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, skw); 1437 if (skw != sks) 1438 uma_zfree(V_pf_state_key_z, sks); 1439 if (idx == PF_SK_STACK) 1440 pf_detach_state(s); 1441 return (EEXIST); /* collision! */ 1442 } 1443 } 1444 PF_HASHROW_UNLOCK(ih); 1445 } 1446 uma_zfree(V_pf_state_key_z, sk); 1447 s->key[idx] = cur; 1448 } else { 1449 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1450 s->key[idx] = sk; 1451 } 1452 1453 stateattach: 1454 /* List is sorted, if-bound states before floating. */ 1455 if (s->kif == V_pfi_all) 1456 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1457 else 1458 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1459 1460 if (olds) { 1461 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1462 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1463 key_list[idx]); 1464 olds = NULL; 1465 } 1466 1467 /* 1468 * Attach done. See how should we (or should not?) 1469 * attach a second key. 1470 */ 1471 if (sks == skw) { 1472 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1473 idx = PF_SK_STACK; 1474 sks = NULL; 1475 goto stateattach; 1476 } else if (sks != NULL) { 1477 /* 1478 * Continue attaching with stack key. 1479 */ 1480 sk = sks; 1481 kh = khs; 1482 idx = PF_SK_STACK; 1483 sks = NULL; 1484 goto keyattach; 1485 } 1486 1487 PF_STATE_LOCK(s); 1488 KEYS_UNLOCK(); 1489 1490 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1491 ("%s failure", __func__)); 1492 1493 return (0); 1494 #undef KEYS_UNLOCK 1495 } 1496 1497 static void 1498 pf_detach_state(struct pf_kstate *s) 1499 { 1500 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1501 struct pf_keyhash *kh; 1502 1503 NET_EPOCH_ASSERT(); 1504 MPASS(s->timeout >= PFTM_MAX); 1505 1506 pf_sctp_multihome_detach_addr(s); 1507 1508 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1509 V_pflow_export_state_ptr(s); 1510 1511 if (sks != NULL) { 1512 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1513 PF_HASHROW_LOCK(kh); 1514 if (s->key[PF_SK_STACK] != NULL) 1515 pf_state_key_detach(s, PF_SK_STACK); 1516 /* 1517 * If both point to same key, then we are done. 1518 */ 1519 if (sks == s->key[PF_SK_WIRE]) { 1520 pf_state_key_detach(s, PF_SK_WIRE); 1521 PF_HASHROW_UNLOCK(kh); 1522 return; 1523 } 1524 PF_HASHROW_UNLOCK(kh); 1525 } 1526 1527 if (s->key[PF_SK_WIRE] != NULL) { 1528 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1529 PF_HASHROW_LOCK(kh); 1530 if (s->key[PF_SK_WIRE] != NULL) 1531 pf_state_key_detach(s, PF_SK_WIRE); 1532 PF_HASHROW_UNLOCK(kh); 1533 } 1534 } 1535 1536 static void 1537 pf_state_key_detach(struct pf_kstate *s, int idx) 1538 { 1539 struct pf_state_key *sk = s->key[idx]; 1540 #ifdef INVARIANTS 1541 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1542 1543 PF_HASHROW_ASSERT(kh); 1544 #endif 1545 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1546 s->key[idx] = NULL; 1547 1548 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1549 LIST_REMOVE(sk, entry); 1550 uma_zfree(V_pf_state_key_z, sk); 1551 } 1552 } 1553 1554 static int 1555 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1556 { 1557 struct pf_state_key *sk = mem; 1558 1559 bzero(sk, sizeof(struct pf_state_key_cmp)); 1560 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1561 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1562 1563 return (0); 1564 } 1565 1566 static int 1567 pf_state_key_addr_setup(struct pf_pdesc *pd, 1568 struct pf_state_key_cmp *key, int multi) 1569 { 1570 struct pf_addr *saddr = pd->src; 1571 struct pf_addr *daddr = pd->dst; 1572 #ifdef INET6 1573 struct nd_neighbor_solicit nd; 1574 struct pf_addr *target; 1575 u_short action, reason; 1576 1577 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1578 goto copy; 1579 1580 switch (pd->hdr.icmp6.icmp6_type) { 1581 case ND_NEIGHBOR_SOLICIT: 1582 if (multi) 1583 return (-1); 1584 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1585 return (-1); 1586 target = (struct pf_addr *)&nd.nd_ns_target; 1587 daddr = target; 1588 break; 1589 case ND_NEIGHBOR_ADVERT: 1590 if (multi) 1591 return (-1); 1592 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1593 return (-1); 1594 target = (struct pf_addr *)&nd.nd_ns_target; 1595 saddr = target; 1596 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1597 key->addr[pd->didx].addr32[0] = 0; 1598 key->addr[pd->didx].addr32[1] = 0; 1599 key->addr[pd->didx].addr32[2] = 0; 1600 key->addr[pd->didx].addr32[3] = 0; 1601 daddr = NULL; /* overwritten */ 1602 } 1603 break; 1604 default: 1605 if (multi == PF_ICMP_MULTI_LINK) { 1606 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1607 key->addr[pd->sidx].addr32[1] = 0; 1608 key->addr[pd->sidx].addr32[2] = 0; 1609 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1610 saddr = NULL; /* overwritten */ 1611 } 1612 } 1613 copy: 1614 #endif 1615 if (saddr) 1616 PF_ACPY(&key->addr[pd->sidx], saddr, pd->af); 1617 if (daddr) 1618 PF_ACPY(&key->addr[pd->didx], daddr, pd->af); 1619 1620 return (0); 1621 } 1622 1623 struct pf_state_key * 1624 pf_state_key_setup(struct pf_pdesc *pd, 1625 struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport, 1626 u_int16_t dport) 1627 { 1628 struct pf_state_key *sk; 1629 1630 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1631 if (sk == NULL) 1632 return (NULL); 1633 1634 if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)sk, 1635 0)) { 1636 uma_zfree(V_pf_state_key_z, sk); 1637 return (NULL); 1638 } 1639 1640 sk->port[pd->sidx] = sport; 1641 sk->port[pd->didx] = dport; 1642 sk->proto = pd->proto; 1643 sk->af = pd->af; 1644 1645 return (sk); 1646 } 1647 1648 struct pf_state_key * 1649 pf_state_key_clone(const struct pf_state_key *orig) 1650 { 1651 struct pf_state_key *sk; 1652 1653 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1654 if (sk == NULL) 1655 return (NULL); 1656 1657 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1658 1659 return (sk); 1660 } 1661 1662 int 1663 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1664 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1665 { 1666 struct pf_idhash *ih; 1667 struct pf_kstate *cur; 1668 int error; 1669 1670 NET_EPOCH_ASSERT(); 1671 1672 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1673 ("%s: sks not pristine", __func__)); 1674 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1675 ("%s: skw not pristine", __func__)); 1676 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1677 1678 s->kif = kif; 1679 s->orig_kif = orig_kif; 1680 1681 if (s->id == 0 && s->creatorid == 0) { 1682 s->id = alloc_unr64(&V_pf_stateid); 1683 s->id = htobe64(s->id); 1684 s->creatorid = V_pf_status.hostid; 1685 } 1686 1687 /* Returns with ID locked on success. */ 1688 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1689 return (error); 1690 1691 ih = &V_pf_idhash[PF_IDHASH(s)]; 1692 PF_HASHROW_ASSERT(ih); 1693 LIST_FOREACH(cur, &ih->states, entry) 1694 if (cur->id == s->id && cur->creatorid == s->creatorid) 1695 break; 1696 1697 if (cur != NULL) { 1698 s->timeout = PFTM_UNLINKED; 1699 PF_HASHROW_UNLOCK(ih); 1700 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1701 printf("pf: state ID collision: " 1702 "id: %016llx creatorid: %08x\n", 1703 (unsigned long long)be64toh(s->id), 1704 ntohl(s->creatorid)); 1705 } 1706 pf_detach_state(s); 1707 return (EEXIST); 1708 } 1709 LIST_INSERT_HEAD(&ih->states, s, entry); 1710 /* One for keys, one for ID hash. */ 1711 refcount_init(&s->refs, 2); 1712 1713 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1714 if (V_pfsync_insert_state_ptr != NULL) 1715 V_pfsync_insert_state_ptr(s); 1716 1717 /* Returns locked. */ 1718 return (0); 1719 } 1720 1721 /* 1722 * Find state by ID: returns with locked row on success. 1723 */ 1724 struct pf_kstate * 1725 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1726 { 1727 struct pf_idhash *ih; 1728 struct pf_kstate *s; 1729 1730 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1731 1732 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1733 1734 PF_HASHROW_LOCK(ih); 1735 LIST_FOREACH(s, &ih->states, entry) 1736 if (s->id == id && s->creatorid == creatorid) 1737 break; 1738 1739 if (s == NULL) 1740 PF_HASHROW_UNLOCK(ih); 1741 1742 return (s); 1743 } 1744 1745 /* 1746 * Find state by key. 1747 * Returns with ID hash slot locked on success. 1748 */ 1749 static struct pf_kstate * 1750 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1751 u_int dir) 1752 { 1753 struct pf_keyhash *kh; 1754 struct pf_state_key *sk; 1755 struct pf_kstate *s; 1756 int idx; 1757 1758 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1759 1760 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1761 1762 PF_HASHROW_LOCK(kh); 1763 LIST_FOREACH(sk, &kh->keys, entry) 1764 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1765 break; 1766 if (sk == NULL) { 1767 PF_HASHROW_UNLOCK(kh); 1768 return (NULL); 1769 } 1770 1771 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1772 1773 /* List is sorted, if-bound states before floating ones. */ 1774 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1775 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1776 PF_STATE_LOCK(s); 1777 PF_HASHROW_UNLOCK(kh); 1778 if (__predict_false(s->timeout >= PFTM_MAX)) { 1779 /* 1780 * State is either being processed by 1781 * pf_unlink_state() in an other thread, or 1782 * is scheduled for immediate expiry. 1783 */ 1784 PF_STATE_UNLOCK(s); 1785 return (NULL); 1786 } 1787 return (s); 1788 } 1789 PF_HASHROW_UNLOCK(kh); 1790 1791 return (NULL); 1792 } 1793 1794 /* 1795 * Returns with ID hash slot locked on success. 1796 */ 1797 struct pf_kstate * 1798 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1799 { 1800 struct pf_keyhash *kh; 1801 struct pf_state_key *sk; 1802 struct pf_kstate *s, *ret = NULL; 1803 int idx, inout = 0; 1804 1805 if (more != NULL) 1806 *more = 0; 1807 1808 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1809 1810 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1811 1812 PF_HASHROW_LOCK(kh); 1813 LIST_FOREACH(sk, &kh->keys, entry) 1814 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1815 break; 1816 if (sk == NULL) { 1817 PF_HASHROW_UNLOCK(kh); 1818 return (NULL); 1819 } 1820 switch (dir) { 1821 case PF_IN: 1822 idx = PF_SK_WIRE; 1823 break; 1824 case PF_OUT: 1825 idx = PF_SK_STACK; 1826 break; 1827 case PF_INOUT: 1828 idx = PF_SK_WIRE; 1829 inout = 1; 1830 break; 1831 default: 1832 panic("%s: dir %u", __func__, dir); 1833 } 1834 second_run: 1835 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1836 if (more == NULL) { 1837 PF_STATE_LOCK(s); 1838 PF_HASHROW_UNLOCK(kh); 1839 return (s); 1840 } 1841 1842 if (ret) 1843 (*more)++; 1844 else { 1845 ret = s; 1846 PF_STATE_LOCK(s); 1847 } 1848 } 1849 if (inout == 1) { 1850 inout = 0; 1851 idx = PF_SK_STACK; 1852 goto second_run; 1853 } 1854 PF_HASHROW_UNLOCK(kh); 1855 1856 return (ret); 1857 } 1858 1859 /* 1860 * FIXME 1861 * This routine is inefficient -- locks the state only to unlock immediately on 1862 * return. 1863 * It is racy -- after the state is unlocked nothing stops other threads from 1864 * removing it. 1865 */ 1866 bool 1867 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 1868 { 1869 struct pf_kstate *s; 1870 1871 s = pf_find_state_all(key, dir, NULL); 1872 if (s != NULL) { 1873 PF_STATE_UNLOCK(s); 1874 return (true); 1875 } 1876 return (false); 1877 } 1878 1879 struct pf_udp_mapping * 1880 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 1881 struct pf_addr *nat_addr, uint16_t nat_port) 1882 { 1883 struct pf_udp_mapping *mapping; 1884 1885 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 1886 if (mapping == NULL) 1887 return (NULL); 1888 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 1889 mapping->endpoints[0].port = src_port; 1890 mapping->endpoints[0].af = af; 1891 mapping->endpoints[0].mapping = mapping; 1892 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 1893 mapping->endpoints[1].port = nat_port; 1894 mapping->endpoints[1].af = af; 1895 mapping->endpoints[1].mapping = mapping; 1896 refcount_init(&mapping->refs, 1); 1897 return (mapping); 1898 } 1899 1900 int 1901 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 1902 { 1903 struct pf_udpendpointhash *h0, *h1; 1904 struct pf_udp_endpoint *endpoint; 1905 int ret = EEXIST; 1906 1907 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1908 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1909 if (h0 == h1) { 1910 PF_HASHROW_LOCK(h0); 1911 } else if (h0 < h1) { 1912 PF_HASHROW_LOCK(h0); 1913 PF_HASHROW_LOCK(h1); 1914 } else { 1915 PF_HASHROW_LOCK(h1); 1916 PF_HASHROW_LOCK(h0); 1917 } 1918 1919 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 1920 if (bcmp(endpoint, &mapping->endpoints[0], 1921 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1922 break; 1923 } 1924 if (endpoint != NULL) 1925 goto cleanup; 1926 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 1927 if (bcmp(endpoint, &mapping->endpoints[1], 1928 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1929 break; 1930 } 1931 if (endpoint != NULL) 1932 goto cleanup; 1933 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 1934 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 1935 ret = 0; 1936 1937 cleanup: 1938 if (h0 != h1) { 1939 PF_HASHROW_UNLOCK(h0); 1940 PF_HASHROW_UNLOCK(h1); 1941 } else { 1942 PF_HASHROW_UNLOCK(h0); 1943 } 1944 return (ret); 1945 } 1946 1947 void 1948 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 1949 { 1950 /* refcount is synchronized on the source endpoint's row lock */ 1951 struct pf_udpendpointhash *h0, *h1; 1952 1953 if (mapping == NULL) 1954 return; 1955 1956 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1957 PF_HASHROW_LOCK(h0); 1958 if (refcount_release(&mapping->refs)) { 1959 LIST_REMOVE(&mapping->endpoints[0], entry); 1960 PF_HASHROW_UNLOCK(h0); 1961 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1962 PF_HASHROW_LOCK(h1); 1963 LIST_REMOVE(&mapping->endpoints[1], entry); 1964 PF_HASHROW_UNLOCK(h1); 1965 1966 uma_zfree(V_pf_udp_mapping_z, mapping); 1967 } else { 1968 PF_HASHROW_UNLOCK(h0); 1969 } 1970 } 1971 1972 1973 struct pf_udp_mapping * 1974 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 1975 { 1976 struct pf_udpendpointhash *uh; 1977 struct pf_udp_endpoint *endpoint; 1978 1979 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 1980 1981 PF_HASHROW_LOCK(uh); 1982 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 1983 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 1984 bcmp(endpoint, &endpoint->mapping->endpoints[0], 1985 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1986 break; 1987 } 1988 if (endpoint == NULL) { 1989 PF_HASHROW_UNLOCK(uh); 1990 return (NULL); 1991 } 1992 refcount_acquire(&endpoint->mapping->refs); 1993 PF_HASHROW_UNLOCK(uh); 1994 return (endpoint->mapping); 1995 } 1996 /* END state table stuff */ 1997 1998 static void 1999 pf_send(struct pf_send_entry *pfse) 2000 { 2001 2002 PF_SENDQ_LOCK(); 2003 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2004 PF_SENDQ_UNLOCK(); 2005 swi_sched(V_pf_swi_cookie, 0); 2006 } 2007 2008 static bool 2009 pf_isforlocal(struct mbuf *m, int af) 2010 { 2011 switch (af) { 2012 #ifdef INET 2013 case AF_INET: { 2014 struct ip *ip = mtod(m, struct ip *); 2015 2016 return (in_localip(ip->ip_dst)); 2017 } 2018 #endif 2019 #ifdef INET6 2020 case AF_INET6: { 2021 struct ip6_hdr *ip6; 2022 struct in6_ifaddr *ia; 2023 ip6 = mtod(m, struct ip6_hdr *); 2024 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2025 if (ia == NULL) 2026 return (false); 2027 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2028 } 2029 #endif 2030 } 2031 2032 return (false); 2033 } 2034 2035 int 2036 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2037 int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type) 2038 { 2039 /* 2040 * ICMP types marked with PF_OUT are typically responses to 2041 * PF_IN, and will match states in the opposite direction. 2042 * PF_IN ICMP types need to match a state with that type. 2043 */ 2044 *icmp_dir = PF_OUT; 2045 *multi = PF_ICMP_MULTI_LINK; 2046 /* Queries (and responses) */ 2047 switch (pd->af) { 2048 #ifdef INET 2049 case AF_INET: 2050 switch (type) { 2051 case ICMP_ECHO: 2052 *icmp_dir = PF_IN; 2053 case ICMP_ECHOREPLY: 2054 *virtual_type = ICMP_ECHO; 2055 *virtual_id = pd->hdr.icmp.icmp_id; 2056 break; 2057 2058 case ICMP_TSTAMP: 2059 *icmp_dir = PF_IN; 2060 case ICMP_TSTAMPREPLY: 2061 *virtual_type = ICMP_TSTAMP; 2062 *virtual_id = pd->hdr.icmp.icmp_id; 2063 break; 2064 2065 case ICMP_IREQ: 2066 *icmp_dir = PF_IN; 2067 case ICMP_IREQREPLY: 2068 *virtual_type = ICMP_IREQ; 2069 *virtual_id = pd->hdr.icmp.icmp_id; 2070 break; 2071 2072 case ICMP_MASKREQ: 2073 *icmp_dir = PF_IN; 2074 case ICMP_MASKREPLY: 2075 *virtual_type = ICMP_MASKREQ; 2076 *virtual_id = pd->hdr.icmp.icmp_id; 2077 break; 2078 2079 case ICMP_IPV6_WHEREAREYOU: 2080 *icmp_dir = PF_IN; 2081 case ICMP_IPV6_IAMHERE: 2082 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2083 *virtual_id = 0; /* Nothing sane to match on! */ 2084 break; 2085 2086 case ICMP_MOBILE_REGREQUEST: 2087 *icmp_dir = PF_IN; 2088 case ICMP_MOBILE_REGREPLY: 2089 *virtual_type = ICMP_MOBILE_REGREQUEST; 2090 *virtual_id = 0; /* Nothing sane to match on! */ 2091 break; 2092 2093 case ICMP_ROUTERSOLICIT: 2094 *icmp_dir = PF_IN; 2095 case ICMP_ROUTERADVERT: 2096 *virtual_type = ICMP_ROUTERSOLICIT; 2097 *virtual_id = 0; /* Nothing sane to match on! */ 2098 break; 2099 2100 /* These ICMP types map to other connections */ 2101 case ICMP_UNREACH: 2102 case ICMP_SOURCEQUENCH: 2103 case ICMP_REDIRECT: 2104 case ICMP_TIMXCEED: 2105 case ICMP_PARAMPROB: 2106 /* These will not be used, but set them anyway */ 2107 *icmp_dir = PF_IN; 2108 *virtual_type = type; 2109 *virtual_id = 0; 2110 HTONS(*virtual_type); 2111 return (1); /* These types match to another state */ 2112 2113 /* 2114 * All remaining ICMP types get their own states, 2115 * and will only match in one direction. 2116 */ 2117 default: 2118 *icmp_dir = PF_IN; 2119 *virtual_type = type; 2120 *virtual_id = 0; 2121 break; 2122 } 2123 break; 2124 #endif /* INET */ 2125 #ifdef INET6 2126 case AF_INET6: 2127 switch (type) { 2128 case ICMP6_ECHO_REQUEST: 2129 *icmp_dir = PF_IN; 2130 case ICMP6_ECHO_REPLY: 2131 *virtual_type = ICMP6_ECHO_REQUEST; 2132 *virtual_id = pd->hdr.icmp6.icmp6_id; 2133 break; 2134 2135 case MLD_LISTENER_QUERY: 2136 case MLD_LISTENER_REPORT: { 2137 /* 2138 * Listener Report can be sent by clients 2139 * without an associated Listener Query. 2140 * In addition to that, when Report is sent as a 2141 * reply to a Query its source and destination 2142 * address are different. 2143 */ 2144 *icmp_dir = PF_IN; 2145 *virtual_type = MLD_LISTENER_QUERY; 2146 *virtual_id = 0; 2147 break; 2148 } 2149 case MLD_MTRACE: 2150 *icmp_dir = PF_IN; 2151 case MLD_MTRACE_RESP: 2152 *virtual_type = MLD_MTRACE; 2153 *virtual_id = 0; /* Nothing sane to match on! */ 2154 break; 2155 2156 case ND_NEIGHBOR_SOLICIT: 2157 *icmp_dir = PF_IN; 2158 case ND_NEIGHBOR_ADVERT: { 2159 *virtual_type = ND_NEIGHBOR_SOLICIT; 2160 *virtual_id = 0; 2161 break; 2162 } 2163 2164 /* 2165 * These ICMP types map to other connections. 2166 * ND_REDIRECT can't be in this list because the triggering 2167 * packet header is optional. 2168 */ 2169 case ICMP6_DST_UNREACH: 2170 case ICMP6_PACKET_TOO_BIG: 2171 case ICMP6_TIME_EXCEEDED: 2172 case ICMP6_PARAM_PROB: 2173 /* These will not be used, but set them anyway */ 2174 *icmp_dir = PF_IN; 2175 *virtual_type = type; 2176 *virtual_id = 0; 2177 HTONS(*virtual_type); 2178 return (1); /* These types match to another state */ 2179 /* 2180 * All remaining ICMP6 types get their own states, 2181 * and will only match in one direction. 2182 */ 2183 default: 2184 *icmp_dir = PF_IN; 2185 *virtual_type = type; 2186 *virtual_id = 0; 2187 break; 2188 } 2189 break; 2190 #endif /* INET6 */ 2191 } 2192 HTONS(*virtual_type); 2193 return (0); /* These types match to their own state */ 2194 } 2195 2196 void 2197 pf_intr(void *v) 2198 { 2199 struct epoch_tracker et; 2200 struct pf_send_head queue; 2201 struct pf_send_entry *pfse, *next; 2202 2203 CURVNET_SET((struct vnet *)v); 2204 2205 PF_SENDQ_LOCK(); 2206 queue = V_pf_sendqueue; 2207 STAILQ_INIT(&V_pf_sendqueue); 2208 PF_SENDQ_UNLOCK(); 2209 2210 NET_EPOCH_ENTER(et); 2211 2212 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2213 switch (pfse->pfse_type) { 2214 #ifdef INET 2215 case PFSE_IP: { 2216 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2217 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2218 ("%s: rcvif != loif", __func__)); 2219 2220 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2221 pfse->pfse_m->m_pkthdr.csum_flags |= 2222 CSUM_IP_VALID | CSUM_IP_CHECKED; 2223 ip_input(pfse->pfse_m); 2224 } else { 2225 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2226 NULL); 2227 } 2228 break; 2229 } 2230 case PFSE_ICMP: 2231 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2232 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2233 break; 2234 #endif /* INET */ 2235 #ifdef INET6 2236 case PFSE_IP6: 2237 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2238 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2239 ("%s: rcvif != loif", __func__)); 2240 2241 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL | 2242 M_LOOP; 2243 ip6_input(pfse->pfse_m); 2244 } else { 2245 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2246 NULL, NULL); 2247 } 2248 break; 2249 case PFSE_ICMP6: 2250 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2251 pfse->icmpopts.code, pfse->icmpopts.mtu); 2252 break; 2253 #endif /* INET6 */ 2254 default: 2255 panic("%s: unknown type", __func__); 2256 } 2257 free(pfse, M_PFTEMP); 2258 } 2259 NET_EPOCH_EXIT(et); 2260 CURVNET_RESTORE(); 2261 } 2262 2263 #define pf_purge_thread_period (hz / 10) 2264 2265 #ifdef PF_WANT_32_TO_64_COUNTER 2266 static void 2267 pf_status_counter_u64_periodic(void) 2268 { 2269 2270 PF_RULES_RASSERT(); 2271 2272 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2273 return; 2274 } 2275 2276 for (int i = 0; i < FCNT_MAX; i++) { 2277 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2278 } 2279 } 2280 2281 static void 2282 pf_kif_counter_u64_periodic(void) 2283 { 2284 struct pfi_kkif *kif; 2285 size_t r, run; 2286 2287 PF_RULES_RASSERT(); 2288 2289 if (__predict_false(V_pf_allkifcount == 0)) { 2290 return; 2291 } 2292 2293 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2294 return; 2295 } 2296 2297 run = V_pf_allkifcount / 10; 2298 if (run < 5) 2299 run = 5; 2300 2301 for (r = 0; r < run; r++) { 2302 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2303 if (kif == NULL) { 2304 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2305 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2306 break; 2307 } 2308 2309 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2310 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2311 2312 for (int i = 0; i < 2; i++) { 2313 for (int j = 0; j < 2; j++) { 2314 for (int k = 0; k < 2; k++) { 2315 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2316 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2317 } 2318 } 2319 } 2320 } 2321 } 2322 2323 static void 2324 pf_rule_counter_u64_periodic(void) 2325 { 2326 struct pf_krule *rule; 2327 size_t r, run; 2328 2329 PF_RULES_RASSERT(); 2330 2331 if (__predict_false(V_pf_allrulecount == 0)) { 2332 return; 2333 } 2334 2335 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2336 return; 2337 } 2338 2339 run = V_pf_allrulecount / 10; 2340 if (run < 5) 2341 run = 5; 2342 2343 for (r = 0; r < run; r++) { 2344 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2345 if (rule == NULL) { 2346 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2347 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2348 break; 2349 } 2350 2351 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2352 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2353 2354 pf_counter_u64_periodic(&rule->evaluations); 2355 for (int i = 0; i < 2; i++) { 2356 pf_counter_u64_periodic(&rule->packets[i]); 2357 pf_counter_u64_periodic(&rule->bytes[i]); 2358 } 2359 } 2360 } 2361 2362 static void 2363 pf_counter_u64_periodic_main(void) 2364 { 2365 PF_RULES_RLOCK_TRACKER; 2366 2367 V_pf_counter_periodic_iter++; 2368 2369 PF_RULES_RLOCK(); 2370 pf_counter_u64_critical_enter(); 2371 pf_status_counter_u64_periodic(); 2372 pf_kif_counter_u64_periodic(); 2373 pf_rule_counter_u64_periodic(); 2374 pf_counter_u64_critical_exit(); 2375 PF_RULES_RUNLOCK(); 2376 } 2377 #else 2378 #define pf_counter_u64_periodic_main() do { } while (0) 2379 #endif 2380 2381 void 2382 pf_purge_thread(void *unused __unused) 2383 { 2384 struct epoch_tracker et; 2385 2386 VNET_ITERATOR_DECL(vnet_iter); 2387 2388 sx_xlock(&pf_end_lock); 2389 while (pf_end_threads == 0) { 2390 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2391 2392 VNET_LIST_RLOCK(); 2393 NET_EPOCH_ENTER(et); 2394 VNET_FOREACH(vnet_iter) { 2395 CURVNET_SET(vnet_iter); 2396 2397 /* Wait until V_pf_default_rule is initialized. */ 2398 if (V_pf_vnet_active == 0) { 2399 CURVNET_RESTORE(); 2400 continue; 2401 } 2402 2403 pf_counter_u64_periodic_main(); 2404 2405 /* 2406 * Process 1/interval fraction of the state 2407 * table every run. 2408 */ 2409 V_pf_purge_idx = 2410 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2411 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2412 2413 /* 2414 * Purge other expired types every 2415 * PFTM_INTERVAL seconds. 