1 /*- 2 * Copyright (c) 2001 Daniel Hartmeier 3 * Copyright (c) 2002 - 2008 Henning Brauer 4 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * - Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * - Redistributions in binary form must reproduce the above 14 * copyright notice, this list of conditions and the following 15 * disclaimer in the documentation and/or other materials provided 16 * with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 21 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 22 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 24 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 25 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 26 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 28 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 * 31 * Effort sponsored in part by the Defense Advanced Research Projects 32 * Agency (DARPA) and Air Force Research Laboratory, Air Force 33 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 34 * 35 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 36 */ 37 38 #include <sys/cdefs.h> 39 __FBSDID("$FreeBSD$"); 40 41 #include "opt_inet.h" 42 #include "opt_inet6.h" 43 #include "opt_bpf.h" 44 #include "opt_pf.h" 45 46 #include <sys/param.h> 47 #include <sys/bus.h> 48 #include <sys/endian.h> 49 #include <sys/hash.h> 50 #include <sys/interrupt.h> 51 #include <sys/kernel.h> 52 #include <sys/kthread.h> 53 #include <sys/limits.h> 54 #include <sys/mbuf.h> 55 #include <sys/md5.h> 56 #include <sys/random.h> 57 #include <sys/refcount.h> 58 #include <sys/socket.h> 59 #include <sys/sysctl.h> 60 #include <sys/taskqueue.h> 61 #include <sys/ucred.h> 62 63 #include <net/if.h> 64 #include <net/if_var.h> 65 #include <net/if_types.h> 66 #include <net/route.h> 67 #include <net/radix_mpath.h> 68 #include <net/vnet.h> 69 70 #include <net/pfvar.h> 71 #include <net/if_pflog.h> 72 #include <net/if_pfsync.h> 73 74 #include <netinet/in_pcb.h> 75 #include <netinet/in_var.h> 76 #include <netinet/ip.h> 77 #include <netinet/ip_fw.h> 78 #include <netinet/ip_icmp.h> 79 #include <netinet/icmp_var.h> 80 #include <netinet/ip_var.h> 81 #include <netinet/tcp.h> 82 #include <netinet/tcp_fsm.h> 83 #include <netinet/tcp_seq.h> 84 #include <netinet/tcp_timer.h> 85 #include <netinet/tcp_var.h> 86 #include <netinet/udp.h> 87 #include <netinet/udp_var.h> 88 89 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */ 90 91 #ifdef INET6 92 #include <netinet/ip6.h> 93 #include <netinet/icmp6.h> 94 #include <netinet6/nd6.h> 95 #include <netinet6/ip6_var.h> 96 #include <netinet6/in6_pcb.h> 97 #endif /* INET6 */ 98 99 #include <machine/in_cksum.h> 100 #include <security/mac/mac_framework.h> 101 102 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 103 104 /* 105 * Global variables 106 */ 107 108 /* state tables */ 109 VNET_DEFINE(struct pf_altqqueue, pf_altqs[2]); 110 VNET_DEFINE(struct pf_palist, pf_pabuf); 111 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 112 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 113 VNET_DEFINE(struct pf_kstatus, pf_status); 114 115 VNET_DEFINE(u_int32_t, ticket_altqs_active); 116 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 117 VNET_DEFINE(int, altqs_inactive_open); 118 VNET_DEFINE(u_int32_t, ticket_pabuf); 119 120 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 121 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 122 VNET_DEFINE(u_char, pf_tcp_secret[16]); 123 #define V_pf_tcp_secret VNET(pf_tcp_secret) 124 VNET_DEFINE(int, pf_tcp_secret_init); 125 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 126 VNET_DEFINE(int, pf_tcp_iss_off); 127 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 128 129 /* 130 * Queue for pf_intr() sends. 131 */ 132 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 133 struct pf_send_entry { 134 STAILQ_ENTRY(pf_send_entry) pfse_next; 135 struct mbuf *pfse_m; 136 enum { 137 PFSE_IP, 138 PFSE_IP6, 139 PFSE_ICMP, 140 PFSE_ICMP6, 141 } pfse_type; 142 struct { 143 int type; 144 int code; 145 int mtu; 146 } icmpopts; 147 }; 148 149 STAILQ_HEAD(pf_send_head, pf_send_entry); 150 static VNET_DEFINE(struct pf_send_head, pf_sendqueue); 151 #define V_pf_sendqueue VNET(pf_sendqueue) 152 153 static struct mtx pf_sendqueue_mtx; 154 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 155 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 156 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 157 158 /* 159 * Queue for pf_overload_task() tasks. 160 */ 161 struct pf_overload_entry { 162 SLIST_ENTRY(pf_overload_entry) next; 163 struct pf_addr addr; 164 sa_family_t af; 165 uint8_t dir; 166 struct pf_rule *rule; 167 }; 168 169 SLIST_HEAD(pf_overload_head, pf_overload_entry); 170 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue); 171 #define V_pf_overloadqueue VNET(pf_overloadqueue) 172 static VNET_DEFINE(struct task, pf_overloadtask); 173 #define V_pf_overloadtask VNET(pf_overloadtask) 174 175 static struct mtx pf_overloadqueue_mtx; 176 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 177 "pf overload/flush queue", MTX_DEF); 178 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 179 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 180 181 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules); 182 struct mtx pf_unlnkdrules_mtx; 183 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 184 MTX_DEF); 185 186 static VNET_DEFINE(uma_zone_t, pf_sources_z); 187 #define V_pf_sources_z VNET(pf_sources_z) 188 uma_zone_t pf_mtag_z; 189 VNET_DEFINE(uma_zone_t, pf_state_z); 190 VNET_DEFINE(uma_zone_t, pf_state_key_z); 191 192 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]); 193 #define PFID_CPUBITS 8 194 #define PFID_CPUSHIFT (sizeof(uint64_t) * NBBY - PFID_CPUBITS) 195 #define PFID_CPUMASK ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT) 196 #define PFID_MAXID (~PFID_CPUMASK) 197 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU); 198 199 static void pf_src_tree_remove_state(struct pf_state *); 200 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 201 u_int32_t); 202 static void pf_add_threshold(struct pf_threshold *); 203 static int pf_check_threshold(struct pf_threshold *); 204 205 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 206 u_int16_t *, u_int16_t *, struct pf_addr *, 207 u_int16_t, u_int8_t, sa_family_t); 208 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 209 struct tcphdr *, struct pf_state_peer *); 210 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 211 struct pf_addr *, struct pf_addr *, u_int16_t, 212 u_int16_t *, u_int16_t *, u_int16_t *, 213 u_int16_t *, u_int8_t, sa_family_t); 214 static void pf_send_tcp(struct mbuf *, 215 const struct pf_rule *, sa_family_t, 216 const struct pf_addr *, const struct pf_addr *, 217 u_int16_t, u_int16_t, u_int32_t, u_int32_t, 218 u_int8_t, u_int16_t, u_int16_t, u_int8_t, int, 219 u_int16_t, struct ifnet *); 220 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 221 sa_family_t, struct pf_rule *); 222 static void pf_detach_state(struct pf_state *); 223 static int pf_state_key_attach(struct pf_state_key *, 224 struct pf_state_key *, struct pf_state *); 225 static void pf_state_key_detach(struct pf_state *, int); 226 static int pf_state_key_ctor(void *, int, void *, int); 227 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 228 static int pf_test_rule(struct pf_rule **, struct pf_state **, 229 int, struct pfi_kif *, struct mbuf *, int, 230 struct pf_pdesc *, struct pf_rule **, 231 struct pf_ruleset **, struct inpcb *); 232 static int pf_create_state(struct pf_rule *, struct pf_rule *, 233 struct pf_rule *, struct pf_pdesc *, 234 struct pf_src_node *, struct pf_state_key *, 235 struct pf_state_key *, struct mbuf *, int, 236 u_int16_t, u_int16_t, int *, struct pfi_kif *, 237 struct pf_state **, int, u_int16_t, u_int16_t, 238 int); 239 static int pf_test_fragment(struct pf_rule **, int, 240 struct pfi_kif *, struct mbuf *, void *, 241 struct pf_pdesc *, struct pf_rule **, 242 struct pf_ruleset **); 243 static int pf_tcp_track_full(struct pf_state_peer *, 244 struct pf_state_peer *, struct pf_state **, 245 struct pfi_kif *, struct mbuf *, int, 246 struct pf_pdesc *, u_short *, int *); 247 static int pf_tcp_track_sloppy(struct pf_state_peer *, 248 struct pf_state_peer *, struct pf_state **, 249 struct pf_pdesc *, u_short *); 250 static int pf_test_state_tcp(struct pf_state **, int, 251 struct pfi_kif *, struct mbuf *, int, 252 void *, struct pf_pdesc *, u_short *); 253 static int pf_test_state_udp(struct pf_state **, int, 254 struct pfi_kif *, struct mbuf *, int, 255 void *, struct pf_pdesc *); 256 static int pf_test_state_icmp(struct pf_state **, int, 257 struct pfi_kif *, struct mbuf *, int, 258 void *, struct pf_pdesc *, u_short *); 259 static int pf_test_state_other(struct pf_state **, int, 260 struct pfi_kif *, struct mbuf *, struct pf_pdesc *); 261 static u_int8_t pf_get_wscale(struct mbuf *, int, u_int16_t, 262 sa_family_t); 263 static u_int16_t pf_get_mss(struct mbuf *, int, u_int16_t, 264 sa_family_t); 265 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 266 int, u_int16_t); 267 static int pf_check_proto_cksum(struct mbuf *, int, int, 268 u_int8_t, sa_family_t); 269 static void pf_print_state_parts(struct pf_state *, 270 struct pf_state_key *, struct pf_state_key *); 271 static int pf_addr_wrap_neq(struct pf_addr_wrap *, 272 struct pf_addr_wrap *); 273 static struct pf_state *pf_find_state(struct pfi_kif *, 274 struct pf_state_key_cmp *, u_int); 275 static int pf_src_connlimit(struct pf_state **); 276 static void pf_overload_task(void *v, int pending); 277 static int pf_insert_src_node(struct pf_src_node **, 278 struct pf_rule *, struct pf_addr *, sa_family_t); 279 static u_int pf_purge_expired_states(u_int, int); 280 static void pf_purge_unlinked_rules(void); 281 static int pf_mtag_uminit(void *, int, int); 282 static void pf_mtag_free(struct m_tag *); 283 #ifdef INET 284 static void pf_route(struct mbuf **, struct pf_rule *, int, 285 struct ifnet *, struct pf_state *, 286 struct pf_pdesc *); 287 #endif /* INET */ 288 #ifdef INET6 289 static void pf_change_a6(struct pf_addr *, u_int16_t *, 290 struct pf_addr *, u_int8_t); 291 static void pf_route6(struct mbuf **, struct pf_rule *, int, 292 struct ifnet *, struct pf_state *, 293 struct pf_pdesc *); 294 #endif /* INET6 */ 295 296 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 297 298 VNET_DECLARE(int, pf_end_threads); 299 300 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 301 302 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 303 (pd)->pf_mtag->flags & PF_PACKET_LOOPED) 304 305 #define STATE_LOOKUP(i, k, d, s, pd) \ 306 do { \ 307 (s) = pf_find_state((i), (k), (d)); \ 308 if ((s) == NULL) \ 309 return (PF_DROP); \ 310 if (PACKET_LOOPED(pd)) \ 311 return (PF_PASS); \ 312 if ((d) == PF_OUT && \ 313 (((s)->rule.ptr->rt == PF_ROUTETO && \ 314 (s)->rule.ptr->direction == PF_OUT) || \ 315 ((s)->rule.ptr->rt == PF_REPLYTO && \ 316 (s)->rule.ptr->direction == PF_IN)) && \ 317 (s)->rt_kif != NULL && \ 318 (s)->rt_kif != (i)) \ 319 return (PF_PASS); \ 320 } while (0) 321 322 #define BOUND_IFACE(r, k) \ 323 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all 324 325 #define STATE_INC_COUNTERS(s) \ 326 do { \ 327 counter_u64_add(s->rule.ptr->states_cur, 1); \ 328 counter_u64_add(s->rule.ptr->states_tot, 1); \ 329 if (s->anchor.ptr != NULL) { \ 330 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 331 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 332 } \ 333 if (s->nat_rule.ptr != NULL) { \ 334 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 335 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 336 } \ 337 } while (0) 338 339 #define STATE_DEC_COUNTERS(s) \ 340 do { \ 341 if (s->nat_rule.ptr != NULL) \ 342 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 343 if (s->anchor.ptr != NULL) \ 344 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 345 counter_u64_add(s->rule.ptr->states_cur, -1); \ 346 } while (0) 347 348 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 349 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 350 VNET_DEFINE(struct pf_idhash *, pf_idhash); 351 VNET_DEFINE(struct pf_srchash *, pf_srchash); 352 353 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)"); 354 355 u_long pf_hashmask; 356 u_long pf_srchashmask; 357 static u_long pf_hashsize; 358 static u_long pf_srchashsize; 359 360 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 361 &pf_hashsize, 0, "Size of pf(4) states hashtable"); 362 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 363 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); 364 365 VNET_DEFINE(void *, pf_swi_cookie); 366 367 VNET_DEFINE(uint32_t, pf_hashseed); 368 #define V_pf_hashseed VNET(pf_hashseed) 369 370 int 371 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 372 { 373 374 switch (af) { 375 #ifdef INET 376 case AF_INET: 377 if (a->addr32[0] > b->addr32[0]) 378 return (1); 379 if (a->addr32[0] < b->addr32[0]) 380 return (-1); 381 break; 382 #endif /* INET */ 383 #ifdef INET6 384 case AF_INET6: 385 if (a->addr32[3] > b->addr32[3]) 386 return (1); 387 if (a->addr32[3] < b->addr32[3]) 388 return (-1); 389 if (a->addr32[2] > b->addr32[2]) 390 return (1); 391 if (a->addr32[2] < b->addr32[2]) 392 return (-1); 393 if (a->addr32[1] > b->addr32[1]) 394 return (1); 395 if (a->addr32[1] < b->addr32[1]) 396 return (-1); 397 if (a->addr32[0] > b->addr32[0]) 398 return (1); 399 if (a->addr32[0] < b->addr32[0]) 400 return (-1); 401 break; 402 #endif /* INET6 */ 403 default: 404 panic("%s: unknown address family %u", __func__, af); 405 } 406 return (0); 407 } 408 409 static __inline uint32_t 410 pf_hashkey(struct pf_state_key *sk) 411 { 412 uint32_t h; 413 414 h = murmur3_32_hash32((uint32_t *)sk, 415 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 416 V_pf_hashseed); 417 418 return (h & pf_hashmask); 419 } 420 421 static __inline uint32_t 422 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 423 { 424 uint32_t h; 425 426 switch (af) { 427 case AF_INET: 428 h = murmur3_32_hash32((uint32_t *)&addr->v4, 429 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 430 break; 431 case AF_INET6: 432 h = murmur3_32_hash32((uint32_t *)&addr->v6, 433 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 434 break; 435 default: 436 panic("%s: unknown address family %u", __func__, af); 437 } 438 439 return (h & pf_srchashmask); 440 } 441 442 #ifdef ALTQ 443 static int 444 pf_state_hash(struct pf_state *s) 445 { 446 u_int32_t hv = (intptr_t)s / sizeof(*s); 447 448 hv ^= crc32(&s->src, sizeof(s->src)); 449 hv ^= crc32(&s->dst, sizeof(s->dst)); 450 if (hv == 0) 451 hv = 1; 452 return (hv); 453 } 454 #endif 455 456 #ifdef INET6 457 void 458 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 459 { 460 switch (af) { 461 #ifdef INET 462 case AF_INET: 463 dst->addr32[0] = src->addr32[0]; 464 break; 465 #endif /* INET */ 466 case AF_INET6: 467 dst->addr32[0] = src->addr32[0]; 468 dst->addr32[1] = src->addr32[1]; 469 dst->addr32[2] = src->addr32[2]; 470 dst->addr32[3] = src->addr32[3]; 471 break; 472 } 473 } 474 #endif /* INET6 */ 475 476 static void 477 pf_init_threshold(struct pf_threshold *threshold, 478 u_int32_t limit, u_int32_t seconds) 479 { 480 threshold->limit = limit * PF_THRESHOLD_MULT; 481 threshold->seconds = seconds; 482 threshold->count = 0; 483 threshold->last = time_uptime; 484 } 485 486 static void 487 pf_add_threshold(struct pf_threshold *threshold) 488 { 489 u_int32_t t = time_uptime, diff = t - threshold->last; 490 491 if (diff >= threshold->seconds) 492 threshold->count = 0; 493 else 494 threshold->count -= threshold->count * diff / 495 threshold->seconds; 496 threshold->count += PF_THRESHOLD_MULT; 497 threshold->last = t; 498 } 499 500 static int 501 pf_check_threshold(struct pf_threshold *threshold) 502 { 503 return (threshold->count > threshold->limit); 504 } 505 506 static int 507 pf_src_connlimit(struct pf_state **state) 508 { 509 struct pf_overload_entry *pfoe; 510 int bad = 0; 511 512 PF_STATE_LOCK_ASSERT(*state); 513 514 (*state)->src_node->conn++; 515 (*state)->src.tcp_est = 1; 516 pf_add_threshold(&(*state)->src_node->conn_rate); 517 518 if ((*state)->rule.ptr->max_src_conn && 519 (*state)->rule.ptr->max_src_conn < 520 (*state)->src_node->conn) { 521 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 522 bad++; 523 } 524 525 if ((*state)->rule.ptr->max_src_conn_rate.limit && 526 pf_check_threshold(&(*state)->src_node->conn_rate)) { 527 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 528 bad++; 529 } 530 531 if (!bad) 532 return (0); 533 534 /* Kill this state. */ 535 (*state)->timeout = PFTM_PURGE; 536 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; 537 538 if ((*state)->rule.ptr->overload_tbl == NULL) 539 return (1); 540 541 /* Schedule overloading and flushing task. */ 542 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 543 if (pfoe == NULL) 544 return (1); /* too bad :( */ 545 546 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 547 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 548 pfoe->rule = (*state)->rule.ptr; 549 pfoe->dir = (*state)->direction; 550 PF_OVERLOADQ_LOCK(); 551 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 552 PF_OVERLOADQ_UNLOCK(); 553 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 554 555 return (1); 556 } 557 558 static void 559 pf_overload_task(void *v, int pending) 560 { 561 struct pf_overload_head queue; 562 struct pfr_addr p; 563 struct pf_overload_entry *pfoe, *pfoe1; 564 uint32_t killed = 0; 565 566 CURVNET_SET((struct vnet *)v); 567 568 PF_OVERLOADQ_LOCK(); 569 queue = V_pf_overloadqueue; 570 SLIST_INIT(&V_pf_overloadqueue); 571 PF_OVERLOADQ_UNLOCK(); 572 573 bzero(&p, sizeof(p)); 574 SLIST_FOREACH(pfoe, &queue, next) { 575 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 576 if (V_pf_status.debug >= PF_DEBUG_MISC) { 577 printf("%s: blocking address ", __func__); 578 pf_print_host(&pfoe->addr, 0, pfoe->af); 579 printf("\n"); 580 } 581 582 p.pfra_af = pfoe->af; 583 switch (pfoe->af) { 584 #ifdef INET 585 case AF_INET: 586 p.pfra_net = 32; 587 p.pfra_ip4addr = pfoe->addr.v4; 588 break; 589 #endif 590 #ifdef INET6 591 case AF_INET6: 592 p.pfra_net = 128; 593 p.pfra_ip6addr = pfoe->addr.v6; 594 break; 595 #endif 596 } 597 598 PF_RULES_WLOCK(); 599 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 600 PF_RULES_WUNLOCK(); 601 } 602 603 /* 604 * Remove those entries, that don't need flushing. 605 */ 606 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 607 if (pfoe->rule->flush == 0) { 608 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 609 free(pfoe, M_PFTEMP); 610 } else 611 counter_u64_add( 612 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 613 614 /* If nothing to flush, return. */ 615 if (SLIST_EMPTY(&queue)) { 616 CURVNET_RESTORE(); 617 return; 618 } 619 620 for (int i = 0; i <= pf_hashmask; i++) { 621 struct pf_idhash *ih = &V_pf_idhash[i]; 622 struct pf_state_key *sk; 623 struct pf_state *s; 624 625 PF_HASHROW_LOCK(ih); 626 LIST_FOREACH(s, &ih->states, entry) { 627 sk = s->key[PF_SK_WIRE]; 628 SLIST_FOREACH(pfoe, &queue, next) 629 if (sk->af == pfoe->af && 630 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 631 pfoe->rule == s->rule.ptr) && 632 ((pfoe->dir == PF_OUT && 633 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 634 (pfoe->dir == PF_IN && 635 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 636 s->timeout = PFTM_PURGE; 637 s->src.state = s->dst.state = TCPS_CLOSED; 638 killed++; 639 } 640 } 641 PF_HASHROW_UNLOCK(ih); 642 } 643 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 644 free(pfoe, M_PFTEMP); 645 if (V_pf_status.debug >= PF_DEBUG_MISC) 646 printf("%s: %u states killed", __func__, killed); 647 648 CURVNET_RESTORE(); 649 } 650 651 /* 652 * Can return locked on failure, so that we can consistently 653 * allocate and insert a new one. 654 */ 655 struct pf_src_node * 656 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af, 657 int returnlocked) 658 { 659 struct pf_srchash *sh; 660 struct pf_src_node *n; 661 662 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 663 664 sh = &V_pf_srchash[pf_hashsrc(src, af)]; 665 PF_HASHROW_LOCK(sh); 666 LIST_FOREACH(n, &sh->nodes, entry) 667 if (n->rule.ptr == rule && n->af == af && 668 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 669 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 670 break; 671 if (n != NULL) { 672 n->states++; 673 PF_HASHROW_UNLOCK(sh); 674 } else if (returnlocked == 0) 675 PF_HASHROW_UNLOCK(sh); 676 677 return (n); 678 } 679 680 static int 681 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule, 682 struct pf_addr *src, sa_family_t af) 683 { 684 685 KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK || 686 rule->rpool.opts & PF_POOL_STICKYADDR), 687 ("%s for non-tracking rule %p", __func__, rule)); 688 689 if (*sn == NULL) 690 *sn = pf_find_src_node(src, rule, af, 1); 691 692 if (*sn == NULL) { 693 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 694 695 PF_HASHROW_ASSERT(sh); 696 697 if (!rule->max_src_nodes || 698 counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes) 699 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 700 else 701 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 702 1); 703 if ((*sn) == NULL) { 704 PF_HASHROW_UNLOCK(sh); 705 return (-1); 706 } 707 708 pf_init_threshold(&(*sn)->conn_rate, 709 rule->max_src_conn_rate.limit, 710 rule->max_src_conn_rate.seconds); 711 712 (*sn)->af = af; 713 (*sn)->rule.ptr = rule; 714 PF_ACPY(&(*sn)->addr, src, af); 715 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 716 (*sn)->creation = time_uptime; 717 (*sn)->ruletype = rule->action; 718 (*sn)->states = 1; 719 if ((*sn)->rule.ptr != NULL) 720 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 721 PF_HASHROW_UNLOCK(sh); 722 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 723 } else { 724 if (rule->max_src_states && 725 (*sn)->states >= rule->max_src_states) { 726 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 727 1); 728 return (-1); 729 } 730 } 731 return (0); 732 } 733 734 void 735 pf_unlink_src_node(struct pf_src_node *src) 736 { 737 738 PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]); 739 LIST_REMOVE(src, entry); 740 if (src->rule.ptr) 741 counter_u64_add(src->rule.ptr->src_nodes, -1); 742 } 743 744 u_int 745 pf_free_src_nodes(struct pf_src_node_list *head) 746 { 747 struct pf_src_node *sn, *tmp; 748 u_int count = 0; 749 750 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 751 uma_zfree(V_pf_sources_z, sn); 752 count++; 753 } 754 755 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 756 757 return (count); 758 } 759 760 void 761 pf_mtag_initialize() 762 { 763 764 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 765 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 766 UMA_ALIGN_PTR, 0); 767 } 768 769 /* Per-vnet data storage structures initialization. */ 770 void 771 pf_initialize() 772 { 773 struct pf_keyhash *kh; 774 struct pf_idhash *ih; 775 struct pf_srchash *sh; 776 u_int i; 777 778 if (pf_hashsize == 0 || !powerof2(pf_hashsize)) 779 pf_hashsize = PF_HASHSIZ; 780 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) 