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[MAXCPU+1]; /* incls one global table */ 134 135 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[MAXCPU]; 355 struct pf_state_tree_id tree_id[MAXCPU]; 356 struct pf_state_queue state_list[MAXCPU]; 357 358 RB_GENERATE(pf_src_tree, pf_src_node, entry, pf_src_compare); 359 RB_GENERATE(pf_state_tree, pf_state_key, entry, pf_state_compare_key); 360 RB_GENERATE(pf_state_rtree, pf_state_key, entry, pf_state_compare_rkey); 361 RB_GENERATE(pf_state_tree_id, pf_state, entry_id, pf_state_compare_id); 362 363 static __inline int 364 pf_src_compare(struct pf_src_node *a, struct pf_src_node *b) 365 { 366 int diff; 367 368 if (a->rule.ptr > b->rule.ptr) 369 return (1); 370 if (a->rule.ptr < b->rule.ptr) 371 return (-1); 372 if ((diff = a->af - b->af) != 0) 373 return (diff); 374 switch (a->af) { 375 #ifdef INET 376 case AF_INET: 377 if (a->addr.addr32[0] > b->addr.addr32[0]) 378 return (1); 379 if (a->addr.addr32[0] < b->addr.addr32[0]) 380 return (-1); 381 break; 382 #endif /* INET */ 383 #ifdef INET6 384 case AF_INET6: 385 if (a->addr.addr32[3] > b->addr.addr32[3]) 386 return (1); 387 if (a->addr.addr32[3] < b->addr.addr32[3]) 388 return (-1); 389 if (a->addr.addr32[2] > b->addr.addr32[2]) 390 return (1); 391 if (a->addr.addr32[2] < b->addr.addr32[2]) 392 return (-1); 393 if (a->addr.addr32[1] > b->addr.addr32[1]) 394 return (1); 395 if (a->addr.addr32[1] < b->addr.addr32[1]) 396 return (-1); 397 if (a->addr.addr32[0] > b->addr.addr32[0]) 398 return (1); 399 if (a->addr.addr32[0] < b->addr.addr32[0]) 400 return (-1); 401 break; 402 #endif /* INET6 */ 403 } 404 return (0); 405 } 406 407 u_int32_t 408 pf_state_hash(struct pf_state_key *sk) 409 { 410 u_int32_t hv = (u_int32_t)(((intptr_t)sk >> 6) ^ ((intptr_t)sk >> 15)); 411 if (hv == 0) /* disallow 0 */ 412 hv = 1; 413 return(hv); 414 } 415 416 #ifdef INET6 417 void 418 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 419 { 420 switch (af) { 421 #ifdef INET 422 case AF_INET: 423 dst->addr32[0] = src->addr32[0]; 424 break; 425 #endif /* INET */ 426 case AF_INET6: 427 dst->addr32[0] = src->addr32[0]; 428 dst->addr32[1] = src->addr32[1]; 429 dst->addr32[2] = src->addr32[2]; 430 dst->addr32[3] = src->addr32[3]; 431 break; 432 } 433 } 434 #endif /* INET6 */ 435 436 void 437 pf_init_threshold(struct pf_threshold *threshold, 438 u_int32_t limit, u_int32_t seconds) 439 { 440 threshold->limit = limit * PF_THRESHOLD_MULT; 441 threshold->seconds = seconds; 442 threshold->count = 0; 443 threshold->last = time_second; 444 } 445 446 void 447 pf_add_threshold(struct pf_threshold *threshold) 448 { 449 u_int32_t t = time_second, diff = t - threshold->last; 450 451 if (diff >= threshold->seconds) 452 threshold->count = 0; 453 else 454 threshold->count -= threshold->count * diff / 455 threshold->seconds; 456 threshold->count += PF_THRESHOLD_MULT; 457 threshold->last = t; 458 } 459 460 int 461 pf_check_threshold(struct pf_threshold *threshold) 462 { 463 return (threshold->count > threshold->limit); 464 } 465 466 int 467 pf_src_connlimit(struct pf_state *state) 468 { 469 int bad = 0; 470 int cpu = mycpu->gd_cpuid; 471 472 state->src_node->conn++; 473 state->src.tcp_est = 1; 474 pf_add_threshold(&state->src_node->conn_rate); 475 476 if (state->rule.ptr->max_src_conn && 477 state->rule.ptr->max_src_conn < 478 state->src_node->conn) { 479 pf_status.lcounters[LCNT_SRCCONN]++; 480 bad++; 481 } 482 483 if (state->rule.ptr->max_src_conn_rate.limit && 484 pf_check_threshold(&state->src_node->conn_rate)) { 485 pf_status.lcounters[LCNT_SRCCONNRATE]++; 486 bad++; 487 } 488 489 if (!bad) 490 return 0; 491 492 if (state->rule.ptr->overload_tbl) { 493 struct pfr_addr p; 494 u_int32_t killed = 0; 495 496 pf_status.lcounters[LCNT_OVERLOAD_TABLE]++; 497 if (pf_status.debug >= PF_DEBUG_MISC) { 498 kprintf("pf_src_connlimit: blocking address "); 499 pf_print_host(&state->src_node->addr, 0, 500 state->key[PF_SK_WIRE]->af); 501 } 502 503 bzero(&p, sizeof(p)); 504 p.pfra_af = state->key[PF_SK_WIRE]->af; 505 switch (state->key[PF_SK_WIRE]->af) { 506 #ifdef INET 507 case AF_INET: 508 p.pfra_net = 32; 509 p.pfra_ip4addr = state->src_node->addr.v4; 510 break; 511 #endif /* INET */ 512 #ifdef INET6 513 case AF_INET6: 514 p.pfra_net = 128; 515 p.pfra_ip6addr = state->src_node->addr.v6; 516 break; 517 #endif /* INET6 */ 518 } 519 520 pfr_insert_kentry(state->rule.ptr->overload_tbl, 521 &p, time_second); 522 523 /* kill existing states if that's required. */ 524 if (state->rule.ptr->flush) { 525 struct pf_state_key *sk; 526 struct pf_state *st; 527 528 pf_status.lcounters[LCNT_OVERLOAD_FLUSH]++; 529 RB_FOREACH(st, pf_state_tree_id, &tree_id[cpu]) { 530 sk = st->key[PF_SK_WIRE]; 531 /* 532 * Kill states from this source. (Only those 533 * from the same rule if PF_FLUSH_GLOBAL is not 534 * set). (Only on current cpu). 535 */ 536 if (sk->af == 537 state->key[PF_SK_WIRE]->af && 538 ((state->direction == PF_OUT && 539 PF_AEQ(&state->src_node->addr, 540 &sk->addr[0], sk->af)) || 541 (state->direction == PF_IN && 542 PF_AEQ(&state->src_node->addr, 543 &sk->addr[1], sk->af))) && 544 (state->rule.ptr->flush & 545 PF_FLUSH_GLOBAL || 546 state->rule.ptr == st->rule.ptr)) { 547 st->timeout = PFTM_PURGE; 548 st->src.state = st->dst.state = 549 TCPS_CLOSED; 550 killed++; 551 } 552 } 553 if (pf_status.debug >= PF_DEBUG_MISC) 554 kprintf(", %u states killed", killed); 555 } 556 if (pf_status.debug >= PF_DEBUG_MISC) 557 kprintf("\n"); 558 } 559 560 /* kill this state */ 561 state->timeout = PFTM_PURGE; 562 state->src.state = state->dst.state = TCPS_CLOSED; 563 564 return 1; 565 } 566 567 int 568 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule, 569 struct pf_addr *src, sa_family_t af) 570 { 571 struct pf_src_node k; 572 int cpu = mycpu->gd_cpuid; 573 574 bzero(&k, sizeof(k)); /* avoid gcc warnings */ 575 if (*sn == NULL) { 576 k.af = af; 577 PF_ACPY(&k.addr, src, af); 578 if (rule->rule_flag & PFRULE_RULESRCTRACK || 579 rule->rpool.opts & PF_POOL_STICKYADDR) 580 k.rule.ptr = rule; 581 else 582 k.rule.ptr = NULL; 583 pf_status.scounters[SCNT_SRC_NODE_SEARCH]++; 584 *sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k); 585 } 586 if (*sn == NULL) { 587 if (!rule->max_src_nodes || 588 rule->src_nodes < rule->max_src_nodes) 589 (*sn) = kmalloc(sizeof(struct pf_src_node), 590 M_PFSRCTREEPL, M_NOWAIT|M_ZERO); 591 else 592 pf_status.lcounters[LCNT_SRCNODES]++; 593 if ((*sn) == NULL) 594 return (-1); 595 596 pf_init_threshold(&(*sn)->conn_rate, 597 rule->max_src_conn_rate.limit, 598 rule->max_src_conn_rate.seconds); 599 600 (*sn)->af = af; 601 if (rule->rule_flag & PFRULE_RULESRCTRACK || 602 rule->rpool.opts & PF_POOL_STICKYADDR) 603 (*sn)->rule.ptr = rule; 604 else 605 (*sn)->rule.ptr = NULL; 606 PF_ACPY(&(*sn)->addr, src, af); 607 if (RB_INSERT(pf_src_tree, 608 &tree_src_tracking[cpu], *sn) != NULL) { 609 if (pf_status.debug >= PF_DEBUG_MISC) { 610 kprintf("pf: src_tree insert failed: "); 611 pf_print_host(&(*sn)->addr, 0, af); 612 kprintf("\n"); 613 } 614 kfree(*sn, M_PFSRCTREEPL); 615 return (-1); 616 } 617 618 /* 619 * Atomic op required to increment src_nodes in the rule 620 * because we hold a shared token here (decrements will use 621 * an exclusive token). 622 */ 623 (*sn)->creation = time_second; 624 (*sn)->ruletype = rule->action; 625 if ((*sn)->rule.ptr != NULL) 626 atomic_add_int(&(*sn)->rule.ptr->src_nodes, 1); 627 pf_status.scounters[SCNT_SRC_NODE_INSERT]++; 628 atomic_add_int(&pf_status.src_nodes, 1); 629 } else { 630 if (rule->max_src_states && 631 (*sn)->states >= rule->max_src_states) { 632 pf_status.lcounters[LCNT_SRCSTATES]++; 633 return (-1); 634 } 635 } 636 return (0); 637 } 638 639 /* 640 * state table (indexed by the pf_state_key structure), normal RBTREE 641 * comparison. 642 */ 643 static __inline int 644 pf_state_compare_key(struct pf_state_key *a, struct pf_state_key *b) 645 { 646 int diff; 647 648 if ((diff = a->proto - b->proto) != 0) 649 return (diff); 650 if ((diff = a->af - b->af) != 0) 651 return (diff); 652 switch (a->af) { 653 #ifdef INET 654 case AF_INET: 655 if (a->addr[0].addr32[0] > b->addr[0].addr32[0]) 656 return (1); 657 if (a->addr[0].addr32[0] < b->addr[0].addr32[0]) 658 return (-1); 659 if (a->addr[1].addr32[0] > b->addr[1].addr32[0]) 660 return (1); 661 if (a->addr[1].addr32[0] < b->addr[1].addr32[0]) 662 return (-1); 663 break; 664 #endif /* INET */ 665 #ifdef INET6 666 case AF_INET6: 667 if (a->addr[0].addr32[3] > b->addr[0].addr32[3]) 668 return (1); 669 if (a->addr[0].addr32[3] < b->addr[0].addr32[3]) 670 return (-1); 671 if (a->addr[1].addr32[3] > b->addr[1].addr32[3]) 672 return (1); 673 if (a->addr[1].addr32[3] < b->addr[1].addr32[3]) 674 return (-1); 675 if (a->addr[0].addr32[2] > b->addr[0].addr32[2]) 676 return (1); 677 if (a->addr[0].addr32[2] < b->addr[0].addr32[2]) 678 return (-1); 679 if (a->addr[1].addr32[2] > b->addr[1].addr32[2]) 680 return (1); 681 if (a->addr[1].addr32[2] < b->addr[1].addr32[2]) 682 return (-1); 683 if (a->addr[0].addr32[1] > b->addr[0].addr32[1]) 684 return (1); 685 if (a->addr[0].addr32[1] < b->addr[0].addr32[1]) 686 return (-1); 687 if (a->addr[1].addr32[1] > b->addr[1].addr32[1]) 688 return (1); 689 if (a->addr[1].addr32[1] < b->addr[1].addr32[1]) 690 return (-1); 691 if (a->addr[0].addr32[0] > b->addr[0].addr32[0]) 692 return (1); 693 if (a->addr[0].addr32[0] < b->addr[0].addr32[0]) 694 return (-1); 695 if (a->addr[1].addr32[0] > b->addr[1].addr32[0]) 696 return (1); 697 if (a->addr[1].addr32[0] < b->addr[1].addr32[0]) 698 return (-1); 699 break; 700 #endif /* INET6 */ 701 } 702 703 if ((diff = a->port[0] - b->port[0]) != 0) 704 return (diff); 705 if ((diff = a->port[1] - b->port[1]) != 0) 706 return (diff); 707 708 return (0); 709 } 710 711 /* 712 * Used for RB_FIND only, compare in the reverse direction. The 713 * element to be reversed is always (a), since we obviously can't 714 * reverse the state tree depicted by (b). 715 */ 716 static __inline int 717 pf_state_compare_rkey(struct pf_state_key *a, struct pf_state_key *b) 718 { 719 int diff; 720 721 if ((diff = a->proto - b->proto) != 0) 722 return (diff); 723 if ((diff = a->af - b->af) != 0) 724 return (diff); 725 switch (a->af) { 726 #ifdef INET 727 case AF_INET: 728 if (a->addr[1].addr32[0] > b->addr[0].addr32[0]) 729 return (1); 730 if (a->addr[1].addr32[0] < b->addr[0].addr32[0]) 731 return (-1); 732 if (a->addr[0].addr32[0] > b->addr[1].addr32[0]) 733 return (1); 734 if (a->addr[0].addr32[0] < b->addr[1].addr32[0]) 735 return (-1); 736 break; 737 #endif /* INET */ 738 #ifdef INET6 739 case AF_INET6: 740 if (a->addr[1].addr32[3] > b->addr[0].addr32[3]) 741 return (1); 742 if (a->addr[1].addr32[3] < b->addr[0].addr32[3]) 743 return (-1); 744 if (a->addr[0].addr32[3] > b->addr[1].addr32[3]) 745 return (1); 746 if (a->addr[0].addr32[3] < b->addr[1].addr32[3]) 747 return (-1); 748 if (a->addr[1].addr32[2] > b->addr[0].addr32[2]) 749 return (1); 750 if (a->addr[1].addr32[2] < b->addr[0].addr32[2]) 751 return (-1); 752 if (a->addr[0].addr32[2] > b->addr[1].addr32[2]) 753 return (1); 754 if (a->addr[0].addr32[2] < b->addr[1].addr32[2]) 755 return (-1); 756 if (a->addr[1].addr32[1] > b->addr[0].addr32[1]) 757 return (1); 758 if (a->addr[1].addr32[1] < b->addr[0].addr32[1]) 759 return (-1); 760 if (a->addr[0].addr32[1] > b->addr[1].addr32[1]) 761 return (1); 762 if (a->addr[0].addr32[1] < b->addr[1].addr32[1]) 763 return (-1); 764 if (a->addr[1].addr32[0] > b->addr[0].addr32[0]) 765 return (1); 766 if (a->addr[1].addr32[0] < b->addr[0].addr32[0]) 767 return (-1); 768 if (a->addr[0].addr32[0] > b->addr[1].addr32[0]) 769 return (1); 770 if (a->addr[0].addr32[0] < b->addr[1].addr32[0]) 771 return (-1); 772 break; 773 #endif /* INET6 */ 774 } 775 776 if ((diff = a->port[1] - b->port[0]) != 0) 777 return (diff); 778 if ((diff = a->port[0] - b->port[1]) != 0) 779 return (diff); 780 781 return (0); 782 } 783 784 static __inline int 785 pf_state_compare_id(struct pf_state *a, struct pf_state *b) 786 { 787 if (a->id > b->id) 788 return (1); 789 if (a->id < b->id) 790 return (-1); 791 if (a->creatorid > b->creatorid) 792 return (1); 793 if (a->creatorid < b->creatorid) 794 return (-1); 795 796 return (0); 797 } 798 799 int 800 pf_state_key_attach(struct pf_state_key *sk, struct pf_state *s, int idx) 801 { 802 struct pf_state_item *si; 803 struct pf_state_key *cur; 804 int cpu; 805 int error; 806 807 /* 808 * PFSTATE_STACK_GLOBAL is set when the state might not hash to the 809 * current cpu. The keys are managed on the global statetbl tree 810 * for this case. Only translations (RDR, NAT) can cause this. 811 * 812 * When this flag is not set we must still check the global statetbl 813 * for a collision, and if we find one we set the HALF_DUPLEX flag 814 * in the state. 815 */ 816 if (s->state_flags & PFSTATE_STACK_GLOBAL) { 817 cpu = MAXCPU; 818 lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE); 819 } else { 820 cpu = mycpu->gd_cpuid; 821 lockmgr(&pf_global_statetbl_lock, LK_SHARED); 822 } 823 KKASSERT(s->key[idx] == NULL); /* XXX handle this? */ 824 825 if (pf_status.debug >= PF_DEBUG_MISC) { 826 kprintf("state_key attach cpu %d (%08x:%d) %s (%08x:%d)\n", 827 cpu, 828 ntohl(sk->addr[0].addr32[0]), ntohs(sk->port[0]), 829 (idx == PF_SK_WIRE ? "->" : "<-"), 830 ntohl(sk->addr[1].addr32[0]), ntohs(sk->port[1])); 831 } 832 833 /* 834 * Check whether (e.g.) a PASS rule being put on a per-cpu tree 835 * collides with a translation rule on the global tree. This is 836 * NOT an error. We *WANT* to establish state for this case so the 837 * packet path is short-cutted and doesn't need to scan the ruleset 838 * on every packet. But the established state will only see one 839 * side of a two-way packet conversation. To prevent this from 840 * causing problems (e.g. generating a RST), we force PFSTATE_SLOPPY 841 * to be set on the established state. 842 * 843 * A collision against RDR state can only occur with a PASS IN in the 844 * opposite direction or a PASS OUT in the forwards direction. This 845 * is because RDRs are processed on the input side. 846 * 847 * A collision against NAT state can only occur with a PASS IN in the 848 * forwards direction or a PASS OUT in the opposite direction. This 849 * is because NATs are processed on the output side. 850 * 851 * In both situations we need to do a reverse addr/port test because 852 * the PASS IN or PASS OUT only establishes if it doesn't match the 853 * established RDR state in the forwards direction. The direction 854 * flag has to be ignored (it will be one way for a PASS IN and the 855 * other way for a PASS OUT). 856 * 857 * pf_global_statetbl_lock will be locked shared when testing and 858 * not entering into the global state table. 859 */ 860 if (cpu != MAXCPU && 861 (cur = RB_FIND(pf_state_rtree, 862 (struct pf_state_rtree *)&pf_statetbl[MAXCPU], 863 sk)) != NULL) { 864 TAILQ_FOREACH(si, &cur->states, entry) { 865 /* 866 * NOTE: We must ignore direction mismatches. 867 */ 868 if (si->s->kif == s->kif) { 869 s->state_flags |= PFSTATE_HALF_DUPLEX | 870 PFSTATE_SLOPPY; 871 if (pf_status.debug >= PF_DEBUG_MISC) { 872 kprintf( 873 "pf: %s key attach collision " 874 "on %s: ", 875 (idx == PF_SK_WIRE) ? 876 "wire" : "stack", 877 s->kif->pfik_name); 878 pf_print_state_parts(s, 879 (idx == PF_SK_WIRE) ? sk : NULL, 880 (idx == PF_SK_STACK) ? sk : NULL); 881 kprintf("\n"); 882 } 883 break; 884 } 885 } 886 } 887 888 /* 889 * Enter into either the per-cpu or the global state table. 890 * 891 * pf_global_statetbl_lock will be locked exclusively when entering 892 * into the global state table. 893 */ 894 if ((cur = RB_INSERT(pf_state_tree, &pf_statetbl[cpu], sk)) != NULL) { 895 /* key exists. check for same kif, if none, add to key */ 896 TAILQ_FOREACH(si, &cur->states, entry) { 897 if (si->s->kif == s->kif && 898 si->s->direction == s->direction) { 899 if (pf_status.debug >= PF_DEBUG_MISC) { 900 kprintf( 901 "pf: %s key attach failed on %s: ", 902 (idx == PF_SK_WIRE) ? 903 "wire" : "stack", 904 s->kif->pfik_name); 905 pf_print_state_parts(s, 906 (idx == PF_SK_WIRE) ? sk : NULL, 907 (idx == PF_SK_STACK) ? sk : NULL); 908 kprintf("\n"); 909 } 910 kfree(sk, M_PFSTATEKEYPL); 911 error = -1; 912 goto failed; /* collision! */ 913 } 914 } 915 kfree(sk, M_PFSTATEKEYPL); 916 917 s->key[idx] = cur; 918 } else { 919 s->key[idx] = sk; 920 } 921 922 if ((si = kmalloc(sizeof(struct pf_state_item), 923 M_PFSTATEITEMPL, M_NOWAIT)) == NULL) { 924 pf_state_key_detach(s, idx); 925 error = -1; 926 goto failed; /* collision! */ 927 } 928 si->s = s; 929 930 /* list is sorted, if-bound states before floating */ 931 if (s->kif == pfi_all) 932 TAILQ_INSERT_TAIL(&s->key[idx]->states, si, entry); 933 else 934 TAILQ_INSERT_HEAD(&s->key[idx]->states, si, entry); 935 936 error = 0; 937 failed: 938 lockmgr(&pf_global_statetbl_lock, LK_RELEASE); 939 return error; 940 } 941 942 /* 943 * NOTE: Can only be called indirectly via the purge thread with pf_token 944 * exclusively locked. 945 */ 946 void 947 pf_detach_state(struct pf_state *s) 948 { 949 if (s->key[PF_SK_WIRE] == s->key[PF_SK_STACK]) 950 s->key[PF_SK_WIRE] = NULL; 951 952 if (s->key[PF_SK_STACK] != NULL) 953 pf_state_key_detach(s, PF_SK_STACK); 954 955 if (s->key[PF_SK_WIRE] != NULL) 956 pf_state_key_detach(s, PF_SK_WIRE); 957 } 958 959 /* 960 * NOTE: Can only be called indirectly via the purge thread with pf_token 961 * exclusively locked. 962 */ 963 void 964 pf_state_key_detach(struct pf_state *s, int idx) 965 { 966 struct pf_state_item *si; 967 int cpu; 968 969 /* 970 * PFSTATE_STACK_GLOBAL is set for translations when the translated 971 * address/port is not localized to the same cpu that the untranslated 972 * address/port is on. The wire pf_state_key is managed on the global 973 * statetbl tree for this case. 974 */ 975 if (s->state_flags & PFSTATE_STACK_GLOBAL) { 976 cpu = MAXCPU; 977 lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE); 978 } else { 979 cpu = mycpu->gd_cpuid; 980 } 981 982 si = TAILQ_FIRST(&s->key[idx]->states); 983 while (si && si->s != s) 984 si = TAILQ_NEXT(si, entry); 985 986 if (si) { 987 TAILQ_REMOVE(&s->key[idx]->states, si, entry); 988 kfree(si, M_PFSTATEITEMPL); 989 } 990 991 if (TAILQ_EMPTY(&s->key[idx]->states)) { 992 RB_REMOVE(pf_state_tree, &pf_statetbl[cpu], s->key[idx]); 993 if (s->key[idx]->reverse) 994 s->key[idx]->reverse->reverse = NULL; 995 if (s->key[idx]->inp) 996 s->key[idx]->inp->inp_pf_sk = NULL; 997 kfree(s->key[idx], M_PFSTATEKEYPL); 998 } 999 s->key[idx] = NULL; 1000 1001 if (s->state_flags & PFSTATE_STACK_GLOBAL) 1002 lockmgr(&pf_global_statetbl_lock, LK_RELEASE); 1003 } 1004 1005 struct pf_state_key * 1006 pf_alloc_state_key(int pool_flags) 1007 { 1008 struct pf_state_key *sk; 1009 1010 sk = kmalloc(sizeof(struct pf_state_key), M_PFSTATEKEYPL, pool_flags); 1011 if (sk) { 1012 TAILQ_INIT(&sk->states); 1013 } 1014 return (sk); 1015 } 1016 1017 int 1018 pf_state_key_setup(struct pf_pdesc *pd, struct pf_rule *nr, 1019 struct pf_state_key **skw, struct pf_state_key **sks, 1020 struct pf_state_key **skp, struct pf_state_key **nkp, 1021 struct pf_addr *saddr, struct pf_addr *daddr, 1022 u_int16_t sport, u_int16_t dport) 1023 { 1024 KKASSERT((*skp == NULL && *nkp == NULL)); 1025 1026 if ((*skp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL) 1027 return (ENOMEM); 1028 1029 PF_ACPY(&(*skp)->addr[pd->sidx], saddr, pd->af); 1030 PF_ACPY(&(*skp)->addr[pd->didx], daddr, pd->af); 1031 (*skp)->port[pd->sidx] = sport; 1032 (*skp)->port[pd->didx] = dport; 1033 (*skp)->proto = pd->proto; 1034 (*skp)->af = pd->af; 1035 1036 if (nr != NULL) { 1037 if ((*nkp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL) 1038 return (ENOMEM); /* caller must handle cleanup */ 1039 1040 /* XXX maybe just bcopy and TAILQ_INIT(&(*nkp)->states) */ 1041 PF_ACPY(&(*nkp)->addr[0], &(*skp)->addr[0], pd->af); 1042 PF_ACPY(&(*nkp)->addr[1], &(*skp)->addr[1], pd->af); 1043 (*nkp)->port[0] = (*skp)->port[0]; 1044 (*nkp)->port[1] = (*skp)->port[1]; 1045 (*nkp)->proto = pd->proto; 1046 (*nkp)->af = pd->af; 1047 } else { 1048 *nkp = *skp; 1049 } 1050 1051 if (pd->dir == PF_IN) { 1052 *skw = *skp; 1053 *sks = *nkp; 1054 } else { 1055 *sks = *skp; 1056 *skw = *nkp; 1057 } 1058 return (0); 1059 } 1060 1061 /* 1062 * Insert pf_state with one or two state keys (allowing a reverse path lookup 1063 * which is used by NAT). In the NAT case skw is the initiator (?) and 1064 * sks is the target. 1065 */ 1066 int 1067 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw, 1068 struct pf_state_key *sks, struct pf_state *s) 1069 { 1070 int cpu = mycpu->gd_cpuid; 1071 1072 s->kif = kif; 1073 s->cpuid = cpu; 1074 1075 if (skw == sks) { 1076 if (pf_state_key_attach(skw, s, PF_SK_WIRE)) 1077 return (-1); 1078 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1079 } else { 1080 /* 1081 skw->reverse = sks; 1082 sks->reverse = skw; 1083 */ 1084 if (pf_state_key_attach(skw, s, PF_SK_WIRE)) { 1085 kfree(sks, M_PFSTATEKEYPL); 1086 return (-1); 1087 } 1088 if (pf_state_key_attach(sks, s, PF_SK_STACK)) { 1089 pf_state_key_detach(s, PF_SK_WIRE); 1090 return (-1); 1091 } 1092 } 1093 1094 if (s->id == 0 && s->creatorid == 0) { 1095 u_int64_t sid; 1096 1097 #if __SIZEOF_LONG__ == 8 1098 sid = atomic_fetchadd_long(&pf_status.stateid, 1); 1099 #else 1100 spin_lock(&pf_spin); 1101 sid = pf_status.stateid++; 1102 spin_unlock(&pf_spin); 1103 #endif 1104 s->id = htobe64(sid); 1105 s->creatorid = pf_status.hostid; 1106 } 1107 1108 /* 1109 * Calculate hash code for altq 1110 */ 1111 s->hash = crc32(s->key[PF_SK_WIRE], PF_STATE_KEY_HASH_LENGTH); 1112 1113 if (RB_INSERT(pf_state_tree_id, &tree_id[cpu], s) != NULL) { 1114 if (pf_status.debug >= PF_DEBUG_MISC) { 1115 kprintf("pf: state insert failed: " 1116 "id: %016jx creatorid: %08x", 1117 (uintmax_t)be64toh(s->id), ntohl(s->creatorid)); 1118 if (s->sync_flags & PFSTATE_FROMSYNC) 1119 kprintf(" (from sync)"); 1120 kprintf("\n"); 1121 } 1122 pf_detach_state(s); 1123 return (-1); 1124 } 1125 TAILQ_INSERT_TAIL(&state_list[cpu], s, entry_list); 1126 pf_status.fcounters[FCNT_STATE_INSERT]++; 1127 atomic_add_int(&pf_status.states, 1); 1128 pfi_kif_ref(kif, PFI_KIF_REF_STATE); 1129 pfsync_insert_state(s); 1130 return (0); 1131 } 1132 1133 struct pf_state * 1134 pf_find_state_byid(struct pf_state_cmp *key) 1135 { 1136 int cpu = mycpu->gd_cpuid; 1137 1138 pf_status.fcounters[FCNT_STATE_SEARCH]++; 1139 1140 return (RB_FIND(pf_state_tree_id, &tree_id[cpu], 1141 (struct pf_state *)key)); 1142 } 1143 1144 /* 1145 * WARNING! May return a state structure that was localized to another cpu, 1146 * destruction is typically protected by the callers pf_token. 1147 * The element can only be destroyed 1148 */ 1149 struct pf_state * 1150 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir, 1151 struct mbuf *m) 1152 { 1153 struct pf_state_key *skey = (void *)key; 1154 struct pf_state_key *sk; 1155 struct pf_state_item *si; 1156 struct pf_state *s; 1157 int cpu = mycpu->gd_cpuid; 1158 int globalstl = 0; 1159 1160 pf_status.fcounters[FCNT_STATE_SEARCH]++; 1161 1162 if (dir == PF_OUT && m->m_pkthdr.pf.statekey && 1163 ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->reverse) { 1164 sk = ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->reverse; 1165 } else { 1166 sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey); 1167 if (sk == NULL) { 1168 lockmgr(&pf_global_statetbl_lock, LK_SHARED); 1169 sk = RB_FIND(pf_state_tree, &pf_statetbl[MAXCPU], skey); 1170 if (sk == NULL) { 1171 lockmgr(&pf_global_statetbl_lock, LK_RELEASE); 1172 return (NULL); 1173 } 1174 globalstl = 1; 1175 } 1176 if (dir == PF_OUT && m->m_pkthdr.pf.statekey) { 1177 ((struct pf_state_key *) 1178 m->m_pkthdr.pf.statekey)->reverse = sk; 1179 sk->reverse = m->m_pkthdr.pf.statekey; 1180 } 1181 } 1182 if (dir == PF_OUT) 1183 m->m_pkthdr.pf.statekey = NULL; 1184 1185 /* list is sorted, if-bound states before floating ones */ 1186 TAILQ_FOREACH(si, &sk->states, entry) { 1187 if ((si->s->kif == pfi_all || si->s->kif == kif) && 1188 sk == (dir == PF_IN ? si->s->key[PF_SK_WIRE] : 1189 si->s->key[PF_SK_STACK])) { 1190 break; 1191 } 1192 } 1193 1194 /* 1195 * Extract state before potentially releasing the global statetbl 1196 * lock. Ignore the state if the create is still in-progress as 1197 * it can be deleted out from under us by the owning localized cpu. 1198 * However, if CREATEINPROG is not set, state can only be deleted 1199 * by the purge thread which we are protected from via our shared 1200 * pf_token. 