2416 */ 2417 if (V_pf_purge_idx == 0) { 2418 /* 2419 * Order is important: 2420 * - states and src nodes reference rules 2421 * - states and rules reference kifs 2422 */ 2423 pf_purge_expired_fragments(); 2424 pf_purge_expired_src_nodes(); 2425 pf_purge_unlinked_rules(); 2426 pfi_kkif_purge(); 2427 } 2428 CURVNET_RESTORE(); 2429 } 2430 NET_EPOCH_EXIT(et); 2431 VNET_LIST_RUNLOCK(); 2432 } 2433 2434 pf_end_threads++; 2435 sx_xunlock(&pf_end_lock); 2436 kproc_exit(0); 2437 } 2438 2439 void 2440 pf_unload_vnet_purge(void) 2441 { 2442 2443 /* 2444 * To cleanse up all kifs and rules we need 2445 * two runs: first one clears reference flags, 2446 * then pf_purge_expired_states() doesn't 2447 * raise them, and then second run frees. 2448 */ 2449 pf_purge_unlinked_rules(); 2450 pfi_kkif_purge(); 2451 2452 /* 2453 * Now purge everything. 2454 */ 2455 pf_purge_expired_states(0, V_pf_hashmask); 2456 pf_purge_fragments(UINT_MAX); 2457 pf_purge_expired_src_nodes(); 2458 2459 /* 2460 * Now all kifs & rules should be unreferenced, 2461 * thus should be successfully freed. 2462 */ 2463 pf_purge_unlinked_rules(); 2464 pfi_kkif_purge(); 2465 } 2466 2467 u_int32_t 2468 pf_state_expires(const struct pf_kstate *state) 2469 { 2470 u_int32_t timeout; 2471 u_int32_t start; 2472 u_int32_t end; 2473 u_int32_t states; 2474 2475 /* handle all PFTM_* > PFTM_MAX here */ 2476 if (state->timeout == PFTM_PURGE) 2477 return (time_uptime); 2478 KASSERT(state->timeout != PFTM_UNLINKED, 2479 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2480 KASSERT((state->timeout < PFTM_MAX), 2481 ("pf_state_expires: timeout > PFTM_MAX")); 2482 timeout = state->rule->timeout[state->timeout]; 2483 if (!timeout) 2484 timeout = V_pf_default_rule.timeout[state->timeout]; 2485 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2486 if (start && state->rule != &V_pf_default_rule) { 2487 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2488 states = counter_u64_fetch(state->rule->states_cur); 2489 } else { 2490 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2491 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2492 states = V_pf_status.states; 2493 } 2494 if (end && states > start && start < end) { 2495 if (states < end) { 2496 timeout = (u_int64_t)timeout * (end - states) / 2497 (end - start); 2498 return ((state->expire / 1000) + timeout); 2499 } 2500 else 2501 return (time_uptime); 2502 } 2503 return ((state->expire / 1000) + timeout); 2504 } 2505 2506 void 2507 pf_purge_expired_src_nodes(void) 2508 { 2509 struct pf_ksrc_node_list freelist; 2510 struct pf_srchash *sh; 2511 struct pf_ksrc_node *cur, *next; 2512 int i; 2513 2514 LIST_INIT(&freelist); 2515 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2516 PF_HASHROW_LOCK(sh); 2517 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2518 if (cur->states == 0 && cur->expire <= time_uptime) { 2519 pf_unlink_src_node(cur); 2520 LIST_INSERT_HEAD(&freelist, cur, entry); 2521 } else if (cur->rule != NULL) 2522 cur->rule->rule_ref |= PFRULE_REFS; 2523 PF_HASHROW_UNLOCK(sh); 2524 } 2525 2526 pf_free_src_nodes(&freelist); 2527 2528 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2529 } 2530 2531 static void 2532 pf_src_tree_remove_state(struct pf_kstate *s) 2533 { 2534 struct pf_ksrc_node *sn; 2535 uint32_t timeout; 2536 2537 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2538 s->rule->timeout[PFTM_SRC_NODE] : 2539 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2540 2541 if (s->src_node != NULL) { 2542 sn = s->src_node; 2543 PF_SRC_NODE_LOCK(sn); 2544 if (s->src.tcp_est) 2545 --sn->conn; 2546 if (--sn->states == 0) 2547 sn->expire = time_uptime + timeout; 2548 PF_SRC_NODE_UNLOCK(sn); 2549 } 2550 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2551 sn = s->nat_src_node; 2552 PF_SRC_NODE_LOCK(sn); 2553 if (--sn->states == 0) 2554 sn->expire = time_uptime + timeout; 2555 PF_SRC_NODE_UNLOCK(sn); 2556 } 2557 s->src_node = s->nat_src_node = NULL; 2558 } 2559 2560 /* 2561 * Unlink and potentilly free a state. Function may be 2562 * called with ID hash row locked, but always returns 2563 * unlocked, since it needs to go through key hash locking. 2564 */ 2565 int 2566 pf_unlink_state(struct pf_kstate *s) 2567 { 2568 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2569 2570 NET_EPOCH_ASSERT(); 2571 PF_HASHROW_ASSERT(ih); 2572 2573 if (s->timeout == PFTM_UNLINKED) { 2574 /* 2575 * State is being processed 2576 * by pf_unlink_state() in 2577 * an other thread. 2578 */ 2579 PF_HASHROW_UNLOCK(ih); 2580 return (0); /* XXXGL: undefined actually */ 2581 } 2582 2583 if (s->src.state == PF_TCPS_PROXY_DST) { 2584 /* XXX wire key the right one? */ 2585 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2586 &s->key[PF_SK_WIRE]->addr[1], 2587 &s->key[PF_SK_WIRE]->addr[0], 2588 s->key[PF_SK_WIRE]->port[1], 2589 s->key[PF_SK_WIRE]->port[0], 2590 s->src.seqhi, s->src.seqlo + 1, 2591 TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0, 2592 s->act.rtableid); 2593 } 2594 2595 LIST_REMOVE(s, entry); 2596 pf_src_tree_remove_state(s); 2597 2598 if (V_pfsync_delete_state_ptr != NULL) 2599 V_pfsync_delete_state_ptr(s); 2600 2601 STATE_DEC_COUNTERS(s); 2602 2603 s->timeout = PFTM_UNLINKED; 2604 2605 /* Ensure we remove it from the list of halfopen states, if needed. */ 2606 if (s->key[PF_SK_STACK] != NULL && 2607 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2608 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2609 2610 PF_HASHROW_UNLOCK(ih); 2611 2612 pf_detach_state(s); 2613 2614 pf_udp_mapping_release(s->udp_mapping); 2615 2616 /* pf_state_insert() initialises refs to 2 */ 2617 return (pf_release_staten(s, 2)); 2618 } 2619 2620 struct pf_kstate * 2621 pf_alloc_state(int flags) 2622 { 2623 2624 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2625 } 2626 2627 void 2628 pf_free_state(struct pf_kstate *cur) 2629 { 2630 struct pf_krule_item *ri; 2631 2632 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2633 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2634 cur->timeout)); 2635 2636 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2637 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2638 free(ri, M_PF_RULE_ITEM); 2639 } 2640 2641 pf_normalize_tcp_cleanup(cur); 2642 uma_zfree(V_pf_state_z, cur); 2643 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2644 } 2645 2646 /* 2647 * Called only from pf_purge_thread(), thus serialized. 2648 */ 2649 static u_int 2650 pf_purge_expired_states(u_int i, int maxcheck) 2651 { 2652 struct pf_idhash *ih; 2653 struct pf_kstate *s; 2654 struct pf_krule_item *mrm; 2655 size_t count __unused; 2656 2657 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2658 2659 /* 2660 * Go through hash and unlink states that expire now. 2661 */ 2662 while (maxcheck > 0) { 2663 count = 0; 2664 ih = &V_pf_idhash[i]; 2665 2666 /* only take the lock if we expect to do work */ 2667 if (!LIST_EMPTY(&ih->states)) { 2668 relock: 2669 PF_HASHROW_LOCK(ih); 2670 LIST_FOREACH(s, &ih->states, entry) { 2671 if (pf_state_expires(s) <= time_uptime) { 2672 V_pf_status.states -= 2673 pf_unlink_state(s); 2674 goto relock; 2675 } 2676 s->rule->rule_ref |= PFRULE_REFS; 2677 if (s->nat_rule != NULL) 2678 s->nat_rule->rule_ref |= PFRULE_REFS; 2679 if (s->anchor != NULL) 2680 s->anchor->rule_ref |= PFRULE_REFS; 2681 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2682 SLIST_FOREACH(mrm, &s->match_rules, entry) 2683 mrm->r->rule_ref |= PFRULE_REFS; 2684 if (s->rt_kif) 2685 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2686 count++; 2687 } 2688 PF_HASHROW_UNLOCK(ih); 2689 } 2690 2691 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2692 2693 /* Return when we hit end of hash. */ 2694 if (++i > V_pf_hashmask) { 2695 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2696 return (0); 2697 } 2698 2699 maxcheck--; 2700 } 2701 2702 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2703 2704 return (i); 2705 } 2706 2707 static void 2708 pf_purge_unlinked_rules(void) 2709 { 2710 struct pf_krulequeue tmpq; 2711 struct pf_krule *r, *r1; 2712 2713 /* 2714 * If we have overloading task pending, then we'd 2715 * better skip purging this time. There is a tiny 2716 * probability that overloading task references 2717 * an already unlinked rule. 2718 */ 2719 PF_OVERLOADQ_LOCK(); 2720 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2721 PF_OVERLOADQ_UNLOCK(); 2722 return; 2723 } 2724 PF_OVERLOADQ_UNLOCK(); 2725 2726 /* 2727 * Do naive mark-and-sweep garbage collecting of old rules. 2728 * Reference flag is raised by pf_purge_expired_states() 2729 * and pf_purge_expired_src_nodes(). 2730 * 2731 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2732 * use a temporary queue. 2733 */ 2734 TAILQ_INIT(&tmpq); 2735 PF_UNLNKDRULES_LOCK(); 2736 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2737 if (!(r->rule_ref & PFRULE_REFS)) { 2738 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2739 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2740 } else 2741 r->rule_ref &= ~PFRULE_REFS; 2742 } 2743 PF_UNLNKDRULES_UNLOCK(); 2744 2745 if (!TAILQ_EMPTY(&tmpq)) { 2746 PF_CONFIG_LOCK(); 2747 PF_RULES_WLOCK(); 2748 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2749 TAILQ_REMOVE(&tmpq, r, entries); 2750 pf_free_rule(r); 2751 } 2752 PF_RULES_WUNLOCK(); 2753 PF_CONFIG_UNLOCK(); 2754 } 2755 } 2756 2757 void 2758 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2759 { 2760 switch (af) { 2761 #ifdef INET 2762 case AF_INET: { 2763 u_int32_t a = ntohl(addr->addr32[0]); 2764 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2765 (a>>8)&255, a&255); 2766 if (p) { 2767 p = ntohs(p); 2768 printf(":%u", p); 2769 } 2770 break; 2771 } 2772 #endif /* INET */ 2773 #ifdef INET6 2774 case AF_INET6: { 2775 u_int16_t b; 2776 u_int8_t i, curstart, curend, maxstart, maxend; 2777 curstart = curend = maxstart = maxend = 255; 2778 for (i = 0; i < 8; i++) { 2779 if (!addr->addr16[i]) { 2780 if (curstart == 255) 2781 curstart = i; 2782 curend = i; 2783 } else { 2784 if ((curend - curstart) > 2785 (maxend - maxstart)) { 2786 maxstart = curstart; 2787 maxend = curend; 2788 } 2789 curstart = curend = 255; 2790 } 2791 } 2792 if ((curend - curstart) > 2793 (maxend - maxstart)) { 2794 maxstart = curstart; 2795 maxend = curend; 2796 } 2797 for (i = 0; i < 8; i++) { 2798 if (i >= maxstart && i <= maxend) { 2799 if (i == 0) 2800 printf(":"); 2801 if (i == maxend) 2802 printf(":"); 2803 } else { 2804 b = ntohs(addr->addr16[i]); 2805 printf("%x", b); 2806 if (i < 7) 2807 printf(":"); 2808 } 2809 } 2810 if (p) { 2811 p = ntohs(p); 2812 printf("[%u]", p); 2813 } 2814 break; 2815 } 2816 #endif /* INET6 */ 2817 } 2818 } 2819 2820 void 2821 pf_print_state(struct pf_kstate *s) 2822 { 2823 pf_print_state_parts(s, NULL, NULL); 2824 } 2825 2826 static void 2827 pf_print_state_parts(struct pf_kstate *s, 2828 struct pf_state_key *skwp, struct pf_state_key *sksp) 2829 { 2830 struct pf_state_key *skw, *sks; 2831 u_int8_t proto, dir; 2832 2833 /* Do our best to fill these, but they're skipped if NULL */ 2834 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2835 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2836 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2837 dir = s ? s->direction : 0; 2838 2839 switch (proto) { 2840 case IPPROTO_IPV4: 2841 printf("IPv4"); 2842 break; 2843 case IPPROTO_IPV6: 2844 printf("IPv6"); 2845 break; 2846 case IPPROTO_TCP: 2847 printf("TCP"); 2848 break; 2849 case IPPROTO_UDP: 2850 printf("UDP"); 2851 break; 2852 case IPPROTO_ICMP: 2853 printf("ICMP"); 2854 break; 2855 case IPPROTO_ICMPV6: 2856 printf("ICMPv6"); 2857 break; 2858 default: 2859 printf("%u", proto); 2860 break; 2861 } 2862 switch (dir) { 2863 case PF_IN: 2864 printf(" in"); 2865 break; 2866 case PF_OUT: 2867 printf(" out"); 2868 break; 2869 } 2870 if (skw) { 2871 printf(" wire: "); 2872 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2873 printf(" "); 2874 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2875 } 2876 if (sks) { 2877 printf(" stack: "); 2878 if (sks != skw) { 2879 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2880 printf(" "); 2881 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2882 } else 2883 printf("-"); 2884 } 2885 if (s) { 2886 if (proto == IPPROTO_TCP) { 2887 printf(" [lo=%u high=%u win=%u modulator=%u", 2888 s->src.seqlo, s->src.seqhi, 2889 s->src.max_win, s->src.seqdiff); 2890 if (s->src.wscale && s->dst.wscale) 2891 printf(" wscale=%u", 2892 s->src.wscale & PF_WSCALE_MASK); 2893 printf("]"); 2894 printf(" [lo=%u high=%u win=%u modulator=%u", 2895 s->dst.seqlo, s->dst.seqhi, 2896 s->dst.max_win, s->dst.seqdiff); 2897 if (s->src.wscale && s->dst.wscale) 2898 printf(" wscale=%u", 2899 s->dst.wscale & PF_WSCALE_MASK); 2900 printf("]"); 2901 } 2902 printf(" %u:%u", s->src.state, s->dst.state); 2903 if (s->rule) 2904 printf(" @%d", s->rule->nr); 2905 } 2906 } 2907 2908 void 2909 pf_print_flags(u_int8_t f) 2910 { 2911 if (f) 2912 printf(" "); 2913 if (f & TH_FIN) 2914 printf("F"); 2915 if (f & TH_SYN) 2916 printf("S"); 2917 if (f & TH_RST) 2918 printf("R"); 2919 if (f & TH_PUSH) 2920 printf("P"); 2921 if (f & TH_ACK) 2922 printf("A"); 2923 if (f & TH_URG) 2924 printf("U"); 2925 if (f & TH_ECE) 2926 printf("E"); 2927 if (f & TH_CWR) 2928 printf("W"); 2929 } 2930 2931 #define PF_SET_SKIP_STEPS(i) \ 2932 do { \ 2933 while (head[i] != cur) { \ 2934 head[i]->skip[i] = cur; \ 2935 head[i] = TAILQ_NEXT(head[i], entries); \ 2936 } \ 2937 } while (0) 2938 2939 void 2940 pf_calc_skip_steps(struct pf_krulequeue *rules) 2941 { 2942 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2943 int i; 2944 2945 cur = TAILQ_FIRST(rules); 2946 prev = cur; 2947 for (i = 0; i < PF_SKIP_COUNT; ++i) 2948 head[i] = cur; 2949 while (cur != NULL) { 2950 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2951 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2952 if (cur->direction != prev->direction) 2953 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2954 if (cur->af != prev->af) 2955 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2956 if (cur->proto != prev->proto) 2957 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2958 if (cur->src.neg != prev->src.neg || 2959 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2960 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2961 if (cur->dst.neg != prev->dst.neg || 2962 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2963 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2964 if (cur->src.port[0] != prev->src.port[0] || 2965 cur->src.port[1] != prev->src.port[1] || 2966 cur->src.port_op != prev->src.port_op) 2967 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2968 if (cur->dst.port[0] != prev->dst.port[0] || 2969 cur->dst.port[1] != prev->dst.port[1] || 2970 cur->dst.port_op != prev->dst.port_op) 2971 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2972 2973 prev = cur; 2974 cur = TAILQ_NEXT(cur, entries); 2975 } 2976 for (i = 0; i < PF_SKIP_COUNT; ++i) 2977 PF_SET_SKIP_STEPS(i); 2978 } 2979 2980 int 2981 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2982 { 2983 if (aw1->type != aw2->type) 2984 return (1); 2985 switch (aw1->type) { 2986 case PF_ADDR_ADDRMASK: 2987 case PF_ADDR_RANGE: 2988 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2989 return (1); 2990 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2991 return (1); 2992 return (0); 2993 case PF_ADDR_DYNIFTL: 2994 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2995 case PF_ADDR_NOROUTE: 2996 case PF_ADDR_URPFFAILED: 2997 return (0); 2998 case PF_ADDR_TABLE: 2999 return (aw1->p.tbl != aw2->p.tbl); 3000 default: 3001 printf("invalid address type: %d\n", aw1->type); 3002 return (1); 3003 } 3004 } 3005 3006 /** 3007 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3008 * header isn't always a full checksum. In some cases (i.e. output) it's a 3009 * pseudo-header checksum, which is a partial checksum over src/dst IP 3010 * addresses, protocol number and length. 3011 * 3012 * That means we have the following cases: 3013 * * Input or forwarding: we don't have TSO, the checksum fields are full 3014 * checksums, we need to update the checksum whenever we change anything. 3015 * * Output (i.e. the checksum is a pseudo-header checksum): 3016 * x The field being updated is src/dst address or affects the length of 3017 * the packet. We need to update the pseudo-header checksum (note that this 3018 * checksum is not ones' complement). 3019 * x Some other field is being modified (e.g. src/dst port numbers): We 3020 * don't have to update anything. 3021 **/ 3022 u_int16_t 3023 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3024 { 3025 u_int32_t x; 3026 3027 x = cksum + old - new; 3028 x = (x + (x >> 16)) & 0xffff; 3029 3030 /* optimise: eliminate a branch when not udp */ 3031 if (udp && cksum == 0x0000) 3032 return cksum; 3033 if (udp && x == 0x0000) 3034 x = 0xffff; 3035 3036 return (u_int16_t)(x); 3037 } 3038 3039 static void 3040 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 3041 u_int8_t udp) 3042 { 3043 u_int16_t old = htons(hi ? (*f << 8) : *f); 3044 u_int16_t new = htons(hi ? ( v << 8) : v); 3045 3046 if (*f == v) 3047 return; 3048 3049 *f = v; 3050 3051 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3052 return; 3053 3054 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 3055 } 3056 3057 void 3058 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 3059 bool hi, u_int8_t udp) 3060 { 3061 u_int8_t *fb = (u_int8_t *)f; 3062 u_int8_t *vb = (u_int8_t *)&v; 3063 3064 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3065 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3066 } 3067 3068 void 3069 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 3070 bool hi, u_int8_t udp) 3071 { 3072 u_int8_t *fb = (u_int8_t *)f; 3073 u_int8_t *vb = (u_int8_t *)&v; 3074 3075 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3076 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3077 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3078 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3079 } 3080 3081 u_int16_t 3082 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3083 u_int16_t new, u_int8_t udp) 3084 { 3085 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3086 return (cksum); 3087 3088 return (pf_cksum_fixup(cksum, old, new, udp)); 3089 } 3090 3091 static void 3092 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 3093 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 3094 sa_family_t af) 3095 { 3096 struct pf_addr ao; 3097 u_int16_t po = *p; 3098 3099 PF_ACPY(&ao, a, af); 3100 PF_ACPY(a, an, af); 3101 3102 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3103 *pc = ~*pc; 3104 3105 *p = pn; 3106 3107 switch (af) { 3108 #ifdef INET 3109 case AF_INET: 3110 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3111 ao.addr16[0], an->addr16[0], 0), 3112 ao.addr16[1], an->addr16[1], 0); 3113 *p = pn; 3114 3115 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 3116 ao.addr16[0], an->addr16[0], u), 3117 ao.addr16[1], an->addr16[1], u); 3118 3119 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3120 break; 3121 #endif /* INET */ 3122 #ifdef INET6 3123 case AF_INET6: 3124 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3125 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3126 pf_cksum_fixup(pf_cksum_fixup(*pc, 3127 ao.addr16[0], an->addr16[0], u), 3128 ao.addr16[1], an->addr16[1], u), 3129 ao.addr16[2], an->addr16[2], u), 3130 ao.addr16[3], an->addr16[3], u), 3131 ao.addr16[4], an->addr16[4], u), 3132 ao.addr16[5], an->addr16[5], u), 3133 ao.addr16[6], an->addr16[6], u), 3134 ao.addr16[7], an->addr16[7], u); 3135 3136 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3137 break; 3138 #endif /* INET6 */ 3139 } 3140 3141 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3142 CSUM_DELAY_DATA_IPV6)) { 3143 *pc = ~*pc; 3144 if (! *pc) 3145 *pc = 0xffff; 3146 } 3147 } 3148 3149 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3150 void 3151 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3152 { 3153 u_int32_t ao; 3154 3155 memcpy(&ao, a, sizeof(ao)); 3156 memcpy(a, &an, sizeof(u_int32_t)); 3157 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3158 ao % 65536, an % 65536, u); 3159 } 3160 3161 void 3162 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3163 { 3164 u_int32_t ao; 3165 3166 memcpy(&ao, a, sizeof(ao)); 3167 memcpy(a, &an, sizeof(u_int32_t)); 3168 3169 *c = pf_proto_cksum_fixup(m, 3170 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3171 ao % 65536, an % 65536, udp); 3172 } 3173 3174 #ifdef INET6 3175 static void 3176 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3177 { 3178 struct pf_addr ao; 3179 3180 PF_ACPY(&ao, a, AF_INET6); 3181 PF_ACPY(a, an, AF_INET6); 3182 3183 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3184 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3185 pf_cksum_fixup(pf_cksum_fixup(*c, 3186 ao.addr16[0], an->addr16[0], u), 3187 ao.addr16[1], an->addr16[1], u), 3188 ao.addr16[2], an->addr16[2], u), 3189 ao.addr16[3], an->addr16[3], u), 3190 ao.addr16[4], an->addr16[4], u), 3191 ao.addr16[5], an->addr16[5], u), 3192 ao.addr16[6], an->addr16[6], u), 3193 ao.addr16[7], an->addr16[7], u); 3194 } 3195 #endif /* INET6 */ 3196 3197 static void 3198 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3199 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3200 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3201 { 3202 struct pf_addr oia, ooa; 3203 3204 PF_ACPY(&oia, ia, af); 3205 if (oa) 3206 PF_ACPY(&ooa, oa, af); 3207 3208 /* Change inner protocol port, fix inner protocol checksum. */ 3209 if (ip != NULL) { 3210 u_int16_t oip = *ip; 3211 u_int32_t opc; 3212 3213 if (pc != NULL) 3214 opc = *pc; 3215 *ip = np; 3216 if (pc != NULL) 3217 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3218 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3219 if (pc != NULL) 3220 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3221 } 3222 /* Change inner ip address, fix inner ip and icmp checksums. */ 3223 PF_ACPY(ia, na, af); 3224 switch (af) { 3225 #ifdef INET 3226 case AF_INET: { 3227 u_int32_t oh2c = *h2c; 3228 3229 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3230 oia.addr16[0], ia->addr16[0], 0), 3231 oia.addr16[1], ia->addr16[1], 0); 3232 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3233 oia.addr16[0], ia->addr16[0], 0), 3234 oia.addr16[1], ia->addr16[1], 0); 3235 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3236 break; 3237 } 3238 #endif /* INET */ 3239 #ifdef INET6 3240 case AF_INET6: 3241 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3242 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3243 pf_cksum_fixup(pf_cksum_fixup(*ic, 3244 oia.addr16[0], ia->addr16[0], u), 3245 oia.addr16[1], ia->addr16[1], u), 3246 oia.addr16[2], ia->addr16[2], u), 3247 oia.addr16[3], ia->addr16[3], u), 3248 oia.addr16[4], ia->addr16[4], u), 3249 oia.addr16[5], ia->addr16[5], u), 3250 oia.addr16[6], ia->addr16[6], u), 3251 oia.addr16[7], ia->addr16[7], u); 3252 break; 3253 #endif /* INET6 */ 3254 } 3255 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3256 if (oa) { 3257 PF_ACPY(oa, na, af); 3258 switch (af) { 3259 #ifdef INET 3260 case AF_INET: 3261 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3262 ooa.addr16[0], oa->addr16[0], 0), 3263 ooa.addr16[1], oa->addr16[1], 0); 3264 break; 3265 #endif /* INET */ 3266 #ifdef INET6 3267 case AF_INET6: 3268 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3269 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3270 pf_cksum_fixup(pf_cksum_fixup(*ic, 3271 ooa.addr16[0], oa->addr16[0], u), 3272 ooa.addr16[1], oa->addr16[1], u), 3273 ooa.addr16[2], oa->addr16[2], u), 3274 ooa.addr16[3], oa->addr16[3], u), 3275 ooa.addr16[4], oa->addr16[4], u), 3276 ooa.addr16[5], oa->addr16[5], u), 3277 ooa.addr16[6], oa->addr16[6], u), 3278 ooa.addr16[7], oa->addr16[7], u); 3279 break; 3280 #endif /* INET6 */ 3281 } 3282 } 3283 } 3284 3285 /* 3286 * Need to modulate the sequence numbers in the TCP SACK option 3287 * (credits to Krzysztof Pfaff for report and patch) 3288 */ 3289 static int 3290 pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th, 3291 struct pf_state_peer *dst) 3292 { 3293 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3294 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3295 