1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD$"); 42 43 #include "opt_bpf.h" 44 #include "opt_inet.h" 45 #include "opt_inet6.h" 46 #include "opt_pf.h" 47 #include "opt_sctp.h" 48 49 #include <sys/param.h> 50 #include <sys/bus.h> 51 #include <sys/endian.h> 52 #include <sys/gsb_crc32.h> 53 #include <sys/hash.h> 54 #include <sys/interrupt.h> 55 #include <sys/kernel.h> 56 #include <sys/kthread.h> 57 #include <sys/limits.h> 58 #include <sys/mbuf.h> 59 #include <sys/md5.h> 60 #include <sys/random.h> 61 #include <sys/refcount.h> 62 #include <sys/socket.h> 63 #include <sys/sysctl.h> 64 #include <sys/taskqueue.h> 65 #include <sys/ucred.h> 66 67 #include <net/if.h> 68 #include <net/if_var.h> 69 #include <net/if_types.h> 70 #include <net/if_vlan_var.h> 71 #include <net/route.h> 72 #include <net/route/nhop.h> 73 #include <net/vnet.h> 74 75 #include <net/pfil.h> 76 #include <net/pfvar.h> 77 #include <net/if_pflog.h> 78 #include <net/if_pfsync.h> 79 80 #include <netinet/in_pcb.h> 81 #include <netinet/in_var.h> 82 #include <netinet/in_fib.h> 83 #include <netinet/ip.h> 84 #include <netinet/ip_fw.h> 85 #include <netinet/ip_icmp.h> 86 #include <netinet/icmp_var.h> 87 #include <netinet/ip_var.h> 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/udp.h> 94 #include <netinet/udp_var.h> 95 96 #ifdef INET6 97 #include <netinet/ip6.h> 98 #include <netinet/icmp6.h> 99 #include <netinet6/nd6.h> 100 #include <netinet6/ip6_var.h> 101 #include <netinet6/in6_pcb.h> 102 #include <netinet6/in6_fib.h> 103 #include <netinet6/scope6_var.h> 104 #endif /* INET6 */ 105 106 #if defined(SCTP) || defined(SCTP_SUPPORT) 107 #include <netinet/sctp_crc32.h> 108 #endif 109 110 #include <machine/in_cksum.h> 111 #include <security/mac/mac_framework.h> 112 113 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 114 115 /* 116 * Global variables 117 */ 118 119 /* state tables */ 120 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 121 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 122 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 123 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 124 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 125 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 126 VNET_DEFINE(struct pf_kstatus, pf_status); 127 128 VNET_DEFINE(u_int32_t, ticket_altqs_active); 129 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 130 VNET_DEFINE(int, altqs_inactive_open); 131 VNET_DEFINE(u_int32_t, ticket_pabuf); 132 133 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 134 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 135 VNET_DEFINE(u_char, pf_tcp_secret[16]); 136 #define V_pf_tcp_secret VNET(pf_tcp_secret) 137 VNET_DEFINE(int, pf_tcp_secret_init); 138 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 139 VNET_DEFINE(int, pf_tcp_iss_off); 140 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 141 VNET_DECLARE(int, pf_vnet_active); 142 #define V_pf_vnet_active VNET(pf_vnet_active) 143 144 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 145 #define V_pf_purge_idx VNET(pf_purge_idx) 146 147 /* 148 * Queue for pf_intr() sends. 149 */ 150 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 151 struct pf_send_entry { 152 STAILQ_ENTRY(pf_send_entry) pfse_next; 153 struct mbuf *pfse_m; 154 enum { 155 PFSE_IP, 156 PFSE_IP6, 157 PFSE_ICMP, 158 PFSE_ICMP6, 159 } pfse_type; 160 struct { 161 int type; 162 int code; 163 int mtu; 164 } icmpopts; 165 }; 166 167 STAILQ_HEAD(pf_send_head, pf_send_entry); 168 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 169 #define V_pf_sendqueue VNET(pf_sendqueue) 170 171 static struct mtx pf_sendqueue_mtx; 172 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 173 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 174 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 175 176 /* 177 * Queue for pf_overload_task() tasks. 178 */ 179 struct pf_overload_entry { 180 SLIST_ENTRY(pf_overload_entry) next; 181 struct pf_addr addr; 182 sa_family_t af; 183 uint8_t dir; 184 struct pf_krule *rule; 185 }; 186 187 SLIST_HEAD(pf_overload_head, pf_overload_entry); 188 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 189 #define V_pf_overloadqueue VNET(pf_overloadqueue) 190 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 191 #define V_pf_overloadtask VNET(pf_overloadtask) 192 193 static struct mtx pf_overloadqueue_mtx; 194 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 195 "pf overload/flush queue", MTX_DEF); 196 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 197 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 198 199 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 200 struct mtx pf_unlnkdrules_mtx; 201 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 202 MTX_DEF); 203 204 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 205 #define V_pf_sources_z VNET(pf_sources_z) 206 uma_zone_t pf_mtag_z; 207 VNET_DEFINE(uma_zone_t, pf_state_z); 208 VNET_DEFINE(uma_zone_t, pf_state_key_z); 209 210 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]); 211 #define PFID_CPUBITS 8 212 #define PFID_CPUSHIFT (sizeof(uint64_t) * NBBY - PFID_CPUBITS) 213 #define PFID_CPUMASK ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT) 214 #define PFID_MAXID (~PFID_CPUMASK) 215 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU); 216 217 static void pf_src_tree_remove_state(struct pf_state *); 218 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 219 u_int32_t); 220 static void pf_add_threshold(struct pf_threshold *); 221 static int pf_check_threshold(struct pf_threshold *); 222 223 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 224 u_int16_t *, u_int16_t *, struct pf_addr *, 225 u_int16_t, u_int8_t, sa_family_t); 226 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 227 struct tcphdr *, struct pf_state_peer *); 228 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 229 struct pf_addr *, struct pf_addr *, u_int16_t, 230 u_int16_t *, u_int16_t *, u_int16_t *, 231 u_int16_t *, u_int8_t, sa_family_t); 232 static void pf_send_tcp(struct mbuf *, 233 const struct pf_krule *, sa_family_t, 234 const struct pf_addr *, const struct pf_addr *, 235 u_int16_t, u_int16_t, u_int32_t, u_int32_t, 236 u_int8_t, u_int16_t, u_int16_t, u_int8_t, int, 237 u_int16_t, struct ifnet *); 238 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 239 sa_family_t, struct pf_krule *); 240 static void pf_detach_state(struct pf_state *); 241 static int pf_state_key_attach(struct pf_state_key *, 242 struct pf_state_key *, struct pf_state *); 243 static void pf_state_key_detach(struct pf_state *, int); 244 static int pf_state_key_ctor(void *, int, void *, int); 245 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 246 static int pf_test_rule(struct pf_krule **, struct pf_state **, 247 int, struct pfi_kkif *, struct mbuf *, int, 248 struct pf_pdesc *, struct pf_krule **, 249 struct pf_kruleset **, struct inpcb *); 250 static int pf_create_state(struct pf_krule *, struct pf_krule *, 251 struct pf_krule *, struct pf_pdesc *, 252 struct pf_ksrc_node *, struct pf_state_key *, 253 struct pf_state_key *, struct mbuf *, int, 254 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 255 struct pf_state **, int, u_int16_t, u_int16_t, 256 int); 257 static int pf_test_fragment(struct pf_krule **, int, 258 struct pfi_kkif *, struct mbuf *, void *, 259 struct pf_pdesc *, struct pf_krule **, 260 struct pf_kruleset **); 261 static int pf_tcp_track_full(struct pf_state_peer *, 262 struct pf_state_peer *, struct pf_state **, 263 struct pfi_kkif *, struct mbuf *, int, 264 struct pf_pdesc *, u_short *, int *); 265 static int pf_tcp_track_sloppy(struct pf_state_peer *, 266 struct pf_state_peer *, struct pf_state **, 267 struct pf_pdesc *, u_short *); 268 static int pf_test_state_tcp(struct pf_state **, int, 269 struct pfi_kkif *, struct mbuf *, int, 270 void *, struct pf_pdesc *, u_short *); 271 static int pf_test_state_udp(struct pf_state **, int, 272 struct pfi_kkif *, struct mbuf *, int, 273 void *, struct pf_pdesc *); 274 static int pf_test_state_icmp(struct pf_state **, int, 275 struct pfi_kkif *, struct mbuf *, int, 276 void *, struct pf_pdesc *, u_short *); 277 static int pf_test_state_other(struct pf_state **, int, 278 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 279 static u_int8_t pf_get_wscale(struct mbuf *, int, u_int16_t, 280 sa_family_t); 281 static u_int16_t pf_get_mss(struct mbuf *, int, u_int16_t, 282 sa_family_t); 283 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 284 int, u_int16_t); 285 static int pf_check_proto_cksum(struct mbuf *, int, int, 286 u_int8_t, sa_family_t); 287 static void pf_print_state_parts(struct pf_state *, 288 struct pf_state_key *, struct pf_state_key *); 289 static int pf_addr_wrap_neq(struct pf_addr_wrap *, 290 struct pf_addr_wrap *); 291 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 292 bool, u_int8_t); 293 static struct pf_state *pf_find_state(struct pfi_kkif *, 294 struct pf_state_key_cmp *, u_int); 295 static int pf_src_connlimit(struct pf_state **); 296 static void pf_overload_task(void *v, int pending); 297 static int pf_insert_src_node(struct pf_ksrc_node **, 298 struct pf_krule *, struct pf_addr *, sa_family_t); 299 static u_int pf_purge_expired_states(u_int, int); 300 static void pf_purge_unlinked_rules(void); 301 static int pf_mtag_uminit(void *, int, int); 302 static void pf_mtag_free(struct m_tag *); 303 #ifdef INET 304 static void pf_route(struct mbuf **, struct pf_krule *, int, 305 struct ifnet *, struct pf_state *, 306 struct pf_pdesc *, struct inpcb *); 307 #endif /* INET */ 308 #ifdef INET6 309 static void pf_change_a6(struct pf_addr *, u_int16_t *, 310 struct pf_addr *, u_int8_t); 311 static void pf_route6(struct mbuf **, struct pf_krule *, int, 312 struct ifnet *, struct pf_state *, 313 struct pf_pdesc *, struct inpcb *); 314 #endif /* INET6 */ 315 316 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 317 318 extern int pf_end_threads; 319 extern struct proc *pf_purge_proc; 320 321 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 322 323 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 324 (pd)->pf_mtag->flags & PF_PACKET_LOOPED) 325 326 #define STATE_LOOKUP(i, k, d, s, pd) \ 327 do { \ 328 (s) = pf_find_state((i), (k), (d)); \ 329 if ((s) == NULL) \ 330 return (PF_DROP); \ 331 if (PACKET_LOOPED(pd)) \ 332 return (PF_PASS); \ 333 if ((d) == PF_OUT && \ 334 (((s)->rule.ptr->rt == PF_ROUTETO && \ 335 (s)->rule.ptr->direction == PF_OUT) || \ 336 ((s)->rule.ptr->rt == PF_REPLYTO && \ 337 (s)->rule.ptr->direction == PF_IN)) && \ 338 (s)->rt_kif != NULL && \ 339 (s)->rt_kif != (i)) \ 340 return (PF_PASS); \ 341 } while (0) 342 343 #define BOUND_IFACE(r, k) \ 344 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all 345 346 #define STATE_INC_COUNTERS(s) \ 347 do { \ 348 counter_u64_add(s->rule.ptr->states_cur, 1); \ 349 counter_u64_add(s->rule.ptr->states_tot, 1); \ 350 if (s->anchor.ptr != NULL) { \ 351 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 352 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 353 } \ 354 if (s->nat_rule.ptr != NULL) { \ 355 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 356 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 357 } \ 358 } while (0) 359 360 #define STATE_DEC_COUNTERS(s) \ 361 do { \ 362 if (s->nat_rule.ptr != NULL) \ 363 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 364 if (s->anchor.ptr != NULL) \ 365 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 366 counter_u64_add(s->rule.ptr->states_cur, -1); \ 367 } while (0) 368 369 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 370 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 371 VNET_DEFINE(struct pf_idhash *, pf_idhash); 372 VNET_DEFINE(struct pf_srchash *, pf_srchash); 373 374 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 375 "pf(4)"); 376 377 u_long pf_hashmask; 378 u_long pf_srchashmask; 379 static u_long pf_hashsize; 380 static u_long pf_srchashsize; 381 u_long pf_ioctl_maxcount = 65535; 382 383 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 384 &pf_hashsize, 0, "Size of pf(4) states hashtable"); 385 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 386 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); 387 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 388 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 389 390 VNET_DEFINE(void *, pf_swi_cookie); 391 VNET_DEFINE(struct intr_event *, pf_swi_ie); 392 393 VNET_DEFINE(uint32_t, pf_hashseed); 394 #define V_pf_hashseed VNET(pf_hashseed) 395 396 int 397 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 398 { 399 400 switch (af) { 401 #ifdef INET 402 case AF_INET: 403 if (a->addr32[0] > b->addr32[0]) 404 return (1); 405 if (a->addr32[0] < b->addr32[0]) 406 return (-1); 407 break; 408 #endif /* INET */ 409 #ifdef INET6 410 case AF_INET6: 411 if (a->addr32[3] > b->addr32[3]) 412 return (1); 413 if (a->addr32[3] < b->addr32[3]) 414 return (-1); 415 if (a->addr32[2] > b->addr32[2]) 416 return (1); 417 if (a->addr32[2] < b->addr32[2]) 418 return (-1); 419 if (a->addr32[1] > b->addr32[1]) 420 return (1); 421 if (a->addr32[1] < b->addr32[1]) 422 return (-1); 423 if (a->addr32[0] > b->addr32[0]) 424 return (1); 425 if (a->addr32[0] < b->addr32[0]) 426 return (-1); 427 break; 428 #endif /* INET6 */ 429 default: 430 panic("%s: unknown address family %u", __func__, af); 431 } 432 return (0); 433 } 434 435 static __inline uint32_t 436 pf_hashkey(struct pf_state_key *sk) 437 { 438 uint32_t h; 439 440 h = murmur3_32_hash32((uint32_t *)sk, 441 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 442 V_pf_hashseed); 443 444 return (h & pf_hashmask); 445 } 446 447 static __inline uint32_t 448 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 449 { 450 uint32_t h; 451 452 switch (af) { 453 case AF_INET: 454 h = murmur3_32_hash32((uint32_t *)&addr->v4, 455 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 456 break; 457 case AF_INET6: 458 h = murmur3_32_hash32((uint32_t *)&addr->v6, 459 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 460 break; 461 default: 462 panic("%s: unknown address family %u", __func__, af); 463 } 464 465 return (h & pf_srchashmask); 466 } 467 468 #ifdef ALTQ 469 static int 470 pf_state_hash(struct pf_state *s) 471 { 472 u_int32_t hv = (intptr_t)s / sizeof(*s); 473 474 hv ^= crc32(&s->src, sizeof(s->src)); 475 hv ^= crc32(&s->dst, sizeof(s->dst)); 476 if (hv == 0) 477 hv = 1; 478 return (hv); 479 } 480 #endif 481 482 #ifdef INET6 483 void 484 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 485 { 486 switch (af) { 487 #ifdef INET 488 case AF_INET: 489 dst->addr32[0] = src->addr32[0]; 490 break; 491 #endif /* INET */ 492 case AF_INET6: 493 dst->addr32[0] = src->addr32[0]; 494 dst->addr32[1] = src->addr32[1]; 495 dst->addr32[2] = src->addr32[2]; 496 dst->addr32[3] = src->addr32[3]; 497 break; 498 } 499 } 500 #endif /* INET6 */ 501 502 static void 503 pf_init_threshold(struct pf_threshold *threshold, 504 u_int32_t limit, u_int32_t seconds) 505 { 506 threshold->limit = limit * PF_THRESHOLD_MULT; 507 threshold->seconds = seconds; 508 threshold->count = 0; 509 threshold->last = time_uptime; 510 } 511 512 static void 513 pf_add_threshold(struct pf_threshold *threshold) 514 { 515 u_int32_t t = time_uptime, diff = t - threshold->last; 516 517 if (diff >= threshold->seconds) 518 threshold->count = 0; 519 else 520 threshold->count -= threshold->count * diff / 521 threshold->seconds; 522 threshold->count += PF_THRESHOLD_MULT; 523 threshold->last = t; 524 } 525 526 static int 527 pf_check_threshold(struct pf_threshold *threshold) 528 { 529 return (threshold->count > threshold->limit); 530 } 531 532 static int 533 pf_src_connlimit(struct pf_state **state) 534 { 535 struct pf_overload_entry *pfoe; 536 int bad = 0; 537 538 PF_STATE_LOCK_ASSERT(*state); 539 540 (*state)->src_node->conn++; 541 (*state)->src.tcp_est = 1; 542 pf_add_threshold(&(*state)->src_node->conn_rate); 543 544 if ((*state)->rule.ptr->max_src_conn && 545 (*state)->rule.ptr->max_src_conn < 546 (*state)->src_node->conn) { 547 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 548 bad++; 549 } 550 551 if ((*state)->rule.ptr->max_src_conn_rate.limit && 552 pf_check_threshold(&(*state)->src_node->conn_rate)) { 553 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 554 bad++; 555 } 556 557 if (!bad) 558 return (0); 559 560 /* Kill this state. */ 561 (*state)->timeout = PFTM_PURGE; 562 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; 563 564 if ((*state)->rule.ptr->overload_tbl == NULL) 565 return (1); 566 567 /* Schedule overloading and flushing task. */ 568 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 569 if (pfoe == NULL) 570 return (1); /* too bad :( */ 571 572 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 573 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 574 pfoe->rule = (*state)->rule.ptr; 575 pfoe->dir = (*state)->direction; 576 PF_OVERLOADQ_LOCK(); 577 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 578 PF_OVERLOADQ_UNLOCK(); 579 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 580 581 return (1); 582 } 583 584 static void 585 pf_overload_task(void *v, int pending) 586 { 587 struct pf_overload_head queue; 588 struct pfr_addr p; 589 struct pf_overload_entry *pfoe, *pfoe1; 590 uint32_t killed = 0; 591 592 CURVNET_SET((struct vnet *)v); 593 594 PF_OVERLOADQ_LOCK(); 595 queue = V_pf_overloadqueue; 596 SLIST_INIT(&V_pf_overloadqueue); 597 PF_OVERLOADQ_UNLOCK(); 598 599 bzero(&p, sizeof(p)); 600 SLIST_FOREACH(pfoe, &queue, next) { 601 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 602 if (V_pf_status.debug >= PF_DEBUG_MISC) { 603 printf("%s: blocking address ", __func__); 604 pf_print_host(&pfoe->addr, 0, pfoe->af); 605 printf("\n"); 606 } 607 608 p.pfra_af = pfoe->af; 609 switch (pfoe->af) { 610 #ifdef INET 611 case AF_INET: 612 p.pfra_net = 32; 613 p.pfra_ip4addr = pfoe->addr.v4; 614 break; 615 #endif 616 #ifdef INET6 617 case AF_INET6: 618 p.pfra_net = 128; 619 p.pfra_ip6addr = pfoe->addr.v6; 620 break; 621 #endif 622 } 623 624 PF_RULES_WLOCK(); 625 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 626 PF_RULES_WUNLOCK(); 627 } 628 629 /* 630 * Remove those entries, that don't need flushing. 631 */ 632 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 633 if (pfoe->rule->flush == 0) { 634 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 635 free(pfoe, M_PFTEMP); 636 } else 637 counter_u64_add( 638 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 639 640 /* If nothing to flush, return. */ 641 if (SLIST_EMPTY(&queue)) { 642 CURVNET_RESTORE(); 643 return; 644 } 645 646 for (int i = 0; i <= pf_hashmask; i++) { 647 struct pf_idhash *ih = &V_pf_idhash[i]; 648 struct pf_state_key *sk; 649 struct pf_state *s; 650 651 PF_HASHROW_LOCK(ih); 652 LIST_FOREACH(s, &ih->states, entry) { 653 sk = s->key[PF_SK_WIRE]; 654 SLIST_FOREACH(pfoe, &queue, next) 655 if (sk->af == pfoe->af && 656 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 657 pfoe->rule == s->rule.ptr) && 658 ((pfoe->dir == PF_OUT && 659 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 660 (pfoe->dir == PF_IN && 661 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 662 s->timeout = PFTM_PURGE; 663 s->src.state = s->dst.state = TCPS_CLOSED; 664 killed++; 665 } 666 } 667 PF_HASHROW_UNLOCK(ih); 668 } 669 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 670 free(pfoe, M_PFTEMP); 671 if (V_pf_status.debug >= PF_DEBUG_MISC) 672 printf("%s: %u states killed", __func__, killed); 673 674 CURVNET_RESTORE(); 675 } 676 677 /* 678 * Can return locked on failure, so that we can consistently 679 * allocate and insert a new one. 680 */ 681 struct pf_ksrc_node * 682 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 683 int returnlocked) 684 { 685 struct pf_srchash *sh; 686 struct pf_ksrc_node *n; 687 688 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 689 690 sh = &V_pf_srchash[pf_hashsrc(src, af)]; 691 PF_HASHROW_LOCK(sh); 692 LIST_FOREACH(n, &sh->nodes, entry) 693 if (n->rule.ptr == rule && n->af == af && 694 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 695 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 696 break; 697 if (n != NULL) { 698 n->states++; 699 PF_HASHROW_UNLOCK(sh); 700 } else if (returnlocked == 0) 701 PF_HASHROW_UNLOCK(sh); 702 703 return (n); 704 } 705 706 static void 707 pf_free_src_node(struct pf_ksrc_node *sn) 708 { 709 710 for (int i = 0; i < 2; i++) { 711 counter_u64_free(sn->bytes[i]); 712 counter_u64_free(sn->packets[i]); 713 } 714 uma_zfree(V_pf_sources_z, sn); 715 } 716 717 static int 718 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 719 struct pf_addr *src, sa_family_t af) 720 { 721 722 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 723 rule->rpool.opts & PF_POOL_STICKYADDR), 724 ("%s for non-tracking rule %p", __func__, rule)); 725 726 if (*sn == NULL) 727 *sn = pf_find_src_node(src, rule, af, 1); 728 729 if (*sn == NULL) { 730 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 731 732 PF_HASHROW_ASSERT(sh); 733 734 if (!rule->max_src_nodes || 735 counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes) 736 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 737 else 738 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 739 1); 740 if ((*sn) == NULL) { 741 PF_HASHROW_UNLOCK(sh); 742 return (-1); 743 } 744 745 for (int i = 0; i < 2; i++) { 746 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 747 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 748 749 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 750 pf_free_src_node(*sn); 751 PF_HASHROW_UNLOCK(sh); 752 return (-1); 753 } 754 } 755 756 pf_init_threshold(&(*sn)->conn_rate, 757 rule->max_src_conn_rate.limit, 758 rule->max_src_conn_rate.seconds); 759 760 (*sn)->af = af; 761 (*sn)->rule.ptr = rule; 762 PF_ACPY(&(*sn)->addr, src, af); 763 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 764 (*sn)->creation = time_uptime; 765 (*sn)->ruletype = rule->action; 766 (*sn)->states = 1; 767 if ((*sn)->rule.ptr != NULL) 768 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 769 PF_HASHROW_UNLOCK(sh); 770 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 771 } else { 772 if (rule->max_src_states && 773 (*sn)->states >= rule->max_src_states) { 774 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 775 1); 776 return (-1); 777 } 778 } 779 return (0); 780 } 781 782 void 783 pf_unlink_src_node(struct pf_ksrc_node *src) 784 { 785 786 PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]); 787 LIST_REMOVE(src, entry); 788 if (src->rule.ptr) 789 counter_u64_add(src->rule.ptr->src_nodes, -1); 790 } 791 792 u_int 793 pf_free_src_nodes(struct pf_ksrc_node_list *head) 794 { 795 struct pf_ksrc_node *sn, *tmp; 796 u_int count = 0; 797 798 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 799 pf_free_src_node(sn); 800 count++; 801 } 802 803 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 804 805 return (count); 806 } 807 808 void 809 pf_mtag_initialize() 810 { 811 812 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 813 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 814 UMA_ALIGN_PTR, 0); 815 } 816 817 /* Per-vnet data storage structures initialization. */ 818 void 819 pf_initialize() 820 { 821 struct pf_keyhash *kh; 822 struct pf_idhash *ih; 823 struct pf_srchash *sh; 824 u_int i; 825 826 if (pf_hashsize == 0 || !powerof2(pf_hashsize)) 827 pf_hashsize = PF_HASHSIZ; 828 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) 829 pf_srchashsize = PF_SRCHASHSIZ; 830 831 V_pf_hashseed = arc4random(); 832 833 /* States and state keys storage. */ 834 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state), 835 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 836 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 837 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 838 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 839 840 V_pf_state_key_z = uma_zcreate("pf state keys", 841 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 842 UMA_ALIGN_PTR, 0); 843 844 V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), 845 M_PFHASH, M_NOWAIT | M_ZERO); 846 V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), 847 M_PFHASH, M_NOWAIT | M_ZERO); 848 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 849 printf("pf: Unable to allocate memory for " 850 "state_hashsize %lu.