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 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 155 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 156 157 /* 158 * Queue for pf_overload_task() tasks. 159 */ 160 struct pf_overload_entry { 161 SLIST_ENTRY(pf_overload_entry) next; 162 struct pf_addr addr; 163 sa_family_t af; 164 uint8_t dir; 165 struct pf_rule *rule; 166 }; 167 168 SLIST_HEAD(pf_overload_head, pf_overload_entry); 169 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue); 170 #define V_pf_overloadqueue VNET(pf_overloadqueue) 171 static VNET_DEFINE(struct task, pf_overloadtask); 172 #define V_pf_overloadtask VNET(pf_overloadtask) 173 174 static struct mtx pf_overloadqueue_mtx; 175 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 176 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 177 178 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules); 179 struct mtx pf_unlnkdrules_mtx; 180 181 static VNET_DEFINE(uma_zone_t, pf_sources_z); 182 #define V_pf_sources_z VNET(pf_sources_z) 183 uma_zone_t pf_mtag_z; 184 VNET_DEFINE(uma_zone_t, pf_state_z); 185 VNET_DEFINE(uma_zone_t, pf_state_key_z); 186 187 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]); 188 #define PFID_CPUBITS 8 189 #define PFID_CPUSHIFT (sizeof(uint64_t) * NBBY - PFID_CPUBITS) 190 #define PFID_CPUMASK ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT) 191 #define PFID_MAXID (~PFID_CPUMASK) 192 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU); 193 194 static void pf_src_tree_remove_state(struct pf_state *); 195 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 196 u_int32_t); 197 static void pf_add_threshold(struct pf_threshold *); 198 static int pf_check_threshold(struct pf_threshold *); 199 200 static void pf_change_ap(struct pf_addr *, u_int16_t *, 201 u_int16_t *, u_int16_t *, struct pf_addr *, 202 u_int16_t, u_int8_t, sa_family_t); 203 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 204 struct tcphdr *, struct pf_state_peer *); 205 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 206 struct pf_addr *, struct pf_addr *, u_int16_t, 207 u_int16_t *, u_int16_t *, u_int16_t *, 208 u_int16_t *, u_int8_t, sa_family_t); 209 static void pf_send_tcp(struct mbuf *, 210 const struct pf_rule *, sa_family_t, 211 const struct pf_addr *, const struct pf_addr *, 212 u_int16_t, u_int16_t, u_int32_t, u_int32_t, 213 u_int8_t, u_int16_t, u_int16_t, u_int8_t, int, 214 u_int16_t, struct ifnet *); 215 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 216 sa_family_t, struct pf_rule *); 217 static void pf_detach_state(struct pf_state *); 218 static int pf_state_key_attach(struct pf_state_key *, 219 struct pf_state_key *, struct pf_state *); 220 static void pf_state_key_detach(struct pf_state *, int); 221 static int pf_state_key_ctor(void *, int, void *, int); 222 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 223 static int pf_test_rule(struct pf_rule **, struct pf_state **, 224 int, struct pfi_kif *, struct mbuf *, int, 225 struct pf_pdesc *, struct pf_rule **, 226 struct pf_ruleset **, struct inpcb *); 227 static int pf_create_state(struct pf_rule *, struct pf_rule *, 228 struct pf_rule *, struct pf_pdesc *, 229 struct pf_src_node *, struct pf_state_key *, 230 struct pf_state_key *, struct mbuf *, int, 231 u_int16_t, u_int16_t, int *, struct pfi_kif *, 232 struct pf_state **, int, u_int16_t, u_int16_t, 233 int); 234 static int pf_test_fragment(struct pf_rule **, int, 235 struct pfi_kif *, struct mbuf *, void *, 236 struct pf_pdesc *, struct pf_rule **, 237 struct pf_ruleset **); 238 static int pf_tcp_track_full(struct pf_state_peer *, 239 struct pf_state_peer *, struct pf_state **, 240 struct pfi_kif *, struct mbuf *, int, 241 struct pf_pdesc *, u_short *, int *); 242 static int pf_tcp_track_sloppy(struct pf_state_peer *, 243 struct pf_state_peer *, struct pf_state **, 244 struct pf_pdesc *, u_short *); 245 static int pf_test_state_tcp(struct pf_state **, int, 246 struct pfi_kif *, struct mbuf *, int, 247 void *, struct pf_pdesc *, u_short *); 248 static int pf_test_state_udp(struct pf_state **, int, 249 struct pfi_kif *, struct mbuf *, int, 250 void *, struct pf_pdesc *); 251 static int pf_test_state_icmp(struct pf_state **, int, 252 struct pfi_kif *, struct mbuf *, int, 253 void *, struct pf_pdesc *, u_short *); 254 static int pf_test_state_other(struct pf_state **, int, 255 struct pfi_kif *, struct mbuf *, struct pf_pdesc *); 256 static u_int8_t pf_get_wscale(struct mbuf *, int, u_int16_t, 257 sa_family_t); 258 static u_int16_t pf_get_mss(struct mbuf *, int, u_int16_t, 259 sa_family_t); 260 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 261 int, u_int16_t); 262 static int pf_check_proto_cksum(struct mbuf *, int, int, 263 u_int8_t, sa_family_t); 264 static void pf_print_state_parts(struct pf_state *, 265 struct pf_state_key *, struct pf_state_key *); 266 static int pf_addr_wrap_neq(struct pf_addr_wrap *, 267 struct pf_addr_wrap *); 268 static struct pf_state *pf_find_state(struct pfi_kif *, 269 struct pf_state_key_cmp *, u_int); 270 static int pf_src_connlimit(struct pf_state **); 271 static void pf_overload_task(void *v, int pending); 272 static int pf_insert_src_node(struct pf_src_node **, 273 struct pf_rule *, struct pf_addr *, sa_family_t); 274 static u_int pf_purge_expired_states(u_int, int); 275 static void pf_purge_unlinked_rules(void); 276 static int pf_mtag_uminit(void *, int, int); 277 static void pf_mtag_free(struct m_tag *); 278 #ifdef INET 279 static void pf_route(struct mbuf **, struct pf_rule *, int, 280 struct ifnet *, struct pf_state *, 281 struct pf_pdesc *); 282 #endif /* INET */ 283 #ifdef INET6 284 static void pf_change_a6(struct pf_addr *, u_int16_t *, 285 struct pf_addr *, u_int8_t); 286 static void pf_route6(struct mbuf **, struct pf_rule *, int, 287 struct ifnet *, struct pf_state *, 288 struct pf_pdesc *); 289 #endif /* INET6 */ 290 291 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 292 293 VNET_DECLARE(int, pf_end_threads); 294 295 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 296 297 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 298 (pd)->pf_mtag->flags & PF_PACKET_LOOPED) 299 300 #define STATE_LOOKUP(i, k, d, s, pd) \ 301 do { \ 302 (s) = pf_find_state((i), (k), (d)); \ 303 if ((s) == NULL) \ 304 return (PF_DROP); \ 305 if (PACKET_LOOPED(pd)) \ 306 return (PF_PASS); \ 307 if ((d) == PF_OUT && \ 308 (((s)->rule.ptr->rt == PF_ROUTETO && \ 309 (s)->rule.ptr->direction == PF_OUT) || \ 310 ((s)->rule.ptr->rt == PF_REPLYTO && \ 311 (s)->rule.ptr->direction == PF_IN)) && \ 312 (s)->rt_kif != NULL && \ 313 (s)->rt_kif != (i)) \ 314 return (PF_PASS); \ 315 } while (0) 316 317 #define BOUND_IFACE(r, k) \ 318 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all 319 320 #define STATE_INC_COUNTERS(s) \ 321 do { \ 322 counter_u64_add(s->rule.ptr->states_cur, 1); \ 323 counter_u64_add(s->rule.ptr->states_tot, 1); \ 324 if (s->anchor.ptr != NULL) { \ 325 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 326 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 327 } \ 328 if (s->nat_rule.ptr != NULL) { \ 329 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 330 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 331 } \ 332 } while (0) 333 334 #define STATE_DEC_COUNTERS(s) \ 335 do { \ 336 if (s->nat_rule.ptr != NULL) \ 337 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 338 if (s->anchor.ptr != NULL) \ 339 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 340 counter_u64_add(s->rule.ptr->states_cur, -1); \ 341 } while (0) 342 343 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 344 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 345 VNET_DEFINE(struct pf_idhash *, pf_idhash); 346 VNET_DEFINE(struct pf_srchash *, pf_srchash); 347 348 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)"); 349 350 u_long pf_hashmask; 351 u_long pf_srchashmask; 352 static u_long pf_hashsize; 353 static u_long pf_srchashsize; 354 355 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 356 &pf_hashsize, 0, "Size of pf(4) states hashtable"); 357 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 358 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); 359 360 VNET_DEFINE(void *, pf_swi_cookie); 361 362 VNET_DEFINE(uint32_t, pf_hashseed); 363 #define V_pf_hashseed VNET(pf_hashseed) 364 365 int 366 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 367 { 368 369 switch (af) { 370 #ifdef INET 371 case AF_INET: 372 if (a->addr32[0] > b->addr32[0]) 373 return (1); 374 if (a->addr32[0] < b->addr32[0]) 375 return (-1); 376 break; 377 #endif /* INET */ 378 #ifdef INET6 379 case AF_INET6: 380 if (a->addr32[3] > b->addr32[3]) 381 return (1); 382 if (a->addr32[3] < b->addr32[3]) 383 return (-1); 384 if (a->addr32[2] > b->addr32[2]) 385 return (1); 386 if (a->addr32[2] < b->addr32[2]) 387 return (-1); 388 if (a->addr32[1] > b->addr32[1]) 389 return (1); 390 if (a->addr32[1] < b->addr32[1]) 391 return (-1); 392 if (a->addr32[0] > b->addr32[0]) 393 return (1); 394 if (a->addr32[0] < b->addr32[0]) 395 return (-1); 396 break; 397 #endif /* INET6 */ 398 default: 399 panic("%s: unknown address family %u", __func__, af); 400 } 401 return (0); 402 } 403 404 static __inline uint32_t 405 pf_hashkey(struct pf_state_key *sk) 406 { 407 uint32_t h; 408 409 h = murmur3_32_hash32((uint32_t *)sk, 410 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 411 V_pf_hashseed); 412 413 return (h & pf_hashmask); 414 } 415 416 static __inline uint32_t 417 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 418 { 419 uint32_t h; 420 421 switch (af) { 422 case AF_INET: 423 h = murmur3_32_hash32((uint32_t *)&addr->v4, 424 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 425 break; 426 case AF_INET6: 427 h = murmur3_32_hash32((uint32_t *)&addr->v6, 428 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 429 break; 430 default: 431 panic("%s: unknown address family %u", __func__, af); 432 } 433 434 return (h & pf_srchashmask); 435 } 436 437 #ifdef INET6 438 void 439 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 440 { 441 switch (af) { 442 #ifdef INET 443 case AF_INET: 444 dst->addr32[0] = src->addr32[0]; 445 break; 446 #endif /* INET */ 447 case AF_INET6: 448 dst->addr32[0] = src->addr32[0]; 449 dst->addr32[1] = src->addr32[1]; 450 dst->addr32[2] = src->addr32[2]; 451 dst->addr32[3] = src->addr32[3]; 452 break; 453 } 454 } 455 #endif /* INET6 */ 456 457 static void 458 pf_init_threshold(struct pf_threshold *threshold, 459 u_int32_t limit, u_int32_t seconds) 460 { 461 threshold->limit = limit * PF_THRESHOLD_MULT; 462 threshold->seconds = seconds; 463 threshold->count = 0; 464 threshold->last = time_uptime; 465 } 466 467 static void 468 pf_add_threshold(struct pf_threshold *threshold) 469 { 470 u_int32_t t = time_uptime, diff = t - threshold->last; 471 472 if (diff >= threshold->seconds) 473 threshold->count = 0; 474 else 475 threshold->count -= threshold->count * diff / 476 threshold->seconds; 477 threshold->count += PF_THRESHOLD_MULT; 478 threshold->last = t; 479 } 480 481 static int 482 pf_check_threshold(struct pf_threshold *threshold) 483 { 484 return (threshold->count > threshold->limit); 485 } 486 487 static int 488 pf_src_connlimit(struct pf_state **state) 489 { 490 struct pf_overload_entry *pfoe; 491 int bad = 0; 492 493 PF_STATE_LOCK_ASSERT(*state); 494 495 (*state)->src_node->conn++; 496 (*state)->src.tcp_est = 1; 497 pf_add_threshold(&(*state)->src_node->conn_rate); 498 499 if ((*state)->rule.ptr->max_src_conn && 500 (*state)->rule.ptr->max_src_conn < 501 (*state)->src_node->conn) { 502 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 503 bad++; 504 } 505 506 if ((*state)->rule.ptr->max_src_conn_rate.limit && 507 pf_check_threshold(&(*state)->src_node->conn_rate)) { 508 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 509 bad++; 510 } 511 512 if (!bad) 513 return (0); 514 515 /* Kill this state. */ 516 (*state)->timeout = PFTM_PURGE; 517 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; 518 519 if ((*state)->rule.ptr->overload_tbl == NULL) 520 return (1); 521 522 /* Schedule overloading and flushing task. */ 523 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 524 if (pfoe == NULL) 525 return (1); /* too bad :( */ 526 527 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 528 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 529 pfoe->rule = (*state)->rule.ptr; 530 pfoe->dir = (*state)->direction; 531 PF_OVERLOADQ_LOCK(); 532 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 533 PF_OVERLOADQ_UNLOCK(); 534 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 535 536 return (1); 537 } 538 539 static void 540 pf_overload_task(void *v, int pending) 541 { 542 struct pf_overload_head queue; 543 struct pfr_addr p; 544 struct pf_overload_entry *pfoe, *pfoe1; 545 uint32_t killed = 0; 546 547 CURVNET_SET((struct vnet *)v); 548 549 PF_OVERLOADQ_LOCK(); 550 queue = V_pf_overloadqueue; 551 SLIST_INIT(&V_pf_overloadqueue); 552 PF_OVERLOADQ_UNLOCK(); 553 554 bzero(&p, sizeof(p)); 555 SLIST_FOREACH(pfoe, &queue, next) { 556 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 557 if (V_pf_status.debug >= PF_DEBUG_MISC) { 558 printf("%s: blocking address ", __func__); 559 pf_print_host(&pfoe->addr, 0, pfoe->af); 560 printf("\n"); 561 } 562 563 p.pfra_af = pfoe->af; 564 switch (pfoe->af) { 565 #ifdef INET 566 case AF_INET: 567 p.pfra_net = 32; 568 p.pfra_ip4addr = pfoe->addr.v4; 569 break; 570 #endif 571 #ifdef INET6 572 case AF_INET6: 573 p.pfra_net = 128; 574 p.pfra_ip6addr = pfoe->addr.v6; 575 break; 576 #endif 577 } 578 579 PF_RULES_WLOCK(); 580 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 581 PF_RULES_WUNLOCK(); 582 } 583 584 /* 585 * Remove those entries, that don't need flushing. 586 */ 587 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 588 if (pfoe->rule->flush == 0) { 589 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 590 free(pfoe, M_PFTEMP); 591 } else 592 counter_u64_add( 593 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 594 595 /* If nothing to flush, return. */ 596 if (SLIST_EMPTY(&queue)) { 597 CURVNET_RESTORE(); 598 return; 599 } 600 601 for (int i = 0; i <= pf_hashmask; i++) { 602 struct pf_idhash *ih = &V_pf_idhash[i]; 603 struct pf_state_key *sk; 604 struct pf_state *s; 605 606 PF_HASHROW_LOCK(ih); 607 LIST_FOREACH(s, &ih->states, entry) { 608 sk = s->key[PF_SK_WIRE]; 609 SLIST_FOREACH(pfoe, &queue, next) 610 if (sk->af == pfoe->af && 611 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 612 pfoe->rule == s->rule.ptr) && 613 ((pfoe->dir == PF_OUT && 614 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 615 (pfoe->dir == PF_IN && 616 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 617 s->timeout = PFTM_PURGE; 618 s->src.state = s->dst.state = TCPS_CLOSED; 619 killed++; 620 } 621 } 622 PF_HASHROW_UNLOCK(ih); 623 } 624 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 625 free(pfoe, M_PFTEMP); 626 if (V_pf_status.debug >= PF_DEBUG_MISC) 627 printf("%s: %u states killed", __func__, killed); 628 629 CURVNET_RESTORE(); 630 } 631 632 /* 633 * Can return locked on failure, so that we can consistently 634 * allocate and insert a new one. 635 */ 636 struct pf_src_node * 637 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af, 638 int returnlocked) 639 { 640 struct pf_srchash *sh; 641 struct pf_src_node *n; 642 643 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 644 645 sh = &V_pf_srchash[pf_hashsrc(src, af)]; 646 PF_HASHROW_LOCK(sh); 647 LIST_FOREACH(n, &sh->nodes, entry) 648 if (n->rule.ptr == rule && n->af == af && 649 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 650 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 651 break; 652 if (n != NULL) { 653 n->states++; 654 PF_HASHROW_UNLOCK(sh); 655 } else if (returnlocked == 0) 656 PF_HASHROW_UNLOCK(sh); 657 658 return (n); 659 } 660 661 static int 662 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule, 663 struct pf_addr *src, sa_family_t af) 664 { 665 666 KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK || 667 rule->rpool.opts & PF_POOL_STICKYADDR), 668 ("%s for non-tracking rule %p", __func__, rule)); 669 670 if (*sn == NULL) 671 *sn = pf_find_src_node(src, rule, af, 1); 672 673 if (*sn == NULL) { 674 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 675 676 PF_HASHROW_ASSERT(sh); 677 678 if (!rule->max_src_nodes || 679 counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes) 680 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 681 else 682 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 683 1); 684 if ((*sn) == NULL) { 685 PF_HASHROW_UNLOCK(sh); 686 return (-1); 687 } 688 689 pf_init_threshold(&(*sn)->conn_rate, 690 rule->max_src_conn_rate.limit, 691 rule->max_src_conn_rate.seconds); 692 693 (*sn)->af = af; 694 (*sn)->rule.ptr = rule; 695 PF_ACPY(&(*sn)->addr, src, af); 696 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 697 (*sn)->creation = time_uptime; 698 (*sn)->ruletype = rule->action; 699 (*sn)->states = 1; 700 if ((*sn)->rule.ptr != NULL) 701 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 702 PF_HASHROW_UNLOCK(sh); 703 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 704 } else { 705 if (rule->max_src_states && 706 (*sn)->states >= rule->max_src_states) { 707 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 708 1); 709 return (-1); 710 } 711 } 712 return (0); 713 } 714 715 void 716 pf_unlink_src_node(struct pf_src_node *src) 717 { 718 719 PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]); 720 LIST_REMOVE(src, entry); 721 if (src->rule.ptr) 722 counter_u64_add(src->rule.ptr->src_nodes, -1); 723 } 724 725 u_int 726 pf_free_src_nodes(struct pf_src_node_list *head) 727 { 728 struct pf_src_node *sn, *tmp; 729 u_int count = 0; 730 731 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 732 uma_zfree(V_pf_sources_z, sn); 733 count++; 734 } 735 736 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 737 738 return (count); 739 } 740 741 void 742 pf_mtag_initialize() 743 { 744 745 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 746 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 747 UMA_ALIGN_PTR, 0); 748 } 749 750 /* Per-vnet data storage structures initialization. */ 751 void 752 pf_initialize() 753 { 754 struct pf_keyhash *kh; 755 struct pf_idhash *ih; 756 struct pf_srchash *sh; 757 u_int i; 758 759 if (pf_hashsize == 0 || !powerof2(pf_hashsize)) 760 pf_hashsize = PF_HASHSIZ; 761 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) 762 pf_srchashsize = PF_HASHSIZ / 4; 763 764 V_pf_hashseed = arc4random(); 765 766 /* States and state keys storage. */ 767 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state), 768 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 769 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 770 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 771 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 772 773 V_pf_state_key_z = uma_zcreate("pf state keys", 774 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 775 UMA_ALIGN_PTR, 0); 776 V_pf_keyhash = malloc(pf_hashsize * sizeof(struct pf_keyhash), 777 M_PFHASH, M_WAITOK | M_ZERO); 778 V_pf_idhash = malloc(pf_hashsize * sizeof(struct pf_idhash), 779 M_PFHASH, M_WAITOK | M_ZERO); 780 pf_hashmask = pf_hashsize - 1; 781 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 782 i++, kh++, ih++) { 783 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 784 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 785 } 786 787 /* Source nodes. */ 788 V_pf_sources_z = uma_zcreate("pf source nodes", 789 sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 790 0); 791 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 792 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 793 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 794 V_pf_srchash = malloc(pf_srchashsize * sizeof(struct pf_srchash), 795 M_PFHASH, M_WAITOK|M_ZERO); 796 pf_srchashmask = pf_srchashsize - 1; 797 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) 798 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 799 800 /* ALTQ */ 801 TAILQ_INIT(&V_pf_altqs[0]); 802 TAILQ_INIT(&V_pf_altqs[1]); 803 TAILQ_INIT(&V_pf_pabuf); 804 V_pf_altqs_active = &V_pf_altqs[0]; 805 V_pf_altqs_inactive = &V_pf_altqs[1]; 806 807 808 /* Send & overload+flush queues. */ 809 STAILQ_INIT(&V_pf_sendqueue); 810 SLIST_INIT(&V_pf_overloadqueue); 811 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 812 mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF); 813 mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL, 814 MTX_DEF); 815 816 /* Unlinked, but may be referenced rules. */ 817 TAILQ_INIT(&V_pf_unlinked_rules); 818 mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF); 819 } 820 821 void 822 pf_mtag_cleanup() 823 { 824 825 uma_zdestroy(pf_mtag_z); 826 } 827 828 void 829 pf_cleanup() 830 { 831 struct pf_keyhash *kh; 832 struct pf_idhash *ih; 833 struct pf_srchash *sh; 834 struct pf_send_entry *pfse, *next; 835 u_int i; 836 837 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 838 i++, kh++, ih++) { 839 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 840 __func__)); 841 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 842 __func__)); 843 mtx_destroy(&kh->lock); 844 mtx_destroy(&ih->lock); 845 } 846 free(V_pf_keyhash, M_PFHASH); 847 free(V_pf_idhash, M_PFHASH); 848 849 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 850 KASSERT(LIST_EMPTY(&sh->nodes), 851 ("%s: source node hash not empty", __func__)); 852 mtx_destroy(&sh->lock); 853 } 854 free(V_pf_srchash, M_PFHASH); 855 856 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 857 m_freem(pfse->pfse_m); 858 free(pfse, M_PFTEMP); 859 } 860 861 mtx_destroy(&pf_sendqueue_mtx); 862 mtx_destroy(&pf_overloadqueue_mtx); 863 mtx_destroy(&pf_unlnkdrules_mtx); 864 865 uma_zdestroy(V_pf_sources_z); 866 uma_zdestroy(V_pf_state_z); 867 uma_zdestroy(V_pf_state_key_z); 868 } 869 870 static int 871 pf_mtag_uminit(void *mem, int size, int how) 872 { 873 struct m_tag *t; 874 875 t = (struct m_tag *)mem; 876 t->m_tag_cookie = MTAG_ABI_COMPAT; 877 t->m_tag_id = PACKET_TAG_PF; 878 t->m_tag_len = sizeof(struct pf_mtag); 879 t->m_tag_free = pf_mtag_free; 880 881 return (0); 882 } 883 884 static void 885 pf_mtag_free(struct m_tag *t) 886 { 887 888 uma_zfree(pf_mtag_z, t); 889 } 890 891 struct pf_mtag * 892 pf_get_mtag(struct mbuf *m) 893 { 894 struct m_tag *mtag; 895 896 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 897 return ((struct pf_mtag *)(mtag + 1)); 898 899 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 900 if (mtag == NULL) 901 return (NULL); 902 bzero(mtag + 1, sizeof(struct pf_mtag)); 903 m_tag_prepend(m, mtag); 904 905 return ((struct pf_mtag *)(mtag + 1)); 906 } 907 908 static int 909 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 910 struct pf_state *s) 911 { 912 struct pf_keyhash *khs, *khw, *kh; 913 struct pf_state_key *sk, *cur; 914 struct pf_state *si, *olds = NULL; 915 int idx; 916 917 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 918 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 919 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 920 921 /* 922 * We need to lock hash slots of both keys. To avoid deadlock 923 * we always lock the slot with lower address first. Unlock order 924 * isn't important. 