *saddr, struct pf_addr *daddr, 3139 u_int16_t sport, u_int16_t dport) 3140 { 3141 struct pf_rule *r = NULL; 3142 3143 if (direction == PF_OUT) { 3144 r = pf_match_translation(pd, m, off, direction, kif, saddr, 3145 sport, daddr, dport, PF_RULESET_BINAT); 3146 if (r == NULL) 3147 r = pf_match_translation(pd, m, off, direction, kif, 3148 saddr, sport, daddr, dport, PF_RULESET_NAT); 3149 } else { 3150 r = pf_match_translation(pd, m, off, direction, kif, saddr, 3151 sport, daddr, dport, PF_RULESET_RDR); 3152 if (r == NULL) 3153 r = pf_match_translation(pd, m, off, direction, kif, 3154 saddr, sport, daddr, dport, PF_RULESET_BINAT); 3155 } 3156 3157 if (r != NULL) { 3158 struct pf_addr *naddr; 3159 u_int16_t *nport; 3160 3161 if (pf_state_key_setup(pd, r, skw, sks, skp, nkp, 3162 saddr, daddr, sport, dport)) 3163 return r; 3164 3165 /* XXX We only modify one side for now. */ 3166 naddr = &(*nkp)->addr[1]; 3167 nport = &(*nkp)->port[1]; 3168 3169 /* 3170 * NOTE: Currently all translations will clear 3171 * BRIDGE_MBUF_TAGGED, telling the bridge to 3172 * ignore the original input encapsulation. 3173 */ 3174 switch (r->action) { 3175 case PF_NONAT: 3176 case PF_NOBINAT: 3177 case PF_NORDR: 3178 return (NULL); 3179 case PF_NAT: 3180 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 3181 if (pf_get_sport(pd, pd->af, pd->proto, r, 3182 saddr, daddr, sport, dport, 3183 naddr, nport, r->rpool.proxy_port[0], 3184 r->rpool.proxy_port[1], sn)) { 3185 DPFPRINTF(PF_DEBUG_MISC, 3186 ("pf: NAT proxy port allocation " 3187 "(%u-%u) failed\n", 3188 r->rpool.proxy_port[0], 3189 r->rpool.proxy_port[1])); 3190 return (NULL); 3191 } 3192 break; 3193 case PF_BINAT: 3194 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 3195 switch (direction) { 3196 case PF_OUT: 3197 if (r->rpool.cur->addr.type == PF_ADDR_DYNIFTL){ 3198 switch (pd->af) { 3199 #ifdef INET 3200 case AF_INET: 3201 if (r->rpool.cur->addr.p.dyn-> 3202 pfid_acnt4 < 1) 3203 return (NULL); 3204 PF_POOLMASK(naddr, 3205 &r->rpool.cur->addr.p.dyn-> 3206 pfid_addr4, 3207 &r->rpool.cur->addr.p.dyn-> 3208 pfid_mask4, 3209 saddr, AF_INET); 3210 break; 3211 #endif /* INET */ 3212 #ifdef INET6 3213 case AF_INET6: 3214 if (r->rpool.cur->addr.p.dyn-> 3215 pfid_acnt6 < 1) 3216 return (NULL); 3217 PF_POOLMASK(naddr, 3218 &r->rpool.cur->addr.p.dyn-> 3219 pfid_addr6, 3220 &r->rpool.cur->addr.p.dyn-> 3221 pfid_mask6, 3222 saddr, AF_INET6); 3223 break; 3224 #endif /* INET6 */ 3225 } 3226 } else 3227 PF_POOLMASK(naddr, 3228 &r->rpool.cur->addr.v.a.addr, 3229 &r->rpool.cur->addr.v.a.mask, 3230 saddr, pd->af); 3231 break; 3232 case PF_IN: 3233 if (r->src.addr.type == PF_ADDR_DYNIFTL) { 3234 switch (pd->af) { 3235 #ifdef INET 3236 case AF_INET: 3237 if (r->src.addr.p.dyn-> 3238 pfid_acnt4 < 1) 3239 return (NULL); 3240 PF_POOLMASK(naddr, 3241 &r->src.addr.p.dyn-> 3242 pfid_addr4, 3243 &r->src.addr.p.dyn-> 3244 pfid_mask4, 3245 daddr, AF_INET); 3246 break; 3247 #endif /* INET */ 3248 #ifdef INET6 3249 case AF_INET6: 3250 if (r->src.addr.p.dyn-> 3251 pfid_acnt6 < 1) 3252 return (NULL); 3253 PF_POOLMASK(naddr, 3254 &r->src.addr.p.dyn-> 3255 pfid_addr6, 3256 &r->src.addr.p.dyn-> 3257 pfid_mask6, 3258 daddr, AF_INET6); 3259 break; 3260 #endif /* INET6 */ 3261 } 3262 } else 3263 PF_POOLMASK(naddr, 3264 &r->src.addr.v.a.addr, 3265 &r->src.addr.v.a.mask, daddr, 3266 pd->af); 3267 break; 3268 } 3269 break; 3270 case PF_RDR: { 3271 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 3272 if (pf_map_addr(pd->af, r, saddr, naddr, NULL, sn)) 3273 return (NULL); 3274 if ((r->rpool.opts & PF_POOL_TYPEMASK) == 3275 PF_POOL_BITMASK) 3276 PF_POOLMASK(naddr, naddr, 3277 &r->rpool.cur->addr.v.a.mask, daddr, 3278 pd->af); 3279 3280 if (r->rpool.proxy_port[1]) { 3281 u_int32_t tmp_nport; 3282 3283 tmp_nport = ((ntohs(dport) - 3284 ntohs(r->dst.port[0])) % 3285 (r->rpool.proxy_port[1] - 3286 r->rpool.proxy_port[0] + 1)) + 3287 r->rpool.proxy_port[0]; 3288 3289 /* wrap around if necessary */ 3290 if (tmp_nport > 65535) 3291 tmp_nport -= 65535; 3292 *nport = htons((u_int16_t)tmp_nport); 3293 } else if (r->rpool.proxy_port[0]) { 3294 *nport = htons(r->rpool.proxy_port[0]); 3295 } 3296 pd->not_cpu_localized = 1; 3297 break; 3298 } 3299 default: 3300 return (NULL); 3301 } 3302 } 3303 3304 return (r); 3305 } 3306 3307 struct netmsg_hashlookup { 3308 struct netmsg_base base; 3309 struct inpcb **nm_pinp; 3310 struct inpcbinfo *nm_pcbinfo; 3311 struct pf_addr *nm_saddr; 3312 struct pf_addr *nm_daddr; 3313 uint16_t nm_sport; 3314 uint16_t nm_dport; 3315 sa_family_t nm_af; 3316 }; 3317 3318 #ifdef PF_SOCKET_LOOKUP_DOMSG 3319 static void 3320 in_pcblookup_hash_handler(netmsg_t msg) 3321 { 3322 struct netmsg_hashlookup *rmsg = (struct netmsg_hashlookup *)msg; 3323 3324 if (rmsg->nm_af == AF_INET) 3325 *rmsg->nm_pinp = in_pcblookup_hash(rmsg->nm_pcbinfo, 3326 rmsg->nm_saddr->v4, rmsg->nm_sport, rmsg->nm_daddr->v4, 3327 rmsg->nm_dport, INPLOOKUP_WILDCARD, NULL); 3328 #ifdef INET6 3329 else 3330 *rmsg->nm_pinp = in6_pcblookup_hash(rmsg->nm_pcbinfo, 3331 &rmsg->nm_saddr->v6, rmsg->nm_sport, &rmsg->nm_daddr->v6, 3332 rmsg->nm_dport, INPLOOKUP_WILDCARD, NULL); 3333 #endif /* INET6 */ 3334 lwkt_replymsg(&rmsg->base.lmsg, 0); 3335 } 3336 #endif /* PF_SOCKET_LOOKUP_DOMSG */ 3337 3338 int 3339 pf_socket_lookup(int direction, struct pf_pdesc *pd) 3340 { 3341 struct pf_addr *saddr, *daddr; 3342 u_int16_t sport, dport; 3343 struct inpcbinfo *pi; 3344 struct inpcb *inp; 3345 struct netmsg_hashlookup *msg = NULL; 3346 #ifdef PF_SOCKET_LOOKUP_DOMSG 3347 struct netmsg_hashlookup msg0; 3348 #endif 3349 int pi_cpu = 0; 3350 3351 if (pd == NULL) 3352 return (-1); 3353 pd->lookup.uid = UID_MAX; 3354 pd->lookup.gid = GID_MAX; 3355 pd->lookup.pid = NO_PID; 3356 if (direction == PF_IN) { 3357 saddr = pd->src; 3358 daddr = pd->dst; 3359 } else { 3360 saddr = pd->dst; 3361 daddr = pd->src; 3362 } 3363 switch (pd->proto) { 3364 case IPPROTO_TCP: 3365 if (pd->hdr.tcp == NULL) 3366 return (-1); 3367 sport = pd->hdr.tcp->th_sport; 3368 dport = pd->hdr.tcp->th_dport; 3369 3370 pi_cpu = tcp_addrcpu(saddr->v4.s_addr, sport, daddr->v4.s_addr, dport); 3371 pi = &tcbinfo[pi_cpu]; 3372 /* 3373 * Our netstack runs lockless on MP systems 3374 * (only for TCP connections at the moment). 3375 * 3376 * As we are not allowed to read another CPU's tcbinfo, 3377 * we have to ask that CPU via remote call to search the 3378 * table for us. 3379 * 3380 * Prepare a msg iff data belongs to another CPU. 3381 */ 3382 if (pi_cpu != mycpu->gd_cpuid) { 3383 #ifdef PF_SOCKET_LOOKUP_DOMSG 3384 /* 3385 * NOTE: 3386 * 3387 * Following lwkt_domsg() is dangerous and could 3388 * lockup the network system, e.g. 3389 * 3390 * On 2 CPU system: 3391 * netisr0 domsg to netisr1 (due to lookup) 3392 * netisr1 domsg to netisr0 (due to lookup) 3393 * 3394 * We simply return -1 here, since we are probably 3395 * called before NAT, so the TCP packet should 3396 * already be on the correct CPU. 3397 */ 3398 msg = &msg0; 3399 netmsg_init(&msg->base, NULL, &curthread->td_msgport, 3400 0, in_pcblookup_hash_handler); 3401 msg->nm_pinp = &inp; 3402 msg->nm_pcbinfo = pi; 3403 msg->nm_saddr = saddr; 3404 msg->nm_sport = sport; 3405 msg->nm_daddr = daddr; 3406 msg->nm_dport = dport; 3407 msg->nm_af = pd->af; 3408 #else /* !PF_SOCKET_LOOKUP_DOMSG */ 3409 kprintf("pf_socket_lookup: tcp packet not on the " 3410 "correct cpu %d, cur cpu %d\n", 3411 pi_cpu, mycpuid); 3412 print_backtrace(-1); 3413 return -1; 3414 #endif /* PF_SOCKET_LOOKUP_DOMSG */ 3415 } 3416 break; 3417 case IPPROTO_UDP: 3418 if (pd->hdr.udp == NULL) 3419 return (-1); 3420 sport = pd->hdr.udp->uh_sport; 3421 dport = pd->hdr.udp->uh_dport; 3422 pi = &udbinfo[mycpuid]; 3423 break; 3424 default: 3425 return (-1); 3426 } 3427 if (direction != PF_IN) { 3428 u_int16_t p; 3429 3430 p = sport; 3431 sport = dport; 3432 dport = p; 3433 } 3434 switch (pd->af) { 3435 #ifdef INET6 3436 case AF_INET6: 3437 /* 3438 * Query other CPU, second part 3439 * 3440 * msg only gets initialized when: 3441 * 1) packet is TCP 3442 * 2) the info belongs to another CPU 3443 * 3444 * Use some switch/case magic to avoid code duplication. 3445 */ 3446 if (msg == NULL) { 3447 inp = in6_pcblookup_hash(pi, &saddr->v6, sport, 3448 &daddr->v6, dport, INPLOOKUP_WILDCARD, NULL); 3449 3450 if (inp == NULL) 3451 return (-1); 3452 break; 3453 } 3454 /* FALLTHROUGH if SMP and on other CPU */ 3455 #endif /* INET6 */ 3456 case AF_INET: 3457 if (msg != NULL) { 3458 lwkt_domsg(netisr_cpuport(pi_cpu), 3459 &msg->base.lmsg, 0); 3460 } else 3461 { 3462 inp = in_pcblookup_hash(pi, saddr->v4, sport, daddr->v4, 3463 dport, INPLOOKUP_WILDCARD, NULL); 3464 } 3465 if (inp == NULL) 3466 return (-1); 3467 break; 3468 3469 default: 3470 return (-1); 3471 } 3472 pd->lookup.uid = inp->inp_socket->so_cred->cr_uid; 3473 pd->lookup.gid = inp->inp_socket->so_cred->cr_groups[0]; 3474 return (1); 3475 } 3476 3477 u_int8_t 3478 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3479 { 3480 int hlen; 3481 u_int8_t hdr[60]; 3482 u_int8_t *opt, optlen; 3483 u_int8_t wscale = 0; 3484 3485 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3486 if (hlen <= sizeof(struct tcphdr)) 3487 return (0); 3488 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3489 return (0); 3490 opt = hdr + sizeof(struct tcphdr); 3491 hlen -= sizeof(struct tcphdr); 3492 while (hlen >= 3) { 3493 switch (*opt) { 3494 case TCPOPT_EOL: 3495 case TCPOPT_NOP: 3496 ++opt; 3497 --hlen; 3498 break; 3499 case TCPOPT_WINDOW: 3500 wscale = opt[2]; 3501 if (wscale > TCP_MAX_WINSHIFT) 3502 wscale = TCP_MAX_WINSHIFT; 3503 wscale |= PF_WSCALE_FLAG; 3504 /* FALLTHROUGH */ 3505 default: 3506 optlen = opt[1]; 3507 if (optlen < 2) 3508 optlen = 2; 3509 hlen -= optlen; 3510 opt += optlen; 3511 break; 3512 } 3513 } 3514 return (wscale); 3515 } 3516 3517 u_int16_t 3518 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3519 { 3520 int hlen; 3521 u_int8_t hdr[60]; 3522 u_int8_t *opt, optlen; 3523 u_int16_t mss = tcp_mssdflt; 3524 3525 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3526 if (hlen <= sizeof(struct tcphdr)) 3527 return (0); 3528 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3529 return (0); 3530 opt = hdr + sizeof(struct tcphdr); 3531 hlen -= sizeof(struct tcphdr); 3532 while (hlen >= TCPOLEN_MAXSEG) { 3533 switch (*opt) { 3534 case TCPOPT_EOL: 3535 case TCPOPT_NOP: 3536 ++opt; 3537 --hlen; 3538 break; 3539 case TCPOPT_MAXSEG: 3540 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 3541 /* FALLTHROUGH */ 3542 default: 3543 optlen = opt[1]; 3544 if (optlen < 2) 3545 optlen = 2; 3546 hlen -= optlen; 3547 opt += optlen; 3548 break; 3549 } 3550 } 3551 return (mss); 3552 } 3553 3554 u_int16_t 3555 pf_calc_mss(struct pf_addr *addr, sa_family_t af, u_int16_t offer) 3556 { 3557 #ifdef INET 3558 struct sockaddr_in *dst; 3559 struct route ro; 3560 #endif /* INET */ 3561 #ifdef INET6 3562 struct sockaddr_in6 *dst6; 3563 struct route_in6 ro6; 3564 #endif /* INET6 */ 3565 struct rtentry *rt = NULL; 3566 int hlen = 0; 3567 u_int16_t mss = tcp_mssdflt; 3568 3569 switch (af) { 3570 #ifdef INET 3571 case AF_INET: 3572 hlen = sizeof(struct ip); 3573 bzero(&ro, sizeof(ro)); 3574 dst = (struct sockaddr_in *)&ro.ro_dst; 3575 dst->sin_family = AF_INET; 3576 dst->sin_len = sizeof(*dst); 3577 dst->sin_addr = addr->v4; 3578 rtalloc_ign(&ro, (RTF_CLONING | RTF_PRCLONING)); 3579 rt = ro.ro_rt; 3580 break; 3581 #endif /* INET */ 3582 #ifdef INET6 3583 case AF_INET6: 3584 hlen = sizeof(struct ip6_hdr); 3585 bzero(&ro6, sizeof(ro6)); 3586 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst; 3587 dst6->sin6_family = AF_INET6; 3588 dst6->sin6_len = sizeof(*dst6); 3589 dst6->sin6_addr = addr->v6; 3590 rtalloc_ign((struct route *)&ro6, (RTF_CLONING | RTF_PRCLONING)); 3591 rt = ro6.ro_rt; 3592 break; 3593 #endif /* INET6 */ 3594 } 3595 3596 if (rt && rt->rt_ifp) { 3597 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr); 3598 mss = max(tcp_mssdflt, mss); 3599 RTFREE(rt); 3600 } 3601 mss = min(mss, offer); 3602 mss = max(mss, 64); /* sanity - at least max opt space */ 3603 return (mss); 3604 } 3605 3606 void 3607 pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr) 3608 { 3609 struct pf_rule *r = s->rule.ptr; 3610 3611 s->rt_kif = NULL; 3612 if (!r->rt || r->rt == PF_FASTROUTE) 3613 return; 3614 switch (s->key[PF_SK_WIRE]->af) { 3615 #ifdef INET 3616 case AF_INET: 3617 pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL, 3618 &s->nat_src_node); 3619 s->rt_kif = r->rpool.cur->kif; 3620 break; 3621 #endif /* INET */ 3622 #ifdef INET6 3623 case AF_INET6: 3624 pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL, 3625 &s->nat_src_node); 3626 s->rt_kif = r->rpool.cur->kif; 3627 break; 3628 #endif /* INET6 */ 3629 } 3630 } 3631 3632 u_int32_t 3633 pf_tcp_iss(struct pf_pdesc *pd) 3634 { 3635 MD5_CTX ctx; 3636 u_int32_t digest[4]; 3637 3638 if (pf_tcp_secret_init == 0) { 3639 lwkt_gettoken(&pf_gtoken); 3640 if (pf_tcp_secret_init == 0) { 3641 karc4rand(pf_tcp_secret, sizeof(pf_tcp_secret)); 3642 MD5Init(&pf_tcp_secret_ctx); 3643 MD5Update(&pf_tcp_secret_ctx, pf_tcp_secret, 3644 sizeof(pf_tcp_secret)); 3645 pf_tcp_secret_init = 1; 3646 } 3647 lwkt_reltoken(&pf_gtoken); 3648 } 3649 ctx = pf_tcp_secret_ctx; 3650 3651 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short)); 3652 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short)); 3653 if (pd->af == AF_INET6) { 3654 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 3655 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 3656 } else { 3657 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 3658 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 3659 } 3660 MD5Final((u_char *)digest, &ctx); 3661 pf_tcp_iss_off += 4096; 3662 3663 return (digest[0] + pd->hdr.tcp->th_seq + pf_tcp_iss_off); 3664 } 3665 3666 int 3667 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction, 3668 struct pfi_kif *kif, struct mbuf *m, int off, void *h, 3669 struct pf_pdesc *pd, struct pf_rule **am, struct pf_ruleset **rsm, 3670 struct ifqueue *ifq, struct inpcb *inp) 3671 { 3672 struct pf_rule *nr = NULL; 3673 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 3674 sa_family_t af = pd->af; 3675 struct pf_rule *r, *a = NULL; 3676 struct pf_ruleset *ruleset = NULL; 3677 struct pf_src_node *nsn = NULL; 3678 struct tcphdr *th = pd->hdr.tcp; 3679 struct pf_state_key *skw = NULL, *sks = NULL; 3680 struct pf_state_key *sk = NULL, *nk = NULL; 3681 u_short reason; 3682 int rewrite = 0, hdrlen = 0; 3683 int tag = -1, rtableid = -1; 3684 int asd = 0; 3685 int match = 0; 3686 int state_icmp = 0; 3687 u_int16_t sport = 0, dport = 0; 3688 u_int16_t bproto_sum = 0, bip_sum = 0; 3689 u_int8_t icmptype = 0, icmpcode = 0; 3690 3691 3692 if (direction == PF_IN && pf_check_congestion(ifq)) { 3693 REASON_SET(&reason, PFRES_CONGEST); 3694 return (PF_DROP); 3695 } 3696 3697 if (inp != NULL) 3698 pd->lookup.done = pf_socket_lookup(direction, pd); 3699 else if (debug_pfugidhack) { 3700 DPFPRINTF(PF_DEBUG_MISC, ("pf: unlocked lookup\n")); 3701 pd->lookup.done = pf_socket_lookup(direction, pd); 3702 } 3703 3704 switch (pd->proto) { 3705 case IPPROTO_TCP: 3706 sport = th->th_sport; 3707 dport = th->th_dport; 3708 hdrlen = sizeof(*th); 3709 break; 3710 case IPPROTO_UDP: 3711 sport = pd->hdr.udp->uh_sport; 3712 dport = pd->hdr.udp->uh_dport; 3713 hdrlen = sizeof(*pd->hdr.udp); 3714 break; 3715 #ifdef INET 3716 case IPPROTO_ICMP: 3717 if (pd->af != AF_INET) 3718 break; 3719 sport = dport = pd->hdr.icmp->icmp_id; 3720 hdrlen = sizeof(*pd->hdr.icmp); 3721 icmptype = pd->hdr.icmp->icmp_type; 3722 icmpcode = pd->hdr.icmp->icmp_code; 3723 3724 if (icmptype == ICMP_UNREACH || 3725 icmptype == ICMP_SOURCEQUENCH || 3726 icmptype == ICMP_REDIRECT || 3727 icmptype == ICMP_TIMXCEED || 3728 icmptype == ICMP_PARAMPROB) 3729 state_icmp++; 3730 break; 3731 #endif /* INET */ 3732 #ifdef INET6 3733 case IPPROTO_ICMPV6: 3734 if (af != AF_INET6) 3735 break; 3736 sport = dport = pd->hdr.icmp6->icmp6_id; 3737 hdrlen = sizeof(*pd->hdr.icmp6); 3738 icmptype = pd->hdr.icmp6->icmp6_type; 3739 icmpcode = pd->hdr.icmp6->icmp6_code; 3740 3741 if (icmptype == ICMP6_DST_UNREACH || 3742 icmptype == ICMP6_PACKET_TOO_BIG || 3743 icmptype == ICMP6_TIME_EXCEEDED || 3744 icmptype == ICMP6_PARAM_PROB) 3745 state_icmp++; 3746 break; 3747 #endif /* INET6 */ 3748 default: 3749 sport = dport = hdrlen = 0; 3750 break; 3751 } 3752 3753 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3754 3755 /* check packet for BINAT/NAT/RDR */ 3756 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, 3757 &skw, &sks, &sk, &nk, saddr, daddr, sport, dport)) != NULL) { 3758 if (nk == NULL || sk == NULL) { 3759 REASON_SET(&reason, PFRES_MEMORY); 3760 goto cleanup; 3761 } 3762 3763 if (pd->ip_sum) 3764 bip_sum = *pd->ip_sum; 3765 3766 m->m_flags &= ~M_HASH; 3767 switch (pd->proto) { 3768 case IPPROTO_TCP: 3769 bproto_sum = th->th_sum; 3770 pd->proto_sum = &th->th_sum; 3771 3772 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3773 nk->port[pd->sidx] != sport) { 3774 pf_change_ap(saddr, &th->th_sport, pd->ip_sum, 3775 &th->th_sum, &nk->addr[pd->sidx], 3776 nk->port[pd->sidx], 0, af); 3777 pd->sport = &th->th_sport; 3778 sport = th->th_sport; 3779 } 3780 3781 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3782 nk->port[pd->didx] != dport) { 3783 pf_change_ap(daddr, &th->th_dport, pd->ip_sum, 3784 &th->th_sum, &nk->addr[pd->didx], 3785 nk->port[pd->didx], 0, af); 3786 dport = th->th_dport; 3787 pd->dport = &th->th_dport; 3788 } 3789 rewrite++; 3790 break; 3791 case IPPROTO_UDP: 3792 bproto_sum = pd->hdr.udp->uh_sum; 3793 pd->proto_sum = &pd->hdr.udp->uh_sum; 3794 3795 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3796 nk->port[pd->sidx] != sport) { 3797 pf_change_ap(saddr, &pd->hdr.udp->uh_sport, 3798 pd->ip_sum, &pd->hdr.udp->uh_sum, 3799 &nk->addr[pd->sidx], 3800 nk->port[pd->sidx], 1, af); 3801 sport = pd->hdr.udp->uh_sport; 3802 pd->sport = &pd->hdr.udp->uh_sport; 3803 } 3804 3805 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3806 nk->port[pd->didx] != dport) { 3807 pf_change_ap(daddr, &pd->hdr.udp->uh_dport, 3808 pd->ip_sum, &pd->hdr.udp->uh_sum, 3809 &nk->addr[pd->didx], 3810 nk->port[pd->didx], 1, af); 3811 dport = pd->hdr.udp->uh_dport; 3812 pd->dport = &pd->hdr.udp->uh_dport; 3813 } 3814 rewrite++; 3815 break; 3816 #ifdef INET 3817 case IPPROTO_ICMP: 3818 nk->port[0] = nk->port[1]; 3819 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 3820 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 3821 nk->addr[pd->sidx].v4.s_addr, 0); 3822 3823 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 3824 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 3825 nk->addr[pd->didx].v4.s_addr, 0); 3826 3827 if (nk->port[1] != pd->hdr.icmp->icmp_id) { 3828 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup( 3829 pd->hdr.icmp->icmp_cksum, sport, 3830 nk->port[1], 0); 3831 pd->hdr.icmp->icmp_id = nk->port[1]; 3832 pd->sport = &pd->hdr.icmp->icmp_id; 3833 } 3834 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 3835 break; 3836 #endif /* INET */ 3837 #ifdef INET6 3838 case IPPROTO_ICMPV6: 3839 nk->port[0] = nk->port[1]; 3840 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 3841 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum, 3842 &nk->addr[pd->sidx], 0); 3843 3844 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 3845 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum, 3846 &nk->addr[pd->didx], 0); 3847 rewrite++; 3848 break; 3849 #endif /* INET */ 3850 default: 3851 switch (af) { 3852 #ifdef INET 3853 case AF_INET: 3854 if (PF_ANEQ(saddr, 3855 &nk->addr[pd->sidx], AF_INET)) 3856 pf_change_a(&saddr->v4.s_addr, 3857 pd->ip_sum, 3858 nk->addr[pd->sidx].v4.s_addr, 0); 3859 3860 if (PF_ANEQ(daddr, 3861 &nk->addr[pd->didx], AF_INET)) 3862 pf_change_a(&daddr->v4.s_addr, 3863 pd->ip_sum, 3864 nk->addr[pd->didx].v4.s_addr, 0); 3865 break; 3866 #endif /* INET */ 3867 #ifdef INET6 3868 case AF_INET6: 3869 if (PF_ANEQ(saddr, 3870 &nk->addr[pd->sidx], AF_INET6)) 3871 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 3872 3873 if (PF_ANEQ(daddr, 3874 &nk->addr[pd->didx], AF_INET6)) 3875 PF_ACPY(saddr, &nk->addr[pd->didx], af); 3876 break; 3877 #endif /* INET */ 3878 } 3879 break; 3880 } 3881 if (nr->natpass) 3882 r = NULL; 3883 pd->nat_rule = nr; 3884 } 3885 3886 while (r != NULL) { 3887 r->evaluations++; 3888 if (pfi_kif_match(r->kif, kif) == r->ifnot) 3889 r = r->skip[PF_SKIP_IFP].ptr; 3890 else if (r->direction && r->direction != direction) 3891 r = r->skip[PF_SKIP_DIR].ptr; 3892 else if (r->af && r->af != af) 3893 r = r->skip[PF_SKIP_AF].ptr; 3894 else if (r->proto && r->proto != pd->proto) 3895 r = r->skip[PF_SKIP_PROTO].ptr; 3896 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 3897 r->src.neg, kif)) 3898 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3899 /* tcp/udp only. port_op always 0 in other cases */ 3900 else if (r->src.port_op && !pf_match_port(r->src.port_op, 3901 r->src.port[0], r->src.port[1], sport)) 3902 r = r->skip[PF_SKIP_SRC_PORT].ptr; 3903 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 3904 r->dst.neg, NULL)) 3905 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3906 /* tcp/udp only. port_op always 0 in other cases */ 3907 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 3908 r->dst.port[0], r->dst.port[1], dport)) 3909 r = r->skip[PF_SKIP_DST_PORT].ptr; 3910 /* icmp only. type always 0 in other cases */ 3911 else if (r->type && r->type != icmptype + 1) 3912 r = TAILQ_NEXT(r, entries); 3913 /* icmp only. type always 0 in other cases */ 3914 else if (r->code && r->code != icmpcode + 1) 3915 r = TAILQ_NEXT(r, entries); 3916 else if (r->tos && !(r->tos == pd->tos)) 3917 r = TAILQ_NEXT(r, entries); 3918 else if (r->rule_flag & PFRULE_FRAGMENT) 3919 r = TAILQ_NEXT(r, entries); 3920 else if (pd->proto == IPPROTO_TCP && 3921 (r->flagset & th->th_flags) != r->flags) 3922 r = TAILQ_NEXT(r, entries); 3923 /* tcp/udp only. uid.op always 0 in other cases */ 3924 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 3925 pf_socket_lookup(direction, pd), 1)) && 3926 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 3927 pd->lookup.uid)) 3928 r = TAILQ_NEXT(r, entries); 3929 /* tcp/udp only. gid.op always 0 in other cases */ 3930 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 3931 pf_socket_lookup(direction, pd), 1)) && 3932 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 3933 pd->lookup.gid)) 3934 r = TAILQ_NEXT(r, entries); 3935 else if (r->prob && 3936 r->prob <= karc4random()) 3937 r = TAILQ_NEXT(r, entries); 3938 else if (r->match_tag && !pf_match_tag(m, r, &tag)) 3939 r = TAILQ_NEXT(r, entries); 3940 else if (r->os_fingerprint != PF_OSFP_ANY && 3941 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 3942 pf_osfp_fingerprint(pd, m, off, th), 3943 r->os_fingerprint))) 3944 r = TAILQ_NEXT(r, entries); 3945 else { 3946 if (r->tag) 3947 tag = r->tag; 3948 if (r->rtableid >= 0) 3949 rtableid = r->rtableid; 3950 if (r->anchor == NULL) { 3951 match = 1; 3952 *rm = r; 3953 *am = a; 3954 *rsm = ruleset; 3955 if ((*rm)->quick) 3956 break; 3957 r = TAILQ_NEXT(r, entries); 3958 } else 3959 pf_step_into_anchor(&asd, &ruleset, 3960 PF_RULESET_FILTER, &r, &a, &match); 3961 } 3962 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset, 3963 PF_RULESET_FILTER, &r, &a, &match)) 3964 break; 3965 } 3966 r = *rm; 3967 a = *am; 3968 ruleset = *rsm; 3969 3970 REASON_SET(&reason, PFRES_MATCH); 3971 3972 if (r->log || (nr != NULL && nr->log)) { 3973 if (rewrite) 3974 m_copyback(m, off, hdrlen, pd->hdr.any); 3975 PFLOG_PACKET(kif, h, m, af, direction, reason, r->log ? r : nr, 3976 a, ruleset, pd); 3977 } 3978 3979 if ((r->action == PF_DROP) && 3980 ((r->rule_flag & PFRULE_RETURNRST) || 3981 (r->rule_flag & PFRULE_RETURNICMP) || 3982 (r->rule_flag & PFRULE_RETURN))) { 3983 /* undo NAT changes, if they have taken place */ 3984 if (nr != NULL) { 3985 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3986 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3987 if (pd->sport) 3988 *pd->sport = sk->port[pd->sidx]; 3989 if (pd->dport) 3990 *pd->dport = sk->port[pd->didx]; 3991 if (pd->proto_sum) 3992 *pd->proto_sum = bproto_sum; 3993 if (pd->ip_sum) 3994 *pd->ip_sum = bip_sum; 3995 m_copyback(m, off, hdrlen, pd->hdr.any); 3996 } 3997 if (pd->proto == IPPROTO_TCP && 3998 ((r->rule_flag & PFRULE_RETURNRST) || 3999 (r->rule_flag & PFRULE_RETURN)) && 4000 !(th->th_flags & TH_RST)) { 4001 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 4002 int len = 0; 4003 struct ip *h4; 4004 #ifdef INET6 4005 struct ip6_hdr *h6; 4006 #endif 4007 switch (af) { 4008 case AF_INET: 4009 h4 = mtod(m, struct ip *); 4010 len = h4->ip_len - off; 4011 break; 4012 #ifdef INET6 4013 case AF_INET6: 4014 h6 = mtod(m, struct ip6_hdr *); 4015 len = h6->ip6_plen - (off - sizeof(*h6)); 4016 break; 4017 #endif 4018 } 4019 4020 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 4021 REASON_SET(&reason, PFRES_PROTCKSUM); 4022 else { 4023 if (th->th_flags & TH_SYN) 4024 ack++; 4025 if (th->th_flags & TH_FIN) 4026 ack++; 4027 pf_send_tcp(r, af, pd->dst, 4028 pd->src, th->th_dport, th->th_sport, 4029 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 4030 r->return_ttl, 1, 0, pd->eh, kif->pfik_ifp); 4031 } 4032 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 4033 r->return_icmp) 4034 pf_send_icmp(m, r->return_icmp >> 8, 4035 r->return_icmp & 255, af, r); 4036 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 4037 r->return_icmp6) 4038 pf_send_icmp(m, r->return_icmp6 >> 8, 4039 r->return_icmp6 & 255, af, r); 4040 } 4041 4042 if (r->action == PF_DROP) 4043 goto cleanup; 4044 4045 if (pf_tag_packet(m, tag, rtableid)) { 4046 REASON_SET(&reason, PFRES_MEMORY); 4047 goto cleanup; 4048 } 4049 4050 if (!state_icmp && (r->keep_state || nr != NULL || 4051 (pd->flags & PFDESC_TCP_NORM))) { 4052 int action; 4053 action = pf_create_state(r, nr, a, pd, nsn, skw, sks, nk, sk, m, 4054 off, sport, dport, &rewrite, kif, sm, tag, bproto_sum, 4055 bip_sum, hdrlen); 4056 if (action != PF_PASS) 4057 return (action); 4058 } 4059 4060 /* copy back packet headers if we performed NAT operations */ 4061 if (rewrite) 4062 m_copyback(m, off, hdrlen, pd->hdr.any); 4063 4064 return (PF_PASS); 4065 4066 cleanup: 4067 if (sk != NULL) 4068 kfree(sk, M_PFSTATEKEYPL); 4069 if (nk != NULL) 4070 kfree(nk, M_PFSTATEKEYPL); 4071 return (PF_DROP); 4072 } 4073 4074 static __inline int 4075 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a, 4076 struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *skw, 4077 struct pf_state_key *sks, struct pf_state_key *nk, struct pf_state_key *sk, 4078 struct mbuf *m, int off, u_int16_t sport, u_int16_t dport, int *rewrite, 4079 struct pfi_kif *kif, struct pf_state **sm, int tag, u_int16_t bproto_sum, 4080 u_int16_t bip_sum, int hdrlen) 4081 { 4082 struct pf_state *s = NULL; 4083 struct pf_src_node *sn = NULL; 4084 struct tcphdr *th = pd->hdr.tcp; 4085 u_int16_t mss = tcp_mssdflt; 4086 u_short reason; 4087 int cpu = mycpu->gd_cpuid; 4088 4089 /* check maximums */ 4090 if (r->max_states && (r->states_cur >= r->max_states)) { 4091 PF_INC_LCOUNTER(LCNT_STATES); 4092 REASON_SET(&reason, PFRES_MAXSTATES); 4093 return (PF_DROP); 4094 } 4095 /* src node for filter rule */ 4096 if ((r->rule_flag & PFRULE_SRCTRACK || 4097 r->rpool.opts & PF_POOL_STICKYADDR) && 4098 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { 4099 REASON_SET(&reason, PFRES_SRCLIMIT); 4100 goto csfailed; 4101 } 4102 /* src node for translation rule */ 4103 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 4104 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { 4105 REASON_SET(&reason, PFRES_SRCLIMIT); 4106 goto csfailed; 4107 } 4108 s = kmalloc(sizeof(struct pf_state), M_PFSTATEPL, M_NOWAIT|M_ZERO); 4109 if (s == NULL) { 4110 REASON_SET(&reason, PFRES_MEMORY); 4111 goto csfailed; 4112 } 4113 lockinit(&s->lk, "pfstlk", 0, 0); 4114 s->id = 0; /* XXX Do we really need that? not in OpenBSD */ 4115 s->creatorid = 0; 4116 s->rule.ptr = r; 4117 s->nat_rule.ptr = nr; 4118 s->anchor.ptr = a; 4119 s->state_flags = PFSTATE_CREATEINPROG; 4120 STATE_INC_COUNTERS(s); 4121 if (r->allow_opts) 4122 s->state_flags |= PFSTATE_ALLOWOPTS; 4123 if (r->rule_flag & PFRULE_STATESLOPPY) 4124 s->state_flags |= PFSTATE_SLOPPY; 4125 if (pd->not_cpu_localized) 4126 s->state_flags |= PFSTATE_STACK_GLOBAL; 4127 4128 s->log = r->log & PF_LOG_ALL; 4129 if (nr != NULL) 4130 s->log |= nr->log & PF_LOG_ALL; 4131 switch (pd->proto) { 4132 case IPPROTO_TCP: 4133 s->src.seqlo = ntohl(th->th_seq); 4134 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 4135 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4136 r->keep_state == PF_STATE_MODULATE) { 4137 /* Generate sequence number modulator */ 4138 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 4139 0) 4140 s->src.seqdiff = 1; 4141 pf_change_a(&th->th_seq, &th->th_sum, 4142 htonl(s->src.seqlo + s->src.seqdiff), 0); 4143 *rewrite = 1; 4144 } else 4145 s->src.seqdiff = 0; 4146 if (th->th_flags & TH_SYN) { 4147 s->src.seqhi++; 4148 s->src.wscale = pf_get_wscale(m, off, 4149 th->th_off, pd->af); 4150 } 4151 s->src.max_win = MAX(ntohs(th->th_win), 1); 4152 if (s->src.wscale & PF_WSCALE_MASK) { 4153 /* Remove scale factor from initial window */ 4154 int win = s->src.max_win; 4155 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 4156 s->src.max_win = (win - 1) >> 4157 (s->src.wscale & PF_WSCALE_MASK); 4158 } 4159 if (th->th_flags & TH_FIN) 