1201 */ 1202 if (si) { 1203 s = si->s; 1204 if (s && (s->state_flags & PFSTATE_CREATEINPROG)) 1205 s = NULL; 1206 } else { 1207 s = NULL; 1208 } 1209 if (globalstl) 1210 lockmgr(&pf_global_statetbl_lock, LK_RELEASE); 1211 return s; 1212 } 1213 1214 /* 1215 * WARNING! May return a state structure that was localized to another cpu, 1216 * destruction is typically protected by the callers pf_token. 1217 */ 1218 struct pf_state * 1219 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1220 { 1221 struct pf_state_key *skey = (void *)key; 1222 struct pf_state_key *sk; 1223 struct pf_state_item *si, *ret = NULL; 1224 struct pf_state *s; 1225 int cpu = mycpu->gd_cpuid; 1226 int globalstl = 0; 1227 1228 pf_status.fcounters[FCNT_STATE_SEARCH]++; 1229 1230 sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey); 1231 if (sk == NULL) { 1232 lockmgr(&pf_global_statetbl_lock, LK_SHARED); 1233 sk = RB_FIND(pf_state_tree, &pf_statetbl[MAXCPU], skey); 1234 globalstl = 1; 1235 } 1236 if (sk != NULL) { 1237 TAILQ_FOREACH(si, &sk->states, entry) 1238 if (dir == PF_INOUT || 1239 (sk == (dir == PF_IN ? si->s->key[PF_SK_WIRE] : 1240 si->s->key[PF_SK_STACK]))) { 1241 if (more == NULL) { 1242 ret = si; 1243 break; 1244 } 1245 if (ret) 1246 (*more)++; 1247 else 1248 ret = si; 1249 } 1250 } 1251 1252 /* 1253 * Extract state before potentially releasing the global statetbl 1254 * lock. Ignore the state if the create is still in-progress as 1255 * it can be deleted out from under us by the owning localized cpu. 1256 * However, if CREATEINPROG is not set, state can only be deleted 1257 * by the purge thread which we are protected from via our shared 1258 * pf_token. 1259 */ 1260 if (ret) { 1261 s = ret->s; 1262 if (s && (s->state_flags & PFSTATE_CREATEINPROG)) 1263 s = NULL; 1264 } else { 1265 s = NULL; 1266 } 1267 if (globalstl) 1268 lockmgr(&pf_global_statetbl_lock, LK_RELEASE); 1269 return s; 1270 } 1271 1272 /* END state table stuff */ 1273 1274 void 1275 pf_purge_thread(void *v) 1276 { 1277 globaldata_t save_gd = mycpu; 1278 int nloops = 0; 1279 int locked = 0; 1280 int nn; 1281 int endingit; 1282 1283 for (;;) { 1284 tsleep(pf_purge_thread, PWAIT, "pftm", 1 * hz); 1285 1286 endingit = pf_end_threads; 1287 1288 for (nn = 0; nn < ncpus; ++nn) { 1289 lwkt_setcpu_self(globaldata_find(nn)); 1290 1291 lwkt_gettoken(&pf_token); 1292 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE); 1293 crit_enter(); 1294 1295 /* 1296 * process a fraction of the state table every second 1297 */ 1298 if(!pf_purge_expired_states( 1299 1 + (pf_status.states / 1300 pf_default_rule.timeout[ 1301 PFTM_INTERVAL]), 0)) { 1302 pf_purge_expired_states( 1303 1 + (pf_status.states / 1304 pf_default_rule.timeout[ 1305 PFTM_INTERVAL]), 1); 1306 } 1307 1308 /* 1309 * purge other expired types every PFTM_INTERVAL 1310 * seconds 1311 */ 1312 if (++nloops >= 1313 pf_default_rule.timeout[PFTM_INTERVAL]) { 1314 pf_purge_expired_fragments(); 1315 if (!pf_purge_expired_src_nodes(locked)) { 1316 pf_purge_expired_src_nodes(1); 1317 } 1318 nloops = 0; 1319 } 1320 1321 /* 1322 * If terminating the thread, clean everything out 1323 * (on all cpus). 1324 */ 1325 if (endingit) { 1326 pf_purge_expired_states(pf_status.states, 0); 1327 pf_purge_expired_fragments(); 1328 pf_purge_expired_src_nodes(1); 1329 } 1330 1331 crit_exit(); 1332 lockmgr(&pf_consistency_lock, LK_RELEASE); 1333 lwkt_reltoken(&pf_token); 1334 } 1335 lwkt_setcpu_self(save_gd); 1336 if (endingit) 1337 break; 1338 } 1339 1340 /* 1341 * Thread termination 1342 */ 1343 pf_end_threads++; 1344 wakeup(pf_purge_thread); 1345 kthread_exit(); 1346 } 1347 1348 u_int32_t 1349 pf_state_expires(const struct pf_state *state) 1350 { 1351 u_int32_t timeout; 1352 u_int32_t start; 1353 u_int32_t end; 1354 u_int32_t states; 1355 1356 /* handle all PFTM_* > PFTM_MAX here */ 1357 if (state->timeout == PFTM_PURGE) 1358 return (time_second); 1359 if (state->timeout == PFTM_UNTIL_PACKET) 1360 return (0); 1361 KKASSERT(state->timeout != PFTM_UNLINKED); 1362 KKASSERT(state->timeout < PFTM_MAX); 1363 timeout = state->rule.ptr->timeout[state->timeout]; 1364 if (!timeout) 1365 timeout = pf_default_rule.timeout[state->timeout]; 1366 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1367 if (start) { 1368 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1369 states = state->rule.ptr->states_cur; 1370 } else { 1371 start = pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1372 end = pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1373 states = pf_status.states; 1374 } 1375 if (end && states > start && start < end) { 1376 if (states < end) 1377 return (state->expire + timeout * (end - states) / 1378 (end - start)); 1379 else 1380 return (time_second); 1381 } 1382 return (state->expire + timeout); 1383 } 1384 1385 /* 1386 * (called with exclusive pf_token) 1387 */ 1388 int 1389 pf_purge_expired_src_nodes(int waslocked) 1390 { 1391 struct pf_src_node *cur, *next; 1392 int locked = waslocked; 1393 int cpu = mycpu->gd_cpuid; 1394 1395 for (cur = RB_MIN(pf_src_tree, &tree_src_tracking[cpu]); 1396 cur; 1397 cur = next) { 1398 next = RB_NEXT(pf_src_tree, &tree_src_tracking[cpu], cur); 1399 1400 if (cur->states <= 0 && cur->expire <= time_second) { 1401 if (!locked) { 1402 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE); 1403 next = RB_NEXT(pf_src_tree, 1404 &tree_src_tracking[cpu], cur); 1405 locked = 1; 1406 } 1407 if (cur->rule.ptr != NULL) { 1408 /* 1409 * decrements in rule should be ok, token is 1410 * held exclusively in this code path. 1411 */ 1412 cur->rule.ptr->src_nodes--; 1413 if (cur->rule.ptr->states_cur <= 0 && 1414 cur->rule.ptr->max_src_nodes <= 0) 1415 pf_rm_rule(NULL, cur->rule.ptr); 1416 } 1417 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], cur); 1418 pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++; 1419 atomic_add_int(&pf_status.src_nodes, -1); 1420 kfree(cur, M_PFSRCTREEPL); 1421 } 1422 } 1423 if (locked && !waslocked) 1424 lockmgr(&pf_consistency_lock, LK_RELEASE); 1425 return(1); 1426 } 1427 1428 void 1429 pf_src_tree_remove_state(struct pf_state *s) 1430 { 1431 u_int32_t timeout; 1432 1433 if (s->src_node != NULL) { 1434 if (s->src.tcp_est) 1435 --s->src_node->conn; 1436 if (--s->src_node->states <= 0) { 1437 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE]; 1438 if (!timeout) { 1439 timeout = 1440 pf_default_rule.timeout[PFTM_SRC_NODE]; 1441 } 1442 s->src_node->expire = time_second + timeout; 1443 } 1444 } 1445 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 1446 if (--s->nat_src_node->states <= 0) { 1447 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE]; 1448 if (!timeout) 1449 timeout = 1450 pf_default_rule.timeout[PFTM_SRC_NODE]; 1451 s->nat_src_node->expire = time_second + timeout; 1452 } 1453 } 1454 s->src_node = s->nat_src_node = NULL; 1455 } 1456 1457 /* callers should be at crit_enter() */ 1458 void 1459 pf_unlink_state(struct pf_state *cur) 1460 { 1461 int cpu = mycpu->gd_cpuid; 1462 1463 if (cur->src.state == PF_TCPS_PROXY_DST) { 1464 /* XXX wire key the right one? */ 1465 pf_send_tcp(cur->rule.ptr, cur->key[PF_SK_WIRE]->af, 1466 &cur->key[PF_SK_WIRE]->addr[1], 1467 &cur->key[PF_SK_WIRE]->addr[0], 1468 cur->key[PF_SK_WIRE]->port[1], 1469 cur->key[PF_SK_WIRE]->port[0], 1470 cur->src.seqhi, cur->src.seqlo + 1, 1471 TH_RST|TH_ACK, 0, 0, 0, 1, cur->tag, NULL, NULL); 1472 } 1473 RB_REMOVE(pf_state_tree_id, &tree_id[cpu], cur); 1474 if (cur->creatorid == pf_status.hostid) 1475 pfsync_delete_state(cur); 1476 cur->timeout = PFTM_UNLINKED; 1477 pf_src_tree_remove_state(cur); 1478 pf_detach_state(cur); 1479 } 1480 1481 static struct pf_state *purge_cur[MAXCPU]; 1482 1483 /* 1484 * callers should be at crit_enter() and hold pf_consistency_lock exclusively. 1485 * pf_token must also be held exclusively. 1486 */ 1487 void 1488 pf_free_state(struct pf_state *cur) 1489 { 1490 int cpu = mycpu->gd_cpuid; 1491 1492 KKASSERT(cur->cpuid == cpu); 1493 1494 if (pfsyncif != NULL && 1495 (pfsyncif->sc_bulk_send_next == cur || 1496 pfsyncif->sc_bulk_terminator == cur)) 1497 return; 1498 KKASSERT(cur->timeout == PFTM_UNLINKED); 1499 /* 1500 * decrements in rule should be ok, token is 1501 * held exclusively in this code path. 1502 */ 1503 if (--cur->rule.ptr->states_cur <= 0 && 1504 cur->rule.ptr->src_nodes <= 0) 1505 pf_rm_rule(NULL, cur->rule.ptr); 1506 if (cur->nat_rule.ptr != NULL) { 1507 if (--cur->nat_rule.ptr->states_cur <= 0 && 1508 cur->nat_rule.ptr->src_nodes <= 0) { 1509 pf_rm_rule(NULL, cur->nat_rule.ptr); 1510 } 1511 } 1512 if (cur->anchor.ptr != NULL) { 1513 if (--cur->anchor.ptr->states_cur <= 0) 1514 pf_rm_rule(NULL, cur->anchor.ptr); 1515 } 1516 pf_normalize_tcp_cleanup(cur); 1517 pfi_kif_unref(cur->kif, PFI_KIF_REF_STATE); 1518 1519 /* 1520 * We may be freeing pf_purge_expired_states()'s saved scan entry, 1521 * adjust it if necessary. 1522 */ 1523 if (purge_cur[cpu] == cur) { 1524 kprintf("PURGE CONFLICT\n"); 1525 purge_cur[cpu] = TAILQ_NEXT(purge_cur[cpu], entry_list); 1526 } 1527 TAILQ_REMOVE(&state_list[cpu], cur, entry_list); 1528 if (cur->tag) 1529 pf_tag_unref(cur->tag); 1530 kfree(cur, M_PFSTATEPL); 1531 pf_status.fcounters[FCNT_STATE_REMOVALS]++; 1532 atomic_add_int(&pf_status.states, -1); 1533 } 1534 1535 int 1536 pf_purge_expired_states(u_int32_t maxcheck, int waslocked) 1537 { 1538 struct pf_state *cur; 1539 int locked = waslocked; 1540 int cpu = mycpu->gd_cpuid; 1541 1542 while (maxcheck--) { 1543 /* 1544 * Wrap to start of list when we hit the end 1545 */ 1546 cur = purge_cur[cpu]; 1547 if (cur == NULL) { 1548 cur = TAILQ_FIRST(&state_list[cpu]); 1549 if (cur == NULL) 1550 break; /* list empty */ 1551 } 1552 1553 /* 1554 * Setup next (purge_cur) while we process this one. If 1555 * we block and something else deletes purge_cur, 1556 * pf_free_state() will adjust it further ahead. 1557 */ 1558 purge_cur[cpu] = TAILQ_NEXT(cur, entry_list); 1559 1560 if (cur->timeout == PFTM_UNLINKED) { 1561 /* free unlinked state */ 1562 if (! locked) { 1563 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE); 1564 locked = 1; 1565 } 1566 pf_free_state(cur); 1567 } else if (pf_state_expires(cur) <= time_second) { 1568 /* unlink and free expired state */ 1569 pf_unlink_state(cur); 1570 if (! locked) { 1571 if (!lockmgr(&pf_consistency_lock, LK_EXCLUSIVE)) 1572 return (0); 1573 locked = 1; 1574 } 1575 pf_free_state(cur); 1576 } 1577 } 1578 1579 if (locked) 1580 lockmgr(&pf_consistency_lock, LK_RELEASE); 1581 return (1); 1582 } 1583 1584 int 1585 pf_tbladdr_setup(struct pf_ruleset *rs, struct pf_addr_wrap *aw) 1586 { 1587 if (aw->type != PF_ADDR_TABLE) 1588 return (0); 1589 if ((aw->p.tbl = pfr_attach_table(rs, aw->v.tblname)) == NULL) 1590 return (1); 1591 return (0); 1592 } 1593 1594 void 1595 pf_tbladdr_remove(struct pf_addr_wrap *aw) 1596 { 1597 if (aw->type != PF_ADDR_TABLE || aw->p.tbl == NULL) 1598 return; 1599 pfr_detach_table(aw->p.tbl); 1600 aw->p.tbl = NULL; 1601 } 1602 1603 void 1604 pf_tbladdr_copyout(struct pf_addr_wrap *aw) 1605 { 1606 struct pfr_ktable *kt = aw->p.tbl; 1607 1608 if (aw->type != PF_ADDR_TABLE || kt == NULL) 1609 return; 1610 if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) 1611 kt = kt->pfrkt_root; 1612 aw->p.tbl = NULL; 1613 aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ? 1614 kt->pfrkt_cnt : -1; 1615 } 1616 1617 void 1618 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 1619 { 1620 switch (af) { 1621 #ifdef INET 1622 case AF_INET: { 1623 u_int32_t a = ntohl(addr->addr32[0]); 1624 kprintf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 1625 (a>>8)&255, a&255); 1626 if (p) { 1627 p = ntohs(p); 1628 kprintf(":%u", p); 1629 } 1630 break; 1631 } 1632 #endif /* INET */ 1633 #ifdef INET6 1634 case AF_INET6: { 1635 u_int16_t b; 1636 u_int8_t i, curstart = 255, curend = 0, 1637 maxstart = 0, maxend = 0; 1638 for (i = 0; i < 8; i++) { 1639 if (!addr->addr16[i]) { 1640 if (curstart == 255) 1641 curstart = i; 1642 else 1643 curend = i; 1644 } else { 1645 if (curstart) { 1646 if ((curend - curstart) > 1647 (maxend - maxstart)) { 1648 maxstart = curstart; 1649 maxend = curend; 1650 curstart = 255; 1651 } 1652 } 1653 } 1654 } 1655 for (i = 0; i < 8; i++) { 1656 if (i >= maxstart && i <= maxend) { 1657 if (maxend != 7) { 1658 if (i == maxstart) 1659 kprintf(":"); 1660 } else { 1661 if (i == maxend) 1662 kprintf(":"); 1663 } 1664 } else { 1665 b = ntohs(addr->addr16[i]); 1666 kprintf("%x", b); 1667 if (i < 7) 1668 kprintf(":"); 1669 } 1670 } 1671 if (p) { 1672 p = ntohs(p); 1673 kprintf("[%u]", p); 1674 } 1675 break; 1676 } 1677 #endif /* INET6 */ 1678 } 1679 } 1680 1681 void 1682 pf_print_state(struct pf_state *s) 1683 { 1684 pf_print_state_parts(s, NULL, NULL); 1685 } 1686 1687 void 1688 pf_print_state_parts(struct pf_state *s, 1689 struct pf_state_key *skwp, struct pf_state_key *sksp) 1690 { 1691 struct pf_state_key *skw, *sks; 1692 u_int8_t proto, dir; 1693 1694 /* Do our best to fill these, but they're skipped if NULL */ 1695 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 1696 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 1697 proto = skw ? skw->proto : (sks ? sks->proto : 0); 1698 dir = s ? s->direction : 0; 1699 1700 switch (proto) { 1701 case IPPROTO_TCP: 1702 kprintf("TCP "); 1703 break; 1704 case IPPROTO_UDP: 1705 kprintf("UDP "); 1706 break; 1707 case IPPROTO_ICMP: 1708 kprintf("ICMP "); 1709 break; 1710 case IPPROTO_ICMPV6: 1711 kprintf("ICMPV6 "); 1712 break; 1713 default: 1714 kprintf("%u ", skw->proto); 1715 break; 1716 } 1717 switch (dir) { 1718 case PF_IN: 1719 kprintf(" in"); 1720 break; 1721 case PF_OUT: 1722 kprintf(" out"); 1723 break; 1724 } 1725 if (skw) { 1726 kprintf(" wire: "); 1727 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 1728 kprintf(" "); 1729 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 1730 } 1731 if (sks) { 1732 kprintf(" stack: "); 1733 if (sks != skw) { 1734 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 1735 kprintf(" "); 1736 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 1737 } else 1738 kprintf("-"); 1739 } 1740 if (s) { 1741 if (proto == IPPROTO_TCP) { 1742 kprintf(" [lo=%u high=%u win=%u modulator=%u", 1743 s->src.seqlo, s->src.seqhi, 1744 s->src.max_win, s->src.seqdiff); 1745 if (s->src.wscale && s->dst.wscale) 1746 kprintf(" wscale=%u", 1747 s->src.wscale & PF_WSCALE_MASK); 1748 kprintf("]"); 1749 kprintf(" [lo=%u high=%u win=%u modulator=%u", 1750 s->dst.seqlo, s->dst.seqhi, 1751 s->dst.max_win, s->dst.seqdiff); 1752 if (s->src.wscale && s->dst.wscale) 1753 kprintf(" wscale=%u", 1754 s->dst.wscale & PF_WSCALE_MASK); 1755 kprintf("]"); 1756 } 1757 kprintf(" %u:%u", s->src.state, s->dst.state); 1758 } 1759 } 1760 1761 void 1762 pf_print_flags(u_int8_t f) 1763 { 1764 if (f) 1765 kprintf(" "); 1766 if (f & TH_FIN) 1767 kprintf("F"); 1768 if (f & TH_SYN) 1769 kprintf("S"); 1770 if (f & TH_RST) 1771 kprintf("R"); 1772 if (f & TH_PUSH) 1773 kprintf("P"); 1774 if (f & TH_ACK) 1775 kprintf("A"); 1776 if (f & TH_URG) 1777 kprintf("U"); 1778 if (f & TH_ECE) 1779 kprintf("E"); 1780 if (f & TH_CWR) 1781 kprintf("W"); 1782 } 1783 1784 #define PF_SET_SKIP_STEPS(i) \ 1785 do { \ 1786 while (head[i] != cur) { \ 1787 head[i]->skip[i].ptr = cur; \ 1788 head[i] = TAILQ_NEXT(head[i], entries); \ 1789 } \ 1790 } while (0) 1791 1792 void 1793 pf_calc_skip_steps(struct pf_rulequeue *rules) 1794 { 1795 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT]; 1796 int i; 1797 1798 cur = TAILQ_FIRST(rules); 1799 prev = cur; 1800 for (i = 0; i < PF_SKIP_COUNT; ++i) 1801 head[i] = cur; 1802 while (cur != NULL) { 1803 1804 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 1805 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 1806 if (cur->direction != prev->direction) 1807 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 1808 if (cur->af != prev->af) 1809 PF_SET_SKIP_STEPS(PF_SKIP_AF); 1810 if (cur->proto != prev->proto) 1811 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 1812 if (cur->src.neg != prev->src.neg || 1813 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 1814 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 1815 if (cur->src.port[0] != prev->src.port[0] || 1816 cur->src.port[1] != prev->src.port[1] || 1817 cur->src.port_op != prev->src.port_op) 1818 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 1819 if (cur->dst.neg != prev->dst.neg || 1820 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 1821 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 1822 if (cur->dst.port[0] != prev->dst.port[0] || 1823 cur->dst.port[1] != prev->dst.port[1] || 1824 cur->dst.port_op != prev->dst.port_op) 1825 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 1826 1827 prev = cur; 1828 cur = TAILQ_NEXT(cur, entries); 1829 } 1830 for (i = 0; i < PF_SKIP_COUNT; ++i) 1831 PF_SET_SKIP_STEPS(i); 1832 } 1833 1834 int 1835 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 1836 { 1837 if (aw1->type != aw2->type) 1838 return (1); 1839 switch (aw1->type) { 1840 case PF_ADDR_ADDRMASK: 1841 case PF_ADDR_RANGE: 1842 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0)) 1843 return (1); 1844 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0)) 1845 return (1); 1846 return (0); 1847 case PF_ADDR_DYNIFTL: 1848 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 1849 case PF_ADDR_NOROUTE: 1850 case PF_ADDR_URPFFAILED: 1851 return (0); 1852 case PF_ADDR_TABLE: 1853 return (aw1->p.tbl != aw2->p.tbl); 1854 case PF_ADDR_RTLABEL: 1855 return (aw1->v.rtlabel != aw2->v.rtlabel); 1856 default: 1857 kprintf("invalid address type: %d\n", aw1->type); 1858 return (1); 1859 } 1860 } 1861 1862 u_int16_t 1863 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 1864 { 1865 u_int32_t l; 1866 1867 if (udp && !cksum) 1868 return (0x0000); 1869 l = cksum + old - new; 1870 l = (l >> 16) + (l & 65535); 1871 l = l & 65535; 1872 if (udp && !l) 1873 return (0xFFFF); 1874 return (l); 1875 } 1876 1877 void 1878 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc, 1879 struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af) 1880 { 1881 struct pf_addr ao; 1882 u_int16_t po = *p; 1883 1884 PF_ACPY(&ao, a, af); 1885 PF_ACPY(a, an, af); 1886 1887 *p = pn; 1888 1889 switch (af) { 1890 #ifdef INET 1891 case AF_INET: 1892 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 1893 ao.addr16[0], an->addr16[0], 0), 1894 ao.addr16[1], an->addr16[1], 0); 1895 *p = pn; 1896 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 1897 ao.addr16[0], an->addr16[0], u), 1898 ao.addr16[1], an->addr16[1], u), 1899 po, pn, u); 1900 break; 1901 #endif /* INET */ 1902 #ifdef INET6 1903 case AF_INET6: 1904 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 1905 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 1906 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 1907 ao.addr16[0], an->addr16[0], u), 1908 ao.addr16[1], an->addr16[1], u), 1909 ao.addr16[2], an->addr16[2], u), 1910 ao.addr16[3], an->addr16[3], u), 1911 ao.addr16[4], an->addr16[4], u), 1912 ao.addr16[5], an->addr16[5], u), 1913 ao.addr16[6], an->addr16[6], u), 1914 ao.addr16[7], an->addr16[7], u), 1915 po, pn, u); 1916 break; 1917 #endif /* INET6 */ 1918 } 1919 } 1920 1921 1922 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 1923 void 1924 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 1925 { 1926 u_int32_t ao; 1927 1928 memcpy(&ao, a, sizeof(ao)); 1929 memcpy(a, &an, sizeof(u_int32_t)); 1930 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 1931 ao % 65536, an % 65536, u); 1932 } 1933 1934 #ifdef INET6 1935 void 1936 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 1937 { 1938 struct pf_addr ao; 1939 1940 PF_ACPY(&ao, a, AF_INET6); 1941 PF_ACPY(a, an, AF_INET6); 1942 1943 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 1944 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 1945 pf_cksum_fixup(pf_cksum_fixup(*c, 1946 ao.addr16[0], an->addr16[0], u), 1947 ao.addr16[1], an->addr16[1], u), 1948 ao.addr16[2], an->addr16[2], u), 1949 ao.addr16[3], an->addr16[3], u), 1950 ao.addr16[4], an->addr16[4], u), 1951 ao.addr16[5], an->addr16[5], u), 1952 ao.addr16[6], an->addr16[6], u), 1953 ao.addr16[7], an->addr16[7], u); 1954 } 1955 #endif /* INET6 */ 1956 1957 void 1958 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 1959 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 1960 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 1961 { 1962 struct pf_addr oia, ooa; 1963 1964 PF_ACPY(&oia, ia, af); 1965 if (oa) 1966 PF_ACPY(&ooa, oa, af); 1967 1968 /* Change inner protocol port, fix inner protocol checksum. */ 1969 if (ip != NULL) { 1970 u_int16_t oip = *ip; 1971 u_int32_t opc = 0; 1972 1973 if (pc != NULL) 1974 opc = *pc; 1975 *ip = np; 1976 if (pc != NULL) 1977 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 1978 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 1979 if (pc != NULL) 1980 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 1981 } 1982 /* Change inner ip address, fix inner ip and icmp checksums. */ 1983 PF_ACPY(ia, na, af); 1984 switch (af) { 1985 #ifdef INET 1986 case AF_INET: { 1987 u_int32_t oh2c = *h2c; 1988 1989 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 1990 oia.addr16[0], ia->addr16[0], 0), 1991 oia.addr16[1], ia->addr16[1], 0); 1992 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 1993 oia.addr16[0], ia->addr16[0], 0), 1994 oia.addr16[1], ia->addr16[1], 0); 1995 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 1996 break; 1997 } 1998 #endif /* INET */ 1999 #ifdef INET6 2000 case AF_INET6: 2001 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2002 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2003 pf_cksum_fixup(pf_cksum_fixup(*ic, 2004 oia.addr16[0], ia->addr16[0], u), 2005 oia.addr16[1], ia->addr16[1], u), 2006 oia.addr16[2], ia->addr16[2], u), 2007 oia.addr16[3], ia->addr16[3], u), 2008 oia.addr16[4], ia->addr16[4], u), 2009 oia.addr16[5], ia->addr16[5], u), 2010 oia.addr16[6], ia->addr16[6], u), 2011 oia.addr16[7], ia->addr16[7], u); 2012 break; 2013 #endif /* INET6 */ 2014 } 2015 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2016 if (oa) { 2017 PF_ACPY(oa, na, af); 2018 switch (af) { 2019 #ifdef INET 2020 case AF_INET: 2021 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2022 ooa.addr16[0], oa->addr16[0], 0), 2023 ooa.addr16[1], oa->addr16[1], 0); 2024 break; 2025 #endif /* INET */ 2026 #ifdef INET6 2027 case AF_INET6: 2028 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2029 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2030 pf_cksum_fixup(pf_cksum_fixup(*ic, 2031 ooa.addr16[0], oa->addr16[0], u), 2032 ooa.addr16[1], oa->addr16[1], u), 2033 ooa.addr16[2], oa->addr16[2], u), 2034 ooa.addr16[3], oa->addr16[3], u), 2035 ooa.addr16[4], oa->addr16[4], u), 2036 ooa.addr16[5], oa->addr16[5], u), 2037 ooa.addr16[6], oa->addr16[6], u), 2038 ooa.addr16[7], oa->addr16[7], u); 2039 break; 2040 #endif /* INET6 */ 2041 } 2042 } 2043 } 2044 2045 2046 /* 2047 * Need to modulate the sequence numbers in the TCP SACK option 2048 * (credits to Krzysztof Pfaff for report and patch) 2049 */ 2050 int 2051 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2052 struct tcphdr *th, struct pf_state_peer *dst) 2053 { 2054 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2055 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2056 int copyback = 0, i, olen; 2057 struct raw_sackblock sack; 2058 2059 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2060 if (hlen < TCPOLEN_SACKLEN || 2061 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2062 return 0; 2063 2064 while (hlen >= TCPOLEN_SACKLEN) { 2065 olen = opt[1]; 2066 switch (*opt) { 2067 case TCPOPT_EOL: /* FALLTHROUGH */ 2068 case TCPOPT_NOP: 2069 opt++; 2070 hlen--; 2071 break; 2072 case TCPOPT_SACK: 2073 if (olen > hlen) 2074 olen = hlen; 2075 if (olen >= TCPOLEN_SACKLEN) { 2076 for (i = 2; i + TCPOLEN_SACK <= olen; 2077 i += TCPOLEN_SACK) { 2078 memcpy(&sack, &opt[i], sizeof(sack)); 2079 pf_change_a(&sack.rblk_start, &th->th_sum, 2080 htonl(ntohl(sack.rblk_start) - 2081 dst->seqdiff), 0); 2082 pf_change_a(&sack.rblk_end, &th->th_sum, 2083 htonl(ntohl(sack.rblk_end) - 2084 dst->seqdiff), 0); 2085 memcpy(&opt[i], &sack, sizeof(sack)); 2086 } 2087 copyback = 1; 2088 } 2089 /* FALLTHROUGH */ 2090 default: 2091 if (olen < 2) 2092 olen = 2; 2093 hlen -= olen; 2094 opt += olen; 2095 } 2096 } 2097 2098 if (copyback) 2099 m_copyback(m, off + sizeof(*th), thoptlen, opts); 2100 return (copyback); 2101 } 2102 2103 void 2104 pf_send_tcp(const struct pf_rule *r, sa_family_t af, 2105 const struct pf_addr *saddr, const struct pf_addr *daddr, 2106 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2107 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2108 u_int16_t rtag, struct ether_header *eh, struct ifnet *ifp) 2109 { 2110 struct mbuf *m; 2111 int len = 0, tlen; 2112 #ifdef INET 2113 struct ip *h = NULL; 2114 #endif /* INET */ 2115 #ifdef INET6 2116 struct ip6_hdr *h6 = NULL; 2117 #endif /* INET6 */ 2118 struct tcphdr *th = NULL; 2119 char *opt; 2120 2121 ASSERT_LWKT_TOKEN_HELD(&pf_token); 2122 2123 /* maximum segment size tcp option */ 2124 tlen = sizeof(struct tcphdr); 2125 if (mss) 2126 tlen += 4; 2127 2128 switch (af) { 2129 #ifdef INET 2130 case AF_INET: 2131 len = sizeof(struct ip) + tlen; 2132 break; 2133 #endif /* INET */ 2134 #ifdef INET6 2135 case AF_INET6: 2136 len = sizeof(struct ip6_hdr) + tlen; 2137 break; 2138 #endif /* INET6 */ 2139 } 2140 2141 /* 2142 * Create outgoing mbuf. 2143 * 2144 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags, 2145 * so make sure pf.flags is clear. 