925 * 926 * We also need to lock ID hash slot before dropping key 927 * locks. On success we return with ID hash slot locked. 928 */ 929 930 if (skw == sks) { 931 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 932 PF_HASHROW_LOCK(khs); 933 } else { 934 khs = &V_pf_keyhash[pf_hashkey(sks)]; 935 khw = &V_pf_keyhash[pf_hashkey(skw)]; 936 if (khs == khw) { 937 PF_HASHROW_LOCK(khs); 938 } else if (khs < khw) { 939 PF_HASHROW_LOCK(khs); 940 PF_HASHROW_LOCK(khw); 941 } else { 942 PF_HASHROW_LOCK(khw); 943 PF_HASHROW_LOCK(khs); 944 } 945 } 946 947 #define KEYS_UNLOCK() do { \ 948 if (khs != khw) { \ 949 PF_HASHROW_UNLOCK(khs); \ 950 PF_HASHROW_UNLOCK(khw); \ 951 } else \ 952 PF_HASHROW_UNLOCK(khs); \ 953 } while (0) 954 955 /* 956 * First run: start with wire key. 957 */ 958 sk = skw; 959 kh = khw; 960 idx = PF_SK_WIRE; 961 962 keyattach: 963 LIST_FOREACH(cur, &kh->keys, entry) 964 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 965 break; 966 967 if (cur != NULL) { 968 /* Key exists. Check for same kif, if none, add to key. */ 969 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 970 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 971 972 PF_HASHROW_LOCK(ih); 973 if (si->kif == s->kif && 974 si->direction == s->direction) { 975 if (sk->proto == IPPROTO_TCP && 976 si->src.state >= TCPS_FIN_WAIT_2 && 977 si->dst.state >= TCPS_FIN_WAIT_2) { 978 /* 979 * New state matches an old >FIN_WAIT_2 980 * state. We can't drop key hash locks, 981 * thus we can't unlink it properly. 982 * 983 * As a workaround we drop it into 984 * TCPS_CLOSED state, schedule purge 985 * ASAP and push it into the very end 986 * of the slot TAILQ, so that it won't 987 * conflict with our new state. 988 */ 989 si->src.state = si->dst.state = 990 TCPS_CLOSED; 991 si->timeout = PFTM_PURGE; 992 olds = si; 993 } else { 994 if (V_pf_status.debug >= PF_DEBUG_MISC) { 995 printf("pf: %s key attach " 996 "failed on %s: ", 997 (idx == PF_SK_WIRE) ? 998 "wire" : "stack", 999 s->kif->pfik_name); 1000 pf_print_state_parts(s, 1001 (idx == PF_SK_WIRE) ? 1002 sk : NULL, 1003 (idx == PF_SK_STACK) ? 1004 sk : NULL); 1005 printf(", existing: "); 1006 pf_print_state_parts(si, 1007 (idx == PF_SK_WIRE) ? 1008 sk : NULL, 1009 (idx == PF_SK_STACK) ? 1010 sk : NULL); 1011 printf("\n"); 1012 } 1013 PF_HASHROW_UNLOCK(ih); 1014 KEYS_UNLOCK(); 1015 uma_zfree(V_pf_state_key_z, sk); 1016 if (idx == PF_SK_STACK) 1017 pf_detach_state(s); 1018 return (EEXIST); /* collision! */ 1019 } 1020 } 1021 PF_HASHROW_UNLOCK(ih); 1022 } 1023 uma_zfree(V_pf_state_key_z, sk); 1024 s->key[idx] = cur; 1025 } else { 1026 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1027 s->key[idx] = sk; 1028 } 1029 1030 stateattach: 1031 /* List is sorted, if-bound states before floating. */ 1032 if (s->kif == V_pfi_all) 1033 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1034 else 1035 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1036 1037 if (olds) { 1038 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1039 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1040 key_list[idx]); 1041 olds = NULL; 1042 } 1043 1044 /* 1045 * Attach done. See how should we (or should not?) 1046 * attach a second key. 1047 */ 1048 if (sks == skw) { 1049 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1050 idx = PF_SK_STACK; 1051 sks = NULL; 1052 goto stateattach; 1053 } else if (sks != NULL) { 1054 /* 1055 * Continue attaching with stack key. 1056 */ 1057 sk = sks; 1058 kh = khs; 1059 idx = PF_SK_STACK; 1060 sks = NULL; 1061 goto keyattach; 1062 } 1063 1064 PF_STATE_LOCK(s); 1065 KEYS_UNLOCK(); 1066 1067 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1068 ("%s failure", __func__)); 1069 1070 return (0); 1071 #undef KEYS_UNLOCK 1072 } 1073 1074 static void 1075 pf_detach_state(struct pf_state *s) 1076 { 1077 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1078 struct pf_keyhash *kh; 1079 1080 if (sks != NULL) { 1081 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1082 PF_HASHROW_LOCK(kh); 1083 if (s->key[PF_SK_STACK] != NULL) 1084 pf_state_key_detach(s, PF_SK_STACK); 1085 /* 1086 * If both point to same key, then we are done. 1087 */ 1088 if (sks == s->key[PF_SK_WIRE]) { 1089 pf_state_key_detach(s, PF_SK_WIRE); 1090 PF_HASHROW_UNLOCK(kh); 1091 return; 1092 } 1093 PF_HASHROW_UNLOCK(kh); 1094 } 1095 1096 if (s->key[PF_SK_WIRE] != NULL) { 1097 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1098 PF_HASHROW_LOCK(kh); 1099 if (s->key[PF_SK_WIRE] != NULL) 1100 pf_state_key_detach(s, PF_SK_WIRE); 1101 PF_HASHROW_UNLOCK(kh); 1102 } 1103 } 1104 1105 static void 1106 pf_state_key_detach(struct pf_state *s, int idx) 1107 { 1108 struct pf_state_key *sk = s->key[idx]; 1109 #ifdef INVARIANTS 1110 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1111 1112 PF_HASHROW_ASSERT(kh); 1113 #endif 1114 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1115 s->key[idx] = NULL; 1116 1117 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1118 LIST_REMOVE(sk, entry); 1119 uma_zfree(V_pf_state_key_z, sk); 1120 } 1121 } 1122 1123 static int 1124 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1125 { 1126 struct pf_state_key *sk = mem; 1127 1128 bzero(sk, sizeof(struct pf_state_key_cmp)); 1129 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1130 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1131 1132 return (0); 1133 } 1134 1135 struct pf_state_key * 1136 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1137 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1138 { 1139 struct pf_state_key *sk; 1140 1141 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1142 if (sk == NULL) 1143 return (NULL); 1144 1145 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1146 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1147 sk->port[pd->sidx] = sport; 1148 sk->port[pd->didx] = dport; 1149 sk->proto = pd->proto; 1150 sk->af = pd->af; 1151 1152 return (sk); 1153 } 1154 1155 struct pf_state_key * 1156 pf_state_key_clone(struct pf_state_key *orig) 1157 { 1158 struct pf_state_key *sk; 1159 1160 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1161 if (sk == NULL) 1162 return (NULL); 1163 1164 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1165 1166 return (sk); 1167 } 1168 1169 int 1170 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw, 1171 struct pf_state_key *sks, struct pf_state *s) 1172 { 1173 struct pf_idhash *ih; 1174 struct pf_state *cur; 1175 int error; 1176 1177 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1178 ("%s: sks not pristine", __func__)); 1179 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1180 ("%s: skw not pristine", __func__)); 1181 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1182 1183 s->kif = kif; 1184 1185 if (s->id == 0 && s->creatorid == 0) { 1186 /* XXX: should be atomic, but probability of collision low */ 1187 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID) 1188 V_pf_stateid[curcpu] = 1; 1189 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT; 1190 s->id = htobe64(s->id); 1191 s->creatorid = V_pf_status.hostid; 1192 } 1193 1194 /* Returns with ID locked on success. */ 1195 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1196 return (error); 1197 1198 ih = &V_pf_idhash[PF_IDHASH(s)]; 1199 PF_HASHROW_ASSERT(ih); 1200 LIST_FOREACH(cur, &ih->states, entry) 1201 if (cur->id == s->id && cur->creatorid == s->creatorid) 1202 break; 1203 1204 if (cur != NULL) { 1205 PF_HASHROW_UNLOCK(ih); 1206 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1207 printf("pf: state ID collision: " 1208 "id: %016llx creatorid: %08x\n", 1209 (unsigned long long)be64toh(s->id), 1210 ntohl(s->creatorid)); 1211 } 1212 pf_detach_state(s); 1213 return (EEXIST); 1214 } 1215 LIST_INSERT_HEAD(&ih->states, s, entry); 1216 /* One for keys, one for ID hash. */ 1217 refcount_init(&s->refs, 2); 1218 1219 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1220 if (pfsync_insert_state_ptr != NULL) 1221 pfsync_insert_state_ptr(s); 1222 1223 /* Returns locked. */ 1224 return (0); 1225 } 1226 1227 /* 1228 * Find state by ID: returns with locked row on success. 1229 */ 1230 struct pf_state * 1231 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1232 { 1233 struct pf_idhash *ih; 1234 struct pf_state *s; 1235 1236 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1237 1238 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; 1239 1240 PF_HASHROW_LOCK(ih); 1241 LIST_FOREACH(s, &ih->states, entry) 1242 if (s->id == id && s->creatorid == creatorid) 1243 break; 1244 1245 if (s == NULL) 1246 PF_HASHROW_UNLOCK(ih); 1247 1248 return (s); 1249 } 1250 1251 /* 1252 * Find state by key. 1253 * Returns with ID hash slot locked on success. 1254 */ 1255 static struct pf_state * 1256 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir) 1257 { 1258 struct pf_keyhash *kh; 1259 struct pf_state_key *sk; 1260 struct pf_state *s; 1261 int idx; 1262 1263 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1264 1265 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1266 1267 PF_HASHROW_LOCK(kh); 1268 LIST_FOREACH(sk, &kh->keys, entry) 1269 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1270 break; 1271 if (sk == NULL) { 1272 PF_HASHROW_UNLOCK(kh); 1273 return (NULL); 1274 } 1275 1276 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1277 1278 /* List is sorted, if-bound states before floating ones. */ 1279 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1280 if (s->kif == V_pfi_all || s->kif == kif) { 1281 PF_STATE_LOCK(s); 1282 PF_HASHROW_UNLOCK(kh); 1283 if (s->timeout >= PFTM_MAX) { 1284 /* 1285 * State is either being processed by 1286 * pf_unlink_state() in an other thread, or 1287 * is scheduled for immediate expiry. 1288 */ 1289 PF_STATE_UNLOCK(s); 1290 return (NULL); 1291 } 1292 return (s); 1293 } 1294 PF_HASHROW_UNLOCK(kh); 1295 1296 return (NULL); 1297 } 1298 1299 struct pf_state * 1300 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1301 { 1302 struct pf_keyhash *kh; 1303 struct pf_state_key *sk; 1304 struct pf_state *s, *ret = NULL; 1305 int idx, inout = 0; 1306 1307 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1308 1309 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1310 1311 PF_HASHROW_LOCK(kh); 1312 LIST_FOREACH(sk, &kh->keys, entry) 1313 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1314 break; 1315 if (sk == NULL) { 1316 PF_HASHROW_UNLOCK(kh); 1317 return (NULL); 1318 } 1319 switch (dir) { 1320 case PF_IN: 1321 idx = PF_SK_WIRE; 1322 break; 1323 case PF_OUT: 1324 idx = PF_SK_STACK; 1325 break; 1326 case PF_INOUT: 1327 idx = PF_SK_WIRE; 1328 inout = 1; 1329 break; 1330 default: 1331 panic("%s: dir %u", __func__, dir); 1332 } 1333 second_run: 1334 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1335 if (more == NULL) { 1336 PF_HASHROW_UNLOCK(kh); 1337 return (s); 1338 } 1339 1340 if (ret) 1341 (*more)++; 1342 else 1343 ret = s; 1344 } 1345 if (inout == 1) { 1346 inout = 0; 1347 idx = PF_SK_STACK; 1348 goto second_run; 1349 } 1350 PF_HASHROW_UNLOCK(kh); 1351 1352 return (ret); 1353 } 1354 1355 /* END state table stuff */ 1356 1357 static void 1358 pf_send(struct pf_send_entry *pfse) 1359 { 1360 1361 PF_SENDQ_LOCK(); 1362 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1363 PF_SENDQ_UNLOCK(); 1364 swi_sched(V_pf_swi_cookie, 0); 1365 } 1366 1367 void 1368 pf_intr(void *v) 1369 { 1370 struct pf_send_head queue; 1371 struct pf_send_entry *pfse, *next; 1372 1373 CURVNET_SET((struct vnet *)v); 1374 1375 PF_SENDQ_LOCK(); 1376 queue = V_pf_sendqueue; 1377 STAILQ_INIT(&V_pf_sendqueue); 1378 PF_SENDQ_UNLOCK(); 1379 1380 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1381 switch (pfse->pfse_type) { 1382 #ifdef INET 1383 case PFSE_IP: 1384 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL); 1385 break; 1386 case PFSE_ICMP: 1387 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1388 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1389 break; 1390 #endif /* INET */ 1391 #ifdef INET6 1392 case PFSE_IP6: 1393 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL, 1394 NULL); 1395 break; 1396 case PFSE_ICMP6: 1397 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 1398 pfse->icmpopts.code, pfse->icmpopts.mtu); 1399 break; 1400 #endif /* INET6 */ 1401 default: 1402 panic("%s: unknown type", __func__); 1403 } 1404 free(pfse, M_PFTEMP); 1405 } 1406 CURVNET_RESTORE(); 1407 } 1408 1409 void 1410 pf_purge_thread(void *v) 1411 { 1412 u_int idx = 0; 1413 1414 CURVNET_SET((struct vnet *)v); 1415 1416 for (;;) { 1417 PF_RULES_RLOCK(); 1418 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10); 1419 1420 if (V_pf_end_threads) { 1421 /* 1422 * To cleanse up all kifs and rules we need 1423 * two runs: first one clears reference flags, 1424 * then pf_purge_expired_states() doesn't 1425 * raise them, and then second run frees. 1426 */ 1427 PF_RULES_RUNLOCK(); 1428 pf_purge_unlinked_rules(); 1429 pfi_kif_purge(); 1430 1431 /* 1432 * Now purge everything. 1433 */ 1434 pf_purge_expired_states(0, pf_hashmask); 1435 pf_purge_expired_fragments(); 1436 pf_purge_expired_src_nodes(); 1437 1438 /* 1439 * Now all kifs & rules should be unreferenced, 1440 * thus should be successfully freed. 1441 */ 1442 pf_purge_unlinked_rules(); 1443 pfi_kif_purge(); 1444 1445 /* 1446 * Announce success and exit. 1447 */ 1448 PF_RULES_RLOCK(); 1449 V_pf_end_threads++; 1450 PF_RULES_RUNLOCK(); 1451 wakeup(pf_purge_thread); 1452 kproc_exit(0); 1453 } 1454 PF_RULES_RUNLOCK(); 1455 1456 /* Process 1/interval fraction of the state table every run. */ 1457 idx = pf_purge_expired_states(idx, pf_hashmask / 1458 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1459 1460 /* Purge other expired types every PFTM_INTERVAL seconds. */ 1461 if (idx == 0) { 1462 /* 1463 * Order is important: 1464 * - states and src nodes reference rules 1465 * - states and rules reference kifs 1466 */ 1467 pf_purge_expired_fragments(); 1468 pf_purge_expired_src_nodes(); 1469 pf_purge_unlinked_rules(); 1470 pfi_kif_purge(); 1471 } 1472 } 1473 /* not reached */ 1474 CURVNET_RESTORE(); 1475 } 1476 1477 u_int32_t 1478 pf_state_expires(const struct pf_state *state) 1479 { 1480 u_int32_t timeout; 1481 u_int32_t start; 1482 u_int32_t end; 1483 u_int32_t states; 1484 1485 /* handle all PFTM_* > PFTM_MAX here */ 1486 if (state->timeout == PFTM_PURGE) 1487 return (time_uptime); 1488 KASSERT(state->timeout != PFTM_UNLINKED, 1489 ("pf_state_expires: timeout == PFTM_UNLINKED")); 1490 KASSERT((state->timeout < PFTM_MAX), 1491 ("pf_state_expires: timeout > PFTM_MAX")); 1492 timeout = state->rule.ptr->timeout[state->timeout]; 1493 if (!timeout) 1494 timeout = V_pf_default_rule.timeout[state->timeout]; 1495 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1496 if (start) { 1497 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1498 states = counter_u64_fetch(state->rule.ptr->states_cur); 1499 } else { 1500 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1501 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1502 states = V_pf_status.states; 1503 } 1504 if (end && states > start && start < end) { 1505 if (states < end) 1506 return (state->expire + timeout * (end - states) / 1507 (end - start)); 1508 else 1509 return (time_uptime); 1510 } 1511 return (state->expire + timeout); 1512 } 1513 1514 void 1515 pf_purge_expired_src_nodes() 1516 { 1517 struct pf_src_node_list freelist; 1518 struct pf_srchash *sh; 1519 struct pf_src_node *cur, *next; 1520 int i; 1521 1522 LIST_INIT(&freelist); 1523 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1524 PF_HASHROW_LOCK(sh); 1525 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 1526 if (cur->states == 0 && cur->expire <= time_uptime) { 1527 pf_unlink_src_node(cur); 1528 LIST_INSERT_HEAD(&freelist, cur, entry); 1529 } else if (cur->rule.ptr != NULL) 1530 cur->rule.ptr->rule_flag |= PFRULE_REFS; 1531 PF_HASHROW_UNLOCK(sh); 1532 } 1533 1534 pf_free_src_nodes(&freelist); 1535 1536 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 1537 } 1538 1539 static void 1540 pf_src_tree_remove_state(struct pf_state *s) 1541 { 1542 struct pf_src_node *sn; 1543 struct pf_srchash *sh; 1544 uint32_t timeout; 1545 1546 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 1547 s->rule.ptr->timeout[PFTM_SRC_NODE] : 1548 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 1549 1550 if (s->src_node != NULL) { 1551 sn = s->src_node; 1552 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 1553 PF_HASHROW_LOCK(sh); 1554 if (s->src.tcp_est) 1555 --sn->conn; 1556 if (--sn->states == 0) 1557 sn->expire = time_uptime + timeout; 1558 PF_HASHROW_UNLOCK(sh); 1559 } 1560 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 1561 sn = s->nat_src_node; 1562 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 1563 PF_HASHROW_LOCK(sh); 1564 if (--sn->states == 0) 1565 sn->expire = time_uptime + timeout; 1566 PF_HASHROW_UNLOCK(sh); 1567 } 1568 s->src_node = s->nat_src_node = NULL; 1569 } 1570 1571 /* 1572 * Unlink and potentilly free a state. Function may be 1573 * called with ID hash row locked, but always returns 1574 * unlocked, since it needs to go through key hash locking. 1575 */ 1576 int 1577 pf_unlink_state(struct pf_state *s, u_int flags) 1578 { 1579 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 1580 1581 if ((flags & PF_ENTER_LOCKED) == 0) 1582 PF_HASHROW_LOCK(ih); 1583 else 1584 PF_HASHROW_ASSERT(ih); 1585 1586 if (s->timeout == PFTM_UNLINKED) { 1587 /* 1588 * State is being processed 1589 * by pf_unlink_state() in 1590 * an other thread. 1591 */ 1592 PF_HASHROW_UNLOCK(ih); 1593 return (0); /* XXXGL: undefined actually */ 1594 } 1595 1596 if (s->src.state == PF_TCPS_PROXY_DST) { 1597 /* XXX wire key the right one? */ 1598 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af, 1599 &s->key[PF_SK_WIRE]->addr[1], 1600 &s->key[PF_SK_WIRE]->addr[0], 1601 s->key[PF_SK_WIRE]->port[1], 1602 s->key[PF_SK_WIRE]->port[0], 1603 s->src.seqhi, s->src.seqlo + 1, 1604 TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL); 1605 } 1606 1607 LIST_REMOVE(s, entry); 1608 pf_src_tree_remove_state(s); 1609 1610 if (pfsync_delete_state_ptr != NULL) 1611 pfsync_delete_state_ptr(s); 1612 1613 STATE_DEC_COUNTERS(s); 1614 1615 s->timeout = PFTM_UNLINKED; 1616 1617 PF_HASHROW_UNLOCK(ih); 1618 1619 pf_detach_state(s); 1620 refcount_release(&s->refs); 1621 1622 return (pf_release_state(s)); 1623 } 1624 1625 void 1626 pf_free_state(struct pf_state *cur) 1627 { 1628 1629 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 1630 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 1631 cur->timeout)); 1632 1633 pf_normalize_tcp_cleanup(cur); 1634 uma_zfree(V_pf_state_z, cur); 1635 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 1636 } 1637 1638 /* 1639 * Called only from pf_purge_thread(), thus serialized. 1640 */ 1641 static u_int 1642 pf_purge_expired_states(u_int i, int maxcheck) 1643 { 1644 struct pf_idhash *ih; 1645 struct pf_state *s; 1646 1647 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1648 1649 /* 1650 * Go through hash and unlink states that expire now. 1651 */ 1652 while (maxcheck > 0) { 1653 1654 ih = &V_pf_idhash[i]; 1655 relock: 1656 PF_HASHROW_LOCK(ih); 1657 LIST_FOREACH(s, &ih->states, entry) { 1658 if (pf_state_expires(s) <= time_uptime) { 1659 V_pf_status.states -= 1660 pf_unlink_state(s, PF_ENTER_LOCKED); 1661 goto relock; 1662 } 1663 s->rule.ptr->rule_flag |= PFRULE_REFS; 1664 if (s->nat_rule.ptr != NULL) 1665 s->nat_rule.ptr->rule_flag |= PFRULE_REFS; 1666 if (s->anchor.ptr != NULL) 1667 s->anchor.ptr->rule_flag |= PFRULE_REFS; 1668 s->kif->pfik_flags |= PFI_IFLAG_REFS; 1669 if (s->rt_kif) 1670 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 1671 } 1672 PF_HASHROW_UNLOCK(ih); 1673 1674 /* Return when we hit end of hash. */ 1675 if (++i > pf_hashmask) { 1676 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1677 return (0); 1678 } 1679 1680 maxcheck--; 1681 } 1682 1683 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1684 1685 return (i); 1686 } 1687 1688 static void 1689 pf_purge_unlinked_rules() 1690 { 1691 struct pf_rulequeue tmpq; 1692 struct pf_rule *r, *r1; 1693 1694 /* 1695 * If we have overloading task pending, then we'd 1696 * better skip purging this time. There is a tiny 1697 * probability that overloading task references 1698 * an already unlinked rule. 1699 */ 1700 PF_OVERLOADQ_LOCK(); 1701 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 1702 PF_OVERLOADQ_UNLOCK(); 1703 return; 1704 } 1705 PF_OVERLOADQ_UNLOCK(); 1706 1707 /* 1708 * Do naive mark-and-sweep garbage collecting of old rules. 1709 * Reference flag is raised by pf_purge_expired_states() 1710 * and pf_purge_expired_src_nodes(). 1711 * 1712 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 1713 * use a temporary queue. 1714 */ 1715 TAILQ_INIT(&tmpq); 1716 PF_UNLNKDRULES_LOCK(); 1717 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 1718 if (!