int copyback = 0, i, olen; 3296 struct sackblk sack; 3297 3298 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3299 if (hlen < TCPOLEN_SACKLEN || 3300 !pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3301 return 0; 3302 3303 while (hlen >= TCPOLEN_SACKLEN) { 3304 size_t startoff = opt - opts; 3305 olen = opt[1]; 3306 switch (*opt) { 3307 case TCPOPT_EOL: /* FALLTHROUGH */ 3308 case TCPOPT_NOP: 3309 opt++; 3310 hlen--; 3311 break; 3312 case TCPOPT_SACK: 3313 if (olen > hlen) 3314 olen = hlen; 3315 if (olen >= TCPOLEN_SACKLEN) { 3316 for (i = 2; i + TCPOLEN_SACK <= olen; 3317 i += TCPOLEN_SACK) { 3318 memcpy(&sack, &opt[i], sizeof(sack)); 3319 pf_patch_32_unaligned(pd->m, 3320 &th->th_sum, &sack.start, 3321 htonl(ntohl(sack.start) - dst->seqdiff), 3322 PF_ALGNMNT(startoff), 3323 0); 3324 pf_patch_32_unaligned(pd->m, &th->th_sum, 3325 &sack.end, 3326 htonl(ntohl(sack.end) - dst->seqdiff), 3327 PF_ALGNMNT(startoff), 3328 0); 3329 memcpy(&opt[i], &sack, sizeof(sack)); 3330 } 3331 copyback = 1; 3332 } 3333 /* FALLTHROUGH */ 3334 default: 3335 if (olen < 2) 3336 olen = 2; 3337 hlen -= olen; 3338 opt += olen; 3339 } 3340 } 3341 3342 if (copyback) 3343 m_copyback(pd->m, pd->off + sizeof(*th), thoptlen, (caddr_t)opts); 3344 return (copyback); 3345 } 3346 3347 struct mbuf * 3348 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3349 const struct pf_addr *saddr, const struct pf_addr *daddr, 3350 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3351 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3352 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3353 { 3354 struct mbuf *m; 3355 int len, tlen; 3356 #ifdef INET 3357 struct ip *h = NULL; 3358 #endif /* INET */ 3359 #ifdef INET6 3360 struct ip6_hdr *h6 = NULL; 3361 #endif /* INET6 */ 3362 struct tcphdr *th; 3363 char *opt; 3364 struct pf_mtag *pf_mtag; 3365 3366 len = 0; 3367 th = NULL; 3368 3369 /* maximum segment size tcp option */ 3370 tlen = sizeof(struct tcphdr); 3371 if (mss) 3372 tlen += 4; 3373 3374 switch (af) { 3375 #ifdef INET 3376 case AF_INET: 3377 len = sizeof(struct ip) + tlen; 3378 break; 3379 #endif /* INET */ 3380 #ifdef INET6 3381 case AF_INET6: 3382 len = sizeof(struct ip6_hdr) + tlen; 3383 break; 3384 #endif /* INET6 */ 3385 } 3386 3387 m = m_gethdr(M_NOWAIT, MT_DATA); 3388 if (m == NULL) 3389 return (NULL); 3390 3391 #ifdef MAC 3392 mac_netinet_firewall_send(m); 3393 #endif 3394 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 3395 m_freem(m); 3396 return (NULL); 3397 } 3398 m->m_flags |= mbuf_flags; 3399 pf_mtag->tag = mtag_tag; 3400 pf_mtag->flags = mtag_flags; 3401 3402 if (rtableid >= 0) 3403 M_SETFIB(m, rtableid); 3404 3405 #ifdef ALTQ 3406 if (r != NULL && r->qid) { 3407 pf_mtag->qid = r->qid; 3408 3409 /* add hints for ecn */ 3410 pf_mtag->hdr = mtod(m, struct ip *); 3411 } 3412 #endif /* ALTQ */ 3413 m->m_data += max_linkhdr; 3414 m->m_pkthdr.len = m->m_len = len; 3415 /* The rest of the stack assumes a rcvif, so provide one. 3416 * This is a locally generated packet, so .. close enough. */ 3417 m->m_pkthdr.rcvif = V_loif; 3418 bzero(m->m_data, len); 3419 switch (af) { 3420 #ifdef INET 3421 case AF_INET: 3422 h = mtod(m, struct ip *); 3423 3424 /* IP header fields included in the TCP checksum */ 3425 h->ip_p = IPPROTO_TCP; 3426 h->ip_len = htons(tlen); 3427 h->ip_src.s_addr = saddr->v4.s_addr; 3428 h->ip_dst.s_addr = daddr->v4.s_addr; 3429 3430 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3431 break; 3432 #endif /* INET */ 3433 #ifdef INET6 3434 case AF_INET6: 3435 h6 = mtod(m, struct ip6_hdr *); 3436 3437 /* IP header fields included in the TCP checksum */ 3438 h6->ip6_nxt = IPPROTO_TCP; 3439 h6->ip6_plen = htons(tlen); 3440 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3441 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3442 3443 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3444 break; 3445 #endif /* INET6 */ 3446 } 3447 3448 /* TCP header */ 3449 th->th_sport = sport; 3450 th->th_dport = dport; 3451 th->th_seq = htonl(seq); 3452 th->th_ack = htonl(ack); 3453 th->th_off = tlen >> 2; 3454 th->th_flags = tcp_flags; 3455 th->th_win = htons(win); 3456 3457 if (mss) { 3458 opt = (char *)(th + 1); 3459 opt[0] = TCPOPT_MAXSEG; 3460 opt[1] = 4; 3461 HTONS(mss); 3462 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 3463 } 3464 3465 switch (af) { 3466 #ifdef INET 3467 case AF_INET: 3468 /* TCP checksum */ 3469 th->th_sum = in_cksum(m, len); 3470 3471 /* Finish the IP header */ 3472 h->ip_v = 4; 3473 h->ip_hl = sizeof(*h) >> 2; 3474 h->ip_tos = IPTOS_LOWDELAY; 3475 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 3476 h->ip_len = htons(len); 3477 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3478 h->ip_sum = 0; 3479 break; 3480 #endif /* INET */ 3481 #ifdef INET6 3482 case AF_INET6: 3483 /* TCP checksum */ 3484 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3485 sizeof(struct ip6_hdr), tlen); 3486 3487 h6->ip6_vfc |= IPV6_VERSION; 3488 h6->ip6_hlim = IPV6_DEFHLIM; 3489 break; 3490 #endif /* INET6 */ 3491 } 3492 3493 return (m); 3494 } 3495 3496 static void 3497 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3498 uint8_t ttl, int rtableid) 3499 { 3500 struct mbuf *m; 3501 #ifdef INET 3502 struct ip *h = NULL; 3503 #endif /* INET */ 3504 #ifdef INET6 3505 struct ip6_hdr *h6 = NULL; 3506 #endif /* INET6 */ 3507 struct sctphdr *hdr; 3508 struct sctp_chunkhdr *chunk; 3509 struct pf_send_entry *pfse; 3510 int off = 0; 3511 3512 MPASS(af == pd->af); 3513 3514 m = m_gethdr(M_NOWAIT, MT_DATA); 3515 if (m == NULL) 3516 return; 3517 3518 m->m_data += max_linkhdr; 3519 m->m_flags |= M_SKIP_FIREWALL; 3520 /* The rest of the stack assumes a rcvif, so provide one. 3521 * This is a locally generated packet, so .. close enough. */ 3522 m->m_pkthdr.rcvif = V_loif; 3523 3524 /* IPv4|6 header */ 3525 switch (af) { 3526 #ifdef INET 3527 case AF_INET: 3528 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3529 3530 h = mtod(m, struct ip *); 3531 3532 /* IP header fields included in the TCP checksum */ 3533 3534 h->ip_p = IPPROTO_SCTP; 3535 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3536 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3537 h->ip_src = pd->dst->v4; 3538 h->ip_dst = pd->src->v4; 3539 3540 off += sizeof(struct ip); 3541 break; 3542 #endif /* INET */ 3543 #ifdef INET6 3544 case AF_INET6: 3545 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3546 3547 h6 = mtod(m, struct ip6_hdr *); 3548 3549 /* IP header fields included in the TCP checksum */ 3550 h6->ip6_vfc |= IPV6_VERSION; 3551 h6->ip6_nxt = IPPROTO_SCTP; 3552 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3553 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3554 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3555 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3556 3557 off += sizeof(struct ip6_hdr); 3558 break; 3559 #endif /* INET6 */ 3560 } 3561 3562 /* SCTP header */ 3563 hdr = mtodo(m, off); 3564 3565 hdr->src_port = pd->hdr.sctp.dest_port; 3566 hdr->dest_port = pd->hdr.sctp.src_port; 3567 hdr->v_tag = pd->sctp_initiate_tag; 3568 hdr->checksum = 0; 3569 3570 /* Abort chunk. */ 3571 off += sizeof(struct sctphdr); 3572 chunk = mtodo(m, off); 3573 3574 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3575 chunk->chunk_length = htons(sizeof(*chunk)); 3576 3577 /* SCTP checksum */ 3578 off += sizeof(*chunk); 3579 m->m_pkthdr.len = m->m_len = off; 3580 3581 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 3582 3583 if (rtableid >= 0) 3584 M_SETFIB(m, rtableid); 3585 3586 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3587 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3588 if (pfse == NULL) { 3589 m_freem(m); 3590 return; 3591 } 3592 3593 switch (af) { 3594 #ifdef INET 3595 case AF_INET: 3596 pfse->pfse_type = PFSE_IP; 3597 break; 3598 #endif /* INET */ 3599 #ifdef INET6 3600 case AF_INET6: 3601 pfse->pfse_type = PFSE_IP6; 3602 break; 3603 #endif /* INET6 */ 3604 } 3605 3606 pfse->pfse_m = m; 3607 pf_send(pfse); 3608 } 3609 3610 void 3611 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3612 const struct pf_addr *saddr, const struct pf_addr *daddr, 3613 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3614 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3615 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3616 { 3617 struct pf_send_entry *pfse; 3618 struct mbuf *m; 3619 3620 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3621 win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, rtableid); 3622 if (m == NULL) 3623 return; 3624 3625 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3626 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3627 if (pfse == NULL) { 3628 m_freem(m); 3629 return; 3630 } 3631 3632 switch (af) { 3633 #ifdef INET 3634 case AF_INET: 3635 pfse->pfse_type = PFSE_IP; 3636 break; 3637 #endif /* INET */ 3638 #ifdef INET6 3639 case AF_INET6: 3640 pfse->pfse_type = PFSE_IP6; 3641 break; 3642 #endif /* INET6 */ 3643 } 3644 3645 pfse->pfse_m = m; 3646 pf_send(pfse); 3647 } 3648 3649 static void 3650 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3651 struct pf_state_key *sk, struct tcphdr *th, 3652 u_int16_t bproto_sum, u_int16_t bip_sum, 3653 u_short *reason, int rtableid) 3654 { 3655 struct pf_addr * const saddr = pd->src; 3656 struct pf_addr * const daddr = pd->dst; 3657 3658 /* undo NAT changes, if they have taken place */ 3659 if (nr != NULL) { 3660 PF_ACPY(saddr, &sk->addr[pd->sidx], pd->af); 3661 PF_ACPY(daddr, &sk->addr[pd->didx], pd->af); 3662 if (pd->sport) 3663 *pd->sport = sk->port[pd->sidx]; 3664 if (pd->dport) 3665 *pd->dport = sk->port[pd->didx]; 3666 if (pd->proto_sum) 3667 *pd->proto_sum = bproto_sum; 3668 if (pd->ip_sum) 3669 *pd->ip_sum = bip_sum; 3670 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 3671 } 3672 if (pd->proto == IPPROTO_TCP && 3673 ((r->rule_flag & PFRULE_RETURNRST) || 3674 (r->rule_flag & PFRULE_RETURN)) && 3675 !(th->th_flags & TH_RST)) { 3676 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3677 3678 if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off, 3679 IPPROTO_TCP, pd->af)) 3680 REASON_SET(reason, PFRES_PROTCKSUM); 3681 else { 3682 if (th->th_flags & TH_SYN) 3683 ack++; 3684 if (th->th_flags & TH_FIN) 3685 ack++; 3686 pf_send_tcp(r, pd->af, pd->dst, 3687 pd->src, th->th_dport, th->th_sport, 3688 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3689 r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid); 3690 } 3691 } else if (pd->proto == IPPROTO_SCTP && 3692 (r->rule_flag & PFRULE_RETURN)) { 3693 pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid); 3694 } else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET && 3695 r->return_icmp) 3696 pf_send_icmp(pd->m, r->return_icmp >> 8, 3697 r->return_icmp & 255, pd->af, r, rtableid); 3698 else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 && 3699 r->return_icmp6) 3700 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 3701 r->return_icmp6 & 255, pd->af, r, rtableid); 3702 } 3703 3704 static int 3705 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3706 { 3707 struct m_tag *mtag; 3708 u_int8_t mpcp; 3709 3710 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3711 if (mtag == NULL) 3712 return (0); 3713 3714 if (prio == PF_PRIO_ZERO) 3715 prio = 0; 3716 3717 mpcp = *(uint8_t *)(mtag + 1); 3718 3719 return (mpcp == prio); 3720 } 3721 3722 static int 3723 pf_icmp_to_bandlim(uint8_t type) 3724 { 3725 switch (type) { 3726 case ICMP_ECHO: 3727 case ICMP_ECHOREPLY: 3728 return (BANDLIM_ICMP_ECHO); 3729 case ICMP_TSTAMP: 3730 case ICMP_TSTAMPREPLY: 3731 return (BANDLIM_ICMP_TSTAMP); 3732 case ICMP_UNREACH: 3733 default: 3734 return (BANDLIM_ICMP_UNREACH); 3735 } 3736 } 3737 3738 static void 3739 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3740 struct pf_krule *r, int rtableid) 3741 { 3742 struct pf_send_entry *pfse; 3743 struct mbuf *m0; 3744 struct pf_mtag *pf_mtag; 3745 3746 /* ICMP packet rate limitation. */ 3747 switch (af) { 3748 #ifdef INET6 3749 case AF_INET6: 3750 if (icmp6_ratelimit(NULL, type, code)) 3751 return; 3752 break; 3753 #endif 3754 #ifdef INET 3755 case AF_INET: 3756 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3757 return; 3758 break; 3759 #endif 3760 } 3761 3762 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3763 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3764 if (pfse == NULL) 3765 return; 3766 3767 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3768 free(pfse, M_PFTEMP); 3769 return; 3770 } 3771 3772 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3773 free(pfse, M_PFTEMP); 3774 return; 3775 } 3776 /* XXX: revisit */ 3777 m0->m_flags |= M_SKIP_FIREWALL; 3778 3779 if (rtableid >= 0) 3780 M_SETFIB(m0, rtableid); 3781 3782 #ifdef ALTQ 3783 if (r->qid) { 3784 pf_mtag->qid = r->qid; 3785 /* add hints for ecn */ 3786 pf_mtag->hdr = mtod(m0, struct ip *); 3787 } 3788 #endif /* ALTQ */ 3789 3790 switch (af) { 3791 #ifdef INET 3792 case AF_INET: 3793 pfse->pfse_type = PFSE_ICMP; 3794 break; 3795 #endif /* INET */ 3796 #ifdef INET6 3797 case AF_INET6: 3798 pfse->pfse_type = PFSE_ICMP6; 3799 break; 3800 #endif /* INET6 */ 3801 } 3802 pfse->pfse_m = m0; 3803 pfse->icmpopts.type = type; 3804 pfse->icmpopts.code = code; 3805 pf_send(pfse); 3806 } 3807 3808 /* 3809 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3810 * If n is 0, they match if they are equal. If n is != 0, they match if they 3811 * are different. 3812 */ 3813 int 3814 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3815 struct pf_addr *b, sa_family_t af) 3816 { 3817 int match = 0; 3818 3819 switch (af) { 3820 #ifdef INET 3821 case AF_INET: 3822 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 3823 match++; 3824 break; 3825 #endif /* INET */ 3826 #ifdef INET6 3827 case AF_INET6: 3828 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 3829 match++; 3830 break; 3831 #endif /* INET6 */ 3832 } 3833 if (match) { 3834 if (n) 3835 return (0); 3836 else 3837 return (1); 3838 } else { 3839 if (n) 3840 return (1); 3841 else 3842 return (0); 3843 } 3844 } 3845 3846 /* 3847 * Return 1 if b <= a <= e, otherwise return 0. 3848 */ 3849 int 3850 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3851 struct pf_addr *a, sa_family_t af) 3852 { 3853 switch (af) { 3854 #ifdef INET 3855 case AF_INET: 3856 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3857 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3858 return (0); 3859 break; 3860 #endif /* INET */ 3861 #ifdef INET6 3862 case AF_INET6: { 3863 int i; 3864 3865 /* check a >= b */ 3866 for (i = 0; i < 4; ++i) 3867 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3868 break; 3869 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3870 return (0); 3871 /* check a <= e */ 3872 for (i = 0; i < 4; ++i) 3873 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3874 break; 3875 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3876 return (0); 3877 break; 3878 } 3879 #endif /* INET6 */ 3880 } 3881 return (1); 3882 } 3883 3884 static int 3885 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3886 { 3887 switch (op) { 3888 case PF_OP_IRG: 3889 return ((p > a1) && (p < a2)); 3890 case PF_OP_XRG: 3891 return ((p < a1) || (p > a2)); 3892 case PF_OP_RRG: 3893 return ((p >= a1) && (p <= a2)); 3894 case PF_OP_EQ: 3895 return (p == a1); 3896 case PF_OP_NE: 3897 return (p != a1); 3898 case PF_OP_LT: 3899 return (p < a1); 3900 case PF_OP_LE: 3901 return (p <= a1); 3902 case PF_OP_GT: 3903 return (p > a1); 3904 case PF_OP_GE: 3905 return (p >= a1); 3906 } 3907 return (0); /* never reached */ 3908 } 3909 3910 int 3911 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3912 { 3913 NTOHS(a1); 3914 NTOHS(a2); 3915 NTOHS(p); 3916 return (pf_match(op, a1, a2, p)); 3917 } 3918 3919 static int 3920 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3921 { 3922 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3923 return (0); 3924 return (pf_match(op, a1, a2, u)); 3925 } 3926 3927 static int 3928 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3929 { 3930 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3931 return (0); 3932 return (pf_match(op, a1, a2, g)); 3933 } 3934 3935 int 3936 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3937 { 3938 if (*tag == -1) 3939 *tag = mtag; 3940 3941 return ((!r->match_tag_not && r->match_tag == *tag) || 3942 (r->match_tag_not && r->match_tag != *tag)); 3943 } 3944 3945 static int 3946 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 3947 { 3948 struct ifnet *ifp = m->m_pkthdr.rcvif; 3949 struct pfi_kkif *kif; 3950 3951 if (ifp == NULL) 3952 return (0); 3953 3954 kif = (struct pfi_kkif *)ifp->if_pf_kif; 3955 3956 if (kif == NULL) { 3957 DPFPRINTF(PF_DEBUG_URGENT, 3958 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 3959 r->rcv_ifname)); 3960 return (0); 3961 } 3962 3963 return (pfi_kkif_match(r->rcv_kif, kif)); 3964 } 3965 3966 int 3967 pf_tag_packet(struct pf_pdesc *pd, int tag) 3968 { 3969 3970 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3971 3972 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) 3973 return (ENOMEM); 3974 3975 pd->pf_mtag->tag = tag; 3976 3977 return (0); 3978 } 3979 3980 #define PF_ANCHOR_STACKSIZE 32 3981 struct pf_kanchor_stackframe { 3982 struct pf_kruleset *rs; 3983 struct pf_krule *r; /* XXX: + match bit */ 3984 struct pf_kanchor *child; 3985 }; 3986 3987 /* 3988 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3989 */ 3990 #define PF_ANCHORSTACK_MATCH 0x00000001 3991 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 3992 3993 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3994 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 3995 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3996 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3997 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3998 } while (0) 3999 4000 void 4001 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4002 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4003 int *match) 4004 { 4005 struct pf_kanchor_stackframe *f; 4006 4007 PF_RULES_RASSERT(); 4008 4009 if (match) 4010 *match = 0; 4011 if (*depth >= PF_ANCHOR_STACKSIZE) { 4012 printf("%s: anchor stack overflow on %s\n", 4013 __func__, (*r)->anchor->name); 4014 *r = TAILQ_NEXT(*r, entries); 4015 return; 4016 } else if (*depth == 0 && a != NULL) 4017 *a = *r; 4018 f = stack + (*depth)++; 4019 f->rs = *rs; 4020 f->r = *r; 4021 if ((*r)->anchor_wildcard) { 4022 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4023 4024 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4025 *r = NULL; 4026 return; 4027 } 4028 *rs = &f->child->ruleset; 4029 } else { 4030 f->child = NULL; 4031 *rs = &(*r)->anchor->ruleset; 4032 } 4033 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4034 } 4035 4036 int 4037 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4038 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4039 int *match) 4040 { 4041 struct pf_kanchor_stackframe *f; 4042 struct pf_krule *fr; 4043 int quick = 0; 4044 4045 PF_RULES_RASSERT(); 4046 4047 do { 4048 if (*depth <= 0) 4049 break; 4050 f = stack + *depth - 1; 4051 fr = PF_ANCHOR_RULE(f); 4052 if (f->child != NULL) { 4053 /* 4054 * This block traverses through 4055 * a wildcard anchor. 4056 */ 4057 if (match != NULL && *match) { 4058 /* 4059 * If any of "*" matched, then 4060 * "foo/ *" matched, mark frame 4061 * appropriately. 4062 */ 4063 PF_ANCHOR_SET_MATCH(f); 4064 *match = 0; 4065 } 4066 f->child = RB_NEXT(pf_kanchor_node, 4067 &fr->anchor->children, f->child); 4068 if (f->child != NULL) { 4069 *rs = &f->child->ruleset; 4070 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4071 if (*r == NULL) 4072 continue; 4073 else 4074 break; 4075 } 4076 } 4077 (*depth)--; 4078 if (*depth == 0 && a != NULL) 4079 *a = NULL; 4080 *rs = f->rs; 4081 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 4082 quick = fr->quick; 4083 *r = TAILQ_NEXT(fr, entries); 4084 } while (*r == NULL); 4085 4086 return (quick); 4087 } 4088 4089 struct pf_keth_anchor_stackframe { 4090 struct pf_keth_ruleset *rs; 4091 struct pf_keth_rule *r; /* XXX: + match bit */ 4092 struct pf_keth_anchor *child; 4093 }; 4094 4095 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4096 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4097 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4098 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4099 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4100 } while (0) 4101 4102 void 4103 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4104 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4105 struct pf_keth_rule **a, int *match) 4106 { 4107 struct pf_keth_anchor_stackframe *f; 4108 4109 NET_EPOCH_ASSERT(); 4110 4111 if (match) 4112 *match = 0; 4113 if (*depth >= PF_ANCHOR_STACKSIZE) { 4114 printf("%s: anchor stack overflow on %s\n", 4115 __func__, (*r)->anchor->name); 4116 *r = TAILQ_NEXT(*r, entries); 4117 return; 4118 } else if (*depth == 0 && a != NULL) 4119 *a = *r; 4120 f = stack + (*depth)++; 4121 f->rs = *rs; 4122 f->r = *r; 4123 if ((*r)->anchor_wildcard) { 4124 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4125 4126 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4127 *r = NULL; 4128 return; 4129 } 4130 *rs = &f->child->ruleset; 4131 } else { 4132 f->child = NULL; 4133 *rs = &(*r)->anchor->ruleset; 4134 } 4135 *r = TAILQ_FIRST((*rs)->active.rules); 4136 } 4137 4138 int 4139 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4140 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4141 struct pf_keth_rule **a, int *match) 4142 { 4143 struct pf_keth_anchor_stackframe *f; 4144 struct pf_keth_rule *fr; 4145 int quick = 0; 4146 4147 NET_EPOCH_ASSERT(); 4148 4149 do { 4150 if (*depth <= 0) 4151 break; 4152 f = stack + *depth - 1; 4153 fr = PF_ETH_ANCHOR_RULE(f); 4154 if (f->child != NULL) { 4155 /* 4156 * This block traverses through 4157 * a wildcard anchor. 4158 */ 4159 if (match != NULL && *match) { 4160 /* 4161 * If any of "*" matched, then 4162 * "foo/ *" matched, mark frame 4163 * appropriately. 4164 */ 4165 PF_ETH_ANCHOR_SET_MATCH(f); 4166 *match = 0; 4167 } 4168 f->child = RB_NEXT(pf_keth_anchor_node, 4169 &fr->anchor->children, f->child); 4170 if (f->child != NULL) { 4171 *rs = &f->child->ruleset; 4172 *r = TAILQ_FIRST((*rs)->active.rules); 4173 if (*r == NULL) 4174 continue; 4175 else 4176 break; 4177 } 4178 } 4179 (*depth)--; 4180 if (*depth == 0 && a != NULL) 4181 *a = NULL; 4182 *rs = f->rs; 4183 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4184 quick = fr->quick; 4185 *r = TAILQ_NEXT(fr, entries); 4186 } while (*r == NULL); 4187 4188 return (quick); 4189 } 4190 4191 #ifdef INET6 4192 void 4193 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4194 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4195 { 4196 switch (af) { 4197 #ifdef INET 4198 case AF_INET: 4199 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4200 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4201 break; 4202 #endif /* INET */ 4203 case AF_INET6: 4204 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4205 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4206 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4207 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4208 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4209 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4210 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4211 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4212 break; 4213 } 4214 } 4215 4216 void 4217 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4218 { 4219 switch (af) { 4220 #ifdef INET 4221 case AF_INET: 4222 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4223 break; 4224 #endif /* INET */ 4225 case AF_INET6: 4226 if (addr->addr32[3] == 0xffffffff) { 4227 addr->addr32[3] = 0; 4228 if (addr->addr32[2] == 0xffffffff) { 4229 addr->addr32[2] = 0; 4230 if (addr->addr32[1] == 0xffffffff) { 4231 addr->addr32[1] = 0; 4232 addr->addr32[0] = 4233 htonl(ntohl(addr->addr32[0]) + 1); 4234 } else 4235 addr->addr32[1] = 4236 htonl(ntohl(addr->addr32[1]) + 1); 4237 } else 4238 addr->addr32[2] = 4239 htonl(ntohl(addr->addr32[2]) + 1); 4240 } else 4241 addr->addr32[3] = 4242 htonl(ntohl(addr->addr32[3]) + 1); 4243 break; 4244 } 4245 } 4246 #endif /* INET6 */ 4247 4248 void 4249 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4250 { 4251 /* 4252 * Modern rules use the same flags in rules as they do in states. 4253 */ 4254 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4255 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4256 4257 /* 4258 * Old-style scrub rules have different flags which need to be translated. 