2146 */ 2147 m = m_gethdr(M_NOWAIT, MT_HEADER); 2148 if (m == NULL) { 2149 return; 2150 } 2151 if (tag) 2152 m->m_pkthdr.fw_flags |= PF_MBUF_TAGGED; 2153 m->m_pkthdr.pf.flags = 0; 2154 m->m_pkthdr.pf.tag = rtag; 2155 /* XXX Recheck when upgrading to > 4.4 */ 2156 m->m_pkthdr.pf.statekey = NULL; 2157 if (r != NULL && r->rtableid >= 0) 2158 m->m_pkthdr.pf.rtableid = r->rtableid; 2159 2160 #ifdef ALTQ 2161 if (r != NULL && r->qid) { 2162 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE; 2163 m->m_pkthdr.pf.qid = r->qid; 2164 m->m_pkthdr.pf.ecn_af = af; 2165 m->m_pkthdr.pf.hdr = mtod(m, struct ip *); 2166 } 2167 #endif /* ALTQ */ 2168 m->m_data += max_linkhdr; 2169 m->m_pkthdr.len = m->m_len = len; 2170 m->m_pkthdr.rcvif = NULL; 2171 bzero(m->m_data, len); 2172 switch (af) { 2173 #ifdef INET 2174 case AF_INET: 2175 h = mtod(m, struct ip *); 2176 2177 /* IP header fields included in the TCP checksum */ 2178 h->ip_p = IPPROTO_TCP; 2179 h->ip_len = tlen; 2180 h->ip_src.s_addr = saddr->v4.s_addr; 2181 h->ip_dst.s_addr = daddr->v4.s_addr; 2182 2183 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2184 break; 2185 #endif /* INET */ 2186 #ifdef INET6 2187 case AF_INET6: 2188 h6 = mtod(m, struct ip6_hdr *); 2189 2190 /* IP header fields included in the TCP checksum */ 2191 h6->ip6_nxt = IPPROTO_TCP; 2192 h6->ip6_plen = htons(tlen); 2193 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2194 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2195 2196 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2197 break; 2198 #endif /* INET6 */ 2199 } 2200 2201 /* TCP header */ 2202 th->th_sport = sport; 2203 th->th_dport = dport; 2204 th->th_seq = htonl(seq); 2205 th->th_ack = htonl(ack); 2206 th->th_off = tlen >> 2; 2207 th->th_flags = flags; 2208 th->th_win = htons(win); 2209 2210 if (mss) { 2211 opt = (char *)(th + 1); 2212 opt[0] = TCPOPT_MAXSEG; 2213 opt[1] = 4; 2214 mss = htons(mss); 2215 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2216 } 2217 2218 switch (af) { 2219 #ifdef INET 2220 case AF_INET: 2221 /* TCP checksum */ 2222 th->th_sum = in_cksum(m, len); 2223 2224 /* Finish the IP header */ 2225 h->ip_v = 4; 2226 h->ip_hl = sizeof(*h) >> 2; 2227 h->ip_tos = IPTOS_LOWDELAY; 2228 h->ip_len = len; 2229 h->ip_off = path_mtu_discovery ? IP_DF : 0; 2230 h->ip_ttl = ttl ? ttl : ip_defttl; 2231 h->ip_sum = 0; 2232 if (eh == NULL) { 2233 lwkt_reltoken(&pf_token); 2234 ip_output(m, NULL, NULL, 0, NULL, NULL); 2235 lwkt_gettoken(&pf_token); 2236 } else { 2237 struct route ro; 2238 struct rtentry rt; 2239 struct ether_header *e = (void *)ro.ro_dst.sa_data; 2240 2241 if (ifp == NULL) { 2242 m_freem(m); 2243 return; 2244 } 2245 rt.rt_ifp = ifp; 2246 ro.ro_rt = &rt; 2247 ro.ro_dst.sa_len = sizeof(ro.ro_dst); 2248 ro.ro_dst.sa_family = pseudo_AF_HDRCMPLT; 2249 bcopy(eh->ether_dhost, e->ether_shost, ETHER_ADDR_LEN); 2250 bcopy(eh->ether_shost, e->ether_dhost, ETHER_ADDR_LEN); 2251 e->ether_type = eh->ether_type; 2252 /* XXX_IMPORT: later */ 2253 lwkt_reltoken(&pf_token); 2254 ip_output(m, NULL, &ro, 0, NULL, NULL); 2255 lwkt_gettoken(&pf_token); 2256 } 2257 break; 2258 #endif /* INET */ 2259 #ifdef INET6 2260 case AF_INET6: 2261 /* TCP checksum */ 2262 th->th_sum = in6_cksum(m, IPPROTO_TCP, 2263 sizeof(struct ip6_hdr), tlen); 2264 2265 h6->ip6_vfc |= IPV6_VERSION; 2266 h6->ip6_hlim = IPV6_DEFHLIM; 2267 2268 lwkt_reltoken(&pf_token); 2269 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 2270 lwkt_gettoken(&pf_token); 2271 break; 2272 #endif /* INET6 */ 2273 } 2274 } 2275 2276 void 2277 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 2278 struct pf_rule *r) 2279 { 2280 struct mbuf *m0; 2281 2282 /* 2283 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags, 2284 * so make sure pf.flags is clear. 2285 */ 2286 if ((m0 = m_copy(m, 0, M_COPYALL)) == NULL) 2287 return; 2288 2289 m0->m_pkthdr.fw_flags |= PF_MBUF_TAGGED; 2290 m0->m_pkthdr.pf.flags = 0; 2291 /* XXX Re-Check when Upgrading to > 4.4 */ 2292 m0->m_pkthdr.pf.statekey = NULL; 2293 2294 if (r->rtableid >= 0) 2295 m0->m_pkthdr.pf.rtableid = r->rtableid; 2296 2297 #ifdef ALTQ 2298 if (r->qid) { 2299 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE; 2300 m0->m_pkthdr.pf.qid = r->qid; 2301 m0->m_pkthdr.pf.ecn_af = af; 2302 m0->m_pkthdr.pf.hdr = mtod(m0, struct ip *); 2303 } 2304 #endif /* ALTQ */ 2305 2306 switch (af) { 2307 #ifdef INET 2308 case AF_INET: 2309 icmp_error(m0, type, code, 0, 0); 2310 break; 2311 #endif /* INET */ 2312 #ifdef INET6 2313 case AF_INET6: 2314 icmp6_error(m0, type, code, 0); 2315 break; 2316 #endif /* INET6 */ 2317 } 2318 } 2319 2320 /* 2321 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 2322 * If n is 0, they match if they are equal. If n is != 0, they match if they 2323 * are different. 2324 */ 2325 int 2326 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 2327 struct pf_addr *b, sa_family_t af) 2328 { 2329 int match = 0; 2330 2331 switch (af) { 2332 #ifdef INET 2333 case AF_INET: 2334 if ((a->addr32[0] & m->addr32[0]) == 2335 (b->addr32[0] & m->addr32[0])) 2336 match++; 2337 break; 2338 #endif /* INET */ 2339 #ifdef INET6 2340 case AF_INET6: 2341 if (((a->addr32[0] & m->addr32[0]) == 2342 (b->addr32[0] & m->addr32[0])) && 2343 ((a->addr32[1] & m->addr32[1]) == 2344 (b->addr32[1] & m->addr32[1])) && 2345 ((a->addr32[2] & m->addr32[2]) == 2346 (b->addr32[2] & m->addr32[2])) && 2347 ((a->addr32[3] & m->addr32[3]) == 2348 (b->addr32[3] & m->addr32[3]))) 2349 match++; 2350 break; 2351 #endif /* INET6 */ 2352 } 2353 if (match) { 2354 if (n) 2355 return (0); 2356 else 2357 return (1); 2358 } else { 2359 if (n) 2360 return (1); 2361 else 2362 return (0); 2363 } 2364 } 2365 2366 /* 2367 * Return 1 if b <= a <= e, otherwise return 0. 2368 */ 2369 int 2370 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 2371 struct pf_addr *a, sa_family_t af) 2372 { 2373 switch (af) { 2374 #ifdef INET 2375 case AF_INET: 2376 if ((a->addr32[0] < b->addr32[0]) || 2377 (a->addr32[0] > e->addr32[0])) 2378 return (0); 2379 break; 2380 #endif /* INET */ 2381 #ifdef INET6 2382 case AF_INET6: { 2383 int i; 2384 2385 /* check a >= b */ 2386 for (i = 0; i < 4; ++i) 2387 if (a->addr32[i] > b->addr32[i]) 2388 break; 2389 else if (a->addr32[i] < b->addr32[i]) 2390 return (0); 2391 /* check a <= e */ 2392 for (i = 0; i < 4; ++i) 2393 if (a->addr32[i] < e->addr32[i]) 2394 break; 2395 else if (a->addr32[i] > e->addr32[i]) 2396 return (0); 2397 break; 2398 } 2399 #endif /* INET6 */ 2400 } 2401 return (1); 2402 } 2403 2404 int 2405 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 2406 { 2407 switch (op) { 2408 case PF_OP_IRG: 2409 return ((p > a1) && (p < a2)); 2410 case PF_OP_XRG: 2411 return ((p < a1) || (p > a2)); 2412 case PF_OP_RRG: 2413 return ((p >= a1) && (p <= a2)); 2414 case PF_OP_EQ: 2415 return (p == a1); 2416 case PF_OP_NE: 2417 return (p != a1); 2418 case PF_OP_LT: 2419 return (p < a1); 2420 case PF_OP_LE: 2421 return (p <= a1); 2422 case PF_OP_GT: 2423 return (p > a1); 2424 case PF_OP_GE: 2425 return (p >= a1); 2426 } 2427 return (0); /* never reached */ 2428 } 2429 2430 int 2431 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 2432 { 2433 a1 = ntohs(a1); 2434 a2 = ntohs(a2); 2435 p = ntohs(p); 2436 return (pf_match(op, a1, a2, p)); 2437 } 2438 2439 int 2440 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 2441 { 2442 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2443 return (0); 2444 return (pf_match(op, a1, a2, u)); 2445 } 2446 2447 int 2448 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 2449 { 2450 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2451 return (0); 2452 return (pf_match(op, a1, a2, g)); 2453 } 2454 2455 int 2456 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag) 2457 { 2458 if (*tag == -1) 2459 *tag = m->m_pkthdr.pf.tag; 2460 2461 return ((!r->match_tag_not && r->match_tag == *tag) || 2462 (r->match_tag_not && r->match_tag != *tag)); 2463 } 2464 2465 int 2466 pf_tag_packet(struct mbuf *m, int tag, int rtableid) 2467 { 2468 if (tag <= 0 && rtableid < 0) 2469 return (0); 2470 2471 if (tag > 0) 2472 m->m_pkthdr.pf.tag = tag; 2473 if (rtableid >= 0) 2474 m->m_pkthdr.pf.rtableid = rtableid; 2475 2476 return (0); 2477 } 2478 2479 void 2480 pf_step_into_anchor(int *depth, struct pf_ruleset **rs, int n, 2481 struct pf_rule **r, struct pf_rule **a, int *match) 2482 { 2483 struct pf_anchor_stackframe *f; 2484 2485 (*r)->anchor->match = 0; 2486 if (match) 2487 *match = 0; 2488 if (*depth >= NELEM(pf_anchor_stack)) { 2489 kprintf("pf_step_into_anchor: stack overflow\n"); 2490 *r = TAILQ_NEXT(*r, entries); 2491 return; 2492 } else if (*depth == 0 && a != NULL) 2493 *a = *r; 2494 f = pf_anchor_stack + (*depth)++; 2495 f->rs = *rs; 2496 f->r = *r; 2497 if ((*r)->anchor_wildcard) { 2498 f->parent = &(*r)->anchor->children; 2499 if ((f->child = RB_MIN(pf_anchor_node, f->parent)) == 2500 NULL) { 2501 *r = NULL; 2502 return; 2503 } 2504 *rs = &f->child->ruleset; 2505 } else { 2506 f->parent = NULL; 2507 f->child = NULL; 2508 *rs = &(*r)->anchor->ruleset; 2509 } 2510 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2511 } 2512 2513 int 2514 pf_step_out_of_anchor(int *depth, struct pf_ruleset **rs, int n, 2515 struct pf_rule **r, struct pf_rule **a, int *match) 2516 { 2517 struct pf_anchor_stackframe *f; 2518 int quick = 0; 2519 2520 do { 2521 if (*depth <= 0) 2522 break; 2523 f = pf_anchor_stack + *depth - 1; 2524 if (f->parent != NULL && f->child != NULL) { 2525 if (f->child->match || 2526 (match != NULL && *match)) { 2527 f->r->anchor->match = 1; 2528 *match = 0; 2529 } 2530 f->child = RB_NEXT(pf_anchor_node, f->parent, f->child); 2531 if (f->child != NULL) { 2532 *rs = &f->child->ruleset; 2533 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2534 if (*r == NULL) 2535 continue; 2536 else 2537 break; 2538 } 2539 } 2540 (*depth)--; 2541 if (*depth == 0 && a != NULL) 2542 *a = NULL; 2543 *rs = f->rs; 2544 if (f->r->anchor->match || (match != NULL && *match)) 2545 quick = f->r->quick; 2546 *r = TAILQ_NEXT(f->r, entries); 2547 } while (*r == NULL); 2548 2549 return (quick); 2550 } 2551 2552 #ifdef INET6 2553 void 2554 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 2555 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 2556 { 2557 switch (af) { 2558 #ifdef INET 2559 case AF_INET: 2560 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 2561 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 2562 break; 2563 #endif /* INET */ 2564 case AF_INET6: 2565 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 2566 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 2567 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 2568 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 2569 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 2570 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 2571 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 2572 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 2573 break; 2574 } 2575 } 2576 2577 void 2578 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 2579 { 2580 switch (af) { 2581 #ifdef INET 2582 case AF_INET: 2583 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 2584 break; 2585 #endif /* INET */ 2586 case AF_INET6: 2587 if (addr->addr32[3] == 0xffffffff) { 2588 addr->addr32[3] = 0; 2589 if (addr->addr32[2] == 0xffffffff) { 2590 addr->addr32[2] = 0; 2591 if (addr->addr32[1] == 0xffffffff) { 2592 addr->addr32[1] = 0; 2593 addr->addr32[0] = 2594 htonl(ntohl(addr->addr32[0]) + 1); 2595 } else 2596 addr->addr32[1] = 2597 htonl(ntohl(addr->addr32[1]) + 1); 2598 } else 2599 addr->addr32[2] = 2600 htonl(ntohl(addr->addr32[2]) + 1); 2601 } else 2602 addr->addr32[3] = 2603 htonl(ntohl(addr->addr32[3]) + 1); 2604 break; 2605 } 2606 } 2607 #endif /* INET6 */ 2608 2609 #define mix(a,b,c) \ 2610 do { \ 2611 a -= b; a -= c; a ^= (c >> 13); \ 2612 b -= c; b -= a; b ^= (a << 8); \ 2613 c -= a; c -= b; c ^= (b >> 13); \ 2614 a -= b; a -= c; a ^= (c >> 12); \ 2615 b -= c; b -= a; b ^= (a << 16); \ 2616 c -= a; c -= b; c ^= (b >> 5); \ 2617 a -= b; a -= c; a ^= (c >> 3); \ 2618 b -= c; b -= a; b ^= (a << 10); \ 2619 c -= a; c -= b; c ^= (b >> 15); \ 2620 } while (0) 2621 2622 /* 2623 * hash function based on bridge_hash in if_bridge.c 2624 */ 2625 void 2626 pf_hash(struct pf_addr *inaddr, struct pf_addr *hash, 2627 struct pf_poolhashkey *key, sa_family_t af) 2628 { 2629 u_int32_t a = 0x9e3779b9, b = 0x9e3779b9, c = key->key32[0]; 2630 2631 switch (af) { 2632 #ifdef INET 2633 case AF_INET: 2634 a += inaddr->addr32[0]; 2635 b += key->key32[1]; 2636 mix(a, b, c); 2637 hash->addr32[0] = c + key->key32[2]; 2638 break; 2639 #endif /* INET */ 2640 #ifdef INET6 2641 case AF_INET6: 2642 a += inaddr->addr32[0]; 2643 b += inaddr->addr32[2]; 2644 mix(a, b, c); 2645 hash->addr32[0] = c; 2646 a += inaddr->addr32[1]; 2647 b += inaddr->addr32[3]; 2648 c += key->key32[1]; 2649 mix(a, b, c); 2650 hash->addr32[1] = c; 2651 a += inaddr->addr32[2]; 2652 b += inaddr->addr32[1]; 2653 c += key->key32[2]; 2654 mix(a, b, c); 2655 hash->addr32[2] = c; 2656 a += inaddr->addr32[3]; 2657 b += inaddr->addr32[0]; 2658 c += key->key32[3]; 2659 mix(a, b, c); 2660 hash->addr32[3] = c; 2661 break; 2662 #endif /* INET6 */ 2663 } 2664 } 2665 2666 int 2667 pf_map_addr(sa_family_t af, struct pf_rule *r, struct pf_addr *saddr, 2668 struct pf_addr *naddr, struct pf_addr *init_addr, struct pf_src_node **sn) 2669 { 2670 unsigned char hash[16]; 2671 struct pf_pool *rpool = &r->rpool; 2672 struct pf_pooladdr *acur = rpool->cur; 2673 struct pf_pooladdr *cur; 2674 struct pf_addr *raddr; 2675 struct pf_addr *rmask; 2676 struct pf_addr counter; 2677 struct pf_src_node k; 2678 int cpu = mycpu->gd_cpuid; 2679 int tblidx; 2680 2681 bzero(hash, sizeof(hash)); /* avoid gcc warnings */ 2682 2683 /* 2684 * NOTE! rpool->cur and rpool->tblidx can be iterators and thus 2685 * may represent a SMP race due to the shared nature of the 2686 * rpool structure. We allow the race and ensure that updates 2687 * do not create a fatal condition. 2688 */ 2689 cpu_ccfence(); 2690 cur = acur; 2691 raddr = &cur->addr.v.a.addr; 2692 rmask = &cur->addr.v.a.mask; 2693 2694 if (*sn == NULL && r->rpool.opts & PF_POOL_STICKYADDR && 2695 (r->rpool.opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) { 2696 k.af = af; 2697 PF_ACPY(&k.addr, saddr, af); 2698 if (r->rule_flag & PFRULE_RULESRCTRACK || 2699 r->rpool.opts & PF_POOL_STICKYADDR) 2700 k.rule.ptr = r; 2701 else 2702 k.rule.ptr = NULL; 2703 pf_status.scounters[SCNT_SRC_NODE_SEARCH]++; 2704 *sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k); 2705 if (*sn != NULL && !PF_AZERO(&(*sn)->raddr, af)) { 2706 PF_ACPY(naddr, &(*sn)->raddr, af); 2707 if (pf_status.debug >= PF_DEBUG_MISC) { 2708 kprintf("pf_map_addr: src tracking maps "); 2709 pf_print_host(&k.addr, 0, af); 2710 kprintf(" to "); 2711 pf_print_host(naddr, 0, af); 2712 kprintf("\n"); 2713 } 2714 return (0); 2715 } 2716 } 2717 2718 if (cur->addr.type == PF_ADDR_NOROUTE) 2719 return (1); 2720 if (cur->addr.type == PF_ADDR_DYNIFTL) { 2721 switch (af) { 2722 #ifdef INET 2723 case AF_INET: 2724 if (cur->addr.p.dyn->pfid_acnt4 < 1 && 2725 (rpool->opts & PF_POOL_TYPEMASK) != 2726 PF_POOL_ROUNDROBIN) 2727 return (1); 2728 raddr = &cur->addr.p.dyn->pfid_addr4; 2729 rmask = &cur->addr.p.dyn->pfid_mask4; 2730 break; 2731 #endif /* INET */ 2732 #ifdef INET6 2733 case AF_INET6: 2734 if (cur->addr.p.dyn->pfid_acnt6 < 1 && 2735 (rpool->opts & PF_POOL_TYPEMASK) != 2736 PF_POOL_ROUNDROBIN) 2737 return (1); 2738 raddr = &cur->addr.p.dyn->pfid_addr6; 2739 rmask = &cur->addr.p.dyn->pfid_mask6; 2740 break; 2741 #endif /* INET6 */ 2742 } 2743 } else if (cur->addr.type == PF_ADDR_TABLE) { 2744 if ((rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_ROUNDROBIN) 2745 return (1); /* unsupported */ 2746 } else { 2747 raddr = &cur->addr.v.a.addr; 2748 rmask = &cur->addr.v.a.mask; 2749 } 2750 2751 switch (rpool->opts & PF_POOL_TYPEMASK) { 2752 case PF_POOL_NONE: 2753 PF_ACPY(naddr, raddr, af); 2754 break; 2755 case PF_POOL_BITMASK: 2756 PF_POOLMASK(naddr, raddr, rmask, saddr, af); 2757 break; 2758 case PF_POOL_RANDOM: 2759 if (init_addr != NULL && PF_AZERO(init_addr, af)) { 2760 switch (af) { 2761 #ifdef INET 2762 case AF_INET: 2763 counter.addr32[0] = htonl(karc4random()); 2764 break; 2765 #endif /* INET */ 2766 #ifdef INET6 2767 case AF_INET6: 2768 if (rmask->addr32[3] != 0xffffffff) 2769 counter.addr32[3] = 2770 htonl(karc4random()); 2771 else 2772 break; 2773 if (rmask->addr32[2] != 0xffffffff) 2774 counter.addr32[2] = 2775 htonl(karc4random()); 2776 else 2777 break; 2778 if (rmask->addr32[1] != 0xffffffff) 2779 counter.addr32[1] = 2780 htonl(karc4random()); 2781 else 2782 break; 2783 if (rmask->addr32[0] != 0xffffffff) 2784 counter.addr32[0] = 2785 htonl(karc4random()); 2786 break; 2787 #endif /* INET6 */ 2788 } 2789 PF_POOLMASK(naddr, raddr, rmask, &counter, af); 2790 PF_ACPY(init_addr, naddr, af); 2791 2792 } else { 2793 counter = rpool->counter; 2794 cpu_ccfence(); 2795 PF_AINC(&counter, af); 2796 PF_POOLMASK(naddr, raddr, rmask, &counter, af); 2797 rpool->counter = counter; 2798 } 2799 break; 2800 case PF_POOL_SRCHASH: 2801 pf_hash(saddr, (struct pf_addr *)&hash, &rpool->key, af); 2802 PF_POOLMASK(naddr, raddr, rmask, (struct pf_addr *)&hash, af); 2803 break; 2804 case PF_POOL_ROUNDROBIN: 2805 tblidx = rpool->tblidx; 2806 counter = rpool->counter; 2807 if (cur->addr.type == PF_ADDR_TABLE) { 2808 if (!pfr_pool_get(cur->addr.p.tbl, 2809 &tblidx, &counter, 2810 &raddr, &rmask, af)) { 2811 goto get_addr; 2812 } 2813 } else if (cur->addr.type == PF_ADDR_DYNIFTL) { 2814 if (!pfr_pool_get(cur->addr.p.dyn->pfid_kt, 2815 &tblidx, &counter, 2816 &raddr, &rmask, af)) { 2817 goto get_addr; 2818 } 2819 } else if (pf_match_addr(0, raddr, rmask, 2820 &counter, af)) { 2821 goto get_addr; 2822 } 2823 2824 try_next: 2825 if ((cur = TAILQ_NEXT(cur, entries)) == NULL) 2826 cur = TAILQ_FIRST(&rpool->list); 2827 if (cur->addr.type == PF_ADDR_TABLE) { 2828 tblidx = -1; 2829 if (pfr_pool_get(cur->addr.p.tbl, 2830 &tblidx, &counter, 2831 &raddr, &rmask, af)) { 2832 /* table contains no address of type 'af' */ 2833 if (cur != acur) 2834 goto try_next; 2835 return (1); 2836 } 2837 } else if (cur->addr.type == PF_ADDR_DYNIFTL) { 2838 tblidx = -1; 2839 if (pfr_pool_get(cur->addr.p.dyn->pfid_kt, 2840 &tblidx, &counter, 2841 &raddr, &rmask, af)) { 2842 /* table contains no address of type 'af' */ 2843 if (cur != acur) 2844 goto try_next; 2845 return (1); 2846 } 2847 } else { 2848 raddr = &cur->addr.v.a.addr; 2849 rmask = &cur->addr.v.a.mask; 2850 PF_ACPY(&counter, raddr, af); 2851 } 2852 2853 get_addr: 2854 rpool->cur = cur; 2855 rpool->tblidx = tblidx; 2856 PF_ACPY(naddr, &counter, af); 2857 if (init_addr != NULL && PF_AZERO(init_addr, af)) 2858 PF_ACPY(init_addr, naddr, af); 2859 PF_AINC(&counter, af); 2860 rpool->counter = counter; 2861 break; 2862 } 2863 if (*sn != NULL) 2864 PF_ACPY(&(*sn)->raddr, naddr, af); 2865 2866 if (pf_status.debug >= PF_DEBUG_MISC && 2867 (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) { 2868 kprintf("pf_map_addr: selected address "); 2869 pf_print_host(naddr, 0, af); 2870 kprintf("\n"); 2871 } 2872 2873 return (0); 2874 } 2875 2876 int 2877 pf_get_sport(struct pf_pdesc *pd, sa_family_t af, 2878 u_int8_t proto, struct pf_rule *r, 2879 struct pf_addr *saddr, struct pf_addr *daddr, 2880 u_int16_t sport, u_int16_t dport, 2881 struct pf_addr *naddr, u_int16_t *nport, 2882 u_int16_t low, u_int16_t high, struct pf_src_node **sn) 2883 { 2884 struct pf_state_key_cmp key; 2885 struct pf_addr init_addr; 2886 u_int16_t cut; 2887 u_int32_t toeplitz_sport; 2888 2889 bzero(&init_addr, sizeof(init_addr)); 2890 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn)) 2891 return (1); 2892 2893 if (proto == IPPROTO_ICMP) { 2894 low = 1; 2895 high = 65535; 2896 } 2897 2898 bzero(&key, sizeof(key)); 2899 key.af = af; 2900 key.proto = proto; 2901 key.port[0] = dport; 2902 PF_ACPY(&key.addr[0], daddr, key.af); 2903 2904 do { 2905 PF_ACPY(&key.addr[1], naddr, key.af); 2906 2907 /* 2908 * We want to select a port that calculates to a toeplitz hash 2909 * that masks to the same cpu, otherwise the response may 2910 * not see the new state. 