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 #include "opt_bpf.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_pf.h" 45 #include "opt_sctp.h" 46 47 #include <sys/param.h> 48 #include <sys/bus.h> 49 #include <sys/endian.h> 50 #include <sys/gsb_crc32.h> 51 #include <sys/hash.h> 52 #include <sys/interrupt.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/limits.h> 56 #include <sys/mbuf.h> 57 #include <sys/md5.h> 58 #include <sys/random.h> 59 #include <sys/refcount.h> 60 #include <sys/sdt.h> 61 #include <sys/socket.h> 62 #include <sys/sysctl.h> 63 #include <sys/taskqueue.h> 64 #include <sys/ucred.h> 65 66 #include <net/if.h> 67 #include <net/if_var.h> 68 #include <net/if_private.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 /* dummynet */ 97 #include <netinet/ip_dummynet.h> 98 #include <netinet/ip_fw.h> 99 #include <netpfil/ipfw/dn_heap.h> 100 #include <netpfil/ipfw/ip_fw_private.h> 101 #include <netpfil/ipfw/ip_dn_private.h> 102 103 #ifdef INET6 104 #include <netinet/ip6.h> 105 #include <netinet/icmp6.h> 106 #include <netinet6/nd6.h> 107 #include <netinet6/ip6_var.h> 108 #include <netinet6/in6_pcb.h> 109 #include <netinet6/in6_fib.h> 110 #include <netinet6/scope6_var.h> 111 #endif /* INET6 */ 112 113 #include <netinet/sctp_header.h> 114 #include <netinet/sctp_crc32.h> 115 116 #include <machine/in_cksum.h> 117 #include <security/mac/mac_framework.h> 118 119 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 120 121 SDT_PROVIDER_DEFINE(pf); 122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 123 "struct pf_kstate *"); 124 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 125 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 126 "struct pf_kstate *"); 127 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 128 "struct pfi_kkif *"); 129 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 130 "struct pf_krule *", "struct mbuf *", "int"); 131 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 132 "struct pf_sctp_source *"); 133 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 134 "struct pf_kstate *", "struct pf_sctp_source *"); 135 136 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 137 "struct mbuf *"); 138 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 139 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 140 "int", "struct pf_keth_rule *", "char *"); 141 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 142 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 143 "int", "struct pf_keth_rule *"); 144 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 145 146 /* 147 * Global variables 148 */ 149 150 /* state tables */ 151 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 152 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 153 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 154 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 155 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 156 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 157 VNET_DEFINE(struct pf_kstatus, pf_status); 158 159 VNET_DEFINE(u_int32_t, ticket_altqs_active); 160 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 161 VNET_DEFINE(int, altqs_inactive_open); 162 VNET_DEFINE(u_int32_t, ticket_pabuf); 163 164 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 165 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 166 VNET_DEFINE(u_char, pf_tcp_secret[16]); 167 #define V_pf_tcp_secret VNET(pf_tcp_secret) 168 VNET_DEFINE(int, pf_tcp_secret_init); 169 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 170 VNET_DEFINE(int, pf_tcp_iss_off); 171 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 172 VNET_DECLARE(int, pf_vnet_active); 173 #define V_pf_vnet_active VNET(pf_vnet_active) 174 175 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 176 #define V_pf_purge_idx VNET(pf_purge_idx) 177 178 #ifdef PF_WANT_32_TO_64_COUNTER 179 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 180 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 181 182 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 183 VNET_DEFINE(size_t, pf_allrulecount); 184 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 185 #endif 186 187 struct pf_sctp_endpoint; 188 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 189 struct pf_sctp_source { 190 sa_family_t af; 191 struct pf_addr addr; 192 TAILQ_ENTRY(pf_sctp_source) entry; 193 }; 194 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 195 struct pf_sctp_endpoint 196 { 197 uint32_t v_tag; 198 struct pf_sctp_sources sources; 199 RB_ENTRY(pf_sctp_endpoint) entry; 200 }; 201 static int 202 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 203 { 204 return (a->v_tag - b->v_tag); 205 } 206 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 207 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 208 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 209 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 210 static struct mtx_padalign pf_sctp_endpoints_mtx; 211 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 212 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 213 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 214 215 /* 216 * Queue for pf_intr() sends. 217 */ 218 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 219 struct pf_send_entry { 220 STAILQ_ENTRY(pf_send_entry) pfse_next; 221 struct mbuf *pfse_m; 222 enum { 223 PFSE_IP, 224 PFSE_IP6, 225 PFSE_ICMP, 226 PFSE_ICMP6, 227 } pfse_type; 228 struct { 229 int type; 230 int code; 231 int mtu; 232 } icmpopts; 233 }; 234 235 STAILQ_HEAD(pf_send_head, pf_send_entry); 236 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 237 #define V_pf_sendqueue VNET(pf_sendqueue) 238 239 static struct mtx_padalign pf_sendqueue_mtx; 240 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 241 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 242 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 243 244 /* 245 * Queue for pf_overload_task() tasks. 246 */ 247 struct pf_overload_entry { 248 SLIST_ENTRY(pf_overload_entry) next; 249 struct pf_addr addr; 250 sa_family_t af; 251 uint8_t dir; 252 struct pf_krule *rule; 253 }; 254 255 SLIST_HEAD(pf_overload_head, pf_overload_entry); 256 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 257 #define V_pf_overloadqueue VNET(pf_overloadqueue) 258 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 259 #define V_pf_overloadtask VNET(pf_overloadtask) 260 261 static struct mtx_padalign pf_overloadqueue_mtx; 262 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 263 "pf overload/flush queue", MTX_DEF); 264 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 265 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 266 267 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 268 struct mtx_padalign pf_unlnkdrules_mtx; 269 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 270 MTX_DEF); 271 272 struct sx pf_config_lock; 273 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 274 275 struct mtx_padalign pf_table_stats_lock; 276 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 277 MTX_DEF); 278 279 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 280 #define V_pf_sources_z VNET(pf_sources_z) 281 uma_zone_t pf_mtag_z; 282 VNET_DEFINE(uma_zone_t, pf_state_z); 283 VNET_DEFINE(uma_zone_t, pf_state_key_z); 284 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 285 286 VNET_DEFINE(struct unrhdr64, pf_stateid); 287 288 static void pf_src_tree_remove_state(struct pf_kstate *); 289 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 290 u_int32_t); 291 static void pf_add_threshold(struct pf_threshold *); 292 static int pf_check_threshold(struct pf_threshold *); 293 294 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 295 u_int16_t *, u_int16_t *, struct pf_addr *, 296 u_int16_t, u_int8_t, sa_family_t); 297 static int pf_modulate_sack(struct pf_pdesc *, 298 struct tcphdr *, struct pf_state_peer *); 299 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 300 int *, u_int16_t *, u_int16_t *); 301 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 302 struct pf_addr *, struct pf_addr *, u_int16_t, 303 u_int16_t *, u_int16_t *, u_int16_t *, 304 u_int16_t *, u_int8_t, sa_family_t); 305 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 306 sa_family_t, struct pf_krule *, int); 307 static void pf_detach_state(struct pf_kstate *); 308 static int pf_state_key_attach(struct pf_state_key *, 309 struct pf_state_key *, struct pf_kstate *); 310 static void pf_state_key_detach(struct pf_kstate *, int); 311 static int pf_state_key_ctor(void *, int, void *, int); 312 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 313 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 314 struct pf_mtag *pf_mtag); 315 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 316 struct pf_krule *, struct mbuf **); 317 static int pf_dummynet_route(struct pf_pdesc *, 318 struct pf_kstate *, struct pf_krule *, 319 struct ifnet *, struct sockaddr *, struct mbuf **); 320 static int pf_test_eth_rule(int, struct pfi_kkif *, 321 struct mbuf **); 322 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 323 struct pf_pdesc *, struct pf_krule **, 324 struct pf_kruleset **, struct inpcb *); 325 static int pf_create_state(struct pf_krule *, struct pf_krule *, 326 struct pf_krule *, struct pf_pdesc *, 327 struct pf_ksrc_node *, struct pf_state_key *, 328 struct pf_state_key *, 329 u_int16_t, u_int16_t, int *, 330 struct pf_kstate **, int, u_int16_t, u_int16_t, 331 struct pf_krule_slist *, struct pf_udp_mapping *); 332 static int pf_state_key_addr_setup(struct pf_pdesc *, 333 struct pf_state_key_cmp *, int); 334 static int pf_tcp_track_full(struct pf_kstate **, 335 struct pf_pdesc *, u_short *, int *); 336 static int pf_tcp_track_sloppy(struct pf_kstate **, 337 struct pf_pdesc *, u_short *); 338 static int pf_test_state_tcp(struct pf_kstate **, 339 struct pf_pdesc *, u_short *); 340 static int pf_test_state_udp(struct pf_kstate **, 341 struct pf_pdesc *); 342 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 343 struct pf_pdesc *, struct pf_kstate **, 344 int, u_int16_t, u_int16_t, 345 int, int *, int, int); 346 static int pf_test_state_icmp(struct pf_kstate **, 347 struct pf_pdesc *, u_short *); 348 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 349 static void pf_sctp_multihome_delayed(struct pf_pdesc *, 350 struct pfi_kkif *, struct pf_kstate *, int); 351 static int pf_test_state_sctp(struct pf_kstate **, 352 struct pf_pdesc *, u_short *); 353 static int pf_test_state_other(struct pf_kstate **, 354 struct pf_pdesc *); 355 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 356 int, u_int16_t); 357 static int pf_check_proto_cksum(struct mbuf *, int, int, 358 u_int8_t, sa_family_t); 359 static int pf_walk_option6(struct mbuf *, int, int, uint32_t *, 360 u_short *); 361 static void pf_print_state_parts(struct pf_kstate *, 362 struct pf_state_key *, struct pf_state_key *); 363 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 364 bool, u_int8_t); 365 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 366 const struct pf_state_key_cmp *, u_int); 367 static int pf_src_connlimit(struct pf_kstate **); 368 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 369 static void pf_counters_inc(int, struct pf_pdesc *, 370 struct pf_kstate *, struct pf_krule *, 371 struct pf_krule *); 372 static void pf_overload_task(void *v, int pending); 373 static u_short pf_insert_src_node(struct pf_ksrc_node **, 374 struct pf_krule *, struct pf_addr *, sa_family_t); 375 static u_int pf_purge_expired_states(u_int, int); 376 static void pf_purge_unlinked_rules(void); 377 static int pf_mtag_uminit(void *, int, int); 378 static void pf_mtag_free(struct m_tag *); 379 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 380 int, struct pf_state_key *); 381 #ifdef INET 382 static void pf_route(struct mbuf **, struct pf_krule *, 383 struct ifnet *, struct pf_kstate *, 384 struct pf_pdesc *, struct inpcb *); 385 #endif /* INET */ 386 #ifdef INET6 387 static void pf_change_a6(struct pf_addr *, u_int16_t *, 388 struct pf_addr *, u_int8_t); 389 static void pf_route6(struct mbuf **, struct pf_krule *, 390 struct ifnet *, struct pf_kstate *, 391 struct pf_pdesc *, struct inpcb *); 392 #endif /* INET6 */ 393 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 394 395 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 396 397 extern int pf_end_threads; 398 extern struct proc *pf_purge_proc; 399 400 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 401 402 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK }; 403 404 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 405 do { \ 406 struct pf_state_key *nk; \ 407 if ((pd->dir) == PF_OUT) \ 408 nk = (_s)->key[PF_SK_STACK]; \ 409 else \ 410 nk = (_s)->key[PF_SK_WIRE]; \ 411 pf_packet_rework_nat(_m, _pd, _off, nk); \ 412 } while (0) 413 414 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 415 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 416 417 #define STATE_LOOKUP(k, s, pd) \ 418 do { \ 419 (s) = pf_find_state((pd->kif), (k), (pd->dir)); \ 420 SDT_PROBE5(pf, ip, state, lookup, pd->kif, k, (pd->dir), pd, (s)); \ 421 if ((s) == NULL) \ 422 return (PF_DROP); \ 423 if (PACKET_LOOPED(pd)) \ 424 return (PF_PASS); \ 425 } while (0) 426 427 static struct pfi_kkif * 428 BOUND_IFACE(struct pf_kstate *st, struct pfi_kkif *k) 429 { 430 SDT_PROBE2(pf, ip, , bound_iface, st, k); 431 432 /* Floating unless otherwise specified. */ 433 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 434 return (V_pfi_all); 435 436 /* 437 * Initially set to all, because we don't know what interface we'll be 438 * sending this out when we create the state. 439 */ 440 if (st->rule->rt == PF_REPLYTO) 441 return (V_pfi_all); 442 443 /* Don't overrule the interface for states created on incoming packets. */ 444 if (st->direction == PF_IN) 445 return (k); 446 447 /* No route-to, so don't overrule. */ 448 if (st->rt != PF_ROUTETO) 449 return (k); 450 451 /* Bind to the route-to interface. */ 452 return (st->rt_kif); 453 } 454 455 #define STATE_INC_COUNTERS(s) \ 456 do { \ 457 struct pf_krule_item *mrm; \ 458 counter_u64_add(s->rule->states_cur, 1); \ 459 counter_u64_add(s->rule->states_tot, 1); \ 460 if (s->anchor != NULL) { \ 461 counter_u64_add(s->anchor->states_cur, 1); \ 462 counter_u64_add(s->anchor->states_tot, 1); \ 463 } \ 464 if (s->nat_rule != NULL) { \ 465 counter_u64_add(s->nat_rule->states_cur, 1);\ 466 counter_u64_add(s->nat_rule->states_tot, 1);\ 467 } \ 468 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 469 counter_u64_add(mrm->r->states_cur, 1); \ 470 counter_u64_add(mrm->r->states_tot, 1); \ 471 } \ 472 } while (0) 473 474 #define STATE_DEC_COUNTERS(s) \ 475 do { \ 476 struct pf_krule_item *mrm; \ 477 if (s->nat_rule != NULL) \ 478 counter_u64_add(s->nat_rule->states_cur, -1);\ 479 if (s->anchor != NULL) \ 480 counter_u64_add(s->anchor->states_cur, -1); \ 481 counter_u64_add(s->rule->states_cur, -1); \ 482 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 483 counter_u64_add(mrm->r->states_cur, -1); \ 484 } while (0) 485 486 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 487 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 488 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 489 VNET_DEFINE(struct pf_idhash *, pf_idhash); 490 VNET_DEFINE(struct pf_srchash *, pf_srchash); 491 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 492 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 493 494 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 495 "pf(4)"); 496 497 VNET_DEFINE(u_long, pf_hashmask); 498 VNET_DEFINE(u_long, pf_srchashmask); 499 VNET_DEFINE(u_long, pf_udpendpointhashmask); 500 VNET_DEFINE_STATIC(u_long, pf_hashsize); 501 #define V_pf_hashsize VNET(pf_hashsize) 502 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 503 #define V_pf_srchashsize VNET(pf_srchashsize) 504 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 505 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 506 u_long pf_ioctl_maxcount = 65535; 507 508 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 509 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 510 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 511 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 512 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 513 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 514 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 515 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 516 517 VNET_DEFINE(void *, pf_swi_cookie); 518 VNET_DEFINE(struct intr_event *, pf_swi_ie); 519 520 VNET_DEFINE(uint32_t, pf_hashseed); 521 #define V_pf_hashseed VNET(pf_hashseed) 522 523 static void 524 pf_sctp_checksum(struct mbuf *m, int off) 525 { 526 uint32_t sum = 0; 527 528 /* Zero out the checksum, to enable recalculation. */ 529 m_copyback(m, off + offsetof(struct sctphdr, checksum), 530 sizeof(sum), (caddr_t)&sum); 531 532 sum = sctp_calculate_cksum(m, off); 533 534 m_copyback(m, off + offsetof(struct sctphdr, checksum), 535 sizeof(sum), (caddr_t)&sum); 536 } 537 538 int 539 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 540 { 541 542 switch (af) { 543 #ifdef INET 544 case AF_INET: 545 if (a->addr32[0] > b->addr32[0]) 546 return (1); 547 if (a->addr32[0] < b->addr32[0]) 548 return (-1); 549 break; 550 #endif /* INET */ 551 #ifdef INET6 552 case AF_INET6: 553 if (a->addr32[3] > b->addr32[3]) 554 return (1); 555 if (a->addr32[3] < b->addr32[3]) 556 return (-1); 557 if (a->addr32[2] > b->addr32[2]) 558 return (1); 559 if (a->addr32[2] < b->addr32[2]) 560 return (-1); 561 if (a->addr32[1] > b->addr32[1]) 562 return (1); 563 if (a->addr32[1] < b->addr32[1]) 564 return (-1); 565 if (a->addr32[0] > b->addr32[0]) 566 return (1); 567 if (a->addr32[0] < b->addr32[0]) 568 return (-1); 569 break; 570 #endif /* INET6 */ 571 } 572 return (0); 573 } 574 575 static bool 576 pf_is_loopback(sa_family_t af, struct pf_addr *addr) 577 { 578 switch (af) { 579 #ifdef INET 580 case AF_INET: 581 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 582 #endif 583 case AF_INET6: 584 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 585 default: 586 panic("Unknown af %d", af); 587 } 588 } 589 590 static void 591 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 592 struct pf_state_key *nk) 593 { 594 595 switch (pd->proto) { 596 case IPPROTO_TCP: { 597 struct tcphdr *th = &pd->hdr.tcp; 598 599 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 600 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 601 &th->th_sum, &nk->addr[pd->sidx], 602 nk->port[pd->sidx], 0, pd->af); 603 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 604 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 605 &th->th_sum, &nk->addr[pd->didx], 606 nk->port[pd->didx], 0, pd->af); 607 m_copyback(m, off, sizeof(*th), (caddr_t)th); 608 break; 609 } 610 case IPPROTO_UDP: { 611 struct udphdr *uh = &pd->hdr.udp; 612 613 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 614 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 615 &uh->uh_sum, &nk->addr[pd->sidx], 616 nk->port[pd->sidx], 1, pd->af); 617 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 618 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 619 &uh->uh_sum, &nk->addr[pd->didx], 620 nk->port[pd->didx], 1, pd->af); 621 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 622 break; 623 } 624 case IPPROTO_SCTP: { 625 struct sctphdr *sh = &pd->hdr.sctp; 626 uint16_t checksum = 0; 627 628 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 629 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 630 &checksum, &nk->addr[pd->sidx], 631 nk->port[pd->sidx], 1, pd->af); 632 } 633 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 634 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 635 &checksum, &nk->addr[pd->didx], 636 nk->port[pd->didx], 1, pd->af); 637 } 638 639 break; 640 } 641 case IPPROTO_ICMP: { 642 struct icmp *ih = &pd->hdr.icmp; 643 644 if (nk->port[pd->sidx] != ih->icmp_id) { 645 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 646 ih->icmp_cksum, ih->icmp_id, 647 nk->port[pd->sidx], 0); 648 ih->icmp_id = nk->port[pd->sidx]; 649 pd->sport = &ih->icmp_id; 650 651 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 652 } 653 /* FALLTHROUGH */ 654 } 655 default: 656 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 657 switch (pd->af) { 658 case AF_INET: 659 pf_change_a(&pd->src->v4.s_addr, 660 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 661 0); 662 break; 663 case AF_INET6: 664 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 665 break; 666 } 667 } 668 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 669 switch (pd->af) { 670 case AF_INET: 671 pf_change_a(&pd->dst->v4.s_addr, 672 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 673 0); 674 break; 675 case AF_INET6: 676 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 677 break; 678 } 679 } 680 break; 681 } 682 } 683 684 static __inline uint32_t 685 pf_hashkey(const struct pf_state_key *sk) 686 { 687 uint32_t h; 688 689 h = murmur3_32_hash32((const uint32_t *)sk, 690 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 691 V_pf_hashseed); 692 693 return (h & V_pf_hashmask); 694 } 695 696 static __inline uint32_t 697 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 698 { 699 uint32_t h; 700 701 switch (af) { 702 case AF_INET: 703 h = murmur3_32_hash32((uint32_t *)&addr->v4, 704 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 705 break; 706 case AF_INET6: 707 h = murmur3_32_hash32((uint32_t *)&addr->v6, 708 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 709 break; 710 } 711 712 return (h & V_pf_srchashmask); 713 } 714 715 static inline uint32_t 716 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 717 { 718 uint32_t h; 719 720 h = murmur3_32_hash32((uint32_t *)endpoint, 721 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 722 V_pf_hashseed); 723 return (h & V_pf_udpendpointhashmask); 724 } 725 726 #ifdef ALTQ 727 static int 728 pf_state_hash(struct pf_kstate *s) 729 { 730 u_int32_t hv = (intptr_t)s / sizeof(*s); 731 732 hv ^= crc32(&s->src, sizeof(s->src)); 733 hv ^= crc32(&s->dst, sizeof(s->dst)); 734 if (hv == 0) 735 hv = 1; 736 return (hv); 737 } 738 #endif 739 740 static __inline void 741 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 742 { 743 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 744 s->dst.state = newstate; 745 if (which == PF_PEER_DST) 746 return; 747 if (s->src.state == newstate) 748 return; 749 if (s->creatorid == V_pf_status.hostid && 750 s->key[PF_SK_STACK] != NULL && 751 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 752 !(TCPS_HAVEESTABLISHED(s->src.state) || 753 s->src.state == TCPS_CLOSED) && 754 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 755 atomic_add_32(&V_pf_status.states_halfopen, -1); 756 757 s->src.state = newstate; 758 } 759 760 #ifdef INET6 761 void 762 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 763 { 764 switch (af) { 765 #ifdef INET 766 case AF_INET: 767 memcpy(&dst->v4, &src->v4, sizeof(dst->v4)); 768 break; 769 #endif /* INET */ 770 case AF_INET6: 771 memcpy(&dst->v6, &src->v6, sizeof(dst->v6)); 772 break; 773 } 774 } 775 #endif /* INET6 */ 776 777 static void 778 pf_init_threshold(struct pf_threshold *threshold, 779 u_int32_t limit, u_int32_t seconds) 780 { 781 threshold->limit = limit * PF_THRESHOLD_MULT; 782 threshold->seconds = seconds; 783 threshold->count = 0; 784 threshold->last = time_uptime; 785 } 786 787 static void 788 pf_add_threshold(struct pf_threshold *threshold) 789 { 790 u_int32_t t = time_uptime, diff = t - threshold->last; 791 792 if (diff >= threshold->seconds) 793 threshold->count = 0; 794 else 795 threshold->count -= threshold->count * diff / 796 threshold->seconds; 797 threshold->count += PF_THRESHOLD_MULT; 798 threshold->last = t; 799 } 800 801 static int 802 pf_check_threshold(struct pf_threshold *threshold) 803 { 804 return (threshold->count > threshold->limit); 805 } 806 807 static int 808 pf_src_connlimit(struct pf_kstate **state) 809 { 810 struct pf_overload_entry *pfoe; 811 int bad = 0; 812 813 PF_STATE_LOCK_ASSERT(*state); 814 /* 815 * XXXKS: The src node is accessed unlocked! 816 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 817 */ 818 819 (*state)->src_node->conn++; 820 (*state)->src.tcp_est = 1; 821 pf_add_threshold(&(*state)->src_node->conn_rate); 822 823 if ((*state)->rule->max_src_conn && 824 (*state)->rule->max_src_conn < 825 (*state)->src_node->conn) { 826 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 827 bad++; 828 } 829 830 if ((*state)->rule->max_src_conn_rate.limit && 831 pf_check_threshold(&(*state)->src_node->conn_rate)) { 832 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 833 bad++; 834 } 835 836 if (!bad) 837 return (0); 838 839 /* Kill this state. */ 840 (*state)->timeout = PFTM_PURGE; 841 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 842 843 if ((*state)->rule->overload_tbl == NULL) 844 return (1); 845 846 /* Schedule overloading and flushing task. */ 847 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 848 if (pfoe == NULL) 849 return (1); /* too bad :( */ 850 851 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 852 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 853 pfoe->rule = (*state)->rule; 854 pfoe->dir = (*state)->direction; 855 PF_OVERLOADQ_LOCK(); 856 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 857 PF_OVERLOADQ_UNLOCK(); 858 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 859 860 return (1); 861 } 862 863 static void 864 pf_overload_task(void *v, int pending) 865 { 866 struct pf_overload_head queue; 867 struct pfr_addr p; 868 struct pf_overload_entry *pfoe, *pfoe1; 869 uint32_t killed = 0; 870 871 CURVNET_SET((struct vnet *)v); 872 873 PF_OVERLOADQ_LOCK(); 874 queue = V_pf_overloadqueue; 875 SLIST_INIT(&V_pf_overloadqueue); 876 PF_OVERLOADQ_UNLOCK(); 877 878 bzero(&p, sizeof(p)); 879 SLIST_FOREACH(pfoe, &queue, next) { 880 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 881 if (V_pf_status.debug >= PF_DEBUG_MISC) { 882 printf("%s: blocking address ", __func__); 883 pf_print_host(&pfoe->addr, 0, pfoe->af); 884 printf("\n"); 885 } 886 887 p.pfra_af = pfoe->af; 888 switch (pfoe->af) { 889 #ifdef INET 890 case AF_INET: 891 p.pfra_net = 32; 892 p.pfra_ip4addr = pfoe->addr.v4; 893 break; 894 #endif 895 #ifdef INET6 896 case AF_INET6: 897 p.pfra_net = 128; 898 p.pfra_ip6addr = pfoe->addr.v6; 899 break; 900 #endif 901 } 902 903 PF_RULES_WLOCK(); 904 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 905 PF_RULES_WUNLOCK(); 906 } 907 908 /* 909 * Remove those entries, that don't need flushing. 910 */ 911 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 912 if (pfoe->rule->flush == 0) { 913 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 914 free(pfoe, M_PFTEMP); 915 } else 916 counter_u64_add( 917 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 918 919 /* If nothing to flush, return. */ 920 if (SLIST_EMPTY(&queue)) { 921 CURVNET_RESTORE(); 922 return; 923 } 924 925 for (int i = 0; i <= V_pf_hashmask; i++) { 926 struct pf_idhash *ih = &V_pf_idhash[i]; 927 struct pf_state_key *sk; 928 struct pf_kstate *s; 929 930 PF_HASHROW_LOCK(ih); 931 LIST_FOREACH(s, &ih->states, entry) { 932 sk = s->key[PF_SK_WIRE]; 933 SLIST_FOREACH(pfoe, &queue, next) 934 if (sk->af == pfoe->af && 935 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 936 pfoe->rule == s->rule) && 937 ((pfoe->dir == PF_OUT && 938 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 939 (pfoe->dir == PF_IN && 940 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 941 s->timeout = PFTM_PURGE; 942 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 943 killed++; 944 } 945 } 946 PF_HASHROW_UNLOCK(ih); 947 } 948 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 949 free(pfoe, M_PFTEMP); 950 if (V_pf_status.debug >= PF_DEBUG_MISC) 951 printf("%s: %u states killed", __func__, killed); 952 953 CURVNET_RESTORE(); 954 } 955 956 /* 957 * Can return locked on failure, so that we can consistently 958 * allocate and insert a new one. 959 */ 960 struct pf_ksrc_node * 961 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 962 struct pf_srchash **sh, bool returnlocked) 963 { 964 struct pf_ksrc_node *n; 965 966 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 967 968 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 969 PF_HASHROW_LOCK(*sh); 970 LIST_FOREACH(n, &(*sh)->nodes, entry) 971 if (n->rule == rule && n->af == af && 972 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 973 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 974 break; 975 976 if (n != NULL) { 977 n->states++; 978 PF_HASHROW_UNLOCK(*sh); 979 } else if (returnlocked == false) 980 PF_HASHROW_UNLOCK(*sh); 981 982 return (n); 983 } 984 985 static void 986 pf_free_src_node(struct pf_ksrc_node *sn) 987 { 988 989 for (int i = 0; i < 2; i++) { 990 counter_u64_free(sn->bytes[i]); 991 counter_u64_free(sn->packets[i]); 992 } 993 uma_zfree(V_pf_sources_z, sn); 994 } 995 996 static u_short 997 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 998 struct pf_addr *src, sa_family_t af) 999 { 1000 u_short reason = 0; 1001 struct pf_srchash *sh = NULL; 1002 1003 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 1004 rule->rpool.opts & PF_POOL_STICKYADDR), 1005 ("%s for non-tracking rule %p", __func__, rule)); 1006 1007 if (*sn == NULL) 1008 *sn = pf_find_src_node(src, rule, af, &sh, true); 1009 1010 if (*sn == NULL) { 1011 PF_HASHROW_ASSERT(sh); 1012 1013 if (rule->max_src_nodes && 1014 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 1015 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1016 PF_HASHROW_UNLOCK(sh); 1017 reason = PFRES_SRCLIMIT; 1018 goto done; 1019 } 1020 1021 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1022 if ((*sn) == NULL) { 1023 PF_HASHROW_UNLOCK(sh); 1024 reason = PFRES_MEMORY; 1025 goto done; 1026 } 1027 1028 for (int i = 0; i < 2; i++) { 1029 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1030 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1031 1032 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1033 pf_free_src_node(*sn); 1034 PF_HASHROW_UNLOCK(sh); 1035 reason = PFRES_MEMORY; 1036 goto done; 1037 } 1038 } 1039 1040 pf_init_threshold(&(*sn)->conn_rate, 1041 rule->max_src_conn_rate.limit, 1042 rule->max_src_conn_rate.seconds); 1043 1044 MPASS((*sn)->lock == NULL); 1045 (*sn)->lock = &sh->lock; 1046 1047 (*sn)->af = af; 1048 (*sn)->rule = rule; 1049 PF_ACPY(&(*sn)->addr, src, af); 1050 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 1051 (*sn)->creation = time_uptime; 1052 (*sn)->ruletype = rule->action; 1053 (*sn)->states = 1; 1054 if ((*sn)->rule != NULL) 1055 counter_u64_add((*sn)->rule->src_nodes, 1); 1056 PF_HASHROW_UNLOCK(sh); 1057 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1058 } else { 1059 if (rule->max_src_states && 1060 (*sn)->states >= rule->max_src_states) { 1061 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1062 1); 1063 reason = PFRES_SRCLIMIT; 1064 goto done; 1065 } 1066 } 1067 done: 1068 return (reason); 1069 } 1070 1071 void 1072 pf_unlink_src_node(struct pf_ksrc_node *src) 1073 { 1074 PF_SRC_NODE_LOCK_ASSERT(src); 1075 1076 LIST_REMOVE(src, entry); 1077 if (src->rule) 1078 counter_u64_add(src->rule->src_nodes, -1); 1079 } 1080 1081 u_int 1082 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1083 { 1084 struct pf_ksrc_node *sn, *tmp; 1085 u_int count = 0; 1086 1087 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1088 pf_free_src_node(sn); 1089 count++; 1090 } 1091 1092 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1093 1094 return (count); 1095 } 1096 1097 void 1098 pf_mtag_initialize(void) 1099 { 1100 1101 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1102 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1103 UMA_ALIGN_PTR, 0); 1104 } 1105 1106 /* Per-vnet data storage structures initialization. */ 1107 void 1108 pf_initialize(void) 1109 { 1110 struct pf_keyhash *kh; 1111 struct pf_idhash *ih; 1112 struct pf_srchash *sh; 1113 struct pf_udpendpointhash *uh; 1114 u_int i; 1115 1116 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1117 V_pf_hashsize = PF_HASHSIZ; 1118 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1119 V_pf_srchashsize = PF_SRCHASHSIZ; 1120 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1121 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1122 1123 V_pf_hashseed = arc4random(); 1124 1125 /* States and state keys storage. */ 1126 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1127 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1128 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1129 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1130 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1131 1132 V_pf_state_key_z = uma_zcreate("pf state keys", 1133 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1134 UMA_ALIGN_PTR, 0); 1135 1136 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1137 M_PFHASH, M_NOWAIT | M_ZERO); 1138 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1139 M_PFHASH, M_NOWAIT | M_ZERO); 1140 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1141 printf("pf: Unable to allocate memory for " 1142 "state_hashsize %lu.