4160 s->src.seqhi++; 4161 s->dst.seqhi = 1; 4162 s->dst.max_win = 1; 4163 s->src.state = TCPS_SYN_SENT; 4164 s->dst.state = TCPS_CLOSED; 4165 s->timeout = PFTM_TCP_FIRST_PACKET; 4166 break; 4167 case IPPROTO_UDP: 4168 s->src.state = PFUDPS_SINGLE; 4169 s->dst.state = PFUDPS_NO_TRAFFIC; 4170 s->timeout = PFTM_UDP_FIRST_PACKET; 4171 break; 4172 case IPPROTO_ICMP: 4173 #ifdef INET6 4174 case IPPROTO_ICMPV6: 4175 #endif 4176 s->timeout = PFTM_ICMP_FIRST_PACKET; 4177 break; 4178 default: 4179 s->src.state = PFOTHERS_SINGLE; 4180 s->dst.state = PFOTHERS_NO_TRAFFIC; 4181 s->timeout = PFTM_OTHER_FIRST_PACKET; 4182 } 4183 4184 s->creation = time_second; 4185 s->expire = time_second; 4186 4187 if (sn != NULL) { 4188 s->src_node = sn; 4189 s->src_node->states++; 4190 } 4191 if (nsn != NULL) { 4192 /* XXX We only modify one side for now. */ 4193 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 4194 s->nat_src_node = nsn; 4195 s->nat_src_node->states++; 4196 } 4197 if (pd->proto == IPPROTO_TCP) { 4198 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m, 4199 off, pd, th, &s->src, &s->dst)) { 4200 REASON_SET(&reason, PFRES_MEMORY); 4201 pf_src_tree_remove_state(s); 4202 STATE_DEC_COUNTERS(s); 4203 kfree(s, M_PFSTATEPL); 4204 return (PF_DROP); 4205 } 4206 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub && 4207 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 4208 &s->src, &s->dst, rewrite)) { 4209 /* This really shouldn't happen!!! */ 4210 DPFPRINTF(PF_DEBUG_URGENT, 4211 ("pf_normalize_tcp_stateful failed on first pkt")); 4212 pf_normalize_tcp_cleanup(s); 4213 pf_src_tree_remove_state(s); 4214 STATE_DEC_COUNTERS(s); 4215 kfree(s, M_PFSTATEPL); 4216 return (PF_DROP); 4217 } 4218 } 4219 s->direction = pd->dir; 4220 4221 if (sk == NULL && pf_state_key_setup(pd, nr, &skw, &sks, &sk, &nk, 4222 pd->src, pd->dst, sport, dport)) { 4223 REASON_SET(&reason, PFRES_MEMORY); 4224 goto csfailed; 4225 } 4226 4227 if (pf_state_insert(BOUND_IFACE(r, kif), skw, sks, s)) { 4228 if (pd->proto == IPPROTO_TCP) 4229 pf_normalize_tcp_cleanup(s); 4230 REASON_SET(&reason, PFRES_STATEINS); 4231 pf_src_tree_remove_state(s); 4232 STATE_DEC_COUNTERS(s); 4233 kfree(s, M_PFSTATEPL); 4234 return (PF_DROP); 4235 } else 4236 *sm = s; 4237 4238 pf_set_rt_ifp(s, pd->src); /* needs s->state_key set */ 4239 if (tag > 0) { 4240 pf_tag_ref(tag); 4241 s->tag = tag; 4242 } 4243 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 4244 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 4245 s->src.state = PF_TCPS_PROXY_SRC; 4246 /* undo NAT changes, if they have taken place */ 4247 if (nr != NULL) { 4248 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 4249 if (pd->dir == PF_OUT) 4250 skt = s->key[PF_SK_STACK]; 4251 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 4252 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 4253 if (pd->sport) 4254 *pd->sport = skt->port[pd->sidx]; 4255 if (pd->dport) 4256 *pd->dport = skt->port[pd->didx]; 4257 if (pd->proto_sum) 4258 *pd->proto_sum = bproto_sum; 4259 if (pd->ip_sum) 4260 *pd->ip_sum = bip_sum; 4261 m->m_flags &= ~M_HASH; 4262 m_copyback(m, off, hdrlen, pd->hdr.any); 4263 } 4264 s->src.seqhi = htonl(karc4random()); 4265 /* Find mss option */ 4266 mss = pf_get_mss(m, off, th->th_off, pd->af); 4267 mss = pf_calc_mss(pd->src, pd->af, mss); 4268 mss = pf_calc_mss(pd->dst, pd->af, mss); 4269 s->src.mss = mss; 4270 s->state_flags &= ~PFSTATE_CREATEINPROG; 4271 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 4272 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 4273 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL, NULL); 4274 REASON_SET(&reason, PFRES_SYNPROXY); 4275 return (PF_SYNPROXY_DROP); 4276 } 4277 4278 s->state_flags &= ~PFSTATE_CREATEINPROG; 4279 return (PF_PASS); 4280 4281 csfailed: 4282 if (sk != NULL) 4283 kfree(sk, M_PFSTATEKEYPL); 4284 if (nk != NULL) 4285 kfree(nk, M_PFSTATEKEYPL); 4286 4287 if (sn != NULL && sn->states == 0 && sn->expire == 0) { 4288 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], sn); 4289 PF_INC_SCOUNTER(SCNT_SRC_NODE_REMOVALS); 4290 atomic_add_int(&pf_status.src_nodes, -1); 4291 kfree(sn, M_PFSRCTREEPL); 4292 } 4293 if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0) { 4294 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], nsn); 4295 PF_INC_SCOUNTER(SCNT_SRC_NODE_REMOVALS); 4296 atomic_add_int(&pf_status.src_nodes, -1); 4297 kfree(nsn, M_PFSRCTREEPL); 4298 } 4299 if (s) { 4300 pf_src_tree_remove_state(s); 4301 STATE_DEC_COUNTERS(s); 4302 kfree(s, M_PFSTATEPL); 4303 } 4304 4305 return (PF_DROP); 4306 } 4307 4308 int 4309 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif, 4310 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am, 4311 struct pf_ruleset **rsm) 4312 { 4313 struct pf_rule *r, *a = NULL; 4314 struct pf_ruleset *ruleset = NULL; 4315 sa_family_t af = pd->af; 4316 u_short reason; 4317 int tag = -1; 4318 int asd = 0; 4319 int match = 0; 4320 4321 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4322 while (r != NULL) { 4323 r->evaluations++; 4324 if (pfi_kif_match(r->kif, kif) == r->ifnot) 4325 r = r->skip[PF_SKIP_IFP].ptr; 4326 else if (r->direction && r->direction != direction) 4327 r = r->skip[PF_SKIP_DIR].ptr; 4328 else if (r->af && r->af != af) 4329 r = r->skip[PF_SKIP_AF].ptr; 4330 else if (r->proto && r->proto != pd->proto) 4331 r = r->skip[PF_SKIP_PROTO].ptr; 4332 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 4333 r->src.neg, kif)) 4334 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4335 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 4336 r->dst.neg, NULL)) 4337 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4338 else if (r->tos && !(r->tos == pd->tos)) 4339 r = TAILQ_NEXT(r, entries); 4340 else if (r->os_fingerprint != PF_OSFP_ANY) 4341 r = TAILQ_NEXT(r, entries); 4342 else if (pd->proto == IPPROTO_UDP && 4343 (r->src.port_op || r->dst.port_op)) 4344 r = TAILQ_NEXT(r, entries); 4345 else if (pd->proto == IPPROTO_TCP && 4346 (r->src.port_op || r->dst.port_op || r->flagset)) 4347 r = TAILQ_NEXT(r, entries); 4348 else if ((pd->proto == IPPROTO_ICMP || 4349 pd->proto == IPPROTO_ICMPV6) && 4350 (r->type || r->code)) 4351 r = TAILQ_NEXT(r, entries); 4352 else if (r->prob && r->prob <= karc4random()) 4353 r = TAILQ_NEXT(r, entries); 4354 else if (r->match_tag && !pf_match_tag(m, r, &tag)) 4355 r = TAILQ_NEXT(r, entries); 4356 else { 4357 if (r->anchor == NULL) { 4358 match = 1; 4359 *rm = r; 4360 *am = a; 4361 *rsm = ruleset; 4362 if ((*rm)->quick) 4363 break; 4364 r = TAILQ_NEXT(r, entries); 4365 } else 4366 pf_step_into_anchor(&asd, &ruleset, 4367 PF_RULESET_FILTER, &r, &a, &match); 4368 } 4369 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset, 4370 PF_RULESET_FILTER, &r, &a, &match)) 4371 break; 4372 } 4373 r = *rm; 4374 a = *am; 4375 ruleset = *rsm; 4376 4377 REASON_SET(&reason, PFRES_MATCH); 4378 4379 if (r->log) 4380 PFLOG_PACKET(kif, h, m, af, direction, reason, r, a, ruleset, 4381 pd); 4382 4383 if (r->action != PF_PASS) 4384 return (PF_DROP); 4385 4386 if (pf_tag_packet(m, tag, -1)) { 4387 REASON_SET(&reason, PFRES_MEMORY); 4388 return (PF_DROP); 4389 } 4390 4391 return (PF_PASS); 4392 } 4393 4394 /* 4395 * Called with state locked 4396 */ 4397 int 4398 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst, 4399 struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off, 4400 struct pf_pdesc *pd, u_short *reason, int *copyback) 4401 { 4402 struct tcphdr *th = pd->hdr.tcp; 4403 u_int16_t win = ntohs(th->th_win); 4404 u_int32_t ack, end, seq, orig_seq; 4405 u_int8_t sws, dws; 4406 int ackskew; 4407 4408 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 4409 sws = src->wscale & PF_WSCALE_MASK; 4410 dws = dst->wscale & PF_WSCALE_MASK; 4411 } else { 4412 sws = dws = 0; 4413 } 4414 4415 /* 4416 * Sequence tracking algorithm from Guido van Rooij's paper: 4417 * http://www.madison-gurkha.com/publications/tcp_filtering/ 4418 * tcp_filtering.ps 4419 */ 4420 4421 orig_seq = seq = ntohl(th->th_seq); 4422 if (src->seqlo == 0) { 4423 /* First packet from this end. Set its state */ 4424 4425 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) && 4426 src->scrub == NULL) { 4427 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 4428 REASON_SET(reason, PFRES_MEMORY); 4429 return (PF_DROP); 4430 } 4431 } 4432 4433 /* Deferred generation of sequence number modulator */ 4434 if (dst->seqdiff && !src->seqdiff) { 4435 /* use random iss for the TCP server */ 4436 while ((src->seqdiff = karc4random() - seq) == 0) 4437 ; 4438 ack = ntohl(th->th_ack) - dst->seqdiff; 4439 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq + 4440 src->seqdiff), 0); 4441 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0); 4442 *copyback = 1; 4443 } else { 4444 ack = ntohl(th->th_ack); 4445 } 4446 4447 end = seq + pd->p_len; 4448 if (th->th_flags & TH_SYN) { 4449 end++; 4450 (*state)->sync_flags |= PFSTATE_GOT_SYN2; 4451 if (dst->wscale & PF_WSCALE_FLAG) { 4452 src->wscale = pf_get_wscale(m, off, th->th_off, 4453 pd->af); 4454 if (src->wscale & PF_WSCALE_FLAG) { 4455 /* Remove scale factor from initial 4456 * window */ 4457 sws = src->wscale & PF_WSCALE_MASK; 4458 win = ((u_int32_t)win + (1 << sws) - 1) 4459 >> sws; 4460 dws = dst->wscale & PF_WSCALE_MASK; 4461 } else { 4462 /* fixup other window */ 4463 dst->max_win <<= dst->wscale & 4464 PF_WSCALE_MASK; 4465 /* in case of a retrans SYN|ACK */ 4466 dst->wscale = 0; 4467 } 4468 } 4469 } 4470 if (th->th_flags & TH_FIN) 4471 end++; 4472 4473 src->seqlo = seq; 4474 if (src->state < TCPS_SYN_SENT) 4475 src->state = TCPS_SYN_SENT; 4476 4477 /* 4478 * May need to slide the window (seqhi may have been set by 4479 * the crappy stack check or if we picked up the connection 4480 * after establishment) 4481 */ 4482 if (src->seqhi == 1 || 4483 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 4484 src->seqhi = end + MAX(1, dst->max_win << dws); 4485 if (win > src->max_win) 4486 src->max_win = win; 4487 4488 } else { 4489 ack = ntohl(th->th_ack) - dst->seqdiff; 4490 if (src->seqdiff) { 4491 /* Modulate sequence numbers */ 4492 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq + 4493 src->seqdiff), 0); 4494 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0); 4495 *copyback = 1; 4496 } 4497 end = seq + pd->p_len; 4498 if (th->th_flags & TH_SYN) 4499 end++; 4500 if (th->th_flags & TH_FIN) 4501 end++; 4502 } 4503 4504 if ((th->th_flags & TH_ACK) == 0) { 4505 /* Let it pass through the ack skew check */ 4506 ack = dst->seqlo; 4507 } else if ((ack == 0 && 4508 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 4509 /* broken tcp stacks do not set ack */ 4510 (dst->state < TCPS_SYN_SENT)) { 4511 /* 4512 * Many stacks (ours included) will set the ACK number in an 4513 * FIN|ACK if the SYN times out -- no sequence to ACK. 4514 */ 4515 ack = dst->seqlo; 4516 } 4517 4518 if (seq == end) { 4519 /* Ease sequencing restrictions on no data packets */ 4520 seq = src->seqlo; 4521 end = seq; 4522 } 4523 4524 ackskew = dst->seqlo - ack; 4525 4526 4527 /* 4528 * Need to demodulate the sequence numbers in any TCP SACK options 4529 * (Selective ACK). We could optionally validate the SACK values 4530 * against the current ACK window, either forwards or backwards, but 4531 * I'm not confident that SACK has been implemented properly 4532 * everywhere. It wouldn't surprise me if several stacks accidently 4533 * SACK too far backwards of previously ACKed data. There really aren't 4534 * any security implications of bad SACKing unless the target stack 4535 * doesn't validate the option length correctly. Someone trying to 4536 * spoof into a TCP connection won't bother blindly sending SACK 4537 * options anyway. 4538 */ 4539 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 4540 if (pf_modulate_sack(m, off, pd, th, dst)) 4541 *copyback = 1; 4542 } 4543 4544 4545 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 4546 if (SEQ_GEQ(src->seqhi, end) && 4547 /* Last octet inside other's window space */ 4548 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 4549 /* Retrans: not more than one window back */ 4550 (ackskew >= -MAXACKWINDOW) && 4551 /* Acking not more than one reassembled fragment backwards */ 4552 (ackskew <= (MAXACKWINDOW << sws)) && 4553 /* Acking not more than one window forward */ 4554 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 4555 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 4556 (pd->flags & PFDESC_IP_REAS) == 0)) { 4557 /* Require an exact/+1 sequence match on resets when possible */ 4558 4559 if (dst->scrub || src->scrub) { 4560 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4561 *state, src, dst, copyback)) 4562 return (PF_DROP); 4563 } 4564 4565 /* update max window */ 4566 if (src->max_win < win) 4567 src->max_win = win; 4568 /* synchronize sequencing */ 4569 if (SEQ_GT(end, src->seqlo)) 4570 src->seqlo = end; 4571 /* slide the window of what the other end can send */ 4572 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4573 dst->seqhi = ack + MAX((win << sws), 1); 4574 4575 4576 /* update states */ 4577 if (th->th_flags & TH_SYN) 4578 if (src->state < TCPS_SYN_SENT) 4579 src->state = TCPS_SYN_SENT; 4580 if (th->th_flags & TH_FIN) 4581 if (src->state < TCPS_CLOSING) 4582 src->state = TCPS_CLOSING; 4583 if (th->th_flags & TH_ACK) { 4584 if (dst->state == TCPS_SYN_SENT) { 4585 dst->state = TCPS_ESTABLISHED; 4586 if (src->state == TCPS_ESTABLISHED && 4587 (*state)->src_node != NULL && 4588 pf_src_connlimit(*state)) { 4589 REASON_SET(reason, PFRES_SRCLIMIT); 4590 return (PF_DROP); 4591 } 4592 } else if (dst->state == TCPS_CLOSING) 4593 dst->state = TCPS_FIN_WAIT_2; 4594 } 4595 if (th->th_flags & TH_RST) 4596 src->state = dst->state = TCPS_TIME_WAIT; 4597 4598 /* update expire time */ 4599 (*state)->expire = time_second; 4600 if (src->state >= TCPS_FIN_WAIT_2 && 4601 dst->state >= TCPS_FIN_WAIT_2) 4602 (*state)->timeout = PFTM_TCP_CLOSED; 4603 else if (src->state >= TCPS_CLOSING && 4604 dst->state >= TCPS_CLOSING) 4605 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4606 else if (src->state < TCPS_ESTABLISHED || 4607 dst->state < TCPS_ESTABLISHED) 4608 (*state)->timeout = PFTM_TCP_OPENING; 4609 else if (src->state >= TCPS_CLOSING || 4610 dst->state >= TCPS_CLOSING) 4611 (*state)->timeout = PFTM_TCP_CLOSING; 4612 else 4613 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4614 4615 /* Fall through to PASS packet */ 4616 4617 } else if ((dst->state < TCPS_SYN_SENT || 4618 dst->state >= TCPS_FIN_WAIT_2 || 4619 src->state >= TCPS_FIN_WAIT_2) && 4620 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 4621 /* Within a window forward of the originating packet */ 4622 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 4623 /* Within a window backward of the originating packet */ 4624 4625 /* 4626 * This currently handles three situations: 4627 * 1) Stupid stacks will shotgun SYNs before their peer 4628 * replies. 