2911 * 2912 * We can still do this even if the kernel is disregarding 2913 * the hash and vectoring the packets to a specific cpu, 2914 * but it will reduce the number of ports we can use. 2915 */ 2916 switch(af) { 2917 case AF_INET: 2918 toeplitz_sport = 2919 toeplitz_piecemeal_port(sport) ^ 2920 toeplitz_piecemeal_addr(saddr->v4.s_addr) ^ 2921 toeplitz_piecemeal_addr(naddr->v4.s_addr); 2922 break; 2923 case AF_INET6: 2924 /* XXX TODO XXX */ 2925 default: 2926 /* XXX TODO XXX */ 2927 toeplitz_sport = 0; 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 ((toeplitz_piecemeal_port(key.port[1]) ^ 2995 toeplitz_sport) & ncpus2_mask) { 2996 continue; 2997 } 2998 if (pf_find_state_all(&key, PF_IN, NULL) == 2999 NULL && !in_baddynamic(tmp, proto)) { 3000 if (proto == IPPROTO_UDP) 3001 pd->not_cpu_localized = 1; 3002 *nport = htons(tmp); 3003 return (0); 3004 } 3005 } 3006 for (tmp = cut - 1; tmp >= low; --(tmp)) { 3007 key.port[1] = htons(tmp); 3008 if ((toeplitz_piecemeal_port(key.port[1]) ^ 3009 toeplitz_sport) & ncpus2_mask) { 3010 continue; 3011 } 3012 if (pf_find_state_all(&key, PF_IN, NULL) == 3013 NULL && !in_baddynamic(tmp, proto)) { 3014 if (proto == IPPROTO_UDP) 3015 pd->not_cpu_localized = 1; 3016 *nport = htons(tmp); 3017 return (0); 3018 } 3019 } 3020 } 3021 3022 /* 3023 * Next address 3024 */ 3025 switch (r->rpool.opts & PF_POOL_TYPEMASK) { 3026 case PF_POOL_RANDOM: 3027 case PF_POOL_ROUNDROBIN: 3028 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn)) 3029 return (1); 3030 break; 3031 case PF_POOL_NONE: 3032 case PF_POOL_SRCHASH: 3033 case PF_POOL_BITMASK: 3034 default: 3035 return (1); 3036 } 3037 } while (! PF_AEQ(&init_addr, naddr, af) ); 3038 return (1); /* none available */ 3039 } 3040 3041 struct pf_rule * 3042 pf_match_translation(struct pf_pdesc *pd, struct mbuf *m, int off, 3043 int direction, struct pfi_kif *kif, struct pf_addr *saddr, u_int16_t sport, 3044 struct pf_addr *daddr, u_int16_t dport, int rs_num) 3045 { 3046 struct pf_rule *r, *rm = NULL; 3047 struct pf_ruleset *ruleset = NULL; 3048 int tag = -1; 3049 int rtableid = -1; 3050 int asd = 0; 3051 3052 r = TAILQ_FIRST(pf_main_ruleset.rules[rs_num].active.ptr); 3053 while (r && rm == NULL) { 3054 struct pf_rule_addr *src = NULL, *dst = NULL; 3055 struct pf_addr_wrap *xdst = NULL; 3056 struct pf_pooladdr *cur; 3057 3058 if (r->action == PF_BINAT && direction == PF_IN) { 3059 src = &r->dst; 3060 cur = r->rpool.cur; /* SMP race possible */ 3061 cpu_ccfence(); 3062 if (cur) 3063 xdst = &cur->addr; 3064 } else { 3065 src = &r->src; 3066 dst = &r->dst; 3067 } 3068 3069 r->evaluations++; 3070 if (pfi_kif_match(r->kif, kif) == r->ifnot) 3071 r = r->skip[PF_SKIP_IFP].ptr; 3072 else if (r->direction && r->direction != direction) 3073 r = r->skip[PF_SKIP_DIR].ptr; 3074 else if (r->af && r->af != pd->af) 3075 r = r->skip[PF_SKIP_AF].ptr; 3076 else if (r->proto && r->proto != pd->proto) 3077 r = r->skip[PF_SKIP_PROTO].ptr; 3078 else if (PF_MISMATCHAW(&src->addr, saddr, pd->af, 3079 src->neg, kif)) 3080 r = r->skip[src == &r->src ? PF_SKIP_SRC_ADDR : 3081 PF_SKIP_DST_ADDR].ptr; 3082 else if (src->port_op && !pf_match_port(src->port_op, 3083 src->port[0], src->port[1], sport)) 3084 r = r->skip[src == &r->src ? PF_SKIP_SRC_PORT : 3085 PF_SKIP_DST_PORT].ptr; 3086 else if (dst != NULL && 3087 PF_MISMATCHAW(&dst->addr, daddr, pd->af, dst->neg, NULL)) 3088 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3089 else if (xdst != NULL && PF_MISMATCHAW(xdst, daddr, pd->af, 3090 0, NULL)) 3091 r = TAILQ_NEXT(r, entries); 3092 else if (dst != NULL && dst->port_op && 3093 !pf_match_port(dst->port_op, dst->port[0], 3094 dst->port[1], dport)) 3095 r = r->skip[PF_SKIP_DST_PORT].ptr; 3096 else if (r->match_tag && !pf_match_tag(m, r, &tag)) 3097 r = TAILQ_NEXT(r, entries); 3098 else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto != 3099 IPPROTO_TCP || !pf_osfp_match(pf_osfp_fingerprint(pd, m, 3100 off, pd->hdr.tcp), r->os_fingerprint))) 3101 r = TAILQ_NEXT(r, entries); 3102 else { 3103 if (r->tag) 3104 tag = r->tag; 3105 if (r->rtableid >= 0) 3106 rtableid = r->rtableid; 3107 if (r->anchor == NULL) { 3108 rm = r; 3109 } else 3110 pf_step_into_anchor(&asd, &ruleset, rs_num, 3111 &r, NULL, NULL); 3112 } 3113 if (r == NULL) 3114 pf_step_out_of_anchor(&asd, &ruleset, rs_num, &r, 3115 NULL, NULL); 3116 } 3117 if (pf_tag_packet(m, tag, rtableid)) 3118 return (NULL); 3119 if (rm != NULL && (rm->action == PF_NONAT || 3120 rm->action == PF_NORDR || rm->action == PF_NOBINAT)) 3121 return (NULL); 3122 return (rm); 3123 } 3124 3125 struct pf_rule * 3126 pf_get_translation(struct pf_pdesc *pd, struct mbuf *m, int off, int direction, 3127 struct pfi_kif *kif, struct pf_src_node **sn, 3128 struct pf_state_key **skw, struct pf_state_key **sks, 3129 struct pf_state_key **skp, struct pf_state_key **nkp, 3130 struct pf_addr *saddr, struct pf_addr *daddr, 3131 u_int16_t sport, u_int16_t dport) 3132 { 3133 struct pf_rule *r = NULL; 3134 3135 if (direction == PF_OUT) { 3136 r = pf_match_translation(pd, m, off, direction, kif, saddr, 3137 sport, daddr, dport, PF_RULESET_BINAT); 3138 if (r == NULL) 3139 r = pf_match_translation(pd, m, off, direction, kif, 3140 saddr, sport, daddr, dport, PF_RULESET_NAT); 3141 } else { 3142 r = pf_match_translation(pd, m, off, direction, kif, saddr, 3143 sport, daddr, dport, PF_RULESET_RDR); 3144 if (r == NULL) 3145 r = pf_match_translation(pd, m, off, direction, kif, 3146 saddr, sport, daddr, dport, PF_RULESET_BINAT); 3147 } 3148 3149 if (r != NULL) { 3150 struct pf_addr *naddr; 3151 u_int16_t *nport; 3152 3153 if (pf_state_key_setup(pd, r, skw, sks, skp, nkp, 3154 saddr, daddr, sport, dport)) 3155 return r; 3156 3157 /* XXX We only modify one side for now. */ 3158 naddr = &(*nkp)->addr[1]; 3159 nport = &(*nkp)->port[1]; 3160 3161 /* 3162 * NOTE: Currently all translations will clear 3163 * BRIDGE_MBUF_TAGGED, telling the bridge to 3164 * ignore the original input encapsulation. 3165 */ 3166 switch (r->action) { 3167 case PF_NONAT: 3168 case PF_NOBINAT: 3169 case PF_NORDR: 3170 return (NULL); 3171 case PF_NAT: 3172 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 3173 if (pf_get_sport(pd, pd->af, pd->proto, r, 3174 saddr, daddr, sport, dport, 3175 naddr, nport, r->rpool.proxy_port[0], 3176 r->rpool.proxy_port[1], sn)) { 3177 DPFPRINTF(PF_DEBUG_MISC, 3178 ("pf: NAT proxy port allocation " 3179 "(%u-%u) failed\n", 3180 r->rpool.proxy_port[0], 3181 r->rpool.proxy_port[1])); 3182 return (NULL); 3183 } 3184 break; 3185 case PF_BINAT: 3186 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 3187 switch (direction) { 3188 case PF_OUT: 3189 if (r->rpool.cur->addr.type == PF_ADDR_DYNIFTL){ 3190 switch (pd->af) { 3191 #ifdef INET 3192 case AF_INET: 3193 if (r->rpool.cur->addr.p.dyn-> 3194 pfid_acnt4 < 1) 3195 return (NULL); 3196 PF_POOLMASK(naddr, 3197 &r->rpool.cur->addr.p.dyn-> 3198 pfid_addr4, 3199 &r->rpool.cur->addr.p.dyn-> 3200 pfid_mask4, 3201 saddr, AF_INET); 3202 break; 3203 #endif /* INET */ 3204 #ifdef INET6 3205 case AF_INET6: 3206 if (r->rpool.cur->addr.p.dyn-> 3207 pfid_acnt6 < 1) 3208 return (NULL); 3209 PF_POOLMASK(naddr, 3210 &r->rpool.cur->addr.p.dyn-> 3211 pfid_addr6, 3212 &r->rpool.cur->addr.p.dyn-> 3213 pfid_mask6, 3214 saddr, AF_INET6); 3215 break; 3216 #endif /* INET6 */ 3217 } 3218 } else 3219 PF_POOLMASK(naddr, 3220 &r->rpool.cur->addr.v.a.addr, 3221 &r->rpool.cur->addr.v.a.mask, 3222 saddr, pd->af); 3223 break; 3224 case PF_IN: 3225 if (r->src.addr.type == PF_ADDR_DYNIFTL) { 3226 switch (pd->af) { 3227 #ifdef INET 3228 case AF_INET: 3229 if (r->src.addr.p.dyn-> 3230 pfid_acnt4 < 1) 3231 return (NULL); 3232 PF_POOLMASK(naddr, 3233 &r->src.addr.p.dyn-> 3234 pfid_addr4, 3235 &r->src.addr.p.dyn-> 3236 pfid_mask4, 3237 daddr, AF_INET); 3238 break; 3239 #endif /* INET */ 3240 #ifdef INET6 3241 case AF_INET6: 3242 if (r->src.addr.p.dyn-> 3243 pfid_acnt6 < 1) 3244 return (NULL); 3245 PF_POOLMASK(naddr, 3246 &r->src.addr.p.dyn-> 3247 pfid_addr6, 3248 &r->src.addr.p.dyn-> 3249 pfid_mask6, 3250 daddr, AF_INET6); 3251 break; 3252 #endif /* INET6 */ 3253 } 3254 } else 3255 PF_POOLMASK(naddr, 3256 &r->src.addr.v.a.addr, 3257 &r->src.addr.v.a.mask, daddr, 3258 pd->af); 3259 break; 3260 } 3261 break; 3262 case PF_RDR: { 3263 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 3264 if (pf_map_addr(pd->af, r, saddr, naddr, NULL, sn)) 3265 return (NULL); 3266 if ((r->rpool.opts & PF_POOL_TYPEMASK) == 3267 PF_POOL_BITMASK) 3268 PF_POOLMASK(naddr, naddr, 3269 &r->rpool.cur->addr.v.a.mask, daddr, 3270 pd->af); 3271 3272 if (r->rpool.proxy_port[1]) { 3273 u_int32_t tmp_nport; 3274 3275 tmp_nport = ((ntohs(dport) - 3276 ntohs(r->dst.port[0])) % 3277 (r->rpool.proxy_port[1] - 3278 r->rpool.proxy_port[0] + 1)) + 3279 r->rpool.proxy_port[0]; 3280 3281 /* wrap around if necessary */ 3282 if (tmp_nport > 65535) 3283 tmp_nport -= 65535; 3284 *nport = htons((u_int16_t)tmp_nport); 3285 } else if (r->rpool.proxy_port[0]) { 3286 *nport = htons(r->rpool.proxy_port[0]); 3287 } 3288 pd->not_cpu_localized = 1; 3289 break; 3290 } 3291 default: 3292 return (NULL); 3293 } 3294 } 3295 3296 return (r); 3297 } 3298 3299 struct netmsg_hashlookup { 3300 struct netmsg_base base; 3301 struct inpcb **nm_pinp; 3302 struct inpcbinfo *nm_pcbinfo; 3303 struct pf_addr *nm_saddr; 3304 struct pf_addr *nm_daddr; 3305 uint16_t nm_sport; 3306 uint16_t nm_dport; 3307 sa_family_t nm_af; 3308 }; 3309 3310 #ifdef PF_SOCKET_LOOKUP_DOMSG 3311 static void 3312 in_pcblookup_hash_handler(netmsg_t msg) 3313 { 3314 struct netmsg_hashlookup *rmsg = (struct netmsg_hashlookup *)msg; 3315 3316 if (rmsg->nm_af == AF_INET) 3317 *rmsg->nm_pinp = in_pcblookup_hash(rmsg->nm_pcbinfo, 3318 rmsg->nm_saddr->v4, rmsg->nm_sport, rmsg->nm_daddr->v4, 3319 rmsg->nm_dport, INPLOOKUP_WILDCARD, NULL); 3320 #ifdef INET6 3321 else 3322 *rmsg->nm_pinp = in6_pcblookup_hash(rmsg->nm_pcbinfo, 3323 &rmsg->nm_saddr->v6, rmsg->nm_sport, &rmsg->nm_daddr->v6, 3324 rmsg->nm_dport, INPLOOKUP_WILDCARD, NULL); 3325 #endif /* INET6 */ 3326 lwkt_replymsg(&rmsg->base.lmsg, 0); 3327 } 3328 #endif /* PF_SOCKET_LOOKUP_DOMSG */ 3329 3330 int 3331 pf_socket_lookup(int direction, struct pf_pdesc *pd) 3332 { 3333 struct pf_addr *saddr, *daddr; 3334 u_int16_t sport, dport; 3335 struct inpcbinfo *pi; 3336 struct inpcb *inp; 3337 struct netmsg_hashlookup *msg = NULL; 3338 #ifdef PF_SOCKET_LOOKUP_DOMSG 3339 struct netmsg_hashlookup msg0; 3340 #endif 3341 int pi_cpu = 0; 3342 3343 if (pd == NULL) 3344 return (-1); 3345 pd->lookup.uid = UID_MAX; 3346 pd->lookup.gid = GID_MAX; 3347 pd->lookup.pid = NO_PID; 3348 if (direction == PF_IN) { 3349 saddr = pd->src; 3350 daddr = pd->dst; 3351 } else { 3352 saddr = pd->dst; 3353 daddr = pd->src; 3354 } 3355 switch (pd->proto) { 3356 case IPPROTO_TCP: 3357 if (pd->hdr.tcp == NULL) 3358 return (-1); 3359 sport = pd->hdr.tcp->th_sport; 3360 dport = pd->hdr.tcp->th_dport; 3361 3362 pi_cpu = tcp_addrcpu(saddr->v4.s_addr, sport, daddr->v4.s_addr, dport); 3363 pi = &tcbinfo[pi_cpu]; 3364 /* 3365 * Our netstack runs lockless on MP systems 3366 * (only for TCP connections at the moment). 3367 * 3368 * As we are not allowed to read another CPU's tcbinfo, 3369 * we have to ask that CPU via remote call to search the 3370 * table for us. 3371 * 3372 * Prepare a msg iff data belongs to another CPU. 3373 */ 3374 if (pi_cpu != mycpu->gd_cpuid) { 3375 #ifdef PF_SOCKET_LOOKUP_DOMSG 3376 /* 3377 * NOTE: 3378 * 3379 * Following lwkt_domsg() is dangerous and could 3380 * lockup the network system, e.g. 3381 * 3382 * On 2 CPU system: 3383 * netisr0 domsg to netisr1 (due to lookup) 3384 * netisr1 domsg to netisr0 (due to lookup) 3385 * 3386 * We simply return -1 here, since we are probably 3387 * called before NAT, so the TCP packet should 3388 * already be on the correct CPU. 3389 */ 3390 msg = &msg0; 3391 netmsg_init(&msg->base, NULL, &curthread->td_msgport, 3392 0, in_pcblookup_hash_handler); 3393 msg->nm_pinp = &inp; 3394 msg->nm_pcbinfo = pi; 3395 msg->nm_saddr = saddr; 3396 msg->nm_sport = sport; 3397 msg->nm_daddr = daddr; 3398 msg->nm_dport = dport; 3399 msg->nm_af = pd->af; 3400 #else /* !PF_SOCKET_LOOKUP_DOMSG */ 3401 kprintf("pf_socket_lookup: tcp packet not on the " 3402 "correct cpu %d, cur cpu %d\n", 3403 pi_cpu, mycpuid); 3404 print_backtrace(-1); 3405 return -1; 3406 #endif /* PF_SOCKET_LOOKUP_DOMSG */ 3407 } 3408 break; 3409 case IPPROTO_UDP: 3410 if (pd->hdr.udp == NULL) 3411 return (-1); 3412 sport = pd->hdr.udp->uh_sport; 3413 dport = pd->hdr.udp->uh_dport; 3414 pi = &udbinfo[mycpuid]; 3415 break; 3416 default: 3417 return (-1); 3418 } 3419 if (direction != PF_IN) { 3420 u_int16_t p; 3421 3422 p = sport; 3423 sport = dport; 3424 dport = p; 3425 } 3426 switch (pd->af) { 3427 #ifdef INET6 3428 case AF_INET6: 3429 /* 3430 * Query other CPU, second part 3431 * 3432 * msg only gets initialized when: 3433 * 1) packet is TCP 3434 * 2) the info belongs to another CPU 3435 * 3436 * Use some switch/case magic to avoid code duplication. 3437 */ 3438 if (msg == NULL) { 3439 inp = in6_pcblookup_hash(pi, &saddr->v6, sport, 3440 &daddr->v6, dport, INPLOOKUP_WILDCARD, NULL); 3441 3442 if (inp == NULL) 3443 return (-1); 3444 break; 3445 } 3446 /* FALLTHROUGH if SMP and on other CPU */ 3447 #endif /* INET6 */ 3448 case AF_INET: 3449 if (msg != NULL) { 3450 lwkt_domsg(netisr_cpuport(pi_cpu), 3451 &msg->base.lmsg, 0); 3452 } else 3453 { 3454 inp = in_pcblookup_hash(pi, saddr->v4, sport, daddr->v4, 3455 dport, INPLOOKUP_WILDCARD, NULL); 3456 } 3457 if (inp == NULL) 3458 return (-1); 3459 break; 3460 3461 default: 3462 return (-1); 3463 } 3464 pd->lookup.uid = inp->inp_socket->so_cred->cr_uid; 3465 pd->lookup.gid = inp->inp_socket->so_cred->cr_groups[0]; 3466 return (1); 3467 } 3468 3469 u_int8_t 3470 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3471 { 3472 int hlen; 3473 u_int8_t hdr[60]; 3474 u_int8_t *opt, optlen; 3475 u_int8_t wscale = 0; 3476 3477 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3478 if (hlen <= sizeof(struct tcphdr)) 3479 return (0); 3480 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3481 return (0); 3482 opt = hdr + sizeof(struct tcphdr); 3483 hlen -= sizeof(struct tcphdr); 3484 while (hlen >= 3) { 3485 switch (*opt) { 3486 case TCPOPT_EOL: 3487 case TCPOPT_NOP: 3488 ++opt; 3489 --hlen; 3490 break; 3491 case TCPOPT_WINDOW: 3492 wscale = opt[2]; 3493 if (wscale > TCP_MAX_WINSHIFT) 3494 wscale = TCP_MAX_WINSHIFT; 3495 wscale |= PF_WSCALE_FLAG; 3496 /* FALLTHROUGH */ 3497 default: 3498 optlen = opt[1]; 3499 if (optlen < 2) 3500 optlen = 2; 3501 hlen -= optlen; 3502 opt += optlen; 3503 break; 3504 } 3505 } 3506 return (wscale); 3507 } 3508 3509 u_int16_t 3510 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3511 { 3512 int hlen; 3513 u_int8_t hdr[60]; 3514 u_int8_t *opt, optlen; 3515 u_int16_t mss = tcp_mssdflt; 3516 3517 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3518 if (hlen <= sizeof(struct tcphdr)) 3519 return (0); 3520 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3521 return (0); 3522 opt = hdr + sizeof(struct tcphdr); 3523 hlen -= sizeof(struct tcphdr); 3524 while (hlen >= TCPOLEN_MAXSEG) { 3525 switch (*opt) { 3526 case TCPOPT_EOL: 3527 case TCPOPT_NOP: 3528 ++opt; 3529 --hlen; 3530 break; 3531 case TCPOPT_MAXSEG: 3532 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 3533 /* FALLTHROUGH */ 3534 default: 3535 optlen = opt[1]; 3536 if (optlen < 2) 3537 optlen = 2; 3538 hlen -= optlen; 3539 opt += optlen; 3540 break; 3541 } 3542 } 3543 return (mss); 3544 } 3545 3546 u_int16_t 3547 pf_calc_mss(struct pf_addr *addr, sa_family_t af, u_int16_t offer) 3548 { 3549 #ifdef INET 3550 struct sockaddr_in *dst; 3551 struct route ro; 3552 #endif /* INET */ 3553 #ifdef INET6 3554 struct sockaddr_in6 *dst6; 3555 struct route_in6 ro6; 3556 #endif /* INET6 */ 3557 struct rtentry *rt = NULL; 3558 int hlen = 0; 3559 u_int16_t mss = tcp_mssdflt; 3560 3561 switch (af) { 3562 #ifdef INET 3563 case AF_INET: 3564 hlen = sizeof(struct ip); 3565 bzero(&ro, sizeof(ro)); 3566 dst = (struct sockaddr_in *)&ro.ro_dst; 3567 dst->sin_family = AF_INET; 3568 dst->sin_len = sizeof(*dst); 3569 dst->sin_addr = addr->v4; 3570 rtalloc_ign(&ro, (RTF_CLONING | RTF_PRCLONING)); 3571 rt = ro.ro_rt; 3572 break; 3573 #endif /* INET */ 3574 #ifdef INET6 3575 case AF_INET6: 3576 hlen = sizeof(struct ip6_hdr); 3577 bzero(&ro6, sizeof(ro6)); 3578 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst; 3579 dst6->sin6_family = AF_INET6; 3580 dst6->sin6_len = sizeof(*dst6); 3581 dst6->sin6_addr = addr->v6; 3582 rtalloc_ign((struct route *)&ro6, (RTF_CLONING | RTF_PRCLONING)); 3583 rt = ro6.ro_rt; 3584 break; 3585 #endif /* INET6 */ 3586 } 3587 3588 if (rt && rt->rt_ifp) { 3589 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr); 3590 mss = max(tcp_mssdflt, mss); 3591 RTFREE(rt); 3592 } 3593 mss = min(mss, offer); 3594 mss = max(mss, 64); /* sanity - at least max opt space */ 3595 return (mss); 3596 } 3597 3598 void 3599 pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr) 3600 { 3601 struct pf_rule *r = s->rule.ptr; 3602 3603 s->rt_kif = NULL; 3604 if (!r->rt || r->rt == PF_FASTROUTE) 3605 return; 3606 switch (s->key[PF_SK_WIRE]->af) { 3607 #ifdef INET 3608 case AF_INET: 3609 pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL, 3610 &s->nat_src_node); 3611 s->rt_kif = r->rpool.cur->kif; 3612 break; 3613 #endif /* INET */ 3614 #ifdef INET6 3615 case AF_INET6: 3616 pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL, 3617 &s->nat_src_node); 3618 s->rt_kif = r->rpool.cur->kif; 3619 break; 3620 #endif /* INET6 */ 3621 } 3622 } 3623 3624 u_int32_t 3625 pf_tcp_iss(struct pf_pdesc *pd) 3626 { 3627 MD5_CTX ctx; 3628 u_int32_t digest[4]; 3629 3630 if (pf_tcp_secret_init == 0) { 3631 lwkt_gettoken(&pf_gtoken); 3632 if (pf_tcp_secret_init == 0) { 3633 karc4rand(pf_tcp_secret, sizeof(pf_tcp_secret)); 3634 MD5Init(&pf_tcp_secret_ctx); 3635 MD5Update(&pf_tcp_secret_ctx, pf_tcp_secret, 3636 sizeof(pf_tcp_secret)); 3637 pf_tcp_secret_init = 1; 3638 } 3639 lwkt_reltoken(&pf_gtoken); 3640 } 3641 ctx = pf_tcp_secret_ctx; 3642 3643 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short)); 3644 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short)); 3645 if (pd->af == AF_INET6) { 3646 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 3647 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 3648 } else { 3649 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 3650 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 3651 } 3652 MD5Final((u_char *)digest, &ctx); 3653 pf_tcp_iss_off += 4096; 3654 3655 return (digest[0] + pd->hdr.tcp->th_seq + pf_tcp_iss_off); 3656 } 3657 3658 int 3659 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction, 3660 struct pfi_kif *kif, struct mbuf *m, int off, void *h, 3661 struct pf_pdesc *pd, struct pf_rule **am, struct pf_ruleset **rsm, 3662 struct ifqueue *ifq, struct inpcb *inp) 3663 { 3664 struct pf_rule *nr = NULL; 3665 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 3666 sa_family_t af = pd->af; 3667 struct pf_rule *r, *a = NULL; 3668 struct pf_ruleset *ruleset = NULL; 3669 struct pf_src_node *nsn = NULL; 3670 struct tcphdr *th = pd->hdr.tcp; 3671 struct pf_state_key *skw = NULL, *sks = NULL; 3672 struct pf_state_key *sk = NULL, *nk = NULL; 3673 u_short reason; 3674 int rewrite = 0, hdrlen = 0; 3675 int tag = -1, rtableid = -1; 3676 int asd = 0; 3677 int match = 0; 3678 int state_icmp = 0; 3679 u_int16_t sport = 0, dport = 0; 3680 u_int16_t bproto_sum = 0, bip_sum = 0; 3681 u_int8_t icmptype = 0, icmpcode = 0; 3682 3683 3684 if (direction == PF_IN && pf_check_congestion(ifq)) { 3685 REASON_SET(&reason, PFRES_CONGEST); 3686 return (PF_DROP); 3687 } 3688 3689 if (inp != NULL) 3690 pd->lookup.done = pf_socket_lookup(direction, pd); 3691 else if (debug_pfugidhack) { 3692 DPFPRINTF(PF_DEBUG_MISC, ("pf: unlocked lookup\n")); 3693 pd->lookup.done = pf_socket_lookup(direction, pd); 3694 } 3695 3696 switch (pd->proto) { 3697 case IPPROTO_TCP: 3698 sport = th->th_sport; 3699 dport = th->th_dport; 3700 hdrlen = sizeof(*th); 3701 break; 3702 case IPPROTO_UDP: 3703 sport = pd->hdr.udp->uh_sport; 3704 dport = pd->hdr.udp->uh_dport; 3705 hdrlen = sizeof(*pd->hdr.udp); 3706 break; 3707 #ifdef INET 3708 case IPPROTO_ICMP: 3709 if (pd->af != AF_INET) 3710 break; 3711 sport = dport = pd->hdr.icmp->icmp_id; 3712 hdrlen = sizeof(*pd->hdr.icmp); 3713 icmptype = pd->hdr.icmp->icmp_type; 3714 icmpcode = pd->hdr.icmp->icmp_code; 3715 3716 if (icmptype == ICMP_UNREACH || 3717 icmptype == ICMP_SOURCEQUENCH || 3718 icmptype == ICMP_REDIRECT || 3719 icmptype == ICMP_TIMXCEED || 3720 icmptype == ICMP_PARAMPROB) 3721 state_icmp++; 3722 break; 3723 #endif /* INET */ 3724 #ifdef INET6 3725 case IPPROTO_ICMPV6: 3726 if (af != AF_INET6) 3727 break; 3728 sport = dport = pd->hdr.icmp6->icmp6_id; 3729 hdrlen = sizeof(*pd->hdr.icmp6); 3730 icmptype = pd->hdr.icmp6->icmp6_type; 3731 icmpcode = pd->hdr.icmp6->icmp6_code; 3732 3733 if (icmptype == ICMP6_DST_UNREACH || 3734 icmptype == ICMP6_PACKET_TOO_BIG || 3735 icmptype == ICMP6_TIME_EXCEEDED || 3736 icmptype == ICMP6_PARAM_PROB) 3737 state_icmp++; 3738 break; 3739 #endif /* INET6 */ 3740 default: 3741 sport = dport = hdrlen = 0; 3742 break; 3743 } 3744 3745 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3746 3747 /* check packet for BINAT/NAT/RDR */ 3748 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, 3749 &skw, &sks, &sk, &nk, saddr, daddr, sport, dport)) != NULL) { 3750 if (nk == NULL || sk == NULL) { 3751 REASON_SET(&reason, PFRES_MEMORY); 3752 goto cleanup; 3753 } 3754 3755 if (pd->ip_sum) 3756 bip_sum = *pd->ip_sum; 3757 3758 m->m_flags &= ~M_HASH; 3759 switch (pd->proto) { 3760 case IPPROTO_TCP: 3761 bproto_sum = th->th_sum; 3762 pd->proto_sum = &th->th_sum; 3763 3764 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3765 nk->port[pd->sidx] != sport) { 3766 pf_change_ap(saddr, &th->th_sport, pd->ip_sum, 3767 &th->th_sum, &nk->addr[pd->sidx], 3768 nk->port[pd->sidx], 0, af); 3769 pd->sport = &th->th_sport; 3770 sport = th->th_sport; 3771 } 3772 3773 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3774 nk->port[pd->didx] != dport) { 3775 pf_change_ap(daddr, &th->th_dport, pd->ip_sum, 3776 &th->th_sum, &nk->addr[pd->didx], 3777 nk->port[pd->didx], 0, af); 3778 dport = th->th_dport; 3779 pd->dport = &th->th_dport; 3780 } 3781 rewrite++; 3782 break; 3783 case IPPROTO_UDP: 3784 bproto_sum = pd->hdr.udp->uh_sum; 3785 pd->proto_sum = &pd->hdr.udp->uh_sum; 3786 3787 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3788 nk->port[pd->sidx] != sport) { 3789 pf_change_ap(saddr, &pd->hdr.udp->uh_sport, 3790 pd->ip_sum, &pd->hdr.udp->uh_sum, 3791 &nk->addr[pd->sidx], 3792 nk->port[pd->sidx], 1, af); 3793 sport = pd->hdr.udp->uh_sport; 3794 pd->sport = &pd->hdr.udp->uh_sport; 3795 } 3796 3797 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3798 nk->port[pd->didx] != dport) { 3799 pf_change_ap(daddr, &pd->hdr.udp->uh_dport, 3800 pd->ip_sum, &pd->hdr.udp->uh_sum, 3801 &nk->addr[pd->didx], 3802 nk->port[pd->didx], 1, af); 3803 dport = pd->hdr.udp->uh_dport; 3804 pd->dport = &pd->hdr.udp->uh_dport; 3805 } 3806 rewrite++; 3807 break; 3808 #ifdef INET 3809 case IPPROTO_ICMP: 3810 nk->port[0] = nk->port[1]; 3811 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 3812 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 3813 nk->addr[pd->sidx].v4.s_addr, 0); 3814 3815 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 3816 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 3817 nk->addr[pd->didx].v4.s_addr, 0); 3818 3819 if (nk->port[1] != pd->hdr.icmp->icmp_id) { 3820 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup( 3821 pd->hdr.icmp->icmp_cksum, sport, 3822 nk->port[1], 0); 3823 pd->hdr.icmp->icmp_id = nk->port[1]; 3824 pd->sport = &pd->hdr.icmp->icmp_id; 3825 } 3826 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 3827 break; 3828 #endif /* INET */ 3829 #ifdef INET6 3830 case IPPROTO_ICMPV6: 3831 nk->port[0] = nk->port[1]; 3832 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 3833 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum, 3834 &nk->addr[pd->sidx], 0); 3835 3836 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 3837 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum, 3838 &nk->addr[pd->didx], 0); 3839 rewrite++; 3840 break; 3841 #endif /* INET */ 3842 default: 3843 switch (af) { 3844 #ifdef INET 3845 case AF_INET: 3846 if (PF_ANEQ(saddr, 3847 &nk->addr[pd->sidx], AF_INET)) 3848 pf_change_a(&saddr->v4.s_addr, 3849 pd->ip_sum, 3850 nk->addr[pd->sidx].v4.s_addr, 0); 3851 3852 if (PF_ANEQ(daddr, 3853 &nk->addr[pd->didx], AF_INET)) 3854 pf_change_a(&daddr->v4.s_addr, 3855 pd->ip_sum, 3856 nk->addr[pd->didx].v4.s_addr, 0); 3857 break; 3858 #endif /* INET */ 3859 #ifdef INET6 3860 case AF_INET6: 3861 if (PF_ANEQ(saddr, 3862 &nk->addr[pd->sidx], AF_INET6)) 3863 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 3864 3865 if (PF_ANEQ(daddr, 3866 &nk->addr[pd->didx], AF_INET6)) 3867 PF_ACPY(saddr, &nk->addr[pd->didx], af); 3868 break; 3869 #endif /* INET */ 3870 } 3871 break; 3872 } 3873 if (nr->natpass) 3874 r = NULL; 3875 pd->nat_rule = nr; 3876 } 3877 3878 while (r != NULL) { 3879 r->evaluations++; 3880 if (pfi_kif_match(r->kif, kif) == r->ifnot) 3881 r = r->skip[PF_SKIP_IFP].ptr; 3882 else if (r->direction && r->direction != direction) 3883 r = r->skip[PF_SKIP_DIR].ptr; 3884 else if (r->af && r->af != af) 3885 r = r->skip[PF_SKIP_AF].ptr; 3886 else if (r->proto && r->proto != pd->proto) 3887 r = r->skip[PF_SKIP_PROTO].ptr; 3888 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 3889 r->src.neg, kif)) 3890 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3891 /* tcp/udp only. port_op always 0 in other cases */ 3892 else if (r->src.port_op && !pf_match_port(r->src.port_op, 3893 r->src.port[0], r->src.port[1], sport)) 3894 r = r->skip[PF_SKIP_SRC_PORT].ptr; 3895 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 3896 r->dst.neg, NULL)) 3897 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3898 /* tcp/udp only. port_op always 0 in other cases */ 3899 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 3900 r->dst.port[0], r->dst.port[1], dport)) 3901 r = r->skip[PF_SKIP_DST_PORT].ptr; 3902 /* icmp only. type always 0 in other cases */ 3903 else if (r->type && r->type != icmptype + 1) 3904 r = TAILQ_NEXT(r, entries); 3905 /* icmp only. type always 0 in other cases */ 3906 else if (r->code && r->code != icmpcode + 1) 3907 r = TAILQ_NEXT(r, entries); 3908 else if (r->tos && !