4259 */ 4260 if (r->rule_flag & PFRULE_RANDOMID) 4261 a->flags |= PFSTATE_RANDOMID; 4262 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4263 a->flags |= PFSTATE_SETTOS; 4264 a->set_tos = r->set_tos; 4265 } 4266 4267 if (r->qid) 4268 a->qid = r->qid; 4269 if (r->pqid) 4270 a->pqid = r->pqid; 4271 if (r->rtableid >= 0) 4272 a->rtableid = r->rtableid; 4273 a->log |= r->log; 4274 if (r->min_ttl) 4275 a->min_ttl = r->min_ttl; 4276 if (r->max_mss) 4277 a->max_mss = r->max_mss; 4278 if (r->dnpipe) 4279 a->dnpipe = r->dnpipe; 4280 if (r->dnrpipe) 4281 a->dnrpipe = r->dnrpipe; 4282 if (r->dnpipe || r->dnrpipe) { 4283 if (r->free_flags & PFRULE_DN_IS_PIPE) 4284 a->flags |= PFSTATE_DN_IS_PIPE; 4285 else 4286 a->flags &= ~PFSTATE_DN_IS_PIPE; 4287 } 4288 if (r->scrub_flags & PFSTATE_SETPRIO) { 4289 a->set_prio[0] = r->set_prio[0]; 4290 a->set_prio[1] = r->set_prio[1]; 4291 } 4292 } 4293 4294 int 4295 pf_socket_lookup(struct pf_pdesc *pd) 4296 { 4297 struct pf_addr *saddr, *daddr; 4298 u_int16_t sport, dport; 4299 struct inpcbinfo *pi; 4300 struct inpcb *inp; 4301 4302 pd->lookup.uid = UID_MAX; 4303 pd->lookup.gid = GID_MAX; 4304 4305 switch (pd->proto) { 4306 case IPPROTO_TCP: 4307 sport = pd->hdr.tcp.th_sport; 4308 dport = pd->hdr.tcp.th_dport; 4309 pi = &V_tcbinfo; 4310 break; 4311 case IPPROTO_UDP: 4312 sport = pd->hdr.udp.uh_sport; 4313 dport = pd->hdr.udp.uh_dport; 4314 pi = &V_udbinfo; 4315 break; 4316 default: 4317 return (-1); 4318 } 4319 if (pd->dir == PF_IN) { 4320 saddr = pd->src; 4321 daddr = pd->dst; 4322 } else { 4323 u_int16_t p; 4324 4325 p = sport; 4326 sport = dport; 4327 dport = p; 4328 saddr = pd->dst; 4329 daddr = pd->src; 4330 } 4331 switch (pd->af) { 4332 #ifdef INET 4333 case AF_INET: 4334 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4335 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4336 if (inp == NULL) { 4337 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4338 daddr->v4, dport, INPLOOKUP_WILDCARD | 4339 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4340 if (inp == NULL) 4341 return (-1); 4342 } 4343 break; 4344 #endif /* INET */ 4345 #ifdef INET6 4346 case AF_INET6: 4347 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4348 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4349 if (inp == NULL) { 4350 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4351 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4352 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4353 if (inp == NULL) 4354 return (-1); 4355 } 4356 break; 4357 #endif /* INET6 */ 4358 } 4359 INP_RLOCK_ASSERT(inp); 4360 pd->lookup.uid = inp->inp_cred->cr_uid; 4361 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4362 INP_RUNLOCK(inp); 4363 4364 return (1); 4365 } 4366 4367 u_int8_t 4368 pf_get_wscale(struct pf_pdesc *pd) 4369 { 4370 struct tcphdr *th = &pd->hdr.tcp; 4371 int hlen; 4372 u_int8_t hdr[60]; 4373 u_int8_t *opt, optlen; 4374 u_int8_t wscale = 0; 4375 4376 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4377 if (hlen <= sizeof(struct tcphdr)) 4378 return (0); 4379 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4380 return (0); 4381 opt = hdr + sizeof(struct tcphdr); 4382 hlen -= sizeof(struct tcphdr); 4383 while (hlen >= 3) { 4384 switch (*opt) { 4385 case TCPOPT_EOL: 4386 case TCPOPT_NOP: 4387 ++opt; 4388 --hlen; 4389 break; 4390 case TCPOPT_WINDOW: 4391 wscale = opt[2]; 4392 if (wscale > TCP_MAX_WINSHIFT) 4393 wscale = TCP_MAX_WINSHIFT; 4394 wscale |= PF_WSCALE_FLAG; 4395 /* FALLTHROUGH */ 4396 default: 4397 optlen = opt[1]; 4398 if (optlen < 2) 4399 optlen = 2; 4400 hlen -= optlen; 4401 opt += optlen; 4402 break; 4403 } 4404 } 4405 return (wscale); 4406 } 4407 4408 u_int16_t 4409 pf_get_mss(struct pf_pdesc *pd) 4410 { 4411 struct tcphdr *th = &pd->hdr.tcp; 4412 int hlen; 4413 u_int8_t hdr[60]; 4414 u_int8_t *opt, optlen; 4415 u_int16_t mss = V_tcp_mssdflt; 4416 4417 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4418 if (hlen <= sizeof(struct tcphdr)) 4419 return (0); 4420 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4421 return (0); 4422 opt = hdr + sizeof(struct tcphdr); 4423 hlen -= sizeof(struct tcphdr); 4424 while (hlen >= TCPOLEN_MAXSEG) { 4425 switch (*opt) { 4426 case TCPOPT_EOL: 4427 case TCPOPT_NOP: 4428 ++opt; 4429 --hlen; 4430 break; 4431 case TCPOPT_MAXSEG: 4432 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 4433 NTOHS(mss); 4434 /* FALLTHROUGH */ 4435 default: 4436 optlen = opt[1]; 4437 if (optlen < 2) 4438 optlen = 2; 4439 hlen -= optlen; 4440 opt += optlen; 4441 break; 4442 } 4443 } 4444 return (mss); 4445 } 4446 4447 static u_int16_t 4448 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 4449 { 4450 struct nhop_object *nh; 4451 #ifdef INET6 4452 struct in6_addr dst6; 4453 uint32_t scopeid; 4454 #endif /* INET6 */ 4455 int hlen = 0; 4456 uint16_t mss = 0; 4457 4458 NET_EPOCH_ASSERT(); 4459 4460 switch (af) { 4461 #ifdef INET 4462 case AF_INET: 4463 hlen = sizeof(struct ip); 4464 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 4465 if (nh != NULL) 4466 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4467 break; 4468 #endif /* INET */ 4469 #ifdef INET6 4470 case AF_INET6: 4471 hlen = sizeof(struct ip6_hdr); 4472 in6_splitscope(&addr->v6, &dst6, &scopeid); 4473 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4474 if (nh != NULL) 4475 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4476 break; 4477 #endif /* INET6 */ 4478 } 4479 4480 mss = max(V_tcp_mssdflt, mss); 4481 mss = min(mss, offer); 4482 mss = max(mss, 64); /* sanity - at least max opt space */ 4483 return (mss); 4484 } 4485 4486 static u_int32_t 4487 pf_tcp_iss(struct pf_pdesc *pd) 4488 { 4489 MD5_CTX ctx; 4490 u_int32_t digest[4]; 4491 4492 if (V_pf_tcp_secret_init == 0) { 4493 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4494 MD5Init(&V_pf_tcp_secret_ctx); 4495 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4496 sizeof(V_pf_tcp_secret)); 4497 V_pf_tcp_secret_init = 1; 4498 } 4499 4500 ctx = V_pf_tcp_secret_ctx; 4501 4502 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4503 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4504 switch (pd->af) { 4505 case AF_INET6: 4506 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4507 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4508 break; 4509 case AF_INET: 4510 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4511 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4512 break; 4513 } 4514 MD5Final((u_char *)digest, &ctx); 4515 V_pf_tcp_iss_off += 4096; 4516 #define ISN_RANDOM_INCREMENT (4096 - 1) 4517 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4518 V_pf_tcp_iss_off); 4519 #undef ISN_RANDOM_INCREMENT 4520 } 4521 4522 static bool 4523 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4524 { 4525 bool match = true; 4526 4527 /* Always matches if not set */ 4528 if (! r->isset) 4529 return (!r->neg); 4530 4531 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4532 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4533 match = false; 4534 break; 4535 } 4536 } 4537 4538 return (match ^ r->neg); 4539 } 4540 4541 static int 4542 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4543 { 4544 if (*tag == -1) 4545 *tag = mtag; 4546 4547 return ((!r->match_tag_not && r->match_tag == *tag) || 4548 (r->match_tag_not && r->match_tag != *tag)); 4549 } 4550 4551 static void 4552 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4553 { 4554 /* If we don't have the interface drop the packet. */ 4555 if (ifp == NULL) { 4556 m_freem(m); 4557 return; 4558 } 4559 4560 switch (ifp->if_type) { 4561 case IFT_ETHER: 4562 case IFT_XETHER: 4563 case IFT_L2VLAN: 4564 case IFT_BRIDGE: 4565 case IFT_IEEE8023ADLAG: 4566 break; 4567 default: 4568 m_freem(m); 4569 return; 4570 } 4571 4572 ifp->if_transmit(ifp, m); 4573 } 4574 4575 static int 4576 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4577 { 4578 #ifdef INET 4579 struct ip ip; 4580 #endif 4581 #ifdef INET6 4582 struct ip6_hdr ip6; 4583 #endif 4584 struct mbuf *m = *m0; 4585 struct ether_header *e; 4586 struct pf_keth_rule *r, *rm, *a = NULL; 4587 struct pf_keth_ruleset *ruleset = NULL; 4588 struct pf_mtag *mtag; 4589 struct pf_keth_ruleq *rules; 4590 struct pf_addr *src = NULL, *dst = NULL; 4591 struct pfi_kkif *bridge_to; 4592 sa_family_t af = 0; 4593 uint16_t proto; 4594 int asd = 0, match = 0; 4595 int tag = -1; 4596 uint8_t action; 4597 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4598 4599 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4600 NET_EPOCH_ASSERT(); 4601 4602 PF_RULES_RLOCK_TRACKER; 4603 4604 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4605 4606 mtag = pf_find_mtag(m); 4607 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4608 /* Dummynet re-injects packets after they've 4609 * completed their delay. We've already 4610 * processed them, so pass unconditionally. */ 4611 4612 /* But only once. We may see the packet multiple times (e.g. 4613 * PFIL_IN/PFIL_OUT). */ 4614 pf_dummynet_flag_remove(m, mtag); 4615 4616 return (PF_PASS); 4617 } 4618 4619 ruleset = V_pf_keth; 4620 rules = ck_pr_load_ptr(&ruleset->active.rules); 4621 r = TAILQ_FIRST(rules); 4622 rm = NULL; 4623 4624 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 4625 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 4626 DPFPRINTF(PF_DEBUG_URGENT, 4627 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 4628 ", pullup failed\n")); 4629 return (PF_DROP); 4630 } 4631 e = mtod(m, struct ether_header *); 4632 proto = ntohs(e->ether_type); 4633 4634 switch (proto) { 4635 #ifdef INET 4636 case ETHERTYPE_IP: { 4637 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4638 sizeof(ip))) 4639 return (PF_DROP); 4640 4641 af = AF_INET; 4642 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4643 (caddr_t)&ip); 4644 src = (struct pf_addr *)&ip.ip_src; 4645 dst = (struct pf_addr *)&ip.ip_dst; 4646 break; 4647 } 4648 #endif /* INET */ 4649 #ifdef INET6 4650 case ETHERTYPE_IPV6: { 4651 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4652 sizeof(ip6))) 4653 return (PF_DROP); 4654 4655 af = AF_INET6; 4656 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4657 (caddr_t)&ip6); 4658 src = (struct pf_addr *)&ip6.ip6_src; 4659 dst = (struct pf_addr *)&ip6.ip6_dst; 4660 break; 4661 } 4662 #endif /* INET6 */ 4663 } 4664 4665 PF_RULES_RLOCK(); 4666 4667 while (r != NULL) { 4668 counter_u64_add(r->evaluations, 1); 4669 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4670 4671 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4672 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4673 "kif"); 4674 r = r->skip[PFE_SKIP_IFP].ptr; 4675 } 4676 else if (r->direction && r->direction != dir) { 4677 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4678 "dir"); 4679 r = r->skip[PFE_SKIP_DIR].ptr; 4680 } 4681 else if (r->proto && r->proto != proto) { 4682 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4683 "proto"); 4684 r = r->skip[PFE_SKIP_PROTO].ptr; 4685 } 4686 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4687 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4688 "src"); 4689 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4690 } 4691 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4692 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4693 "dst"); 4694 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4695 } 4696 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4697 r->ipsrc.neg, kif, M_GETFIB(m))) { 4698 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4699 "ip_src"); 4700 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4701 } 4702 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4703 r->ipdst.neg, kif, M_GETFIB(m))) { 4704 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4705 "ip_dst"); 4706 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4707 } 4708 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4709 mtag ? mtag->tag : 0)) { 4710 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4711 "match_tag"); 4712 r = TAILQ_NEXT(r, entries); 4713 } 4714 else { 4715 if (r->tag) 4716 tag = r->tag; 4717 if (r->anchor == NULL) { 4718 /* Rule matches */ 4719 rm = r; 4720 4721 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4722 4723 if (r->quick) 4724 break; 4725 4726 r = TAILQ_NEXT(r, entries); 4727 } else { 4728 pf_step_into_keth_anchor(anchor_stack, &asd, 4729 &ruleset, &r, &a, &match); 4730 } 4731 } 4732 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4733 &ruleset, &r, &a, &match)) 4734 break; 4735 } 4736 4737 r = rm; 4738 4739 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4740 4741 /* Default to pass. */ 4742 if (r == NULL) { 4743 PF_RULES_RUNLOCK(); 4744 return (PF_PASS); 4745 } 4746 4747 /* Execute action. */ 4748 counter_u64_add(r->packets[dir == PF_OUT], 1); 4749 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4750 pf_update_timestamp(r); 4751 4752 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4753 if (r->action == PF_DROP) { 4754 PF_RULES_RUNLOCK(); 4755 return (PF_DROP); 4756 } 4757 4758 if (tag > 0) { 4759 if (mtag == NULL) 4760 mtag = pf_get_mtag(m); 4761 if (mtag == NULL) { 4762 PF_RULES_RUNLOCK(); 4763 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4764 return (PF_DROP); 4765 } 4766 mtag->tag = tag; 4767 } 4768 4769 if (r->qid != 0) { 4770 if (mtag == NULL) 4771 mtag = pf_get_mtag(m); 4772 if (mtag == NULL) { 4773 PF_RULES_RUNLOCK(); 4774 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4775 return (PF_DROP); 4776 } 4777 mtag->qid = r->qid; 4778 } 4779 4780 action = r->action; 4781 bridge_to = r->bridge_to; 4782 4783 /* Dummynet */ 4784 if (r->dnpipe) { 4785 struct ip_fw_args dnflow; 4786 4787 /* Drop packet if dummynet is not loaded. */ 4788 if (ip_dn_io_ptr == NULL) { 4789 PF_RULES_RUNLOCK(); 4790 m_freem(m); 4791 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4792 return (PF_DROP); 4793 } 4794 if (mtag == NULL) 4795 mtag = pf_get_mtag(m); 4796 if (mtag == NULL) { 4797 PF_RULES_RUNLOCK(); 4798 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4799 return (PF_DROP); 4800 } 4801 4802 bzero(&dnflow, sizeof(dnflow)); 4803 4804 /* We don't have port numbers here, so we set 0. That means 4805 * that we'll be somewhat limited in distinguishing flows (i.e. 4806 * only based on IP addresses, not based on port numbers), but 4807 * it's better than nothing. */ 4808 dnflow.f_id.dst_port = 0; 4809 dnflow.f_id.src_port = 0; 4810 dnflow.f_id.proto = 0; 4811 4812 dnflow.rule.info = r->dnpipe; 4813 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4814 if (r->dnflags & PFRULE_DN_IS_PIPE) 4815 dnflow.rule.info |= IPFW_IS_PIPE; 4816 4817 dnflow.f_id.extra = dnflow.rule.info; 4818 4819 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4820 dnflow.flags |= IPFW_ARGS_ETHER; 4821 dnflow.ifp = kif->pfik_ifp; 4822 4823 switch (af) { 4824 case AF_INET: 4825 dnflow.f_id.addr_type = 4; 4826 dnflow.f_id.src_ip = src->v4.s_addr; 4827 dnflow.f_id.dst_ip = dst->v4.s_addr; 4828 break; 4829 case AF_INET6: 4830 dnflow.flags |= IPFW_ARGS_IP6; 4831 dnflow.f_id.addr_type = 6; 4832 dnflow.f_id.src_ip6 = src->v6; 4833 dnflow.f_id.dst_ip6 = dst->v6; 4834 break; 4835 } 4836 4837 PF_RULES_RUNLOCK(); 4838 4839 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4840 ip_dn_io_ptr(m0, &dnflow); 4841 if (*m0 != NULL) 4842 pf_dummynet_flag_remove(m, mtag); 4843 } else { 4844 PF_RULES_RUNLOCK(); 4845 } 4846 4847 if (action == PF_PASS && bridge_to) { 4848 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4849 *m0 = NULL; /* We've eaten the packet. */ 4850 } 4851 4852 return (action); 4853 } 4854 4855 #define PF_TEST_ATTRIB(t, a)\ 4856 do { \ 4857 if (t) { \ 4858 r = a; \ 4859 goto nextrule; \ 4860 } \ 4861 } while (0) 4862 4863 static int 4864 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, 4865 struct pf_pdesc *pd, struct pf_krule **am, 4866 struct pf_kruleset **rsm, struct inpcb *inp) 4867 { 4868 struct pf_krule *nr = NULL; 4869 struct pf_addr * const saddr = pd->src; 4870 struct pf_addr * const daddr = pd->dst; 4871 struct pf_krule *r, *a = NULL; 4872 struct pf_kruleset *ruleset = NULL; 4873 struct pf_krule_slist match_rules; 4874 struct pf_krule_item *ri; 4875 struct pf_ksrc_node *nsn = NULL; 4876 struct tcphdr *th = &pd->hdr.tcp; 4877 struct pf_state_key *sk = NULL, *nk = NULL; 4878 u_short reason, transerror; 4879 int rewrite = 0; 4880 int tag = -1; 4881 int asd = 0; 4882 int match = 0; 4883 int state_icmp = 0, icmp_dir, multi; 4884 u_int16_t sport = 0, dport = 0, virtual_type, virtual_id; 4885 u_int16_t bproto_sum = 0, bip_sum = 0; 4886 u_int8_t icmptype = 0, icmpcode = 0; 4887 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4888 struct pf_udp_mapping *udp_mapping = NULL; 4889 4890 PF_RULES_RASSERT(); 4891 4892 SLIST_INIT(&match_rules); 4893 4894 if (inp != NULL) { 4895 INP_LOCK_ASSERT(inp); 4896 pd->lookup.uid = inp->inp_cred->cr_uid; 4897 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4898 pd->lookup.done = 1; 4899 } 4900 4901 switch (pd->virtual_proto) { 4902 case IPPROTO_TCP: 4903 sport = th->th_sport; 4904 dport = th->th_dport; 4905 break; 4906 case IPPROTO_UDP: 4907 sport = pd->hdr.udp.uh_sport; 4908 dport = pd->hdr.udp.uh_dport; 4909 break; 4910 case IPPROTO_SCTP: 4911 sport = pd->hdr.sctp.src_port; 4912 dport = pd->hdr.sctp.dest_port; 4913 break; 4914 #ifdef INET 4915 case IPPROTO_ICMP: 4916 MPASS(pd->af == AF_INET); 4917 icmptype = pd->hdr.icmp.icmp_type; 4918 icmpcode = pd->hdr.icmp.icmp_code; 4919 state_icmp = pf_icmp_mapping(pd, icmptype, 4920 &icmp_dir, &multi, &virtual_id, &virtual_type); 4921 if (icmp_dir == PF_IN) { 4922 sport = virtual_id; 4923 dport = virtual_type; 4924 } else { 4925 sport = virtual_type; 4926 dport = virtual_id; 4927 } 4928 break; 4929 #endif /* INET */ 4930 #ifdef INET6 4931 case IPPROTO_ICMPV6: 4932 MPASS(pd->af == AF_INET6); 4933 icmptype = pd->hdr.icmp6.icmp6_type; 4934 icmpcode = pd->hdr.icmp6.icmp6_code; 4935 state_icmp = pf_icmp_mapping(pd, icmptype, 4936 &icmp_dir, &multi, &virtual_id, &virtual_type); 4937 if (icmp_dir == PF_IN) { 4938 sport = virtual_id; 4939 dport = virtual_type; 4940 } else { 4941 sport = virtual_type; 4942 dport = virtual_id; 4943 } 4944 4945 break; 4946 #endif /* INET6 */ 4947 default: 4948 sport = dport = 0; 4949 break; 4950 } 4951 4952 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4953 4954 /* check packet for BINAT/NAT/RDR */ 4955 transerror = pf_get_translation(pd, pd->off, &nsn, &sk, 4956 &nk, saddr, daddr, sport, dport, anchor_stack, &nr, &udp_mapping); 4957 switch (transerror) { 4958 default: 4959 /* A translation error occurred. */ 4960 REASON_SET(&reason, transerror); 4961 goto cleanup; 4962 case PFRES_MAX: 4963 /* No match. */ 4964 break; 4965 case PFRES_MATCH: 4966 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4967 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4968 4969 if (nr->log) { 4970 PFLOG_PACKET(PF_PASS, PFRES_MATCH, nr, a, 4971 ruleset, pd, 1); 4972 } 4973 4974 if (pd->ip_sum) 4975 bip_sum = *pd->ip_sum; 4976 4977 switch (pd->proto) { 4978 case IPPROTO_TCP: 4979 bproto_sum = th->th_sum; 4980 pd->proto_sum = &th->th_sum; 4981 4982 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], pd->af) || 4983 nk->port[pd->sidx] != sport) { 4984 pf_change_ap(pd->m, saddr, &th->th_sport, pd->ip_sum, 4985 &th->th_sum, &nk->addr[pd->sidx], 4986 nk->port[pd->sidx], 0, pd->af); 4987 pd->sport = &th->th_sport; 4988 sport = th->th_sport; 4989 } 4990 4991 if (PF_ANEQ(daddr, &nk->addr[pd->didx], pd->af) || 4992 nk->port[pd->didx] != dport) { 4993 pf_change_ap(pd->m, daddr, &th->th_dport, pd->ip_sum, 4994 &th->th_sum, &nk->addr[pd->didx], 4995 nk->port[pd->didx], 0, pd->af); 4996 dport = th->th_dport; 4997 pd->dport = &th->th_dport; 4998 } 4999 rewrite++; 5000 break; 5001 case IPPROTO_UDP: 5002 bproto_sum = pd->hdr.udp.uh_sum; 5003 pd->proto_sum = &pd->hdr.udp.uh_sum; 5004 5005 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], pd->af) || 5006 nk->port[pd->sidx] != sport) { 5007 pf_change_ap(pd->m, saddr, &pd->hdr.udp.uh_sport, 5008 pd->ip_sum, &pd->hdr.udp.uh_sum, 5009 &nk->addr[pd->sidx], 5010 nk->port[pd->sidx], 1, pd->af); 5011 sport = pd->hdr.udp.uh_sport; 5012 pd->sport = &pd->hdr.udp.uh_sport; 5013 } 5014 5015 if (PF_ANEQ(daddr, &nk->addr[pd->didx], pd->af) || 5016 nk->port[pd->didx] != dport) { 5017 pf_change_ap(pd->m, daddr, &pd->hdr.udp.uh_dport, 5018 pd->ip_sum, &pd->hdr.udp.uh_sum, 5019 &nk->addr[pd->didx], 5020 nk->port[pd->didx], 1, pd->af); 5021 dport = pd->hdr.udp.uh_dport; 5022 pd->dport = &pd->hdr.udp.uh_dport; 5023 } 5024 rewrite++; 5025 break; 5026 case IPPROTO_SCTP: { 5027 uint16_t checksum = 0; 5028 5029 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], pd->af) || 5030 nk->port[pd->sidx] != sport) { 5031 pf_change_ap(pd->m, saddr, &pd->hdr.sctp.src_port, 5032 pd->ip_sum, &checksum, 5033 &nk->addr[pd->sidx], 5034 nk->port[pd->sidx], 1, pd->af); 5035 } 5036 if (PF_ANEQ(daddr, &nk->addr[pd->didx], pd->af) || 5037 nk->port[pd->didx] != dport) { 5038 pf_change_ap(pd->m, daddr, &pd->hdr.sctp.dest_port, 5039 pd->ip_sum, &checksum, 5040 &nk->addr[pd->didx], 5041 nk->port[pd->didx], 1, pd->af); 5042 } 5043 break; 5044 } 5045 #ifdef INET 5046 case IPPROTO_ICMP: 5047 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 5048 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 5049 nk->addr[pd->sidx].v4.s_addr, 0); 5050 5051 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 5052 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 5053 nk->addr[pd->didx].v4.s_addr, 0); 5054 5055 if (virtual_type == htons(ICMP_ECHO) && 5056 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5057 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5058 pd->hdr.icmp.icmp_cksum, sport, 5059 nk->port[pd->sidx], 0); 5060 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5061 pd->sport = &pd->hdr.icmp.icmp_id; 5062 } 5063 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5064 break; 5065 #endif /* INET */ 5066 #ifdef INET6 5067 case IPPROTO_ICMPV6: 5068 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 5069 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 5070 &nk->addr[pd->sidx], 0); 5071 5072 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 5073 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 5074 &nk->addr[pd->didx], 0); 5075 rewrite++; 5076 break; 5077 #endif /* INET */ 5078 default: 5079 switch (pd->af) { 5080 #ifdef INET 5081 case AF_INET: 5082 if (PF_ANEQ(saddr, 5083 &nk->addr[pd->sidx], AF_INET)) 5084 pf_change_a(&saddr->v4.s_addr, 5085 pd->ip_sum, 5086 nk->addr[pd->sidx].v4.s_addr, 0); 5087 5088 if (PF_ANEQ(daddr, 5089 &nk->addr[pd->didx], AF_INET)) 5090 pf_change_a(&daddr->v4.s_addr, 5091 pd->ip_sum, 5092 nk->addr[pd->didx].v4.s_addr, 0); 5093 break; 5094 #endif /* INET */ 5095 #ifdef INET6 5096 case AF_INET6: 5097 if (PF_ANEQ(saddr, 5098 &nk->addr[pd->sidx], AF_INET6)) 5099 PF_ACPY(saddr, &nk->addr[pd->sidx], pd->af); 5100 5101 if (PF_ANEQ(daddr, 5102 &nk->addr[pd->didx], AF_INET6)) 5103 PF_ACPY(daddr, &nk->addr[pd->didx], pd->af); 5104 break; 5105 #endif /* INET */ 5106 } 5107 break; 5108 } 5109 if (nr->natpass) 5110 r = NULL; 5111 } 5112 5113 while (r != NULL) { 5114 pf_counter_u64_add(&r->evaluations, 1); 5115 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot, 5116 r->skip[PF_SKIP_IFP]); 5117 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5118 r->skip[PF_SKIP_DIR]); 5119 PF_TEST_ATTRIB(r->af && r->af != pd->af, 5120 r->skip[PF_SKIP_AF]); 5121 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5122 r->skip[PF_SKIP_PROTO]); 5123 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, saddr, pd->af, 5124 r->src.neg, pd->kif, M_GETFIB(pd->m)), 5125 r->skip[PF_SKIP_SRC_ADDR]); 5126 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, daddr, pd->af, 5127 r->dst.neg, NULL, M_GETFIB(pd->m)), 5128 r->skip[PF_SKIP_DST_ADDR]); 5129 switch (pd->virtual_proto) { 5130 case PF_VPROTO_FRAGMENT: 5131 /* tcp/udp only. port_op always 0 in other cases */ 5132 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5133 TAILQ_NEXT(r, entries)); 5134 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5135 TAILQ_NEXT(r, entries)); 5136 /* icmp only. type/code always 0 in other cases */ 5137 PF_TEST_ATTRIB((r->type || r->code), 5138 TAILQ_NEXT(r, entries)); 5139 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5140 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5141 TAILQ_NEXT(r, entries)); 5142 break; 5143 5144 case IPPROTO_TCP: 5145 PF_TEST_ATTRIB((r->flagset & th->th_flags) != r->flags, 5146 TAILQ_NEXT(r, entries)); 5147 /* FALLTHROUGH */ 5148 case IPPROTO_SCTP: 5149 case IPPROTO_UDP: 5150 /* tcp/udp only. port_op always 0 in other cases */ 5151 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5152 r->src.port[0], r->src.port[1], sport), 5153 r->skip[PF_SKIP_SRC_PORT]); 5154 /* tcp/udp only. port_op always 0 in other cases */ 5155 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5156 r->dst.port[0], r->dst.port[1], dport), 5157 r->skip[PF_SKIP_DST_PORT]); 5158 /* tcp/udp