(r->rule_flag & PFRULE_REFS)) { 1719 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 1720 TAILQ_INSERT_TAIL(&tmpq, r, entries); 1721 } else 1722 r->rule_flag &= ~PFRULE_REFS; 1723 } 1724 PF_UNLNKDRULES_UNLOCK(); 1725 1726 if (!TAILQ_EMPTY(&tmpq)) { 1727 PF_RULES_WLOCK(); 1728 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 1729 TAILQ_REMOVE(&tmpq, r, entries); 1730 pf_free_rule(r); 1731 } 1732 PF_RULES_WUNLOCK(); 1733 } 1734 } 1735 1736 void 1737 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 1738 { 1739 switch (af) { 1740 #ifdef INET 1741 case AF_INET: { 1742 u_int32_t a = ntohl(addr->addr32[0]); 1743 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 1744 (a>>8)&255, a&255); 1745 if (p) { 1746 p = ntohs(p); 1747 printf(":%u", p); 1748 } 1749 break; 1750 } 1751 #endif /* INET */ 1752 #ifdef INET6 1753 case AF_INET6: { 1754 u_int16_t b; 1755 u_int8_t i, curstart, curend, maxstart, maxend; 1756 curstart = curend = maxstart = maxend = 255; 1757 for (i = 0; i < 8; i++) { 1758 if (!addr->addr16[i]) { 1759 if (curstart == 255) 1760 curstart = i; 1761 curend = i; 1762 } else { 1763 if ((curend - curstart) > 1764 (maxend - maxstart)) { 1765 maxstart = curstart; 1766 maxend = curend; 1767 } 1768 curstart = curend = 255; 1769 } 1770 } 1771 if ((curend - curstart) > 1772 (maxend - maxstart)) { 1773 maxstart = curstart; 1774 maxend = curend; 1775 } 1776 for (i = 0; i < 8; i++) { 1777 if (i >= maxstart && i <= maxend) { 1778 if (i == 0) 1779 printf(":"); 1780 if (i == maxend) 1781 printf(":"); 1782 } else { 1783 b = ntohs(addr->addr16[i]); 1784 printf("%x", b); 1785 if (i < 7) 1786 printf(":"); 1787 } 1788 } 1789 if (p) { 1790 p = ntohs(p); 1791 printf("[%u]", p); 1792 } 1793 break; 1794 } 1795 #endif /* INET6 */ 1796 } 1797 } 1798 1799 void 1800 pf_print_state(struct pf_state *s) 1801 { 1802 pf_print_state_parts(s, NULL, NULL); 1803 } 1804 1805 static void 1806 pf_print_state_parts(struct pf_state *s, 1807 struct pf_state_key *skwp, struct pf_state_key *sksp) 1808 { 1809 struct pf_state_key *skw, *sks; 1810 u_int8_t proto, dir; 1811 1812 /* Do our best to fill these, but they're skipped if NULL */ 1813 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 1814 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 1815 proto = skw ? skw->proto : (sks ? sks->proto : 0); 1816 dir = s ? s->direction : 0; 1817 1818 switch (proto) { 1819 case IPPROTO_IPV4: 1820 printf("IPv4"); 1821 break; 1822 case IPPROTO_IPV6: 1823 printf("IPv6"); 1824 break; 1825 case IPPROTO_TCP: 1826 printf("TCP"); 1827 break; 1828 case IPPROTO_UDP: 1829 printf("UDP"); 1830 break; 1831 case IPPROTO_ICMP: 1832 printf("ICMP"); 1833 break; 1834 case IPPROTO_ICMPV6: 1835 printf("ICMPv6"); 1836 break; 1837 default: 1838 printf("%u", skw->proto); 1839 break; 1840 } 1841 switch (dir) { 1842 case PF_IN: 1843 printf(" in"); 1844 break; 1845 case PF_OUT: 1846 printf(" out"); 1847 break; 1848 } 1849 if (skw) { 1850 printf(" wire: "); 1851 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 1852 printf(" "); 1853 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 1854 } 1855 if (sks) { 1856 printf(" stack: "); 1857 if (sks != skw) { 1858 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 1859 printf(" "); 1860 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 1861 } else 1862 printf("-"); 1863 } 1864 if (s) { 1865 if (proto == IPPROTO_TCP) { 1866 printf(" [lo=%u high=%u win=%u modulator=%u", 1867 s->src.seqlo, s->src.seqhi, 1868 s->src.max_win, s->src.seqdiff); 1869 if (s->src.wscale && s->dst.wscale) 1870 printf(" wscale=%u", 1871 s->src.wscale & PF_WSCALE_MASK); 1872 printf("]"); 1873 printf(" [lo=%u high=%u win=%u modulator=%u", 1874 s->dst.seqlo, s->dst.seqhi, 1875 s->dst.max_win, s->dst.seqdiff); 1876 if (s->src.wscale && s->dst.wscale) 1877 printf(" wscale=%u", 1878 s->dst.wscale & PF_WSCALE_MASK); 1879 printf("]"); 1880 } 1881 printf(" %u:%u", s->src.state, s->dst.state); 1882 } 1883 } 1884 1885 void 1886 pf_print_flags(u_int8_t f) 1887 { 1888 if (f) 1889 printf(" "); 1890 if (f & TH_FIN) 1891 printf("F"); 1892 if (f & TH_SYN) 1893 printf("S"); 1894 if (f & TH_RST) 1895 printf("R"); 1896 if (f & TH_PUSH) 1897 printf("P"); 1898 if (f & TH_ACK) 1899 printf("A"); 1900 if (f & TH_URG) 1901 printf("U"); 1902 if (f & TH_ECE) 1903 printf("E"); 1904 if (f & TH_CWR) 1905 printf("W"); 1906 } 1907 1908 #define PF_SET_SKIP_STEPS(i) \ 1909 do { \ 1910 while (head[i] != cur) { \ 1911 head[i]->skip[i].ptr = cur; \ 1912 head[i] = TAILQ_NEXT(head[i], entries); \ 1913 } \ 1914 } while (0) 1915 1916 void 1917 pf_calc_skip_steps(struct pf_rulequeue *rules) 1918 { 1919 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT]; 1920 int i; 1921 1922 cur = TAILQ_FIRST(rules); 1923 prev = cur; 1924 for (i = 0; i < PF_SKIP_COUNT; ++i) 1925 head[i] = cur; 1926 while (cur != NULL) { 1927 1928 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 1929 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 1930 if (cur->direction != prev->direction) 1931 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 1932 if (cur->af != prev->af) 1933 PF_SET_SKIP_STEPS(PF_SKIP_AF); 1934 if (cur->proto != prev->proto) 1935 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 1936 if (cur->src.neg != prev->src.neg || 1937 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 1938 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 1939 if (cur->src.port[0] != prev->src.port[0] || 1940 cur->src.port[1] != prev->src.port[1] || 1941 cur->src.port_op != prev->src.port_op) 1942 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 1943 if (cur->dst.neg != prev->dst.neg || 1944 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 1945 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 1946 if (cur->dst.port[0] != prev->dst.port[0] || 1947 cur->dst.port[1] != prev->dst.port[1] || 1948 cur->dst.port_op != prev->dst.port_op) 1949 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 1950 1951 prev = cur; 1952 cur = TAILQ_NEXT(cur, entries); 1953 } 1954 for (i = 0; i < PF_SKIP_COUNT; ++i) 1955 PF_SET_SKIP_STEPS(i); 1956 } 1957 1958 static int 1959 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 1960 { 1961 if (aw1->type != aw2->type) 1962 return (1); 1963 switch (aw1->type) { 1964 case PF_ADDR_ADDRMASK: 1965 case PF_ADDR_RANGE: 1966 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0)) 1967 return (1); 1968 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0)) 1969 return (1); 1970 return (0); 1971 case PF_ADDR_DYNIFTL: 1972 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 1973 case PF_ADDR_NOROUTE: 1974 case PF_ADDR_URPFFAILED: 1975 return (0); 1976 case PF_ADDR_TABLE: 1977 return (aw1->p.tbl != aw2->p.tbl); 1978 default: 1979 printf("invalid address type: %d\n", aw1->type); 1980 return (1); 1981 } 1982 } 1983 1984 u_int16_t 1985 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 1986 { 1987 u_int32_t l; 1988 1989 if (udp && !cksum) 1990 return (0x0000); 1991 l = cksum + old - new; 1992 l = (l >> 16) + (l & 65535); 1993 l = l & 65535; 1994 if (udp && !l) 1995 return (0xFFFF); 1996 return (l); 1997 } 1998 1999 static void 2000 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc, 2001 struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af) 2002 { 2003 struct pf_addr ao; 2004 u_int16_t po = *p; 2005 2006 PF_ACPY(&ao, a, af); 2007 PF_ACPY(a, an, af); 2008 2009 *p = pn; 2010 2011 switch (af) { 2012 #ifdef INET 2013 case AF_INET: 2014 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2015 ao.addr16[0], an->addr16[0], 0), 2016 ao.addr16[1], an->addr16[1], 0); 2017 *p = pn; 2018 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 2019 ao.addr16[0], an->addr16[0], u), 2020 ao.addr16[1], an->addr16[1], u), 2021 po, pn, u); 2022 break; 2023 #endif /* INET */ 2024 #ifdef INET6 2025 case AF_INET6: 2026 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2027 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2028 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 2029 ao.addr16[0], an->addr16[0], u), 2030 ao.addr16[1], an->addr16[1], u), 2031 ao.addr16[2], an->addr16[2], u), 2032 ao.addr16[3], an->addr16[3], u), 2033 ao.addr16[4], an->addr16[4], u), 2034 ao.addr16[5], an->addr16[5], u), 2035 ao.addr16[6], an->addr16[6], u), 2036 ao.addr16[7], an->addr16[7], u), 2037 po, pn, u); 2038 break; 2039 #endif /* INET6 */ 2040 } 2041 } 2042 2043 2044 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2045 void 2046 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2047 { 2048 u_int32_t ao; 2049 2050 memcpy(&ao, a, sizeof(ao)); 2051 memcpy(a, &an, sizeof(u_int32_t)); 2052 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2053 ao % 65536, an % 65536, u); 2054 } 2055 2056 #ifdef INET6 2057 static void 2058 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2059 { 2060 struct pf_addr ao; 2061 2062 PF_ACPY(&ao, a, AF_INET6); 2063 PF_ACPY(a, an, AF_INET6); 2064 2065 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2066 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2067 pf_cksum_fixup(pf_cksum_fixup(*c, 2068 ao.addr16[0], an->addr16[0], u), 2069 ao.addr16[1], an->addr16[1], u), 2070 ao.addr16[2], an->addr16[2], u), 2071 ao.addr16[3], an->addr16[3], u), 2072 ao.addr16[4], an->addr16[4], u), 2073 ao.addr16[5], an->addr16[5], u), 2074 ao.addr16[6], an->addr16[6], u), 2075 ao.addr16[7], an->addr16[7], u); 2076 } 2077 #endif /* INET6 */ 2078 2079 static void 2080 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2081 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2082 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2083 { 2084 struct pf_addr oia, ooa; 2085 2086 PF_ACPY(&oia, ia, af); 2087 if (oa) 2088 PF_ACPY(&ooa, oa, af); 2089 2090 /* Change inner protocol port, fix inner protocol checksum. */ 2091 if (ip != NULL) { 2092 u_int16_t oip = *ip; 2093 u_int32_t opc; 2094 2095 if (pc != NULL) 2096 opc = *pc; 2097 *ip = np; 2098 if (pc != NULL) 2099 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2100 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2101 if (pc != NULL) 2102 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2103 } 2104 /* Change inner ip address, fix inner ip and icmp checksums. */ 2105 PF_ACPY(ia, na, af); 2106 switch (af) { 2107 #ifdef INET 2108 case AF_INET: { 2109 u_int32_t oh2c = *h2c; 2110 2111 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2112 oia.addr16[0], ia->addr16[0], 0), 2113 oia.addr16[1], ia->addr16[1], 0); 2114 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2115 oia.addr16[0], ia->addr16[0], 0), 2116 oia.addr16[1], ia->addr16[1], 0); 2117 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2118 break; 2119 } 2120 #endif /* INET */ 2121 #ifdef INET6 2122 case AF_INET6: 2123 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2124 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2125 pf_cksum_fixup(pf_cksum_fixup(*ic, 2126 oia.addr16[0], ia->addr16[0], u), 2127 oia.addr16[1], ia->addr16[1], u), 2128 oia.addr16[2], ia->addr16[2], u), 2129 oia.addr16[3], ia->addr16[3], u), 2130 oia.addr16[4], ia->addr16[4], u), 2131 oia.addr16[5], ia->addr16[5], u), 2132 oia.addr16[6], ia->addr16[6], u), 2133 oia.addr16[7], ia->addr16[7], u); 2134 break; 2135 #endif /* INET6 */ 2136 } 2137 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2138 if (oa) { 2139 PF_ACPY(oa, na, af); 2140 switch (af) { 2141 #ifdef INET 2142 case AF_INET: 2143 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2144 ooa.addr16[0], oa->addr16[0], 0), 2145 ooa.addr16[1], oa->addr16[1], 0); 2146 break; 2147 #endif /* INET */ 2148 #ifdef INET6 2149 case AF_INET6: 2150 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2151 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2152 pf_cksum_fixup(pf_cksum_fixup(*ic, 2153 ooa.addr16[0], oa->addr16[0], u), 2154 ooa.addr16[1], oa->addr16[1], u), 2155 ooa.addr16[2], oa->addr16[2], u), 2156 ooa.addr16[3], oa->addr16[3], u), 2157 ooa.addr16[4], oa->addr16[4], u), 2158 ooa.addr16[5], oa->addr16[5], u), 2159 ooa.addr16[6], oa->addr16[6], u), 2160 ooa.addr16[7], oa->addr16[7], u); 2161 break; 2162 #endif /* INET6 */ 2163 } 2164 } 2165 } 2166 2167 2168 /* 2169 * Need to modulate the sequence numbers in the TCP SACK option 2170 * (credits to Krzysztof Pfaff for report and patch) 2171 */ 2172 static int 2173 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2174 struct tcphdr *th, struct pf_state_peer *dst) 2175 { 2176 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2177 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2178 int copyback = 0, i, olen; 2179 struct sackblk sack; 2180 2181 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2182 if (hlen < TCPOLEN_SACKLEN || 2183 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2184 return 0; 2185 2186 while (hlen >= TCPOLEN_SACKLEN) { 2187 olen = opt[1]; 2188 switch (*opt) { 2189 case TCPOPT_EOL: /* FALLTHROUGH */ 2190 case TCPOPT_NOP: 2191 opt++; 2192 hlen--; 2193 break; 2194 case TCPOPT_SACK: 2195 if (olen > hlen) 2196 olen = hlen; 2197 if (olen >= TCPOLEN_SACKLEN) { 2198 for (i = 2; i + TCPOLEN_SACK <= olen; 2199 i += TCPOLEN_SACK) { 2200 memcpy(&sack, &opt[i], sizeof(sack)); 2201 pf_change_a(&sack.start, &th->th_sum, 2202 htonl(ntohl(sack.start) - 2203 dst->seqdiff), 0); 2204 pf_change_a(&sack.end, &th->th_sum, 2205 htonl(ntohl(sack.end) - 2206 dst->seqdiff), 0); 2207 memcpy(&opt[i], &sack, sizeof(sack)); 2208 } 2209 copyback = 1; 2210 } 2211 /* FALLTHROUGH */ 2212 default: 2213 if (olen < 2) 2214 olen = 2; 2215 hlen -= olen; 2216 opt += olen; 2217 } 2218 } 2219 2220 if (copyback) 2221 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2222 return (copyback); 2223 } 2224 2225 static void 2226 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af, 2227 const struct pf_addr *saddr, const struct pf_addr *daddr, 2228 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2229 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2230 u_int16_t rtag, struct ifnet *ifp) 2231 { 2232 struct pf_send_entry *pfse; 2233 struct mbuf *m; 2234 int len, tlen; 2235 #ifdef INET 2236 struct ip *h = NULL; 2237 #endif /* INET */ 2238 #ifdef INET6 2239 struct ip6_hdr *h6 = NULL; 2240 #endif /* INET6 */ 2241 struct tcphdr *th; 2242 char *opt; 2243 struct pf_mtag *pf_mtag; 2244 2245 len = 0; 2246 th = NULL; 2247 2248 /* maximum segment size tcp option */ 2249 tlen = sizeof(struct tcphdr); 2250 if (mss) 2251 tlen += 4; 2252 2253 switch (af) { 2254 #ifdef INET 2255 case AF_INET: 2256 len = sizeof(struct ip) + tlen; 2257 break; 2258 #endif /* INET */ 2259 #ifdef INET6 2260 case AF_INET6: 2261 len = sizeof(struct ip6_hdr) + tlen; 2262 break; 2263 #endif /* INET6 */ 2264 default: 2265 panic("%s: unsupported af %d", __func__, af); 2266 } 2267 2268 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2269 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2270 if (pfse == NULL) 2271 return; 2272 m = m_gethdr(M_NOWAIT, MT_DATA); 2273 if (m == NULL) { 2274 free(pfse, M_PFTEMP); 2275 return; 2276 } 2277 #ifdef MAC 2278 mac_netinet_firewall_send(m); 2279 #endif 2280 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2281 free(pfse, M_PFTEMP); 2282 m_freem(m); 2283 return; 2284 } 2285 if (tag) 2286 m->m_flags |= M_SKIP_FIREWALL; 2287 pf_mtag->tag = rtag; 2288 2289 if (r != NULL && r->rtableid >= 0) 2290 M_SETFIB(m, r->rtableid); 2291 2292 #ifdef ALTQ 2293 if (r != NULL && r->qid) { 2294 pf_mtag->qid = r->qid; 2295 2296 /* add hints for ecn */ 2297 pf_mtag->hdr = mtod(m, struct ip *); 2298 } 2299 #endif /* ALTQ */ 2300 m->m_data += max_linkhdr; 2301 m->m_pkthdr.len = m->m_len = len; 2302 m->m_pkthdr.rcvif = NULL; 2303 bzero(m->m_data, len); 2304 switch (af) { 2305 #ifdef INET 2306 case AF_INET: 2307 h = mtod(m, struct ip *); 2308 2309 /* IP header fields included in the TCP checksum */ 2310 h->ip_p = IPPROTO_TCP; 2311 h->ip_len = htons(tlen); 2312 h->ip_src.s_addr = saddr->v4.s_addr; 2313 h->ip_dst.s_addr = daddr->v4.s_addr; 2314 2315 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2316 break; 2317 #endif /* INET */ 2318 #ifdef INET6 2319 case AF_INET6: 2320 h6 = mtod(m, struct ip6_hdr *); 2321 2322 /* IP header fields included in the TCP checksum */ 2323 h6->ip6_nxt = IPPROTO_TCP; 2324 h6->ip6_plen = htons(tlen); 2325 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2326 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2327 2328 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2329 break; 2330 #endif /* INET6 */ 2331 } 2332 2333 /* TCP header */ 2334 th->th_sport = sport; 2335 th->th_dport = dport; 2336 th->th_seq = htonl(seq); 2337 th->th_ack = htonl(ack); 2338 th->th_off = tlen >> 2; 2339 th->th_flags = flags; 2340 th->th_win = htons(win); 2341 2342 if (mss) { 2343 opt = (char *)(th + 1); 2344 opt[0] = TCPOPT_MAXSEG; 2345 opt[1] = 4; 2346 HTONS(mss); 2347 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2348 } 2349 2350 switch (af) { 2351 #ifdef INET 2352 case AF_INET: 2353 /* TCP checksum */ 2354 th->th_sum = in_cksum(m, len); 2355 2356 /* Finish the IP header */ 2357 h->ip_v = 4; 2358 h->ip_hl = sizeof(*h) >> 2; 2359 h->ip_tos = IPTOS_LOWDELAY; 2360 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 2361 h->ip_len = htons(len); 2362 h->ip_ttl = ttl ? ttl : V_ip_defttl; 2363 h->ip_sum = 0; 2364 2365 pfse->pfse_type = PFSE_IP; 2366 break; 2367 #endif /* INET */ 2368 #ifdef INET6 2369 case AF_INET6: 2370 /* TCP checksum */ 2371 th->th_sum = in6_cksum(m, IPPROTO_TCP, 2372 sizeof(struct ip6_hdr), tlen); 2373 2374 h6->ip6_vfc |= IPV6_VERSION; 2375 h6->ip6_hlim = IPV6_DEFHLIM; 2376 2377 pfse->pfse_type = PFSE_IP6; 2378 break; 2379 #endif /* INET6 */ 2380 } 2381 pfse->pfse_m = m; 2382 pf_send(pfse); 2383 } 2384 2385 static void 2386 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 2387 struct pf_rule *r) 2388 { 2389 struct pf_send_entry *pfse; 2390 struct mbuf *m0; 2391 struct pf_mtag *pf_mtag; 2392 2393 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2394 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2395 if (pfse == NULL) 2396 return; 2397 2398 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 2399 free(pfse, M_PFTEMP); 2400 return; 2401 } 2402 2403 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 2404 free(pfse, M_PFTEMP); 2405 return; 2406 } 2407 /* XXX: revisit */ 2408 m0->m_flags |= M_SKIP_FIREWALL; 2409 2410 if (r->rtableid >= 0) 2411 M_SETFIB(m0, r->rtableid); 2412 2413 #ifdef ALTQ 2414 if (r->qid) { 2415 pf_mtag->qid = r->qid; 2416 /* add hints for ecn */ 2417 pf_mtag->hdr = mtod(m0, struct ip *); 2418 } 2419 #endif /* ALTQ */ 2420 2421 switch (af) { 2422 #ifdef INET 2423 case AF_INET: 2424 pfse->pfse_type = PFSE_ICMP; 2425 break; 2426 #endif /* INET */ 2427 #ifdef INET6 2428 case AF_INET6: 2429 pfse->pfse_type = PFSE_ICMP6; 2430 break; 2431 #endif /* INET6 */ 2432 } 2433 pfse->pfse_m = m0; 2434 pfse->icmpopts.type = type; 2435 pfse->icmpopts.code = code; 2436 pf_send(pfse); 2437 } 2438 2439 /* 2440 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 2441 * If n is 0, they match if they are equal. If n is != 0, they match if they 2442 * are different. 2443 */ 2444 int 2445 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 2446 struct pf_addr *b, sa_family_t af) 2447 { 2448 int match = 0; 2449 2450 switch (af) { 2451 #ifdef INET 2452 case AF_INET: 2453 if ((a->addr32[0] & m->addr32[0]) == 2454 (b->addr32[0] & m->addr32[0])) 2455 match++; 2456 break; 2457 #endif /* INET */ 2458 #ifdef INET6 2459 case AF_INET6: 2460 if (((a->addr32[0] & m->addr32[0]) == 2461 (b->addr32[0] & m->addr32[0])) && 2462 ((a->addr32[1] & m->addr32[1]) == 2463 (b->addr32[1] & m->addr32[1])) && 2464 ((a->addr32[2] & m->addr32[2]) == 2465 (b->addr32[2] & m->addr32[2])) && 2466 ((a->addr32[3] & m->addr32[3]) == 2467 (b->addr32[3] & m->addr32[3]))) 2468 match++; 2469 break; 2470 #endif /* INET6 */ 2471 } 2472 if (match) { 2473 if (n) 2474 return (0); 2475 else 2476 return (1); 2477 } else { 2478 if (n) 2479 return (1); 2480 else 2481 return (0); 2482 } 2483 } 2484 2485 /* 2486 * Return 1 if b <= a <= e, otherwise return 0. 2487 */ 2488 int 2489 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 2490 struct pf_addr *a, sa_family_t af) 2491 { 2492 switch (af) { 2493 #ifdef INET 2494 case AF_INET: 2495 if ((a->addr32[0] < b->addr32[0]) || 2496 (a->addr32[0] > e->addr32[0])) 2497 return (0); 2498 break; 2499 #endif /* INET */ 2500 #ifdef INET6 2501 case AF_INET6: { 2502 int i; 2503 2504 /* check a >= b */ 2505 for (i = 0; i < 4; ++i) 2506 if (a->addr32[i] > b->addr32[i]) 2507 break; 2508 else if (a->addr32[i] < b->addr32[i]) 2509 return (0); 2510 /* check a <= e */ 2511 for (i = 0; i < 4; ++i) 2512 if (a->addr32[i] < e->addr32[i]) 2513 break; 2514 else if (a->addr32[i] > e->addr32[i]) 2515 return (0); 2516 break; 2517 } 2518 #endif /* INET6 */ 2519 } 2520 return (1); 2521 } 2522 2523 static int 2524 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 2525 { 2526 switch (op) { 2527 case PF_OP_IRG: 2528 return ((p > a1) && (p < a2)); 2529 case PF_OP_XRG: 2530 return ((p < a1) || (p > a2)); 2531 case PF_OP_RRG: 2532 return ((p >= a1) && (p <= a2)); 2533 case PF_OP_EQ: 2534 return (p == a1); 2535 case PF_OP_NE: 2536 return (p != a1); 2537 case PF_OP_LT: 2538 return (p < a1); 2539 case PF_OP_LE: 2540 return (p <= a1); 2541 case PF_OP_GT: 2542 return (p > a1); 2543 case PF_OP_GE: 2544 return (p >= a1); 2545 } 2546 return (0); /* never reached */ 2547 } 2548 2549 int 2550 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 2551 { 2552 NTOHS(a1); 2553 NTOHS(a2); 2554 NTOHS(p); 2555 return (pf_match(op, a1, a2, p)); 2556 } 2557 2558 static int 2559 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 2560 { 2561 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2562 return (0); 2563 return (pf_match(op, a1, a2, u)); 2564 } 2565 2566 static int 2567 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 2568 { 2569 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2570 return (0); 2571 return (pf_match(op, a1, a2, g)); 2572 } 2573 2574 int 2575 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag) 2576 { 2577 if (*tag == -1) 2578 *tag = mtag; 2579 2580 return ((!r->match_tag_not && r->match_tag == *tag) || 2581 (r->match_tag_not && r->match_tag != *tag)); 2582 } 2583 2584 int 2585 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 2586 { 2587 2588 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 2589 2590 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 2591 return (ENOMEM); 2592 2593 pd->pf_mtag->tag = tag; 2594 2595 return (0); 2596 } 2597 2598 #define PF_ANCHOR_STACKSIZE 32 2599 struct pf_anchor_stackframe { 2600 struct pf_ruleset *rs; 2601 struct pf_rule *r; /* XXX: + match bit */ 2602 struct pf_anchor *child; 2603 }; 2604 2605 /* 2606 * XXX: We rely on malloc(9) returning pointer aligned addresses. 