\n", V_pf_hashsize); 1143 1144 free(V_pf_keyhash, M_PFHASH); 1145 free(V_pf_idhash, M_PFHASH); 1146 1147 V_pf_hashsize = PF_HASHSIZ; 1148 V_pf_keyhash = mallocarray(V_pf_hashsize, 1149 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1150 V_pf_idhash = mallocarray(V_pf_hashsize, 1151 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1152 } 1153 1154 V_pf_hashmask = V_pf_hashsize - 1; 1155 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1156 i++, kh++, ih++) { 1157 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1158 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1159 } 1160 1161 /* Source nodes. */ 1162 V_pf_sources_z = uma_zcreate("pf source nodes", 1163 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1164 0); 1165 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1166 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1167 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1168 1169 V_pf_srchash = mallocarray(V_pf_srchashsize, 1170 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1171 if (V_pf_srchash == NULL) { 1172 printf("pf: Unable to allocate memory for " 1173 "source_hashsize %lu.\n", V_pf_srchashsize); 1174 1175 V_pf_srchashsize = PF_SRCHASHSIZ; 1176 V_pf_srchash = mallocarray(V_pf_srchashsize, 1177 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1178 } 1179 1180 V_pf_srchashmask = V_pf_srchashsize - 1; 1181 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1182 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1183 1184 1185 /* UDP endpoint mappings. */ 1186 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1187 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1188 UMA_ALIGN_PTR, 0); 1189 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1190 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1191 if (V_pf_udpendpointhash == NULL) { 1192 printf("pf: Unable to allocate memory for " 1193 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1194 1195 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1196 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1197 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1198 } 1199 1200 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1201 for (i = 0, uh = V_pf_udpendpointhash; 1202 i <= V_pf_udpendpointhashmask; 1203 i++, uh++) { 1204 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1205 MTX_DEF | MTX_DUPOK); 1206 } 1207 1208 /* ALTQ */ 1209 TAILQ_INIT(&V_pf_altqs[0]); 1210 TAILQ_INIT(&V_pf_altqs[1]); 1211 TAILQ_INIT(&V_pf_altqs[2]); 1212 TAILQ_INIT(&V_pf_altqs[3]); 1213 TAILQ_INIT(&V_pf_pabuf); 1214 V_pf_altqs_active = &V_pf_altqs[0]; 1215 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1216 V_pf_altqs_inactive = &V_pf_altqs[2]; 1217 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1218 1219 /* Send & overload+flush queues. */ 1220 STAILQ_INIT(&V_pf_sendqueue); 1221 SLIST_INIT(&V_pf_overloadqueue); 1222 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1223 1224 /* Unlinked, but may be referenced rules. */ 1225 TAILQ_INIT(&V_pf_unlinked_rules); 1226 } 1227 1228 void 1229 pf_mtag_cleanup(void) 1230 { 1231 1232 uma_zdestroy(pf_mtag_z); 1233 } 1234 1235 void 1236 pf_cleanup(void) 1237 { 1238 struct pf_keyhash *kh; 1239 struct pf_idhash *ih; 1240 struct pf_srchash *sh; 1241 struct pf_udpendpointhash *uh; 1242 struct pf_send_entry *pfse, *next; 1243 u_int i; 1244 1245 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1246 i <= V_pf_hashmask; 1247 i++, kh++, ih++) { 1248 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1249 __func__)); 1250 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1251 __func__)); 1252 mtx_destroy(&kh->lock); 1253 mtx_destroy(&ih->lock); 1254 } 1255 free(V_pf_keyhash, M_PFHASH); 1256 free(V_pf_idhash, M_PFHASH); 1257 1258 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1259 KASSERT(LIST_EMPTY(&sh->nodes), 1260 ("%s: source node hash not empty", __func__)); 1261 mtx_destroy(&sh->lock); 1262 } 1263 free(V_pf_srchash, M_PFHASH); 1264 1265 for (i = 0, uh = V_pf_udpendpointhash; 1266 i <= V_pf_udpendpointhashmask; 1267 i++, uh++) { 1268 KASSERT(LIST_EMPTY(&uh->endpoints), 1269 ("%s: udp endpoint hash not empty", __func__)); 1270 mtx_destroy(&uh->lock); 1271 } 1272 free(V_pf_udpendpointhash, M_PFHASH); 1273 1274 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1275 m_freem(pfse->pfse_m); 1276 free(pfse, M_PFTEMP); 1277 } 1278 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1279 1280 uma_zdestroy(V_pf_sources_z); 1281 uma_zdestroy(V_pf_state_z); 1282 uma_zdestroy(V_pf_state_key_z); 1283 uma_zdestroy(V_pf_udp_mapping_z); 1284 } 1285 1286 static int 1287 pf_mtag_uminit(void *mem, int size, int how) 1288 { 1289 struct m_tag *t; 1290 1291 t = (struct m_tag *)mem; 1292 t->m_tag_cookie = MTAG_ABI_COMPAT; 1293 t->m_tag_id = PACKET_TAG_PF; 1294 t->m_tag_len = sizeof(struct pf_mtag); 1295 t->m_tag_free = pf_mtag_free; 1296 1297 return (0); 1298 } 1299 1300 static void 1301 pf_mtag_free(struct m_tag *t) 1302 { 1303 1304 uma_zfree(pf_mtag_z, t); 1305 } 1306 1307 struct pf_mtag * 1308 pf_get_mtag(struct mbuf *m) 1309 { 1310 struct m_tag *mtag; 1311 1312 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1313 return ((struct pf_mtag *)(mtag + 1)); 1314 1315 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1316 if (mtag == NULL) 1317 return (NULL); 1318 bzero(mtag + 1, sizeof(struct pf_mtag)); 1319 m_tag_prepend(m, mtag); 1320 1321 return ((struct pf_mtag *)(mtag + 1)); 1322 } 1323 1324 static int 1325 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1326 struct pf_kstate *s) 1327 { 1328 struct pf_keyhash *khs, *khw, *kh; 1329 struct pf_state_key *sk, *cur; 1330 struct pf_kstate *si, *olds = NULL; 1331 int idx; 1332 1333 NET_EPOCH_ASSERT(); 1334 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1335 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1336 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1337 1338 /* 1339 * We need to lock hash slots of both keys. To avoid deadlock 1340 * we always lock the slot with lower address first. Unlock order 1341 * isn't important. 1342 * 1343 * We also need to lock ID hash slot before dropping key 1344 * locks. On success we return with ID hash slot locked. 1345 */ 1346 1347 if (skw == sks) { 1348 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1349 PF_HASHROW_LOCK(khs); 1350 } else { 1351 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1352 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1353 if (khs == khw) { 1354 PF_HASHROW_LOCK(khs); 1355 } else if (khs < khw) { 1356 PF_HASHROW_LOCK(khs); 1357 PF_HASHROW_LOCK(khw); 1358 } else { 1359 PF_HASHROW_LOCK(khw); 1360 PF_HASHROW_LOCK(khs); 1361 } 1362 } 1363 1364 #define KEYS_UNLOCK() do { \ 1365 if (khs != khw) { \ 1366 PF_HASHROW_UNLOCK(khs); \ 1367 PF_HASHROW_UNLOCK(khw); \ 1368 } else \ 1369 PF_HASHROW_UNLOCK(khs); \ 1370 } while (0) 1371 1372 /* 1373 * First run: start with wire key. 1374 */ 1375 sk = skw; 1376 kh = khw; 1377 idx = PF_SK_WIRE; 1378 1379 MPASS(s->lock == NULL); 1380 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1381 1382 keyattach: 1383 LIST_FOREACH(cur, &kh->keys, entry) 1384 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1385 break; 1386 1387 if (cur != NULL) { 1388 /* Key exists. Check for same kif, if none, add to key. */ 1389 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1390 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1391 1392 PF_HASHROW_LOCK(ih); 1393 if (si->kif == s->kif && 1394 si->direction == s->direction) { 1395 if (sk->proto == IPPROTO_TCP && 1396 si->src.state >= TCPS_FIN_WAIT_2 && 1397 si->dst.state >= TCPS_FIN_WAIT_2) { 1398 /* 1399 * New state matches an old >FIN_WAIT_2 1400 * state. We can't drop key hash locks, 1401 * thus we can't unlink it properly. 1402 * 1403 * As a workaround we drop it into 1404 * TCPS_CLOSED state, schedule purge 1405 * ASAP and push it into the very end 1406 * of the slot TAILQ, so that it won't 1407 * conflict with our new state. 1408 */ 1409 pf_set_protostate(si, PF_PEER_BOTH, 1410 TCPS_CLOSED); 1411 si->timeout = PFTM_PURGE; 1412 olds = si; 1413 } else { 1414 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1415 printf("pf: %s key attach " 1416 "failed on %s: ", 1417 (idx == PF_SK_WIRE) ? 1418 "wire" : "stack", 1419 s->kif->pfik_name); 1420 pf_print_state_parts(s, 1421 (idx == PF_SK_WIRE) ? 1422 sk : NULL, 1423 (idx == PF_SK_STACK) ? 1424 sk : NULL); 1425 printf(", existing: "); 1426 pf_print_state_parts(si, 1427 (idx == PF_SK_WIRE) ? 1428 sk : NULL, 1429 (idx == PF_SK_STACK) ? 1430 sk : NULL); 1431 printf("\n"); 1432 } 1433 s->timeout = PFTM_UNLINKED; 1434 PF_HASHROW_UNLOCK(ih); 1435 KEYS_UNLOCK(); 1436 uma_zfree(V_pf_state_key_z, sk); 1437 if (idx == PF_SK_STACK) 1438 pf_detach_state(s); 1439 return (EEXIST); /* collision! */ 1440 } 1441 } 1442 PF_HASHROW_UNLOCK(ih); 1443 } 1444 uma_zfree(V_pf_state_key_z, sk); 1445 s->key[idx] = cur; 1446 } else { 1447 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1448 s->key[idx] = sk; 1449 } 1450 1451 stateattach: 1452 /* List is sorted, if-bound states before floating. */ 1453 if (s->kif == V_pfi_all) 1454 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1455 else 1456 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1457 1458 if (olds) { 1459 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1460 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1461 key_list[idx]); 1462 olds = NULL; 1463 } 1464 1465 /* 1466 * Attach done. See how should we (or should not?) 1467 * attach a second key. 1468 */ 1469 if (sks == skw) { 1470 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1471 idx = PF_SK_STACK; 1472 sks = NULL; 1473 goto stateattach; 1474 } else if (sks != NULL) { 1475 /* 1476 * Continue attaching with stack key. 1477 */ 1478 sk = sks; 1479 kh = khs; 1480 idx = PF_SK_STACK; 1481 sks = NULL; 1482 goto keyattach; 1483 } 1484 1485 PF_STATE_LOCK(s); 1486 KEYS_UNLOCK(); 1487 1488 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1489 ("%s failure", __func__)); 1490 1491 return (0); 1492 #undef KEYS_UNLOCK 1493 } 1494 1495 static void 1496 pf_detach_state(struct pf_kstate *s) 1497 { 1498 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1499 struct pf_keyhash *kh; 1500 1501 NET_EPOCH_ASSERT(); 1502 MPASS(s->timeout >= PFTM_MAX); 1503 1504 pf_sctp_multihome_detach_addr(s); 1505 1506 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1507 V_pflow_export_state_ptr(s); 1508 1509 if (sks != NULL) { 1510 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1511 PF_HASHROW_LOCK(kh); 1512 if (s->key[PF_SK_STACK] != NULL) 1513 pf_state_key_detach(s, PF_SK_STACK); 1514 /* 1515 * If both point to same key, then we are done. 1516 */ 1517 if (sks == s->key[PF_SK_WIRE]) { 1518 pf_state_key_detach(s, PF_SK_WIRE); 1519 PF_HASHROW_UNLOCK(kh); 1520 return; 1521 } 1522 PF_HASHROW_UNLOCK(kh); 1523 } 1524 1525 if (s->key[PF_SK_WIRE] != NULL) { 1526 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1527 PF_HASHROW_LOCK(kh); 1528 if (s->key[PF_SK_WIRE] != NULL) 1529 pf_state_key_detach(s, PF_SK_WIRE); 1530 PF_HASHROW_UNLOCK(kh); 1531 } 1532 } 1533 1534 static void 1535 pf_state_key_detach(struct pf_kstate *s, int idx) 1536 { 1537 struct pf_state_key *sk = s->key[idx]; 1538 #ifdef INVARIANTS 1539 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1540 1541 PF_HASHROW_ASSERT(kh); 1542 #endif 1543 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1544 s->key[idx] = NULL; 1545 1546 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1547 LIST_REMOVE(sk, entry); 1548 uma_zfree(V_pf_state_key_z, sk); 1549 } 1550 } 1551 1552 static int 1553 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1554 { 1555 struct pf_state_key *sk = mem; 1556 1557 bzero(sk, sizeof(struct pf_state_key_cmp)); 1558 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1559 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1560 1561 return (0); 1562 } 1563 1564 static int 1565 pf_state_key_addr_setup(struct pf_pdesc *pd, 1566 struct pf_state_key_cmp *key, int multi) 1567 { 1568 struct pf_addr *saddr = pd->src; 1569 struct pf_addr *daddr = pd->dst; 1570 #ifdef INET6 1571 struct nd_neighbor_solicit nd; 1572 struct pf_addr *target; 1573 u_short action, reason; 1574 1575 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1576 goto copy; 1577 1578 switch (pd->hdr.icmp6.icmp6_type) { 1579 case ND_NEIGHBOR_SOLICIT: 1580 if (multi) 1581 return (-1); 1582 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1583 return (-1); 1584 target = (struct pf_addr *)&nd.nd_ns_target; 1585 daddr = target; 1586 break; 1587 case ND_NEIGHBOR_ADVERT: 1588 if (multi) 1589 return (-1); 1590 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1591 return (-1); 1592 target = (struct pf_addr *)&nd.nd_ns_target; 1593 saddr = target; 1594 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1595 key->addr[pd->didx].addr32[0] = 0; 1596 key->addr[pd->didx].addr32[1] = 0; 1597 key->addr[pd->didx].addr32[2] = 0; 1598 key->addr[pd->didx].addr32[3] = 0; 1599 daddr = NULL; /* overwritten */ 1600 } 1601 break; 1602 default: 1603 if (multi == PF_ICMP_MULTI_LINK) { 1604 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1605 key->addr[pd->sidx].addr32[1] = 0; 1606 key->addr[pd->sidx].addr32[2] = 0; 1607 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1608 saddr = NULL; /* overwritten */ 1609 } 1610 } 1611 copy: 1612 #endif 1613 if (saddr) 1614 PF_ACPY(&key->addr[pd->sidx], saddr, pd->af); 1615 if (daddr) 1616 PF_ACPY(&key->addr[pd->didx], daddr, pd->af); 1617 1618 return (0); 1619 } 1620 1621 struct pf_state_key * 1622 pf_state_key_setup(struct pf_pdesc *pd, 1623 struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport, 1624 u_int16_t dport) 1625 { 1626 struct pf_state_key *sk; 1627 1628 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1629 if (sk == NULL) 1630 return (NULL); 1631 1632 if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)sk, 1633 0)) { 1634 uma_zfree(V_pf_state_key_z, sk); 1635 return (NULL); 1636 } 1637 1638 sk->port[pd->sidx] = sport; 1639 sk->port[pd->didx] = dport; 1640 sk->proto = pd->proto; 1641 sk->af = pd->af; 1642 1643 return (sk); 1644 } 1645 1646 struct pf_state_key * 1647 pf_state_key_clone(const struct pf_state_key *orig) 1648 { 1649 struct pf_state_key *sk; 1650 1651 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1652 if (sk == NULL) 1653 return (NULL); 1654 1655 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1656 1657 return (sk); 1658 } 1659 1660 int 1661 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1662 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1663 { 1664 struct pf_idhash *ih; 1665 struct pf_kstate *cur; 1666 int error; 1667 1668 NET_EPOCH_ASSERT(); 1669 1670 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1671 ("%s: sks not pristine", __func__)); 1672 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1673 ("%s: skw not pristine", __func__)); 1674 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1675 1676 s->kif = kif; 1677 s->orig_kif = orig_kif; 1678 1679 if (s->id == 0 && s->creatorid == 0) { 1680 s->id = alloc_unr64(&V_pf_stateid); 1681 s->id = htobe64(s->id); 1682 s->creatorid = V_pf_status.hostid; 1683 } 1684 1685 /* Returns with ID locked on success. */ 1686 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1687 return (error); 1688 1689 ih = &V_pf_idhash[PF_IDHASH(s)]; 1690 PF_HASHROW_ASSERT(ih); 1691 LIST_FOREACH(cur, &ih->states, entry) 1692 if (cur->id == s->id && cur->creatorid == s->creatorid) 1693 break; 1694 1695 if (cur != NULL) { 1696 s->timeout = PFTM_UNLINKED; 1697 PF_HASHROW_UNLOCK(ih); 1698 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1699 printf("pf: state ID collision: " 1700 "id: %016llx creatorid: %08x\n", 1701 (unsigned long long)be64toh(s->id), 1702 ntohl(s->creatorid)); 1703 } 1704 pf_detach_state(s); 1705 return (EEXIST); 1706 } 1707 LIST_INSERT_HEAD(&ih->states, s, entry); 1708 /* One for keys, one for ID hash. */ 1709 refcount_init(&s->refs, 2); 1710 1711 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1712 if (V_pfsync_insert_state_ptr != NULL) 1713 V_pfsync_insert_state_ptr(s); 1714 1715 /* Returns locked. */ 1716 return (0); 1717 } 1718 1719 /* 1720 * Find state by ID: returns with locked row on success. 1721 */ 1722 struct pf_kstate * 1723 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1724 { 1725 struct pf_idhash *ih; 1726 struct pf_kstate *s; 1727 1728 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1729 1730 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1731 1732 PF_HASHROW_LOCK(ih); 1733 LIST_FOREACH(s, &ih->states, entry) 1734 if (s->id == id && s->creatorid == creatorid) 1735 break; 1736 1737 if (s == NULL) 1738 PF_HASHROW_UNLOCK(ih); 1739 1740 return (s); 1741 } 1742 1743 /* 1744 * Find state by key. 1745 * Returns with ID hash slot locked on success. 1746 */ 1747 static struct pf_kstate * 1748 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1749 u_int dir) 1750 { 1751 struct pf_keyhash *kh; 1752 struct pf_state_key *sk; 1753 struct pf_kstate *s; 1754 int idx; 1755 1756 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1757 1758 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1759 1760 PF_HASHROW_LOCK(kh); 1761 LIST_FOREACH(sk, &kh->keys, entry) 1762 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1763 break; 1764 if (sk == NULL) { 1765 PF_HASHROW_UNLOCK(kh); 1766 return (NULL); 1767 } 1768 1769 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1770 1771 /* List is sorted, if-bound states before floating ones. */ 1772 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1773 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1774 PF_STATE_LOCK(s); 1775 PF_HASHROW_UNLOCK(kh); 1776 if (__predict_false(s->timeout >= PFTM_MAX)) { 1777 /* 1778 * State is either being processed by 1779 * pf_unlink_state() in an other thread, or 1780 * is scheduled for immediate expiry. 1781 */ 1782 PF_STATE_UNLOCK(s); 1783 return (NULL); 1784 } 1785 return (s); 1786 } 1787 PF_HASHROW_UNLOCK(kh); 1788 1789 return (NULL); 1790 } 1791 1792 /* 1793 * Returns with ID hash slot locked on success. 1794 */ 1795 struct pf_kstate * 1796 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1797 { 1798 struct pf_keyhash *kh; 1799 struct pf_state_key *sk; 1800 struct pf_kstate *s, *ret = NULL; 1801 int idx, inout = 0; 1802 1803 if (more != NULL) 1804 *more = 0; 1805 1806 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1807 1808 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1809 1810 PF_HASHROW_LOCK(kh); 1811 LIST_FOREACH(sk, &kh->keys, entry) 1812 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1813 break; 1814 if (sk == NULL) { 1815 PF_HASHROW_UNLOCK(kh); 1816 return (NULL); 1817 } 1818 switch (dir) { 1819 case PF_IN: 1820 idx = PF_SK_WIRE; 1821 break; 1822 case PF_OUT: 1823 idx = PF_SK_STACK; 1824 break; 1825 case PF_INOUT: 1826 idx = PF_SK_WIRE; 1827 inout = 1; 1828 break; 1829 default: 1830 panic("%s: dir %u", __func__, dir); 1831 } 1832 second_run: 1833 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1834 if (more == NULL) { 1835 PF_STATE_LOCK(s); 1836 PF_HASHROW_UNLOCK(kh); 1837 return (s); 1838 } 1839 1840 if (ret) 1841 (*more)++; 1842 else { 1843 ret = s; 1844 PF_STATE_LOCK(s); 1845 } 1846 } 1847 if (inout == 1) { 1848 inout = 0; 1849 idx = PF_SK_STACK; 1850 goto second_run; 1851 } 1852 PF_HASHROW_UNLOCK(kh); 1853 1854 return (ret); 1855 } 1856 1857 /* 1858 * FIXME 1859 * This routine is inefficient -- locks the state only to unlock immediately on 1860 * return. 1861 * It is racy -- after the state is unlocked nothing stops other threads from 1862 * removing it. 1863 */ 1864 bool 1865 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 1866 { 1867 struct pf_kstate *s; 1868 1869 s = pf_find_state_all(key, dir, NULL); 1870 if (s != NULL) { 1871 PF_STATE_UNLOCK(s); 1872 return (true); 1873 } 1874 return (false); 1875 } 1876 1877 struct pf_udp_mapping * 1878 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 1879 struct pf_addr *nat_addr, uint16_t nat_port) 1880 { 1881 struct pf_udp_mapping *mapping; 1882 1883 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 1884 if (mapping == NULL) 1885 return (NULL); 1886 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 1887 mapping->endpoints[0].port = src_port; 1888 mapping->endpoints[0].af = af; 1889 mapping->endpoints[0].mapping = mapping; 1890 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 1891 mapping->endpoints[1].port = nat_port; 1892 mapping->endpoints[1].af = af; 1893 mapping->endpoints[1].mapping = mapping; 1894 refcount_init(&mapping->refs, 1); 1895 return (mapping); 1896 } 1897 1898 int 1899 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 1900 { 1901 struct pf_udpendpointhash *h0, *h1; 1902 struct pf_udp_endpoint *endpoint; 1903 int ret = EEXIST; 1904 1905 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1906 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1907 if (h0 == h1) { 1908 PF_HASHROW_LOCK(h0); 1909 } else if (h0 < h1) { 1910 PF_HASHROW_LOCK(h0); 1911 PF_HASHROW_LOCK(h1); 1912 } else { 1913 PF_HASHROW_LOCK(h1); 1914 PF_HASHROW_LOCK(h0); 1915 } 1916 1917 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 1918 if (bcmp(endpoint, &mapping->endpoints[0], 1919 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1920 break; 1921 } 1922 if (endpoint != NULL) 1923 goto cleanup; 1924 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 1925 if (bcmp(endpoint, &mapping->endpoints[1], 1926 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1927 break; 1928 } 1929 if (endpoint != NULL) 1930 goto cleanup; 1931 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 1932 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 1933 ret = 0; 1934 1935 cleanup: 1936 if (h0 != h1) { 1937 PF_HASHROW_UNLOCK(h0); 1938 PF_HASHROW_UNLOCK(h1); 1939 } else { 1940 PF_HASHROW_UNLOCK(h0); 1941 } 1942 return (ret); 1943 } 1944 1945 void 1946 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 1947 { 1948 /* refcount is synchronized on the source endpoint's row lock */ 1949 struct pf_udpendpointhash *h0, *h1; 1950 1951 if (mapping == NULL) 1952 return; 1953 1954 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1955 PF_HASHROW_LOCK(h0); 1956 if (refcount_release(&mapping->refs)) { 1957 LIST_REMOVE(&mapping->endpoints[0], entry); 1958 PF_HASHROW_UNLOCK(h0); 1959 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1960 PF_HASHROW_LOCK(h1); 1961 LIST_REMOVE(&mapping->endpoints[1], entry); 1962 PF_HASHROW_UNLOCK(h1); 1963 1964 uma_zfree(V_pf_udp_mapping_z, mapping); 1965 } else { 1966 PF_HASHROW_UNLOCK(h0); 1967 } 1968 } 1969 1970 1971 struct pf_udp_mapping * 1972 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 1973 { 1974 struct pf_udpendpointhash *uh; 1975 struct pf_udp_endpoint *endpoint; 1976 1977 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 1978 1979 PF_HASHROW_LOCK(uh); 1980 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 1981 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 1982 bcmp(endpoint, &endpoint->mapping->endpoints[0], 1983 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1984 break; 1985 } 1986 if (endpoint == NULL) { 1987 PF_HASHROW_UNLOCK(uh); 1988 return (NULL); 1989 } 1990 refcount_acquire(&endpoint->mapping->refs); 1991 PF_HASHROW_UNLOCK(uh); 1992 return (endpoint->mapping); 1993 } 1994 /* END state table stuff */ 1995 1996 static void 1997 pf_send(struct pf_send_entry *pfse) 1998 { 1999 2000 PF_SENDQ_LOCK(); 2001 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2002 PF_SENDQ_UNLOCK(); 2003 swi_sched(V_pf_swi_cookie, 0); 2004 } 2005 2006 static bool 2007 pf_isforlocal(struct mbuf *m, int af) 2008 { 2009 switch (af) { 2010 #ifdef INET 2011 case AF_INET: { 2012 struct ip *ip = mtod(m, struct ip *); 2013 2014 return (in_localip(ip->ip_dst)); 2015 } 2016 #endif 2017 #ifdef INET6 2018 case AF_INET6: { 2019 struct ip6_hdr *ip6; 2020 struct in6_ifaddr *ia; 2021 ip6 = mtod(m, struct ip6_hdr *); 2022 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2023 if (ia == NULL) 2024 return (false); 2025 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2026 } 2027 #endif 2028 } 2029 2030 return (false); 2031 } 2032 2033 int 2034 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2035 int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type) 2036 { 2037 /* 2038 * ICMP types marked with PF_OUT are typically responses to 2039 * PF_IN, and will match states in the opposite direction. 2040 * PF_IN ICMP types need to match a state with that type. 2041 */ 2042 *icmp_dir = PF_OUT; 2043 *multi = PF_ICMP_MULTI_LINK; 2044 /* Queries (and responses) */ 2045 switch (pd->af) { 2046 #ifdef INET 2047 case AF_INET: 2048 switch (type) { 2049 case ICMP_ECHO: 2050 *icmp_dir = PF_IN; 2051 case ICMP_ECHOREPLY: 2052 *virtual_type = ICMP_ECHO; 2053 *virtual_id = pd->hdr.icmp.icmp_id; 2054 break; 2055 2056 case ICMP_TSTAMP: 2057 *icmp_dir = PF_IN; 2058 case ICMP_TSTAMPREPLY: 2059 *virtual_type = ICMP_TSTAMP; 2060 *virtual_id = pd->hdr.icmp.icmp_id; 2061 break; 2062 2063 case ICMP_IREQ: 2064 *icmp_dir = PF_IN; 2065 case ICMP_IREQREPLY: 2066 *virtual_type = ICMP_IREQ; 2067 *virtual_id = pd->hdr.icmp.icmp_id; 2068 break; 2069 2070 case ICMP_MASKREQ: 2071 *icmp_dir = PF_IN; 2072 case ICMP_MASKREPLY: 2073 *virtual_type = ICMP_MASKREQ; 2074 *virtual_id = pd->hdr.icmp.icmp_id; 2075 break; 2076 2077 case ICMP_IPV6_WHEREAREYOU: 2078 *icmp_dir = PF_IN; 2079 case ICMP_IPV6_IAMHERE: 2080 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2081 *virtual_id = 0; /* Nothing sane to match on! */ 2082 break; 2083 2084 case ICMP_MOBILE_REGREQUEST: 2085 *icmp_dir = PF_IN; 2086 case ICMP_MOBILE_REGREPLY: 2087 *virtual_type = ICMP_MOBILE_REGREQUEST; 2088 *virtual_id = 0; /* Nothing sane to match on! */ 2089 break; 2090 2091 case ICMP_ROUTERSOLICIT: 2092 *icmp_dir = PF_IN; 2093 case ICMP_ROUTERADVERT: 2094 *virtual_type = ICMP_ROUTERSOLICIT; 2095 *virtual_id = 0; /* Nothing sane to match on! */ 2096 break; 2097 2098 /* These ICMP types map to other connections */ 2099 case ICMP_UNREACH: 2100 case ICMP_SOURCEQUENCH: 2101 case ICMP_REDIRECT: 2102 case ICMP_TIMXCEED: 2103 case ICMP_PARAMPROB: 2104 /* These will not be used, but set them anyway */ 2105 *icmp_dir = PF_IN; 2106 *virtual_type = type; 2107 *virtual_id = 0; 2108 HTONS(*virtual_type); 2109 return (1); /* These types match to another state */ 2110 2111 /* 2112 * All remaining ICMP types get their own states, 2113 * and will only match in one direction. 2114 */ 2115 default: 2116 *icmp_dir = PF_IN; 2117 *virtual_type = type; 2118 *virtual_id = 0; 2119 break; 2120 } 2121 break; 2122 #endif /* INET */ 2123 #ifdef INET6 2124 case AF_INET6: 2125 switch (type) { 2126 case ICMP6_ECHO_REQUEST: 2127 *icmp_dir = PF_IN; 2128 case ICMP6_ECHO_REPLY: 2129 *virtual_type = ICMP6_ECHO_REQUEST; 2130 *virtual_id = pd->hdr.icmp6.icmp6_id; 2131 break; 2132 2133 case MLD_LISTENER_QUERY: 2134 case MLD_LISTENER_REPORT: { 2135 /* 2136 * Listener Report can be sent by clients 2137 * without an associated Listener Query. 2138 * In addition to that, when Report is sent as a 2139 * reply to a Query its source and destination 2140 * address are different. 2141 */ 2142 *icmp_dir = PF_IN; 2143 *virtual_type = MLD_LISTENER_QUERY; 2144 *virtual_id = 0; 2145 break; 2146 } 2147 case MLD_MTRACE: 2148 *icmp_dir = PF_IN; 2149 case MLD_MTRACE_RESP: 2150 *virtual_type = MLD_MTRACE; 2151 *virtual_id = 0; /* Nothing sane to match on! */ 2152 break; 2153 2154 case ND_NEIGHBOR_SOLICIT: 2155 *icmp_dir = PF_IN; 2156 case ND_NEIGHBOR_ADVERT: { 2157 *virtual_type = ND_NEIGHBOR_SOLICIT; 2158 *virtual_id = 0; 2159 break; 2160 } 2161 2162 /* 2163 * These ICMP types map to other connections. 2164 * ND_REDIRECT can't be in this list because the triggering 2165 * packet header is optional. 2166 */ 2167 case ICMP6_DST_UNREACH: 2168 case ICMP6_PACKET_TOO_BIG: 2169 case ICMP6_TIME_EXCEEDED: 2170 case ICMP6_PARAM_PROB: 2171 /* These will not be used, but set them anyway */ 2172 *icmp_dir = PF_IN; 2173 *virtual_type = type; 2174 *virtual_id = 0; 2175 HTONS(*virtual_type); 2176 return (1); /* These types match to another state */ 2177 /* 2178 * All remaining ICMP6 types get their own states, 2179 * and will only match in one direction. 2180 */ 2181 default: 2182 *icmp_dir = PF_IN; 2183 *virtual_type = type; 2184 *virtual_id = 0; 2185 break; 2186 } 2187 break; 2188 #endif /* INET6 */ 2189 } 2190 HTONS(*virtual_type); 2191 return (0); /* These types match to their own state */ 2192 } 2193 2194 void 2195 pf_intr(void *v) 2196 { 2197 struct epoch_tracker et; 2198 struct pf_send_head queue; 2199 struct pf_send_entry *pfse, *next; 2200 2201 CURVNET_SET((struct vnet *)v); 2202 2203 PF_SENDQ_LOCK(); 2204 queue = V_pf_sendqueue; 2205 STAILQ_INIT(&V_pf_sendqueue); 2206 PF_SENDQ_UNLOCK(); 2207 2208 NET_EPOCH_ENTER(et); 2209 2210 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2211 switch (pfse->pfse_type) { 2212 #ifdef INET 2213 case PFSE_IP: { 2214 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2215 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2216 ("%s: rcvif != loif", __func__)); 2217 2218 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2219 pfse->pfse_m->m_pkthdr.csum_flags |= 2220 CSUM_IP_VALID | CSUM_IP_CHECKED; 2221 ip_input(pfse->pfse_m); 2222 } else { 2223 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2224 NULL); 2225 } 2226 break; 2227 } 2228 case PFSE_ICMP: 2229 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2230 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2231 break; 2232 #endif /* INET */ 2233 #ifdef INET6 2234 case PFSE_IP6: 2235 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2236 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2237 ("%s: rcvif != loif", __func__)); 2238 2239 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL | 2240 M_LOOP; 2241 ip6_input(pfse->pfse_m); 2242 } else { 2243 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2244 NULL, NULL); 2245 } 2246 break; 2247 case PFSE_ICMP6: 2248 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2249 pfse->icmpopts.code, pfse->icmpopts.mtu); 2250 break; 2251 #endif /* INET6 */ 2252 default: 2253 panic("%s: unknown type", __func__); 2254 } 2255 free(pfse, M_PFTEMP); 2256 } 2257 NET_EPOCH_EXIT(et); 2258 CURVNET_RESTORE(); 2259 } 2260 2261 #define pf_purge_thread_period (hz / 10) 2262 2263 #ifdef PF_WANT_32_TO_64_COUNTER 2264 static void 2265 pf_status_counter_u64_periodic(void) 2266 { 2267 2268 PF_RULES_RASSERT(); 2269 2270 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2271 return; 2272 } 2273 2274 for (int i = 0; i < FCNT_MAX; i++) { 2275 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2276 } 2277 } 2278 2279 static void 2280 pf_kif_counter_u64_periodic(void) 2281 { 2282 struct pfi_kkif *kif; 2283 size_t r, run; 2284 2285 PF_RULES_RASSERT(); 2286 2287 if (__predict_false(V_pf_allkifcount == 0)) { 2288 return; 2289 } 2290 2291 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2292 return; 2293 } 2294 2295 run = V_pf_allkifcount / 10; 2296 if (run < 5) 2297 run = 5; 2298 2299 for (r = 0; r < run; r++) { 2300 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2301 if (kif == NULL) { 2302 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2303 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2304 break; 2305 } 2306 2307 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2308 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2309 2310 for (int i = 0; i < 2; i++) { 2311 for (int j = 0; j < 2; j++) { 2312 for (int k = 0; k < 2; k++) { 2313 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2314 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2315 } 2316 } 2317 } 2318 } 2319 } 2320 2321 static void 2322 pf_rule_counter_u64_periodic(void) 2323 { 2324 struct pf_krule *rule; 2325 size_t r, run; 2326 2327 PF_RULES_RASSERT(); 2328 2329 if (__predict_false(V_pf_allrulecount == 0)) { 2330 return; 2331 } 2332 2333 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2334 return; 2335 } 2336 2337 run = V_pf_allrulecount / 10; 2338 if (run < 5) 2339 run = 5; 2340 2341 for (r = 0; r < run; r++) { 2342 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2343 if (rule == NULL) { 2344 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2345 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2346 break; 2347 } 2348 2349 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2350 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2351 2352 pf_counter_u64_periodic(&rule->evaluations); 2353 for (int i = 0; i < 2; i++) { 2354 pf_counter_u64_periodic(&rule->packets[i]); 2355 pf_counter_u64_periodic(&rule->bytes[i]); 2356 } 2357 } 2358 } 2359 2360 static void 2361 pf_counter_u64_periodic_main(void) 2362 { 2363 PF_RULES_RLOCK_TRACKER; 2364 2365 V_pf_counter_periodic_iter++; 2366 2367 PF_RULES_RLOCK(); 2368 pf_counter_u64_critical_enter(); 2369 pf_status_counter_u64_periodic(); 2370 pf_kif_counter_u64_periodic(); 2371 pf_rule_counter_u64_periodic(); 2372 pf_counter_u64_critical_exit(); 2373 PF_RULES_RUNLOCK(); 2374 } 2375 #else 2376 #define pf_counter_u64_periodic_main() do { } while (0) 2377 #endif 2378 2379 void 2380 pf_purge_thread(void *unused __unused) 2381 { 2382 struct epoch_tracker et; 2383 2384 VNET_ITERATOR_DECL(vnet_iter); 2385 2386 sx_xlock(&pf_end_lock); 2387 while (pf_end_threads == 0) { 2388 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2389 2390 VNET_LIST_RLOCK(); 2391 NET_EPOCH_ENTER(et); 2392 VNET_FOREACH(vnet_iter) { 2393 CURVNET_SET(vnet_iter); 2394 2395 /* Wait until V_pf_default_rule is initialized. */ 2396 if (V_pf_vnet_active == 0) { 2397 CURVNET_RESTORE(); 2398 continue; 2399 } 2400 2401 pf_counter_u64_periodic_main(); 2402 2403 /* 2404 * Process 1/interval fraction of the state 2405 * table every run. 2406 */ 2407 V_pf_purge_idx = 2408 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2409 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2410 2411 /* 2412 * Purge other expired types every 2413 * PFTM_INTERVAL seconds. 