only. uid.op always 0 in other cases */ 5159 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5160 pf_socket_lookup(pd), 1)) && 5161 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5162 pd->lookup.uid), 5163 TAILQ_NEXT(r, entries)); 5164 /* tcp/udp only. gid.op always 0 in other cases */ 5165 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5166 pf_socket_lookup(pd), 1)) && 5167 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5168 pd->lookup.gid), 5169 TAILQ_NEXT(r, entries)); 5170 break; 5171 5172 case IPPROTO_ICMP: 5173 case IPPROTO_ICMPV6: 5174 /* icmp only. type always 0 in other cases */ 5175 PF_TEST_ATTRIB(r->type && r->type != icmptype + 1, 5176 TAILQ_NEXT(r, entries)); 5177 /* icmp only. type always 0 in other cases */ 5178 PF_TEST_ATTRIB(r->code && r->code != icmpcode + 1, 5179 TAILQ_NEXT(r, entries)); 5180 break; 5181 5182 default: 5183 break; 5184 } 5185 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5186 TAILQ_NEXT(r, entries)); 5187 PF_TEST_ATTRIB(r->prio && 5188 !pf_match_ieee8021q_pcp(r->prio, pd->m), 5189 TAILQ_NEXT(r, entries)); 5190 PF_TEST_ATTRIB(r->prob && 5191 r->prob <= arc4random(), 5192 TAILQ_NEXT(r, entries)); 5193 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r, &tag, 5194 pd->pf_mtag ? pd->pf_mtag->tag : 0), 5195 TAILQ_NEXT(r, entries)); 5196 PF_TEST_ATTRIB(r->rcv_kif && !pf_match_rcvif(pd->m, r), 5197 TAILQ_NEXT(r, entries)); 5198 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5199 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5200 TAILQ_NEXT(r, entries)); 5201 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5202 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5203 pf_osfp_fingerprint(pd, th), 5204 r->os_fingerprint)), 5205 TAILQ_NEXT(r, entries)); 5206 /* FALLTHROUGH */ 5207 if (r->tag) 5208 tag = r->tag; 5209 if (r->anchor == NULL) { 5210 if (r->action == PF_MATCH) { 5211 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5212 if (ri == NULL) { 5213 REASON_SET(&reason, PFRES_MEMORY); 5214 goto cleanup; 5215 } 5216 ri->r = r; 5217 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5218 pf_counter_u64_critical_enter(); 5219 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5220 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5221 pf_counter_u64_critical_exit(); 5222 pf_rule_to_actions(r, &pd->act); 5223 if (r->log || pd->act.log & PF_LOG_MATCHES) 5224 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5225 a, ruleset, pd, 1); 5226 } else { 5227 match = 1; 5228 *rm = r; 5229 *am = a; 5230 *rsm = ruleset; 5231 if (pd->act.log & PF_LOG_MATCHES) 5232 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5233 a, ruleset, pd, 1); 5234 } 5235 if ((*rm)->quick) 5236 break; 5237 r = TAILQ_NEXT(r, entries); 5238 } else 5239 pf_step_into_anchor(anchor_stack, &asd, 5240 &ruleset, PF_RULESET_FILTER, &r, &a, 5241 &match); 5242 nextrule: 5243 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5244 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5245 break; 5246 } 5247 r = *rm; 5248 a = *am; 5249 ruleset = *rsm; 5250 5251 REASON_SET(&reason, PFRES_MATCH); 5252 5253 /* apply actions for last matching pass/block rule */ 5254 pf_rule_to_actions(r, &pd->act); 5255 5256 if (r->log || pd->act.log & PF_LOG_MATCHES) { 5257 if (rewrite) 5258 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5259 PFLOG_PACKET(r->action, reason, r, a, ruleset, pd, 1); 5260 } 5261 5262 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5263 (r->action == PF_DROP) && 5264 ((r->rule_flag & PFRULE_RETURNRST) || 5265 (r->rule_flag & PFRULE_RETURNICMP) || 5266 (r->rule_flag & PFRULE_RETURN))) { 5267 pf_return(r, nr, pd, sk, th, bproto_sum, 5268 bip_sum, &reason, r->rtableid); 5269 } 5270 5271 if (r->action == PF_DROP) 5272 goto cleanup; 5273 5274 if (tag > 0 && pf_tag_packet(pd, tag)) { 5275 REASON_SET(&reason, PFRES_MEMORY); 5276 goto cleanup; 5277 } 5278 if (pd->act.rtableid >= 0) 5279 M_SETFIB(pd->m, pd->act.rtableid); 5280 5281 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5282 (!state_icmp && (r->keep_state || nr != NULL || 5283 (pd->flags & PFDESC_TCP_NORM)))) { 5284 int action; 5285 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, 5286 sport, dport, &rewrite, sm, tag, bproto_sum, bip_sum, 5287 &match_rules, udp_mapping); 5288 if (action != PF_PASS) { 5289 pf_udp_mapping_release(udp_mapping); 5290 if (action == PF_DROP && 5291 (r->rule_flag & PFRULE_RETURN)) 5292 pf_return(r, nr, pd, sk, th, 5293 bproto_sum, bip_sum, &reason, 5294 pd->act.rtableid); 5295 return (action); 5296 } 5297 } else { 5298 while ((ri = SLIST_FIRST(&match_rules))) { 5299 SLIST_REMOVE_HEAD(&match_rules, entry); 5300 free(ri, M_PF_RULE_ITEM); 5301 } 5302 5303 uma_zfree(V_pf_state_key_z, sk); 5304 uma_zfree(V_pf_state_key_z, nk); 5305 pf_udp_mapping_release(udp_mapping); 5306 } 5307 5308 /* copy back packet headers if we performed NAT operations */ 5309 if (rewrite) 5310 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5311 5312 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5313 pd->dir == PF_OUT && 5314 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) 5315 /* 5316 * We want the state created, but we dont 5317 * want to send this in case a partner 5318 * firewall has to know about it to allow 5319 * replies through it. 5320 */ 5321 return (PF_DEFER); 5322 5323 return (PF_PASS); 5324 5325 cleanup: 5326 while ((ri = SLIST_FIRST(&match_rules))) { 5327 SLIST_REMOVE_HEAD(&match_rules, entry); 5328 free(ri, M_PF_RULE_ITEM); 5329 } 5330 5331 uma_zfree(V_pf_state_key_z, sk); 5332 uma_zfree(V_pf_state_key_z, nk); 5333 pf_udp_mapping_release(udp_mapping); 5334 5335 return (PF_DROP); 5336 } 5337 5338 static int 5339 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 5340 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 5341 struct pf_state_key *sk, u_int16_t sport, 5342 u_int16_t dport, int *rewrite, struct pf_kstate **sm, 5343 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, 5344 struct pf_krule_slist *match_rules, struct pf_udp_mapping *udp_mapping) 5345 { 5346 struct pf_kstate *s = NULL; 5347 struct pf_ksrc_node *sn = NULL; 5348 struct tcphdr *th = &pd->hdr.tcp; 5349 u_int16_t mss = V_tcp_mssdflt; 5350 u_short reason, sn_reason; 5351 struct pf_krule_item *ri; 5352 5353 /* check maximums */ 5354 if (r->max_states && 5355 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 5356 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 5357 REASON_SET(&reason, PFRES_MAXSTATES); 5358 goto csfailed; 5359 } 5360 /* src node for filter rule */ 5361 if ((r->rule_flag & PFRULE_SRCTRACK || 5362 r->rpool.opts & PF_POOL_STICKYADDR) && 5363 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 5364 REASON_SET(&reason, sn_reason); 5365 goto csfailed; 5366 } 5367 /* src node for translation rule */ 5368 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 5369 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 5370 pd->af)) != 0 ) { 5371 REASON_SET(&reason, sn_reason); 5372 goto csfailed; 5373 } 5374 s = pf_alloc_state(M_NOWAIT); 5375 if (s == NULL) { 5376 REASON_SET(&reason, PFRES_MEMORY); 5377 goto csfailed; 5378 } 5379 s->rule = r; 5380 s->nat_rule = nr; 5381 s->anchor = a; 5382 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 5383 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 5384 5385 STATE_INC_COUNTERS(s); 5386 if (r->allow_opts) 5387 s->state_flags |= PFSTATE_ALLOWOPTS; 5388 if (r->rule_flag & PFRULE_STATESLOPPY) 5389 s->state_flags |= PFSTATE_SLOPPY; 5390 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 5391 s->state_flags |= PFSTATE_SCRUB_TCP; 5392 if ((r->rule_flag & PFRULE_PFLOW) || 5393 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 5394 s->state_flags |= PFSTATE_PFLOW; 5395 5396 s->act.log = pd->act.log & PF_LOG_ALL; 5397 s->sync_state = PFSYNC_S_NONE; 5398 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 5399 5400 if (nr != NULL) 5401 s->act.log |= nr->log & PF_LOG_ALL; 5402 switch (pd->proto) { 5403 case IPPROTO_TCP: 5404 s->src.seqlo = ntohl(th->th_seq); 5405 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 5406 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 5407 r->keep_state == PF_STATE_MODULATE) { 5408 /* Generate sequence number modulator */ 5409 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 5410 0) 5411 s->src.seqdiff = 1; 5412 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, 5413 htonl(s->src.seqlo + s->src.seqdiff), 0); 5414 *rewrite = 1; 5415 } else 5416 s->src.seqdiff = 0; 5417 if (th->th_flags & TH_SYN) { 5418 s->src.seqhi++; 5419 s->src.wscale = pf_get_wscale(pd); 5420 } 5421 s->src.max_win = MAX(ntohs(th->th_win), 1); 5422 if (s->src.wscale & PF_WSCALE_MASK) { 5423 /* Remove scale factor from initial window */ 5424 int win = s->src.max_win; 5425 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 5426 s->src.max_win = (win - 1) >> 5427 (s->src.wscale & PF_WSCALE_MASK); 5428 } 5429 if (th->th_flags & TH_FIN) 5430 s->src.seqhi++; 5431 s->dst.seqhi = 1; 5432 s->dst.max_win = 1; 5433 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 5434 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 5435 s->timeout = PFTM_TCP_FIRST_PACKET; 5436 atomic_add_32(&V_pf_status.states_halfopen, 1); 5437 break; 5438 case IPPROTO_UDP: 5439 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 5440 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 5441 s->timeout = PFTM_UDP_FIRST_PACKET; 5442 break; 5443 case IPPROTO_SCTP: 5444 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 5445 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 5446 s->timeout = PFTM_SCTP_FIRST_PACKET; 5447 break; 5448 case IPPROTO_ICMP: 5449 #ifdef INET6 5450 case IPPROTO_ICMPV6: 5451 #endif 5452 s->timeout = PFTM_ICMP_FIRST_PACKET; 5453 break; 5454 default: 5455 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 5456 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 5457 s->timeout = PFTM_OTHER_FIRST_PACKET; 5458 } 5459 5460 if (r->rt) { 5461 /* pf_map_addr increases the reason counters */ 5462 if ((reason = pf_map_addr_sn(pd->af, r, pd->src, &s->rt_addr, 5463 &s->rt_kif, NULL, &sn)) != 0) 5464 goto csfailed; 5465 s->rt = r->rt; 5466 } 5467 5468 s->creation = s->expire = pf_get_uptime(); 5469 5470 if (sn != NULL) 5471 s->src_node = sn; 5472 if (nsn != NULL) { 5473 /* XXX We only modify one side for now. */ 5474 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 5475 s->nat_src_node = nsn; 5476 } 5477 if (pd->proto == IPPROTO_TCP) { 5478 if (s->state_flags & PFSTATE_SCRUB_TCP && 5479 pf_normalize_tcp_init(pd, th, &s->src, &s->dst)) { 5480 REASON_SET(&reason, PFRES_MEMORY); 5481 goto csfailed; 5482 } 5483 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 5484 pf_normalize_tcp_stateful(pd, &reason, th, s, 5485 &s->src, &s->dst, rewrite)) { 5486 /* This really shouldn't happen!!! */ 5487 DPFPRINTF(PF_DEBUG_URGENT, 5488 ("pf_normalize_tcp_stateful failed on first " 5489 "pkt\n")); 5490 goto csfailed; 5491 } 5492 } else if (pd->proto == IPPROTO_SCTP) { 5493 if (pf_normalize_sctp_init(pd, &s->src, &s->dst)) 5494 goto csfailed; 5495 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 5496 goto csfailed; 5497 } 5498 s->direction = pd->dir; 5499 5500 /* 5501 * sk/nk could already been setup by pf_get_translation(). 5502 */ 5503 if (nr == NULL) { 5504 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 5505 __func__, nr, sk, nk)); 5506 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 5507 if (sk == NULL) 5508 goto csfailed; 5509 nk = sk; 5510 } else 5511 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 5512 __func__, nr, sk, nk)); 5513 5514 /* Swap sk/nk for PF_OUT. */ 5515 if (pf_state_insert(BOUND_IFACE(s, pd->kif), pd->kif, 5516 (pd->dir == PF_IN) ? sk : nk, 5517 (pd->dir == PF_IN) ? nk : sk, s)) { 5518 REASON_SET(&reason, PFRES_STATEINS); 5519 goto drop; 5520 } else 5521 *sm = s; 5522 5523 if (tag > 0) 5524 s->tag = tag; 5525 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 5526 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 5527 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 5528 /* undo NAT changes, if they have taken place */ 5529 if (nr != NULL) { 5530 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 5531 if (pd->dir == PF_OUT) 5532 skt = s->key[PF_SK_STACK]; 5533 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 5534 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 5535 if (pd->sport) 5536 *pd->sport = skt->port[pd->sidx]; 5537 if (pd->dport) 5538 *pd->dport = skt->port[pd->didx]; 5539 if (pd->proto_sum) 5540 *pd->proto_sum = bproto_sum; 5541 if (pd->ip_sum) 5542 *pd->ip_sum = bip_sum; 5543 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5544 } 5545 s->src.seqhi = htonl(arc4random()); 5546 /* Find mss option */ 5547 int rtid = M_GETFIB(pd->m); 5548 mss = pf_get_mss(pd); 5549 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5550 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5551 s->src.mss = mss; 5552 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5553 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5554 TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0, 5555 pd->act.rtableid); 5556 REASON_SET(&reason, PFRES_SYNPROXY); 5557 return (PF_SYNPROXY_DROP); 5558 } 5559 5560 s->udp_mapping = udp_mapping; 5561 5562 return (PF_PASS); 5563 5564 csfailed: 5565 while ((ri = SLIST_FIRST(match_rules))) { 5566 SLIST_REMOVE_HEAD(match_rules, entry); 5567 free(ri, M_PF_RULE_ITEM); 5568 } 5569 5570 uma_zfree(V_pf_state_key_z, sk); 5571 uma_zfree(V_pf_state_key_z, nk); 5572 5573 if (sn != NULL) { 5574 PF_SRC_NODE_LOCK(sn); 5575 if (--sn->states == 0 && sn->expire == 0) { 5576 pf_unlink_src_node(sn); 5577 uma_zfree(V_pf_sources_z, sn); 5578 counter_u64_add( 5579 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5580 } 5581 PF_SRC_NODE_UNLOCK(sn); 5582 } 5583 5584 if (nsn != sn && nsn != NULL) { 5585 PF_SRC_NODE_LOCK(nsn); 5586 if (--nsn->states == 0 && nsn->expire == 0) { 5587 pf_unlink_src_node(nsn); 5588 uma_zfree(V_pf_sources_z, nsn); 5589 counter_u64_add( 5590 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5591 } 5592 PF_SRC_NODE_UNLOCK(nsn); 5593 } 5594 5595 drop: 5596 if (s != NULL) { 5597 pf_src_tree_remove_state(s); 5598 s->timeout = PFTM_UNLINKED; 5599 STATE_DEC_COUNTERS(s); 5600 pf_free_state(s); 5601 } 5602 5603 return (PF_DROP); 5604 } 5605 5606 static int 5607 pf_tcp_track_full(struct pf_kstate **state, struct pf_pdesc *pd, 5608 u_short *reason, int *copyback) 5609 { 5610 struct tcphdr *th = &pd->hdr.tcp; 5611 struct pf_state_peer *src, *dst; 5612 u_int16_t win = ntohs(th->th_win); 5613 u_int32_t ack, end, data_end, seq, orig_seq; 5614 u_int8_t sws, dws, psrc, pdst; 5615 int ackskew; 5616 5617 if (pd->dir == (*state)->direction) { 5618 src = &(*state)->src; 5619 dst = &(*state)->dst; 5620 psrc = PF_PEER_SRC; 5621 pdst = PF_PEER_DST; 5622 } else { 5623 src = &(*state)->dst; 5624 dst = &(*state)->src; 5625 psrc = PF_PEER_DST; 5626 pdst = PF_PEER_SRC; 5627 } 5628 5629 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5630 sws = src->wscale & PF_WSCALE_MASK; 5631 dws = dst->wscale & PF_WSCALE_MASK; 5632 } else 5633 sws = dws = 0; 5634 5635 /* 5636 * Sequence tracking algorithm from Guido van Rooij's paper: 5637 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5638 * tcp_filtering.ps 5639 */ 5640 5641 orig_seq = seq = ntohl(th->th_seq); 5642 if (src->seqlo == 0) { 5643 /* First packet from this end. Set its state */ 5644 5645 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5646 src->scrub == NULL) { 5647 if (pf_normalize_tcp_init(pd, th, src, dst)) { 5648 REASON_SET(reason, PFRES_MEMORY); 5649 return (PF_DROP); 5650 } 5651 } 5652 5653 /* Deferred generation of sequence number modulator */ 5654 if (dst->seqdiff && !src->seqdiff) { 5655 /* use random iss for the TCP server */ 5656 while ((src->seqdiff = arc4random() - seq) == 0) 5657 ; 5658 ack = ntohl(th->th_ack) - dst->seqdiff; 5659 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 5660 src->seqdiff), 0); 5661 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5662 *copyback = 1; 5663 } else { 5664 ack = ntohl(th->th_ack); 5665 } 5666 5667 end = seq + pd->p_len; 5668 if (th->th_flags & TH_SYN) { 5669 end++; 5670 if (dst->wscale & PF_WSCALE_FLAG) { 5671 src->wscale = pf_get_wscale(pd); 5672 if (src->wscale & PF_WSCALE_FLAG) { 5673 /* Remove scale factor from initial 5674 * window */ 5675 sws = src->wscale & PF_WSCALE_MASK; 5676 win = ((u_int32_t)win + (1 << sws) - 1) 5677 >> sws; 5678 dws = dst->wscale & PF_WSCALE_MASK; 5679 } else { 5680 /* fixup other window */ 5681 dst->max_win = MIN(TCP_MAXWIN, 5682 (u_int32_t)dst->max_win << 5683 (dst->wscale & PF_WSCALE_MASK)); 5684 /* in case of a retrans SYN|ACK */ 5685 dst->wscale = 0; 5686 } 5687 } 5688 } 5689 data_end = end; 5690 if (th->th_flags & TH_FIN) 5691 end++; 5692 5693 src->seqlo = seq; 5694 if (src->state < TCPS_SYN_SENT) 5695 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5696 5697 /* 5698 * May need to slide the window (seqhi may have been set by 5699 * the crappy stack check or if we picked up the connection 5700 * after establishment) 5701 */ 5702 if (src->seqhi == 1 || 5703 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5704 src->seqhi = end + MAX(1, dst->max_win << dws); 5705 if (win > src->max_win) 5706 src->max_win = win; 5707 5708 } else { 5709 ack = ntohl(th->th_ack) - dst->seqdiff; 5710 if (src->seqdiff) { 5711 /* Modulate sequence numbers */ 5712 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 5713 src->seqdiff), 0); 5714 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5715 *copyback = 1; 5716 } 5717 end = seq + pd->p_len; 5718 if (th->th_flags & TH_SYN) 5719 end++; 5720 data_end = end; 5721 if (th->th_flags & TH_FIN) 5722 end++; 5723 } 5724 5725 if ((th->th_flags & TH_ACK) == 0) { 5726 /* Let it pass through the ack skew check */ 5727 ack = dst->seqlo; 5728 } else if ((ack == 0 && 5729 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5730 /* broken tcp stacks do not set ack */ 5731 (dst->state < TCPS_SYN_SENT)) { 5732 /* 5733 * Many stacks (ours included) will set the ACK number in an 5734 * FIN|ACK if the SYN times out -- no sequence to ACK. 5735 */ 5736 ack = dst->seqlo; 5737 } 5738 5739 if (seq == end) { 5740 /* Ease sequencing restrictions on no data packets */ 5741 seq = src->seqlo; 5742 data_end = end = seq; 5743 } 5744 5745 ackskew = dst->seqlo - ack; 5746 5747 /* 5748 * Need to demodulate the sequence numbers in any TCP SACK options 5749 * (Selective ACK). We could optionally validate the SACK values 5750 * against the current ACK window, either forwards or backwards, but 5751 * I'm not confident that SACK has been implemented properly 5752 * everywhere. It wouldn't surprise me if several stacks accidentally 5753 * SACK too far backwards of previously ACKed data. There really aren't 5754 * any security implications of bad SACKing unless the target stack 5755 * doesn't validate the option length correctly. Someone trying to 5756 * spoof into a TCP connection won't bother blindly sending SACK 5757 * options anyway. 5758 */ 5759 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5760 if (pf_modulate_sack(pd, th, dst)) 5761 *copyback = 1; 5762 } 5763 5764 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5765 if (SEQ_GEQ(src->seqhi, data_end) && 5766 /* Last octet inside other's window space */ 5767 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5768 /* Retrans: not more than one window back */ 5769 (ackskew >= -MAXACKWINDOW) && 5770 /* Acking not more than one reassembled fragment backwards */ 5771 (ackskew <= (MAXACKWINDOW << sws)) && 5772 /* Acking not more than one window forward */ 5773 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5774 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 5775 /* Require an exact/+1 sequence match on resets when possible */ 5776 5777 if (dst->scrub || src->scrub) { 5778 if (pf_normalize_tcp_stateful(pd, reason, th, 5779 *state, src, dst, copyback)) 5780 return (PF_DROP); 5781 } 5782 5783 /* update max window */ 5784 if (src->max_win < win) 5785 src->max_win = win; 5786 /* synchronize sequencing */ 5787 if (SEQ_GT(end, src->seqlo)) 5788 src->seqlo = end; 5789 /* slide the window of what the other end can send */ 5790 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5791 dst->seqhi = ack + MAX((win << sws), 1); 5792 5793 /* update states */ 5794 if (th->th_flags & TH_SYN) 5795 if (src->state < TCPS_SYN_SENT) 5796 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5797 if (th->th_flags & TH_FIN) 5798 if (src->state < TCPS_CLOSING) 5799 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5800 if (th->th_flags & TH_ACK) { 5801 if (dst->state == TCPS_SYN_SENT) { 5802 pf_set_protostate(*state, pdst, 5803 TCPS_ESTABLISHED); 5804 if (src->state == TCPS_ESTABLISHED && 5805 (*state)->src_node != NULL && 5806 pf_src_connlimit(state)) { 5807 REASON_SET(reason, PFRES_SRCLIMIT); 5808 return (PF_DROP); 5809 } 5810 } else if (dst->state == TCPS_CLOSING) 5811 pf_set_protostate(*state, pdst, 5812 TCPS_FIN_WAIT_2); 5813 } 5814 if (th->th_flags & TH_RST) 5815 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5816 5817 /* update expire time */ 5818 (*state)->expire = pf_get_uptime(); 5819 if (src->state >= TCPS_FIN_WAIT_2 && 5820 dst->state >= TCPS_FIN_WAIT_2) 5821 (*state)->timeout = PFTM_TCP_CLOSED; 5822 else if (src->state >= TCPS_CLOSING && 5823 dst->state >= TCPS_CLOSING) 5824 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5825 else if (src->state < TCPS_ESTABLISHED || 5826 dst->state < TCPS_ESTABLISHED) 5827 (*state)->timeout = PFTM_TCP_OPENING; 5828 else if (src->state >= TCPS_CLOSING || 5829 dst->state >= TCPS_CLOSING) 5830 (*state)->timeout = PFTM_TCP_CLOSING; 5831 else 5832 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5833 5834 /* Fall through to PASS packet */ 5835 5836 } else if ((dst->state < TCPS_SYN_SENT || 5837 dst->state >= TCPS_FIN_WAIT_2 || 5838 src->state >= TCPS_FIN_WAIT_2) && 5839 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 5840 /* Within a window forward of the originating packet */ 5841 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5842 /* Within a window backward of the originating packet */ 5843 5844 /* 5845 * This currently handles three situations: 5846 * 1) Stupid stacks will shotgun SYNs before their peer 5847 * replies. 5848 * 2) When PF catches an already established stream (the 5849 * firewall rebooted, the state table was flushed, routes 5850 * changed...) 5851 * 3) Packets get funky immediately after the connection 5852 * closes (this should catch Solaris spurious ACK|FINs 5853 * that web servers like to spew after a close) 5854 * 5855 * This must be a little more careful than the above code 5856 * since packet floods will also be caught here. We don't 5857 * update the TTL here to mitigate the damage of a packet 5858 * flood and so the same code can handle awkward establishment 5859 * and a loosened connection close. 5860 * In the establishment case, a correct peer response will 5861 * validate the connection, go through the normal state code 5862 * and keep updating the state TTL. 5863 */ 5864 5865 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5866 printf("pf: loose state match: "); 5867 pf_print_state(*state); 5868 pf_print_flags(th->th_flags); 5869 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5870 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5871 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5872 (unsigned long long)(*state)->packets[1], 5873 pd->dir == PF_IN ? "in" : "out", 5874 pd->dir == (*state)->direction ? "fwd" : "rev"); 5875 } 5876 5877 if (dst->scrub || src->scrub) { 5878 if (pf_normalize_tcp_stateful(pd, reason, th, 5879 *state, src, dst, copyback)) 5880 return (PF_DROP); 5881 } 5882 5883 /* update max window */ 5884 if (src->max_win < win) 5885 src->max_win = win; 5886 /* synchronize sequencing */ 5887 if (SEQ_GT(end, src->seqlo)) 5888 src->seqlo = end; 5889 /* slide the window of what the other end can send */ 5890 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5891 dst->seqhi = ack + MAX((win << sws), 1); 5892 5893 /* 5894 * Cannot set dst->seqhi here since this could be a shotgunned 5895 * SYN and not an already established connection. 5896 */ 5897 5898 if (th->th_flags & TH_FIN) 5899 if (src->state < TCPS_CLOSING) 5900 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5901 if (th->th_flags & TH_RST) 5902 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5903 5904 /* Fall through to PASS packet */ 5905 5906 } else { 5907 if ((*state)->dst.state == TCPS_SYN_SENT && 5908 (*state)->src.state == TCPS_SYN_SENT) { 5909 /* Send RST for state mismatches during handshake */ 5910 if (!