\n", pf_hashsize); 851 852 free(V_pf_keyhash, M_PFHASH); 853 free(V_pf_idhash, M_PFHASH); 854 855 pf_hashsize = PF_HASHSIZ; 856 V_pf_keyhash = mallocarray(pf_hashsize, 857 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 858 V_pf_idhash = mallocarray(pf_hashsize, 859 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 860 } 861 862 pf_hashmask = pf_hashsize - 1; 863 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 864 i++, kh++, ih++) { 865 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 866 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 867 } 868 869 /* Source nodes. */ 870 V_pf_sources_z = uma_zcreate("pf source nodes", 871 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 872 0); 873 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 874 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 875 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 876 877 V_pf_srchash = mallocarray(pf_srchashsize, 878 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 879 if (V_pf_srchash == NULL) { 880 printf("pf: Unable to allocate memory for " 881 "source_hashsize %lu.\n", pf_srchashsize); 882 883 pf_srchashsize = PF_SRCHASHSIZ; 884 V_pf_srchash = mallocarray(pf_srchashsize, 885 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 886 } 887 888 pf_srchashmask = pf_srchashsize - 1; 889 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) 890 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 891 892 /* ALTQ */ 893 TAILQ_INIT(&V_pf_altqs[0]); 894 TAILQ_INIT(&V_pf_altqs[1]); 895 TAILQ_INIT(&V_pf_altqs[2]); 896 TAILQ_INIT(&V_pf_altqs[3]); 897 TAILQ_INIT(&V_pf_pabuf); 898 V_pf_altqs_active = &V_pf_altqs[0]; 899 V_pf_altq_ifs_active = &V_pf_altqs[1]; 900 V_pf_altqs_inactive = &V_pf_altqs[2]; 901 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 902 903 /* Send & overload+flush queues. */ 904 STAILQ_INIT(&V_pf_sendqueue); 905 SLIST_INIT(&V_pf_overloadqueue); 906 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 907 908 /* Unlinked, but may be referenced rules. */ 909 TAILQ_INIT(&V_pf_unlinked_rules); 910 } 911 912 void 913 pf_mtag_cleanup() 914 { 915 916 uma_zdestroy(pf_mtag_z); 917 } 918 919 void 920 pf_cleanup() 921 { 922 struct pf_keyhash *kh; 923 struct pf_idhash *ih; 924 struct pf_srchash *sh; 925 struct pf_send_entry *pfse, *next; 926 u_int i; 927 928 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 929 i++, kh++, ih++) { 930 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 931 __func__)); 932 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 933 __func__)); 934 mtx_destroy(&kh->lock); 935 mtx_destroy(&ih->lock); 936 } 937 free(V_pf_keyhash, M_PFHASH); 938 free(V_pf_idhash, M_PFHASH); 939 940 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 941 KASSERT(LIST_EMPTY(&sh->nodes), 942 ("%s: source node hash not empty", __func__)); 943 mtx_destroy(&sh->lock); 944 } 945 free(V_pf_srchash, M_PFHASH); 946 947 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 948 m_freem(pfse->pfse_m); 949 free(pfse, M_PFTEMP); 950 } 951 952 uma_zdestroy(V_pf_sources_z); 953 uma_zdestroy(V_pf_state_z); 954 uma_zdestroy(V_pf_state_key_z); 955 } 956 957 static int 958 pf_mtag_uminit(void *mem, int size, int how) 959 { 960 struct m_tag *t; 961 962 t = (struct m_tag *)mem; 963 t->m_tag_cookie = MTAG_ABI_COMPAT; 964 t->m_tag_id = PACKET_TAG_PF; 965 t->m_tag_len = sizeof(struct pf_mtag); 966 t->m_tag_free = pf_mtag_free; 967 968 return (0); 969 } 970 971 static void 972 pf_mtag_free(struct m_tag *t) 973 { 974 975 uma_zfree(pf_mtag_z, t); 976 } 977 978 struct pf_mtag * 979 pf_get_mtag(struct mbuf *m) 980 { 981 struct m_tag *mtag; 982 983 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 984 return ((struct pf_mtag *)(mtag + 1)); 985 986 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 987 if (mtag == NULL) 988 return (NULL); 989 bzero(mtag + 1, sizeof(struct pf_mtag)); 990 m_tag_prepend(m, mtag); 991 992 return ((struct pf_mtag *)(mtag + 1)); 993 } 994 995 static int 996 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 997 struct pf_state *s) 998 { 999 struct pf_keyhash *khs, *khw, *kh; 1000 struct pf_state_key *sk, *cur; 1001 struct pf_state *si, *olds = NULL; 1002 int idx; 1003 1004 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1005 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1006 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1007 1008 /* 1009 * We need to lock hash slots of both keys. To avoid deadlock 1010 * we always lock the slot with lower address first. Unlock order 1011 * isn't important. 1012 * 1013 * We also need to lock ID hash slot before dropping key 1014 * locks. On success we return with ID hash slot locked. 1015 */ 1016 1017 if (skw == sks) { 1018 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1019 PF_HASHROW_LOCK(khs); 1020 } else { 1021 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1022 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1023 if (khs == khw) { 1024 PF_HASHROW_LOCK(khs); 1025 } else if (khs < khw) { 1026 PF_HASHROW_LOCK(khs); 1027 PF_HASHROW_LOCK(khw); 1028 } else { 1029 PF_HASHROW_LOCK(khw); 1030 PF_HASHROW_LOCK(khs); 1031 } 1032 } 1033 1034 #define KEYS_UNLOCK() do { \ 1035 if (khs != khw) { \ 1036 PF_HASHROW_UNLOCK(khs); \ 1037 PF_HASHROW_UNLOCK(khw); \ 1038 } else \ 1039 PF_HASHROW_UNLOCK(khs); \ 1040 } while (0) 1041 1042 /* 1043 * First run: start with wire key. 1044 */ 1045 sk = skw; 1046 kh = khw; 1047 idx = PF_SK_WIRE; 1048 1049 keyattach: 1050 LIST_FOREACH(cur, &kh->keys, entry) 1051 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1052 break; 1053 1054 if (cur != NULL) { 1055 /* Key exists. Check for same kif, if none, add to key. */ 1056 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1057 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1058 1059 PF_HASHROW_LOCK(ih); 1060 if (si->kif == s->kif && 1061 si->direction == s->direction) { 1062 if (sk->proto == IPPROTO_TCP && 1063 si->src.state >= TCPS_FIN_WAIT_2 && 1064 si->dst.state >= TCPS_FIN_WAIT_2) { 1065 /* 1066 * New state matches an old >FIN_WAIT_2 1067 * state. We can't drop key hash locks, 1068 * thus we can't unlink it properly. 1069 * 1070 * As a workaround we drop it into 1071 * TCPS_CLOSED state, schedule purge 1072 * ASAP and push it into the very end 1073 * of the slot TAILQ, so that it won't 1074 * conflict with our new state. 1075 */ 1076 si->src.state = si->dst.state = 1077 TCPS_CLOSED; 1078 si->timeout = PFTM_PURGE; 1079 olds = si; 1080 } else { 1081 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1082 printf("pf: %s key attach " 1083 "failed on %s: ", 1084 (idx == PF_SK_WIRE) ? 1085 "wire" : "stack", 1086 s->kif->pfik_name); 1087 pf_print_state_parts(s, 1088 (idx == PF_SK_WIRE) ? 1089 sk : NULL, 1090 (idx == PF_SK_STACK) ? 1091 sk : NULL); 1092 printf(", existing: "); 1093 pf_print_state_parts(si, 1094 (idx == PF_SK_WIRE) ? 1095 sk : NULL, 1096 (idx == PF_SK_STACK) ? 1097 sk : NULL); 1098 printf("\n"); 1099 } 1100 PF_HASHROW_UNLOCK(ih); 1101 KEYS_UNLOCK(); 1102 uma_zfree(V_pf_state_key_z, sk); 1103 if (idx == PF_SK_STACK) 1104 pf_detach_state(s); 1105 return (EEXIST); /* collision! */ 1106 } 1107 } 1108 PF_HASHROW_UNLOCK(ih); 1109 } 1110 uma_zfree(V_pf_state_key_z, sk); 1111 s->key[idx] = cur; 1112 } else { 1113 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1114 s->key[idx] = sk; 1115 } 1116 1117 stateattach: 1118 /* List is sorted, if-bound states before floating. */ 1119 if (s->kif == V_pfi_all) 1120 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1121 else 1122 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1123 1124 if (olds) { 1125 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1126 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1127 key_list[idx]); 1128 olds = NULL; 1129 } 1130 1131 /* 1132 * Attach done. See how should we (or should not?) 1133 * attach a second key. 1134 */ 1135 if (sks == skw) { 1136 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1137 idx = PF_SK_STACK; 1138 sks = NULL; 1139 goto stateattach; 1140 } else if (sks != NULL) { 1141 /* 1142 * Continue attaching with stack key. 1143 */ 1144 sk = sks; 1145 kh = khs; 1146 idx = PF_SK_STACK; 1147 sks = NULL; 1148 goto keyattach; 1149 } 1150 1151 PF_STATE_LOCK(s); 1152 KEYS_UNLOCK(); 1153 1154 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1155 ("%s failure", __func__)); 1156 1157 return (0); 1158 #undef KEYS_UNLOCK 1159 } 1160 1161 static void 1162 pf_detach_state(struct pf_state *s) 1163 { 1164 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1165 struct pf_keyhash *kh; 1166 1167 if (sks != NULL) { 1168 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1169 PF_HASHROW_LOCK(kh); 1170 if (s->key[PF_SK_STACK] != NULL) 1171 pf_state_key_detach(s, PF_SK_STACK); 1172 /* 1173 * If both point to same key, then we are done. 1174 */ 1175 if (sks == s->key[PF_SK_WIRE]) { 1176 pf_state_key_detach(s, PF_SK_WIRE); 1177 PF_HASHROW_UNLOCK(kh); 1178 return; 1179 } 1180 PF_HASHROW_UNLOCK(kh); 1181 } 1182 1183 if (s->key[PF_SK_WIRE] != NULL) { 1184 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1185 PF_HASHROW_LOCK(kh); 1186 if (s->key[PF_SK_WIRE] != NULL) 1187 pf_state_key_detach(s, PF_SK_WIRE); 1188 PF_HASHROW_UNLOCK(kh); 1189 } 1190 } 1191 1192 static void 1193 pf_state_key_detach(struct pf_state *s, int idx) 1194 { 1195 struct pf_state_key *sk = s->key[idx]; 1196 #ifdef INVARIANTS 1197 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1198 1199 PF_HASHROW_ASSERT(kh); 1200 #endif 1201 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1202 s->key[idx] = NULL; 1203 1204 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1205 LIST_REMOVE(sk, entry); 1206 uma_zfree(V_pf_state_key_z, sk); 1207 } 1208 } 1209 1210 static int 1211 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1212 { 1213 struct pf_state_key *sk = mem; 1214 1215 bzero(sk, sizeof(struct pf_state_key_cmp)); 1216 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1217 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1218 1219 return (0); 1220 } 1221 1222 struct pf_state_key * 1223 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1224 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1225 { 1226 struct pf_state_key *sk; 1227 1228 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1229 if (sk == NULL) 1230 return (NULL); 1231 1232 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1233 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1234 sk->port[pd->sidx] = sport; 1235 sk->port[pd->didx] = dport; 1236 sk->proto = pd->proto; 1237 sk->af = pd->af; 1238 1239 return (sk); 1240 } 1241 1242 struct pf_state_key * 1243 pf_state_key_clone(struct pf_state_key *orig) 1244 { 1245 struct pf_state_key *sk; 1246 1247 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1248 if (sk == NULL) 1249 return (NULL); 1250 1251 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1252 1253 return (sk); 1254 } 1255 1256 int 1257 pf_state_insert(struct pfi_kkif *kif, struct pf_state_key *skw, 1258 struct pf_state_key *sks, struct pf_state *s) 1259 { 1260 struct pf_idhash *ih; 1261 struct pf_state *cur; 1262 int error; 1263 1264 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1265 ("%s: sks not pristine", __func__)); 1266 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1267 ("%s: skw not pristine", __func__)); 1268 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1269 1270 s->kif = kif; 1271 1272 if (s->id == 0 && s->creatorid == 0) { 1273 /* XXX: should be atomic, but probability of collision low */ 1274 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID) 1275 V_pf_stateid[curcpu] = 1; 1276 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT; 1277 s->id = htobe64(s->id); 1278 s->creatorid = V_pf_status.hostid; 1279 } 1280 1281 /* Returns with ID locked on success. */ 1282 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1283 return (error); 1284 1285 ih = &V_pf_idhash[PF_IDHASH(s)]; 1286 PF_HASHROW_ASSERT(ih); 1287 LIST_FOREACH(cur, &ih->states, entry) 1288 if (cur->id == s->id && cur->creatorid == s->creatorid) 1289 break; 1290 1291 if (cur != NULL) { 1292 PF_HASHROW_UNLOCK(ih); 1293 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1294 printf("pf: state ID collision: " 1295 "id: %016llx creatorid: %08x\n", 1296 (unsigned long long)be64toh(s->id), 1297 ntohl(s->creatorid)); 1298 } 1299 pf_detach_state(s); 1300 return (EEXIST); 1301 } 1302 LIST_INSERT_HEAD(&ih->states, s, entry); 1303 /* One for keys, one for ID hash. */ 1304 refcount_init(&s->refs, 2); 1305 1306 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1307 if (V_pfsync_insert_state_ptr != NULL) 1308 V_pfsync_insert_state_ptr(s); 1309 1310 /* Returns locked. */ 1311 return (0); 1312 } 1313 1314 /* 1315 * Find state by ID: returns with locked row on success. 1316 */ 1317 struct pf_state * 1318 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1319 { 1320 struct pf_idhash *ih; 1321 struct pf_state *s; 1322 1323 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1324 1325 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; 1326 1327 PF_HASHROW_LOCK(ih); 1328 LIST_FOREACH(s, &ih->states, entry) 1329 if (s->id == id && s->creatorid == creatorid) 1330 break; 1331 1332 if (s == NULL) 1333 PF_HASHROW_UNLOCK(ih); 1334 1335 return (s); 1336 } 1337 1338 /* 1339 * Find state by key. 1340 * Returns with ID hash slot locked on success. 1341 */ 1342 static struct pf_state * 1343 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir) 1344 { 1345 struct pf_keyhash *kh; 1346 struct pf_state_key *sk; 1347 struct pf_state *s; 1348 int idx; 1349 1350 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1351 1352 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1353 1354 PF_HASHROW_LOCK(kh); 1355 LIST_FOREACH(sk, &kh->keys, entry) 1356 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1357 break; 1358 if (sk == NULL) { 1359 PF_HASHROW_UNLOCK(kh); 1360 return (NULL); 1361 } 1362 1363 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1364 1365 /* List is sorted, if-bound states before floating ones. */ 1366 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1367 if (s->kif == V_pfi_all || s->kif == kif) { 1368 PF_STATE_LOCK(s); 1369 PF_HASHROW_UNLOCK(kh); 1370 if (s->timeout >= PFTM_MAX) { 1371 /* 1372 * State is either being processed by 1373 * pf_unlink_state() in an other thread, or 1374 * is scheduled for immediate expiry. 1375 */ 1376 PF_STATE_UNLOCK(s); 1377 return (NULL); 1378 } 1379 return (s); 1380 } 1381 PF_HASHROW_UNLOCK(kh); 1382 1383 return (NULL); 1384 } 1385 1386 struct pf_state * 1387 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1388 { 1389 struct pf_keyhash *kh; 1390 struct pf_state_key *sk; 1391 struct pf_state *s, *ret = NULL; 1392 int idx, inout = 0; 1393 1394 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1395 1396 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1397 1398 PF_HASHROW_LOCK(kh); 1399 LIST_FOREACH(sk, &kh->keys, entry) 1400 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1401 break; 1402 if (sk == NULL) { 1403 PF_HASHROW_UNLOCK(kh); 1404 return (NULL); 1405 } 1406 switch (dir) { 1407 case PF_IN: 1408 idx = PF_SK_WIRE; 1409 break; 1410 case PF_OUT: 1411 idx = PF_SK_STACK; 1412 break; 1413 case PF_INOUT: 1414 idx = PF_SK_WIRE; 1415 inout = 1; 1416 break; 1417 default: 1418 panic("%s: dir %u", __func__, dir); 1419 } 1420 second_run: 1421 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1422 if (more == NULL) { 1423 PF_HASHROW_UNLOCK(kh); 1424 return (s); 1425 } 1426 1427 if (ret) 1428 (*more)++; 1429 else 1430 ret = s; 1431 } 1432 if (inout == 1) { 1433 inout = 0; 1434 idx = PF_SK_STACK; 1435 goto second_run; 1436 } 1437 PF_HASHROW_UNLOCK(kh); 1438 1439 return (ret); 1440 } 1441 1442 /* END state table stuff */ 1443 1444 static void 1445 pf_send(struct pf_send_entry *pfse) 1446 { 1447 1448 PF_SENDQ_LOCK(); 1449 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1450 PF_SENDQ_UNLOCK(); 1451 swi_sched(V_pf_swi_cookie, 0); 1452 } 1453 1454 void 1455 pf_intr(void *v) 1456 { 1457 struct epoch_tracker et; 1458 struct pf_send_head queue; 1459 struct pf_send_entry *pfse, *next; 1460 1461 CURVNET_SET((struct vnet *)v); 1462 1463 PF_SENDQ_LOCK(); 1464 queue = V_pf_sendqueue; 1465 STAILQ_INIT(&V_pf_sendqueue); 1466 PF_SENDQ_UNLOCK(); 1467 1468 NET_EPOCH_ENTER(et); 1469 1470 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1471 switch (pfse->pfse_type) { 1472 #ifdef INET 1473 case PFSE_IP: 1474 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL); 1475 break; 1476 case PFSE_ICMP: 1477 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1478 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1479 break; 1480 #endif /* INET */ 1481 #ifdef INET6 1482 case PFSE_IP6: 1483 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL, 1484 NULL); 1485 break; 1486 case PFSE_ICMP6: 1487 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 1488 pfse->icmpopts.code, pfse->icmpopts.mtu); 1489 break; 1490 #endif /* INET6 */ 1491 default: 1492 panic("%s: unknown type", __func__); 1493 } 1494 free(pfse, M_PFTEMP); 1495 } 1496 NET_EPOCH_EXIT(et); 1497 CURVNET_RESTORE(); 1498 } 1499 1500 void 1501 pf_purge_thread(void *unused __unused) 1502 { 1503 VNET_ITERATOR_DECL(vnet_iter); 1504 1505 sx_xlock(&pf_end_lock); 1506 while (pf_end_threads == 0) { 1507 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10); 1508 1509 VNET_LIST_RLOCK(); 1510 VNET_FOREACH(vnet_iter) { 1511 CURVNET_SET(vnet_iter); 1512 1513 /* Wait until V_pf_default_rule is initialized. */ 1514 if (V_pf_vnet_active == 0) { 1515 CURVNET_RESTORE(); 1516 continue; 1517 } 1518 1519 /* 1520 * Process 1/interval fraction of the state 1521 * table every run. 1522 */ 1523 V_pf_purge_idx = 1524 pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / 1525 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1526 1527 /* 1528 * Purge other expired types every 1529 * PFTM_INTERVAL seconds. 1530 */ 1531 if (V_pf_purge_idx == 0) { 1532 /* 1533 * Order is important: 1534 * - states and src nodes reference rules 1535 * - states and rules reference kifs 1536 */ 1537 pf_purge_expired_fragments(); 1538 pf_purge_expired_src_nodes(); 1539 pf_purge_unlinked_rules(); 1540 pfi_kkif_purge(); 1541 } 1542 CURVNET_RESTORE(); 1543 } 1544 VNET_LIST_RUNLOCK(); 1545 } 1546 1547 pf_end_threads++; 1548 sx_xunlock(&pf_end_lock); 1549 kproc_exit(0); 1550 } 1551 1552 void 1553 pf_unload_vnet_purge(void) 1554 { 1555 1556 /* 1557 * To cleanse up all kifs and rules we need 1558 * two runs: first one clears reference flags, 1559 * then pf_purge_expired_states() doesn't 1560 * raise them, and then second run frees. 1561 */ 1562 pf_purge_unlinked_rules(); 1563 pfi_kkif_purge(); 1564 1565 /* 1566 * Now purge everything. 1567 */ 1568 pf_purge_expired_states(0, pf_hashmask); 1569 pf_purge_fragments(UINT_MAX); 1570 pf_purge_expired_src_nodes(); 1571 1572 /* 1573 * Now all kifs & rules should be unreferenced, 1574 * thus should be successfully freed. 1575 */ 1576 pf_purge_unlinked_rules(); 1577 pfi_kkif_purge(); 1578 } 1579 1580 u_int32_t 1581 pf_state_expires(const struct pf_state *state) 1582 { 1583 u_int32_t timeout; 1584 u_int32_t start; 1585 u_int32_t end; 1586 u_int32_t states; 1587 1588 /* handle all PFTM_* > PFTM_MAX here */ 1589 if (state->timeout == PFTM_PURGE) 1590 return (time_uptime); 1591 KASSERT(state->timeout != PFTM_UNLINKED, 1592 ("pf_state_expires: timeout == PFTM_UNLINKED")); 1593 KASSERT((state->timeout < PFTM_MAX), 1594 ("pf_state_expires: timeout > PFTM_MAX")); 1595 timeout = state->rule.ptr->timeout[state->timeout]; 1596 if (!timeout) 1597 timeout = V_pf_default_rule.timeout[state->timeout]; 1598 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1599 if (start && state->rule.ptr != &V_pf_default_rule) { 1600 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1601 states = counter_u64_fetch(state->rule.ptr->states_cur); 1602 } else { 1603 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1604 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1605 states = V_pf_status.states; 1606 } 1607 if (end && states > start && start < end) { 1608 if (states < end) { 1609 timeout = (u_int64_t)timeout * (end - states) / 1610 (end - start); 1611 return (state->expire + timeout); 1612 } 1613 else 1614 return (time_uptime); 1615 } 1616 return (state->expire + timeout); 1617 } 1618 1619 void 1620 pf_purge_expired_src_nodes() 1621 { 1622 struct pf_ksrc_node_list freelist; 1623 struct pf_srchash *sh; 1624 struct pf_ksrc_node *cur, *next; 1625 int i; 1626 1627 LIST_INIT(&freelist); 1628 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1629 PF_HASHROW_LOCK(sh); 1630 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 1631 if (cur->states == 0 && cur->expire <= time_uptime) { 1632 pf_unlink_src_node(cur); 1633 LIST_INSERT_HEAD(&freelist, cur, entry); 1634 } else if (cur->rule.ptr != NULL) 1635 cur->rule.ptr->rule_flag |= PFRULE_REFS; 1636 PF_HASHROW_UNLOCK(sh); 1637 } 1638 1639 pf_free_src_nodes(&freelist); 1640 1641 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 1642 } 1643 1644 static void 1645 pf_src_tree_remove_state(struct pf_state *s) 1646 { 1647 struct pf_ksrc_node *sn; 1648 struct pf_srchash *sh; 1649 uint32_t timeout; 1650 1651 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 1652 s->rule.ptr->timeout[PFTM_SRC_NODE] : 1653 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 1654 1655 if (s->src_node != NULL) { 1656 sn = s->src_node; 1657 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 1658 PF_HASHROW_LOCK(sh); 1659 if (s->src.tcp_est) 1660 --sn->conn; 1661 if (--sn->states == 0) 1662 sn->expire = time_uptime + timeout; 1663 PF_HASHROW_UNLOCK(sh); 1664 } 1665 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 1666 sn = s->nat_src_node; 1667 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 1668 PF_HASHROW_LOCK(sh); 1669 if (--sn->states == 0) 1670 sn->expire = time_uptime + timeout; 1671 PF_HASHROW_UNLOCK(sh); 1672 } 1673 s->src_node = s->nat_src_node = NULL; 1674 } 1675 1676 /* 1677 * Unlink and potentilly free a state. Function may be 1678 * called with ID hash row locked, but always returns 1679 * unlocked, since it needs to go through key hash locking. 1680 */ 1681 int 1682 pf_unlink_state(struct pf_state *s, u_int flags) 1683 { 1684 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 1685 1686 if ((flags & PF_ENTER_LOCKED) == 0) 1687 PF_HASHROW_LOCK(ih); 1688 else 1689 PF_HASHROW_ASSERT(ih); 1690 1691 if (s->timeout == PFTM_UNLINKED) { 1692 /* 1693 * State is being processed 1694 * by pf_unlink_state() in 1695 * an other thread. 1696 */ 1697 PF_HASHROW_UNLOCK(ih); 1698 return (0); /* XXXGL: undefined actually */ 1699 } 1700 1701 if (s->src.state == PF_TCPS_PROXY_DST) { 1702 /* XXX wire key the right one? */ 1703 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af, 1704 &s->key[PF_SK_WIRE]->addr[1], 1705 &s->key[PF_SK_WIRE]->addr[0], 1706 s->key[PF_SK_WIRE]->port[1], 1707 s->key[PF_SK_WIRE]->port[0], 1708 s->src.seqhi, s->src.seqlo + 1, 1709 TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL); 1710 } 1711 1712 LIST_REMOVE(s, entry); 1713 pf_src_tree_remove_state(s); 1714 1715 if (V_pfsync_delete_state_ptr != NULL) 1716 V_pfsync_delete_state_ptr(s); 1717 1718 STATE_DEC_COUNTERS(s); 1719 1720 s->timeout = PFTM_UNLINKED; 1721 1722 PF_HASHROW_UNLOCK(ih); 1723 1724 pf_detach_state(s); 1725 /* pf_state_insert() initialises refs to 2, so we can never release the 1726 * last reference here, only in pf_release_state(). */ 1727 (void)refcount_release(&s->refs); 1728 1729 return (pf_release_state(s)); 1730 } 1731 1732 void 1733 pf_free_state(struct pf_state *cur) 1734 { 1735 1736 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 1737 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 1738 cur->timeout)); 1739 1740 for (int i = 0; i < 2; i++) { 1741 counter_u64_free(cur->bytes[i]); 1742 counter_u64_free(cur->packets[i]); 1743 } 1744 1745 pf_normalize_tcp_cleanup(cur); 1746 uma_zfree(V_pf_state_z, cur); 1747 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 1748 } 1749 1750 /* 1751 * Called only from pf_purge_thread(), thus serialized. 