781 pf_srchashsize = PF_HASHSIZ / 4; 782 783 V_pf_hashseed = arc4random(); 784 785 /* States and state keys storage. */ 786 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state), 787 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 788 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 789 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 790 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 791 792 V_pf_state_key_z = uma_zcreate("pf state keys", 793 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 794 UMA_ALIGN_PTR, 0); 795 V_pf_keyhash = malloc(pf_hashsize * sizeof(struct pf_keyhash), 796 M_PFHASH, M_WAITOK | M_ZERO); 797 V_pf_idhash = malloc(pf_hashsize * sizeof(struct pf_idhash), 798 M_PFHASH, M_WAITOK | M_ZERO); 799 pf_hashmask = pf_hashsize - 1; 800 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 801 i++, kh++, ih++) { 802 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 803 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 804 } 805 806 /* Source nodes. */ 807 V_pf_sources_z = uma_zcreate("pf source nodes", 808 sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 809 0); 810 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 811 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 812 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 813 V_pf_srchash = malloc(pf_srchashsize * sizeof(struct pf_srchash), 814 M_PFHASH, M_WAITOK|M_ZERO); 815 pf_srchashmask = pf_srchashsize - 1; 816 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) 817 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 818 819 /* ALTQ */ 820 TAILQ_INIT(&V_pf_altqs[0]); 821 TAILQ_INIT(&V_pf_altqs[1]); 822 TAILQ_INIT(&V_pf_pabuf); 823 V_pf_altqs_active = &V_pf_altqs[0]; 824 V_pf_altqs_inactive = &V_pf_altqs[1]; 825 826 /* Send & overload+flush queues. */ 827 STAILQ_INIT(&V_pf_sendqueue); 828 SLIST_INIT(&V_pf_overloadqueue); 829 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 830 831 /* Unlinked, but may be referenced rules. */ 832 TAILQ_INIT(&V_pf_unlinked_rules); 833 } 834 835 void 836 pf_mtag_cleanup() 837 { 838 839 uma_zdestroy(pf_mtag_z); 840 } 841 842 void 843 pf_cleanup() 844 { 845 struct pf_keyhash *kh; 846 struct pf_idhash *ih; 847 struct pf_srchash *sh; 848 struct pf_send_entry *pfse, *next; 849 u_int i; 850 851 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 852 i++, kh++, ih++) { 853 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 854 __func__)); 855 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 856 __func__)); 857 mtx_destroy(&kh->lock); 858 mtx_destroy(&ih->lock); 859 } 860 free(V_pf_keyhash, M_PFHASH); 861 free(V_pf_idhash, M_PFHASH); 862 863 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 864 KASSERT(LIST_EMPTY(&sh->nodes), 865 ("%s: source node hash not empty", __func__)); 866 mtx_destroy(&sh->lock); 867 } 868 free(V_pf_srchash, M_PFHASH); 869 870 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 871 m_freem(pfse->pfse_m); 872 free(pfse, M_PFTEMP); 873 } 874 875 uma_zdestroy(V_pf_sources_z); 876 uma_zdestroy(V_pf_state_z); 877 uma_zdestroy(V_pf_state_key_z); 878 } 879 880 static int 881 pf_mtag_uminit(void *mem, int size, int how) 882 { 883 struct m_tag *t; 884 885 t = (struct m_tag *)mem; 886 t->m_tag_cookie = MTAG_ABI_COMPAT; 887 t->m_tag_id = PACKET_TAG_PF; 888 t->m_tag_len = sizeof(struct pf_mtag); 889 t->m_tag_free = pf_mtag_free; 890 891 return (0); 892 } 893 894 static void 895 pf_mtag_free(struct m_tag *t) 896 { 897 898 uma_zfree(pf_mtag_z, t); 899 } 900 901 struct pf_mtag * 902 pf_get_mtag(struct mbuf *m) 903 { 904 struct m_tag *mtag; 905 906 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 907 return ((struct pf_mtag *)(mtag + 1)); 908 909 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 910 if (mtag == NULL) 911 return (NULL); 912 bzero(mtag + 1, sizeof(struct pf_mtag)); 913 m_tag_prepend(m, mtag); 914 915 return ((struct pf_mtag *)(mtag + 1)); 916 } 917 918 static int 919 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 920 struct pf_state *s) 921 { 922 struct pf_keyhash *khs, *khw, *kh; 923 struct pf_state_key *sk, *cur; 924 struct pf_state *si, *olds = NULL; 925 int idx; 926 927 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 928 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 929 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 930 931 /* 932 * We need to lock hash slots of both keys. To avoid deadlock 933 * we always lock the slot with lower address first. Unlock order 934 * isn't important. 935 * 936 * We also need to lock ID hash slot before dropping key 937 * locks. On success we return with ID hash slot locked. 938 */ 939 940 if (skw == sks) { 941 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 942 PF_HASHROW_LOCK(khs); 943 } else { 944 khs = &V_pf_keyhash[pf_hashkey(sks)]; 945 khw = &V_pf_keyhash[pf_hashkey(skw)]; 946 if (khs == khw) { 947 PF_HASHROW_LOCK(khs); 948 } else if (khs < khw) { 949 PF_HASHROW_LOCK(khs); 950 PF_HASHROW_LOCK(khw); 951 } else { 952 PF_HASHROW_LOCK(khw); 953 PF_HASHROW_LOCK(khs); 954 } 955 } 956 957 #define KEYS_UNLOCK() do { \ 958 if (khs != khw) { \ 959 PF_HASHROW_UNLOCK(khs); \ 960 PF_HASHROW_UNLOCK(khw); \ 961 } else \ 962 PF_HASHROW_UNLOCK(khs); \ 963 } while (0) 964 965 /* 966 * First run: start with wire key. 967 */ 968 sk = skw; 969 kh = khw; 970 idx = PF_SK_WIRE; 971 972 keyattach: 973 LIST_FOREACH(cur, &kh->keys, entry) 974 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 975 break; 976 977 if (cur != NULL) { 978 /* Key exists. Check for same kif, if none, add to key. */ 979 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 980 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 981 982 PF_HASHROW_LOCK(ih); 983 if (si->kif == s->kif && 984 si->direction == s->direction) { 985 if (sk->proto == IPPROTO_TCP && 986 si->src.state >= TCPS_FIN_WAIT_2 && 987 si->dst.state >= TCPS_FIN_WAIT_2) { 988 /* 989 * New state matches an old >FIN_WAIT_2 990 * state. We can't drop key hash locks, 991 * thus we can't unlink it properly. 992 * 993 * As a workaround we drop it into 994 * TCPS_CLOSED state, schedule purge 995 * ASAP and push it into the very end 996 * of the slot TAILQ, so that it won't 997 * conflict with our new state. 998 */ 999 si->src.state = si->dst.state = 1000 TCPS_CLOSED; 1001 si->timeout = PFTM_PURGE; 1002 olds = si; 1003 } else { 1004 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1005 printf("pf: %s key attach " 1006 "failed on %s: ", 1007 (idx == PF_SK_WIRE) ? 1008 "wire" : "stack", 1009 s->kif->pfik_name); 1010 pf_print_state_parts(s, 1011 (idx == PF_SK_WIRE) ? 1012 sk : NULL, 1013 (idx == PF_SK_STACK) ? 1014 sk : NULL); 1015 printf(", existing: "); 1016 pf_print_state_parts(si, 1017 (idx == PF_SK_WIRE) ? 1018 sk : NULL, 1019 (idx == PF_SK_STACK) ? 1020 sk : NULL); 1021 printf("\n"); 1022 } 1023 PF_HASHROW_UNLOCK(ih); 1024 KEYS_UNLOCK(); 1025 uma_zfree(V_pf_state_key_z, sk); 1026 if (idx == PF_SK_STACK) 1027 pf_detach_state(s); 1028 return (EEXIST); /* collision! */ 1029 } 1030 } 1031 PF_HASHROW_UNLOCK(ih); 1032 } 1033 uma_zfree(V_pf_state_key_z, sk); 1034 s->key[idx] = cur; 1035 } else { 1036 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1037 s->key[idx] = sk; 1038 } 1039 1040 stateattach: 1041 /* List is sorted, if-bound states before floating. */ 1042 if (s->kif == V_pfi_all) 1043 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1044 else 1045 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1046 1047 if (olds) { 1048 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1049 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1050 key_list[idx]); 1051 olds = NULL; 1052 } 1053 1054 /* 1055 * Attach done. See how should we (or should not?) 1056 * attach a second key. 1057 */ 1058 if (sks == skw) { 1059 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1060 idx = PF_SK_STACK; 1061 sks = NULL; 1062 goto stateattach; 1063 } else if (sks != NULL) { 1064 /* 1065 * Continue attaching with stack key. 1066 */ 1067 sk = sks; 1068 kh = khs; 1069 idx = PF_SK_STACK; 1070 sks = NULL; 1071 goto keyattach; 1072 } 1073 1074 PF_STATE_LOCK(s); 1075 KEYS_UNLOCK(); 1076 1077 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1078 ("%s failure", __func__)); 1079 1080 return (0); 1081 #undef KEYS_UNLOCK 1082 } 1083 1084 static void 1085 pf_detach_state(struct pf_state *s) 1086 { 1087 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1088 struct pf_keyhash *kh; 1089 1090 if (sks != NULL) { 1091 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1092 PF_HASHROW_LOCK(kh); 1093 if (s->key[PF_SK_STACK] != NULL) 1094 pf_state_key_detach(s, PF_SK_STACK); 1095 /* 1096 * If both point to same key, then we are done. 1097 */ 1098 if (sks == s->key[PF_SK_WIRE]) { 1099 pf_state_key_detach(s, PF_SK_WIRE); 1100 PF_HASHROW_UNLOCK(kh); 1101 return; 1102 } 1103 PF_HASHROW_UNLOCK(kh); 1104 } 1105 1106 if (s->key[PF_SK_WIRE] != NULL) { 1107 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1108 PF_HASHROW_LOCK(kh); 1109 if (s->key[PF_SK_WIRE] != NULL) 1110 pf_state_key_detach(s, PF_SK_WIRE); 1111 PF_HASHROW_UNLOCK(kh); 1112 } 1113 } 1114 1115 static void 1116 pf_state_key_detach(struct pf_state *s, int idx) 1117 { 1118 struct pf_state_key *sk = s->key[idx]; 1119 #ifdef INVARIANTS 1120 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1121 1122 PF_HASHROW_ASSERT(kh); 1123 #endif 1124 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1125 s->key[idx] = NULL; 1126 1127 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1128 LIST_REMOVE(sk, entry); 1129 uma_zfree(V_pf_state_key_z, sk); 1130 } 1131 } 1132 1133 static int 1134 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1135 { 1136 struct pf_state_key *sk = mem; 1137 1138 bzero(sk, sizeof(struct pf_state_key_cmp)); 1139 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1140 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1141 1142 return (0); 1143 } 1144 1145 struct pf_state_key * 1146 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1147 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1148 { 1149 struct pf_state_key *sk; 1150 1151 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1152 if (sk == NULL) 1153 return (NULL); 1154 1155 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1156 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1157 sk->port[pd->sidx] = sport; 1158 sk->port[pd->didx] = dport; 1159 sk->proto = pd->proto; 1160 sk->af = pd->af; 1161 1162 return (sk); 1163 } 1164 1165 struct pf_state_key * 1166 pf_state_key_clone(struct pf_state_key *orig) 1167 { 1168 struct pf_state_key *sk; 1169 1170 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1171 if (sk == NULL) 1172 return (NULL); 1173 1174 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1175 1176 return (sk); 1177 } 1178 1179 int 1180 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw, 1181 struct pf_state_key *sks, struct pf_state *s) 1182 { 1183 struct pf_idhash *ih; 1184 struct pf_state *cur; 1185 int error; 1186 1187 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1188 ("%s: sks not pristine", __func__)); 1189 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1190 ("%s: skw not pristine", __func__)); 1191 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1192 1193 s->kif = kif; 1194 1195 if (s->id == 0 && s->creatorid == 0) { 1196 /* XXX: should be atomic, but probability of collision low */ 1197 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID) 1198 V_pf_stateid[curcpu] = 1; 1199 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT; 1200 s->id = htobe64(s->id); 1201 s->creatorid = V_pf_status.hostid; 1202 } 1203 1204 /* Returns with ID locked on success. */ 1205 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1206 return (error); 1207 1208 ih = &V_pf_idhash[PF_IDHASH(s)]; 1209 PF_HASHROW_ASSERT(ih); 1210 LIST_FOREACH(cur, &ih->states, entry) 1211 if (cur->id == s->id && cur->creatorid == s->creatorid) 1212 break; 1213 1214 if (cur != NULL) { 1215 PF_HASHROW_UNLOCK(ih); 1216 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1217 printf("pf: state ID collision: " 1218 "id: %016llx creatorid: %08x\n", 1219 (unsigned long long)be64toh(s->id), 1220 ntohl(s->creatorid)); 1221 } 1222 pf_detach_state(s); 1223 return (EEXIST); 1224 } 1225 LIST_INSERT_HEAD(&ih->states, s, entry); 1226 /* One for keys, one for ID hash. */ 1227 refcount_init(&s->refs, 2); 1228 1229 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1230 if (pfsync_insert_state_ptr != NULL) 1231 pfsync_insert_state_ptr(s); 1232 1233 /* Returns locked. */ 1234 return (0); 1235 } 1236 1237 /* 1238 * Find state by ID: returns with locked row on success. 1239 */ 1240 struct pf_state * 1241 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1242 { 1243 struct pf_idhash *ih; 1244 struct pf_state *s; 1245 1246 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1247 1248 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; 1249 1250 PF_HASHROW_LOCK(ih); 1251 LIST_FOREACH(s, &ih->states, entry) 1252 if (s->id == id && s->creatorid == creatorid) 1253 break; 1254 1255 if (s == NULL) 1256 PF_HASHROW_UNLOCK(ih); 1257 1258 return (s); 1259 } 1260 1261 /* 1262 * Find state by key. 1263 * Returns with ID hash slot locked on success. 1264 */ 1265 static struct pf_state * 1266 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir) 1267 { 1268 struct pf_keyhash *kh; 1269 struct pf_state_key *sk; 1270 struct pf_state *s; 1271 int idx; 1272 1273 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1274 1275 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1276 1277 PF_HASHROW_LOCK(kh); 1278 LIST_FOREACH(sk, &kh->keys, entry) 1279 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1280 break; 1281 if (sk == NULL) { 1282 PF_HASHROW_UNLOCK(kh); 1283 return (NULL); 1284 } 1285 1286 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1287 1288 /* List is sorted, if-bound states before floating ones. */ 1289 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1290 if (s->kif == V_pfi_all || s->kif == kif) { 1291 PF_STATE_LOCK(s); 1292 PF_HASHROW_UNLOCK(kh); 1293 if (s->timeout >= PFTM_MAX) { 1294 /* 1295 * State is either being processed by 1296 * pf_unlink_state() in an other thread, or 1297 * is scheduled for immediate expiry. 1298 */ 1299 PF_STATE_UNLOCK(s); 1300 return (NULL); 1301 } 1302 return (s); 1303 } 1304 PF_HASHROW_UNLOCK(kh); 1305 1306 return (NULL); 1307 } 1308 1309 struct pf_state * 1310 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1311 { 1312 struct pf_keyhash *kh; 1313 struct pf_state_key *sk; 1314 struct pf_state *s, *ret = NULL; 1315 int idx, inout = 0; 1316 1317 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1318 1319 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1320 1321 PF_HASHROW_LOCK(kh); 1322 LIST_FOREACH(sk, &kh->keys, entry) 1323 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1324 break; 1325 if (sk == NULL) { 1326 PF_HASHROW_UNLOCK(kh); 1327 return (NULL); 1328 } 1329 switch (dir) { 1330 case PF_IN: 1331 idx = PF_SK_WIRE; 1332 break; 1333 case PF_OUT: 1334 idx = PF_SK_STACK; 1335 break; 1336 case PF_INOUT: 1337 idx = PF_SK_WIRE; 1338 inout = 1; 1339 break; 1340 default: 1341 panic("%s: dir %u", __func__, dir); 1342 } 1343 second_run: 1344 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1345 if (more == NULL) { 1346 PF_HASHROW_UNLOCK(kh); 1347 return (s); 1348 } 1349 1350 if (ret) 1351 (*more)++; 1352 else 1353 ret = s; 1354 } 1355 if (inout == 1) { 1356 inout = 0; 1357 idx = PF_SK_STACK; 1358 goto second_run; 1359 } 1360 PF_HASHROW_UNLOCK(kh); 1361 1362 return (ret); 1363 } 1364 1365 /* END state table stuff */ 1366 1367 static void 1368 pf_send(struct pf_send_entry *pfse) 1369 { 1370 1371 PF_SENDQ_LOCK(); 1372 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1373 PF_SENDQ_UNLOCK(); 1374 swi_sched(V_pf_swi_cookie, 0); 1375 } 1376 1377 void 1378 pf_intr(void *v) 1379 { 1380 struct pf_send_head queue; 1381 struct pf_send_entry *pfse, *next; 1382 1383 CURVNET_SET((struct vnet *)v); 1384 1385 PF_SENDQ_LOCK(); 1386 queue = V_pf_sendqueue; 1387 STAILQ_INIT(&V_pf_sendqueue); 1388 PF_SENDQ_UNLOCK(); 1389 1390 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1391 switch (pfse->pfse_type) { 1392 #ifdef INET 1393 case PFSE_IP: 1394 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL); 1395 break; 1396 case PFSE_ICMP: 1397 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1398 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1399 break; 1400 #endif /* INET */ 1401 #ifdef INET6 1402 case PFSE_IP6: 1403 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL, 1404 NULL); 1405 break; 1406 case PFSE_ICMP6: 1407 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 1408 pfse->icmpopts.code, pfse->icmpopts.mtu); 1409 break; 1410 #endif /* INET6 */ 1411 default: 1412 panic("%s: unknown type", __func__); 1413 } 1414 free(pfse, M_PFTEMP); 1415 } 1416 CURVNET_RESTORE(); 1417 } 1418 1419 void 1420 pf_purge_thread(void *v) 1421 { 1422 u_int idx = 0; 1423 1424 CURVNET_SET((struct vnet *)v); 1425 1426 for (;;) { 1427 PF_RULES_RLOCK(); 1428 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10); 1429 1430 if (V_pf_end_threads) { 1431 /* 1432 * To cleanse up all kifs and rules we need 1433 * two runs: first one clears reference flags, 1434 * then pf_purge_expired_states() doesn't 1435 * raise them, and then second run frees. 1436 */ 1437 PF_RULES_RUNLOCK(); 1438 pf_purge_unlinked_rules(); 1439 pfi_kif_purge(); 1440 1441 /* 1442 * Now purge everything. 1443 */ 1444 pf_purge_expired_states(0, pf_hashmask); 1445 pf_purge_expired_fragments(); 1446 pf_purge_expired_src_nodes(); 1447 1448 /* 1449 * Now all kifs & rules should be unreferenced, 1450 * thus should be successfully freed. 1451 */ 1452 pf_purge_unlinked_rules(); 1453 pfi_kif_purge(); 1454 1455 /* 1456 * Announce success and exit. 1457 */ 1458 PF_RULES_RLOCK(); 1459 V_pf_end_threads++; 1460 PF_RULES_RUNLOCK(); 1461 wakeup(pf_purge_thread); 1462 kproc_exit(0); 1463 } 1464 PF_RULES_RUNLOCK(); 1465 1466 /* Process 1/interval fraction of the state table every run. */ 1467 idx = pf_purge_expired_states(idx, pf_hashmask / 1468 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1469 1470 /* Purge other expired types every PFTM_INTERVAL seconds. */ 1471 if (idx == 0) { 1472 /* 1473 * Order is important: 1474 * - states and src nodes reference rules 1475 * - states and rules reference kifs 1476 */ 1477 pf_purge_expired_fragments(); 1478 pf_purge_expired_src_nodes(); 1479 pf_purge_unlinked_rules(); 1480 pfi_kif_purge(); 1481 } 1482 } 1483 /* not reached */ 1484 CURVNET_RESTORE(); 1485 } 1486 1487 u_int32_t 1488 pf_state_expires(const struct pf_state *state) 1489 { 1490 u_int32_t timeout; 1491 u_int32_t start; 1492 u_int32_t end; 1493 u_int32_t states; 1494 1495 /* handle all PFTM_* > PFTM_MAX here */ 1496 if (state->timeout == PFTM_PURGE) 1497 return (time_uptime); 1498 KASSERT(state->timeout != PFTM_UNLINKED, 1499 ("pf_state_expires: timeout == PFTM_UNLINKED")); 1500 KASSERT((state->timeout < PFTM_MAX), 1501 ("pf_state_expires: timeout > PFTM_MAX")); 1502 timeout = state->rule.ptr->timeout[state->timeout]; 1503 if (!timeout) 1504 timeout = V_pf_default_rule.timeout[state->timeout]; 1505 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1506 if (start) { 1507 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1508 states = counter_u64_fetch(state->rule.ptr->states_cur); 1509 } else { 1510 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1511 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1512 states = V_pf_status.states; 1513 } 1514 if (end && states > start && start < end) { 1515 if (states < end) 1516 return (state->expire + timeout * (end - states) / 1517 (end - start)); 1518 else 1519 return (time_uptime); 1520 } 1521 return (state->expire + timeout); 1522 } 1523 1524 void 1525 pf_purge_expired_src_nodes() 1526 { 1527 struct pf_src_node_list freelist; 1528 struct pf_srchash *sh; 1529 struct pf_src_node *cur, *next; 1530 int i; 1531 1532 LIST_INIT(&freelist); 1533 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1534 PF_HASHROW_LOCK(sh); 1535 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 1536 if (cur->states == 0 && cur->expire <= time_uptime) { 1537 pf_unlink_src_node(cur); 1538 LIST_INSERT_HEAD(&freelist, cur, entry); 1539 } else if (cur->rule.ptr != NULL) 1540 cur->rule.ptr->rule_flag |= PFRULE_REFS; 1541 PF_HASHROW_UNLOCK(sh); 1542 } 1543 1544 pf_free_src_nodes(&freelist); 1545 1546 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 1547 } 1548 1549 static void 1550 pf_src_tree_remove_state(struct pf_state *s) 1551 { 1552 struct pf_src_node *sn; 1553 struct pf_srchash *sh; 1554 uint32_t timeout; 1555 1556 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 1557 s->rule.ptr->timeout[PFTM_SRC_NODE] : 1558 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 1559 1560 if (s->src_node != NULL) { 1561 sn = s->src_node; 1562 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 1563 PF_HASHROW_LOCK(sh); 1564 if (s->src.tcp_est) 1565 --sn->conn; 1566 if (--sn->states == 0) 1567 sn->expire = time_uptime + timeout; 1568 PF_HASHROW_UNLOCK(sh); 1569 } 1570 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 1571 sn = s->nat_src_node; 1572 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 1573 PF_HASHROW_LOCK(sh); 1574 if (--sn->states == 0) 1575 sn->expire = time_uptime + timeout; 1576 PF_HASHROW_UNLOCK(sh); 1577 } 1578 s->src_node = s->nat_src_node = NULL; 1579 } 1580 1581 /* 1582 * Unlink and potentilly free a state. Function may be 1583 * called with ID hash row locked, but always returns 1584 * unlocked, since it needs to go through key hash locking. 1585 */ 1586 int 1587 pf_unlink_state(struct pf_state *s, u_int flags) 1588 { 1589 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 1590 1591 if ((flags & PF_ENTER_LOCKED) == 0) 1592 PF_HASHROW_LOCK(ih); 1593 else 1594 PF_HASHROW_ASSERT(ih); 1595 1596 if (s->timeout == PFTM_UNLINKED) { 1597 /* 1598 * State is being processed 1599 * by pf_unlink_state() in 1600 * an other thread. 1601 */ 1602 PF_HASHROW_UNLOCK(ih); 1603 return (0); /* XXXGL: undefined actually */ 1604 } 1605 1606 if (s->src.state == PF_TCPS_PROXY_DST) { 1607 /* XXX wire key the right one? */ 1608 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af, 1609 &s->key[PF_SK_WIRE]->addr[1], 1610 &s->key[PF_SK_WIRE]->addr[0], 1611 s->key[PF_SK_WIRE]->port[1], 1612 s->key[PF_SK_WIRE]->port[0], 1613 s->src.seqhi, s->src.seqlo + 1, 1614 TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL); 1615 } 1616 1617 LIST_REMOVE(s, entry); 1618 pf_src_tree_remove_state(s); 1619 1620 if (pfsync_delete_state_ptr != NULL) 1621 pfsync_delete_state_ptr(s); 1622 1623 STATE_DEC_COUNTERS(s); 1624 1625 s->timeout = PFTM_UNLINKED; 1626 1627 PF_HASHROW_UNLOCK(ih); 1628 1629 pf_detach_state(s); 1630 refcount_release(&s->refs); 1631 1632 return (pf_release_state(s)); 1633 } 1634 1635 void 1636 pf_free_state(struct pf_state *cur) 1637 { 1638 1639 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 1640 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 1641 cur->timeout)); 1642 1643 pf_normalize_tcp_cleanup(cur); 1644 uma_zfree(V_pf_state_z, cur); 1645 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 1646 } 1647 1648 /* 1649 * Called only from pf_purge_thread(), thus serialized. 