2414 */ 2415 if (V_pf_purge_idx == 0) { 2416 /* 2417 * Order is important: 2418 * - states and src nodes reference rules 2419 * - states and rules reference kifs 2420 */ 2421 pf_purge_expired_fragments(); 2422 pf_purge_expired_src_nodes(); 2423 pf_purge_unlinked_rules(); 2424 pfi_kkif_purge(); 2425 } 2426 CURVNET_RESTORE(); 2427 } 2428 NET_EPOCH_EXIT(et); 2429 VNET_LIST_RUNLOCK(); 2430 } 2431 2432 pf_end_threads++; 2433 sx_xunlock(&pf_end_lock); 2434 kproc_exit(0); 2435 } 2436 2437 void 2438 pf_unload_vnet_purge(void) 2439 { 2440 2441 /* 2442 * To cleanse up all kifs and rules we need 2443 * two runs: first one clears reference flags, 2444 * then pf_purge_expired_states() doesn't 2445 * raise them, and then second run frees. 2446 */ 2447 pf_purge_unlinked_rules(); 2448 pfi_kkif_purge(); 2449 2450 /* 2451 * Now purge everything. 2452 */ 2453 pf_purge_expired_states(0, V_pf_hashmask); 2454 pf_purge_fragments(UINT_MAX); 2455 pf_purge_expired_src_nodes(); 2456 2457 /* 2458 * Now all kifs & rules should be unreferenced, 2459 * thus should be successfully freed. 2460 */ 2461 pf_purge_unlinked_rules(); 2462 pfi_kkif_purge(); 2463 } 2464 2465 u_int32_t 2466 pf_state_expires(const struct pf_kstate *state) 2467 { 2468 u_int32_t timeout; 2469 u_int32_t start; 2470 u_int32_t end; 2471 u_int32_t states; 2472 2473 /* handle all PFTM_* > PFTM_MAX here */ 2474 if (state->timeout == PFTM_PURGE) 2475 return (time_uptime); 2476 KASSERT(state->timeout != PFTM_UNLINKED, 2477 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2478 KASSERT((state->timeout < PFTM_MAX), 2479 ("pf_state_expires: timeout > PFTM_MAX")); 2480 timeout = state->rule->timeout[state->timeout]; 2481 if (!timeout) 2482 timeout = V_pf_default_rule.timeout[state->timeout]; 2483 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2484 if (start && state->rule != &V_pf_default_rule) { 2485 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2486 states = counter_u64_fetch(state->rule->states_cur); 2487 } else { 2488 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2489 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2490 states = V_pf_status.states; 2491 } 2492 if (end && states > start && start < end) { 2493 if (states < end) { 2494 timeout = (u_int64_t)timeout * (end - states) / 2495 (end - start); 2496 return ((state->expire / 1000) + timeout); 2497 } 2498 else 2499 return (time_uptime); 2500 } 2501 return ((state->expire / 1000) + timeout); 2502 } 2503 2504 void 2505 pf_purge_expired_src_nodes(void) 2506 { 2507 struct pf_ksrc_node_list freelist; 2508 struct pf_srchash *sh; 2509 struct pf_ksrc_node *cur, *next; 2510 int i; 2511 2512 LIST_INIT(&freelist); 2513 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2514 PF_HASHROW_LOCK(sh); 2515 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2516 if (cur->states == 0 && cur->expire <= time_uptime) { 2517 pf_unlink_src_node(cur); 2518 LIST_INSERT_HEAD(&freelist, cur, entry); 2519 } else if (cur->rule != NULL) 2520 cur->rule->rule_ref |= PFRULE_REFS; 2521 PF_HASHROW_UNLOCK(sh); 2522 } 2523 2524 pf_free_src_nodes(&freelist); 2525 2526 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2527 } 2528 2529 static void 2530 pf_src_tree_remove_state(struct pf_kstate *s) 2531 { 2532 struct pf_ksrc_node *sn; 2533 uint32_t timeout; 2534 2535 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2536 s->rule->timeout[PFTM_SRC_NODE] : 2537 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2538 2539 if (s->src_node != NULL) { 2540 sn = s->src_node; 2541 PF_SRC_NODE_LOCK(sn); 2542 if (s->src.tcp_est) 2543 --sn->conn; 2544 if (--sn->states == 0) 2545 sn->expire = time_uptime + timeout; 2546 PF_SRC_NODE_UNLOCK(sn); 2547 } 2548 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2549 sn = s->nat_src_node; 2550 PF_SRC_NODE_LOCK(sn); 2551 if (--sn->states == 0) 2552 sn->expire = time_uptime + timeout; 2553 PF_SRC_NODE_UNLOCK(sn); 2554 } 2555 s->src_node = s->nat_src_node = NULL; 2556 } 2557 2558 /* 2559 * Unlink and potentilly free a state. Function may be 2560 * called with ID hash row locked, but always returns 2561 * unlocked, since it needs to go through key hash locking. 2562 */ 2563 int 2564 pf_unlink_state(struct pf_kstate *s) 2565 { 2566 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2567 2568 NET_EPOCH_ASSERT(); 2569 PF_HASHROW_ASSERT(ih); 2570 2571 if (s->timeout == PFTM_UNLINKED) { 2572 /* 2573 * State is being processed 2574 * by pf_unlink_state() in 2575 * an other thread. 2576 */ 2577 PF_HASHROW_UNLOCK(ih); 2578 return (0); /* XXXGL: undefined actually */ 2579 } 2580 2581 if (s->src.state == PF_TCPS_PROXY_DST) { 2582 /* XXX wire key the right one? */ 2583 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2584 &s->key[PF_SK_WIRE]->addr[1], 2585 &s->key[PF_SK_WIRE]->addr[0], 2586 s->key[PF_SK_WIRE]->port[1], 2587 s->key[PF_SK_WIRE]->port[0], 2588 s->src.seqhi, s->src.seqlo + 1, 2589 TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0, 2590 s->act.rtableid); 2591 } 2592 2593 LIST_REMOVE(s, entry); 2594 pf_src_tree_remove_state(s); 2595 2596 if (V_pfsync_delete_state_ptr != NULL) 2597 V_pfsync_delete_state_ptr(s); 2598 2599 STATE_DEC_COUNTERS(s); 2600 2601 s->timeout = PFTM_UNLINKED; 2602 2603 /* Ensure we remove it from the list of halfopen states, if needed. */ 2604 if (s->key[PF_SK_STACK] != NULL && 2605 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2606 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2607 2608 PF_HASHROW_UNLOCK(ih); 2609 2610 pf_detach_state(s); 2611 2612 pf_udp_mapping_release(s->udp_mapping); 2613 2614 /* pf_state_insert() initialises refs to 2 */ 2615 return (pf_release_staten(s, 2)); 2616 } 2617 2618 struct pf_kstate * 2619 pf_alloc_state(int flags) 2620 { 2621 2622 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2623 } 2624 2625 void 2626 pf_free_state(struct pf_kstate *cur) 2627 { 2628 struct pf_krule_item *ri; 2629 2630 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2631 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2632 cur->timeout)); 2633 2634 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2635 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2636 free(ri, M_PF_RULE_ITEM); 2637 } 2638 2639 pf_normalize_tcp_cleanup(cur); 2640 uma_zfree(V_pf_state_z, cur); 2641 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2642 } 2643 2644 /* 2645 * Called only from pf_purge_thread(), thus serialized. 2646 */ 2647 static u_int 2648 pf_purge_expired_states(u_int i, int maxcheck) 2649 { 2650 struct pf_idhash *ih; 2651 struct pf_kstate *s; 2652 struct pf_krule_item *mrm; 2653 size_t count __unused; 2654 2655 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2656 2657 /* 2658 * Go through hash and unlink states that expire now. 2659 */ 2660 while (maxcheck > 0) { 2661 count = 0; 2662 ih = &V_pf_idhash[i]; 2663 2664 /* only take the lock if we expect to do work */ 2665 if (!LIST_EMPTY(&ih->states)) { 2666 relock: 2667 PF_HASHROW_LOCK(ih); 2668 LIST_FOREACH(s, &ih->states, entry) { 2669 if (pf_state_expires(s) <= time_uptime) { 2670 V_pf_status.states -= 2671 pf_unlink_state(s); 2672 goto relock; 2673 } 2674 s->rule->rule_ref |= PFRULE_REFS; 2675 if (s->nat_rule != NULL) 2676 s->nat_rule->rule_ref |= PFRULE_REFS; 2677 if (s->anchor != NULL) 2678 s->anchor->rule_ref |= PFRULE_REFS; 2679 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2680 SLIST_FOREACH(mrm, &s->match_rules, entry) 2681 mrm->r->rule_ref |= PFRULE_REFS; 2682 if (s->rt_kif) 2683 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2684 count++; 2685 } 2686 PF_HASHROW_UNLOCK(ih); 2687 } 2688 2689 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2690 2691 /* Return when we hit end of hash. */ 2692 if (++i > V_pf_hashmask) { 2693 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2694 return (0); 2695 } 2696 2697 maxcheck--; 2698 } 2699 2700 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2701 2702 return (i); 2703 } 2704 2705 static void 2706 pf_purge_unlinked_rules(void) 2707 { 2708 struct pf_krulequeue tmpq; 2709 struct pf_krule *r, *r1; 2710 2711 /* 2712 * If we have overloading task pending, then we'd 2713 * better skip purging this time. There is a tiny 2714 * probability that overloading task references 2715 * an already unlinked rule. 2716 */ 2717 PF_OVERLOADQ_LOCK(); 2718 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2719 PF_OVERLOADQ_UNLOCK(); 2720 return; 2721 } 2722 PF_OVERLOADQ_UNLOCK(); 2723 2724 /* 2725 * Do naive mark-and-sweep garbage collecting of old rules. 2726 * Reference flag is raised by pf_purge_expired_states() 2727 * and pf_purge_expired_src_nodes(). 2728 * 2729 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2730 * use a temporary queue. 2731 */ 2732 TAILQ_INIT(&tmpq); 2733 PF_UNLNKDRULES_LOCK(); 2734 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2735 if (!(r->rule_ref & PFRULE_REFS)) { 2736 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2737 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2738 } else 2739 r->rule_ref &= ~PFRULE_REFS; 2740 } 2741 PF_UNLNKDRULES_UNLOCK(); 2742 2743 if (!TAILQ_EMPTY(&tmpq)) { 2744 PF_CONFIG_LOCK(); 2745 PF_RULES_WLOCK(); 2746 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2747 TAILQ_REMOVE(&tmpq, r, entries); 2748 pf_free_rule(r); 2749 } 2750 PF_RULES_WUNLOCK(); 2751 PF_CONFIG_UNLOCK(); 2752 } 2753 } 2754 2755 void 2756 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2757 { 2758 switch (af) { 2759 #ifdef INET 2760 case AF_INET: { 2761 u_int32_t a = ntohl(addr->addr32[0]); 2762 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2763 (a>>8)&255, a&255); 2764 if (p) { 2765 p = ntohs(p); 2766 printf(":%u", p); 2767 } 2768 break; 2769 } 2770 #endif /* INET */ 2771 #ifdef INET6 2772 case AF_INET6: { 2773 u_int16_t b; 2774 u_int8_t i, curstart, curend, maxstart, maxend; 2775 curstart = curend = maxstart = maxend = 255; 2776 for (i = 0; i < 8; i++) { 2777 if (!addr->addr16[i]) { 2778 if (curstart == 255) 2779 curstart = i; 2780 curend = i; 2781 } else { 2782 if ((curend - curstart) > 2783 (maxend - maxstart)) { 2784 maxstart = curstart; 2785 maxend = curend; 2786 } 2787 curstart = curend = 255; 2788 } 2789 } 2790 if ((curend - curstart) > 2791 (maxend - maxstart)) { 2792 maxstart = curstart; 2793 maxend = curend; 2794 } 2795 for (i = 0; i < 8; i++) { 2796 if (i >= maxstart && i <= maxend) { 2797 if (i == 0) 2798 printf(":"); 2799 if (i == maxend) 2800 printf(":"); 2801 } else { 2802 b = ntohs(addr->addr16[i]); 2803 printf("%x", b); 2804 if (i < 7) 2805 printf(":"); 2806 } 2807 } 2808 if (p) { 2809 p = ntohs(p); 2810 printf("[%u]", p); 2811 } 2812 break; 2813 } 2814 #endif /* INET6 */ 2815 } 2816 } 2817 2818 void 2819 pf_print_state(struct pf_kstate *s) 2820 { 2821 pf_print_state_parts(s, NULL, NULL); 2822 } 2823 2824 static void 2825 pf_print_state_parts(struct pf_kstate *s, 2826 struct pf_state_key *skwp, struct pf_state_key *sksp) 2827 { 2828 struct pf_state_key *skw, *sks; 2829 u_int8_t proto, dir; 2830 2831 /* Do our best to fill these, but they're skipped if NULL */ 2832 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2833 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2834 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2835 dir = s ? s->direction : 0; 2836 2837 switch (proto) { 2838 case IPPROTO_IPV4: 2839 printf("IPv4"); 2840 break; 2841 case IPPROTO_IPV6: 2842 printf("IPv6"); 2843 break; 2844 case IPPROTO_TCP: 2845 printf("TCP"); 2846 break; 2847 case IPPROTO_UDP: 2848 printf("UDP"); 2849 break; 2850 case IPPROTO_ICMP: 2851 printf("ICMP"); 2852 break; 2853 case IPPROTO_ICMPV6: 2854 printf("ICMPv6"); 2855 break; 2856 default: 2857 printf("%u", proto); 2858 break; 2859 } 2860 switch (dir) { 2861 case PF_IN: 2862 printf(" in"); 2863 break; 2864 case PF_OUT: 2865 printf(" out"); 2866 break; 2867 } 2868 if (skw) { 2869 printf(" wire: "); 2870 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2871 printf(" "); 2872 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2873 } 2874 if (sks) { 2875 printf(" stack: "); 2876 if (sks != skw) { 2877 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2878 printf(" "); 2879 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2880 } else 2881 printf("-"); 2882 } 2883 if (s) { 2884 if (proto == IPPROTO_TCP) { 2885 printf(" [lo=%u high=%u win=%u modulator=%u", 2886 s->src.seqlo, s->src.seqhi, 2887 s->src.max_win, s->src.seqdiff); 2888 if (s->src.wscale && s->dst.wscale) 2889 printf(" wscale=%u", 2890 s->src.wscale & PF_WSCALE_MASK); 2891 printf("]"); 2892 printf(" [lo=%u high=%u win=%u modulator=%u", 2893 s->dst.seqlo, s->dst.seqhi, 2894 s->dst.max_win, s->dst.seqdiff); 2895 if (s->src.wscale && s->dst.wscale) 2896 printf(" wscale=%u", 2897 s->dst.wscale & PF_WSCALE_MASK); 2898 printf("]"); 2899 } 2900 printf(" %u:%u", s->src.state, s->dst.state); 2901 if (s->rule) 2902 printf(" @%d", s->rule->nr); 2903 } 2904 } 2905 2906 void 2907 pf_print_flags(u_int8_t f) 2908 { 2909 if (f) 2910 printf(" "); 2911 if (f & TH_FIN) 2912 printf("F"); 2913 if (f & TH_SYN) 2914 printf("S"); 2915 if (f & TH_RST) 2916 printf("R"); 2917 if (f & TH_PUSH) 2918 printf("P"); 2919 if (f & TH_ACK) 2920 printf("A"); 2921 if (f & TH_URG) 2922 printf("U"); 2923 if (f & TH_ECE) 2924 printf("E"); 2925 if (f & TH_CWR) 2926 printf("W"); 2927 } 2928 2929 #define PF_SET_SKIP_STEPS(i) \ 2930 do { \ 2931 while (head[i] != cur) { \ 2932 head[i]->skip[i] = cur; \ 2933 head[i] = TAILQ_NEXT(head[i], entries); \ 2934 } \ 2935 } while (0) 2936 2937 void 2938 pf_calc_skip_steps(struct pf_krulequeue *rules) 2939 { 2940 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2941 int i; 2942 2943 cur = TAILQ_FIRST(rules); 2944 prev = cur; 2945 for (i = 0; i < PF_SKIP_COUNT; ++i) 2946 head[i] = cur; 2947 while (cur != NULL) { 2948 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2949 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2950 if (cur->direction != prev->direction) 2951 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2952 if (cur->af != prev->af) 2953 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2954 if (cur->proto != prev->proto) 2955 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2956 if (cur->src.neg != prev->src.neg || 2957 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2958 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2959 if (cur->dst.neg != prev->dst.neg || 2960 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2961 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2962 if (cur->src.port[0] != prev->src.port[0] || 2963 cur->src.port[1] != prev->src.port[1] || 2964 cur->src.port_op != prev->src.port_op) 2965 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2966 if (cur->dst.port[0] != prev->dst.port[0] || 2967 cur->dst.port[1] != prev->dst.port[1] || 2968 cur->dst.port_op != prev->dst.port_op) 2969 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2970 2971 prev = cur; 2972 cur = TAILQ_NEXT(cur, entries); 2973 } 2974 for (i = 0; i < PF_SKIP_COUNT; ++i) 2975 PF_SET_SKIP_STEPS(i); 2976 } 2977 2978 int 2979 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2980 { 2981 if (aw1->type != aw2->type) 2982 return (1); 2983 switch (aw1->type) { 2984 case PF_ADDR_ADDRMASK: 2985 case PF_ADDR_RANGE: 2986 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2987 return (1); 2988 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2989 return (1); 2990 return (0); 2991 case PF_ADDR_DYNIFTL: 2992 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2993 case PF_ADDR_NOROUTE: 2994 case PF_ADDR_URPFFAILED: 2995 return (0); 2996 case PF_ADDR_TABLE: 2997 return (aw1->p.tbl != aw2->p.tbl); 2998 default: 2999 printf("invalid address type: %d\n", aw1->type); 3000 return (1); 3001 } 3002 } 3003 3004 /** 3005 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3006 * header isn't always a full checksum. In some cases (i.e. output) it's a 3007 * pseudo-header checksum, which is a partial checksum over src/dst IP 3008 * addresses, protocol number and length. 3009 * 3010 * That means we have the following cases: 3011 * * Input or forwarding: we don't have TSO, the checksum fields are full 3012 * checksums, we need to update the checksum whenever we change anything. 3013 * * Output (i.e. the checksum is a pseudo-header checksum): 3014 * x The field being updated is src/dst address or affects the length of 3015 * the packet. We need to update the pseudo-header checksum (note that this 3016 * checksum is not ones' complement). 3017 * x Some other field is being modified (e.g. src/dst port numbers): We 3018 * don't have to update anything. 3019 **/ 3020 u_int16_t 3021 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3022 { 3023 u_int32_t x; 3024 3025 x = cksum + old - new; 3026 x = (x + (x >> 16)) & 0xffff; 3027 3028 /* optimise: eliminate a branch when not udp */ 3029 if (udp && cksum == 0x0000) 3030 return cksum; 3031 if (udp && x == 0x0000) 3032 x = 0xffff; 3033 3034 return (u_int16_t)(x); 3035 } 3036 3037 static void 3038 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 3039 u_int8_t udp) 3040 { 3041 u_int16_t old = htons(hi ? (*f << 8) : *f); 3042 u_int16_t new = htons(hi ? ( v << 8) : v); 3043 3044 if (*f == v) 3045 return; 3046 3047 *f = v; 3048 3049 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3050 return; 3051 3052 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 3053 } 3054 3055 void 3056 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 3057 bool hi, u_int8_t udp) 3058 { 3059 u_int8_t *fb = (u_int8_t *)f; 3060 u_int8_t *vb = (u_int8_t *)&v; 3061 3062 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3063 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3064 } 3065 3066 void 3067 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 3068 bool hi, u_int8_t udp) 3069 { 3070 u_int8_t *fb = (u_int8_t *)f; 3071 u_int8_t *vb = (u_int8_t *)&v; 3072 3073 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3074 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3075 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3076 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3077 } 3078 3079 u_int16_t 3080 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3081 u_int16_t new, u_int8_t udp) 3082 { 3083 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3084 return (cksum); 3085 3086 return (pf_cksum_fixup(cksum, old, new, udp)); 3087 } 3088 3089 static void 3090 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 3091 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 3092 sa_family_t af) 3093 { 3094 struct pf_addr ao; 3095 u_int16_t po = *p; 3096 3097 PF_ACPY(&ao, a, af); 3098 PF_ACPY(a, an, af); 3099 3100 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3101 *pc = ~*pc; 3102 3103 *p = pn; 3104 3105 switch (af) { 3106 #ifdef INET 3107 case AF_INET: 3108 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3109 ao.addr16[0], an->addr16[0], 0), 3110 ao.addr16[1], an->addr16[1], 0); 3111 *p = pn; 3112 3113 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 3114 ao.addr16[0], an->addr16[0], u), 3115 ao.addr16[1], an->addr16[1], u); 3116 3117 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3118 break; 3119 #endif /* INET */ 3120 #ifdef INET6 3121 case AF_INET6: 3122 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3123 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3124 pf_cksum_fixup(pf_cksum_fixup(*pc, 3125 ao.addr16[0], an->addr16[0], u), 3126 ao.addr16[1], an->addr16[1], u), 3127 ao.addr16[2], an->addr16[2], u), 3128 ao.addr16[3], an->addr16[3], u), 3129 ao.addr16[4], an->addr16[4], u), 3130 ao.addr16[5], an->addr16[5], u), 3131 ao.addr16[6], an->addr16[6], u), 3132 ao.addr16[7], an->addr16[7], u); 3133 3134 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3135 break; 3136 #endif /* INET6 */ 3137 } 3138 3139 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3140 CSUM_DELAY_DATA_IPV6)) { 3141 *pc = ~*pc; 3142 if (! *pc) 3143 *pc = 0xffff; 3144 } 3145 } 3146 3147 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3148 void 3149 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3150 { 3151 u_int32_t ao; 3152 3153 memcpy(&ao, a, sizeof(ao)); 3154 memcpy(a, &an, sizeof(u_int32_t)); 3155 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3156 ao % 65536, an % 65536, u); 3157 } 3158 3159 void 3160 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3161 { 3162 u_int32_t ao; 3163 3164 memcpy(&ao, a, sizeof(ao)); 3165 memcpy(a, &an, sizeof(u_int32_t)); 3166 3167 *c = pf_proto_cksum_fixup(m, 3168 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3169 ao % 65536, an % 65536, udp); 3170 } 3171 3172 #ifdef INET6 3173 static void 3174 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3175 { 3176 struct pf_addr ao; 3177 3178 PF_ACPY(&ao, a, AF_INET6); 3179 PF_ACPY(a, an, AF_INET6); 3180 3181 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3182 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3183 pf_cksum_fixup(pf_cksum_fixup(*c, 3184 ao.addr16[0], an->addr16[0], u), 3185 ao.addr16[1], an->addr16[1], u), 3186 ao.addr16[2], an->addr16[2], u), 3187 ao.addr16[3], an->addr16[3], u), 3188 ao.addr16[4], an->addr16[4], u), 3189 ao.addr16[5], an->addr16[5], u), 3190 ao.addr16[6], an->addr16[6], u), 3191 ao.addr16[7], an->addr16[7], u); 3192 } 3193 #endif /* INET6 */ 3194 3195 static void 3196 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3197 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3198 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3199 { 3200 struct pf_addr oia, ooa; 3201 3202 PF_ACPY(&oia, ia, af); 3203 if (oa) 3204 PF_ACPY(&ooa, oa, af); 3205 3206 /* Change inner protocol port, fix inner protocol checksum. */ 3207 if (ip != NULL) { 3208 u_int16_t oip = *ip; 3209 u_int32_t opc; 3210 3211 if (pc != NULL) 3212 opc = *pc; 3213 *ip = np; 3214 if (pc != NULL) 3215 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3216 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3217 if (pc != NULL) 3218 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3219 } 3220 /* Change inner ip address, fix inner ip and icmp checksums. */ 3221 PF_ACPY(ia, na, af); 3222 switch (af) { 3223 #ifdef INET 3224 case AF_INET: { 3225 u_int32_t oh2c = *h2c; 3226 3227 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3228 oia.addr16[0], ia->addr16[0], 0), 3229 oia.addr16[1], ia->addr16[1], 0); 3230 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3231 oia.addr16[0], ia->addr16[0], 0), 3232 oia.addr16[1], ia->addr16[1], 0); 3233 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3234 break; 3235 } 3236 #endif /* INET */ 3237 #ifdef INET6 3238 case AF_INET6: 3239 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3240 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3241 pf_cksum_fixup(pf_cksum_fixup(*ic, 3242 oia.addr16[0], ia->addr16[0], u), 3243 oia.addr16[1], ia->addr16[1], u), 3244 oia.addr16[2], ia->addr16[2], u), 3245 oia.addr16[3], ia->addr16[3], u), 3246 oia.addr16[4], ia->addr16[4], u), 3247 oia.addr16[5], ia->addr16[5], u), 3248 oia.addr16[6], ia->addr16[6], u), 3249 oia.addr16[7], ia->addr16[7], u); 3250 break; 3251 #endif /* INET6 */ 3252 } 3253 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3254 if (oa) { 3255 PF_ACPY(oa, na, af); 3256 switch (af) { 3257 #ifdef INET 3258 case AF_INET: 3259 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3260 ooa.addr16[0], oa->addr16[0], 0), 3261 ooa.addr16[1], oa->addr16[1], 0); 3262 break; 3263 #endif /* INET */ 3264 #ifdef INET6 3265 case AF_INET6: 3266 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3267 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3268 pf_cksum_fixup(pf_cksum_fixup(*ic, 3269 ooa.addr16[0], oa->addr16[0], u), 3270 ooa.addr16[1], oa->addr16[1], u), 3271 ooa.addr16[2], oa->addr16[2], u), 3272 ooa.addr16[3], oa->addr16[3], u), 3273 ooa.addr16[4], oa->addr16[4], u), 3274 ooa.addr16[5], oa->addr16[5], u), 3275 ooa.addr16[6], oa->addr16[6], u), 3276 ooa.addr16[7], oa->addr16[7], u); 3277 break; 3278 #endif /* INET6 */ 3279 } 3280 } 3281 } 3282 3283 /* 3284 * Need to modulate the sequence numbers in the TCP SACK option 3285 * (credits to Krzysztof Pfaff for report and patch) 3286 */ 3287 static int 3288 pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th, 3289 struct pf_state_peer *dst) 3290 { 3291 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3292 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3293 int copyback = 0, i, olen; 3294 struct sackblk sack; 3295 3296 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3297 if (hlen < TCPOLEN_SACKLEN || 3298 !pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3299 return 0; 3300 3301 while (hlen >= TCPOLEN_SACKLEN) { 3302 size_t startoff = opt - opts; 3303 olen = opt[1]; 3304 switch (*opt) { 3305 case TCPOPT_EOL: /* FALLTHROUGH */ 3306 case TCPOPT_NOP: 3307 opt++; 3308 hlen--; 3309 break; 3310 case TCPOPT_SACK: 3311 if (olen > hlen) 3312 olen = hlen; 3313 if (olen >= TCPOLEN_SACKLEN) { 3314 for (i = 2; i + TCPOLEN_SACK <= olen; 3315 i += TCPOLEN_SACK) { 3316 memcpy(&sack, &opt[i], sizeof(sack)); 3317 pf_patch_32_unaligned(pd->m, 3318 &th->th_sum, &sack.start, 3319 htonl(ntohl(sack.start) - dst->seqdiff), 3320 PF_ALGNMNT(startoff), 3321 0); 3322 pf_patch_32_unaligned(pd->m, &th->th_sum, 3323 &sack.end, 3324 htonl(ntohl(sack.end) - dst->seqdiff), 3325 PF_ALGNMNT(startoff), 3326 0); 3327 memcpy(&opt[i], &sack, sizeof(sack)); 3328 } 3329 copyback = 1; 3330 } 3331 /* FALLTHROUGH */ 3332 default: 3333 if (olen < 2) 3334 olen = 2; 3335 hlen -= olen; 3336 opt += olen; 3337 } 3338 } 3339 3340 if (copyback) 3341 m_copyback(pd->m, pd->off + sizeof(*th), thoptlen, (caddr_t)opts); 3342 return (copyback); 3343 } 3344 3345 struct mbuf * 3346 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3347 const struct pf_addr *saddr, const struct pf_addr *daddr, 3348 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3349 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3350 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3351 { 3352 struct mbuf *m; 3353 int len, tlen; 3354 #ifdef INET 3355 struct ip *h = NULL; 3356 #endif /* INET */ 3357 #ifdef INET6 3358 struct ip6_hdr *h6 = NULL; 3359 #endif /* INET6 */ 3360 struct tcphdr *th; 3361 char *opt; 3362 struct pf_mtag *pf_mtag; 3363 3364 len = 0; 3365 th = NULL; 3366 3367 /* maximum segment size tcp option */ 3368 tlen = sizeof(struct tcphdr); 3369 if (mss) 3370 tlen += 4; 3371 3372 switch (af) { 3373 #ifdef INET 3374 case AF_INET: 3375 len = sizeof(struct ip) + tlen; 3376 break; 3377 #endif /* INET */ 3378 #ifdef INET6 3379 case AF_INET6: 3380 len = sizeof(struct ip6_hdr) + tlen; 3381 break; 3382 #endif /* INET6 */ 3383 } 3384 3385 m = m_gethdr(M_NOWAIT, MT_DATA); 3386 if (m == NULL) 3387 return (NULL); 3388 3389 #ifdef MAC 3390 mac_netinet_firewall_send(m); 3391 #endif 3392 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 3393 m_freem(m); 3394 return (NULL); 3395 } 3396 m->m_flags |= mbuf_flags; 3397 pf_mtag->tag = mtag_tag; 3398 pf_mtag->flags = mtag_flags; 3399 3400 if (rtableid >= 0) 3401 M_SETFIB(m, rtableid); 3402 3403 #ifdef ALTQ 3404 if (r != NULL && r->qid) { 3405 pf_mtag->qid = r->qid; 3406 3407 /* add hints for ecn */ 3408 pf_mtag->hdr = mtod(m, struct ip *); 3409 } 3410 #endif /* ALTQ */ 3411 m->m_data += max_linkhdr; 3412 m->m_pkthdr.len = m->m_len = len; 3413 /* The rest of the stack assumes a rcvif, so provide one. 3414 * This is a locally generated packet, so .. close enough. */ 3415 m->m_pkthdr.rcvif = V_loif; 3416 bzero(m->m_data, len); 3417 switch (af) { 3418 #ifdef INET 3419 case AF_INET: 3420 h = mtod(m, struct ip *); 3421 3422 /* IP header fields included in the TCP checksum */ 3423 h->ip_p = IPPROTO_TCP; 3424 h->ip_len = htons(tlen); 3425 h->ip_src.s_addr = saddr->v4.s_addr; 3426 h->ip_dst.s_addr = daddr->v4.s_addr; 3427 3428 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3429 break; 3430 #endif /* INET */ 3431 #ifdef INET6 3432 case AF_INET6: 3433 h6 = mtod(m, struct ip6_hdr *); 3434 3435 /* IP header fields included in the TCP checksum */ 3436 h6->ip6_nxt = IPPROTO_TCP; 3437 h6->ip6_plen = htons(tlen); 3438 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3439 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3440 3441 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3442 break; 3443 #endif /* INET6 */ 3444 } 3445 3446 /* TCP header */ 3447 th->th_sport = sport; 3448 th->th_dport = dport; 3449 th->th_seq = htonl(seq); 3450 th->th_ack = htonl(ack); 3451 th->th_off = tlen >> 2; 3452 th->th_flags = tcp_flags; 3453 th->th_win = htons(win); 3454 3455 if (mss) { 3456 opt = (char *)(th + 1); 3457 opt[0] = TCPOPT_MAXSEG; 3458 opt[1] = 4; 3459 HTONS(mss); 3460 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 3461 } 3462 3463 switch (af) { 3464 #ifdef INET 3465 case AF_INET: 3466 /* TCP checksum */ 3467 th->th_sum = in_cksum(m, len); 3468 3469 /* Finish the IP header */ 3470 h->ip_v = 4; 3471 h->ip_hl = sizeof(*h) >> 2; 3472 h->ip_tos = IPTOS_LOWDELAY; 3473 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 3474 h->ip_len = htons(len); 3475 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3476 h->ip_sum = 0; 3477 break; 3478 #endif /* INET */ 3479 #ifdef INET6 3480 case AF_INET6: 3481 /* TCP checksum */ 3482 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3483 sizeof(struct ip6_hdr), tlen); 3484 3485 h6->ip6_vfc |= IPV6_VERSION; 3486 h6->ip6_hlim = IPV6_DEFHLIM; 3487 break; 3488 #endif /* INET6 */ 3489 } 3490 3491 return (m); 3492 } 3493 3494 static void 3495 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3496 uint8_t ttl, int rtableid) 3497 { 3498 struct mbuf *m; 3499 #ifdef INET 3500 struct ip *h = NULL; 3501 #endif /* INET */ 3502 #ifdef INET6 3503 struct ip6_hdr *h6 = NULL; 3504 #endif /* INET6 */ 3505 struct sctphdr *hdr; 3506 struct sctp_chunkhdr *chunk; 3507 struct pf_send_entry *pfse; 3508 int off = 0; 3509 3510 MPASS(af == pd->af); 3511 3512 m = m_gethdr(M_NOWAIT, MT_DATA); 3513 if (m == NULL) 3514 return; 3515 3516 m->m_data += max_linkhdr; 3517 m->m_flags |= M_SKIP_FIREWALL; 3518 /* The rest of the stack assumes a rcvif, so provide one. 3519 * This is a locally generated packet, so .. close enough. */ 3520 m->m_pkthdr.rcvif = V_loif; 3521 3522 /* IPv4|6 header */ 3523 switch (af) { 3524 #ifdef INET 3525 case AF_INET: 3526 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3527 3528 h = mtod(m, struct ip *); 3529 3530 /* IP header fields included in the TCP checksum */ 3531 3532 h->ip_p = IPPROTO_SCTP; 3533 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3534 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3535 h->ip_src = pd->dst->v4; 3536 h->ip_dst = pd->src->v4; 3537 3538 off += sizeof(struct ip); 3539 break; 3540 #endif /* INET */ 3541 #ifdef INET6 3542 case AF_INET6: 3543 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3544 3545 h6 = mtod(m, struct ip6_hdr *); 3546 3547 /* IP header fields included in the TCP checksum */ 3548 h6->ip6_vfc |= IPV6_VERSION; 3549 h6->ip6_nxt = IPPROTO_SCTP; 3550 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3551 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3552 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3553 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3554 3555 off += sizeof(struct ip6_hdr); 3556 break; 3557 #endif /* INET6 */ 3558 } 3559 3560 /* SCTP header */ 3561 hdr = mtodo(m, off); 3562 3563 hdr->src_port = pd->hdr.sctp.dest_port; 3564 hdr->dest_port = pd->hdr.sctp.src_port; 3565 hdr->v_tag = pd->sctp_initiate_tag; 3566 hdr->checksum = 0; 3567 3568 /* Abort chunk. */ 3569 off += sizeof(struct sctphdr); 3570 chunk = mtodo(m, off); 3571 3572 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3573 chunk->chunk_length = htons(sizeof(*chunk)); 3574 3575 /* SCTP checksum */ 3576 off += sizeof(*chunk); 3577 m->m_pkthdr.len = m->m_len = off; 3578 3579 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 3580 3581 if (rtableid >= 0) 3582 M_SETFIB(m, rtableid); 3583 3584 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3585 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3586 if (pfse == NULL) { 3587 m_freem(m); 3588 return; 3589 } 3590 3591 switch (af) { 3592 #ifdef INET 3593 case AF_INET: 3594 pfse->pfse_type = PFSE_IP; 3595 break; 3596 #endif /* INET */ 3597 #ifdef INET6 3598 case AF_INET6: 3599 pfse->pfse_type = PFSE_IP6; 3600 break; 3601 #endif /* INET6 */ 3602 } 3603 3604 pfse->pfse_m = m; 3605 pf_send(pfse); 3606 } 3607 3608 void 3609 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3610 const struct pf_addr *saddr, const struct pf_addr *daddr, 3611 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3612 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3613 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3614 { 3615 struct pf_send_entry *pfse; 3616 struct mbuf *m; 3617 3618 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3619 win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, rtableid); 3620 if (m == NULL) 3621 return; 3622 3623 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3624 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3625 if (pfse == NULL) { 3626 m_freem(m); 3627 return; 3628 } 3629 3630 switch (af) { 3631 #ifdef INET 3632 case AF_INET: 3633 pfse->pfse_type = PFSE_IP; 3634 break; 3635 #endif /* INET */ 3636 #ifdef INET6 3637 case AF_INET6: 3638 pfse->pfse_type = PFSE_IP6; 3639 break; 3640 #endif /* INET6 */ 3641 } 3642 3643 pfse->pfse_m = m; 3644 pf_send(pfse); 3645 } 3646 3647 static void 3648 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3649 struct pf_state_key *sk, struct tcphdr *th, 3650 u_int16_t bproto_sum, u_int16_t bip_sum, 3651 u_short *reason, int rtableid) 3652 { 3653 struct pf_addr * const saddr = pd->src; 3654 struct pf_addr * const daddr = pd->dst; 3655 3656 /* undo NAT changes, if they have taken place */ 3657 if (nr != NULL) { 3658 PF_ACPY(saddr, &sk->addr[pd->sidx], pd->af); 3659 PF_ACPY(daddr, &sk->addr[pd->didx], pd->af); 3660 if (pd->sport) 3661 *pd->sport = sk->port[pd->sidx]; 3662 if (pd->dport) 3663 *pd->dport = sk->port[pd->didx]; 3664 if (pd->proto_sum) 3665 *pd->proto_sum = bproto_sum; 3666 if (pd->ip_sum) 3667 *pd->ip_sum = bip_sum; 3668 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 3669 } 3670 if (pd->proto == IPPROTO_TCP && 3671 ((r->rule_flag & PFRULE_RETURNRST) || 3672 (r->rule_flag & PFRULE_RETURN)) && 3673 !