4629 * 2) When PF catches an already established stream (the 4630 * firewall rebooted, the state table was flushed, routes 4631 * changed...) 4632 * 3) Packets get funky immediately after the connection 4633 * closes (this should catch Solaris spurious ACK|FINs 4634 * that web servers like to spew after a close) 4635 * 4636 * This must be a little more careful than the above code 4637 * since packet floods will also be caught here. We don't 4638 * update the TTL here to mitigate the damage of a packet 4639 * flood and so the same code can handle awkward establishment 4640 * and a loosened connection close. 4641 * In the establishment case, a correct peer response will 4642 * validate the connection, go through the normal state code 4643 * and keep updating the state TTL. 4644 */ 4645 4646 if (pf_status.debug >= PF_DEBUG_MISC) { 4647 kprintf("pf: loose state match: "); 4648 pf_print_state(*state); 4649 pf_print_flags(th->th_flags); 4650 kprintf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4651 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, pd->p_len, 4652 ackskew, (unsigned long long)(*state)->packets[0], 4653 (unsigned long long)(*state)->packets[1], 4654 pd->dir == PF_IN ? "in" : "out", 4655 pd->dir == (*state)->direction ? "fwd" : "rev"); 4656 } 4657 4658 if (dst->scrub || src->scrub) { 4659 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4660 *state, src, dst, copyback)) 4661 return (PF_DROP); 4662 } 4663 4664 /* update max window */ 4665 if (src->max_win < win) 4666 src->max_win = win; 4667 /* synchronize sequencing */ 4668 if (SEQ_GT(end, src->seqlo)) 4669 src->seqlo = end; 4670 /* slide the window of what the other end can send */ 4671 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4672 dst->seqhi = ack + MAX((win << sws), 1); 4673 4674 /* 4675 * Cannot set dst->seqhi here since this could be a shotgunned 4676 * SYN and not an already established connection. 4677 */ 4678 4679 if (th->th_flags & TH_FIN) 4680 if (src->state < TCPS_CLOSING) 4681 src->state = TCPS_CLOSING; 4682 if (th->th_flags & TH_RST) 4683 src->state = dst->state = TCPS_TIME_WAIT; 4684 4685 /* Fall through to PASS packet */ 4686 4687 } else if ((*state)->pickup_mode == PF_PICKUPS_HASHONLY || 4688 ((*state)->pickup_mode == PF_PICKUPS_ENABLED && 4689 ((*state)->sync_flags & PFSTATE_GOT_SYN_MASK) != 4690 PFSTATE_GOT_SYN_MASK)) { 4691 /* 4692 * If pickup mode is hash only, do not fail on sequence checks. 4693 * 4694 * If pickup mode is enabled and we did not see the SYN in 4695 * both direction, do not fail on sequence checks because 4696 * we do not have complete information on window scale. 4697 * 4698 * Adjust expiration and fall through to PASS packet. 4699 * XXX Add a FIN check to reduce timeout? 4700 */ 4701 (*state)->expire = time_second; 4702 } else { 4703 /* 4704 * Failure processing 4705 */ 4706 if ((*state)->dst.state == TCPS_SYN_SENT && 4707 (*state)->src.state == TCPS_SYN_SENT) { 4708 /* Send RST for state mismatches during handshake */ 4709 if (!(th->th_flags & TH_RST)) 4710 pf_send_tcp((*state)->rule.ptr, pd->af, 4711 pd->dst, pd->src, th->th_dport, 4712 th->th_sport, ntohl(th->th_ack), 0, 4713 TH_RST, 0, 0, 4714 (*state)->rule.ptr->return_ttl, 1, 0, 4715 pd->eh, kif->pfik_ifp); 4716 src->seqlo = 0; 4717 src->seqhi = 1; 4718 src->max_win = 1; 4719 } else if (pf_status.debug >= PF_DEBUG_MISC) { 4720 kprintf("pf: BAD state: "); 4721 pf_print_state(*state); 4722 pf_print_flags(th->th_flags); 4723 kprintf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4724 "pkts=%llu:%llu dir=%s,%s\n", 4725 seq, orig_seq, ack, pd->p_len, ackskew, 4726 (unsigned long long)(*state)->packets[0], 4727 (unsigned long long)(*state)->packets[1], 4728 pd->dir == PF_IN ? "in" : "out", 4729 pd->dir == (*state)->direction ? "fwd" : "rev"); 4730 kprintf("pf: State failure on: %c %c %c %c | %c %c\n", 4731 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 4732 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 4733 ' ': '2', 4734 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 4735 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 4736 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 4737 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 4738 } 4739 REASON_SET(reason, PFRES_BADSTATE); 4740 return (PF_DROP); 4741 } 4742 4743 return (PF_PASS); 4744 } 4745 4746 /* 4747 * Called with state locked 4748 */ 4749 int 4750 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst, 4751 struct pf_state **state, struct pf_pdesc *pd, u_short *reason) 4752 { 4753 struct tcphdr *th = pd->hdr.tcp; 4754 4755 if (th->th_flags & TH_SYN) 4756 if (src->state < TCPS_SYN_SENT) 4757 src->state = TCPS_SYN_SENT; 4758 if (th->th_flags & TH_FIN) 4759 if (src->state < TCPS_CLOSING) 4760 src->state = TCPS_CLOSING; 4761 if (th->th_flags & TH_ACK) { 4762 if (dst->state == TCPS_SYN_SENT) { 4763 dst->state = TCPS_ESTABLISHED; 4764 if (src->state == TCPS_ESTABLISHED && 4765 (*state)->src_node != NULL && 4766 pf_src_connlimit(*state)) { 4767 REASON_SET(reason, PFRES_SRCLIMIT); 4768 return (PF_DROP); 4769 } 4770 } else if (dst->state == TCPS_CLOSING) { 4771 dst->state = TCPS_FIN_WAIT_2; 4772 } else if (src->state == TCPS_SYN_SENT && 4773 dst->state < TCPS_SYN_SENT) { 4774 /* 4775 * Handle a special sloppy case where we only see one 4776 * half of the connection. If there is a ACK after 4777 * the initial SYN without ever seeing a packet from 4778 * the destination, set the connection to established. 4779 */ 4780 dst->state = src->state = TCPS_ESTABLISHED; 4781 if ((*state)->src_node != NULL && 4782 pf_src_connlimit(*state)) { 4783 REASON_SET(reason, PFRES_SRCLIMIT); 4784 return (PF_DROP); 4785 } 4786 } else if (src->state == TCPS_CLOSING && 4787 dst->state == TCPS_ESTABLISHED && 4788 dst->seqlo == 0) { 4789 /* 4790 * Handle the closing of half connections where we 4791 * don't see the full bidirectional FIN/ACK+ACK 4792 * handshake. 4793 */ 4794 dst->state = TCPS_CLOSING; 4795 } 4796 } 4797 if (th->th_flags & TH_RST) 4798 src->state = dst->state = TCPS_TIME_WAIT; 4799 4800 /* update expire time */ 4801 (*state)->expire = time_second; 4802 if (src->state >= TCPS_FIN_WAIT_2 && 4803 dst->state >= TCPS_FIN_WAIT_2) 4804 (*state)->timeout = PFTM_TCP_CLOSED; 4805 else if (src->state >= TCPS_CLOSING && 4806 dst->state >= TCPS_CLOSING) 4807 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4808 else if (src->state < TCPS_ESTABLISHED || 4809 dst->state < TCPS_ESTABLISHED) 4810 (*state)->timeout = PFTM_TCP_OPENING; 4811 else if (src->state >= TCPS_CLOSING || 4812 dst->state >= TCPS_CLOSING) 4813 (*state)->timeout = PFTM_TCP_CLOSING; 4814 else 4815 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4816 4817 return (PF_PASS); 4818 } 4819 4820 /* 4821 * Test TCP connection state. Caller must hold the state locked. 4822 */ 4823 int 4824 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif, 4825 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 4826 u_short *reason) 4827 { 4828 struct pf_state_key_cmp key; 4829 struct tcphdr *th = pd->hdr.tcp; 4830 int copyback = 0; 4831 int error; 4832 struct pf_state_peer *src, *dst; 4833 struct pf_state_key *sk; 4834 4835 bzero(&key, sizeof(key)); 4836 key.af = pd->af; 4837 key.proto = IPPROTO_TCP; 4838 if (direction == PF_IN) { /* wire side, straight */ 4839 PF_ACPY(&key.addr[0], pd->src, key.af); 4840 PF_ACPY(&key.addr[1], pd->dst, key.af); 4841 key.port[0] = th->th_sport; 4842 key.port[1] = th->th_dport; 4843 if (pf_status.debug >= PF_DEBUG_MISC) { 4844 kprintf("test-tcp IN (%08x:%d) -> (%08x:%d)\n", 4845 ntohl(key.addr[0].addr32[0]), 4846 ntohs(key.port[0]), 4847 ntohl(key.addr[1].addr32[0]), 4848 ntohs(key.port[1])); 4849 } 4850 } else { /* stack side, reverse */ 4851 PF_ACPY(&key.addr[1], pd->src, key.af); 4852 PF_ACPY(&key.addr[0], pd->dst, key.af); 4853 key.port[1] = th->th_sport; 4854 key.port[0] = th->th_dport; 4855 if (pf_status.debug >= PF_DEBUG_MISC) { 4856 kprintf("test-tcp OUT (%08x:%d) <- (%08x:%d)\n", 4857 ntohl(key.addr[0].addr32[0]), 4858 ntohs(key.port[0]), 4859 ntohl(key.addr[1].addr32[0]), 4860 ntohs(key.port[1])); 4861 } 4862 } 4863 4864 STATE_LOOKUP(kif, &key, direction, *state, m); 4865 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 4866 4867 if (direction == (*state)->direction) { 4868 src = &(*state)->src; 4869 dst = &(*state)->dst; 4870 } else { 4871 src = &(*state)->dst; 4872 dst = &(*state)->src; 4873 } 4874 4875 sk = (*state)->key[pd->didx]; 4876 4877 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 4878 if (direction != (*state)->direction) { 4879 REASON_SET(reason, PFRES_SYNPROXY); 4880 FAIL (PF_SYNPROXY_DROP); 4881 } 4882 if (th->th_flags & TH_SYN) { 4883 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 4884 REASON_SET(reason, PFRES_SYNPROXY); 4885 FAIL (PF_DROP); 4886 } 4887 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 4888 pd->src, th->th_dport, th->th_sport, 4889 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 4890 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 4891 0, NULL, NULL); 4892 REASON_SET(reason, PFRES_SYNPROXY); 4893 FAIL (PF_SYNPROXY_DROP); 4894 } else if (!(th->th_flags & TH_ACK) || 4895 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4896 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4897 REASON_SET(reason, PFRES_SYNPROXY); 4898 FAIL (PF_DROP); 4899 } else if ((*state)->src_node != NULL && 4900 pf_src_connlimit(*state)) { 4901 REASON_SET(reason, PFRES_SRCLIMIT); 4902 FAIL (PF_DROP); 4903 } else 4904 (*state)->src.state = PF_TCPS_PROXY_DST; 4905 } 4906 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 4907 if (direction == (*state)->direction) { 4908 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 4909 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4910 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4911 REASON_SET(reason, PFRES_SYNPROXY); 4912 FAIL (PF_DROP); 4913 } 4914 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 4915 if ((*state)->dst.seqhi == 1) 4916 (*state)->dst.seqhi = htonl(karc4random()); 4917 pf_send_tcp((*state)->rule.ptr, pd->af, 4918 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4919 sk->port[pd->sidx], sk->port[pd->didx], 4920 (*state)->dst.seqhi, 0, TH_SYN, 0, 4921 (*state)->src.mss, 0, 0, (*state)->tag, NULL, NULL); 4922 REASON_SET(reason, PFRES_SYNPROXY); 4923 FAIL (PF_SYNPROXY_DROP); 4924 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 4925 (TH_SYN|TH_ACK)) || 4926 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 4927 REASON_SET(reason, PFRES_SYNPROXY); 4928 FAIL (PF_DROP); 4929 } else { 4930 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 4931 (*state)->dst.seqlo = ntohl(th->th_seq); 4932 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 4933 pd->src, th->th_dport, th->th_sport, 4934 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 4935 TH_ACK, (*state)->src.max_win, 0, 0, 0, 4936 (*state)->tag, NULL, NULL); 4937 pf_send_tcp((*state)->rule.ptr, pd->af, 4938 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4939 sk->port[pd->sidx], sk->port[pd->didx], 4940 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 4941 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 4942 0, NULL, NULL); 4943 (*state)->src.seqdiff = (*state)->dst.seqhi - 4944 (*state)->src.seqlo; 4945 (*state)->dst.seqdiff = (*state)->src.seqhi - 4946 (*state)->dst.seqlo; 4947 (*state)->src.seqhi = (*state)->src.seqlo + 4948 (*state)->dst.max_win; 4949 (*state)->dst.seqhi = (*state)->dst.seqlo + 4950 (*state)->src.max_win; 4951 (*state)->src.wscale = (*state)->dst.wscale = 0; 4952 (*state)->src.state = (*state)->dst.state = 4953 TCPS_ESTABLISHED; 4954 REASON_SET(reason, PFRES_SYNPROXY); 4955 FAIL (PF_SYNPROXY_DROP); 4956 } 4957 } 4958 4959 /* 4960 * Check for connection (addr+port pair) reuse. We can't actually 4961 * unlink the state if we don't own it. 4962 */ 4963 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) && 4964 dst->state >= TCPS_FIN_WAIT_2 && 4965 src->state >= TCPS_FIN_WAIT_2) { 4966 if (pf_status.debug >= PF_DEBUG_MISC) { 4967 kprintf("pf: state reuse "); 4968 pf_print_state(*state); 4969 pf_print_flags(th->th_flags); 4970 kprintf("\n"); 4971 } 4972 /* XXX make sure it's the same direction ?? */ 4973 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; 4974 if ((*state)->cpuid == mycpu->gd_cpuid) { 4975 pf_unlink_state(*state); 4976 *state = NULL; 4977 } else { 4978 (*state)->timeout = PFTM_PURGE; 4979 } 4980 FAIL (PF_DROP); 4981 } 4982 4983 if ((*state)->state_flags & PFSTATE_SLOPPY) { 4984 if (pf_tcp_track_sloppy(src, dst, state, pd, 4985 reason) == PF_DROP) { 4986 FAIL (PF_DROP); 4987 } 4988 } else { 4989 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, 4990 reason, ©back) == PF_DROP) { 4991 FAIL (PF_DROP); 4992 } 4993 } 4994 4995 /* translate source/destination address, if necessary */ 4996 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4997 struct pf_state_key *nk = (*state)->key[pd->didx]; 4998 4999 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5000 nk->port[pd->sidx] != th->th_sport) { 5001 /* 5002 * The translated source address may be completely 5003 * unrelated to the saved link header, make sure 5004 * a bridge doesn't try to use it. 5005 */ 5006 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 5007 pf_change_ap(pd->src, &th->th_sport, pd->ip_sum, 5008 &th->th_sum, &nk->addr[pd->sidx], 5009 nk->port[pd->sidx], 0, pd->af); 5010 } 5011 5012 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5013 nk->port[pd->didx] != th->th_dport) { 5014 /* 5015 * If we don't redispatch the packet will go into 5016 * the protocol stack on the wrong cpu for the 5017 * post-translated address. 5018 */ 5019 pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum, 5020 &th->th_sum, &nk->addr[pd->didx], 5021 nk->port[pd->didx], 0, pd->af); 5022 } 5023 copyback = 1; 5024 } 5025 5026 /* Copyback sequence modulation or stateful scrub changes if needed */ 5027 if (copyback) { 5028 m->m_flags &= ~M_HASH; 5029 m_copyback(m, off, sizeof(*th), (caddr_t)th); 5030 } 5031 5032 pfsync_update_state(*state); 5033 error = PF_PASS; 5034 done: 5035 if (*state) 5036 lockmgr(&(*state)->lk, LK_RELEASE); 5037 return (error); 5038 } 5039 5040 /* 5041 * Test UDP connection state. Caller must hold the state locked. 