1650 */ 1651 static u_int 1652 pf_purge_expired_states(u_int i, int maxcheck) 1653 { 1654 struct pf_idhash *ih; 1655 struct pf_state *s; 1656 1657 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1658 1659 /* 1660 * Go through hash and unlink states that expire now. 1661 */ 1662 while (maxcheck > 0) { 1663 1664 ih = &V_pf_idhash[i]; 1665 relock: 1666 PF_HASHROW_LOCK(ih); 1667 LIST_FOREACH(s, &ih->states, entry) { 1668 if (pf_state_expires(s) <= time_uptime) { 1669 V_pf_status.states -= 1670 pf_unlink_state(s, PF_ENTER_LOCKED); 1671 goto relock; 1672 } 1673 s->rule.ptr->rule_flag |= PFRULE_REFS; 1674 if (s->nat_rule.ptr != NULL) 1675 s->nat_rule.ptr->rule_flag |= PFRULE_REFS; 1676 if (s->anchor.ptr != NULL) 1677 s->anchor.ptr->rule_flag |= PFRULE_REFS; 1678 s->kif->pfik_flags |= PFI_IFLAG_REFS; 1679 if (s->rt_kif) 1680 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 1681 } 1682 PF_HASHROW_UNLOCK(ih); 1683 1684 /* Return when we hit end of hash. */ 1685 if (++i > pf_hashmask) { 1686 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1687 return (0); 1688 } 1689 1690 maxcheck--; 1691 } 1692 1693 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1694 1695 return (i); 1696 } 1697 1698 static void 1699 pf_purge_unlinked_rules() 1700 { 1701 struct pf_rulequeue tmpq; 1702 struct pf_rule *r, *r1; 1703 1704 /* 1705 * If we have overloading task pending, then we'd 1706 * better skip purging this time. There is a tiny 1707 * probability that overloading task references 1708 * an already unlinked rule. 1709 */ 1710 PF_OVERLOADQ_LOCK(); 1711 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 1712 PF_OVERLOADQ_UNLOCK(); 1713 return; 1714 } 1715 PF_OVERLOADQ_UNLOCK(); 1716 1717 /* 1718 * Do naive mark-and-sweep garbage collecting of old rules. 1719 * Reference flag is raised by pf_purge_expired_states() 1720 * and pf_purge_expired_src_nodes(). 1721 * 1722 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 1723 * use a temporary queue. 1724 */ 1725 TAILQ_INIT(&tmpq); 1726 PF_UNLNKDRULES_LOCK(); 1727 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 1728 if (!(r->rule_flag & PFRULE_REFS)) { 1729 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 1730 TAILQ_INSERT_TAIL(&tmpq, r, entries); 1731 } else 1732 r->rule_flag &= ~PFRULE_REFS; 1733 } 1734 PF_UNLNKDRULES_UNLOCK(); 1735 1736 if (!TAILQ_EMPTY(&tmpq)) { 1737 PF_RULES_WLOCK(); 1738 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 1739 TAILQ_REMOVE(&tmpq, r, entries); 1740 pf_free_rule(r); 1741 } 1742 PF_RULES_WUNLOCK(); 1743 } 1744 } 1745 1746 void 1747 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 1748 { 1749 switch (af) { 1750 #ifdef INET 1751 case AF_INET: { 1752 u_int32_t a = ntohl(addr->addr32[0]); 1753 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 1754 (a>>8)&255, a&255); 1755 if (p) { 1756 p = ntohs(p); 1757 printf(":%u", p); 1758 } 1759 break; 1760 } 1761 #endif /* INET */ 1762 #ifdef INET6 1763 case AF_INET6: { 1764 u_int16_t b; 1765 u_int8_t i, curstart, curend, maxstart, maxend; 1766 curstart = curend = maxstart = maxend = 255; 1767 for (i = 0; i < 8; i++) { 1768 if (!addr->addr16[i]) { 1769 if (curstart == 255) 1770 curstart = i; 1771 curend = i; 1772 } else { 1773 if ((curend - curstart) > 1774 (maxend - maxstart)) { 1775 maxstart = curstart; 1776 maxend = curend; 1777 } 1778 curstart = curend = 255; 1779 } 1780 } 1781 if ((curend - curstart) > 1782 (maxend - maxstart)) { 1783 maxstart = curstart; 1784 maxend = curend; 1785 } 1786 for (i = 0; i < 8; i++) { 1787 if (i >= maxstart && i <= maxend) { 1788 if (i == 0) 1789 printf(":"); 1790 if (i == maxend) 1791 printf(":"); 1792 } else { 1793 b = ntohs(addr->addr16[i]); 1794 printf("%x", b); 1795 if (i < 7) 1796 printf(":"); 1797 } 1798 } 1799 if (p) { 1800 p = ntohs(p); 1801 printf("[%u]", p); 1802 } 1803 break; 1804 } 1805 #endif /* INET6 */ 1806 } 1807 } 1808 1809 void 1810 pf_print_state(struct pf_state *s) 1811 { 1812 pf_print_state_parts(s, NULL, NULL); 1813 } 1814 1815 static void 1816 pf_print_state_parts(struct pf_state *s, 1817 struct pf_state_key *skwp, struct pf_state_key *sksp) 1818 { 1819 struct pf_state_key *skw, *sks; 1820 u_int8_t proto, dir; 1821 1822 /* Do our best to fill these, but they're skipped if NULL */ 1823 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 1824 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 1825 proto = skw ? skw->proto : (sks ? sks->proto : 0); 1826 dir = s ? s->direction : 0; 1827 1828 switch (proto) { 1829 case IPPROTO_IPV4: 1830 printf("IPv4"); 1831 break; 1832 case IPPROTO_IPV6: 1833 printf("IPv6"); 1834 break; 1835 case IPPROTO_TCP: 1836 printf("TCP"); 1837 break; 1838 case IPPROTO_UDP: 1839 printf("UDP"); 1840 break; 1841 case IPPROTO_ICMP: 1842 printf("ICMP"); 1843 break; 1844 case IPPROTO_ICMPV6: 1845 printf("ICMPv6"); 1846 break; 1847 default: 1848 printf("%u", skw->proto); 1849 break; 1850 } 1851 switch (dir) { 1852 case PF_IN: 1853 printf(" in"); 1854 break; 1855 case PF_OUT: 1856 printf(" out"); 1857 break; 1858 } 1859 if (skw) { 1860 printf(" wire: "); 1861 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 1862 printf(" "); 1863 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 1864 } 1865 if (sks) { 1866 printf(" stack: "); 1867 if (sks != skw) { 1868 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 1869 printf(" "); 1870 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 1871 } else 1872 printf("-"); 1873 } 1874 if (s) { 1875 if (proto == IPPROTO_TCP) { 1876 printf(" [lo=%u high=%u win=%u modulator=%u", 1877 s->src.seqlo, s->src.seqhi, 1878 s->src.max_win, s->src.seqdiff); 1879 if (s->src.wscale && s->dst.wscale) 1880 printf(" wscale=%u", 1881 s->src.wscale & PF_WSCALE_MASK); 1882 printf("]"); 1883 printf(" [lo=%u high=%u win=%u modulator=%u", 1884 s->dst.seqlo, s->dst.seqhi, 1885 s->dst.max_win, s->dst.seqdiff); 1886 if (s->src.wscale && s->dst.wscale) 1887 printf(" wscale=%u", 1888 s->dst.wscale & PF_WSCALE_MASK); 1889 printf("]"); 1890 } 1891 printf(" %u:%u", s->src.state, s->dst.state); 1892 } 1893 } 1894 1895 void 1896 pf_print_flags(u_int8_t f) 1897 { 1898 if (f) 1899 printf(" "); 1900 if (f & TH_FIN) 1901 printf("F"); 1902 if (f & TH_SYN) 1903 printf("S"); 1904 if (f & TH_RST) 1905 printf("R"); 1906 if (f & TH_PUSH) 1907 printf("P"); 1908 if (f & TH_ACK) 1909 printf("A"); 1910 if (f & TH_URG) 1911 printf("U"); 1912 if (f & TH_ECE) 1913 printf("E"); 1914 if (f & TH_CWR) 1915 printf("W"); 1916 } 1917 1918 #define PF_SET_SKIP_STEPS(i) \ 1919 do { \ 1920 while (head[i] != cur) { \ 1921 head[i]->skip[i].ptr = cur; \ 1922 head[i] = TAILQ_NEXT(head[i], entries); \ 1923 } \ 1924 } while (0) 1925 1926 void 1927 pf_calc_skip_steps(struct pf_rulequeue *rules) 1928 { 1929 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT]; 1930 int i; 1931 1932 cur = TAILQ_FIRST(rules); 1933 prev = cur; 1934 for (i = 0; i < PF_SKIP_COUNT; ++i) 1935 head[i] = cur; 1936 while (cur != NULL) { 1937 1938 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 1939 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 1940 if (cur->direction != prev->direction) 1941 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 1942 if (cur->af != prev->af) 1943 PF_SET_SKIP_STEPS(PF_SKIP_AF); 1944 if (cur->proto != prev->proto) 1945 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 1946 if (cur->src.neg != prev->src.neg || 1947 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 1948 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 1949 if (cur->src.port[0] != prev->src.port[0] || 1950 cur->src.port[1] != prev->src.port[1] || 1951 cur->src.port_op != prev->src.port_op) 1952 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 1953 if (cur->dst.neg != prev->dst.neg || 1954 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 1955 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 1956 if (cur->dst.port[0] != prev->dst.port[0] || 1957 cur->dst.port[1] != prev->dst.port[1] || 1958 cur->dst.port_op != prev->dst.port_op) 1959 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 1960 1961 prev = cur; 1962 cur = TAILQ_NEXT(cur, entries); 1963 } 1964 for (i = 0; i < PF_SKIP_COUNT; ++i) 1965 PF_SET_SKIP_STEPS(i); 1966 } 1967 1968 static int 1969 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 1970 { 1971 if (aw1->type != aw2->type) 1972 return (1); 1973 switch (aw1->type) { 1974 case PF_ADDR_ADDRMASK: 1975 case PF_ADDR_RANGE: 1976 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0)) 1977 return (1); 1978 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0)) 1979 return (1); 1980 return (0); 1981 case PF_ADDR_DYNIFTL: 1982 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 1983 case PF_ADDR_NOROUTE: 1984 case PF_ADDR_URPFFAILED: 1985 return (0); 1986 case PF_ADDR_TABLE: 1987 return (aw1->p.tbl != aw2->p.tbl); 1988 default: 1989 printf("invalid address type: %d\n", aw1->type); 1990 return (1); 1991 } 1992 } 1993 1994 /** 1995 * Checksum updates are a little complicated because the checksum in the TCP/UDP 1996 * header isn't always a full checksum. In some cases (i.e. output) it's a 1997 * pseudo-header checksum, which is a partial checksum over src/dst IP 1998 * addresses, protocol number and length. 1999 * 2000 * That means we have the following cases: 2001 * * Input or forwarding: we don't have TSO, the checksum fields are full 2002 * checksums, we need to update the checksum whenever we change anything. 2003 * * Output (i.e. the checksum is a pseudo-header checksum): 2004 * x The field being updated is src/dst address or affects the length of 2005 * the packet. We need to update the pseudo-header checksum (note that this 2006 * checksum is not ones' complement). 2007 * x Some other field is being modified (e.g. src/dst port numbers): We 2008 * don't have to update anything. 2009 **/ 2010 u_int16_t 2011 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 2012 { 2013 u_int32_t l; 2014 2015 if (udp && !cksum) 2016 return (0x0000); 2017 l = cksum + old - new; 2018 l = (l >> 16) + (l & 65535); 2019 l = l & 65535; 2020 if (udp && !l) 2021 return (0xFFFF); 2022 return (l); 2023 } 2024 2025 u_int16_t 2026 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 2027 u_int16_t new, u_int8_t udp) 2028 { 2029 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2030 return (cksum); 2031 2032 return (pf_cksum_fixup(cksum, old, new, udp)); 2033 } 2034 2035 static void 2036 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 2037 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 2038 sa_family_t af) 2039 { 2040 struct pf_addr ao; 2041 u_int16_t po = *p; 2042 2043 PF_ACPY(&ao, a, af); 2044 PF_ACPY(a, an, af); 2045 2046 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2047 *pc = ~*pc; 2048 2049 *p = pn; 2050 2051 switch (af) { 2052 #ifdef INET 2053 case AF_INET: 2054 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2055 ao.addr16[0], an->addr16[0], 0), 2056 ao.addr16[1], an->addr16[1], 0); 2057 *p = pn; 2058 2059 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 2060 ao.addr16[0], an->addr16[0], u), 2061 ao.addr16[1], an->addr16[1], u); 2062 2063 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2064 break; 2065 #endif /* INET */ 2066 #ifdef INET6 2067 case AF_INET6: 2068 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2069 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2070 pf_cksum_fixup(pf_cksum_fixup(*pc, 2071 ao.addr16[0], an->addr16[0], u), 2072 ao.addr16[1], an->addr16[1], u), 2073 ao.addr16[2], an->addr16[2], u), 2074 ao.addr16[3], an->addr16[3], u), 2075 ao.addr16[4], an->addr16[4], u), 2076 ao.addr16[5], an->addr16[5], u), 2077 ao.addr16[6], an->addr16[6], u), 2078 ao.addr16[7], an->addr16[7], u); 2079 2080 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2081 break; 2082 #endif /* INET6 */ 2083 } 2084 2085 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 2086 CSUM_DELAY_DATA_IPV6)) { 2087 *pc = ~*pc; 2088 if (! *pc) 2089 *pc = 0xffff; 2090 } 2091 } 2092 2093 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2094 void 2095 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2096 { 2097 u_int32_t ao; 2098 2099 memcpy(&ao, a, sizeof(ao)); 2100 memcpy(a, &an, sizeof(u_int32_t)); 2101 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2102 ao % 65536, an % 65536, u); 2103 } 2104 2105 void 2106 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 2107 { 2108 u_int32_t ao; 2109 2110 memcpy(&ao, a, sizeof(ao)); 2111 memcpy(a, &an, sizeof(u_int32_t)); 2112 2113 *c = pf_proto_cksum_fixup(m, 2114 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 2115 ao % 65536, an % 65536, udp); 2116 } 2117 2118 #ifdef INET6 2119 static void 2120 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2121 { 2122 struct pf_addr ao; 2123 2124 PF_ACPY(&ao, a, AF_INET6); 2125 PF_ACPY(a, an, AF_INET6); 2126 2127 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2128 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2129 pf_cksum_fixup(pf_cksum_fixup(*c, 2130 ao.addr16[0], an->addr16[0], u), 2131 ao.addr16[1], an->addr16[1], u), 2132 ao.addr16[2], an->addr16[2], u), 2133 ao.addr16[3], an->addr16[3], u), 2134 ao.addr16[4], an->addr16[4], u), 2135 ao.addr16[5], an->addr16[5], u), 2136 ao.addr16[6], an->addr16[6], u), 2137 ao.addr16[7], an->addr16[7], u); 2138 } 2139 #endif /* INET6 */ 2140 2141 static void 2142 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2143 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2144 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2145 { 2146 struct pf_addr oia, ooa; 2147 2148 PF_ACPY(&oia, ia, af); 2149 if (oa) 2150 PF_ACPY(&ooa, oa, af); 2151 2152 /* Change inner protocol port, fix inner protocol checksum. */ 2153 if (ip != NULL) { 2154 u_int16_t oip = *ip; 2155 u_int32_t opc; 2156 2157 if (pc != NULL) 2158 opc = *pc; 2159 *ip = np; 2160 if (pc != NULL) 2161 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2162 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2163 if (pc != NULL) 2164 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2165 } 2166 /* Change inner ip address, fix inner ip and icmp checksums. */ 2167 PF_ACPY(ia, na, af); 2168 switch (af) { 2169 #ifdef INET 2170 case AF_INET: { 2171 u_int32_t oh2c = *h2c; 2172 2173 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2174 oia.addr16[0], ia->addr16[0], 0), 2175 oia.addr16[1], ia->addr16[1], 0); 2176 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2177 oia.addr16[0], ia->addr16[0], 0), 2178 oia.addr16[1], ia->addr16[1], 0); 2179 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2180 break; 2181 } 2182 #endif /* INET */ 2183 #ifdef INET6 2184 case AF_INET6: 2185 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2186 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2187 pf_cksum_fixup(pf_cksum_fixup(*ic, 2188 oia.addr16[0], ia->addr16[0], u), 2189 oia.addr16[1], ia->addr16[1], u), 2190 oia.addr16[2], ia->addr16[2], u), 2191 oia.addr16[3], ia->addr16[3], u), 2192 oia.addr16[4], ia->addr16[4], u), 2193 oia.addr16[5], ia->addr16[5], u), 2194 oia.addr16[6], ia->addr16[6], u), 2195 oia.addr16[7], ia->addr16[7], u); 2196 break; 2197 #endif /* INET6 */ 2198 } 2199 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2200 if (oa) { 2201 PF_ACPY(oa, na, af); 2202 switch (af) { 2203 #ifdef INET 2204 case AF_INET: 2205 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2206 ooa.addr16[0], oa->addr16[0], 0), 2207 ooa.addr16[1], oa->addr16[1], 0); 2208 break; 2209 #endif /* INET */ 2210 #ifdef INET6 2211 case AF_INET6: 2212 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2213 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2214 pf_cksum_fixup(pf_cksum_fixup(*ic, 2215 ooa.addr16[0], oa->addr16[0], u), 2216 ooa.addr16[1], oa->addr16[1], u), 2217 ooa.addr16[2], oa->addr16[2], u), 2218 ooa.addr16[3], oa->addr16[3], u), 2219 ooa.addr16[4], oa->addr16[4], u), 2220 ooa.addr16[5], oa->addr16[5], u), 2221 ooa.addr16[6], oa->addr16[6], u), 2222 ooa.addr16[7], oa->addr16[7], u); 2223 break; 2224 #endif /* INET6 */ 2225 } 2226 } 2227 } 2228 2229 2230 /* 2231 * Need to modulate the sequence numbers in the TCP SACK option 2232 * (credits to Krzysztof Pfaff for report and patch) 2233 */ 2234 static int 2235 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2236 struct tcphdr *th, struct pf_state_peer *dst) 2237 { 2238 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2239 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2240 int copyback = 0, i, olen; 2241 struct sackblk sack; 2242 2243 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2244 if (hlen < TCPOLEN_SACKLEN || 2245 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2246 return 0; 2247 2248 while (hlen >= TCPOLEN_SACKLEN) { 2249 olen = opt[1]; 2250 switch (*opt) { 2251 case TCPOPT_EOL: /* FALLTHROUGH */ 2252 case TCPOPT_NOP: 2253 opt++; 2254 hlen--; 2255 break; 2256 case TCPOPT_SACK: 2257 if (olen > hlen) 2258 olen = hlen; 2259 if (olen >= TCPOLEN_SACKLEN) { 2260 for (i = 2; i + TCPOLEN_SACK <= olen; 2261 i += TCPOLEN_SACK) { 2262 memcpy(&sack, &opt[i], sizeof(sack)); 2263 pf_change_proto_a(m, &sack.start, &th->th_sum, 2264 htonl(ntohl(sack.start) - dst->seqdiff), 0); 2265 pf_change_proto_a(m, &sack.end, &th->th_sum, 2266 htonl(ntohl(sack.end) - dst->seqdiff), 0); 2267 memcpy(&opt[i], &sack, sizeof(sack)); 2268 } 2269 copyback = 1; 2270 } 2271 /* FALLTHROUGH */ 2272 default: 2273 if (olen < 2) 2274 olen = 2; 2275 hlen -= olen; 2276 opt += olen; 2277 } 2278 } 2279 2280 if (copyback) 2281 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2282 return (copyback); 2283 } 2284 2285 static void 2286 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af, 2287 const struct pf_addr *saddr, const struct pf_addr *daddr, 2288 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2289 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2290 u_int16_t rtag, struct ifnet *ifp) 2291 { 2292 struct pf_send_entry *pfse; 2293 struct mbuf *m; 2294 int len, tlen; 2295 #ifdef INET 2296 struct ip *h = NULL; 2297 #endif /* INET */ 2298 #ifdef INET6 2299 struct ip6_hdr *h6 = NULL; 2300 #endif /* INET6 */ 2301 struct tcphdr *th; 2302 char *opt; 2303 struct pf_mtag *pf_mtag; 2304 2305 len = 0; 2306 th = NULL; 2307 2308 /* maximum segment size tcp option */ 2309 tlen = sizeof(struct tcphdr); 2310 if (mss) 2311 tlen += 4; 2312 2313 switch (af) { 2314 #ifdef INET 2315 case AF_INET: 2316 len = sizeof(struct ip) + tlen; 2317 break; 2318 #endif /* INET */ 2319 #ifdef INET6 2320 case AF_INET6: 2321 len = sizeof(struct ip6_hdr) + tlen; 2322 break; 2323 #endif /* INET6 */ 2324 default: 2325 panic("%s: unsupported af %d", __func__, af); 2326 } 2327 2328 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2329 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2330 if (pfse == NULL) 2331 return; 2332 m = m_gethdr(M_NOWAIT, MT_DATA); 2333 if (m == NULL) { 2334 free(pfse, M_PFTEMP); 2335 return; 2336 } 2337 #ifdef MAC 2338 mac_netinet_firewall_send(m); 2339 #endif 2340 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2341 free(pfse, M_PFTEMP); 2342 m_freem(m); 2343 return; 2344 } 2345 if (tag) 2346 m->m_flags |= M_SKIP_FIREWALL; 2347 pf_mtag->tag = rtag; 2348 2349 if (r != NULL && r->rtableid >= 0) 2350 M_SETFIB(m, r->rtableid); 2351 2352 #ifdef ALTQ 2353 if (r != NULL && r->qid) { 2354 pf_mtag->qid = r->qid; 2355 2356 /* add hints for ecn */ 2357 pf_mtag->hdr = mtod(m, struct ip *); 2358 } 2359 #endif /* ALTQ */ 2360 m->m_data += max_linkhdr; 2361 m->m_pkthdr.len = m->m_len = len; 2362 m->m_pkthdr.rcvif = NULL; 2363 bzero(m->m_data, len); 2364 switch (af) { 2365 #ifdef INET 2366 case AF_INET: 2367 h = mtod(m, struct ip *); 2368 2369 /* IP header fields included in the TCP checksum */ 2370 h->ip_p = IPPROTO_TCP; 2371 h->ip_len = htons(tlen); 2372 h->ip_src.s_addr = saddr->v4.s_addr; 2373 h->ip_dst.s_addr = daddr->v4.s_addr; 2374 2375 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2376 break; 2377 #endif /* INET */ 2378 #ifdef INET6 2379 case AF_INET6: 2380 h6 = mtod(m, struct ip6_hdr *); 2381 2382 /* IP header fields included in the TCP checksum */ 2383 h6->ip6_nxt = IPPROTO_TCP; 2384 h6->ip6_plen = htons(tlen); 2385 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2386 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2387 2388 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2389 break; 2390 #endif /* INET6 */ 2391 } 2392 2393 /* TCP header */ 2394 th->th_sport = sport; 2395 th->th_dport = dport; 2396 th->th_seq = htonl(seq); 2397 th->th_ack = htonl(ack); 2398 th->th_off = tlen >> 2; 2399 th->th_flags = flags; 2400 th->th_win = htons(win); 2401 2402 if (mss) { 2403 opt = (char *)(th + 1); 2404 opt[0] = TCPOPT_MAXSEG; 2405 opt[1] = 4; 2406 HTONS(mss); 2407 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2408 } 2409 2410 switch (af) { 2411 #ifdef INET 2412 case AF_INET: 2413 /* TCP checksum */ 2414 th->th_sum = in_cksum(m, len); 2415 2416 /* Finish the IP header */ 2417 h->ip_v = 4; 2418 h->ip_hl = sizeof(*h) >> 2; 2419 h->ip_tos = IPTOS_LOWDELAY; 2420 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 2421 h->ip_len = htons(len); 2422 h->ip_ttl = ttl ? ttl : V_ip_defttl; 2423 h->ip_sum = 0; 2424 2425 pfse->pfse_type = PFSE_IP; 2426 break; 2427 #endif /* INET */ 2428 #ifdef INET6 2429 case AF_INET6: 2430 /* TCP checksum */ 2431 th->th_sum = in6_cksum(m, IPPROTO_TCP, 2432 sizeof(struct ip6_hdr), tlen); 2433 2434 h6->ip6_vfc |= IPV6_VERSION; 2435 h6->ip6_hlim = IPV6_DEFHLIM; 2436 2437 pfse->pfse_type = PFSE_IP6; 2438 break; 2439 #endif /* INET6 */ 2440 } 2441 pfse->pfse_m = m; 2442 pf_send(pfse); 2443 } 2444 2445 static void 2446 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 2447 struct pf_rule *r) 2448 { 2449 struct pf_send_entry *pfse; 2450 struct mbuf *m0; 2451 struct pf_mtag *pf_mtag; 2452 2453 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2454 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2455 if (pfse == NULL) 2456 return; 2457 2458 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 2459 free(pfse, M_PFTEMP); 2460 return; 2461 } 2462 2463 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 2464 free(pfse, M_PFTEMP); 2465 return; 2466 } 2467 /* XXX: revisit */ 2468 m0->m_flags |= M_SKIP_FIREWALL; 2469 2470 if (r->rtableid >= 0) 2471 M_SETFIB(m0, r->rtableid); 2472 2473 #ifdef ALTQ 2474 if (r->qid) { 2475 pf_mtag->qid = r->qid; 2476 /* add hints for ecn */ 2477 pf_mtag->hdr = mtod(m0, struct ip *); 2478 } 2479 #endif /* ALTQ */ 2480 2481 switch (af) { 2482 #ifdef INET 2483 case AF_INET: 2484 pfse->pfse_type = PFSE_ICMP; 2485 break; 2486 #endif /* INET */ 2487 #ifdef INET6 2488 case AF_INET6: 2489 pfse->pfse_type = PFSE_ICMP6; 2490 break; 2491 #endif /* INET6 */ 2492 } 2493 pfse->pfse_m = m0; 2494 pfse->icmpopts.type = type; 2495 pfse->icmpopts.code = code; 2496 pf_send(pfse); 2497 } 2498 2499 /* 2500 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 2501 * If n is 0, they match if they are equal. If n is != 0, they match if they 2502 * are different. 