(th->th_flags & TH_RST)) 5911 pf_send_tcp((*state)->rule, pd->af, 5912 pd->dst, pd->src, th->th_dport, 5913 th->th_sport, ntohl(th->th_ack), 0, 5914 TH_RST, 0, 0, 5915 (*state)->rule->return_ttl, M_SKIP_FIREWALL, 5916 0, 0, (*state)->act.rtableid); 5917 src->seqlo = 0; 5918 src->seqhi = 1; 5919 src->max_win = 1; 5920 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5921 printf("pf: BAD state: "); 5922 pf_print_state(*state); 5923 pf_print_flags(th->th_flags); 5924 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5925 "pkts=%llu:%llu dir=%s,%s\n", 5926 seq, orig_seq, ack, pd->p_len, ackskew, 5927 (unsigned long long)(*state)->packets[0], 5928 (unsigned long long)(*state)->packets[1], 5929 pd->dir == PF_IN ? "in" : "out", 5930 pd->dir == (*state)->direction ? "fwd" : "rev"); 5931 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5932 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 5933 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5934 ' ': '2', 5935 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5936 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5937 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 5938 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5939 } 5940 REASON_SET(reason, PFRES_BADSTATE); 5941 return (PF_DROP); 5942 } 5943 5944 return (PF_PASS); 5945 } 5946 5947 static int 5948 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5949 { 5950 struct tcphdr *th = &pd->hdr.tcp; 5951 struct pf_state_peer *src, *dst; 5952 u_int8_t psrc, pdst; 5953 5954 if (pd->dir == (*state)->direction) { 5955 src = &(*state)->src; 5956 dst = &(*state)->dst; 5957 psrc = PF_PEER_SRC; 5958 pdst = PF_PEER_DST; 5959 } else { 5960 src = &(*state)->dst; 5961 dst = &(*state)->src; 5962 psrc = PF_PEER_DST; 5963 pdst = PF_PEER_SRC; 5964 } 5965 5966 if (th->th_flags & TH_SYN) 5967 if (src->state < TCPS_SYN_SENT) 5968 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5969 if (th->th_flags & TH_FIN) 5970 if (src->state < TCPS_CLOSING) 5971 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5972 if (th->th_flags & TH_ACK) { 5973 if (dst->state == TCPS_SYN_SENT) { 5974 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 5975 if (src->state == TCPS_ESTABLISHED && 5976 (*state)->src_node != NULL && 5977 pf_src_connlimit(state)) { 5978 REASON_SET(reason, PFRES_SRCLIMIT); 5979 return (PF_DROP); 5980 } 5981 } else if (dst->state == TCPS_CLOSING) { 5982 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 5983 } else if (src->state == TCPS_SYN_SENT && 5984 dst->state < TCPS_SYN_SENT) { 5985 /* 5986 * Handle a special sloppy case where we only see one 5987 * half of the connection. If there is a ACK after 5988 * the initial SYN without ever seeing a packet from 5989 * the destination, set the connection to established. 5990 */ 5991 pf_set_protostate(*state, PF_PEER_BOTH, 5992 TCPS_ESTABLISHED); 5993 dst->state = src->state = TCPS_ESTABLISHED; 5994 if ((*state)->src_node != NULL && 5995 pf_src_connlimit(state)) { 5996 REASON_SET(reason, PFRES_SRCLIMIT); 5997 return (PF_DROP); 5998 } 5999 } else if (src->state == TCPS_CLOSING && 6000 dst->state == TCPS_ESTABLISHED && 6001 dst->seqlo == 0) { 6002 /* 6003 * Handle the closing of half connections where we 6004 * don't see the full bidirectional FIN/ACK+ACK 6005 * handshake. 6006 */ 6007 pf_set_protostate(*state, pdst, TCPS_CLOSING); 6008 } 6009 } 6010 if (th->th_flags & TH_RST) 6011 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6012 6013 /* update expire time */ 6014 (*state)->expire = pf_get_uptime(); 6015 if (src->state >= TCPS_FIN_WAIT_2 && 6016 dst->state >= TCPS_FIN_WAIT_2) 6017 (*state)->timeout = PFTM_TCP_CLOSED; 6018 else if (src->state >= TCPS_CLOSING && 6019 dst->state >= TCPS_CLOSING) 6020 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6021 else if (src->state < TCPS_ESTABLISHED || 6022 dst->state < TCPS_ESTABLISHED) 6023 (*state)->timeout = PFTM_TCP_OPENING; 6024 else if (src->state >= TCPS_CLOSING || 6025 dst->state >= TCPS_CLOSING) 6026 (*state)->timeout = PFTM_TCP_CLOSING; 6027 else 6028 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6029 6030 return (PF_PASS); 6031 } 6032 6033 static int 6034 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 6035 { 6036 struct pf_state_key *sk = (*state)->key[pd->didx]; 6037 struct tcphdr *th = &pd->hdr.tcp; 6038 6039 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 6040 if (pd->dir != (*state)->direction) { 6041 REASON_SET(reason, PFRES_SYNPROXY); 6042 return (PF_SYNPROXY_DROP); 6043 } 6044 if (th->th_flags & TH_SYN) { 6045 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 6046 REASON_SET(reason, PFRES_SYNPROXY); 6047 return (PF_DROP); 6048 } 6049 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6050 pd->src, th->th_dport, th->th_sport, 6051 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 6052 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 6053 M_SKIP_FIREWALL, 0, 0, (*state)->act.rtableid); 6054 REASON_SET(reason, PFRES_SYNPROXY); 6055 return (PF_SYNPROXY_DROP); 6056 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6057 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6058 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6059 REASON_SET(reason, PFRES_SYNPROXY); 6060 return (PF_DROP); 6061 } else if ((*state)->src_node != NULL && 6062 pf_src_connlimit(state)) { 6063 REASON_SET(reason, PFRES_SRCLIMIT); 6064 return (PF_DROP); 6065 } else 6066 pf_set_protostate(*state, PF_PEER_SRC, 6067 PF_TCPS_PROXY_DST); 6068 } 6069 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 6070 if (pd->dir == (*state)->direction) { 6071 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 6072 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6073 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6074 REASON_SET(reason, PFRES_SYNPROXY); 6075 return (PF_DROP); 6076 } 6077 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 6078 if ((*state)->dst.seqhi == 1) 6079 (*state)->dst.seqhi = htonl(arc4random()); 6080 pf_send_tcp((*state)->rule, pd->af, 6081 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6082 sk->port[pd->sidx], sk->port[pd->didx], 6083 (*state)->dst.seqhi, 0, TH_SYN, 0, 6084 (*state)->src.mss, 0, 6085 (*state)->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0, 6086 (*state)->tag, 0, (*state)->act.rtableid); 6087 REASON_SET(reason, PFRES_SYNPROXY); 6088 return (PF_SYNPROXY_DROP); 6089 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 6090 (TH_SYN|TH_ACK)) || 6091 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 6092 REASON_SET(reason, PFRES_SYNPROXY); 6093 return (PF_DROP); 6094 } else { 6095 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 6096 (*state)->dst.seqlo = ntohl(th->th_seq); 6097 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6098 pd->src, th->th_dport, th->th_sport, 6099 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6100 TH_ACK, (*state)->src.max_win, 0, 0, 0, 6101 (*state)->tag, 0, (*state)->act.rtableid); 6102 pf_send_tcp((*state)->rule, pd->af, 6103 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6104 sk->port[pd->sidx], sk->port[pd->didx], 6105 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 6106 TH_ACK, (*state)->dst.max_win, 0, 0, 6107 M_SKIP_FIREWALL, 0, 0, (*state)->act.rtableid); 6108 (*state)->src.seqdiff = (*state)->dst.seqhi - 6109 (*state)->src.seqlo; 6110 (*state)->dst.seqdiff = (*state)->src.seqhi - 6111 (*state)->dst.seqlo; 6112 (*state)->src.seqhi = (*state)->src.seqlo + 6113 (*state)->dst.max_win; 6114 (*state)->dst.seqhi = (*state)->dst.seqlo + 6115 (*state)->src.max_win; 6116 (*state)->src.wscale = (*state)->dst.wscale = 0; 6117 pf_set_protostate(*state, PF_PEER_BOTH, 6118 TCPS_ESTABLISHED); 6119 REASON_SET(reason, PFRES_SYNPROXY); 6120 return (PF_SYNPROXY_DROP); 6121 } 6122 } 6123 6124 return (PF_PASS); 6125 } 6126 6127 static int 6128 pf_test_state_tcp(struct pf_kstate **state, struct pf_pdesc *pd, 6129 u_short *reason) 6130 { 6131 struct pf_state_key_cmp key; 6132 struct tcphdr *th = &pd->hdr.tcp; 6133 int copyback = 0; 6134 int action; 6135 struct pf_state_peer *src, *dst; 6136 6137 bzero(&key, sizeof(key)); 6138 key.af = pd->af; 6139 key.proto = IPPROTO_TCP; 6140 if (pd->dir == PF_IN) { /* wire side, straight */ 6141 PF_ACPY(&key.addr[0], pd->src, key.af); 6142 PF_ACPY(&key.addr[1], pd->dst, key.af); 6143 key.port[0] = th->th_sport; 6144 key.port[1] = th->th_dport; 6145 } else { /* stack side, reverse */ 6146 PF_ACPY(&key.addr[1], pd->src, key.af); 6147 PF_ACPY(&key.addr[0], pd->dst, key.af); 6148 key.port[1] = th->th_sport; 6149 key.port[0] = th->th_dport; 6150 } 6151 6152 STATE_LOOKUP(&key, *state, pd); 6153 6154 if (pd->dir == (*state)->direction) { 6155 src = &(*state)->src; 6156 dst = &(*state)->dst; 6157 } else { 6158 src = &(*state)->dst; 6159 dst = &(*state)->src; 6160 } 6161 6162 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 6163 return (action); 6164 6165 if (dst->state >= TCPS_FIN_WAIT_2 && 6166 src->state >= TCPS_FIN_WAIT_2 && 6167 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 6168 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6169 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6170 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6171 printf("pf: state reuse "); 6172 pf_print_state(*state); 6173 pf_print_flags(th->th_flags); 6174 printf("\n"); 6175 } 6176 /* XXX make sure it's the same direction ?? */ 6177 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6178 pf_unlink_state(*state); 6179 *state = NULL; 6180 return (PF_DROP); 6181 } 6182 6183 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6184 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 6185 return (PF_DROP); 6186 } else { 6187 if (pf_tcp_track_full(state, pd, reason, 6188 ©back) == PF_DROP) 6189 return (PF_DROP); 6190 } 6191 6192 /* translate source/destination address, if necessary */ 6193 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6194 struct pf_state_key *nk = (*state)->key[pd->didx]; 6195 6196 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6197 nk->port[pd->sidx] != th->th_sport) 6198 pf_change_ap(pd->m, pd->src, &th->th_sport, 6199 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 6200 nk->port[pd->sidx], 0, pd->af); 6201 6202 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6203 nk->port[pd->didx] != th->th_dport) 6204 pf_change_ap(pd->m, pd->dst, &th->th_dport, 6205 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 6206 nk->port[pd->didx], 0, pd->af); 6207 copyback = 1; 6208 } 6209 6210 /* Copyback sequence modulation or stateful scrub changes if needed */ 6211 if (copyback) 6212 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th); 6213 6214 return (PF_PASS); 6215 } 6216 6217 static int 6218 pf_test_state_udp(struct pf_kstate **state, struct pf_pdesc *pd) 6219 { 6220 struct pf_state_peer *src, *dst; 6221 struct pf_state_key_cmp key; 6222 struct udphdr *uh = &pd->hdr.udp; 6223 uint8_t psrc, pdst; 6224 6225 bzero(&key, sizeof(key)); 6226 key.af = pd->af; 6227 key.proto = IPPROTO_UDP; 6228 if (pd->dir == PF_IN) { /* wire side, straight */ 6229 PF_ACPY(&key.addr[0], pd->src, key.af); 6230 PF_ACPY(&key.addr[1], pd->dst, key.af); 6231 key.port[0] = uh->uh_sport; 6232 key.port[1] = uh->uh_dport; 6233 } else { /* stack side, reverse */ 6234 PF_ACPY(&key.addr[1], pd->src, key.af); 6235 PF_ACPY(&key.addr[0], pd->dst, key.af); 6236 key.port[1] = uh->uh_sport; 6237 key.port[0] = uh->uh_dport; 6238 } 6239 6240 STATE_LOOKUP(&key, *state, pd); 6241 6242 if (pd->dir == (*state)->direction) { 6243 src = &(*state)->src; 6244 dst = &(*state)->dst; 6245 psrc = PF_PEER_SRC; 6246 pdst = PF_PEER_DST; 6247 } else { 6248 src = &(*state)->dst; 6249 dst = &(*state)->src; 6250 psrc = PF_PEER_DST; 6251 pdst = PF_PEER_SRC; 6252 } 6253 6254 /* update states */ 6255 if (src->state < PFUDPS_SINGLE) 6256 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 6257 if (dst->state == PFUDPS_SINGLE) 6258 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 6259 6260 /* update expire time */ 6261 (*state)->expire = pf_get_uptime(); 6262 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 6263 (*state)->timeout = PFTM_UDP_MULTIPLE; 6264 else 6265 (*state)->timeout = PFTM_UDP_SINGLE; 6266 6267 /* translate source/destination address, if necessary */ 6268 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6269 struct pf_state_key *nk = (*state)->key[pd->didx]; 6270 6271 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6272 nk->port[pd->sidx] != uh->uh_sport) 6273 pf_change_ap(pd->m, pd->src, &uh->uh_sport, pd->ip_sum, 6274 &uh->uh_sum, &nk->addr[pd->sidx], 6275 nk->port[pd->sidx], 1, pd->af); 6276 6277 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6278 nk->port[pd->didx] != uh->uh_dport) 6279 pf_change_ap(pd->m, pd->dst, &uh->uh_dport, pd->ip_sum, 6280 &uh->uh_sum, &nk->addr[pd->didx], 6281 nk->port[pd->didx], 1, pd->af); 6282 m_copyback(pd->m, pd->off, sizeof(*uh), (caddr_t)uh); 6283 } 6284 6285 return (PF_PASS); 6286 } 6287 6288 static int 6289 pf_test_state_sctp(struct pf_kstate **state, struct pf_pdesc *pd, 6290 u_short *reason) 6291 { 6292 struct pf_state_key_cmp key; 6293 struct pf_state_peer *src, *dst; 6294 struct sctphdr *sh = &pd->hdr.sctp; 6295 u_int8_t psrc; //, pdst; 6296 6297 bzero(&key, sizeof(key)); 6298 key.af = pd->af; 6299 key.proto = IPPROTO_SCTP; 6300 if (pd->dir == PF_IN) { /* wire side, straight */ 6301 PF_ACPY(&key.addr[0], pd->src, key.af); 6302 PF_ACPY(&key.addr[1], pd->dst, key.af); 6303 key.port[0] = sh->src_port; 6304 key.port[1] = sh->dest_port; 6305 } else { /* stack side, reverse */ 6306 PF_ACPY(&key.addr[1], pd->src, key.af); 6307 PF_ACPY(&key.addr[0], pd->dst, key.af); 6308 key.port[1] = sh->src_port; 6309 key.port[0] = sh->dest_port; 6310 } 6311 6312 STATE_LOOKUP(&key, *state, pd); 6313 6314 if (pd->dir == (*state)->direction) { 6315 src = &(*state)->src; 6316 dst = &(*state)->dst; 6317 psrc = PF_PEER_SRC; 6318 } else { 6319 src = &(*state)->dst; 6320 dst = &(*state)->src; 6321 psrc = PF_PEER_DST; 6322 } 6323 6324 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 6325 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 6326 pd->sctp_flags & PFDESC_SCTP_INIT) { 6327 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 6328 pf_unlink_state(*state); 6329 *state = NULL; 6330 return (PF_DROP); 6331 } 6332 6333 /* Track state. */ 6334 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 6335 if (src->state < SCTP_COOKIE_WAIT) { 6336 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 6337 (*state)->timeout = PFTM_SCTP_OPENING; 6338 } 6339 } 6340 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 6341 MPASS(dst->scrub != NULL); 6342 if (dst->scrub->pfss_v_tag == 0) 6343 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 6344 } 6345 6346 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 6347 if (src->state < SCTP_ESTABLISHED) { 6348 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 6349 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 6350 } 6351 } 6352 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 6353 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6354 if (src->state < SCTP_SHUTDOWN_PENDING) { 6355 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 6356 (*state)->timeout = PFTM_SCTP_CLOSING; 6357 } 6358 } 6359 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6360 pf_set_protostate(*state, psrc, SCTP_CLOSED); 6361 (*state)->timeout = PFTM_SCTP_CLOSED; 6362 } 6363 6364 if (src->scrub != NULL) { 6365 if (src->scrub->pfss_v_tag == 0) { 6366 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 6367 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 6368 return (PF_DROP); 6369 } 6370 6371 (*state)->expire = pf_get_uptime(); 6372 6373 /* translate source/destination address, if necessary */ 6374 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6375 uint16_t checksum = 0; 6376 struct pf_state_key *nk = (*state)->key[pd->didx]; 6377 6378 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6379 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 6380 pf_change_ap(pd->m, pd->src, &pd->hdr.sctp.src_port, 6381 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 6382 nk->port[pd->sidx], 1, pd->af); 6383 } 6384 6385 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6386 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 6387 pf_change_ap(pd->m, pd->dst, &pd->hdr.sctp.dest_port, 6388 pd->ip_sum, &checksum, &nk->addr[pd->didx], 6389 nk->port[pd->didx], 1, pd->af); 6390 } 6391 } 6392 6393 return (PF_PASS); 6394 } 6395 6396 static void 6397 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 6398 { 6399 struct pf_sctp_endpoint key; 6400 struct pf_sctp_endpoint *ep; 6401 struct pf_state_key *sks = s->key[PF_SK_STACK]; 6402 struct pf_sctp_source *i, *tmp; 6403 6404 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 6405 return; 6406 6407 PF_SCTP_ENDPOINTS_LOCK(); 6408 6409 key.v_tag = s->dst.scrub->pfss_v_tag; 6410 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6411 if (ep != NULL) { 6412 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6413 if (pf_addr_cmp(&i->addr, 6414 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 6415 s->key[PF_SK_WIRE]->af) == 0) { 6416 SDT_PROBE3(pf, sctp, multihome, remove, 6417 key.v_tag, s, i); 6418 TAILQ_REMOVE(&ep->sources, i, entry); 6419 free(i, M_PFTEMP); 6420 break; 6421 } 6422 } 6423 6424 if (TAILQ_EMPTY(&ep->sources)) { 6425 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6426 free(ep, M_PFTEMP); 6427 } 6428 } 6429 6430 /* Other direction. */ 6431 key.v_tag = s->src.scrub->pfss_v_tag; 6432 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6433 if (ep != NULL) { 6434 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6435 if (pf_addr_cmp(&i->addr, 6436 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6437 s->key[PF_SK_WIRE]->af) == 0) { 6438 SDT_PROBE3(pf, sctp, multihome, remove, 6439 key.v_tag, s, i); 6440 TAILQ_REMOVE(&ep->sources, i, entry); 6441 free(i, M_PFTEMP); 6442 break; 6443 } 6444 } 6445 6446 if (TAILQ_EMPTY(&ep->sources)) { 6447 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6448 free(ep, M_PFTEMP); 6449 } 6450 } 6451 6452 PF_SCTP_ENDPOINTS_UNLOCK(); 6453 } 6454 6455 static void 6456 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6457 { 6458 struct pf_sctp_endpoint key = { 6459 .v_tag = v_tag, 6460 }; 6461 struct pf_sctp_source *i; 6462 struct pf_sctp_endpoint *ep; 6463 6464 PF_SCTP_ENDPOINTS_LOCK(); 6465 6466 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6467 if (ep == NULL) { 6468 ep = malloc(sizeof(struct pf_sctp_endpoint), 6469 M_PFTEMP, M_NOWAIT); 6470 if (ep == NULL) { 6471 PF_SCTP_ENDPOINTS_UNLOCK(); 6472 return; 6473 } 6474 6475 ep->v_tag = v_tag; 6476 TAILQ_INIT(&ep->sources); 6477 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6478 } 6479 6480 /* Avoid inserting duplicates. */ 6481 TAILQ_FOREACH(i, &ep->sources, entry) { 6482 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6483 PF_SCTP_ENDPOINTS_UNLOCK(); 6484 return; 6485 } 6486 } 6487 6488 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6489 if (i == NULL) { 6490 PF_SCTP_ENDPOINTS_UNLOCK(); 6491 return; 6492 } 6493 6494 i->af = pd->af; 6495 memcpy(&i->addr, a, sizeof(*a)); 6496 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6497 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 6498 6499 PF_SCTP_ENDPOINTS_UNLOCK(); 6500 } 6501 6502 static void 6503 pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif, 6504 struct pf_kstate *s, int action) 6505 { 6506 struct pf_sctp_multihome_job *j, *tmp; 6507 struct pf_sctp_source *i; 6508 int ret __unused; 6509 struct pf_kstate *sm = NULL; 6510 struct pf_krule *ra = NULL; 6511 struct pf_krule *r = &V_pf_default_rule; 6512 struct pf_kruleset *rs = NULL; 6513 bool do_extra = true; 6514 6515 PF_RULES_RLOCK_TRACKER; 6516 6517 again: 6518 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6519 if (s == NULL || action != PF_PASS) 6520 goto free; 6521 6522 /* Confirm we don't recurse here. */ 6523 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6524 6525 switch (j->op) { 6526 case SCTP_ADD_IP_ADDRESS: { 6527 uint32_t v_tag = pd->sctp_initiate_tag; 6528 6529 if (v_tag == 0) { 6530 if (s->direction == pd->dir) 6531 v_tag = s->src.scrub->pfss_v_tag; 6532 else 6533 v_tag = s->dst.scrub->pfss_v_tag; 6534 } 6535 6536 /* 6537 * Avoid duplicating states. We'll already have 6538 * created a state based on the source address of 6539 * the packet, but SCTP endpoints may also list this 6540 * address again in the INIT(_ACK) parameters. 6541 */ 6542 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6543 break; 6544 } 6545 6546 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6547 PF_RULES_RLOCK(); 6548 sm = NULL; 6549 /* 6550 * New connections need to be floating, because 6551 * we cannot know what interfaces it will use. 6552 * That's why we pass V_pfi_all rather than kif. 6553 */ 6554 j->pd.kif = V_pfi_all; 6555 ret = pf_test_rule(&r, &sm, 6556 &j->pd, &ra, &rs, NULL); 6557 PF_RULES_RUNLOCK(); 6558 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret); 6559 if (ret != PF_DROP && sm != NULL) { 6560 /* Inherit v_tag values. */ 6561 if (sm->direction == s->direction) { 6562 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6563 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6564 } else { 6565 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6566 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6567 } 6568 PF_STATE_UNLOCK(sm); 6569 } else { 6570 /* If we try duplicate inserts? */ 6571 break; 6572 } 6573 6574 /* Only add the address if we've actually allowed the state. */ 6575 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6576 6577 if (! do_extra) { 6578 break; 6579 } 6580 /* 6581 * We need to do this for each of our source addresses. 6582 * Find those based on the verification tag. 6583 */ 6584 struct pf_sctp_endpoint key = { 6585 .v_tag = pd->hdr.sctp.v_tag, 6586 }; 6587 struct pf_sctp_endpoint *ep; 6588 6589 PF_SCTP_ENDPOINTS_LOCK(); 6590 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6591 if (ep == NULL) { 6592 PF_SCTP_ENDPOINTS_UNLOCK(); 6593 break; 6594 } 6595 MPASS(ep != NULL); 6596 6597 TAILQ_FOREACH(i, &ep->sources, entry) { 6598 struct pf_sctp_multihome_job *nj; 6599 6600 /* SCTP can intermingle IPv4 and IPv6. */ 6601 if (i->af != pd->af) 6602 continue; 6603 6604 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6605 if (! nj) { 6606 continue; 6607 } 6608 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6609 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6610 nj->pd.src = &nj->src; 6611 // New destination address! 6612 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6613 nj->pd.dst = &nj->dst; 6614 nj->pd.m = j->pd.m; 6615 nj->op = j->op; 6616 6617 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6618 } 6619 PF_SCTP_ENDPOINTS_UNLOCK(); 6620 6621 break; 6622 } 6623 case SCTP_DEL_IP_ADDRESS: { 6624 struct pf_state_key_cmp key; 6625 uint8_t psrc; 6626 6627 bzero(&key, sizeof(key)); 6628 key.af = j->pd.af; 6629 key.proto = IPPROTO_SCTP; 6630 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6631 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6632 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6633 key.port[0] = j->pd.hdr.sctp.src_port; 6634 key.port[1] = j->pd.hdr.sctp.dest_port; 6635 } else { /* stack side, reverse */ 6636 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6637 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6638 key.port[1] = j->pd.hdr.sctp.src_port; 6639 key.port[0] = j->pd.hdr.sctp.dest_port; 6640 } 6641 6642 sm = pf_find_state(kif, &key, j->pd.dir); 6643 if (sm != NULL) { 6644 PF_STATE_LOCK_ASSERT(sm); 6645 if (j->pd.dir == sm->direction) { 6646 psrc = PF_PEER_SRC; 6647 } else { 6648 psrc = PF_PEER_DST; 6649 } 6650 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6651 sm->timeout = PFTM_SCTP_CLOSING; 6652 PF_STATE_UNLOCK(sm); 6653 } 6654 break; 6655 default: 6656 panic("Unknown op %#x", j->op); 6657 } 6658 } 6659 6660 free: 6661 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6662 free(j, M_PFTEMP); 6663 } 6664 6665 /* We may have inserted extra work while processing the list. */ 6666 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6667 do_extra = false; 6668 goto again; 6669 } 6670 } 6671 6672 static int 6673 pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op) 6674 { 6675 int off = 0; 6676 struct pf_sctp_multihome_job *job; 6677 6678 while (off < len) { 6679 struct sctp_paramhdr h; 6680 6681 if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, NULL, 6682 pd->af)) 6683 return (PF_DROP); 6684 6685 /* Parameters are at least 4 bytes. */ 6686 if (ntohs(h.param_length) < 4) 6687 return (PF_DROP); 6688 6689 switch (ntohs(h.param_type)) { 6690 case SCTP_IPV4_ADDRESS: { 6691 struct in_addr t; 6692 6693 if (ntohs(h.param_length) != 6694 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6695 return (PF_DROP); 6696 6697 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 6698 NULL, NULL, pd->af)) 6699 return (PF_DROP); 6700 6701 if (in_nullhost(t)) 6702 t.s_addr = pd->src->v4.s_addr; 6703 6704 /* 6705 * We hold the state lock (idhash) here, which means 6706 * that we can't acquire the keyhash, or we'll get a 6707 * LOR (and potentially double-lock things too). We also 6708 * can't release the state lock here, so instead we'll 6709 * enqueue this for async handling. 