5042 */ 5043 int 5044 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif, 5045 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 5046 { 5047 struct pf_state_peer *src, *dst; 5048 struct pf_state_key_cmp key; 5049 struct udphdr *uh = pd->hdr.udp; 5050 5051 bzero(&key, sizeof(key)); 5052 key.af = pd->af; 5053 key.proto = IPPROTO_UDP; 5054 if (direction == PF_IN) { /* wire side, straight */ 5055 PF_ACPY(&key.addr[0], pd->src, key.af); 5056 PF_ACPY(&key.addr[1], pd->dst, key.af); 5057 key.port[0] = uh->uh_sport; 5058 key.port[1] = uh->uh_dport; 5059 } else { /* stack side, reverse */ 5060 PF_ACPY(&key.addr[1], pd->src, key.af); 5061 PF_ACPY(&key.addr[0], pd->dst, key.af); 5062 key.port[1] = uh->uh_sport; 5063 key.port[0] = uh->uh_dport; 5064 } 5065 5066 STATE_LOOKUP(kif, &key, direction, *state, m); 5067 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5068 5069 if (direction == (*state)->direction) { 5070 src = &(*state)->src; 5071 dst = &(*state)->dst; 5072 } else { 5073 src = &(*state)->dst; 5074 dst = &(*state)->src; 5075 } 5076 5077 /* update states */ 5078 if (src->state < PFUDPS_SINGLE) 5079 src->state = PFUDPS_SINGLE; 5080 if (dst->state == PFUDPS_SINGLE) 5081 dst->state = PFUDPS_MULTIPLE; 5082 5083 /* update expire time */ 5084 (*state)->expire = time_second; 5085 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 5086 (*state)->timeout = PFTM_UDP_MULTIPLE; 5087 else 5088 (*state)->timeout = PFTM_UDP_SINGLE; 5089 5090 /* translate source/destination address, if necessary */ 5091 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5092 struct pf_state_key *nk = (*state)->key[pd->didx]; 5093 5094 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5095 nk->port[pd->sidx] != uh->uh_sport) { 5096 /* 5097 * The translated source address may be completely 5098 * unrelated to the saved link header, make sure 5099 * a bridge doesn't try to use it. 5100 */ 5101 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 5102 m->m_flags &= ~M_HASH; 5103 pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum, 5104 &uh->uh_sum, &nk->addr[pd->sidx], 5105 nk->port[pd->sidx], 1, pd->af); 5106 } 5107 5108 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5109 nk->port[pd->didx] != uh->uh_dport) { 5110 /* 5111 * If we don't redispatch the packet will go into 5112 * the protocol stack on the wrong cpu for the 5113 * post-translated address. 5114 */ 5115 m->m_flags &= ~M_HASH; 5116 pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum, 5117 &uh->uh_sum, &nk->addr[pd->didx], 5118 nk->port[pd->didx], 1, pd->af); 5119 } 5120 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 5121 } 5122 5123 pfsync_update_state(*state); 5124 lockmgr(&(*state)->lk, LK_RELEASE); 5125 return (PF_PASS); 5126 } 5127 5128 /* 5129 * Test ICMP connection state. Caller must hold the state locked. 5130 */ 5131 int 5132 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif, 5133 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 5134 u_short *reason) 5135 { 5136 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 5137 u_int16_t icmpid = 0, *icmpsum = NULL; 5138 u_int8_t icmptype = 0; 5139 int state_icmp = 0; 5140 int error; 5141 struct pf_state_key_cmp key; 5142 5143 bzero(&key, sizeof(key)); 5144 5145 switch (pd->proto) { 5146 #ifdef INET 5147 case IPPROTO_ICMP: 5148 icmptype = pd->hdr.icmp->icmp_type; 5149 icmpid = pd->hdr.icmp->icmp_id; 5150 icmpsum = &pd->hdr.icmp->icmp_cksum; 5151 5152 if (icmptype == ICMP_UNREACH || 5153 icmptype == ICMP_SOURCEQUENCH || 5154 icmptype == ICMP_REDIRECT || 5155 icmptype == ICMP_TIMXCEED || 5156 icmptype == ICMP_PARAMPROB) 5157 state_icmp++; 5158 break; 5159 #endif /* INET */ 5160 #ifdef INET6 5161 case IPPROTO_ICMPV6: 5162 icmptype = pd->hdr.icmp6->icmp6_type; 5163 icmpid = pd->hdr.icmp6->icmp6_id; 5164 icmpsum = &pd->hdr.icmp6->icmp6_cksum; 5165 5166 if (icmptype == ICMP6_DST_UNREACH || 5167 icmptype == ICMP6_PACKET_TOO_BIG || 5168 icmptype == ICMP6_TIME_EXCEEDED || 5169 icmptype == ICMP6_PARAM_PROB) 5170 state_icmp++; 5171 break; 5172 #endif /* INET6 */ 5173 } 5174 5175 if (!state_icmp) { 5176 5177 /* 5178 * ICMP query/reply message not related to a TCP/UDP packet. 5179 * Search for an ICMP state. 5180 */ 5181 key.af = pd->af; 5182 key.proto = pd->proto; 5183 key.port[0] = key.port[1] = icmpid; 5184 if (direction == PF_IN) { /* wire side, straight */ 5185 PF_ACPY(&key.addr[0], pd->src, key.af); 5186 PF_ACPY(&key.addr[1], pd->dst, key.af); 5187 } else { /* stack side, reverse */ 5188 PF_ACPY(&key.addr[1], pd->src, key.af); 5189 PF_ACPY(&key.addr[0], pd->dst, key.af); 5190 } 5191 5192 STATE_LOOKUP(kif, &key, direction, *state, m); 5193 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5194 5195 (*state)->expire = time_second; 5196 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 5197 5198 /* translate source/destination address, if necessary */ 5199 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5200 struct pf_state_key *nk = (*state)->key[pd->didx]; 5201 5202 switch (pd->af) { 5203 #ifdef INET 5204 case AF_INET: 5205 if (PF_ANEQ(pd->src, 5206 &nk->addr[pd->sidx], AF_INET)) 5207 pf_change_a(&saddr->v4.s_addr, 5208 pd->ip_sum, 5209 nk->addr[pd->sidx].v4.s_addr, 0); 5210 5211 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 5212 AF_INET)) 5213 pf_change_a(&daddr->v4.s_addr, 5214 pd->ip_sum, 5215 nk->addr[pd->didx].v4.s_addr, 0); 5216 5217 if (nk->port[0] != 5218 pd->hdr.icmp->icmp_id) { 5219 pd->hdr.icmp->icmp_cksum = 5220 pf_cksum_fixup( 5221 pd->hdr.icmp->icmp_cksum, icmpid, 5222 nk->port[pd->sidx], 0); 5223 pd->hdr.icmp->icmp_id = 5224 nk->port[pd->sidx]; 5225 } 5226 5227 m->m_flags &= ~M_HASH; 5228 m_copyback(m, off, ICMP_MINLEN, 5229 (caddr_t)pd->hdr.icmp); 5230 break; 5231 #endif /* INET */ 5232 #ifdef INET6 5233 case AF_INET6: 5234 if (PF_ANEQ(pd->src, 5235 &nk->addr[pd->sidx], AF_INET6)) 5236 pf_change_a6(saddr, 5237 &pd->hdr.icmp6->icmp6_cksum, 5238 &nk->addr[pd->sidx], 0); 5239 5240 if (PF_ANEQ(pd->dst, 5241 &nk->addr[pd->didx], AF_INET6)) 5242 pf_change_a6(daddr, 5243 &pd->hdr.icmp6->icmp6_cksum, 5244 &nk->addr[pd->didx], 0); 5245 5246 m->m_flags &= ~M_HASH; 5247 m_copyback(m, off, 5248 sizeof(struct icmp6_hdr), 5249 (caddr_t)pd->hdr.icmp6); 5250 break; 5251 #endif /* INET6 */ 5252 } 5253 } 5254 } else { 5255 /* 5256 * ICMP error message in response to a TCP/UDP packet. 5257 * Extract the inner TCP/UDP header and search for that state. 5258 */ 5259 5260 struct pf_pdesc pd2; 5261 #ifdef INET 5262 struct ip h2; 5263 #endif /* INET */ 5264 #ifdef INET6 5265 struct ip6_hdr h2_6; 5266 int terminal = 0; 5267 #endif /* INET6 */ 5268 int ipoff2; 5269 int off2; 5270 5271 pd2.not_cpu_localized = 1; 5272 pd2.af = pd->af; 5273 /* Payload packet is from the opposite direction. */ 5274 pd2.sidx = (direction == PF_IN) ? 1 : 0; 5275 pd2.didx = (direction == PF_IN) ? 0 : 1; 5276 switch (pd->af) { 5277 #ifdef INET 5278 case AF_INET: 5279 /* offset of h2 in mbuf chain */ 5280 ipoff2 = off + ICMP_MINLEN; 5281 5282 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 5283 NULL, reason, pd2.af)) { 5284 DPFPRINTF(PF_DEBUG_MISC, 5285 ("pf: ICMP error message too short " 5286 "(ip)\n")); 5287 FAIL (PF_DROP); 5288 } 5289 /* 5290 * ICMP error messages don't refer to non-first 5291 * fragments 5292 */ 5293 if (h2.ip_off & htons(IP_OFFMASK)) { 5294 REASON_SET(reason, PFRES_FRAG); 5295 FAIL (PF_DROP); 5296 } 5297 5298 /* offset of protocol header that follows h2 */ 5299 off2 = ipoff2 + (h2.ip_hl << 2); 5300 5301 pd2.proto = h2.ip_p; 5302 pd2.src = (struct pf_addr *)&h2.ip_src; 5303 pd2.dst = (struct pf_addr *)&h2.ip_dst; 5304 pd2.ip_sum = &h2.ip_sum; 5305 break; 5306 #endif /* INET */ 5307 #ifdef INET6 5308 case AF_INET6: 5309 ipoff2 = off + sizeof(struct icmp6_hdr); 5310 5311 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 5312 NULL, reason, pd2.af)) { 5313 DPFPRINTF(PF_DEBUG_MISC, 5314 ("pf: ICMP error message too short " 5315 "(ip6)\n")); 5316 FAIL (PF_DROP); 5317 } 5318 pd2.proto = h2_6.ip6_nxt; 5319 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 5320 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 5321 pd2.ip_sum = NULL; 5322 off2 = ipoff2 + sizeof(h2_6); 5323 do { 5324 switch (pd2.proto) { 5325 case IPPROTO_FRAGMENT: 5326 /* 5327 * ICMPv6 error messages for 5328 * non-first fragments 5329 */ 5330 REASON_SET(reason, PFRES_FRAG); 5331 FAIL (PF_DROP); 5332 case IPPROTO_AH: 5333 case IPPROTO_HOPOPTS: 5334 case IPPROTO_ROUTING: 5335 case IPPROTO_DSTOPTS: { 5336 /* get next header and header length */ 5337 struct ip6_ext opt6; 5338 5339 if (!pf_pull_hdr(m, off2, &opt6, 5340 sizeof(opt6), NULL, reason, 5341 pd2.af)) { 5342 DPFPRINTF(PF_DEBUG_MISC, 5343 ("pf: ICMPv6 short opt\n")); 5344 FAIL (PF_DROP); 5345 } 5346 if (pd2.proto == IPPROTO_AH) 5347 off2 += (opt6.ip6e_len + 2) * 4; 5348 else 5349 off2 += (opt6.ip6e_len + 1) * 8; 5350 pd2.proto = opt6.ip6e_nxt; 5351 /* goto the next header */ 5352 break; 5353 } 5354 default: 5355 terminal++; 5356 break; 5357 } 5358 } while (!terminal); 5359 break; 5360 #endif /* INET6 */ 5361 default: 5362 DPFPRINTF(PF_DEBUG_MISC, 5363 ("pf: ICMP AF %d unknown (ip6)\n", pd->af)); 5364 FAIL (PF_DROP); 5365 break; 5366 } 5367 5368 switch (pd2.proto) { 5369 case IPPROTO_TCP: { 5370 struct tcphdr th; 5371 u_int32_t seq; 5372 struct pf_state_peer *src, *dst; 5373 u_int8_t dws; 5374 int copyback = 0; 5375 5376 /* 5377 * Only the first 8 bytes of the TCP header can be 5378 * expected. Don't access any TCP header fields after 5379 * th_seq, an ackskew test is not possible. 5380 */ 5381 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 5382 pd2.af)) { 5383 DPFPRINTF(PF_DEBUG_MISC, 5384 ("pf: ICMP error message too short " 5385 "(tcp)\n")); 5386 FAIL (PF_DROP); 5387 } 5388 5389 key.af = pd2.af; 5390 key.proto = IPPROTO_TCP; 5391 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5392 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5393 key.port[pd2.sidx] = th.th_sport; 5394 key.port[pd2.didx] = th.th_dport; 5395 5396 STATE_LOOKUP(kif, &key, direction, *state, m); 5397 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5398 5399 if (direction == (*state)->direction) { 5400 src = &(*state)->dst; 5401 dst = &(*state)->src; 5402 } else { 5403 src = &(*state)->src; 5404 dst = &(*state)->dst; 5405 } 5406 5407 if (src->wscale && dst->wscale) 5408 dws = dst->wscale & PF_WSCALE_MASK; 5409 else 5410 dws = 0; 5411 5412 /* Demodulate sequence number */ 5413 seq = ntohl(th.th_seq) - src->seqdiff; 5414 if (src->seqdiff) { 5415 pf_change_a(&th.th_seq, icmpsum, 5416 htonl(seq), 0); 5417 copyback = 1; 5418 } 5419 5420 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 5421 (!SEQ_GEQ(src->seqhi, seq) || 5422 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 5423 if (pf_status.debug >= PF_DEBUG_MISC) { 5424 kprintf("pf: BAD ICMP %d:%d ", 5425 icmptype, pd->hdr.icmp->icmp_code); 5426 pf_print_host(pd->src, 0, pd->af); 5427 kprintf(" -> "); 5428 pf_print_host(pd->dst, 0, pd->af); 5429 kprintf(" state: "); 5430 pf_print_state(*state); 5431 kprintf(" seq=%u\n", seq); 5432 } 5433 REASON_SET(reason, PFRES_BADSTATE); 5434 FAIL (PF_DROP); 5435 } else { 5436 if (pf_status.debug >= PF_DEBUG_MISC) { 5437 kprintf("pf: OK ICMP %d:%d ", 5438 icmptype, pd->hdr.icmp->icmp_code); 5439 pf_print_host(pd->src, 0, pd->af); 5440 kprintf(" -> "); 5441 pf_print_host(pd->dst, 0, pd->af); 5442 kprintf(" state: "); 5443 pf_print_state(*state); 5444 kprintf(" seq=%u\n", seq); 5445 } 5446 } 5447 5448 /* translate source/destination address, if necessary */ 5449 if ((*state)->key[PF_SK_WIRE] != 5450 (*state)->key[PF_SK_STACK]) { 5451 struct pf_state_key *nk = 5452 (*state)->key[pd->didx]; 5453 5454 if (PF_ANEQ(pd2.src, 5455 &nk->addr[pd2.sidx], pd2.af) || 5456 nk->port[pd2.sidx] != th.th_sport) 5457 pf_change_icmp(pd2.src, &th.th_sport, 5458 daddr, &nk->addr[pd2.sidx], 5459 nk->port[pd2.sidx], NULL, 5460 pd2.ip_sum, icmpsum, 5461 pd->ip_sum, 0, pd2.af); 5462 5463 if (PF_ANEQ(pd2.dst, 5464 &nk->addr[pd2.didx], pd2.af) || 5465 nk->port[pd2.didx] != th.th_dport) 5466 pf_change_icmp(pd2.dst, &th.th_dport, 5467 NULL, /* XXX Inbound NAT? */ 5468 &nk->addr[pd2.didx], 5469 nk->port[pd2.didx], NULL, 5470 pd2.ip_sum, icmpsum, 5471 pd->ip_sum, 0, pd2.af); 5472 copyback = 1; 5473 } 5474 5475 if (copyback) { 5476 switch (pd2.af) { 5477 #ifdef INET 5478 case AF_INET: 5479 m_copyback(m, off, ICMP_MINLEN, 5480 (caddr_t)pd->hdr.icmp); 5481 m_copyback(m, ipoff2, sizeof(h2), 5482 (caddr_t)&h2); 5483 break; 5484 #endif /* INET */ 5485 #ifdef INET6 5486 case AF_INET6: 5487 m_copyback(m, off, 5488 sizeof(struct icmp6_hdr), 5489 (caddr_t)pd->hdr.icmp6); 5490 m_copyback(m, ipoff2, sizeof(h2_6), 5491 (caddr_t)&h2_6); 5492 break; 5493 #endif /* INET6 */ 5494 } 5495 m->m_flags &= ~M_HASH; 5496 m_copyback(m, off2, 8, (caddr_t)&th); 5497 } 5498 break; 5499 } 5500 case IPPROTO_UDP: { 5501 struct udphdr uh; 5502 5503 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 5504 NULL, reason, pd2.af)) { 5505 DPFPRINTF(PF_DEBUG_MISC, 5506 ("pf: ICMP error message too short " 5507 "(udp)\n")); 5508 return (PF_DROP); 5509 } 5510 5511 key.af = pd2.af; 5512 key.proto = IPPROTO_UDP; 5513 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5514 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5515 key.port[pd2.sidx] = uh.uh_sport; 5516 key.port[pd2.didx] = uh.uh_dport; 5517 5518 STATE_LOOKUP(kif, &key, direction, *state, m); 5519 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5520 5521 /* translate source/destination address, if necessary */ 5522 if ((*state)->key[PF_SK_WIRE] != 5523 (*state)->key[PF_SK_STACK]) { 5524 struct pf_state_key *nk = 5525 (*state)->key[pd->didx]; 5526 5527 if (PF_ANEQ(pd2.src, 5528 &nk->addr[pd2.sidx], pd2.af) || 5529 nk->port[pd2.sidx] != uh.uh_sport) 5530 pf_change_icmp(pd2.src, &uh.uh_sport, 5531 daddr, &nk->addr[pd2.sidx], 5532 nk->port[pd2.sidx], &uh.uh_sum, 5533 pd2.ip_sum, icmpsum, 5534 pd->ip_sum, 1, pd2.af); 5535 5536 if (PF_ANEQ(pd2.dst, 5537 &nk->addr[pd2.didx], pd2.af) || 5538 nk->port[pd2.didx] != uh.uh_dport) 5539 pf_change_icmp(pd2.dst, &uh.uh_dport, 5540 NULL, /* XXX Inbound NAT? */ 5541 &nk->addr[pd2.didx], 5542 nk->port[pd2.didx], &uh.uh_sum, 5543 pd2.ip_sum, icmpsum, 5544 pd->ip_sum, 1, pd2.af); 5545 5546 switch (pd2.af) { 5547 #ifdef INET 5548 case AF_INET: 5549 m_copyback(m, off, ICMP_MINLEN, 5550 (caddr_t)pd->hdr.icmp); 5551 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5552 break; 5553 #endif /* INET */ 5554 #ifdef INET6 5555 case AF_INET6: 5556 m_copyback(m, off, 5557 sizeof(struct icmp6_hdr), 5558 (caddr_t)pd->hdr.icmp6); 5559 m_copyback(m, ipoff2, sizeof(h2_6), 5560 (caddr_t)&h2_6); 5561 break; 5562 #endif /* INET6 */ 5563 } 5564 m->m_flags &= ~M_HASH; 5565 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 5566 } 5567 break; 5568 } 5569 #ifdef INET 5570 case IPPROTO_ICMP: { 5571 struct icmp iih; 5572 5573 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 5574 NULL, reason, pd2.af)) { 5575 DPFPRINTF(PF_DEBUG_MISC, 5576 ("pf: ICMP error message too short i" 5577 "(icmp)\n")); 5578 return (PF_DROP); 5579 } 5580 5581 key.af = pd2.af; 5582 key.proto = IPPROTO_ICMP; 5583 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5584 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5585 key.port[0] = key.port[1] = iih.icmp_id; 5586 5587 STATE_LOOKUP(kif, &key, direction, *state, m); 5588 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5589 5590 /* translate source/destination address, if necessary */ 5591 if ((*state)->key[PF_SK_WIRE] != 5592 (*state)->key[PF_SK_STACK]) { 5593 struct pf_state_key *nk = 5594 (*state)->key[pd->didx]; 5595 5596 if (PF_ANEQ(pd2.src, 5597 &nk->addr[pd2.sidx], pd2.af) || 5598 nk->port[pd2.sidx] != iih.icmp_id) 5599 pf_change_icmp(pd2.src, &iih.icmp_id, 5600 daddr, &nk->addr[pd2.sidx], 5601 nk->port[pd2.sidx], NULL, 5602 pd2.ip_sum, icmpsum, 5603 pd->ip_sum, 0, AF_INET); 5604 5605 if (PF_ANEQ(pd2.dst, 5606 &nk->addr[pd2.didx], pd2.af) || 5607 nk->port[pd2.didx] != iih.icmp_id) 5608 pf_change_icmp(pd2.dst, &iih.icmp_id, 5609 NULL, /* XXX Inbound NAT? */ 5610 &nk->addr[pd2.didx], 5611 nk->port[pd2.didx], NULL, 5612 pd2.ip_sum, icmpsum, 5613 pd->ip_sum, 0, AF_INET); 5614 5615 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 5616 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5617 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 5618 m->m_flags &= ~M_HASH; 5619 } 5620 break; 5621 } 5622 #endif /* INET */ 5623 #ifdef INET6 5624 case IPPROTO_ICMPV6: { 5625 struct icmp6_hdr iih; 5626 5627 if (!pf_pull_hdr(m, off2, &iih, 5628 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 5629 DPFPRINTF(PF_DEBUG_MISC, 5630 ("pf: ICMP error message too short " 5631 "(icmp6)\n")); 5632 FAIL (PF_DROP); 5633 } 5634 5635 key.af = pd2.af; 5636 key.proto = IPPROTO_ICMPV6; 5637 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5638 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5639 key.port[0] = key.port[1] = iih.icmp6_id; 5640 5641 STATE_LOOKUP(kif, &key, direction, *state, m); 5642 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5643 5644 /* translate source/destination address, if necessary */ 5645 if ((*state)->key[PF_SK_WIRE] != 5646 (*state)->key[PF_SK_STACK]) { 5647 struct pf_state_key *nk = 5648 (*state)->key[pd->didx]; 5649 5650 if (PF_ANEQ(pd2.src, 5651 &nk->addr[pd2.sidx], pd2.af) || 5652 nk->port[pd2.sidx] != iih.icmp6_id) 5653 pf_change_icmp(pd2.src, &iih.icmp6_id, 5654 daddr, &nk->addr[pd2.sidx], 5655 nk->port[pd2.sidx], NULL, 5656 pd2.ip_sum, icmpsum, 5657 pd->ip_sum, 0, AF_INET6); 5658 5659 if (PF_ANEQ(pd2.dst, 5660 &nk->addr[pd2.didx], pd2.af) || 5661 nk->port[pd2.didx] != iih.icmp6_id) 5662 pf_change_icmp(pd2.dst, &iih.icmp6_id, 5663 NULL, /* XXX Inbound NAT? */ 5664 &nk->addr[pd2.didx], 5665 nk->port[pd2.didx], NULL, 5666 pd2.ip_sum, icmpsum, 5667 pd->ip_sum, 0, AF_INET6); 5668 5669 m_copyback(m, off, sizeof(struct icmp6_hdr), 5670 (caddr_t)pd->hdr.icmp6); 5671 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 5672 m_copyback(m, off2, sizeof(struct icmp6_hdr), 5673 (caddr_t)&iih); 5674 m->m_flags &= ~M_HASH; 5675 } 5676 break; 5677 } 5678 #endif /* INET6 */ 5679 default: { 5680 key.af = pd2.af; 5681 key.proto = pd2.proto; 5682 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5683 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5684 key.port[0] = key.port[1] = 0; 5685 5686 STATE_LOOKUP(kif, &key, direction, *state, m); 5687 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5688 5689 /* translate source/destination address, if necessary */ 5690 if ((*state)->key[PF_SK_WIRE] != 5691 (*state)->key[PF_SK_STACK]) { 5692 struct pf_state_key *nk = 5693 (*state)->key[pd->didx]; 5694 5695 if (PF_ANEQ(pd2.src, 5696 &nk->addr[pd2.sidx], pd2.af)) 5697 pf_change_icmp(pd2.src, NULL, daddr, 5698 &nk->addr[pd2.sidx], 0, NULL, 5699 pd2.ip_sum, icmpsum, 5700 pd->ip_sum, 0, pd2.af); 5701 5702 if (PF_ANEQ(pd2.dst, 5703 &nk->addr[pd2.didx], pd2.af)) 5704 pf_change_icmp(pd2.src, NULL, 5705 NULL, /* XXX Inbound NAT? */ 5706 &nk->addr[pd2.didx], 0, NULL, 5707 pd2.ip_sum, icmpsum, 5708 pd->ip_sum, 0, pd2.af); 5709 5710 switch (pd2.af) { 5711 #ifdef INET 5712 case AF_INET: 5713 m_copyback(m, off, ICMP_MINLEN, 5714 (caddr_t)pd->hdr.icmp); 5715 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5716 m->m_flags &= ~M_HASH; 5717 break; 5718 #endif /* INET */ 5719 #ifdef INET6 5720 case AF_INET6: 5721 m_copyback(m, off, 5722 sizeof(struct icmp6_hdr), 5723 (caddr_t)pd->hdr.icmp6); 5724 m_copyback(m, ipoff2, sizeof(h2_6), 5725 (caddr_t)&h2_6); 5726 m->m_flags &= ~M_HASH; 5727 break; 5728 #endif /* INET6 */ 5729 } 5730 } 5731 break; 5732 } 5733 } 5734 } 5735 5736 pfsync_update_state(*state); 5737 error = PF_PASS; 5738 done: 5739 if (*state) 5740 lockmgr(&(*state)->lk, LK_RELEASE); 5741 return (error); 5742 } 5743 5744 /* 5745 * Test other connection state. Caller must hold the state locked. 5746 */ 5747 int 5748 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif, 5749 struct mbuf *m, struct pf_pdesc *pd) 5750 { 5751 struct pf_state_peer *src, *dst; 5752 struct pf_state_key_cmp key; 5753 5754 bzero(&key, sizeof(key)); 5755 key.af = pd->af; 5756 key.proto = pd->proto; 5757 if (direction == PF_IN) { 5758 PF_ACPY(&key.addr[0], pd->src, key.af); 5759 PF_ACPY(&key.addr[1], pd->dst, key.af); 5760 key.port[0] = key.port[1] = 0; 5761 } else { 5762 PF_ACPY(&key.addr[1], pd->src, key.af); 5763 PF_ACPY(&key.addr[0], pd->dst, key.af); 5764 key.port[1] = key.port[0] = 0; 5765 } 5766 5767 STATE_LOOKUP(kif, &key, direction, *state, m); 5768 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5769 5770 if (direction == (*state)->direction) { 5771 src = &(*state)->src; 5772 dst = &(*state)->dst; 5773 } else { 5774 src = &(*state)->dst; 5775 dst = &(*state)->src; 5776 } 5777 5778 /* update states */ 5779 if (src->state < PFOTHERS_SINGLE) 5780 src->state = PFOTHERS_SINGLE; 5781 if (dst->state == PFOTHERS_SINGLE) 5782 dst->state = PFOTHERS_MULTIPLE; 5783 5784 /* update expire time */ 5785 (*state)->expire = time_second; 5786 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 5787 (*state)->timeout = PFTM_OTHER_MULTIPLE; 5788 else 5789 (*state)->timeout = PFTM_OTHER_SINGLE; 5790 5791 /* translate source/destination address, if necessary */ 5792 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5793 struct pf_state_key *nk = (*state)->key[pd->didx]; 5794 5795 KKASSERT(nk); 5796 KKASSERT(pd); 5797 KKASSERT(pd->src); 5798 KKASSERT(pd->dst); 5799 switch (pd->af) { 5800 #ifdef INET 5801 case AF_INET: 5802 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5803 pf_change_a(&pd->src->v4.s_addr, 5804 pd->ip_sum, 5805 nk->addr[pd->sidx].v4.s_addr, 5806 0); 5807 5808 5809 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5810 pf_change_a(&pd->dst->v4.s_addr, 5811 pd->ip_sum, 5812 nk->addr[pd->didx].v4.s_addr, 5813 0); 5814 5815 break; 5816 #endif /* INET */ 5817 #ifdef INET6 5818 case AF_INET6: 5819 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5820 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5821 5822 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5823 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5824 #endif /* INET6 */ 5825 } 5826 } 5827 5828 pfsync_update_state(*state); 5829 lockmgr(&(*state)->lk, LK_RELEASE); 5830 return (PF_PASS); 5831 } 5832 5833 /* 5834 * ipoff and off are measured from the start of the mbuf chain. 5835 * h must be at "ipoff" on the mbuf chain. 5836 */ 5837 void * 5838 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 5839 u_short *actionp, u_short *reasonp, sa_family_t af) 5840 { 5841 switch (af) { 5842 #ifdef INET 5843 case AF_INET: { 5844 struct ip *h = mtod(m, struct ip *); 5845 u_int16_t fragoff = (h->ip_off & IP_OFFMASK) << 3; 5846 5847 if (fragoff) { 5848 if (fragoff >= len) 5849 ACTION_SET(actionp, PF_PASS); 5850 else { 5851 ACTION_SET(actionp, PF_DROP); 5852 REASON_SET(reasonp, PFRES_FRAG); 5853 } 5854 return (NULL); 5855 } 5856 if (m->m_pkthdr.len < off + len || 5857 h->ip_len < off + len) { 5858 ACTION_SET(actionp, PF_DROP); 5859 REASON_SET(reasonp, PFRES_SHORT); 5860 return (NULL); 5861 } 5862 break; 5863 } 5864 #endif /* INET */ 5865 #ifdef INET6 5866 case AF_INET6: { 5867 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 5868 5869 if (m->m_pkthdr.len < off + len || 5870 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 5871 (unsigned)(off + len)) { 5872 ACTION_SET(actionp, PF_DROP); 5873 REASON_SET(reasonp, PFRES_SHORT); 5874 return (NULL); 5875 } 5876 break; 5877 } 5878 #endif /* INET6 */ 5879 } 5880 m_copydata(m, off, len, p); 5881 return (p); 5882 } 5883 5884 int 5885 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif) 5886 { 5887 struct sockaddr_in *dst; 5888 int ret = 1; 5889 int check_mpath; 5890 #ifdef INET6 5891 struct sockaddr_in6 *dst6; 5892 struct route_in6 ro; 5893 #else 5894 struct route ro; 5895 #endif 5896 struct radix_node *rn; 5897 struct rtentry *rt; 5898 struct ifnet *ifp; 5899 5900 check_mpath = 0; 5901 bzero(&ro, sizeof(ro)); 5902 switch (af) { 5903 case AF_INET: 5904 dst = satosin(&ro.ro_dst); 5905 dst->sin_family = AF_INET; 5906 dst->sin_len = sizeof(*dst); 5907 dst->sin_addr = addr->v4; 5908 break; 5909 #ifdef INET6 5910 case AF_INET6: 5911 /* 5912 * Skip check for addresses with embedded interface scope, 5913 * as they would always match anyway. 5914 */ 5915 if (IN6_IS_SCOPE_EMBED(&addr->v6)) 5916 goto out; 5917 dst6 = (struct sockaddr_in6 *)&ro.ro_dst; 5918 dst6->sin6_family = AF_INET6; 5919 dst6->sin6_len = sizeof(*dst6); 5920 dst6->sin6_addr = addr->v6; 5921 break; 5922 #endif /* INET6 */ 5923 default: 5924 return (0); 5925 } 5926 5927 /* Skip checks for ipsec interfaces */ 5928 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 5929 goto out; 5930 5931 rtalloc_ign((struct route *)&ro, 0); 5932 5933 if (ro.ro_rt != NULL) { 5934 /* No interface given, this is a no-route check */ 5935 if (kif == NULL) 5936 goto out; 5937 5938 if (kif->pfik_ifp == NULL) { 5939 ret = 0; 5940 goto out; 5941 } 5942 5943 /* Perform uRPF check if passed input interface */ 5944 ret = 0; 5945 rn = (struct radix_node *)ro.ro_rt; 5946 do { 5947 rt = (struct rtentry *)rn; 5948 ifp = rt->rt_ifp; 5949 5950 if (kif->pfik_ifp == ifp) 5951 ret = 1; 5952 rn = NULL; 5953 } while (check_mpath == 1 && rn != NULL && ret == 0); 5954 } else 5955 ret = 0; 5956 out: 5957 if (ro.ro_rt != NULL) 5958 RTFREE(ro.ro_rt); 5959 return (ret); 5960 } 5961 5962 int 5963 pf_rtlabel_match(struct pf_addr *addr, sa_family_t af, struct pf_addr_wrap *aw) 5964 { 5965 struct sockaddr_in *dst; 5966 #ifdef INET6 5967 struct sockaddr_in6 *dst6; 5968 struct route_in6 ro; 5969 #else 5970 struct route ro; 5971 #endif 5972 int ret = 0; 5973 5974 ASSERT_LWKT_TOKEN_HELD(&pf_token); 5975 5976 bzero(&ro, sizeof(ro)); 5977 switch (af) { 5978 case AF_INET: 5979 dst = satosin(&ro.ro_dst); 5980 dst->sin_family = AF_INET; 5981 dst->sin_len = sizeof(*dst); 5982 dst->sin_addr = addr->v4; 5983 break; 5984 #ifdef INET6 5985 case AF_INET6: 5986 dst6 = (struct sockaddr_in6 *)&ro.ro_dst; 5987 dst6->sin6_family = AF_INET6; 5988 dst6->sin6_len = sizeof(*dst6); 5989 dst6->sin6_addr = addr->v6; 5990 break; 5991 #endif /* INET6 */ 5992 default: 5993 return (0); 5994 } 5995 5996 rtalloc_ign((struct route *)&ro, (RTF_CLONING | RTF_PRCLONING)); 5997 5998 if (ro.ro_rt != NULL) { 5999 RTFREE(ro.ro_rt); 6000 } 6001 6002 return (ret); 6003 } 6004 6005 #ifdef INET 6006 void 6007 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, 6008 struct pf_state *s, struct pf_pdesc *pd) 6009 { 6010 struct mbuf *m0, *m1; 6011 struct route iproute; 6012 struct route *ro = NULL; 6013 struct sockaddr_in *dst; 6014 struct ip *ip; 6015 struct ifnet *ifp = NULL; 6016 struct pf_addr naddr; 6017 struct pf_src_node *sn = NULL; 6018 int error = 0; 6019 int sw_csum; 6020 6021 ASSERT_LWKT_TOKEN_HELD(&pf_token); 6022 6023 if (m == NULL || *m == NULL || r == NULL || 6024 (dir != PF_IN && dir != PF_OUT) || oifp == NULL) 6025 panic("pf_route: invalid parameters"); 6026 6027 if (((*m)->m_pkthdr.fw_flags & PF_MBUF_ROUTED) == 0) { 6028 (*m)->m_pkthdr.fw_flags |= PF_MBUF_ROUTED; 6029 (*m)->m_pkthdr.pf.routed = 1; 6030 } else { 6031 if ((*m)->m_pkthdr.pf.routed++ > 3) { 6032 m0 = *m; 6033 *m = NULL; 6034 goto bad; 6035 } 6036 } 6037 6038 if (r->rt == PF_DUPTO) { 6039 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { 6040 return; 6041 } 6042 } else { 6043 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 6044 return; 6045 } 6046 m0 = *m; 6047 } 6048 6049 if (m0->m_len < sizeof(struct ip)) { 6050 DPFPRINTF(PF_DEBUG_URGENT, 6051 ("pf_route: m0->m_len < sizeof(struct ip)\n")); 6052 goto bad; 6053 } 6054 6055 ip = mtod(m0, struct ip *); 6056 6057 ro = &iproute; 6058 bzero((caddr_t)ro, sizeof(*ro)); 6059 dst = satosin(&ro->ro_dst); 6060 dst->sin_family = AF_INET; 6061 dst->sin_len = sizeof(*dst); 6062 dst->sin_addr = ip->ip_dst; 6063 6064 if (r->rt == PF_FASTROUTE) { 6065 rtalloc(ro); 6066 if (ro->ro_rt == 0) { 6067 ipstat.ips_noroute++; 6068 goto bad; 6069 } 6070 6071 ifp = ro->ro_rt->rt_ifp; 6072 ro->ro_rt->rt_use++; 6073 6074 if (ro->ro_rt->rt_flags & RTF_GATEWAY) 6075 dst = satosin(ro->ro_rt->rt_gateway); 6076 } else { 6077 if (TAILQ_EMPTY(&r->rpool.list)) { 6078 DPFPRINTF(PF_DEBUG_URGENT, 6079 ("pf_route: TAILQ_EMPTY(&r->rpool.list)\n")); 6080 goto bad; 6081 } 6082 if (s == NULL) { 6083 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 6084 &naddr, NULL, &sn); 6085 if (!PF_AZERO(&naddr, AF_INET)) 6086 dst->sin_addr.s_addr = naddr.v4.s_addr; 6087 ifp = r->rpool.cur->kif ? 6088 r->rpool.cur->kif->pfik_ifp : NULL; 6089 } else { 6090 if (!PF_AZERO(&s->rt_addr, AF_INET)) 6091 dst->sin_addr.s_addr = 6092 s->rt_addr.v4.s_addr; 6093 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6094 } 6095 } 6096 if (ifp == NULL) 6097 goto bad; 6098 6099 if (oifp != ifp) { 6100 if (pf_test(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) { 6101 goto bad; 6102 } else if (m0 == NULL) { 6103 goto done; 6104 } 6105 if (m0->m_len < sizeof(struct ip)) { 6106 DPFPRINTF(PF_DEBUG_URGENT, 6107 ("pf_route: m0->m_len < sizeof(struct ip)\n")); 6108 goto bad; 6109 } 6110 ip = mtod(m0, struct ip *); 6111 } 6112 6113 /* Copied from FreeBSD 5.1-CURRENT ip_output. */ 6114 m0->m_pkthdr.csum_flags |= CSUM_IP; 6115 sw_csum = m0->m_pkthdr.csum_flags & ~ifp->if_hwassist; 6116 if (sw_csum & CSUM_DELAY_DATA) { 6117 in_delayed_cksum(m0); 6118 sw_csum &= ~CSUM_DELAY_DATA; 6119 } 6120 m0->m_pkthdr.csum_flags &= ifp->if_hwassist; 6121 m0->m_pkthdr.csum_iphlen = (ip->ip_hl << 2); 6122 6123 /* 6124 * WARNING! We cannot fragment if the packet was modified from an 6125 * original which expected to be using TSO. In this 6126 * situation we pray that the target interface is 6127 * compatible with the originating interface. 