(r->tos == pd->tos)) 3909 r = TAILQ_NEXT(r, entries); 3910 else if (r->rule_flag & PFRULE_FRAGMENT) 3911 r = TAILQ_NEXT(r, entries); 3912 else if (pd->proto == IPPROTO_TCP && 3913 (r->flagset & th->th_flags) != r->flags) 3914 r = TAILQ_NEXT(r, entries); 3915 /* tcp/udp only. uid.op always 0 in other cases */ 3916 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 3917 pf_socket_lookup(direction, pd), 1)) && 3918 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 3919 pd->lookup.uid)) 3920 r = TAILQ_NEXT(r, entries); 3921 /* tcp/udp only. gid.op always 0 in other cases */ 3922 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 3923 pf_socket_lookup(direction, pd), 1)) && 3924 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 3925 pd->lookup.gid)) 3926 r = TAILQ_NEXT(r, entries); 3927 else if (r->prob && 3928 r->prob <= karc4random()) 3929 r = TAILQ_NEXT(r, entries); 3930 else if (r->match_tag && !pf_match_tag(m, r, &tag)) 3931 r = TAILQ_NEXT(r, entries); 3932 else if (r->os_fingerprint != PF_OSFP_ANY && 3933 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 3934 pf_osfp_fingerprint(pd, m, off, th), 3935 r->os_fingerprint))) 3936 r = TAILQ_NEXT(r, entries); 3937 else { 3938 if (r->tag) 3939 tag = r->tag; 3940 if (r->rtableid >= 0) 3941 rtableid = r->rtableid; 3942 if (r->anchor == NULL) { 3943 match = 1; 3944 *rm = r; 3945 *am = a; 3946 *rsm = ruleset; 3947 if ((*rm)->quick) 3948 break; 3949 r = TAILQ_NEXT(r, entries); 3950 } else 3951 pf_step_into_anchor(&asd, &ruleset, 3952 PF_RULESET_FILTER, &r, &a, &match); 3953 } 3954 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset, 3955 PF_RULESET_FILTER, &r, &a, &match)) 3956 break; 3957 } 3958 r = *rm; 3959 a = *am; 3960 ruleset = *rsm; 3961 3962 REASON_SET(&reason, PFRES_MATCH); 3963 3964 if (r->log || (nr != NULL && nr->log)) { 3965 if (rewrite) 3966 m_copyback(m, off, hdrlen, pd->hdr.any); 3967 PFLOG_PACKET(kif, h, m, af, direction, reason, r->log ? r : nr, 3968 a, ruleset, pd); 3969 } 3970 3971 if ((r->action == PF_DROP) && 3972 ((r->rule_flag & PFRULE_RETURNRST) || 3973 (r->rule_flag & PFRULE_RETURNICMP) || 3974 (r->rule_flag & PFRULE_RETURN))) { 3975 /* undo NAT changes, if they have taken place */ 3976 if (nr != NULL) { 3977 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3978 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3979 if (pd->sport) 3980 *pd->sport = sk->port[pd->sidx]; 3981 if (pd->dport) 3982 *pd->dport = sk->port[pd->didx]; 3983 if (pd->proto_sum) 3984 *pd->proto_sum = bproto_sum; 3985 if (pd->ip_sum) 3986 *pd->ip_sum = bip_sum; 3987 m_copyback(m, off, hdrlen, pd->hdr.any); 3988 } 3989 if (pd->proto == IPPROTO_TCP && 3990 ((r->rule_flag & PFRULE_RETURNRST) || 3991 (r->rule_flag & PFRULE_RETURN)) && 3992 !(th->th_flags & TH_RST)) { 3993 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3994 int len = 0; 3995 struct ip *h4; 3996 #ifdef INET6 3997 struct ip6_hdr *h6; 3998 #endif 3999 switch (af) { 4000 case AF_INET: 4001 h4 = mtod(m, struct ip *); 4002 len = h4->ip_len - off; 4003 break; 4004 #ifdef INET6 4005 case AF_INET6: 4006 h6 = mtod(m, struct ip6_hdr *); 4007 len = h6->ip6_plen - (off - sizeof(*h6)); 4008 break; 4009 #endif 4010 } 4011 4012 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 4013 REASON_SET(&reason, PFRES_PROTCKSUM); 4014 else { 4015 if (th->th_flags & TH_SYN) 4016 ack++; 4017 if (th->th_flags & TH_FIN) 4018 ack++; 4019 pf_send_tcp(r, af, pd->dst, 4020 pd->src, th->th_dport, th->th_sport, 4021 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 4022 r->return_ttl, 1, 0, pd->eh, kif->pfik_ifp); 4023 } 4024 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 4025 r->return_icmp) 4026 pf_send_icmp(m, r->return_icmp >> 8, 4027 r->return_icmp & 255, af, r); 4028 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 4029 r->return_icmp6) 4030 pf_send_icmp(m, r->return_icmp6 >> 8, 4031 r->return_icmp6 & 255, af, r); 4032 } 4033 4034 if (r->action == PF_DROP) 4035 goto cleanup; 4036 4037 if (pf_tag_packet(m, tag, rtableid)) { 4038 REASON_SET(&reason, PFRES_MEMORY); 4039 goto cleanup; 4040 } 4041 4042 if (!state_icmp && (r->keep_state || nr != NULL || 4043 (pd->flags & PFDESC_TCP_NORM))) { 4044 int action; 4045 action = pf_create_state(r, nr, a, pd, nsn, skw, sks, nk, sk, m, 4046 off, sport, dport, &rewrite, kif, sm, tag, bproto_sum, 4047 bip_sum, hdrlen); 4048 if (action != PF_PASS) 4049 return (action); 4050 } 4051 4052 /* copy back packet headers if we performed NAT operations */ 4053 if (rewrite) 4054 m_copyback(m, off, hdrlen, pd->hdr.any); 4055 4056 return (PF_PASS); 4057 4058 cleanup: 4059 if (sk != NULL) 4060 kfree(sk, M_PFSTATEKEYPL); 4061 if (nk != NULL) 4062 kfree(nk, M_PFSTATEKEYPL); 4063 return (PF_DROP); 4064 } 4065 4066 static __inline int 4067 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a, 4068 struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *skw, 4069 struct pf_state_key *sks, struct pf_state_key *nk, struct pf_state_key *sk, 4070 struct mbuf *m, int off, u_int16_t sport, u_int16_t dport, int *rewrite, 4071 struct pfi_kif *kif, struct pf_state **sm, int tag, u_int16_t bproto_sum, 4072 u_int16_t bip_sum, int hdrlen) 4073 { 4074 struct pf_state *s = NULL; 4075 struct pf_src_node *sn = NULL; 4076 struct tcphdr *th = pd->hdr.tcp; 4077 u_int16_t mss = tcp_mssdflt; 4078 u_short reason; 4079 int cpu = mycpu->gd_cpuid; 4080 4081 /* check maximums */ 4082 if (r->max_states && (r->states_cur >= r->max_states)) { 4083 pf_status.lcounters[LCNT_STATES]++; 4084 REASON_SET(&reason, PFRES_MAXSTATES); 4085 return (PF_DROP); 4086 } 4087 /* src node for filter rule */ 4088 if ((r->rule_flag & PFRULE_SRCTRACK || 4089 r->rpool.opts & PF_POOL_STICKYADDR) && 4090 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { 4091 REASON_SET(&reason, PFRES_SRCLIMIT); 4092 goto csfailed; 4093 } 4094 /* src node for translation rule */ 4095 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 4096 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { 4097 REASON_SET(&reason, PFRES_SRCLIMIT); 4098 goto csfailed; 4099 } 4100 s = kmalloc(sizeof(struct pf_state), M_PFSTATEPL, M_NOWAIT|M_ZERO); 4101 if (s == NULL) { 4102 REASON_SET(&reason, PFRES_MEMORY); 4103 goto csfailed; 4104 } 4105 lockinit(&s->lk, "pfstlk", 0, 0); 4106 s->id = 0; /* XXX Do we really need that? not in OpenBSD */ 4107 s->creatorid = 0; 4108 s->rule.ptr = r; 4109 s->nat_rule.ptr = nr; 4110 s->anchor.ptr = a; 4111 s->state_flags = PFSTATE_CREATEINPROG; 4112 STATE_INC_COUNTERS(s); 4113 if (r->allow_opts) 4114 s->state_flags |= PFSTATE_ALLOWOPTS; 4115 if (r->rule_flag & PFRULE_STATESLOPPY) 4116 s->state_flags |= PFSTATE_SLOPPY; 4117 if (pd->not_cpu_localized) 4118 s->state_flags |= PFSTATE_STACK_GLOBAL; 4119 4120 s->log = r->log & PF_LOG_ALL; 4121 if (nr != NULL) 4122 s->log |= nr->log & PF_LOG_ALL; 4123 switch (pd->proto) { 4124 case IPPROTO_TCP: 4125 s->src.seqlo = ntohl(th->th_seq); 4126 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 4127 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4128 r->keep_state == PF_STATE_MODULATE) { 4129 /* Generate sequence number modulator */ 4130 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 4131 0) 4132 s->src.seqdiff = 1; 4133 pf_change_a(&th->th_seq, &th->th_sum, 4134 htonl(s->src.seqlo + s->src.seqdiff), 0); 4135 *rewrite = 1; 4136 } else 4137 s->src.seqdiff = 0; 4138 if (th->th_flags & TH_SYN) { 4139 s->src.seqhi++; 4140 s->src.wscale = pf_get_wscale(m, off, 4141 th->th_off, pd->af); 4142 } 4143 s->src.max_win = MAX(ntohs(th->th_win), 1); 4144 if (s->src.wscale & PF_WSCALE_MASK) { 4145 /* Remove scale factor from initial window */ 4146 int win = s->src.max_win; 4147 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 4148 s->src.max_win = (win - 1) >> 4149 (s->src.wscale & PF_WSCALE_MASK); 4150 } 4151 if (th->th_flags & TH_FIN) 4152 s->src.seqhi++; 4153 s->dst.seqhi = 1; 4154 s->dst.max_win = 1; 4155 s->src.state = TCPS_SYN_SENT; 4156 s->dst.state = TCPS_CLOSED; 4157 s->timeout = PFTM_TCP_FIRST_PACKET; 4158 break; 4159 case IPPROTO_UDP: 4160 s->src.state = PFUDPS_SINGLE; 4161 s->dst.state = PFUDPS_NO_TRAFFIC; 4162 s->timeout = PFTM_UDP_FIRST_PACKET; 4163 break; 4164 case IPPROTO_ICMP: 4165 #ifdef INET6 4166 case IPPROTO_ICMPV6: 4167 #endif 4168 s->timeout = PFTM_ICMP_FIRST_PACKET; 4169 break; 4170 default: 4171 s->src.state = PFOTHERS_SINGLE; 4172 s->dst.state = PFOTHERS_NO_TRAFFIC; 4173 s->timeout = PFTM_OTHER_FIRST_PACKET; 4174 } 4175 4176 s->creation = time_second; 4177 s->expire = time_second; 4178 4179 if (sn != NULL) { 4180 s->src_node = sn; 4181 s->src_node->states++; 4182 } 4183 if (nsn != NULL) { 4184 /* XXX We only modify one side for now. */ 4185 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 4186 s->nat_src_node = nsn; 4187 s->nat_src_node->states++; 4188 } 4189 if (pd->proto == IPPROTO_TCP) { 4190 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m, 4191 off, pd, th, &s->src, &s->dst)) { 4192 REASON_SET(&reason, PFRES_MEMORY); 4193 pf_src_tree_remove_state(s); 4194 STATE_DEC_COUNTERS(s); 4195 kfree(s, M_PFSTATEPL); 4196 return (PF_DROP); 4197 } 4198 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub && 4199 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 4200 &s->src, &s->dst, rewrite)) { 4201 /* This really shouldn't happen!!! */ 4202 DPFPRINTF(PF_DEBUG_URGENT, 4203 ("pf_normalize_tcp_stateful failed on first pkt")); 4204 pf_normalize_tcp_cleanup(s); 4205 pf_src_tree_remove_state(s); 4206 STATE_DEC_COUNTERS(s); 4207 kfree(s, M_PFSTATEPL); 4208 return (PF_DROP); 4209 } 4210 } 4211 s->direction = pd->dir; 4212 4213 if (sk == NULL && pf_state_key_setup(pd, nr, &skw, &sks, &sk, &nk, 4214 pd->src, pd->dst, sport, dport)) { 4215 REASON_SET(&reason, PFRES_MEMORY); 4216 goto csfailed; 4217 } 4218 4219 if (pf_state_insert(BOUND_IFACE(r, kif), skw, sks, s)) { 4220 if (pd->proto == IPPROTO_TCP) 4221 pf_normalize_tcp_cleanup(s); 4222 REASON_SET(&reason, PFRES_STATEINS); 4223 pf_src_tree_remove_state(s); 4224 STATE_DEC_COUNTERS(s); 4225 kfree(s, M_PFSTATEPL); 4226 return (PF_DROP); 4227 } else 4228 *sm = s; 4229 4230 pf_set_rt_ifp(s, pd->src); /* needs s->state_key set */ 4231 if (tag > 0) { 4232 pf_tag_ref(tag); 4233 s->tag = tag; 4234 } 4235 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 4236 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 4237 s->src.state = PF_TCPS_PROXY_SRC; 4238 /* undo NAT changes, if they have taken place */ 4239 if (nr != NULL) { 4240 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 4241 if (pd->dir == PF_OUT) 4242 skt = s->key[PF_SK_STACK]; 4243 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 4244 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 4245 if (pd->sport) 4246 *pd->sport = skt->port[pd->sidx]; 4247 if (pd->dport) 4248 *pd->dport = skt->port[pd->didx]; 4249 if (pd->proto_sum) 4250 *pd->proto_sum = bproto_sum; 4251 if (pd->ip_sum) 4252 *pd->ip_sum = bip_sum; 4253 m->m_flags &= ~M_HASH; 4254 m_copyback(m, off, hdrlen, pd->hdr.any); 4255 } 4256 s->src.seqhi = htonl(karc4random()); 4257 /* Find mss option */ 4258 mss = pf_get_mss(m, off, th->th_off, pd->af); 4259 mss = pf_calc_mss(pd->src, pd->af, mss); 4260 mss = pf_calc_mss(pd->dst, pd->af, mss); 4261 s->src.mss = mss; 4262 s->state_flags &= ~PFSTATE_CREATEINPROG; 4263 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 4264 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 4265 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL, NULL); 4266 REASON_SET(&reason, PFRES_SYNPROXY); 4267 return (PF_SYNPROXY_DROP); 4268 } 4269 4270 s->state_flags &= ~PFSTATE_CREATEINPROG; 4271 return (PF_PASS); 4272 4273 csfailed: 4274 if (sk != NULL) 4275 kfree(sk, M_PFSTATEKEYPL); 4276 if (nk != NULL) 4277 kfree(nk, M_PFSTATEKEYPL); 4278 4279 if (sn != NULL && sn->states == 0 && sn->expire == 0) { 4280 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], sn); 4281 pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++; 4282 atomic_add_int(&pf_status.src_nodes, -1); 4283 kfree(sn, M_PFSRCTREEPL); 4284 } 4285 if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0) { 4286 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], nsn); 4287 pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++; 4288 atomic_add_int(&pf_status.src_nodes, -1); 4289 kfree(nsn, M_PFSRCTREEPL); 4290 } 4291 if (s) { 4292 pf_src_tree_remove_state(s); 4293 STATE_DEC_COUNTERS(s); 4294 kfree(s, M_PFSTATEPL); 4295 } 4296 4297 return (PF_DROP); 4298 } 4299 4300 int 4301 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif, 4302 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am, 4303 struct pf_ruleset **rsm) 4304 { 4305 struct pf_rule *r, *a = NULL; 4306 struct pf_ruleset *ruleset = NULL; 4307 sa_family_t af = pd->af; 4308 u_short reason; 4309 int tag = -1; 4310 int asd = 0; 4311 int match = 0; 4312 4313 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4314 while (r != NULL) { 4315 r->evaluations++; 4316 if (pfi_kif_match(r->kif, kif) == r->ifnot) 4317 r = r->skip[PF_SKIP_IFP].ptr; 4318 else if (r->direction && r->direction != direction) 4319 r = r->skip[PF_SKIP_DIR].ptr; 4320 else if (r->af && r->af != af) 4321 r = r->skip[PF_SKIP_AF].ptr; 4322 else if (r->proto && r->proto != pd->proto) 4323 r = r->skip[PF_SKIP_PROTO].ptr; 4324 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 4325 r->src.neg, kif)) 4326 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4327 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 4328 r->dst.neg, NULL)) 4329 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4330 else if (r->tos && !(r->tos == pd->tos)) 4331 r = TAILQ_NEXT(r, entries); 4332 else if (r->os_fingerprint != PF_OSFP_ANY) 4333 r = TAILQ_NEXT(r, entries); 4334 else if (pd->proto == IPPROTO_UDP && 4335 (r->src.port_op || r->dst.port_op)) 4336 r = TAILQ_NEXT(r, entries); 4337 else if (pd->proto == IPPROTO_TCP && 4338 (r->src.port_op || r->dst.port_op || r->flagset)) 4339 r = TAILQ_NEXT(r, entries); 4340 else if ((pd->proto == IPPROTO_ICMP || 4341 pd->proto == IPPROTO_ICMPV6) && 4342 (r->type || r->code)) 4343 r = TAILQ_NEXT(r, entries); 4344 else if (r->prob && r->prob <= karc4random()) 4345 r = TAILQ_NEXT(r, entries); 4346 else if (r->match_tag && !pf_match_tag(m, r, &tag)) 4347 r = TAILQ_NEXT(r, entries); 4348 else { 4349 if (r->anchor == NULL) { 4350 match = 1; 4351 *rm = r; 4352 *am = a; 4353 *rsm = ruleset; 4354 if ((*rm)->quick) 4355 break; 4356 r = TAILQ_NEXT(r, entries); 4357 } else 4358 pf_step_into_anchor(&asd, &ruleset, 4359 PF_RULESET_FILTER, &r, &a, &match); 4360 } 4361 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset, 4362 PF_RULESET_FILTER, &r, &a, &match)) 4363 break; 4364 } 4365 r = *rm; 4366 a = *am; 4367 ruleset = *rsm; 4368 4369 REASON_SET(&reason, PFRES_MATCH); 4370 4371 if (r->log) 4372 PFLOG_PACKET(kif, h, m, af, direction, reason, r, a, ruleset, 4373 pd); 4374 4375 if (r->action != PF_PASS) 4376 return (PF_DROP); 4377 4378 if (pf_tag_packet(m, tag, -1)) { 4379 REASON_SET(&reason, PFRES_MEMORY); 4380 return (PF_DROP); 4381 } 4382 4383 return (PF_PASS); 4384 } 4385 4386 /* 4387 * Called with state locked 4388 */ 4389 int 4390 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst, 4391 struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off, 4392 struct pf_pdesc *pd, u_short *reason, int *copyback) 4393 { 4394 struct tcphdr *th = pd->hdr.tcp; 4395 u_int16_t win = ntohs(th->th_win); 4396 u_int32_t ack, end, seq, orig_seq; 4397 u_int8_t sws, dws; 4398 int ackskew; 4399 4400 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 4401 sws = src->wscale & PF_WSCALE_MASK; 4402 dws = dst->wscale & PF_WSCALE_MASK; 4403 } else { 4404 sws = dws = 0; 4405 } 4406 4407 /* 4408 * Sequence tracking algorithm from Guido van Rooij's paper: 4409 * http://www.madison-gurkha.com/publications/tcp_filtering/ 4410 * tcp_filtering.ps 4411 */ 4412 4413 orig_seq = seq = ntohl(th->th_seq); 4414 if (src->seqlo == 0) { 4415 /* First packet from this end. Set its state */ 4416 4417 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) && 4418 src->scrub == NULL) { 4419 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 4420 REASON_SET(reason, PFRES_MEMORY); 4421 return (PF_DROP); 4422 } 4423 } 4424 4425 /* Deferred generation of sequence number modulator */ 4426 if (dst->seqdiff && !src->seqdiff) { 4427 /* use random iss for the TCP server */ 4428 while ((src->seqdiff = karc4random() - seq) == 0) 4429 ; 4430 ack = ntohl(th->th_ack) - dst->seqdiff; 4431 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq + 4432 src->seqdiff), 0); 4433 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0); 4434 *copyback = 1; 4435 } else { 4436 ack = ntohl(th->th_ack); 4437 } 4438 4439 end = seq + pd->p_len; 4440 if (th->th_flags & TH_SYN) { 4441 end++; 4442 (*state)->sync_flags |= PFSTATE_GOT_SYN2; 4443 if (dst->wscale & PF_WSCALE_FLAG) { 4444 src->wscale = pf_get_wscale(m, off, th->th_off, 4445 pd->af); 4446 if (src->wscale & PF_WSCALE_FLAG) { 4447 /* Remove scale factor from initial 4448 * window */ 4449 sws = src->wscale & PF_WSCALE_MASK; 4450 win = ((u_int32_t)win + (1 << sws) - 1) 4451 >> sws; 4452 dws = dst->wscale & PF_WSCALE_MASK; 4453 } else { 4454 /* fixup other window */ 4455 dst->max_win <<= dst->wscale & 4456 PF_WSCALE_MASK; 4457 /* in case of a retrans SYN|ACK */ 4458 dst->wscale = 0; 4459 } 4460 } 4461 } 4462 if (th->th_flags & TH_FIN) 4463 end++; 4464 4465 src->seqlo = seq; 4466 if (src->state < TCPS_SYN_SENT) 4467 src->state = TCPS_SYN_SENT; 4468 4469 /* 4470 * May need to slide the window (seqhi may have been set by 4471 * the crappy stack check or if we picked up the connection 4472 * after establishment) 4473 */ 4474 if (src->seqhi == 1 || 4475 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 4476 src->seqhi = end + MAX(1, dst->max_win << dws); 4477 if (win > src->max_win) 4478 src->max_win = win; 4479 4480 } else { 4481 ack = ntohl(th->th_ack) - dst->seqdiff; 4482 if (src->seqdiff) { 4483 /* Modulate sequence numbers */ 4484 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq + 4485 src->seqdiff), 0); 4486 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0); 4487 *copyback = 1; 4488 } 4489 end = seq + pd->p_len; 4490 if (th->th_flags & TH_SYN) 4491 end++; 4492 if (th->th_flags & TH_FIN) 4493 end++; 4494 } 4495 4496 if ((th->th_flags & TH_ACK) == 0) { 4497 /* Let it pass through the ack skew check */ 4498 ack = dst->seqlo; 4499 } else if ((ack == 0 && 4500 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 4501 /* broken tcp stacks do not set ack */ 4502 (dst->state < TCPS_SYN_SENT)) { 4503 /* 4504 * Many stacks (ours included) will set the ACK number in an 4505 * FIN|ACK if the SYN times out -- no sequence to ACK. 4506 */ 4507 ack = dst->seqlo; 4508 } 4509 4510 if (seq == end) { 4511 /* Ease sequencing restrictions on no data packets */ 4512 seq = src->seqlo; 4513 end = seq; 4514 } 4515 4516 ackskew = dst->seqlo - ack; 4517 4518 4519 /* 4520 * Need to demodulate the sequence numbers in any TCP SACK options 4521 * (Selective ACK). We could optionally validate the SACK values 4522 * against the current ACK window, either forwards or backwards, but 4523 * I'm not confident that SACK has been implemented properly 4524 * everywhere. It wouldn't surprise me if several stacks accidently 4525 * SACK too far backwards of previously ACKed data. There really aren't 4526 * any security implications of bad SACKing unless the target stack 4527 * doesn't validate the option length correctly. Someone trying to 4528 * spoof into a TCP connection won't bother blindly sending SACK 4529 * options anyway. 4530 */ 4531 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 4532 if (pf_modulate_sack(m, off, pd, th, dst)) 4533 *copyback = 1; 4534 } 4535 4536 4537 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 4538 if (SEQ_GEQ(src->seqhi, end) && 4539 /* Last octet inside other's window space */ 4540 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 4541 /* Retrans: not more than one window back */ 4542 (ackskew >= -MAXACKWINDOW) && 4543 /* Acking not more than one reassembled fragment backwards */ 4544 (ackskew <= (MAXACKWINDOW << sws)) && 4545 /* Acking not more than one window forward */ 4546 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 4547 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 4548 (pd->flags & PFDESC_IP_REAS) == 0)) { 4549 /* Require an exact/+1 sequence match on resets when possible */ 4550 4551 if (dst->scrub || src->scrub) { 4552 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4553 *state, src, dst, copyback)) 4554 return (PF_DROP); 4555 } 4556 4557 /* update max window */ 4558 if (src->max_win < win) 4559 src->max_win = win; 4560 /* synchronize sequencing */ 4561 if (SEQ_GT(end, src->seqlo)) 4562 src->seqlo = end; 4563 /* slide the window of what the other end can send */ 4564 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4565 dst->seqhi = ack + MAX((win << sws), 1); 4566 4567 4568 /* update states */ 4569 if (th->th_flags & TH_SYN) 4570 if (src->state < TCPS_SYN_SENT) 4571 src->state = TCPS_SYN_SENT; 4572 if (th->th_flags & TH_FIN) 4573 if (src->state < TCPS_CLOSING) 4574 src->state = TCPS_CLOSING; 4575 if (th->th_flags & TH_ACK) { 4576 if (dst->state == TCPS_SYN_SENT) { 4577 dst->state = TCPS_ESTABLISHED; 4578 if (src->state == TCPS_ESTABLISHED && 4579 (*state)->src_node != NULL && 4580 pf_src_connlimit(*state)) { 4581 REASON_SET(reason, PFRES_SRCLIMIT); 4582 return (PF_DROP); 4583 } 4584 } else if (dst->state == TCPS_CLOSING) 4585 dst->state = TCPS_FIN_WAIT_2; 4586 } 4587 if (th->th_flags & TH_RST) 4588 src->state = dst->state = TCPS_TIME_WAIT; 4589 4590 /* update expire time */ 4591 (*state)->expire = time_second; 4592 if (src->state >= TCPS_FIN_WAIT_2 && 4593 dst->state >= TCPS_FIN_WAIT_2) 4594 (*state)->timeout = PFTM_TCP_CLOSED; 4595 else if (src->state >= TCPS_CLOSING && 4596 dst->state >= TCPS_CLOSING) 4597 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4598 else if (src->state < TCPS_ESTABLISHED || 4599 dst->state < TCPS_ESTABLISHED) 4600 (*state)->timeout = PFTM_TCP_OPENING; 4601 else if (src->state >= TCPS_CLOSING || 4602 dst->state >= TCPS_CLOSING) 4603 (*state)->timeout = PFTM_TCP_CLOSING; 4604 else 4605 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4606 4607 /* Fall through to PASS packet */ 4608 4609 } else if ((dst->state < TCPS_SYN_SENT || 4610 dst->state >= TCPS_FIN_WAIT_2 || 4611 src->state >= TCPS_FIN_WAIT_2) && 4612 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 4613 /* Within a window forward of the originating packet */ 4614 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 4615 /* Within a window backward of the originating packet */ 4616 4617 /* 4618 * This currently handles three situations: 4619 * 1) Stupid stacks will shotgun SYNs before their peer 4620 * replies. 4621 * 2) When PF catches an already established stream (the 4622 * firewall rebooted, the state table was flushed, routes 4623 * changed...) 4624 * 3) Packets get funky immediately after the connection 4625 * closes (this should catch Solaris spurious ACK|FINs 4626 * that web servers like to spew after a close) 4627 * 4628 * This must be a little more careful than the above code 4629 * since packet floods will also be caught here. We don't 4630 * update the TTL here to mitigate the damage of a packet 4631 * flood and so the same code can handle awkward establishment 4632 * and a loosened connection close. 