1752 */ 1753 static u_int 1754 pf_purge_expired_states(u_int i, int maxcheck) 1755 { 1756 struct pf_idhash *ih; 1757 struct pf_state *s; 1758 1759 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1760 1761 /* 1762 * Go through hash and unlink states that expire now. 1763 */ 1764 while (maxcheck > 0) { 1765 ih = &V_pf_idhash[i]; 1766 1767 /* only take the lock if we expect to do work */ 1768 if (!LIST_EMPTY(&ih->states)) { 1769 relock: 1770 PF_HASHROW_LOCK(ih); 1771 LIST_FOREACH(s, &ih->states, entry) { 1772 if (pf_state_expires(s) <= time_uptime) { 1773 V_pf_status.states -= 1774 pf_unlink_state(s, PF_ENTER_LOCKED); 1775 goto relock; 1776 } 1777 s->rule.ptr->rule_flag |= PFRULE_REFS; 1778 if (s->nat_rule.ptr != NULL) 1779 s->nat_rule.ptr->rule_flag |= PFRULE_REFS; 1780 if (s->anchor.ptr != NULL) 1781 s->anchor.ptr->rule_flag |= PFRULE_REFS; 1782 s->kif->pfik_flags |= PFI_IFLAG_REFS; 1783 if (s->rt_kif) 1784 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 1785 } 1786 PF_HASHROW_UNLOCK(ih); 1787 } 1788 1789 /* Return when we hit end of hash. */ 1790 if (++i > pf_hashmask) { 1791 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1792 return (0); 1793 } 1794 1795 maxcheck--; 1796 } 1797 1798 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1799 1800 return (i); 1801 } 1802 1803 static void 1804 pf_purge_unlinked_rules() 1805 { 1806 struct pf_krulequeue tmpq; 1807 struct pf_krule *r, *r1; 1808 1809 /* 1810 * If we have overloading task pending, then we'd 1811 * better skip purging this time. There is a tiny 1812 * probability that overloading task references 1813 * an already unlinked rule. 1814 */ 1815 PF_OVERLOADQ_LOCK(); 1816 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 1817 PF_OVERLOADQ_UNLOCK(); 1818 return; 1819 } 1820 PF_OVERLOADQ_UNLOCK(); 1821 1822 /* 1823 * Do naive mark-and-sweep garbage collecting of old rules. 1824 * Reference flag is raised by pf_purge_expired_states() 1825 * and pf_purge_expired_src_nodes(). 1826 * 1827 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 1828 * use a temporary queue. 1829 */ 1830 TAILQ_INIT(&tmpq); 1831 PF_UNLNKDRULES_LOCK(); 1832 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 1833 if (!(r->rule_flag & PFRULE_REFS)) { 1834 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 1835 TAILQ_INSERT_TAIL(&tmpq, r, entries); 1836 } else 1837 r->rule_flag &= ~PFRULE_REFS; 1838 } 1839 PF_UNLNKDRULES_UNLOCK(); 1840 1841 if (!TAILQ_EMPTY(&tmpq)) { 1842 PF_RULES_WLOCK(); 1843 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 1844 TAILQ_REMOVE(&tmpq, r, entries); 1845 pf_free_rule(r); 1846 } 1847 PF_RULES_WUNLOCK(); 1848 } 1849 } 1850 1851 void 1852 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 1853 { 1854 switch (af) { 1855 #ifdef INET 1856 case AF_INET: { 1857 u_int32_t a = ntohl(addr->addr32[0]); 1858 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 1859 (a>>8)&255, a&255); 1860 if (p) { 1861 p = ntohs(p); 1862 printf(":%u", p); 1863 } 1864 break; 1865 } 1866 #endif /* INET */ 1867 #ifdef INET6 1868 case AF_INET6: { 1869 u_int16_t b; 1870 u_int8_t i, curstart, curend, maxstart, maxend; 1871 curstart = curend = maxstart = maxend = 255; 1872 for (i = 0; i < 8; i++) { 1873 if (!addr->addr16[i]) { 1874 if (curstart == 255) 1875 curstart = i; 1876 curend = i; 1877 } else { 1878 if ((curend - curstart) > 1879 (maxend - maxstart)) { 1880 maxstart = curstart; 1881 maxend = curend; 1882 } 1883 curstart = curend = 255; 1884 } 1885 } 1886 if ((curend - curstart) > 1887 (maxend - maxstart)) { 1888 maxstart = curstart; 1889 maxend = curend; 1890 } 1891 for (i = 0; i < 8; i++) { 1892 if (i >= maxstart && i <= maxend) { 1893 if (i == 0) 1894 printf(":"); 1895 if (i == maxend) 1896 printf(":"); 1897 } else { 1898 b = ntohs(addr->addr16[i]); 1899 printf("%x", b); 1900 if (i < 7) 1901 printf(":"); 1902 } 1903 } 1904 if (p) { 1905 p = ntohs(p); 1906 printf("[%u]", p); 1907 } 1908 break; 1909 } 1910 #endif /* INET6 */ 1911 } 1912 } 1913 1914 void 1915 pf_print_state(struct pf_state *s) 1916 { 1917 pf_print_state_parts(s, NULL, NULL); 1918 } 1919 1920 static void 1921 pf_print_state_parts(struct pf_state *s, 1922 struct pf_state_key *skwp, struct pf_state_key *sksp) 1923 { 1924 struct pf_state_key *skw, *sks; 1925 u_int8_t proto, dir; 1926 1927 /* Do our best to fill these, but they're skipped if NULL */ 1928 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 1929 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 1930 proto = skw ? skw->proto : (sks ? sks->proto : 0); 1931 dir = s ? s->direction : 0; 1932 1933 switch (proto) { 1934 case IPPROTO_IPV4: 1935 printf("IPv4"); 1936 break; 1937 case IPPROTO_IPV6: 1938 printf("IPv6"); 1939 break; 1940 case IPPROTO_TCP: 1941 printf("TCP"); 1942 break; 1943 case IPPROTO_UDP: 1944 printf("UDP"); 1945 break; 1946 case IPPROTO_ICMP: 1947 printf("ICMP"); 1948 break; 1949 case IPPROTO_ICMPV6: 1950 printf("ICMPv6"); 1951 break; 1952 default: 1953 printf("%u", proto); 1954 break; 1955 } 1956 switch (dir) { 1957 case PF_IN: 1958 printf(" in"); 1959 break; 1960 case PF_OUT: 1961 printf(" out"); 1962 break; 1963 } 1964 if (skw) { 1965 printf(" wire: "); 1966 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 1967 printf(" "); 1968 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 1969 } 1970 if (sks) { 1971 printf(" stack: "); 1972 if (sks != skw) { 1973 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 1974 printf(" "); 1975 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 1976 } else 1977 printf("-"); 1978 } 1979 if (s) { 1980 if (proto == IPPROTO_TCP) { 1981 printf(" [lo=%u high=%u win=%u modulator=%u", 1982 s->src.seqlo, s->src.seqhi, 1983 s->src.max_win, s->src.seqdiff); 1984 if (s->src.wscale && s->dst.wscale) 1985 printf(" wscale=%u", 1986 s->src.wscale & PF_WSCALE_MASK); 1987 printf("]"); 1988 printf(" [lo=%u high=%u win=%u modulator=%u", 1989 s->dst.seqlo, s->dst.seqhi, 1990 s->dst.max_win, s->dst.seqdiff); 1991 if (s->src.wscale && s->dst.wscale) 1992 printf(" wscale=%u", 1993 s->dst.wscale & PF_WSCALE_MASK); 1994 printf("]"); 1995 } 1996 printf(" %u:%u", s->src.state, s->dst.state); 1997 } 1998 } 1999 2000 void 2001 pf_print_flags(u_int8_t f) 2002 { 2003 if (f) 2004 printf(" "); 2005 if (f & TH_FIN) 2006 printf("F"); 2007 if (f & TH_SYN) 2008 printf("S"); 2009 if (f & TH_RST) 2010 printf("R"); 2011 if (f & TH_PUSH) 2012 printf("P"); 2013 if (f & TH_ACK) 2014 printf("A"); 2015 if (f & TH_URG) 2016 printf("U"); 2017 if (f & TH_ECE) 2018 printf("E"); 2019 if (f & TH_CWR) 2020 printf("W"); 2021 } 2022 2023 #define PF_SET_SKIP_STEPS(i) \ 2024 do { \ 2025 while (head[i] != cur) { \ 2026 head[i]->skip[i].ptr = cur; \ 2027 head[i] = TAILQ_NEXT(head[i], entries); \ 2028 } \ 2029 } while (0) 2030 2031 void 2032 pf_calc_skip_steps(struct pf_krulequeue *rules) 2033 { 2034 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2035 int i; 2036 2037 cur = TAILQ_FIRST(rules); 2038 prev = cur; 2039 for (i = 0; i < PF_SKIP_COUNT; ++i) 2040 head[i] = cur; 2041 while (cur != NULL) { 2042 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2043 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2044 if (cur->direction != prev->direction) 2045 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2046 if (cur->af != prev->af) 2047 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2048 if (cur->proto != prev->proto) 2049 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2050 if (cur->src.neg != prev->src.neg || 2051 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2052 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2053 if (cur->src.port[0] != prev->src.port[0] || 2054 cur->src.port[1] != prev->src.port[1] || 2055 cur->src.port_op != prev->src.port_op) 2056 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2057 if (cur->dst.neg != prev->dst.neg || 2058 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2059 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2060 if (cur->dst.port[0] != prev->dst.port[0] || 2061 cur->dst.port[1] != prev->dst.port[1] || 2062 cur->dst.port_op != prev->dst.port_op) 2063 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2064 2065 prev = cur; 2066 cur = TAILQ_NEXT(cur, entries); 2067 } 2068 for (i = 0; i < PF_SKIP_COUNT; ++i) 2069 PF_SET_SKIP_STEPS(i); 2070 } 2071 2072 static int 2073 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2074 { 2075 if (aw1->type != aw2->type) 2076 return (1); 2077 switch (aw1->type) { 2078 case PF_ADDR_ADDRMASK: 2079 case PF_ADDR_RANGE: 2080 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2081 return (1); 2082 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2083 return (1); 2084 return (0); 2085 case PF_ADDR_DYNIFTL: 2086 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2087 case PF_ADDR_NOROUTE: 2088 case PF_ADDR_URPFFAILED: 2089 return (0); 2090 case PF_ADDR_TABLE: 2091 return (aw1->p.tbl != aw2->p.tbl); 2092 default: 2093 printf("invalid address type: %d\n", aw1->type); 2094 return (1); 2095 } 2096 } 2097 2098 /** 2099 * Checksum updates are a little complicated because the checksum in the TCP/UDP 2100 * header isn't always a full checksum. In some cases (i.e. output) it's a 2101 * pseudo-header checksum, which is a partial checksum over src/dst IP 2102 * addresses, protocol number and length. 2103 * 2104 * That means we have the following cases: 2105 * * Input or forwarding: we don't have TSO, the checksum fields are full 2106 * checksums, we need to update the checksum whenever we change anything. 2107 * * Output (i.e. the checksum is a pseudo-header checksum): 2108 * x The field being updated is src/dst address or affects the length of 2109 * the packet. We need to update the pseudo-header checksum (note that this 2110 * checksum is not ones' complement). 2111 * x Some other field is being modified (e.g. src/dst port numbers): We 2112 * don't have to update anything. 2113 **/ 2114 u_int16_t 2115 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 2116 { 2117 u_int32_t x; 2118 2119 x = cksum + old - new; 2120 x = (x + (x >> 16)) & 0xffff; 2121 2122 /* optimise: eliminate a branch when not udp */ 2123 if (udp && cksum == 0x0000) 2124 return cksum; 2125 if (udp && x == 0x0000) 2126 x = 0xffff; 2127 2128 return (u_int16_t)(x); 2129 } 2130 2131 static void 2132 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 2133 u_int8_t udp) 2134 { 2135 u_int16_t old = htons(hi ? (*f << 8) : *f); 2136 u_int16_t new = htons(hi ? ( v << 8) : v); 2137 2138 if (*f == v) 2139 return; 2140 2141 *f = v; 2142 2143 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2144 return; 2145 2146 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 2147 } 2148 2149 void 2150 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 2151 bool hi, u_int8_t udp) 2152 { 2153 u_int8_t *fb = (u_int8_t *)f; 2154 u_int8_t *vb = (u_int8_t *)&v; 2155 2156 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2157 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2158 } 2159 2160 void 2161 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 2162 bool hi, u_int8_t udp) 2163 { 2164 u_int8_t *fb = (u_int8_t *)f; 2165 u_int8_t *vb = (u_int8_t *)&v; 2166 2167 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2168 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2169 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2170 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2171 } 2172 2173 u_int16_t 2174 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 2175 u_int16_t new, u_int8_t udp) 2176 { 2177 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2178 return (cksum); 2179 2180 return (pf_cksum_fixup(cksum, old, new, udp)); 2181 } 2182 2183 static void 2184 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 2185 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 2186 sa_family_t af) 2187 { 2188 struct pf_addr ao; 2189 u_int16_t po = *p; 2190 2191 PF_ACPY(&ao, a, af); 2192 PF_ACPY(a, an, af); 2193 2194 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2195 *pc = ~*pc; 2196 2197 *p = pn; 2198 2199 switch (af) { 2200 #ifdef INET 2201 case AF_INET: 2202 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2203 ao.addr16[0], an->addr16[0], 0), 2204 ao.addr16[1], an->addr16[1], 0); 2205 *p = pn; 2206 2207 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 2208 ao.addr16[0], an->addr16[0], u), 2209 ao.addr16[1], an->addr16[1], u); 2210 2211 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2212 break; 2213 #endif /* INET */ 2214 #ifdef INET6 2215 case AF_INET6: 2216 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2217 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2218 pf_cksum_fixup(pf_cksum_fixup(*pc, 2219 ao.addr16[0], an->addr16[0], u), 2220 ao.addr16[1], an->addr16[1], u), 2221 ao.addr16[2], an->addr16[2], u), 2222 ao.addr16[3], an->addr16[3], u), 2223 ao.addr16[4], an->addr16[4], u), 2224 ao.addr16[5], an->addr16[5], u), 2225 ao.addr16[6], an->addr16[6], u), 2226 ao.addr16[7], an->addr16[7], u); 2227 2228 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2229 break; 2230 #endif /* INET6 */ 2231 } 2232 2233 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 2234 CSUM_DELAY_DATA_IPV6)) { 2235 *pc = ~*pc; 2236 if (! *pc) 2237 *pc = 0xffff; 2238 } 2239 } 2240 2241 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2242 void 2243 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2244 { 2245 u_int32_t ao; 2246 2247 memcpy(&ao, a, sizeof(ao)); 2248 memcpy(a, &an, sizeof(u_int32_t)); 2249 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2250 ao % 65536, an % 65536, u); 2251 } 2252 2253 void 2254 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 2255 { 2256 u_int32_t ao; 2257 2258 memcpy(&ao, a, sizeof(ao)); 2259 memcpy(a, &an, sizeof(u_int32_t)); 2260 2261 *c = pf_proto_cksum_fixup(m, 2262 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 2263 ao % 65536, an % 65536, udp); 2264 } 2265 2266 #ifdef INET6 2267 static void 2268 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2269 { 2270 struct pf_addr ao; 2271 2272 PF_ACPY(&ao, a, AF_INET6); 2273 PF_ACPY(a, an, AF_INET6); 2274 2275 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2276 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2277 pf_cksum_fixup(pf_cksum_fixup(*c, 2278 ao.addr16[0], an->addr16[0], u), 2279 ao.addr16[1], an->addr16[1], u), 2280 ao.addr16[2], an->addr16[2], u), 2281 ao.addr16[3], an->addr16[3], u), 2282 ao.addr16[4], an->addr16[4], u), 2283 ao.addr16[5], an->addr16[5], u), 2284 ao.addr16[6], an->addr16[6], u), 2285 ao.addr16[7], an->addr16[7], u); 2286 } 2287 #endif /* INET6 */ 2288 2289 static void 2290 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2291 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2292 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2293 { 2294 struct pf_addr oia, ooa; 2295 2296 PF_ACPY(&oia, ia, af); 2297 if (oa) 2298 PF_ACPY(&ooa, oa, af); 2299 2300 /* Change inner protocol port, fix inner protocol checksum. */ 2301 if (ip != NULL) { 2302 u_int16_t oip = *ip; 2303 u_int32_t opc; 2304 2305 if (pc != NULL) 2306 opc = *pc; 2307 *ip = np; 2308 if (pc != NULL) 2309 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2310 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2311 if (pc != NULL) 2312 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2313 } 2314 /* Change inner ip address, fix inner ip and icmp checksums. */ 2315 PF_ACPY(ia, na, af); 2316 switch (af) { 2317 #ifdef INET 2318 case AF_INET: { 2319 u_int32_t oh2c = *h2c; 2320 2321 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2322 oia.addr16[0], ia->addr16[0], 0), 2323 oia.addr16[1], ia->addr16[1], 0); 2324 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2325 oia.addr16[0], ia->addr16[0], 0), 2326 oia.addr16[1], ia->addr16[1], 0); 2327 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2328 break; 2329 } 2330 #endif /* INET */ 2331 #ifdef INET6 2332 case AF_INET6: 2333 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2334 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2335 pf_cksum_fixup(pf_cksum_fixup(*ic, 2336 oia.addr16[0], ia->addr16[0], u), 2337 oia.addr16[1], ia->addr16[1], u), 2338 oia.addr16[2], ia->addr16[2], u), 2339 oia.addr16[3], ia->addr16[3], u), 2340 oia.addr16[4], ia->addr16[4], u), 2341 oia.addr16[5], ia->addr16[5], u), 2342 oia.addr16[6], ia->addr16[6], u), 2343 oia.addr16[7], ia->addr16[7], u); 2344 break; 2345 #endif /* INET6 */ 2346 } 2347 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2348 if (oa) { 2349 PF_ACPY(oa, na, af); 2350 switch (af) { 2351 #ifdef INET 2352 case AF_INET: 2353 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2354 ooa.addr16[0], oa->addr16[0], 0), 2355 ooa.addr16[1], oa->addr16[1], 0); 2356 break; 2357 #endif /* INET */ 2358 #ifdef INET6 2359 case AF_INET6: 2360 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2361 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2362 pf_cksum_fixup(pf_cksum_fixup(*ic, 2363 ooa.addr16[0], oa->addr16[0], u), 2364 ooa.addr16[1], oa->addr16[1], u), 2365 ooa.addr16[2], oa->addr16[2], u), 2366 ooa.addr16[3], oa->addr16[3], u), 2367 ooa.addr16[4], oa->addr16[4], u), 2368 ooa.addr16[5], oa->addr16[5], u), 2369 ooa.addr16[6], oa->addr16[6], u), 2370 ooa.addr16[7], oa->addr16[7], u); 2371 break; 2372 #endif /* INET6 */ 2373 } 2374 } 2375 } 2376 2377 /* 2378 * Need to modulate the sequence numbers in the TCP SACK option 2379 * (credits to Krzysztof Pfaff for report and patch) 2380 */ 2381 static int 2382 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2383 struct tcphdr *th, struct pf_state_peer *dst) 2384 { 2385 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2386 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2387 int copyback = 0, i, olen; 2388 struct sackblk sack; 2389 2390 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2391 if (hlen < TCPOLEN_SACKLEN || 2392 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2393 return 0; 2394 2395 while (hlen >= TCPOLEN_SACKLEN) { 2396 size_t startoff = opt - opts; 2397 olen = opt[1]; 2398 switch (*opt) { 2399 case TCPOPT_EOL: /* FALLTHROUGH */ 2400 case TCPOPT_NOP: 2401 opt++; 2402 hlen--; 2403 break; 2404 case TCPOPT_SACK: 2405 if (olen > hlen) 2406 olen = hlen; 2407 if (olen >= TCPOLEN_SACKLEN) { 2408 for (i = 2; i + TCPOLEN_SACK <= olen; 2409 i += TCPOLEN_SACK) { 2410 memcpy(&sack, &opt[i], sizeof(sack)); 2411 pf_patch_32_unaligned(m, 2412 &th->th_sum, &sack.start, 2413 htonl(ntohl(sack.start) - dst->seqdiff), 2414 PF_ALGNMNT(startoff), 2415 0); 2416 pf_patch_32_unaligned(m, &th->th_sum, 2417 &sack.end, 2418 htonl(ntohl(sack.end) - dst->seqdiff), 2419 PF_ALGNMNT(startoff), 2420 0); 2421 memcpy(&opt[i], &sack, sizeof(sack)); 2422 } 2423 copyback = 1; 2424 } 2425 /* FALLTHROUGH */ 2426 default: 2427 if (olen < 2) 2428 olen = 2; 2429 hlen -= olen; 2430 opt += olen; 2431 } 2432 } 2433 2434 if (copyback) 2435 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2436 return (copyback); 2437 } 2438 2439 static void 2440 pf_send_tcp(struct mbuf *replyto, const struct pf_krule *r, sa_family_t af, 2441 const struct pf_addr *saddr, const struct pf_addr *daddr, 2442 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2443 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2444 u_int16_t rtag, struct ifnet *ifp) 2445 { 2446 struct pf_send_entry *pfse; 2447 struct mbuf *m; 2448 int len, tlen; 2449 #ifdef INET 2450 struct ip *h = NULL; 2451 #endif /* INET */ 2452 #ifdef INET6 2453 struct ip6_hdr *h6 = NULL; 2454 #endif /* INET6 */ 2455 struct tcphdr *th; 2456 char *opt; 2457 struct pf_mtag *pf_mtag; 2458 2459 len = 0; 2460 th = NULL; 2461 2462 /* maximum segment size tcp option */ 2463 tlen = sizeof(struct tcphdr); 2464 if (mss) 2465 tlen += 4; 2466 2467 switch (af) { 2468 #ifdef INET 2469 case AF_INET: 2470 len = sizeof(struct ip) + tlen; 2471 break; 2472 #endif /* INET */ 2473 #ifdef INET6 2474 case AF_INET6: 2475 len = sizeof(struct ip6_hdr) + tlen; 2476 break; 2477 #endif /* INET6 */ 2478 default: 2479 panic("%s: unsupported af %d", __func__, af); 2480 } 2481 2482 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2483 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2484 if (pfse == NULL) 2485 return; 2486 m = m_gethdr(M_NOWAIT, MT_DATA); 2487 if (m == NULL) { 2488 free(pfse, M_PFTEMP); 2489 return; 2490 } 2491 #ifdef MAC 2492 mac_netinet_firewall_send(m); 2493 #endif 2494 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2495 free(pfse, M_PFTEMP); 2496 m_freem(m); 2497 return; 2498 } 2499 if (tag) 2500 m->m_flags |= M_SKIP_FIREWALL; 2501 pf_mtag->tag = rtag; 2502 2503 if (r != NULL && r->rtableid >= 0) 2504 M_SETFIB(m, r->rtableid); 2505 2506 #ifdef ALTQ 2507 if (r != NULL && r->qid) { 2508 pf_mtag->qid = r->qid; 2509 2510 /* add hints for ecn */ 2511 pf_mtag->hdr = mtod(m, struct ip *); 2512 } 2513 #endif /* ALTQ */ 2514 m->m_data += max_linkhdr; 2515 m->m_pkthdr.len = m->m_len = len; 2516 m->m_pkthdr.rcvif = NULL; 2517 bzero(m->m_data, len); 2518 switch (af) { 2519 #ifdef INET 2520 case AF_INET: 2521 h = mtod(m, struct ip *); 2522 2523 /* IP header fields included in the TCP checksum */ 2524 h->ip_p = IPPROTO_TCP; 2525 h->ip_len = htons(tlen); 2526 h->ip_src.s_addr = saddr->v4.s_addr; 2527 h->ip_dst.s_addr = daddr->v4.s_addr; 2528 2529 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2530 break; 2531 #endif /* INET */ 2532 #ifdef INET6 2533 case AF_INET6: 2534 h6 = mtod(m, struct ip6_hdr *); 2535 2536 /* IP header fields included in the TCP checksum */ 2537 h6->ip6_nxt = IPPROTO_TCP; 2538 h6->ip6_plen = htons(tlen); 2539 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2540 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2541 2542 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2543 break; 2544 #endif /* INET6 */ 2545 } 2546 2547 /* TCP header */ 2548 th->th_sport = sport; 2549 th->th_dport = dport; 2550 th->th_seq = htonl(seq); 2551 th->th_ack = htonl(ack); 2552 th->th_off = tlen >> 2; 2553 th->th_flags = flags; 2554 th->th_win = htons(win); 2555 2556 if (mss) { 2557 opt = (char *)(th + 1); 2558 opt[0] = TCPOPT_MAXSEG; 2559 opt[1] = 4; 2560 HTONS(mss); 2561 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2562 } 2563 2564 switch (af) { 2565 #ifdef INET 2566 case AF_INET: 2567 /* TCP checksum */ 2568 th->th_sum = in_cksum(m, len); 2569 2570 /* Finish the IP header */ 2571 h->ip_v = 4; 2572 h->ip_hl = sizeof(*h) >> 2; 2573 h->ip_tos = IPTOS_LOWDELAY; 2574 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 2575 h->ip_len = htons(len); 2576 h->ip_ttl = ttl ? ttl : V_ip_defttl; 2577 h->ip_sum = 0; 2578 2579 pfse->pfse_type = PFSE_IP; 2580 break; 2581 #endif /* INET */ 2582 #ifdef INET6 2583 case AF_INET6: 2584 /* TCP checksum */ 2585 th->th_sum = in6_cksum(m, IPPROTO_TCP, 2586 sizeof(struct ip6_hdr), tlen); 2587 2588 h6->ip6_vfc |= IPV6_VERSION; 2589 h6->ip6_hlim = IPV6_DEFHLIM; 2590 2591 pfse->pfse_type = PFSE_IP6; 2592 break; 2593 #endif /* INET6 */ 2594 } 2595 pfse->pfse_m = m; 2596 pf_send(pfse); 2597 } 2598 2599 static void 2600 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 2601 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 2602 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 2603 u_short *reason) 2604 { 2605 struct pf_addr * const saddr = pd->src; 2606 struct pf_addr * const daddr = pd->dst; 2607 sa_family_t af = pd->af; 2608 2609 /* undo NAT changes, if they have taken place */ 2610 if (nr != NULL) { 2611 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 2612 PF_ACPY(daddr, &sk->addr[pd->didx], af); 2613 if (pd->sport) 2614 *pd->sport = sk->port[pd->sidx]; 2615 if (pd->dport) 2616 *pd->dport = sk->port[pd->didx]; 2617 if (pd->proto_sum) 2618 *pd->proto_sum = bproto_sum; 2619 if (pd->ip_sum) 2620 *pd->ip_sum = bip_sum; 2621 m_copyback(m, off, hdrlen, pd->hdr.any); 2622 } 2623 if (pd->proto == IPPROTO_TCP && 2624 ((r->rule_flag & PFRULE_RETURNRST) || 2625 (r->rule_flag & PFRULE_RETURN)) && 2626 !