2503 */ 2504 int 2505 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 2506 struct pf_addr *b, sa_family_t af) 2507 { 2508 int match = 0; 2509 2510 switch (af) { 2511 #ifdef INET 2512 case AF_INET: 2513 if ((a->addr32[0] & m->addr32[0]) == 2514 (b->addr32[0] & m->addr32[0])) 2515 match++; 2516 break; 2517 #endif /* INET */ 2518 #ifdef INET6 2519 case AF_INET6: 2520 if (((a->addr32[0] & m->addr32[0]) == 2521 (b->addr32[0] & m->addr32[0])) && 2522 ((a->addr32[1] & m->addr32[1]) == 2523 (b->addr32[1] & m->addr32[1])) && 2524 ((a->addr32[2] & m->addr32[2]) == 2525 (b->addr32[2] & m->addr32[2])) && 2526 ((a->addr32[3] & m->addr32[3]) == 2527 (b->addr32[3] & m->addr32[3]))) 2528 match++; 2529 break; 2530 #endif /* INET6 */ 2531 } 2532 if (match) { 2533 if (n) 2534 return (0); 2535 else 2536 return (1); 2537 } else { 2538 if (n) 2539 return (1); 2540 else 2541 return (0); 2542 } 2543 } 2544 2545 /* 2546 * Return 1 if b <= a <= e, otherwise return 0. 2547 */ 2548 int 2549 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 2550 struct pf_addr *a, sa_family_t af) 2551 { 2552 switch (af) { 2553 #ifdef INET 2554 case AF_INET: 2555 if ((a->addr32[0] < b->addr32[0]) || 2556 (a->addr32[0] > e->addr32[0])) 2557 return (0); 2558 break; 2559 #endif /* INET */ 2560 #ifdef INET6 2561 case AF_INET6: { 2562 int i; 2563 2564 /* check a >= b */ 2565 for (i = 0; i < 4; ++i) 2566 if (a->addr32[i] > b->addr32[i]) 2567 break; 2568 else if (a->addr32[i] < b->addr32[i]) 2569 return (0); 2570 /* check a <= e */ 2571 for (i = 0; i < 4; ++i) 2572 if (a->addr32[i] < e->addr32[i]) 2573 break; 2574 else if (a->addr32[i] > e->addr32[i]) 2575 return (0); 2576 break; 2577 } 2578 #endif /* INET6 */ 2579 } 2580 return (1); 2581 } 2582 2583 static int 2584 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 2585 { 2586 switch (op) { 2587 case PF_OP_IRG: 2588 return ((p > a1) && (p < a2)); 2589 case PF_OP_XRG: 2590 return ((p < a1) || (p > a2)); 2591 case PF_OP_RRG: 2592 return ((p >= a1) && (p <= a2)); 2593 case PF_OP_EQ: 2594 return (p == a1); 2595 case PF_OP_NE: 2596 return (p != a1); 2597 case PF_OP_LT: 2598 return (p < a1); 2599 case PF_OP_LE: 2600 return (p <= a1); 2601 case PF_OP_GT: 2602 return (p > a1); 2603 case PF_OP_GE: 2604 return (p >= a1); 2605 } 2606 return (0); /* never reached */ 2607 } 2608 2609 int 2610 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 2611 { 2612 NTOHS(a1); 2613 NTOHS(a2); 2614 NTOHS(p); 2615 return (pf_match(op, a1, a2, p)); 2616 } 2617 2618 static int 2619 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 2620 { 2621 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2622 return (0); 2623 return (pf_match(op, a1, a2, u)); 2624 } 2625 2626 static int 2627 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 2628 { 2629 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2630 return (0); 2631 return (pf_match(op, a1, a2, g)); 2632 } 2633 2634 int 2635 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag) 2636 { 2637 if (*tag == -1) 2638 *tag = mtag; 2639 2640 return ((!r->match_tag_not && r->match_tag == *tag) || 2641 (r->match_tag_not && r->match_tag != *tag)); 2642 } 2643 2644 int 2645 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 2646 { 2647 2648 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 2649 2650 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 2651 return (ENOMEM); 2652 2653 pd->pf_mtag->tag = tag; 2654 2655 return (0); 2656 } 2657 2658 #define PF_ANCHOR_STACKSIZE 32 2659 struct pf_anchor_stackframe { 2660 struct pf_ruleset *rs; 2661 struct pf_rule *r; /* XXX: + match bit */ 2662 struct pf_anchor *child; 2663 }; 2664 2665 /* 2666 * XXX: We rely on malloc(9) returning pointer aligned addresses. 2667 */ 2668 #define PF_ANCHORSTACK_MATCH 0x00000001 2669 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 2670 2671 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 2672 #define PF_ANCHOR_RULE(f) (struct pf_rule *) \ 2673 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 2674 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 2675 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 2676 } while (0) 2677 2678 void 2679 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth, 2680 struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a, 2681 int *match) 2682 { 2683 struct pf_anchor_stackframe *f; 2684 2685 PF_RULES_RASSERT(); 2686 2687 if (match) 2688 *match = 0; 2689 if (*depth >= PF_ANCHOR_STACKSIZE) { 2690 printf("%s: anchor stack overflow on %s\n", 2691 __func__, (*r)->anchor->name); 2692 *r = TAILQ_NEXT(*r, entries); 2693 return; 2694 } else if (*depth == 0 && a != NULL) 2695 *a = *r; 2696 f = stack + (*depth)++; 2697 f->rs = *rs; 2698 f->r = *r; 2699 if ((*r)->anchor_wildcard) { 2700 struct pf_anchor_node *parent = &(*r)->anchor->children; 2701 2702 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) { 2703 *r = NULL; 2704 return; 2705 } 2706 *rs = &f->child->ruleset; 2707 } else { 2708 f->child = NULL; 2709 *rs = &(*r)->anchor->ruleset; 2710 } 2711 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2712 } 2713 2714 int 2715 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth, 2716 struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a, 2717 int *match) 2718 { 2719 struct pf_anchor_stackframe *f; 2720 struct pf_rule *fr; 2721 int quick = 0; 2722 2723 PF_RULES_RASSERT(); 2724 2725 do { 2726 if (*depth <= 0) 2727 break; 2728 f = stack + *depth - 1; 2729 fr = PF_ANCHOR_RULE(f); 2730 if (f->child != NULL) { 2731 struct pf_anchor_node *parent; 2732 2733 /* 2734 * This block traverses through 2735 * a wildcard anchor. 2736 */ 2737 parent = &fr->anchor->children; 2738 if (match != NULL && *match) { 2739 /* 2740 * If any of "*" matched, then 2741 * "foo/ *" matched, mark frame 2742 * appropriately. 2743 */ 2744 PF_ANCHOR_SET_MATCH(f); 2745 *match = 0; 2746 } 2747 f->child = RB_NEXT(pf_anchor_node, parent, f->child); 2748 if (f->child != NULL) { 2749 *rs = &f->child->ruleset; 2750 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2751 if (*r == NULL) 2752 continue; 2753 else 2754 break; 2755 } 2756 } 2757 (*depth)--; 2758 if (*depth == 0 && a != NULL) 2759 *a = NULL; 2760 *rs = f->rs; 2761 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 2762 quick = fr->quick; 2763 *r = TAILQ_NEXT(fr, entries); 2764 } while (*r == NULL); 2765 2766 return (quick); 2767 } 2768 2769 #ifdef INET6 2770 void 2771 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 2772 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 2773 { 2774 switch (af) { 2775 #ifdef INET 2776 case AF_INET: 2777 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 2778 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 2779 break; 2780 #endif /* INET */ 2781 case AF_INET6: 2782 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 2783 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 2784 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 2785 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 2786 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 2787 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 2788 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 2789 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 2790 break; 2791 } 2792 } 2793 2794 void 2795 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 2796 { 2797 switch (af) { 2798 #ifdef INET 2799 case AF_INET: 2800 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 2801 break; 2802 #endif /* INET */ 2803 case AF_INET6: 2804 if (addr->addr32[3] == 0xffffffff) { 2805 addr->addr32[3] = 0; 2806 if (addr->addr32[2] == 0xffffffff) { 2807 addr->addr32[2] = 0; 2808 if (addr->addr32[1] == 0xffffffff) { 2809 addr->addr32[1] = 0; 2810 addr->addr32[0] = 2811 htonl(ntohl(addr->addr32[0]) + 1); 2812 } else 2813 addr->addr32[1] = 2814 htonl(ntohl(addr->addr32[1]) + 1); 2815 } else 2816 addr->addr32[2] = 2817 htonl(ntohl(addr->addr32[2]) + 1); 2818 } else 2819 addr->addr32[3] = 2820 htonl(ntohl(addr->addr32[3]) + 1); 2821 break; 2822 } 2823 } 2824 #endif /* INET6 */ 2825 2826 int 2827 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m) 2828 { 2829 struct pf_addr *saddr, *daddr; 2830 u_int16_t sport, dport; 2831 struct inpcbinfo *pi; 2832 struct inpcb *inp; 2833 2834 pd->lookup.uid = UID_MAX; 2835 pd->lookup.gid = GID_MAX; 2836 2837 switch (pd->proto) { 2838 case IPPROTO_TCP: 2839 if (pd->hdr.tcp == NULL) 2840 return (-1); 2841 sport = pd->hdr.tcp->th_sport; 2842 dport = pd->hdr.tcp->th_dport; 2843 pi = &V_tcbinfo; 2844 break; 2845 case IPPROTO_UDP: 2846 if (pd->hdr.udp == NULL) 2847 return (-1); 2848 sport = pd->hdr.udp->uh_sport; 2849 dport = pd->hdr.udp->uh_dport; 2850 pi = &V_udbinfo; 2851 break; 2852 default: 2853 return (-1); 2854 } 2855 if (direction == PF_IN) { 2856 saddr = pd->src; 2857 daddr = pd->dst; 2858 } else { 2859 u_int16_t p; 2860 2861 p = sport; 2862 sport = dport; 2863 dport = p; 2864 saddr = pd->dst; 2865 daddr = pd->src; 2866 } 2867 switch (pd->af) { 2868 #ifdef INET 2869 case AF_INET: 2870 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 2871 dport, INPLOOKUP_RLOCKPCB, NULL, m); 2872 if (inp == NULL) { 2873 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 2874 daddr->v4, dport, INPLOOKUP_WILDCARD | 2875 INPLOOKUP_RLOCKPCB, NULL, m); 2876 if (inp == NULL) 2877 return (-1); 2878 } 2879 break; 2880 #endif /* INET */ 2881 #ifdef INET6 2882 case AF_INET6: 2883 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 2884 dport, INPLOOKUP_RLOCKPCB, NULL, m); 2885 if (inp == NULL) { 2886 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 2887 &daddr->v6, dport, INPLOOKUP_WILDCARD | 2888 INPLOOKUP_RLOCKPCB, NULL, m); 2889 if (inp == NULL) 2890 return (-1); 2891 } 2892 break; 2893 #endif /* INET6 */ 2894 2895 default: 2896 return (-1); 2897 } 2898 INP_RLOCK_ASSERT(inp); 2899 pd->lookup.uid = inp->inp_cred->cr_uid; 2900 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 2901 INP_RUNLOCK(inp); 2902 2903 return (1); 2904 } 2905 2906 static u_int8_t 2907 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 2908 { 2909 int hlen; 2910 u_int8_t hdr[60]; 2911 u_int8_t *opt, optlen; 2912 u_int8_t wscale = 0; 2913 2914 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 2915 if (hlen <= sizeof(struct tcphdr)) 2916 return (0); 2917 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 2918 return (0); 2919 opt = hdr + sizeof(struct tcphdr); 2920 hlen -= sizeof(struct tcphdr); 2921 while (hlen >= 3) { 2922 switch (*opt) { 2923 case TCPOPT_EOL: 2924 case TCPOPT_NOP: 2925 ++opt; 2926 --hlen; 2927 break; 2928 case TCPOPT_WINDOW: 2929 wscale = opt[2]; 2930 if (wscale > TCP_MAX_WINSHIFT) 2931 wscale = TCP_MAX_WINSHIFT; 2932 wscale |= PF_WSCALE_FLAG; 2933 /* FALLTHROUGH */ 2934 default: 2935 optlen = opt[1]; 2936 if (optlen < 2) 2937 optlen = 2; 2938 hlen -= optlen; 2939 opt += optlen; 2940 break; 2941 } 2942 } 2943 return (wscale); 2944 } 2945 2946 static u_int16_t 2947 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 2948 { 2949 int hlen; 2950 u_int8_t hdr[60]; 2951 u_int8_t *opt, optlen; 2952 u_int16_t mss = V_tcp_mssdflt; 2953 2954 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 2955 if (hlen <= sizeof(struct tcphdr)) 2956 return (0); 2957 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 2958 return (0); 2959 opt = hdr + sizeof(struct tcphdr); 2960 hlen -= sizeof(struct tcphdr); 2961 while (hlen >= TCPOLEN_MAXSEG) { 2962 switch (*opt) { 2963 case TCPOPT_EOL: 2964 case TCPOPT_NOP: 2965 ++opt; 2966 --hlen; 2967 break; 2968 case TCPOPT_MAXSEG: 2969 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 2970 NTOHS(mss); 2971 /* FALLTHROUGH */ 2972 default: 2973 optlen = opt[1]; 2974 if (optlen < 2) 2975 optlen = 2; 2976 hlen -= optlen; 2977 opt += optlen; 2978 break; 2979 } 2980 } 2981 return (mss); 2982 } 2983 2984 static u_int16_t 2985 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 2986 { 2987 #ifdef INET 2988 struct sockaddr_in *dst; 2989 struct route ro; 2990 #endif /* INET */ 2991 #ifdef INET6 2992 struct sockaddr_in6 *dst6; 2993 struct route_in6 ro6; 2994 #endif /* INET6 */ 2995 struct rtentry *rt = NULL; 2996 int hlen = 0; 2997 u_int16_t mss = V_tcp_mssdflt; 2998 2999 switch (af) { 3000 #ifdef INET 3001 case AF_INET: 3002 hlen = sizeof(struct ip); 3003 bzero(&ro, sizeof(ro)); 3004 dst = (struct sockaddr_in *)&ro.ro_dst; 3005 dst->sin_family = AF_INET; 3006 dst->sin_len = sizeof(*dst); 3007 dst->sin_addr = addr->v4; 3008 in_rtalloc_ign(&ro, 0, rtableid); 3009 rt = ro.ro_rt; 3010 break; 3011 #endif /* INET */ 3012 #ifdef INET6 3013 case AF_INET6: 3014 hlen = sizeof(struct ip6_hdr); 3015 bzero(&ro6, sizeof(ro6)); 3016 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst; 3017 dst6->sin6_family = AF_INET6; 3018 dst6->sin6_len = sizeof(*dst6); 3019 dst6->sin6_addr = addr->v6; 3020 in6_rtalloc_ign(&ro6, 0, rtableid); 3021 rt = ro6.ro_rt; 3022 break; 3023 #endif /* INET6 */ 3024 } 3025 3026 if (rt && rt->rt_ifp) { 3027 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr); 3028 mss = max(V_tcp_mssdflt, mss); 3029 RTFREE(rt); 3030 } 3031 mss = min(mss, offer); 3032 mss = max(mss, 64); /* sanity - at least max opt space */ 3033 return (mss); 3034 } 3035 3036 static u_int32_t 3037 pf_tcp_iss(struct pf_pdesc *pd) 3038 { 3039 MD5_CTX ctx; 3040 u_int32_t digest[4]; 3041 3042 if (V_pf_tcp_secret_init == 0) { 3043 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 3044 MD5Init(&V_pf_tcp_secret_ctx); 3045 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 3046 sizeof(V_pf_tcp_secret)); 3047 V_pf_tcp_secret_init = 1; 3048 } 3049 3050 ctx = V_pf_tcp_secret_ctx; 3051 3052 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short)); 3053 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short)); 3054 if (pd->af == AF_INET6) { 3055 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 3056 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 3057 } else { 3058 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 3059 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 3060 } 3061 MD5Final((u_char *)digest, &ctx); 3062 V_pf_tcp_iss_off += 4096; 3063 #define ISN_RANDOM_INCREMENT (4096 - 1) 3064 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 3065 V_pf_tcp_iss_off); 3066 #undef ISN_RANDOM_INCREMENT 3067 } 3068 3069 static int 3070 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction, 3071 struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, 3072 struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp) 3073 { 3074 struct pf_rule *nr = NULL; 3075 struct pf_addr * const saddr = pd->src; 3076 struct pf_addr * const daddr = pd->dst; 3077 sa_family_t af = pd->af; 3078 struct pf_rule *r, *a = NULL; 3079 struct pf_ruleset *ruleset = NULL; 3080 struct pf_src_node *nsn = NULL; 3081 struct tcphdr *th = pd->hdr.tcp; 3082 struct pf_state_key *sk = NULL, *nk = NULL; 3083 u_short reason; 3084 int rewrite = 0, hdrlen = 0; 3085 int tag = -1, rtableid = -1; 3086 int asd = 0; 3087 int match = 0; 3088 int state_icmp = 0; 3089 u_int16_t sport = 0, dport = 0; 3090 u_int16_t bproto_sum = 0, bip_sum = 0; 3091 u_int8_t icmptype = 0, icmpcode = 0; 3092 struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3093 3094 PF_RULES_RASSERT(); 3095 3096 if (inp != NULL) { 3097 INP_LOCK_ASSERT(inp); 3098 pd->lookup.uid = inp->inp_cred->cr_uid; 3099 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3100 pd->lookup.done = 1; 3101 } 3102 3103 switch (pd->proto) { 3104 case IPPROTO_TCP: 3105 sport = th->th_sport; 3106 dport = th->th_dport; 3107 hdrlen = sizeof(*th); 3108 break; 3109 case IPPROTO_UDP: 3110 sport = pd->hdr.udp->uh_sport; 3111 dport = pd->hdr.udp->uh_dport; 3112 hdrlen = sizeof(*pd->hdr.udp); 3113 break; 3114 #ifdef INET 3115 case IPPROTO_ICMP: 3116 if (pd->af != AF_INET) 3117 break; 3118 sport = dport = pd->hdr.icmp->icmp_id; 3119 hdrlen = sizeof(*pd->hdr.icmp); 3120 icmptype = pd->hdr.icmp->icmp_type; 3121 icmpcode = pd->hdr.icmp->icmp_code; 3122 3123 if (icmptype == ICMP_UNREACH || 3124 icmptype == ICMP_SOURCEQUENCH || 3125 icmptype == ICMP_REDIRECT || 3126 icmptype == ICMP_TIMXCEED || 3127 icmptype == ICMP_PARAMPROB) 3128 state_icmp++; 3129 break; 3130 #endif /* INET */ 3131 #ifdef INET6 3132 case IPPROTO_ICMPV6: 3133 if (af != AF_INET6) 3134 break; 3135 sport = dport = pd->hdr.icmp6->icmp6_id; 3136 hdrlen = sizeof(*pd->hdr.icmp6); 3137 icmptype = pd->hdr.icmp6->icmp6_type; 3138 icmpcode = pd->hdr.icmp6->icmp6_code; 3139 3140 if (icmptype == ICMP6_DST_UNREACH || 3141 icmptype == ICMP6_PACKET_TOO_BIG || 3142 icmptype == ICMP6_TIME_EXCEEDED || 3143 icmptype == ICMP6_PARAM_PROB) 3144 state_icmp++; 3145 break; 3146 #endif /* INET6 */ 3147 default: 3148 sport = dport = hdrlen = 0; 3149 break; 3150 } 3151 3152 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3153 3154 /* check packet for BINAT/NAT/RDR */ 3155 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk, 3156 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 3157 KASSERT(sk != NULL, ("%s: null sk", __func__)); 3158 KASSERT(nk != NULL, ("%s: null nk", __func__)); 3159 3160 if (pd->ip_sum) 3161 bip_sum = *pd->ip_sum; 3162 3163 switch (pd->proto) { 3164 case IPPROTO_TCP: 3165 bproto_sum = th->th_sum; 3166 pd->proto_sum = &th->th_sum; 3167 3168 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3169 nk->port[pd->sidx] != sport) { 3170 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 3171 &th->th_sum, &nk->addr[pd->sidx], 3172 nk->port[pd->sidx], 0, af); 3173 pd->sport = &th->th_sport; 3174 sport = th->th_sport; 3175 } 3176 3177 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3178 nk->port[pd->didx] != dport) { 3179 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 3180 &th->th_sum, &nk->addr[pd->didx], 3181 nk->port[pd->didx], 0, af); 3182 dport = th->th_dport; 3183 pd->dport = &th->th_dport; 3184 } 3185 rewrite++; 3186 break; 3187 case IPPROTO_UDP: 3188 bproto_sum = pd->hdr.udp->uh_sum; 3189 pd->proto_sum = &pd->hdr.udp->uh_sum; 3190 3191 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3192 nk->port[pd->sidx] != sport) { 3193 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport, 3194 pd->ip_sum, &pd->hdr.udp->uh_sum, 3195 &nk->addr[pd->sidx], 3196 nk->port[pd->sidx], 1, af); 3197 sport = pd->hdr.udp->uh_sport; 3198 pd->sport = &pd->hdr.udp->uh_sport; 3199 } 3200 3201 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3202 nk->port[pd->didx] != dport) { 3203 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport, 3204 pd->ip_sum, &pd->hdr.udp->uh_sum, 3205 &nk->addr[pd->didx], 3206 nk->port[pd->didx], 1, af); 3207 dport = pd->hdr.udp->uh_dport; 3208 pd->dport = &pd->hdr.udp->uh_dport; 3209 } 3210 rewrite++; 3211 break; 3212 #ifdef INET 3213 case IPPROTO_ICMP: 3214 nk->port[0] = nk->port[1]; 3215 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 3216 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 3217 nk->addr[pd->sidx].v4.s_addr, 0); 3218 3219 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 3220 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 3221 nk->addr[pd->didx].v4.s_addr, 0); 3222 3223 if (nk->port[1] != pd->hdr.icmp->icmp_id) { 3224 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup( 3225 pd->hdr.icmp->icmp_cksum, sport, 3226 nk->port[1], 0); 3227 pd->hdr.icmp->icmp_id = nk->port[1]; 3228 pd->sport = &pd->hdr.icmp->icmp_id; 3229 } 3230 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 3231 break; 3232 #endif /* INET */ 3233 #ifdef INET6 3234 case IPPROTO_ICMPV6: 3235 nk->port[0] = nk->port[1]; 3236 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 3237 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum, 3238 &nk->addr[pd->sidx], 0); 3239 3240 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 3241 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum, 3242 &nk->addr[pd->didx], 0); 3243 rewrite++; 3244 break; 3245 #endif /* INET */ 3246 default: 3247 switch (af) { 3248 #ifdef INET 3249 case AF_INET: 3250 if (PF_ANEQ(saddr, 3251 &nk->addr[pd->sidx], AF_INET)) 3252 pf_change_a(&saddr->v4.s_addr, 3253 pd->ip_sum, 3254 nk->addr[pd->sidx].v4.s_addr, 0); 3255 3256 if (PF_ANEQ(daddr, 3257 &nk->addr[pd->didx], AF_INET)) 3258 pf_change_a(&daddr->v4.s_addr, 3259 pd->ip_sum, 3260 nk->addr[pd->didx].v4.s_addr, 0); 3261 break; 3262 #endif /* INET */ 3263 #ifdef INET6 3264 case AF_INET6: 3265 if (PF_ANEQ(saddr, 3266 &nk->addr[pd->sidx], AF_INET6)) 3267 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 3268 3269 if (PF_ANEQ(daddr, 3270 &nk->addr[pd->didx], AF_INET6)) 3271 PF_ACPY(saddr, &nk->addr[pd->didx], af); 3272 break; 3273 #endif /* INET */ 3274 } 3275 break; 3276 } 3277 if (nr->natpass) 3278 r = NULL; 3279 pd->nat_rule = nr; 3280 } 3281 3282 while (r != NULL) { 3283 r->evaluations++; 3284 if (pfi_kif_match(r->kif, kif) == r->ifnot) 3285 r = r->skip[PF_SKIP_IFP].ptr; 3286 else if (r->direction && r->direction != direction) 3287 r = r->skip[PF_SKIP_DIR].ptr; 3288 else if (r->af && r->af != af) 3289 r = r->skip[PF_SKIP_AF].ptr; 3290 else if (r->proto && r->proto != pd->proto) 3291 r = r->skip[PF_SKIP_PROTO].ptr; 3292 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 3293 r->src.neg, kif, M_GETFIB(m))) 3294 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3295 /* tcp/udp only. port_op always 0 in other cases */ 3296 else if (r->src.port_op && !pf_match_port(r->src.port_op, 3297 r->src.port[0], r->src.port[1], sport)) 3298 r = r->skip[PF_SKIP_SRC_PORT].ptr; 3299 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 3300 r->dst.neg, NULL, M_GETFIB(m))) 3301 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3302 /* tcp/udp only. port_op always 0 in other cases */ 3303 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 3304 r->dst.port[0], r->dst.port[1], dport)) 3305 r = r->skip[PF_SKIP_DST_PORT].ptr; 3306 /* icmp only. type always 0 in other cases */ 3307 else if (r->type && r->type != icmptype + 1) 3308 r = TAILQ_NEXT(r, entries); 3309 /* icmp only. type always 0 in other cases */ 3310 else if (r->code && r->code != icmpcode + 1) 3311 r = TAILQ_NEXT(r, entries); 3312 else if (r->tos && !