4633 * In the establishment case, a correct peer response will 4634 * validate the connection, go through the normal state code 4635 * and keep updating the state TTL. 4636 */ 4637 4638 if (pf_status.debug >= PF_DEBUG_MISC) { 4639 kprintf("pf: loose state match: "); 4640 pf_print_state(*state); 4641 pf_print_flags(th->th_flags); 4642 kprintf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4643 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, pd->p_len, 4644 ackskew, (unsigned long long)(*state)->packets[0], 4645 (unsigned long long)(*state)->packets[1], 4646 pd->dir == PF_IN ? "in" : "out", 4647 pd->dir == (*state)->direction ? "fwd" : "rev"); 4648 } 4649 4650 if (dst->scrub || src->scrub) { 4651 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4652 *state, src, dst, copyback)) 4653 return (PF_DROP); 4654 } 4655 4656 /* update max window */ 4657 if (src->max_win < win) 4658 src->max_win = win; 4659 /* synchronize sequencing */ 4660 if (SEQ_GT(end, src->seqlo)) 4661 src->seqlo = end; 4662 /* slide the window of what the other end can send */ 4663 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4664 dst->seqhi = ack + MAX((win << sws), 1); 4665 4666 /* 4667 * Cannot set dst->seqhi here since this could be a shotgunned 4668 * SYN and not an already established connection. 4669 */ 4670 4671 if (th->th_flags & TH_FIN) 4672 if (src->state < TCPS_CLOSING) 4673 src->state = TCPS_CLOSING; 4674 if (th->th_flags & TH_RST) 4675 src->state = dst->state = TCPS_TIME_WAIT; 4676 4677 /* Fall through to PASS packet */ 4678 4679 } else if ((*state)->pickup_mode == PF_PICKUPS_HASHONLY || 4680 ((*state)->pickup_mode == PF_PICKUPS_ENABLED && 4681 ((*state)->sync_flags & PFSTATE_GOT_SYN_MASK) != 4682 PFSTATE_GOT_SYN_MASK)) { 4683 /* 4684 * If pickup mode is hash only, do not fail on sequence checks. 4685 * 4686 * If pickup mode is enabled and we did not see the SYN in 4687 * both direction, do not fail on sequence checks because 4688 * we do not have complete information on window scale. 4689 * 4690 * Adjust expiration and fall through to PASS packet. 4691 * XXX Add a FIN check to reduce timeout? 4692 */ 4693 (*state)->expire = time_second; 4694 } else { 4695 /* 4696 * Failure processing 4697 */ 4698 if ((*state)->dst.state == TCPS_SYN_SENT && 4699 (*state)->src.state == TCPS_SYN_SENT) { 4700 /* Send RST for state mismatches during handshake */ 4701 if (!(th->th_flags & TH_RST)) 4702 pf_send_tcp((*state)->rule.ptr, pd->af, 4703 pd->dst, pd->src, th->th_dport, 4704 th->th_sport, ntohl(th->th_ack), 0, 4705 TH_RST, 0, 0, 4706 (*state)->rule.ptr->return_ttl, 1, 0, 4707 pd->eh, kif->pfik_ifp); 4708 src->seqlo = 0; 4709 src->seqhi = 1; 4710 src->max_win = 1; 4711 } else if (pf_status.debug >= PF_DEBUG_MISC) { 4712 kprintf("pf: BAD state: "); 4713 pf_print_state(*state); 4714 pf_print_flags(th->th_flags); 4715 kprintf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4716 "pkts=%llu:%llu dir=%s,%s\n", 4717 seq, orig_seq, ack, pd->p_len, ackskew, 4718 (unsigned long long)(*state)->packets[0], 4719 (unsigned long long)(*state)->packets[1], 4720 pd->dir == PF_IN ? "in" : "out", 4721 pd->dir == (*state)->direction ? "fwd" : "rev"); 4722 kprintf("pf: State failure on: %c %c %c %c | %c %c\n", 4723 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 4724 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 4725 ' ': '2', 4726 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 4727 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 4728 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 4729 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 4730 } 4731 REASON_SET(reason, PFRES_BADSTATE); 4732 return (PF_DROP); 4733 } 4734 4735 return (PF_PASS); 4736 } 4737 4738 /* 4739 * Called with state locked 4740 */ 4741 int 4742 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst, 4743 struct pf_state **state, struct pf_pdesc *pd, u_short *reason) 4744 { 4745 struct tcphdr *th = pd->hdr.tcp; 4746 4747 if (th->th_flags & TH_SYN) 4748 if (src->state < TCPS_SYN_SENT) 4749 src->state = TCPS_SYN_SENT; 4750 if (th->th_flags & TH_FIN) 4751 if (src->state < TCPS_CLOSING) 4752 src->state = TCPS_CLOSING; 4753 if (th->th_flags & TH_ACK) { 4754 if (dst->state == TCPS_SYN_SENT) { 4755 dst->state = TCPS_ESTABLISHED; 4756 if (src->state == TCPS_ESTABLISHED && 4757 (*state)->src_node != NULL && 4758 pf_src_connlimit(*state)) { 4759 REASON_SET(reason, PFRES_SRCLIMIT); 4760 return (PF_DROP); 4761 } 4762 } else if (dst->state == TCPS_CLOSING) { 4763 dst->state = TCPS_FIN_WAIT_2; 4764 } else if (src->state == TCPS_SYN_SENT && 4765 dst->state < TCPS_SYN_SENT) { 4766 /* 4767 * Handle a special sloppy case where we only see one 4768 * half of the connection. If there is a ACK after 4769 * the initial SYN without ever seeing a packet from 4770 * the destination, set the connection to established. 4771 */ 4772 dst->state = src->state = TCPS_ESTABLISHED; 4773 if ((*state)->src_node != NULL && 4774 pf_src_connlimit(*state)) { 4775 REASON_SET(reason, PFRES_SRCLIMIT); 4776 return (PF_DROP); 4777 } 4778 } else if (src->state == TCPS_CLOSING && 4779 dst->state == TCPS_ESTABLISHED && 4780 dst->seqlo == 0) { 4781 /* 4782 * Handle the closing of half connections where we 4783 * don't see the full bidirectional FIN/ACK+ACK 4784 * handshake. 4785 */ 4786 dst->state = TCPS_CLOSING; 4787 } 4788 } 4789 if (th->th_flags & TH_RST) 4790 src->state = dst->state = TCPS_TIME_WAIT; 4791 4792 /* update expire time */ 4793 (*state)->expire = time_second; 4794 if (src->state >= TCPS_FIN_WAIT_2 && 4795 dst->state >= TCPS_FIN_WAIT_2) 4796 (*state)->timeout = PFTM_TCP_CLOSED; 4797 else if (src->state >= TCPS_CLOSING && 4798 dst->state >= TCPS_CLOSING) 4799 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4800 else if (src->state < TCPS_ESTABLISHED || 4801 dst->state < TCPS_ESTABLISHED) 4802 (*state)->timeout = PFTM_TCP_OPENING; 4803 else if (src->state >= TCPS_CLOSING || 4804 dst->state >= TCPS_CLOSING) 4805 (*state)->timeout = PFTM_TCP_CLOSING; 4806 else 4807 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4808 4809 return (PF_PASS); 4810 } 4811 4812 /* 4813 * Test TCP connection state. Caller must hold the state locked. 4814 */ 4815 int 4816 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif, 4817 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 4818 u_short *reason) 4819 { 4820 struct pf_state_key_cmp key; 4821 struct tcphdr *th = pd->hdr.tcp; 4822 int copyback = 0; 4823 int error; 4824 struct pf_state_peer *src, *dst; 4825 struct pf_state_key *sk; 4826 4827 bzero(&key, sizeof(key)); 4828 key.af = pd->af; 4829 key.proto = IPPROTO_TCP; 4830 if (direction == PF_IN) { /* wire side, straight */ 4831 PF_ACPY(&key.addr[0], pd->src, key.af); 4832 PF_ACPY(&key.addr[1], pd->dst, key.af); 4833 key.port[0] = th->th_sport; 4834 key.port[1] = th->th_dport; 4835 if (pf_status.debug >= PF_DEBUG_MISC) { 4836 kprintf("test-tcp IN (%08x:%d) -> (%08x:%d)\n", 4837 ntohl(key.addr[0].addr32[0]), 4838 ntohs(key.port[0]), 4839 ntohl(key.addr[1].addr32[0]), 4840 ntohs(key.port[1])); 4841 } 4842 } else { /* stack side, reverse */ 4843 PF_ACPY(&key.addr[1], pd->src, key.af); 4844 PF_ACPY(&key.addr[0], pd->dst, key.af); 4845 key.port[1] = th->th_sport; 4846 key.port[0] = th->th_dport; 4847 if (pf_status.debug >= PF_DEBUG_MISC) { 4848 kprintf("test-tcp OUT (%08x:%d) <- (%08x:%d)\n", 4849 ntohl(key.addr[0].addr32[0]), 4850 ntohs(key.port[0]), 4851 ntohl(key.addr[1].addr32[0]), 4852 ntohs(key.port[1])); 4853 } 4854 } 4855 4856 STATE_LOOKUP(kif, &key, direction, *state, m); 4857 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 4858 4859 if (direction == (*state)->direction) { 4860 src = &(*state)->src; 4861 dst = &(*state)->dst; 4862 } else { 4863 src = &(*state)->dst; 4864 dst = &(*state)->src; 4865 } 4866 4867 sk = (*state)->key[pd->didx]; 4868 4869 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 4870 if (direction != (*state)->direction) { 4871 REASON_SET(reason, PFRES_SYNPROXY); 4872 FAIL (PF_SYNPROXY_DROP); 4873 } 4874 if (th->th_flags & TH_SYN) { 4875 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 4876 REASON_SET(reason, PFRES_SYNPROXY); 4877 FAIL (PF_DROP); 4878 } 4879 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 4880 pd->src, th->th_dport, th->th_sport, 4881 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 4882 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 4883 0, NULL, NULL); 4884 REASON_SET(reason, PFRES_SYNPROXY); 4885 FAIL (PF_SYNPROXY_DROP); 4886 } else if (!(th->th_flags & TH_ACK) || 4887 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4888 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4889 REASON_SET(reason, PFRES_SYNPROXY); 4890 FAIL (PF_DROP); 4891 } else if ((*state)->src_node != NULL && 4892 pf_src_connlimit(*state)) { 4893 REASON_SET(reason, PFRES_SRCLIMIT); 4894 FAIL (PF_DROP); 4895 } else 4896 (*state)->src.state = PF_TCPS_PROXY_DST; 4897 } 4898 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 4899 if (direction == (*state)->direction) { 4900 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 4901 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4902 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4903 REASON_SET(reason, PFRES_SYNPROXY); 4904 FAIL (PF_DROP); 4905 } 4906 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 4907 if ((*state)->dst.seqhi == 1) 4908 (*state)->dst.seqhi = htonl(karc4random()); 4909 pf_send_tcp((*state)->rule.ptr, pd->af, 4910 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4911 sk->port[pd->sidx], sk->port[pd->didx], 4912 (*state)->dst.seqhi, 0, TH_SYN, 0, 4913 (*state)->src.mss, 0, 0, (*state)->tag, NULL, NULL); 4914 REASON_SET(reason, PFRES_SYNPROXY); 4915 FAIL (PF_SYNPROXY_DROP); 4916 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 4917 (TH_SYN|TH_ACK)) || 4918 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 4919 REASON_SET(reason, PFRES_SYNPROXY); 4920 FAIL (PF_DROP); 4921 } else { 4922 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 4923 (*state)->dst.seqlo = ntohl(th->th_seq); 4924 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 4925 pd->src, th->th_dport, th->th_sport, 4926 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 4927 TH_ACK, (*state)->src.max_win, 0, 0, 0, 4928 (*state)->tag, NULL, NULL); 4929 pf_send_tcp((*state)->rule.ptr, pd->af, 4930 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4931 sk->port[pd->sidx], sk->port[pd->didx], 4932 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 4933 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 4934 0, NULL, NULL); 4935 (*state)->src.seqdiff = (*state)->dst.seqhi - 4936 (*state)->src.seqlo; 4937 (*state)->dst.seqdiff = (*state)->src.seqhi - 4938 (*state)->dst.seqlo; 4939 (*state)->src.seqhi = (*state)->src.seqlo + 4940 (*state)->dst.max_win; 4941 (*state)->dst.seqhi = (*state)->dst.seqlo + 4942 (*state)->src.max_win; 4943 (*state)->src.wscale = (*state)->dst.wscale = 0; 4944 (*state)->src.state = (*state)->dst.state = 4945 TCPS_ESTABLISHED; 4946 REASON_SET(reason, PFRES_SYNPROXY); 4947 FAIL (PF_SYNPROXY_DROP); 4948 } 4949 } 4950 4951 /* 4952 * Check for connection (addr+port pair) reuse. We can't actually 4953 * unlink the state if we don't own it. 4954 */ 4955 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) && 4956 dst->state >= TCPS_FIN_WAIT_2 && 4957 src->state >= TCPS_FIN_WAIT_2) { 4958 if (pf_status.debug >= PF_DEBUG_MISC) { 4959 kprintf("pf: state reuse "); 4960 pf_print_state(*state); 4961 pf_print_flags(th->th_flags); 4962 kprintf("\n"); 4963 } 4964 /* XXX make sure it's the same direction ?? */ 4965 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; 4966 if ((*state)->cpuid == mycpu->gd_cpuid) { 4967 pf_unlink_state(*state); 4968 *state = NULL; 4969 } else { 4970 (*state)->timeout = PFTM_PURGE; 4971 } 4972 FAIL (PF_DROP); 4973 } 4974 4975 if ((*state)->state_flags & PFSTATE_SLOPPY) { 4976 if (pf_tcp_track_sloppy(src, dst, state, pd, 4977 reason) == PF_DROP) { 4978 FAIL (PF_DROP); 4979 } 4980 } else { 4981 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, 4982 reason, ©back) == PF_DROP) { 4983 FAIL (PF_DROP); 4984 } 4985 } 4986 4987 /* translate source/destination address, if necessary */ 4988 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4989 struct pf_state_key *nk = (*state)->key[pd->didx]; 4990 4991 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4992 nk->port[pd->sidx] != th->th_sport) { 4993 /* 4994 * The translated source address may be completely 4995 * unrelated to the saved link header, make sure 4996 * a bridge doesn't try to use it. 4997 */ 4998 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 4999 pf_change_ap(pd->src, &th->th_sport, pd->ip_sum, 5000 &th->th_sum, &nk->addr[pd->sidx], 5001 nk->port[pd->sidx], 0, pd->af); 5002 } 5003 5004 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5005 nk->port[pd->didx] != th->th_dport) { 5006 /* 5007 * If we don't redispatch the packet will go into 5008 * the protocol stack on the wrong cpu for the 5009 * post-translated address. 5010 */ 5011 pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum, 5012 &th->th_sum, &nk->addr[pd->didx], 5013 nk->port[pd->didx], 0, pd->af); 5014 } 5015 copyback = 1; 5016 } 5017 5018 /* Copyback sequence modulation or stateful scrub changes if needed */ 5019 if (copyback) { 5020 m->m_flags &= ~M_HASH; 5021 m_copyback(m, off, sizeof(*th), (caddr_t)th); 5022 } 5023 5024 pfsync_update_state(*state); 5025 error = PF_PASS; 5026 done: 5027 if (*state) 5028 lockmgr(&(*state)->lk, LK_RELEASE); 5029 return (error); 5030 } 5031 5032 /* 5033 * Test UDP connection state. Caller must hold the state locked. 5034 */ 5035 int 5036 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif, 5037 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 5038 { 5039 struct pf_state_peer *src, *dst; 5040 struct pf_state_key_cmp key; 5041 struct udphdr *uh = pd->hdr.udp; 5042 5043 bzero(&key, sizeof(key)); 5044 key.af = pd->af; 5045 key.proto = IPPROTO_UDP; 5046 if (direction == PF_IN) { /* wire side, straight */ 5047 PF_ACPY(&key.addr[0], pd->src, key.af); 5048 PF_ACPY(&key.addr[1], pd->dst, key.af); 5049 key.port[0] = uh->uh_sport; 5050 key.port[1] = uh->uh_dport; 5051 } else { /* stack side, reverse */ 5052 PF_ACPY(&key.addr[1], pd->src, key.af); 5053 PF_ACPY(&key.addr[0], pd->dst, key.af); 5054 key.port[1] = uh->uh_sport; 5055 key.port[0] = uh->uh_dport; 5056 } 5057 5058 STATE_LOOKUP(kif, &key, direction, *state, m); 5059 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5060 5061 if (direction == (*state)->direction) { 5062 src = &(*state)->src; 5063 dst = &(*state)->dst; 5064 } else { 5065 src = &(*state)->dst; 5066 dst = &(*state)->src; 5067 } 5068 5069 /* update states */ 5070 if (src->state < PFUDPS_SINGLE) 5071 src->state = PFUDPS_SINGLE; 5072 if (dst->state == PFUDPS_SINGLE) 5073 dst->state = PFUDPS_MULTIPLE; 5074 5075 /* update expire time */ 5076 (*state)->expire = time_second; 5077 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 5078 (*state)->timeout = PFTM_UDP_MULTIPLE; 5079 else 5080 (*state)->timeout = PFTM_UDP_SINGLE; 5081 5082 /* translate source/destination address, if necessary */ 5083 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5084 struct pf_state_key *nk = (*state)->key[pd->didx]; 5085 5086 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5087 nk->port[pd->sidx] != uh->uh_sport) { 5088 /* 5089 * The translated source address may be completely 5090 * unrelated to the saved link header, make sure 5091 * a bridge doesn't try to use it. 5092 */ 5093 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED; 5094 m->m_flags &= ~M_HASH; 5095 pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum, 5096 &uh->uh_sum, &nk->addr[pd->sidx], 5097 nk->port[pd->sidx], 1, pd->af); 5098 } 5099 5100 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5101 nk->port[pd->didx] != uh->uh_dport) { 5102 /* 5103 * If we don't redispatch the packet will go into 5104 * the protocol stack on the wrong cpu for the 5105 * post-translated address. 5106 */ 5107 m->m_flags &= ~M_HASH; 5108 pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum, 5109 &uh->uh_sum, &nk->addr[pd->didx], 5110 nk->port[pd->didx], 1, pd->af); 5111 } 5112 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 5113 } 5114 5115 pfsync_update_state(*state); 5116 lockmgr(&(*state)->lk, LK_RELEASE); 5117 return (PF_PASS); 5118 } 5119 5120 /* 5121 * Test ICMP connection state. Caller must hold the state locked. 5122 */ 5123 int 5124 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif, 5125 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 5126 u_short *reason) 5127 { 5128 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 5129 u_int16_t icmpid = 0, *icmpsum = NULL; 5130 u_int8_t icmptype = 0; 5131 int state_icmp = 0; 5132 int error; 5133 struct pf_state_key_cmp key; 5134 5135 bzero(&key, sizeof(key)); 5136 5137 switch (pd->proto) { 5138 #ifdef INET 5139 case IPPROTO_ICMP: 5140 icmptype = pd->hdr.icmp->icmp_type; 5141 icmpid = pd->hdr.icmp->icmp_id; 5142 icmpsum = &pd->hdr.icmp->icmp_cksum; 5143 5144 if (icmptype == ICMP_UNREACH || 5145 icmptype == ICMP_SOURCEQUENCH || 5146 icmptype == ICMP_REDIRECT || 5147 icmptype == ICMP_TIMXCEED || 5148 icmptype == ICMP_PARAMPROB) 5149 state_icmp++; 5150 break; 5151 #endif /* INET */ 5152 #ifdef INET6 5153 case IPPROTO_ICMPV6: 5154 icmptype = pd->hdr.icmp6->icmp6_type; 5155 icmpid = pd->hdr.icmp6->icmp6_id; 5156 icmpsum = &pd->hdr.icmp6->icmp6_cksum; 5157 5158 if (icmptype == ICMP6_DST_UNREACH || 5159 icmptype == ICMP6_PACKET_TOO_BIG || 5160 icmptype == ICMP6_TIME_EXCEEDED || 5161 icmptype == ICMP6_PARAM_PROB) 5162 state_icmp++; 5163 break; 5164 #endif /* INET6 */ 5165 } 5166 5167 if (!state_icmp) { 5168 5169 /* 5170 * ICMP query/reply message not related to a TCP/UDP packet. 5171 * Search for an ICMP state. 5172 */ 5173 key.af = pd->af; 5174 key.proto = pd->proto; 5175 key.port[0] = key.port[1] = icmpid; 5176 if (direction == PF_IN) { /* wire side, straight */ 5177 PF_ACPY(&key.addr[0], pd->src, key.af); 5178 PF_ACPY(&key.addr[1], pd->dst, key.af); 5179 } else { /* stack side, reverse */ 5180 PF_ACPY(&key.addr[1], pd->src, key.af); 5181 PF_ACPY(&key.addr[0], pd->dst, key.af); 5182 } 5183 5184 STATE_LOOKUP(kif, &key, direction, *state, m); 5185 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5186 5187 (*state)->expire = time_second; 5188 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 5189 5190 /* translate source/destination address, if necessary */ 5191 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5192 struct pf_state_key *nk = (*state)->key[pd->didx]; 5193 5194 switch (pd->af) { 5195 #ifdef INET 5196 case AF_INET: 5197 if (PF_ANEQ(pd->src, 5198 &nk->addr[pd->sidx], AF_INET)) 5199 pf_change_a(&saddr->v4.s_addr, 5200 pd->ip_sum, 5201 nk->addr[pd->sidx].v4.s_addr, 0); 5202 5203 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 5204 AF_INET)) 5205 pf_change_a(&daddr->v4.s_addr, 5206 pd->ip_sum, 5207 nk->addr[pd->didx].v4.s_addr, 0); 5208 5209 if (nk->port[0] != 5210 pd->hdr.icmp->icmp_id) { 5211 pd->hdr.icmp->icmp_cksum = 5212 pf_cksum_fixup( 5213 pd->hdr.icmp->icmp_cksum, icmpid, 5214 nk->port[pd->sidx], 0); 5215 pd->hdr.icmp->icmp_id = 5216 nk->port[pd->sidx]; 5217 } 5218 5219 m->m_flags &= ~M_HASH; 5220 m_copyback(m, off, ICMP_MINLEN, 5221 (caddr_t)pd->hdr.icmp); 5222 break; 5223 #endif /* INET */ 5224 #ifdef INET6 5225 case AF_INET6: 5226 if (PF_ANEQ(pd->src, 5227 &nk->addr[pd->sidx], AF_INET6)) 5228 pf_change_a6(saddr, 5229 &pd->hdr.icmp6->icmp6_cksum, 5230 &nk->addr[pd->sidx], 0); 5231 5232 if (PF_ANEQ(pd->dst, 5233 &nk->addr[pd->didx], AF_INET6)) 5234 pf_change_a6(daddr, 5235 &pd->hdr.icmp6->icmp6_cksum, 5236 &nk->addr[pd->didx], 0); 5237 5238 m->m_flags &= ~M_HASH; 5239 m_copyback(m, off, 5240 sizeof(struct icmp6_hdr), 5241 (caddr_t)pd->hdr.icmp6); 5242 break; 5243 #endif /* INET6 */ 5244 } 5245 } 5246 } else { 5247 /* 5248 * ICMP error message in response to a TCP/UDP packet. 5249 * Extract the inner TCP/UDP header and search for that state. 5250 */ 5251 5252 struct pf_pdesc pd2; 5253 #ifdef INET 5254 struct ip h2; 5255 #endif /* INET */ 5256 #ifdef INET6 5257 struct ip6_hdr h2_6; 5258 int terminal = 0; 5259 #endif /* INET6 */ 5260 int ipoff2; 5261 int off2; 5262 5263 pd2.not_cpu_localized = 1; 5264 pd2.af = pd->af; 5265 /* Payload packet is from the opposite direction. */ 5266 pd2.sidx = (direction == PF_IN) ? 1 : 0; 5267 pd2.didx = (direction == PF_IN) ? 0 : 1; 5268 switch (pd->af) { 5269 #ifdef INET 5270 case AF_INET: 5271 /* offset of h2 in mbuf chain */ 5272 ipoff2 = off + ICMP_MINLEN; 5273 5274 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 5275 NULL, reason, pd2.af)) { 5276 DPFPRINTF(PF_DEBUG_MISC, 5277 ("pf: ICMP error message too short " 5278 "(ip)\n")); 5279 FAIL (PF_DROP); 5280 } 5281 /* 5282 * ICMP error messages don't refer to non-first 5283 * fragments 5284 */ 5285 if (h2.ip_off & htons(IP_OFFMASK)) { 5286 REASON_SET(reason, PFRES_FRAG); 5287 FAIL (PF_DROP); 5288 } 5289 5290 /* offset of protocol header that follows h2 */ 5291 off2 = ipoff2 + (h2.ip_hl << 2); 5292 5293 pd2.proto = h2.ip_p; 5294 pd2.src = (struct pf_addr *)&h2.ip_src; 5295 pd2.dst = (struct pf_addr *)&h2.ip_dst; 5296 pd2.ip_sum = &h2.ip_sum; 5297 break; 5298 #endif /* INET */ 5299 #ifdef INET6 5300 case AF_INET6: 5301 ipoff2 = off + sizeof(struct icmp6_hdr); 5302 5303 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 5304 NULL, reason, pd2.af)) { 5305 DPFPRINTF(PF_DEBUG_MISC, 5306 ("pf: ICMP error message too short " 5307 "(ip6)\n")); 5308 FAIL (PF_DROP); 5309 } 5310 pd2.proto = h2_6.ip6_nxt; 5311 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 5312 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 5313 pd2.ip_sum = NULL; 5314 off2 = ipoff2 + sizeof(h2_6); 5315 do { 5316 switch (pd2.proto) { 5317 case IPPROTO_FRAGMENT: 5318 /* 5319 * ICMPv6 error messages for 5320 * non-first fragments 5321 */ 5322 REASON_SET(reason, PFRES_FRAG); 5323 FAIL (PF_DROP); 5324 case IPPROTO_AH: 5325 case IPPROTO_HOPOPTS: 5326 case IPPROTO_ROUTING: 5327 case IPPROTO_DSTOPTS: { 5328 /* get next header and header length */ 5329 struct ip6_ext opt6; 5330 5331 if (!pf_pull_hdr(m, off2, &opt6, 5332 sizeof(opt6), NULL, reason, 5333 pd2.af)) { 5334 DPFPRINTF(PF_DEBUG_MISC, 5335 ("pf: ICMPv6 short opt\n")); 5336 FAIL (PF_DROP); 5337 } 5338 if (pd2.proto == IPPROTO_AH) 5339 off2 += (opt6.ip6e_len + 2) * 4; 5340 else 5341 off2 += (opt6.ip6e_len + 1) * 8; 5342 pd2.proto = opt6.ip6e_nxt; 5343 /* goto the next header */ 5344 break; 5345 } 5346 default: 5347 terminal++; 5348 break; 5349 } 5350 } while (!terminal); 5351 break; 5352 #endif /* INET6 */ 5353 default: 5354 DPFPRINTF(PF_DEBUG_MISC, 5355 ("pf: ICMP AF %d unknown (ip6)\n", pd->af)); 5356 FAIL (PF_DROP); 5357 break; 5358 } 5359 5360 switch (pd2.proto) { 5361 case IPPROTO_TCP: { 5362 struct tcphdr th; 5363 u_int32_t seq; 5364 struct pf_state_peer *src, *dst; 5365 u_int8_t dws; 5366 int copyback = 0; 5367 5368 /* 5369 * Only the first 8 bytes of the TCP header can be 5370 * expected. Don't access any TCP header fields after 5371 * th_seq, an ackskew test is not possible. 