(th->th_flags & TH_RST)) { 3674 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3675 3676 if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off, 3677 IPPROTO_TCP, pd->af)) 3678 REASON_SET(reason, PFRES_PROTCKSUM); 3679 else { 3680 if (th->th_flags & TH_SYN) 3681 ack++; 3682 if (th->th_flags & TH_FIN) 3683 ack++; 3684 pf_send_tcp(r, pd->af, pd->dst, 3685 pd->src, th->th_dport, th->th_sport, 3686 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3687 r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid); 3688 } 3689 } else if (pd->proto == IPPROTO_SCTP && 3690 (r->rule_flag & PFRULE_RETURN)) { 3691 pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid); 3692 } else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET && 3693 r->return_icmp) 3694 pf_send_icmp(pd->m, r->return_icmp >> 8, 3695 r->return_icmp & 255, pd->af, r, rtableid); 3696 else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 && 3697 r->return_icmp6) 3698 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 3699 r->return_icmp6 & 255, pd->af, r, rtableid); 3700 } 3701 3702 static int 3703 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3704 { 3705 struct m_tag *mtag; 3706 u_int8_t mpcp; 3707 3708 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3709 if (mtag == NULL) 3710 return (0); 3711 3712 if (prio == PF_PRIO_ZERO) 3713 prio = 0; 3714 3715 mpcp = *(uint8_t *)(mtag + 1); 3716 3717 return (mpcp == prio); 3718 } 3719 3720 static int 3721 pf_icmp_to_bandlim(uint8_t type) 3722 { 3723 switch (type) { 3724 case ICMP_ECHO: 3725 case ICMP_ECHOREPLY: 3726 return (BANDLIM_ICMP_ECHO); 3727 case ICMP_TSTAMP: 3728 case ICMP_TSTAMPREPLY: 3729 return (BANDLIM_ICMP_TSTAMP); 3730 case ICMP_UNREACH: 3731 default: 3732 return (BANDLIM_ICMP_UNREACH); 3733 } 3734 } 3735 3736 static void 3737 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3738 struct pf_krule *r, int rtableid) 3739 { 3740 struct pf_send_entry *pfse; 3741 struct mbuf *m0; 3742 struct pf_mtag *pf_mtag; 3743 3744 /* ICMP packet rate limitation. */ 3745 switch (af) { 3746 #ifdef INET6 3747 case AF_INET6: 3748 if (icmp6_ratelimit(NULL, type, code)) 3749 return; 3750 break; 3751 #endif 3752 #ifdef INET 3753 case AF_INET: 3754 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3755 return; 3756 break; 3757 #endif 3758 } 3759 3760 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3761 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3762 if (pfse == NULL) 3763 return; 3764 3765 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3766 free(pfse, M_PFTEMP); 3767 return; 3768 } 3769 3770 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3771 free(pfse, M_PFTEMP); 3772 return; 3773 } 3774 /* XXX: revisit */ 3775 m0->m_flags |= M_SKIP_FIREWALL; 3776 3777 if (rtableid >= 0) 3778 M_SETFIB(m0, rtableid); 3779 3780 #ifdef ALTQ 3781 if (r->qid) { 3782 pf_mtag->qid = r->qid; 3783 /* add hints for ecn */ 3784 pf_mtag->hdr = mtod(m0, struct ip *); 3785 } 3786 #endif /* ALTQ */ 3787 3788 switch (af) { 3789 #ifdef INET 3790 case AF_INET: 3791 pfse->pfse_type = PFSE_ICMP; 3792 break; 3793 #endif /* INET */ 3794 #ifdef INET6 3795 case AF_INET6: 3796 pfse->pfse_type = PFSE_ICMP6; 3797 break; 3798 #endif /* INET6 */ 3799 } 3800 pfse->pfse_m = m0; 3801 pfse->icmpopts.type = type; 3802 pfse->icmpopts.code = code; 3803 pf_send(pfse); 3804 } 3805 3806 /* 3807 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3808 * If n is 0, they match if they are equal. If n is != 0, they match if they 3809 * are different. 3810 */ 3811 int 3812 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3813 struct pf_addr *b, sa_family_t af) 3814 { 3815 int match = 0; 3816 3817 switch (af) { 3818 #ifdef INET 3819 case AF_INET: 3820 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 3821 match++; 3822 break; 3823 #endif /* INET */ 3824 #ifdef INET6 3825 case AF_INET6: 3826 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 3827 match++; 3828 break; 3829 #endif /* INET6 */ 3830 } 3831 if (match) { 3832 if (n) 3833 return (0); 3834 else 3835 return (1); 3836 } else { 3837 if (n) 3838 return (1); 3839 else 3840 return (0); 3841 } 3842 } 3843 3844 /* 3845 * Return 1 if b <= a <= e, otherwise return 0. 3846 */ 3847 int 3848 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3849 struct pf_addr *a, sa_family_t af) 3850 { 3851 switch (af) { 3852 #ifdef INET 3853 case AF_INET: 3854 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3855 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3856 return (0); 3857 break; 3858 #endif /* INET */ 3859 #ifdef INET6 3860 case AF_INET6: { 3861 int i; 3862 3863 /* check a >= b */ 3864 for (i = 0; i < 4; ++i) 3865 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3866 break; 3867 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3868 return (0); 3869 /* check a <= e */ 3870 for (i = 0; i < 4; ++i) 3871 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3872 break; 3873 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3874 return (0); 3875 break; 3876 } 3877 #endif /* INET6 */ 3878 } 3879 return (1); 3880 } 3881 3882 static int 3883 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3884 { 3885 switch (op) { 3886 case PF_OP_IRG: 3887 return ((p > a1) && (p < a2)); 3888 case PF_OP_XRG: 3889 return ((p < a1) || (p > a2)); 3890 case PF_OP_RRG: 3891 return ((p >= a1) && (p <= a2)); 3892 case PF_OP_EQ: 3893 return (p == a1); 3894 case PF_OP_NE: 3895 return (p != a1); 3896 case PF_OP_LT: 3897 return (p < a1); 3898 case PF_OP_LE: 3899 return (p <= a1); 3900 case PF_OP_GT: 3901 return (p > a1); 3902 case PF_OP_GE: 3903 return (p >= a1); 3904 } 3905 return (0); /* never reached */ 3906 } 3907 3908 int 3909 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3910 { 3911 NTOHS(a1); 3912 NTOHS(a2); 3913 NTOHS(p); 3914 return (pf_match(op, a1, a2, p)); 3915 } 3916 3917 static int 3918 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3919 { 3920 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3921 return (0); 3922 return (pf_match(op, a1, a2, u)); 3923 } 3924 3925 static int 3926 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3927 { 3928 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3929 return (0); 3930 return (pf_match(op, a1, a2, g)); 3931 } 3932 3933 int 3934 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3935 { 3936 if (*tag == -1) 3937 *tag = mtag; 3938 3939 return ((!r->match_tag_not && r->match_tag == *tag) || 3940 (r->match_tag_not && r->match_tag != *tag)); 3941 } 3942 3943 static int 3944 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 3945 { 3946 struct ifnet *ifp = m->m_pkthdr.rcvif; 3947 struct pfi_kkif *kif; 3948 3949 if (ifp == NULL) 3950 return (0); 3951 3952 kif = (struct pfi_kkif *)ifp->if_pf_kif; 3953 3954 if (kif == NULL) { 3955 DPFPRINTF(PF_DEBUG_URGENT, 3956 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 3957 r->rcv_ifname)); 3958 return (0); 3959 } 3960 3961 return (pfi_kkif_match(r->rcv_kif, kif)); 3962 } 3963 3964 int 3965 pf_tag_packet(struct pf_pdesc *pd, int tag) 3966 { 3967 3968 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3969 3970 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) 3971 return (ENOMEM); 3972 3973 pd->pf_mtag->tag = tag; 3974 3975 return (0); 3976 } 3977 3978 #define PF_ANCHOR_STACKSIZE 32 3979 struct pf_kanchor_stackframe { 3980 struct pf_kruleset *rs; 3981 struct pf_krule *r; /* XXX: + match bit */ 3982 struct pf_kanchor *child; 3983 }; 3984 3985 /* 3986 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3987 */ 3988 #define PF_ANCHORSTACK_MATCH 0x00000001 3989 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 3990 3991 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3992 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 3993 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3994 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3995 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3996 } while (0) 3997 3998 void 3999 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4000 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4001 int *match) 4002 { 4003 struct pf_kanchor_stackframe *f; 4004 4005 PF_RULES_RASSERT(); 4006 4007 if (match) 4008 *match = 0; 4009 if (*depth >= PF_ANCHOR_STACKSIZE) { 4010 printf("%s: anchor stack overflow on %s\n", 4011 __func__, (*r)->anchor->name); 4012 *r = TAILQ_NEXT(*r, entries); 4013 return; 4014 } else if (*depth == 0 && a != NULL) 4015 *a = *r; 4016 f = stack + (*depth)++; 4017 f->rs = *rs; 4018 f->r = *r; 4019 if ((*r)->anchor_wildcard) { 4020 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4021 4022 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4023 *r = NULL; 4024 return; 4025 } 4026 *rs = &f->child->ruleset; 4027 } else { 4028 f->child = NULL; 4029 *rs = &(*r)->anchor->ruleset; 4030 } 4031 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4032 } 4033 4034 int 4035 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4036 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4037 int *match) 4038 { 4039 struct pf_kanchor_stackframe *f; 4040 struct pf_krule *fr; 4041 int quick = 0; 4042 4043 PF_RULES_RASSERT(); 4044 4045 do { 4046 if (*depth <= 0) 4047 break; 4048 f = stack + *depth - 1; 4049 fr = PF_ANCHOR_RULE(f); 4050 if (f->child != NULL) { 4051 /* 4052 * This block traverses through 4053 * a wildcard anchor. 4054 */ 4055 if (match != NULL && *match) { 4056 /* 4057 * If any of "*" matched, then 4058 * "foo/ *" matched, mark frame 4059 * appropriately. 4060 */ 4061 PF_ANCHOR_SET_MATCH(f); 4062 *match = 0; 4063 } 4064 f->child = RB_NEXT(pf_kanchor_node, 4065 &fr->anchor->children, f->child); 4066 if (f->child != NULL) { 4067 *rs = &f->child->ruleset; 4068 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4069 if (*r == NULL) 4070 continue; 4071 else 4072 break; 4073 } 4074 } 4075 (*depth)--; 4076 if (*depth == 0 && a != NULL) 4077 *a = NULL; 4078 *rs = f->rs; 4079 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 4080 quick = fr->quick; 4081 *r = TAILQ_NEXT(fr, entries); 4082 } while (*r == NULL); 4083 4084 return (quick); 4085 } 4086 4087 struct pf_keth_anchor_stackframe { 4088 struct pf_keth_ruleset *rs; 4089 struct pf_keth_rule *r; /* XXX: + match bit */ 4090 struct pf_keth_anchor *child; 4091 }; 4092 4093 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4094 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4095 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4096 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4097 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4098 } while (0) 4099 4100 void 4101 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4102 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4103 struct pf_keth_rule **a, int *match) 4104 { 4105 struct pf_keth_anchor_stackframe *f; 4106 4107 NET_EPOCH_ASSERT(); 4108 4109 if (match) 4110 *match = 0; 4111 if (*depth >= PF_ANCHOR_STACKSIZE) { 4112 printf("%s: anchor stack overflow on %s\n", 4113 __func__, (*r)->anchor->name); 4114 *r = TAILQ_NEXT(*r, entries); 4115 return; 4116 } else if (*depth == 0 && a != NULL) 4117 *a = *r; 4118 f = stack + (*depth)++; 4119 f->rs = *rs; 4120 f->r = *r; 4121 if ((*r)->anchor_wildcard) { 4122 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4123 4124 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4125 *r = NULL; 4126 return; 4127 } 4128 *rs = &f->child->ruleset; 4129 } else { 4130 f->child = NULL; 4131 *rs = &(*r)->anchor->ruleset; 4132 } 4133 *r = TAILQ_FIRST((*rs)->active.rules); 4134 } 4135 4136 int 4137 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4138 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4139 struct pf_keth_rule **a, int *match) 4140 { 4141 struct pf_keth_anchor_stackframe *f; 4142 struct pf_keth_rule *fr; 4143 int quick = 0; 4144 4145 NET_EPOCH_ASSERT(); 4146 4147 do { 4148 if (*depth <= 0) 4149 break; 4150 f = stack + *depth - 1; 4151 fr = PF_ETH_ANCHOR_RULE(f); 4152 if (f->child != NULL) { 4153 /* 4154 * This block traverses through 4155 * a wildcard anchor. 4156 */ 4157 if (match != NULL && *match) { 4158 /* 4159 * If any of "*" matched, then 4160 * "foo/ *" matched, mark frame 4161 * appropriately. 4162 */ 4163 PF_ETH_ANCHOR_SET_MATCH(f); 4164 *match = 0; 4165 } 4166 f->child = RB_NEXT(pf_keth_anchor_node, 4167 &fr->anchor->children, f->child); 4168 if (f->child != NULL) { 4169 *rs = &f->child->ruleset; 4170 *r = TAILQ_FIRST((*rs)->active.rules); 4171 if (*r == NULL) 4172 continue; 4173 else 4174 break; 4175 } 4176 } 4177 (*depth)--; 4178 if (*depth == 0 && a != NULL) 4179 *a = NULL; 4180 *rs = f->rs; 4181 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4182 quick = fr->quick; 4183 *r = TAILQ_NEXT(fr, entries); 4184 } while (*r == NULL); 4185 4186 return (quick); 4187 } 4188 4189 #ifdef INET6 4190 void 4191 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4192 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4193 { 4194 switch (af) { 4195 #ifdef INET 4196 case AF_INET: 4197 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4198 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4199 break; 4200 #endif /* INET */ 4201 case AF_INET6: 4202 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4203 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4204 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4205 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4206 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4207 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4208 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4209 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4210 break; 4211 } 4212 } 4213 4214 void 4215 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4216 { 4217 switch (af) { 4218 #ifdef INET 4219 case AF_INET: 4220 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4221 break; 4222 #endif /* INET */ 4223 case AF_INET6: 4224 if (addr->addr32[3] == 0xffffffff) { 4225 addr->addr32[3] = 0; 4226 if (addr->addr32[2] == 0xffffffff) { 4227 addr->addr32[2] = 0; 4228 if (addr->addr32[1] == 0xffffffff) { 4229 addr->addr32[1] = 0; 4230 addr->addr32[0] = 4231 htonl(ntohl(addr->addr32[0]) + 1); 4232 } else 4233 addr->addr32[1] = 4234 htonl(ntohl(addr->addr32[1]) + 1); 4235 } else 4236 addr->addr32[2] = 4237 htonl(ntohl(addr->addr32[2]) + 1); 4238 } else 4239 addr->addr32[3] = 4240 htonl(ntohl(addr->addr32[3]) + 1); 4241 break; 4242 } 4243 } 4244 #endif /* INET6 */ 4245 4246 void 4247 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4248 { 4249 /* 4250 * Modern rules use the same flags in rules as they do in states. 4251 */ 4252 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4253 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4254 4255 /* 4256 * Old-style scrub rules have different flags which need to be translated. 4257 */ 4258 if (r->rule_flag & PFRULE_RANDOMID) 4259 a->flags |= PFSTATE_RANDOMID; 4260 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4261 a->flags |= PFSTATE_SETTOS; 4262 a->set_tos = r->set_tos; 4263 } 4264 4265 if (r->qid) 4266 a->qid = r->qid; 4267 if (r->pqid) 4268 a->pqid = r->pqid; 4269 if (r->rtableid >= 0) 4270 a->rtableid = r->rtableid; 4271 a->log |= r->log; 4272 if (r->min_ttl) 4273 a->min_ttl = r->min_ttl; 4274 if (r->max_mss) 4275 a->max_mss = r->max_mss; 4276 if (r->dnpipe) 4277 a->dnpipe = r->dnpipe; 4278 if (r->dnrpipe) 4279 a->dnrpipe = r->dnrpipe; 4280 if (r->dnpipe || r->dnrpipe) { 4281 if (r->free_flags & PFRULE_DN_IS_PIPE) 4282 a->flags |= PFSTATE_DN_IS_PIPE; 4283 else 4284 a->flags &= ~PFSTATE_DN_IS_PIPE; 4285 } 4286 if (r->scrub_flags & PFSTATE_SETPRIO) { 4287 a->set_prio[0] = r->set_prio[0]; 4288 a->set_prio[1] = r->set_prio[1]; 4289 } 4290 } 4291 4292 int 4293 pf_socket_lookup(struct pf_pdesc *pd) 4294 { 4295 struct pf_addr *saddr, *daddr; 4296 u_int16_t sport, dport; 4297 struct inpcbinfo *pi; 4298 struct inpcb *inp; 4299 4300 pd->lookup.uid = UID_MAX; 4301 pd->lookup.gid = GID_MAX; 4302 4303 switch (pd->proto) { 4304 case IPPROTO_TCP: 4305 sport = pd->hdr.tcp.th_sport; 4306 dport = pd->hdr.tcp.th_dport; 4307 pi = &V_tcbinfo; 4308 break; 4309 case IPPROTO_UDP: 4310 sport = pd->hdr.udp.uh_sport; 4311 dport = pd->hdr.udp.uh_dport; 4312 pi = &V_udbinfo; 4313 break; 4314 default: 4315 return (-1); 4316 } 4317 if (pd->dir == PF_IN) { 4318 saddr = pd->src; 4319 daddr = pd->dst; 4320 } else { 4321 u_int16_t p; 4322 4323 p = sport; 4324 sport = dport; 4325 dport = p; 4326 saddr = pd->dst; 4327 daddr = pd->src; 4328 } 4329 switch (pd->af) { 4330 #ifdef INET 4331 case AF_INET: 4332 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4333 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4334 if (inp == NULL) { 4335 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4336 daddr->v4, dport, INPLOOKUP_WILDCARD | 4337 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4338 if (inp == NULL) 4339 return (-1); 4340 } 4341 break; 4342 #endif /* INET */ 4343 #ifdef INET6 4344 case AF_INET6: 4345 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4346 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4347 if (inp == NULL) { 4348 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4349 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4350 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4351 if (inp == NULL) 4352 return (-1); 4353 } 4354 break; 4355 #endif /* INET6 */ 4356 } 4357 INP_RLOCK_ASSERT(inp); 4358 pd->lookup.uid = inp->inp_cred->cr_uid; 4359 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4360 INP_RUNLOCK(inp); 4361 4362 return (1); 4363 } 4364 4365 u_int8_t 4366 pf_get_wscale(struct pf_pdesc *pd) 4367 { 4368 struct tcphdr *th = &pd->hdr.tcp; 4369 int hlen; 4370 u_int8_t hdr[60]; 4371 u_int8_t *opt, optlen; 4372 u_int8_t wscale = 0; 4373 4374 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4375 if (hlen <= sizeof(struct tcphdr)) 4376 return (0); 4377 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4378 return (0); 4379 opt = hdr + sizeof(struct tcphdr); 4380 hlen -= sizeof(struct tcphdr); 4381 while (hlen >= 3) { 4382 switch (*opt) { 4383 case TCPOPT_EOL: 4384 case TCPOPT_NOP: 4385 ++opt; 4386 --hlen; 4387 break; 4388 case TCPOPT_WINDOW: 4389 wscale = opt[2]; 4390 if (wscale > TCP_MAX_WINSHIFT) 4391 wscale = TCP_MAX_WINSHIFT; 4392 wscale |= PF_WSCALE_FLAG; 4393 /* FALLTHROUGH */ 4394 default: 4395 optlen = opt[1]; 4396 if (optlen < 2) 4397 optlen = 2; 4398 hlen -= optlen; 4399 opt += optlen; 4400 break; 4401 } 4402 } 4403 return (wscale); 4404 } 4405 4406 u_int16_t 4407 pf_get_mss(struct pf_pdesc *pd) 4408 { 4409 struct tcphdr *th = &pd->hdr.tcp; 4410 int hlen; 4411 u_int8_t hdr[60]; 4412 u_int8_t *opt, optlen; 4413 u_int16_t mss = V_tcp_mssdflt; 4414 4415 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4416 if (hlen <= sizeof(struct tcphdr)) 4417 return (0); 4418 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4419 return (0); 4420 opt = hdr + sizeof(struct tcphdr); 4421 hlen -= sizeof(struct tcphdr); 4422 while (hlen >= TCPOLEN_MAXSEG) { 4423 switch (*opt) { 4424 case TCPOPT_EOL: 4425 case TCPOPT_NOP: 4426 ++opt; 4427 --hlen; 4428 break; 4429 case TCPOPT_MAXSEG: 4430 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 4431 NTOHS(mss); 4432 /* FALLTHROUGH */ 4433 default: 4434 optlen = opt[1]; 4435 if (optlen < 2) 4436 optlen = 2; 4437 hlen -= optlen; 4438 opt += optlen; 4439 break; 4440 } 4441 } 4442 return (mss); 4443 } 4444 4445 static u_int16_t 4446 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 4447 { 4448 struct nhop_object *nh; 4449 #ifdef INET6 4450 struct in6_addr dst6; 4451 uint32_t scopeid; 4452 #endif /* INET6 */ 4453 int hlen = 0; 4454 uint16_t mss = 0; 4455 4456 NET_EPOCH_ASSERT(); 4457 4458 switch (af) { 4459 #ifdef INET 4460 case AF_INET: 4461 hlen = sizeof(struct ip); 4462 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 4463 if (nh != NULL) 4464 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4465 break; 4466 #endif /* INET */ 4467 #ifdef INET6 4468 case AF_INET6: 4469 hlen = sizeof(struct ip6_hdr); 4470 in6_splitscope(&addr->v6, &dst6, &scopeid); 4471 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4472 if (nh != NULL) 4473 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4474 break; 4475 #endif /* INET6 */ 4476 } 4477 4478 mss = max(V_tcp_mssdflt, mss); 4479 mss = min(mss, offer); 4480 mss = max(mss, 64); /* sanity - at least max opt space */ 4481 return (mss); 4482 } 4483 4484 static u_int32_t 4485 pf_tcp_iss(struct pf_pdesc *pd) 4486 { 4487 MD5_CTX ctx; 4488 u_int32_t digest[4]; 4489 4490 if (V_pf_tcp_secret_init == 0) { 4491 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4492 MD5Init(&V_pf_tcp_secret_ctx); 4493 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4494 sizeof(V_pf_tcp_secret)); 4495 V_pf_tcp_secret_init = 1; 4496 } 4497 4498 ctx = V_pf_tcp_secret_ctx; 4499 4500 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4501 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4502 switch (pd->af) { 4503 case AF_INET6: 4504 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4505 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4506 break; 4507 case AF_INET: 4508 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4509 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4510 break; 4511 } 4512 MD5Final((u_char *)digest, &ctx); 4513 V_pf_tcp_iss_off += 4096; 4514 #define ISN_RANDOM_INCREMENT (4096 - 1) 4515 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4516 V_pf_tcp_iss_off); 4517 #undef ISN_RANDOM_INCREMENT 4518 } 4519 4520 static bool 4521 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4522 { 4523 bool match = true; 4524 4525 /* Always matches if not set */ 4526 if (! r->isset) 4527 return (!r->neg); 4528 4529 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4530 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4531 match = false; 4532 break; 4533 } 4534 } 4535 4536 return (match ^ r->neg); 4537 } 4538 4539 static int 4540 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4541 { 4542 if (*tag == -1) 4543 *tag = mtag; 4544 4545 return ((!r->match_tag_not && r->match_tag == *tag) || 4546 (r->match_tag_not && r->match_tag != *tag)); 4547 } 4548 4549 static void 4550 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4551 { 4552 /* If we don't have the interface drop the packet. */ 4553 if (ifp == NULL) { 4554 m_freem(m); 4555 return; 4556 } 4557 4558 switch (ifp->if_type) { 4559 case IFT_ETHER: 4560 case IFT_XETHER: 4561 case IFT_L2VLAN: 4562 case IFT_BRIDGE: 4563 case IFT_IEEE8023ADLAG: 4564 break; 4565 default: 4566 m_freem(m); 4567 return; 4568 } 4569 4570 ifp->if_transmit(ifp, m); 4571 } 4572 4573 static int 4574 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4575 { 4576 #ifdef INET 4577 struct ip ip; 4578 #endif 4579 #ifdef INET6 4580 struct ip6_hdr ip6; 4581 #endif 4582 struct mbuf *m = *m0; 4583 struct ether_header *e; 4584 struct pf_keth_rule *r, *rm, *a = NULL; 4585 struct pf_keth_ruleset *ruleset = NULL; 4586 struct pf_mtag *mtag; 4587 struct pf_keth_ruleq *rules; 4588 struct pf_addr *src = NULL, *dst = NULL; 4589 struct pfi_kkif *bridge_to; 4590 sa_family_t af = 0; 4591 uint16_t proto; 4592 int asd = 0, match = 0; 4593 int tag = -1; 4594 uint8_t action; 4595 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4596 4597 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4598 NET_EPOCH_ASSERT(); 4599 4600 PF_RULES_RLOCK_TRACKER; 4601 4602 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4603 4604 mtag = pf_find_mtag(m); 4605 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4606 /* Dummynet re-injects packets after they've 4607 * completed their delay. We've already 4608 * processed them, so pass unconditionally. */ 4609 4610 /* But only once. We may see the packet multiple times (e.g. 4611 * PFIL_IN/PFIL_OUT). */ 4612 pf_dummynet_flag_remove(m, mtag); 4613 4614 return (PF_PASS); 4615 } 4616 4617 ruleset = V_pf_keth; 4618 rules = ck_pr_load_ptr(&ruleset->active.rules); 4619 r = TAILQ_FIRST(rules); 4620 rm = NULL; 4621 4622 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 4623 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 4624 DPFPRINTF(PF_DEBUG_URGENT, 4625 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 4626 ", pullup failed\n")); 4627 return (PF_DROP); 4628 } 4629 e = mtod(m, struct ether_header *); 4630 proto = ntohs(e->ether_type); 4631 4632 switch (proto) { 4633 #ifdef INET 4634 case ETHERTYPE_IP: { 4635 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4636 sizeof(ip))) 4637 return (PF_DROP); 4638 4639 af = AF_INET; 4640 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4641 (caddr_t)&ip); 4642 src = (struct pf_addr *)&ip.ip_src; 4643 dst = (struct pf_addr *)&ip.ip_dst; 4644 break; 4645 } 4646 #endif /* INET */ 4647 #ifdef INET6 4648 case ETHERTYPE_IPV6: { 4649 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4650 sizeof(ip6))) 4651 return (PF_DROP); 4652 4653 af = AF_INET6; 4654 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4655 (caddr_t)&ip6); 4656 src = (struct pf_addr *)&ip6.ip6_src; 4657 dst = (struct pf_addr *)&ip6.ip6_dst; 4658 break; 4659 } 4660 #endif /* INET6 */ 4661 } 4662 4663 PF_RULES_RLOCK(); 4664 4665 while (r != NULL) { 4666 counter_u64_add(r->evaluations, 1); 4667 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4668 4669 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4670 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4671 "kif"); 4672 r = r->skip[PFE_SKIP_IFP].ptr; 4673 } 4674 else if (r->direction && r->direction != dir) { 4675 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4676 "dir"); 4677 r = r->skip[PFE_SKIP_DIR].ptr; 4678 } 4679 else if (r->proto && r->proto != proto) { 4680 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4681 "proto"); 4682 r = r->skip[PFE_SKIP_PROTO].ptr; 4683 } 4684 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4685 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4686 "src"); 4687 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4688 } 4689 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4690 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4691 "dst"); 4692 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4693 } 4694 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4695 r->ipsrc.neg, kif, M_GETFIB(m))) { 4696 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4697 "ip_src"); 4698 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4699 } 4700 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4701 r->ipdst.neg, kif, M_GETFIB(m))) { 4702 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4703 "ip_dst"); 4704 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4705 } 4706 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4707 mtag ? mtag->tag : 0)) { 4708 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4709 "match_tag"); 4710 r = TAILQ_NEXT(r, entries); 4711 } 4712 else { 4713 if (r->tag) 4714 tag = r->tag; 4715 if (r->anchor == NULL) { 4716 /* Rule matches */ 4717 rm = r; 4718 4719 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4720 4721 if (r->quick) 4722 break; 4723 4724 r = TAILQ_NEXT(r, entries); 4725 } else { 4726 pf_step_into_keth_anchor(anchor_stack, &asd, 4727 &ruleset, &r, &a, &match); 4728 } 4729 } 4730 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4731 &ruleset, &r, &a, &match)) 4732 break; 4733 } 4734 4735 r = rm; 4736 4737 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4738 4739 /* Default to pass. */ 4740 if (r == NULL) { 4741 PF_RULES_RUNLOCK(); 4742 return (PF_PASS); 4743 } 4744 4745 /* Execute action. */ 4746 counter_u64_add(r->packets[dir == PF_OUT], 1); 4747 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4748 pf_update_timestamp(r); 4749 4750 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4751 if (r->action == PF_DROP) { 4752 PF_RULES_RUNLOCK(); 4753 return (PF_DROP); 4754 } 4755 4756 if (tag > 0) { 4757 if (mtag == NULL) 4758 mtag = pf_get_mtag(m); 4759 if (mtag == NULL) { 4760 PF_RULES_RUNLOCK(); 4761 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4762 return (PF_DROP); 4763 } 4764 mtag->tag = tag; 4765 } 4766 4767 if (r->qid != 0) { 4768 if (mtag == NULL) 4769 mtag = pf_get_mtag(m); 4770 if (mtag == NULL) { 4771 PF_RULES_RUNLOCK(); 4772 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4773 return (PF_DROP); 4774 } 4775 mtag->qid = r->qid; 4776 } 4777 4778 action = r->action; 4779 bridge_to = r->bridge_to; 4780 4781 /* Dummynet */ 4782 if (r->dnpipe) { 4783 struct ip_fw_args dnflow; 4784 4785 /* Drop packet if dummynet is not loaded. */ 4786 if (ip_dn_io_ptr == NULL) { 4787 PF_RULES_RUNLOCK(); 4788 m_freem(m); 4789 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4790 return (PF_DROP); 4791 } 4792 if (mtag == NULL) 4793 mtag = pf_get_mtag(m); 4794 if (mtag == NULL) { 4795 PF_RULES_RUNLOCK(); 4796 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4797 return (PF_DROP); 4798 } 4799 4800 bzero(&dnflow, sizeof(dnflow)); 4801 4802 /* We don't have port numbers here, so we set 0. That means 4803 * that we'll be somewhat limited in distinguishing flows (i.e. 4804 * only based on IP addresses, not based on port numbers), but 4805 * it's better than nothing. */ 4806 dnflow.f_id.dst_port = 0; 4807 dnflow.f_id.src_port = 0; 4808 dnflow.f_id.proto = 0; 4809 4810 dnflow.rule.info = r->dnpipe; 4811 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4812 if (r->dnflags & PFRULE_DN_IS_PIPE) 4813 dnflow.rule.info |= IPFW_IS_PIPE; 4814 4815 dnflow.f_id.extra = dnflow.rule.info; 4816 4817 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4818 dnflow.flags |= IPFW_ARGS_ETHER; 4819 dnflow.ifp = kif->pfik_ifp; 4820 4821 switch (af) { 4822 case AF_INET: 4823 dnflow.f_id.addr_type = 4; 4824 dnflow.f_id.src_ip = src->v4.s_addr; 4825 dnflow.f_id.dst_ip = dst->v4.s_addr; 4826 break; 4827 case AF_INET6: 4828 dnflow.flags |= IPFW_ARGS_IP6; 4829 dnflow.f_id.addr_type = 6; 4830 dnflow.f_id.src_ip6 = src->v6; 4831 dnflow.f_id.dst_ip6 = dst->v6; 4832 break; 4833 } 4834 4835 PF_RULES_RUNLOCK(); 4836 4837 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4838 ip_dn_io_ptr(m0, &dnflow); 4839 if (*m0 != NULL) 4840 pf_dummynet_flag_remove(m, mtag); 4841 } else { 4842 PF_RULES_RUNLOCK(); 4843 } 4844 4845 if (action == PF_PASS && bridge_to) { 4846 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4847 *m0 = NULL; /* We've eaten the packet. */ 4848 } 4849 4850 return (action); 4851 } 4852 4853 #define PF_TEST_ATTRIB(t, a)\ 4854 do { \ 4855 if (t) { \ 4856 r = a; \ 4857 goto nextrule; \ 4858 } \ 4859 } while (0) 4860 4861 static int 4862 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, 4863 struct pf_pdesc *pd, struct pf_krule **am, 4864 struct pf_kruleset **rsm, struct inpcb *inp) 4865 { 4866 struct pf_krule *nr = NULL; 4867 struct pf_addr * const saddr = pd->src; 4868 struct pf_addr * const daddr = pd->dst; 4869 struct pf_krule *r, *a = NULL; 4870 struct pf_kruleset *ruleset = NULL; 4871 struct pf_krule_slist match_rules; 4872 struct pf_krule_item *ri; 4873 struct pf_ksrc_node *nsn = NULL; 4874 struct tcphdr *th = &pd->hdr.tcp; 4875 struct pf_state_key *sk = NULL, *nk = NULL; 4876 u_short reason, transerror; 4877 int rewrite = 0; 4878 int tag = -1; 4879 int asd = 0; 4880 int match = 0; 4881 int state_icmp = 0, icmp_dir, multi; 4882 u_int16_t sport = 0, dport = 0, virtual_type, virtual_id; 4883 u_int16_t bproto_sum = 0, bip_sum = 0; 4884 u_int8_t icmptype = 0, icmpcode = 0; 4885 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4886 struct pf_udp_mapping *udp_mapping = NULL; 4887 4888 PF_RULES_RASSERT(); 4889 4890 SLIST_INIT(&match_rules); 4891 4892 if (inp != NULL) { 4893 INP_LOCK_ASSERT(inp); 4894 pd->lookup.uid = inp->inp_cred->cr_uid; 4895 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4896 pd->lookup.done = 1; 4897 } 4898 4899 switch (pd->virtual_proto) { 4900 case IPPROTO_TCP: 4901 sport = th->th_sport; 4902 dport = th->th_dport; 4903 break; 4904 case IPPROTO_UDP: 4905 sport = pd->hdr.udp.uh_sport; 4906 dport = pd->hdr.udp.uh_dport; 4907 break; 4908 case IPPROTO_SCTP: 4909 sport = pd->hdr.sctp.src_port; 