2607 */ 2608 #define PF_ANCHORSTACK_MATCH 0x00000001 2609 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 2610 2611 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 2612 #define PF_ANCHOR_RULE(f) (struct pf_rule *) \ 2613 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 2614 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 2615 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 2616 } while (0) 2617 2618 void 2619 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth, 2620 struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a, 2621 int *match) 2622 { 2623 struct pf_anchor_stackframe *f; 2624 2625 PF_RULES_RASSERT(); 2626 2627 if (match) 2628 *match = 0; 2629 if (*depth >= PF_ANCHOR_STACKSIZE) { 2630 printf("%s: anchor stack overflow on %s\n", 2631 __func__, (*r)->anchor->name); 2632 *r = TAILQ_NEXT(*r, entries); 2633 return; 2634 } else if (*depth == 0 && a != NULL) 2635 *a = *r; 2636 f = stack + (*depth)++; 2637 f->rs = *rs; 2638 f->r = *r; 2639 if ((*r)->anchor_wildcard) { 2640 struct pf_anchor_node *parent = &(*r)->anchor->children; 2641 2642 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) { 2643 *r = NULL; 2644 return; 2645 } 2646 *rs = &f->child->ruleset; 2647 } else { 2648 f->child = NULL; 2649 *rs = &(*r)->anchor->ruleset; 2650 } 2651 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2652 } 2653 2654 int 2655 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth, 2656 struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a, 2657 int *match) 2658 { 2659 struct pf_anchor_stackframe *f; 2660 struct pf_rule *fr; 2661 int quick = 0; 2662 2663 PF_RULES_RASSERT(); 2664 2665 do { 2666 if (*depth <= 0) 2667 break; 2668 f = stack + *depth - 1; 2669 fr = PF_ANCHOR_RULE(f); 2670 if (f->child != NULL) { 2671 struct pf_anchor_node *parent; 2672 2673 /* 2674 * This block traverses through 2675 * a wildcard anchor. 2676 */ 2677 parent = &fr->anchor->children; 2678 if (match != NULL && *match) { 2679 /* 2680 * If any of "*" matched, then 2681 * "foo/ *" matched, mark frame 2682 * appropriately. 2683 */ 2684 PF_ANCHOR_SET_MATCH(f); 2685 *match = 0; 2686 } 2687 f->child = RB_NEXT(pf_anchor_node, parent, f->child); 2688 if (f->child != NULL) { 2689 *rs = &f->child->ruleset; 2690 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2691 if (*r == NULL) 2692 continue; 2693 else 2694 break; 2695 } 2696 } 2697 (*depth)--; 2698 if (*depth == 0 && a != NULL) 2699 *a = NULL; 2700 *rs = f->rs; 2701 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 2702 quick = fr->quick; 2703 *r = TAILQ_NEXT(fr, entries); 2704 } while (*r == NULL); 2705 2706 return (quick); 2707 } 2708 2709 #ifdef INET6 2710 void 2711 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 2712 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 2713 { 2714 switch (af) { 2715 #ifdef INET 2716 case AF_INET: 2717 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 2718 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 2719 break; 2720 #endif /* INET */ 2721 case AF_INET6: 2722 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 2723 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 2724 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 2725 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 2726 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 2727 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 2728 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 2729 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 2730 break; 2731 } 2732 } 2733 2734 void 2735 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 2736 { 2737 switch (af) { 2738 #ifdef INET 2739 case AF_INET: 2740 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 2741 break; 2742 #endif /* INET */ 2743 case AF_INET6: 2744 if (addr->addr32[3] == 0xffffffff) { 2745 addr->addr32[3] = 0; 2746 if (addr->addr32[2] == 0xffffffff) { 2747 addr->addr32[2] = 0; 2748 if (addr->addr32[1] == 0xffffffff) { 2749 addr->addr32[1] = 0; 2750 addr->addr32[0] = 2751 htonl(ntohl(addr->addr32[0]) + 1); 2752 } else 2753 addr->addr32[1] = 2754 htonl(ntohl(addr->addr32[1]) + 1); 2755 } else 2756 addr->addr32[2] = 2757 htonl(ntohl(addr->addr32[2]) + 1); 2758 } else 2759 addr->addr32[3] = 2760 htonl(ntohl(addr->addr32[3]) + 1); 2761 break; 2762 } 2763 } 2764 #endif /* INET6 */ 2765 2766 int 2767 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m) 2768 { 2769 struct pf_addr *saddr, *daddr; 2770 u_int16_t sport, dport; 2771 struct inpcbinfo *pi; 2772 struct inpcb *inp; 2773 2774 pd->lookup.uid = UID_MAX; 2775 pd->lookup.gid = GID_MAX; 2776 2777 switch (pd->proto) { 2778 case IPPROTO_TCP: 2779 if (pd->hdr.tcp == NULL) 2780 return (-1); 2781 sport = pd->hdr.tcp->th_sport; 2782 dport = pd->hdr.tcp->th_dport; 2783 pi = &V_tcbinfo; 2784 break; 2785 case IPPROTO_UDP: 2786 if (pd->hdr.udp == NULL) 2787 return (-1); 2788 sport = pd->hdr.udp->uh_sport; 2789 dport = pd->hdr.udp->uh_dport; 2790 pi = &V_udbinfo; 2791 break; 2792 default: 2793 return (-1); 2794 } 2795 if (direction == PF_IN) { 2796 saddr = pd->src; 2797 daddr = pd->dst; 2798 } else { 2799 u_int16_t p; 2800 2801 p = sport; 2802 sport = dport; 2803 dport = p; 2804 saddr = pd->dst; 2805 daddr = pd->src; 2806 } 2807 switch (pd->af) { 2808 #ifdef INET 2809 case AF_INET: 2810 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 2811 dport, INPLOOKUP_RLOCKPCB, NULL, m); 2812 if (inp == NULL) { 2813 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 2814 daddr->v4, dport, INPLOOKUP_WILDCARD | 2815 INPLOOKUP_RLOCKPCB, NULL, m); 2816 if (inp == NULL) 2817 return (-1); 2818 } 2819 break; 2820 #endif /* INET */ 2821 #ifdef INET6 2822 case AF_INET6: 2823 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 2824 dport, INPLOOKUP_RLOCKPCB, NULL, m); 2825 if (inp == NULL) { 2826 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 2827 &daddr->v6, dport, INPLOOKUP_WILDCARD | 2828 INPLOOKUP_RLOCKPCB, NULL, m); 2829 if (inp == NULL) 2830 return (-1); 2831 } 2832 break; 2833 #endif /* INET6 */ 2834 2835 default: 2836 return (-1); 2837 } 2838 INP_RLOCK_ASSERT(inp); 2839 pd->lookup.uid = inp->inp_cred->cr_uid; 2840 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 2841 INP_RUNLOCK(inp); 2842 2843 return (1); 2844 } 2845 2846 static u_int8_t 2847 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 2848 { 2849 int hlen; 2850 u_int8_t hdr[60]; 2851 u_int8_t *opt, optlen; 2852 u_int8_t wscale = 0; 2853 2854 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 2855 if (hlen <= sizeof(struct tcphdr)) 2856 return (0); 2857 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 2858 return (0); 2859 opt = hdr + sizeof(struct tcphdr); 2860 hlen -= sizeof(struct tcphdr); 2861 while (hlen >= 3) { 2862 switch (*opt) { 2863 case TCPOPT_EOL: 2864 case TCPOPT_NOP: 2865 ++opt; 2866 --hlen; 2867 break; 2868 case TCPOPT_WINDOW: 2869 wscale = opt[2]; 2870 if (wscale > TCP_MAX_WINSHIFT) 2871 wscale = TCP_MAX_WINSHIFT; 2872 wscale |= PF_WSCALE_FLAG; 2873 /* FALLTHROUGH */ 2874 default: 2875 optlen = opt[1]; 2876 if (optlen < 2) 2877 optlen = 2; 2878 hlen -= optlen; 2879 opt += optlen; 2880 break; 2881 } 2882 } 2883 return (wscale); 2884 } 2885 2886 static u_int16_t 2887 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 2888 { 2889 int hlen; 2890 u_int8_t hdr[60]; 2891 u_int8_t *opt, optlen; 2892 u_int16_t mss = V_tcp_mssdflt; 2893 2894 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 2895 if (hlen <= sizeof(struct tcphdr)) 2896 return (0); 2897 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 2898 return (0); 2899 opt = hdr + sizeof(struct tcphdr); 2900 hlen -= sizeof(struct tcphdr); 2901 while (hlen >= TCPOLEN_MAXSEG) { 2902 switch (*opt) { 2903 case TCPOPT_EOL: 2904 case TCPOPT_NOP: 2905 ++opt; 2906 --hlen; 2907 break; 2908 case TCPOPT_MAXSEG: 2909 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 2910 NTOHS(mss); 2911 /* FALLTHROUGH */ 2912 default: 2913 optlen = opt[1]; 2914 if (optlen < 2) 2915 optlen = 2; 2916 hlen -= optlen; 2917 opt += optlen; 2918 break; 2919 } 2920 } 2921 return (mss); 2922 } 2923 2924 static u_int16_t 2925 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 2926 { 2927 #ifdef INET 2928 struct sockaddr_in *dst; 2929 struct route ro; 2930 #endif /* INET */ 2931 #ifdef INET6 2932 struct sockaddr_in6 *dst6; 2933 struct route_in6 ro6; 2934 #endif /* INET6 */ 2935 struct rtentry *rt = NULL; 2936 int hlen = 0; 2937 u_int16_t mss = V_tcp_mssdflt; 2938 2939 switch (af) { 2940 #ifdef INET 2941 case AF_INET: 2942 hlen = sizeof(struct ip); 2943 bzero(&ro, sizeof(ro)); 2944 dst = (struct sockaddr_in *)&ro.ro_dst; 2945 dst->sin_family = AF_INET; 2946 dst->sin_len = sizeof(*dst); 2947 dst->sin_addr = addr->v4; 2948 in_rtalloc_ign(&ro, 0, rtableid); 2949 rt = ro.ro_rt; 2950 break; 2951 #endif /* INET */ 2952 #ifdef INET6 2953 case AF_INET6: 2954 hlen = sizeof(struct ip6_hdr); 2955 bzero(&ro6, sizeof(ro6)); 2956 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst; 2957 dst6->sin6_family = AF_INET6; 2958 dst6->sin6_len = sizeof(*dst6); 2959 dst6->sin6_addr = addr->v6; 2960 in6_rtalloc_ign(&ro6, 0, rtableid); 2961 rt = ro6.ro_rt; 2962 break; 2963 #endif /* INET6 */ 2964 } 2965 2966 if (rt && rt->rt_ifp) { 2967 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr); 2968 mss = max(V_tcp_mssdflt, mss); 2969 RTFREE(rt); 2970 } 2971 mss = min(mss, offer); 2972 mss = max(mss, 64); /* sanity - at least max opt space */ 2973 return (mss); 2974 } 2975 2976 static u_int32_t 2977 pf_tcp_iss(struct pf_pdesc *pd) 2978 { 2979 MD5_CTX ctx; 2980 u_int32_t digest[4]; 2981 2982 if (V_pf_tcp_secret_init == 0) { 2983 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 2984 MD5Init(&V_pf_tcp_secret_ctx); 2985 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 2986 sizeof(V_pf_tcp_secret)); 2987 V_pf_tcp_secret_init = 1; 2988 } 2989 2990 ctx = V_pf_tcp_secret_ctx; 2991 2992 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short)); 2993 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short)); 2994 if (pd->af == AF_INET6) { 2995 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 2996 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 2997 } else { 2998 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 2999 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 3000 } 3001 MD5Final((u_char *)digest, &ctx); 3002 V_pf_tcp_iss_off += 4096; 3003 #define ISN_RANDOM_INCREMENT (4096 - 1) 3004 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 3005 V_pf_tcp_iss_off); 3006 #undef ISN_RANDOM_INCREMENT 3007 } 3008 3009 static int 3010 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction, 3011 struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, 3012 struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp) 3013 { 3014 struct pf_rule *nr = NULL; 3015 struct pf_addr * const saddr = pd->src; 3016 struct pf_addr * const daddr = pd->dst; 3017 sa_family_t af = pd->af; 3018 struct pf_rule *r, *a = NULL; 3019 struct pf_ruleset *ruleset = NULL; 3020 struct pf_src_node *nsn = NULL; 3021 struct tcphdr *th = pd->hdr.tcp; 3022 struct pf_state_key *sk = NULL, *nk = NULL; 3023 u_short reason; 3024 int rewrite = 0, hdrlen = 0; 3025 int tag = -1, rtableid = -1; 3026 int asd = 0; 3027 int match = 0; 3028 int state_icmp = 0; 3029 u_int16_t sport = 0, dport = 0; 3030 u_int16_t bproto_sum = 0, bip_sum = 0; 3031 u_int8_t icmptype = 0, icmpcode = 0; 3032 struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3033 3034 PF_RULES_RASSERT(); 3035 3036 if (inp != NULL) { 3037 INP_LOCK_ASSERT(inp); 3038 pd->lookup.uid = inp->inp_cred->cr_uid; 3039 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3040 pd->lookup.done = 1; 3041 } 3042 3043 switch (pd->proto) { 3044 case IPPROTO_TCP: 3045 sport = th->th_sport; 3046 dport = th->th_dport; 3047 hdrlen = sizeof(*th); 3048 break; 3049 case IPPROTO_UDP: 3050 sport = pd->hdr.udp->uh_sport; 3051 dport = pd->hdr.udp->uh_dport; 3052 hdrlen = sizeof(*pd->hdr.udp); 3053 break; 3054 #ifdef INET 3055 case IPPROTO_ICMP: 3056 if (pd->af != AF_INET) 3057 break; 3058 sport = dport = pd->hdr.icmp->icmp_id; 3059 hdrlen = sizeof(*pd->hdr.icmp); 3060 icmptype = pd->hdr.icmp->icmp_type; 3061 icmpcode = pd->hdr.icmp->icmp_code; 3062 3063 if (icmptype == ICMP_UNREACH || 3064 icmptype == ICMP_SOURCEQUENCH || 3065 icmptype == ICMP_REDIRECT || 3066 icmptype == ICMP_TIMXCEED || 3067 icmptype == ICMP_PARAMPROB) 3068 state_icmp++; 3069 break; 3070 #endif /* INET */ 3071 #ifdef INET6 3072 case IPPROTO_ICMPV6: 3073 if (af != AF_INET6) 3074 break; 3075 sport = dport = pd->hdr.icmp6->icmp6_id; 3076 hdrlen = sizeof(*pd->hdr.icmp6); 3077 icmptype = pd->hdr.icmp6->icmp6_type; 3078 icmpcode = pd->hdr.icmp6->icmp6_code; 3079 3080 if (icmptype == ICMP6_DST_UNREACH || 3081 icmptype == ICMP6_PACKET_TOO_BIG || 3082 icmptype == ICMP6_TIME_EXCEEDED || 3083 icmptype == ICMP6_PARAM_PROB) 3084 state_icmp++; 3085 break; 3086 #endif /* INET6 */ 3087 default: 3088 sport = dport = hdrlen = 0; 3089 break; 3090 } 3091 3092 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3093 3094 /* check packet for BINAT/NAT/RDR */ 3095 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk, 3096 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 3097 KASSERT(sk != NULL, ("%s: null sk", __func__)); 3098 KASSERT(nk != NULL, ("%s: null nk", __func__)); 3099 3100 if (pd->ip_sum) 3101 bip_sum = *pd->ip_sum; 3102 3103 switch (pd->proto) { 3104 case IPPROTO_TCP: 3105 bproto_sum = th->th_sum; 3106 pd->proto_sum = &th->th_sum; 3107 3108 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3109 nk->port[pd->sidx] != sport) { 3110 pf_change_ap(saddr, &th->th_sport, pd->ip_sum, 3111 &th->th_sum, &nk->addr[pd->sidx], 3112 nk->port[pd->sidx], 0, af); 3113 pd->sport = &th->th_sport; 3114 sport = th->th_sport; 3115 } 3116 3117 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3118 nk->port[pd->didx] != dport) { 3119 pf_change_ap(daddr, &th->th_dport, pd->ip_sum, 3120 &th->th_sum, &nk->addr[pd->didx], 3121 nk->port[pd->didx], 0, af); 3122 dport = th->th_dport; 3123 pd->dport = &th->th_dport; 3124 } 3125 rewrite++; 3126 break; 3127 case IPPROTO_UDP: 3128 bproto_sum = pd->hdr.udp->uh_sum; 3129 pd->proto_sum = &pd->hdr.udp->uh_sum; 3130 3131 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3132 nk->port[pd->sidx] != sport) { 3133 pf_change_ap(saddr, &pd->hdr.udp->uh_sport, 3134 pd->ip_sum, &pd->hdr.udp->uh_sum, 3135 &nk->addr[pd->sidx], 3136 nk->port[pd->sidx], 1, af); 3137 sport = pd->hdr.udp->uh_sport; 3138 pd->sport = &pd->hdr.udp->uh_sport; 3139 } 3140 3141 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3142 nk->port[pd->didx] != dport) { 3143 pf_change_ap(daddr, &pd->hdr.udp->uh_dport, 3144 pd->ip_sum, &pd->hdr.udp->uh_sum, 3145 &nk->addr[pd->didx], 3146 nk->port[pd->didx], 1, af); 3147 dport = pd->hdr.udp->uh_dport; 3148 pd->dport = &pd->hdr.udp->uh_dport; 3149 } 3150 rewrite++; 3151 break; 3152 #ifdef INET 3153 case IPPROTO_ICMP: 3154 nk->port[0] = nk->port[1]; 3155 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 3156 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 3157 nk->addr[pd->sidx].v4.s_addr, 0); 3158 3159 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 3160 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 3161 nk->addr[pd->didx].v4.s_addr, 0); 3162 3163 if (nk->port[1] != pd->hdr.icmp->icmp_id) { 3164 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup( 3165 pd->hdr.icmp->icmp_cksum, sport, 3166 nk->port[1], 0); 3167 pd->hdr.icmp->icmp_id = nk->port[1]; 3168 pd->sport = &pd->hdr.icmp->icmp_id; 3169 } 3170 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 3171 break; 3172 #endif /* INET */ 3173 #ifdef INET6 3174 case IPPROTO_ICMPV6: 3175 nk->port[0] = nk->port[1]; 3176 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 3177 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum, 3178 &nk->addr[pd->sidx], 0); 3179 3180 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 3181 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum, 3182 &nk->addr[pd->didx], 0); 3183 rewrite++; 3184 break; 3185 #endif /* INET */ 3186 default: 3187 switch (af) { 3188 #ifdef INET 3189 case AF_INET: 3190 if (PF_ANEQ(saddr, 3191 &nk->addr[pd->sidx], AF_INET)) 3192 pf_change_a(&saddr->v4.s_addr, 3193 pd->ip_sum, 3194 nk->addr[pd->sidx].v4.s_addr, 0); 3195 3196 if (PF_ANEQ(daddr, 3197 &nk->addr[pd->didx], AF_INET)) 3198 pf_change_a(&daddr->v4.s_addr, 3199 pd->ip_sum, 3200 nk->addr[pd->didx].v4.s_addr, 0); 3201 break; 3202 #endif /* INET */ 3203 #ifdef INET6 3204 case AF_INET6: 3205 if (PF_ANEQ(saddr, 3206 &nk->addr[pd->sidx], AF_INET6)) 3207 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 3208 3209 if (PF_ANEQ(daddr, 3210 &nk->addr[pd->didx], AF_INET6)) 3211 PF_ACPY(saddr, &nk->addr[pd->didx], af); 3212 break; 3213 #endif /* INET */ 3214 } 3215 break; 3216 } 3217 if (nr->natpass) 3218 r = NULL; 3219 pd->nat_rule = nr; 3220 } 3221 3222 while (r != NULL) { 3223 r->evaluations++; 3224 if (pfi_kif_match(r->kif, kif) == r->ifnot) 3225 r = r->skip[PF_SKIP_IFP].ptr; 3226 else if (r->direction && r->direction != direction) 3227 r = r->skip[PF_SKIP_DIR].ptr; 3228 else if (r->af && r->af != af) 3229 r = r->skip[PF_SKIP_AF].ptr; 3230 else if (r->proto && r->proto != pd->proto) 3231 r = r->skip[PF_SKIP_PROTO].ptr; 3232 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 3233 r->src.neg, kif, M_GETFIB(m))) 3234 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3235 /* tcp/udp only. port_op always 0 in other cases */ 3236 else if (r->src.port_op && !pf_match_port(r->src.port_op, 3237 r->src.port[0], r->src.port[1], sport)) 3238 r = r->skip[PF_SKIP_SRC_PORT].ptr; 3239 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 3240 r->dst.neg, NULL, M_GETFIB(m))) 3241 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3242 /* tcp/udp only. port_op always 0 in other cases */ 3243 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 3244 r->dst.port[0], r->dst.port[1], dport)) 3245 r = r->skip[PF_SKIP_DST_PORT].ptr; 3246 /* icmp only. type always 0 in other cases */ 3247 else if (r->type && r->type != icmptype + 1) 3248 r = TAILQ_NEXT(r, entries); 3249 /* icmp only. type always 0 in other cases */ 3250 else if (r->code && r->code != icmpcode + 1) 3251 r = TAILQ_NEXT(r, entries); 3252 else if (r->tos && !(r->tos == pd->tos)) 3253 r = TAILQ_NEXT(r, entries); 3254 else if (r->rule_flag & PFRULE_FRAGMENT) 3255 r = TAILQ_NEXT(r, entries); 3256 else if (pd->proto == IPPROTO_TCP && 3257 (r->flagset & th->th_flags) != r->flags) 3258 r = TAILQ_NEXT(r, entries); 3259 /* tcp/udp only. uid.op always 0 in other cases */ 3260 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 3261 pf_socket_lookup(direction, pd, m), 1)) && 3262 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 3263 pd->lookup.uid)) 3264 r = TAILQ_NEXT(r, entries); 3265 /* tcp/udp only. gid.op always 0 in other cases */ 3266 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 3267 pf_socket_lookup(direction, pd, m), 1)) && 3268 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 3269 pd->lookup.gid)) 3270 r = TAILQ_NEXT(r, entries); 3271 else if (r->prob && 3272 r->prob <= arc4random()) 3273 r = TAILQ_NEXT(r, entries); 3274 else if (r->match_tag && !pf_match_tag(m, r, &tag, 3275 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 3276 r = TAILQ_NEXT(r, entries); 3277 else if (r->os_fingerprint != PF_OSFP_ANY && 3278 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 3279 pf_osfp_fingerprint(pd, m, off, th), 3280 r->os_fingerprint))) 3281 r = TAILQ_NEXT(r, entries); 3282 else { 3283 if (r->tag) 3284 tag = r->tag; 3285 if (r->rtableid >= 0) 3286 rtableid = r->rtableid; 3287 if (r->anchor == NULL) { 3288 match = 1; 3289 *rm = r; 3290 *am = a; 3291 *rsm = ruleset; 3292 if ((*rm)->quick) 3293 break; 3294 r = TAILQ_NEXT(r, entries); 3295 } else 3296 pf_step_into_anchor(anchor_stack, &asd, 3297 &ruleset, PF_RULESET_FILTER, &r, &a, 3298 &match); 3299 } 3300 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 3301 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 3302 break; 3303 } 3304 r = *rm; 3305 a = *am; 3306 ruleset = *rsm; 3307 3308 REASON_SET(&reason, PFRES_MATCH); 3309 3310 if (r->log || (nr != NULL && nr->log)) { 3311 if (rewrite) 3312 m_copyback(m, off, hdrlen, pd->hdr.any); 3313 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a, 3314 ruleset, pd, 1); 3315 } 3316 3317 if ((r->action == PF_DROP) && 3318 ((r->rule_flag & PFRULE_RETURNRST) || 3319 (r->rule_flag & PFRULE_RETURNICMP) || 3320 (r->rule_flag & PFRULE_RETURN))) { 3321 /* undo NAT changes, if they have taken place */ 3322 if (nr != NULL) { 3323 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3324 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3325 if (pd->sport) 3326 *pd->sport = sk->port[pd->sidx]; 3327 if (pd->dport) 3328 *pd->dport = sk->port[pd->didx]; 3329 if (pd->proto_sum) 3330 *pd->proto_sum = bproto_sum; 3331 if (pd->ip_sum) 3332 *pd->ip_sum = bip_sum; 3333 m_copyback(m, off, hdrlen, pd->hdr.any); 3334 } 3335 if (pd->proto == IPPROTO_TCP && 3336 ((r->rule_flag & PFRULE_RETURNRST) || 3337 (r->rule_flag & PFRULE_RETURN)) && 3338 !