(r->tos == pd->tos)) 3313 r = TAILQ_NEXT(r, entries); 3314 else if (r->rule_flag & PFRULE_FRAGMENT) 3315 r = TAILQ_NEXT(r, entries); 3316 else if (pd->proto == IPPROTO_TCP && 3317 (r->flagset & th->th_flags) != r->flags) 3318 r = TAILQ_NEXT(r, entries); 3319 /* tcp/udp only. uid.op always 0 in other cases */ 3320 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 3321 pf_socket_lookup(direction, pd, m), 1)) && 3322 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 3323 pd->lookup.uid)) 3324 r = TAILQ_NEXT(r, entries); 3325 /* tcp/udp only. gid.op always 0 in other cases */ 3326 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 3327 pf_socket_lookup(direction, pd, m), 1)) && 3328 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 3329 pd->lookup.gid)) 3330 r = TAILQ_NEXT(r, entries); 3331 else if (r->prob && 3332 r->prob <= arc4random()) 3333 r = TAILQ_NEXT(r, entries); 3334 else if (r->match_tag && !pf_match_tag(m, r, &tag, 3335 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 3336 r = TAILQ_NEXT(r, entries); 3337 else if (r->os_fingerprint != PF_OSFP_ANY && 3338 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 3339 pf_osfp_fingerprint(pd, m, off, th), 3340 r->os_fingerprint))) 3341 r = TAILQ_NEXT(r, entries); 3342 else { 3343 if (r->tag) 3344 tag = r->tag; 3345 if (r->rtableid >= 0) 3346 rtableid = r->rtableid; 3347 if (r->anchor == NULL) { 3348 match = 1; 3349 *rm = r; 3350 *am = a; 3351 *rsm = ruleset; 3352 if ((*rm)->quick) 3353 break; 3354 r = TAILQ_NEXT(r, entries); 3355 } else 3356 pf_step_into_anchor(anchor_stack, &asd, 3357 &ruleset, PF_RULESET_FILTER, &r, &a, 3358 &match); 3359 } 3360 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 3361 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 3362 break; 3363 } 3364 r = *rm; 3365 a = *am; 3366 ruleset = *rsm; 3367 3368 REASON_SET(&reason, PFRES_MATCH); 3369 3370 if (r->log || (nr != NULL && nr->log)) { 3371 if (rewrite) 3372 m_copyback(m, off, hdrlen, pd->hdr.any); 3373 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a, 3374 ruleset, pd, 1); 3375 } 3376 3377 if ((r->action == PF_DROP) && 3378 ((r->rule_flag & PFRULE_RETURNRST) || 3379 (r->rule_flag & PFRULE_RETURNICMP) || 3380 (r->rule_flag & PFRULE_RETURN))) { 3381 /* undo NAT changes, if they have taken place */ 3382 if (nr != NULL) { 3383 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3384 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3385 if (pd->sport) 3386 *pd->sport = sk->port[pd->sidx]; 3387 if (pd->dport) 3388 *pd->dport = sk->port[pd->didx]; 3389 if (pd->proto_sum) 3390 *pd->proto_sum = bproto_sum; 3391 if (pd->ip_sum) 3392 *pd->ip_sum = bip_sum; 3393 m_copyback(m, off, hdrlen, pd->hdr.any); 3394 } 3395 if (pd->proto == IPPROTO_TCP && 3396 ((r->rule_flag & PFRULE_RETURNRST) || 3397 (r->rule_flag & PFRULE_RETURN)) && 3398 !(th->th_flags & TH_RST)) { 3399 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3400 int len = 0; 3401 #ifdef INET 3402 struct ip *h4; 3403 #endif 3404 #ifdef INET6 3405 struct ip6_hdr *h6; 3406 #endif 3407 3408 switch (af) { 3409 #ifdef INET 3410 case AF_INET: 3411 h4 = mtod(m, struct ip *); 3412 len = ntohs(h4->ip_len) - off; 3413 break; 3414 #endif 3415 #ifdef INET6 3416 case AF_INET6: 3417 h6 = mtod(m, struct ip6_hdr *); 3418 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3419 break; 3420 #endif 3421 } 3422 3423 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3424 REASON_SET(&reason, PFRES_PROTCKSUM); 3425 else { 3426 if (th->th_flags & TH_SYN) 3427 ack++; 3428 if (th->th_flags & TH_FIN) 3429 ack++; 3430 pf_send_tcp(m, r, af, pd->dst, 3431 pd->src, th->th_dport, th->th_sport, 3432 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3433 r->return_ttl, 1, 0, kif->pfik_ifp); 3434 } 3435 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3436 r->return_icmp) 3437 pf_send_icmp(m, r->return_icmp >> 8, 3438 r->return_icmp & 255, af, r); 3439 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3440 r->return_icmp6) 3441 pf_send_icmp(m, r->return_icmp6 >> 8, 3442 r->return_icmp6 & 255, af, r); 3443 } 3444 3445 if (r->action == PF_DROP) 3446 goto cleanup; 3447 3448 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 3449 REASON_SET(&reason, PFRES_MEMORY); 3450 goto cleanup; 3451 } 3452 if (rtableid >= 0) 3453 M_SETFIB(m, rtableid); 3454 3455 if (!state_icmp && (r->keep_state || nr != NULL || 3456 (pd->flags & PFDESC_TCP_NORM))) { 3457 int action; 3458 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 3459 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 3460 hdrlen); 3461 if (action != PF_PASS) 3462 return (action); 3463 } else { 3464 if (sk != NULL) 3465 uma_zfree(V_pf_state_key_z, sk); 3466 if (nk != NULL) 3467 uma_zfree(V_pf_state_key_z, nk); 3468 } 3469 3470 /* copy back packet headers if we performed NAT operations */ 3471 if (rewrite) 3472 m_copyback(m, off, hdrlen, pd->hdr.any); 3473 3474 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 3475 direction == PF_OUT && 3476 pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m)) 3477 /* 3478 * We want the state created, but we dont 3479 * want to send this in case a partner 3480 * firewall has to know about it to allow 3481 * replies through it. 3482 */ 3483 return (PF_DEFER); 3484 3485 return (PF_PASS); 3486 3487 cleanup: 3488 if (sk != NULL) 3489 uma_zfree(V_pf_state_key_z, sk); 3490 if (nk != NULL) 3491 uma_zfree(V_pf_state_key_z, nk); 3492 return (PF_DROP); 3493 } 3494 3495 static int 3496 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a, 3497 struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk, 3498 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 3499 u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm, 3500 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen) 3501 { 3502 struct pf_state *s = NULL; 3503 struct pf_src_node *sn = NULL; 3504 struct tcphdr *th = pd->hdr.tcp; 3505 u_int16_t mss = V_tcp_mssdflt; 3506 u_short reason; 3507 3508 /* check maximums */ 3509 if (r->max_states && 3510 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 3511 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 3512 REASON_SET(&reason, PFRES_MAXSTATES); 3513 return (PF_DROP); 3514 } 3515 /* src node for filter rule */ 3516 if ((r->rule_flag & PFRULE_SRCTRACK || 3517 r->rpool.opts & PF_POOL_STICKYADDR) && 3518 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { 3519 REASON_SET(&reason, PFRES_SRCLIMIT); 3520 goto csfailed; 3521 } 3522 /* src node for translation rule */ 3523 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 3524 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { 3525 REASON_SET(&reason, PFRES_SRCLIMIT); 3526 goto csfailed; 3527 } 3528 s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO); 3529 if (s == NULL) { 3530 REASON_SET(&reason, PFRES_MEMORY); 3531 goto csfailed; 3532 } 3533 s->rule.ptr = r; 3534 s->nat_rule.ptr = nr; 3535 s->anchor.ptr = a; 3536 STATE_INC_COUNTERS(s); 3537 if (r->allow_opts) 3538 s->state_flags |= PFSTATE_ALLOWOPTS; 3539 if (r->rule_flag & PFRULE_STATESLOPPY) 3540 s->state_flags |= PFSTATE_SLOPPY; 3541 s->log = r->log & PF_LOG_ALL; 3542 s->sync_state = PFSYNC_S_NONE; 3543 if (nr != NULL) 3544 s->log |= nr->log & PF_LOG_ALL; 3545 switch (pd->proto) { 3546 case IPPROTO_TCP: 3547 s->src.seqlo = ntohl(th->th_seq); 3548 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 3549 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 3550 r->keep_state == PF_STATE_MODULATE) { 3551 /* Generate sequence number modulator */ 3552 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 3553 0) 3554 s->src.seqdiff = 1; 3555 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 3556 htonl(s->src.seqlo + s->src.seqdiff), 0); 3557 *rewrite = 1; 3558 } else 3559 s->src.seqdiff = 0; 3560 if (th->th_flags & TH_SYN) { 3561 s->src.seqhi++; 3562 s->src.wscale = pf_get_wscale(m, off, 3563 th->th_off, pd->af); 3564 } 3565 s->src.max_win = MAX(ntohs(th->th_win), 1); 3566 if (s->src.wscale & PF_WSCALE_MASK) { 3567 /* Remove scale factor from initial window */ 3568 int win = s->src.max_win; 3569 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 3570 s->src.max_win = (win - 1) >> 3571 (s->src.wscale & PF_WSCALE_MASK); 3572 } 3573 if (th->th_flags & TH_FIN) 3574 s->src.seqhi++; 3575 s->dst.seqhi = 1; 3576 s->dst.max_win = 1; 3577 s->src.state = TCPS_SYN_SENT; 3578 s->dst.state = TCPS_CLOSED; 3579 s->timeout = PFTM_TCP_FIRST_PACKET; 3580 break; 3581 case IPPROTO_UDP: 3582 s->src.state = PFUDPS_SINGLE; 3583 s->dst.state = PFUDPS_NO_TRAFFIC; 3584 s->timeout = PFTM_UDP_FIRST_PACKET; 3585 break; 3586 case IPPROTO_ICMP: 3587 #ifdef INET6 3588 case IPPROTO_ICMPV6: 3589 #endif 3590 s->timeout = PFTM_ICMP_FIRST_PACKET; 3591 break; 3592 default: 3593 s->src.state = PFOTHERS_SINGLE; 3594 s->dst.state = PFOTHERS_NO_TRAFFIC; 3595 s->timeout = PFTM_OTHER_FIRST_PACKET; 3596 } 3597 3598 if (r->rt && r->rt != PF_FASTROUTE) { 3599 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) { 3600 REASON_SET(&reason, PFRES_MAPFAILED); 3601 pf_src_tree_remove_state(s); 3602 STATE_DEC_COUNTERS(s); 3603 uma_zfree(V_pf_state_z, s); 3604 goto csfailed; 3605 } 3606 s->rt_kif = r->rpool.cur->kif; 3607 } 3608 3609 s->creation = time_uptime; 3610 s->expire = time_uptime; 3611 3612 if (sn != NULL) 3613 s->src_node = sn; 3614 if (nsn != NULL) { 3615 /* XXX We only modify one side for now. */ 3616 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 3617 s->nat_src_node = nsn; 3618 } 3619 if (pd->proto == IPPROTO_TCP) { 3620 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m, 3621 off, pd, th, &s->src, &s->dst)) { 3622 REASON_SET(&reason, PFRES_MEMORY); 3623 pf_src_tree_remove_state(s); 3624 STATE_DEC_COUNTERS(s); 3625 uma_zfree(V_pf_state_z, s); 3626 return (PF_DROP); 3627 } 3628 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub && 3629 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 3630 &s->src, &s->dst, rewrite)) { 3631 /* This really shouldn't happen!!! */ 3632 DPFPRINTF(PF_DEBUG_URGENT, 3633 ("pf_normalize_tcp_stateful failed on first pkt")); 3634 pf_normalize_tcp_cleanup(s); 3635 pf_src_tree_remove_state(s); 3636 STATE_DEC_COUNTERS(s); 3637 uma_zfree(V_pf_state_z, s); 3638 return (PF_DROP); 3639 } 3640 } 3641 s->direction = pd->dir; 3642 3643 /* 3644 * sk/nk could already been setup by pf_get_translation(). 3645 */ 3646 if (nr == NULL) { 3647 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 3648 __func__, nr, sk, nk)); 3649 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 3650 if (sk == NULL) 3651 goto csfailed; 3652 nk = sk; 3653 } else 3654 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 3655 __func__, nr, sk, nk)); 3656 3657 /* Swap sk/nk for PF_OUT. */ 3658 if (pf_state_insert(BOUND_IFACE(r, kif), 3659 (pd->dir == PF_IN) ? sk : nk, 3660 (pd->dir == PF_IN) ? nk : sk, s)) { 3661 if (pd->proto == IPPROTO_TCP) 3662 pf_normalize_tcp_cleanup(s); 3663 REASON_SET(&reason, PFRES_STATEINS); 3664 pf_src_tree_remove_state(s); 3665 STATE_DEC_COUNTERS(s); 3666 uma_zfree(V_pf_state_z, s); 3667 return (PF_DROP); 3668 } else 3669 *sm = s; 3670 3671 if (tag > 0) 3672 s->tag = tag; 3673 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 3674 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 3675 s->src.state = PF_TCPS_PROXY_SRC; 3676 /* undo NAT changes, if they have taken place */ 3677 if (nr != NULL) { 3678 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 3679 if (pd->dir == PF_OUT) 3680 skt = s->key[PF_SK_STACK]; 3681 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 3682 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 3683 if (pd->sport) 3684 *pd->sport = skt->port[pd->sidx]; 3685 if (pd->dport) 3686 *pd->dport = skt->port[pd->didx]; 3687 if (pd->proto_sum) 3688 *pd->proto_sum = bproto_sum; 3689 if (pd->ip_sum) 3690 *pd->ip_sum = bip_sum; 3691 m_copyback(m, off, hdrlen, pd->hdr.any); 3692 } 3693 s->src.seqhi = htonl(arc4random()); 3694 /* Find mss option */ 3695 int rtid = M_GETFIB(m); 3696 mss = pf_get_mss(m, off, th->th_off, pd->af); 3697 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 3698 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 3699 s->src.mss = mss; 3700 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport, 3701 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 3702 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL); 3703 REASON_SET(&reason, PFRES_SYNPROXY); 3704 return (PF_SYNPROXY_DROP); 3705 } 3706 3707 return (PF_PASS); 3708 3709 csfailed: 3710 if (sk != NULL) 3711 uma_zfree(V_pf_state_key_z, sk); 3712 if (nk != NULL) 3713 uma_zfree(V_pf_state_key_z, nk); 3714 3715 if (sn != NULL) { 3716 struct pf_srchash *sh; 3717 3718 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 3719 PF_HASHROW_LOCK(sh); 3720 if (--sn->states == 0 && sn->expire == 0) { 3721 pf_unlink_src_node(sn); 3722 uma_zfree(V_pf_sources_z, sn); 3723 counter_u64_add( 3724 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 3725 } 3726 PF_HASHROW_UNLOCK(sh); 3727 } 3728 3729 if (nsn != sn && nsn != NULL) { 3730 struct pf_srchash *sh; 3731 3732 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)]; 3733 PF_HASHROW_LOCK(sh); 3734 if (--nsn->states == 0 && nsn->expire == 0) { 3735 pf_unlink_src_node(nsn); 3736 uma_zfree(V_pf_sources_z, nsn); 3737 counter_u64_add( 3738 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 3739 } 3740 PF_HASHROW_UNLOCK(sh); 3741 } 3742 3743 return (PF_DROP); 3744 } 3745 3746 static int 3747 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif, 3748 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am, 3749 struct pf_ruleset **rsm) 3750 { 3751 struct pf_rule *r, *a = NULL; 3752 struct pf_ruleset *ruleset = NULL; 3753 sa_family_t af = pd->af; 3754 u_short reason; 3755 int tag = -1; 3756 int asd = 0; 3757 int match = 0; 3758 struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3759 3760 PF_RULES_RASSERT(); 3761 3762 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3763 while (r != NULL) { 3764 r->evaluations++; 3765 if (pfi_kif_match(r->kif, kif) == r->ifnot) 3766 r = r->skip[PF_SKIP_IFP].ptr; 3767 else if (r->direction && r->direction != direction) 3768 r = r->skip[PF_SKIP_DIR].ptr; 3769 else if (r->af && r->af != af) 3770 r = r->skip[PF_SKIP_AF].ptr; 3771 else if (r->proto && r->proto != pd->proto) 3772 r = r->skip[PF_SKIP_PROTO].ptr; 3773 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 3774 r->src.neg, kif, M_GETFIB(m))) 3775 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3776 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 3777 r->dst.neg, NULL, M_GETFIB(m))) 3778 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3779 else if (r->tos && !(r->tos == pd->tos)) 3780 r = TAILQ_NEXT(r, entries); 3781 else if (r->os_fingerprint != PF_OSFP_ANY) 3782 r = TAILQ_NEXT(r, entries); 3783 else if (pd->proto == IPPROTO_UDP && 3784 (r->src.port_op || r->dst.port_op)) 3785 r = TAILQ_NEXT(r, entries); 3786 else if (pd->proto == IPPROTO_TCP && 3787 (r->src.port_op || r->dst.port_op || r->flagset)) 3788 r = TAILQ_NEXT(r, entries); 3789 else if ((pd->proto == IPPROTO_ICMP || 3790 pd->proto == IPPROTO_ICMPV6) && 3791 (r->type || r->code)) 3792 r = TAILQ_NEXT(r, entries); 3793 else if (r->prob && r->prob <= 3794 (arc4random() % (UINT_MAX - 1) + 1)) 3795 r = TAILQ_NEXT(r, entries); 3796 else if (r->match_tag && !pf_match_tag(m, r, &tag, 3797 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 3798 r = TAILQ_NEXT(r, entries); 3799 else { 3800 if (r->anchor == NULL) { 3801 match = 1; 3802 *rm = r; 3803 *am = a; 3804 *rsm = ruleset; 3805 if ((*rm)->quick) 3806 break; 3807 r = TAILQ_NEXT(r, entries); 3808 } else 3809 pf_step_into_anchor(anchor_stack, &asd, 3810 &ruleset, PF_RULESET_FILTER, &r, &a, 3811 &match); 3812 } 3813 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 3814 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 3815 break; 3816 } 3817 r = *rm; 3818 a = *am; 3819 ruleset = *rsm; 3820 3821 REASON_SET(&reason, PFRES_MATCH); 3822 3823 if (r->log) 3824 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd, 3825 1); 3826 3827 if (r->action != PF_PASS) 3828 return (PF_DROP); 3829 3830 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 3831 REASON_SET(&reason, PFRES_MEMORY); 3832 return (PF_DROP); 3833 } 3834 3835 return (PF_PASS); 3836 } 3837 3838 static int 3839 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst, 3840 struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off, 3841 struct pf_pdesc *pd, u_short *reason, int *copyback) 3842 { 3843 struct tcphdr *th = pd->hdr.tcp; 3844 u_int16_t win = ntohs(th->th_win); 3845 u_int32_t ack, end, seq, orig_seq; 3846 u_int8_t sws, dws; 3847 int ackskew; 3848 3849 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 3850 sws = src->wscale & PF_WSCALE_MASK; 3851 dws = dst->wscale & PF_WSCALE_MASK; 3852 } else 3853 sws = dws = 0; 3854 3855 /* 3856 * Sequence tracking algorithm from Guido van Rooij's paper: 3857 * http://www.madison-gurkha.com/publications/tcp_filtering/ 3858 * tcp_filtering.ps 3859 */ 3860 3861 orig_seq = seq = ntohl(th->th_seq); 3862 if (src->seqlo == 0) { 3863 /* First packet from this end. Set its state */ 3864 3865 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) && 3866 src->scrub == NULL) { 3867 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 3868 REASON_SET(reason, PFRES_MEMORY); 3869 return (PF_DROP); 3870 } 3871 } 3872 3873 /* Deferred generation of sequence number modulator */ 3874 if (dst->seqdiff && !src->seqdiff) { 3875 /* use random iss for the TCP server */ 3876 while ((src->seqdiff = arc4random() - seq) == 0) 3877 ; 3878 ack = ntohl(th->th_ack) - dst->seqdiff; 3879 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 3880 src->seqdiff), 0); 3881 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 3882 *copyback = 1; 3883 } else { 3884 ack = ntohl(th->th_ack); 3885 } 3886 3887 end = seq + pd->p_len; 3888 if (th->th_flags & TH_SYN) { 3889 end++; 3890 if (dst->wscale & PF_WSCALE_FLAG) { 3891 src->wscale = pf_get_wscale(m, off, th->th_off, 3892 pd->af); 3893 if (src->wscale & PF_WSCALE_FLAG) { 3894 /* Remove scale factor from initial 3895 * window */ 3896 sws = src->wscale & PF_WSCALE_MASK; 3897 win = ((u_int32_t)win + (1 << sws) - 1) 3898 >> sws; 3899 dws = dst->wscale & PF_WSCALE_MASK; 3900 } else { 3901 /* fixup other window */ 3902 dst->max_win <<= dst->wscale & 3903 PF_WSCALE_MASK; 3904 /* in case of a retrans SYN|ACK */ 3905 dst->wscale = 0; 3906 } 3907 } 3908 } 3909 if (th->th_flags & TH_FIN) 3910 end++; 3911 3912 src->seqlo = seq; 3913 if (src->state < TCPS_SYN_SENT) 3914 src->state = TCPS_SYN_SENT; 3915 3916 /* 3917 * May need to slide the window (seqhi may have been set by 3918 * the crappy stack check or if we picked up the connection 3919 * after establishment) 3920 */ 3921 if (src->seqhi == 1 || 3922 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 3923 src->seqhi = end + MAX(1, dst->max_win << dws); 3924 if (win > src->max_win) 3925 src->max_win = win; 3926 3927 } else { 3928 ack = ntohl(th->th_ack) - dst->seqdiff; 3929 if (src->seqdiff) { 3930 /* Modulate sequence numbers */ 3931 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 3932 src->seqdiff), 0); 3933 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 3934 *copyback = 1; 3935 } 3936 end = seq + pd->p_len; 3937 if (th->th_flags & TH_SYN) 3938 end++; 3939 if (th->th_flags & TH_FIN) 3940 end++; 3941 } 3942 3943 if ((th->th_flags & TH_ACK) == 0) { 3944 /* Let it pass through the ack skew check */ 3945 ack = dst->seqlo; 3946 } else if ((ack == 0 && 3947 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 3948 /* broken tcp stacks do not set ack */ 3949 (dst->state < TCPS_SYN_SENT)) { 3950 /* 3951 * Many stacks (ours included) will set the ACK number in an 3952 * FIN|ACK if the SYN times out -- no sequence to ACK. 3953 */ 3954 ack = dst->seqlo; 3955 } 3956 3957 if (seq == end) { 3958 /* Ease sequencing restrictions on no data packets */ 3959 seq = src->seqlo; 3960 end = seq; 3961 } 3962 3963 ackskew = dst->seqlo - ack; 3964 3965 3966 /* 3967 * Need to demodulate the sequence numbers in any TCP SACK options 3968 * (Selective ACK). We could optionally validate the SACK values 3969 * against the current ACK window, either forwards or backwards, but 3970 * I'm not confident that SACK has been implemented properly 3971 * everywhere. It wouldn't surprise me if several stacks accidently 3972 * SACK too far backwards of previously ACKed data. There really aren't 3973 * any security implications of bad SACKing unless the target stack 3974 * doesn't validate the option length correctly. Someone trying to 3975 * spoof into a TCP connection won't bother blindly sending SACK 3976 * options anyway. 3977 */ 3978 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 3979 if (pf_modulate_sack(m, off, pd, th, dst)) 3980 *copyback = 1; 3981 } 3982 3983 3984 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 3985 if (SEQ_GEQ(src->seqhi, end) && 3986 /* Last octet inside other's window space */ 3987 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 3988 /* Retrans: not more than one window back */ 3989 (ackskew >= -MAXACKWINDOW) && 3990 /* Acking not more than one reassembled fragment backwards */ 3991 (ackskew <= (MAXACKWINDOW << sws)) && 3992 /* Acking not more than one window forward */ 3993 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 3994 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 3995 (pd->flags & PFDESC_IP_REAS) == 0)) { 3996 /* Require an exact/+1 sequence match on resets when possible */ 3997 3998 if (dst->scrub || src->scrub) { 3999 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4000 *state, src, dst, copyback)) 4001 return (PF_DROP); 4002 } 4003 4004 /* update max window */ 4005 if (src->max_win < win) 4006 src->max_win = win; 4007 /* synchronize sequencing */ 4008 if (SEQ_GT(end, src->seqlo)) 4009 src->seqlo = end; 4010 /* slide the window of what the other end can send */ 4011 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4012 dst->seqhi = ack + MAX((win << sws), 1); 4013 4014 4015 /* update states */ 4016 if (th->th_flags & TH_SYN) 4017 if (src->state < TCPS_SYN_SENT) 4018 src->state = TCPS_SYN_SENT; 4019 if (th->th_flags & TH_FIN) 4020 if (src->state < TCPS_CLOSING) 4021 src->state = TCPS_CLOSING; 4022 if (th->th_flags & TH_ACK) { 4023 if (dst->state == TCPS_SYN_SENT) { 4024 dst->state = TCPS_ESTABLISHED; 4025 if (src->state == TCPS_ESTABLISHED && 4026 (*state)->src_node != NULL && 4027 pf_src_connlimit(state)) { 4028 REASON_SET(reason, PFRES_SRCLIMIT); 4029 return (PF_DROP); 4030 } 4031 } else if (dst->state == TCPS_CLOSING) 4032 dst->state = TCPS_FIN_WAIT_2; 4033 } 4034 if (th->th_flags & TH_RST) 4035 src->state = dst->state = TCPS_TIME_WAIT; 4036 4037 /* update expire time */ 4038 (*state)->expire = time_uptime; 4039 if (src->state >= TCPS_FIN_WAIT_2 && 4040 dst->state >= TCPS_FIN_WAIT_2) 4041 (*state)->timeout = PFTM_TCP_CLOSED; 4042 else if (src->state >= TCPS_CLOSING && 4043 dst->state >= TCPS_CLOSING) 4044 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4045 else if (src->state < TCPS_ESTABLISHED || 4046 dst->state < TCPS_ESTABLISHED) 4047 (*state)->timeout = PFTM_TCP_OPENING; 4048 else if (src->state >= TCPS_CLOSING || 4049 dst->state >= TCPS_CLOSING) 4050 (*state)->timeout = PFTM_TCP_CLOSING; 4051 else 4052 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4053 4054 /* Fall through to PASS packet */ 4055 4056 } else if ((dst->state < TCPS_SYN_SENT || 4057 dst->state >= TCPS_FIN_WAIT_2 || 4058 src->state >= TCPS_FIN_WAIT_2) && 4059 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 4060 /* Within a window forward of the originating packet */ 4061 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 4062 /* Within a window backward of the originating packet */ 4063 4064 /* 4065 * This currently handles three situations: 4066 * 1) Stupid stacks will shotgun SYNs before their peer 4067 * replies. 4068 * 2) When PF catches an already established stream (the 4069 * firewall rebooted, the state table was flushed, routes 4070 * changed...) 4071 * 3) Packets get funky immediately after the connection 4072 * closes (this should catch Solaris spurious ACK|FINs 4073 * that web servers like to spew after a close) 4074 * 4075 * This must be a little more careful than the above code 4076 * since packet floods will also be caught here. We don't 4077 * update the TTL here to mitigate the damage of a packet 4078 * flood and so the same code can handle awkward establishment 4079 * and a loosened connection close. 