5372 */ 5373 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 5374 pd2.af)) { 5375 DPFPRINTF(PF_DEBUG_MISC, 5376 ("pf: ICMP error message too short " 5377 "(tcp)\n")); 5378 FAIL (PF_DROP); 5379 } 5380 5381 key.af = pd2.af; 5382 key.proto = IPPROTO_TCP; 5383 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5384 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5385 key.port[pd2.sidx] = th.th_sport; 5386 key.port[pd2.didx] = th.th_dport; 5387 5388 STATE_LOOKUP(kif, &key, direction, *state, m); 5389 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5390 5391 if (direction == (*state)->direction) { 5392 src = &(*state)->dst; 5393 dst = &(*state)->src; 5394 } else { 5395 src = &(*state)->src; 5396 dst = &(*state)->dst; 5397 } 5398 5399 if (src->wscale && dst->wscale) 5400 dws = dst->wscale & PF_WSCALE_MASK; 5401 else 5402 dws = 0; 5403 5404 /* Demodulate sequence number */ 5405 seq = ntohl(th.th_seq) - src->seqdiff; 5406 if (src->seqdiff) { 5407 pf_change_a(&th.th_seq, icmpsum, 5408 htonl(seq), 0); 5409 copyback = 1; 5410 } 5411 5412 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 5413 (!SEQ_GEQ(src->seqhi, seq) || 5414 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 5415 if (pf_status.debug >= PF_DEBUG_MISC) { 5416 kprintf("pf: BAD ICMP %d:%d ", 5417 icmptype, pd->hdr.icmp->icmp_code); 5418 pf_print_host(pd->src, 0, pd->af); 5419 kprintf(" -> "); 5420 pf_print_host(pd->dst, 0, pd->af); 5421 kprintf(" state: "); 5422 pf_print_state(*state); 5423 kprintf(" seq=%u\n", seq); 5424 } 5425 REASON_SET(reason, PFRES_BADSTATE); 5426 FAIL (PF_DROP); 5427 } else { 5428 if (pf_status.debug >= PF_DEBUG_MISC) { 5429 kprintf("pf: OK ICMP %d:%d ", 5430 icmptype, pd->hdr.icmp->icmp_code); 5431 pf_print_host(pd->src, 0, pd->af); 5432 kprintf(" -> "); 5433 pf_print_host(pd->dst, 0, pd->af); 5434 kprintf(" state: "); 5435 pf_print_state(*state); 5436 kprintf(" seq=%u\n", seq); 5437 } 5438 } 5439 5440 /* translate source/destination address, if necessary */ 5441 if ((*state)->key[PF_SK_WIRE] != 5442 (*state)->key[PF_SK_STACK]) { 5443 struct pf_state_key *nk = 5444 (*state)->key[pd->didx]; 5445 5446 if (PF_ANEQ(pd2.src, 5447 &nk->addr[pd2.sidx], pd2.af) || 5448 nk->port[pd2.sidx] != th.th_sport) 5449 pf_change_icmp(pd2.src, &th.th_sport, 5450 daddr, &nk->addr[pd2.sidx], 5451 nk->port[pd2.sidx], NULL, 5452 pd2.ip_sum, icmpsum, 5453 pd->ip_sum, 0, pd2.af); 5454 5455 if (PF_ANEQ(pd2.dst, 5456 &nk->addr[pd2.didx], pd2.af) || 5457 nk->port[pd2.didx] != th.th_dport) 5458 pf_change_icmp(pd2.dst, &th.th_dport, 5459 NULL, /* XXX Inbound NAT? */ 5460 &nk->addr[pd2.didx], 5461 nk->port[pd2.didx], NULL, 5462 pd2.ip_sum, icmpsum, 5463 pd->ip_sum, 0, pd2.af); 5464 copyback = 1; 5465 } 5466 5467 if (copyback) { 5468 switch (pd2.af) { 5469 #ifdef INET 5470 case AF_INET: 5471 m_copyback(m, off, ICMP_MINLEN, 5472 (caddr_t)pd->hdr.icmp); 5473 m_copyback(m, ipoff2, sizeof(h2), 5474 (caddr_t)&h2); 5475 break; 5476 #endif /* INET */ 5477 #ifdef INET6 5478 case AF_INET6: 5479 m_copyback(m, off, 5480 sizeof(struct icmp6_hdr), 5481 (caddr_t)pd->hdr.icmp6); 5482 m_copyback(m, ipoff2, sizeof(h2_6), 5483 (caddr_t)&h2_6); 5484 break; 5485 #endif /* INET6 */ 5486 } 5487 m->m_flags &= ~M_HASH; 5488 m_copyback(m, off2, 8, (caddr_t)&th); 5489 } 5490 break; 5491 } 5492 case IPPROTO_UDP: { 5493 struct udphdr uh; 5494 5495 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 5496 NULL, reason, pd2.af)) { 5497 DPFPRINTF(PF_DEBUG_MISC, 5498 ("pf: ICMP error message too short " 5499 "(udp)\n")); 5500 return (PF_DROP); 5501 } 5502 5503 key.af = pd2.af; 5504 key.proto = IPPROTO_UDP; 5505 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5506 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5507 key.port[pd2.sidx] = uh.uh_sport; 5508 key.port[pd2.didx] = uh.uh_dport; 5509 5510 STATE_LOOKUP(kif, &key, direction, *state, m); 5511 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5512 5513 /* translate source/destination address, if necessary */ 5514 if ((*state)->key[PF_SK_WIRE] != 5515 (*state)->key[PF_SK_STACK]) { 5516 struct pf_state_key *nk = 5517 (*state)->key[pd->didx]; 5518 5519 if (PF_ANEQ(pd2.src, 5520 &nk->addr[pd2.sidx], pd2.af) || 5521 nk->port[pd2.sidx] != uh.uh_sport) 5522 pf_change_icmp(pd2.src, &uh.uh_sport, 5523 daddr, &nk->addr[pd2.sidx], 5524 nk->port[pd2.sidx], &uh.uh_sum, 5525 pd2.ip_sum, icmpsum, 5526 pd->ip_sum, 1, pd2.af); 5527 5528 if (PF_ANEQ(pd2.dst, 5529 &nk->addr[pd2.didx], pd2.af) || 5530 nk->port[pd2.didx] != uh.uh_dport) 5531 pf_change_icmp(pd2.dst, &uh.uh_dport, 5532 NULL, /* XXX Inbound NAT? */ 5533 &nk->addr[pd2.didx], 5534 nk->port[pd2.didx], &uh.uh_sum, 5535 pd2.ip_sum, icmpsum, 5536 pd->ip_sum, 1, pd2.af); 5537 5538 switch (pd2.af) { 5539 #ifdef INET 5540 case AF_INET: 5541 m_copyback(m, off, ICMP_MINLEN, 5542 (caddr_t)pd->hdr.icmp); 5543 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5544 break; 5545 #endif /* INET */ 5546 #ifdef INET6 5547 case AF_INET6: 5548 m_copyback(m, off, 5549 sizeof(struct icmp6_hdr), 5550 (caddr_t)pd->hdr.icmp6); 5551 m_copyback(m, ipoff2, sizeof(h2_6), 5552 (caddr_t)&h2_6); 5553 break; 5554 #endif /* INET6 */ 5555 } 5556 m->m_flags &= ~M_HASH; 5557 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 5558 } 5559 break; 5560 } 5561 #ifdef INET 5562 case IPPROTO_ICMP: { 5563 struct icmp iih; 5564 5565 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 5566 NULL, reason, pd2.af)) { 5567 DPFPRINTF(PF_DEBUG_MISC, 5568 ("pf: ICMP error message too short i" 5569 "(icmp)\n")); 5570 return (PF_DROP); 5571 } 5572 5573 key.af = pd2.af; 5574 key.proto = IPPROTO_ICMP; 5575 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5576 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5577 key.port[0] = key.port[1] = iih.icmp_id; 5578 5579 STATE_LOOKUP(kif, &key, direction, *state, m); 5580 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5581 5582 /* translate source/destination address, if necessary */ 5583 if ((*state)->key[PF_SK_WIRE] != 5584 (*state)->key[PF_SK_STACK]) { 5585 struct pf_state_key *nk = 5586 (*state)->key[pd->didx]; 5587 5588 if (PF_ANEQ(pd2.src, 5589 &nk->addr[pd2.sidx], pd2.af) || 5590 nk->port[pd2.sidx] != iih.icmp_id) 5591 pf_change_icmp(pd2.src, &iih.icmp_id, 5592 daddr, &nk->addr[pd2.sidx], 5593 nk->port[pd2.sidx], NULL, 5594 pd2.ip_sum, icmpsum, 5595 pd->ip_sum, 0, AF_INET); 5596 5597 if (PF_ANEQ(pd2.dst, 5598 &nk->addr[pd2.didx], pd2.af) || 5599 nk->port[pd2.didx] != iih.icmp_id) 5600 pf_change_icmp(pd2.dst, &iih.icmp_id, 5601 NULL, /* XXX Inbound NAT? */ 5602 &nk->addr[pd2.didx], 5603 nk->port[pd2.didx], NULL, 5604 pd2.ip_sum, icmpsum, 5605 pd->ip_sum, 0, AF_INET); 5606 5607 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 5608 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5609 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 5610 m->m_flags &= ~M_HASH; 5611 } 5612 break; 5613 } 5614 #endif /* INET */ 5615 #ifdef INET6 5616 case IPPROTO_ICMPV6: { 5617 struct icmp6_hdr iih; 5618 5619 if (!pf_pull_hdr(m, off2, &iih, 5620 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 5621 DPFPRINTF(PF_DEBUG_MISC, 5622 ("pf: ICMP error message too short " 5623 "(icmp6)\n")); 5624 FAIL (PF_DROP); 5625 } 5626 5627 key.af = pd2.af; 5628 key.proto = IPPROTO_ICMPV6; 5629 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5630 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5631 key.port[0] = key.port[1] = iih.icmp6_id; 5632 5633 STATE_LOOKUP(kif, &key, direction, *state, m); 5634 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5635 5636 /* translate source/destination address, if necessary */ 5637 if ((*state)->key[PF_SK_WIRE] != 5638 (*state)->key[PF_SK_STACK]) { 5639 struct pf_state_key *nk = 5640 (*state)->key[pd->didx]; 5641 5642 if (PF_ANEQ(pd2.src, 5643 &nk->addr[pd2.sidx], pd2.af) || 5644 nk->port[pd2.sidx] != iih.icmp6_id) 5645 pf_change_icmp(pd2.src, &iih.icmp6_id, 5646 daddr, &nk->addr[pd2.sidx], 5647 nk->port[pd2.sidx], NULL, 5648 pd2.ip_sum, icmpsum, 5649 pd->ip_sum, 0, AF_INET6); 5650 5651 if (PF_ANEQ(pd2.dst, 5652 &nk->addr[pd2.didx], pd2.af) || 5653 nk->port[pd2.didx] != iih.icmp6_id) 5654 pf_change_icmp(pd2.dst, &iih.icmp6_id, 5655 NULL, /* XXX Inbound NAT? */ 5656 &nk->addr[pd2.didx], 5657 nk->port[pd2.didx], NULL, 5658 pd2.ip_sum, icmpsum, 5659 pd->ip_sum, 0, AF_INET6); 5660 5661 m_copyback(m, off, sizeof(struct icmp6_hdr), 5662 (caddr_t)pd->hdr.icmp6); 5663 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 5664 m_copyback(m, off2, sizeof(struct icmp6_hdr), 5665 (caddr_t)&iih); 5666 m->m_flags &= ~M_HASH; 5667 } 5668 break; 5669 } 5670 #endif /* INET6 */ 5671 default: { 5672 key.af = pd2.af; 5673 key.proto = pd2.proto; 5674 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5675 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5676 key.port[0] = key.port[1] = 0; 5677 5678 STATE_LOOKUP(kif, &key, direction, *state, m); 5679 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5680 5681 /* translate source/destination address, if necessary */ 5682 if ((*state)->key[PF_SK_WIRE] != 5683 (*state)->key[PF_SK_STACK]) { 5684 struct pf_state_key *nk = 5685 (*state)->key[pd->didx]; 5686 5687 if (PF_ANEQ(pd2.src, 5688 &nk->addr[pd2.sidx], pd2.af)) 5689 pf_change_icmp(pd2.src, NULL, daddr, 5690 &nk->addr[pd2.sidx], 0, NULL, 5691 pd2.ip_sum, icmpsum, 5692 pd->ip_sum, 0, pd2.af); 5693 5694 if (PF_ANEQ(pd2.dst, 5695 &nk->addr[pd2.didx], pd2.af)) 5696 pf_change_icmp(pd2.src, NULL, 5697 NULL, /* XXX Inbound NAT? */ 5698 &nk->addr[pd2.didx], 0, NULL, 5699 pd2.ip_sum, icmpsum, 5700 pd->ip_sum, 0, pd2.af); 5701 5702 switch (pd2.af) { 5703 #ifdef INET 5704 case AF_INET: 5705 m_copyback(m, off, ICMP_MINLEN, 5706 (caddr_t)pd->hdr.icmp); 5707 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5708 m->m_flags &= ~M_HASH; 5709 break; 5710 #endif /* INET */ 5711 #ifdef INET6 5712 case AF_INET6: 5713 m_copyback(m, off, 5714 sizeof(struct icmp6_hdr), 5715 (caddr_t)pd->hdr.icmp6); 5716 m_copyback(m, ipoff2, sizeof(h2_6), 5717 (caddr_t)&h2_6); 5718 m->m_flags &= ~M_HASH; 5719 break; 5720 #endif /* INET6 */ 5721 } 5722 } 5723 break; 5724 } 5725 } 5726 } 5727 5728 pfsync_update_state(*state); 5729 error = PF_PASS; 5730 done: 5731 if (*state) 5732 lockmgr(&(*state)->lk, LK_RELEASE); 5733 return (error); 5734 } 5735 5736 /* 5737 * Test other connection state. Caller must hold the state locked. 5738 */ 5739 int 5740 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif, 5741 struct mbuf *m, struct pf_pdesc *pd) 5742 { 5743 struct pf_state_peer *src, *dst; 5744 struct pf_state_key_cmp key; 5745 5746 bzero(&key, sizeof(key)); 5747 key.af = pd->af; 5748 key.proto = pd->proto; 5749 if (direction == PF_IN) { 5750 PF_ACPY(&key.addr[0], pd->src, key.af); 5751 PF_ACPY(&key.addr[1], pd->dst, key.af); 5752 key.port[0] = key.port[1] = 0; 5753 } else { 5754 PF_ACPY(&key.addr[1], pd->src, key.af); 5755 PF_ACPY(&key.addr[0], pd->dst, key.af); 5756 key.port[1] = key.port[0] = 0; 5757 } 5758 5759 STATE_LOOKUP(kif, &key, direction, *state, m); 5760 lockmgr(&(*state)->lk, LK_EXCLUSIVE); 5761 5762 if (direction == (*state)->direction) { 5763 src = &(*state)->src; 5764 dst = &(*state)->dst; 5765 } else { 5766 src = &(*state)->dst; 5767 dst = &(*state)->src; 5768 } 5769 5770 /* update states */ 5771 if (src->state < PFOTHERS_SINGLE) 5772 src->state = PFOTHERS_SINGLE; 5773 if (dst->state == PFOTHERS_SINGLE) 5774 dst->state = PFOTHERS_MULTIPLE; 5775 5776 /* update expire time */ 5777 (*state)->expire = time_second; 5778 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 5779 (*state)->timeout = PFTM_OTHER_MULTIPLE; 5780 else 5781 (*state)->timeout = PFTM_OTHER_SINGLE; 5782 5783 /* translate source/destination address, if necessary */ 5784 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5785 struct pf_state_key *nk = (*state)->key[pd->didx]; 5786 5787 KKASSERT(nk); 5788 KKASSERT(pd); 5789 KKASSERT(pd->src); 5790 KKASSERT(pd->dst); 5791 switch (pd->af) { 5792 #ifdef INET 5793 case AF_INET: 5794 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5795 pf_change_a(&pd->src->v4.s_addr, 5796 pd->ip_sum, 5797 nk->addr[pd->sidx].v4.s_addr, 5798 0); 5799 5800 5801 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5802 pf_change_a(&pd->dst->v4.s_addr, 5803 pd->ip_sum, 5804 nk->addr[pd->didx].v4.s_addr, 5805 0); 5806 5807 break; 5808 #endif /* INET */ 5809 #ifdef INET6 5810 case AF_INET6: 5811 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5812 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5813 5814 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5815 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5816 #endif /* INET6 */ 5817 } 5818 } 5819 5820 pfsync_update_state(*state); 5821 lockmgr(&(*state)->lk, LK_RELEASE); 5822 return (PF_PASS); 5823 } 5824 5825 /* 5826 * ipoff and off are measured from the start of the mbuf chain. 5827 * h must be at "ipoff" on the mbuf chain. 5828 */ 5829 void * 5830 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 5831 u_short *actionp, u_short *reasonp, sa_family_t af) 5832 { 5833 switch (af) { 5834 #ifdef INET 5835 case AF_INET: { 5836 struct ip *h = mtod(m, struct ip *); 5837 u_int16_t fragoff = (h->ip_off & IP_OFFMASK) << 3; 5838 5839 if (fragoff) { 5840 if (fragoff >= len) 5841 ACTION_SET(actionp, PF_PASS); 5842 else { 5843 ACTION_SET(actionp, PF_DROP); 5844 REASON_SET(reasonp, PFRES_FRAG); 5845 } 5846 return (NULL); 5847 } 5848 if (m->m_pkthdr.len < off + len || 5849 h->ip_len < off + len) { 5850 ACTION_SET(actionp, PF_DROP); 5851 REASON_SET(reasonp, PFRES_SHORT); 5852 return (NULL); 5853 } 5854 break; 5855 } 5856 #endif /* INET */ 5857 #ifdef INET6 5858 case AF_INET6: { 5859 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 5860 5861 if (m->m_pkthdr.len < off + len || 5862 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 5863 (unsigned)(off + len)) { 5864 ACTION_SET(actionp, PF_DROP); 5865 REASON_SET(reasonp, PFRES_SHORT); 5866 return (NULL); 5867 } 5868 break; 5869 } 5870 #endif /* INET6 */ 5871 } 5872 m_copydata(m, off, len, p); 5873 return (p); 5874 } 5875 5876 int 5877 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif) 5878 { 5879 struct sockaddr_in *dst; 5880 int ret = 1; 5881 int check_mpath; 5882 #ifdef INET6 5883 struct sockaddr_in6 *dst6; 5884 struct route_in6 ro; 5885 #else 5886 struct route ro; 5887 #endif 5888 struct radix_node *rn; 5889 struct rtentry *rt; 5890 struct ifnet *ifp; 5891 5892 check_mpath = 0; 5893 bzero(&ro, sizeof(ro)); 5894 switch (af) { 5895 case AF_INET: 5896 dst = satosin(&ro.ro_dst); 5897 dst->sin_family = AF_INET; 5898 dst->sin_len = sizeof(*dst); 5899 dst->sin_addr = addr->v4; 5900 break; 5901 #ifdef INET6 5902 case AF_INET6: 5903 dst6 = (struct sockaddr_in6 *)&ro.ro_dst; 5904 dst6->sin6_family = AF_INET6; 5905 dst6->sin6_len = sizeof(*dst6); 5906 dst6->sin6_addr = addr->v6; 5907 break; 5908 #endif /* INET6 */ 5909 default: 5910 return (0); 5911 } 5912 5913 /* Skip checks for ipsec interfaces */ 5914 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 5915 goto out; 5916 5917 rtalloc_ign((struct route *)&ro, 0); 5918 5919 if (ro.ro_rt != NULL) { 5920 /* No interface given, this is a no-route check */ 5921 if (kif == NULL) 5922 goto out; 5923 5924 if (kif->pfik_ifp == NULL) { 5925 ret = 0; 5926 goto out; 5927 } 5928 5929 /* Perform uRPF check if passed input interface */ 5930 ret = 0; 5931 rn = (struct radix_node *)ro.ro_rt; 5932 do { 5933 rt = (struct rtentry *)rn; 5934 ifp = rt->rt_ifp; 5935 5936 if (kif->pfik_ifp == ifp) 5937 ret = 1; 5938 rn = NULL; 5939 } while (check_mpath == 1 && rn != NULL && ret == 0); 5940 } else 5941 ret = 0; 5942 out: 5943 if (ro.ro_rt != NULL) 5944 RTFREE(ro.ro_rt); 5945 return (ret); 5946 } 5947 5948 int 5949 pf_rtlabel_match(struct pf_addr *addr, sa_family_t af, struct pf_addr_wrap *aw) 5950 { 5951 struct sockaddr_in *dst; 5952 #ifdef INET6 5953 struct sockaddr_in6 *dst6; 5954 struct route_in6 ro; 5955 #else 5956 struct route ro; 5957 #endif 5958 int ret = 0; 5959 5960 ASSERT_LWKT_TOKEN_HELD(&pf_token); 5961 5962 bzero(&ro, sizeof(ro)); 5963 switch (af) { 5964 case AF_INET: 5965 dst = satosin(&ro.ro_dst); 5966 dst->sin_family = AF_INET; 5967 dst->sin_len = sizeof(*dst); 5968 dst->sin_addr = addr->v4; 5969 break; 5970 #ifdef INET6 5971 case AF_INET6: 5972 dst6 = (struct sockaddr_in6 *)&ro.ro_dst; 5973 dst6->sin6_family = AF_INET6; 5974 dst6->sin6_len = sizeof(*dst6); 5975 dst6->sin6_addr = addr->v6; 5976 break; 5977 #endif /* INET6 */ 5978 default: 5979 return (0); 5980 } 5981 5982 rtalloc_ign((struct route *)&ro, (RTF_CLONING | RTF_PRCLONING)); 5983 5984 if (ro.ro_rt != NULL) { 5985 RTFREE(ro.ro_rt); 5986 } 5987 5988 return (ret); 5989 } 5990 5991 #ifdef INET 5992 void 5993 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, 5994 struct pf_state *s, struct pf_pdesc *pd) 5995 { 5996 struct mbuf *m0, *m1; 5997 struct route iproute; 5998 struct route *ro = NULL; 5999 struct sockaddr_in *dst; 6000 struct ip *ip; 6001 struct ifnet *ifp = NULL; 6002 struct pf_addr naddr; 6003 struct pf_src_node *sn = NULL; 6004 int error = 0; 6005 int sw_csum; 6006 #ifdef IPSEC 6007 struct m_tag *mtag; 6008 #endif /* IPSEC */ 6009 6010 ASSERT_LWKT_TOKEN_HELD(&pf_token); 6011 6012 if (m == NULL || *m == NULL || r == NULL || 6013 (dir != PF_IN && dir != PF_OUT) || oifp == NULL) 6014 panic("pf_route: invalid parameters"); 6015 6016 if (((*m)->m_pkthdr.fw_flags & PF_MBUF_ROUTED) == 0) { 6017 (*m)->m_pkthdr.fw_flags |= PF_MBUF_ROUTED; 6018 (*m)->m_pkthdr.pf.routed = 1; 6019 } else { 6020 if ((*m)->m_pkthdr.pf.routed++ > 3) { 6021 m0 = *m; 6022 *m = NULL; 6023 goto bad; 6024 } 6025 } 6026 6027 if (r->rt == PF_DUPTO) { 6028 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { 6029 return; 6030 } 6031 } else { 6032 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 6033 return; 6034 } 6035 m0 = *m; 6036 } 6037 6038 if (m0->m_len < sizeof(struct ip)) { 6039 DPFPRINTF(PF_DEBUG_URGENT, 6040 ("pf_route: m0->m_len < sizeof(struct ip)\n")); 6041 goto bad; 6042 } 6043 6044 ip = mtod(m0, struct ip *); 6045 6046 ro = &iproute; 6047 bzero((caddr_t)ro, sizeof(*ro)); 6048 dst = satosin(&ro->ro_dst); 6049 dst->sin_family = AF_INET; 6050 dst->sin_len = sizeof(*dst); 6051 dst->sin_addr = ip->ip_dst; 6052 6053 if (r->rt == PF_FASTROUTE) { 6054 rtalloc(ro); 6055 if (ro->ro_rt == 0) { 6056 ipstat.ips_noroute++; 6057 goto bad; 6058 } 6059 6060 ifp = ro->ro_rt->rt_ifp; 6061 ro->ro_rt->rt_use++; 6062 6063 if (ro->ro_rt->rt_flags & RTF_GATEWAY) 6064 dst = satosin(ro->ro_rt->rt_gateway); 6065 } else { 6066 if (TAILQ_EMPTY(&r->rpool.list)) { 6067 DPFPRINTF(PF_DEBUG_URGENT, 6068 ("pf_route: TAILQ_EMPTY(&r->rpool.list)\n")); 6069 goto bad; 6070 } 6071 if (s == NULL) { 6072 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 6073 &naddr, NULL, &sn); 6074 if (!PF_AZERO(&naddr, AF_INET)) 6075 dst->sin_addr.s_addr = naddr.v4.s_addr; 6076 ifp = r->rpool.cur->kif ? 6077 r->rpool.cur->kif->pfik_ifp : NULL; 6078 } else { 6079 if (!PF_AZERO(&s->rt_addr, AF_INET)) 6080 dst->sin_addr.s_addr = 6081 s->rt_addr.v4.s_addr; 6082 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6083 } 6084 } 6085 if (ifp == NULL) 6086 goto bad; 6087 6088 if (oifp != ifp) { 6089 if (pf_test(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) { 6090 goto bad; 6091 } else if (m0 == NULL) { 6092 goto done; 6093 } 6094 if (m0->m_len < sizeof(struct ip)) { 6095 DPFPRINTF(PF_DEBUG_URGENT, 6096 ("pf_route: m0->m_len < sizeof(struct ip)\n")); 6097 goto bad; 6098 } 6099 ip = mtod(m0, struct ip *); 6100 } 6101 6102 /* Copied from FreeBSD 5.1-CURRENT ip_output. */ 6103 m0->m_pkthdr.csum_flags |= CSUM_IP; 6104 sw_csum = m0->m_pkthdr.csum_flags & ~ifp->if_hwassist; 6105 if (sw_csum & CSUM_DELAY_DATA) { 6106 in_delayed_cksum(m0); 6107 sw_csum &= ~CSUM_DELAY_DATA; 6108 } 6109 m0->m_pkthdr.csum_flags &= ifp->if_hwassist; 6110 m0->m_pkthdr.csum_iphlen = (ip->ip_hl << 2); 6111 6112 /* 6113 * WARNING! We cannot fragment if the packet was modified from an 6114 * original which expected to be using TSO. In this 6115 * situation we pray that the target interface is 6116 * compatible with the originating interface. 6117 */ 6118 if (ip->ip_len <= ifp->if_mtu || 6119 (m0->m_pkthdr.csum_flags & CSUM_TSO) || 6120 ((ifp->if_hwassist & CSUM_FRAGMENT) && 6121 (ip->ip_off & IP_DF) == 0)) { 6122 ip->ip_len = htons(ip->ip_len); 6123 ip->ip_off = htons(ip->ip_off); 6124 ip->ip_sum = 0; 6125 if (sw_csum & CSUM_DELAY_IP) { 6126 /* From KAME */ 6127 if (ip->ip_v == IPVERSION && 6128 (ip->ip_hl << 2) == sizeof(*ip)) { 6129 ip->ip_sum = in_cksum_hdr(ip); 6130 } else { 6131 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 6132 } 6133 } 6134 lwkt_reltoken(&pf_token); 6135 error = ifp->if_output(ifp, m0, sintosa(dst), ro->ro_rt); 6136 lwkt_gettoken(&pf_token); 6137 goto done; 6138 } 6139 6140 /* 6141 * Too large for interface; fragment if possible. 6142 * Must be able to put at least 8 bytes per fragment. 6143 */ 6144 if (ip->ip_off & IP_DF) { 6145 ipstat.ips_cantfrag++; 6146 if (r->rt != PF_DUPTO) { 6147 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 6148 ifp->if_mtu); 6149 goto done; 6150 } else 6151 goto bad; 6152 } 6153 6154 m1 = m0; 6155 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist, sw_csum); 6156 if (error) { 6157 goto bad; 6158 } 6159 6160 for (m0 = m1; m0; m0 = m1) { 6161 m1 = m0->m_nextpkt; 6162 m0->m_nextpkt = 0; 6163 if (error == 0) { 6164 lwkt_reltoken(&pf_token); 6165 error = (*ifp->if_output)(ifp, m0, sintosa(dst), 6166 NULL); 6167 lwkt_gettoken(&pf_token); 6168 } else 6169 m_freem(m0); 6170 } 6171 6172 if (error == 0) 6173 ipstat.ips_fragmented++; 6174 6175 done: 6176 if (r->rt != PF_DUPTO) 6177 *m = NULL; 6178 if (ro == &iproute && ro->ro_rt) 6179 RTFREE(ro->ro_rt); 6180 return; 6181 6182 bad: 6183 m_freem(m0); 6184 goto done; 6185 } 6186 #endif /* INET */ 6187 6188 #ifdef INET6 6189 void 6190 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, 6191 struct pf_state *s, struct pf_pdesc *pd) 6192 { 6193 struct mbuf *m0; 6194 struct route_in6 ip6route; 6195 struct route_in6 *ro; 6196 struct sockaddr_in6 *dst; 6197 struct ip6_hdr *ip6; 6198 struct ifnet *ifp = NULL; 6199 struct pf_addr naddr; 6200 struct pf_src_node *sn = NULL; 6201 6202 if (m == NULL || *m == NULL || r == NULL || 6203 (dir != PF_IN && dir != PF_OUT) || oifp == NULL) 6204 panic("pf_route6: invalid parameters"); 6205 6206 if (((*m)->m_pkthdr.fw_flags & PF_MBUF_ROUTED) == 0) { 6207 (*m)->m_pkthdr.fw_flags |= PF_MBUF_ROUTED; 6208 (*m)->m_pkthdr.pf.routed = 1; 6209 } else { 6210 if ((*m)->m_pkthdr.pf.routed++ > 3) { 6211 m0 = *m; 6212 *m = NULL; 6213 goto bad; 6214 } 6215 } 6216 6217 if (r->rt == PF_DUPTO) { 6218 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) 6219 return; 6220 } else { 6221 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) 6222 return; 6223 m0 = *m; 6224 } 6225 6226 if (m0->m_len < sizeof(struct ip6_hdr)) { 6227 DPFPRINTF(PF_DEBUG_URGENT, 6228 ("pf_route6: m0->m_len < sizeof(struct ip6_hdr)\n")); 6229 goto bad; 6230 } 6231 ip6 = mtod(m0, struct ip6_hdr *); 6232 6233 ro = &ip6route; 6234 bzero((caddr_t)ro, sizeof(*ro)); 6235 dst = (struct sockaddr_in6 *)&ro->ro_dst; 6236 dst->sin6_family = AF_INET6; 6237 dst->sin6_len = sizeof(*dst); 6238 dst->sin6_addr = ip6->ip6_dst; 6239 6240 /* 6241 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags, 6242 * so make sure pf.flags is clear. 6243 * 6244 * Cheat. XXX why only in the v6 case??? 6245 */ 6246 if (r->rt == PF_FASTROUTE) { 6247 m0->m_pkthdr.fw_flags |= PF_MBUF_TAGGED; 6248 m0->m_pkthdr.pf.flags = 0; 6249 /* XXX Re-Check when Upgrading to > 4.4 */ 6250 m0->m_pkthdr.pf.statekey = NULL; 6251 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL); 6252 return; 6253 } 6254 6255 if (TAILQ_EMPTY(&r->rpool.list)) { 6256 DPFPRINTF(PF_DEBUG_URGENT, 6257 ("pf_route6: TAILQ_EMPTY(&r->rpool.list)\n")); 6258 goto bad; 6259 } 6260 if (s == NULL) { 6261 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 6262 &naddr, NULL, &sn); 6263 if (!PF_AZERO(&naddr, AF_INET6)) 6264 PF_ACPY((struct pf_addr *)&dst->sin6_addr, 6265 &naddr, AF_INET6); 6266 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 6267 } else { 6268 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 6269 PF_ACPY((struct pf_addr *)&dst->sin6_addr, 6270 &s->rt_addr, AF_INET6); 6271 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6272 } 6273 if (ifp == NULL) 6274 goto bad; 6275 6276 if (oifp != ifp) { 6277 if (pf_test6(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) { 6278 goto bad; 6279 } else if (m0 == NULL) { 6280 goto done; 6281 } 6282 if (m0->m_len < sizeof(struct ip6_hdr)) { 6283 DPFPRINTF(PF_DEBUG_URGENT, 6284 ("pf_route6: m0->m_len < sizeof(struct ip6_hdr)\n")); 6285 goto bad; 6286 } 6287 ip6 = mtod(m0, struct ip6_hdr *); 6288 } 6289 6290 /* 6291 * If the packet is too large for the outgoing interface, 6292 * send back an icmp6 error. 6293 */ 6294 if (IN6_IS_ADDR_LINKLOCAL(&dst->sin6_addr)) 6295 dst->sin6_addr.s6_addr16[1] = htons(ifp->if_index); 6296 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 6297 nd6_output(ifp, ifp, m0, dst, NULL); 6298 } else { 6299 in6_ifstat_inc(ifp, ifs6_in_toobig); 6300 if (r->rt != PF_DUPTO) 6301 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 6302 else 6303 goto bad; 6304 } 6305 6306 done: 6307 if (r->rt != PF_DUPTO) 6308 *m = NULL; 6309 return; 6310 6311 bad: 6312 m_freem(m0); 6313 goto done; 6314 } 6315 #endif /* INET6 */ 6316 6317 6318 /* 6319 * check protocol (tcp/udp/icmp/icmp6) checksum and set mbuf flag 6320 * off is the offset where the protocol header starts 6321 * len is the total length of protocol header plus payload 6322 * returns 0 when the checksum is valid, otherwise returns 1. 