(th->th_flags & TH_RST)) { 3339 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3340 int len = 0; 3341 #ifdef INET 3342 struct ip *h4; 3343 #endif 3344 #ifdef INET6 3345 struct ip6_hdr *h6; 3346 #endif 3347 3348 switch (af) { 3349 #ifdef INET 3350 case AF_INET: 3351 h4 = mtod(m, struct ip *); 3352 len = ntohs(h4->ip_len) - off; 3353 break; 3354 #endif 3355 #ifdef INET6 3356 case AF_INET6: 3357 h6 = mtod(m, struct ip6_hdr *); 3358 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3359 break; 3360 #endif 3361 } 3362 3363 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3364 REASON_SET(&reason, PFRES_PROTCKSUM); 3365 else { 3366 if (th->th_flags & TH_SYN) 3367 ack++; 3368 if (th->th_flags & TH_FIN) 3369 ack++; 3370 pf_send_tcp(m, r, af, pd->dst, 3371 pd->src, th->th_dport, th->th_sport, 3372 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3373 r->return_ttl, 1, 0, kif->pfik_ifp); 3374 } 3375 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3376 r->return_icmp) 3377 pf_send_icmp(m, r->return_icmp >> 8, 3378 r->return_icmp & 255, af, r); 3379 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3380 r->return_icmp6) 3381 pf_send_icmp(m, r->return_icmp6 >> 8, 3382 r->return_icmp6 & 255, af, r); 3383 } 3384 3385 if (r->action == PF_DROP) 3386 goto cleanup; 3387 3388 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 3389 REASON_SET(&reason, PFRES_MEMORY); 3390 goto cleanup; 3391 } 3392 if (rtableid >= 0) 3393 M_SETFIB(m, rtableid); 3394 3395 if (!state_icmp && (r->keep_state || nr != NULL || 3396 (pd->flags & PFDESC_TCP_NORM))) { 3397 int action; 3398 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 3399 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 3400 hdrlen); 3401 if (action != PF_PASS) 3402 return (action); 3403 } else { 3404 if (sk != NULL) 3405 uma_zfree(V_pf_state_key_z, sk); 3406 if (nk != NULL) 3407 uma_zfree(V_pf_state_key_z, nk); 3408 } 3409 3410 /* copy back packet headers if we performed NAT operations */ 3411 if (rewrite) 3412 m_copyback(m, off, hdrlen, pd->hdr.any); 3413 3414 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 3415 direction == PF_OUT && 3416 pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m)) 3417 /* 3418 * We want the state created, but we dont 3419 * want to send this in case a partner 3420 * firewall has to know about it to allow 3421 * replies through it. 3422 */ 3423 return (PF_DEFER); 3424 3425 return (PF_PASS); 3426 3427 cleanup: 3428 if (sk != NULL) 3429 uma_zfree(V_pf_state_key_z, sk); 3430 if (nk != NULL) 3431 uma_zfree(V_pf_state_key_z, nk); 3432 return (PF_DROP); 3433 } 3434 3435 static int 3436 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a, 3437 struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk, 3438 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 3439 u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm, 3440 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen) 3441 { 3442 struct pf_state *s = NULL; 3443 struct pf_src_node *sn = NULL; 3444 struct tcphdr *th = pd->hdr.tcp; 3445 u_int16_t mss = V_tcp_mssdflt; 3446 u_short reason; 3447 3448 /* check maximums */ 3449 if (r->max_states && 3450 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 3451 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 3452 REASON_SET(&reason, PFRES_MAXSTATES); 3453 return (PF_DROP); 3454 } 3455 /* src node for filter rule */ 3456 if ((r->rule_flag & PFRULE_SRCTRACK || 3457 r->rpool.opts & PF_POOL_STICKYADDR) && 3458 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { 3459 REASON_SET(&reason, PFRES_SRCLIMIT); 3460 goto csfailed; 3461 } 3462 /* src node for translation rule */ 3463 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 3464 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { 3465 REASON_SET(&reason, PFRES_SRCLIMIT); 3466 goto csfailed; 3467 } 3468 s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO); 3469 if (s == NULL) { 3470 REASON_SET(&reason, PFRES_MEMORY); 3471 goto csfailed; 3472 } 3473 s->rule.ptr = r; 3474 s->nat_rule.ptr = nr; 3475 s->anchor.ptr = a; 3476 STATE_INC_COUNTERS(s); 3477 if (r->allow_opts) 3478 s->state_flags |= PFSTATE_ALLOWOPTS; 3479 if (r->rule_flag & PFRULE_STATESLOPPY) 3480 s->state_flags |= PFSTATE_SLOPPY; 3481 s->log = r->log & PF_LOG_ALL; 3482 s->sync_state = PFSYNC_S_NONE; 3483 if (nr != NULL) 3484 s->log |= nr->log & PF_LOG_ALL; 3485 switch (pd->proto) { 3486 case IPPROTO_TCP: 3487 s->src.seqlo = ntohl(th->th_seq); 3488 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 3489 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 3490 r->keep_state == PF_STATE_MODULATE) { 3491 /* Generate sequence number modulator */ 3492 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 3493 0) 3494 s->src.seqdiff = 1; 3495 pf_change_a(&th->th_seq, &th->th_sum, 3496 htonl(s->src.seqlo + s->src.seqdiff), 0); 3497 *rewrite = 1; 3498 } else 3499 s->src.seqdiff = 0; 3500 if (th->th_flags & TH_SYN) { 3501 s->src.seqhi++; 3502 s->src.wscale = pf_get_wscale(m, off, 3503 th->th_off, pd->af); 3504 } 3505 s->src.max_win = MAX(ntohs(th->th_win), 1); 3506 if (s->src.wscale & PF_WSCALE_MASK) { 3507 /* Remove scale factor from initial window */ 3508 int win = s->src.max_win; 3509 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 3510 s->src.max_win = (win - 1) >> 3511 (s->src.wscale & PF_WSCALE_MASK); 3512 } 3513 if (th->th_flags & TH_FIN) 3514 s->src.seqhi++; 3515 s->dst.seqhi = 1; 3516 s->dst.max_win = 1; 3517 s->src.state = TCPS_SYN_SENT; 3518 s->dst.state = TCPS_CLOSED; 3519 s->timeout = PFTM_TCP_FIRST_PACKET; 3520 break; 3521 case IPPROTO_UDP: 3522 s->src.state = PFUDPS_SINGLE; 3523 s->dst.state = PFUDPS_NO_TRAFFIC; 3524 s->timeout = PFTM_UDP_FIRST_PACKET; 3525 break; 3526 case IPPROTO_ICMP: 3527 #ifdef INET6 3528 case IPPROTO_ICMPV6: 3529 #endif 3530 s->timeout = PFTM_ICMP_FIRST_PACKET; 3531 break; 3532 default: 3533 s->src.state = PFOTHERS_SINGLE; 3534 s->dst.state = PFOTHERS_NO_TRAFFIC; 3535 s->timeout = PFTM_OTHER_FIRST_PACKET; 3536 } 3537 3538 if (r->rt && r->rt != PF_FASTROUTE) { 3539 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) { 3540 REASON_SET(&reason, PFRES_MAPFAILED); 3541 pf_src_tree_remove_state(s); 3542 STATE_DEC_COUNTERS(s); 3543 uma_zfree(V_pf_state_z, s); 3544 goto csfailed; 3545 } 3546 s->rt_kif = r->rpool.cur->kif; 3547 } 3548 3549 s->creation = time_uptime; 3550 s->expire = time_uptime; 3551 3552 if (sn != NULL) 3553 s->src_node = sn; 3554 if (nsn != NULL) { 3555 /* XXX We only modify one side for now. */ 3556 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 3557 s->nat_src_node = nsn; 3558 } 3559 if (pd->proto == IPPROTO_TCP) { 3560 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m, 3561 off, pd, th, &s->src, &s->dst)) { 3562 REASON_SET(&reason, PFRES_MEMORY); 3563 pf_src_tree_remove_state(s); 3564 STATE_DEC_COUNTERS(s); 3565 uma_zfree(V_pf_state_z, s); 3566 return (PF_DROP); 3567 } 3568 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub && 3569 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 3570 &s->src, &s->dst, rewrite)) { 3571 /* This really shouldn't happen!!! */ 3572 DPFPRINTF(PF_DEBUG_URGENT, 3573 ("pf_normalize_tcp_stateful failed on first pkt")); 3574 pf_normalize_tcp_cleanup(s); 3575 pf_src_tree_remove_state(s); 3576 STATE_DEC_COUNTERS(s); 3577 uma_zfree(V_pf_state_z, s); 3578 return (PF_DROP); 3579 } 3580 } 3581 s->direction = pd->dir; 3582 3583 /* 3584 * sk/nk could already been setup by pf_get_translation(). 3585 */ 3586 if (nr == NULL) { 3587 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 3588 __func__, nr, sk, nk)); 3589 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 3590 if (sk == NULL) 3591 goto csfailed; 3592 nk = sk; 3593 } else 3594 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 3595 __func__, nr, sk, nk)); 3596 3597 /* Swap sk/nk for PF_OUT. */ 3598 if (pf_state_insert(BOUND_IFACE(r, kif), 3599 (pd->dir == PF_IN) ? sk : nk, 3600 (pd->dir == PF_IN) ? nk : sk, s)) { 3601 if (pd->proto == IPPROTO_TCP) 3602 pf_normalize_tcp_cleanup(s); 3603 REASON_SET(&reason, PFRES_STATEINS); 3604 pf_src_tree_remove_state(s); 3605 STATE_DEC_COUNTERS(s); 3606 uma_zfree(V_pf_state_z, s); 3607 return (PF_DROP); 3608 } else 3609 *sm = s; 3610 3611 if (tag > 0) 3612 s->tag = tag; 3613 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 3614 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 3615 s->src.state = PF_TCPS_PROXY_SRC; 3616 /* undo NAT changes, if they have taken place */ 3617 if (nr != NULL) { 3618 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 3619 if (pd->dir == PF_OUT) 3620 skt = s->key[PF_SK_STACK]; 3621 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 3622 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 3623 if (pd->sport) 3624 *pd->sport = skt->port[pd->sidx]; 3625 if (pd->dport) 3626 *pd->dport = skt->port[pd->didx]; 3627 if (pd->proto_sum) 3628 *pd->proto_sum = bproto_sum; 3629 if (pd->ip_sum) 3630 *pd->ip_sum = bip_sum; 3631 m_copyback(m, off, hdrlen, pd->hdr.any); 3632 } 3633 s->src.seqhi = htonl(arc4random()); 3634 /* Find mss option */ 3635 int rtid = M_GETFIB(m); 3636 mss = pf_get_mss(m, off, th->th_off, pd->af); 3637 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 3638 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 3639 s->src.mss = mss; 3640 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport, 3641 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 3642 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL); 3643 REASON_SET(&reason, PFRES_SYNPROXY); 3644 return (PF_SYNPROXY_DROP); 3645 } 3646 3647 return (PF_PASS); 3648 3649 csfailed: 3650 if (sk != NULL) 3651 uma_zfree(V_pf_state_key_z, sk); 3652 if (nk != NULL) 3653 uma_zfree(V_pf_state_key_z, nk); 3654 3655 if (sn != NULL) { 3656 struct pf_srchash *sh; 3657 3658 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 3659 PF_HASHROW_LOCK(sh); 3660 if (--sn->states == 0 && sn->expire == 0) { 3661 pf_unlink_src_node(sn); 3662 uma_zfree(V_pf_sources_z, sn); 3663 counter_u64_add( 3664 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 3665 } 3666 PF_HASHROW_UNLOCK(sh); 3667 } 3668 3669 if (nsn != sn && nsn != NULL) { 3670 struct pf_srchash *sh; 3671 3672 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)]; 3673 PF_HASHROW_LOCK(sh); 3674 if (--nsn->states == 1 && nsn->expire == 0) { 3675 pf_unlink_src_node(nsn); 3676 uma_zfree(V_pf_sources_z, nsn); 3677 counter_u64_add( 3678 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 3679 } 3680 PF_HASHROW_UNLOCK(sh); 3681 } 3682 3683 return (PF_DROP); 3684 } 3685 3686 static int 3687 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif, 3688 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am, 3689 struct pf_ruleset **rsm) 3690 { 3691 struct pf_rule *r, *a = NULL; 3692 struct pf_ruleset *ruleset = NULL; 3693 sa_family_t af = pd->af; 3694 u_short reason; 3695 int tag = -1; 3696 int asd = 0; 3697 int match = 0; 3698 struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3699 3700 PF_RULES_RASSERT(); 3701 3702 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3703 while (r != NULL) { 3704 r->evaluations++; 3705 if (pfi_kif_match(r->kif, kif) == r->ifnot) 3706 r = r->skip[PF_SKIP_IFP].ptr; 3707 else if (r->direction && r->direction != direction) 3708 r = r->skip[PF_SKIP_DIR].ptr; 3709 else if (r->af && r->af != af) 3710 r = r->skip[PF_SKIP_AF].ptr; 3711 else if (r->proto && r->proto != pd->proto) 3712 r = r->skip[PF_SKIP_PROTO].ptr; 3713 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 3714 r->src.neg, kif, M_GETFIB(m))) 3715 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3716 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 3717 r->dst.neg, NULL, M_GETFIB(m))) 3718 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3719 else if (r->tos && !(r->tos == pd->tos)) 3720 r = TAILQ_NEXT(r, entries); 3721 else if (r->os_fingerprint != PF_OSFP_ANY) 3722 r = TAILQ_NEXT(r, entries); 3723 else if (pd->proto == IPPROTO_UDP && 3724 (r->src.port_op || r->dst.port_op)) 3725 r = TAILQ_NEXT(r, entries); 3726 else if (pd->proto == IPPROTO_TCP && 3727 (r->src.port_op || r->dst.port_op || r->flagset)) 3728 r = TAILQ_NEXT(r, entries); 3729 else if ((pd->proto == IPPROTO_ICMP || 3730 pd->proto == IPPROTO_ICMPV6) && 3731 (r->type || r->code)) 3732 r = TAILQ_NEXT(r, entries); 3733 else if (r->prob && r->prob <= 3734 (arc4random() % (UINT_MAX - 1) + 1)) 3735 r = TAILQ_NEXT(r, entries); 3736 else if (r->match_tag && !pf_match_tag(m, r, &tag, 3737 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 3738 r = TAILQ_NEXT(r, entries); 3739 else { 3740 if (r->anchor == NULL) { 3741 match = 1; 3742 *rm = r; 3743 *am = a; 3744 *rsm = ruleset; 3745 if ((*rm)->quick) 3746 break; 3747 r = TAILQ_NEXT(r, entries); 3748 } else 3749 pf_step_into_anchor(anchor_stack, &asd, 3750 &ruleset, PF_RULESET_FILTER, &r, &a, 3751 &match); 3752 } 3753 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 3754 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 3755 break; 3756 } 3757 r = *rm; 3758 a = *am; 3759 ruleset = *rsm; 3760 3761 REASON_SET(&reason, PFRES_MATCH); 3762 3763 if (r->log) 3764 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd, 3765 1); 3766 3767 if (r->action != PF_PASS) 3768 return (PF_DROP); 3769 3770 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 3771 REASON_SET(&reason, PFRES_MEMORY); 3772 return (PF_DROP); 3773 } 3774 3775 return (PF_PASS); 3776 } 3777 3778 static int 3779 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst, 3780 struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off, 3781 struct pf_pdesc *pd, u_short *reason, int *copyback) 3782 { 3783 struct tcphdr *th = pd->hdr.tcp; 3784 u_int16_t win = ntohs(th->th_win); 3785 u_int32_t ack, end, seq, orig_seq; 3786 u_int8_t sws, dws; 3787 int ackskew; 3788 3789 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 3790 sws = src->wscale & PF_WSCALE_MASK; 3791 dws = dst->wscale & PF_WSCALE_MASK; 3792 } else 3793 sws = dws = 0; 3794 3795 /* 3796 * Sequence tracking algorithm from Guido van Rooij's paper: 3797 * http://www.madison-gurkha.com/publications/tcp_filtering/ 3798 * tcp_filtering.ps 3799 */ 3800 3801 orig_seq = seq = ntohl(th->th_seq); 3802 if (src->seqlo == 0) { 3803 /* First packet from this end. Set its state */ 3804 3805 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) && 3806 src->scrub == NULL) { 3807 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 3808 REASON_SET(reason, PFRES_MEMORY); 3809 return (PF_DROP); 3810 } 3811 } 3812 3813 /* Deferred generation of sequence number modulator */ 3814 if (dst->seqdiff && !src->seqdiff) { 3815 /* use random iss for the TCP server */ 3816 while ((src->seqdiff = arc4random() - seq) == 0) 3817 ; 3818 ack = ntohl(th->th_ack) - dst->seqdiff; 3819 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq + 3820 src->seqdiff), 0); 3821 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0); 3822 *copyback = 1; 3823 } else { 3824 ack = ntohl(th->th_ack); 3825 } 3826 3827 end = seq + pd->p_len; 3828 if (th->th_flags & TH_SYN) { 3829 end++; 3830 if (dst->wscale & PF_WSCALE_FLAG) { 3831 src->wscale = pf_get_wscale(m, off, th->th_off, 3832 pd->af); 3833 if (src->wscale & PF_WSCALE_FLAG) { 3834 /* Remove scale factor from initial 3835 * window */ 3836 sws = src->wscale & PF_WSCALE_MASK; 3837 win = ((u_int32_t)win + (1 << sws) - 1) 3838 >> sws; 3839 dws = dst->wscale & PF_WSCALE_MASK; 3840 } else { 3841 /* fixup other window */ 3842 dst->max_win <<= dst->wscale & 3843 PF_WSCALE_MASK; 3844 /* in case of a retrans SYN|ACK */ 3845 dst->wscale = 0; 3846 } 3847 } 3848 } 3849 if (th->th_flags & TH_FIN) 3850 end++; 3851 3852 src->seqlo = seq; 3853 if (src->state < TCPS_SYN_SENT) 3854 src->state = TCPS_SYN_SENT; 3855 3856 /* 3857 * May need to slide the window (seqhi may have been set by 3858 * the crappy stack check or if we picked up the connection 3859 * after establishment) 3860 */ 3861 if (src->seqhi == 1 || 3862 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 3863 src->seqhi = end + MAX(1, dst->max_win << dws); 3864 if (win > src->max_win) 3865 src->max_win = win; 3866 3867 } else { 3868 ack = ntohl(th->th_ack) - dst->seqdiff; 3869 if (src->seqdiff) { 3870 /* Modulate sequence numbers */ 3871 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq + 3872 src->seqdiff), 0); 3873 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0); 3874 *copyback = 1; 3875 } 3876 end = seq + pd->p_len; 3877 if (th->th_flags & TH_SYN) 3878 end++; 3879 if (th->th_flags & TH_FIN) 3880 end++; 3881 } 3882 3883 if ((th->th_flags & TH_ACK) == 0) { 3884 /* Let it pass through the ack skew check */ 3885 ack = dst->seqlo; 3886 } else if ((ack == 0 && 3887 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 3888 /* broken tcp stacks do not set ack */ 3889 (dst->state < TCPS_SYN_SENT)) { 3890 /* 3891 * Many stacks (ours included) will set the ACK number in an 3892 * FIN|ACK if the SYN times out -- no sequence to ACK. 3893 */ 3894 ack = dst->seqlo; 3895 } 3896 3897 if (seq == end) { 3898 /* Ease sequencing restrictions on no data packets */ 3899 seq = src->seqlo; 3900 end = seq; 3901 } 3902 3903 ackskew = dst->seqlo - ack; 3904 3905 3906 /* 3907 * Need to demodulate the sequence numbers in any TCP SACK options 3908 * (Selective ACK). We could optionally validate the SACK values 3909 * against the current ACK window, either forwards or backwards, but 3910 * I'm not confident that SACK has been implemented properly 3911 * everywhere. It wouldn't surprise me if several stacks accidently 3912 * SACK too far backwards of previously ACKed data. There really aren't 3913 * any security implications of bad SACKing unless the target stack 3914 * doesn't validate the option length correctly. Someone trying to 3915 * spoof into a TCP connection won't bother blindly sending SACK 3916 * options anyway. 3917 */ 3918 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 3919 if (pf_modulate_sack(m, off, pd, th, dst)) 3920 *copyback = 1; 3921 } 3922 3923 3924 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 3925 if (SEQ_GEQ(src->seqhi, end) && 3926 /* Last octet inside other's window space */ 3927 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 3928 /* Retrans: not more than one window back */ 3929 (ackskew >= -MAXACKWINDOW) && 3930 /* Acking not more than one reassembled fragment backwards */ 3931 (ackskew <= (MAXACKWINDOW << sws)) && 3932 /* Acking not more than one window forward */ 3933 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 3934 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 3935 (pd->flags & PFDESC_IP_REAS) == 0)) { 3936 /* Require an exact/+1 sequence match on resets when possible */ 3937 3938 if (dst->scrub || src->scrub) { 3939 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 3940 *state, src, dst, copyback)) 3941 return (PF_DROP); 3942 } 3943 3944 /* update max window */ 3945 if (src->max_win < win) 3946 src->max_win = win; 3947 /* synchronize sequencing */ 3948 if (SEQ_GT(end, src->seqlo)) 3949 src->seqlo = end; 3950 /* slide the window of what the other end can send */ 3951 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 3952 dst->seqhi = ack + MAX((win << sws), 1); 3953 3954 3955 /* update states */ 3956 if (th->th_flags & TH_SYN) 3957 if (src->state < TCPS_SYN_SENT) 3958 src->state = TCPS_SYN_SENT; 3959 if (th->th_flags & TH_FIN) 3960 if (src->state < TCPS_CLOSING) 3961 src->state = TCPS_CLOSING; 3962 if (th->th_flags & TH_ACK) { 3963 if (dst->state == TCPS_SYN_SENT) { 3964 dst->state = TCPS_ESTABLISHED; 3965 if (src->state == TCPS_ESTABLISHED && 3966 (*state)->src_node != NULL && 3967 pf_src_connlimit(state)) { 3968 REASON_SET(reason, PFRES_SRCLIMIT); 3969 return (PF_DROP); 3970 } 3971 } else if (dst->state == TCPS_CLOSING) 3972 dst->state = TCPS_FIN_WAIT_2; 3973 } 3974 if (th->th_flags & TH_RST) 3975 src->state = dst->state = TCPS_TIME_WAIT; 3976 3977 /* update expire time */ 3978 (*state)->expire = time_uptime; 3979 if (src->state >= TCPS_FIN_WAIT_2 && 3980 dst->state >= TCPS_FIN_WAIT_2) 3981 (*state)->timeout = PFTM_TCP_CLOSED; 3982 else if (src->state >= TCPS_CLOSING && 3983 dst->state >= TCPS_CLOSING) 3984 (*state)->timeout = PFTM_TCP_FIN_WAIT; 3985 else if (src->state < TCPS_ESTABLISHED || 3986 dst->state < TCPS_ESTABLISHED) 3987 (*state)->timeout = PFTM_TCP_OPENING; 3988 else if (src->state >= TCPS_CLOSING || 3989 dst->state >= TCPS_CLOSING) 3990 (*state)->timeout = PFTM_TCP_CLOSING; 3991 else 3992 (*state)->timeout = PFTM_TCP_ESTABLISHED; 3993 3994 /* Fall through to PASS packet */ 3995 3996 } else if ((dst->state < TCPS_SYN_SENT || 3997 dst->state >= TCPS_FIN_WAIT_2 || 3998 src->state >= TCPS_FIN_WAIT_2) && 3999 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 4000 /* Within a window forward of the originating packet */ 4001 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 4002 /* Within a window backward of the originating packet */ 4003 4004 /* 4005 * This currently handles three situations: 4006 * 1) Stupid stacks will shotgun SYNs before their peer 4007 * replies. 4008 * 2) When PF catches an already established stream (the 4009 * firewall rebooted, the state table was flushed, routes 4010 * changed...) 4011 * 3) Packets get funky immediately after the connection 4012 * closes (this should catch Solaris spurious ACK|FINs 4013 * that web servers like to spew after a close) 4014 * 4015 * This must be a little more careful than the above code 4016 * since packet floods will also be caught here. We don't 4017 * update the TTL here to mitigate the damage of a packet 4018 * flood and so the same code can handle awkward establishment 4019 * and a loosened connection close. 4020 * In the establishment case, a correct peer response will 4021 * validate the connection, go through the normal state code 4022 * and keep updating the state TTL. 4023 */ 4024 4025 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4026 printf("pf: loose state match: "); 4027 pf_print_state(*state); 4028 pf_print_flags(th->th_flags); 4029 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4030 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 4031 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 4032 (unsigned long long)(*state)->packets[1], 4033 pd->dir == PF_IN ? "in" : "out", 4034 pd->dir == (*state)->direction ? "fwd" : "rev"); 4035 } 4036 4037 if (dst->scrub || src->scrub) { 4038 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4039 *state, src, dst, copyback)) 4040 return (PF_DROP); 4041 } 4042 4043 /* update max window */ 4044 if (src->max_win < win) 4045 src->max_win = win; 4046 /* synchronize sequencing */ 4047 if (SEQ_GT(end, src->seqlo)) 4048 src->seqlo = end; 4049 /* slide the window of what the other end can send */ 4050 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4051 dst->seqhi = ack + MAX((win << sws), 1); 4052 4053 /* 4054 * Cannot set dst->seqhi here since this could be a shotgunned 4055 * SYN and not an already established connection. 4056 */ 4057 4058 if (th->th_flags & TH_FIN) 4059 if (src->state < TCPS_CLOSING) 4060 src->state = TCPS_CLOSING; 4061 if (th->th_flags & TH_RST) 4062 src->state = dst->state = TCPS_TIME_WAIT; 4063 4064 /* Fall through to PASS packet */ 4065 4066 } else { 4067 if ((*state)->dst.state == TCPS_SYN_SENT && 4068 (*state)->src.state == TCPS_SYN_SENT) { 4069 /* Send RST for state mismatches during handshake */ 4070 if (!