(th->th_flags & TH_RST)) { 2627 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 2628 int len = 0; 2629 #ifdef INET 2630 struct ip *h4; 2631 #endif 2632 #ifdef INET6 2633 struct ip6_hdr *h6; 2634 #endif 2635 2636 switch (af) { 2637 #ifdef INET 2638 case AF_INET: 2639 h4 = mtod(m, struct ip *); 2640 len = ntohs(h4->ip_len) - off; 2641 break; 2642 #endif 2643 #ifdef INET6 2644 case AF_INET6: 2645 h6 = mtod(m, struct ip6_hdr *); 2646 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 2647 break; 2648 #endif 2649 } 2650 2651 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 2652 REASON_SET(reason, PFRES_PROTCKSUM); 2653 else { 2654 if (th->th_flags & TH_SYN) 2655 ack++; 2656 if (th->th_flags & TH_FIN) 2657 ack++; 2658 pf_send_tcp(m, r, af, pd->dst, 2659 pd->src, th->th_dport, th->th_sport, 2660 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 2661 r->return_ttl, 1, 0, kif->pfik_ifp); 2662 } 2663 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 2664 r->return_icmp) 2665 pf_send_icmp(m, r->return_icmp >> 8, 2666 r->return_icmp & 255, af, r); 2667 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 2668 r->return_icmp6) 2669 pf_send_icmp(m, r->return_icmp6 >> 8, 2670 r->return_icmp6 & 255, af, r); 2671 } 2672 2673 static int 2674 pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio) 2675 { 2676 struct m_tag *mtag; 2677 2678 KASSERT(prio <= PF_PRIO_MAX, 2679 ("%s with invalid pcp", __func__)); 2680 2681 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL); 2682 if (mtag == NULL) { 2683 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT, 2684 sizeof(uint8_t), M_NOWAIT); 2685 if (mtag == NULL) 2686 return (ENOMEM); 2687 m_tag_prepend(m, mtag); 2688 } 2689 2690 *(uint8_t *)(mtag + 1) = prio; 2691 return (0); 2692 } 2693 2694 static int 2695 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 2696 { 2697 struct m_tag *mtag; 2698 u_int8_t mpcp; 2699 2700 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 2701 if (mtag == NULL) 2702 return (0); 2703 2704 if (prio == PF_PRIO_ZERO) 2705 prio = 0; 2706 2707 mpcp = *(uint8_t *)(mtag + 1); 2708 2709 return (mpcp == prio); 2710 } 2711 2712 static void 2713 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 2714 struct pf_krule *r) 2715 { 2716 struct pf_send_entry *pfse; 2717 struct mbuf *m0; 2718 struct pf_mtag *pf_mtag; 2719 2720 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2721 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2722 if (pfse == NULL) 2723 return; 2724 2725 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 2726 free(pfse, M_PFTEMP); 2727 return; 2728 } 2729 2730 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 2731 free(pfse, M_PFTEMP); 2732 return; 2733 } 2734 /* XXX: revisit */ 2735 m0->m_flags |= M_SKIP_FIREWALL; 2736 2737 if (r->rtableid >= 0) 2738 M_SETFIB(m0, r->rtableid); 2739 2740 #ifdef ALTQ 2741 if (r->qid) { 2742 pf_mtag->qid = r->qid; 2743 /* add hints for ecn */ 2744 pf_mtag->hdr = mtod(m0, struct ip *); 2745 } 2746 #endif /* ALTQ */ 2747 2748 switch (af) { 2749 #ifdef INET 2750 case AF_INET: 2751 pfse->pfse_type = PFSE_ICMP; 2752 break; 2753 #endif /* INET */ 2754 #ifdef INET6 2755 case AF_INET6: 2756 pfse->pfse_type = PFSE_ICMP6; 2757 break; 2758 #endif /* INET6 */ 2759 } 2760 pfse->pfse_m = m0; 2761 pfse->icmpopts.type = type; 2762 pfse->icmpopts.code = code; 2763 pf_send(pfse); 2764 } 2765 2766 /* 2767 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 2768 * If n is 0, they match if they are equal. If n is != 0, they match if they 2769 * are different. 2770 */ 2771 int 2772 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 2773 struct pf_addr *b, sa_family_t af) 2774 { 2775 int match = 0; 2776 2777 switch (af) { 2778 #ifdef INET 2779 case AF_INET: 2780 if ((a->addr32[0] & m->addr32[0]) == 2781 (b->addr32[0] & m->addr32[0])) 2782 match++; 2783 break; 2784 #endif /* INET */ 2785 #ifdef INET6 2786 case AF_INET6: 2787 if (((a->addr32[0] & m->addr32[0]) == 2788 (b->addr32[0] & m->addr32[0])) && 2789 ((a->addr32[1] & m->addr32[1]) == 2790 (b->addr32[1] & m->addr32[1])) && 2791 ((a->addr32[2] & m->addr32[2]) == 2792 (b->addr32[2] & m->addr32[2])) && 2793 ((a->addr32[3] & m->addr32[3]) == 2794 (b->addr32[3] & m->addr32[3]))) 2795 match++; 2796 break; 2797 #endif /* INET6 */ 2798 } 2799 if (match) { 2800 if (n) 2801 return (0); 2802 else 2803 return (1); 2804 } else { 2805 if (n) 2806 return (1); 2807 else 2808 return (0); 2809 } 2810 } 2811 2812 /* 2813 * Return 1 if b <= a <= e, otherwise return 0. 2814 */ 2815 int 2816 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 2817 struct pf_addr *a, sa_family_t af) 2818 { 2819 switch (af) { 2820 #ifdef INET 2821 case AF_INET: 2822 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 2823 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 2824 return (0); 2825 break; 2826 #endif /* INET */ 2827 #ifdef INET6 2828 case AF_INET6: { 2829 int i; 2830 2831 /* check a >= b */ 2832 for (i = 0; i < 4; ++i) 2833 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 2834 break; 2835 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 2836 return (0); 2837 /* check a <= e */ 2838 for (i = 0; i < 4; ++i) 2839 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 2840 break; 2841 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 2842 return (0); 2843 break; 2844 } 2845 #endif /* INET6 */ 2846 } 2847 return (1); 2848 } 2849 2850 static int 2851 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 2852 { 2853 switch (op) { 2854 case PF_OP_IRG: 2855 return ((p > a1) && (p < a2)); 2856 case PF_OP_XRG: 2857 return ((p < a1) || (p > a2)); 2858 case PF_OP_RRG: 2859 return ((p >= a1) && (p <= a2)); 2860 case PF_OP_EQ: 2861 return (p == a1); 2862 case PF_OP_NE: 2863 return (p != a1); 2864 case PF_OP_LT: 2865 return (p < a1); 2866 case PF_OP_LE: 2867 return (p <= a1); 2868 case PF_OP_GT: 2869 return (p > a1); 2870 case PF_OP_GE: 2871 return (p >= a1); 2872 } 2873 return (0); /* never reached */ 2874 } 2875 2876 int 2877 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 2878 { 2879 NTOHS(a1); 2880 NTOHS(a2); 2881 NTOHS(p); 2882 return (pf_match(op, a1, a2, p)); 2883 } 2884 2885 static int 2886 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 2887 { 2888 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2889 return (0); 2890 return (pf_match(op, a1, a2, u)); 2891 } 2892 2893 static int 2894 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 2895 { 2896 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2897 return (0); 2898 return (pf_match(op, a1, a2, g)); 2899 } 2900 2901 int 2902 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 2903 { 2904 if (*tag == -1) 2905 *tag = mtag; 2906 2907 return ((!r->match_tag_not && r->match_tag == *tag) || 2908 (r->match_tag_not && r->match_tag != *tag)); 2909 } 2910 2911 int 2912 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 2913 { 2914 2915 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 2916 2917 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 2918 return (ENOMEM); 2919 2920 pd->pf_mtag->tag = tag; 2921 2922 return (0); 2923 } 2924 2925 #define PF_ANCHOR_STACKSIZE 32 2926 struct pf_kanchor_stackframe { 2927 struct pf_kruleset *rs; 2928 struct pf_krule *r; /* XXX: + match bit */ 2929 struct pf_kanchor *child; 2930 }; 2931 2932 /* 2933 * XXX: We rely on malloc(9) returning pointer aligned addresses. 2934 */ 2935 #define PF_ANCHORSTACK_MATCH 0x00000001 2936 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 2937 2938 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 2939 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 2940 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 2941 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 2942 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 2943 } while (0) 2944 2945 void 2946 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 2947 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 2948 int *match) 2949 { 2950 struct pf_kanchor_stackframe *f; 2951 2952 PF_RULES_RASSERT(); 2953 2954 if (match) 2955 *match = 0; 2956 if (*depth >= PF_ANCHOR_STACKSIZE) { 2957 printf("%s: anchor stack overflow on %s\n", 2958 __func__, (*r)->anchor->name); 2959 *r = TAILQ_NEXT(*r, entries); 2960 return; 2961 } else if (*depth == 0 && a != NULL) 2962 *a = *r; 2963 f = stack + (*depth)++; 2964 f->rs = *rs; 2965 f->r = *r; 2966 if ((*r)->anchor_wildcard) { 2967 struct pf_kanchor_node *parent = &(*r)->anchor->children; 2968 2969 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 2970 *r = NULL; 2971 return; 2972 } 2973 *rs = &f->child->ruleset; 2974 } else { 2975 f->child = NULL; 2976 *rs = &(*r)->anchor->ruleset; 2977 } 2978 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2979 } 2980 2981 int 2982 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 2983 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 2984 int *match) 2985 { 2986 struct pf_kanchor_stackframe *f; 2987 struct pf_krule *fr; 2988 int quick = 0; 2989 2990 PF_RULES_RASSERT(); 2991 2992 do { 2993 if (*depth <= 0) 2994 break; 2995 f = stack + *depth - 1; 2996 fr = PF_ANCHOR_RULE(f); 2997 if (f->child != NULL) { 2998 struct pf_kanchor_node *parent; 2999 3000 /* 3001 * This block traverses through 3002 * a wildcard anchor. 3003 */ 3004 parent = &fr->anchor->children; 3005 if (match != NULL && *match) { 3006 /* 3007 * If any of "*" matched, then 3008 * "foo/ *" matched, mark frame 3009 * appropriately. 3010 */ 3011 PF_ANCHOR_SET_MATCH(f); 3012 *match = 0; 3013 } 3014 f->child = RB_NEXT(pf_kanchor_node, parent, f->child); 3015 if (f->child != NULL) { 3016 *rs = &f->child->ruleset; 3017 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3018 if (*r == NULL) 3019 continue; 3020 else 3021 break; 3022 } 3023 } 3024 (*depth)--; 3025 if (*depth == 0 && a != NULL) 3026 *a = NULL; 3027 *rs = f->rs; 3028 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 3029 quick = fr->quick; 3030 *r = TAILQ_NEXT(fr, entries); 3031 } while (*r == NULL); 3032 3033 return (quick); 3034 } 3035 3036 #ifdef INET6 3037 void 3038 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3039 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3040 { 3041 switch (af) { 3042 #ifdef INET 3043 case AF_INET: 3044 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3045 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3046 break; 3047 #endif /* INET */ 3048 case AF_INET6: 3049 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3050 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3051 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3052 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3053 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3054 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3055 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3056 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3057 break; 3058 } 3059 } 3060 3061 void 3062 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3063 { 3064 switch (af) { 3065 #ifdef INET 3066 case AF_INET: 3067 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3068 break; 3069 #endif /* INET */ 3070 case AF_INET6: 3071 if (addr->addr32[3] == 0xffffffff) { 3072 addr->addr32[3] = 0; 3073 if (addr->addr32[2] == 0xffffffff) { 3074 addr->addr32[2] = 0; 3075 if (addr->addr32[1] == 0xffffffff) { 3076 addr->addr32[1] = 0; 3077 addr->addr32[0] = 3078 htonl(ntohl(addr->addr32[0]) + 1); 3079 } else 3080 addr->addr32[1] = 3081 htonl(ntohl(addr->addr32[1]) + 1); 3082 } else 3083 addr->addr32[2] = 3084 htonl(ntohl(addr->addr32[2]) + 1); 3085 } else 3086 addr->addr32[3] = 3087 htonl(ntohl(addr->addr32[3]) + 1); 3088 break; 3089 } 3090 } 3091 #endif /* INET6 */ 3092 3093 int 3094 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m) 3095 { 3096 struct pf_addr *saddr, *daddr; 3097 u_int16_t sport, dport; 3098 struct inpcbinfo *pi; 3099 struct inpcb *inp; 3100 3101 pd->lookup.uid = UID_MAX; 3102 pd->lookup.gid = GID_MAX; 3103 3104 switch (pd->proto) { 3105 case IPPROTO_TCP: 3106 if (pd->hdr.tcp == NULL) 3107 return (-1); 3108 sport = pd->hdr.tcp->th_sport; 3109 dport = pd->hdr.tcp->th_dport; 3110 pi = &V_tcbinfo; 3111 break; 3112 case IPPROTO_UDP: 3113 if (pd->hdr.udp == NULL) 3114 return (-1); 3115 sport = pd->hdr.udp->uh_sport; 3116 dport = pd->hdr.udp->uh_dport; 3117 pi = &V_udbinfo; 3118 break; 3119 default: 3120 return (-1); 3121 } 3122 if (direction == PF_IN) { 3123 saddr = pd->src; 3124 daddr = pd->dst; 3125 } else { 3126 u_int16_t p; 3127 3128 p = sport; 3129 sport = dport; 3130 dport = p; 3131 saddr = pd->dst; 3132 daddr = pd->src; 3133 } 3134 switch (pd->af) { 3135 #ifdef INET 3136 case AF_INET: 3137 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 3138 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3139 if (inp == NULL) { 3140 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 3141 daddr->v4, dport, INPLOOKUP_WILDCARD | 3142 INPLOOKUP_RLOCKPCB, NULL, m); 3143 if (inp == NULL) 3144 return (-1); 3145 } 3146 break; 3147 #endif /* INET */ 3148 #ifdef INET6 3149 case AF_INET6: 3150 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 3151 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3152 if (inp == NULL) { 3153 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 3154 &daddr->v6, dport, INPLOOKUP_WILDCARD | 3155 INPLOOKUP_RLOCKPCB, NULL, m); 3156 if (inp == NULL) 3157 return (-1); 3158 } 3159 break; 3160 #endif /* INET6 */ 3161 3162 default: 3163 return (-1); 3164 } 3165 INP_RLOCK_ASSERT(inp); 3166 pd->lookup.uid = inp->inp_cred->cr_uid; 3167 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3168 INP_RUNLOCK(inp); 3169 3170 return (1); 3171 } 3172 3173 static u_int8_t 3174 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3175 { 3176 int hlen; 3177 u_int8_t hdr[60]; 3178 u_int8_t *opt, optlen; 3179 u_int8_t wscale = 0; 3180 3181 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3182 if (hlen <= sizeof(struct tcphdr)) 3183 return (0); 3184 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3185 return (0); 3186 opt = hdr + sizeof(struct tcphdr); 3187 hlen -= sizeof(struct tcphdr); 3188 while (hlen >= 3) { 3189 switch (*opt) { 3190 case TCPOPT_EOL: 3191 case TCPOPT_NOP: 3192 ++opt; 3193 --hlen; 3194 break; 3195 case TCPOPT_WINDOW: 3196 wscale = opt[2]; 3197 if (wscale > TCP_MAX_WINSHIFT) 3198 wscale = TCP_MAX_WINSHIFT; 3199 wscale |= PF_WSCALE_FLAG; 3200 /* FALLTHROUGH */ 3201 default: 3202 optlen = opt[1]; 3203 if (optlen < 2) 3204 optlen = 2; 3205 hlen -= optlen; 3206 opt += optlen; 3207 break; 3208 } 3209 } 3210 return (wscale); 3211 } 3212 3213 static u_int16_t 3214 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3215 { 3216 int hlen; 3217 u_int8_t hdr[60]; 3218 u_int8_t *opt, optlen; 3219 u_int16_t mss = V_tcp_mssdflt; 3220 3221 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3222 if (hlen <= sizeof(struct tcphdr)) 3223 return (0); 3224 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3225 return (0); 3226 opt = hdr + sizeof(struct tcphdr); 3227 hlen -= sizeof(struct tcphdr); 3228 while (hlen >= TCPOLEN_MAXSEG) { 3229 switch (*opt) { 3230 case TCPOPT_EOL: 3231 case TCPOPT_NOP: 3232 ++opt; 3233 --hlen; 3234 break; 3235 case TCPOPT_MAXSEG: 3236 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 3237 NTOHS(mss); 3238 /* FALLTHROUGH */ 3239 default: 3240 optlen = opt[1]; 3241 if (optlen < 2) 3242 optlen = 2; 3243 hlen -= optlen; 3244 opt += optlen; 3245 break; 3246 } 3247 } 3248 return (mss); 3249 } 3250 3251 static u_int16_t 3252 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 3253 { 3254 struct nhop_object *nh; 3255 #ifdef INET6 3256 struct in6_addr dst6; 3257 uint32_t scopeid; 3258 #endif /* INET6 */ 3259 int hlen = 0; 3260 uint16_t mss = 0; 3261 3262 NET_EPOCH_ASSERT(); 3263 3264 switch (af) { 3265 #ifdef INET 3266 case AF_INET: 3267 hlen = sizeof(struct ip); 3268 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 3269 if (nh != NULL) 3270 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3271 break; 3272 #endif /* INET */ 3273 #ifdef INET6 3274 case AF_INET6: 3275 hlen = sizeof(struct ip6_hdr); 3276 in6_splitscope(&addr->v6, &dst6, &scopeid); 3277 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 3278 if (nh != NULL) 3279 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3280 break; 3281 #endif /* INET6 */ 3282 } 3283 3284 mss = max(V_tcp_mssdflt, mss); 3285 mss = min(mss, offer); 3286 mss = max(mss, 64); /* sanity - at least max opt space */ 3287 return (mss); 3288 } 3289 3290 static u_int32_t 3291 pf_tcp_iss(struct pf_pdesc *pd) 3292 { 3293 MD5_CTX ctx; 3294 u_int32_t digest[4]; 3295 3296 if (V_pf_tcp_secret_init == 0) { 3297 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 3298 MD5Init(&V_pf_tcp_secret_ctx); 3299 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 3300 sizeof(V_pf_tcp_secret)); 3301 V_pf_tcp_secret_init = 1; 3302 } 3303 3304 ctx = V_pf_tcp_secret_ctx; 3305 3306 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short)); 3307 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short)); 3308 if (pd->af == AF_INET6) { 3309 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 3310 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 3311 } else { 3312 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 3313 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 3314 } 3315 MD5Final((u_char *)digest, &ctx); 3316 V_pf_tcp_iss_off += 4096; 3317 #define ISN_RANDOM_INCREMENT (4096 - 1) 3318 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 3319 V_pf_tcp_iss_off); 3320 #undef ISN_RANDOM_INCREMENT 3321 } 3322 3323 static int 3324 pf_test_rule(struct pf_krule **rm, struct pf_state **sm, int direction, 3325 struct pfi_kkif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, 3326 struct pf_krule **am, struct pf_kruleset **rsm, struct inpcb *inp) 3327 { 3328 struct pf_krule *nr = NULL; 3329 struct pf_addr * const saddr = pd->src; 3330 struct pf_addr * const daddr = pd->dst; 3331 sa_family_t af = pd->af; 3332 struct pf_krule *r, *a = NULL; 3333 struct pf_kruleset *ruleset = NULL; 3334 struct pf_ksrc_node *nsn = NULL; 3335 struct tcphdr *th = pd->hdr.tcp; 3336 struct pf_state_key *sk = NULL, *nk = NULL; 3337 u_short reason; 3338 int rewrite = 0, hdrlen = 0; 3339 int tag = -1, rtableid = -1; 3340 int asd = 0; 3341 int match = 0; 3342 int state_icmp = 0; 3343 u_int16_t sport = 0, dport = 0; 3344 u_int16_t bproto_sum = 0, bip_sum = 0; 3345 u_int8_t icmptype = 0, icmpcode = 0; 3346 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3347 3348 PF_RULES_RASSERT(); 3349 3350 if (inp != NULL) { 3351 INP_LOCK_ASSERT(inp); 3352 pd->lookup.uid = inp->inp_cred->cr_uid; 3353 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3354 pd->lookup.done = 1; 3355 } 3356 3357 switch (pd->proto) { 3358 case IPPROTO_TCP: 3359 sport = th->th_sport; 3360 dport = th->th_dport; 3361 hdrlen = sizeof(*th); 3362 break; 3363 case IPPROTO_UDP: 3364 sport = pd->hdr.udp->uh_sport; 3365 dport = pd->hdr.udp->uh_dport; 3366 hdrlen = sizeof(*pd->hdr.udp); 3367 break; 3368 #ifdef INET 3369 case IPPROTO_ICMP: 3370 if (pd->af != AF_INET) 3371 break; 3372 sport = dport = pd->hdr.icmp->icmp_id; 3373 hdrlen = sizeof(*pd->hdr.icmp); 3374 icmptype = pd->hdr.icmp->icmp_type; 3375 icmpcode = pd->hdr.icmp->icmp_code; 3376 3377 if (icmptype == ICMP_UNREACH || 3378 icmptype == ICMP_SOURCEQUENCH || 3379 icmptype == ICMP_REDIRECT || 3380 icmptype == ICMP_TIMXCEED || 3381 icmptype == ICMP_PARAMPROB) 3382 state_icmp++; 3383 break; 3384 #endif /* INET */ 3385 #ifdef INET6 3386 case IPPROTO_ICMPV6: 3387 if (af != AF_INET6) 3388 break; 3389 sport = dport = pd->hdr.icmp6->icmp6_id; 3390 hdrlen = sizeof(*pd->hdr.icmp6); 3391 icmptype = pd->hdr.icmp6->icmp6_type; 3392 icmpcode = pd->hdr.icmp6->icmp6_code; 3393 3394 if (icmptype == ICMP6_DST_UNREACH || 3395 icmptype == ICMP6_PACKET_TOO_BIG || 3396 icmptype == ICMP6_TIME_EXCEEDED || 3397 icmptype == ICMP6_PARAM_PROB) 3398 state_icmp++; 3399 break; 3400 #endif /* INET6 */ 3401 default: 3402 sport = dport = hdrlen = 0; 3403 break; 3404 } 3405 3406 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3407 3408 /* check packet for BINAT/NAT/RDR */ 3409 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk, 3410 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 3411 KASSERT(sk != NULL, ("%s: null sk", __func__)); 3412 KASSERT(nk != NULL, ("%s: null nk", __func__)); 3413 3414 if (pd->ip_sum) 3415 bip_sum = *pd->ip_sum; 3416 3417 switch (pd->proto) { 3418 case IPPROTO_TCP: 3419 bproto_sum = th->th_sum; 3420 pd->proto_sum = &th->th_sum; 3421 3422 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3423 nk->port[pd->sidx] != sport) { 3424 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 3425 &th->th_sum, &nk->addr[pd->sidx], 3426 nk->port[pd->sidx], 0, af); 3427 pd->sport = &th->th_sport; 3428 sport = th->th_sport; 3429 } 3430 3431 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3432 nk->port[pd->didx] != dport) { 3433 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 3434 &th->th_sum, &nk->addr[pd->didx], 3435 nk->port[pd->didx], 0, af); 3436 dport = th->th_dport; 3437 pd->dport = &th->th_dport; 3438 } 3439 rewrite++; 3440 break; 3441 case IPPROTO_UDP: 3442 bproto_sum = pd->hdr.udp->uh_sum; 3443 pd->proto_sum = &pd->hdr.udp->uh_sum; 3444 3445 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3446 nk->port[pd->sidx] != sport) { 3447 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport, 3448 pd->ip_sum, &pd->hdr.udp->uh_sum, 3449 &nk->addr[pd->sidx], 3450 nk->port[pd->sidx], 1, af); 3451 sport = pd->hdr.udp->uh_sport; 3452 pd->sport = &pd->hdr.udp->uh_sport; 3453 } 3454 3455 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3456 nk->port[pd->didx] != dport) { 3457 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport, 3458 pd->ip_sum, &pd->hdr.udp->uh_sum, 3459 &nk->addr[pd->didx], 3460 nk->port[pd->didx], 1, af); 3461 dport = pd->hdr.udp->uh_dport; 3462 pd->dport = &pd->hdr.udp->uh_dport; 3463 } 3464 rewrite++; 3465 break; 3466 #ifdef INET 3467 case IPPROTO_ICMP: 3468 nk->port[0] = nk->port[1]; 3469 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 3470 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 3471 nk->addr[pd->sidx].v4.s_addr, 0); 3472 3473 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 3474 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 3475 nk->addr[pd->didx].v4.s_addr, 0); 3476 3477 if (nk->port[1] != pd->hdr.icmp->icmp_id) { 3478 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup( 3479 pd->hdr.icmp->icmp_cksum, sport, 3480 nk->port[1], 0); 3481 pd->hdr.icmp->icmp_id = nk->port[1]; 3482 pd->sport = &pd->hdr.icmp->icmp_id; 3483 } 3484 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 3485 break; 3486 #endif /* INET */ 3487 #ifdef INET6 3488 case IPPROTO_ICMPV6: 3489 nk->port[0] = nk->port[1]; 3490 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 3491 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum, 3492 &nk->addr[pd->sidx], 0); 3493 3494 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 3495 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum, 3496 &nk->addr[pd->didx], 0); 3497 rewrite++; 3498 break; 3499 #endif /* INET */ 3500 default: 3501 switch (af) { 3502 #ifdef INET 3503 case AF_INET: 3504 if (PF_ANEQ(saddr, 3505 &nk->addr[pd->sidx], AF_INET)) 3506 pf_change_a(&saddr->v4.s_addr, 3507 pd->ip_sum, 3508 nk->addr[pd->sidx].v4.s_addr, 0); 3509 3510 if (PF_ANEQ(daddr, 3511 &nk->addr[pd->didx], AF_INET)) 3512 pf_change_a(&daddr->v4.s_addr, 3513 pd->ip_sum, 3514 nk->addr[pd->didx].v4.s_addr, 0); 3515 break; 3516 #endif /* INET */ 3517 #ifdef INET6 3518 case AF_INET6: 3519 if (PF_ANEQ(saddr, 3520 &nk->addr[pd->sidx], AF_INET6)) 3521 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 3522 3523 if (PF_ANEQ(daddr, 3524 &nk->addr[pd->didx], AF_INET6)) 3525 PF_ACPY(daddr, &nk->addr[pd->didx], af); 3526 break; 3527 #endif /* INET */ 3528 } 3529 break; 3530 } 3531 if (nr->natpass) 3532 r = NULL; 3533 pd->nat_rule = nr; 3534 } 3535 3536 while (r != NULL) { 3537 counter_u64_add(r->evaluations, 1); 3538 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 3539 r = r->skip[PF_SKIP_IFP].ptr; 3540 else if (r->direction && r->direction != direction) 3541 r = r->skip[PF_SKIP_DIR].ptr; 3542 else if (r->af && r->af != af) 3543 r = r->skip[PF_SKIP_AF].ptr; 3544 else if (r->proto && r->proto != pd->proto) 3545 r = r->skip[PF_SKIP_PROTO].ptr; 3546 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 3547 r->src.neg, kif, M_GETFIB(m))) 3548 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3549 /* tcp/udp only. port_op always 0 in other cases */ 3550 else if (r->src.port_op && !pf_match_port(r->src.port_op, 3551 r->src.port[0], r->src.port[1], sport)) 3552 r = r->skip[PF_SKIP_SRC_PORT].ptr; 3553 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 3554 r->dst.neg, NULL, M_GETFIB(m))) 3555 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3556 /* tcp/udp only. port_op always 0 in other cases */ 3557 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 3558 r->dst.port[0], r->dst.port[1], dport)) 3559 r = r->skip[PF_SKIP_DST_PORT].ptr; 3560 /* icmp only. type always 0 in other cases */ 3561 else if (r->type && r->type != icmptype + 1) 3562 r = TAILQ_NEXT(r, entries); 3563 /* icmp only. type always 0 in other cases */ 3564 else if (r->code && r->code != icmpcode + 1) 3565 r = TAILQ_NEXT(r, entries); 3566 else if (r->tos && !