4910 dport = pd->hdr.sctp.dest_port; 4911 break; 4912 #ifdef INET 4913 case IPPROTO_ICMP: 4914 MPASS(pd->af == AF_INET); 4915 icmptype = pd->hdr.icmp.icmp_type; 4916 icmpcode = pd->hdr.icmp.icmp_code; 4917 state_icmp = pf_icmp_mapping(pd, icmptype, 4918 &icmp_dir, &multi, &virtual_id, &virtual_type); 4919 if (icmp_dir == PF_IN) { 4920 sport = virtual_id; 4921 dport = virtual_type; 4922 } else { 4923 sport = virtual_type; 4924 dport = virtual_id; 4925 } 4926 break; 4927 #endif /* INET */ 4928 #ifdef INET6 4929 case IPPROTO_ICMPV6: 4930 MPASS(pd->af == AF_INET6); 4931 icmptype = pd->hdr.icmp6.icmp6_type; 4932 icmpcode = pd->hdr.icmp6.icmp6_code; 4933 state_icmp = pf_icmp_mapping(pd, icmptype, 4934 &icmp_dir, &multi, &virtual_id, &virtual_type); 4935 if (icmp_dir == PF_IN) { 4936 sport = virtual_id; 4937 dport = virtual_type; 4938 } else { 4939 sport = virtual_type; 4940 dport = virtual_id; 4941 } 4942 4943 break; 4944 #endif /* INET6 */ 4945 default: 4946 sport = dport = 0; 4947 break; 4948 } 4949 4950 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4951 4952 /* check packet for BINAT/NAT/RDR */ 4953 transerror = pf_get_translation(pd, pd->off, &nsn, &sk, 4954 &nk, saddr, daddr, sport, dport, anchor_stack, &nr, &udp_mapping); 4955 switch (transerror) { 4956 default: 4957 /* A translation error occurred. */ 4958 REASON_SET(&reason, transerror); 4959 goto cleanup; 4960 case PFRES_MAX: 4961 /* No match. */ 4962 break; 4963 case PFRES_MATCH: 4964 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4965 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4966 4967 if (nr->log) { 4968 PFLOG_PACKET(PF_PASS, PFRES_MATCH, nr, a, 4969 ruleset, pd, 1); 4970 } 4971 4972 if (pd->ip_sum) 4973 bip_sum = *pd->ip_sum; 4974 4975 switch (pd->proto) { 4976 case IPPROTO_TCP: 4977 bproto_sum = th->th_sum; 4978 pd->proto_sum = &th->th_sum; 4979 4980 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], pd->af) || 4981 nk->port[pd->sidx] != sport) { 4982 pf_change_ap(pd->m, saddr, &th->th_sport, pd->ip_sum, 4983 &th->th_sum, &nk->addr[pd->sidx], 4984 nk->port[pd->sidx], 0, pd->af); 4985 pd->sport = &th->th_sport; 4986 sport = th->th_sport; 4987 } 4988 4989 if (PF_ANEQ(daddr, &nk->addr[pd->didx], pd->af) || 4990 nk->port[pd->didx] != dport) { 4991 pf_change_ap(pd->m, daddr, &th->th_dport, pd->ip_sum, 4992 &th->th_sum, &nk->addr[pd->didx], 4993 nk->port[pd->didx], 0, pd->af); 4994 dport = th->th_dport; 4995 pd->dport = &th->th_dport; 4996 } 4997 rewrite++; 4998 break; 4999 case IPPROTO_UDP: 5000 bproto_sum = pd->hdr.udp.uh_sum; 5001 pd->proto_sum = &pd->hdr.udp.uh_sum; 5002 5003 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], pd->af) || 5004 nk->port[pd->sidx] != sport) { 5005 pf_change_ap(pd->m, saddr, &pd->hdr.udp.uh_sport, 5006 pd->ip_sum, &pd->hdr.udp.uh_sum, 5007 &nk->addr[pd->sidx], 5008 nk->port[pd->sidx], 1, pd->af); 5009 sport = pd->hdr.udp.uh_sport; 5010 pd->sport = &pd->hdr.udp.uh_sport; 5011 } 5012 5013 if (PF_ANEQ(daddr, &nk->addr[pd->didx], pd->af) || 5014 nk->port[pd->didx] != dport) { 5015 pf_change_ap(pd->m, daddr, &pd->hdr.udp.uh_dport, 5016 pd->ip_sum, &pd->hdr.udp.uh_sum, 5017 &nk->addr[pd->didx], 5018 nk->port[pd->didx], 1, pd->af); 5019 dport = pd->hdr.udp.uh_dport; 5020 pd->dport = &pd->hdr.udp.uh_dport; 5021 } 5022 rewrite++; 5023 break; 5024 case IPPROTO_SCTP: { 5025 uint16_t checksum = 0; 5026 5027 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], pd->af) || 5028 nk->port[pd->sidx] != sport) { 5029 pf_change_ap(pd->m, saddr, &pd->hdr.sctp.src_port, 5030 pd->ip_sum, &checksum, 5031 &nk->addr[pd->sidx], 5032 nk->port[pd->sidx], 1, pd->af); 5033 } 5034 if (PF_ANEQ(daddr, &nk->addr[pd->didx], pd->af) || 5035 nk->port[pd->didx] != dport) { 5036 pf_change_ap(pd->m, daddr, &pd->hdr.sctp.dest_port, 5037 pd->ip_sum, &checksum, 5038 &nk->addr[pd->didx], 5039 nk->port[pd->didx], 1, pd->af); 5040 } 5041 break; 5042 } 5043 #ifdef INET 5044 case IPPROTO_ICMP: 5045 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 5046 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 5047 nk->addr[pd->sidx].v4.s_addr, 0); 5048 5049 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 5050 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 5051 nk->addr[pd->didx].v4.s_addr, 0); 5052 5053 if (virtual_type == htons(ICMP_ECHO) && 5054 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5055 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5056 pd->hdr.icmp.icmp_cksum, sport, 5057 nk->port[pd->sidx], 0); 5058 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5059 pd->sport = &pd->hdr.icmp.icmp_id; 5060 } 5061 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5062 break; 5063 #endif /* INET */ 5064 #ifdef INET6 5065 case IPPROTO_ICMPV6: 5066 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 5067 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 5068 &nk->addr[pd->sidx], 0); 5069 5070 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 5071 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 5072 &nk->addr[pd->didx], 0); 5073 rewrite++; 5074 break; 5075 #endif /* INET */ 5076 default: 5077 switch (pd->af) { 5078 #ifdef INET 5079 case AF_INET: 5080 if (PF_ANEQ(saddr, 5081 &nk->addr[pd->sidx], AF_INET)) 5082 pf_change_a(&saddr->v4.s_addr, 5083 pd->ip_sum, 5084 nk->addr[pd->sidx].v4.s_addr, 0); 5085 5086 if (PF_ANEQ(daddr, 5087 &nk->addr[pd->didx], AF_INET)) 5088 pf_change_a(&daddr->v4.s_addr, 5089 pd->ip_sum, 5090 nk->addr[pd->didx].v4.s_addr, 0); 5091 break; 5092 #endif /* INET */ 5093 #ifdef INET6 5094 case AF_INET6: 5095 if (PF_ANEQ(saddr, 5096 &nk->addr[pd->sidx], AF_INET6)) 5097 PF_ACPY(saddr, &nk->addr[pd->sidx], pd->af); 5098 5099 if (PF_ANEQ(daddr, 5100 &nk->addr[pd->didx], AF_INET6)) 5101 PF_ACPY(daddr, &nk->addr[pd->didx], pd->af); 5102 break; 5103 #endif /* INET */ 5104 } 5105 break; 5106 } 5107 if (nr->natpass) 5108 r = NULL; 5109 } 5110 5111 while (r != NULL) { 5112 pf_counter_u64_add(&r->evaluations, 1); 5113 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot, 5114 r->skip[PF_SKIP_IFP]); 5115 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5116 r->skip[PF_SKIP_DIR]); 5117 PF_TEST_ATTRIB(r->af && r->af != pd->af, 5118 r->skip[PF_SKIP_AF]); 5119 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5120 r->skip[PF_SKIP_PROTO]); 5121 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, saddr, pd->af, 5122 r->src.neg, pd->kif, M_GETFIB(pd->m)), 5123 r->skip[PF_SKIP_SRC_ADDR]); 5124 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, daddr, pd->af, 5125 r->dst.neg, NULL, M_GETFIB(pd->m)), 5126 r->skip[PF_SKIP_DST_ADDR]); 5127 switch (pd->virtual_proto) { 5128 case PF_VPROTO_FRAGMENT: 5129 /* tcp/udp only. port_op always 0 in other cases */ 5130 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5131 TAILQ_NEXT(r, entries)); 5132 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5133 TAILQ_NEXT(r, entries)); 5134 /* icmp only. type/code always 0 in other cases */ 5135 PF_TEST_ATTRIB((r->type || r->code), 5136 TAILQ_NEXT(r, entries)); 5137 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5138 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5139 TAILQ_NEXT(r, entries)); 5140 break; 5141 5142 case IPPROTO_TCP: 5143 PF_TEST_ATTRIB((r->flagset & th->th_flags) != r->flags, 5144 TAILQ_NEXT(r, entries)); 5145 /* FALLTHROUGH */ 5146 case IPPROTO_SCTP: 5147 case IPPROTO_UDP: 5148 /* tcp/udp only. port_op always 0 in other cases */ 5149 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5150 r->src.port[0], r->src.port[1], sport), 5151 r->skip[PF_SKIP_SRC_PORT]); 5152 /* tcp/udp only. port_op always 0 in other cases */ 5153 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5154 r->dst.port[0], r->dst.port[1], dport), 5155 r->skip[PF_SKIP_DST_PORT]); 5156 /* tcp/udp only. uid.op always 0 in other cases */ 5157 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5158 pf_socket_lookup(pd), 1)) && 5159 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5160 pd->lookup.uid), 5161 TAILQ_NEXT(r, entries)); 5162 /* tcp/udp only. gid.op always 0 in other cases */ 5163 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5164 pf_socket_lookup(pd), 1)) && 5165 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5166 pd->lookup.gid), 5167 TAILQ_NEXT(r, entries)); 5168 break; 5169 5170 case IPPROTO_ICMP: 5171 case IPPROTO_ICMPV6: 5172 /* icmp only. type always 0 in other cases */ 5173 PF_TEST_ATTRIB(r->type && r->type != icmptype + 1, 5174 TAILQ_NEXT(r, entries)); 5175 /* icmp only. type always 0 in other cases */ 5176 PF_TEST_ATTRIB(r->code && r->code != icmpcode + 1, 5177 TAILQ_NEXT(r, entries)); 5178 break; 5179 5180 default: 5181 break; 5182 } 5183 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5184 TAILQ_NEXT(r, entries)); 5185 PF_TEST_ATTRIB(r->prio && 5186 !pf_match_ieee8021q_pcp(r->prio, pd->m), 5187 TAILQ_NEXT(r, entries)); 5188 PF_TEST_ATTRIB(r->prob && 5189 r->prob <= arc4random(), 5190 TAILQ_NEXT(r, entries)); 5191 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r, &tag, 5192 pd->pf_mtag ? pd->pf_mtag->tag : 0), 5193 TAILQ_NEXT(r, entries)); 5194 PF_TEST_ATTRIB(r->rcv_kif && !pf_match_rcvif(pd->m, r), 5195 TAILQ_NEXT(r, entries)); 5196 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5197 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5198 TAILQ_NEXT(r, entries)); 5199 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5200 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5201 pf_osfp_fingerprint(pd, th), 5202 r->os_fingerprint)), 5203 TAILQ_NEXT(r, entries)); 5204 /* FALLTHROUGH */ 5205 if (r->tag) 5206 tag = r->tag; 5207 if (r->anchor == NULL) { 5208 if (r->action == PF_MATCH) { 5209 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5210 if (ri == NULL) { 5211 REASON_SET(&reason, PFRES_MEMORY); 5212 goto cleanup; 5213 } 5214 ri->r = r; 5215 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5216 pf_counter_u64_critical_enter(); 5217 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5218 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5219 pf_counter_u64_critical_exit(); 5220 pf_rule_to_actions(r, &pd->act); 5221 if (r->log || pd->act.log & PF_LOG_MATCHES) 5222 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5223 a, ruleset, pd, 1); 5224 } else { 5225 match = 1; 5226 *rm = r; 5227 *am = a; 5228 *rsm = ruleset; 5229 if (pd->act.log & PF_LOG_MATCHES) 5230 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5231 a, ruleset, pd, 1); 5232 } 5233 if ((*rm)->quick) 5234 break; 5235 r = TAILQ_NEXT(r, entries); 5236 } else 5237 pf_step_into_anchor(anchor_stack, &asd, 5238 &ruleset, PF_RULESET_FILTER, &r, &a, 5239 &match); 5240 nextrule: 5241 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5242 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5243 break; 5244 } 5245 r = *rm; 5246 a = *am; 5247 ruleset = *rsm; 5248 5249 REASON_SET(&reason, PFRES_MATCH); 5250 5251 /* apply actions for last matching pass/block rule */ 5252 pf_rule_to_actions(r, &pd->act); 5253 5254 if (r->log || pd->act.log & PF_LOG_MATCHES) { 5255 if (rewrite) 5256 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5257 PFLOG_PACKET(r->action, reason, r, a, ruleset, pd, 1); 5258 } 5259 5260 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5261 (r->action == PF_DROP) && 5262 ((r->rule_flag & PFRULE_RETURNRST) || 5263 (r->rule_flag & PFRULE_RETURNICMP) || 5264 (r->rule_flag & PFRULE_RETURN))) { 5265 pf_return(r, nr, pd, sk, th, bproto_sum, 5266 bip_sum, &reason, r->rtableid); 5267 } 5268 5269 if (r->action == PF_DROP) 5270 goto cleanup; 5271 5272 if (tag > 0 && pf_tag_packet(pd, tag)) { 5273 REASON_SET(&reason, PFRES_MEMORY); 5274 goto cleanup; 5275 } 5276 if (pd->act.rtableid >= 0) 5277 M_SETFIB(pd->m, pd->act.rtableid); 5278 5279 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5280 (!state_icmp && (r->keep_state || nr != NULL || 5281 (pd->flags & PFDESC_TCP_NORM)))) { 5282 int action; 5283 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, 5284 sport, dport, &rewrite, sm, tag, bproto_sum, bip_sum, 5285 &match_rules, udp_mapping); 5286 if (action != PF_PASS) { 5287 pf_udp_mapping_release(udp_mapping); 5288 if (action == PF_DROP && 5289 (r->rule_flag & PFRULE_RETURN)) 5290 pf_return(r, nr, pd, sk, th, 5291 bproto_sum, bip_sum, &reason, 5292 pd->act.rtableid); 5293 return (action); 5294 } 5295 } else { 5296 while ((ri = SLIST_FIRST(&match_rules))) { 5297 SLIST_REMOVE_HEAD(&match_rules, entry); 5298 free(ri, M_PF_RULE_ITEM); 5299 } 5300 5301 uma_zfree(V_pf_state_key_z, sk); 5302 uma_zfree(V_pf_state_key_z, nk); 5303 pf_udp_mapping_release(udp_mapping); 5304 } 5305 5306 /* copy back packet headers if we performed NAT operations */ 5307 if (rewrite) 5308 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5309 5310 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5311 pd->dir == PF_OUT && 5312 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) 5313 /* 5314 * We want the state created, but we dont 5315 * want to send this in case a partner 5316 * firewall has to know about it to allow 5317 * replies through it. 5318 */ 5319 return (PF_DEFER); 5320 5321 return (PF_PASS); 5322 5323 cleanup: 5324 while ((ri = SLIST_FIRST(&match_rules))) { 5325 SLIST_REMOVE_HEAD(&match_rules, entry); 5326 free(ri, M_PF_RULE_ITEM); 5327 } 5328 5329 uma_zfree(V_pf_state_key_z, sk); 5330 uma_zfree(V_pf_state_key_z, nk); 5331 pf_udp_mapping_release(udp_mapping); 5332 5333 return (PF_DROP); 5334 } 5335 5336 static int 5337 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 5338 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 5339 struct pf_state_key *sk, u_int16_t sport, 5340 u_int16_t dport, int *rewrite, struct pf_kstate **sm, 5341 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, 5342 struct pf_krule_slist *match_rules, struct pf_udp_mapping *udp_mapping) 5343 { 5344 struct pf_kstate *s = NULL; 5345 struct pf_ksrc_node *sn = NULL; 5346 struct tcphdr *th = &pd->hdr.tcp; 5347 u_int16_t mss = V_tcp_mssdflt; 5348 u_short reason, sn_reason; 5349 struct pf_krule_item *ri; 5350 5351 /* check maximums */ 5352 if (r->max_states && 5353 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 5354 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 5355 REASON_SET(&reason, PFRES_MAXSTATES); 5356 goto csfailed; 5357 } 5358 /* src node for filter rule */ 5359 if ((r->rule_flag & PFRULE_SRCTRACK || 5360 r->rpool.opts & PF_POOL_STICKYADDR) && 5361 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 5362 REASON_SET(&reason, sn_reason); 5363 goto csfailed; 5364 } 5365 /* src node for translation rule */ 5366 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 5367 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 5368 pd->af)) != 0 ) { 5369 REASON_SET(&reason, sn_reason); 5370 goto csfailed; 5371 } 5372 s = pf_alloc_state(M_NOWAIT); 5373 if (s == NULL) { 5374 REASON_SET(&reason, PFRES_MEMORY); 5375 goto csfailed; 5376 } 5377 s->rule = r; 5378 s->nat_rule = nr; 5379 s->anchor = a; 5380 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 5381 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 5382 5383 STATE_INC_COUNTERS(s); 5384 if (r->allow_opts) 5385 s->state_flags |= PFSTATE_ALLOWOPTS; 5386 if (r->rule_flag & PFRULE_STATESLOPPY) 5387 s->state_flags |= PFSTATE_SLOPPY; 5388 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 5389 s->state_flags |= PFSTATE_SCRUB_TCP; 5390 if ((r->rule_flag & PFRULE_PFLOW) || 5391 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 5392 s->state_flags |= PFSTATE_PFLOW; 5393 5394 s->act.log = pd->act.log & PF_LOG_ALL; 5395 s->sync_state = PFSYNC_S_NONE; 5396 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 5397 5398 if (nr != NULL) 5399 s->act.log |= nr->log & PF_LOG_ALL; 5400 switch (pd->proto) { 5401 case IPPROTO_TCP: 5402 s->src.seqlo = ntohl(th->th_seq); 5403 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 5404 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 5405 r->keep_state == PF_STATE_MODULATE) { 5406 /* Generate sequence number modulator */ 5407 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 5408 0) 5409 s->src.seqdiff = 1; 5410 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, 5411 htonl(s->src.seqlo + s->src.seqdiff), 0); 5412 *rewrite = 1; 5413 } else 5414 s->src.seqdiff = 0; 5415 if (th->th_flags & TH_SYN) { 5416 s->src.seqhi++; 5417 s->src.wscale = pf_get_wscale(pd); 5418 } 5419 s->src.max_win = MAX(ntohs(th->th_win), 1); 5420 if (s->src.wscale & PF_WSCALE_MASK) { 5421 /* Remove scale factor from initial window */ 5422 int win = s->src.max_win; 5423 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 5424 s->src.max_win = (win - 1) >> 5425 (s->src.wscale & PF_WSCALE_MASK); 5426 } 5427 if (th->th_flags & TH_FIN) 5428 s->src.seqhi++; 5429 s->dst.seqhi = 1; 5430 s->dst.max_win = 1; 5431 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 5432 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 5433 s->timeout = PFTM_TCP_FIRST_PACKET; 5434 atomic_add_32(&V_pf_status.states_halfopen, 1); 5435 break; 5436 case IPPROTO_UDP: 5437 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 5438 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 5439 s->timeout = PFTM_UDP_FIRST_PACKET; 5440 break; 5441 case IPPROTO_SCTP: 5442 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 5443 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 5444 s->timeout = PFTM_SCTP_FIRST_PACKET; 5445 break; 5446 case IPPROTO_ICMP: 5447 #ifdef INET6 5448 case IPPROTO_ICMPV6: 5449 #endif 5450 s->timeout = PFTM_ICMP_FIRST_PACKET; 5451 break; 5452 default: 5453 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 5454 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 5455 s->timeout = PFTM_OTHER_FIRST_PACKET; 5456 } 5457 5458 if (r->rt) { 5459 /* pf_map_addr increases the reason counters */ 5460 if ((reason = pf_map_addr_sn(pd->af, r, pd->src, &s->rt_addr, 5461 &s->rt_kif, NULL, &sn)) != 0) 5462 goto csfailed; 5463 s->rt = r->rt; 5464 } 5465 5466 s->creation = s->expire = pf_get_uptime(); 5467 5468 if (sn != NULL) 5469 s->src_node = sn; 5470 if (nsn != NULL) { 5471 /* XXX We only modify one side for now. */ 5472 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 5473 s->nat_src_node = nsn; 5474 } 5475 if (pd->proto == IPPROTO_TCP) { 5476 if (s->state_flags & PFSTATE_SCRUB_TCP && 5477 pf_normalize_tcp_init(pd, th, &s->src, &s->dst)) { 5478 REASON_SET(&reason, PFRES_MEMORY); 5479 goto csfailed; 5480 } 5481 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 5482 pf_normalize_tcp_stateful(pd, &reason, th, s, 5483 &s->src, &s->dst, rewrite)) { 5484 /* This really shouldn't happen!!! */ 5485 DPFPRINTF(PF_DEBUG_URGENT, 5486 ("pf_normalize_tcp_stateful failed on first " 5487 "pkt\n")); 5488 goto csfailed; 5489 } 5490 } else if (pd->proto == IPPROTO_SCTP) { 5491 if (pf_normalize_sctp_init(pd, &s->src, &s->dst)) 5492 goto csfailed; 5493 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 5494 goto csfailed; 5495 } 5496 s->direction = pd->dir; 5497 5498 /* 5499 * sk/nk could already been setup by pf_get_translation(). 5500 */ 5501 if (nr == NULL) { 5502 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 5503 __func__, nr, sk, nk)); 5504 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 5505 if (sk == NULL) 5506 goto csfailed; 5507 nk = sk; 5508 } else 5509 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 5510 __func__, nr, sk, nk)); 5511 5512 /* Swap sk/nk for PF_OUT. */ 5513 if (pf_state_insert(BOUND_IFACE(s, pd->kif), pd->kif, 5514 (pd->dir == PF_IN) ? sk : nk, 5515 (pd->dir == PF_IN) ? nk : sk, s)) { 5516 REASON_SET(&reason, PFRES_STATEINS); 5517 goto drop; 5518 } else 5519 *sm = s; 5520 5521 if (tag > 0) 5522 s->tag = tag; 5523 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 5524 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 5525 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 5526 /* undo NAT changes, if they have taken place */ 5527 if (nr != NULL) { 5528 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 5529 if (pd->dir == PF_OUT) 5530 skt = s->key[PF_SK_STACK]; 5531 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 5532 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 5533 if (pd->sport) 5534 *pd->sport = skt->port[pd->sidx]; 5535 if (pd->dport) 5536 *pd->dport = skt->port[pd->didx]; 5537 if (pd->proto_sum) 5538 *pd->proto_sum = bproto_sum; 5539 if (pd->ip_sum) 5540 *pd->ip_sum = bip_sum; 5541 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5542 } 5543 s->src.seqhi = htonl(arc4random()); 5544 /* Find mss option */ 5545 int rtid = M_GETFIB(pd->m); 5546 mss = pf_get_mss(pd); 5547 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5548 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5549 s->src.mss = mss; 5550 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5551 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5552 TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0, 5553 pd->act.rtableid); 5554 REASON_SET(&reason, PFRES_SYNPROXY); 5555 return (PF_SYNPROXY_DROP); 5556 } 5557 5558 s->udp_mapping = udp_mapping; 5559 5560 return (PF_PASS); 5561 5562 csfailed: 5563 while ((ri = SLIST_FIRST(match_rules))) { 5564 SLIST_REMOVE_HEAD(match_rules, entry); 5565 free(ri, M_PF_RULE_ITEM); 5566 } 5567 5568 uma_zfree(V_pf_state_key_z, sk); 5569 uma_zfree(V_pf_state_key_z, nk); 5570 5571 if (sn != NULL) { 5572 PF_SRC_NODE_LOCK(sn); 5573 if (--sn->states == 0 && sn->expire == 0) { 5574 pf_unlink_src_node(sn); 5575 uma_zfree(V_pf_sources_z, sn); 5576 counter_u64_add( 5577 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5578 } 5579 PF_SRC_NODE_UNLOCK(sn); 5580 } 5581 5582 if (nsn != sn && nsn != NULL) { 5583 PF_SRC_NODE_LOCK(nsn); 5584 if (--nsn->states == 0 && nsn->expire == 0) { 5585 pf_unlink_src_node(nsn); 5586 uma_zfree(V_pf_sources_z, nsn); 5587 counter_u64_add( 5588 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5589 } 5590 PF_SRC_NODE_UNLOCK(nsn); 5591 } 5592 5593 drop: 5594 if (s != NULL) { 5595 pf_src_tree_remove_state(s); 5596 s->timeout = PFTM_UNLINKED; 5597 STATE_DEC_COUNTERS(s); 5598 pf_free_state(s); 5599 } 5600 5601 return (PF_DROP); 5602 } 5603 5604 static int 5605 pf_tcp_track_full(struct pf_kstate **state, struct pf_pdesc *pd, 5606 u_short *reason, int *copyback) 5607 { 5608 struct tcphdr *th = &pd->hdr.tcp; 5609 struct pf_state_peer *src, *dst; 5610 u_int16_t win = ntohs(th->th_win); 5611 u_int32_t ack, end, data_end, seq, orig_seq; 5612 u_int8_t sws, dws, psrc, pdst; 5613 int ackskew; 5614 5615 if (pd->dir == (*state)->direction) { 5616 src = &(*state)->src; 5617 dst = &(*state)->dst; 5618 psrc = PF_PEER_SRC; 5619 pdst = PF_PEER_DST; 5620 } else { 5621 src = &(*state)->dst; 5622 dst = &(*state)->src; 5623 psrc = PF_PEER_DST; 5624 pdst = PF_PEER_SRC; 5625 } 5626 5627 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5628 sws = src->wscale & PF_WSCALE_MASK; 5629 dws = dst->wscale & PF_WSCALE_MASK; 5630 } else 5631 sws = dws = 0; 5632 5633 /* 5634 * Sequence tracking algorithm from Guido van Rooij's paper: 5635 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5636 * tcp_filtering.ps 5637 */ 5638 5639 orig_seq = seq = ntohl(th->th_seq); 5640 if (src->seqlo == 0) { 5641 /* First packet from this end. Set its state */ 5642 5643 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5644 src->scrub == NULL) { 5645 if (pf_normalize_tcp_init(pd, th, src, dst)) { 5646 REASON_SET(reason, PFRES_MEMORY); 5647 return (PF_DROP); 5648 } 5649 } 5650 5651 /* Deferred generation of sequence number modulator */ 5652 if (dst->seqdiff && !src->seqdiff) { 5653 /* use random iss for the TCP server */ 5654 while ((src->seqdiff = arc4random() - seq) == 0) 5655 ; 5656 ack = ntohl(th->th_ack) - dst->seqdiff; 5657 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 5658 src->seqdiff), 0); 5659 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5660 *copyback = 1; 5661 } else { 5662 ack = ntohl(th->th_ack); 5663 } 5664 5665 end = seq + pd->p_len; 5666 if (th->th_flags & TH_SYN) { 5667 end++; 5668 if (dst->wscale & PF_WSCALE_FLAG) { 5669 src->wscale = pf_get_wscale(pd); 5670 if (src->wscale & PF_WSCALE_FLAG) { 5671 /* Remove scale factor from initial 5672 * window */ 5673 sws = src->wscale & PF_WSCALE_MASK; 5674 win = ((u_int32_t)win + (1 << sws) - 1) 5675 >> sws; 5676 dws = dst->wscale & PF_WSCALE_MASK; 5677 } else { 5678 /* fixup other window */ 5679 dst->max_win = MIN(TCP_MAXWIN, 5680 (u_int32_t)dst->max_win << 5681 (dst->wscale & PF_WSCALE_MASK)); 5682 /* in case of a retrans SYN|ACK */ 5683 dst->wscale = 0; 5684 } 5685 } 5686 } 5687 data_end = end; 5688 if (th->th_flags & TH_FIN) 5689 end++; 5690 5691 src->seqlo = seq; 5692 if (src->state < TCPS_SYN_SENT) 5693 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5694 5695 /* 5696 * May need to slide the window (seqhi may have been set by 5697 * the crappy stack check or if we picked up the connection 5698 * after establishment) 5699 */ 5700 if (src->seqhi == 1 || 5701 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5702 src->seqhi = end + MAX(1, dst->max_win << dws); 5703 if (win > src->max_win) 5704 src->max_win = win; 5705 5706 } else { 5707 ack = ntohl(th->th_ack) - dst->seqdiff; 5708 if (src->seqdiff) { 5709 /* Modulate sequence numbers */ 5710 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 5711 src->seqdiff), 0); 5712 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5713 *copyback = 1; 5714 } 5715 end = seq + pd->p_len; 5716 if (th->th_flags & TH_SYN) 5717 end++; 5718 data_end = end; 5719 if (th->th_flags & TH_FIN) 5720 end++; 5721 } 5722 5723 if ((th->th_flags & TH_ACK) == 0) { 5724 /* Let it pass through the ack skew check */ 5725 ack = dst->seqlo; 5726 } else if ((ack == 0 && 5727 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5728 /* broken tcp stacks do not set ack */ 5729 (dst->state < TCPS_SYN_SENT)) { 5730 /* 5731 * Many stacks (ours included) will set the ACK number in an 5732 * FIN|ACK if the SYN times out -- no sequence to ACK. 5733 */ 5734 ack = dst->seqlo; 5735 } 5736 5737 if (seq == end) { 5738 /* Ease sequencing restrictions on no data packets */ 5739 seq = src->seqlo; 5740 data_end = end = seq; 5741 } 5742 5743 ackskew = dst->seqlo - ack; 5744 5745 /* 5746 * Need to demodulate the sequence numbers in any TCP SACK options 5747 * (Selective ACK). We could optionally validate the SACK values 5748 * against the current ACK window, either forwards or backwards, but 5749 * I'm not confident that SACK has been implemented properly 5750 * everywhere. It wouldn't surprise me if several stacks accidentally 5751 * SACK too far backwards of previously ACKed data. There really aren't 5752 * any security implications of bad SACKing unless the target stack 5753 * doesn't validate the option length correctly. Someone trying to 5754 * spoof into a TCP connection won't bother blindly sending SACK 5755 * options anyway. 5756 */ 5757 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5758 if (pf_modulate_sack(pd, th, dst)) 5759 *copyback = 1; 5760 } 5761 5762 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5763 if (SEQ_GEQ(src->seqhi, data_end) && 5764 /* Last octet inside other's window space */ 5765 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5766 /* Retrans: not more than one window back */ 5767 (ackskew >= -MAXACKWINDOW) && 5768 /* Acking not more than one reassembled fragment backwards */ 5769 (ackskew <= (MAXACKWINDOW << sws)) && 5770 /* Acking not more than one window forward */ 5771 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5772 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 5773 /* Require an exact/+1 sequence match on resets when possible */ 5774 5775 if (dst->scrub || src->scrub) { 5776 if (pf_normalize_tcp_stateful(pd, reason, th, 5777 *state, src, dst, copyback)) 5778 return (PF_DROP); 5779 } 5780 5781 /* update max window */ 5782 if (src->max_win < win) 5783 src->max_win = win; 5784 /* synchronize sequencing */ 5785 if (SEQ_GT(end, src->seqlo)) 5786 src->seqlo = end; 5787 /* slide the window of what the other end can send */ 5788 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5789 dst->seqhi = ack + MAX((win << sws), 1); 5790 5791 /* update states */ 5792 if (th->th_flags & TH_SYN) 5793 if (src->state < TCPS_SYN_SENT) 5794 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5795 if (th->th_flags & TH_FIN) 5796 if (src->state < TCPS_CLOSING) 5797 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5798 if (th->th_flags & TH_ACK) { 5799 if (dst->state == TCPS_SYN_SENT) { 5800 pf_set_protostate(*state, pdst, 5801 TCPS_ESTABLISHED); 5802 if (src->state == TCPS_ESTABLISHED && 5803 (*state)->src_node != NULL && 5804 pf_src_connlimit(state)) { 5805 REASON_SET(reason, PFRES_SRCLIMIT); 5806 return (PF_DROP); 5807 } 5808 } else if (dst->state == TCPS_CLOSING) 5809 pf_set_protostate(*state, pdst, 5810 TCPS_FIN_WAIT_2); 5811 } 5812 if (th->th_flags & TH_RST) 5813 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5814 5815 /* update expire time */ 5816 (*state)->expire = pf_get_uptime(); 5817 if (src->state >= TCPS_FIN_WAIT_2 && 5818 dst->state >= TCPS_FIN_WAIT_2) 5819 (*state)->timeout = PFTM_TCP_CLOSED; 5820 else if (src->state >= TCPS_CLOSING && 5821 dst->state >= TCPS_CLOSING) 5822 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5823 else if (src->state < TCPS_ESTABLISHED || 5824 dst->state < TCPS_ESTABLISHED) 5825 (*state)->timeout = PFTM_TCP_OPENING; 5826 else if (src->state >= TCPS_CLOSING || 5827 dst->state >= TCPS_CLOSING) 5828 (*state)->timeout = PFTM_TCP_CLOSING; 5829 else 5830 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5831 5832 /* Fall through to PASS packet */ 5833 5834 } else if ((dst->state < TCPS_SYN_SENT || 5835 dst->state >= TCPS_FIN_WAIT_2 || 5836 src->state >= TCPS_FIN_WAIT_2) && 5837 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 5838 /* Within a window forward of the originating packet */ 5839 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5840 /* Within a window backward of the originating packet */ 5841 5842 /* 5843 * This currently handles three situations: 5844 * 1) Stupid stacks will shotgun SYNs before their peer 5845 * replies. 5846 * 2) When PF catches an already established stream (the 5847 * firewall rebooted, the state table was flushed, routes 5848 * changed...) 5849 * 3) Packets get funky immediately after the connection 5850 * closes (this should catch Solaris spurious ACK|FINs 5851 * that web servers like to spew after a close) 5852 * 5853 * This must be a little more careful than the above code 5854 * since packet floods will also be caught here. We don't 5855 * update the TTL here to mitigate the damage of a packet 5856 * flood and so the same code can handle awkward establishment 5857 * and a loosened connection close. 5858 * In the establishment case, a correct peer response will 5859 * validate the connection, go through the normal state code 5860 * and keep updating the state TTL. 5861 */ 5862 5863 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5864 printf("pf: loose state match: "); 5865 pf_print_state(*state); 5866 pf_print_flags(th->th_flags); 5867 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5868 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5869 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5870 (unsigned long long)(*state)->packets[1], 5871 pd->dir == PF_IN ? "in" : "out", 5872 pd->dir == (*state)->direction ? "fwd" : "rev"); 5873 } 5874 5875 if (dst->scrub || src->scrub) { 5876 if (pf_normalize_tcp_stateful(pd, reason, th, 5877 *state, src, dst, copyback)) 5878 return (PF_DROP); 5879 } 5880 5881 /* update max window */ 5882 if (src->max_win < win) 5883 src->max_win = win; 5884 /* synchronize sequencing */ 5885 if (SEQ_GT(end, src->seqlo)) 5886 src->seqlo = end; 5887 /* slide the window of what the other end can send */ 5888 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5889 dst->seqhi = ack + MAX((win << sws), 1); 5890 5891 /* 5892 * Cannot set dst->seqhi here since this could be a shotgunned 5893 * SYN and not an already established connection. 5894 */ 5895 5896 if (th->th_flags & TH_FIN) 5897 if (src->state < TCPS_CLOSING) 5898 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5899 if (th->th_flags & TH_RST) 5900 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5901 5902 /* Fall through to PASS packet */ 5903 5904 } else { 5905 if ((*state)->dst.state == TCPS_SYN_SENT && 5906 (*state)->src.state == TCPS_SYN_SENT) { 5907 /* Send RST for state mismatches during handshake */ 5908 if (!