4080 * In the establishment case, a correct peer response will 4081 * validate the connection, go through the normal state code 4082 * and keep updating the state TTL. 4083 */ 4084 4085 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4086 printf("pf: loose state match: "); 4087 pf_print_state(*state); 4088 pf_print_flags(th->th_flags); 4089 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4090 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 4091 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 4092 (unsigned long long)(*state)->packets[1], 4093 pd->dir == PF_IN ? "in" : "out", 4094 pd->dir == (*state)->direction ? "fwd" : "rev"); 4095 } 4096 4097 if (dst->scrub || src->scrub) { 4098 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4099 *state, src, dst, copyback)) 4100 return (PF_DROP); 4101 } 4102 4103 /* update max window */ 4104 if (src->max_win < win) 4105 src->max_win = win; 4106 /* synchronize sequencing */ 4107 if (SEQ_GT(end, src->seqlo)) 4108 src->seqlo = end; 4109 /* slide the window of what the other end can send */ 4110 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4111 dst->seqhi = ack + MAX((win << sws), 1); 4112 4113 /* 4114 * Cannot set dst->seqhi here since this could be a shotgunned 4115 * SYN and not an already established connection. 4116 */ 4117 4118 if (th->th_flags & TH_FIN) 4119 if (src->state < TCPS_CLOSING) 4120 src->state = TCPS_CLOSING; 4121 if (th->th_flags & TH_RST) 4122 src->state = dst->state = TCPS_TIME_WAIT; 4123 4124 /* Fall through to PASS packet */ 4125 4126 } else { 4127 if ((*state)->dst.state == TCPS_SYN_SENT && 4128 (*state)->src.state == TCPS_SYN_SENT) { 4129 /* Send RST for state mismatches during handshake */ 4130 if (!(th->th_flags & TH_RST)) 4131 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4132 pd->dst, pd->src, th->th_dport, 4133 th->th_sport, ntohl(th->th_ack), 0, 4134 TH_RST, 0, 0, 4135 (*state)->rule.ptr->return_ttl, 1, 0, 4136 kif->pfik_ifp); 4137 src->seqlo = 0; 4138 src->seqhi = 1; 4139 src->max_win = 1; 4140 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 4141 printf("pf: BAD state: "); 4142 pf_print_state(*state); 4143 pf_print_flags(th->th_flags); 4144 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4145 "pkts=%llu:%llu dir=%s,%s\n", 4146 seq, orig_seq, ack, pd->p_len, ackskew, 4147 (unsigned long long)(*state)->packets[0], 4148 (unsigned long long)(*state)->packets[1], 4149 pd->dir == PF_IN ? "in" : "out", 4150 pd->dir == (*state)->direction ? "fwd" : "rev"); 4151 printf("pf: State failure on: %c %c %c %c | %c %c\n", 4152 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 4153 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 4154 ' ': '2', 4155 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 4156 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 4157 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 4158 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 4159 } 4160 REASON_SET(reason, PFRES_BADSTATE); 4161 return (PF_DROP); 4162 } 4163 4164 return (PF_PASS); 4165 } 4166 4167 static int 4168 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst, 4169 struct pf_state **state, struct pf_pdesc *pd, u_short *reason) 4170 { 4171 struct tcphdr *th = pd->hdr.tcp; 4172 4173 if (th->th_flags & TH_SYN) 4174 if (src->state < TCPS_SYN_SENT) 4175 src->state = TCPS_SYN_SENT; 4176 if (th->th_flags & TH_FIN) 4177 if (src->state < TCPS_CLOSING) 4178 src->state = TCPS_CLOSING; 4179 if (th->th_flags & TH_ACK) { 4180 if (dst->state == TCPS_SYN_SENT) { 4181 dst->state = TCPS_ESTABLISHED; 4182 if (src->state == TCPS_ESTABLISHED && 4183 (*state)->src_node != NULL && 4184 pf_src_connlimit(state)) { 4185 REASON_SET(reason, PFRES_SRCLIMIT); 4186 return (PF_DROP); 4187 } 4188 } else if (dst->state == TCPS_CLOSING) { 4189 dst->state = TCPS_FIN_WAIT_2; 4190 } else if (src->state == TCPS_SYN_SENT && 4191 dst->state < TCPS_SYN_SENT) { 4192 /* 4193 * Handle a special sloppy case where we only see one 4194 * half of the connection. If there is a ACK after 4195 * the initial SYN without ever seeing a packet from 4196 * the destination, set the connection to established. 4197 */ 4198 dst->state = src->state = TCPS_ESTABLISHED; 4199 if ((*state)->src_node != NULL && 4200 pf_src_connlimit(state)) { 4201 REASON_SET(reason, PFRES_SRCLIMIT); 4202 return (PF_DROP); 4203 } 4204 } else if (src->state == TCPS_CLOSING && 4205 dst->state == TCPS_ESTABLISHED && 4206 dst->seqlo == 0) { 4207 /* 4208 * Handle the closing of half connections where we 4209 * don't see the full bidirectional FIN/ACK+ACK 4210 * handshake. 4211 */ 4212 dst->state = TCPS_CLOSING; 4213 } 4214 } 4215 if (th->th_flags & TH_RST) 4216 src->state = dst->state = TCPS_TIME_WAIT; 4217 4218 /* update expire time */ 4219 (*state)->expire = time_uptime; 4220 if (src->state >= TCPS_FIN_WAIT_2 && 4221 dst->state >= TCPS_FIN_WAIT_2) 4222 (*state)->timeout = PFTM_TCP_CLOSED; 4223 else if (src->state >= TCPS_CLOSING && 4224 dst->state >= TCPS_CLOSING) 4225 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4226 else if (src->state < TCPS_ESTABLISHED || 4227 dst->state < TCPS_ESTABLISHED) 4228 (*state)->timeout = PFTM_TCP_OPENING; 4229 else if (src->state >= TCPS_CLOSING || 4230 dst->state >= TCPS_CLOSING) 4231 (*state)->timeout = PFTM_TCP_CLOSING; 4232 else 4233 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4234 4235 return (PF_PASS); 4236 } 4237 4238 static int 4239 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif, 4240 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 4241 u_short *reason) 4242 { 4243 struct pf_state_key_cmp key; 4244 struct tcphdr *th = pd->hdr.tcp; 4245 int copyback = 0; 4246 struct pf_state_peer *src, *dst; 4247 struct pf_state_key *sk; 4248 4249 bzero(&key, sizeof(key)); 4250 key.af = pd->af; 4251 key.proto = IPPROTO_TCP; 4252 if (direction == PF_IN) { /* wire side, straight */ 4253 PF_ACPY(&key.addr[0], pd->src, key.af); 4254 PF_ACPY(&key.addr[1], pd->dst, key.af); 4255 key.port[0] = th->th_sport; 4256 key.port[1] = th->th_dport; 4257 } else { /* stack side, reverse */ 4258 PF_ACPY(&key.addr[1], pd->src, key.af); 4259 PF_ACPY(&key.addr[0], pd->dst, key.af); 4260 key.port[1] = th->th_sport; 4261 key.port[0] = th->th_dport; 4262 } 4263 4264 STATE_LOOKUP(kif, &key, direction, *state, pd); 4265 4266 if (direction == (*state)->direction) { 4267 src = &(*state)->src; 4268 dst = &(*state)->dst; 4269 } else { 4270 src = &(*state)->dst; 4271 dst = &(*state)->src; 4272 } 4273 4274 sk = (*state)->key[pd->didx]; 4275 4276 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 4277 if (direction != (*state)->direction) { 4278 REASON_SET(reason, PFRES_SYNPROXY); 4279 return (PF_SYNPROXY_DROP); 4280 } 4281 if (th->th_flags & TH_SYN) { 4282 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 4283 REASON_SET(reason, PFRES_SYNPROXY); 4284 return (PF_DROP); 4285 } 4286 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, 4287 pd->src, th->th_dport, th->th_sport, 4288 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 4289 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL); 4290 REASON_SET(reason, PFRES_SYNPROXY); 4291 return (PF_SYNPROXY_DROP); 4292 } else if (!(th->th_flags & TH_ACK) || 4293 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4294 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4295 REASON_SET(reason, PFRES_SYNPROXY); 4296 return (PF_DROP); 4297 } else if ((*state)->src_node != NULL && 4298 pf_src_connlimit(state)) { 4299 REASON_SET(reason, PFRES_SRCLIMIT); 4300 return (PF_DROP); 4301 } else 4302 (*state)->src.state = PF_TCPS_PROXY_DST; 4303 } 4304 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 4305 if (direction == (*state)->direction) { 4306 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 4307 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4308 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4309 REASON_SET(reason, PFRES_SYNPROXY); 4310 return (PF_DROP); 4311 } 4312 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 4313 if ((*state)->dst.seqhi == 1) 4314 (*state)->dst.seqhi = htonl(arc4random()); 4315 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4316 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4317 sk->port[pd->sidx], sk->port[pd->didx], 4318 (*state)->dst.seqhi, 0, TH_SYN, 0, 4319 (*state)->src.mss, 0, 0, (*state)->tag, NULL); 4320 REASON_SET(reason, PFRES_SYNPROXY); 4321 return (PF_SYNPROXY_DROP); 4322 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 4323 (TH_SYN|TH_ACK)) || 4324 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 4325 REASON_SET(reason, PFRES_SYNPROXY); 4326 return (PF_DROP); 4327 } else { 4328 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 4329 (*state)->dst.seqlo = ntohl(th->th_seq); 4330 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, 4331 pd->src, th->th_dport, th->th_sport, 4332 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 4333 TH_ACK, (*state)->src.max_win, 0, 0, 0, 4334 (*state)->tag, NULL); 4335 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4336 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4337 sk->port[pd->sidx], sk->port[pd->didx], 4338 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 4339 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL); 4340 (*state)->src.seqdiff = (*state)->dst.seqhi - 4341 (*state)->src.seqlo; 4342 (*state)->dst.seqdiff = (*state)->src.seqhi - 4343 (*state)->dst.seqlo; 4344 (*state)->src.seqhi = (*state)->src.seqlo + 4345 (*state)->dst.max_win; 4346 (*state)->dst.seqhi = (*state)->dst.seqlo + 4347 (*state)->src.max_win; 4348 (*state)->src.wscale = (*state)->dst.wscale = 0; 4349 (*state)->src.state = (*state)->dst.state = 4350 TCPS_ESTABLISHED; 4351 REASON_SET(reason, PFRES_SYNPROXY); 4352 return (PF_SYNPROXY_DROP); 4353 } 4354 } 4355 4356 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) && 4357 dst->state >= TCPS_FIN_WAIT_2 && 4358 src->state >= TCPS_FIN_WAIT_2) { 4359 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4360 printf("pf: state reuse "); 4361 pf_print_state(*state); 4362 pf_print_flags(th->th_flags); 4363 printf("\n"); 4364 } 4365 /* XXX make sure it's the same direction ?? */ 4366 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; 4367 pf_unlink_state(*state, PF_ENTER_LOCKED); 4368 *state = NULL; 4369 return (PF_DROP); 4370 } 4371 4372 if ((*state)->state_flags & PFSTATE_SLOPPY) { 4373 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP) 4374 return (PF_DROP); 4375 } else { 4376 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason, 4377 ©back) == PF_DROP) 4378 return (PF_DROP); 4379 } 4380 4381 /* translate source/destination address, if necessary */ 4382 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4383 struct pf_state_key *nk = (*state)->key[pd->didx]; 4384 4385 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4386 nk->port[pd->sidx] != th->th_sport) 4387 pf_change_ap(m, pd->src, &th->th_sport, 4388 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 4389 nk->port[pd->sidx], 0, pd->af); 4390 4391 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 4392 nk->port[pd->didx] != th->th_dport) 4393 pf_change_ap(m, pd->dst, &th->th_dport, 4394 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 4395 nk->port[pd->didx], 0, pd->af); 4396 copyback = 1; 4397 } 4398 4399 /* Copyback sequence modulation or stateful scrub changes if needed */ 4400 if (copyback) 4401 m_copyback(m, off, sizeof(*th), (caddr_t)th); 4402 4403 return (PF_PASS); 4404 } 4405 4406 static int 4407 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif, 4408 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 4409 { 4410 struct pf_state_peer *src, *dst; 4411 struct pf_state_key_cmp key; 4412 struct udphdr *uh = pd->hdr.udp; 4413 4414 bzero(&key, sizeof(key)); 4415 key.af = pd->af; 4416 key.proto = IPPROTO_UDP; 4417 if (direction == PF_IN) { /* wire side, straight */ 4418 PF_ACPY(&key.addr[0], pd->src, key.af); 4419 PF_ACPY(&key.addr[1], pd->dst, key.af); 4420 key.port[0] = uh->uh_sport; 4421 key.port[1] = uh->uh_dport; 4422 } else { /* stack side, reverse */ 4423 PF_ACPY(&key.addr[1], pd->src, key.af); 4424 PF_ACPY(&key.addr[0], pd->dst, key.af); 4425 key.port[1] = uh->uh_sport; 4426 key.port[0] = uh->uh_dport; 4427 } 4428 4429 STATE_LOOKUP(kif, &key, direction, *state, pd); 4430 4431 if (direction == (*state)->direction) { 4432 src = &(*state)->src; 4433 dst = &(*state)->dst; 4434 } else { 4435 src = &(*state)->dst; 4436 dst = &(*state)->src; 4437 } 4438 4439 /* update states */ 4440 if (src->state < PFUDPS_SINGLE) 4441 src->state = PFUDPS_SINGLE; 4442 if (dst->state == PFUDPS_SINGLE) 4443 dst->state = PFUDPS_MULTIPLE; 4444 4445 /* update expire time */ 4446 (*state)->expire = time_uptime; 4447 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 4448 (*state)->timeout = PFTM_UDP_MULTIPLE; 4449 else 4450 (*state)->timeout = PFTM_UDP_SINGLE; 4451 4452 /* translate source/destination address, if necessary */ 4453 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4454 struct pf_state_key *nk = (*state)->key[pd->didx]; 4455 4456 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4457 nk->port[pd->sidx] != uh->uh_sport) 4458 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 4459 &uh->uh_sum, &nk->addr[pd->sidx], 4460 nk->port[pd->sidx], 1, pd->af); 4461 4462 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 4463 nk->port[pd->didx] != uh->uh_dport) 4464 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 4465 &uh->uh_sum, &nk->addr[pd->didx], 4466 nk->port[pd->didx], 1, pd->af); 4467 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 4468 } 4469 4470 return (PF_PASS); 4471 } 4472 4473 static int 4474 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif, 4475 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 4476 { 4477 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 4478 u_int16_t icmpid = 0, *icmpsum; 4479 u_int8_t icmptype; 4480 int state_icmp = 0; 4481 struct pf_state_key_cmp key; 4482 4483 bzero(&key, sizeof(key)); 4484 switch (pd->proto) { 4485 #ifdef INET 4486 case IPPROTO_ICMP: 4487 icmptype = pd->hdr.icmp->icmp_type; 4488 icmpid = pd->hdr.icmp->icmp_id; 4489 icmpsum = &pd->hdr.icmp->icmp_cksum; 4490 4491 if (icmptype == ICMP_UNREACH || 4492 icmptype == ICMP_SOURCEQUENCH || 4493 icmptype == ICMP_REDIRECT || 4494 icmptype == ICMP_TIMXCEED || 4495 icmptype == ICMP_PARAMPROB) 4496 state_icmp++; 4497 break; 4498 #endif /* INET */ 4499 #ifdef INET6 4500 case IPPROTO_ICMPV6: 4501 icmptype = pd->hdr.icmp6->icmp6_type; 4502 icmpid = pd->hdr.icmp6->icmp6_id; 4503 icmpsum = &pd->hdr.icmp6->icmp6_cksum; 4504 4505 if (icmptype == ICMP6_DST_UNREACH || 4506 icmptype == ICMP6_PACKET_TOO_BIG || 4507 icmptype == ICMP6_TIME_EXCEEDED || 4508 icmptype == ICMP6_PARAM_PROB) 4509 state_icmp++; 4510 break; 4511 #endif /* INET6 */ 4512 } 4513 4514 if (!state_icmp) { 4515 4516 /* 4517 * ICMP query/reply message not related to a TCP/UDP packet. 4518 * Search for an ICMP state. 4519 */ 4520 key.af = pd->af; 4521 key.proto = pd->proto; 4522 key.port[0] = key.port[1] = icmpid; 4523 if (direction == PF_IN) { /* wire side, straight */ 4524 PF_ACPY(&key.addr[0], pd->src, key.af); 4525 PF_ACPY(&key.addr[1], pd->dst, key.af); 4526 } else { /* stack side, reverse */ 4527 PF_ACPY(&key.addr[1], pd->src, key.af); 4528 PF_ACPY(&key.addr[0], pd->dst, key.af); 4529 } 4530 4531 STATE_LOOKUP(kif, &key, direction, *state, pd); 4532 4533 (*state)->expire = time_uptime; 4534 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 4535 4536 /* translate source/destination address, if necessary */ 4537 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4538 struct pf_state_key *nk = (*state)->key[pd->didx]; 4539 4540 switch (pd->af) { 4541 #ifdef INET 4542 case AF_INET: 4543 if (PF_ANEQ(pd->src, 4544 &nk->addr[pd->sidx], AF_INET)) 4545 pf_change_a(&saddr->v4.s_addr, 4546 pd->ip_sum, 4547 nk->addr[pd->sidx].v4.s_addr, 0); 4548 4549 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 4550 AF_INET)) 4551 pf_change_a(&daddr->v4.s_addr, 4552 pd->ip_sum, 4553 nk->addr[pd->didx].v4.s_addr, 0); 4554 4555 if (nk->port[0] != 4556 pd->hdr.icmp->icmp_id) { 4557 pd->hdr.icmp->icmp_cksum = 4558 pf_cksum_fixup( 4559 pd->hdr.icmp->icmp_cksum, icmpid, 4560 nk->port[pd->sidx], 0); 4561 pd->hdr.icmp->icmp_id = 4562 nk->port[pd->sidx]; 4563 } 4564 4565 m_copyback(m, off, ICMP_MINLEN, 4566 (caddr_t )pd->hdr.icmp); 4567 break; 4568 #endif /* INET */ 4569 #ifdef INET6 4570 case AF_INET6: 4571 if (PF_ANEQ(pd->src, 4572 &nk->addr[pd->sidx], AF_INET6)) 4573 pf_change_a6(saddr, 4574 &pd->hdr.icmp6->icmp6_cksum, 4575 &nk->addr[pd->sidx], 0); 4576 4577 if (PF_ANEQ(pd->dst, 4578 &nk->addr[pd->didx], AF_INET6)) 4579 pf_change_a6(daddr, 4580 &pd->hdr.icmp6->icmp6_cksum, 4581 &nk->addr[pd->didx], 0); 4582 4583 m_copyback(m, off, sizeof(struct icmp6_hdr), 4584 (caddr_t )pd->hdr.icmp6); 4585 break; 4586 #endif /* INET6 */ 4587 } 4588 } 4589 return (PF_PASS); 4590 4591 } else { 4592 /* 4593 * ICMP error message in response to a TCP/UDP packet. 4594 * Extract the inner TCP/UDP header and search for that state. 4595 */ 4596 4597 struct pf_pdesc pd2; 4598 bzero(&pd2, sizeof pd2); 4599 #ifdef INET 4600 struct ip h2; 4601 #endif /* INET */ 4602 #ifdef INET6 4603 struct ip6_hdr h2_6; 4604 int terminal = 0; 4605 #endif /* INET6 */ 4606 int ipoff2 = 0; 4607 int off2 = 0; 4608 4609 pd2.af = pd->af; 4610 /* Payload packet is from the opposite direction. */ 4611 pd2.sidx = (direction == PF_IN) ? 1 : 0; 4612 pd2.didx = (direction == PF_IN) ? 0 : 1; 4613 switch (pd->af) { 4614 #ifdef INET 4615 case AF_INET: 4616 /* offset of h2 in mbuf chain */ 4617 ipoff2 = off + ICMP_MINLEN; 4618 4619 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 4620 NULL, reason, pd2.af)) { 4621 DPFPRINTF(PF_DEBUG_MISC, 4622 ("pf: ICMP error message too short " 4623 "(ip)\n")); 4624 return (PF_DROP); 4625 } 4626 /* 4627 * ICMP error messages don't refer to non-first 4628 * fragments 4629 */ 4630 if (h2.ip_off & htons(IP_OFFMASK)) { 4631 REASON_SET(reason, PFRES_FRAG); 4632 return (PF_DROP); 4633 } 4634 4635 /* offset of protocol header that follows h2 */ 4636 off2 = ipoff2 + (h2.ip_hl << 2); 4637 4638 pd2.proto = h2.ip_p; 4639 pd2.src = (struct pf_addr *)&h2.ip_src; 4640 pd2.dst = (struct pf_addr *)&h2.ip_dst; 4641 pd2.ip_sum = &h2.ip_sum; 4642 break; 4643 #endif /* INET */ 4644 #ifdef INET6 4645 case AF_INET6: 4646 ipoff2 = off + sizeof(struct icmp6_hdr); 4647 4648 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 4649 NULL, reason, pd2.af)) { 4650 DPFPRINTF(PF_DEBUG_MISC, 4651 ("pf: ICMP error message too short " 4652 "(ip6)\n")); 4653 return (PF_DROP); 4654 } 4655 pd2.proto = h2_6.ip6_nxt; 4656 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 4657 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 4658 pd2.ip_sum = NULL; 4659 off2 = ipoff2 + sizeof(h2_6); 4660 do { 4661 switch (pd2.proto) { 4662 case IPPROTO_FRAGMENT: 4663 /* 4664 * ICMPv6 error messages for 4665 * non-first fragments 4666 */ 4667 REASON_SET(reason, PFRES_FRAG); 4668 return (PF_DROP); 4669 case IPPROTO_AH: 4670 case IPPROTO_HOPOPTS: 4671 case IPPROTO_ROUTING: 4672 case IPPROTO_DSTOPTS: { 4673 /* get next header and header length */ 4674 struct ip6_ext opt6; 4675 4676 if (!pf_pull_hdr(m, off2, &opt6, 4677 sizeof(opt6), NULL, reason, 4678 pd2.af)) { 4679 DPFPRINTF(PF_DEBUG_MISC, 4680 ("pf: ICMPv6 short opt\n")); 4681 return (PF_DROP); 4682 } 4683 if (pd2.proto == IPPROTO_AH) 4684 off2 += (opt6.ip6e_len + 2) * 4; 4685 else 4686 off2 += (opt6.ip6e_len + 1) * 8; 4687 pd2.proto = opt6.ip6e_nxt; 4688 /* goto the next header */ 4689 break; 4690 } 4691 default: 4692 terminal++; 4693 break; 4694 } 4695 } while (!terminal); 4696 break; 4697 #endif /* INET6 */ 4698 } 4699 4700 switch (pd2.proto) { 4701 case IPPROTO_TCP: { 4702 struct tcphdr th; 4703 u_int32_t seq; 4704 struct pf_state_peer *src, *dst; 4705 u_int8_t dws; 4706 int copyback = 0; 4707 4708 /* 4709 * Only the first 8 bytes of the TCP header can be 4710 * expected. Don't access any TCP header fields after 4711 * th_seq, an ackskew test is not possible. 