(r->tos == pd->tos)) 3567 r = TAILQ_NEXT(r, entries); 3568 else if (r->rule_flag & PFRULE_FRAGMENT) 3569 r = TAILQ_NEXT(r, entries); 3570 else if (pd->proto == IPPROTO_TCP && 3571 (r->flagset & th->th_flags) != r->flags) 3572 r = TAILQ_NEXT(r, entries); 3573 /* tcp/udp only. uid.op always 0 in other cases */ 3574 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 3575 pf_socket_lookup(direction, pd, m), 1)) && 3576 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 3577 pd->lookup.uid)) 3578 r = TAILQ_NEXT(r, entries); 3579 /* tcp/udp only. gid.op always 0 in other cases */ 3580 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 3581 pf_socket_lookup(direction, pd, m), 1)) && 3582 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 3583 pd->lookup.gid)) 3584 r = TAILQ_NEXT(r, entries); 3585 else if (r->prio && 3586 !pf_match_ieee8021q_pcp(r->prio, m)) 3587 r = TAILQ_NEXT(r, entries); 3588 else if (r->prob && 3589 r->prob <= arc4random()) 3590 r = TAILQ_NEXT(r, entries); 3591 else if (r->match_tag && !pf_match_tag(m, r, &tag, 3592 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 3593 r = TAILQ_NEXT(r, entries); 3594 else if (r->os_fingerprint != PF_OSFP_ANY && 3595 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 3596 pf_osfp_fingerprint(pd, m, off, th), 3597 r->os_fingerprint))) 3598 r = TAILQ_NEXT(r, entries); 3599 else { 3600 if (r->tag) 3601 tag = r->tag; 3602 if (r->rtableid >= 0) 3603 rtableid = r->rtableid; 3604 if (r->anchor == NULL) { 3605 match = 1; 3606 *rm = r; 3607 *am = a; 3608 *rsm = ruleset; 3609 if ((*rm)->quick) 3610 break; 3611 r = TAILQ_NEXT(r, entries); 3612 } else 3613 pf_step_into_anchor(anchor_stack, &asd, 3614 &ruleset, PF_RULESET_FILTER, &r, &a, 3615 &match); 3616 } 3617 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 3618 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 3619 break; 3620 } 3621 r = *rm; 3622 a = *am; 3623 ruleset = *rsm; 3624 3625 REASON_SET(&reason, PFRES_MATCH); 3626 3627 if (r->log || (nr != NULL && nr->log)) { 3628 if (rewrite) 3629 m_copyback(m, off, hdrlen, pd->hdr.any); 3630 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a, 3631 ruleset, pd, 1); 3632 } 3633 3634 if ((r->action == PF_DROP) && 3635 ((r->rule_flag & PFRULE_RETURNRST) || 3636 (r->rule_flag & PFRULE_RETURNICMP) || 3637 (r->rule_flag & PFRULE_RETURN))) { 3638 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 3639 bip_sum, hdrlen, &reason); 3640 } 3641 3642 if (r->action == PF_DROP) 3643 goto cleanup; 3644 3645 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 3646 REASON_SET(&reason, PFRES_MEMORY); 3647 goto cleanup; 3648 } 3649 if (rtableid >= 0) 3650 M_SETFIB(m, rtableid); 3651 3652 if (!state_icmp && (r->keep_state || nr != NULL || 3653 (pd->flags & PFDESC_TCP_NORM))) { 3654 int action; 3655 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 3656 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 3657 hdrlen); 3658 if (action != PF_PASS) { 3659 if (action == PF_DROP && 3660 (r->rule_flag & PFRULE_RETURN)) 3661 pf_return(r, nr, pd, sk, off, m, th, kif, 3662 bproto_sum, bip_sum, hdrlen, &reason); 3663 return (action); 3664 } 3665 } else { 3666 if (sk != NULL) 3667 uma_zfree(V_pf_state_key_z, sk); 3668 if (nk != NULL) 3669 uma_zfree(V_pf_state_key_z, nk); 3670 } 3671 3672 /* copy back packet headers if we performed NAT operations */ 3673 if (rewrite) 3674 m_copyback(m, off, hdrlen, pd->hdr.any); 3675 3676 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 3677 direction == PF_OUT && 3678 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 3679 /* 3680 * We want the state created, but we dont 3681 * want to send this in case a partner 3682 * firewall has to know about it to allow 3683 * replies through it. 3684 */ 3685 return (PF_DEFER); 3686 3687 return (PF_PASS); 3688 3689 cleanup: 3690 if (sk != NULL) 3691 uma_zfree(V_pf_state_key_z, sk); 3692 if (nk != NULL) 3693 uma_zfree(V_pf_state_key_z, nk); 3694 return (PF_DROP); 3695 } 3696 3697 static int 3698 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 3699 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 3700 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 3701 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_state **sm, 3702 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen) 3703 { 3704 struct pf_state *s = NULL; 3705 struct pf_ksrc_node *sn = NULL; 3706 struct tcphdr *th = pd->hdr.tcp; 3707 u_int16_t mss = V_tcp_mssdflt; 3708 u_short reason; 3709 3710 /* check maximums */ 3711 if (r->max_states && 3712 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 3713 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 3714 REASON_SET(&reason, PFRES_MAXSTATES); 3715 goto csfailed; 3716 } 3717 /* src node for filter rule */ 3718 if ((r->rule_flag & PFRULE_SRCTRACK || 3719 r->rpool.opts & PF_POOL_STICKYADDR) && 3720 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { 3721 REASON_SET(&reason, PFRES_SRCLIMIT); 3722 goto csfailed; 3723 } 3724 /* src node for translation rule */ 3725 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 3726 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { 3727 REASON_SET(&reason, PFRES_SRCLIMIT); 3728 goto csfailed; 3729 } 3730 s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO); 3731 if (s == NULL) { 3732 REASON_SET(&reason, PFRES_MEMORY); 3733 goto csfailed; 3734 } 3735 for (int i = 0; i < 2; i++) { 3736 s->bytes[i] = counter_u64_alloc(M_NOWAIT); 3737 s->packets[i] = counter_u64_alloc(M_NOWAIT); 3738 3739 if (s->bytes[i] == NULL || s->packets[i] == NULL) { 3740 pf_free_state(s); 3741 REASON_SET(&reason, PFRES_MEMORY); 3742 goto csfailed; 3743 } 3744 } 3745 s->rule.ptr = r; 3746 s->nat_rule.ptr = nr; 3747 s->anchor.ptr = a; 3748 STATE_INC_COUNTERS(s); 3749 if (r->allow_opts) 3750 s->state_flags |= PFSTATE_ALLOWOPTS; 3751 if (r->rule_flag & PFRULE_STATESLOPPY) 3752 s->state_flags |= PFSTATE_SLOPPY; 3753 s->log = r->log & PF_LOG_ALL; 3754 s->sync_state = PFSYNC_S_NONE; 3755 if (nr != NULL) 3756 s->log |= nr->log & PF_LOG_ALL; 3757 switch (pd->proto) { 3758 case IPPROTO_TCP: 3759 s->src.seqlo = ntohl(th->th_seq); 3760 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 3761 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 3762 r->keep_state == PF_STATE_MODULATE) { 3763 /* Generate sequence number modulator */ 3764 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 3765 0) 3766 s->src.seqdiff = 1; 3767 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 3768 htonl(s->src.seqlo + s->src.seqdiff), 0); 3769 *rewrite = 1; 3770 } else 3771 s->src.seqdiff = 0; 3772 if (th->th_flags & TH_SYN) { 3773 s->src.seqhi++; 3774 s->src.wscale = pf_get_wscale(m, off, 3775 th->th_off, pd->af); 3776 } 3777 s->src.max_win = MAX(ntohs(th->th_win), 1); 3778 if (s->src.wscale & PF_WSCALE_MASK) { 3779 /* Remove scale factor from initial window */ 3780 int win = s->src.max_win; 3781 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 3782 s->src.max_win = (win - 1) >> 3783 (s->src.wscale & PF_WSCALE_MASK); 3784 } 3785 if (th->th_flags & TH_FIN) 3786 s->src.seqhi++; 3787 s->dst.seqhi = 1; 3788 s->dst.max_win = 1; 3789 s->src.state = TCPS_SYN_SENT; 3790 s->dst.state = TCPS_CLOSED; 3791 s->timeout = PFTM_TCP_FIRST_PACKET; 3792 break; 3793 case IPPROTO_UDP: 3794 s->src.state = PFUDPS_SINGLE; 3795 s->dst.state = PFUDPS_NO_TRAFFIC; 3796 s->timeout = PFTM_UDP_FIRST_PACKET; 3797 break; 3798 case IPPROTO_ICMP: 3799 #ifdef INET6 3800 case IPPROTO_ICMPV6: 3801 #endif 3802 s->timeout = PFTM_ICMP_FIRST_PACKET; 3803 break; 3804 default: 3805 s->src.state = PFOTHERS_SINGLE; 3806 s->dst.state = PFOTHERS_NO_TRAFFIC; 3807 s->timeout = PFTM_OTHER_FIRST_PACKET; 3808 } 3809 3810 if (r->rt) { 3811 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) { 3812 REASON_SET(&reason, PFRES_MAPFAILED); 3813 pf_src_tree_remove_state(s); 3814 STATE_DEC_COUNTERS(s); 3815 uma_zfree(V_pf_state_z, s); 3816 goto csfailed; 3817 } 3818 s->rt_kif = r->rpool.cur->kif; 3819 } 3820 3821 s->creation = time_uptime; 3822 s->expire = time_uptime; 3823 3824 if (sn != NULL) 3825 s->src_node = sn; 3826 if (nsn != NULL) { 3827 /* XXX We only modify one side for now. */ 3828 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 3829 s->nat_src_node = nsn; 3830 } 3831 if (pd->proto == IPPROTO_TCP) { 3832 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m, 3833 off, pd, th, &s->src, &s->dst)) { 3834 REASON_SET(&reason, PFRES_MEMORY); 3835 pf_src_tree_remove_state(s); 3836 STATE_DEC_COUNTERS(s); 3837 uma_zfree(V_pf_state_z, s); 3838 return (PF_DROP); 3839 } 3840 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub && 3841 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 3842 &s->src, &s->dst, rewrite)) { 3843 /* This really shouldn't happen!!! */ 3844 DPFPRINTF(PF_DEBUG_URGENT, 3845 ("pf_normalize_tcp_stateful failed on first " 3846 "pkt\n")); 3847 pf_normalize_tcp_cleanup(s); 3848 pf_src_tree_remove_state(s); 3849 STATE_DEC_COUNTERS(s); 3850 uma_zfree(V_pf_state_z, s); 3851 return (PF_DROP); 3852 } 3853 } 3854 s->direction = pd->dir; 3855 3856 /* 3857 * sk/nk could already been setup by pf_get_translation(). 3858 */ 3859 if (nr == NULL) { 3860 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 3861 __func__, nr, sk, nk)); 3862 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 3863 if (sk == NULL) 3864 goto csfailed; 3865 nk = sk; 3866 } else 3867 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 3868 __func__, nr, sk, nk)); 3869 3870 /* Swap sk/nk for PF_OUT. */ 3871 if (pf_state_insert(BOUND_IFACE(r, kif), 3872 (pd->dir == PF_IN) ? sk : nk, 3873 (pd->dir == PF_IN) ? nk : sk, s)) { 3874 if (pd->proto == IPPROTO_TCP) 3875 pf_normalize_tcp_cleanup(s); 3876 REASON_SET(&reason, PFRES_STATEINS); 3877 pf_src_tree_remove_state(s); 3878 STATE_DEC_COUNTERS(s); 3879 uma_zfree(V_pf_state_z, s); 3880 return (PF_DROP); 3881 } else 3882 *sm = s; 3883 3884 if (tag > 0) 3885 s->tag = tag; 3886 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 3887 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 3888 s->src.state = PF_TCPS_PROXY_SRC; 3889 /* undo NAT changes, if they have taken place */ 3890 if (nr != NULL) { 3891 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 3892 if (pd->dir == PF_OUT) 3893 skt = s->key[PF_SK_STACK]; 3894 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 3895 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 3896 if (pd->sport) 3897 *pd->sport = skt->port[pd->sidx]; 3898 if (pd->dport) 3899 *pd->dport = skt->port[pd->didx]; 3900 if (pd->proto_sum) 3901 *pd->proto_sum = bproto_sum; 3902 if (pd->ip_sum) 3903 *pd->ip_sum = bip_sum; 3904 m_copyback(m, off, hdrlen, pd->hdr.any); 3905 } 3906 s->src.seqhi = htonl(arc4random()); 3907 /* Find mss option */ 3908 int rtid = M_GETFIB(m); 3909 mss = pf_get_mss(m, off, th->th_off, pd->af); 3910 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 3911 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 3912 s->src.mss = mss; 3913 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport, 3914 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 3915 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL); 3916 REASON_SET(&reason, PFRES_SYNPROXY); 3917 return (PF_SYNPROXY_DROP); 3918 } 3919 3920 return (PF_PASS); 3921 3922 csfailed: 3923 if (sk != NULL) 3924 uma_zfree(V_pf_state_key_z, sk); 3925 if (nk != NULL) 3926 uma_zfree(V_pf_state_key_z, nk); 3927 3928 if (sn != NULL) { 3929 struct pf_srchash *sh; 3930 3931 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 3932 PF_HASHROW_LOCK(sh); 3933 if (--sn->states == 0 && sn->expire == 0) { 3934 pf_unlink_src_node(sn); 3935 uma_zfree(V_pf_sources_z, sn); 3936 counter_u64_add( 3937 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 3938 } 3939 PF_HASHROW_UNLOCK(sh); 3940 } 3941 3942 if (nsn != sn && nsn != NULL) { 3943 struct pf_srchash *sh; 3944 3945 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)]; 3946 PF_HASHROW_LOCK(sh); 3947 if (--nsn->states == 0 && nsn->expire == 0) { 3948 pf_unlink_src_node(nsn); 3949 uma_zfree(V_pf_sources_z, nsn); 3950 counter_u64_add( 3951 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 3952 } 3953 PF_HASHROW_UNLOCK(sh); 3954 } 3955 3956 return (PF_DROP); 3957 } 3958 3959 static int 3960 pf_test_fragment(struct pf_krule **rm, int direction, struct pfi_kkif *kif, 3961 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 3962 struct pf_kruleset **rsm) 3963 { 3964 struct pf_krule *r, *a = NULL; 3965 struct pf_kruleset *ruleset = NULL; 3966 sa_family_t af = pd->af; 3967 u_short reason; 3968 int tag = -1; 3969 int asd = 0; 3970 int match = 0; 3971 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3972 3973 PF_RULES_RASSERT(); 3974 3975 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3976 while (r != NULL) { 3977 counter_u64_add(r->evaluations, 1); 3978 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 3979 r = r->skip[PF_SKIP_IFP].ptr; 3980 else if (r->direction && r->direction != direction) 3981 r = r->skip[PF_SKIP_DIR].ptr; 3982 else if (r->af && r->af != af) 3983 r = r->skip[PF_SKIP_AF].ptr; 3984 else if (r->proto && r->proto != pd->proto) 3985 r = r->skip[PF_SKIP_PROTO].ptr; 3986 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 3987 r->src.neg, kif, M_GETFIB(m))) 3988 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3989 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 3990 r->dst.neg, NULL, M_GETFIB(m))) 3991 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3992 else if (r->tos && !(r->tos == pd->tos)) 3993 r = TAILQ_NEXT(r, entries); 3994 else if (r->os_fingerprint != PF_OSFP_ANY) 3995 r = TAILQ_NEXT(r, entries); 3996 else if (pd->proto == IPPROTO_UDP && 3997 (r->src.port_op || r->dst.port_op)) 3998 r = TAILQ_NEXT(r, entries); 3999 else if (pd->proto == IPPROTO_TCP && 4000 (r->src.port_op || r->dst.port_op || r->flagset)) 4001 r = TAILQ_NEXT(r, entries); 4002 else if ((pd->proto == IPPROTO_ICMP || 4003 pd->proto == IPPROTO_ICMPV6) && 4004 (r->type || r->code)) 4005 r = TAILQ_NEXT(r, entries); 4006 else if (r->prio && 4007 !pf_match_ieee8021q_pcp(r->prio, m)) 4008 r = TAILQ_NEXT(r, entries); 4009 else if (r->prob && r->prob <= 4010 (arc4random() % (UINT_MAX - 1) + 1)) 4011 r = TAILQ_NEXT(r, entries); 4012 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4013 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4014 r = TAILQ_NEXT(r, entries); 4015 else { 4016 if (r->anchor == NULL) { 4017 match = 1; 4018 *rm = r; 4019 *am = a; 4020 *rsm = ruleset; 4021 if ((*rm)->quick) 4022 break; 4023 r = TAILQ_NEXT(r, entries); 4024 } else 4025 pf_step_into_anchor(anchor_stack, &asd, 4026 &ruleset, PF_RULESET_FILTER, &r, &a, 4027 &match); 4028 } 4029 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4030 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4031 break; 4032 } 4033 r = *rm; 4034 a = *am; 4035 ruleset = *rsm; 4036 4037 REASON_SET(&reason, PFRES_MATCH); 4038 4039 if (r->log) 4040 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd, 4041 1); 4042 4043 if (r->action != PF_PASS) 4044 return (PF_DROP); 4045 4046 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4047 REASON_SET(&reason, PFRES_MEMORY); 4048 return (PF_DROP); 4049 } 4050 4051 return (PF_PASS); 4052 } 4053 4054 static int 4055 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst, 4056 struct pf_state **state, struct pfi_kkif *kif, struct mbuf *m, int off, 4057 struct pf_pdesc *pd, u_short *reason, int *copyback) 4058 { 4059 struct tcphdr *th = pd->hdr.tcp; 4060 u_int16_t win = ntohs(th->th_win); 4061 u_int32_t ack, end, seq, orig_seq; 4062 u_int8_t sws, dws; 4063 int ackskew; 4064 4065 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 4066 sws = src->wscale & PF_WSCALE_MASK; 4067 dws = dst->wscale & PF_WSCALE_MASK; 4068 } else 4069 sws = dws = 0; 4070 4071 /* 4072 * Sequence tracking algorithm from Guido van Rooij's paper: 4073 * http://www.madison-gurkha.com/publications/tcp_filtering/ 4074 * tcp_filtering.ps 4075 */ 4076 4077 orig_seq = seq = ntohl(th->th_seq); 4078 if (src->seqlo == 0) { 4079 /* First packet from this end. Set its state */ 4080 4081 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) && 4082 src->scrub == NULL) { 4083 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 4084 REASON_SET(reason, PFRES_MEMORY); 4085 return (PF_DROP); 4086 } 4087 } 4088 4089 /* Deferred generation of sequence number modulator */ 4090 if (dst->seqdiff && !src->seqdiff) { 4091 /* use random iss for the TCP server */ 4092 while ((src->seqdiff = arc4random() - seq) == 0) 4093 ; 4094 ack = ntohl(th->th_ack) - dst->seqdiff; 4095 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 4096 src->seqdiff), 0); 4097 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 4098 *copyback = 1; 4099 } else { 4100 ack = ntohl(th->th_ack); 4101 } 4102 4103 end = seq + pd->p_len; 4104 if (th->th_flags & TH_SYN) { 4105 end++; 4106 if (dst->wscale & PF_WSCALE_FLAG) { 4107 src->wscale = pf_get_wscale(m, off, th->th_off, 4108 pd->af); 4109 if (src->wscale & PF_WSCALE_FLAG) { 4110 /* Remove scale factor from initial 4111 * window */ 4112 sws = src->wscale & PF_WSCALE_MASK; 4113 win = ((u_int32_t)win + (1 << sws) - 1) 4114 >> sws; 4115 dws = dst->wscale & PF_WSCALE_MASK; 4116 } else { 4117 /* fixup other window */ 4118 dst->max_win <<= dst->wscale & 4119 PF_WSCALE_MASK; 4120 /* in case of a retrans SYN|ACK */ 4121 dst->wscale = 0; 4122 } 4123 } 4124 } 4125 if (th->th_flags & TH_FIN) 4126 end++; 4127 4128 src->seqlo = seq; 4129 if (src->state < TCPS_SYN_SENT) 4130 src->state = TCPS_SYN_SENT; 4131 4132 /* 4133 * May need to slide the window (seqhi may have been set by 4134 * the crappy stack check or if we picked up the connection 4135 * after establishment) 4136 */ 4137 if (src->seqhi == 1 || 4138 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 4139 src->seqhi = end + MAX(1, dst->max_win << dws); 4140 if (win > src->max_win) 4141 src->max_win = win; 4142 4143 } else { 4144 ack = ntohl(th->th_ack) - dst->seqdiff; 4145 if (src->seqdiff) { 4146 /* Modulate sequence numbers */ 4147 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 4148 src->seqdiff), 0); 4149 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 4150 *copyback = 1; 4151 } 4152 end = seq + pd->p_len; 4153 if (th->th_flags & TH_SYN) 4154 end++; 4155 if (th->th_flags & TH_FIN) 4156 end++; 4157 } 4158 4159 if ((th->th_flags & TH_ACK) == 0) { 4160 /* Let it pass through the ack skew check */ 4161 ack = dst->seqlo; 4162 } else if ((ack == 0 && 4163 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 4164 /* broken tcp stacks do not set ack */ 4165 (dst->state < TCPS_SYN_SENT)) { 4166 /* 4167 * Many stacks (ours included) will set the ACK number in an 4168 * FIN|ACK if the SYN times out -- no sequence to ACK. 4169 */ 4170 ack = dst->seqlo; 4171 } 4172 4173 if (seq == end) { 4174 /* Ease sequencing restrictions on no data packets */ 4175 seq = src->seqlo; 4176 end = seq; 4177 } 4178 4179 ackskew = dst->seqlo - ack; 4180 4181 /* 4182 * Need to demodulate the sequence numbers in any TCP SACK options 4183 * (Selective ACK). We could optionally validate the SACK values 4184 * against the current ACK window, either forwards or backwards, but 4185 * I'm not confident that SACK has been implemented properly 4186 * everywhere. It wouldn't surprise me if several stacks accidentally 4187 * SACK too far backwards of previously ACKed data. There really aren't 4188 * any security implications of bad SACKing unless the target stack 4189 * doesn't validate the option length correctly. Someone trying to 4190 * spoof into a TCP connection won't bother blindly sending SACK 4191 * options anyway. 4192 */ 4193 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 4194 if (pf_modulate_sack(m, off, pd, th, dst)) 4195 *copyback = 1; 4196 } 4197 4198 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 4199 if (SEQ_GEQ(src->seqhi, end) && 4200 /* Last octet inside other's window space */ 4201 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 4202 /* Retrans: not more than one window back */ 4203 (ackskew >= -MAXACKWINDOW) && 4204 /* Acking not more than one reassembled fragment backwards */ 4205 (ackskew <= (MAXACKWINDOW << sws)) && 4206 /* Acking not more than one window forward */ 4207 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 4208 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 4209 (pd->flags & PFDESC_IP_REAS) == 0)) { 4210 /* Require an exact/+1 sequence match on resets when possible */ 4211 4212 if (dst->scrub || src->scrub) { 4213 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4214 *state, src, dst, copyback)) 4215 return (PF_DROP); 4216 } 4217 4218 /* update max window */ 4219 if (src->max_win < win) 4220 src->max_win = win; 4221 /* synchronize sequencing */ 4222 if (SEQ_GT(end, src->seqlo)) 4223 src->seqlo = end; 4224 /* slide the window of what the other end can send */ 4225 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4226 dst->seqhi = ack + MAX((win << sws), 1); 4227 4228 /* update states */ 4229 if (th->th_flags & TH_SYN) 4230 if (src->state < TCPS_SYN_SENT) 4231 src->state = TCPS_SYN_SENT; 4232 if (th->th_flags & TH_FIN) 4233 if (src->state < TCPS_CLOSING) 4234 src->state = TCPS_CLOSING; 4235 if (th->th_flags & TH_ACK) { 4236 if (dst->state == TCPS_SYN_SENT) { 4237 dst->state = TCPS_ESTABLISHED; 4238 if (src->state == TCPS_ESTABLISHED && 4239 (*state)->src_node != NULL && 4240 pf_src_connlimit(state)) { 4241 REASON_SET(reason, PFRES_SRCLIMIT); 4242 return (PF_DROP); 4243 } 4244 } else if (dst->state == TCPS_CLOSING) 4245 dst->state = TCPS_FIN_WAIT_2; 4246 } 4247 if (th->th_flags & TH_RST) 4248 src->state = dst->state = TCPS_TIME_WAIT; 4249 4250 /* update expire time */ 4251 (*state)->expire = time_uptime; 4252 if (src->state >= TCPS_FIN_WAIT_2 && 4253 dst->state >= TCPS_FIN_WAIT_2) 4254 (*state)->timeout = PFTM_TCP_CLOSED; 4255 else if (src->state >= TCPS_CLOSING && 4256 dst->state >= TCPS_CLOSING) 4257 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4258 else if (src->state < TCPS_ESTABLISHED || 4259 dst->state < TCPS_ESTABLISHED) 4260 (*state)->timeout = PFTM_TCP_OPENING; 4261 else if (src->state >= TCPS_CLOSING || 4262 dst->state >= TCPS_CLOSING) 4263 (*state)->timeout = PFTM_TCP_CLOSING; 4264 else 4265 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4266 4267 /* Fall through to PASS packet */ 4268 4269 } else if ((dst->state < TCPS_SYN_SENT || 4270 dst->state >= TCPS_FIN_WAIT_2 || 4271 src->state >= TCPS_FIN_WAIT_2) && 4272 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 4273 /* Within a window forward of the originating packet */ 4274 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 4275 /* Within a window backward of the originating packet */ 4276 4277 /* 4278 * This currently handles three situations: 4279 * 1) Stupid stacks will shotgun SYNs before their peer 4280 * replies. 