6710 * There's a relatively small race here, in that a 6711 * packet using the new addresses could arrive already, 6712 * but that's just though luck for it. 6713 */ 6714 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6715 if (! job) 6716 return (PF_DROP); 6717 6718 memcpy(&job->pd, pd, sizeof(*pd)); 6719 6720 // New source address! 6721 memcpy(&job->src, &t, sizeof(t)); 6722 job->pd.src = &job->src; 6723 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6724 job->pd.dst = &job->dst; 6725 job->pd.m = pd->m; 6726 job->op = op; 6727 6728 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6729 break; 6730 } 6731 #ifdef INET6 6732 case SCTP_IPV6_ADDRESS: { 6733 struct in6_addr t; 6734 6735 if (ntohs(h.param_length) != 6736 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6737 return (PF_DROP); 6738 6739 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 6740 NULL, NULL, pd->af)) 6741 return (PF_DROP); 6742 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6743 break; 6744 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6745 memcpy(&t, &pd->src->v6, sizeof(t)); 6746 6747 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6748 if (! job) 6749 return (PF_DROP); 6750 6751 memcpy(&job->pd, pd, sizeof(*pd)); 6752 memcpy(&job->src, &t, sizeof(t)); 6753 job->pd.src = &job->src; 6754 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6755 job->pd.dst = &job->dst; 6756 job->pd.m = pd->m; 6757 job->op = op; 6758 6759 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6760 break; 6761 } 6762 #endif 6763 case SCTP_ADD_IP_ADDRESS: { 6764 int ret; 6765 struct sctp_asconf_paramhdr ah; 6766 6767 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 6768 NULL, NULL, pd->af)) 6769 return (PF_DROP); 6770 6771 ret = pf_multihome_scan(start + off + sizeof(ah), 6772 ntohs(ah.ph.param_length) - sizeof(ah), pd, 6773 SCTP_ADD_IP_ADDRESS); 6774 if (ret != PF_PASS) 6775 return (ret); 6776 break; 6777 } 6778 case SCTP_DEL_IP_ADDRESS: { 6779 int ret; 6780 struct sctp_asconf_paramhdr ah; 6781 6782 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 6783 NULL, NULL, pd->af)) 6784 return (PF_DROP); 6785 ret = pf_multihome_scan(start + off + sizeof(ah), 6786 ntohs(ah.ph.param_length) - sizeof(ah), pd, 6787 SCTP_DEL_IP_ADDRESS); 6788 if (ret != PF_PASS) 6789 return (ret); 6790 break; 6791 } 6792 default: 6793 break; 6794 } 6795 6796 off += roundup(ntohs(h.param_length), 4); 6797 } 6798 6799 return (PF_PASS); 6800 } 6801 int 6802 pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd) 6803 { 6804 start += sizeof(struct sctp_init_chunk); 6805 len -= sizeof(struct sctp_init_chunk); 6806 6807 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 6808 } 6809 6810 int 6811 pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd) 6812 { 6813 start += sizeof(struct sctp_asconf_chunk); 6814 len -= sizeof(struct sctp_asconf_chunk); 6815 6816 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 6817 } 6818 6819 int 6820 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 6821 struct pf_kstate **state, int direction, 6822 u_int16_t icmpid, u_int16_t type, int icmp_dir, 6823 int *iidx, int multi, int inner) 6824 { 6825 key->af = pd->af; 6826 key->proto = pd->proto; 6827 if (icmp_dir == PF_IN) { 6828 *iidx = pd->sidx; 6829 key->port[pd->sidx] = icmpid; 6830 key->port[pd->didx] = type; 6831 } else { 6832 *iidx = pd->didx; 6833 key->port[pd->sidx] = type; 6834 key->port[pd->didx] = icmpid; 6835 } 6836 if (pf_state_key_addr_setup(pd, key, multi)) 6837 return (PF_DROP); 6838 6839 STATE_LOOKUP(key, *state, pd); 6840 6841 if ((*state)->state_flags & PFSTATE_SLOPPY) 6842 return (-1); 6843 6844 /* Is this ICMP message flowing in right direction? */ 6845 if ((*state)->rule->type && 6846 (((!inner && (*state)->direction == direction) || 6847 (inner && (*state)->direction != direction)) ? 6848 PF_IN : PF_OUT) != icmp_dir) { 6849 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6850 printf("pf: icmp type %d in wrong direction (%d): ", 6851 ntohs(type), icmp_dir); 6852 pf_print_state(*state); 6853 printf("\n"); 6854 } 6855 PF_STATE_UNLOCK(*state); 6856 *state = NULL; 6857 return (PF_DROP); 6858 } 6859 return (-1); 6860 } 6861 6862 static int 6863 pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd, 6864 u_short *reason) 6865 { 6866 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6867 u_int16_t *icmpsum, virtual_id, virtual_type; 6868 u_int8_t icmptype, icmpcode; 6869 int icmp_dir, iidx, ret, multi; 6870 struct pf_state_key_cmp key; 6871 #ifdef INET 6872 u_int16_t icmpid; 6873 #endif 6874 6875 MPASS(*state == NULL); 6876 6877 bzero(&key, sizeof(key)); 6878 switch (pd->proto) { 6879 #ifdef INET 6880 case IPPROTO_ICMP: 6881 icmptype = pd->hdr.icmp.icmp_type; 6882 icmpcode = pd->hdr.icmp.icmp_code; 6883 icmpid = pd->hdr.icmp.icmp_id; 6884 icmpsum = &pd->hdr.icmp.icmp_cksum; 6885 break; 6886 #endif /* INET */ 6887 #ifdef INET6 6888 case IPPROTO_ICMPV6: 6889 icmptype = pd->hdr.icmp6.icmp6_type; 6890 icmpcode = pd->hdr.icmp6.icmp6_code; 6891 #ifdef INET 6892 icmpid = pd->hdr.icmp6.icmp6_id; 6893 #endif 6894 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 6895 break; 6896 #endif /* INET6 */ 6897 } 6898 6899 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi, 6900 &virtual_id, &virtual_type) == 0) { 6901 /* 6902 * ICMP query/reply message not related to a TCP/UDP packet. 6903 * Search for an ICMP state. 6904 */ 6905 ret = pf_icmp_state_lookup(&key, pd, state, pd->dir, 6906 virtual_id, virtual_type, icmp_dir, &iidx, 6907 PF_ICMP_MULTI_NONE, 0); 6908 if (ret >= 0) { 6909 MPASS(*state == NULL); 6910 if (ret == PF_DROP && pd->af == AF_INET6 && 6911 icmp_dir == PF_OUT) { 6912 ret = pf_icmp_state_lookup(&key, pd, state, 6913 pd->dir, virtual_id, virtual_type, 6914 icmp_dir, &iidx, multi, 0); 6915 if (ret >= 0) { 6916 MPASS(*state == NULL); 6917 return (ret); 6918 } 6919 } else 6920 return (ret); 6921 } 6922 6923 (*state)->expire = pf_get_uptime(); 6924 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 6925 6926 /* translate source/destination address, if necessary */ 6927 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6928 struct pf_state_key *nk = (*state)->key[pd->didx]; 6929 6930 switch (pd->af) { 6931 #ifdef INET 6932 case AF_INET: 6933 if (PF_ANEQ(pd->src, 6934 &nk->addr[pd->sidx], AF_INET)) 6935 pf_change_a(&saddr->v4.s_addr, 6936 pd->ip_sum, 6937 nk->addr[pd->sidx].v4.s_addr, 0); 6938 6939 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 6940 AF_INET)) 6941 pf_change_a(&daddr->v4.s_addr, 6942 pd->ip_sum, 6943 nk->addr[pd->didx].v4.s_addr, 0); 6944 6945 if (nk->port[iidx] != 6946 pd->hdr.icmp.icmp_id) { 6947 pd->hdr.icmp.icmp_cksum = 6948 pf_cksum_fixup( 6949 pd->hdr.icmp.icmp_cksum, icmpid, 6950 nk->port[iidx], 0); 6951 pd->hdr.icmp.icmp_id = 6952 nk->port[iidx]; 6953 } 6954 6955 m_copyback(pd->m, pd->off, ICMP_MINLEN, 6956 (caddr_t )&pd->hdr.icmp); 6957 break; 6958 #endif /* INET */ 6959 #ifdef INET6 6960 case AF_INET6: 6961 if (PF_ANEQ(pd->src, 6962 &nk->addr[pd->sidx], AF_INET6)) 6963 pf_change_a6(saddr, 6964 &pd->hdr.icmp6.icmp6_cksum, 6965 &nk->addr[pd->sidx], 0); 6966 6967 if (PF_ANEQ(pd->dst, 6968 &nk->addr[pd->didx], AF_INET6)) 6969 pf_change_a6(daddr, 6970 &pd->hdr.icmp6.icmp6_cksum, 6971 &nk->addr[pd->didx], 0); 6972 6973 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 6974 (caddr_t )&pd->hdr.icmp6); 6975 break; 6976 #endif /* INET6 */ 6977 } 6978 } 6979 return (PF_PASS); 6980 6981 } else { 6982 /* 6983 * ICMP error message in response to a TCP/UDP packet. 6984 * Extract the inner TCP/UDP header and search for that state. 6985 */ 6986 6987 struct pf_pdesc pd2; 6988 bzero(&pd2, sizeof pd2); 6989 #ifdef INET 6990 struct ip h2; 6991 #endif /* INET */ 6992 #ifdef INET6 6993 struct ip6_hdr h2_6; 6994 int fragoff2, extoff2; 6995 u_int32_t jumbolen; 6996 #endif /* INET6 */ 6997 int ipoff2 = 0; 6998 6999 pd2.af = pd->af; 7000 pd2.dir = pd->dir; 7001 /* Payload packet is from the opposite direction. */ 7002 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7003 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7004 pd2.m = pd->m; 7005 switch (pd->af) { 7006 #ifdef INET 7007 case AF_INET: 7008 /* offset of h2 in mbuf chain */ 7009 ipoff2 = pd->off + ICMP_MINLEN; 7010 7011 if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2), 7012 NULL, reason, pd2.af)) { 7013 DPFPRINTF(PF_DEBUG_MISC, 7014 ("pf: ICMP error message too short " 7015 "(ip)\n")); 7016 return (PF_DROP); 7017 } 7018 /* 7019 * ICMP error messages don't refer to non-first 7020 * fragments 7021 */ 7022 if (h2.ip_off & htons(IP_OFFMASK)) { 7023 REASON_SET(reason, PFRES_FRAG); 7024 return (PF_DROP); 7025 } 7026 7027 /* offset of protocol header that follows h2 */ 7028 pd2.off = ipoff2 + (h2.ip_hl << 2); 7029 7030 pd2.proto = h2.ip_p; 7031 pd2.src = (struct pf_addr *)&h2.ip_src; 7032 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7033 pd2.ip_sum = &h2.ip_sum; 7034 break; 7035 #endif /* INET */ 7036 #ifdef INET6 7037 case AF_INET6: 7038 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 7039 7040 if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6), 7041 NULL, reason, pd2.af)) { 7042 DPFPRINTF(PF_DEBUG_MISC, 7043 ("pf: ICMP error message too short " 7044 "(ip6)\n")); 7045 return (PF_DROP); 7046 } 7047 pd2.off = ipoff2; 7048 if (pf_walk_header6(pd->m, &h2_6, &pd2.off, &extoff2, 7049 &fragoff2, &pd2.proto, &jumbolen, 7050 reason) != PF_PASS) 7051 return (PF_DROP); 7052 7053 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7054 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7055 pd2.ip_sum = NULL; 7056 break; 7057 #endif /* INET6 */ 7058 } 7059 7060 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 7061 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7062 printf("pf: BAD ICMP %d:%d outer dst: ", 7063 icmptype, icmpcode); 7064 pf_print_host(pd->src, 0, pd->af); 7065 printf(" -> "); 7066 pf_print_host(pd->dst, 0, pd->af); 7067 printf(" inner src: "); 7068 pf_print_host(pd2.src, 0, pd2.af); 7069 printf(" -> "); 7070 pf_print_host(pd2.dst, 0, pd2.af); 7071 printf("\n"); 7072 } 7073 REASON_SET(reason, PFRES_BADSTATE); 7074 return (PF_DROP); 7075 } 7076 7077 switch (pd2.proto) { 7078 case IPPROTO_TCP: { 7079 struct tcphdr th; 7080 u_int32_t seq; 7081 struct pf_state_peer *src, *dst; 7082 u_int8_t dws; 7083 int copyback = 0; 7084 7085 /* 7086 * Only the first 8 bytes of the TCP header can be 7087 * expected. Don't access any TCP header fields after 7088 * th_seq, an ackskew test is not possible. 7089 */ 7090 if (!pf_pull_hdr(pd->m, pd2.off, &th, 8, NULL, reason, 7091 pd2.af)) { 7092 DPFPRINTF(PF_DEBUG_MISC, 7093 ("pf: ICMP error message too short " 7094 "(tcp)\n")); 7095 return (PF_DROP); 7096 } 7097 7098 key.af = pd2.af; 7099 key.proto = IPPROTO_TCP; 7100 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7101 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7102 key.port[pd2.sidx] = th.th_sport; 7103 key.port[pd2.didx] = th.th_dport; 7104 7105 STATE_LOOKUP(&key, *state, pd); 7106 7107 if (pd->dir == (*state)->direction) { 7108 src = &(*state)->dst; 7109 dst = &(*state)->src; 7110 } else { 7111 src = &(*state)->src; 7112 dst = &(*state)->dst; 7113 } 7114 7115 if (src->wscale && dst->wscale) 7116 dws = dst->wscale & PF_WSCALE_MASK; 7117 else 7118 dws = 0; 7119 7120 /* Demodulate sequence number */ 7121 seq = ntohl(th.th_seq) - src->seqdiff; 7122 if (src->seqdiff) { 7123 pf_change_a(&th.th_seq, icmpsum, 7124 htonl(seq), 0); 7125 copyback = 1; 7126 } 7127 7128 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7129 (!SEQ_GEQ(src->seqhi, seq) || 7130 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7131 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7132 printf("pf: BAD ICMP %d:%d ", 7133 icmptype, icmpcode); 7134 pf_print_host(pd->src, 0, pd->af); 7135 printf(" -> "); 7136 pf_print_host(pd->dst, 0, pd->af); 7137 printf(" state: "); 7138 pf_print_state(*state); 7139 printf(" seq=%u\n", seq); 7140 } 7141 REASON_SET(reason, PFRES_BADSTATE); 7142 return (PF_DROP); 7143 } else { 7144 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7145 printf("pf: OK ICMP %d:%d ", 7146 icmptype, icmpcode); 7147 pf_print_host(pd->src, 0, pd->af); 7148 printf(" -> "); 7149 pf_print_host(pd->dst, 0, pd->af); 7150 printf(" state: "); 7151 pf_print_state(*state); 7152 printf(" seq=%u\n", seq); 7153 } 7154 } 7155 7156 /* translate source/destination address, if necessary */ 7157 if ((*state)->key[PF_SK_WIRE] != 7158 (*state)->key[PF_SK_STACK]) { 7159 struct pf_state_key *nk = 7160 (*state)->key[pd->didx]; 7161 7162 if (PF_ANEQ(pd2.src, 7163 &nk->addr[pd2.sidx], pd2.af) || 7164 nk->port[pd2.sidx] != th.th_sport) 7165 pf_change_icmp(pd2.src, &th.th_sport, 7166 daddr, &nk->addr[pd2.sidx], 7167 nk->port[pd2.sidx], NULL, 7168 pd2.ip_sum, icmpsum, 7169 pd->ip_sum, 0, pd2.af); 7170 7171 if (PF_ANEQ(pd2.dst, 7172 &nk->addr[pd2.didx], pd2.af) || 7173 nk->port[pd2.didx] != th.th_dport) 7174 pf_change_icmp(pd2.dst, &th.th_dport, 7175 saddr, &nk->addr[pd2.didx], 7176 nk->port[pd2.didx], NULL, 7177 pd2.ip_sum, icmpsum, 7178 pd->ip_sum, 0, pd2.af); 7179 copyback = 1; 7180 } 7181 7182 if (copyback) { 7183 switch (pd2.af) { 7184 #ifdef INET 7185 case AF_INET: 7186 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7187 (caddr_t )&pd->hdr.icmp); 7188 m_copyback(pd->m, ipoff2, sizeof(h2), 7189 (caddr_t )&h2); 7190 break; 7191 #endif /* INET */ 7192 #ifdef INET6 7193 case AF_INET6: 7194 m_copyback(pd->m, pd->off, 7195 sizeof(struct icmp6_hdr), 7196 (caddr_t )&pd->hdr.icmp6); 7197 m_copyback(pd->m, ipoff2, sizeof(h2_6), 7198 (caddr_t )&h2_6); 7199 break; 7200 #endif /* INET6 */ 7201 } 7202 m_copyback(pd->m, pd2.off, 8, (caddr_t)&th); 7203 } 7204 7205 return (PF_PASS); 7206 break; 7207 } 7208 case IPPROTO_UDP: { 7209 struct udphdr uh; 7210 7211 if (!pf_pull_hdr(pd->m, pd2.off, &uh, sizeof(uh), 7212 NULL, reason, pd2.af)) { 7213 DPFPRINTF(PF_DEBUG_MISC, 7214 ("pf: ICMP error message too short " 7215 "(udp)\n")); 7216 return (PF_DROP); 7217 } 7218 7219 key.af = pd2.af; 7220 key.proto = IPPROTO_UDP; 7221 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7222 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7223 key.port[pd2.sidx] = uh.uh_sport; 7224 key.port[pd2.didx] = uh.uh_dport; 7225 7226 STATE_LOOKUP(&key, *state, pd); 7227 7228 /* translate source/destination address, if necessary */ 7229 if ((*state)->key[PF_SK_WIRE] != 7230 (*state)->key[PF_SK_STACK]) { 7231 struct pf_state_key *nk = 7232 (*state)->key[pd->didx]; 7233 7234 if (PF_ANEQ(pd2.src, 7235 &nk->addr[pd2.sidx], pd2.af) || 7236 nk->port[pd2.sidx] != uh.uh_sport) 7237 pf_change_icmp(pd2.src, &uh.uh_sport, 7238 daddr, &nk->addr[pd2.sidx], 7239 nk->port[pd2.sidx], &uh.uh_sum, 7240 pd2.ip_sum, icmpsum, 7241 pd->ip_sum, 1, pd2.af); 7242 7243 if (PF_ANEQ(pd2.dst, 7244 &nk->addr[pd2.didx], pd2.af) || 7245 nk->port[pd2.didx] != uh.uh_dport) 7246 pf_change_icmp(pd2.dst, &uh.uh_dport, 7247 saddr, &nk->addr[pd2.didx], 7248 nk->port[pd2.didx], &uh.uh_sum, 7249 pd2.ip_sum, icmpsum, 7250 pd->ip_sum, 1, pd2.af); 7251 7252 switch (pd2.af) { 7253 #ifdef INET 7254 case AF_INET: 7255 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7256 (caddr_t )&pd->hdr.icmp); 7257 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 7258 break; 7259 #endif /* INET */ 7260 #ifdef INET6 7261 case AF_INET6: 7262 m_copyback(pd->m, pd->off, 7263 sizeof(struct icmp6_hdr), 7264 (caddr_t )&pd->hdr.icmp6); 7265 m_copyback(pd->m, ipoff2, sizeof(h2_6), 7266 (caddr_t )&h2_6); 7267 break; 7268 #endif /* INET6 */ 7269 } 7270 m_copyback(pd->m, pd2.off, sizeof(uh), (caddr_t)&uh); 7271 } 7272 return (PF_PASS); 7273 break; 7274 } 7275 #ifdef INET 7276 case IPPROTO_ICMP: { 7277 struct icmp *iih = &pd2.hdr.icmp; 7278 7279 if (pd2.af != AF_INET) { 7280 REASON_SET(reason, PFRES_NORM); 7281 return (PF_DROP); 7282 } 7283 7284 if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN, 7285 NULL, reason, pd2.af)) { 7286 DPFPRINTF(PF_DEBUG_MISC, 7287 ("pf: ICMP error message too short i" 7288 "(icmp)\n")); 7289 return (PF_DROP); 7290 } 7291 7292 icmpid = iih->icmp_id; 7293 pf_icmp_mapping(&pd2, iih->icmp_type, 7294 &icmp_dir, &multi, &virtual_id, &virtual_type); 7295 7296 ret = pf_icmp_state_lookup(&key, &pd2, state, 7297 pd2.dir, virtual_id, virtual_type, 7298 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7299 if (ret >= 0) { 7300 MPASS(*state == NULL); 7301 return (ret); 7302 } 7303 7304 /* translate source/destination address, if necessary */ 7305 if ((*state)->key[PF_SK_WIRE] != 7306 (*state)->key[PF_SK_STACK]) { 7307 struct pf_state_key *nk = 7308 (*state)->key[pd->didx]; 7309 7310 if (PF_ANEQ(pd2.src, 7311 &nk->addr[pd2.sidx], pd2.af) || 7312 (virtual_type == htons(ICMP_ECHO) && 7313 nk->port[iidx] != iih->icmp_id)) 7314 pf_change_icmp(pd2.src, 7315 (virtual_type == htons(ICMP_ECHO)) ? 7316 &iih->icmp_id : NULL, 7317 daddr, &nk->addr[pd2.sidx], 7318 (virtual_type == htons(ICMP_ECHO)) ? 7319 nk->port[iidx] : 0, NULL, 7320 pd2.ip_sum, icmpsum, 7321 pd->ip_sum, 0, AF_INET); 7322 7323 if (PF_ANEQ(pd2.dst, 7324 &nk->addr[pd2.didx], pd2.af)) 7325 pf_change_icmp(pd2.dst, NULL, NULL, 7326 &nk->addr[pd2.didx], 0, NULL, 7327 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 7328 AF_INET); 7329 7330 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 7331 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 7332 m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih); 7333 } 7334 return (PF_PASS); 7335 break; 7336 } 7337 #endif /* INET */ 7338 #ifdef INET6 7339 case IPPROTO_ICMPV6: { 7340 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 7341 7342 if (pd2.af != AF_INET6) { 7343 REASON_SET(reason, PFRES_NORM); 7344 return (PF_DROP); 7345 } 7346 7347 if (!pf_pull_hdr(pd->m, pd2.off, iih, 7348 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 7349 DPFPRINTF(PF_DEBUG_MISC, 7350 ("pf: ICMP error message too short " 7351 "(icmp6)\n")); 7352 return (PF_DROP); 7353 } 7354 7355 pf_icmp_mapping(&pd2, iih->icmp6_type, 7356 &icmp_dir, &multi, &virtual_id, &virtual_type); 7357 7358 ret = pf_icmp_state_lookup(&key, &pd2, state, 7359 pd->dir, virtual_id, virtual_type, 7360 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7361 if (ret >= 0) { 7362 MPASS(*state == NULL); 7363 if (ret == PF_DROP && pd2.af == AF_INET6 && 7364 icmp_dir == PF_OUT) { 7365 ret = pf_icmp_state_lookup(&key, &pd2, 7366 state, pd->dir, 7367 virtual_id, virtual_type, 7368 icmp_dir, &iidx, multi, 1); 7369 if (ret >= 0) { 7370 MPASS(*state == NULL); 7371 return (ret); 7372 } 7373 } else 7374 return (ret); 7375 } 7376 7377 /* translate source/destination address, if necessary */ 7378 if ((*state)->key[PF_SK_WIRE] != 7379 (*state)->key[PF_SK_STACK]) { 7380 struct pf_state_key *nk = 7381 (*state)->key[pd->didx]; 7382 7383 if (PF_ANEQ(pd2.src, 7384 &nk->addr[pd2.sidx], pd2.af) || 7385 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 7386 nk->port[pd2.sidx] != iih->icmp6_id)) 7387 pf_change_icmp(pd2.src, 7388 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7389 ? &iih->icmp6_id : NULL, 7390 daddr, &nk->addr[pd2.sidx], 7391 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7392 ? nk->port[iidx] : 0, NULL, 7393 pd2.ip_sum, icmpsum, 7394 pd->ip_sum, 0, AF_INET6); 7395 7396 if (PF_ANEQ(pd2.dst, 7397 &nk->addr[pd2.didx], pd2.af)) 7398 pf_change_icmp(pd2.dst, NULL, NULL, 7399 &nk->addr[pd2.didx], 0, NULL, 7400 pd2.ip_sum, icmpsum, 7401 pd->ip_sum, 0, AF_INET6); 7402 7403 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 7404 (caddr_t)&pd->hdr.icmp6); 7405 m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 7406 m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr), 7407 (caddr_t)iih); 7408 } 7409 return (PF_PASS); 7410 break; 7411 } 7412 #endif /* INET6 */ 7413 default: { 7414 key.af = pd2.af; 7415 key.proto = pd2.proto; 7416 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7417 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7418 key.port[0] = key.port[1] = 0; 7419 7420 STATE_LOOKUP(&key, *state, pd); 7421 7422 /* translate source/destination address, if necessary */ 7423 if ((*state)->key[PF_SK_WIRE] != 7424 (*state)->key[PF_SK_STACK]) { 7425 struct pf_state_key *nk = 7426 (*state)->key[pd->didx]; 7427 7428 if (PF_ANEQ(pd2.src, 7429 &nk->addr[pd2.sidx], pd2.af)) 7430 pf_change_icmp(pd2.src, NULL, daddr, 7431 &nk->addr[pd2.sidx], 0, NULL, 7432 pd2.ip_sum, icmpsum, 7433 pd->ip_sum, 0, pd2.af); 7434 7435 if (PF_ANEQ(pd2.dst, 7436 &nk->addr[pd2.didx], pd2.af)) 7437 pf_change_icmp(pd2.dst, NULL, saddr, 7438 &nk->addr[pd2.didx], 0, NULL, 7439 pd2.ip_sum, icmpsum, 7440 pd->ip_sum, 0, pd2.af); 7441 7442 switch (pd2.af) { 7443 #ifdef INET 7444 case AF_INET: 7445 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7446 (caddr_t)&pd->hdr.icmp); 7447 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 7448 break; 7449 #endif /* INET */ 7450 #ifdef INET6 7451 case AF_INET6: 7452 m_copyback(pd->m, pd->off, 7453 sizeof(struct icmp6_hdr), 7454 (caddr_t )&pd->hdr.icmp6); 7455 m_copyback(pd->m, ipoff2, sizeof(h2_6), 7456 (caddr_t )&h2_6); 7457 break; 7458 #endif /* INET6 */ 7459 } 7460 } 7461 return (PF_PASS); 7462 break; 7463 } 7464 } 7465 } 7466 } 7467 7468 static int 7469 pf_test_state_other(struct pf_kstate **state, struct pf_pdesc *pd) 7470 { 7471 struct pf_state_peer *src, *dst; 7472 struct pf_state_key_cmp key; 7473 uint8_t psrc, pdst; 7474 7475 bzero(&key, sizeof(key)); 7476 key.af = pd->af; 7477 key.proto = pd->proto; 7478 if (pd->dir == PF_IN) { 7479 PF_ACPY(&key.addr[0], pd->src, key.af); 7480 PF_ACPY(&key.addr[1], pd->dst, key.af); 7481 key.port[0] = key.port[1] = 0; 7482 } else { 7483 PF_ACPY(&key.addr[1], pd->src, key.af); 7484 PF_ACPY(&key.addr[0], pd->dst, key.af); 7485 key.port[1] = key.port[0] = 0; 7486 } 7487 7488 STATE_LOOKUP(&key, *state, pd); 7489 7490 if (pd->dir == (*state)->direction) { 7491 src = &(*state)->src; 7492 dst = &(*state)->dst; 7493 psrc = PF_PEER_SRC; 7494 pdst = PF_PEER_DST; 7495 } else { 7496 src = &(*state)->dst; 7497 dst = &(*state)->src; 7498 psrc = PF_PEER_DST; 7499 pdst = PF_PEER_SRC; 7500 } 7501 7502 /* update states */ 7503 if (src->state < PFOTHERS_SINGLE) 7504 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7505 if (dst->state == PFOTHERS_SINGLE) 7506 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7507 7508 /* update expire time */ 7509 (*state)->expire = pf_get_uptime(); 7510 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7511 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7512 else 7513 (*state)->timeout = PFTM_OTHER_SINGLE; 7514 7515 /* translate source/destination address, if necessary */ 7516 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7517 struct pf_state_key *nk = (*state)->key[pd->didx]; 7518 7519 KASSERT(nk, ("%s: nk is null", __func__)); 7520 KASSERT(pd, ("%s: pd is null", __func__)); 7521 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7522 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7523 switch (pd->af) { 7524 #ifdef INET 7525 case AF_INET: 7526 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7527 pf_change_a(&pd->src->v4.s_addr, 7528 pd->ip_sum, 7529 nk->addr[pd->sidx].v4.s_addr, 7530 0); 7531 7532 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7533 pf_change_a(&pd->dst->v4.s_addr, 7534 pd->ip_sum, 7535 nk->addr[pd->didx].v4.s_addr, 7536 0); 7537 7538 break; 7539 #endif /* INET */ 7540 #ifdef INET6 7541 case AF_INET6: 7542 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6)) 7543 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7544 7545 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6)) 7546 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7547 #endif /* INET6 */ 7548 } 7549 } 7550 return (PF_PASS); 7551 } 7552 7553 /* 7554 * ipoff and off are measured from the start of the mbuf chain. 7555 * h must be at "ipoff" on the mbuf chain. 7556 */ 7557 void * 7558 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 7559 u_short *actionp, u_short *reasonp, sa_family_t af) 7560 { 7561 switch (af) { 7562 #ifdef INET 7563 case AF_INET: { 7564 const struct ip *h = mtod(m, struct ip *); 7565 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7566 7567 if (fragoff) { 7568 if (fragoff >= len) 7569 ACTION_SET(actionp, PF_PASS); 7570 else { 7571 ACTION_SET(actionp, PF_DROP); 7572 REASON_SET(reasonp, PFRES_FRAG); 7573 } 7574 return (NULL); 7575 } 7576 if (m->m_pkthdr.len < off + len || 7577 ntohs(h->ip_len) < off + len) { 7578 ACTION_SET(actionp, PF_DROP); 7579 REASON_SET(reasonp, PFRES_SHORT); 7580 return (NULL); 7581 } 7582 break; 7583 } 7584 #endif /* INET */ 7585 #ifdef INET6 7586 case AF_INET6: { 7587 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7588 7589 if (m->m_pkthdr.len < off + len || 7590 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7591 (unsigned)(off + len)) { 7592 ACTION_SET(actionp, PF_DROP); 7593 REASON_SET(reasonp, PFRES_SHORT); 7594 return (NULL); 7595 } 7596 break; 7597 } 7598 #endif /* INET6 */ 7599 } 7600 m_copydata(m, off, len, p); 7601 return (p); 7602 } 7603 7604 int 7605 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7606 int rtableid) 7607 { 7608 struct ifnet *ifp; 7609 7610 /* 7611 * Skip check for addresses with embedded interface scope, 7612 * as they would always match anyway. 