6128 */ 6129 if (ip->ip_len <= ifp->if_mtu || 6130 (m0->m_pkthdr.csum_flags & CSUM_TSO) || 6131 ((ifp->if_hwassist & CSUM_FRAGMENT) && 6132 (ip->ip_off & IP_DF) == 0)) { 6133 ip->ip_len = htons(ip->ip_len); 6134 ip->ip_off = htons(ip->ip_off); 6135 ip->ip_sum = 0; 6136 if (sw_csum & CSUM_DELAY_IP) { 6137 /* From KAME */ 6138 if (ip->ip_v == IPVERSION && 6139 (ip->ip_hl << 2) == sizeof(*ip)) { 6140 ip->ip_sum = in_cksum_hdr(ip); 6141 } else { 6142 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 6143 } 6144 } 6145 lwkt_reltoken(&pf_token); 6146 error = ifp->if_output(ifp, m0, sintosa(dst), ro->ro_rt); 6147 lwkt_gettoken(&pf_token); 6148 goto done; 6149 } 6150 6151 /* 6152 * Too large for interface; fragment if possible. 6153 * Must be able to put at least 8 bytes per fragment. 6154 */ 6155 if (ip->ip_off & IP_DF) { 6156 ipstat.ips_cantfrag++; 6157 if (r->rt != PF_DUPTO) { 6158 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 6159 ifp->if_mtu); 6160 goto done; 6161 } else 6162 goto bad; 6163 } 6164 6165 m1 = m0; 6166 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist, sw_csum); 6167 if (error) { 6168 goto bad; 6169 } 6170 6171 for (m0 = m1; m0; m0 = m1) { 6172 m1 = m0->m_nextpkt; 6173 m0->m_nextpkt = 0; 6174 if (error == 0) { 6175 lwkt_reltoken(&pf_token); 6176 error = (*ifp->if_output)(ifp, m0, sintosa(dst), 6177 NULL); 6178 lwkt_gettoken(&pf_token); 6179 } else 6180 m_freem(m0); 6181 } 6182 6183 if (error == 0) 6184 ipstat.ips_fragmented++; 6185 6186 done: 6187 if (r->rt != PF_DUPTO) 6188 *m = NULL; 6189 if (ro == &iproute && ro->ro_rt) 6190 RTFREE(ro->ro_rt); 6191 return; 6192 6193 bad: 6194 m_freem(m0); 6195 goto done; 6196 } 6197 #endif /* INET */ 6198 6199 #ifdef INET6 6200 void 6201 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, 6202 struct pf_state *s, struct pf_pdesc *pd) 6203 { 6204 struct mbuf *m0; 6205 struct route_in6 ip6route; 6206 struct route_in6 *ro; 6207 struct sockaddr_in6 *dst; 6208 struct ip6_hdr *ip6; 6209 struct ifnet *ifp = NULL; 6210 struct pf_addr naddr; 6211 struct pf_src_node *sn = NULL; 6212 6213 if (m == NULL || *m == NULL || r == NULL || 6214 (dir != PF_IN && dir != PF_OUT) || oifp == NULL) 6215 panic("pf_route6: invalid parameters"); 6216 6217 if (((*m)->m_pkthdr.fw_flags & PF_MBUF_ROUTED) == 0) { 6218 (*m)->m_pkthdr.fw_flags |= PF_MBUF_ROUTED; 6219 (*m)->m_pkthdr.pf.routed = 1; 6220 } else { 6221 if ((*m)->m_pkthdr.pf.routed++ > 3) { 6222 m0 = *m; 6223 *m = NULL; 6224 goto bad; 6225 } 6226 } 6227 6228 if (r->rt == PF_DUPTO) { 6229 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) 6230 return; 6231 } else { 6232 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) 6233 return; 6234 m0 = *m; 6235 } 6236 6237 if (m0->m_len < sizeof(struct ip6_hdr)) { 6238 DPFPRINTF(PF_DEBUG_URGENT, 6239 ("pf_route6: m0->m_len < sizeof(struct ip6_hdr)\n")); 6240 goto bad; 6241 } 6242 ip6 = mtod(m0, struct ip6_hdr *); 6243 6244 ro = &ip6route; 6245 bzero((caddr_t)ro, sizeof(*ro)); 6246 dst = (struct sockaddr_in6 *)&ro->ro_dst; 6247 dst->sin6_family = AF_INET6; 6248 dst->sin6_len = sizeof(*dst); 6249 dst->sin6_addr = ip6->ip6_dst; 6250 6251 /* 6252 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags, 6253 * so make sure pf.flags is clear. 6254 * 6255 * Cheat. XXX why only in the v6 case??? 6256 */ 6257 if (r->rt == PF_FASTROUTE) { 6258 m0->m_pkthdr.fw_flags |= PF_MBUF_TAGGED; 6259 m0->m_pkthdr.pf.flags = 0; 6260 /* XXX Re-Check when Upgrading to > 4.4 */ 6261 m0->m_pkthdr.pf.statekey = NULL; 6262 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL); 6263 return; 6264 } 6265 6266 if (TAILQ_EMPTY(&r->rpool.list)) { 6267 DPFPRINTF(PF_DEBUG_URGENT, 6268 ("pf_route6: TAILQ_EMPTY(&r->rpool.list)\n")); 6269 goto bad; 6270 } 6271 if (s == NULL) { 6272 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 6273 &naddr, NULL, &sn); 6274 if (!PF_AZERO(&naddr, AF_INET6)) 6275 PF_ACPY((struct pf_addr *)&dst->sin6_addr, 6276 &naddr, AF_INET6); 6277 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 6278 } else { 6279 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 6280 PF_ACPY((struct pf_addr *)&dst->sin6_addr, 6281 &s->rt_addr, AF_INET6); 6282 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6283 } 6284 if (ifp == NULL) 6285 goto bad; 6286 6287 if (oifp != ifp) { 6288 if (pf_test6(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) { 6289 goto bad; 6290 } else if (m0 == NULL) { 6291 goto done; 6292 } 6293 if (m0->m_len < sizeof(struct ip6_hdr)) { 6294 DPFPRINTF(PF_DEBUG_URGENT, 6295 ("pf_route6: m0->m_len < sizeof(struct ip6_hdr)\n")); 6296 goto bad; 6297 } 6298 ip6 = mtod(m0, struct ip6_hdr *); 6299 } 6300 6301 /* 6302 * If the packet is too large for the outgoing interface, 6303 * send back an icmp6 error. 6304 */ 6305 if (IN6_IS_SCOPE_EMBED(&dst->sin6_addr)) 6306 dst->sin6_addr.s6_addr16[1] = htons(ifp->if_index); 6307 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 6308 nd6_output(ifp, ifp, m0, dst, NULL); 6309 } else { 6310 in6_ifstat_inc(ifp, ifs6_in_toobig); 6311 if (r->rt != PF_DUPTO) 6312 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 6313 else 6314 goto bad; 6315 } 6316 6317 done: 6318 if (r->rt != PF_DUPTO) 6319 *m = NULL; 6320 return; 6321 6322 bad: 6323 m_freem(m0); 6324 goto done; 6325 } 6326 #endif /* INET6 */ 6327 6328 6329 /* 6330 * check protocol (tcp/udp/icmp/icmp6) checksum and set mbuf flag 6331 * off is the offset where the protocol header starts 6332 * len is the total length of protocol header plus payload 6333 * returns 0 when the checksum is valid, otherwise returns 1. 6334 */ 6335 /* 6336 * XXX 6337 * FreeBSD supports cksum offload for the following drivers. 6338 * em(4), gx(4), lge(4), nge(4), ti(4), xl(4) 6339 * If we can make full use of it we would outperform ipfw/ipfilter in 6340 * very heavy traffic. 6341 * I have not tested 'cause I don't have NICs that supports cksum offload. 6342 * (There might be problems. Typical phenomena would be 6343 * 1. No route message for UDP packet. 6344 * 2. No connection acceptance from external hosts regardless of rule set.) 6345 */ 6346 int 6347 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, 6348 sa_family_t af) 6349 { 6350 u_int16_t sum = 0; 6351 int hw_assist = 0; 6352 struct ip *ip; 6353 6354 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 6355 return (1); 6356 if (m->m_pkthdr.len < off + len) 6357 return (1); 6358 6359 switch (p) { 6360 case IPPROTO_TCP: 6361 case IPPROTO_UDP: 6362 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 6363 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 6364 sum = m->m_pkthdr.csum_data; 6365 } else { 6366 ip = mtod(m, struct ip *); 6367 sum = in_pseudo(ip->ip_src.s_addr, 6368 ip->ip_dst.s_addr, htonl((u_short)len + 6369 m->m_pkthdr.csum_data + p)); 6370 } 6371 sum ^= 0xffff; 6372 ++hw_assist; 6373 } 6374 break; 6375 case IPPROTO_ICMP: 6376 #ifdef INET6 6377 case IPPROTO_ICMPV6: 6378 #endif /* INET6 */ 6379 break; 6380 default: 6381 return (1); 6382 } 6383 6384 if (!hw_assist) { 6385 switch (af) { 6386 case AF_INET: 6387 if (p == IPPROTO_ICMP) { 6388 if (m->m_len < off) 6389 return (1); 6390 m->m_data += off; 6391 m->m_len -= off; 6392 sum = in_cksum(m, len); 6393 m->m_data -= off; 6394 m->m_len += off; 6395 } else { 6396 if (m->m_len < sizeof(struct ip)) 6397 return (1); 6398 sum = in_cksum_range(m, p, off, len); 6399 if (sum == 0) { 6400 m->m_pkthdr.csum_flags |= 6401 (CSUM_DATA_VALID | 6402 CSUM_PSEUDO_HDR); 6403 m->m_pkthdr.csum_data = 0xffff; 6404 } 6405 } 6406 break; 6407 #ifdef INET6 6408 case AF_INET6: 6409 if (m->m_len < sizeof(struct ip6_hdr)) 6410 return (1); 6411 sum = in6_cksum(m, p, off, len); 6412 /* 6413 * XXX 6414 * IPv6 H/W cksum off-load not supported yet! 6415 * 6416 * if (sum == 0) { 6417 * m->m_pkthdr.csum_flags |= 6418 * (CSUM_DATA_VALID|CSUM_PSEUDO_HDR); 6419 * m->m_pkthdr.csum_data = 0xffff; 6420 *} 6421 */ 6422 break; 6423 #endif /* INET6 */ 6424 default: 6425 return (1); 6426 } 6427 } 6428 if (sum) { 6429 switch (p) { 6430 case IPPROTO_TCP: 6431 tcpstat.tcps_rcvbadsum++; 6432 break; 6433 case IPPROTO_UDP: 6434 udp_stat.udps_badsum++; 6435 break; 6436 case IPPROTO_ICMP: 6437 icmpstat.icps_checksum++; 6438 break; 6439 #ifdef INET6 6440 case IPPROTO_ICMPV6: 6441 icmp6stat.icp6s_checksum++; 6442 break; 6443 #endif /* INET6 */ 6444 } 6445 return (1); 6446 } 6447 return (0); 6448 } 6449 6450 struct pf_divert * 6451 pf_find_divert(struct mbuf *m) 6452 { 6453 struct m_tag *mtag; 6454 6455 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) 6456 return (NULL); 6457 6458 return ((struct pf_divert *)(mtag + 1)); 6459 } 6460 6461 struct pf_divert * 6462 pf_get_divert(struct mbuf *m) 6463 { 6464 struct m_tag *mtag; 6465 6466 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) { 6467 mtag = m_tag_get(PACKET_TAG_PF_DIVERT, sizeof(struct pf_divert), 6468 M_NOWAIT); 6469 if (mtag == NULL) 6470 return (NULL); 6471 bzero(mtag + 1, sizeof(struct pf_divert)); 6472 m_tag_prepend(m, mtag); 6473 } 6474 6475 return ((struct pf_divert *)(mtag + 1)); 6476 } 6477 6478 #ifdef INET 6479 6480 /* 6481 * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE 6482 */ 6483 int 6484 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, 6485 struct ether_header *eh, struct inpcb *inp) 6486 { 6487 struct pfi_kif *kif; 6488 u_short action, reason = 0, log = 0; 6489 struct mbuf *m = *m0; 6490 struct ip *h = NULL; 6491 struct pf_rule *a = NULL, *r = &pf_default_rule, *tr, *nr; 6492 struct pf_state *s = NULL; 6493 struct pf_ruleset *ruleset = NULL; 6494 struct pf_pdesc pd; 6495 int off, dirndx; 6496 #ifdef ALTQ 6497 int pqid = 0; 6498 #endif 6499 6500 if (m->m_pkthdr.fw_flags & IPFW_MBUF_CONTINUE) { 6501 /* Skip us; continue in ipfw. */ 6502 return (PF_PASS); 6503 } 6504 6505 if (!pf_status.running) 6506 return (PF_PASS); 6507 6508 memset(&pd, 0, sizeof(pd)); 6509 #ifdef foo 6510 if (ifp->if_type == IFT_CARP && ifp->if_carpdev) 6511 kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif; 6512 else 6513 #endif 6514 kif = (struct pfi_kif *)ifp->if_pf_kif; 6515 6516 if (kif == NULL) { 6517 DPFPRINTF(PF_DEBUG_URGENT, 6518 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 6519 return (PF_DROP); 6520 } 6521 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6522 return (PF_PASS); 6523 6524 #ifdef DIAGNOSTIC 6525 if ((m->m_flags & M_PKTHDR) == 0) 6526 panic("non-M_PKTHDR is passed to pf_test"); 6527 #endif /* DIAGNOSTIC */ 6528 6529 if (m->m_pkthdr.len < (int)sizeof(*h)) { 6530 action = PF_DROP; 6531 REASON_SET(&reason, PFRES_SHORT); 6532 log = 1; 6533 goto done; 6534 } 6535 6536 /* 6537 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags, 6538 * so make sure pf.flags is clear. 6539 */ 6540 if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED) 6541 return (PF_PASS); 6542 m->m_pkthdr.pf.flags = 0; 6543 /* Re-Check when updating to > 4.4 */ 6544 m->m_pkthdr.pf.statekey = NULL; 6545 6546 /* We do IP header normalization and packet reassembly here */ 6547 if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { 6548 action = PF_DROP; 6549 goto done; 6550 } 6551 m = *m0; /* pf_normalize messes with m0 */ 6552 h = mtod(m, struct ip *); 6553 6554 off = h->ip_hl << 2; 6555 if (off < (int)sizeof(*h)) { 6556 action = PF_DROP; 6557 REASON_SET(&reason, PFRES_SHORT); 6558 log = 1; 6559 goto done; 6560 } 6561 6562 pd.src = (struct pf_addr *)&h->ip_src; 6563 pd.dst = (struct pf_addr *)&h->ip_dst; 6564 pd.sport = pd.dport = NULL; 6565 pd.ip_sum = &h->ip_sum; 6566 pd.proto_sum = NULL; 6567 pd.proto = h->ip_p; 6568 pd.dir = dir; 6569 pd.sidx = (dir == PF_IN) ? 0 : 1; 6570 pd.didx = (dir == PF_IN) ? 1 : 0; 6571 pd.af = AF_INET; 6572 pd.tos = h->ip_tos; 6573 pd.tot_len = h->ip_len; 6574 pd.eh = eh; 6575 6576 /* handle fragments that didn't get reassembled by normalization */ 6577 if (h->ip_off & (IP_MF | IP_OFFMASK)) { 6578 action = pf_test_fragment(&r, dir, kif, m, h, 6579 &pd, &a, &ruleset); 6580 goto done; 6581 } 6582 6583 switch (h->ip_p) { 6584 6585 case IPPROTO_TCP: { 6586 struct tcphdr th; 6587 6588 pd.hdr.tcp = &th; 6589 if (!pf_pull_hdr(m, off, &th, sizeof(th), 6590 &action, &reason, AF_INET)) { 6591 log = action != PF_PASS; 6592 goto done; 6593 } 6594 pd.p_len = pd.tot_len - off - (th.th_off << 2); 6595 #ifdef ALTQ 6596 if ((th.th_flags & TH_ACK) && pd.p_len == 0) 6597 pqid = 1; 6598 #endif 6599 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6600 if (action == PF_DROP) 6601 goto done; 6602 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6603 &reason); 6604 if (action == PF_PASS) { 6605 r = s->rule.ptr; 6606 a = s->anchor.ptr; 6607 log = s->log; 6608 } else if (s == NULL) { 6609 action = pf_test_rule(&r, &s, dir, kif, 6610 m, off, h, &pd, &a, 6611 &ruleset, NULL, inp); 6612 } 6613 break; 6614 } 6615 6616 case IPPROTO_UDP: { 6617 struct udphdr uh; 6618 6619 pd.hdr.udp = &uh; 6620 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 6621 &action, &reason, AF_INET)) { 6622 log = action != PF_PASS; 6623 goto done; 6624 } 6625 if (uh.uh_dport == 0 || 6626 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 6627 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 6628 action = PF_DROP; 6629 REASON_SET(&reason, PFRES_SHORT); 6630 goto done; 6631 } 6632 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 6633 if (action == PF_PASS) { 6634 r = s->rule.ptr; 6635 a = s->anchor.ptr; 6636 log = s->log; 6637 } else if (s == NULL) { 6638 action = pf_test_rule(&r, &s, dir, kif, 6639 m, off, h, &pd, &a, 6640 &ruleset, NULL, inp); 6641 } 6642 break; 6643 } 6644 6645 case IPPROTO_ICMP: { 6646 struct icmp ih; 6647 6648 pd.hdr.icmp = &ih; 6649 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN, 6650 &action, &reason, AF_INET)) { 6651 log = action != PF_PASS; 6652 goto done; 6653 } 6654 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, 6655 &reason); 6656 if (action == PF_PASS) { 6657 r = s->rule.ptr; 6658 a = s->anchor.ptr; 6659 log = s->log; 6660 } else if (s == NULL) { 6661 action = pf_test_rule(&r, &s, dir, kif, 6662 m, off, h, &pd, &a, 6663 &ruleset, NULL, inp); 6664 } 6665 break; 6666 } 6667 6668 default: 6669 action = pf_test_state_other(&s, dir, kif, m, &pd); 6670 if (action == PF_PASS) { 6671 r = s->rule.ptr; 6672 a = s->anchor.ptr; 6673 log = s->log; 6674 } else if (s == NULL) { 6675 action = pf_test_rule(&r, &s, dir, kif, m, off, h, 6676 &pd, &a, &ruleset, NULL, inp); 6677 } 6678 break; 6679 } 6680 6681 done: 6682 if (action == PF_PASS && h->ip_hl > 5 && 6683 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 6684 action = PF_DROP; 6685 REASON_SET(&reason, PFRES_IPOPTIONS); 6686 log = 1; 6687 DPFPRINTF(PF_DEBUG_MISC, 6688 ("pf: dropping packet with ip options\n")); 6689 } 6690 6691 if ((s && s->tag) || r->rtableid) 6692 pf_tag_packet(m, s ? s->tag : 0, r->rtableid); 6693 6694 #if 0 6695 if (dir == PF_IN && s && s->key[PF_SK_STACK]) 6696 m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK]; 6697 #endif 6698 6699 #ifdef ALTQ 6700 /* 6701 * Generate a hash code and qid request for ALTQ. A qid of 0 6702 * is allowed and will cause altq to select the default queue. 