4281 * 2) When PF catches an already established stream (the 4282 * firewall rebooted, the state table was flushed, routes 4283 * changed...) 4284 * 3) Packets get funky immediately after the connection 4285 * closes (this should catch Solaris spurious ACK|FINs 4286 * that web servers like to spew after a close) 4287 * 4288 * This must be a little more careful than the above code 4289 * since packet floods will also be caught here. We don't 4290 * update the TTL here to mitigate the damage of a packet 4291 * flood and so the same code can handle awkward establishment 4292 * and a loosened connection close. 4293 * In the establishment case, a correct peer response will 4294 * validate the connection, go through the normal state code 4295 * and keep updating the state TTL. 4296 */ 4297 4298 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4299 printf("pf: loose state match: "); 4300 pf_print_state(*state); 4301 pf_print_flags(th->th_flags); 4302 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4303 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 4304 pd->p_len, ackskew, 4305 (unsigned long long)counter_u64_fetch((*state)->packets[0]), 4306 (unsigned long long)counter_u64_fetch((*state)->packets[1]), 4307 pd->dir == PF_IN ? "in" : "out", 4308 pd->dir == (*state)->direction ? "fwd" : "rev"); 4309 } 4310 4311 if (dst->scrub || src->scrub) { 4312 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4313 *state, src, dst, copyback)) 4314 return (PF_DROP); 4315 } 4316 4317 /* update max window */ 4318 if (src->max_win < win) 4319 src->max_win = win; 4320 /* synchronize sequencing */ 4321 if (SEQ_GT(end, src->seqlo)) 4322 src->seqlo = end; 4323 /* slide the window of what the other end can send */ 4324 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4325 dst->seqhi = ack + MAX((win << sws), 1); 4326 4327 /* 4328 * Cannot set dst->seqhi here since this could be a shotgunned 4329 * SYN and not an already established connection. 4330 */ 4331 4332 if (th->th_flags & TH_FIN) 4333 if (src->state < TCPS_CLOSING) 4334 src->state = TCPS_CLOSING; 4335 if (th->th_flags & TH_RST) 4336 src->state = dst->state = TCPS_TIME_WAIT; 4337 4338 /* Fall through to PASS packet */ 4339 4340 } else { 4341 if ((*state)->dst.state == TCPS_SYN_SENT && 4342 (*state)->src.state == TCPS_SYN_SENT) { 4343 /* Send RST for state mismatches during handshake */ 4344 if (!(th->th_flags & TH_RST)) 4345 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4346 pd->dst, pd->src, th->th_dport, 4347 th->th_sport, ntohl(th->th_ack), 0, 4348 TH_RST, 0, 0, 4349 (*state)->rule.ptr->return_ttl, 1, 0, 4350 kif->pfik_ifp); 4351 src->seqlo = 0; 4352 src->seqhi = 1; 4353 src->max_win = 1; 4354 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 4355 printf("pf: BAD state: "); 4356 pf_print_state(*state); 4357 pf_print_flags(th->th_flags); 4358 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4359 "pkts=%llu:%llu dir=%s,%s\n", 4360 seq, orig_seq, ack, pd->p_len, ackskew, 4361 (unsigned long long)counter_u64_fetch((*state)->packets[0]), 4362 (unsigned long long)counter_u64_fetch((*state)->packets[1]), 4363 pd->dir == PF_IN ? "in" : "out", 4364 pd->dir == (*state)->direction ? "fwd" : "rev"); 4365 printf("pf: State failure on: %c %c %c %c | %c %c\n", 4366 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 4367 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 4368 ' ': '2', 4369 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 4370 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 4371 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 4372 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 4373 } 4374 REASON_SET(reason, PFRES_BADSTATE); 4375 return (PF_DROP); 4376 } 4377 4378 return (PF_PASS); 4379 } 4380 4381 static int 4382 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst, 4383 struct pf_state **state, struct pf_pdesc *pd, u_short *reason) 4384 { 4385 struct tcphdr *th = pd->hdr.tcp; 4386 4387 if (th->th_flags & TH_SYN) 4388 if (src->state < TCPS_SYN_SENT) 4389 src->state = TCPS_SYN_SENT; 4390 if (th->th_flags & TH_FIN) 4391 if (src->state < TCPS_CLOSING) 4392 src->state = TCPS_CLOSING; 4393 if (th->th_flags & TH_ACK) { 4394 if (dst->state == TCPS_SYN_SENT) { 4395 dst->state = TCPS_ESTABLISHED; 4396 if (src->state == TCPS_ESTABLISHED && 4397 (*state)->src_node != NULL && 4398 pf_src_connlimit(state)) { 4399 REASON_SET(reason, PFRES_SRCLIMIT); 4400 return (PF_DROP); 4401 } 4402 } else if (dst->state == TCPS_CLOSING) { 4403 dst->state = TCPS_FIN_WAIT_2; 4404 } else if (src->state == TCPS_SYN_SENT && 4405 dst->state < TCPS_SYN_SENT) { 4406 /* 4407 * Handle a special sloppy case where we only see one 4408 * half of the connection. If there is a ACK after 4409 * the initial SYN without ever seeing a packet from 4410 * the destination, set the connection to established. 4411 */ 4412 dst->state = src->state = TCPS_ESTABLISHED; 4413 if ((*state)->src_node != NULL && 4414 pf_src_connlimit(state)) { 4415 REASON_SET(reason, PFRES_SRCLIMIT); 4416 return (PF_DROP); 4417 } 4418 } else if (src->state == TCPS_CLOSING && 4419 dst->state == TCPS_ESTABLISHED && 4420 dst->seqlo == 0) { 4421 /* 4422 * Handle the closing of half connections where we 4423 * don't see the full bidirectional FIN/ACK+ACK 4424 * handshake. 4425 */ 4426 dst->state = TCPS_CLOSING; 4427 } 4428 } 4429 if (th->th_flags & TH_RST) 4430 src->state = dst->state = TCPS_TIME_WAIT; 4431 4432 /* update expire time */ 4433 (*state)->expire = time_uptime; 4434 if (src->state >= TCPS_FIN_WAIT_2 && 4435 dst->state >= TCPS_FIN_WAIT_2) 4436 (*state)->timeout = PFTM_TCP_CLOSED; 4437 else if (src->state >= TCPS_CLOSING && 4438 dst->state >= TCPS_CLOSING) 4439 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4440 else if (src->state < TCPS_ESTABLISHED || 4441 dst->state < TCPS_ESTABLISHED) 4442 (*state)->timeout = PFTM_TCP_OPENING; 4443 else if (src->state >= TCPS_CLOSING || 4444 dst->state >= TCPS_CLOSING) 4445 (*state)->timeout = PFTM_TCP_CLOSING; 4446 else 4447 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4448 4449 return (PF_PASS); 4450 } 4451 4452 static int 4453 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kkif *kif, 4454 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 4455 u_short *reason) 4456 { 4457 struct pf_state_key_cmp key; 4458 struct tcphdr *th = pd->hdr.tcp; 4459 int copyback = 0; 4460 struct pf_state_peer *src, *dst; 4461 struct pf_state_key *sk; 4462 4463 bzero(&key, sizeof(key)); 4464 key.af = pd->af; 4465 key.proto = IPPROTO_TCP; 4466 if (direction == PF_IN) { /* wire side, straight */ 4467 PF_ACPY(&key.addr[0], pd->src, key.af); 4468 PF_ACPY(&key.addr[1], pd->dst, key.af); 4469 key.port[0] = th->th_sport; 4470 key.port[1] = th->th_dport; 4471 } else { /* stack side, reverse */ 4472 PF_ACPY(&key.addr[1], pd->src, key.af); 4473 PF_ACPY(&key.addr[0], pd->dst, key.af); 4474 key.port[1] = th->th_sport; 4475 key.port[0] = th->th_dport; 4476 } 4477 4478 STATE_LOOKUP(kif, &key, direction, *state, pd); 4479 4480 if (direction == (*state)->direction) { 4481 src = &(*state)->src; 4482 dst = &(*state)->dst; 4483 } else { 4484 src = &(*state)->dst; 4485 dst = &(*state)->src; 4486 } 4487 4488 sk = (*state)->key[pd->didx]; 4489 4490 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 4491 if (direction != (*state)->direction) { 4492 REASON_SET(reason, PFRES_SYNPROXY); 4493 return (PF_SYNPROXY_DROP); 4494 } 4495 if (th->th_flags & TH_SYN) { 4496 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 4497 REASON_SET(reason, PFRES_SYNPROXY); 4498 return (PF_DROP); 4499 } 4500 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, 4501 pd->src, th->th_dport, th->th_sport, 4502 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 4503 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL); 4504 REASON_SET(reason, PFRES_SYNPROXY); 4505 return (PF_SYNPROXY_DROP); 4506 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 4507 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4508 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4509 REASON_SET(reason, PFRES_SYNPROXY); 4510 return (PF_DROP); 4511 } else if ((*state)->src_node != NULL && 4512 pf_src_connlimit(state)) { 4513 REASON_SET(reason, PFRES_SRCLIMIT); 4514 return (PF_DROP); 4515 } else 4516 (*state)->src.state = PF_TCPS_PROXY_DST; 4517 } 4518 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 4519 if (direction == (*state)->direction) { 4520 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 4521 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4522 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4523 REASON_SET(reason, PFRES_SYNPROXY); 4524 return (PF_DROP); 4525 } 4526 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 4527 if ((*state)->dst.seqhi == 1) 4528 (*state)->dst.seqhi = htonl(arc4random()); 4529 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4530 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4531 sk->port[pd->sidx], sk->port[pd->didx], 4532 (*state)->dst.seqhi, 0, TH_SYN, 0, 4533 (*state)->src.mss, 0, 0, (*state)->tag, NULL); 4534 REASON_SET(reason, PFRES_SYNPROXY); 4535 return (PF_SYNPROXY_DROP); 4536 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 4537 (TH_SYN|TH_ACK)) || 4538 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 4539 REASON_SET(reason, PFRES_SYNPROXY); 4540 return (PF_DROP); 4541 } else { 4542 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 4543 (*state)->dst.seqlo = ntohl(th->th_seq); 4544 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, 4545 pd->src, th->th_dport, th->th_sport, 4546 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 4547 TH_ACK, (*state)->src.max_win, 0, 0, 0, 4548 (*state)->tag, NULL); 4549 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4550 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4551 sk->port[pd->sidx], sk->port[pd->didx], 4552 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 4553 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL); 4554 (*state)->src.seqdiff = (*state)->dst.seqhi - 4555 (*state)->src.seqlo; 4556 (*state)->dst.seqdiff = (*state)->src.seqhi - 4557 (*state)->dst.seqlo; 4558 (*state)->src.seqhi = (*state)->src.seqlo + 4559 (*state)->dst.max_win; 4560 (*state)->dst.seqhi = (*state)->dst.seqlo + 4561 (*state)->src.max_win; 4562 (*state)->src.wscale = (*state)->dst.wscale = 0; 4563 (*state)->src.state = (*state)->dst.state = 4564 TCPS_ESTABLISHED; 4565 REASON_SET(reason, PFRES_SYNPROXY); 4566 return (PF_SYNPROXY_DROP); 4567 } 4568 } 4569 4570 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) && 4571 dst->state >= TCPS_FIN_WAIT_2 && 4572 src->state >= TCPS_FIN_WAIT_2) { 4573 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4574 printf("pf: state reuse "); 4575 pf_print_state(*state); 4576 pf_print_flags(th->th_flags); 4577 printf("\n"); 4578 } 4579 /* XXX make sure it's the same direction ?? */ 4580 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; 4581 pf_unlink_state(*state, PF_ENTER_LOCKED); 4582 *state = NULL; 4583 return (PF_DROP); 4584 } 4585 4586 if ((*state)->state_flags & PFSTATE_SLOPPY) { 4587 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP) 4588 return (PF_DROP); 4589 } else { 4590 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason, 4591 ©back) == PF_DROP) 4592 return (PF_DROP); 4593 } 4594 4595 /* translate source/destination address, if necessary */ 4596 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4597 struct pf_state_key *nk = (*state)->key[pd->didx]; 4598 4599 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4600 nk->port[pd->sidx] != th->th_sport) 4601 pf_change_ap(m, pd->src, &th->th_sport, 4602 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 4603 nk->port[pd->sidx], 0, pd->af); 4604 4605 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 4606 nk->port[pd->didx] != th->th_dport) 4607 pf_change_ap(m, pd->dst, &th->th_dport, 4608 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 4609 nk->port[pd->didx], 0, pd->af); 4610 copyback = 1; 4611 } 4612 4613 /* Copyback sequence modulation or stateful scrub changes if needed */ 4614 if (copyback) 4615 m_copyback(m, off, sizeof(*th), (caddr_t)th); 4616 4617 return (PF_PASS); 4618 } 4619 4620 static int 4621 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kkif *kif, 4622 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 4623 { 4624 struct pf_state_peer *src, *dst; 4625 struct pf_state_key_cmp key; 4626 struct udphdr *uh = pd->hdr.udp; 4627 4628 bzero(&key, sizeof(key)); 4629 key.af = pd->af; 4630 key.proto = IPPROTO_UDP; 4631 if (direction == PF_IN) { /* wire side, straight */ 4632 PF_ACPY(&key.addr[0], pd->src, key.af); 4633 PF_ACPY(&key.addr[1], pd->dst, key.af); 4634 key.port[0] = uh->uh_sport; 4635 key.port[1] = uh->uh_dport; 4636 } else { /* stack side, reverse */ 4637 PF_ACPY(&key.addr[1], pd->src, key.af); 4638 PF_ACPY(&key.addr[0], pd->dst, key.af); 4639 key.port[1] = uh->uh_sport; 4640 key.port[0] = uh->uh_dport; 4641 } 4642 4643 STATE_LOOKUP(kif, &key, direction, *state, pd); 4644 4645 if (direction == (*state)->direction) { 4646 src = &(*state)->src; 4647 dst = &(*state)->dst; 4648 } else { 4649 src = &(*state)->dst; 4650 dst = &(*state)->src; 4651 } 4652 4653 /* update states */ 4654 if (src->state < PFUDPS_SINGLE) 4655 src->state = PFUDPS_SINGLE; 4656 if (dst->state == PFUDPS_SINGLE) 4657 dst->state = PFUDPS_MULTIPLE; 4658 4659 /* update expire time */ 4660 (*state)->expire = time_uptime; 4661 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 4662 (*state)->timeout = PFTM_UDP_MULTIPLE; 4663 else 4664 (*state)->timeout = PFTM_UDP_SINGLE; 4665 4666 /* translate source/destination address, if necessary */ 4667 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4668 struct pf_state_key *nk = (*state)->key[pd->didx]; 4669 4670 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4671 nk->port[pd->sidx] != uh->uh_sport) 4672 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 4673 &uh->uh_sum, &nk->addr[pd->sidx], 4674 nk->port[pd->sidx], 1, pd->af); 4675 4676 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 4677 nk->port[pd->didx] != uh->uh_dport) 4678 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 4679 &uh->uh_sum, &nk->addr[pd->didx], 4680 nk->port[pd->didx], 1, pd->af); 4681 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 4682 } 4683 4684 return (PF_PASS); 4685 } 4686 4687 static int 4688 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kkif *kif, 4689 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 4690 { 4691 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 4692 u_int16_t icmpid = 0, *icmpsum; 4693 u_int8_t icmptype, icmpcode; 4694 int state_icmp = 0; 4695 struct pf_state_key_cmp key; 4696 4697 bzero(&key, sizeof(key)); 4698 switch (pd->proto) { 4699 #ifdef INET 4700 case IPPROTO_ICMP: 4701 icmptype = pd->hdr.icmp->icmp_type; 4702 icmpcode = pd->hdr.icmp->icmp_code; 4703 icmpid = pd->hdr.icmp->icmp_id; 4704 icmpsum = &pd->hdr.icmp->icmp_cksum; 4705 4706 if (icmptype == ICMP_UNREACH || 4707 icmptype == ICMP_SOURCEQUENCH || 4708 icmptype == ICMP_REDIRECT || 4709 icmptype == ICMP_TIMXCEED || 4710 icmptype == ICMP_PARAMPROB) 4711 state_icmp++; 4712 break; 4713 #endif /* INET */ 4714 #ifdef INET6 4715 case IPPROTO_ICMPV6: 4716 icmptype = pd->hdr.icmp6->icmp6_type; 4717 icmpcode = pd->hdr.icmp6->icmp6_code; 4718 icmpid = pd->hdr.icmp6->icmp6_id; 4719 icmpsum = &pd->hdr.icmp6->icmp6_cksum; 4720 4721 if (icmptype == ICMP6_DST_UNREACH || 4722 icmptype == ICMP6_PACKET_TOO_BIG || 4723 icmptype == ICMP6_TIME_EXCEEDED || 4724 icmptype == ICMP6_PARAM_PROB) 4725 state_icmp++; 4726 break; 4727 #endif /* INET6 */ 4728 } 4729 4730 if (!state_icmp) { 4731 /* 4732 * ICMP query/reply message not related to a TCP/UDP packet. 4733 * Search for an ICMP state. 4734 */ 4735 key.af = pd->af; 4736 key.proto = pd->proto; 4737 key.port[0] = key.port[1] = icmpid; 4738 if (direction == PF_IN) { /* wire side, straight */ 4739 PF_ACPY(&key.addr[0], pd->src, key.af); 4740 PF_ACPY(&key.addr[1], pd->dst, key.af); 4741 } else { /* stack side, reverse */ 4742 PF_ACPY(&key.addr[1], pd->src, key.af); 4743 PF_ACPY(&key.addr[0], pd->dst, key.af); 4744 } 4745 4746 STATE_LOOKUP(kif, &key, direction, *state, pd); 4747 4748 (*state)->expire = time_uptime; 4749 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 4750 4751 /* translate source/destination address, if necessary */ 4752 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4753 struct pf_state_key *nk = (*state)->key[pd->didx]; 4754 4755 switch (pd->af) { 4756 #ifdef INET 4757 case AF_INET: 4758 if (PF_ANEQ(pd->src, 4759 &nk->addr[pd->sidx], AF_INET)) 4760 pf_change_a(&saddr->v4.s_addr, 4761 pd->ip_sum, 4762 nk->addr[pd->sidx].v4.s_addr, 0); 4763 4764 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 4765 AF_INET)) 4766 pf_change_a(&daddr->v4.s_addr, 4767 pd->ip_sum, 4768 nk->addr[pd->didx].v4.s_addr, 0); 4769 4770 if (nk->port[0] != 4771 pd->hdr.icmp->icmp_id) { 4772 pd->hdr.icmp->icmp_cksum = 4773 pf_cksum_fixup( 4774 pd->hdr.icmp->icmp_cksum, icmpid, 4775 nk->port[pd->sidx], 0); 4776 pd->hdr.icmp->icmp_id = 4777 nk->port[pd->sidx]; 4778 } 4779 4780 m_copyback(m, off, ICMP_MINLEN, 4781 (caddr_t )pd->hdr.icmp); 4782 break; 4783 #endif /* INET */ 4784 #ifdef INET6 4785 case AF_INET6: 4786 if (PF_ANEQ(pd->src, 4787 &nk->addr[pd->sidx], AF_INET6)) 4788 pf_change_a6(saddr, 4789 &pd->hdr.icmp6->icmp6_cksum, 4790 &nk->addr[pd->sidx], 0); 4791 4792 if (PF_ANEQ(pd->dst, 4793 &nk->addr[pd->didx], AF_INET6)) 4794 pf_change_a6(daddr, 4795 &pd->hdr.icmp6->icmp6_cksum, 4796 &nk->addr[pd->didx], 0); 4797 4798 m_copyback(m, off, sizeof(struct icmp6_hdr), 4799 (caddr_t )pd->hdr.icmp6); 4800 break; 4801 #endif /* INET6 */ 4802 } 4803 } 4804 return (PF_PASS); 4805 4806 } else { 4807 /* 4808 * ICMP error message in response to a TCP/UDP packet. 4809 * Extract the inner TCP/UDP header and search for that state. 4810 */ 4811 4812 struct pf_pdesc pd2; 4813 bzero(&pd2, sizeof pd2); 4814 #ifdef INET 4815 struct ip h2; 4816 #endif /* INET */ 4817 #ifdef INET6 4818 struct ip6_hdr h2_6; 4819 int terminal = 0; 4820 #endif /* INET6 */ 4821 int ipoff2 = 0; 4822 int off2 = 0; 4823 4824 pd2.af = pd->af; 4825 /* Payload packet is from the opposite direction. */ 4826 pd2.sidx = (direction == PF_IN) ? 1 : 0; 4827 pd2.didx = (direction == PF_IN) ? 0 : 1; 4828 switch (pd->af) { 4829 #ifdef INET 4830 case AF_INET: 4831 /* offset of h2 in mbuf chain */ 4832 ipoff2 = off + ICMP_MINLEN; 4833 4834 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 4835 NULL, reason, pd2.af)) { 4836 DPFPRINTF(PF_DEBUG_MISC, 4837 ("pf: ICMP error message too short " 4838 "(ip)\n")); 4839 return (PF_DROP); 4840 } 4841 /* 4842 * ICMP error messages don't refer to non-first 4843 * fragments 4844 */ 4845 if (h2.ip_off & htons(IP_OFFMASK)) { 4846 REASON_SET(reason, PFRES_FRAG); 4847 return (PF_DROP); 4848 } 4849 4850 /* offset of protocol header that follows h2 */ 4851 off2 = ipoff2 + (h2.ip_hl << 2); 4852 4853 pd2.proto = h2.ip_p; 4854 pd2.src = (struct pf_addr *)&h2.ip_src; 4855 pd2.dst = (struct pf_addr *)&h2.ip_dst; 4856 pd2.ip_sum = &h2.ip_sum; 4857 break; 4858 #endif /* INET */ 4859 #ifdef INET6 4860 case AF_INET6: 4861 ipoff2 = off + sizeof(struct icmp6_hdr); 4862 4863 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 4864 NULL, reason, pd2.af)) { 4865 DPFPRINTF(PF_DEBUG_MISC, 4866 ("pf: ICMP error message too short " 4867 "(ip6)\n")); 4868 return (PF_DROP); 4869 } 4870 pd2.proto = h2_6.ip6_nxt; 4871 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 4872 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 4873 pd2.ip_sum = NULL; 4874 off2 = ipoff2 + sizeof(h2_6); 4875 do { 4876 switch (pd2.proto) { 4877 case IPPROTO_FRAGMENT: 4878 /* 4879 * ICMPv6 error messages for 4880 * non-first fragments 4881 */ 4882 REASON_SET(reason, PFRES_FRAG); 4883 return (PF_DROP); 4884 case IPPROTO_AH: 4885 case IPPROTO_HOPOPTS: 4886 case IPPROTO_ROUTING: 4887 case IPPROTO_DSTOPTS: { 4888 /* get next header and header length */ 4889 struct ip6_ext opt6; 4890 4891 if (!pf_pull_hdr(m, off2, &opt6, 4892 sizeof(opt6), NULL, reason, 4893 pd2.af)) { 4894 DPFPRINTF(PF_DEBUG_MISC, 4895 ("pf: ICMPv6 short opt\n")); 4896 return (PF_DROP); 4897 } 4898 if (pd2.proto == IPPROTO_AH) 4899 off2 += (opt6.ip6e_len + 2) * 4; 4900 else 4901 off2 += (opt6.ip6e_len + 1) * 8; 4902 pd2.proto = opt6.ip6e_nxt; 4903 /* goto the next header */ 4904 break; 4905 } 4906 default: 4907 terminal++; 4908 break; 4909 } 4910 } while (!terminal); 4911 break; 4912 #endif /* INET6 */ 4913 } 4914 4915 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 4916 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4917 printf("pf: BAD ICMP %d:%d outer dst: ", 4918 icmptype, icmpcode); 4919 pf_print_host(pd->src, 0, pd->af); 4920 printf(" -> "); 4921 pf_print_host(pd->dst, 0, pd->af); 4922 printf(" inner src: "); 4923 pf_print_host(pd2.src, 0, pd2.af); 4924 printf(" -> "); 4925 pf_print_host(pd2.dst, 0, pd2.af); 4926 printf("\n"); 4927 } 4928 REASON_SET(reason, PFRES_BADSTATE); 4929 return (PF_DROP); 4930 } 4931 4932 switch (pd2.proto) { 4933 case IPPROTO_TCP: { 4934 struct tcphdr th; 4935 u_int32_t seq; 4936 struct pf_state_peer *src, *dst; 4937 u_int8_t dws; 4938 int copyback = 0; 4939 4940 /* 4941 * Only the first 8 bytes of the TCP header can be 4942 * expected. Don't access any TCP header fields after 4943 * th_seq, an ackskew test is not possible. 