(th->th_flags & TH_RST)) 4071 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4072 pd->dst, pd->src, th->th_dport, 4073 th->th_sport, ntohl(th->th_ack), 0, 4074 TH_RST, 0, 0, 4075 (*state)->rule.ptr->return_ttl, 1, 0, 4076 kif->pfik_ifp); 4077 src->seqlo = 0; 4078 src->seqhi = 1; 4079 src->max_win = 1; 4080 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 4081 printf("pf: BAD state: "); 4082 pf_print_state(*state); 4083 pf_print_flags(th->th_flags); 4084 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4085 "pkts=%llu:%llu dir=%s,%s\n", 4086 seq, orig_seq, ack, pd->p_len, ackskew, 4087 (unsigned long long)(*state)->packets[0], 4088 (unsigned long long)(*state)->packets[1], 4089 pd->dir == PF_IN ? "in" : "out", 4090 pd->dir == (*state)->direction ? "fwd" : "rev"); 4091 printf("pf: State failure on: %c %c %c %c | %c %c\n", 4092 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 4093 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 4094 ' ': '2', 4095 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 4096 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 4097 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 4098 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 4099 } 4100 REASON_SET(reason, PFRES_BADSTATE); 4101 return (PF_DROP); 4102 } 4103 4104 return (PF_PASS); 4105 } 4106 4107 static int 4108 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst, 4109 struct pf_state **state, struct pf_pdesc *pd, u_short *reason) 4110 { 4111 struct tcphdr *th = pd->hdr.tcp; 4112 4113 if (th->th_flags & TH_SYN) 4114 if (src->state < TCPS_SYN_SENT) 4115 src->state = TCPS_SYN_SENT; 4116 if (th->th_flags & TH_FIN) 4117 if (src->state < TCPS_CLOSING) 4118 src->state = TCPS_CLOSING; 4119 if (th->th_flags & TH_ACK) { 4120 if (dst->state == TCPS_SYN_SENT) { 4121 dst->state = TCPS_ESTABLISHED; 4122 if (src->state == TCPS_ESTABLISHED && 4123 (*state)->src_node != NULL && 4124 pf_src_connlimit(state)) { 4125 REASON_SET(reason, PFRES_SRCLIMIT); 4126 return (PF_DROP); 4127 } 4128 } else if (dst->state == TCPS_CLOSING) { 4129 dst->state = TCPS_FIN_WAIT_2; 4130 } else if (src->state == TCPS_SYN_SENT && 4131 dst->state < TCPS_SYN_SENT) { 4132 /* 4133 * Handle a special sloppy case where we only see one 4134 * half of the connection. If there is a ACK after 4135 * the initial SYN without ever seeing a packet from 4136 * the destination, set the connection to established. 4137 */ 4138 dst->state = src->state = TCPS_ESTABLISHED; 4139 if ((*state)->src_node != NULL && 4140 pf_src_connlimit(state)) { 4141 REASON_SET(reason, PFRES_SRCLIMIT); 4142 return (PF_DROP); 4143 } 4144 } else if (src->state == TCPS_CLOSING && 4145 dst->state == TCPS_ESTABLISHED && 4146 dst->seqlo == 0) { 4147 /* 4148 * Handle the closing of half connections where we 4149 * don't see the full bidirectional FIN/ACK+ACK 4150 * handshake. 4151 */ 4152 dst->state = TCPS_CLOSING; 4153 } 4154 } 4155 if (th->th_flags & TH_RST) 4156 src->state = dst->state = TCPS_TIME_WAIT; 4157 4158 /* update expire time */ 4159 (*state)->expire = time_uptime; 4160 if (src->state >= TCPS_FIN_WAIT_2 && 4161 dst->state >= TCPS_FIN_WAIT_2) 4162 (*state)->timeout = PFTM_TCP_CLOSED; 4163 else if (src->state >= TCPS_CLOSING && 4164 dst->state >= TCPS_CLOSING) 4165 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4166 else if (src->state < TCPS_ESTABLISHED || 4167 dst->state < TCPS_ESTABLISHED) 4168 (*state)->timeout = PFTM_TCP_OPENING; 4169 else if (src->state >= TCPS_CLOSING || 4170 dst->state >= TCPS_CLOSING) 4171 (*state)->timeout = PFTM_TCP_CLOSING; 4172 else 4173 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4174 4175 return (PF_PASS); 4176 } 4177 4178 static int 4179 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif, 4180 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 4181 u_short *reason) 4182 { 4183 struct pf_state_key_cmp key; 4184 struct tcphdr *th = pd->hdr.tcp; 4185 int copyback = 0; 4186 struct pf_state_peer *src, *dst; 4187 struct pf_state_key *sk; 4188 4189 bzero(&key, sizeof(key)); 4190 key.af = pd->af; 4191 key.proto = IPPROTO_TCP; 4192 if (direction == PF_IN) { /* wire side, straight */ 4193 PF_ACPY(&key.addr[0], pd->src, key.af); 4194 PF_ACPY(&key.addr[1], pd->dst, key.af); 4195 key.port[0] = th->th_sport; 4196 key.port[1] = th->th_dport; 4197 } else { /* stack side, reverse */ 4198 PF_ACPY(&key.addr[1], pd->src, key.af); 4199 PF_ACPY(&key.addr[0], pd->dst, key.af); 4200 key.port[1] = th->th_sport; 4201 key.port[0] = th->th_dport; 4202 } 4203 4204 STATE_LOOKUP(kif, &key, direction, *state, pd); 4205 4206 if (direction == (*state)->direction) { 4207 src = &(*state)->src; 4208 dst = &(*state)->dst; 4209 } else { 4210 src = &(*state)->dst; 4211 dst = &(*state)->src; 4212 } 4213 4214 sk = (*state)->key[pd->didx]; 4215 4216 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 4217 if (direction != (*state)->direction) { 4218 REASON_SET(reason, PFRES_SYNPROXY); 4219 return (PF_SYNPROXY_DROP); 4220 } 4221 if (th->th_flags & TH_SYN) { 4222 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 4223 REASON_SET(reason, PFRES_SYNPROXY); 4224 return (PF_DROP); 4225 } 4226 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, 4227 pd->src, th->th_dport, th->th_sport, 4228 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 4229 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL); 4230 REASON_SET(reason, PFRES_SYNPROXY); 4231 return (PF_SYNPROXY_DROP); 4232 } else if (!(th->th_flags & TH_ACK) || 4233 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4234 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4235 REASON_SET(reason, PFRES_SYNPROXY); 4236 return (PF_DROP); 4237 } else if ((*state)->src_node != NULL && 4238 pf_src_connlimit(state)) { 4239 REASON_SET(reason, PFRES_SRCLIMIT); 4240 return (PF_DROP); 4241 } else 4242 (*state)->src.state = PF_TCPS_PROXY_DST; 4243 } 4244 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 4245 if (direction == (*state)->direction) { 4246 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 4247 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4248 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4249 REASON_SET(reason, PFRES_SYNPROXY); 4250 return (PF_DROP); 4251 } 4252 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 4253 if ((*state)->dst.seqhi == 1) 4254 (*state)->dst.seqhi = htonl(arc4random()); 4255 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4256 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4257 sk->port[pd->sidx], sk->port[pd->didx], 4258 (*state)->dst.seqhi, 0, TH_SYN, 0, 4259 (*state)->src.mss, 0, 0, (*state)->tag, NULL); 4260 REASON_SET(reason, PFRES_SYNPROXY); 4261 return (PF_SYNPROXY_DROP); 4262 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 4263 (TH_SYN|TH_ACK)) || 4264 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 4265 REASON_SET(reason, PFRES_SYNPROXY); 4266 return (PF_DROP); 4267 } else { 4268 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 4269 (*state)->dst.seqlo = ntohl(th->th_seq); 4270 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, 4271 pd->src, th->th_dport, th->th_sport, 4272 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 4273 TH_ACK, (*state)->src.max_win, 0, 0, 0, 4274 (*state)->tag, NULL); 4275 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4276 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4277 sk->port[pd->sidx], sk->port[pd->didx], 4278 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 4279 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL); 4280 (*state)->src.seqdiff = (*state)->dst.seqhi - 4281 (*state)->src.seqlo; 4282 (*state)->dst.seqdiff = (*state)->src.seqhi - 4283 (*state)->dst.seqlo; 4284 (*state)->src.seqhi = (*state)->src.seqlo + 4285 (*state)->dst.max_win; 4286 (*state)->dst.seqhi = (*state)->dst.seqlo + 4287 (*state)->src.max_win; 4288 (*state)->src.wscale = (*state)->dst.wscale = 0; 4289 (*state)->src.state = (*state)->dst.state = 4290 TCPS_ESTABLISHED; 4291 REASON_SET(reason, PFRES_SYNPROXY); 4292 return (PF_SYNPROXY_DROP); 4293 } 4294 } 4295 4296 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) && 4297 dst->state >= TCPS_FIN_WAIT_2 && 4298 src->state >= TCPS_FIN_WAIT_2) { 4299 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4300 printf("pf: state reuse "); 4301 pf_print_state(*state); 4302 pf_print_flags(th->th_flags); 4303 printf("\n"); 4304 } 4305 /* XXX make sure it's the same direction ?? */ 4306 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; 4307 pf_unlink_state(*state, PF_ENTER_LOCKED); 4308 *state = NULL; 4309 return (PF_DROP); 4310 } 4311 4312 if ((*state)->state_flags & PFSTATE_SLOPPY) { 4313 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP) 4314 return (PF_DROP); 4315 } else { 4316 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason, 4317 ©back) == PF_DROP) 4318 return (PF_DROP); 4319 } 4320 4321 /* translate source/destination address, if necessary */ 4322 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4323 struct pf_state_key *nk = (*state)->key[pd->didx]; 4324 4325 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4326 nk->port[pd->sidx] != th->th_sport) 4327 pf_change_ap(pd->src, &th->th_sport, pd->ip_sum, 4328 &th->th_sum, &nk->addr[pd->sidx], 4329 nk->port[pd->sidx], 0, pd->af); 4330 4331 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 4332 nk->port[pd->didx] != th->th_dport) 4333 pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum, 4334 &th->th_sum, &nk->addr[pd->didx], 4335 nk->port[pd->didx], 0, pd->af); 4336 copyback = 1; 4337 } 4338 4339 /* Copyback sequence modulation or stateful scrub changes if needed */ 4340 if (copyback) 4341 m_copyback(m, off, sizeof(*th), (caddr_t)th); 4342 4343 return (PF_PASS); 4344 } 4345 4346 static int 4347 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif, 4348 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 4349 { 4350 struct pf_state_peer *src, *dst; 4351 struct pf_state_key_cmp key; 4352 struct udphdr *uh = pd->hdr.udp; 4353 4354 bzero(&key, sizeof(key)); 4355 key.af = pd->af; 4356 key.proto = IPPROTO_UDP; 4357 if (direction == PF_IN) { /* wire side, straight */ 4358 PF_ACPY(&key.addr[0], pd->src, key.af); 4359 PF_ACPY(&key.addr[1], pd->dst, key.af); 4360 key.port[0] = uh->uh_sport; 4361 key.port[1] = uh->uh_dport; 4362 } else { /* stack side, reverse */ 4363 PF_ACPY(&key.addr[1], pd->src, key.af); 4364 PF_ACPY(&key.addr[0], pd->dst, key.af); 4365 key.port[1] = uh->uh_sport; 4366 key.port[0] = uh->uh_dport; 4367 } 4368 4369 STATE_LOOKUP(kif, &key, direction, *state, pd); 4370 4371 if (direction == (*state)->direction) { 4372 src = &(*state)->src; 4373 dst = &(*state)->dst; 4374 } else { 4375 src = &(*state)->dst; 4376 dst = &(*state)->src; 4377 } 4378 4379 /* update states */ 4380 if (src->state < PFUDPS_SINGLE) 4381 src->state = PFUDPS_SINGLE; 4382 if (dst->state == PFUDPS_SINGLE) 4383 dst->state = PFUDPS_MULTIPLE; 4384 4385 /* update expire time */ 4386 (*state)->expire = time_uptime; 4387 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 4388 (*state)->timeout = PFTM_UDP_MULTIPLE; 4389 else 4390 (*state)->timeout = PFTM_UDP_SINGLE; 4391 4392 /* translate source/destination address, if necessary */ 4393 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4394 struct pf_state_key *nk = (*state)->key[pd->didx]; 4395 4396 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4397 nk->port[pd->sidx] != uh->uh_sport) 4398 pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum, 4399 &uh->uh_sum, &nk->addr[pd->sidx], 4400 nk->port[pd->sidx], 1, pd->af); 4401 4402 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 4403 nk->port[pd->didx] != uh->uh_dport) 4404 pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum, 4405 &uh->uh_sum, &nk->addr[pd->didx], 4406 nk->port[pd->didx], 1, pd->af); 4407 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 4408 } 4409 4410 return (PF_PASS); 4411 } 4412 4413 static int 4414 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif, 4415 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 4416 { 4417 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 4418 u_int16_t icmpid = 0, *icmpsum; 4419 u_int8_t icmptype; 4420 int state_icmp = 0; 4421 struct pf_state_key_cmp key; 4422 4423 bzero(&key, sizeof(key)); 4424 switch (pd->proto) { 4425 #ifdef INET 4426 case IPPROTO_ICMP: 4427 icmptype = pd->hdr.icmp->icmp_type; 4428 icmpid = pd->hdr.icmp->icmp_id; 4429 icmpsum = &pd->hdr.icmp->icmp_cksum; 4430 4431 if (icmptype == ICMP_UNREACH || 4432 icmptype == ICMP_SOURCEQUENCH || 4433 icmptype == ICMP_REDIRECT || 4434 icmptype == ICMP_TIMXCEED || 4435 icmptype == ICMP_PARAMPROB) 4436 state_icmp++; 4437 break; 4438 #endif /* INET */ 4439 #ifdef INET6 4440 case IPPROTO_ICMPV6: 4441 icmptype = pd->hdr.icmp6->icmp6_type; 4442 icmpid = pd->hdr.icmp6->icmp6_id; 4443 icmpsum = &pd->hdr.icmp6->icmp6_cksum; 4444 4445 if (icmptype == ICMP6_DST_UNREACH || 4446 icmptype == ICMP6_PACKET_TOO_BIG || 4447 icmptype == ICMP6_TIME_EXCEEDED || 4448 icmptype == ICMP6_PARAM_PROB) 4449 state_icmp++; 4450 break; 4451 #endif /* INET6 */ 4452 } 4453 4454 if (!state_icmp) { 4455 4456 /* 4457 * ICMP query/reply message not related to a TCP/UDP packet. 4458 * Search for an ICMP state. 4459 */ 4460 key.af = pd->af; 4461 key.proto = pd->proto; 4462 key.port[0] = key.port[1] = icmpid; 4463 if (direction == PF_IN) { /* wire side, straight */ 4464 PF_ACPY(&key.addr[0], pd->src, key.af); 4465 PF_ACPY(&key.addr[1], pd->dst, key.af); 4466 } else { /* stack side, reverse */ 4467 PF_ACPY(&key.addr[1], pd->src, key.af); 4468 PF_ACPY(&key.addr[0], pd->dst, key.af); 4469 } 4470 4471 STATE_LOOKUP(kif, &key, direction, *state, pd); 4472 4473 (*state)->expire = time_uptime; 4474 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 4475 4476 /* translate source/destination address, if necessary */ 4477 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4478 struct pf_state_key *nk = (*state)->key[pd->didx]; 4479 4480 switch (pd->af) { 4481 #ifdef INET 4482 case AF_INET: 4483 if (PF_ANEQ(pd->src, 4484 &nk->addr[pd->sidx], AF_INET)) 4485 pf_change_a(&saddr->v4.s_addr, 4486 pd->ip_sum, 4487 nk->addr[pd->sidx].v4.s_addr, 0); 4488 4489 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 4490 AF_INET)) 4491 pf_change_a(&daddr->v4.s_addr, 4492 pd->ip_sum, 4493 nk->addr[pd->didx].v4.s_addr, 0); 4494 4495 if (nk->port[0] != 4496 pd->hdr.icmp->icmp_id) { 4497 pd->hdr.icmp->icmp_cksum = 4498 pf_cksum_fixup( 4499 pd->hdr.icmp->icmp_cksum, icmpid, 4500 nk->port[pd->sidx], 0); 4501 pd->hdr.icmp->icmp_id = 4502 nk->port[pd->sidx]; 4503 } 4504 4505 m_copyback(m, off, ICMP_MINLEN, 4506 (caddr_t )pd->hdr.icmp); 4507 break; 4508 #endif /* INET */ 4509 #ifdef INET6 4510 case AF_INET6: 4511 if (PF_ANEQ(pd->src, 4512 &nk->addr[pd->sidx], AF_INET6)) 4513 pf_change_a6(saddr, 4514 &pd->hdr.icmp6->icmp6_cksum, 4515 &nk->addr[pd->sidx], 0); 4516 4517 if (PF_ANEQ(pd->dst, 4518 &nk->addr[pd->didx], AF_INET6)) 4519 pf_change_a6(daddr, 4520 &pd->hdr.icmp6->icmp6_cksum, 4521 &nk->addr[pd->didx], 0); 4522 4523 m_copyback(m, off, sizeof(struct icmp6_hdr), 4524 (caddr_t )pd->hdr.icmp6); 4525 break; 4526 #endif /* INET6 */ 4527 } 4528 } 4529 return (PF_PASS); 4530 4531 } else { 4532 /* 4533 * ICMP error message in response to a TCP/UDP packet. 4534 * Extract the inner TCP/UDP header and search for that state. 4535 */ 4536 4537 struct pf_pdesc pd2; 4538 bzero(&pd2, sizeof pd2); 4539 #ifdef INET 4540 struct ip h2; 4541 #endif /* INET */ 4542 #ifdef INET6 4543 struct ip6_hdr h2_6; 4544 int terminal = 0; 4545 #endif /* INET6 */ 4546 int ipoff2 = 0; 4547 int off2 = 0; 4548 4549 pd2.af = pd->af; 4550 /* Payload packet is from the opposite direction. */ 4551 pd2.sidx = (direction == PF_IN) ? 1 : 0; 4552 pd2.didx = (direction == PF_IN) ? 0 : 1; 4553 switch (pd->af) { 4554 #ifdef INET 4555 case AF_INET: 4556 /* offset of h2 in mbuf chain */ 4557 ipoff2 = off + ICMP_MINLEN; 4558 4559 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 4560 NULL, reason, pd2.af)) { 4561 DPFPRINTF(PF_DEBUG_MISC, 4562 ("pf: ICMP error message too short " 4563 "(ip)\n")); 4564 return (PF_DROP); 4565 } 4566 /* 4567 * ICMP error messages don't refer to non-first 4568 * fragments 4569 */ 4570 if (h2.ip_off & htons(IP_OFFMASK)) { 4571 REASON_SET(reason, PFRES_FRAG); 4572 return (PF_DROP); 4573 } 4574 4575 /* offset of protocol header that follows h2 */ 4576 off2 = ipoff2 + (h2.ip_hl << 2); 4577 4578 pd2.proto = h2.ip_p; 4579 pd2.src = (struct pf_addr *)&h2.ip_src; 4580 pd2.dst = (struct pf_addr *)&h2.ip_dst; 4581 pd2.ip_sum = &h2.ip_sum; 4582 break; 4583 #endif /* INET */ 4584 #ifdef INET6 4585 case AF_INET6: 4586 ipoff2 = off + sizeof(struct icmp6_hdr); 4587 4588 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 4589 NULL, reason, pd2.af)) { 4590 DPFPRINTF(PF_DEBUG_MISC, 4591 ("pf: ICMP error message too short " 4592 "(ip6)\n")); 4593 return (PF_DROP); 4594 } 4595 pd2.proto = h2_6.ip6_nxt; 4596 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 4597 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 4598 pd2.ip_sum = NULL; 4599 off2 = ipoff2 + sizeof(h2_6); 4600 do { 4601 switch (pd2.proto) { 4602 case IPPROTO_FRAGMENT: 4603 /* 4604 * ICMPv6 error messages for 4605 * non-first fragments 4606 */ 4607 REASON_SET(reason, PFRES_FRAG); 4608 return (PF_DROP); 4609 case IPPROTO_AH: 4610 case IPPROTO_HOPOPTS: 4611 case IPPROTO_ROUTING: 4612 case IPPROTO_DSTOPTS: { 4613 /* get next header and header length */ 4614 struct ip6_ext opt6; 4615 4616 if (!pf_pull_hdr(m, off2, &opt6, 4617 sizeof(opt6), NULL, reason, 4618 pd2.af)) { 4619 DPFPRINTF(PF_DEBUG_MISC, 4620 ("pf: ICMPv6 short opt\n")); 4621 return (PF_DROP); 4622 } 4623 if (pd2.proto == IPPROTO_AH) 4624 off2 += (opt6.ip6e_len + 2) * 4; 4625 else 4626 off2 += (opt6.ip6e_len + 1) * 8; 4627 pd2.proto = opt6.ip6e_nxt; 4628 /* goto the next header */ 4629 break; 4630 } 4631 default: 4632 terminal++; 4633 break; 4634 } 4635 } while (!terminal); 4636 break; 4637 #endif /* INET6 */ 4638 } 4639 4640 switch (pd2.proto) { 4641 case IPPROTO_TCP: { 4642 struct tcphdr th; 4643 u_int32_t seq; 4644 struct pf_state_peer *src, *dst; 4645 u_int8_t dws; 4646 int copyback = 0; 4647 4648 /* 4649 * Only the first 8 bytes of the TCP header can be 4650 * expected. Don't access any TCP header fields after 4651 * th_seq, an ackskew test is not possible. 4652 */ 4653 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 4654 pd2.af)) { 4655 DPFPRINTF(PF_DEBUG_MISC, 4656 ("pf: ICMP error message too short " 4657 "(tcp)\n")); 4658 return (PF_DROP); 4659 } 4660 4661 key.af = pd2.af; 4662 key.proto = IPPROTO_TCP; 4663 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4664 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4665 key.port[pd2.sidx] = th.th_sport; 4666 key.port[pd2.didx] = th.th_dport; 4667 4668 STATE_LOOKUP(kif, &key, direction, *state, pd); 4669 4670 if (direction == (*state)->direction) { 4671 src = &(*state)->dst; 4672 dst = &(*state)->src; 4673 } else { 4674 src = &(*state)->src; 4675 dst = &(*state)->dst; 4676 } 4677 4678 if (src->wscale && dst->wscale) 4679 dws = dst->wscale & PF_WSCALE_MASK; 4680 else 4681 dws = 0; 4682 4683 /* Demodulate sequence number */ 4684 seq = ntohl(th.th_seq) - src->seqdiff; 4685 if (src->seqdiff) { 4686 pf_change_a(&th.th_seq, icmpsum, 4687 htonl(seq), 0); 4688 copyback = 1; 4689 } 4690 4691 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 4692 (!SEQ_GEQ(src->seqhi, seq) || 4693 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 4694 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4695 printf("pf: BAD ICMP %d:%d ", 4696 icmptype, pd->hdr.icmp->icmp_code); 4697 pf_print_host(pd->src, 0, pd->af); 4698 printf(" -> "); 4699 pf_print_host(pd->dst, 0, pd->af); 4700 printf(" state: "); 4701 pf_print_state(*state); 4702 printf(" seq=%u\n", seq); 4703 } 4704 REASON_SET(reason, PFRES_BADSTATE); 4705 return (PF_DROP); 4706 } else { 4707 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4708 printf("pf: OK ICMP %d:%d ", 4709 icmptype, pd->hdr.icmp->icmp_code); 4710 pf_print_host(pd->src, 0, pd->af); 4711 printf(" -> "); 4712 pf_print_host(pd->dst, 0, pd->af); 4713 printf(" state: "); 4714 pf_print_state(*state); 4715 printf(" seq=%u\n", seq); 4716 } 4717 } 4718 4719 /* translate source/destination address, if necessary */ 4720 if ((*state)->key[PF_SK_WIRE] != 4721 (*state)->key[PF_SK_STACK]) { 4722 struct pf_state_key *nk = 4723 (*state)->key[pd->didx]; 4724 4725 if (PF_ANEQ(pd2.src, 4726 &nk->addr[pd2.sidx], pd2.af) || 4727 nk->port[pd2.sidx] != th.th_sport) 4728 pf_change_icmp(pd2.src, &th.th_sport, 4729 daddr, &nk->addr[pd2.sidx], 4730 nk->port[pd2.sidx], NULL, 4731 pd2.ip_sum, icmpsum, 4732 pd->ip_sum, 0, pd2.af); 4733 4734 if (PF_ANEQ(pd2.dst, 4735 &nk->addr[pd2.didx], pd2.af) || 4736 nk->port[pd2.didx] != th.th_dport) 4737 pf_change_icmp(pd2.dst, &th.th_dport, 4738 NULL, /* XXX Inbound NAT? */ 4739 &nk->addr[pd2.didx], 4740 nk->port[pd2.didx], NULL, 4741 pd2.ip_sum, icmpsum, 4742 pd->ip_sum, 0, pd2.af); 4743 copyback = 1; 4744 } 4745 4746 if (copyback) { 4747 switch (pd2.af) { 4748 #ifdef INET 4749 case AF_INET: 4750 m_copyback(m, off, ICMP_MINLEN, 4751 (caddr_t )pd->hdr.icmp); 4752 m_copyback(m, ipoff2, sizeof(h2), 4753 (caddr_t )&h2); 4754 break; 4755 #endif /* INET */ 4756 #ifdef INET6 4757 case AF_INET6: 4758 m_copyback(m, off, 4759 sizeof(struct icmp6_hdr), 4760 (caddr_t )pd->hdr.icmp6); 4761 m_copyback(m, ipoff2, sizeof(h2_6), 4762 (caddr_t )&h2_6); 4763 break; 4764 #endif /* INET6 */ 4765 } 4766 m_copyback(m, off2, 8, (caddr_t)&th); 4767 } 4768 4769 return (PF_PASS); 4770 break; 4771 } 4772 case IPPROTO_UDP: { 4773 struct udphdr uh; 4774 4775 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 4776 NULL, reason, pd2.af)) { 4777 DPFPRINTF(PF_DEBUG_MISC, 4778 ("pf: ICMP error message too short " 4779 "(udp)\n")); 4780 return (PF_DROP); 4781 } 4782 4783 key.af = pd2.af; 4784 key.proto = IPPROTO_UDP; 4785 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4786 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4787 key.port[pd2.sidx] = uh.uh_sport; 4788 key.port[pd2.didx] = uh.uh_dport; 4789 4790 STATE_LOOKUP(kif, &key, direction, *state, pd); 4791 4792 /* translate source/destination address, if necessary */ 4793 if ((*state)->key[PF_SK_WIRE] != 4794 (*state)->key[PF_SK_STACK]) { 4795 struct pf_state_key *nk = 4796 (*state)->key[pd->didx]; 4797 4798 if (PF_ANEQ(pd2.src, 4799 &nk->addr[pd2.sidx], pd2.af) || 4800 