6703 */ 6704 if (action == PF_PASS) { 6705 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE; 6706 if (pqid || (pd.tos & IPTOS_LOWDELAY)) 6707 m->m_pkthdr.pf.qid = r->pqid; 6708 else 6709 m->m_pkthdr.pf.qid = r->qid; 6710 m->m_pkthdr.pf.ecn_af = AF_INET; 6711 m->m_pkthdr.pf.hdr = h; 6712 /* add connection hash for fairq */ 6713 if (s) { 6714 /* for fairq */ 6715 m->m_pkthdr.pf.state_hash = s->hash; 6716 m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED; 6717 } 6718 } 6719 #endif /* ALTQ */ 6720 6721 /* 6722 * connections redirected to loopback should not match sockets 6723 * bound specifically to loopback due to security implications, 6724 * see tcp_input() and in_pcblookup_listen(). 6725 */ 6726 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 6727 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 6728 (s->nat_rule.ptr->action == PF_RDR || 6729 s->nat_rule.ptr->action == PF_BINAT) && 6730 (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) 6731 m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST; 6732 6733 if (dir == PF_IN && action == PF_PASS && r->divert.port) { 6734 struct pf_divert *divert; 6735 6736 if ((divert = pf_get_divert(m))) { 6737 m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED; 6738 divert->port = r->divert.port; 6739 divert->addr.ipv4 = r->divert.addr.v4; 6740 } 6741 } 6742 6743 if (log) { 6744 struct pf_rule *lr; 6745 6746 if (s != NULL && s->nat_rule.ptr != NULL && 6747 s->nat_rule.ptr->log & PF_LOG_ALL) 6748 lr = s->nat_rule.ptr; 6749 else 6750 lr = r; 6751 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, lr, a, ruleset, 6752 &pd); 6753 } 6754 6755 kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len; 6756 kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++; 6757 6758 if (action == PF_PASS || r->action == PF_DROP) { 6759 dirndx = (dir == PF_OUT); 6760 r->packets[dirndx]++; 6761 r->bytes[dirndx] += pd.tot_len; 6762 if (a != NULL) { 6763 a->packets[dirndx]++; 6764 a->bytes[dirndx] += pd.tot_len; 6765 } 6766 if (s != NULL) { 6767 if (s->nat_rule.ptr != NULL) { 6768 s->nat_rule.ptr->packets[dirndx]++; 6769 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; 6770 } 6771 if (s->src_node != NULL) { 6772 s->src_node->packets[dirndx]++; 6773 s->src_node->bytes[dirndx] += pd.tot_len; 6774 } 6775 if (s->nat_src_node != NULL) { 6776 s->nat_src_node->packets[dirndx]++; 6777 s->nat_src_node->bytes[dirndx] += pd.tot_len; 6778 } 6779 dirndx = (dir == s->direction) ? 0 : 1; 6780 s->packets[dirndx]++; 6781 s->bytes[dirndx] += pd.tot_len; 6782 } 6783 tr = r; 6784 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 6785 if (nr != NULL && r == &pf_default_rule) 6786 tr = nr; 6787 if (tr->src.addr.type == PF_ADDR_TABLE) 6788 pfr_update_stats(tr->src.addr.p.tbl, 6789 (s == NULL) ? pd.src : 6790 &s->key[(s->direction == PF_IN)]-> 6791 addr[(s->direction == PF_OUT)], 6792 pd.af, pd.tot_len, dir == PF_OUT, 6793 r->action == PF_PASS, tr->src.neg); 6794 if (tr->dst.addr.type == PF_ADDR_TABLE) 6795 pfr_update_stats(tr->dst.addr.p.tbl, 6796 (s == NULL) ? pd.dst : 6797 &s->key[(s->direction == PF_IN)]-> 6798 addr[(s->direction == PF_IN)], 6799 pd.af, pd.tot_len, dir == PF_OUT, 6800 r->action == PF_PASS, tr->dst.neg); 6801 } 6802 6803 6804 if (action == PF_SYNPROXY_DROP) { 6805 m_freem(*m0); 6806 *m0 = NULL; 6807 action = PF_PASS; 6808 } else if (r->rt) { 6809 /* pf_route can free the mbuf causing *m0 to become NULL */ 6810 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd); 6811 } 6812 6813 return (action); 6814 } 6815 #endif /* INET */ 6816 6817 #ifdef INET6 6818 6819 /* 6820 * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE 6821 */ 6822 int 6823 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, 6824 struct ether_header *eh, struct inpcb *inp) 6825 { 6826 struct pfi_kif *kif; 6827 u_short action, reason = 0, log = 0; 6828 struct mbuf *m = *m0, *n = NULL; 6829 struct ip6_hdr *h = NULL; 6830 struct pf_rule *a = NULL, *r = &pf_default_rule, *tr, *nr; 6831 struct pf_state *s = NULL; 6832 struct pf_ruleset *ruleset = NULL; 6833 struct pf_pdesc pd; 6834 int off, terminal = 0, dirndx, rh_cnt = 0; 6835 6836 if (!pf_status.running) 6837 return (PF_PASS); 6838 6839 memset(&pd, 0, sizeof(pd)); 6840 #ifdef foo 6841 if (ifp->if_type == IFT_CARP && ifp->if_carpdev) 6842 kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif; 6843 else 6844 #endif 6845 kif = (struct pfi_kif *)ifp->if_pf_kif; 6846 6847 if (kif == NULL) { 6848 DPFPRINTF(PF_DEBUG_URGENT, 6849 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 6850 return (PF_DROP); 6851 } 6852 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6853 return (PF_PASS); 6854 6855 #ifdef DIAGNOSTIC 6856 if ((m->m_flags & M_PKTHDR) == 0) 6857 panic("non-M_PKTHDR is passed to pf_test6"); 6858 #endif /* DIAGNOSTIC */ 6859 6860 if (m->m_pkthdr.len < (int)sizeof(*h)) { 6861 action = PF_DROP; 6862 REASON_SET(&reason, PFRES_SHORT); 6863 log = 1; 6864 goto done; 6865 } 6866 6867 /* 6868 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags, 6869 * so make sure pf.flags is clear. 6870 */ 6871 if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED) 6872 return (PF_PASS); 6873 m->m_pkthdr.pf.flags = 0; 6874 /* Re-Check when updating to > 4.4 */ 6875 m->m_pkthdr.pf.statekey = NULL; 6876 6877 /* We do IP header normalization and packet reassembly here */ 6878 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { 6879 action = PF_DROP; 6880 goto done; 6881 } 6882 m = *m0; /* pf_normalize messes with m0 */ 6883 h = mtod(m, struct ip6_hdr *); 6884 6885 #if 1 6886 /* 6887 * we do not support jumbogram yet. if we keep going, zero ip6_plen 6888 * will do something bad, so drop the packet for now. 6889 */ 6890 if (htons(h->ip6_plen) == 0) { 6891 action = PF_DROP; 6892 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 6893 goto done; 6894 } 6895 #endif 6896 6897 pd.src = (struct pf_addr *)&h->ip6_src; 6898 pd.dst = (struct pf_addr *)&h->ip6_dst; 6899 pd.sport = pd.dport = NULL; 6900 pd.ip_sum = NULL; 6901 pd.proto_sum = NULL; 6902 pd.dir = dir; 6903 pd.sidx = (dir == PF_IN) ? 0 : 1; 6904 pd.didx = (dir == PF_IN) ? 1 : 0; 6905 pd.af = AF_INET6; 6906 pd.tos = 0; 6907 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 6908 pd.eh = eh; 6909 6910 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 6911 pd.proto = h->ip6_nxt; 6912 do { 6913 switch (pd.proto) { 6914 case IPPROTO_FRAGMENT: 6915 action = pf_test_fragment(&r, dir, kif, m, h, 6916 &pd, &a, &ruleset); 6917 if (action == PF_DROP) 6918 REASON_SET(&reason, PFRES_FRAG); 6919 goto done; 6920 case IPPROTO_ROUTING: { 6921 struct ip6_rthdr rthdr; 6922 6923 if (rh_cnt++) { 6924 DPFPRINTF(PF_DEBUG_MISC, 6925 ("pf: IPv6 more than one rthdr\n")); 6926 action = PF_DROP; 6927 REASON_SET(&reason, PFRES_IPOPTIONS); 6928 log = 1; 6929 goto done; 6930 } 6931 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 6932 &reason, pd.af)) { 6933 DPFPRINTF(PF_DEBUG_MISC, 6934 ("pf: IPv6 short rthdr\n")); 6935 action = PF_DROP; 6936 REASON_SET(&reason, PFRES_SHORT); 6937 log = 1; 6938 goto done; 6939 } 6940 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 6941 DPFPRINTF(PF_DEBUG_MISC, 6942 ("pf: IPv6 rthdr0\n")); 6943 action = PF_DROP; 6944 REASON_SET(&reason, PFRES_IPOPTIONS); 6945 log = 1; 6946 goto done; 6947 } 6948 /* FALLTHROUGH */ 6949 } 6950 case IPPROTO_AH: 6951 case IPPROTO_HOPOPTS: 6952 case IPPROTO_DSTOPTS: { 6953 /* get next header and header length */ 6954 struct ip6_ext opt6; 6955 6956 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 6957 NULL, &reason, pd.af)) { 6958 DPFPRINTF(PF_DEBUG_MISC, 6959 ("pf: IPv6 short opt\n")); 6960 action = PF_DROP; 6961 log = 1; 6962 goto done; 6963 } 6964 if (pd.proto == IPPROTO_AH) 6965 off += (opt6.ip6e_len + 2) * 4; 6966 else 6967 off += (opt6.ip6e_len + 1) * 8; 6968 pd.proto = opt6.ip6e_nxt; 6969 /* goto the next header */ 6970 break; 6971 } 6972 default: 6973 terminal++; 6974 break; 6975 } 6976 } while (!terminal); 6977 6978 /* if there's no routing header, use unmodified mbuf for checksumming */ 6979 if (!n) 6980 n = m; 6981 6982 switch (pd.proto) { 6983 6984 case IPPROTO_TCP: { 6985 struct tcphdr th; 6986 6987 pd.hdr.tcp = &th; 6988 if (!pf_pull_hdr(m, off, &th, sizeof(th), 6989 &action, &reason, AF_INET6)) { 6990 log = action != PF_PASS; 6991 goto done; 6992 } 6993 pd.p_len = pd.tot_len - off - (th.th_off << 2); 6994 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6995 if (action == PF_DROP) 6996 goto done; 6997 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6998 &reason); 6999 if (action == PF_PASS) { 7000 r = s->rule.ptr; 7001 a = s->anchor.ptr; 7002 log = s->log; 7003 } else if (s == NULL) { 7004 action = pf_test_rule(&r, &s, dir, kif, 7005 m, off, h, &pd, &a, 7006 &ruleset, NULL, inp); 7007 } 7008 break; 7009 } 7010 7011 case IPPROTO_UDP: { 7012 struct udphdr uh; 7013 7014 pd.hdr.udp = &uh; 7015 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 7016 &action, &reason, AF_INET6)) { 7017 log = action != PF_PASS; 7018 goto done; 7019 } 7020 if (uh.uh_dport == 0 || 7021 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 7022 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 7023 action = PF_DROP; 7024 REASON_SET(&reason, PFRES_SHORT); 7025 goto done; 7026 } 7027 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 7028 if (action == PF_PASS) { 7029 r = s->rule.ptr; 7030 a = s->anchor.ptr; 7031 log = s->log; 7032 } else if (s == NULL) { 7033 action = pf_test_rule(&r, &s, dir, kif, 7034 m, off, h, &pd, &a, 7035 &ruleset, NULL, inp); 7036 } 7037 break; 7038 } 7039 7040 case IPPROTO_ICMPV6: { 7041 struct icmp6_hdr ih; 7042 7043 pd.hdr.icmp6 = &ih; 7044 if (!pf_pull_hdr(m, off, &ih, sizeof(ih), 7045 &action, &reason, AF_INET6)) { 7046 log = action != PF_PASS; 7047 goto done; 7048 } 7049 action = pf_test_state_icmp(&s, dir, kif, 7050 m, off, h, &pd, &reason); 7051 if (action == PF_PASS) { 7052 r = s->rule.ptr; 7053 a = s->anchor.ptr; 7054 log = s->log; 7055 } else if (s == NULL) { 7056 action = pf_test_rule(&r, &s, dir, kif, 7057 m, off, h, &pd, &a, 7058 &ruleset, NULL, inp); 7059 } 7060 break; 7061 } 7062 7063 default: 7064 action = pf_test_state_other(&s, dir, kif, m, &pd); 7065 if (action == PF_PASS) { 7066 r = s->rule.ptr; 7067 a = s->anchor.ptr; 7068 log = s->log; 7069 } else if (s == NULL) { 7070 action = pf_test_rule(&r, &s, dir, kif, m, off, h, 7071 &pd, &a, &ruleset, NULL, inp); 7072 } 7073 break; 7074 } 7075 7076 done: 7077 if (n != m) { 7078 m_freem(n); 7079 n = NULL; 7080 } 7081 7082 /* handle dangerous IPv6 extension headers. */ 7083 if (action == PF_PASS && rh_cnt && 7084 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 7085 action = PF_DROP; 7086 REASON_SET(&reason, PFRES_IPOPTIONS); 7087 log = 1; 7088 DPFPRINTF(PF_DEBUG_MISC, 7089 ("pf: dropping packet with dangerous v6 headers\n")); 7090 } 7091 7092 if ((s && s->tag) || r->rtableid) 7093 pf_tag_packet(m, s ? s->tag : 0, r->rtableid); 7094 7095 #if 0 7096 if (dir == PF_IN && s && s->key[PF_SK_STACK]) 7097 m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK]; 7098 #endif 7099 7100 #ifdef ALTQ 7101 /* 7102 * Generate a hash code and qid request for ALTQ. A qid of 0 7103 * is allowed and will cause altq to select the default queue. 7104 */ 7105 if (action == PF_PASS) { 7106 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE; 7107 if (pd.tos & IPTOS_LOWDELAY) 7108 m->m_pkthdr.pf.qid = r->pqid; 7109 else 7110 m->m_pkthdr.pf.qid = r->qid; 7111 m->m_pkthdr.pf.ecn_af = AF_INET6; 7112 m->m_pkthdr.pf.hdr = h; 7113 if (s) { 7114 /* for fairq */ 7115 m->m_pkthdr.pf.state_hash = s->hash; 7116 m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED; 7117 } 7118 } 7119 #endif /* ALTQ */ 7120 7121 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 7122 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 7123 (s->nat_rule.ptr->action == PF_RDR || 7124 s->nat_rule.ptr->action == PF_BINAT) && 7125 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 7126 m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST; 7127 7128 if (dir == PF_IN && action == PF_PASS && r->divert.port) { 7129 struct pf_divert *divert; 7130 7131 if ((divert = pf_get_divert(m))) { 7132 m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED; 7133 divert->port = r->divert.port; 7134 divert->addr.ipv6 = r->divert.addr.v6; 7135 } 7136 } 7137 7138 if (log) { 7139 struct pf_rule *lr; 7140 7141 if (s != NULL && s->nat_rule.ptr != NULL && 7142 s->nat_rule.ptr->log & PF_LOG_ALL) 7143 lr = s->nat_rule.ptr; 7144 else 7145 lr = r; 7146 PFLOG_PACKET(kif, h, m, AF_INET6, dir, reason, lr, a, ruleset, 7147 &pd); 7148 } 7149 7150 kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len; 7151 kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++; 7152 7153 if (action == PF_PASS || r->action == PF_DROP) { 7154 dirndx = (dir == PF_OUT); 7155 r->packets[dirndx]++; 7156 r->bytes[dirndx] += pd.tot_len; 7157 if (a != NULL) { 7158 a->packets[dirndx]++; 7159 a->bytes[dirndx] += pd.tot_len; 7160 } 7161 if (s != NULL) { 7162 if (s->nat_rule.ptr != NULL) { 7163 s->nat_rule.ptr->packets[dirndx]++; 7164 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; 7165 } 7166 if (s->src_node != NULL) { 7167 s->src_node->packets[dirndx]++; 7168 s->src_node->bytes[dirndx] += pd.tot_len; 7169 } 7170 if (s->nat_src_node != NULL) { 7171 s->nat_src_node->packets[dirndx]++; 7172 s->nat_src_node->bytes[dirndx] += pd.tot_len; 7173 } 7174 dirndx = (dir == s->direction) ? 0 : 1; 7175 s->packets[dirndx]++; 7176 s->bytes[dirndx] += pd.tot_len; 7177 } 7178 tr = r; 7179 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 7180 if (nr != NULL && r == &pf_default_rule) 7181 tr = nr; 7182 if (tr->src.addr.type == PF_ADDR_TABLE) 7183 pfr_update_stats(tr->src.addr.p.tbl, 7184 (s == NULL) ? pd.src : 7185 &s->key[(s->direction == PF_IN)]->addr[0], 7186 pd.af, pd.tot_len, dir == PF_OUT, 7187 r->action == PF_PASS, tr->src.neg); 7188 if (tr->dst.addr.type == PF_ADDR_TABLE) 7189 pfr_update_stats(tr->dst.addr.p.tbl, 7190 (s == NULL) ? pd.dst : 7191 &s->key[(s->direction == PF_IN)]->addr[1], 7192 pd.af, pd.tot_len, dir == PF_OUT, 7193 r->action == PF_PASS, tr->dst.neg); 7194 } 7195 7196 7197 if (action == PF_SYNPROXY_DROP) { 7198 m_freem(*m0); 7199 *m0 = NULL; 7200 action = PF_PASS; 7201 } else if (r->rt) 7202 /* pf_route6 can free the mbuf causing *m0 to become NULL */ 7203 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd); 7204 7205 return (action); 7206 } 7207 #endif /* INET6 */ 7208 7209 int 7210 pf_check_congestion(struct ifqueue *ifq) 7211 { 7212 return (0); 7213 } 7214