(th->th_flags & TH_RST)) 5909 pf_send_tcp((*state)->rule, pd->af, 5910 pd->dst, pd->src, th->th_dport, 5911 th->th_sport, ntohl(th->th_ack), 0, 5912 TH_RST, 0, 0, 5913 (*state)->rule->return_ttl, M_SKIP_FIREWALL, 5914 0, 0, (*state)->act.rtableid); 5915 src->seqlo = 0; 5916 src->seqhi = 1; 5917 src->max_win = 1; 5918 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5919 printf("pf: BAD state: "); 5920 pf_print_state(*state); 5921 pf_print_flags(th->th_flags); 5922 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5923 "pkts=%llu:%llu dir=%s,%s\n", 5924 seq, orig_seq, ack, pd->p_len, ackskew, 5925 (unsigned long long)(*state)->packets[0], 5926 (unsigned long long)(*state)->packets[1], 5927 pd->dir == PF_IN ? "in" : "out", 5928 pd->dir == (*state)->direction ? "fwd" : "rev"); 5929 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5930 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 5931 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5932 ' ': '2', 5933 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5934 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5935 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 5936 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5937 } 5938 REASON_SET(reason, PFRES_BADSTATE); 5939 return (PF_DROP); 5940 } 5941 5942 return (PF_PASS); 5943 } 5944 5945 static int 5946 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5947 { 5948 struct tcphdr *th = &pd->hdr.tcp; 5949 struct pf_state_peer *src, *dst; 5950 u_int8_t psrc, pdst; 5951 5952 if (pd->dir == (*state)->direction) { 5953 src = &(*state)->src; 5954 dst = &(*state)->dst; 5955 psrc = PF_PEER_SRC; 5956 pdst = PF_PEER_DST; 5957 } else { 5958 src = &(*state)->dst; 5959 dst = &(*state)->src; 5960 psrc = PF_PEER_DST; 5961 pdst = PF_PEER_SRC; 5962 } 5963 5964 if (th->th_flags & TH_SYN) 5965 if (src->state < TCPS_SYN_SENT) 5966 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5967 if (th->th_flags & TH_FIN) 5968 if (src->state < TCPS_CLOSING) 5969 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5970 if (th->th_flags & TH_ACK) { 5971 if (dst->state == TCPS_SYN_SENT) { 5972 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 5973 if (src->state == TCPS_ESTABLISHED && 5974 (*state)->src_node != NULL && 5975 pf_src_connlimit(state)) { 5976 REASON_SET(reason, PFRES_SRCLIMIT); 5977 return (PF_DROP); 5978 } 5979 } else if (dst->state == TCPS_CLOSING) { 5980 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 5981 } else if (src->state == TCPS_SYN_SENT && 5982 dst->state < TCPS_SYN_SENT) { 5983 /* 5984 * Handle a special sloppy case where we only see one 5985 * half of the connection. If there is a ACK after 5986 * the initial SYN without ever seeing a packet from 5987 * the destination, set the connection to established. 5988 */ 5989 pf_set_protostate(*state, PF_PEER_BOTH, 5990 TCPS_ESTABLISHED); 5991 dst->state = src->state = TCPS_ESTABLISHED; 5992 if ((*state)->src_node != NULL && 5993 pf_src_connlimit(state)) { 5994 REASON_SET(reason, PFRES_SRCLIMIT); 5995 return (PF_DROP); 5996 } 5997 } else if (src->state == TCPS_CLOSING && 5998 dst->state == TCPS_ESTABLISHED && 5999 dst->seqlo == 0) { 6000 /* 6001 * Handle the closing of half connections where we 6002 * don't see the full bidirectional FIN/ACK+ACK 6003 * handshake. 6004 */ 6005 pf_set_protostate(*state, pdst, TCPS_CLOSING); 6006 } 6007 } 6008 if (th->th_flags & TH_RST) 6009 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6010 6011 /* update expire time */ 6012 (*state)->expire = pf_get_uptime(); 6013 if (src->state >= TCPS_FIN_WAIT_2 && 6014 dst->state >= TCPS_FIN_WAIT_2) 6015 (*state)->timeout = PFTM_TCP_CLOSED; 6016 else if (src->state >= TCPS_CLOSING && 6017 dst->state >= TCPS_CLOSING) 6018 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6019 else if (src->state < TCPS_ESTABLISHED || 6020 dst->state < TCPS_ESTABLISHED) 6021 (*state)->timeout = PFTM_TCP_OPENING; 6022 else if (src->state >= TCPS_CLOSING || 6023 dst->state >= TCPS_CLOSING) 6024 (*state)->timeout = PFTM_TCP_CLOSING; 6025 else 6026 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6027 6028 return (PF_PASS); 6029 } 6030 6031 static int 6032 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 6033 { 6034 struct pf_state_key *sk = (*state)->key[pd->didx]; 6035 struct tcphdr *th = &pd->hdr.tcp; 6036 6037 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 6038 if (pd->dir != (*state)->direction) { 6039 REASON_SET(reason, PFRES_SYNPROXY); 6040 return (PF_SYNPROXY_DROP); 6041 } 6042 if (th->th_flags & TH_SYN) { 6043 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 6044 REASON_SET(reason, PFRES_SYNPROXY); 6045 return (PF_DROP); 6046 } 6047 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6048 pd->src, th->th_dport, th->th_sport, 6049 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 6050 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 6051 M_SKIP_FIREWALL, 0, 0, (*state)->act.rtableid); 6052 REASON_SET(reason, PFRES_SYNPROXY); 6053 return (PF_SYNPROXY_DROP); 6054 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6055 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6056 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6057 REASON_SET(reason, PFRES_SYNPROXY); 6058 return (PF_DROP); 6059 } else if ((*state)->src_node != NULL && 6060 pf_src_connlimit(state)) { 6061 REASON_SET(reason, PFRES_SRCLIMIT); 6062 return (PF_DROP); 6063 } else 6064 pf_set_protostate(*state, PF_PEER_SRC, 6065 PF_TCPS_PROXY_DST); 6066 } 6067 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 6068 if (pd->dir == (*state)->direction) { 6069 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 6070 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6071 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6072 REASON_SET(reason, PFRES_SYNPROXY); 6073 return (PF_DROP); 6074 } 6075 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 6076 if ((*state)->dst.seqhi == 1) 6077 (*state)->dst.seqhi = htonl(arc4random()); 6078 pf_send_tcp((*state)->rule, pd->af, 6079 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6080 sk->port[pd->sidx], sk->port[pd->didx], 6081 (*state)->dst.seqhi, 0, TH_SYN, 0, 6082 (*state)->src.mss, 0, 6083 (*state)->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0, 6084 (*state)->tag, 0, (*state)->act.rtableid); 6085 REASON_SET(reason, PFRES_SYNPROXY); 6086 return (PF_SYNPROXY_DROP); 6087 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 6088 (TH_SYN|TH_ACK)) || 6089 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 6090 REASON_SET(reason, PFRES_SYNPROXY); 6091 return (PF_DROP); 6092 } else { 6093 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 6094 (*state)->dst.seqlo = ntohl(th->th_seq); 6095 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6096 pd->src, th->th_dport, th->th_sport, 6097 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6098 TH_ACK, (*state)->src.max_win, 0, 0, 0, 6099 (*state)->tag, 0, (*state)->act.rtableid); 6100 pf_send_tcp((*state)->rule, pd->af, 6101 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6102 sk->port[pd->sidx], sk->port[pd->didx], 6103 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 6104 TH_ACK, (*state)->dst.max_win, 0, 0, 6105 M_SKIP_FIREWALL, 0, 0, (*state)->act.rtableid); 6106 (*state)->src.seqdiff = (*state)->dst.seqhi - 6107 (*state)->src.seqlo; 6108 (*state)->dst.seqdiff = (*state)->src.seqhi - 6109 (*state)->dst.seqlo; 6110 (*state)->src.seqhi = (*state)->src.seqlo + 6111 (*state)->dst.max_win; 6112 (*state)->dst.seqhi = (*state)->dst.seqlo + 6113 (*state)->src.max_win; 6114 (*state)->src.wscale = (*state)->dst.wscale = 0; 6115 pf_set_protostate(*state, PF_PEER_BOTH, 6116 TCPS_ESTABLISHED); 6117 REASON_SET(reason, PFRES_SYNPROXY); 6118 return (PF_SYNPROXY_DROP); 6119 } 6120 } 6121 6122 return (PF_PASS); 6123 } 6124 6125 static int 6126 pf_test_state_tcp(struct pf_kstate **state, struct pf_pdesc *pd, 6127 u_short *reason) 6128 { 6129 struct pf_state_key_cmp key; 6130 struct tcphdr *th = &pd->hdr.tcp; 6131 int copyback = 0; 6132 int action; 6133 struct pf_state_peer *src, *dst; 6134 6135 bzero(&key, sizeof(key)); 6136 key.af = pd->af; 6137 key.proto = IPPROTO_TCP; 6138 if (pd->dir == PF_IN) { /* wire side, straight */ 6139 PF_ACPY(&key.addr[0], pd->src, key.af); 6140 PF_ACPY(&key.addr[1], pd->dst, key.af); 6141 key.port[0] = th->th_sport; 6142 key.port[1] = th->th_dport; 6143 } else { /* stack side, reverse */ 6144 PF_ACPY(&key.addr[1], pd->src, key.af); 6145 PF_ACPY(&key.addr[0], pd->dst, key.af); 6146 key.port[1] = th->th_sport; 6147 key.port[0] = th->th_dport; 6148 } 6149 6150 STATE_LOOKUP(&key, *state, pd); 6151 6152 if (pd->dir == (*state)->direction) { 6153 src = &(*state)->src; 6154 dst = &(*state)->dst; 6155 } else { 6156 src = &(*state)->dst; 6157 dst = &(*state)->src; 6158 } 6159 6160 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 6161 return (action); 6162 6163 if (dst->state >= TCPS_FIN_WAIT_2 && 6164 src->state >= TCPS_FIN_WAIT_2 && 6165 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 6166 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6167 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6168 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6169 printf("pf: state reuse "); 6170 pf_print_state(*state); 6171 pf_print_flags(th->th_flags); 6172 printf("\n"); 6173 } 6174 /* XXX make sure it's the same direction ?? */ 6175 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6176 pf_unlink_state(*state); 6177 *state = NULL; 6178 return (PF_DROP); 6179 } 6180 6181 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6182 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 6183 return (PF_DROP); 6184 } else { 6185 if (pf_tcp_track_full(state, pd, reason, 6186 ©back) == PF_DROP) 6187 return (PF_DROP); 6188 } 6189 6190 /* translate source/destination address, if necessary */ 6191 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6192 struct pf_state_key *nk = (*state)->key[pd->didx]; 6193 6194 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6195 nk->port[pd->sidx] != th->th_sport) 6196 pf_change_ap(pd->m, pd->src, &th->th_sport, 6197 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 6198 nk->port[pd->sidx], 0, pd->af); 6199 6200 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6201 nk->port[pd->didx] != th->th_dport) 6202 pf_change_ap(pd->m, pd->dst, &th->th_dport, 6203 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 6204 nk->port[pd->didx], 0, pd->af); 6205 copyback = 1; 6206 } 6207 6208 /* Copyback sequence modulation or stateful scrub changes if needed */ 6209 if (copyback) 6210 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th); 6211 6212 return (PF_PASS); 6213 } 6214 6215 static int 6216 pf_test_state_udp(struct pf_kstate **state, struct pf_pdesc *pd) 6217 { 6218 struct pf_state_peer *src, *dst; 6219 struct pf_state_key_cmp key; 6220 struct udphdr *uh = &pd->hdr.udp; 6221 uint8_t psrc, pdst; 6222 6223 bzero(&key, sizeof(key)); 6224 key.af = pd->af; 6225 key.proto = IPPROTO_UDP; 6226 if (pd->dir == PF_IN) { /* wire side, straight */ 6227 PF_ACPY(&key.addr[0], pd->src, key.af); 6228 PF_ACPY(&key.addr[1], pd->dst, key.af); 6229 key.port[0] = uh->uh_sport; 6230 key.port[1] = uh->uh_dport; 6231 } else { /* stack side, reverse */ 6232 PF_ACPY(&key.addr[1], pd->src, key.af); 6233 PF_ACPY(&key.addr[0], pd->dst, key.af); 6234 key.port[1] = uh->uh_sport; 6235 key.port[0] = uh->uh_dport; 6236 } 6237 6238 STATE_LOOKUP(&key, *state, pd); 6239 6240 if (pd->dir == (*state)->direction) { 6241 src = &(*state)->src; 6242 dst = &(*state)->dst; 6243 psrc = PF_PEER_SRC; 6244 pdst = PF_PEER_DST; 6245 } else { 6246 src = &(*state)->dst; 6247 dst = &(*state)->src; 6248 psrc = PF_PEER_DST; 6249 pdst = PF_PEER_SRC; 6250 } 6251 6252 /* update states */ 6253 if (src->state < PFUDPS_SINGLE) 6254 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 6255 if (dst->state == PFUDPS_SINGLE) 6256 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 6257 6258 /* update expire time */ 6259 (*state)->expire = pf_get_uptime(); 6260 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 6261 (*state)->timeout = PFTM_UDP_MULTIPLE; 6262 else 6263 (*state)->timeout = PFTM_UDP_SINGLE; 6264 6265 /* translate source/destination address, if necessary */ 6266 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6267 struct pf_state_key *nk = (*state)->key[pd->didx]; 6268 6269 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6270 nk->port[pd->sidx] != uh->uh_sport) 6271 pf_change_ap(pd->m, pd->src, &uh->uh_sport, pd->ip_sum, 6272 &uh->uh_sum, &nk->addr[pd->sidx], 6273 nk->port[pd->sidx], 1, pd->af); 6274 6275 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6276 nk->port[pd->didx] != uh->uh_dport) 6277 pf_change_ap(pd->m, pd->dst, &uh->uh_dport, pd->ip_sum, 6278 &uh->uh_sum, &nk->addr[pd->didx], 6279 nk->port[pd->didx], 1, pd->af); 6280 m_copyback(pd->m, pd->off, sizeof(*uh), (caddr_t)uh); 6281 } 6282 6283 return (PF_PASS); 6284 } 6285 6286 static int 6287 pf_test_state_sctp(struct pf_kstate **state, struct pf_pdesc *pd, 6288 u_short *reason) 6289 { 6290 struct pf_state_key_cmp key; 6291 struct pf_state_peer *src, *dst; 6292 struct sctphdr *sh = &pd->hdr.sctp; 6293 u_int8_t psrc; //, pdst; 6294 6295 bzero(&key, sizeof(key)); 6296 key.af = pd->af; 6297 key.proto = IPPROTO_SCTP; 6298 if (pd->dir == PF_IN) { /* wire side, straight */ 6299 PF_ACPY(&key.addr[0], pd->src, key.af); 6300 PF_ACPY(&key.addr[1], pd->dst, key.af); 6301 key.port[0] = sh->src_port; 6302 key.port[1] = sh->dest_port; 6303 } else { /* stack side, reverse */ 6304 PF_ACPY(&key.addr[1], pd->src, key.af); 6305 PF_ACPY(&key.addr[0], pd->dst, key.af); 6306 key.port[1] = sh->src_port; 6307 key.port[0] = sh->dest_port; 6308 } 6309 6310 STATE_LOOKUP(&key, *state, pd); 6311 6312 if (pd->dir == (*state)->direction) { 6313 src = &(*state)->src; 6314 dst = &(*state)->dst; 6315 psrc = PF_PEER_SRC; 6316 } else { 6317 src = &(*state)->dst; 6318 dst = &(*state)->src; 6319 psrc = PF_PEER_DST; 6320 } 6321 6322 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 6323 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 6324 pd->sctp_flags & PFDESC_SCTP_INIT) { 6325 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 6326 pf_unlink_state(*state); 6327 *state = NULL; 6328 return (PF_DROP); 6329 } 6330 6331 /* Track state. */ 6332 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 6333 if (src->state < SCTP_COOKIE_WAIT) { 6334 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 6335 (*state)->timeout = PFTM_SCTP_OPENING; 6336 } 6337 } 6338 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 6339 MPASS(dst->scrub != NULL); 6340 if (dst->scrub->pfss_v_tag == 0) 6341 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 6342 } 6343 6344 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 6345 if (src->state < SCTP_ESTABLISHED) { 6346 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 6347 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 6348 } 6349 } 6350 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 6351 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6352 if (src->state < SCTP_SHUTDOWN_PENDING) { 6353 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 6354 (*state)->timeout = PFTM_SCTP_CLOSING; 6355 } 6356 } 6357 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6358 pf_set_protostate(*state, psrc, SCTP_CLOSED); 6359 (*state)->timeout = PFTM_SCTP_CLOSED; 6360 } 6361 6362 if (src->scrub != NULL) { 6363 if (src->scrub->pfss_v_tag == 0) { 6364 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 6365 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 6366 return (PF_DROP); 6367 } 6368 6369 (*state)->expire = pf_get_uptime(); 6370 6371 /* translate source/destination address, if necessary */ 6372 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6373 uint16_t checksum = 0; 6374 struct pf_state_key *nk = (*state)->key[pd->didx]; 6375 6376 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6377 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 6378 pf_change_ap(pd->m, pd->src, &pd->hdr.sctp.src_port, 6379 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 6380 nk->port[pd->sidx], 1, pd->af); 6381 } 6382 6383 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6384 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 6385 pf_change_ap(pd->m, pd->dst, &pd->hdr.sctp.dest_port, 6386 pd->ip_sum, &checksum, &nk->addr[pd->didx], 6387 nk->port[pd->didx], 1, pd->af); 6388 } 6389 } 6390 6391 return (PF_PASS); 6392 } 6393 6394 static void 6395 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 6396 { 6397 struct pf_sctp_endpoint key; 6398 struct pf_sctp_endpoint *ep; 6399 struct pf_state_key *sks = s->key[PF_SK_STACK]; 6400 struct pf_sctp_source *i, *tmp; 6401 6402 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 6403 return; 6404 6405 PF_SCTP_ENDPOINTS_LOCK(); 6406 6407 key.v_tag = s->dst.scrub->pfss_v_tag; 6408 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6409 if (ep != NULL) { 6410 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6411 if (pf_addr_cmp(&i->addr, 6412 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 6413 s->key[PF_SK_WIRE]->af) == 0) { 6414 SDT_PROBE3(pf, sctp, multihome, remove, 6415 key.v_tag, s, i); 6416 TAILQ_REMOVE(&ep->sources, i, entry); 6417 free(i, M_PFTEMP); 6418 break; 6419 } 6420 } 6421 6422 if (TAILQ_EMPTY(&ep->sources)) { 6423 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6424 free(ep, M_PFTEMP); 6425 } 6426 } 6427 6428 /* Other direction. */ 6429 key.v_tag = s->src.scrub->pfss_v_tag; 6430 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6431 if (ep != NULL) { 6432 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6433 if (pf_addr_cmp(&i->addr, 6434 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6435 s->key[PF_SK_WIRE]->af) == 0) { 6436 SDT_PROBE3(pf, sctp, multihome, remove, 6437 key.v_tag, s, i); 6438 TAILQ_REMOVE(&ep->sources, i, entry); 6439 free(i, M_PFTEMP); 6440 break; 6441 } 6442 } 6443 6444 if (TAILQ_EMPTY(&ep->sources)) { 6445 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6446 free(ep, M_PFTEMP); 6447 } 6448 } 6449 6450 PF_SCTP_ENDPOINTS_UNLOCK(); 6451 } 6452 6453 static void 6454 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6455 { 6456 struct pf_sctp_endpoint key = { 6457 .v_tag = v_tag, 6458 }; 6459 struct pf_sctp_source *i; 6460 struct pf_sctp_endpoint *ep; 6461 6462 PF_SCTP_ENDPOINTS_LOCK(); 6463 6464 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6465 if (ep == NULL) { 6466 ep = malloc(sizeof(struct pf_sctp_endpoint), 6467 M_PFTEMP, M_NOWAIT); 6468 if (ep == NULL) { 6469 PF_SCTP_ENDPOINTS_UNLOCK(); 6470 return; 6471 } 6472 6473 ep->v_tag = v_tag; 6474 TAILQ_INIT(&ep->sources); 6475 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6476 } 6477 6478 /* Avoid inserting duplicates. */ 6479 TAILQ_FOREACH(i, &ep->sources, entry) { 6480 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6481 PF_SCTP_ENDPOINTS_UNLOCK(); 6482 return; 6483 } 6484 } 6485 6486 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6487 if (i == NULL) { 6488 PF_SCTP_ENDPOINTS_UNLOCK(); 6489 return; 6490 } 6491 6492 i->af = pd->af; 6493 memcpy(&i->addr, a, sizeof(*a)); 6494 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6495 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 6496 6497 PF_SCTP_ENDPOINTS_UNLOCK(); 6498 } 6499 6500 static void 6501 pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif, 6502 struct pf_kstate *s, int action) 6503 { 6504 struct pf_sctp_multihome_job *j, *tmp; 6505 struct pf_sctp_source *i; 6506 int ret __unused; 6507 struct pf_kstate *sm = NULL; 6508 struct pf_krule *ra = NULL; 6509 struct pf_krule *r = &V_pf_default_rule; 6510 struct pf_kruleset *rs = NULL; 6511 bool do_extra = true; 6512 6513 PF_RULES_RLOCK_TRACKER; 6514 6515 again: 6516 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6517 if (s == NULL || action != PF_PASS) 6518 goto free; 6519 6520 /* Confirm we don't recurse here. */ 6521 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6522 6523 switch (j->op) { 6524 case SCTP_ADD_IP_ADDRESS: { 6525 uint32_t v_tag = pd->sctp_initiate_tag; 6526 6527 if (v_tag == 0) { 6528 if (s->direction == pd->dir) 6529 v_tag = s->src.scrub->pfss_v_tag; 6530 else 6531 v_tag = s->dst.scrub->pfss_v_tag; 6532 } 6533 6534 /* 6535 * Avoid duplicating states. We'll already have 6536 * created a state based on the source address of 6537 * the packet, but SCTP endpoints may also list this 6538 * address again in the INIT(_ACK) parameters. 6539 */ 6540 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6541 break; 6542 } 6543 6544 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6545 PF_RULES_RLOCK(); 6546 sm = NULL; 6547 /* 6548 * New connections need to be floating, because 6549 * we cannot know what interfaces it will use. 6550 * That's why we pass V_pfi_all rather than kif. 6551 */ 6552 j->pd.kif = V_pfi_all; 6553 ret = pf_test_rule(&r, &sm, 6554 &j->pd, &ra, &rs, NULL); 6555 PF_RULES_RUNLOCK(); 6556 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret); 6557 if (ret != PF_DROP && sm != NULL) { 6558 /* Inherit v_tag values. */ 6559 if (sm->direction == s->direction) { 6560 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6561 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6562 } else { 6563 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6564 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6565 } 6566 PF_STATE_UNLOCK(sm); 6567 } else { 6568 /* If we try duplicate inserts? */ 6569 break; 6570 } 6571 6572 /* Only add the address if we've actually allowed the state. */ 6573 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6574 6575 if (! do_extra) { 6576 break; 6577 } 6578 /* 6579 * We need to do this for each of our source addresses. 6580 * Find those based on the verification tag. 6581 */ 6582 struct pf_sctp_endpoint key = { 6583 .v_tag = pd->hdr.sctp.v_tag, 6584 }; 6585 struct pf_sctp_endpoint *ep; 6586 6587 PF_SCTP_ENDPOINTS_LOCK(); 6588 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6589 if (ep == NULL) { 6590 PF_SCTP_ENDPOINTS_UNLOCK(); 6591 break; 6592 } 6593 MPASS(ep != NULL); 6594 6595 TAILQ_FOREACH(i, &ep->sources, entry) { 6596 struct pf_sctp_multihome_job *nj; 6597 6598 /* SCTP can intermingle IPv4 and IPv6. */ 6599 if (i->af != pd->af) 6600 continue; 6601 6602 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6603 if (! nj) { 6604 continue; 6605 } 6606 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6607 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6608 nj->pd.src = &nj->src; 6609 // New destination address! 6610 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6611 nj->pd.dst = &nj->dst; 6612 nj->pd.m = j->pd.m; 6613 nj->op = j->op; 6614 6615 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6616 } 6617 PF_SCTP_ENDPOINTS_UNLOCK(); 6618 6619 break; 6620 } 6621 case SCTP_DEL_IP_ADDRESS: { 6622 struct pf_state_key_cmp key; 6623 uint8_t psrc; 6624 6625 bzero(&key, sizeof(key)); 6626 key.af = j->pd.af; 6627 key.proto = IPPROTO_SCTP; 6628 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6629 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6630 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6631 key.port[0] = j->pd.hdr.sctp.src_port; 6632 key.port[1] = j->pd.hdr.sctp.dest_port; 6633 } else { /* stack side, reverse */ 6634 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6635 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6636 key.port[1] = j->pd.hdr.sctp.src_port; 6637 key.port[0] = j->pd.hdr.sctp.dest_port; 6638 } 6639 6640 sm = pf_find_state(kif, &key, j->pd.dir); 6641 if (sm != NULL) { 6642 PF_STATE_LOCK_ASSERT(sm); 6643 if (j->pd.dir == sm->direction) { 6644 psrc = PF_PEER_SRC; 6645 } else { 6646 psrc = PF_PEER_DST; 6647 } 6648 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6649 sm->timeout = PFTM_SCTP_CLOSING; 6650 PF_STATE_UNLOCK(sm); 6651 } 6652 break; 6653 default: 6654 panic("Unknown op %#x", j->op); 6655 } 6656 } 6657 6658 free: 6659 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6660 free(j, M_PFTEMP); 6661 } 6662 6663 /* We may have inserted extra work while processing the list. */ 6664 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6665 do_extra = false; 6666 goto again; 6667 } 6668 } 6669 6670 static int 6671 pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op) 6672 { 6673 int off = 0; 6674 struct pf_sctp_multihome_job *job; 6675 6676 while (off < len) { 6677 struct sctp_paramhdr h; 6678 6679 if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, NULL, 6680 pd->af)) 6681 return (PF_DROP); 6682 6683 /* Parameters are at least 4 bytes. */ 6684 if (ntohs(h.param_length) < 4) 6685 return (PF_DROP); 6686 6687 switch (ntohs(h.param_type)) { 6688 case SCTP_IPV4_ADDRESS: { 6689 struct in_addr t; 6690 6691 if (ntohs(h.param_length) != 6692 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6693 return (PF_DROP); 6694 6695 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 6696 NULL, NULL, pd->af)) 6697 return (PF_DROP); 6698 6699 if (in_nullhost(t)) 6700 t.s_addr = pd->src->v4.s_addr; 6701 6702 /* 6703 * We hold the state lock (idhash) here, which means 6704 * that we can't acquire the keyhash, or we'll get a 6705 * LOR (and potentially double-lock things too). We also 6706 * can't release the state lock here, so instead we'll 6707 * enqueue this for async handling. 6708 * There's a relatively small race here, in that a 6709 * packet using the new addresses could arrive already, 6710 * but that's just though luck for it. 6711 */ 6712 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6713 if (! job) 6714 return (PF_DROP); 6715 6716 memcpy(&job->pd, pd, sizeof(*pd)); 6717 6718 // New source address! 6719 memcpy(&job->src, &t, sizeof(t)); 6720 job->pd.src = &job->src; 6721 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6722 job->pd.dst = &job->dst; 6723 job->pd.m = pd->m; 6724 job->op = op; 6725 6726 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6727 break; 6728 } 6729 #ifdef INET6 6730 case SCTP_IPV6_ADDRESS: { 6731 struct in6_addr t; 6732 6733 if (ntohs(h.param_length) != 6734 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6735 return (PF_DROP); 6736 6737 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 6738 NULL, NULL, pd->af)) 6739 return (PF_DROP); 6740 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6741 break; 6742 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6743 memcpy(&t, &pd->src->v6, sizeof(t)); 6744 6745 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6746 if (! job) 6747 return (PF_DROP); 6748 6749 memcpy(&job->pd, pd, sizeof(*pd)); 6750 memcpy(&job->src, &t, sizeof(t)); 6751 job->pd.src = &job->src; 6752 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6753 job->pd.dst = &job->dst; 6754 job->pd.m = pd->m; 6755 job->op = op; 6756 6757 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6758 break; 6759 } 6760 #endif 6761 case SCTP_ADD_IP_ADDRESS: { 6762 int ret; 6763 struct sctp_asconf_paramhdr ah; 6764 6765 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 6766 NULL, NULL, pd->af)) 6767 return (PF_DROP); 6768 6769 ret = pf_multihome_scan(start + off + sizeof(ah), 6770 ntohs(ah.ph.param_length) - sizeof(ah), pd, 6771 SCTP_ADD_IP_ADDRESS); 6772 if (ret != PF_PASS) 6773 return (ret); 6774 break; 6775 } 6776 case SCTP_DEL_IP_ADDRESS: { 6777 int ret; 6778 struct sctp_asconf_paramhdr ah; 6779 6780 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 6781 NULL, NULL, pd->af)) 6782 return (PF_DROP); 6783 ret = pf_multihome_scan(start + off + sizeof(ah), 6784 ntohs(ah.ph.param_length) - sizeof(ah), pd, 6785 SCTP_DEL_IP_ADDRESS); 6786 if (ret != PF_PASS) 6787 return (ret); 6788 break; 6789 } 6790 default: 6791 break; 6792 } 6793 6794 off += roundup(ntohs(h.param_length), 4); 6795 } 6796 6797 return (PF_PASS); 6798 } 6799 int 6800 pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd) 6801 { 6802 start += sizeof(struct sctp_init_chunk); 6803 len -= sizeof(struct sctp_init_chunk); 6804 6805 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 6806 } 6807 6808 int 6809 pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd) 6810 { 6811 start += sizeof(struct sctp_asconf_chunk); 6812 len -= sizeof(struct sctp_asconf_chunk); 6813 6814 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 6815 } 6816 6817 int 6818 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 6819 struct pf_kstate **state, int direction, 6820 u_int16_t icmpid, u_int16_t type, int icmp_dir, 6821 int *iidx, int multi, int inner) 6822 { 6823 key->af = pd->af; 6824 key->proto = pd->proto; 6825 if (icmp_dir == PF_IN) { 6826 *iidx = pd->sidx; 6827 key->port[pd->sidx] = icmpid; 6828 key->port[pd->didx] = type; 6829 } else { 6830 *iidx = pd->didx; 6831 key->port[pd->sidx] = type; 6832 key->port[pd->didx] = icmpid; 6833 } 6834 if (pf_state_key_addr_setup(pd, key, multi)) 6835 return (PF_DROP); 6836 6837 STATE_LOOKUP(key, *state, pd); 6838 6839 if ((*state)->state_flags & PFSTATE_SLOPPY) 6840 return (-1); 6841 6842 /* Is this ICMP message flowing in right direction? */ 6843 if ((*state)->rule->type && 6844 (((!inner && (*state)->direction == direction) || 6845 (inner && (*state)->direction != direction)) ? 6846 PF_IN : PF_OUT) != icmp_dir) { 6847 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6848 printf("pf: icmp type %d in wrong direction (%d): ", 6849 ntohs(type), icmp_dir); 6850 pf_print_state(*state); 6851 printf("\n"); 6852 } 6853 PF_STATE_UNLOCK(*state); 6854 *state = NULL; 6855 return (PF_DROP); 6856 } 6857 return (-1); 6858 } 6859 6860 static int 6861 pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd, 6862 u_short *reason) 6863 { 6864 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6865 u_int16_t *icmpsum, virtual_id, virtual_type; 6866 u_int8_t icmptype, icmpcode; 6867 int icmp_dir, iidx, ret, multi; 6868 struct pf_state_key_cmp key; 6869 #ifdef INET 6870 u_int16_t icmpid; 6871 #endif 6872 6873 MPASS(*state == NULL); 6874 6875 bzero(&key, sizeof(key)); 6876 switch (pd->proto) { 6877 #ifdef INET 6878 case IPPROTO_ICMP: 6879 icmptype = pd->hdr.icmp.icmp_type; 6880 icmpcode = pd->hdr.icmp.icmp_code; 6881 icmpid = pd->hdr.icmp.icmp_id; 6882 icmpsum = &pd->hdr.icmp.icmp_cksum; 6883 break; 6884 #endif /* INET */ 6885 #ifdef INET6 6886 case IPPROTO_ICMPV6: 6887 icmptype = pd->hdr.icmp6.icmp6_type; 6888 icmpcode = pd->hdr.icmp6.icmp6_code; 6889 #ifdef INET 6890 icmpid = pd->hdr.icmp6.icmp6_id; 6891 #endif 6892 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 6893 break; 6894 #endif /* INET6 */ 6895 } 6896 6897 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi, 6898 &virtual_id, &virtual_type) == 0) { 6899 /* 6900 * ICMP query/reply message not related to a TCP/UDP packet. 6901 * Search for an ICMP state. 6902 */ 6903 ret = pf_icmp_state_lookup(&key, pd, state, pd->dir, 6904 virtual_id, virtual_type, icmp_dir, &iidx, 6905 PF_ICMP_MULTI_NONE, 0); 6906 if (ret >= 0) { 6907 MPASS(*state == NULL); 6908 if (ret == PF_DROP && pd->af == AF_INET6 && 6909 icmp_dir == PF_OUT) { 6910 ret = pf_icmp_state_lookup(&key, pd, state, 6911 pd->dir, virtual_id, virtual_type, 6912 icmp_dir, &iidx, multi, 0); 6913 if (ret >= 0) { 6914 MPASS(*state == NULL); 6915 return (ret); 6916 } 6917 } else 6918 return (ret); 6919 } 6920 6921 (*state)->expire = pf_get_uptime(); 6922 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 6923 6924 /* translate source/destination address, if necessary */ 6925 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6926 struct pf_state_key *nk = (*state)->key[pd->didx]; 6927 6928 switch (pd->af) { 6929 #ifdef INET 6930 case AF_INET: 6931 if (PF_ANEQ(pd->src, 6932 &nk->addr[pd->sidx], AF_INET)) 6933 pf_change_a(&saddr->v4.s_addr, 6934 pd->ip_sum, 6935 nk->addr[pd->sidx].v4.s_addr, 0); 6936 6937 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 6938 AF_INET)) 6939 pf_change_a(&daddr->v4.s_addr, 6940 pd->ip_sum, 6941 nk->addr[pd->didx].v4.s_addr, 0); 6942 6943 if (nk->port[iidx] != 6944 pd->hdr.icmp.icmp_id) { 6945 pd->hdr.icmp.icmp_cksum = 6946 pf_cksum_fixup( 6947 pd->hdr.icmp.icmp_cksum, icmpid, 6948 nk->port[iidx], 0); 6949 pd->hdr.icmp.icmp_id = 6950 nk->port[iidx]; 6951 } 6952 6953 m_copyback(pd->m, pd->off, ICMP_MINLEN, 6954 (caddr_t )&pd->hdr.icmp); 6955 break; 6956 #endif /* INET */ 6957 #ifdef INET6 6958 case AF_INET6: 6959 if (PF_ANEQ(pd->src, 6960 &nk->addr[pd->sidx], AF_INET6)) 6961 pf_change_a6(saddr, 6962 &pd->hdr.icmp6.icmp6_cksum, 6963 &nk->addr[pd->sidx], 0); 6964 6965 if (PF_ANEQ(pd->dst, 6966 &nk->addr[pd->didx], AF_INET6)) 6967 pf_change_a6(daddr, 6968 &pd->hdr.icmp6.icmp6_cksum, 6969 &nk->addr[pd->didx], 0); 6970 6971 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 6972 (caddr_t )&pd->hdr.icmp6); 6973 break; 6974 #endif /* INET6 */ 6975 } 6976 } 6977 return (PF_PASS); 6978 6979 } else { 6980 /* 6981 * ICMP error message in response to a TCP/UDP packet. 6982 * Extract the inner TCP/UDP header and search for that state. 