nk->port[pd2.sidx] != uh.uh_sport) 4801 pf_change_icmp(pd2.src, &uh.uh_sport, 4802 daddr, &nk->addr[pd2.sidx], 4803 nk->port[pd2.sidx], &uh.uh_sum, 4804 pd2.ip_sum, icmpsum, 4805 pd->ip_sum, 1, pd2.af); 4806 4807 if (PF_ANEQ(pd2.dst, 4808 &nk->addr[pd2.didx], pd2.af) || 4809 nk->port[pd2.didx] != uh.uh_dport) 4810 pf_change_icmp(pd2.dst, &uh.uh_dport, 4811 NULL, /* XXX Inbound NAT? */ 4812 &nk->addr[pd2.didx], 4813 nk->port[pd2.didx], &uh.uh_sum, 4814 pd2.ip_sum, icmpsum, 4815 pd->ip_sum, 1, pd2.af); 4816 4817 switch (pd2.af) { 4818 #ifdef INET 4819 case AF_INET: 4820 m_copyback(m, off, ICMP_MINLEN, 4821 (caddr_t )pd->hdr.icmp); 4822 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 4823 break; 4824 #endif /* INET */ 4825 #ifdef INET6 4826 case AF_INET6: 4827 m_copyback(m, off, 4828 sizeof(struct icmp6_hdr), 4829 (caddr_t )pd->hdr.icmp6); 4830 m_copyback(m, ipoff2, sizeof(h2_6), 4831 (caddr_t )&h2_6); 4832 break; 4833 #endif /* INET6 */ 4834 } 4835 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 4836 } 4837 return (PF_PASS); 4838 break; 4839 } 4840 #ifdef INET 4841 case IPPROTO_ICMP: { 4842 struct icmp iih; 4843 4844 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 4845 NULL, reason, pd2.af)) { 4846 DPFPRINTF(PF_DEBUG_MISC, 4847 ("pf: ICMP error message too short i" 4848 "(icmp)\n")); 4849 return (PF_DROP); 4850 } 4851 4852 key.af = pd2.af; 4853 key.proto = IPPROTO_ICMP; 4854 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4855 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4856 key.port[0] = key.port[1] = iih.icmp_id; 4857 4858 STATE_LOOKUP(kif, &key, direction, *state, pd); 4859 4860 /* translate source/destination address, if necessary */ 4861 if ((*state)->key[PF_SK_WIRE] != 4862 (*state)->key[PF_SK_STACK]) { 4863 struct pf_state_key *nk = 4864 (*state)->key[pd->didx]; 4865 4866 if (PF_ANEQ(pd2.src, 4867 &nk->addr[pd2.sidx], pd2.af) || 4868 nk->port[pd2.sidx] != iih.icmp_id) 4869 pf_change_icmp(pd2.src, &iih.icmp_id, 4870 daddr, &nk->addr[pd2.sidx], 4871 nk->port[pd2.sidx], NULL, 4872 pd2.ip_sum, icmpsum, 4873 pd->ip_sum, 0, AF_INET); 4874 4875 if (PF_ANEQ(pd2.dst, 4876 &nk->addr[pd2.didx], pd2.af) || 4877 nk->port[pd2.didx] != iih.icmp_id) 4878 pf_change_icmp(pd2.dst, &iih.icmp_id, 4879 NULL, /* XXX Inbound NAT? */ 4880 &nk->addr[pd2.didx], 4881 nk->port[pd2.didx], NULL, 4882 pd2.ip_sum, icmpsum, 4883 pd->ip_sum, 0, AF_INET); 4884 4885 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 4886 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 4887 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 4888 } 4889 return (PF_PASS); 4890 break; 4891 } 4892 #endif /* INET */ 4893 #ifdef INET6 4894 case IPPROTO_ICMPV6: { 4895 struct icmp6_hdr iih; 4896 4897 if (!pf_pull_hdr(m, off2, &iih, 4898 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 4899 DPFPRINTF(PF_DEBUG_MISC, 4900 ("pf: ICMP error message too short " 4901 "(icmp6)\n")); 4902 return (PF_DROP); 4903 } 4904 4905 key.af = pd2.af; 4906 key.proto = IPPROTO_ICMPV6; 4907 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4908 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4909 key.port[0] = key.port[1] = iih.icmp6_id; 4910 4911 STATE_LOOKUP(kif, &key, direction, *state, pd); 4912 4913 /* translate source/destination address, if necessary */ 4914 if ((*state)->key[PF_SK_WIRE] != 4915 (*state)->key[PF_SK_STACK]) { 4916 struct pf_state_key *nk = 4917 (*state)->key[pd->didx]; 4918 4919 if (PF_ANEQ(pd2.src, 4920 &nk->addr[pd2.sidx], pd2.af) || 4921 nk->port[pd2.sidx] != iih.icmp6_id) 4922 pf_change_icmp(pd2.src, &iih.icmp6_id, 4923 daddr, &nk->addr[pd2.sidx], 4924 nk->port[pd2.sidx], NULL, 4925 pd2.ip_sum, icmpsum, 4926 pd->ip_sum, 0, AF_INET6); 4927 4928 if (PF_ANEQ(pd2.dst, 4929 &nk->addr[pd2.didx], pd2.af) || 4930 nk->port[pd2.didx] != iih.icmp6_id) 4931 pf_change_icmp(pd2.dst, &iih.icmp6_id, 4932 NULL, /* XXX Inbound NAT? */ 4933 &nk->addr[pd2.didx], 4934 nk->port[pd2.didx], NULL, 4935 pd2.ip_sum, icmpsum, 4936 pd->ip_sum, 0, AF_INET6); 4937 4938 m_copyback(m, off, sizeof(struct icmp6_hdr), 4939 (caddr_t)pd->hdr.icmp6); 4940 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 4941 m_copyback(m, off2, sizeof(struct icmp6_hdr), 4942 (caddr_t)&iih); 4943 } 4944 return (PF_PASS); 4945 break; 4946 } 4947 #endif /* INET6 */ 4948 default: { 4949 key.af = pd2.af; 4950 key.proto = pd2.proto; 4951 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4952 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4953 key.port[0] = key.port[1] = 0; 4954 4955 STATE_LOOKUP(kif, &key, direction, *state, pd); 4956 4957 /* translate source/destination address, if necessary */ 4958 if ((*state)->key[PF_SK_WIRE] != 4959 (*state)->key[PF_SK_STACK]) { 4960 struct pf_state_key *nk = 4961 (*state)->key[pd->didx]; 4962 4963 if (PF_ANEQ(pd2.src, 4964 &nk->addr[pd2.sidx], pd2.af)) 4965 pf_change_icmp(pd2.src, NULL, daddr, 4966 &nk->addr[pd2.sidx], 0, NULL, 4967 pd2.ip_sum, icmpsum, 4968 pd->ip_sum, 0, pd2.af); 4969 4970 if (PF_ANEQ(pd2.dst, 4971 &nk->addr[pd2.didx], pd2.af)) 4972 pf_change_icmp(pd2.src, NULL, 4973 NULL, /* XXX Inbound NAT? */ 4974 &nk->addr[pd2.didx], 0, NULL, 4975 pd2.ip_sum, icmpsum, 4976 pd->ip_sum, 0, pd2.af); 4977 4978 switch (pd2.af) { 4979 #ifdef INET 4980 case AF_INET: 4981 m_copyback(m, off, ICMP_MINLEN, 4982 (caddr_t)pd->hdr.icmp); 4983 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 4984 break; 4985 #endif /* INET */ 4986 #ifdef INET6 4987 case AF_INET6: 4988 m_copyback(m, off, 4989 sizeof(struct icmp6_hdr), 4990 (caddr_t )pd->hdr.icmp6); 4991 m_copyback(m, ipoff2, sizeof(h2_6), 4992 (caddr_t )&h2_6); 4993 break; 4994 #endif /* INET6 */ 4995 } 4996 } 4997 return (PF_PASS); 4998 break; 4999 } 5000 } 5001 } 5002 } 5003 5004 static int 5005 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif, 5006 struct mbuf *m, struct pf_pdesc *pd) 5007 { 5008 struct pf_state_peer *src, *dst; 5009 struct pf_state_key_cmp key; 5010 5011 bzero(&key, sizeof(key)); 5012 key.af = pd->af; 5013 key.proto = pd->proto; 5014 if (direction == PF_IN) { 5015 PF_ACPY(&key.addr[0], pd->src, key.af); 5016 PF_ACPY(&key.addr[1], pd->dst, key.af); 5017 key.port[0] = key.port[1] = 0; 5018 } else { 5019 PF_ACPY(&key.addr[1], pd->src, key.af); 5020 PF_ACPY(&key.addr[0], pd->dst, key.af); 5021 key.port[1] = key.port[0] = 0; 5022 } 5023 5024 STATE_LOOKUP(kif, &key, direction, *state, pd); 5025 5026 if (direction == (*state)->direction) { 5027 src = &(*state)->src; 5028 dst = &(*state)->dst; 5029 } else { 5030 src = &(*state)->dst; 5031 dst = &(*state)->src; 5032 } 5033 5034 /* update states */ 5035 if (src->state < PFOTHERS_SINGLE) 5036 src->state = PFOTHERS_SINGLE; 5037 if (dst->state == PFOTHERS_SINGLE) 5038 dst->state = PFOTHERS_MULTIPLE; 5039 5040 /* update expire time */ 5041 (*state)->expire = time_uptime; 5042 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 5043 (*state)->timeout = PFTM_OTHER_MULTIPLE; 5044 else 5045 (*state)->timeout = PFTM_OTHER_SINGLE; 5046 5047 /* translate source/destination address, if necessary */ 5048 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5049 struct pf_state_key *nk = (*state)->key[pd->didx]; 5050 5051 KASSERT(nk, ("%s: nk is null", __func__)); 5052 KASSERT(pd, ("%s: pd is null", __func__)); 5053 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 5054 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 5055 switch (pd->af) { 5056 #ifdef INET 5057 case AF_INET: 5058 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5059 pf_change_a(&pd->src->v4.s_addr, 5060 pd->ip_sum, 5061 nk->addr[pd->sidx].v4.s_addr, 5062 0); 5063 5064 5065 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5066 pf_change_a(&pd->dst->v4.s_addr, 5067 pd->ip_sum, 5068 nk->addr[pd->didx].v4.s_addr, 5069 0); 5070 5071 break; 5072 #endif /* INET */ 5073 #ifdef INET6 5074 case AF_INET6: 5075 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5076 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5077 5078 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5079 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5080 #endif /* INET6 */ 5081 } 5082 } 5083 return (PF_PASS); 5084 } 5085 5086 /* 5087 * ipoff and off are measured from the start of the mbuf chain. 5088 * h must be at "ipoff" on the mbuf chain. 5089 */ 5090 void * 5091 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 5092 u_short *actionp, u_short *reasonp, sa_family_t af) 5093 { 5094 switch (af) { 5095 #ifdef INET 5096 case AF_INET: { 5097 struct ip *h = mtod(m, struct ip *); 5098 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 5099 5100 if (fragoff) { 5101 if (fragoff >= len) 5102 ACTION_SET(actionp, PF_PASS); 5103 else { 5104 ACTION_SET(actionp, PF_DROP); 5105 REASON_SET(reasonp, PFRES_FRAG); 5106 } 5107 return (NULL); 5108 } 5109 if (m->m_pkthdr.len < off + len || 5110 ntohs(h->ip_len) < off + len) { 5111 ACTION_SET(actionp, PF_DROP); 5112 REASON_SET(reasonp, PFRES_SHORT); 5113 return (NULL); 5114 } 5115 break; 5116 } 5117 #endif /* INET */ 5118 #ifdef INET6 5119 case AF_INET6: { 5120 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 5121 5122 if (m->m_pkthdr.len < off + len || 5123 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 5124 (unsigned)(off + len)) { 5125 ACTION_SET(actionp, PF_DROP); 5126 REASON_SET(reasonp, PFRES_SHORT); 5127 return (NULL); 5128 } 5129 break; 5130 } 5131 #endif /* INET6 */ 5132 } 5133 m_copydata(m, off, len, p); 5134 return (p); 5135 } 5136 5137 int 5138 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif, 5139 int rtableid) 5140 { 5141 #ifdef RADIX_MPATH 5142 struct radix_node_head *rnh; 5143 #endif 5144 struct sockaddr_in *dst; 5145 int ret = 1; 5146 int check_mpath; 5147 #ifdef INET6 5148 struct sockaddr_in6 *dst6; 5149 struct route_in6 ro; 5150 #else 5151 struct route ro; 5152 #endif 5153 struct radix_node *rn; 5154 struct rtentry *rt; 5155 struct ifnet *ifp; 5156 5157 check_mpath = 0; 5158 #ifdef RADIX_MPATH 5159 /* XXX: stick to table 0 for now */ 5160 rnh = rt_tables_get_rnh(0, af); 5161 if (rnh != NULL && rn_mpath_capable(rnh)) 5162 check_mpath = 1; 5163 #endif 5164 bzero(&ro, sizeof(ro)); 5165 switch (af) { 5166 case AF_INET: 5167 dst = satosin(&ro.ro_dst); 5168 dst->sin_family = AF_INET; 5169 dst->sin_len = sizeof(*dst); 5170 dst->sin_addr = addr->v4; 5171 break; 5172 #ifdef INET6 5173 case AF_INET6: 5174 /* 5175 * Skip check for addresses with embedded interface scope, 5176 * as they would always match anyway. 5177 */ 5178 if (IN6_IS_SCOPE_EMBED(&addr->v6)) 5179 goto out; 5180 dst6 = (struct sockaddr_in6 *)&ro.ro_dst; 5181 dst6->sin6_family = AF_INET6; 5182 dst6->sin6_len = sizeof(*dst6); 5183 dst6->sin6_addr = addr->v6; 5184 break; 5185 #endif /* INET6 */ 5186 default: 5187 return (0); 5188 } 5189 5190 /* Skip checks for ipsec interfaces */ 5191 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 5192 goto out; 5193 5194 switch (af) { 5195 #ifdef INET6 5196 case AF_INET6: 5197 in6_rtalloc_ign(&ro, 0, rtableid); 5198 break; 5199 #endif 5200 #ifdef INET 5201 case AF_INET: 5202 in_rtalloc_ign((struct route *)&ro, 0, rtableid); 5203 break; 5204 #endif 5205 default: 5206 rtalloc_ign((struct route *)&ro, 0); /* No/default FIB. */ 5207 break; 5208 } 5209 5210 if (ro.ro_rt != NULL) { 5211 /* No interface given, this is a no-route check */ 5212 if (kif == NULL) 5213 goto out; 5214 5215 if (kif->pfik_ifp == NULL) { 5216 ret = 0; 5217 goto out; 5218 } 5219 5220 /* Perform uRPF check if passed input interface */ 5221 ret = 0; 5222 rn = (struct radix_node *)ro.ro_rt; 5223 do { 5224 rt = (struct rtentry *)rn; 5225 ifp = rt->rt_ifp; 5226 5227 if (kif->pfik_ifp == ifp) 5228 ret = 1; 5229 #ifdef RADIX_MPATH 5230 rn = rn_mpath_next(rn); 5231 #endif 5232 } while (check_mpath == 1 && rn != NULL && ret == 0); 5233 } else 5234 ret = 0; 5235 out: 5236 if (ro.ro_rt != NULL) 5237 RTFREE(ro.ro_rt); 5238 return (ret); 5239 } 5240 5241 #ifdef INET 5242 static void 5243 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, 5244 struct pf_state *s, struct pf_pdesc *pd) 5245 { 5246 struct mbuf *m0, *m1; 5247 struct sockaddr_in dst; 5248 struct ip *ip; 5249 struct ifnet *ifp = NULL; 5250 struct pf_addr naddr; 5251 struct pf_src_node *sn = NULL; 5252 int error = 0; 5253 uint16_t ip_len, ip_off; 5254 5255 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 5256 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 5257 __func__)); 5258 5259 if ((pd->pf_mtag == NULL && 5260 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 5261 pd->pf_mtag->routed++ > 3) { 5262 m0 = *m; 5263 *m = NULL; 5264 goto bad_locked; 5265 } 5266 5267 if (r->rt == PF_DUPTO) { 5268 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { 5269 if (s) 5270 PF_STATE_UNLOCK(s); 5271 return; 5272 } 5273 } else { 5274 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 5275 if (s) 5276 PF_STATE_UNLOCK(s); 5277 return; 5278 } 5279 m0 = *m; 5280 } 5281 5282 ip = mtod(m0, struct ip *); 5283 5284 bzero(&dst, sizeof(dst)); 5285 dst.sin_family = AF_INET; 5286 dst.sin_len = sizeof(dst); 5287 dst.sin_addr = ip->ip_dst; 5288 5289 if (r->rt == PF_FASTROUTE) { 5290 struct rtentry *rt; 5291 5292 if (s) 5293 PF_STATE_UNLOCK(s); 5294 rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0)); 5295 if (rt == NULL) { 5296 KMOD_IPSTAT_INC(ips_noroute); 5297 error = EHOSTUNREACH; 5298 goto bad; 5299 } 5300 5301 ifp = rt->rt_ifp; 5302 counter_u64_add(rt->rt_pksent, 1); 5303 5304 if (rt->rt_flags & RTF_GATEWAY) 5305 bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst)); 5306 RTFREE_LOCKED(rt); 5307 } else { 5308 if (TAILQ_EMPTY(&r->rpool.list)) { 5309 DPFPRINTF(PF_DEBUG_URGENT, 5310 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 5311 goto bad_locked; 5312 } 5313 if (s == NULL) { 5314 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 5315 &naddr, NULL, &sn); 5316 if (!PF_AZERO(&naddr, AF_INET)) 5317 dst.sin_addr.s_addr = naddr.v4.s_addr; 5318 ifp = r->rpool.cur->kif ? 5319 r->rpool.cur->kif->pfik_ifp : NULL; 5320 } else { 5321 if (!PF_AZERO(&s->rt_addr, AF_INET)) 5322 dst.sin_addr.s_addr = 5323 s->rt_addr.v4.s_addr; 5324 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5325 PF_STATE_UNLOCK(s); 5326 } 5327 } 5328 if (ifp == NULL) 5329 goto bad; 5330 5331 if (oifp != ifp) { 5332 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS) 5333 goto bad; 5334 else if (m0 == NULL) 5335 goto done; 5336 if (m0->m_len < sizeof(struct ip)) { 5337 DPFPRINTF(PF_DEBUG_URGENT, 5338 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 5339 goto bad; 5340 } 5341 ip = mtod(m0, struct ip *); 5342 } 5343 5344 if (ifp->if_flags & IFF_LOOPBACK) 5345 m0->m_flags |= M_SKIP_FIREWALL; 5346 5347 ip_len = ntohs(ip->ip_len); 5348 ip_off = ntohs(ip->ip_off); 5349 5350 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 5351 m0->m_pkthdr.csum_flags |= CSUM_IP; 5352 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 5353 in_delayed_cksum(m0); 5354 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 5355 } 5356 #ifdef SCTP 5357 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 5358 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 5359 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 5360 } 5361 #endif 5362 5363 /* 5364 * If small enough for interface, or the interface will take 5365 * care of the fragmentation for us, we can just send directly. 5366 */ 5367 if (ip_len <= ifp->if_mtu || 5368 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 5369 ip->ip_sum = 0; 5370 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 5371 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 5372 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 5373 } 5374 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 5375 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); 5376 goto done; 5377 } 5378 5379 /* Balk when DF bit is set or the interface didn't support TSO. */ 5380 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 5381 error = EMSGSIZE; 5382 KMOD_IPSTAT_INC(ips_cantfrag); 5383 if (r->rt != PF_DUPTO) { 5384 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 5385 ifp->if_mtu); 5386 goto done; 5387 } else 5388 goto bad; 5389 } 5390 5391 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 5392 if (error) 5393 goto bad; 5394 5395 for (; m0; m0 = m1) { 5396 m1 = m0->m_nextpkt; 5397 m0->m_nextpkt = NULL; 5398 if (error == 0) { 5399 m_clrprotoflags(m0); 5400 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); 5401 } else 5402 m_freem(m0); 5403 } 5404 5405 if (error == 0) 5406 KMOD_IPSTAT_INC(ips_fragmented); 5407 5408 done: 5409 if (r->rt != PF_DUPTO) 5410 *m = NULL; 5411 return; 5412 5413 bad_locked: 5414 if (s) 5415 PF_STATE_UNLOCK(s); 5416 bad: 5417 m_freem(m0); 5418 goto done; 5419 } 5420 #endif /* INET */ 5421 5422 #ifdef INET6 5423 static void 5424 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, 5425 struct pf_state *s, struct pf_pdesc *pd) 5426 { 5427 struct mbuf *m0; 5428 struct sockaddr_in6 dst; 5429 struct ip6_hdr *ip6; 5430 struct ifnet *ifp = NULL; 5431 struct pf_addr naddr; 5432 struct pf_src_node *sn = NULL; 5433 5434 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 5435 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 5436 __func__)); 5437 5438 if ((pd->pf_mtag == NULL && 5439 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 5440 pd->pf_mtag->routed++ > 3) { 5441 m0 = *m; 5442 *m = NULL; 5443 goto bad_locked; 5444 } 5445 5446 if (r->rt == PF_DUPTO) { 5447 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { 5448 if (s) 5449 PF_STATE_UNLOCK(s); 5450 return; 5451 } 5452 } else { 5453 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 5454 if (s) 5455 PF_STATE_UNLOCK(s); 5456 return; 5457 } 5458 m0 = *m; 5459 } 5460 5461 ip6 = mtod(m0, struct ip6_hdr *); 5462 5463 bzero(&dst, sizeof(dst)); 5464 dst.sin6_family = AF_INET6; 5465 dst.sin6_len = sizeof(dst); 5466 dst.sin6_addr = ip6->ip6_dst; 5467 5468 /* Cheat. XXX why only in the v6 case??? */ 5469 if (r->rt == PF_FASTROUTE) { 5470 if (s) 5471 PF_STATE_UNLOCK(s); 5472 m0->m_flags |= M_SKIP_FIREWALL; 5473 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL); 5474 *m = NULL; 5475 return; 5476 } 5477 5478 if (TAILQ_EMPTY(&r->rpool.list)) { 5479 DPFPRINTF(PF_DEBUG_URGENT, 5480 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 5481 goto bad_locked; 5482 } 5483 if (s == NULL) { 5484 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 5485 &naddr, NULL, &sn); 5486 if (!PF_AZERO(&naddr, AF_INET6)) 5487 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 5488 &naddr, AF_INET6); 5489 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 5490 } else { 5491 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 5492 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 5493 &s->rt_addr, AF_INET6); 5494 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5495 } 5496 5497 if (s) 5498 PF_STATE_UNLOCK(s); 5499 5500 if (ifp == NULL) 5501 goto bad; 5502 5503 if (oifp != ifp) { 5504 if (pf_test6(PF_FWD, ifp, &m0, NULL) != PF_PASS) 5505 goto bad; 5506 else if (m0 == NULL) 5507 goto done; 5508 if (m0->m_len < sizeof(struct ip6_hdr)) { 5509 DPFPRINTF(PF_DEBUG_URGENT, 5510 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 5511 __func__)); 5512 goto bad; 5513 } 5514 ip6 = mtod(m0, struct ip6_hdr *); 5515 } 5516 5517 if (ifp->if_flags & IFF_LOOPBACK) 5518 m0->m_flags |= M_SKIP_FIREWALL; 5519 5520 /* 5521 * If the packet is too large for the outgoing interface, 5522 * send back an icmp6 error. 5523 */ 5524 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 5525 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 5526 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) 5527 nd6_output(ifp, ifp, m0, &dst, NULL); 5528 else { 5529 in6_ifstat_inc(ifp, ifs6_in_toobig); 5530 if (r->rt != PF_DUPTO) 5531 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 5532 else 5533 goto bad; 5534 } 5535 5536 done: 5537 if (r->rt != PF_DUPTO) 5538 *m = NULL; 5539 return; 5540 5541 bad_locked: 5542 if (s) 5543 PF_STATE_UNLOCK(s); 5544 bad: 5545 m_freem(m0); 5546 goto done; 5547 } 5548 #endif /* INET6 */ 5549 5550 /* 5551 * FreeBSD supports cksum offloads for the following drivers. 5552 * em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4), 5553 * ti(4), txp(4), xl(4) 5554 * 5555 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 5556 * network driver performed cksum including pseudo header, need to verify 5557 * csum_data 5558 * CSUM_DATA_VALID : 5559 * network driver performed cksum, needs to additional pseudo header 5560 * cksum computation with partial csum_data(i.e. lack of H/W support for 5561 * pseudo header, for instance hme(4), sk(4) and possibly gem(4)) 5562 * 5563 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 5564 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 5565 * TCP/UDP layer. 5566 * Also, set csum_data to 0xffff to force cksum validation. 