4712 */ 4713 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 4714 pd2.af)) { 4715 DPFPRINTF(PF_DEBUG_MISC, 4716 ("pf: ICMP error message too short " 4717 "(tcp)\n")); 4718 return (PF_DROP); 4719 } 4720 4721 key.af = pd2.af; 4722 key.proto = IPPROTO_TCP; 4723 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4724 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4725 key.port[pd2.sidx] = th.th_sport; 4726 key.port[pd2.didx] = th.th_dport; 4727 4728 STATE_LOOKUP(kif, &key, direction, *state, pd); 4729 4730 if (direction == (*state)->direction) { 4731 src = &(*state)->dst; 4732 dst = &(*state)->src; 4733 } else { 4734 src = &(*state)->src; 4735 dst = &(*state)->dst; 4736 } 4737 4738 if (src->wscale && dst->wscale) 4739 dws = dst->wscale & PF_WSCALE_MASK; 4740 else 4741 dws = 0; 4742 4743 /* Demodulate sequence number */ 4744 seq = ntohl(th.th_seq) - src->seqdiff; 4745 if (src->seqdiff) { 4746 pf_change_a(&th.th_seq, icmpsum, 4747 htonl(seq), 0); 4748 copyback = 1; 4749 } 4750 4751 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 4752 (!SEQ_GEQ(src->seqhi, seq) || 4753 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 4754 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4755 printf("pf: BAD ICMP %d:%d ", 4756 icmptype, pd->hdr.icmp->icmp_code); 4757 pf_print_host(pd->src, 0, pd->af); 4758 printf(" -> "); 4759 pf_print_host(pd->dst, 0, pd->af); 4760 printf(" state: "); 4761 pf_print_state(*state); 4762 printf(" seq=%u\n", seq); 4763 } 4764 REASON_SET(reason, PFRES_BADSTATE); 4765 return (PF_DROP); 4766 } else { 4767 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4768 printf("pf: OK ICMP %d:%d ", 4769 icmptype, pd->hdr.icmp->icmp_code); 4770 pf_print_host(pd->src, 0, pd->af); 4771 printf(" -> "); 4772 pf_print_host(pd->dst, 0, pd->af); 4773 printf(" state: "); 4774 pf_print_state(*state); 4775 printf(" seq=%u\n", seq); 4776 } 4777 } 4778 4779 /* translate source/destination address, if necessary */ 4780 if ((*state)->key[PF_SK_WIRE] != 4781 (*state)->key[PF_SK_STACK]) { 4782 struct pf_state_key *nk = 4783 (*state)->key[pd->didx]; 4784 4785 if (PF_ANEQ(pd2.src, 4786 &nk->addr[pd2.sidx], pd2.af) || 4787 nk->port[pd2.sidx] != th.th_sport) 4788 pf_change_icmp(pd2.src, &th.th_sport, 4789 daddr, &nk->addr[pd2.sidx], 4790 nk->port[pd2.sidx], NULL, 4791 pd2.ip_sum, icmpsum, 4792 pd->ip_sum, 0, pd2.af); 4793 4794 if (PF_ANEQ(pd2.dst, 4795 &nk->addr[pd2.didx], pd2.af) || 4796 nk->port[pd2.didx] != th.th_dport) 4797 pf_change_icmp(pd2.dst, &th.th_dport, 4798 NULL, /* XXX Inbound NAT? */ 4799 &nk->addr[pd2.didx], 4800 nk->port[pd2.didx], NULL, 4801 pd2.ip_sum, icmpsum, 4802 pd->ip_sum, 0, pd2.af); 4803 copyback = 1; 4804 } 4805 4806 if (copyback) { 4807 switch (pd2.af) { 4808 #ifdef INET 4809 case AF_INET: 4810 m_copyback(m, off, ICMP_MINLEN, 4811 (caddr_t )pd->hdr.icmp); 4812 m_copyback(m, ipoff2, sizeof(h2), 4813 (caddr_t )&h2); 4814 break; 4815 #endif /* INET */ 4816 #ifdef INET6 4817 case AF_INET6: 4818 m_copyback(m, off, 4819 sizeof(struct icmp6_hdr), 4820 (caddr_t )pd->hdr.icmp6); 4821 m_copyback(m, ipoff2, sizeof(h2_6), 4822 (caddr_t )&h2_6); 4823 break; 4824 #endif /* INET6 */ 4825 } 4826 m_copyback(m, off2, 8, (caddr_t)&th); 4827 } 4828 4829 return (PF_PASS); 4830 break; 4831 } 4832 case IPPROTO_UDP: { 4833 struct udphdr uh; 4834 4835 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 4836 NULL, reason, pd2.af)) { 4837 DPFPRINTF(PF_DEBUG_MISC, 4838 ("pf: ICMP error message too short " 4839 "(udp)\n")); 4840 return (PF_DROP); 4841 } 4842 4843 key.af = pd2.af; 4844 key.proto = IPPROTO_UDP; 4845 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4846 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4847 key.port[pd2.sidx] = uh.uh_sport; 4848 key.port[pd2.didx] = uh.uh_dport; 4849 4850 STATE_LOOKUP(kif, &key, direction, *state, pd); 4851 4852 /* translate source/destination address, if necessary */ 4853 if ((*state)->key[PF_SK_WIRE] != 4854 (*state)->key[PF_SK_STACK]) { 4855 struct pf_state_key *nk = 4856 (*state)->key[pd->didx]; 4857 4858 if (PF_ANEQ(pd2.src, 4859 &nk->addr[pd2.sidx], pd2.af) || 4860 nk->port[pd2.sidx] != uh.uh_sport) 4861 pf_change_icmp(pd2.src, &uh.uh_sport, 4862 daddr, &nk->addr[pd2.sidx], 4863 nk->port[pd2.sidx], &uh.uh_sum, 4864 pd2.ip_sum, icmpsum, 4865 pd->ip_sum, 1, pd2.af); 4866 4867 if (PF_ANEQ(pd2.dst, 4868 &nk->addr[pd2.didx], pd2.af) || 4869 nk->port[pd2.didx] != uh.uh_dport) 4870 pf_change_icmp(pd2.dst, &uh.uh_dport, 4871 NULL, /* XXX Inbound NAT? */ 4872 &nk->addr[pd2.didx], 4873 nk->port[pd2.didx], &uh.uh_sum, 4874 pd2.ip_sum, icmpsum, 4875 pd->ip_sum, 1, pd2.af); 4876 4877 switch (pd2.af) { 4878 #ifdef INET 4879 case AF_INET: 4880 m_copyback(m, off, ICMP_MINLEN, 4881 (caddr_t )pd->hdr.icmp); 4882 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 4883 break; 4884 #endif /* INET */ 4885 #ifdef INET6 4886 case AF_INET6: 4887 m_copyback(m, off, 4888 sizeof(struct icmp6_hdr), 4889 (caddr_t )pd->hdr.icmp6); 4890 m_copyback(m, ipoff2, sizeof(h2_6), 4891 (caddr_t )&h2_6); 4892 break; 4893 #endif /* INET6 */ 4894 } 4895 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 4896 } 4897 return (PF_PASS); 4898 break; 4899 } 4900 #ifdef INET 4901 case IPPROTO_ICMP: { 4902 struct icmp iih; 4903 4904 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 4905 NULL, reason, pd2.af)) { 4906 DPFPRINTF(PF_DEBUG_MISC, 4907 ("pf: ICMP error message too short i" 4908 "(icmp)\n")); 4909 return (PF_DROP); 4910 } 4911 4912 key.af = pd2.af; 4913 key.proto = IPPROTO_ICMP; 4914 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4915 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4916 key.port[0] = key.port[1] = iih.icmp_id; 4917 4918 STATE_LOOKUP(kif, &key, direction, *state, pd); 4919 4920 /* translate source/destination address, if necessary */ 4921 if ((*state)->key[PF_SK_WIRE] != 4922 (*state)->key[PF_SK_STACK]) { 4923 struct pf_state_key *nk = 4924 (*state)->key[pd->didx]; 4925 4926 if (PF_ANEQ(pd2.src, 4927 &nk->addr[pd2.sidx], pd2.af) || 4928 nk->port[pd2.sidx] != iih.icmp_id) 4929 pf_change_icmp(pd2.src, &iih.icmp_id, 4930 daddr, &nk->addr[pd2.sidx], 4931 nk->port[pd2.sidx], NULL, 4932 pd2.ip_sum, icmpsum, 4933 pd->ip_sum, 0, AF_INET); 4934 4935 if (PF_ANEQ(pd2.dst, 4936 &nk->addr[pd2.didx], pd2.af) || 4937 nk->port[pd2.didx] != iih.icmp_id) 4938 pf_change_icmp(pd2.dst, &iih.icmp_id, 4939 NULL, /* XXX Inbound NAT? */ 4940 &nk->addr[pd2.didx], 4941 nk->port[pd2.didx], NULL, 4942 pd2.ip_sum, icmpsum, 4943 pd->ip_sum, 0, AF_INET); 4944 4945 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 4946 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 4947 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 4948 } 4949 return (PF_PASS); 4950 break; 4951 } 4952 #endif /* INET */ 4953 #ifdef INET6 4954 case IPPROTO_ICMPV6: { 4955 struct icmp6_hdr iih; 4956 4957 if (!pf_pull_hdr(m, off2, &iih, 4958 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 4959 DPFPRINTF(PF_DEBUG_MISC, 4960 ("pf: ICMP error message too short " 4961 "(icmp6)\n")); 4962 return (PF_DROP); 4963 } 4964 4965 key.af = pd2.af; 4966 key.proto = IPPROTO_ICMPV6; 4967 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4968 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4969 key.port[0] = key.port[1] = iih.icmp6_id; 4970 4971 STATE_LOOKUP(kif, &key, direction, *state, pd); 4972 4973 /* translate source/destination address, if necessary */ 4974 if ((*state)->key[PF_SK_WIRE] != 4975 (*state)->key[PF_SK_STACK]) { 4976 struct pf_state_key *nk = 4977 (*state)->key[pd->didx]; 4978 4979 if (PF_ANEQ(pd2.src, 4980 &nk->addr[pd2.sidx], pd2.af) || 4981 nk->port[pd2.sidx] != iih.icmp6_id) 4982 pf_change_icmp(pd2.src, &iih.icmp6_id, 4983 daddr, &nk->addr[pd2.sidx], 4984 nk->port[pd2.sidx], NULL, 4985 pd2.ip_sum, icmpsum, 4986 pd->ip_sum, 0, AF_INET6); 4987 4988 if (PF_ANEQ(pd2.dst, 4989 &nk->addr[pd2.didx], pd2.af) || 4990 nk->port[pd2.didx] != iih.icmp6_id) 4991 pf_change_icmp(pd2.dst, &iih.icmp6_id, 4992 NULL, /* XXX Inbound NAT? */ 4993 &nk->addr[pd2.didx], 4994 nk->port[pd2.didx], NULL, 4995 pd2.ip_sum, icmpsum, 4996 pd->ip_sum, 0, AF_INET6); 4997 4998 m_copyback(m, off, sizeof(struct icmp6_hdr), 4999 (caddr_t)pd->hdr.icmp6); 5000 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 5001 m_copyback(m, off2, sizeof(struct icmp6_hdr), 5002 (caddr_t)&iih); 5003 } 5004 return (PF_PASS); 5005 break; 5006 } 5007 #endif /* INET6 */ 5008 default: { 5009 key.af = pd2.af; 5010 key.proto = pd2.proto; 5011 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5012 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5013 key.port[0] = key.port[1] = 0; 5014 5015 STATE_LOOKUP(kif, &key, direction, *state, pd); 5016 5017 /* translate source/destination address, if necessary */ 5018 if ((*state)->key[PF_SK_WIRE] != 5019 (*state)->key[PF_SK_STACK]) { 5020 struct pf_state_key *nk = 5021 (*state)->key[pd->didx]; 5022 5023 if (PF_ANEQ(pd2.src, 5024 &nk->addr[pd2.sidx], pd2.af)) 5025 pf_change_icmp(pd2.src, NULL, daddr, 5026 &nk->addr[pd2.sidx], 0, NULL, 5027 pd2.ip_sum, icmpsum, 5028 pd->ip_sum, 0, pd2.af); 5029 5030 if (PF_ANEQ(pd2.dst, 5031 &nk->addr[pd2.didx], pd2.af)) 5032 pf_change_icmp(pd2.src, NULL, 5033 NULL, /* XXX Inbound NAT? */ 5034 &nk->addr[pd2.didx], 0, NULL, 5035 pd2.ip_sum, icmpsum, 5036 pd->ip_sum, 0, pd2.af); 5037 5038 switch (pd2.af) { 5039 #ifdef INET 5040 case AF_INET: 5041 m_copyback(m, off, ICMP_MINLEN, 5042 (caddr_t)pd->hdr.icmp); 5043 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5044 break; 5045 #endif /* INET */ 5046 #ifdef INET6 5047 case AF_INET6: 5048 m_copyback(m, off, 5049 sizeof(struct icmp6_hdr), 5050 (caddr_t )pd->hdr.icmp6); 5051 m_copyback(m, ipoff2, sizeof(h2_6), 5052 (caddr_t )&h2_6); 5053 break; 5054 #endif /* INET6 */ 5055 } 5056 } 5057 return (PF_PASS); 5058 break; 5059 } 5060 } 5061 } 5062 } 5063 5064 static int 5065 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif, 5066 struct mbuf *m, struct pf_pdesc *pd) 5067 { 5068 struct pf_state_peer *src, *dst; 5069 struct pf_state_key_cmp key; 5070 5071 bzero(&key, sizeof(key)); 5072 key.af = pd->af; 5073 key.proto = pd->proto; 5074 if (direction == PF_IN) { 5075 PF_ACPY(&key.addr[0], pd->src, key.af); 5076 PF_ACPY(&key.addr[1], pd->dst, key.af); 5077 key.port[0] = key.port[1] = 0; 5078 } else { 5079 PF_ACPY(&key.addr[1], pd->src, key.af); 5080 PF_ACPY(&key.addr[0], pd->dst, key.af); 5081 key.port[1] = key.port[0] = 0; 5082 } 5083 5084 STATE_LOOKUP(kif, &key, direction, *state, pd); 5085 5086 if (direction == (*state)->direction) { 5087 src = &(*state)->src; 5088 dst = &(*state)->dst; 5089 } else { 5090 src = &(*state)->dst; 5091 dst = &(*state)->src; 5092 } 5093 5094 /* update states */ 5095 if (src->state < PFOTHERS_SINGLE) 5096 src->state = PFOTHERS_SINGLE; 5097 if (dst->state == PFOTHERS_SINGLE) 5098 dst->state = PFOTHERS_MULTIPLE; 5099 5100 /* update expire time */ 5101 (*state)->expire = time_uptime; 5102 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 5103 (*state)->timeout = PFTM_OTHER_MULTIPLE; 5104 else 5105 (*state)->timeout = PFTM_OTHER_SINGLE; 5106 5107 /* translate source/destination address, if necessary */ 5108 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5109 struct pf_state_key *nk = (*state)->key[pd->didx]; 5110 5111 KASSERT(nk, ("%s: nk is null", __func__)); 5112 KASSERT(pd, ("%s: pd is null", __func__)); 5113 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 5114 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 5115 switch (pd->af) { 5116 #ifdef INET 5117 case AF_INET: 5118 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5119 pf_change_a(&pd->src->v4.s_addr, 5120 pd->ip_sum, 5121 nk->addr[pd->sidx].v4.s_addr, 5122 0); 5123 5124 5125 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5126 pf_change_a(&pd->dst->v4.s_addr, 5127 pd->ip_sum, 5128 nk->addr[pd->didx].v4.s_addr, 5129 0); 5130 5131 break; 5132 #endif /* INET */ 5133 #ifdef INET6 5134 case AF_INET6: 5135 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5136 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5137 5138 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5139 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5140 #endif /* INET6 */ 5141 } 5142 } 5143 return (PF_PASS); 5144 } 5145 5146 /* 5147 * ipoff and off are measured from the start of the mbuf chain. 5148 * h must be at "ipoff" on the mbuf chain. 5149 */ 5150 void * 5151 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 5152 u_short *actionp, u_short *reasonp, sa_family_t af) 5153 { 5154 switch (af) { 5155 #ifdef INET 5156 case AF_INET: { 5157 struct ip *h = mtod(m, struct ip *); 5158 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 5159 5160 if (fragoff) { 5161 if (fragoff >= len) 5162 ACTION_SET(actionp, PF_PASS); 5163 else { 5164 ACTION_SET(actionp, PF_DROP); 5165 REASON_SET(reasonp, PFRES_FRAG); 5166 } 5167 return (NULL); 5168 } 5169 if (m->m_pkthdr.len < off + len || 5170 ntohs(h->ip_len) < off + len) { 5171 ACTION_SET(actionp, PF_DROP); 5172 REASON_SET(reasonp, PFRES_SHORT); 5173 return (NULL); 5174 } 5175 break; 5176 } 5177 #endif /* INET */ 5178 #ifdef INET6 5179 case AF_INET6: { 5180 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 5181 5182 if (m->m_pkthdr.len < off + len || 5183 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 5184 (unsigned)(off + len)) { 5185 ACTION_SET(actionp, PF_DROP); 5186 REASON_SET(reasonp, PFRES_SHORT); 5187 return (NULL); 5188 } 5189 break; 5190 } 5191 #endif /* INET6 */ 5192 } 5193 m_copydata(m, off, len, p); 5194 return (p); 5195 } 5196 5197 int 5198 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif, 5199 int rtableid) 5200 { 5201 #ifdef RADIX_MPATH 5202 struct radix_node_head *rnh; 5203 #endif 5204 struct sockaddr_in *dst; 5205 int ret = 1; 5206 int check_mpath; 5207 #ifdef INET6 5208 struct sockaddr_in6 *dst6; 5209 struct route_in6 ro; 5210 #else 5211 struct route ro; 5212 #endif 5213 struct radix_node *rn; 5214 struct rtentry *rt; 5215 struct ifnet *ifp; 5216 5217 check_mpath = 0; 5218 #ifdef RADIX_MPATH 5219 /* XXX: stick to table 0 for now */ 5220 rnh = rt_tables_get_rnh(0, af); 5221 if (rnh != NULL && rn_mpath_capable(rnh)) 5222 check_mpath = 1; 5223 #endif 5224 bzero(&ro, sizeof(ro)); 5225 switch (af) { 5226 case AF_INET: 5227 dst = satosin(&ro.ro_dst); 5228 dst->sin_family = AF_INET; 5229 dst->sin_len = sizeof(*dst); 5230 dst->sin_addr = addr->v4; 5231 break; 5232 #ifdef INET6 5233 case AF_INET6: 5234 /* 5235 * Skip check for addresses with embedded interface scope, 5236 * as they would always match anyway. 5237 */ 5238 if (IN6_IS_SCOPE_EMBED(&addr->v6)) 5239 goto out; 5240 dst6 = (struct sockaddr_in6 *)&ro.ro_dst; 5241 dst6->sin6_family = AF_INET6; 5242 dst6->sin6_len = sizeof(*dst6); 5243 dst6->sin6_addr = addr->v6; 5244 break; 5245 #endif /* INET6 */ 5246 default: 5247 return (0); 5248 } 5249 5250 /* Skip checks for ipsec interfaces */ 5251 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 5252 goto out; 5253 5254 switch (af) { 5255 #ifdef INET6 5256 case AF_INET6: 5257 in6_rtalloc_ign(&ro, 0, rtableid); 5258 break; 5259 #endif 5260 #ifdef INET 5261 case AF_INET: 5262 in_rtalloc_ign((struct route *)&ro, 0, rtableid); 5263 break; 5264 #endif 5265 default: 5266 rtalloc_ign((struct route *)&ro, 0); /* No/default FIB. */ 5267 break; 5268 } 5269 5270 if (ro.ro_rt != NULL) { 5271 /* No interface given, this is a no-route check */ 5272 if (kif == NULL) 5273 goto out; 5274 5275 if (kif->pfik_ifp == NULL) { 5276 ret = 0; 5277 goto out; 5278 } 5279 5280 /* Perform uRPF check if passed input interface */ 5281 ret = 0; 5282 rn = (struct radix_node *)ro.ro_rt; 5283 do { 5284 rt = (struct rtentry *)rn; 5285 ifp = rt->rt_ifp; 5286 5287 if (kif->pfik_ifp == ifp) 5288 ret = 1; 5289 #ifdef RADIX_MPATH 5290 rn = rn_mpath_next(rn); 5291 #endif 5292 } while (check_mpath == 1 && rn != NULL && ret == 0); 5293 } else 5294 ret = 0; 5295 out: 5296 if (ro.ro_rt != NULL) 5297 RTFREE(ro.ro_rt); 5298 return (ret); 5299 } 5300 5301 #ifdef INET 5302 static void 5303 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, 5304 struct pf_state *s, struct pf_pdesc *pd) 5305 { 5306 struct mbuf *m0, *m1; 5307 struct sockaddr_in dst; 5308 struct ip *ip; 5309 struct ifnet *ifp = NULL; 5310 struct pf_addr naddr; 5311 struct pf_src_node *sn = NULL; 5312 int error = 0; 5313 uint16_t ip_len, ip_off; 5314 5315 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 5316 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 5317 __func__)); 5318 5319 if ((pd->pf_mtag == NULL && 5320 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 5321 pd->pf_mtag->routed++ > 3) { 5322 m0 = *m; 5323 *m = NULL; 5324 goto bad_locked; 5325 } 5326 5327 if (r->rt == PF_DUPTO) { 5328 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { 5329 if (s) 5330 PF_STATE_UNLOCK(s); 5331 return; 5332 } 5333 } else { 5334 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 5335 if (s) 5336 PF_STATE_UNLOCK(s); 5337 return; 5338 } 5339 m0 = *m; 5340 } 5341 5342 ip = mtod(m0, struct ip *); 5343 5344 bzero(&dst, sizeof(dst)); 5345 dst.sin_family = AF_INET; 5346 dst.sin_len = sizeof(dst); 5347 dst.sin_addr = ip->ip_dst; 5348 5349 if (r->rt == PF_FASTROUTE) { 5350 struct rtentry *rt; 5351 5352 if (s) 5353 PF_STATE_UNLOCK(s); 5354 rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0)); 5355 if (rt == NULL) { 5356 KMOD_IPSTAT_INC(ips_noroute); 5357 error = EHOSTUNREACH; 5358 goto bad; 5359 } 5360 5361 ifp = rt->rt_ifp; 5362 counter_u64_add(rt->rt_pksent, 1); 5363 5364 if (rt->rt_flags & RTF_GATEWAY) 5365 bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst)); 5366 RTFREE_LOCKED(rt); 5367 } else { 5368 if (TAILQ_EMPTY(&r->rpool.list)) { 5369 DPFPRINTF(PF_DEBUG_URGENT, 5370 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 5371 goto bad_locked; 5372 } 5373 if (s == NULL) { 5374 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 5375 &naddr, NULL, &sn); 5376 if (!PF_AZERO(&naddr, AF_INET)) 5377 dst.sin_addr.s_addr = naddr.v4.s_addr; 5378 ifp = r->rpool.cur->kif ? 5379 r->rpool.cur->kif->pfik_ifp : NULL; 5380 } else { 5381 if (!PF_AZERO(&s->rt_addr, AF_INET)) 5382 dst.sin_addr.s_addr = 5383 s->rt_addr.v4.s_addr; 5384 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5385 PF_STATE_UNLOCK(s); 5386 } 5387 } 5388 if (ifp == NULL) 5389 goto bad; 5390 5391 if (oifp != ifp) { 5392 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS) 5393 goto bad; 5394 else if (m0 == NULL) 5395 goto done; 5396 if (m0->m_len < sizeof(struct ip)) { 5397 DPFPRINTF(PF_DEBUG_URGENT, 5398 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 5399 goto bad; 5400 } 5401 ip = mtod(m0, struct ip *); 5402 } 5403 5404 if (ifp->if_flags & IFF_LOOPBACK) 5405 m0->m_flags |= M_SKIP_FIREWALL; 5406 5407 ip_len = ntohs(ip->ip_len); 5408 ip_off = ntohs(ip->ip_off); 5409 5410 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 5411 m0->m_pkthdr.csum_flags |= CSUM_IP; 5412 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 5413 in_delayed_cksum(m0); 5414 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 5415 } 5416 #ifdef SCTP 5417 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 5418 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 5419 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 5420 } 5421 #endif 5422 5423 /* 5424 * If small enough for interface, or the interface will take 5425 * care of the fragmentation for us, we can just send directly. 5426 */ 5427 if (ip_len <= ifp->if_mtu || 5428 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 5429 ip->ip_sum = 0; 5430 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 5431 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 5432 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 5433 } 5434 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 5435 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); 5436 goto done; 5437 } 5438 5439 /* Balk when DF bit is set or the interface didn't support TSO. */ 5440 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 5441 error = EMSGSIZE; 5442 KMOD_IPSTAT_INC(ips_cantfrag); 5443 if (r->rt != PF_DUPTO) { 5444 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 5445 ifp->if_mtu); 5446 goto done; 5447 } else 5448 goto bad; 5449 } 5450 5451 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 5452 if (error) 5453 goto bad; 5454 5455 for (; m0; m0 = m1) { 5456 m1 = m0->m_nextpkt; 5457 m0->m_nextpkt = NULL; 5458 if (error == 0) { 5459 m_clrprotoflags(m0); 5460 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); 5461 } else 5462 m_freem(m0); 5463 } 5464 5465 if (error == 0) 5466 KMOD_IPSTAT_INC(ips_fragmented); 5467 5468 done: 5469 if (r->rt != PF_DUPTO) 5470 *m = NULL; 5471 return; 5472 5473 bad_locked: 5474 if (s) 5475 PF_STATE_UNLOCK(s); 5476 bad: 5477 m_freem(m0); 5478 goto done; 5479 } 5480 #endif /* INET */ 5481 5482 #ifdef INET6 5483 static void 5484 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, 5485 struct pf_state *s, struct pf_pdesc *pd) 5486 { 5487 struct mbuf *m0; 5488 struct sockaddr_in6 dst; 5489 struct ip6_hdr *ip6; 5490 struct ifnet *ifp = NULL; 5491 struct pf_addr naddr; 5492 struct pf_src_node *sn = NULL; 5493 5494 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 5495 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 5496 __func__)); 5497 5498 if ((pd->pf_mtag == NULL && 5499 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 5500 pd->pf_mtag->routed++ > 3) { 5501 m0 = *m; 5502 *m = NULL; 5503 goto bad_locked; 5504 } 5505 5506 if (r->rt == PF_DUPTO) { 5507 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { 5508 if (s) 5509 PF_STATE_UNLOCK(s); 5510 return; 5511 } 5512 } else { 5513 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 5514 if (s) 5515 PF_STATE_UNLOCK(s); 5516 return; 5517 } 5518 m0 = *m; 5519 } 5520 5521 ip6 = mtod(m0, struct ip6_hdr *); 5522 5523 bzero(&dst, sizeof(dst)); 5524 dst.sin6_family = AF_INET6; 5525 dst.sin6_len = sizeof(dst); 5526 dst.sin6_addr = ip6->ip6_dst; 5527 5528 /* Cheat. XXX why only in the v6 case??? */ 5529 if (r->rt == PF_FASTROUTE) { 5530 if (s) 5531 PF_STATE_UNLOCK(s); 5532 m0->m_flags |= M_SKIP_FIREWALL; 5533 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL); 5534 *m = NULL; 5535 return; 5536 } 5537 5538 if (TAILQ_EMPTY(&r->rpool.list)) { 5539 DPFPRINTF(PF_DEBUG_URGENT, 5540 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 5541 goto bad_locked; 5542 } 5543 if (s == NULL) { 5544 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 5545 &naddr, NULL, &sn); 5546 if (!PF_AZERO(&naddr, AF_INET6)) 5547 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 5548 &naddr, AF_INET6); 5549 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 5550 } else { 5551 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 5552 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 5553 &s->rt_addr, AF_INET6); 5554 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5555 } 5556 5557 if (s) 5558 PF_STATE_UNLOCK(s); 5559 5560 if (ifp == NULL) 5561 goto bad; 5562 5563 if (oifp != ifp) { 5564 if (pf_test6(PF_FWD, ifp, &m0, NULL) != PF_PASS) 5565 goto bad; 5566 else if (m0 == NULL) 5567 goto done; 5568 if (m0->m_len < sizeof(struct ip6_hdr)) { 5569 DPFPRINTF(PF_DEBUG_URGENT, 5570 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 5571 __func__)); 5572 goto bad; 5573 } 5574 ip6 = mtod(m0, struct ip6_hdr *); 5575 } 5576 5577 if (ifp->if_flags & IFF_LOOPBACK) 5578 m0->m_flags |= M_SKIP_FIREWALL; 5579 5580 /* 5581 * If the packet is too large for the outgoing interface, 5582 * send back an icmp6 error. 5583 */ 5584 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 5585 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 5586 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) 5587 nd6_output_ifp(ifp, ifp, m0, &dst); 5588 else { 5589 in6_ifstat_inc(ifp, ifs6_in_toobig); 5590 if (r->rt != PF_DUPTO) 5591 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 5592 else 5593 goto bad; 5594 } 5595 5596 done: 5597 if (r->rt != PF_DUPTO) 5598 *m = NULL; 5599 return; 5600 5601 bad_locked: 5602 if (s) 5603 PF_STATE_UNLOCK(s); 5604 bad: 5605 m_freem(m0); 5606 goto done; 5607 } 5608 #endif /* INET6 */ 5609 5610 /* 5611 * FreeBSD supports cksum offloads for the following drivers. 5612 * em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4), 5613 * ti(4), txp(4), xl(4) 5614 * 5615 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 5616 * network driver performed cksum including pseudo header, need to verify 5617 * csum_data 5618 * CSUM_DATA_VALID : 5619 * network driver performed cksum, needs to additional pseudo header 5620 * cksum computation with partial csum_data(i.e. lack of H/W support for 5621 * pseudo header, for instance hme(4), sk(4) and possibly gem(4)) 5622 * 5623 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 5624 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 5625 * TCP/UDP layer. 5626 * Also, set csum_data to 0xffff to force cksum validation. 