6983 */ 6984 6985 struct pf_pdesc pd2; 6986 bzero(&pd2, sizeof pd2); 6987 #ifdef INET 6988 struct ip h2; 6989 #endif /* INET */ 6990 #ifdef INET6 6991 struct ip6_hdr h2_6; 6992 int fragoff2, extoff2; 6993 u_int32_t jumbolen; 6994 #endif /* INET6 */ 6995 int ipoff2 = 0; 6996 6997 pd2.af = pd->af; 6998 pd2.dir = pd->dir; 6999 /* Payload packet is from the opposite direction. */ 7000 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7001 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7002 pd2.m = pd->m; 7003 switch (pd->af) { 7004 #ifdef INET 7005 case AF_INET: 7006 /* offset of h2 in mbuf chain */ 7007 ipoff2 = pd->off + ICMP_MINLEN; 7008 7009 if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2), 7010 NULL, reason, pd2.af)) { 7011 DPFPRINTF(PF_DEBUG_MISC, 7012 ("pf: ICMP error message too short " 7013 "(ip)\n")); 7014 return (PF_DROP); 7015 } 7016 /* 7017 * ICMP error messages don't refer to non-first 7018 * fragments 7019 */ 7020 if (h2.ip_off & htons(IP_OFFMASK)) { 7021 REASON_SET(reason, PFRES_FRAG); 7022 return (PF_DROP); 7023 } 7024 7025 /* offset of protocol header that follows h2 */ 7026 pd2.off = ipoff2 + (h2.ip_hl << 2); 7027 7028 pd2.proto = h2.ip_p; 7029 pd2.src = (struct pf_addr *)&h2.ip_src; 7030 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7031 pd2.ip_sum = &h2.ip_sum; 7032 break; 7033 #endif /* INET */ 7034 #ifdef INET6 7035 case AF_INET6: 7036 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 7037 7038 if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6), 7039 NULL, reason, pd2.af)) { 7040 DPFPRINTF(PF_DEBUG_MISC, 7041 ("pf: ICMP error message too short " 7042 "(ip6)\n")); 7043 return (PF_DROP); 7044 } 7045 pd2.off = ipoff2; 7046 if (pf_walk_header6(pd->m, &h2_6, &pd2.off, &extoff2, 7047 &fragoff2, &pd2.proto, &jumbolen, 7048 reason) != PF_PASS) 7049 return (PF_DROP); 7050 7051 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7052 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7053 pd2.ip_sum = NULL; 7054 break; 7055 #endif /* INET6 */ 7056 } 7057 7058 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 7059 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7060 printf("pf: BAD ICMP %d:%d outer dst: ", 7061 icmptype, icmpcode); 7062 pf_print_host(pd->src, 0, pd->af); 7063 printf(" -> "); 7064 pf_print_host(pd->dst, 0, pd->af); 7065 printf(" inner src: "); 7066 pf_print_host(pd2.src, 0, pd2.af); 7067 printf(" -> "); 7068 pf_print_host(pd2.dst, 0, pd2.af); 7069 printf("\n"); 7070 } 7071 REASON_SET(reason, PFRES_BADSTATE); 7072 return (PF_DROP); 7073 } 7074 7075 switch (pd2.proto) { 7076 case IPPROTO_TCP: { 7077 struct tcphdr th; 7078 u_int32_t seq; 7079 struct pf_state_peer *src, *dst; 7080 u_int8_t dws; 7081 int copyback = 0; 7082 7083 /* 7084 * Only the first 8 bytes of the TCP header can be 7085 * expected. Don't access any TCP header fields after 7086 * th_seq, an ackskew test is not possible. 7087 */ 7088 if (!pf_pull_hdr(pd->m, pd2.off, &th, 8, NULL, reason, 7089 pd2.af)) { 7090 DPFPRINTF(PF_DEBUG_MISC, 7091 ("pf: ICMP error message too short " 7092 "(tcp)\n")); 7093 return (PF_DROP); 7094 } 7095 7096 key.af = pd2.af; 7097 key.proto = IPPROTO_TCP; 7098 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7099 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7100 key.port[pd2.sidx] = th.th_sport; 7101 key.port[pd2.didx] = th.th_dport; 7102 7103 STATE_LOOKUP(&key, *state, pd); 7104 7105 if (pd->dir == (*state)->direction) { 7106 src = &(*state)->dst; 7107 dst = &(*state)->src; 7108 } else { 7109 src = &(*state)->src; 7110 dst = &(*state)->dst; 7111 } 7112 7113 if (src->wscale && dst->wscale) 7114 dws = dst->wscale & PF_WSCALE_MASK; 7115 else 7116 dws = 0; 7117 7118 /* Demodulate sequence number */ 7119 seq = ntohl(th.th_seq) - src->seqdiff; 7120 if (src->seqdiff) { 7121 pf_change_a(&th.th_seq, icmpsum, 7122 htonl(seq), 0); 7123 copyback = 1; 7124 } 7125 7126 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7127 (!SEQ_GEQ(src->seqhi, seq) || 7128 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7129 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7130 printf("pf: BAD ICMP %d:%d ", 7131 icmptype, icmpcode); 7132 pf_print_host(pd->src, 0, pd->af); 7133 printf(" -> "); 7134 pf_print_host(pd->dst, 0, pd->af); 7135 printf(" state: "); 7136 pf_print_state(*state); 7137 printf(" seq=%u\n", seq); 7138 } 7139 REASON_SET(reason, PFRES_BADSTATE); 7140 return (PF_DROP); 7141 } else { 7142 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7143 printf("pf: OK ICMP %d:%d ", 7144 icmptype, icmpcode); 7145 pf_print_host(pd->src, 0, pd->af); 7146 printf(" -> "); 7147 pf_print_host(pd->dst, 0, pd->af); 7148 printf(" state: "); 7149 pf_print_state(*state); 7150 printf(" seq=%u\n", seq); 7151 } 7152 } 7153 7154 /* translate source/destination address, if necessary */ 7155 if ((*state)->key[PF_SK_WIRE] != 7156 (*state)->key[PF_SK_STACK]) { 7157 struct pf_state_key *nk = 7158 (*state)->key[pd->didx]; 7159 7160 if (PF_ANEQ(pd2.src, 7161 &nk->addr[pd2.sidx], pd2.af) || 7162 nk->port[pd2.sidx] != th.th_sport) 7163 pf_change_icmp(pd2.src, &th.th_sport, 7164 daddr, &nk->addr[pd2.sidx], 7165 nk->port[pd2.sidx], NULL, 7166 pd2.ip_sum, icmpsum, 7167 pd->ip_sum, 0, pd2.af); 7168 7169 if (PF_ANEQ(pd2.dst, 7170 &nk->addr[pd2.didx], pd2.af) || 7171 nk->port[pd2.didx] != th.th_dport) 7172 pf_change_icmp(pd2.dst, &th.th_dport, 7173 saddr, &nk->addr[pd2.didx], 7174 nk->port[pd2.didx], NULL, 7175 pd2.ip_sum, icmpsum, 7176 pd->ip_sum, 0, pd2.af); 7177 copyback = 1; 7178 } 7179 7180 if (copyback) { 7181 switch (pd2.af) { 7182 #ifdef INET 7183 case AF_INET: 7184 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7185 (caddr_t )&pd->hdr.icmp); 7186 m_copyback(pd->m, ipoff2, sizeof(h2), 7187 (caddr_t )&h2); 7188 break; 7189 #endif /* INET */ 7190 #ifdef INET6 7191 case AF_INET6: 7192 m_copyback(pd->m, pd->off, 7193 sizeof(struct icmp6_hdr), 7194 (caddr_t )&pd->hdr.icmp6); 7195 m_copyback(pd->m, ipoff2, sizeof(h2_6), 7196 (caddr_t )&h2_6); 7197 break; 7198 #endif /* INET6 */ 7199 } 7200 m_copyback(pd->m, pd2.off, 8, (caddr_t)&th); 7201 } 7202 7203 return (PF_PASS); 7204 break; 7205 } 7206 case IPPROTO_UDP: { 7207 struct udphdr uh; 7208 7209 if (!pf_pull_hdr(pd->m, pd2.off, &uh, sizeof(uh), 7210 NULL, reason, pd2.af)) { 7211 DPFPRINTF(PF_DEBUG_MISC, 7212 ("pf: ICMP error message too short " 7213 "(udp)\n")); 7214 return (PF_DROP); 7215 } 7216 7217 key.af = pd2.af; 7218 key.proto = IPPROTO_UDP; 7219 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7220 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7221 key.port[pd2.sidx] = uh.uh_sport; 7222 key.port[pd2.didx] = uh.uh_dport; 7223 7224 STATE_LOOKUP(&key, *state, pd); 7225 7226 /* translate source/destination address, if necessary */ 7227 if ((*state)->key[PF_SK_WIRE] != 7228 (*state)->key[PF_SK_STACK]) { 7229 struct pf_state_key *nk = 7230 (*state)->key[pd->didx]; 7231 7232 if (PF_ANEQ(pd2.src, 7233 &nk->addr[pd2.sidx], pd2.af) || 7234 nk->port[pd2.sidx] != uh.uh_sport) 7235 pf_change_icmp(pd2.src, &uh.uh_sport, 7236 daddr, &nk->addr[pd2.sidx], 7237 nk->port[pd2.sidx], &uh.uh_sum, 7238 pd2.ip_sum, icmpsum, 7239 pd->ip_sum, 1, pd2.af); 7240 7241 if (PF_ANEQ(pd2.dst, 7242 &nk->addr[pd2.didx], pd2.af) || 7243 nk->port[pd2.didx] != uh.uh_dport) 7244 pf_change_icmp(pd2.dst, &uh.uh_dport, 7245 saddr, &nk->addr[pd2.didx], 7246 nk->port[pd2.didx], &uh.uh_sum, 7247 pd2.ip_sum, icmpsum, 7248 pd->ip_sum, 1, pd2.af); 7249 7250 switch (pd2.af) { 7251 #ifdef INET 7252 case AF_INET: 7253 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7254 (caddr_t )&pd->hdr.icmp); 7255 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 7256 break; 7257 #endif /* INET */ 7258 #ifdef INET6 7259 case AF_INET6: 7260 m_copyback(pd->m, pd->off, 7261 sizeof(struct icmp6_hdr), 7262 (caddr_t )&pd->hdr.icmp6); 7263 m_copyback(pd->m, ipoff2, sizeof(h2_6), 7264 (caddr_t )&h2_6); 7265 break; 7266 #endif /* INET6 */ 7267 } 7268 m_copyback(pd->m, pd2.off, sizeof(uh), (caddr_t)&uh); 7269 } 7270 return (PF_PASS); 7271 break; 7272 } 7273 #ifdef INET 7274 case IPPROTO_ICMP: { 7275 struct icmp *iih = &pd2.hdr.icmp; 7276 7277 if (pd2.af != AF_INET) { 7278 REASON_SET(reason, PFRES_NORM); 7279 return (PF_DROP); 7280 } 7281 7282 if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN, 7283 NULL, reason, pd2.af)) { 7284 DPFPRINTF(PF_DEBUG_MISC, 7285 ("pf: ICMP error message too short i" 7286 "(icmp)\n")); 7287 return (PF_DROP); 7288 } 7289 7290 icmpid = iih->icmp_id; 7291 pf_icmp_mapping(&pd2, iih->icmp_type, 7292 &icmp_dir, &multi, &virtual_id, &virtual_type); 7293 7294 ret = pf_icmp_state_lookup(&key, &pd2, state, 7295 pd2.dir, virtual_id, virtual_type, 7296 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7297 if (ret >= 0) { 7298 MPASS(*state == NULL); 7299 return (ret); 7300 } 7301 7302 /* translate source/destination address, if necessary */ 7303 if ((*state)->key[PF_SK_WIRE] != 7304 (*state)->key[PF_SK_STACK]) { 7305 struct pf_state_key *nk = 7306 (*state)->key[pd->didx]; 7307 7308 if (PF_ANEQ(pd2.src, 7309 &nk->addr[pd2.sidx], pd2.af) || 7310 (virtual_type == htons(ICMP_ECHO) && 7311 nk->port[iidx] != iih->icmp_id)) 7312 pf_change_icmp(pd2.src, 7313 (virtual_type == htons(ICMP_ECHO)) ? 7314 &iih->icmp_id : NULL, 7315 daddr, &nk->addr[pd2.sidx], 7316 (virtual_type == htons(ICMP_ECHO)) ? 7317 nk->port[iidx] : 0, NULL, 7318 pd2.ip_sum, icmpsum, 7319 pd->ip_sum, 0, AF_INET); 7320 7321 if (PF_ANEQ(pd2.dst, 7322 &nk->addr[pd2.didx], pd2.af)) 7323 pf_change_icmp(pd2.dst, NULL, NULL, 7324 &nk->addr[pd2.didx], 0, NULL, 7325 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 7326 AF_INET); 7327 7328 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 7329 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 7330 m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih); 7331 } 7332 return (PF_PASS); 7333 break; 7334 } 7335 #endif /* INET */ 7336 #ifdef INET6 7337 case IPPROTO_ICMPV6: { 7338 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 7339 7340 if (pd2.af != AF_INET6) { 7341 REASON_SET(reason, PFRES_NORM); 7342 return (PF_DROP); 7343 } 7344 7345 if (!pf_pull_hdr(pd->m, pd2.off, iih, 7346 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 7347 DPFPRINTF(PF_DEBUG_MISC, 7348 ("pf: ICMP error message too short " 7349 "(icmp6)\n")); 7350 return (PF_DROP); 7351 } 7352 7353 pf_icmp_mapping(&pd2, iih->icmp6_type, 7354 &icmp_dir, &multi, &virtual_id, &virtual_type); 7355 7356 ret = pf_icmp_state_lookup(&key, &pd2, state, 7357 pd->dir, virtual_id, virtual_type, 7358 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7359 if (ret >= 0) { 7360 MPASS(*state == NULL); 7361 if (ret == PF_DROP && pd2.af == AF_INET6 && 7362 icmp_dir == PF_OUT) { 7363 ret = pf_icmp_state_lookup(&key, &pd2, 7364 state, pd->dir, 7365 virtual_id, virtual_type, 7366 icmp_dir, &iidx, multi, 1); 7367 if (ret >= 0) { 7368 MPASS(*state == NULL); 7369 return (ret); 7370 } 7371 } else 7372 return (ret); 7373 } 7374 7375 /* translate source/destination address, if necessary */ 7376 if ((*state)->key[PF_SK_WIRE] != 7377 (*state)->key[PF_SK_STACK]) { 7378 struct pf_state_key *nk = 7379 (*state)->key[pd->didx]; 7380 7381 if (PF_ANEQ(pd2.src, 7382 &nk->addr[pd2.sidx], pd2.af) || 7383 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 7384 nk->port[pd2.sidx] != iih->icmp6_id)) 7385 pf_change_icmp(pd2.src, 7386 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7387 ? &iih->icmp6_id : NULL, 7388 daddr, &nk->addr[pd2.sidx], 7389 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7390 ? nk->port[iidx] : 0, NULL, 7391 pd2.ip_sum, icmpsum, 7392 pd->ip_sum, 0, AF_INET6); 7393 7394 if (PF_ANEQ(pd2.dst, 7395 &nk->addr[pd2.didx], pd2.af)) 7396 pf_change_icmp(pd2.dst, NULL, NULL, 7397 &nk->addr[pd2.didx], 0, NULL, 7398 pd2.ip_sum, icmpsum, 7399 pd->ip_sum, 0, AF_INET6); 7400 7401 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 7402 (caddr_t)&pd->hdr.icmp6); 7403 m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 7404 m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr), 7405 (caddr_t)iih); 7406 } 7407 return (PF_PASS); 7408 break; 7409 } 7410 #endif /* INET6 */ 7411 default: { 7412 key.af = pd2.af; 7413 key.proto = pd2.proto; 7414 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7415 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7416 key.port[0] = key.port[1] = 0; 7417 7418 STATE_LOOKUP(&key, *state, pd); 7419 7420 /* translate source/destination address, if necessary */ 7421 if ((*state)->key[PF_SK_WIRE] != 7422 (*state)->key[PF_SK_STACK]) { 7423 struct pf_state_key *nk = 7424 (*state)->key[pd->didx]; 7425 7426 if (PF_ANEQ(pd2.src, 7427 &nk->addr[pd2.sidx], pd2.af)) 7428 pf_change_icmp(pd2.src, NULL, daddr, 7429 &nk->addr[pd2.sidx], 0, NULL, 7430 pd2.ip_sum, icmpsum, 7431 pd->ip_sum, 0, pd2.af); 7432 7433 if (PF_ANEQ(pd2.dst, 7434 &nk->addr[pd2.didx], pd2.af)) 7435 pf_change_icmp(pd2.dst, NULL, saddr, 7436 &nk->addr[pd2.didx], 0, NULL, 7437 pd2.ip_sum, icmpsum, 7438 pd->ip_sum, 0, pd2.af); 7439 7440 switch (pd2.af) { 7441 #ifdef INET 7442 case AF_INET: 7443 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7444 (caddr_t)&pd->hdr.icmp); 7445 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 7446 break; 7447 #endif /* INET */ 7448 #ifdef INET6 7449 case AF_INET6: 7450 m_copyback(pd->m, pd->off, 7451 sizeof(struct icmp6_hdr), 7452 (caddr_t )&pd->hdr.icmp6); 7453 m_copyback(pd->m, ipoff2, sizeof(h2_6), 7454 (caddr_t )&h2_6); 7455 break; 7456 #endif /* INET6 */ 7457 } 7458 } 7459 return (PF_PASS); 7460 break; 7461 } 7462 } 7463 } 7464 } 7465 7466 static int 7467 pf_test_state_other(struct pf_kstate **state, struct pf_pdesc *pd) 7468 { 7469 struct pf_state_peer *src, *dst; 7470 struct pf_state_key_cmp key; 7471 uint8_t psrc, pdst; 7472 7473 bzero(&key, sizeof(key)); 7474 key.af = pd->af; 7475 key.proto = pd->proto; 7476 if (pd->dir == PF_IN) { 7477 PF_ACPY(&key.addr[0], pd->src, key.af); 7478 PF_ACPY(&key.addr[1], pd->dst, key.af); 7479 key.port[0] = key.port[1] = 0; 7480 } else { 7481 PF_ACPY(&key.addr[1], pd->src, key.af); 7482 PF_ACPY(&key.addr[0], pd->dst, key.af); 7483 key.port[1] = key.port[0] = 0; 7484 } 7485 7486 STATE_LOOKUP(&key, *state, pd); 7487 7488 if (pd->dir == (*state)->direction) { 7489 src = &(*state)->src; 7490 dst = &(*state)->dst; 7491 psrc = PF_PEER_SRC; 7492 pdst = PF_PEER_DST; 7493 } else { 7494 src = &(*state)->dst; 7495 dst = &(*state)->src; 7496 psrc = PF_PEER_DST; 7497 pdst = PF_PEER_SRC; 7498 } 7499 7500 /* update states */ 7501 if (src->state < PFOTHERS_SINGLE) 7502 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7503 if (dst->state == PFOTHERS_SINGLE) 7504 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7505 7506 /* update expire time */ 7507 (*state)->expire = pf_get_uptime(); 7508 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7509 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7510 else 7511 (*state)->timeout = PFTM_OTHER_SINGLE; 7512 7513 /* translate source/destination address, if necessary */ 7514 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7515 struct pf_state_key *nk = (*state)->key[pd->didx]; 7516 7517 KASSERT(nk, ("%s: nk is null", __func__)); 7518 KASSERT(pd, ("%s: pd is null", __func__)); 7519 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7520 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7521 switch (pd->af) { 7522 #ifdef INET 7523 case AF_INET: 7524 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7525 pf_change_a(&pd->src->v4.s_addr, 7526 pd->ip_sum, 7527 nk->addr[pd->sidx].v4.s_addr, 7528 0); 7529 7530 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7531 pf_change_a(&pd->dst->v4.s_addr, 7532 pd->ip_sum, 7533 nk->addr[pd->didx].v4.s_addr, 7534 0); 7535 7536 break; 7537 #endif /* INET */ 7538 #ifdef INET6 7539 case AF_INET6: 7540 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6)) 7541 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7542 7543 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6)) 7544 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7545 #endif /* INET6 */ 7546 } 7547 } 7548 return (PF_PASS); 7549 } 7550 7551 /* 7552 * ipoff and off are measured from the start of the mbuf chain. 7553 * h must be at "ipoff" on the mbuf chain. 7554 */ 7555 void * 7556 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 7557 u_short *actionp, u_short *reasonp, sa_family_t af) 7558 { 7559 switch (af) { 7560 #ifdef INET 7561 case AF_INET: { 7562 const struct ip *h = mtod(m, struct ip *); 7563 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7564 7565 if (fragoff) { 7566 if (fragoff >= len) 7567 ACTION_SET(actionp, PF_PASS); 7568 else { 7569 ACTION_SET(actionp, PF_DROP); 7570 REASON_SET(reasonp, PFRES_FRAG); 7571 } 7572 return (NULL); 7573 } 7574 if (m->m_pkthdr.len < off + len || 7575 ntohs(h->ip_len) < off + len) { 7576 ACTION_SET(actionp, PF_DROP); 7577 REASON_SET(reasonp, PFRES_SHORT); 7578 return (NULL); 7579 } 7580 break; 7581 } 7582 #endif /* INET */ 7583 #ifdef INET6 7584 case AF_INET6: { 7585 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7586 7587 if (m->m_pkthdr.len < off + len || 7588 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7589 (unsigned)(off + len)) { 7590 ACTION_SET(actionp, PF_DROP); 7591 REASON_SET(reasonp, PFRES_SHORT); 7592 return (NULL); 7593 } 7594 break; 7595 } 7596 #endif /* INET6 */ 7597 } 7598 m_copydata(m, off, len, p); 7599 return (p); 7600 } 7601 7602 int 7603 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7604 int rtableid) 7605 { 7606 struct ifnet *ifp; 7607 7608 /* 7609 * Skip check for addresses with embedded interface scope, 7610 * as they would always match anyway. 7611 */ 7612 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7613 return (1); 7614 7615 if (af != AF_INET && af != AF_INET6) 7616 return (0); 7617 7618 if (kif == V_pfi_all) 7619 return (1); 7620 7621 /* Skip checks for ipsec interfaces */ 7622 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7623 return (1); 7624 7625 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7626 7627 switch (af) { 7628 #ifdef INET6 7629 case AF_INET6: 7630 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7631 ifp)); 7632 #endif 7633 #ifdef INET 7634 case AF_INET: 7635 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7636 ifp)); 7637 #endif 7638 } 7639 7640 return (0); 7641 } 7642 7643 #ifdef INET 7644 static void 7645 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7646 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7647 { 7648 struct mbuf *m0, *m1, *md; 7649 struct sockaddr_in dst; 7650 struct ip *ip; 7651 struct pfi_kkif *nkif = NULL; 7652 struct ifnet *ifp = NULL; 7653 struct pf_addr naddr; 7654 int error = 0; 7655 uint16_t ip_len, ip_off; 7656 uint16_t tmp; 7657 int r_rt, r_dir; 7658 7659 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7660 7661 if (s) { 7662 r_rt = s->rt; 7663 r_dir = s->direction; 7664 } else { 7665 r_rt = r->rt; 7666 r_dir = r->direction; 7667 } 7668 7669 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7670 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7671 __func__)); 7672 7673 if ((pd->pf_mtag == NULL && 7674 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7675 pd->pf_mtag->routed++ > 3) { 7676 m0 = *m; 7677 *m = NULL; 7678 goto bad_locked; 7679 } 7680 7681 if (r_rt == PF_DUPTO) { 7682 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7683 if (s == NULL) { 7684 ifp = r->rpool.cur->kif ? 7685 r->rpool.cur->kif->pfik_ifp : NULL; 7686 } else { 7687 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7688 /* If pfsync'd */ 7689 if (ifp == NULL && r->rpool.cur != NULL) 7690 ifp = r->rpool.cur->kif ? 7691 r->rpool.cur->kif->pfik_ifp : NULL; 7692 PF_STATE_UNLOCK(s); 7693 } 7694 if (ifp == oifp) { 7695 /* When the 2nd interface is not skipped */ 7696 return; 7697 } else { 7698 m0 = *m; 7699 *m = NULL; 7700 goto bad; 7701 } 7702 } else { 7703 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7704 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7705 if (s) 7706 PF_STATE_UNLOCK(s); 7707 return; 7708 } 7709 } 7710 } else { 7711 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7712 pf_dummynet(pd, s, r, m); 7713 if (s) 7714 PF_STATE_UNLOCK(s); 7715 return; 7716 } 7717 m0 = *m; 7718 } 7719 7720 ip = mtod(m0, struct ip *); 7721 7722 bzero(&dst, sizeof(dst)); 7723 dst.sin_family = AF_INET; 7724 dst.sin_len = sizeof(dst); 7725 dst.sin_addr = ip->ip_dst; 7726 7727 bzero(&naddr, sizeof(naddr)); 7728 7729 if (s == NULL) { 7730 if (TAILQ_EMPTY(&r->rpool.list)) { 7731 DPFPRINTF(PF_DEBUG_URGENT, 7732 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7733 goto bad_locked; 7734 } 7735 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7736 &naddr, &nkif, NULL); 7737 if (!PF_AZERO(&naddr, AF_INET)) 7738 dst.sin_addr.s_addr = naddr.v4.s_addr; 7739 ifp = nkif ? nkif->pfik_ifp : NULL; 7740 } else { 7741 struct pfi_kkif *kif; 7742 7743 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7744 dst.sin_addr.s_addr = 7745 s->rt_addr.v4.s_addr; 7746 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7747 kif = s->rt_kif; 7748 /* If pfsync'd */ 7749 if (ifp == NULL && r->rpool.cur != NULL) { 7750 ifp = r->rpool.cur->kif ? 7751 r->rpool.cur->kif->pfik_ifp : NULL; 7752 kif = r->rpool.cur->kif; 7753 } 7754 if (ifp != NULL && kif != NULL && 7755 r->rule_flag & PFRULE_IFBOUND && 7756 r->rt == PF_REPLYTO && 7757 s->kif == V_pfi_all) { 7758 s->kif = kif; 7759 s->orig_kif = oifp->if_pf_kif; 7760 } 7761 7762 PF_STATE_UNLOCK(s); 7763 } 7764 7765 if (ifp == NULL) 7766 goto bad; 7767 7768 if (pd->dir == PF_IN) { 7769 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 7770 &pd->act) != PF_PASS) 7771 goto bad; 7772 else if (m0 == NULL) 7773 goto done; 7774 if (m0->m_len < sizeof(struct ip)) { 7775 DPFPRINTF(PF_DEBUG_URGENT, 7776 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7777 goto bad; 7778 } 7779 ip = mtod(m0, struct ip *); 7780 } 7781 7782 if (ifp->if_flags & IFF_LOOPBACK) 7783 m0->m_flags |= M_SKIP_FIREWALL; 7784 7785 ip_len = ntohs(ip->ip_len); 7786 ip_off = ntohs(ip->ip_off); 7787 7788 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7789 m0->m_pkthdr.csum_flags |= CSUM_IP; 7790 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7791 in_delayed_cksum(m0); 7792 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7793 } 7794 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7795 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7796 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7797 } 7798 7799 if (pd->dir == PF_IN) { 7800 /* 7801 * Make sure dummynet gets the correct direction, in case it needs to 7802 * re-inject later. 7803 */ 7804 pd->dir = PF_OUT; 7805 7806 /* 7807 * The following processing is actually the rest of the inbound processing, even 7808 * though we've marked it as outbound (so we don't look through dummynet) and it 7809 * happens after the outbound processing (pf_test(PF_OUT) above). 7810 * Swap the dummynet pipe numbers, because it's going to come to the wrong 7811 * conclusion about what direction it's processing, and we can't fix it or it 7812 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 7813 * decision will pick the right pipe, and everything will mostly work as expected. 7814 */ 7815 tmp = pd->act.dnrpipe; 7816 pd->act.dnrpipe = pd->act.dnpipe; 7817 pd->act.dnpipe = tmp; 7818 } 7819 7820 /* 7821 * If small enough for interface, or the interface will take 7822 * care of the fragmentation for us, we can just send directly. 7823 */ 7824 if (ip_len <= ifp->if_mtu || 7825 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7826 ip->ip_sum = 0; 7827 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7828 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7829 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7830 } 7831 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7832 7833 md = m0; 7834 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7835 if (md != NULL) 7836 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7837 goto done; 7838 } 7839 7840 /* Balk when DF bit is set or the interface didn't support TSO. */ 7841 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7842 error = EMSGSIZE; 7843 KMOD_IPSTAT_INC(ips_cantfrag); 7844 if (r_rt != PF_DUPTO) { 7845 if (s && s->nat_rule != NULL) 7846 PACKET_UNDO_NAT(m0, pd, 7847 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7848 s); 7849 7850 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7851 ifp->if_mtu); 7852 goto done; 7853 } else 7854 goto bad; 7855 } 7856 7857 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7858 if (error) 7859 goto bad; 7860 7861 for (; m0; m0 = m1) { 7862 m1 = m0->m_nextpkt; 7863 m0->m_nextpkt = NULL; 7864 if (error == 0) { 7865 m_clrprotoflags(m0); 7866 md = m0; 7867 pd->pf_mtag = pf_find_mtag(md); 7868 error = pf_dummynet_route(pd, s, r, ifp, 7869 sintosa(&dst), &md); 7870 if (md != NULL) 7871 error = (*ifp->if_output)(ifp, md, 7872 sintosa(&dst), NULL); 7873 } else 7874 m_freem(m0); 7875 } 7876 7877 if (error == 0) 7878 KMOD_IPSTAT_INC(ips_fragmented); 7879 7880 done: 7881 if (r_rt != PF_DUPTO) 7882 *m = NULL; 7883 return; 7884 7885 bad_locked: 7886 if (s) 7887 PF_STATE_UNLOCK(s); 7888 bad: 7889 m_freem(m0); 7890 goto done; 7891 } 7892 #endif /* INET */ 7893 7894 #ifdef INET6 7895 static void 7896 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7897 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7898 { 7899 struct mbuf *m0, *md; 7900 struct sockaddr_in6 dst; 7901 struct ip6_hdr *ip6; 7902 struct pfi_kkif *nkif = NULL; 7903 struct ifnet *ifp = NULL; 7904 struct pf_addr naddr; 7905 int r_rt, r_dir; 7906 7907 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7908 7909 if (s) { 7910 r_rt = s->rt; 7911 r_dir = s->direction; 7912 } else { 7913 r_rt = r->rt; 7914 r_dir = r->direction; 7915 } 7916 7917 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7918 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7919 __func__)); 7920 7921 if ((pd->pf_mtag == NULL && 7922 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7923 pd->pf_mtag->routed++ > 3) { 7924 m0 = *m; 7925 *m = NULL; 7926 goto bad_locked; 7927 } 7928 7929 if (r_rt == PF_DUPTO) { 7930 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7931 if (s == NULL) { 7932 ifp = r->rpool.cur->kif ? 7933 r->rpool.cur->kif->pfik_ifp : NULL; 7934 } else { 7935 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7936 /* If pfsync'd */ 7937 if (ifp == NULL && r->rpool.cur != NULL) 7938 ifp = r->rpool.cur->kif ? 7939 r->rpool.cur->kif->pfik_ifp : NULL; 7940 PF_STATE_UNLOCK(s); 7941 } 7942 if (ifp == oifp) { 7943 /* When the 2nd interface is not skipped */ 7944 return; 7945 } else { 7946 m0 = *m; 7947 *m = NULL; 7948 goto bad; 7949 } 7950 } else { 7951 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7952 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7953 if (s) 7954 PF_STATE_UNLOCK(s); 7955 return; 7956 } 7957 } 7958 } else { 7959 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7960 pf_dummynet(pd, s, r, m); 7961 if (s) 7962 PF_STATE_UNLOCK(s); 7963 return; 7964 } 7965 m0 = *m; 7966 } 7967 7968 ip6 = mtod(m0, struct ip6_hdr *); 7969 7970 bzero(&dst, sizeof(dst)); 7971 dst.sin6_family = AF_INET6; 7972 dst.sin6_len = sizeof(dst); 7973 dst.sin6_addr = ip6->ip6_dst; 7974 7975 bzero(&naddr, sizeof(naddr)); 7976 7977 if (s == NULL) { 7978 if (TAILQ_EMPTY(&r->rpool.list)) { 7979 DPFPRINTF(PF_DEBUG_URGENT, 7980 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7981 goto bad_locked; 7982 } 7983 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 7984 &naddr, &nkif, NULL); 7985 if (!PF_AZERO(&naddr, AF_INET6)) 7986 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7987 &naddr, AF_INET6); 7988 ifp = nkif ? nkif->pfik_ifp : NULL; 7989 } else { 7990 struct pfi_kkif *kif; 7991 7992 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 7993 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7994 &s->rt_addr, AF_INET6); 7995 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7996 kif = s->rt_kif; 7997 /* If pfsync'd */ 7998 if (ifp == NULL && r->rpool.cur != NULL) { 7999 ifp = r->rpool.cur->kif ? 8000 r->rpool.cur->kif->pfik_ifp : NULL; 8001 kif = r->rpool.cur->kif; 8002 } 8003 if (ifp != NULL && kif != NULL && 8004 r->rule_flag & PFRULE_IFBOUND && 8005 r->rt == PF_REPLYTO && 8006 s->kif == V_pfi_all) { 8007 s->kif = kif; 8008 s->orig_kif = oifp->if_pf_kif; 8009 } 8010 } 8011 8012 if (s) 8013 PF_STATE_UNLOCK(s); 8014 8015 if (ifp == NULL) 8016 goto bad; 8017 8018 if (pd->dir == PF_IN) { 8019 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD, ifp, &m0, inp, 8020 &pd->act) != PF_PASS) 8021 goto bad; 8022 else if (m0 == NULL) 8023 goto done; 8024 if (m0->m_len < sizeof(struct ip6_hdr)) { 8025 DPFPRINTF(PF_DEBUG_URGENT, 8026 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 8027 __func__)); 8028 goto bad; 8029 } 8030 ip6 = mtod(m0, struct ip6_hdr *); 8031 } 8032 8033 if (ifp->if_flags & IFF_LOOPBACK) 8034 m0->m_flags |= M_SKIP_FIREWALL; 8035 8036 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 8037 ~ifp->if_hwassist) { 8038 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 8039 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 8040 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 8041 } 8042 8043 /* 8044 * If the packet is too large for the outgoing interface, 8045 * send back an icmp6 error. 8046 */ 8047 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 8048 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 8049 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 8050 md = m0; 8051 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 8052 if (md != NULL) 8053 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 8054 } 8055 else { 8056 in6_ifstat_inc(ifp, ifs6_in_toobig); 8057 if (r_rt != PF_DUPTO) { 8058 if (s && s->nat_rule != NULL) 8059 PACKET_UNDO_NAT(m0, pd, 8060 ((caddr_t)ip6 - m0->m_data) + 8061 sizeof(struct ip6_hdr), s); 8062 8063 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 8064 } else 8065 goto bad; 8066 } 8067 8068 done: 8069 if (r_rt != PF_DUPTO) 8070 *m = NULL; 8071 return; 8072 8073 bad_locked: 8074 if (s) 8075 PF_STATE_UNLOCK(s); 8076 bad: 8077 m_freem(m0); 8078 goto done; 8079 } 8080 #endif /* INET6 */ 8081 8082 /* 8083 * FreeBSD supports cksum offloads for the following drivers. 8084 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 8085 * 8086 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 8087 * network driver performed cksum including pseudo header, need to verify 8088 * csum_data 8089 * CSUM_DATA_VALID : 8090 * network driver performed cksum, needs to additional pseudo header 8091 * cksum computation with partial csum_data(i.e. lack of H/W support for 8092 * pseudo header, for instance sk(4) and possibly gem(4)) 8093 * 8094 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 8095 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 8096 * TCP/UDP layer. 8097 * Also, set csum_data to 0xffff to force cksum validation. 8098 */ 8099 static int 8100 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 8101 { 8102 u_int16_t sum = 0; 8103 int hw_assist = 0; 8104 struct ip *ip; 8105 8106 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 8107 return (1); 8108 if (m->m_pkthdr.len < off + len) 8109 return (1); 8110 8111 switch (p) { 8112 case IPPROTO_TCP: 8113 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8114 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8115 sum = m->m_pkthdr.csum_data; 8116 } else { 8117 ip = mtod(m, struct ip *); 8118 sum = in_pseudo(ip->ip_src.s_addr, 8119 ip->ip_dst.s_addr, htonl((u_short)len + 8120 m->m_pkthdr.csum_data + IPPROTO_TCP)); 8121 } 8122 sum ^= 0xffff; 8123 ++hw_assist; 8124 } 8125 break; 8126 case IPPROTO_UDP: 8127 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8128 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8129 sum = m->m_pkthdr.csum_data; 8130 } else { 8131 ip = mtod(m, struct ip *); 8132 sum = in_pseudo(ip->ip_src.s_addr, 8133 ip->ip_dst.s_addr, htonl((u_short)len + 8134 m->m_pkthdr.csum_data + IPPROTO_UDP)); 8135 } 8136 sum ^= 0xffff; 8137 ++hw_assist; 8138 } 8139 break; 8140 case IPPROTO_ICMP: 8141 #ifdef INET6 8142 case IPPROTO_ICMPV6: 8143 #endif /* INET6 */ 8144 break; 8145 default: 8146 return (1); 8147 } 8148 8149 if (!hw_assist) { 8150 switch (af) { 8151 case AF_INET: 8152 if (p == IPPROTO_ICMP) { 8153 if (m->m_len < off) 8154 return (1); 8155 m->m_data += off; 8156 m->m_len -= off; 8157 sum = in_cksum(m, len); 8158 m->m_data -= off; 8159 m->m_len += off; 8160 } else { 8161 if (m->m_len < sizeof(struct ip)) 8162 return (1); 8163 sum = in4_cksum(m, p, off, len); 8164 } 8165 break; 8166 #ifdef INET6 8167 case AF_INET6: 8168 if (m->m_len < sizeof(struct ip6_hdr)) 8169 return (1); 8170 sum = in6_cksum(m, p, off, len); 8171 break; 8172 #endif /* INET6 */ 8173 } 8174 } 8175 if (sum) { 8176 switch (p) { 8177 case IPPROTO_TCP: 8178 { 8179 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 8180 break; 8181 } 8182 case IPPROTO_UDP: 8183 { 8184 KMOD_UDPSTAT_INC(udps_badsum); 8185 break; 8186 } 8187 #ifdef INET 8188 case IPPROTO_ICMP: 8189 { 8190 KMOD_ICMPSTAT_INC(icps_checksum); 8191 break; 8192 } 8193 #endif 8194 #ifdef INET6 8195 case IPPROTO_ICMPV6: 8196 { 8197 KMOD_ICMP6STAT_INC(icp6s_checksum); 8198 break; 8199 } 8200 #endif /* INET6 */ 8201 } 8202 return (1); 8203 } else { 8204 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 8205 m->m_pkthdr.csum_flags |= 8206 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 8207 m->m_pkthdr.csum_data = 0xffff; 8208 } 8209 } 8210 return (0); 8211 } 8212 8213 static bool 8214 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 8215 const struct pf_kstate *s, struct ip_fw_args *dnflow) 8216 { 8217 int dndir = r->direction; 8218 8219 if (s && dndir == PF_INOUT) { 8220 dndir = s->direction; 8221 } else if (dndir == PF_INOUT) { 8222 /* Assume primary direction. Happens when we've set dnpipe in 8223 * the ethernet level code. */ 8224 dndir = pd->dir; 8225 } 8226 8227 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 8228 return (false); 8229 8230 memset(dnflow, 0, sizeof(*dnflow)); 8231 8232 if (pd->dport != NULL) 8233 dnflow->f_id.dst_port = ntohs(*pd->dport); 8234 if (pd->sport != NULL) 8235 dnflow->f_id.src_port = ntohs(*pd->sport); 8236 8237 if (pd->dir == PF_IN) 8238 dnflow->flags |= IPFW_ARGS_IN; 8239 else 8240 dnflow->flags |= IPFW_ARGS_OUT; 8241 8242 if (pd->dir != dndir && pd->act.dnrpipe) { 8243 dnflow->rule.info = pd->act.dnrpipe; 8244 } 8245 else if (pd->dir == dndir && pd->act.dnpipe) { 8246 dnflow->rule.info = pd->act.dnpipe; 8247 } 8248 else { 8249 return (false); 8250 } 8251 8252 dnflow->rule.info |= IPFW_IS_DUMMYNET; 8253 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 8254 dnflow->rule.info |= IPFW_IS_PIPE; 8255 8256 dnflow->f_id.proto = pd->proto; 8257 dnflow->f_id.extra = dnflow->rule.info; 8258 switch (pd->af) { 8259 case AF_INET: 8260 dnflow->f_id.addr_type = 4; 8261 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 8262 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 8263 break; 8264 case AF_INET6: 8265 dnflow->flags |= IPFW_ARGS_IP6; 8266 dnflow->f_id.addr_type = 6; 8267 dnflow->f_id.src_ip6 = pd->src->v6; 8268 dnflow->f_id.dst_ip6 = pd->dst->v6; 8269 break; 8270 } 8271 8272 return (true); 8273 } 8274 8275 int 8276 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8277 struct inpcb *inp) 8278 { 8279 struct pfi_kkif *kif; 8280 struct mbuf *m = *m0; 8281 8282 M_ASSERTPKTHDR(m); 8283 MPASS(ifp->if_vnet == curvnet); 8284 NET_EPOCH_ASSERT(); 8285 8286 if (!V_pf_status.running) 8287 return (PF_PASS); 8288 8289 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8290 8291 if (kif == NULL) { 8292 DPFPRINTF(PF_DEBUG_URGENT, 8293 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 8294 return (PF_DROP); 8295 } 8296 if (kif->pfik_flags & PFI_IFLAG_SKIP) 8297 return (PF_PASS); 8298 8299 if (m->m_flags & M_SKIP_FIREWALL) 8300 return (PF_PASS); 8301 8302 if (__predict_false(! M_WRITABLE(*m0))) { 8303 m = *m0 = m_unshare(*m0, M_NOWAIT); 8304 if (*m0 == NULL) 8305 return (PF_DROP); 8306 } 8307 8308 /* Stateless! */ 8309 return (pf_test_eth_rule(dir, kif, m0)); 8310 } 8311 8312 static __inline void 8313 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 8314 { 8315 struct m_tag *mtag; 8316 8317 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 8318 8319 /* dummynet adds this tag, but pf does not need it, 8320 * and keeping it creates unexpected behavior, 8321 * e.g. in case of divert(4) usage right after dummynet. */ 8322 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8323 if (mtag != NULL) 8324 m_tag_delete(m, mtag); 8325 } 8326 8327 static int 8328 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 8329 struct pf_krule *r, struct mbuf **m0) 8330 { 8331 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 8332 } 8333 8334 static int 8335 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 8336 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 8337 struct mbuf **m0) 8338 { 8339 NET_EPOCH_ASSERT(); 8340 8341 if (pd->act.dnpipe || pd->act.dnrpipe) { 8342 struct ip_fw_args dnflow; 8343 if (ip_dn_io_ptr == NULL) { 8344 m_freem(*m0); 8345 *m0 = NULL; 8346 return (ENOMEM); 8347 } 8348 8349 if (pd->pf_mtag == NULL && 8350 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 8351 m_freem(*m0); 8352 *m0 = NULL; 8353 return (ENOMEM); 8354 } 8355 8356 if (ifp != NULL) { 8357 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 8358 8359 pd->pf_mtag->if_index = ifp->if_index; 8360 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 8361 8362 MPASS(sa != NULL); 8363 8364 switch (pd->af) { 8365 case AF_INET: 8366 memcpy(&pd->pf_mtag->dst, sa, 8367 sizeof(struct sockaddr_in)); 8368 break; 8369 case AF_INET6: 8370 memcpy(&pd->pf_mtag->dst, sa, 8371 sizeof(struct sockaddr_in6)); 8372 break; 8373 } 8374 } 8375 8376 if (s != NULL && s->nat_rule != NULL && 8377 s->nat_rule->action == PF_RDR && 8378 ( 8379 #ifdef INET 8380 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 8381 #endif 8382 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 8383 /* 8384 * If we're redirecting to loopback mark this packet 8385 * as being local. Otherwise it might get dropped 8386 * if dummynet re-injects. 