4944 */ 4945 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 4946 pd2.af)) { 4947 DPFPRINTF(PF_DEBUG_MISC, 4948 ("pf: ICMP error message too short " 4949 "(tcp)\n")); 4950 return (PF_DROP); 4951 } 4952 4953 key.af = pd2.af; 4954 key.proto = IPPROTO_TCP; 4955 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4956 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4957 key.port[pd2.sidx] = th.th_sport; 4958 key.port[pd2.didx] = th.th_dport; 4959 4960 STATE_LOOKUP(kif, &key, direction, *state, pd); 4961 4962 if (direction == (*state)->direction) { 4963 src = &(*state)->dst; 4964 dst = &(*state)->src; 4965 } else { 4966 src = &(*state)->src; 4967 dst = &(*state)->dst; 4968 } 4969 4970 if (src->wscale && dst->wscale) 4971 dws = dst->wscale & PF_WSCALE_MASK; 4972 else 4973 dws = 0; 4974 4975 /* Demodulate sequence number */ 4976 seq = ntohl(th.th_seq) - src->seqdiff; 4977 if (src->seqdiff) { 4978 pf_change_a(&th.th_seq, icmpsum, 4979 htonl(seq), 0); 4980 copyback = 1; 4981 } 4982 4983 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 4984 (!SEQ_GEQ(src->seqhi, seq) || 4985 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 4986 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4987 printf("pf: BAD ICMP %d:%d ", 4988 icmptype, icmpcode); 4989 pf_print_host(pd->src, 0, pd->af); 4990 printf(" -> "); 4991 pf_print_host(pd->dst, 0, pd->af); 4992 printf(" state: "); 4993 pf_print_state(*state); 4994 printf(" seq=%u\n", seq); 4995 } 4996 REASON_SET(reason, PFRES_BADSTATE); 4997 return (PF_DROP); 4998 } else { 4999 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5000 printf("pf: OK ICMP %d:%d ", 5001 icmptype, icmpcode); 5002 pf_print_host(pd->src, 0, pd->af); 5003 printf(" -> "); 5004 pf_print_host(pd->dst, 0, pd->af); 5005 printf(" state: "); 5006 pf_print_state(*state); 5007 printf(" seq=%u\n", seq); 5008 } 5009 } 5010 5011 /* translate source/destination address, if necessary */ 5012 if ((*state)->key[PF_SK_WIRE] != 5013 (*state)->key[PF_SK_STACK]) { 5014 struct pf_state_key *nk = 5015 (*state)->key[pd->didx]; 5016 5017 if (PF_ANEQ(pd2.src, 5018 &nk->addr[pd2.sidx], pd2.af) || 5019 nk->port[pd2.sidx] != th.th_sport) 5020 pf_change_icmp(pd2.src, &th.th_sport, 5021 daddr, &nk->addr[pd2.sidx], 5022 nk->port[pd2.sidx], NULL, 5023 pd2.ip_sum, icmpsum, 5024 pd->ip_sum, 0, pd2.af); 5025 5026 if (PF_ANEQ(pd2.dst, 5027 &nk->addr[pd2.didx], pd2.af) || 5028 nk->port[pd2.didx] != th.th_dport) 5029 pf_change_icmp(pd2.dst, &th.th_dport, 5030 saddr, &nk->addr[pd2.didx], 5031 nk->port[pd2.didx], NULL, 5032 pd2.ip_sum, icmpsum, 5033 pd->ip_sum, 0, pd2.af); 5034 copyback = 1; 5035 } 5036 5037 if (copyback) { 5038 switch (pd2.af) { 5039 #ifdef INET 5040 case AF_INET: 5041 m_copyback(m, off, ICMP_MINLEN, 5042 (caddr_t )pd->hdr.icmp); 5043 m_copyback(m, ipoff2, sizeof(h2), 5044 (caddr_t )&h2); 5045 break; 5046 #endif /* INET */ 5047 #ifdef INET6 5048 case AF_INET6: 5049 m_copyback(m, off, 5050 sizeof(struct icmp6_hdr), 5051 (caddr_t )pd->hdr.icmp6); 5052 m_copyback(m, ipoff2, sizeof(h2_6), 5053 (caddr_t )&h2_6); 5054 break; 5055 #endif /* INET6 */ 5056 } 5057 m_copyback(m, off2, 8, (caddr_t)&th); 5058 } 5059 5060 return (PF_PASS); 5061 break; 5062 } 5063 case IPPROTO_UDP: { 5064 struct udphdr uh; 5065 5066 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 5067 NULL, reason, pd2.af)) { 5068 DPFPRINTF(PF_DEBUG_MISC, 5069 ("pf: ICMP error message too short " 5070 "(udp)\n")); 5071 return (PF_DROP); 5072 } 5073 5074 key.af = pd2.af; 5075 key.proto = IPPROTO_UDP; 5076 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5077 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5078 key.port[pd2.sidx] = uh.uh_sport; 5079 key.port[pd2.didx] = uh.uh_dport; 5080 5081 STATE_LOOKUP(kif, &key, direction, *state, pd); 5082 5083 /* translate source/destination address, if necessary */ 5084 if ((*state)->key[PF_SK_WIRE] != 5085 (*state)->key[PF_SK_STACK]) { 5086 struct pf_state_key *nk = 5087 (*state)->key[pd->didx]; 5088 5089 if (PF_ANEQ(pd2.src, 5090 &nk->addr[pd2.sidx], pd2.af) || 5091 nk->port[pd2.sidx] != uh.uh_sport) 5092 pf_change_icmp(pd2.src, &uh.uh_sport, 5093 daddr, &nk->addr[pd2.sidx], 5094 nk->port[pd2.sidx], &uh.uh_sum, 5095 pd2.ip_sum, icmpsum, 5096 pd->ip_sum, 1, pd2.af); 5097 5098 if (PF_ANEQ(pd2.dst, 5099 &nk->addr[pd2.didx], pd2.af) || 5100 nk->port[pd2.didx] != uh.uh_dport) 5101 pf_change_icmp(pd2.dst, &uh.uh_dport, 5102 saddr, &nk->addr[pd2.didx], 5103 nk->port[pd2.didx], &uh.uh_sum, 5104 pd2.ip_sum, icmpsum, 5105 pd->ip_sum, 1, pd2.af); 5106 5107 switch (pd2.af) { 5108 #ifdef INET 5109 case AF_INET: 5110 m_copyback(m, off, ICMP_MINLEN, 5111 (caddr_t )pd->hdr.icmp); 5112 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5113 break; 5114 #endif /* INET */ 5115 #ifdef INET6 5116 case AF_INET6: 5117 m_copyback(m, off, 5118 sizeof(struct icmp6_hdr), 5119 (caddr_t )pd->hdr.icmp6); 5120 m_copyback(m, ipoff2, sizeof(h2_6), 5121 (caddr_t )&h2_6); 5122 break; 5123 #endif /* INET6 */ 5124 } 5125 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 5126 } 5127 return (PF_PASS); 5128 break; 5129 } 5130 #ifdef INET 5131 case IPPROTO_ICMP: { 5132 struct icmp iih; 5133 5134 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 5135 NULL, reason, pd2.af)) { 5136 DPFPRINTF(PF_DEBUG_MISC, 5137 ("pf: ICMP error message too short i" 5138 "(icmp)\n")); 5139 return (PF_DROP); 5140 } 5141 5142 key.af = pd2.af; 5143 key.proto = IPPROTO_ICMP; 5144 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5145 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5146 key.port[0] = key.port[1] = iih.icmp_id; 5147 5148 STATE_LOOKUP(kif, &key, direction, *state, pd); 5149 5150 /* translate source/destination address, if necessary */ 5151 if ((*state)->key[PF_SK_WIRE] != 5152 (*state)->key[PF_SK_STACK]) { 5153 struct pf_state_key *nk = 5154 (*state)->key[pd->didx]; 5155 5156 if (PF_ANEQ(pd2.src, 5157 &nk->addr[pd2.sidx], pd2.af) || 5158 nk->port[pd2.sidx] != iih.icmp_id) 5159 pf_change_icmp(pd2.src, &iih.icmp_id, 5160 daddr, &nk->addr[pd2.sidx], 5161 nk->port[pd2.sidx], NULL, 5162 pd2.ip_sum, icmpsum, 5163 pd->ip_sum, 0, AF_INET); 5164 5165 if (PF_ANEQ(pd2.dst, 5166 &nk->addr[pd2.didx], pd2.af) || 5167 nk->port[pd2.didx] != iih.icmp_id) 5168 pf_change_icmp(pd2.dst, &iih.icmp_id, 5169 saddr, &nk->addr[pd2.didx], 5170 nk->port[pd2.didx], NULL, 5171 pd2.ip_sum, icmpsum, 5172 pd->ip_sum, 0, AF_INET); 5173 5174 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 5175 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5176 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 5177 } 5178 return (PF_PASS); 5179 break; 5180 } 5181 #endif /* INET */ 5182 #ifdef INET6 5183 case IPPROTO_ICMPV6: { 5184 struct icmp6_hdr iih; 5185 5186 if (!pf_pull_hdr(m, off2, &iih, 5187 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 5188 DPFPRINTF(PF_DEBUG_MISC, 5189 ("pf: ICMP error message too short " 5190 "(icmp6)\n")); 5191 return (PF_DROP); 5192 } 5193 5194 key.af = pd2.af; 5195 key.proto = IPPROTO_ICMPV6; 5196 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5197 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5198 key.port[0] = key.port[1] = iih.icmp6_id; 5199 5200 STATE_LOOKUP(kif, &key, direction, *state, pd); 5201 5202 /* translate source/destination address, if necessary */ 5203 if ((*state)->key[PF_SK_WIRE] != 5204 (*state)->key[PF_SK_STACK]) { 5205 struct pf_state_key *nk = 5206 (*state)->key[pd->didx]; 5207 5208 if (PF_ANEQ(pd2.src, 5209 &nk->addr[pd2.sidx], pd2.af) || 5210 nk->port[pd2.sidx] != iih.icmp6_id) 5211 pf_change_icmp(pd2.src, &iih.icmp6_id, 5212 daddr, &nk->addr[pd2.sidx], 5213 nk->port[pd2.sidx], NULL, 5214 pd2.ip_sum, icmpsum, 5215 pd->ip_sum, 0, AF_INET6); 5216 5217 if (PF_ANEQ(pd2.dst, 5218 &nk->addr[pd2.didx], pd2.af) || 5219 nk->port[pd2.didx] != iih.icmp6_id) 5220 pf_change_icmp(pd2.dst, &iih.icmp6_id, 5221 saddr, &nk->addr[pd2.didx], 5222 nk->port[pd2.didx], NULL, 5223 pd2.ip_sum, icmpsum, 5224 pd->ip_sum, 0, AF_INET6); 5225 5226 m_copyback(m, off, sizeof(struct icmp6_hdr), 5227 (caddr_t)pd->hdr.icmp6); 5228 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 5229 m_copyback(m, off2, sizeof(struct icmp6_hdr), 5230 (caddr_t)&iih); 5231 } 5232 return (PF_PASS); 5233 break; 5234 } 5235 #endif /* INET6 */ 5236 default: { 5237 key.af = pd2.af; 5238 key.proto = pd2.proto; 5239 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5240 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5241 key.port[0] = key.port[1] = 0; 5242 5243 STATE_LOOKUP(kif, &key, direction, *state, pd); 5244 5245 /* translate source/destination address, if necessary */ 5246 if ((*state)->key[PF_SK_WIRE] != 5247 (*state)->key[PF_SK_STACK]) { 5248 struct pf_state_key *nk = 5249 (*state)->key[pd->didx]; 5250 5251 if (PF_ANEQ(pd2.src, 5252 &nk->addr[pd2.sidx], pd2.af)) 5253 pf_change_icmp(pd2.src, NULL, daddr, 5254 &nk->addr[pd2.sidx], 0, NULL, 5255 pd2.ip_sum, icmpsum, 5256 pd->ip_sum, 0, pd2.af); 5257 5258 if (PF_ANEQ(pd2.dst, 5259 &nk->addr[pd2.didx], pd2.af)) 5260 pf_change_icmp(pd2.dst, NULL, saddr, 5261 &nk->addr[pd2.didx], 0, NULL, 5262 pd2.ip_sum, icmpsum, 5263 pd->ip_sum, 0, pd2.af); 5264 5265 switch (pd2.af) { 5266 #ifdef INET 5267 case AF_INET: 5268 m_copyback(m, off, ICMP_MINLEN, 5269 (caddr_t)pd->hdr.icmp); 5270 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5271 break; 5272 #endif /* INET */ 5273 #ifdef INET6 5274 case AF_INET6: 5275 m_copyback(m, off, 5276 sizeof(struct icmp6_hdr), 5277 (caddr_t )pd->hdr.icmp6); 5278 m_copyback(m, ipoff2, sizeof(h2_6), 5279 (caddr_t )&h2_6); 5280 break; 5281 #endif /* INET6 */ 5282 } 5283 } 5284 return (PF_PASS); 5285 break; 5286 } 5287 } 5288 } 5289 } 5290 5291 static int 5292 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kkif *kif, 5293 struct mbuf *m, struct pf_pdesc *pd) 5294 { 5295 struct pf_state_peer *src, *dst; 5296 struct pf_state_key_cmp key; 5297 5298 bzero(&key, sizeof(key)); 5299 key.af = pd->af; 5300 key.proto = pd->proto; 5301 if (direction == PF_IN) { 5302 PF_ACPY(&key.addr[0], pd->src, key.af); 5303 PF_ACPY(&key.addr[1], pd->dst, key.af); 5304 key.port[0] = key.port[1] = 0; 5305 } else { 5306 PF_ACPY(&key.addr[1], pd->src, key.af); 5307 PF_ACPY(&key.addr[0], pd->dst, key.af); 5308 key.port[1] = key.port[0] = 0; 5309 } 5310 5311 STATE_LOOKUP(kif, &key, direction, *state, pd); 5312 5313 if (direction == (*state)->direction) { 5314 src = &(*state)->src; 5315 dst = &(*state)->dst; 5316 } else { 5317 src = &(*state)->dst; 5318 dst = &(*state)->src; 5319 } 5320 5321 /* update states */ 5322 if (src->state < PFOTHERS_SINGLE) 5323 src->state = PFOTHERS_SINGLE; 5324 if (dst->state == PFOTHERS_SINGLE) 5325 dst->state = PFOTHERS_MULTIPLE; 5326 5327 /* update expire time */ 5328 (*state)->expire = time_uptime; 5329 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 5330 (*state)->timeout = PFTM_OTHER_MULTIPLE; 5331 else 5332 (*state)->timeout = PFTM_OTHER_SINGLE; 5333 5334 /* translate source/destination address, if necessary */ 5335 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5336 struct pf_state_key *nk = (*state)->key[pd->didx]; 5337 5338 KASSERT(nk, ("%s: nk is null", __func__)); 5339 KASSERT(pd, ("%s: pd is null", __func__)); 5340 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 5341 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 5342 switch (pd->af) { 5343 #ifdef INET 5344 case AF_INET: 5345 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5346 pf_change_a(&pd->src->v4.s_addr, 5347 pd->ip_sum, 5348 nk->addr[pd->sidx].v4.s_addr, 5349 0); 5350 5351 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5352 pf_change_a(&pd->dst->v4.s_addr, 5353 pd->ip_sum, 5354 nk->addr[pd->didx].v4.s_addr, 5355 0); 5356 5357 break; 5358 #endif /* INET */ 5359 #ifdef INET6 5360 case AF_INET6: 5361 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5362 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5363 5364 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5365 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5366 #endif /* INET6 */ 5367 } 5368 } 5369 return (PF_PASS); 5370 } 5371 5372 /* 5373 * ipoff and off are measured from the start of the mbuf chain. 5374 * h must be at "ipoff" on the mbuf chain. 5375 */ 5376 void * 5377 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 5378 u_short *actionp, u_short *reasonp, sa_family_t af) 5379 { 5380 switch (af) { 5381 #ifdef INET 5382 case AF_INET: { 5383 struct ip *h = mtod(m, struct ip *); 5384 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 5385 5386 if (fragoff) { 5387 if (fragoff >= len) 5388 ACTION_SET(actionp, PF_PASS); 5389 else { 5390 ACTION_SET(actionp, PF_DROP); 5391 REASON_SET(reasonp, PFRES_FRAG); 5392 } 5393 return (NULL); 5394 } 5395 if (m->m_pkthdr.len < off + len || 5396 ntohs(h->ip_len) < off + len) { 5397 ACTION_SET(actionp, PF_DROP); 5398 REASON_SET(reasonp, PFRES_SHORT); 5399 return (NULL); 5400 } 5401 break; 5402 } 5403 #endif /* INET */ 5404 #ifdef INET6 5405 case AF_INET6: { 5406 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 5407 5408 if (m->m_pkthdr.len < off + len || 5409 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 5410 (unsigned)(off + len)) { 5411 ACTION_SET(actionp, PF_DROP); 5412 REASON_SET(reasonp, PFRES_SHORT); 5413 return (NULL); 5414 } 5415 break; 5416 } 5417 #endif /* INET6 */ 5418 } 5419 m_copydata(m, off, len, p); 5420 return (p); 5421 } 5422 5423 int 5424 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 5425 int rtableid) 5426 { 5427 struct ifnet *ifp; 5428 5429 /* 5430 * Skip check for addresses with embedded interface scope, 5431 * as they would always match anyway. 5432 */ 5433 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 5434 return (1); 5435 5436 if (af != AF_INET && af != AF_INET6) 5437 return (0); 5438 5439 /* Skip checks for ipsec interfaces */ 5440 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 5441 return (1); 5442 5443 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 5444 5445 switch (af) { 5446 #ifdef INET6 5447 case AF_INET6: 5448 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 5449 ifp)); 5450 #endif 5451 #ifdef INET 5452 case AF_INET: 5453 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 5454 ifp)); 5455 #endif 5456 } 5457 5458 return (0); 5459 } 5460 5461 #ifdef INET 5462 static void 5463 pf_route(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp, 5464 struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp) 5465 { 5466 struct mbuf *m0, *m1; 5467 struct sockaddr_in dst; 5468 struct ip *ip; 5469 struct ifnet *ifp = NULL; 5470 struct pf_addr naddr; 5471 struct pf_ksrc_node *sn = NULL; 5472 int error = 0; 5473 uint16_t ip_len, ip_off; 5474 5475 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 5476 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 5477 __func__)); 5478 5479 if ((pd->pf_mtag == NULL && 5480 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 5481 pd->pf_mtag->routed++ > 3) { 5482 m0 = *m; 5483 *m = NULL; 5484 goto bad_locked; 5485 } 5486 5487 if (r->rt == PF_DUPTO) { 5488 if ((pd->pf_mtag->flags & PF_DUPLICATED)) { 5489 if (s == NULL) { 5490 ifp = r->rpool.cur->kif ? 5491 r->rpool.cur->kif->pfik_ifp : NULL; 5492 } else { 5493 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5494 PF_STATE_UNLOCK(s); 5495 } 5496 if (ifp == oifp) { 5497 /* When the 2nd interface is not skipped */ 5498 return; 5499 } else { 5500 m0 = *m; 5501 *m = NULL; 5502 goto bad; 5503 } 5504 } else { 5505 pd->pf_mtag->flags |= PF_DUPLICATED; 5506 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 5507 if (s) 5508 PF_STATE_UNLOCK(s); 5509 return; 5510 } 5511 } 5512 } else { 5513 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 5514 if (s) 5515 PF_STATE_UNLOCK(s); 5516 return; 5517 } 5518 m0 = *m; 5519 } 5520 5521 ip = mtod(m0, struct ip *); 5522 5523 bzero(&dst, sizeof(dst)); 5524 dst.sin_family = AF_INET; 5525 dst.sin_len = sizeof(dst); 5526 dst.sin_addr = ip->ip_dst; 5527 5528 bzero(&naddr, sizeof(naddr)); 5529 5530 if (TAILQ_EMPTY(&r->rpool.list)) { 5531 DPFPRINTF(PF_DEBUG_URGENT, 5532 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 5533 goto bad_locked; 5534 } 5535 if (s == NULL) { 5536 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 5537 &naddr, NULL, &sn); 5538 if (!PF_AZERO(&naddr, AF_INET)) 5539 dst.sin_addr.s_addr = naddr.v4.s_addr; 5540 ifp = r->rpool.cur->kif ? 5541 r->rpool.cur->kif->pfik_ifp : NULL; 5542 } else { 5543 if (!PF_AZERO(&s->rt_addr, AF_INET)) 5544 dst.sin_addr.s_addr = 5545 s->rt_addr.v4.s_addr; 5546 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5547 PF_STATE_UNLOCK(s); 5548 } 5549 if (ifp == NULL) 5550 goto bad; 5551 5552 if (oifp != ifp) { 5553 if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS) 5554 goto bad; 5555 else if (m0 == NULL) 5556 goto done; 5557 if (m0->m_len < sizeof(struct ip)) { 5558 DPFPRINTF(PF_DEBUG_URGENT, 5559 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 5560 goto bad; 5561 } 5562 ip = mtod(m0, struct ip *); 5563 } 5564 5565 if (ifp->if_flags & IFF_LOOPBACK) 5566 m0->m_flags |= M_SKIP_FIREWALL; 5567 5568 ip_len = ntohs(ip->ip_len); 5569 ip_off = ntohs(ip->ip_off); 5570 5571 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 5572 m0->m_pkthdr.csum_flags |= CSUM_IP; 5573 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 5574 in_delayed_cksum(m0); 5575 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 5576 } 5577 #if defined(SCTP) || defined(SCTP_SUPPORT) 5578 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 5579 sctp_delayed_cksum(m0, (uint32_t)(ip->ip_hl << 2)); 5580 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 5581 } 5582 #endif 5583 5584 /* 5585 * If small enough for interface, or the interface will take 5586 * care of the fragmentation for us, we can just send directly. 5587 */ 5588 if (ip_len <= ifp->if_mtu || 5589 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 5590 ip->ip_sum = 0; 5591 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 5592 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 5593 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 5594 } 5595 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 5596 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); 5597 goto done; 5598 } 5599 5600 /* Balk when DF bit is set or the interface didn't support TSO. */ 5601 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 5602 error = EMSGSIZE; 5603 KMOD_IPSTAT_INC(ips_cantfrag); 5604 if (r->rt != PF_DUPTO) { 5605 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 5606 ifp->if_mtu); 5607 goto done; 5608 } else 5609 goto bad; 5610 } 5611 5612 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 5613 if (error) 5614 goto bad; 5615 5616 for (; m0; m0 = m1) { 5617 m1 = m0->m_nextpkt; 5618 m0->m_nextpkt = NULL; 5619 if (error == 0) { 5620 m_clrprotoflags(m0); 5621 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); 5622 } else 5623 m_freem(m0); 5624 } 5625 5626 if (error == 0) 5627 KMOD_IPSTAT_INC(ips_fragmented); 5628 5629 done: 5630 if (r->rt != PF_DUPTO) 5631 *m = NULL; 5632 return; 5633 5634 bad_locked: 5635 if (s) 5636 PF_STATE_UNLOCK(s); 5637 bad: 5638 m_freem(m0); 5639 goto done; 5640 } 5641 #endif /* INET */ 5642 5643 #ifdef INET6 5644 static void 5645 pf_route6(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp, 5646 struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp) 5647 { 5648 struct mbuf *m0; 5649 struct sockaddr_in6 dst; 5650 struct ip6_hdr *ip6; 5651 struct ifnet *ifp = NULL; 5652 struct pf_addr naddr; 5653 struct pf_ksrc_node *sn = NULL; 5654 5655 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 5656 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 5657 __func__)); 5658 5659 if ((pd->pf_mtag == NULL && 5660 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 5661 pd->pf_mtag->routed++ > 3) { 5662 m0 = *m; 5663 *m = NULL; 5664 goto bad_locked; 5665 } 5666 5667 if (r->rt == PF_DUPTO) { 5668 if ((pd->pf_mtag->flags & PF_DUPLICATED)) { 5669 if (s == NULL) { 5670 ifp = r->rpool.cur->kif ? 5671 r->rpool.cur->kif->pfik_ifp : NULL; 5672 } else { 5673 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5674 PF_STATE_UNLOCK(s); 5675 } 5676 if (ifp == oifp) { 5677 /* When the 2nd interface is not skipped */ 5678 return; 5679 } else { 5680 m0 = *m; 5681 *m = NULL; 5682 goto bad; 5683 } 5684 } else { 5685 pd->pf_mtag->flags |= PF_DUPLICATED; 5686 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 5687 if (s) 5688 PF_STATE_UNLOCK(s); 5689 return; 5690 } 5691 } 5692 } else { 5693 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 5694 if (s) 5695 PF_STATE_UNLOCK(s); 5696 return; 5697 } 5698 m0 = *m; 5699 } 5700 5701 ip6 = mtod(m0, struct ip6_hdr *); 5702 5703 bzero(&dst, sizeof(dst)); 5704 dst.sin6_family = AF_INET6; 5705 dst.sin6_len = sizeof(dst); 5706 dst.sin6_addr = ip6->ip6_dst; 5707 5708 bzero(&naddr, sizeof(naddr)); 5709 5710 if (TAILQ_EMPTY(&r->rpool.list)) { 5711 DPFPRINTF(PF_DEBUG_URGENT, 5712 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 5713 goto bad_locked; 5714 } 5715 if (s == NULL) { 5716 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 5717 &naddr, NULL, &sn); 5718 if (!PF_AZERO(&naddr, AF_INET6)) 5719 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 5720 &naddr, AF_INET6); 5721 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 5722 } else { 5723 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 5724 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 5725 &s->rt_addr, AF_INET6); 5726 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5727 } 5728 5729 if (s) 5730 PF_STATE_UNLOCK(s); 5731 5732 if (ifp == NULL) 5733 goto bad; 5734 5735 if (oifp != ifp) { 5736 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS) 5737 goto bad; 5738 else if (m0 == NULL) 5739 goto done; 5740 if (m0->m_len < sizeof(struct ip6_hdr)) { 5741 DPFPRINTF(PF_DEBUG_URGENT, 5742 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 5743 __func__)); 5744 goto bad; 5745 } 5746 ip6 = mtod(m0, struct ip6_hdr *); 5747 } 5748 5749 if (ifp->if_flags & IFF_LOOPBACK) 5750 m0->m_flags |= M_SKIP_FIREWALL; 5751 5752 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 5753 ~ifp->if_hwassist) { 5754 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 5755 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 5756 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 5757 } 5758 5759 /* 5760 * If the packet is too large for the outgoing interface, 5761 * send back an icmp6 error. 5762 */ 5763 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 5764 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 5765 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) 5766 nd6_output_ifp(ifp, ifp, m0, &dst, NULL); 5767 else { 5768 in6_ifstat_inc(ifp, ifs6_in_toobig); 5769 if (r->rt != PF_DUPTO) 5770 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 5771 else 5772 goto bad; 5773 } 5774 5775 done: 5776 if (r->rt != PF_DUPTO) 5777 *m = NULL; 5778 return; 5779 5780 bad_locked: 5781 if (s) 5782 PF_STATE_UNLOCK(s); 5783 bad: 5784 m_freem(m0); 5785 goto done; 5786 } 5787 #endif /* INET6 */ 5788 5789 /* 5790 * FreeBSD supports cksum offloads for the following drivers. 5791 * em(4), fxp(4), lge(4), ndis(4), nge(4), re(4), ti(4), txp(4), xl(4) 5792 * 5793 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 5794 * network driver performed cksum including pseudo header, need to verify 5795 * csum_data 5796 * CSUM_DATA_VALID : 5797 * network driver performed cksum, needs to additional pseudo header 5798 * cksum computation with partial csum_data(i.e. lack of H/W support for 5799 * pseudo header, for instance sk(4) and possibly gem(4)) 5800 * 5801 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 5802 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 5803 * TCP/UDP layer. 5804 * Also, set csum_data to 0xffff to force cksum validation. 