5627 */ 5628 static int 5629 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 5630 { 5631 u_int16_t sum = 0; 5632 int hw_assist = 0; 5633 struct ip *ip; 5634 5635 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 5636 return (1); 5637 if (m->m_pkthdr.len < off + len) 5638 return (1); 5639 5640 switch (p) { 5641 case IPPROTO_TCP: 5642 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 5643 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 5644 sum = m->m_pkthdr.csum_data; 5645 } else { 5646 ip = mtod(m, struct ip *); 5647 sum = in_pseudo(ip->ip_src.s_addr, 5648 ip->ip_dst.s_addr, htonl((u_short)len + 5649 m->m_pkthdr.csum_data + IPPROTO_TCP)); 5650 } 5651 sum ^= 0xffff; 5652 ++hw_assist; 5653 } 5654 break; 5655 case IPPROTO_UDP: 5656 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 5657 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 5658 sum = m->m_pkthdr.csum_data; 5659 } else { 5660 ip = mtod(m, struct ip *); 5661 sum = in_pseudo(ip->ip_src.s_addr, 5662 ip->ip_dst.s_addr, htonl((u_short)len + 5663 m->m_pkthdr.csum_data + IPPROTO_UDP)); 5664 } 5665 sum ^= 0xffff; 5666 ++hw_assist; 5667 } 5668 break; 5669 case IPPROTO_ICMP: 5670 #ifdef INET6 5671 case IPPROTO_ICMPV6: 5672 #endif /* INET6 */ 5673 break; 5674 default: 5675 return (1); 5676 } 5677 5678 if (!hw_assist) { 5679 switch (af) { 5680 case AF_INET: 5681 if (p == IPPROTO_ICMP) { 5682 if (m->m_len < off) 5683 return (1); 5684 m->m_data += off; 5685 m->m_len -= off; 5686 sum = in_cksum(m, len); 5687 m->m_data -= off; 5688 m->m_len += off; 5689 } else { 5690 if (m->m_len < sizeof(struct ip)) 5691 return (1); 5692 sum = in4_cksum(m, p, off, len); 5693 } 5694 break; 5695 #ifdef INET6 5696 case AF_INET6: 5697 if (m->m_len < sizeof(struct ip6_hdr)) 5698 return (1); 5699 sum = in6_cksum(m, p, off, len); 5700 break; 5701 #endif /* INET6 */ 5702 default: 5703 return (1); 5704 } 5705 } 5706 if (sum) { 5707 switch (p) { 5708 case IPPROTO_TCP: 5709 { 5710 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 5711 break; 5712 } 5713 case IPPROTO_UDP: 5714 { 5715 KMOD_UDPSTAT_INC(udps_badsum); 5716 break; 5717 } 5718 #ifdef INET 5719 case IPPROTO_ICMP: 5720 { 5721 KMOD_ICMPSTAT_INC(icps_checksum); 5722 break; 5723 } 5724 #endif 5725 #ifdef INET6 5726 case IPPROTO_ICMPV6: 5727 { 5728 KMOD_ICMP6STAT_INC(icp6s_checksum); 5729 break; 5730 } 5731 #endif /* INET6 */ 5732 } 5733 return (1); 5734 } else { 5735 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 5736 m->m_pkthdr.csum_flags |= 5737 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 5738 m->m_pkthdr.csum_data = 0xffff; 5739 } 5740 } 5741 return (0); 5742 } 5743 5744 5745 #ifdef INET 5746 int 5747 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 5748 { 5749 struct pfi_kif *kif; 5750 u_short action, reason = 0, log = 0; 5751 struct mbuf *m = *m0; 5752 struct ip *h = NULL; 5753 struct m_tag *ipfwtag; 5754 struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 5755 struct pf_state *s = NULL; 5756 struct pf_ruleset *ruleset = NULL; 5757 struct pf_pdesc pd; 5758 int off, dirndx, pqid = 0; 5759 5760 M_ASSERTPKTHDR(m); 5761 5762 if (!V_pf_status.running) 5763 return (PF_PASS); 5764 5765 memset(&pd, 0, sizeof(pd)); 5766 5767 kif = (struct pfi_kif *)ifp->if_pf_kif; 5768 5769 if (kif == NULL) { 5770 DPFPRINTF(PF_DEBUG_URGENT, 5771 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 5772 return (PF_DROP); 5773 } 5774 if (kif->pfik_flags & PFI_IFLAG_SKIP) 5775 return (PF_PASS); 5776 5777 if (m->m_flags & M_SKIP_FIREWALL) 5778 return (PF_PASS); 5779 5780 pd.pf_mtag = pf_find_mtag(m); 5781 5782 PF_RULES_RLOCK(); 5783 5784 if (ip_divert_ptr != NULL && 5785 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { 5786 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); 5787 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) { 5788 if (pd.pf_mtag == NULL && 5789 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 5790 action = PF_DROP; 5791 goto done; 5792 } 5793 pd.pf_mtag->flags |= PF_PACKET_LOOPED; 5794 m_tag_delete(m, ipfwtag); 5795 } 5796 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) { 5797 m->m_flags |= M_FASTFWD_OURS; 5798 pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT; 5799 } 5800 } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { 5801 /* We do IP header normalization and packet reassembly here */ 5802 action = PF_DROP; 5803 goto done; 5804 } 5805 m = *m0; /* pf_normalize messes with m0 */ 5806 h = mtod(m, struct ip *); 5807 5808 off = h->ip_hl << 2; 5809 if (off < (int)sizeof(struct ip)) { 5810 action = PF_DROP; 5811 REASON_SET(&reason, PFRES_SHORT); 5812 log = 1; 5813 goto done; 5814 } 5815 5816 pd.src = (struct pf_addr *)&h->ip_src; 5817 pd.dst = (struct pf_addr *)&h->ip_dst; 5818 pd.sport = pd.dport = NULL; 5819 pd.ip_sum = &h->ip_sum; 5820 pd.proto_sum = NULL; 5821 pd.proto = h->ip_p; 5822 pd.dir = dir; 5823 pd.sidx = (dir == PF_IN) ? 0 : 1; 5824 pd.didx = (dir == PF_IN) ? 1 : 0; 5825 pd.af = AF_INET; 5826 pd.tos = h->ip_tos; 5827 pd.tot_len = ntohs(h->ip_len); 5828 5829 /* handle fragments that didn't get reassembled by normalization */ 5830 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 5831 action = pf_test_fragment(&r, dir, kif, m, h, 5832 &pd, &a, &ruleset); 5833 goto done; 5834 } 5835 5836 switch (h->ip_p) { 5837 5838 case IPPROTO_TCP: { 5839 struct tcphdr th; 5840 5841 pd.hdr.tcp = &th; 5842 if (!pf_pull_hdr(m, off, &th, sizeof(th), 5843 &action, &reason, AF_INET)) { 5844 log = action != PF_PASS; 5845 goto done; 5846 } 5847 pd.p_len = pd.tot_len - off - (th.th_off << 2); 5848 if ((th.th_flags & TH_ACK) && pd.p_len == 0) 5849 pqid = 1; 5850 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 5851 if (action == PF_DROP) 5852 goto done; 5853 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 5854 &reason); 5855 if (action == PF_PASS) { 5856 if (pfsync_update_state_ptr != NULL) 5857 pfsync_update_state_ptr(s); 5858 r = s->rule.ptr; 5859 a = s->anchor.ptr; 5860 log = s->log; 5861 } else if (s == NULL) 5862 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5863 &a, &ruleset, inp); 5864 break; 5865 } 5866 5867 case IPPROTO_UDP: { 5868 struct udphdr uh; 5869 5870 pd.hdr.udp = &uh; 5871 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 5872 &action, &reason, AF_INET)) { 5873 log = action != PF_PASS; 5874 goto done; 5875 } 5876 if (uh.uh_dport == 0 || 5877 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 5878 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 5879 action = PF_DROP; 5880 REASON_SET(&reason, PFRES_SHORT); 5881 goto done; 5882 } 5883 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 5884 if (action == PF_PASS) { 5885 if (pfsync_update_state_ptr != NULL) 5886 pfsync_update_state_ptr(s); 5887 r = s->rule.ptr; 5888 a = s->anchor.ptr; 5889 log = s->log; 5890 } else if (s == NULL) 5891 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5892 &a, &ruleset, inp); 5893 break; 5894 } 5895 5896 case IPPROTO_ICMP: { 5897 struct icmp ih; 5898 5899 pd.hdr.icmp = &ih; 5900 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN, 5901 &action, &reason, AF_INET)) { 5902 log = action != PF_PASS; 5903 goto done; 5904 } 5905 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, 5906 &reason); 5907 if (action == PF_PASS) { 5908 if (pfsync_update_state_ptr != NULL) 5909 pfsync_update_state_ptr(s); 5910 r = s->rule.ptr; 5911 a = s->anchor.ptr; 5912 log = s->log; 5913 } else if (s == NULL) 5914 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5915 &a, &ruleset, inp); 5916 break; 5917 } 5918 5919 #ifdef INET6 5920 case IPPROTO_ICMPV6: { 5921 action = PF_DROP; 5922 DPFPRINTF(PF_DEBUG_MISC, 5923 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 5924 goto done; 5925 } 5926 #endif 5927 5928 default: 5929 action = pf_test_state_other(&s, dir, kif, m, &pd); 5930 if (action == PF_PASS) { 5931 if (pfsync_update_state_ptr != NULL) 5932 pfsync_update_state_ptr(s); 5933 r = s->rule.ptr; 5934 a = s->anchor.ptr; 5935 log = s->log; 5936 } else if (s == NULL) 5937 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5938 &a, &ruleset, inp); 5939 break; 5940 } 5941 5942 done: 5943 PF_RULES_RUNLOCK(); 5944 if (action == PF_PASS && h->ip_hl > 5 && 5945 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 5946 action = PF_DROP; 5947 REASON_SET(&reason, PFRES_IPOPTIONS); 5948 log = r->log; 5949 DPFPRINTF(PF_DEBUG_MISC, 5950 ("pf: dropping packet with ip options\n")); 5951 } 5952 5953 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 5954 action = PF_DROP; 5955 REASON_SET(&reason, PFRES_MEMORY); 5956 } 5957 if (r->rtableid >= 0) 5958 M_SETFIB(m, r->rtableid); 5959 5960 #ifdef ALTQ 5961 if (action == PF_PASS && r->qid) { 5962 if (pd.pf_mtag == NULL && 5963 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 5964 action = PF_DROP; 5965 REASON_SET(&reason, PFRES_MEMORY); 5966 } else { 5967 if (s != NULL) 5968 pd.pf_mtag->qid_hash = pf_state_hash(s); 5969 if (pqid || (pd.tos & IPTOS_LOWDELAY)) 5970 pd.pf_mtag->qid = r->pqid; 5971 else 5972 pd.pf_mtag->qid = r->qid; 5973 /* Add hints for ecn. */ 5974 pd.pf_mtag->hdr = h; 5975 } 5976 5977 } 5978 #endif /* ALTQ */ 5979 5980 /* 5981 * connections redirected to loopback should not match sockets 5982 * bound specifically to loopback due to security implications, 5983 * see tcp_input() and in_pcblookup_listen(). 5984 */ 5985 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 5986 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 5987 (s->nat_rule.ptr->action == PF_RDR || 5988 s->nat_rule.ptr->action == PF_BINAT) && 5989 (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) 5990 m->m_flags |= M_SKIP_FIREWALL; 5991 5992 if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL && 5993 !PACKET_LOOPED(&pd)) { 5994 5995 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 5996 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 5997 if (ipfwtag != NULL) { 5998 ((struct ipfw_rule_ref *)(ipfwtag+1))->info = 5999 ntohs(r->divert.port); 6000 ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir; 6001 6002 if (s) 6003 PF_STATE_UNLOCK(s); 6004 6005 m_tag_prepend(m, ipfwtag); 6006 if (m->m_flags & M_FASTFWD_OURS) { 6007 if (pd.pf_mtag == NULL && 6008 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6009 action = PF_DROP; 6010 REASON_SET(&reason, PFRES_MEMORY); 6011 log = 1; 6012 DPFPRINTF(PF_DEBUG_MISC, 6013 ("pf: failed to allocate tag\n")); 6014 } else { 6015 pd.pf_mtag->flags |= 6016 PF_FASTFWD_OURS_PRESENT; 6017 m->m_flags &= ~M_FASTFWD_OURS; 6018 } 6019 } 6020 ip_divert_ptr(*m0, dir == PF_IN ? DIR_IN : DIR_OUT); 6021 *m0 = NULL; 6022 6023 return (action); 6024 } else { 6025 /* XXX: ipfw has the same behaviour! */ 6026 action = PF_DROP; 6027 REASON_SET(&reason, PFRES_MEMORY); 6028 log = 1; 6029 DPFPRINTF(PF_DEBUG_MISC, 6030 ("pf: failed to allocate divert tag\n")); 6031 } 6032 } 6033 6034 if (log) { 6035 struct pf_rule *lr; 6036 6037 if (s != NULL && s->nat_rule.ptr != NULL && 6038 s->nat_rule.ptr->log & PF_LOG_ALL) 6039 lr = s->nat_rule.ptr; 6040 else 6041 lr = r; 6042 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd, 6043 (s == NULL)); 6044 } 6045 6046 kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len; 6047 kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++; 6048 6049 if (action == PF_PASS || r->action == PF_DROP) { 6050 dirndx = (dir == PF_OUT); 6051 r->packets[dirndx]++; 6052 r->bytes[dirndx] += pd.tot_len; 6053 if (a != NULL) { 6054 a->packets[dirndx]++; 6055 a->bytes[dirndx] += pd.tot_len; 6056 } 6057 if (s != NULL) { 6058 if (s->nat_rule.ptr != NULL) { 6059 s->nat_rule.ptr->packets[dirndx]++; 6060 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; 6061 } 6062 if (s->src_node != NULL) { 6063 s->src_node->packets[dirndx]++; 6064 s->src_node->bytes[dirndx] += pd.tot_len; 6065 } 6066 if (s->nat_src_node != NULL) { 6067 s->nat_src_node->packets[dirndx]++; 6068 s->nat_src_node->bytes[dirndx] += pd.tot_len; 6069 } 6070 dirndx = (dir == s->direction) ? 0 : 1; 6071 s->packets[dirndx]++; 6072 s->bytes[dirndx] += pd.tot_len; 6073 } 6074 tr = r; 6075 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 6076 if (nr != NULL && r == &V_pf_default_rule) 6077 tr = nr; 6078 if (tr->src.addr.type == PF_ADDR_TABLE) 6079 pfr_update_stats(tr->src.addr.p.tbl, 6080 (s == NULL) ? pd.src : 6081 &s->key[(s->direction == PF_IN)]-> 6082 addr[(s->direction == PF_OUT)], 6083 pd.af, pd.tot_len, dir == PF_OUT, 6084 r->action == PF_PASS, tr->src.neg); 6085 if (tr->dst.addr.type == PF_ADDR_TABLE) 6086 pfr_update_stats(tr->dst.addr.p.tbl, 6087 (s == NULL) ? pd.dst : 6088 &s->key[(s->direction == PF_IN)]-> 6089 addr[(s->direction == PF_IN)], 6090 pd.af, pd.tot_len, dir == PF_OUT, 6091 r->action == PF_PASS, tr->dst.neg); 6092 } 6093 6094 switch (action) { 6095 case PF_SYNPROXY_DROP: 6096 m_freem(*m0); 6097 case PF_DEFER: 6098 *m0 = NULL; 6099 action = PF_PASS; 6100 break; 6101 case PF_DROP: 6102 m_freem(*m0); 6103 *m0 = NULL; 6104 break; 6105 default: 6106 /* pf_route() returns unlocked. */ 6107 if (r->rt) { 6108 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd); 6109 return (action); 6110 } 6111 break; 6112 } 6113 if (s) 6114 PF_STATE_UNLOCK(s); 6115 6116 return (action); 6117 } 6118 #endif /* INET */ 6119 6120 #ifdef INET6 6121 int 6122 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 6123 { 6124 struct pfi_kif *kif; 6125 u_short action, reason = 0, log = 0; 6126 struct mbuf *m = *m0, *n = NULL; 6127 struct m_tag *mtag; 6128 struct ip6_hdr *h = NULL; 6129 struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 6130 struct pf_state *s = NULL; 6131 struct pf_ruleset *ruleset = NULL; 6132 struct pf_pdesc pd; 6133 int off, terminal = 0, dirndx, rh_cnt = 0; 6134 int fwdir = dir; 6135 6136 M_ASSERTPKTHDR(m); 6137 6138 /* Detect packet forwarding. 6139 * If the input interface is different from the output interface we're 6140 * forwarding. 6141 * We do need to be careful about bridges. If the 6142 * net.link.bridge.pfil_bridge sysctl is set we can be filtering on a 6143 * bridge, so if the input interface is a bridge member and the output 6144 * interface is its bridge we're not actually forwarding but bridging. 6145 */ 6146 if (dir == PF_OUT && m->m_pkthdr.rcvif && ifp != m->m_pkthdr.rcvif 6147 && (m->m_pkthdr.rcvif->if_bridge == NULL 6148 || m->m_pkthdr.rcvif->if_bridge != ifp->if_softc)) 6149 fwdir = PF_FWD; 6150 6151 if (!V_pf_status.running) 6152 return (PF_PASS); 6153 6154 memset(&pd, 0, sizeof(pd)); 6155 pd.pf_mtag = pf_find_mtag(m); 6156 6157 if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED) 6158 return (PF_PASS); 6159 6160 kif = (struct pfi_kif *)ifp->if_pf_kif; 6161 if (kif == NULL) { 6162 DPFPRINTF(PF_DEBUG_URGENT, 6163 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 6164 return (PF_DROP); 6165 } 6166 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6167 return (PF_PASS); 6168 6169 if (m->m_flags & M_SKIP_FIREWALL) 6170 return (PF_PASS); 6171 6172 PF_RULES_RLOCK(); 6173 6174 /* We do IP header normalization and packet reassembly here */ 6175 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { 6176 action = PF_DROP; 6177 goto done; 6178 } 6179 m = *m0; /* pf_normalize messes with m0 */ 6180 h = mtod(m, struct ip6_hdr *); 6181 6182 #if 1 6183 /* 6184 * we do not support jumbogram yet. if we keep going, zero ip6_plen 6185 * will do something bad, so drop the packet for now. 6186 */ 6187 if (htons(h->ip6_plen) == 0) { 6188 action = PF_DROP; 6189 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 6190 goto done; 6191 } 6192 #endif 6193 6194 pd.src = (struct pf_addr *)&h->ip6_src; 6195 pd.dst = (struct pf_addr *)&h->ip6_dst; 6196 pd.sport = pd.dport = NULL; 6197 pd.ip_sum = NULL; 6198 pd.proto_sum = NULL; 6199 pd.dir = dir; 6200 pd.sidx = (dir == PF_IN) ? 0 : 1; 6201 pd.didx = (dir == PF_IN) ? 1 : 0; 6202 pd.af = AF_INET6; 6203 pd.tos = 0; 6204 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 6205 6206 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 6207 pd.proto = h->ip6_nxt; 6208 do { 6209 switch (pd.proto) { 6210 case IPPROTO_FRAGMENT: 6211 action = pf_test_fragment(&r, dir, kif, m, h, 6212 &pd, &a, &ruleset); 6213 if (action == PF_DROP) 6214 REASON_SET(&reason, PFRES_FRAG); 6215 goto done; 6216 case IPPROTO_ROUTING: { 6217 struct ip6_rthdr rthdr; 6218 6219 if (rh_cnt++) { 6220 DPFPRINTF(PF_DEBUG_MISC, 6221 ("pf: IPv6 more than one rthdr\n")); 6222 action = PF_DROP; 6223 REASON_SET(&reason, PFRES_IPOPTIONS); 6224 log = 1; 6225 goto done; 6226 } 6227 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 6228 &reason, pd.af)) { 6229 DPFPRINTF(PF_DEBUG_MISC, 6230 ("pf: IPv6 short rthdr\n")); 6231 action = PF_DROP; 6232 REASON_SET(&reason, PFRES_SHORT); 6233 log = 1; 6234 goto done; 6235 } 6236 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 6237 DPFPRINTF(PF_DEBUG_MISC, 6238 ("pf: IPv6 rthdr0\n")); 6239 action = PF_DROP; 6240 REASON_SET(&reason, PFRES_IPOPTIONS); 6241 log = 1; 6242 goto done; 6243 } 6244 /* FALLTHROUGH */ 6245 } 6246 case IPPROTO_AH: 6247 case IPPROTO_HOPOPTS: 6248 case IPPROTO_DSTOPTS: { 6249 /* get next header and header length */ 6250 struct ip6_ext opt6; 6251 6252 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 6253 NULL, &reason, pd.af)) { 6254 DPFPRINTF(PF_DEBUG_MISC, 6255 ("pf: IPv6 short opt\n")); 6256 action = PF_DROP; 6257 log = 1; 6258 goto done; 6259 } 6260 if (pd.proto == IPPROTO_AH) 6261 off += (opt6.ip6e_len + 2) * 4; 6262 else 6263 off += (opt6.ip6e_len + 1) * 8; 6264 pd.proto = opt6.ip6e_nxt; 6265 /* goto the next header */ 6266 break; 6267 } 6268 default: 6269 terminal++; 6270 break; 6271 } 6272 } while (!terminal); 6273 6274 /* if there's no routing header, use unmodified mbuf for checksumming */ 6275 if (!n) 6276 n = m; 6277 6278 switch (pd.proto) { 6279 6280 case IPPROTO_TCP: { 6281 struct tcphdr th; 6282 6283 pd.hdr.tcp = &th; 6284 if (!pf_pull_hdr(m, off, &th, sizeof(th), 6285 &action, &reason, AF_INET6)) { 6286 log = action != PF_PASS; 6287 goto done; 6288 } 6289 pd.p_len = pd.tot_len - off - (th.th_off << 2); 6290 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6291 if (action == PF_DROP) 6292 goto done; 6293 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6294 &reason); 6295 if (action == PF_PASS) { 6296 if (pfsync_update_state_ptr != NULL) 6297 pfsync_update_state_ptr(s); 6298 r = s->rule.ptr; 6299 a = s->anchor.ptr; 6300 log = s->log; 6301 } else if (s == NULL) 6302 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6303 &a, &ruleset, inp); 6304 break; 6305 } 6306 6307 case IPPROTO_UDP: { 6308 struct udphdr uh; 6309 6310 pd.hdr.udp = &uh; 6311 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 6312 &action, &reason, AF_INET6)) { 6313 log = action != PF_PASS; 6314 goto done; 6315 } 6316 if (uh.uh_dport == 0 || 6317 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 6318 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 6319 action = PF_DROP; 6320 REASON_SET(&reason, PFRES_SHORT); 6321 goto done; 6322 } 6323 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 6324 if (action == PF_PASS) { 6325 if (pfsync_update_state_ptr != NULL) 6326 pfsync_update_state_ptr(s); 6327 r = s->rule.ptr; 6328 a = s->anchor.ptr; 6329 log = s->log; 6330 } else if (s == NULL) 6331 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6332 &a, &ruleset, inp); 6333 break; 6334 } 6335 6336 case IPPROTO_ICMP: { 6337 action = PF_DROP; 6338 DPFPRINTF(PF_DEBUG_MISC, 6339 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 6340 goto done; 6341 } 6342 6343 case IPPROTO_ICMPV6: { 6344 struct icmp6_hdr ih; 6345 6346 pd.hdr.icmp6 = &ih; 6347 if (!pf_pull_hdr(m, off, &ih, sizeof(ih), 6348 &action, &reason, AF_INET6)) { 6349 log = action != PF_PASS; 6350 goto done; 6351 } 6352 action = pf_test_state_icmp(&s, dir, kif, 6353 m, off, h, &pd, &reason); 6354 if (action == PF_PASS) { 6355 if (pfsync_update_state_ptr != NULL) 6356 pfsync_update_state_ptr(s); 6357 r = s->rule.ptr; 6358 a = s->anchor.ptr; 6359 log = s->log; 6360 } else if (s == NULL) 6361 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6362 &a, &ruleset, inp); 6363 break; 6364 } 6365 6366 default: 6367 action = pf_test_state_other(&s, dir, kif, m, &pd); 6368 if (action == PF_PASS) { 6369 if (pfsync_update_state_ptr != NULL) 6370 pfsync_update_state_ptr(s); 6371 r = s->rule.ptr; 6372 a = s->anchor.ptr; 6373 log = s->log; 6374 } else if (s == NULL) 6375 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6376 &a, &ruleset, inp); 6377 break; 6378 } 6379 6380 done: 6381 PF_RULES_RUNLOCK(); 6382 if (n != m) { 6383 m_freem(n); 6384 n = NULL; 6385 } 6386 6387 /* handle dangerous IPv6 extension headers. */ 6388 if (action == PF_PASS && rh_cnt && 6389 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 6390 action = PF_DROP; 6391 REASON_SET(&reason, PFRES_IPOPTIONS); 6392 log = r->log; 6393 DPFPRINTF(PF_DEBUG_MISC, 6394 ("pf: dropping packet with dangerous v6 headers\n")); 6395 } 6396 6397 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 6398 action = PF_DROP; 6399 REASON_SET(&reason, PFRES_MEMORY); 6400 } 6401 if (r->rtableid >= 0) 6402 M_SETFIB(m, r->rtableid); 6403 6404 #ifdef ALTQ 6405 if (action == PF_PASS && r->qid) { 6406 if (pd.pf_mtag == NULL && 6407 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6408 action = PF_DROP; 6409 REASON_SET(&reason, PFRES_MEMORY); 6410 } else { 6411 if (s != NULL) 6412 pd.pf_mtag->qid_hash = pf_state_hash(s); 6413 if (pd.tos & IPTOS_LOWDELAY) 6414 pd.pf_mtag->qid = r->pqid; 6415 else 6416 pd.pf_mtag->qid = r->qid; 6417 /* Add hints for ecn. */ 6418 pd.pf_mtag->hdr = h; 6419 } 6420 } 6421 #endif /* ALTQ */ 6422 6423 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 6424 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 6425 (s->nat_rule.ptr->action == PF_RDR || 6426 s->nat_rule.ptr->action == PF_BINAT) && 6427 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 6428 m->m_flags |= M_SKIP_FIREWALL; 6429 6430 /* XXX: Anybody working on it?! */ 6431 if (r->divert.port) 6432 printf("pf: divert(9) is not supported for IPv6\n"); 6433 6434 if (log) { 6435 struct pf_rule *lr; 6436 6437 if (s != NULL && s->nat_rule.ptr != NULL && 6438 s->nat_rule.ptr->log & PF_LOG_ALL) 6439 lr = s->nat_rule.ptr; 6440 else 6441 lr = r; 6442 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset, 6443 &pd, (s == NULL)); 6444 } 6445 6446 kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len; 6447 kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++; 6448 6449 if (action == PF_PASS || r->action == PF_DROP) { 6450 dirndx = (dir == PF_OUT); 6451 r->packets[dirndx]++; 6452 r->bytes[dirndx] += pd.tot_len; 6453 if (a != NULL) { 6454 a->packets[dirndx]++; 6455 a->bytes[dirndx] += pd.tot_len; 6456 } 6457 if (s != NULL) { 6458 if (s->nat_rule.ptr != NULL) { 6459 s->nat_rule.ptr->packets[dirndx]++; 6460 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; 6461 } 6462 if (s->src_node != NULL) { 6463 s->src_node->packets[dirndx]++; 6464 s->src_node->bytes[dirndx] += pd.tot_len; 6465 } 6466 if (s->nat_src_node != NULL) { 6467 s->nat_src_node->packets[dirndx]++; 6468 s->nat_src_node->bytes[dirndx] += pd.tot_len; 6469 } 6470 dirndx = (dir == s->direction) ? 0 : 1; 6471 s->packets[dirndx]++; 6472 s->bytes[dirndx] += pd.tot_len; 6473 } 6474 tr = r; 6475 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 6476 if (nr != NULL && r == &V_pf_default_rule) 6477 tr = nr; 6478 if (tr->src.addr.type == PF_ADDR_TABLE) 6479 pfr_update_stats(tr->src.addr.p.tbl, 6480 (s == NULL) ? pd.src : 6481 &s->key[(s->direction == PF_IN)]->addr[0], 6482 pd.af, pd.tot_len, dir == PF_OUT, 6483 r->action == PF_PASS, tr->src.neg); 6484 if (tr->dst.addr.type == PF_ADDR_TABLE) 6485 pfr_update_stats(tr->dst.addr.p.tbl, 6486 (s == NULL) ? pd.dst : 6487 &s->key[(s->direction == PF_IN)]->addr[1], 6488 pd.af, pd.tot_len, dir == PF_OUT, 6489 r->action == PF_PASS, tr->dst.neg); 6490 } 6491 6492 switch (action) { 6493 case PF_SYNPROXY_DROP: 6494 m_freem(*m0); 6495 case PF_DEFER: 6496 *m0 = NULL; 6497 action = PF_PASS; 6498 break; 6499 case PF_DROP: 6500 m_freem(*m0); 6501 *m0 = NULL; 6502 break; 6503 default: 6504 /* pf_route6() returns unlocked. */ 6505 if (r->rt) { 6506 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd); 6507 return (action); 6508 } 6509 break; 6510 } 6511 6512 if (s) 6513 PF_STATE_UNLOCK(s); 6514 6515 /* If reassembled packet passed, create new fragments. */ 6516 if (action == PF_PASS && *m0 && fwdir == PF_FWD && 6517 (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL) 6518 action = pf_refragment6(ifp, m0, mtag); 6519 6520 return (action); 6521 } 6522 #endif /* INET6 */ 6523