6323 */ 6324 /* 6325 * XXX 6326 * FreeBSD supports cksum offload for the following drivers. 6327 * em(4), gx(4), lge(4), nge(4), ti(4), xl(4) 6328 * If we can make full use of it we would outperform ipfw/ipfilter in 6329 * very heavy traffic. 6330 * I have not tested 'cause I don't have NICs that supports cksum offload. 6331 * (There might be problems. Typical phenomena would be 6332 * 1. No route message for UDP packet. 6333 * 2. No connection acceptance from external hosts regardless of rule set.) 6334 */ 6335 int 6336 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, 6337 sa_family_t af) 6338 { 6339 u_int16_t sum = 0; 6340 int hw_assist = 0; 6341 struct ip *ip; 6342 6343 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 6344 return (1); 6345 if (m->m_pkthdr.len < off + len) 6346 return (1); 6347 6348 switch (p) { 6349 case IPPROTO_TCP: 6350 case IPPROTO_UDP: 6351 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 6352 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 6353 sum = m->m_pkthdr.csum_data; 6354 } else { 6355 ip = mtod(m, struct ip *); 6356 sum = in_pseudo(ip->ip_src.s_addr, 6357 ip->ip_dst.s_addr, htonl((u_short)len + 6358 m->m_pkthdr.csum_data + p)); 6359 } 6360 sum ^= 0xffff; 6361 ++hw_assist; 6362 } 6363 break; 6364 case IPPROTO_ICMP: 6365 #ifdef INET6 6366 case IPPROTO_ICMPV6: 6367 #endif /* INET6 */ 6368 break; 6369 default: 6370 return (1); 6371 } 6372 6373 if (!hw_assist) { 6374 switch (af) { 6375 case AF_INET: 6376 if (p == IPPROTO_ICMP) { 6377 if (m->m_len < off) 6378 return (1); 6379 m->m_data += off; 6380 m->m_len -= off; 6381 sum = in_cksum(m, len); 6382 m->m_data -= off; 6383 m->m_len += off; 6384 } else { 6385 if (m->m_len < sizeof(struct ip)) 6386 return (1); 6387 sum = in_cksum_range(m, p, off, len); 6388 if (sum == 0) { 6389 m->m_pkthdr.csum_flags |= 6390 (CSUM_DATA_VALID | 6391 CSUM_PSEUDO_HDR); 6392 m->m_pkthdr.csum_data = 0xffff; 6393 } 6394 } 6395 break; 6396 #ifdef INET6 6397 case AF_INET6: 6398 if (m->m_len < sizeof(struct ip6_hdr)) 6399 return (1); 6400 sum = in6_cksum(m, p, off, len); 6401 /* 6402 * XXX 6403 * IPv6 H/W cksum off-load not supported yet! 6404 * 6405 * if (sum == 0) { 6406 * m->m_pkthdr.csum_flags |= 6407 * (CSUM_DATA_VALID|CSUM_PSEUDO_HDR); 6408 * m->m_pkthdr.csum_data = 0xffff; 6409 *} 6410 */ 6411 break; 6412 #endif /* INET6 */ 6413 default: 6414 return (1); 6415 } 6416 } 6417 if (sum) { 6418 switch (p) { 6419 case IPPROTO_TCP: 6420 tcpstat.tcps_rcvbadsum++; 6421 break; 6422 case IPPROTO_UDP: 6423 udp_stat.udps_badsum++; 6424 break; 6425 case IPPROTO_ICMP: 6426 icmpstat.icps_checksum++; 6427 break; 6428 #ifdef INET6 6429 case IPPROTO_ICMPV6: 6430 icmp6stat.icp6s_checksum++; 6431 break; 6432 #endif /* INET6 */ 6433 } 6434 return (1); 6435 } 6436 return (0); 6437 } 6438 6439 struct pf_divert * 6440 pf_find_divert(struct mbuf *m) 6441 { 6442 struct m_tag *mtag; 6443 6444 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) 6445 return (NULL); 6446 6447 return ((struct pf_divert *)(mtag + 1)); 6448 } 6449 6450 struct pf_divert * 6451 pf_get_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 mtag = m_tag_get(PACKET_TAG_PF_DIVERT, sizeof(struct pf_divert), 6457 M_NOWAIT); 6458 if (mtag == NULL) 6459 return (NULL); 6460 bzero(mtag + 1, sizeof(struct pf_divert)); 6461 m_tag_prepend(m, mtag); 6462 } 6463 6464 return ((struct pf_divert *)(mtag + 1)); 6465 } 6466 6467 #ifdef INET 6468 6469 /* 6470 * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE 6471 */ 6472 int 6473 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, 6474 struct ether_header *eh, struct inpcb *inp) 6475 { 6476 struct pfi_kif *kif; 6477 u_short action, reason = 0, log = 0; 6478 struct mbuf *m = *m0; 6479 struct ip *h = NULL; 6480 struct pf_rule *a = NULL, *r = &pf_default_rule, *tr, *nr; 6481 struct pf_state *s = NULL; 6482 struct pf_ruleset *ruleset = NULL; 6483 struct pf_pdesc pd; 6484 int off, dirndx; 6485 #ifdef ALTQ 6486 int pqid = 0; 6487 #endif 6488 6489 if (!pf_status.running) 6490 return (PF_PASS); 6491 6492 memset(&pd, 0, sizeof(pd)); 6493 #ifdef foo 6494 if (ifp->if_type == IFT_CARP && ifp->if_carpdev) 6495 kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif; 6496 else 6497 #endif 6498 kif = (struct pfi_kif *)ifp->if_pf_kif; 6499 6500 if (kif == NULL) { 6501 DPFPRINTF(PF_DEBUG_URGENT, 6502 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 6503 return (PF_DROP); 6504 } 6505 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6506 return (PF_PASS); 6507 6508 #ifdef DIAGNOSTIC 6509 if ((m->m_flags & M_PKTHDR) == 0) 6510 panic("non-M_PKTHDR is passed to pf_test"); 6511 #endif /* DIAGNOSTIC */ 6512 6513 if (m->m_pkthdr.len < (int)sizeof(*h)) { 6514 action = PF_DROP; 6515 REASON_SET(&reason, PFRES_SHORT); 6516 log = 1; 6517 goto done; 6518 } 6519 6520 /* 6521 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags, 6522 * so make sure pf.flags is clear. 6523 */ 6524 if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED) 6525 return (PF_PASS); 6526 m->m_pkthdr.pf.flags = 0; 6527 /* Re-Check when updating to > 4.4 */ 6528 m->m_pkthdr.pf.statekey = NULL; 6529 6530 /* We do IP header normalization and packet reassembly here */ 6531 if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { 6532 action = PF_DROP; 6533 goto done; 6534 } 6535 m = *m0; /* pf_normalize messes with m0 */ 6536 h = mtod(m, struct ip *); 6537 6538 off = h->ip_hl << 2; 6539 if (off < (int)sizeof(*h)) { 6540 action = PF_DROP; 6541 REASON_SET(&reason, PFRES_SHORT); 6542 log = 1; 6543 goto done; 6544 } 6545 6546 pd.src = (struct pf_addr *)&h->ip_src; 6547 pd.dst = (struct pf_addr *)&h->ip_dst; 6548 pd.sport = pd.dport = NULL; 6549 pd.ip_sum = &h->ip_sum; 6550 pd.proto_sum = NULL; 6551 pd.proto = h->ip_p; 6552 pd.dir = dir; 6553 pd.sidx = (dir == PF_IN) ? 0 : 1; 6554 pd.didx = (dir == PF_IN) ? 1 : 0; 6555 pd.af = AF_INET; 6556 pd.tos = h->ip_tos; 6557 pd.tot_len = h->ip_len; 6558 pd.eh = eh; 6559 6560 /* handle fragments that didn't get reassembled by normalization */ 6561 if (h->ip_off & (IP_MF | IP_OFFMASK)) { 6562 action = pf_test_fragment(&r, dir, kif, m, h, 6563 &pd, &a, &ruleset); 6564 goto done; 6565 } 6566 6567 switch (h->ip_p) { 6568 6569 case IPPROTO_TCP: { 6570 struct tcphdr th; 6571 6572 pd.hdr.tcp = &th; 6573 if (!pf_pull_hdr(m, off, &th, sizeof(th), 6574 &action, &reason, AF_INET)) { 6575 log = action != PF_PASS; 6576 goto done; 6577 } 6578 pd.p_len = pd.tot_len - off - (th.th_off << 2); 6579 #ifdef ALTQ 6580 if ((th.th_flags & TH_ACK) && pd.p_len == 0) 6581 pqid = 1; 6582 #endif 6583 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6584 if (action == PF_DROP) 6585 goto done; 6586 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6587 &reason); 6588 if (action == PF_PASS) { 6589 r = s->rule.ptr; 6590 a = s->anchor.ptr; 6591 log = s->log; 6592 } else if (s == NULL) { 6593 action = pf_test_rule(&r, &s, dir, kif, 6594 m, off, h, &pd, &a, 6595 &ruleset, NULL, inp); 6596 } 6597 break; 6598 } 6599 6600 case IPPROTO_UDP: { 6601 struct udphdr uh; 6602 6603 pd.hdr.udp = &uh; 6604 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 6605 &action, &reason, AF_INET)) { 6606 log = action != PF_PASS; 6607 goto done; 6608 } 6609 if (uh.uh_dport == 0 || 6610 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 6611 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 6612 action = PF_DROP; 6613 REASON_SET(&reason, PFRES_SHORT); 6614 goto done; 6615 } 6616 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 6617 if (action == PF_PASS) { 6618 r = s->rule.ptr; 6619 a = s->anchor.ptr; 6620 log = s->log; 6621 } else if (s == NULL) { 6622 action = pf_test_rule(&r, &s, dir, kif, 6623 m, off, h, &pd, &a, 6624 &ruleset, NULL, inp); 6625 } 6626 break; 6627 } 6628 6629 case IPPROTO_ICMP: { 6630 struct icmp ih; 6631 6632 pd.hdr.icmp = &ih; 6633 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN, 6634 &action, &reason, AF_INET)) { 6635 log = action != PF_PASS; 6636 goto done; 6637 } 6638 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, 6639 &reason); 6640 if (action == PF_PASS) { 6641 r = s->rule.ptr; 6642 a = s->anchor.ptr; 6643 log = s->log; 6644 } else if (s == NULL) { 6645 action = pf_test_rule(&r, &s, dir, kif, 6646 m, off, h, &pd, &a, 6647 &ruleset, NULL, inp); 6648 } 6649 break; 6650 } 6651 6652 default: 6653 action = pf_test_state_other(&s, dir, kif, m, &pd); 6654 if (action == PF_PASS) { 6655 r = s->rule.ptr; 6656 a = s->anchor.ptr; 6657 log = s->log; 6658 } else if (s == NULL) { 6659 action = pf_test_rule(&r, &s, dir, kif, m, off, h, 6660 &pd, &a, &ruleset, NULL, inp); 6661 } 6662 break; 6663 } 6664 6665 done: 6666 if (action == PF_PASS && h->ip_hl > 5 && 6667 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 6668 action = PF_DROP; 6669 REASON_SET(&reason, PFRES_IPOPTIONS); 6670 log = 1; 6671 DPFPRINTF(PF_DEBUG_MISC, 6672 ("pf: dropping packet with ip options\n")); 6673 } 6674 6675 if ((s && s->tag) || r->rtableid) 6676 pf_tag_packet(m, s ? s->tag : 0, r->rtableid); 6677 6678 #if 0 6679 if (dir == PF_IN && s && s->key[PF_SK_STACK]) 6680 m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK]; 6681 #endif 6682 6683 #ifdef ALTQ 6684 /* 6685 * Generate a hash code and qid request for ALTQ. A qid of 0 6686 * is allowed and will cause altq to select the default queue. 6687 */ 6688 if (action == PF_PASS) { 6689 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE; 6690 if (pqid || (pd.tos & IPTOS_LOWDELAY)) 6691 m->m_pkthdr.pf.qid = r->pqid; 6692 else 6693 m->m_pkthdr.pf.qid = r->qid; 6694 m->m_pkthdr.pf.ecn_af = AF_INET; 6695 m->m_pkthdr.pf.hdr = h; 6696 /* add connection hash for fairq */ 6697 if (s) { 6698 /* for fairq */ 6699 m->m_pkthdr.pf.state_hash = s->hash; 6700 m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED; 6701 } 6702 } 6703 #endif /* ALTQ */ 6704 6705 /* 6706 * connections redirected to loopback should not match sockets 6707 * bound specifically to loopback due to security implications, 6708 * see tcp_input() and in_pcblookup_listen(). 6709 */ 6710 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 6711 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 6712 (s->nat_rule.ptr->action == PF_RDR || 6713 s->nat_rule.ptr->action == PF_BINAT) && 6714 (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) 6715 m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST; 6716 6717 if (dir == PF_IN && action == PF_PASS && r->divert.port) { 6718 struct pf_divert *divert; 6719 6720 if ((divert = pf_get_divert(m))) { 6721 m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED; 6722 divert->port = r->divert.port; 6723 divert->addr.ipv4 = r->divert.addr.v4; 6724 } 6725 } 6726 6727 if (log) { 6728 struct pf_rule *lr; 6729 6730 if (s != NULL && s->nat_rule.ptr != NULL && 6731 s->nat_rule.ptr->log & PF_LOG_ALL) 6732 lr = s->nat_rule.ptr; 6733 else 6734 lr = r; 6735 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, lr, a, ruleset, 6736 &pd); 6737 } 6738 6739 kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len; 6740 kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++; 6741 6742 if (action == PF_PASS || r->action == PF_DROP) { 6743 dirndx = (dir == PF_OUT); 6744 r->packets[dirndx]++; 6745 r->bytes[dirndx] += pd.tot_len; 6746 if (a != NULL) { 6747 a->packets[dirndx]++; 6748 a->bytes[dirndx] += pd.tot_len; 6749 } 6750 if (s != NULL) { 6751 if (s->nat_rule.ptr != NULL) { 6752 s->nat_rule.ptr->packets[dirndx]++; 6753 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; 6754 } 6755 if (s->src_node != NULL) { 6756 s->src_node->packets[dirndx]++; 6757 s->src_node->bytes[dirndx] += pd.tot_len; 6758 } 6759 if (s->nat_src_node != NULL) { 6760 s->nat_src_node->packets[dirndx]++; 6761 s->nat_src_node->bytes[dirndx] += pd.tot_len; 6762 } 6763 dirndx = (dir == s->direction) ? 0 : 1; 6764 s->packets[dirndx]++; 6765 s->bytes[dirndx] += pd.tot_len; 6766 } 6767 tr = r; 6768 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 6769 if (nr != NULL && r == &pf_default_rule) 6770 tr = nr; 6771 if (tr->src.addr.type == PF_ADDR_TABLE) 6772 pfr_update_stats(tr->src.addr.p.tbl, 6773 (s == NULL) ? pd.src : 6774 &s->key[(s->direction == PF_IN)]-> 6775 addr[(s->direction == PF_OUT)], 6776 pd.af, pd.tot_len, dir == PF_OUT, 6777 r->action == PF_PASS, tr->src.neg); 6778 if (tr->dst.addr.type == PF_ADDR_TABLE) 6779 pfr_update_stats(tr->dst.addr.p.tbl, 6780 (s == NULL) ? pd.dst : 6781 &s->key[(s->direction == PF_IN)]-> 6782 addr[(s->direction == PF_IN)], 6783 pd.af, pd.tot_len, dir == PF_OUT, 6784 r->action == PF_PASS, tr->dst.neg); 6785 } 6786 6787 6788 if (action == PF_SYNPROXY_DROP) { 6789 m_freem(*m0); 6790 *m0 = NULL; 6791 action = PF_PASS; 6792 } else if (r->rt) { 6793 /* pf_route can free the mbuf causing *m0 to become NULL */ 6794 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd); 6795 } 6796 6797 return (action); 6798 } 6799 #endif /* INET */ 6800 6801 #ifdef INET6 6802 6803 /* 6804 * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE 6805 */ 6806 int 6807 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, 6808 struct ether_header *eh, struct inpcb *inp) 6809 { 6810 struct pfi_kif *kif; 6811 u_short action, reason = 0, log = 0; 6812 struct mbuf *m = *m0, *n = NULL; 6813 struct ip6_hdr *h = NULL; 6814 struct pf_rule *a = NULL, *r = &pf_default_rule, *tr, *nr; 6815 struct pf_state *s = NULL; 6816 struct pf_ruleset *ruleset = NULL; 6817 struct pf_pdesc pd; 6818 int off, terminal = 0, dirndx, rh_cnt = 0; 6819 6820 if (!pf_status.running) 6821 return (PF_PASS); 6822 6823 memset(&pd, 0, sizeof(pd)); 6824 #ifdef foo 6825 if (ifp->if_type == IFT_CARP && ifp->if_carpdev) 6826 kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif; 6827 else 6828 #endif 6829 kif = (struct pfi_kif *)ifp->if_pf_kif; 6830 6831 if (kif == NULL) { 6832 DPFPRINTF(PF_DEBUG_URGENT, 6833 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 6834 return (PF_DROP); 6835 } 6836 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6837 return (PF_PASS); 6838 6839 #ifdef DIAGNOSTIC 6840 if ((m->m_flags & M_PKTHDR) == 0) 6841 panic("non-M_PKTHDR is passed to pf_test6"); 6842 #endif /* DIAGNOSTIC */ 6843 6844 if (m->m_pkthdr.len < (int)sizeof(*h)) { 6845 action = PF_DROP; 6846 REASON_SET(&reason, PFRES_SHORT); 6847 log = 1; 6848 goto done; 6849 } 6850 6851 /* 6852 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags, 6853 * so make sure pf.flags is clear. 6854 */ 6855 if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED) 6856 return (PF_PASS); 6857 m->m_pkthdr.pf.flags = 0; 6858 /* Re-Check when updating to > 4.4 */ 6859 m->m_pkthdr.pf.statekey = NULL; 6860 6861 /* We do IP header normalization and packet reassembly here */ 6862 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { 6863 action = PF_DROP; 6864 goto done; 6865 } 6866 m = *m0; /* pf_normalize messes with m0 */ 6867 h = mtod(m, struct ip6_hdr *); 6868 6869 #if 1 6870 /* 6871 * we do not support jumbogram yet. if we keep going, zero ip6_plen 6872 * will do something bad, so drop the packet for now. 6873 */ 6874 if (htons(h->ip6_plen) == 0) { 6875 action = PF_DROP; 6876 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 6877 goto done; 6878 } 6879 #endif 6880 6881 pd.src = (struct pf_addr *)&h->ip6_src; 6882 pd.dst = (struct pf_addr *)&h->ip6_dst; 6883 pd.sport = pd.dport = NULL; 6884 pd.ip_sum = NULL; 6885 pd.proto_sum = NULL; 6886 pd.dir = dir; 6887 pd.sidx = (dir == PF_IN) ? 0 : 1; 6888 pd.didx = (dir == PF_IN) ? 1 : 0; 6889 pd.af = AF_INET6; 6890 pd.tos = 0; 6891 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 6892 pd.eh = eh; 6893 6894 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 6895 pd.proto = h->ip6_nxt; 6896 do { 6897 switch (pd.proto) { 6898 case IPPROTO_FRAGMENT: 6899 action = pf_test_fragment(&r, dir, kif, m, h, 6900 &pd, &a, &ruleset); 6901 if (action == PF_DROP) 6902 REASON_SET(&reason, PFRES_FRAG); 6903 goto done; 6904 case IPPROTO_ROUTING: { 6905 struct ip6_rthdr rthdr; 6906 6907 if (rh_cnt++) { 6908 DPFPRINTF(PF_DEBUG_MISC, 6909 ("pf: IPv6 more than one rthdr\n")); 6910 action = PF_DROP; 6911 REASON_SET(&reason, PFRES_IPOPTIONS); 6912 log = 1; 6913 goto done; 6914 } 6915 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 6916 &reason, pd.af)) { 6917 DPFPRINTF(PF_DEBUG_MISC, 6918 ("pf: IPv6 short rthdr\n")); 6919 action = PF_DROP; 6920 REASON_SET(&reason, PFRES_SHORT); 6921 log = 1; 6922 goto done; 6923 } 6924 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 6925 DPFPRINTF(PF_DEBUG_MISC, 6926 ("pf: IPv6 rthdr0\n")); 6927 action = PF_DROP; 6928 REASON_SET(&reason, PFRES_IPOPTIONS); 6929 log = 1; 6930 goto done; 6931 } 6932 /* FALLTHROUGH */ 6933 } 6934 case IPPROTO_AH: 6935 case IPPROTO_HOPOPTS: 6936 case IPPROTO_DSTOPTS: { 6937 /* get next header and header length */ 6938 struct ip6_ext opt6; 6939 6940 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 6941 NULL, &reason, pd.af)) { 6942 DPFPRINTF(PF_DEBUG_MISC, 6943 ("pf: IPv6 short opt\n")); 6944 action = PF_DROP; 6945 log = 1; 6946 goto done; 6947 } 6948 if (pd.proto == IPPROTO_AH) 6949 off += (opt6.ip6e_len + 2) * 4; 6950 else 6951 off += (opt6.ip6e_len + 1) * 8; 6952 pd.proto = opt6.ip6e_nxt; 6953 /* goto the next header */ 6954 break; 6955 } 6956 default: 6957 terminal++; 6958 break; 6959 } 6960 } while (!terminal); 6961 6962 /* if there's no routing header, use unmodified mbuf for checksumming */ 6963 if (!n) 6964 n = m; 6965 6966 switch (pd.proto) { 6967 6968 case IPPROTO_TCP: { 6969 struct tcphdr th; 6970 6971 pd.hdr.tcp = &th; 6972 if (!pf_pull_hdr(m, off, &th, sizeof(th), 6973 &action, &reason, AF_INET6)) { 6974 log = action != PF_PASS; 6975 goto done; 6976 } 6977 pd.p_len = pd.tot_len - off - (th.th_off << 2); 6978 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6979 if (action == PF_DROP) 6980 goto done; 6981 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6982 &reason); 6983 if (action == PF_PASS) { 6984 r = s->rule.ptr; 6985 a = s->anchor.ptr; 6986 log = s->log; 6987 } else if (s == NULL) { 6988 action = pf_test_rule(&r, &s, dir, kif, 6989 m, off, h, &pd, &a, 6990 &ruleset, NULL, inp); 6991 } 6992 break; 6993 } 6994 6995 case IPPROTO_UDP: { 6996 struct udphdr uh; 6997 6998 pd.hdr.udp = &uh; 6999 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 7000 &action, &reason, AF_INET6)) { 7001 log = action != PF_PASS; 7002 goto done; 7003 } 7004 if (uh.uh_dport == 0 || 7005 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 7006 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 7007 action = PF_DROP; 7008 REASON_SET(&reason, PFRES_SHORT); 7009 goto done; 7010 } 7011 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 7012 if (action == PF_PASS) { 7013 r = s->rule.ptr; 7014 a = s->anchor.ptr; 7015 log = s->log; 7016 } else if (s == NULL) { 7017 action = pf_test_rule(&r, &s, dir, kif, 7018 m, off, h, &pd, &a, 7019 &ruleset, NULL, inp); 7020 } 7021 break; 7022 } 7023 7024 case IPPROTO_ICMPV6: { 7025 struct icmp6_hdr ih; 7026 7027 pd.hdr.icmp6 = &ih; 7028 if (!pf_pull_hdr(m, off, &ih, sizeof(ih), 7029 &action, &reason, AF_INET6)) { 7030 log = action != PF_PASS; 7031 goto done; 7032 } 7033 action = pf_test_state_icmp(&s, dir, kif, 7034 m, off, h, &pd, &reason); 7035 if (action == PF_PASS) { 7036 r = s->rule.ptr; 7037 a = s->anchor.ptr; 7038 log = s->log; 7039 } else if (s == NULL) { 7040 action = pf_test_rule(&r, &s, dir, kif, 7041 m, off, h, &pd, &a, 7042 &ruleset, NULL, inp); 7043 } 7044 break; 7045 } 7046 7047 default: 7048 action = pf_test_state_other(&s, dir, kif, m, &pd); 7049 if (action == PF_PASS) { 7050 r = s->rule.ptr; 7051 a = s->anchor.ptr; 7052 log = s->log; 7053 } else if (s == NULL) { 7054 action = pf_test_rule(&r, &s, dir, kif, m, off, h, 7055 &pd, &a, &ruleset, NULL, inp); 7056 } 7057 break; 7058 } 7059 7060 done: 7061 if (n != m) { 7062 m_freem(n); 7063 n = NULL; 7064 } 7065 7066 /* handle dangerous IPv6 extension headers. */ 7067 if (action == PF_PASS && rh_cnt && 7068 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 7069 action = PF_DROP; 7070 REASON_SET(&reason, PFRES_IPOPTIONS); 7071 log = 1; 7072 DPFPRINTF(PF_DEBUG_MISC, 7073 ("pf: dropping packet with dangerous v6 headers\n")); 7074 } 7075 7076 if ((s && s->tag) || r->rtableid) 7077 pf_tag_packet(m, s ? s->tag : 0, r->rtableid); 7078 7079 #if 0 7080 if (dir == PF_IN && s && s->key[PF_SK_STACK]) 7081 m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK]; 7082 #endif 7083 7084 #ifdef ALTQ 7085 /* 7086 * Generate a hash code and qid request for ALTQ. A qid of 0 7087 * is allowed and will cause altq to select the default queue. 7088 */ 7089 if (action == PF_PASS) { 7090 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE; 7091 if (pd.tos & IPTOS_LOWDELAY) 7092 m->m_pkthdr.pf.qid = r->pqid; 7093 else 7094 m->m_pkthdr.pf.qid = r->qid; 7095 m->m_pkthdr.pf.ecn_af = AF_INET6; 7096 m->m_pkthdr.pf.hdr = h; 7097 if (s) { 7098 /* for fairq */ 7099 m->m_pkthdr.pf.state_hash = s->hash; 7100 m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED; 7101 } 7102 } 7103 #endif /* ALTQ */ 7104 7105 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 7106 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 7107 (s->nat_rule.ptr->action == PF_RDR || 7108 s->nat_rule.ptr->action == PF_BINAT) && 7109 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 7110 m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST; 7111 7112 if (dir == PF_IN && action == PF_PASS && r->divert.port) { 7113 struct pf_divert *divert; 7114 7115 if ((divert = pf_get_divert(m))) { 7116 m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED; 7117 divert->port = r->divert.port; 7118 divert->addr.ipv6 = r->divert.addr.v6; 7119 } 7120 } 7121 7122 if (log) { 7123 struct pf_rule *lr; 7124 7125 if (s != NULL && s->nat_rule.ptr != NULL && 7126 s->nat_rule.ptr->log & PF_LOG_ALL) 7127 lr = s->nat_rule.ptr; 7128 else 7129 lr = r; 7130 PFLOG_PACKET(kif, h, m, AF_INET6, dir, reason, lr, a, ruleset, 7131 &pd); 7132 } 7133 7134 kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len; 7135 kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++; 7136 7137 if (action == PF_PASS || r->action == PF_DROP) { 7138 dirndx = (dir == PF_OUT); 7139 r->packets[dirndx]++; 7140 r->bytes[dirndx] += pd.tot_len; 7141 if (a != NULL) { 7142 a->packets[dirndx]++; 7143 a->bytes[dirndx] += pd.tot_len; 7144 } 7145 if (s != NULL) { 7146 if (s->nat_rule.ptr != NULL) { 7147 s->nat_rule.ptr->packets[dirndx]++; 7148 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; 7149 } 7150 if (s->src_node != NULL) { 7151 s->src_node->packets[dirndx]++; 7152 s->src_node->bytes[dirndx] += pd.tot_len; 7153 } 7154 if (s->nat_src_node != NULL) { 7155 s->nat_src_node->packets[dirndx]++; 7156 s->nat_src_node->bytes[dirndx] += pd.tot_len; 7157 } 7158 dirndx = (dir == s->direction) ? 0 : 1; 7159 s->packets[dirndx]++; 7160 s->bytes[dirndx] += pd.tot_len; 7161 } 7162 tr = r; 7163 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 7164 if (nr != NULL && r == &pf_default_rule) 7165 tr = nr; 7166 if (tr->src.addr.type == PF_ADDR_TABLE) 7167 pfr_update_stats(tr->src.addr.p.tbl, 7168 (s == NULL) ? pd.src : 7169 &s->key[(s->direction == PF_IN)]->addr[0], 7170 pd.af, pd.tot_len, dir == PF_OUT, 7171 r->action == PF_PASS, tr->src.neg); 7172 if (tr->dst.addr.type == PF_ADDR_TABLE) 7173 pfr_update_stats(tr->dst.addr.p.tbl, 7174 (s == NULL) ? pd.dst : 7175 &s->key[(s->direction == PF_IN)]->addr[1], 7176 pd.af, pd.tot_len, dir == PF_OUT, 7177 r->action == PF_PASS, tr->dst.neg); 7178 } 7179 7180 7181 if (action == PF_SYNPROXY_DROP) { 7182 m_freem(*m0); 7183 *m0 = NULL; 7184 action = PF_PASS; 7185 } else if (r->rt) 7186 /* pf_route6 can free the mbuf causing *m0 to become NULL */ 7187 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd); 7188 7189 return (action); 7190 } 7191 #endif /* INET6 */ 7192 7193 int 7194 pf_check_congestion(struct ifqueue *ifq) 7195 { 7196 return (0); 7197 } 7198