5805 */ 5806 static int 5807 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 5808 { 5809 u_int16_t sum = 0; 5810 int hw_assist = 0; 5811 struct ip *ip; 5812 5813 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 5814 return (1); 5815 if (m->m_pkthdr.len < off + len) 5816 return (1); 5817 5818 switch (p) { 5819 case IPPROTO_TCP: 5820 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 5821 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 5822 sum = m->m_pkthdr.csum_data; 5823 } else { 5824 ip = mtod(m, struct ip *); 5825 sum = in_pseudo(ip->ip_src.s_addr, 5826 ip->ip_dst.s_addr, htonl((u_short)len + 5827 m->m_pkthdr.csum_data + IPPROTO_TCP)); 5828 } 5829 sum ^= 0xffff; 5830 ++hw_assist; 5831 } 5832 break; 5833 case IPPROTO_UDP: 5834 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 5835 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 5836 sum = m->m_pkthdr.csum_data; 5837 } else { 5838 ip = mtod(m, struct ip *); 5839 sum = in_pseudo(ip->ip_src.s_addr, 5840 ip->ip_dst.s_addr, htonl((u_short)len + 5841 m->m_pkthdr.csum_data + IPPROTO_UDP)); 5842 } 5843 sum ^= 0xffff; 5844 ++hw_assist; 5845 } 5846 break; 5847 case IPPROTO_ICMP: 5848 #ifdef INET6 5849 case IPPROTO_ICMPV6: 5850 #endif /* INET6 */ 5851 break; 5852 default: 5853 return (1); 5854 } 5855 5856 if (!hw_assist) { 5857 switch (af) { 5858 case AF_INET: 5859 if (p == IPPROTO_ICMP) { 5860 if (m->m_len < off) 5861 return (1); 5862 m->m_data += off; 5863 m->m_len -= off; 5864 sum = in_cksum(m, len); 5865 m->m_data -= off; 5866 m->m_len += off; 5867 } else { 5868 if (m->m_len < sizeof(struct ip)) 5869 return (1); 5870 sum = in4_cksum(m, p, off, len); 5871 } 5872 break; 5873 #ifdef INET6 5874 case AF_INET6: 5875 if (m->m_len < sizeof(struct ip6_hdr)) 5876 return (1); 5877 sum = in6_cksum(m, p, off, len); 5878 break; 5879 #endif /* INET6 */ 5880 default: 5881 return (1); 5882 } 5883 } 5884 if (sum) { 5885 switch (p) { 5886 case IPPROTO_TCP: 5887 { 5888 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 5889 break; 5890 } 5891 case IPPROTO_UDP: 5892 { 5893 KMOD_UDPSTAT_INC(udps_badsum); 5894 break; 5895 } 5896 #ifdef INET 5897 case IPPROTO_ICMP: 5898 { 5899 KMOD_ICMPSTAT_INC(icps_checksum); 5900 break; 5901 } 5902 #endif 5903 #ifdef INET6 5904 case IPPROTO_ICMPV6: 5905 { 5906 KMOD_ICMP6STAT_INC(icp6s_checksum); 5907 break; 5908 } 5909 #endif /* INET6 */ 5910 } 5911 return (1); 5912 } else { 5913 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 5914 m->m_pkthdr.csum_flags |= 5915 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 5916 m->m_pkthdr.csum_data = 0xffff; 5917 } 5918 } 5919 return (0); 5920 } 5921 5922 #ifdef INET 5923 int 5924 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 5925 { 5926 struct pfi_kkif *kif; 5927 u_short action, reason = 0, log = 0; 5928 struct mbuf *m = *m0; 5929 struct ip *h = NULL; 5930 struct m_tag *ipfwtag; 5931 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 5932 struct pf_state *s = NULL; 5933 struct pf_kruleset *ruleset = NULL; 5934 struct pf_pdesc pd; 5935 int off, dirndx, pqid = 0; 5936 5937 PF_RULES_RLOCK_TRACKER; 5938 5939 M_ASSERTPKTHDR(m); 5940 5941 if (!V_pf_status.running) 5942 return (PF_PASS); 5943 5944 memset(&pd, 0, sizeof(pd)); 5945 5946 kif = (struct pfi_kkif *)ifp->if_pf_kif; 5947 5948 if (kif == NULL) { 5949 DPFPRINTF(PF_DEBUG_URGENT, 5950 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 5951 return (PF_DROP); 5952 } 5953 if (kif->pfik_flags & PFI_IFLAG_SKIP) 5954 return (PF_PASS); 5955 5956 if (m->m_flags & M_SKIP_FIREWALL) 5957 return (PF_PASS); 5958 5959 pd.pf_mtag = pf_find_mtag(m); 5960 5961 PF_RULES_RLOCK(); 5962 5963 if (ip_divert_ptr != NULL && 5964 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { 5965 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); 5966 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) { 5967 if (pd.pf_mtag == NULL && 5968 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 5969 action = PF_DROP; 5970 goto done; 5971 } 5972 pd.pf_mtag->flags |= PF_PACKET_LOOPED; 5973 m_tag_delete(m, ipfwtag); 5974 } 5975 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) { 5976 m->m_flags |= M_FASTFWD_OURS; 5977 pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT; 5978 } 5979 } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { 5980 /* We do IP header normalization and packet reassembly here */ 5981 action = PF_DROP; 5982 goto done; 5983 } 5984 m = *m0; /* pf_normalize messes with m0 */ 5985 h = mtod(m, struct ip *); 5986 5987 off = h->ip_hl << 2; 5988 if (off < (int)sizeof(struct ip)) { 5989 action = PF_DROP; 5990 REASON_SET(&reason, PFRES_SHORT); 5991 log = 1; 5992 goto done; 5993 } 5994 5995 pd.src = (struct pf_addr *)&h->ip_src; 5996 pd.dst = (struct pf_addr *)&h->ip_dst; 5997 pd.sport = pd.dport = NULL; 5998 pd.ip_sum = &h->ip_sum; 5999 pd.proto_sum = NULL; 6000 pd.proto = h->ip_p; 6001 pd.dir = dir; 6002 pd.sidx = (dir == PF_IN) ? 0 : 1; 6003 pd.didx = (dir == PF_IN) ? 1 : 0; 6004 pd.af = AF_INET; 6005 pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; 6006 pd.tot_len = ntohs(h->ip_len); 6007 6008 /* handle fragments that didn't get reassembled by normalization */ 6009 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 6010 action = pf_test_fragment(&r, dir, kif, m, h, 6011 &pd, &a, &ruleset); 6012 goto done; 6013 } 6014 6015 switch (h->ip_p) { 6016 case IPPROTO_TCP: { 6017 struct tcphdr th; 6018 6019 pd.hdr.tcp = &th; 6020 if (!pf_pull_hdr(m, off, &th, sizeof(th), 6021 &action, &reason, AF_INET)) { 6022 log = action != PF_PASS; 6023 goto done; 6024 } 6025 pd.p_len = pd.tot_len - off - (th.th_off << 2); 6026 if ((th.th_flags & TH_ACK) && pd.p_len == 0) 6027 pqid = 1; 6028 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6029 if (action == PF_DROP) 6030 goto done; 6031 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6032 &reason); 6033 if (action == PF_PASS) { 6034 if (V_pfsync_update_state_ptr != NULL) 6035 V_pfsync_update_state_ptr(s); 6036 r = s->rule.ptr; 6037 a = s->anchor.ptr; 6038 log = s->log; 6039 } else if (s == NULL) 6040 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6041 &a, &ruleset, inp); 6042 break; 6043 } 6044 6045 case IPPROTO_UDP: { 6046 struct udphdr uh; 6047 6048 pd.hdr.udp = &uh; 6049 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 6050 &action, &reason, AF_INET)) { 6051 log = action != PF_PASS; 6052 goto done; 6053 } 6054 if (uh.uh_dport == 0 || 6055 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 6056 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 6057 action = PF_DROP; 6058 REASON_SET(&reason, PFRES_SHORT); 6059 goto done; 6060 } 6061 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 6062 if (action == PF_PASS) { 6063 if (V_pfsync_update_state_ptr != NULL) 6064 V_pfsync_update_state_ptr(s); 6065 r = s->rule.ptr; 6066 a = s->anchor.ptr; 6067 log = s->log; 6068 } else if (s == NULL) 6069 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6070 &a, &ruleset, inp); 6071 break; 6072 } 6073 6074 case IPPROTO_ICMP: { 6075 struct icmp ih; 6076 6077 pd.hdr.icmp = &ih; 6078 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN, 6079 &action, &reason, AF_INET)) { 6080 log = action != PF_PASS; 6081 goto done; 6082 } 6083 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, 6084 &reason); 6085 if (action == PF_PASS) { 6086 if (V_pfsync_update_state_ptr != NULL) 6087 V_pfsync_update_state_ptr(s); 6088 r = s->rule.ptr; 6089 a = s->anchor.ptr; 6090 log = s->log; 6091 } else if (s == NULL) 6092 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6093 &a, &ruleset, inp); 6094 break; 6095 } 6096 6097 #ifdef INET6 6098 case IPPROTO_ICMPV6: { 6099 action = PF_DROP; 6100 DPFPRINTF(PF_DEBUG_MISC, 6101 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 6102 goto done; 6103 } 6104 #endif 6105 6106 default: 6107 action = pf_test_state_other(&s, dir, kif, m, &pd); 6108 if (action == PF_PASS) { 6109 if (V_pfsync_update_state_ptr != NULL) 6110 V_pfsync_update_state_ptr(s); 6111 r = s->rule.ptr; 6112 a = s->anchor.ptr; 6113 log = s->log; 6114 } else if (s == NULL) 6115 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6116 &a, &ruleset, inp); 6117 break; 6118 } 6119 6120 done: 6121 PF_RULES_RUNLOCK(); 6122 if (action == PF_PASS && h->ip_hl > 5 && 6123 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 6124 action = PF_DROP; 6125 REASON_SET(&reason, PFRES_IPOPTIONS); 6126 log = r->log; 6127 DPFPRINTF(PF_DEBUG_MISC, 6128 ("pf: dropping packet with ip options\n")); 6129 } 6130 6131 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 6132 action = PF_DROP; 6133 REASON_SET(&reason, PFRES_MEMORY); 6134 } 6135 if (r->rtableid >= 0) 6136 M_SETFIB(m, r->rtableid); 6137 6138 if (r->scrub_flags & PFSTATE_SETPRIO) { 6139 if (pd.tos & IPTOS_LOWDELAY) 6140 pqid = 1; 6141 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) { 6142 action = PF_DROP; 6143 REASON_SET(&reason, PFRES_MEMORY); 6144 log = 1; 6145 DPFPRINTF(PF_DEBUG_MISC, 6146 ("pf: failed to allocate 802.1q mtag\n")); 6147 } 6148 } 6149 6150 #ifdef ALTQ 6151 if (action == PF_PASS && r->qid) { 6152 if (pd.pf_mtag == NULL && 6153 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6154 action = PF_DROP; 6155 REASON_SET(&reason, PFRES_MEMORY); 6156 } else { 6157 if (s != NULL) 6158 pd.pf_mtag->qid_hash = pf_state_hash(s); 6159 if (pqid || (pd.tos & IPTOS_LOWDELAY)) 6160 pd.pf_mtag->qid = r->pqid; 6161 else 6162 pd.pf_mtag->qid = r->qid; 6163 /* Add hints for ecn. */ 6164 pd.pf_mtag->hdr = h; 6165 } 6166 } 6167 #endif /* ALTQ */ 6168 6169 /* 6170 * connections redirected to loopback should not match sockets 6171 * bound specifically to loopback due to security implications, 6172 * see tcp_input() and in_pcblookup_listen(). 6173 */ 6174 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 6175 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 6176 (s->nat_rule.ptr->action == PF_RDR || 6177 s->nat_rule.ptr->action == PF_BINAT) && 6178 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 6179 m->m_flags |= M_SKIP_FIREWALL; 6180 6181 if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL && 6182 !PACKET_LOOPED(&pd)) { 6183 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 6184 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 6185 if (ipfwtag != NULL) { 6186 ((struct ipfw_rule_ref *)(ipfwtag+1))->info = 6187 ntohs(r->divert.port); 6188 ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir; 6189 6190 if (s) 6191 PF_STATE_UNLOCK(s); 6192 6193 m_tag_prepend(m, ipfwtag); 6194 if (m->m_flags & M_FASTFWD_OURS) { 6195 if (pd.pf_mtag == NULL && 6196 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6197 action = PF_DROP; 6198 REASON_SET(&reason, PFRES_MEMORY); 6199 log = 1; 6200 DPFPRINTF(PF_DEBUG_MISC, 6201 ("pf: failed to allocate tag\n")); 6202 } else { 6203 pd.pf_mtag->flags |= 6204 PF_FASTFWD_OURS_PRESENT; 6205 m->m_flags &= ~M_FASTFWD_OURS; 6206 } 6207 } 6208 ip_divert_ptr(*m0, dir == PF_IN); 6209 *m0 = NULL; 6210 6211 return (action); 6212 } else { 6213 /* XXX: ipfw has the same behaviour! */ 6214 action = PF_DROP; 6215 REASON_SET(&reason, PFRES_MEMORY); 6216 log = 1; 6217 DPFPRINTF(PF_DEBUG_MISC, 6218 ("pf: failed to allocate divert tag\n")); 6219 } 6220 } 6221 6222 if (log) { 6223 struct pf_krule *lr; 6224 6225 if (s != NULL && s->nat_rule.ptr != NULL && 6226 s->nat_rule.ptr->log & PF_LOG_ALL) 6227 lr = s->nat_rule.ptr; 6228 else 6229 lr = r; 6230 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd, 6231 (s == NULL)); 6232 } 6233 6234 counter_u64_add(kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS], 6235 pd.tot_len); 6236 counter_u64_add(kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS], 6237 1); 6238 6239 if (action == PF_PASS || r->action == PF_DROP) { 6240 dirndx = (dir == PF_OUT); 6241 counter_u64_add(r->packets[dirndx], 1); 6242 counter_u64_add(r->bytes[dirndx], pd.tot_len); 6243 if (a != NULL) { 6244 counter_u64_add(a->packets[dirndx], 1); 6245 counter_u64_add(a->bytes[dirndx], pd.tot_len); 6246 } 6247 if (s != NULL) { 6248 if (s->nat_rule.ptr != NULL) { 6249 counter_u64_add(s->nat_rule.ptr->packets[dirndx], 6250 1); 6251 counter_u64_add(s->nat_rule.ptr->bytes[dirndx], 6252 pd.tot_len); 6253 } 6254 if (s->src_node != NULL) { 6255 counter_u64_add(s->src_node->packets[dirndx], 6256 1); 6257 counter_u64_add(s->src_node->bytes[dirndx], 6258 pd.tot_len); 6259 } 6260 if (s->nat_src_node != NULL) { 6261 counter_u64_add(s->nat_src_node->packets[dirndx], 6262 1); 6263 counter_u64_add(s->nat_src_node->bytes[dirndx], 6264 pd.tot_len); 6265 } 6266 dirndx = (dir == s->direction) ? 0 : 1; 6267 counter_u64_add(s->packets[dirndx], 1); 6268 counter_u64_add(s->bytes[dirndx], pd.tot_len); 6269 } 6270 tr = r; 6271 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 6272 if (nr != NULL && r == &V_pf_default_rule) 6273 tr = nr; 6274 if (tr->src.addr.type == PF_ADDR_TABLE) 6275 pfr_update_stats(tr->src.addr.p.tbl, 6276 (s == NULL) ? pd.src : 6277 &s->key[(s->direction == PF_IN)]-> 6278 addr[(s->direction == PF_OUT)], 6279 pd.af, pd.tot_len, dir == PF_OUT, 6280 r->action == PF_PASS, tr->src.neg); 6281 if (tr->dst.addr.type == PF_ADDR_TABLE) 6282 pfr_update_stats(tr->dst.addr.p.tbl, 6283 (s == NULL) ? pd.dst : 6284 &s->key[(s->direction == PF_IN)]-> 6285 addr[(s->direction == PF_IN)], 6286 pd.af, pd.tot_len, dir == PF_OUT, 6287 r->action == PF_PASS, tr->dst.neg); 6288 } 6289 6290 switch (action) { 6291 case PF_SYNPROXY_DROP: 6292 m_freem(*m0); 6293 case PF_DEFER: 6294 *m0 = NULL; 6295 action = PF_PASS; 6296 break; 6297 case PF_DROP: 6298 m_freem(*m0); 6299 *m0 = NULL; 6300 break; 6301 default: 6302 /* pf_route() returns unlocked. */ 6303 if (r->rt) { 6304 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp); 6305 return (action); 6306 } 6307 break; 6308 } 6309 if (s) 6310 PF_STATE_UNLOCK(s); 6311 6312 return (action); 6313 } 6314 #endif /* INET */ 6315 6316 #ifdef INET6 6317 int 6318 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 6319 { 6320 struct pfi_kkif *kif; 6321 u_short action, reason = 0, log = 0; 6322 struct mbuf *m = *m0, *n = NULL; 6323 struct m_tag *mtag; 6324 struct ip6_hdr *h = NULL; 6325 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 6326 struct pf_state *s = NULL; 6327 struct pf_kruleset *ruleset = NULL; 6328 struct pf_pdesc pd; 6329 int off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0; 6330 6331 PF_RULES_RLOCK_TRACKER; 6332 M_ASSERTPKTHDR(m); 6333 6334 if (!V_pf_status.running) 6335 return (PF_PASS); 6336 6337 memset(&pd, 0, sizeof(pd)); 6338 pd.pf_mtag = pf_find_mtag(m); 6339 6340 if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED) 6341 return (PF_PASS); 6342 6343 kif = (struct pfi_kkif *)ifp->if_pf_kif; 6344 if (kif == NULL) { 6345 DPFPRINTF(PF_DEBUG_URGENT, 6346 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 6347 return (PF_DROP); 6348 } 6349 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6350 return (PF_PASS); 6351 6352 if (m->m_flags & M_SKIP_FIREWALL) 6353 return (PF_PASS); 6354 6355 PF_RULES_RLOCK(); 6356 6357 /* We do IP header normalization and packet reassembly here */ 6358 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { 6359 action = PF_DROP; 6360 goto done; 6361 } 6362 m = *m0; /* pf_normalize messes with m0 */ 6363 h = mtod(m, struct ip6_hdr *); 6364 6365 /* 6366 * we do not support jumbogram. if we keep going, zero ip6_plen 6367 * will do something bad, so drop the packet for now. 6368 */ 6369 if (htons(h->ip6_plen) == 0) { 6370 action = PF_DROP; 6371 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 6372 goto done; 6373 } 6374 6375 pd.src = (struct pf_addr *)&h->ip6_src; 6376 pd.dst = (struct pf_addr *)&h->ip6_dst; 6377 pd.sport = pd.dport = NULL; 6378 pd.ip_sum = NULL; 6379 pd.proto_sum = NULL; 6380 pd.dir = dir; 6381 pd.sidx = (dir == PF_IN) ? 0 : 1; 6382 pd.didx = (dir == PF_IN) ? 1 : 0; 6383 pd.af = AF_INET6; 6384 pd.tos = IPV6_DSCP(h); 6385 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 6386 6387 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 6388 pd.proto = h->ip6_nxt; 6389 do { 6390 switch (pd.proto) { 6391 case IPPROTO_FRAGMENT: 6392 action = pf_test_fragment(&r, dir, kif, m, h, 6393 &pd, &a, &ruleset); 6394 if (action == PF_DROP) 6395 REASON_SET(&reason, PFRES_FRAG); 6396 goto done; 6397 case IPPROTO_ROUTING: { 6398 struct ip6_rthdr rthdr; 6399 6400 if (rh_cnt++) { 6401 DPFPRINTF(PF_DEBUG_MISC, 6402 ("pf: IPv6 more than one rthdr\n")); 6403 action = PF_DROP; 6404 REASON_SET(&reason, PFRES_IPOPTIONS); 6405 log = 1; 6406 goto done; 6407 } 6408 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 6409 &reason, pd.af)) { 6410 DPFPRINTF(PF_DEBUG_MISC, 6411 ("pf: IPv6 short rthdr\n")); 6412 action = PF_DROP; 6413 REASON_SET(&reason, PFRES_SHORT); 6414 log = 1; 6415 goto done; 6416 } 6417 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 6418 DPFPRINTF(PF_DEBUG_MISC, 6419 ("pf: IPv6 rthdr0\n")); 6420 action = PF_DROP; 6421 REASON_SET(&reason, PFRES_IPOPTIONS); 6422 log = 1; 6423 goto done; 6424 } 6425 /* FALLTHROUGH */ 6426 } 6427 case IPPROTO_AH: 6428 case IPPROTO_HOPOPTS: 6429 case IPPROTO_DSTOPTS: { 6430 /* get next header and header length */ 6431 struct ip6_ext opt6; 6432 6433 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 6434 NULL, &reason, pd.af)) { 6435 DPFPRINTF(PF_DEBUG_MISC, 6436 ("pf: IPv6 short opt\n")); 6437 action = PF_DROP; 6438 log = 1; 6439 goto done; 6440 } 6441 if (pd.proto == IPPROTO_AH) 6442 off += (opt6.ip6e_len + 2) * 4; 6443 else 6444 off += (opt6.ip6e_len + 1) * 8; 6445 pd.proto = opt6.ip6e_nxt; 6446 /* goto the next header */ 6447 break; 6448 } 6449 default: 6450 terminal++; 6451 break; 6452 } 6453 } while (!terminal); 6454 6455 /* if there's no routing header, use unmodified mbuf for checksumming */ 6456 if (!n) 6457 n = m; 6458 6459 switch (pd.proto) { 6460 case IPPROTO_TCP: { 6461 struct tcphdr th; 6462 6463 pd.hdr.tcp = &th; 6464 if (!pf_pull_hdr(m, off, &th, sizeof(th), 6465 &action, &reason, AF_INET6)) { 6466 log = action != PF_PASS; 6467 goto done; 6468 } 6469 pd.p_len = pd.tot_len - off - (th.th_off << 2); 6470 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6471 if (action == PF_DROP) 6472 goto done; 6473 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6474 &reason); 6475 if (action == PF_PASS) { 6476 if (V_pfsync_update_state_ptr != NULL) 6477 V_pfsync_update_state_ptr(s); 6478 r = s->rule.ptr; 6479 a = s->anchor.ptr; 6480 log = s->log; 6481 } else if (s == NULL) 6482 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6483 &a, &ruleset, inp); 6484 break; 6485 } 6486 6487 case IPPROTO_UDP: { 6488 struct udphdr uh; 6489 6490 pd.hdr.udp = &uh; 6491 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 6492 &action, &reason, AF_INET6)) { 6493 log = action != PF_PASS; 6494 goto done; 6495 } 6496 if (uh.uh_dport == 0 || 6497 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 6498 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 6499 action = PF_DROP; 6500 REASON_SET(&reason, PFRES_SHORT); 6501 goto done; 6502 } 6503 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 6504 if (action == PF_PASS) { 6505 if (V_pfsync_update_state_ptr != NULL) 6506 V_pfsync_update_state_ptr(s); 6507 r = s->rule.ptr; 6508 a = s->anchor.ptr; 6509 log = s->log; 6510 } else if (s == NULL) 6511 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6512 &a, &ruleset, inp); 6513 break; 6514 } 6515 6516 case IPPROTO_ICMP: { 6517 action = PF_DROP; 6518 DPFPRINTF(PF_DEBUG_MISC, 6519 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 6520 goto done; 6521 } 6522 6523 case IPPROTO_ICMPV6: { 6524 struct icmp6_hdr ih; 6525 6526 pd.hdr.icmp6 = &ih; 6527 if (!pf_pull_hdr(m, off, &ih, sizeof(ih), 6528 &action, &reason, AF_INET6)) { 6529 log = action != PF_PASS; 6530 goto done; 6531 } 6532 action = pf_test_state_icmp(&s, dir, kif, 6533 m, off, h, &pd, &reason); 6534 if (action == PF_PASS) { 6535 if (V_pfsync_update_state_ptr != NULL) 6536 V_pfsync_update_state_ptr(s); 6537 r = s->rule.ptr; 6538 a = s->anchor.ptr; 6539 log = s->log; 6540 } else if (s == NULL) 6541 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6542 &a, &ruleset, inp); 6543 break; 6544 } 6545 6546 default: 6547 action = pf_test_state_other(&s, dir, kif, m, &pd); 6548 if (action == PF_PASS) { 6549 if (V_pfsync_update_state_ptr != NULL) 6550 V_pfsync_update_state_ptr(s); 6551 r = s->rule.ptr; 6552 a = s->anchor.ptr; 6553 log = s->log; 6554 } else if (s == NULL) 6555 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6556 &a, &ruleset, inp); 6557 break; 6558 } 6559 6560 done: 6561 PF_RULES_RUNLOCK(); 6562 if (n != m) { 6563 m_freem(n); 6564 n = NULL; 6565 } 6566 6567 /* handle dangerous IPv6 extension headers. */ 6568 if (action == PF_PASS && rh_cnt && 6569 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 6570 action = PF_DROP; 6571 REASON_SET(&reason, PFRES_IPOPTIONS); 6572 log = r->log; 6573 DPFPRINTF(PF_DEBUG_MISC, 6574 ("pf: dropping packet with dangerous v6 headers\n")); 6575 } 6576 6577 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 6578 action = PF_DROP; 6579 REASON_SET(&reason, PFRES_MEMORY); 6580 } 6581 if (r->rtableid >= 0) 6582 M_SETFIB(m, r->rtableid); 6583 6584 if (r->scrub_flags & PFSTATE_SETPRIO) { 6585 if (pd.tos & IPTOS_LOWDELAY) 6586 pqid = 1; 6587 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) { 6588 action = PF_DROP; 6589 REASON_SET(&reason, PFRES_MEMORY); 6590 log = 1; 6591 DPFPRINTF(PF_DEBUG_MISC, 6592 ("pf: failed to allocate 802.1q mtag\n")); 6593 } 6594 } 6595 6596 #ifdef ALTQ 6597 if (action == PF_PASS && r->qid) { 6598 if (pd.pf_mtag == NULL && 6599 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6600 action = PF_DROP; 6601 REASON_SET(&reason, PFRES_MEMORY); 6602 } else { 6603 if (s != NULL) 6604 pd.pf_mtag->qid_hash = pf_state_hash(s); 6605 if (pd.tos & IPTOS_LOWDELAY) 6606 pd.pf_mtag->qid = r->pqid; 6607 else 6608 pd.pf_mtag->qid = r->qid; 6609 /* Add hints for ecn. */ 6610 pd.pf_mtag->hdr = h; 6611 } 6612 } 6613 #endif /* ALTQ */ 6614 6615 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 6616 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 6617 (s->nat_rule.ptr->action == PF_RDR || 6618 s->nat_rule.ptr->action == PF_BINAT) && 6619 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 6620 m->m_flags |= M_SKIP_FIREWALL; 6621 6622 /* XXX: Anybody working on it?! */ 6623 if (r->divert.port) 6624 printf("pf: divert(9) is not supported for IPv6\n"); 6625 6626 if (log) { 6627 struct pf_krule *lr; 6628 6629 if (s != NULL && s->nat_rule.ptr != NULL && 6630 s->nat_rule.ptr->log & PF_LOG_ALL) 6631 lr = s->nat_rule.ptr; 6632 else 6633 lr = r; 6634 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset, 6635 &pd, (s == NULL)); 6636 } 6637 6638 counter_u64_add(kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS], 6639 pd.tot_len); 6640 counter_u64_add(kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS], 6641 1); 6642 6643 if (action == PF_PASS || r->action == PF_DROP) { 6644 dirndx = (dir == PF_OUT); 6645 counter_u64_add(r->packets[dirndx], 1); 6646 counter_u64_add(r->bytes[dirndx], pd.tot_len); 6647 if (a != NULL) { 6648 counter_u64_add(a->packets[dirndx], 1); 6649 counter_u64_add(a->bytes[dirndx], pd.tot_len); 6650 } 6651 if (s != NULL) { 6652 if (s->nat_rule.ptr != NULL) { 6653 counter_u64_add(s->nat_rule.ptr->packets[dirndx], 6654 1); 6655 counter_u64_add(s->nat_rule.ptr->bytes[dirndx], 6656 pd.tot_len); 6657 } 6658 if (s->src_node != NULL) { 6659 counter_u64_add(s->src_node->packets[dirndx], 6660 1); 6661 counter_u64_add(s->src_node->bytes[dirndx], 6662 pd.tot_len); 6663 } 6664 if (s->nat_src_node != NULL) { 6665 counter_u64_add(s->nat_src_node->packets[dirndx], 6666 1); 6667 counter_u64_add(s->nat_src_node->bytes[dirndx], 6668 pd.tot_len); 6669 } 6670 dirndx = (dir == s->direction) ? 0 : 1; 6671 counter_u64_add(s->packets[dirndx], 1); 6672 counter_u64_add(s->bytes[dirndx], pd.tot_len); 6673 } 6674 tr = r; 6675 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 6676 if (nr != NULL && r == &V_pf_default_rule) 6677 tr = nr; 6678 if (tr->src.addr.type == PF_ADDR_TABLE) 6679 pfr_update_stats(tr->src.addr.p.tbl, 6680 (s == NULL) ? pd.src : 6681 &s->key[(s->direction == PF_IN)]->addr[0], 6682 pd.af, pd.tot_len, dir == PF_OUT, 6683 r->action == PF_PASS, tr->src.neg); 6684 if (tr->dst.addr.type == PF_ADDR_TABLE) 6685 pfr_update_stats(tr->dst.addr.p.tbl, 6686 (s == NULL) ? pd.dst : 6687 &s->key[(s->direction == PF_IN)]->addr[1], 6688 pd.af, pd.tot_len, dir == PF_OUT, 6689 r->action == PF_PASS, tr->dst.neg); 6690 } 6691 6692 switch (action) { 6693 case PF_SYNPROXY_DROP: 6694 m_freem(*m0); 6695 case PF_DEFER: 6696 *m0 = NULL; 6697 action = PF_PASS; 6698 break; 6699 case PF_DROP: 6700 m_freem(*m0); 6701 *m0 = NULL; 6702 break; 6703 default: 6704 /* pf_route6() returns unlocked. */ 6705 if (r->rt) { 6706 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp); 6707 return (action); 6708 } 6709 break; 6710 } 6711 6712 if (s) 6713 PF_STATE_UNLOCK(s); 6714 6715 /* If reassembled packet passed, create new fragments. */ 6716 if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) && 6717 (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL) 6718 action = pf_refragment6(ifp, m0, mtag); 6719 6720 return (action); 6721 } 6722 #endif /* INET6 */ 6723