5567 */ 5568 static int 5569 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 5570 { 5571 u_int16_t sum = 0; 5572 int hw_assist = 0; 5573 struct ip *ip; 5574 5575 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 5576 return (1); 5577 if (m->m_pkthdr.len < off + len) 5578 return (1); 5579 5580 switch (p) { 5581 case IPPROTO_TCP: 5582 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 5583 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 5584 sum = m->m_pkthdr.csum_data; 5585 } else { 5586 ip = mtod(m, struct ip *); 5587 sum = in_pseudo(ip->ip_src.s_addr, 5588 ip->ip_dst.s_addr, htonl((u_short)len + 5589 m->m_pkthdr.csum_data + IPPROTO_TCP)); 5590 } 5591 sum ^= 0xffff; 5592 ++hw_assist; 5593 } 5594 break; 5595 case IPPROTO_UDP: 5596 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 5597 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 5598 sum = m->m_pkthdr.csum_data; 5599 } else { 5600 ip = mtod(m, struct ip *); 5601 sum = in_pseudo(ip->ip_src.s_addr, 5602 ip->ip_dst.s_addr, htonl((u_short)len + 5603 m->m_pkthdr.csum_data + IPPROTO_UDP)); 5604 } 5605 sum ^= 0xffff; 5606 ++hw_assist; 5607 } 5608 break; 5609 case IPPROTO_ICMP: 5610 #ifdef INET6 5611 case IPPROTO_ICMPV6: 5612 #endif /* INET6 */ 5613 break; 5614 default: 5615 return (1); 5616 } 5617 5618 if (!hw_assist) { 5619 switch (af) { 5620 case AF_INET: 5621 if (p == IPPROTO_ICMP) { 5622 if (m->m_len < off) 5623 return (1); 5624 m->m_data += off; 5625 m->m_len -= off; 5626 sum = in_cksum(m, len); 5627 m->m_data -= off; 5628 m->m_len += off; 5629 } else { 5630 if (m->m_len < sizeof(struct ip)) 5631 return (1); 5632 sum = in4_cksum(m, p, off, len); 5633 } 5634 break; 5635 #ifdef INET6 5636 case AF_INET6: 5637 if (m->m_len < sizeof(struct ip6_hdr)) 5638 return (1); 5639 sum = in6_cksum(m, p, off, len); 5640 break; 5641 #endif /* INET6 */ 5642 default: 5643 return (1); 5644 } 5645 } 5646 if (sum) { 5647 switch (p) { 5648 case IPPROTO_TCP: 5649 { 5650 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 5651 break; 5652 } 5653 case IPPROTO_UDP: 5654 { 5655 KMOD_UDPSTAT_INC(udps_badsum); 5656 break; 5657 } 5658 #ifdef INET 5659 case IPPROTO_ICMP: 5660 { 5661 KMOD_ICMPSTAT_INC(icps_checksum); 5662 break; 5663 } 5664 #endif 5665 #ifdef INET6 5666 case IPPROTO_ICMPV6: 5667 { 5668 KMOD_ICMP6STAT_INC(icp6s_checksum); 5669 break; 5670 } 5671 #endif /* INET6 */ 5672 } 5673 return (1); 5674 } else { 5675 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 5676 m->m_pkthdr.csum_flags |= 5677 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 5678 m->m_pkthdr.csum_data = 0xffff; 5679 } 5680 } 5681 return (0); 5682 } 5683 5684 5685 #ifdef INET 5686 int 5687 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 5688 { 5689 struct pfi_kif *kif; 5690 u_short action, reason = 0, log = 0; 5691 struct mbuf *m = *m0; 5692 struct ip *h = NULL; 5693 struct m_tag *ipfwtag; 5694 struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 5695 struct pf_state *s = NULL; 5696 struct pf_ruleset *ruleset = NULL; 5697 struct pf_pdesc pd; 5698 int off, dirndx, pqid = 0; 5699 5700 M_ASSERTPKTHDR(m); 5701 5702 if (!V_pf_status.running) 5703 return (PF_PASS); 5704 5705 memset(&pd, 0, sizeof(pd)); 5706 5707 kif = (struct pfi_kif *)ifp->if_pf_kif; 5708 5709 if (kif == NULL) { 5710 DPFPRINTF(PF_DEBUG_URGENT, 5711 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 5712 return (PF_DROP); 5713 } 5714 if (kif->pfik_flags & PFI_IFLAG_SKIP) 5715 return (PF_PASS); 5716 5717 if (m->m_flags & M_SKIP_FIREWALL) 5718 return (PF_PASS); 5719 5720 pd.pf_mtag = pf_find_mtag(m); 5721 5722 PF_RULES_RLOCK(); 5723 5724 if (ip_divert_ptr != NULL && 5725 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { 5726 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); 5727 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) { 5728 if (pd.pf_mtag == NULL && 5729 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 5730 action = PF_DROP; 5731 goto done; 5732 } 5733 pd.pf_mtag->flags |= PF_PACKET_LOOPED; 5734 m_tag_delete(m, ipfwtag); 5735 } 5736 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) { 5737 m->m_flags |= M_FASTFWD_OURS; 5738 pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT; 5739 } 5740 } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { 5741 /* We do IP header normalization and packet reassembly here */ 5742 action = PF_DROP; 5743 goto done; 5744 } 5745 m = *m0; /* pf_normalize messes with m0 */ 5746 h = mtod(m, struct ip *); 5747 5748 off = h->ip_hl << 2; 5749 if (off < (int)sizeof(struct ip)) { 5750 action = PF_DROP; 5751 REASON_SET(&reason, PFRES_SHORT); 5752 log = 1; 5753 goto done; 5754 } 5755 5756 pd.src = (struct pf_addr *)&h->ip_src; 5757 pd.dst = (struct pf_addr *)&h->ip_dst; 5758 pd.sport = pd.dport = NULL; 5759 pd.ip_sum = &h->ip_sum; 5760 pd.proto_sum = NULL; 5761 pd.proto = h->ip_p; 5762 pd.dir = dir; 5763 pd.sidx = (dir == PF_IN) ? 0 : 1; 5764 pd.didx = (dir == PF_IN) ? 1 : 0; 5765 pd.af = AF_INET; 5766 pd.tos = h->ip_tos; 5767 pd.tot_len = ntohs(h->ip_len); 5768 5769 /* handle fragments that didn't get reassembled by normalization */ 5770 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 5771 action = pf_test_fragment(&r, dir, kif, m, h, 5772 &pd, &a, &ruleset); 5773 goto done; 5774 } 5775 5776 switch (h->ip_p) { 5777 5778 case IPPROTO_TCP: { 5779 struct tcphdr th; 5780 5781 pd.hdr.tcp = &th; 5782 if (!pf_pull_hdr(m, off, &th, sizeof(th), 5783 &action, &reason, AF_INET)) { 5784 log = action != PF_PASS; 5785 goto done; 5786 } 5787 pd.p_len = pd.tot_len - off - (th.th_off << 2); 5788 if ((th.th_flags & TH_ACK) && pd.p_len == 0) 5789 pqid = 1; 5790 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 5791 if (action == PF_DROP) 5792 goto done; 5793 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 5794 &reason); 5795 if (action == PF_PASS) { 5796 if (pfsync_update_state_ptr != NULL) 5797 pfsync_update_state_ptr(s); 5798 r = s->rule.ptr; 5799 a = s->anchor.ptr; 5800 log = s->log; 5801 } else if (s == NULL) 5802 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5803 &a, &ruleset, inp); 5804 break; 5805 } 5806 5807 case IPPROTO_UDP: { 5808 struct udphdr uh; 5809 5810 pd.hdr.udp = &uh; 5811 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 5812 &action, &reason, AF_INET)) { 5813 log = action != PF_PASS; 5814 goto done; 5815 } 5816 if (uh.uh_dport == 0 || 5817 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 5818 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 5819 action = PF_DROP; 5820 REASON_SET(&reason, PFRES_SHORT); 5821 goto done; 5822 } 5823 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 5824 if (action == PF_PASS) { 5825 if (pfsync_update_state_ptr != NULL) 5826 pfsync_update_state_ptr(s); 5827 r = s->rule.ptr; 5828 a = s->anchor.ptr; 5829 log = s->log; 5830 } else if (s == NULL) 5831 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5832 &a, &ruleset, inp); 5833 break; 5834 } 5835 5836 case IPPROTO_ICMP: { 5837 struct icmp ih; 5838 5839 pd.hdr.icmp = &ih; 5840 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN, 5841 &action, &reason, AF_INET)) { 5842 log = action != PF_PASS; 5843 goto done; 5844 } 5845 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, 5846 &reason); 5847 if (action == PF_PASS) { 5848 if (pfsync_update_state_ptr != NULL) 5849 pfsync_update_state_ptr(s); 5850 r = s->rule.ptr; 5851 a = s->anchor.ptr; 5852 log = s->log; 5853 } else if (s == NULL) 5854 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5855 &a, &ruleset, inp); 5856 break; 5857 } 5858 5859 #ifdef INET6 5860 case IPPROTO_ICMPV6: { 5861 action = PF_DROP; 5862 DPFPRINTF(PF_DEBUG_MISC, 5863 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 5864 goto done; 5865 } 5866 #endif 5867 5868 default: 5869 action = pf_test_state_other(&s, dir, kif, m, &pd); 5870 if (action == PF_PASS) { 5871 if (pfsync_update_state_ptr != NULL) 5872 pfsync_update_state_ptr(s); 5873 r = s->rule.ptr; 5874 a = s->anchor.ptr; 5875 log = s->log; 5876 } else if (s == NULL) 5877 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5878 &a, &ruleset, inp); 5879 break; 5880 } 5881 5882 done: 5883 PF_RULES_RUNLOCK(); 5884 if (action == PF_PASS && h->ip_hl > 5 && 5885 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 5886 action = PF_DROP; 5887 REASON_SET(&reason, PFRES_IPOPTIONS); 5888 log = 1; 5889 DPFPRINTF(PF_DEBUG_MISC, 5890 ("pf: dropping packet with ip options\n")); 5891 } 5892 5893 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 5894 action = PF_DROP; 5895 REASON_SET(&reason, PFRES_MEMORY); 5896 } 5897 if (r->rtableid >= 0) 5898 M_SETFIB(m, r->rtableid); 5899 5900 #ifdef ALTQ 5901 if (action == PF_PASS && r->qid) { 5902 if (pd.pf_mtag == NULL && 5903 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 5904 action = PF_DROP; 5905 REASON_SET(&reason, PFRES_MEMORY); 5906 } 5907 if (pqid || (pd.tos & IPTOS_LOWDELAY)) 5908 pd.pf_mtag->qid = r->pqid; 5909 else 5910 pd.pf_mtag->qid = r->qid; 5911 /* add hints for ecn */ 5912 pd.pf_mtag->hdr = h; 5913 5914 } 5915 #endif /* ALTQ */ 5916 5917 /* 5918 * connections redirected to loopback should not match sockets 5919 * bound specifically to loopback due to security implications, 5920 * see tcp_input() and in_pcblookup_listen(). 5921 */ 5922 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 5923 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 5924 (s->nat_rule.ptr->action == PF_RDR || 5925 s->nat_rule.ptr->action == PF_BINAT) && 5926 (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) 5927 m->m_flags |= M_SKIP_FIREWALL; 5928 5929 if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL && 5930 !PACKET_LOOPED(&pd)) { 5931 5932 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 5933 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 5934 if (ipfwtag != NULL) { 5935 ((struct ipfw_rule_ref *)(ipfwtag+1))->info = 5936 ntohs(r->divert.port); 5937 ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir; 5938 5939 if (s) 5940 PF_STATE_UNLOCK(s); 5941 5942 m_tag_prepend(m, ipfwtag); 5943 if (m->m_flags & M_FASTFWD_OURS) { 5944 if (pd.pf_mtag == NULL && 5945 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 5946 action = PF_DROP; 5947 REASON_SET(&reason, PFRES_MEMORY); 5948 log = 1; 5949 DPFPRINTF(PF_DEBUG_MISC, 5950 ("pf: failed to allocate tag\n")); 5951 } 5952 pd.pf_mtag->flags |= PF_FASTFWD_OURS_PRESENT; 5953 m->m_flags &= ~M_FASTFWD_OURS; 5954 } 5955 ip_divert_ptr(*m0, dir == PF_IN ? DIR_IN : DIR_OUT); 5956 *m0 = NULL; 5957 5958 return (action); 5959 } else { 5960 /* XXX: ipfw has the same behaviour! */ 5961 action = PF_DROP; 5962 REASON_SET(&reason, PFRES_MEMORY); 5963 log = 1; 5964 DPFPRINTF(PF_DEBUG_MISC, 5965 ("pf: failed to allocate divert tag\n")); 5966 } 5967 } 5968 5969 if (log) { 5970 struct pf_rule *lr; 5971 5972 if (s != NULL && s->nat_rule.ptr != NULL && 5973 s->nat_rule.ptr->log & PF_LOG_ALL) 5974 lr = s->nat_rule.ptr; 5975 else 5976 lr = r; 5977 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd, 5978 (s == NULL)); 5979 } 5980 5981 kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len; 5982 kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++; 5983 5984 if (action == PF_PASS || r->action == PF_DROP) { 5985 dirndx = (dir == PF_OUT); 5986 r->packets[dirndx]++; 5987 r->bytes[dirndx] += pd.tot_len; 5988 if (a != NULL) { 5989 a->packets[dirndx]++; 5990 a->bytes[dirndx] += pd.tot_len; 5991 } 5992 if (s != NULL) { 5993 if (s->nat_rule.ptr != NULL) { 5994 s->nat_rule.ptr->packets[dirndx]++; 5995 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; 5996 } 5997 if (s->src_node != NULL) { 5998 s->src_node->packets[dirndx]++; 5999 s->src_node->bytes[dirndx] += pd.tot_len; 6000 } 6001 if (s->nat_src_node != NULL) { 6002 s->nat_src_node->packets[dirndx]++; 6003 s->nat_src_node->bytes[dirndx] += pd.tot_len; 6004 } 6005 dirndx = (dir == s->direction) ? 0 : 1; 6006 s->packets[dirndx]++; 6007 s->bytes[dirndx] += pd.tot_len; 6008 } 6009 tr = r; 6010 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 6011 if (nr != NULL && r == &V_pf_default_rule) 6012 tr = nr; 6013 if (tr->src.addr.type == PF_ADDR_TABLE) 6014 pfr_update_stats(tr->src.addr.p.tbl, 6015 (s == NULL) ? pd.src : 6016 &s->key[(s->direction == PF_IN)]-> 6017 addr[(s->direction == PF_OUT)], 6018 pd.af, pd.tot_len, dir == PF_OUT, 6019 r->action == PF_PASS, tr->src.neg); 6020 if (tr->dst.addr.type == PF_ADDR_TABLE) 6021 pfr_update_stats(tr->dst.addr.p.tbl, 6022 (s == NULL) ? pd.dst : 6023 &s->key[(s->direction == PF_IN)]-> 6024 addr[(s->direction == PF_IN)], 6025 pd.af, pd.tot_len, dir == PF_OUT, 6026 r->action == PF_PASS, tr->dst.neg); 6027 } 6028 6029 switch (action) { 6030 case PF_SYNPROXY_DROP: 6031 m_freem(*m0); 6032 case PF_DEFER: 6033 *m0 = NULL; 6034 action = PF_PASS; 6035 break; 6036 case PF_DROP: 6037 m_freem(*m0); 6038 *m0 = NULL; 6039 break; 6040 default: 6041 /* pf_route() returns unlocked. */ 6042 if (r->rt) { 6043 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd); 6044 return (action); 6045 } 6046 break; 6047 } 6048 if (s) 6049 PF_STATE_UNLOCK(s); 6050 6051 return (action); 6052 } 6053 #endif /* INET */ 6054 6055 #ifdef INET6 6056 int 6057 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 6058 { 6059 struct pfi_kif *kif; 6060 u_short action, reason = 0, log = 0; 6061 struct mbuf *m = *m0, *n = NULL; 6062 struct m_tag *mtag; 6063 struct ip6_hdr *h = NULL; 6064 struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 6065 struct pf_state *s = NULL; 6066 struct pf_ruleset *ruleset = NULL; 6067 struct pf_pdesc pd; 6068 int off, terminal = 0, dirndx, rh_cnt = 0; 6069 int fwdir = dir; 6070 6071 M_ASSERTPKTHDR(m); 6072 6073 if (dir == PF_OUT && m->m_pkthdr.rcvif && ifp != m->m_pkthdr.rcvif) 6074 fwdir = PF_FWD; 6075 6076 if (!V_pf_status.running) 6077 return (PF_PASS); 6078 6079 memset(&pd, 0, sizeof(pd)); 6080 pd.pf_mtag = pf_find_mtag(m); 6081 6082 if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED) 6083 return (PF_PASS); 6084 6085 kif = (struct pfi_kif *)ifp->if_pf_kif; 6086 if (kif == NULL) { 6087 DPFPRINTF(PF_DEBUG_URGENT, 6088 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 6089 return (PF_DROP); 6090 } 6091 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6092 return (PF_PASS); 6093 6094 if (m->m_flags & M_SKIP_FIREWALL) 6095 return (PF_PASS); 6096 6097 PF_RULES_RLOCK(); 6098 6099 /* We do IP header normalization and packet reassembly here */ 6100 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { 6101 action = PF_DROP; 6102 goto done; 6103 } 6104 m = *m0; /* pf_normalize messes with m0 */ 6105 h = mtod(m, struct ip6_hdr *); 6106 6107 #if 1 6108 /* 6109 * we do not support jumbogram yet. if we keep going, zero ip6_plen 6110 * will do something bad, so drop the packet for now. 6111 */ 6112 if (htons(h->ip6_plen) == 0) { 6113 action = PF_DROP; 6114 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 6115 goto done; 6116 } 6117 #endif 6118 6119 pd.src = (struct pf_addr *)&h->ip6_src; 6120 pd.dst = (struct pf_addr *)&h->ip6_dst; 6121 pd.sport = pd.dport = NULL; 6122 pd.ip_sum = NULL; 6123 pd.proto_sum = NULL; 6124 pd.dir = dir; 6125 pd.sidx = (dir == PF_IN) ? 0 : 1; 6126 pd.didx = (dir == PF_IN) ? 1 : 0; 6127 pd.af = AF_INET6; 6128 pd.tos = 0; 6129 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 6130 6131 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 6132 pd.proto = h->ip6_nxt; 6133 do { 6134 switch (pd.proto) { 6135 case IPPROTO_FRAGMENT: 6136 action = pf_test_fragment(&r, dir, kif, m, h, 6137 &pd, &a, &ruleset); 6138 if (action == PF_DROP) 6139 REASON_SET(&reason, PFRES_FRAG); 6140 goto done; 6141 case IPPROTO_ROUTING: { 6142 struct ip6_rthdr rthdr; 6143 6144 if (rh_cnt++) { 6145 DPFPRINTF(PF_DEBUG_MISC, 6146 ("pf: IPv6 more than one rthdr\n")); 6147 action = PF_DROP; 6148 REASON_SET(&reason, PFRES_IPOPTIONS); 6149 log = 1; 6150 goto done; 6151 } 6152 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 6153 &reason, pd.af)) { 6154 DPFPRINTF(PF_DEBUG_MISC, 6155 ("pf: IPv6 short rthdr\n")); 6156 action = PF_DROP; 6157 REASON_SET(&reason, PFRES_SHORT); 6158 log = 1; 6159 goto done; 6160 } 6161 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 6162 DPFPRINTF(PF_DEBUG_MISC, 6163 ("pf: IPv6 rthdr0\n")); 6164 action = PF_DROP; 6165 REASON_SET(&reason, PFRES_IPOPTIONS); 6166 log = 1; 6167 goto done; 6168 } 6169 /* FALLTHROUGH */ 6170 } 6171 case IPPROTO_AH: 6172 case IPPROTO_HOPOPTS: 6173 case IPPROTO_DSTOPTS: { 6174 /* get next header and header length */ 6175 struct ip6_ext opt6; 6176 6177 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 6178 NULL, &reason, pd.af)) { 6179 DPFPRINTF(PF_DEBUG_MISC, 6180 ("pf: IPv6 short opt\n")); 6181 action = PF_DROP; 6182 log = 1; 6183 goto done; 6184 } 6185 if (pd.proto == IPPROTO_AH) 6186 off += (opt6.ip6e_len + 2) * 4; 6187 else 6188 off += (opt6.ip6e_len + 1) * 8; 6189 pd.proto = opt6.ip6e_nxt; 6190 /* goto the next header */ 6191 break; 6192 } 6193 default: 6194 terminal++; 6195 break; 6196 } 6197 } while (!terminal); 6198 6199 /* if there's no routing header, use unmodified mbuf for checksumming */ 6200 if (!n) 6201 n = m; 6202 6203 switch (pd.proto) { 6204 6205 case IPPROTO_TCP: { 6206 struct tcphdr th; 6207 6208 pd.hdr.tcp = &th; 6209 if (!pf_pull_hdr(m, off, &th, sizeof(th), 6210 &action, &reason, AF_INET6)) { 6211 log = action != PF_PASS; 6212 goto done; 6213 } 6214 pd.p_len = pd.tot_len - off - (th.th_off << 2); 6215 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6216 if (action == PF_DROP) 6217 goto done; 6218 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6219 &reason); 6220 if (action == PF_PASS) { 6221 if (pfsync_update_state_ptr != NULL) 6222 pfsync_update_state_ptr(s); 6223 r = s->rule.ptr; 6224 a = s->anchor.ptr; 6225 log = s->log; 6226 } else if (s == NULL) 6227 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6228 &a, &ruleset, inp); 6229 break; 6230 } 6231 6232 case IPPROTO_UDP: { 6233 struct udphdr uh; 6234 6235 pd.hdr.udp = &uh; 6236 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 6237 &action, &reason, AF_INET6)) { 6238 log = action != PF_PASS; 6239 goto done; 6240 } 6241 if (uh.uh_dport == 0 || 6242 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 6243 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 6244 action = PF_DROP; 6245 REASON_SET(&reason, PFRES_SHORT); 6246 goto done; 6247 } 6248 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 6249 if (action == PF_PASS) { 6250 if (pfsync_update_state_ptr != NULL) 6251 pfsync_update_state_ptr(s); 6252 r = s->rule.ptr; 6253 a = s->anchor.ptr; 6254 log = s->log; 6255 } else if (s == NULL) 6256 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6257 &a, &ruleset, inp); 6258 break; 6259 } 6260 6261 case IPPROTO_ICMP: { 6262 action = PF_DROP; 6263 DPFPRINTF(PF_DEBUG_MISC, 6264 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 6265 goto done; 6266 } 6267 6268 case IPPROTO_ICMPV6: { 6269 struct icmp6_hdr ih; 6270 6271 pd.hdr.icmp6 = &ih; 6272 if (!pf_pull_hdr(m, off, &ih, sizeof(ih), 6273 &action, &reason, AF_INET6)) { 6274 log = action != PF_PASS; 6275 goto done; 6276 } 6277 action = pf_test_state_icmp(&s, dir, kif, 6278 m, off, h, &pd, &reason); 6279 if (action == PF_PASS) { 6280 if (pfsync_update_state_ptr != NULL) 6281 pfsync_update_state_ptr(s); 6282 r = s->rule.ptr; 6283 a = s->anchor.ptr; 6284 log = s->log; 6285 } else if (s == NULL) 6286 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6287 &a, &ruleset, inp); 6288 break; 6289 } 6290 6291 default: 6292 action = pf_test_state_other(&s, dir, kif, m, &pd); 6293 if (action == PF_PASS) { 6294 if (pfsync_update_state_ptr != NULL) 6295 pfsync_update_state_ptr(s); 6296 r = s->rule.ptr; 6297 a = s->anchor.ptr; 6298 log = s->log; 6299 } else if (s == NULL) 6300 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6301 &a, &ruleset, inp); 6302 break; 6303 } 6304 6305 done: 6306 PF_RULES_RUNLOCK(); 6307 if (n != m) { 6308 m_freem(n); 6309 n = NULL; 6310 } 6311 6312 /* handle dangerous IPv6 extension headers. */ 6313 if (action == PF_PASS && rh_cnt && 6314 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 6315 action = PF_DROP; 6316 REASON_SET(&reason, PFRES_IPOPTIONS); 6317 log = 1; 6318 DPFPRINTF(PF_DEBUG_MISC, 6319 ("pf: dropping packet with dangerous v6 headers\n")); 6320 } 6321 6322 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 6323 action = PF_DROP; 6324 REASON_SET(&reason, PFRES_MEMORY); 6325 } 6326 if (r->rtableid >= 0) 6327 M_SETFIB(m, r->rtableid); 6328 6329 #ifdef ALTQ 6330 if (action == PF_PASS && r->qid) { 6331 if (pd.pf_mtag == NULL && 6332 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6333 action = PF_DROP; 6334 REASON_SET(&reason, PFRES_MEMORY); 6335 } 6336 if (pd.tos & IPTOS_LOWDELAY) 6337 pd.pf_mtag->qid = r->pqid; 6338 else 6339 pd.pf_mtag->qid = r->qid; 6340 /* add hints for ecn */ 6341 pd.pf_mtag->hdr = h; 6342 } 6343 #endif /* ALTQ */ 6344 6345 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 6346 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 6347 (s->nat_rule.ptr->action == PF_RDR || 6348 s->nat_rule.ptr->action == PF_BINAT) && 6349 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 6350 m->m_flags |= M_SKIP_FIREWALL; 6351 6352 /* XXX: Anybody working on it?! */ 6353 if (r->divert.port) 6354 printf("pf: divert(9) is not supported for IPv6\n"); 6355 6356 if (log) { 6357 struct pf_rule *lr; 6358 6359 if (s != NULL && s->nat_rule.ptr != NULL && 6360 s->nat_rule.ptr->log & PF_LOG_ALL) 6361 lr = s->nat_rule.ptr; 6362 else 6363 lr = r; 6364 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset, 6365 &pd, (s == NULL)); 6366 } 6367 6368 kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len; 6369 kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++; 6370 6371 if (action == PF_PASS || r->action == PF_DROP) { 6372 dirndx = (dir == PF_OUT); 6373 r->packets[dirndx]++; 6374 r->bytes[dirndx] += pd.tot_len; 6375 if (a != NULL) { 6376 a->packets[dirndx]++; 6377 a->bytes[dirndx] += pd.tot_len; 6378 } 6379 if (s != NULL) { 6380 if (s->nat_rule.ptr != NULL) { 6381 s->nat_rule.ptr->packets[dirndx]++; 6382 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; 6383 } 6384 if (s->src_node != NULL) { 6385 s->src_node->packets[dirndx]++; 6386 s->src_node->bytes[dirndx] += pd.tot_len; 6387 } 6388 if (s->nat_src_node != NULL) { 6389 s->nat_src_node->packets[dirndx]++; 6390 s->nat_src_node->bytes[dirndx] += pd.tot_len; 6391 } 6392 dirndx = (dir == s->direction) ? 0 : 1; 6393 s->packets[dirndx]++; 6394 s->bytes[dirndx] += pd.tot_len; 6395 } 6396 tr = r; 6397 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 6398 if (nr != NULL && r == &V_pf_default_rule) 6399 tr = nr; 6400 if (tr->src.addr.type == PF_ADDR_TABLE) 6401 pfr_update_stats(tr->src.addr.p.tbl, 6402 (s == NULL) ? pd.src : 6403 &s->key[(s->direction == PF_IN)]->addr[0], 6404 pd.af, pd.tot_len, dir == PF_OUT, 6405 r->action == PF_PASS, tr->src.neg); 6406 if (tr->dst.addr.type == PF_ADDR_TABLE) 6407 pfr_update_stats(tr->dst.addr.p.tbl, 6408 (s == NULL) ? pd.dst : 6409 &s->key[(s->direction == PF_IN)]->addr[1], 6410 pd.af, pd.tot_len, dir == PF_OUT, 6411 r->action == PF_PASS, tr->dst.neg); 6412 } 6413 6414 switch (action) { 6415 case PF_SYNPROXY_DROP: 6416 m_freem(*m0); 6417 case PF_DEFER: 6418 *m0 = NULL; 6419 action = PF_PASS; 6420 break; 6421 case PF_DROP: 6422 m_freem(*m0); 6423 *m0 = NULL; 6424 break; 6425 default: 6426 /* pf_route6() returns unlocked. */ 6427 if (r->rt) { 6428 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd); 6429 return (action); 6430 } 6431 break; 6432 } 6433 6434 if (s) 6435 PF_STATE_UNLOCK(s); 6436 6437 /* If reassembled packet passed, create new fragments. */ 6438 if (action == PF_PASS && *m0 && fwdir == PF_FWD && 6439 (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL) 6440 action = pf_refragment6(ifp, m0, mtag); 6441 6442 return (action); 6443 } 6444 #endif /* INET6 */ 6445