8387 */ 8388 (*m0)->m_pkthdr.rcvif = V_loif; 8389 } 8390 8391 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 8392 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 8393 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 8394 ip_dn_io_ptr(m0, &dnflow); 8395 if (*m0 != NULL) { 8396 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8397 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 8398 } 8399 } 8400 } 8401 8402 return (0); 8403 } 8404 8405 #ifdef INET6 8406 static int 8407 pf_walk_option6(struct mbuf *m, int off, int end, uint32_t *jumbolen, 8408 u_short *reason) 8409 { 8410 struct ip6_opt opt; 8411 struct ip6_opt_jumbo jumbo; 8412 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 8413 8414 while (off < end) { 8415 if (!pf_pull_hdr(m, off, &opt.ip6o_type, sizeof(opt.ip6o_type), 8416 NULL, reason, AF_INET6)) { 8417 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt type")); 8418 return (PF_DROP); 8419 } 8420 if (opt.ip6o_type == IP6OPT_PAD1) { 8421 off++; 8422 continue; 8423 } 8424 if (!pf_pull_hdr(m, off, &opt, sizeof(opt), NULL, reason, 8425 AF_INET6)) { 8426 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt")); 8427 return (PF_DROP); 8428 } 8429 if (off + sizeof(opt) + opt.ip6o_len > end) { 8430 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 long opt")); 8431 REASON_SET(reason, PFRES_IPOPTIONS); 8432 return (PF_DROP); 8433 } 8434 switch (opt.ip6o_type) { 8435 case IP6OPT_JUMBO: 8436 if (*jumbolen != 0) { 8437 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple jumbo")); 8438 REASON_SET(reason, PFRES_IPOPTIONS); 8439 return (PF_DROP); 8440 } 8441 if (ntohs(h->ip6_plen) != 0) { 8442 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 bad jumbo plen")); 8443 REASON_SET(reason, PFRES_IPOPTIONS); 8444 return (PF_DROP); 8445 } 8446 if (!pf_pull_hdr(m, off, &jumbo, sizeof(jumbo), NULL, 8447 reason, AF_INET6)) { 8448 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbo")); 8449 return (PF_DROP); 8450 } 8451 memcpy(jumbolen, jumbo.ip6oj_jumbo_len, 8452 sizeof(*jumbolen)); 8453 *jumbolen = ntohl(*jumbolen); 8454 if (*jumbolen < IPV6_MAXPACKET) { 8455 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbolen")); 8456 REASON_SET(reason, PFRES_IPOPTIONS); 8457 return (PF_DROP); 8458 } 8459 break; 8460 default: 8461 break; 8462 } 8463 off += sizeof(opt) + opt.ip6o_len; 8464 } 8465 8466 return (PF_PASS); 8467 } 8468 8469 int 8470 pf_walk_header6(struct mbuf *m, struct ip6_hdr *h, int *off, int *extoff, 8471 int *fragoff, uint8_t *nxt, uint32_t *jumbolen, u_short *reason) 8472 { 8473 struct ip6_frag frag; 8474 struct ip6_ext ext; 8475 struct ip6_rthdr rthdr; 8476 int rthdr_cnt = 0; 8477 8478 *off += sizeof(struct ip6_hdr); 8479 *extoff = *fragoff = 0; 8480 *nxt = h->ip6_nxt; 8481 *jumbolen = 0; 8482 for (;;) { 8483 switch (*nxt) { 8484 case IPPROTO_FRAGMENT: 8485 if (*fragoff != 0) { 8486 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple fragment")); 8487 REASON_SET(reason, PFRES_FRAG); 8488 return (PF_DROP); 8489 } 8490 /* jumbo payload packets cannot be fragmented */ 8491 if (*jumbolen != 0) { 8492 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 fragmented jumbo")); 8493 REASON_SET(reason, PFRES_FRAG); 8494 return (PF_DROP); 8495 } 8496 if (!pf_pull_hdr(m, *off, &frag, sizeof(frag), NULL, 8497 reason, AF_INET6)) { 8498 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short fragment")); 8499 return (PF_DROP); 8500 } 8501 *fragoff = *off; 8502 /* stop walking over non initial fragments */ 8503 if ((frag.ip6f_offlg & IP6F_OFF_MASK) != 0) 8504 return (PF_PASS); 8505 *off += sizeof(frag); 8506 *nxt = frag.ip6f_nxt; 8507 break; 8508 case IPPROTO_ROUTING: 8509 if (rthdr_cnt++) { 8510 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple rthdr")); 8511 REASON_SET(reason, PFRES_IPOPTIONS); 8512 return (PF_DROP); 8513 } 8514 if (!pf_pull_hdr(m, *off, &rthdr, sizeof(rthdr), NULL, 8515 reason, AF_INET6)) { 8516 /* fragments may be short */ 8517 if (*fragoff != 0) { 8518 *off = *fragoff; 8519 *nxt = IPPROTO_FRAGMENT; 8520 return (PF_PASS); 8521 } 8522 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short rthdr")); 8523 return (PF_DROP); 8524 } 8525 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8526 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 rthdr0")); 8527 REASON_SET(reason, PFRES_IPOPTIONS); 8528 return (PF_DROP); 8529 } 8530 /* FALLTHROUGH */ 8531 case IPPROTO_AH: 8532 case IPPROTO_HOPOPTS: 8533 case IPPROTO_DSTOPTS: 8534 if (!pf_pull_hdr(m, *off, &ext, sizeof(ext), NULL, 8535 reason, AF_INET6)) { 8536 /* fragments may be short */ 8537 if (*fragoff != 0) { 8538 *off = *fragoff; 8539 *nxt = IPPROTO_FRAGMENT; 8540 return (PF_PASS); 8541 } 8542 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short exthdr")); 8543 return (PF_DROP); 8544 } 8545 /* reassembly needs the ext header before the frag */ 8546 if (*fragoff == 0) 8547 *extoff = *off; 8548 if (*nxt == IPPROTO_HOPOPTS && *fragoff == 0) { 8549 if (pf_walk_option6(m, *off + sizeof(ext), 8550 *off + (ext.ip6e_len + 1) * 8, jumbolen, 8551 reason) != PF_PASS) 8552 return (PF_DROP); 8553 if (ntohs(h->ip6_plen) == 0 && *jumbolen != 0) { 8554 DPFPRINTF(PF_DEBUG_MISC, 8555 ("IPv6 missing jumbo")); 8556 REASON_SET(reason, PFRES_IPOPTIONS); 8557 return (PF_DROP); 8558 } 8559 } 8560 if (*nxt == IPPROTO_AH) 8561 *off += (ext.ip6e_len + 2) * 4; 8562 else 8563 *off += (ext.ip6e_len + 1) * 8; 8564 *nxt = ext.ip6e_nxt; 8565 break; 8566 case IPPROTO_TCP: 8567 case IPPROTO_UDP: 8568 case IPPROTO_SCTP: 8569 case IPPROTO_ICMPV6: 8570 /* fragments may be short, ignore inner header then */ 8571 if (*fragoff != 0 && ntohs(h->ip6_plen) < *off + 8572 (*nxt == IPPROTO_TCP ? sizeof(struct tcphdr) : 8573 *nxt == IPPROTO_UDP ? sizeof(struct udphdr) : 8574 *nxt == IPPROTO_SCTP ? sizeof(struct sctphdr) : 8575 sizeof(struct icmp6_hdr))) { 8576 *off = *fragoff; 8577 *nxt = IPPROTO_FRAGMENT; 8578 } 8579 /* FALLTHROUGH */ 8580 default: 8581 return (PF_PASS); 8582 } 8583 } 8584 } 8585 #endif 8586 8587 static void 8588 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 8589 { 8590 memset(pd, 0, sizeof(*pd)); 8591 pd->pf_mtag = pf_find_mtag(m); 8592 pd->m = m; 8593 } 8594 8595 static int 8596 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 8597 u_short *action, u_short *reason, struct pfi_kkif *kif, 8598 struct pf_rule_actions *default_actions) 8599 { 8600 pd->af = af; 8601 pd->dir = dir; 8602 pd->kif = kif; 8603 pd->m = *m0; 8604 pd->sidx = (dir == PF_IN) ? 0 : 1; 8605 pd->didx = (dir == PF_IN) ? 1 : 0; 8606 8607 TAILQ_INIT(&pd->sctp_multihome_jobs); 8608 if (default_actions != NULL) 8609 memcpy(&pd->act, default_actions, sizeof(pd->act)); 8610 8611 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 8612 pd->act.dnpipe = pd->pf_mtag->dnpipe; 8613 pd->act.flags = pd->pf_mtag->dnflags; 8614 } 8615 8616 switch (af) { 8617 #ifdef INET 8618 case AF_INET: { 8619 struct ip *h; 8620 8621 if (__predict_false((*m0)->m_len < sizeof(struct ip)) && 8622 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 8623 DPFPRINTF(PF_DEBUG_URGENT, 8624 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 8625 *action = PF_DROP; 8626 REASON_SET(reason, PFRES_SHORT); 8627 return (-1); 8628 } 8629 8630 if (pf_normalize_ip(m0, reason, pd) != PF_PASS) { 8631 /* We do IP header normalization and packet reassembly here */ 8632 *action = PF_DROP; 8633 return (-1); 8634 } 8635 pd->m = *m0; 8636 8637 h = mtod(pd->m, struct ip *); 8638 pd->off = h->ip_hl << 2; 8639 if (pd->off < (int)sizeof(*h)) { 8640 *action = PF_DROP; 8641 REASON_SET(reason, PFRES_SHORT); 8642 return (-1); 8643 } 8644 pd->src = (struct pf_addr *)&h->ip_src; 8645 pd->dst = (struct pf_addr *)&h->ip_dst; 8646 pd->ip_sum = &h->ip_sum; 8647 pd->proto_sum = NULL; 8648 pd->virtual_proto = pd->proto = h->ip_p; 8649 pd->tos = h->ip_tos; 8650 pd->ttl = h->ip_ttl; 8651 pd->tot_len = ntohs(h->ip_len); 8652 pd->act.rtableid = -1; 8653 8654 if (h->ip_hl > 5) /* has options */ 8655 pd->badopts++; 8656 8657 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) 8658 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8659 8660 break; 8661 } 8662 #endif 8663 #ifdef INET6 8664 case AF_INET6: { 8665 struct ip6_hdr *h; 8666 int fragoff; 8667 uint32_t jumbolen; 8668 uint8_t nxt; 8669 8670 if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) && 8671 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 8672 DPFPRINTF(PF_DEBUG_URGENT, 8673 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 8674 ", pullup failed\n")); 8675 *action = PF_DROP; 8676 REASON_SET(reason, PFRES_SHORT); 8677 return (-1); 8678 } 8679 8680 h = mtod(pd->m, struct ip6_hdr *); 8681 pd->off = 0; 8682 if (pf_walk_header6(pd->m, h, &pd->off, &pd->extoff, &fragoff, &nxt, 8683 &jumbolen, reason) != PF_PASS) { 8684 *action = PF_DROP; 8685 return (-1); 8686 } 8687 8688 h = mtod(pd->m, struct ip6_hdr *); 8689 pd->src = (struct pf_addr *)&h->ip6_src; 8690 pd->dst = (struct pf_addr *)&h->ip6_dst; 8691 pd->ip_sum = NULL; 8692 pd->proto_sum = NULL; 8693 pd->tos = IPV6_DSCP(h); 8694 pd->ttl = h->ip6_hlim; 8695 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8696 pd->virtual_proto = pd->proto = h->ip6_nxt; 8697 pd->act.rtableid = -1; 8698 8699 if (fragoff != 0) 8700 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8701 8702 /* 8703 * we do not support jumbogram. if we keep going, zero ip6_plen 8704 * will do something bad, so drop the packet for now. 8705 */ 8706 if (htons(h->ip6_plen) == 0) { 8707 *action = PF_DROP; 8708 return (-1); 8709 } 8710 8711 /* We do IP header normalization and packet reassembly here */ 8712 if (pf_normalize_ip6(m0, fragoff, reason, pd) != 8713 PF_PASS) { 8714 *action = PF_DROP; 8715 return (-1); 8716 } 8717 pd->m = *m0; 8718 if (pd->m == NULL) { 8719 /* packet sits in reassembly queue, no error */ 8720 *action = PF_PASS; 8721 return (-1); 8722 } 8723 8724 /* Update pointers into the packet. */ 8725 h = mtod(pd->m, struct ip6_hdr *); 8726 pd->src = (struct pf_addr *)&h->ip6_src; 8727 pd->dst = (struct pf_addr *)&h->ip6_dst; 8728 8729 /* 8730 * Reassembly may have changed the next protocol from fragment 8731 * to something else, so update. 8732 */ 8733 pd->virtual_proto = pd->proto = h->ip6_nxt; 8734 pd->off = 0; 8735 8736 if (pf_walk_header6(pd->m, h, &pd->off, &pd->extoff, &fragoff, &nxt, 8737 &jumbolen, reason) != PF_PASS) { 8738 *action = PF_DROP; 8739 return (-1); 8740 } 8741 8742 if (fragoff != 0) 8743 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8744 8745 break; 8746 } 8747 #endif 8748 default: 8749 panic("pf_setup_pdesc called with illegal af %u", af); 8750 } 8751 8752 switch (pd->virtual_proto) { 8753 case IPPROTO_TCP: { 8754 struct tcphdr *th = &pd->hdr.tcp; 8755 8756 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), action, 8757 reason, af)) { 8758 *action = PF_DROP; 8759 REASON_SET(reason, PFRES_SHORT); 8760 return (-1); 8761 } 8762 pd->hdrlen = sizeof(*th); 8763 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 8764 pd->sport = &th->th_sport; 8765 pd->dport = &th->th_dport; 8766 break; 8767 } 8768 case IPPROTO_UDP: { 8769 struct udphdr *uh = &pd->hdr.udp; 8770 8771 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), action, 8772 reason, af)) { 8773 *action = PF_DROP; 8774 REASON_SET(reason, PFRES_SHORT); 8775 return (-1); 8776 } 8777 pd->hdrlen = sizeof(*uh); 8778 if (uh->uh_dport == 0 || 8779 ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off || 8780 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 8781 *action = PF_DROP; 8782 REASON_SET(reason, PFRES_SHORT); 8783 return (-1); 8784 } 8785 pd->sport = &uh->uh_sport; 8786 pd->dport = &uh->uh_dport; 8787 break; 8788 } 8789 case IPPROTO_SCTP: { 8790 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 8791 action, reason, af)) { 8792 *action = PF_DROP; 8793 REASON_SET(reason, PFRES_SHORT); 8794 return (-1); 8795 } 8796 pd->hdrlen = sizeof(pd->hdr.sctp); 8797 pd->p_len = pd->tot_len - pd->off; 8798 8799 pd->sport = &pd->hdr.sctp.src_port; 8800 pd->dport = &pd->hdr.sctp.dest_port; 8801 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 8802 *action = PF_DROP; 8803 REASON_SET(reason, PFRES_SHORT); 8804 return (-1); 8805 } 8806 if (pf_scan_sctp(pd) != PF_PASS) { 8807 *action = PF_DROP; 8808 REASON_SET(reason, PFRES_SHORT); 8809 return (-1); 8810 } 8811 break; 8812 } 8813 case IPPROTO_ICMP: { 8814 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 8815 action, reason, af)) { 8816 *action = PF_DROP; 8817 REASON_SET(reason, PFRES_SHORT); 8818 return (-1); 8819 } 8820 pd->hdrlen = ICMP_MINLEN; 8821 break; 8822 } 8823 #ifdef INET6 8824 case IPPROTO_ICMPV6: { 8825 size_t icmp_hlen = sizeof(struct icmp6_hdr); 8826 8827 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 8828 action, reason, af)) { 8829 *action = PF_DROP; 8830 REASON_SET(reason, PFRES_SHORT); 8831 return (-1); 8832 } 8833 /* ICMP headers we look further into to match state */ 8834 switch (pd->hdr.icmp6.icmp6_type) { 8835 case MLD_LISTENER_QUERY: 8836 case MLD_LISTENER_REPORT: 8837 icmp_hlen = sizeof(struct mld_hdr); 8838 break; 8839 case ND_NEIGHBOR_SOLICIT: 8840 case ND_NEIGHBOR_ADVERT: 8841 icmp_hlen = sizeof(struct nd_neighbor_solicit); 8842 break; 8843 } 8844 if (icmp_hlen > sizeof(struct icmp6_hdr) && 8845 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 8846 action, reason, af)) { 8847 *action = PF_DROP; 8848 REASON_SET(reason, PFRES_SHORT); 8849 return (-1); 8850 } 8851 pd->hdrlen = icmp_hlen; 8852 break; 8853 } 8854 #endif 8855 } 8856 return (0); 8857 } 8858 8859 static void 8860 pf_counters_inc(int action, struct pf_pdesc *pd, 8861 struct pf_kstate *s, struct pf_krule *r, struct pf_krule *a) 8862 { 8863 struct pf_krule *tr; 8864 int dir = pd->dir; 8865 int dirndx; 8866 8867 pf_counter_u64_critical_enter(); 8868 pf_counter_u64_add_protected( 8869 &pd->kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8870 pd->tot_len); 8871 pf_counter_u64_add_protected( 8872 &pd->kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8873 1); 8874 8875 if (action == PF_PASS || r->action == PF_DROP) { 8876 dirndx = (dir == PF_OUT); 8877 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8878 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 8879 pf_update_timestamp(r); 8880 8881 if (a != NULL) { 8882 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8883 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 8884 } 8885 if (s != NULL) { 8886 struct pf_krule_item *ri; 8887 8888 if (s->nat_rule != NULL) { 8889 pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx], 8890 1); 8891 pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx], 8892 pd->tot_len); 8893 } 8894 if (s->src_node != NULL) { 8895 counter_u64_add(s->src_node->packets[dirndx], 8896 1); 8897 counter_u64_add(s->src_node->bytes[dirndx], 8898 pd->tot_len); 8899 } 8900 if (s->nat_src_node != NULL) { 8901 counter_u64_add(s->nat_src_node->packets[dirndx], 8902 1); 8903 counter_u64_add(s->nat_src_node->bytes[dirndx], 8904 pd->tot_len); 8905 } 8906 dirndx = (dir == s->direction) ? 0 : 1; 8907 s->packets[dirndx]++; 8908 s->bytes[dirndx] += pd->tot_len; 8909 8910 SLIST_FOREACH(ri, &s->match_rules, entry) { 8911 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8912 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 8913 } 8914 } 8915 8916 tr = r; 8917 if (s != NULL && s->nat_rule != NULL && 8918 r == &V_pf_default_rule) 8919 tr = s->nat_rule; 8920 8921 if (tr->src.addr.type == PF_ADDR_TABLE) 8922 pfr_update_stats(tr->src.addr.p.tbl, 8923 (s == NULL) ? pd->src : 8924 &s->key[(s->direction == PF_IN)]-> 8925 addr[(s->direction == PF_OUT)], 8926 pd->af, pd->tot_len, dir == PF_OUT, 8927 r->action == PF_PASS, tr->src.neg); 8928 if (tr->dst.addr.type == PF_ADDR_TABLE) 8929 pfr_update_stats(tr->dst.addr.p.tbl, 8930 (s == NULL) ? pd->dst : 8931 &s->key[(s->direction == PF_IN)]-> 8932 addr[(s->direction == PF_IN)], 8933 pd->af, pd->tot_len, dir == PF_OUT, 8934 r->action == PF_PASS, tr->dst.neg); 8935 } 8936 pf_counter_u64_critical_exit(); 8937 } 8938 8939 #if defined(INET) || defined(INET6) 8940 int 8941 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8942 struct inpcb *inp, struct pf_rule_actions *default_actions) 8943 { 8944 struct pfi_kkif *kif; 8945 u_short action, reason = 0; 8946 struct m_tag *mtag; 8947 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 8948 struct pf_kstate *s = NULL; 8949 struct pf_kruleset *ruleset = NULL; 8950 struct pf_pdesc pd; 8951 int use_2nd_queue = 0; 8952 uint16_t tag; 8953 uint8_t rt; 8954 8955 PF_RULES_RLOCK_TRACKER; 8956 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8957 M_ASSERTPKTHDR(*m0); 8958 8959 if (!V_pf_status.running) 8960 return (PF_PASS); 8961 8962 PF_RULES_RLOCK(); 8963 8964 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8965 8966 if (__predict_false(kif == NULL)) { 8967 DPFPRINTF(PF_DEBUG_URGENT, 8968 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 8969 PF_RULES_RUNLOCK(); 8970 return (PF_DROP); 8971 } 8972 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8973 PF_RULES_RUNLOCK(); 8974 return (PF_PASS); 8975 } 8976 8977 if ((*m0)->m_flags & M_SKIP_FIREWALL) { 8978 PF_RULES_RUNLOCK(); 8979 return (PF_PASS); 8980 } 8981 8982 #ifdef INET6 8983 /* 8984 * If we end up changing IP addresses (e.g. binat) the stack may get 8985 * confused and fail to send the icmp6 packet too big error. Just send 8986 * it here, before we do any NAT. 8987 */ 8988 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 8989 IN6_LINKMTU(ifp) < pf_max_frag_size(*m0)) { 8990 PF_RULES_RUNLOCK(); 8991 icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 8992 *m0 = NULL; 8993 return (PF_DROP); 8994 } 8995 #endif 8996 8997 if (__predict_false(! M_WRITABLE(*m0))) { 8998 *m0 = m_unshare(*m0, M_NOWAIT); 8999 if (*m0 == NULL) 9000 return (PF_DROP); 9001 } 9002 9003 pf_init_pdesc(&pd, *m0); 9004 9005 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 9006 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9007 9008 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 9009 pd.pf_mtag->if_idxgen); 9010 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 9011 PF_RULES_RUNLOCK(); 9012 m_freem(*m0); 9013 *m0 = NULL; 9014 return (PF_PASS); 9015 } 9016 PF_RULES_RUNLOCK(); 9017 (ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL); 9018 *m0 = NULL; 9019 return (PF_PASS); 9020 } 9021 9022 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 9023 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 9024 /* Dummynet re-injects packets after they've 9025 * completed their delay. We've already 9026 * processed them, so pass unconditionally. */ 9027 9028 /* But only once. We may see the packet multiple times (e.g. 9029 * PFIL_IN/PFIL_OUT). */ 9030 pf_dummynet_flag_remove(pd.m, pd.pf_mtag); 9031 PF_RULES_RUNLOCK(); 9032 9033 return (PF_PASS); 9034 } 9035 9036 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, 9037 kif, default_actions) == -1) { 9038 if (action != PF_PASS) 9039 pd.act.log |= PF_LOG_FORCE; 9040 goto done; 9041 } 9042 9043 if (__predict_false(ip_divert_ptr != NULL) && 9044 ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 9045 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 9046 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 9047 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 9048 if (pd.pf_mtag == NULL && 9049 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 9050 action = PF_DROP; 9051 goto done; 9052 } 9053 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 9054 } 9055 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 9056 pd.m->m_flags |= M_FASTFWD_OURS; 9057 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9058 } 9059 m_tag_delete(pd.m, mtag); 9060 9061 mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL); 9062 if (mtag != NULL) 9063 m_tag_delete(pd.m, mtag); 9064 } 9065 9066 switch (pd.virtual_proto) { 9067 case PF_VPROTO_FRAGMENT: 9068 /* 9069 * handle fragments that aren't reassembled by 9070 * normalization 9071 */ 9072 if (kif == NULL || r == NULL) /* pflog */ 9073 action = PF_DROP; 9074 else 9075 action = pf_test_rule(&r, &s, &pd, &a, 9076 &ruleset, inp); 9077 if (action != PF_PASS) 9078 REASON_SET(&reason, PFRES_FRAG); 9079 break; 9080 9081 case IPPROTO_TCP: { 9082 /* Respond to SYN with a syncookie. */ 9083 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 9084 pd.dir == PF_IN && pf_synflood_check(&pd)) { 9085 pf_syncookie_send(&pd); 9086 action = PF_DROP; 9087 break; 9088 } 9089 9090 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 9091 use_2nd_queue = 1; 9092 action = pf_normalize_tcp(&pd); 9093 if (action == PF_DROP) 9094 goto done; 9095 action = pf_test_state_tcp(&s, &pd, &reason); 9096 if (action == PF_PASS) { 9097 if (V_pfsync_update_state_ptr != NULL) 9098 V_pfsync_update_state_ptr(s); 9099 r = s->rule; 9100 a = s->anchor; 9101 } else if (s == NULL) { 9102 /* Validate remote SYN|ACK, re-create original SYN if 9103 * valid. */ 9104 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 9105 TH_ACK && pf_syncookie_validate(&pd) && 9106 pd.dir == PF_IN) { 9107 struct mbuf *msyn; 9108 9109 msyn = pf_syncookie_recreate_syn(&pd); 9110 if (msyn == NULL) { 9111 action = PF_DROP; 9112 break; 9113 } 9114 9115 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 9116 &pd.act); 9117 m_freem(msyn); 9118 if (action != PF_PASS) 9119 break; 9120 9121 action = pf_test_state_tcp(&s, &pd, &reason); 9122 if (action != PF_PASS || s == NULL) { 9123 action = PF_DROP; 9124 break; 9125 } 9126 9127 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 9128 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 9129 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 9130 action = pf_synproxy(&pd, &s, &reason); 9131 break; 9132 } else { 9133 action = pf_test_rule(&r, &s, &pd, 9134 &a, &ruleset, inp); 9135 } 9136 } 9137 break; 9138 } 9139 9140 case IPPROTO_UDP: { 9141 action = pf_test_state_udp(&s, &pd); 9142 if (action == PF_PASS) { 9143 if (V_pfsync_update_state_ptr != NULL) 9144 V_pfsync_update_state_ptr(s); 9145 r = s->rule; 9146 a = s->anchor; 9147 } else if (s == NULL) 9148 action = pf_test_rule(&r, &s, &pd, 9149 &a, &ruleset, inp); 9150 break; 9151 } 9152 9153 case IPPROTO_SCTP: { 9154 action = pf_normalize_sctp(&pd); 9155 if (action == PF_DROP) 9156 goto done; 9157 action = pf_test_state_sctp(&s, &pd, &reason); 9158 if (action == PF_PASS) { 9159 if (V_pfsync_update_state_ptr != NULL) 9160 V_pfsync_update_state_ptr(s); 9161 r = s->rule; 9162 a = s->anchor; 9163 } else if (s == NULL) { 9164 action = pf_test_rule(&r, &s, 9165 &pd, &a, &ruleset, inp); 9166 } 9167 break; 9168 } 9169 9170 case IPPROTO_ICMP: { 9171 if (af != AF_INET) { 9172 action = PF_DROP; 9173 REASON_SET(&reason, PFRES_NORM); 9174 DPFPRINTF(PF_DEBUG_MISC, 9175 ("dropping IPv6 packet with ICMPv4 payload")); 9176 goto done; 9177 } 9178 action = pf_test_state_icmp(&s, &pd, &reason); 9179 if (action == PF_PASS) { 9180 if (V_pfsync_update_state_ptr != NULL) 9181 V_pfsync_update_state_ptr(s); 9182 r = s->rule; 9183 a = s->anchor; 9184 } else if (s == NULL) 9185 action = pf_test_rule(&r, &s, &pd, 9186 &a, &ruleset, inp); 9187 break; 9188 } 9189 9190 case IPPROTO_ICMPV6: { 9191 if (af != AF_INET6) { 9192 action = PF_DROP; 9193 REASON_SET(&reason, PFRES_NORM); 9194 DPFPRINTF(PF_DEBUG_MISC, 9195 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 9196 goto done; 9197 } 9198 action = pf_test_state_icmp(&s, &pd, &reason); 9199 if (action == PF_PASS) { 9200 if (V_pfsync_update_state_ptr != NULL) 9201 V_pfsync_update_state_ptr(s); 9202 r = s->rule; 9203 a = s->anchor; 9204 } else if (s == NULL) 9205 action = pf_test_rule(&r, &s, &pd, 9206 &a, &ruleset, inp); 9207 break; 9208 } 9209 9210 default: 9211 action = pf_test_state_other(&s, &pd); 9212 if (action == PF_PASS) { 9213 if (V_pfsync_update_state_ptr != NULL) 9214 V_pfsync_update_state_ptr(s); 9215 r = s->rule; 9216 a = s->anchor; 9217 } else if (s == NULL) 9218 action = pf_test_rule(&r, &s, &pd, 9219 &a, &ruleset, inp); 9220 break; 9221 } 9222 9223 done: 9224 PF_RULES_RUNLOCK(); 9225 9226 if (pd.m == NULL) 9227 goto eat_pkt; 9228 9229 if (action == PF_PASS && pd.badopts && 9230 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 9231 action = PF_DROP; 9232 REASON_SET(&reason, PFRES_IPOPTIONS); 9233 pd.act.log = PF_LOG_FORCE; 9234 DPFPRINTF(PF_DEBUG_MISC, 9235 ("pf: dropping packet with dangerous headers\n")); 9236 } 9237 9238 if (s) { 9239 uint8_t log = pd.act.log; 9240 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 9241 pd.act.log |= log; 9242 tag = s->tag; 9243 rt = s->rt; 9244 } else { 9245 tag = r->tag; 9246 rt = r->rt; 9247 } 9248 9249 if (tag > 0 && pf_tag_packet(&pd, tag)) { 9250 action = PF_DROP; 9251 REASON_SET(&reason, PFRES_MEMORY); 9252 } 9253 9254 pf_scrub(&pd); 9255 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 9256 pf_normalize_mss(&pd); 9257 9258 if (pd.act.rtableid >= 0) 9259 M_SETFIB(pd.m, pd.act.rtableid); 9260 9261 if (pd.act.flags & PFSTATE_SETPRIO) { 9262 if (pd.tos & IPTOS_LOWDELAY) 9263 use_2nd_queue = 1; 9264 if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) { 9265 action = PF_DROP; 9266 REASON_SET(&reason, PFRES_MEMORY); 9267 pd.act.log = PF_LOG_FORCE; 9268 DPFPRINTF(PF_DEBUG_MISC, 9269 ("pf: failed to allocate 802.1q mtag\n")); 9270 } 9271 } 9272 9273 #ifdef ALTQ 9274 if (action == PF_PASS && pd.act.qid) { 9275 if (pd.pf_mtag == NULL && 9276 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 9277 action = PF_DROP; 9278 REASON_SET(&reason, PFRES_MEMORY); 9279 } else { 9280 if (s != NULL) 9281 pd.pf_mtag->qid_hash = pf_state_hash(s); 9282 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 9283 pd.pf_mtag->qid = pd.act.pqid; 9284 else 9285 pd.pf_mtag->qid = pd.act.qid; 9286 /* Add hints for ecn. */ 9287 pd.pf_mtag->hdr = mtod(pd.m, void *); 9288 } 9289 } 9290 #endif /* ALTQ */ 9291 9292 /* 9293 * connections redirected to loopback should not match sockets 9294 * bound specifically to loopback due to security implications, 9295 * see tcp_input() and in_pcblookup_listen(). 9296 */ 9297 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9298 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 9299 (s->nat_rule->action == PF_RDR || 9300 s->nat_rule->action == PF_BINAT) && 9301 pf_is_loopback(af, pd.dst)) 9302 pd.m->m_flags |= M_SKIP_FIREWALL; 9303 9304 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 9305 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 9306 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 9307 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 9308 if (mtag != NULL) { 9309 ((struct pf_divert_mtag *)(mtag+1))->port = 9310 ntohs(r->divert.port); 9311 ((struct pf_divert_mtag *)(mtag+1))->idir = 9312 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 9313 PF_DIVERT_MTAG_DIR_OUT; 9314 9315 if (s) 9316 PF_STATE_UNLOCK(s); 9317 9318 m_tag_prepend(pd.m, mtag); 9319 if (pd.m->m_flags & M_FASTFWD_OURS) { 9320 if (pd.pf_mtag == NULL && 9321 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 9322 action = PF_DROP; 9323 REASON_SET(&reason, PFRES_MEMORY); 9324 pd.act.log = PF_LOG_FORCE; 9325 DPFPRINTF(PF_DEBUG_MISC, 9326 ("pf: failed to allocate tag\n")); 9327 } else { 9328 pd.pf_mtag->flags |= 9329 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9330 pd.m->m_flags &= ~M_FASTFWD_OURS; 9331 } 9332 } 9333 ip_divert_ptr(*m0, dir == PF_IN); 9334 *m0 = NULL; 9335 9336 return (action); 9337 } else { 9338 /* XXX: ipfw has the same behaviour! */ 9339 action = PF_DROP; 9340 REASON_SET(&reason, PFRES_MEMORY); 9341 pd.act.log = PF_LOG_FORCE; 9342 DPFPRINTF(PF_DEBUG_MISC, 9343 ("pf: failed to allocate divert tag\n")); 9344 } 9345 } 9346 /* XXX: Anybody working on it?! */ 9347 if (af == AF_INET6 && r->divert.port) 9348 printf("pf: divert(9) is not supported for IPv6\n"); 9349 9350 /* this flag will need revising if the pkt is forwarded */ 9351 if (pd.pf_mtag) 9352 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 9353 9354 if (pd.act.log) { 9355 struct pf_krule *lr; 9356 struct pf_krule_item *ri; 9357 9358 if (s != NULL && s->nat_rule != NULL && 9359 s->nat_rule->log & PF_LOG_ALL) 9360 lr = s->nat_rule; 9361 else 9362 lr = r; 9363 9364 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9365 PFLOG_PACKET(action, reason, lr, a, 9366 ruleset, &pd, (s == NULL)); 9367 if (s) { 9368 SLIST_FOREACH(ri, &s->match_rules, entry) 9369 if (ri->r->log & PF_LOG_ALL) 9370 PFLOG_PACKET(action, 9371 reason, ri->r, a, ruleset, &pd, 0); 9372 } 9373 } 9374 9375 pf_counters_inc(action, &pd, s, r, a); 9376 9377 switch (action) { 9378 case PF_SYNPROXY_DROP: 9379 m_freem(*m0); 9380 case PF_DEFER: 9381 *m0 = NULL; 9382 action = PF_PASS; 9383 break; 9384 case PF_DROP: 9385 m_freem(*m0); 9386 *m0 = NULL; 9387 break; 9388 default: 9389 if (rt) { 9390 switch (af) { 9391 #ifdef INET 9392 case AF_INET: 9393 /* pf_route() returns unlocked. */ 9394 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 9395 break; 9396 #endif 9397 #ifdef INET6 9398 case AF_INET6: 9399 /* pf_route6() returns unlocked. */ 9400 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 9401 break; 9402 #endif 9403 } 9404 goto out; 9405 } 9406 if (pf_dummynet(&pd, s, r, m0) != 0) { 9407 action = PF_DROP; 9408 REASON_SET(&reason, PFRES_MEMORY); 9409 } 9410 break; 9411 } 9412 9413 eat_pkt: 9414 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 9415 9416 if (s && action != PF_DROP) { 9417 if (!s->if_index_in && dir == PF_IN) 9418 s->if_index_in = ifp->if_index; 9419 else if (!s->if_index_out && dir == PF_OUT) 9420 s->if_index_out = ifp->if_index; 9421 } 9422 9423 if (s) 9424 PF_STATE_UNLOCK(s); 9425 9426 #ifdef INET6 9427 /* If reassembled packet passed, create new fragments. */ 9428 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 9429 (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 9430 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 9431 #endif 9432 9433 out: 9434 pf_sctp_multihome_delayed(&pd, kif, s, action); 9435 9436 return (action); 9437 } 9438 #endif /* INET || INET6 */ 9439