7613 */ 7614 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7615 return (1); 7616 7617 if (af != AF_INET && af != AF_INET6) 7618 return (0); 7619 7620 if (kif == V_pfi_all) 7621 return (1); 7622 7623 /* Skip checks for ipsec interfaces */ 7624 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7625 return (1); 7626 7627 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7628 7629 switch (af) { 7630 #ifdef INET6 7631 case AF_INET6: 7632 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7633 ifp)); 7634 #endif 7635 #ifdef INET 7636 case AF_INET: 7637 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7638 ifp)); 7639 #endif 7640 } 7641 7642 return (0); 7643 } 7644 7645 #ifdef INET 7646 static void 7647 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7648 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7649 { 7650 struct mbuf *m0, *m1, *md; 7651 struct sockaddr_in dst; 7652 struct ip *ip; 7653 struct pfi_kkif *nkif = NULL; 7654 struct ifnet *ifp = NULL; 7655 struct pf_addr naddr; 7656 int error = 0; 7657 uint16_t ip_len, ip_off; 7658 uint16_t tmp; 7659 int r_rt, r_dir; 7660 7661 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7662 7663 if (s) { 7664 r_rt = s->rt; 7665 r_dir = s->direction; 7666 } else { 7667 r_rt = r->rt; 7668 r_dir = r->direction; 7669 } 7670 7671 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7672 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7673 __func__)); 7674 7675 if ((pd->pf_mtag == NULL && 7676 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7677 pd->pf_mtag->routed++ > 3) { 7678 m0 = *m; 7679 *m = NULL; 7680 goto bad_locked; 7681 } 7682 7683 if (r_rt == PF_DUPTO) { 7684 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7685 if (s == NULL) { 7686 ifp = r->rpool.cur->kif ? 7687 r->rpool.cur->kif->pfik_ifp : NULL; 7688 } else { 7689 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7690 /* If pfsync'd */ 7691 if (ifp == NULL && r->rpool.cur != NULL) 7692 ifp = r->rpool.cur->kif ? 7693 r->rpool.cur->kif->pfik_ifp : NULL; 7694 PF_STATE_UNLOCK(s); 7695 } 7696 if (ifp == oifp) { 7697 /* When the 2nd interface is not skipped */ 7698 return; 7699 } else { 7700 m0 = *m; 7701 *m = NULL; 7702 goto bad; 7703 } 7704 } else { 7705 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7706 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7707 if (s) 7708 PF_STATE_UNLOCK(s); 7709 return; 7710 } 7711 } 7712 } else { 7713 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7714 pf_dummynet(pd, s, r, m); 7715 if (s) 7716 PF_STATE_UNLOCK(s); 7717 return; 7718 } 7719 m0 = *m; 7720 } 7721 7722 ip = mtod(m0, struct ip *); 7723 7724 bzero(&dst, sizeof(dst)); 7725 dst.sin_family = AF_INET; 7726 dst.sin_len = sizeof(dst); 7727 dst.sin_addr = ip->ip_dst; 7728 7729 bzero(&naddr, sizeof(naddr)); 7730 7731 if (s == NULL) { 7732 if (TAILQ_EMPTY(&r->rpool.list)) { 7733 DPFPRINTF(PF_DEBUG_URGENT, 7734 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7735 goto bad_locked; 7736 } 7737 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7738 &naddr, &nkif, NULL); 7739 if (!PF_AZERO(&naddr, AF_INET)) 7740 dst.sin_addr.s_addr = naddr.v4.s_addr; 7741 ifp = nkif ? nkif->pfik_ifp : NULL; 7742 } else { 7743 struct pfi_kkif *kif; 7744 7745 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7746 dst.sin_addr.s_addr = 7747 s->rt_addr.v4.s_addr; 7748 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7749 kif = s->rt_kif; 7750 /* If pfsync'd */ 7751 if (ifp == NULL && r->rpool.cur != NULL) { 7752 ifp = r->rpool.cur->kif ? 7753 r->rpool.cur->kif->pfik_ifp : NULL; 7754 kif = r->rpool.cur->kif; 7755 } 7756 if (ifp != NULL && kif != NULL && 7757 r->rule_flag & PFRULE_IFBOUND && 7758 r->rt == PF_REPLYTO && 7759 s->kif == V_pfi_all) { 7760 s->kif = kif; 7761 s->orig_kif = oifp->if_pf_kif; 7762 } 7763 7764 PF_STATE_UNLOCK(s); 7765 } 7766 7767 if (ifp == NULL) 7768 goto bad; 7769 7770 if (pd->dir == PF_IN) { 7771 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 7772 &pd->act) != PF_PASS) 7773 goto bad; 7774 else if (m0 == NULL) 7775 goto done; 7776 if (m0->m_len < sizeof(struct ip)) { 7777 DPFPRINTF(PF_DEBUG_URGENT, 7778 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7779 goto bad; 7780 } 7781 ip = mtod(m0, struct ip *); 7782 } 7783 7784 if (ifp->if_flags & IFF_LOOPBACK) 7785 m0->m_flags |= M_SKIP_FIREWALL; 7786 7787 ip_len = ntohs(ip->ip_len); 7788 ip_off = ntohs(ip->ip_off); 7789 7790 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7791 m0->m_pkthdr.csum_flags |= CSUM_IP; 7792 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7793 in_delayed_cksum(m0); 7794 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7795 } 7796 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7797 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7798 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7799 } 7800 7801 if (pd->dir == PF_IN) { 7802 /* 7803 * Make sure dummynet gets the correct direction, in case it needs to 7804 * re-inject later. 7805 */ 7806 pd->dir = PF_OUT; 7807 7808 /* 7809 * The following processing is actually the rest of the inbound processing, even 7810 * though we've marked it as outbound (so we don't look through dummynet) and it 7811 * happens after the outbound processing (pf_test(PF_OUT) above). 7812 * Swap the dummynet pipe numbers, because it's going to come to the wrong 7813 * conclusion about what direction it's processing, and we can't fix it or it 7814 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 7815 * decision will pick the right pipe, and everything will mostly work as expected. 7816 */ 7817 tmp = pd->act.dnrpipe; 7818 pd->act.dnrpipe = pd->act.dnpipe; 7819 pd->act.dnpipe = tmp; 7820 } 7821 7822 /* 7823 * If small enough for interface, or the interface will take 7824 * care of the fragmentation for us, we can just send directly. 7825 */ 7826 if (ip_len <= ifp->if_mtu || 7827 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7828 ip->ip_sum = 0; 7829 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7830 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7831 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7832 } 7833 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7834 7835 md = m0; 7836 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7837 if (md != NULL) 7838 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7839 goto done; 7840 } 7841 7842 /* Balk when DF bit is set or the interface didn't support TSO. */ 7843 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7844 error = EMSGSIZE; 7845 KMOD_IPSTAT_INC(ips_cantfrag); 7846 if (r_rt != PF_DUPTO) { 7847 if (s && s->nat_rule != NULL) 7848 PACKET_UNDO_NAT(m0, pd, 7849 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7850 s); 7851 7852 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7853 ifp->if_mtu); 7854 goto done; 7855 } else 7856 goto bad; 7857 } 7858 7859 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7860 if (error) 7861 goto bad; 7862 7863 for (; m0; m0 = m1) { 7864 m1 = m0->m_nextpkt; 7865 m0->m_nextpkt = NULL; 7866 if (error == 0) { 7867 m_clrprotoflags(m0); 7868 md = m0; 7869 pd->pf_mtag = pf_find_mtag(md); 7870 error = pf_dummynet_route(pd, s, r, ifp, 7871 sintosa(&dst), &md); 7872 if (md != NULL) 7873 error = (*ifp->if_output)(ifp, md, 7874 sintosa(&dst), NULL); 7875 } else 7876 m_freem(m0); 7877 } 7878 7879 if (error == 0) 7880 KMOD_IPSTAT_INC(ips_fragmented); 7881 7882 done: 7883 if (r_rt != PF_DUPTO) 7884 *m = NULL; 7885 return; 7886 7887 bad_locked: 7888 if (s) 7889 PF_STATE_UNLOCK(s); 7890 bad: 7891 m_freem(m0); 7892 goto done; 7893 } 7894 #endif /* INET */ 7895 7896 #ifdef INET6 7897 static void 7898 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7899 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7900 { 7901 struct mbuf *m0, *md; 7902 struct sockaddr_in6 dst; 7903 struct ip6_hdr *ip6; 7904 struct pfi_kkif *nkif = NULL; 7905 struct ifnet *ifp = NULL; 7906 struct pf_addr naddr; 7907 int r_rt, r_dir; 7908 7909 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7910 7911 if (s) { 7912 r_rt = s->rt; 7913 r_dir = s->direction; 7914 } else { 7915 r_rt = r->rt; 7916 r_dir = r->direction; 7917 } 7918 7919 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7920 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7921 __func__)); 7922 7923 if ((pd->pf_mtag == NULL && 7924 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7925 pd->pf_mtag->routed++ > 3) { 7926 m0 = *m; 7927 *m = NULL; 7928 goto bad_locked; 7929 } 7930 7931 if (r_rt == PF_DUPTO) { 7932 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7933 if (s == NULL) { 7934 ifp = r->rpool.cur->kif ? 7935 r->rpool.cur->kif->pfik_ifp : NULL; 7936 } else { 7937 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7938 /* If pfsync'd */ 7939 if (ifp == NULL && r->rpool.cur != NULL) 7940 ifp = r->rpool.cur->kif ? 7941 r->rpool.cur->kif->pfik_ifp : NULL; 7942 PF_STATE_UNLOCK(s); 7943 } 7944 if (ifp == oifp) { 7945 /* When the 2nd interface is not skipped */ 7946 return; 7947 } else { 7948 m0 = *m; 7949 *m = NULL; 7950 goto bad; 7951 } 7952 } else { 7953 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7954 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7955 if (s) 7956 PF_STATE_UNLOCK(s); 7957 return; 7958 } 7959 } 7960 } else { 7961 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7962 pf_dummynet(pd, s, r, m); 7963 if (s) 7964 PF_STATE_UNLOCK(s); 7965 return; 7966 } 7967 m0 = *m; 7968 } 7969 7970 ip6 = mtod(m0, struct ip6_hdr *); 7971 7972 bzero(&dst, sizeof(dst)); 7973 dst.sin6_family = AF_INET6; 7974 dst.sin6_len = sizeof(dst); 7975 dst.sin6_addr = ip6->ip6_dst; 7976 7977 bzero(&naddr, sizeof(naddr)); 7978 7979 if (s == NULL) { 7980 if (TAILQ_EMPTY(&r->rpool.list)) { 7981 DPFPRINTF(PF_DEBUG_URGENT, 7982 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7983 goto bad_locked; 7984 } 7985 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 7986 &naddr, &nkif, NULL); 7987 if (!PF_AZERO(&naddr, AF_INET6)) 7988 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7989 &naddr, AF_INET6); 7990 ifp = nkif ? nkif->pfik_ifp : NULL; 7991 } else { 7992 struct pfi_kkif *kif; 7993 7994 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 7995 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7996 &s->rt_addr, AF_INET6); 7997 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7998 kif = s->rt_kif; 7999 /* If pfsync'd */ 8000 if (ifp == NULL && r->rpool.cur != NULL) { 8001 ifp = r->rpool.cur->kif ? 8002 r->rpool.cur->kif->pfik_ifp : NULL; 8003 kif = r->rpool.cur->kif; 8004 } 8005 if (ifp != NULL && kif != NULL && 8006 r->rule_flag & PFRULE_IFBOUND && 8007 r->rt == PF_REPLYTO && 8008 s->kif == V_pfi_all) { 8009 s->kif = kif; 8010 s->orig_kif = oifp->if_pf_kif; 8011 } 8012 } 8013 8014 if (s) 8015 PF_STATE_UNLOCK(s); 8016 8017 if (ifp == NULL) 8018 goto bad; 8019 8020 if (pd->dir == PF_IN) { 8021 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD, ifp, &m0, inp, 8022 &pd->act) != PF_PASS) 8023 goto bad; 8024 else if (m0 == NULL) 8025 goto done; 8026 if (m0->m_len < sizeof(struct ip6_hdr)) { 8027 DPFPRINTF(PF_DEBUG_URGENT, 8028 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 8029 __func__)); 8030 goto bad; 8031 } 8032 ip6 = mtod(m0, struct ip6_hdr *); 8033 } 8034 8035 if (ifp->if_flags & IFF_LOOPBACK) 8036 m0->m_flags |= M_SKIP_FIREWALL; 8037 8038 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 8039 ~ifp->if_hwassist) { 8040 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 8041 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 8042 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 8043 } 8044 8045 /* 8046 * If the packet is too large for the outgoing interface, 8047 * send back an icmp6 error. 8048 */ 8049 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 8050 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 8051 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 8052 md = m0; 8053 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 8054 if (md != NULL) 8055 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 8056 } 8057 else { 8058 in6_ifstat_inc(ifp, ifs6_in_toobig); 8059 if (r_rt != PF_DUPTO) { 8060 if (s && s->nat_rule != NULL) 8061 PACKET_UNDO_NAT(m0, pd, 8062 ((caddr_t)ip6 - m0->m_data) + 8063 sizeof(struct ip6_hdr), s); 8064 8065 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 8066 } else 8067 goto bad; 8068 } 8069 8070 done: 8071 if (r_rt != PF_DUPTO) 8072 *m = NULL; 8073 return; 8074 8075 bad_locked: 8076 if (s) 8077 PF_STATE_UNLOCK(s); 8078 bad: 8079 m_freem(m0); 8080 goto done; 8081 } 8082 #endif /* INET6 */ 8083 8084 /* 8085 * FreeBSD supports cksum offloads for the following drivers. 8086 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 8087 * 8088 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 8089 * network driver performed cksum including pseudo header, need to verify 8090 * csum_data 8091 * CSUM_DATA_VALID : 8092 * network driver performed cksum, needs to additional pseudo header 8093 * cksum computation with partial csum_data(i.e. lack of H/W support for 8094 * pseudo header, for instance sk(4) and possibly gem(4)) 8095 * 8096 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 8097 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 8098 * TCP/UDP layer. 8099 * Also, set csum_data to 0xffff to force cksum validation. 8100 */ 8101 static int 8102 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 8103 { 8104 u_int16_t sum = 0; 8105 int hw_assist = 0; 8106 struct ip *ip; 8107 8108 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 8109 return (1); 8110 if (m->m_pkthdr.len < off + len) 8111 return (1); 8112 8113 switch (p) { 8114 case IPPROTO_TCP: 8115 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8116 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8117 sum = m->m_pkthdr.csum_data; 8118 } else { 8119 ip = mtod(m, struct ip *); 8120 sum = in_pseudo(ip->ip_src.s_addr, 8121 ip->ip_dst.s_addr, htonl((u_short)len + 8122 m->m_pkthdr.csum_data + IPPROTO_TCP)); 8123 } 8124 sum ^= 0xffff; 8125 ++hw_assist; 8126 } 8127 break; 8128 case IPPROTO_UDP: 8129 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8130 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8131 sum = m->m_pkthdr.csum_data; 8132 } else { 8133 ip = mtod(m, struct ip *); 8134 sum = in_pseudo(ip->ip_src.s_addr, 8135 ip->ip_dst.s_addr, htonl((u_short)len + 8136 m->m_pkthdr.csum_data + IPPROTO_UDP)); 8137 } 8138 sum ^= 0xffff; 8139 ++hw_assist; 8140 } 8141 break; 8142 case IPPROTO_ICMP: 8143 #ifdef INET6 8144 case IPPROTO_ICMPV6: 8145 #endif /* INET6 */ 8146 break; 8147 default: 8148 return (1); 8149 } 8150 8151 if (!hw_assist) { 8152 switch (af) { 8153 case AF_INET: 8154 if (p == IPPROTO_ICMP) { 8155 if (m->m_len < off) 8156 return (1); 8157 m->m_data += off; 8158 m->m_len -= off; 8159 sum = in_cksum(m, len); 8160 m->m_data -= off; 8161 m->m_len += off; 8162 } else { 8163 if (m->m_len < sizeof(struct ip)) 8164 return (1); 8165 sum = in4_cksum(m, p, off, len); 8166 } 8167 break; 8168 #ifdef INET6 8169 case AF_INET6: 8170 if (m->m_len < sizeof(struct ip6_hdr)) 8171 return (1); 8172 sum = in6_cksum(m, p, off, len); 8173 break; 8174 #endif /* INET6 */ 8175 } 8176 } 8177 if (sum) { 8178 switch (p) { 8179 case IPPROTO_TCP: 8180 { 8181 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 8182 break; 8183 } 8184 case IPPROTO_UDP: 8185 { 8186 KMOD_UDPSTAT_INC(udps_badsum); 8187 break; 8188 } 8189 #ifdef INET 8190 case IPPROTO_ICMP: 8191 { 8192 KMOD_ICMPSTAT_INC(icps_checksum); 8193 break; 8194 } 8195 #endif 8196 #ifdef INET6 8197 case IPPROTO_ICMPV6: 8198 { 8199 KMOD_ICMP6STAT_INC(icp6s_checksum); 8200 break; 8201 } 8202 #endif /* INET6 */ 8203 } 8204 return (1); 8205 } else { 8206 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 8207 m->m_pkthdr.csum_flags |= 8208 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 8209 m->m_pkthdr.csum_data = 0xffff; 8210 } 8211 } 8212 return (0); 8213 } 8214 8215 static bool 8216 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 8217 const struct pf_kstate *s, struct ip_fw_args *dnflow) 8218 { 8219 int dndir = r->direction; 8220 8221 if (s && dndir == PF_INOUT) { 8222 dndir = s->direction; 8223 } else if (dndir == PF_INOUT) { 8224 /* Assume primary direction. Happens when we've set dnpipe in 8225 * the ethernet level code. */ 8226 dndir = pd->dir; 8227 } 8228 8229 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 8230 return (false); 8231 8232 memset(dnflow, 0, sizeof(*dnflow)); 8233 8234 if (pd->dport != NULL) 8235 dnflow->f_id.dst_port = ntohs(*pd->dport); 8236 if (pd->sport != NULL) 8237 dnflow->f_id.src_port = ntohs(*pd->sport); 8238 8239 if (pd->dir == PF_IN) 8240 dnflow->flags |= IPFW_ARGS_IN; 8241 else 8242 dnflow->flags |= IPFW_ARGS_OUT; 8243 8244 if (pd->dir != dndir && pd->act.dnrpipe) { 8245 dnflow->rule.info = pd->act.dnrpipe; 8246 } 8247 else if (pd->dir == dndir && pd->act.dnpipe) { 8248 dnflow->rule.info = pd->act.dnpipe; 8249 } 8250 else { 8251 return (false); 8252 } 8253 8254 dnflow->rule.info |= IPFW_IS_DUMMYNET; 8255 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 8256 dnflow->rule.info |= IPFW_IS_PIPE; 8257 8258 dnflow->f_id.proto = pd->proto; 8259 dnflow->f_id.extra = dnflow->rule.info; 8260 switch (pd->af) { 8261 case AF_INET: 8262 dnflow->f_id.addr_type = 4; 8263 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 8264 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 8265 break; 8266 case AF_INET6: 8267 dnflow->flags |= IPFW_ARGS_IP6; 8268 dnflow->f_id.addr_type = 6; 8269 dnflow->f_id.src_ip6 = pd->src->v6; 8270 dnflow->f_id.dst_ip6 = pd->dst->v6; 8271 break; 8272 } 8273 8274 return (true); 8275 } 8276 8277 int 8278 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8279 struct inpcb *inp) 8280 { 8281 struct pfi_kkif *kif; 8282 struct mbuf *m = *m0; 8283 8284 M_ASSERTPKTHDR(m); 8285 MPASS(ifp->if_vnet == curvnet); 8286 NET_EPOCH_ASSERT(); 8287 8288 if (!V_pf_status.running) 8289 return (PF_PASS); 8290 8291 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8292 8293 if (kif == NULL) { 8294 DPFPRINTF(PF_DEBUG_URGENT, 8295 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 8296 return (PF_DROP); 8297 } 8298 if (kif->pfik_flags & PFI_IFLAG_SKIP) 8299 return (PF_PASS); 8300 8301 if (m->m_flags & M_SKIP_FIREWALL) 8302 return (PF_PASS); 8303 8304 if (__predict_false(! M_WRITABLE(*m0))) { 8305 m = *m0 = m_unshare(*m0, M_NOWAIT); 8306 if (*m0 == NULL) 8307 return (PF_DROP); 8308 } 8309 8310 /* Stateless! */ 8311 return (pf_test_eth_rule(dir, kif, m0)); 8312 } 8313 8314 static __inline void 8315 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 8316 { 8317 struct m_tag *mtag; 8318 8319 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 8320 8321 /* dummynet adds this tag, but pf does not need it, 8322 * and keeping it creates unexpected behavior, 8323 * e.g. in case of divert(4) usage right after dummynet. */ 8324 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8325 if (mtag != NULL) 8326 m_tag_delete(m, mtag); 8327 } 8328 8329 static int 8330 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 8331 struct pf_krule *r, struct mbuf **m0) 8332 { 8333 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 8334 } 8335 8336 static int 8337 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 8338 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 8339 struct mbuf **m0) 8340 { 8341 NET_EPOCH_ASSERT(); 8342 8343 if (pd->act.dnpipe || pd->act.dnrpipe) { 8344 struct ip_fw_args dnflow; 8345 if (ip_dn_io_ptr == NULL) { 8346 m_freem(*m0); 8347 *m0 = NULL; 8348 return (ENOMEM); 8349 } 8350 8351 if (pd->pf_mtag == NULL && 8352 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 8353 m_freem(*m0); 8354 *m0 = NULL; 8355 return (ENOMEM); 8356 } 8357 8358 if (ifp != NULL) { 8359 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 8360 8361 pd->pf_mtag->if_index = ifp->if_index; 8362 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 8363 8364 MPASS(sa != NULL); 8365 8366 switch (pd->af) { 8367 case AF_INET: 8368 memcpy(&pd->pf_mtag->dst, sa, 8369 sizeof(struct sockaddr_in)); 8370 break; 8371 case AF_INET6: 8372 memcpy(&pd->pf_mtag->dst, sa, 8373 sizeof(struct sockaddr_in6)); 8374 break; 8375 } 8376 } 8377 8378 if (s != NULL && s->nat_rule != NULL && 8379 s->nat_rule->action == PF_RDR && 8380 ( 8381 #ifdef INET 8382 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 8383 #endif 8384 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 8385 /* 8386 * If we're redirecting to loopback mark this packet 8387 * as being local. Otherwise it might get dropped 8388 * if dummynet re-injects. 8389 */ 8390 (*m0)->m_pkthdr.rcvif = V_loif; 8391 } 8392 8393 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 8394 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 8395 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 8396 ip_dn_io_ptr(m0, &dnflow); 8397 if (*m0 != NULL) { 8398 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8399 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 8400 } 8401 } 8402 } 8403 8404 return (0); 8405 } 8406 8407 #ifdef INET6 8408 static int 8409 pf_walk_option6(struct mbuf *m, int off, int end, uint32_t *jumbolen, 8410 u_short *reason) 8411 { 8412 struct ip6_opt opt; 8413 struct ip6_opt_jumbo jumbo; 8414 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 8415 8416 while (off < end) { 8417 if (!pf_pull_hdr(m, off, &opt.ip6o_type, sizeof(opt.ip6o_type), 8418 NULL, reason, AF_INET6)) { 8419 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt type")); 8420 return (PF_DROP); 8421 } 8422 if (opt.ip6o_type == IP6OPT_PAD1) { 8423 off++; 8424 continue; 8425 } 8426 if (!pf_pull_hdr(m, off, &opt, sizeof(opt), NULL, reason, 8427 AF_INET6)) { 8428 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt")); 8429 return (PF_DROP); 8430 } 8431 if (off + sizeof(opt) + opt.ip6o_len > end) { 8432 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 long opt")); 8433 REASON_SET(reason, PFRES_IPOPTIONS); 8434 return (PF_DROP); 8435 } 8436 switch (opt.ip6o_type) { 8437 case IP6OPT_JUMBO: 8438 if (*jumbolen != 0) { 8439 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple jumbo")); 8440 REASON_SET(reason, PFRES_IPOPTIONS); 8441 return (PF_DROP); 8442 } 8443 if (ntohs(h->ip6_plen) != 0) { 8444 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 bad jumbo plen")); 8445 REASON_SET(reason, PFRES_IPOPTIONS); 8446 return (PF_DROP); 8447 } 8448 if (!pf_pull_hdr(m, off, &jumbo, sizeof(jumbo), NULL, 8449 reason, AF_INET6)) { 8450 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbo")); 8451 return (PF_DROP); 8452 } 8453 memcpy(jumbolen, jumbo.ip6oj_jumbo_len, 8454 sizeof(*jumbolen)); 8455 *jumbolen = ntohl(*jumbolen); 8456 if (*jumbolen < IPV6_MAXPACKET) { 8457 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbolen")); 8458 REASON_SET(reason, PFRES_IPOPTIONS); 8459 return (PF_DROP); 8460 } 8461 break; 8462 default: 8463 break; 8464 } 8465 off += sizeof(opt) + opt.ip6o_len; 8466 } 8467 8468 return (PF_PASS); 8469 } 8470 8471 int 8472 pf_walk_header6(struct mbuf *m, struct ip6_hdr *h, int *off, int *extoff, 8473 int *fragoff, uint8_t *nxt, uint32_t *jumbolen, u_short *reason) 8474 { 8475 struct ip6_frag frag; 8476 struct ip6_ext ext; 8477 struct ip6_rthdr rthdr; 8478 int rthdr_cnt = 0; 8479 8480 *off += sizeof(struct ip6_hdr); 8481 *extoff = *fragoff = 0; 8482 *nxt = h->ip6_nxt; 8483 *jumbolen = 0; 8484 for (;;) { 8485 switch (*nxt) { 8486 case IPPROTO_FRAGMENT: 8487 if (*fragoff != 0) { 8488 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple fragment")); 8489 REASON_SET(reason, PFRES_FRAG); 8490 return (PF_DROP); 8491 } 8492 /* jumbo payload packets cannot be fragmented */ 8493 if (*jumbolen != 0) { 8494 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 fragmented jumbo")); 8495 REASON_SET(reason, PFRES_FRAG); 8496 return (PF_DROP); 8497 } 8498 if (!pf_pull_hdr(m, *off, &frag, sizeof(frag), NULL, 8499 reason, AF_INET6)) { 8500 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short fragment")); 8501 return (PF_DROP); 8502 } 8503 *fragoff = *off; 8504 /* stop walking over non initial fragments */ 8505 if ((frag.ip6f_offlg & IP6F_OFF_MASK) != 0) 8506 return (PF_PASS); 8507 *off += sizeof(frag); 8508 *nxt = frag.ip6f_nxt; 8509 break; 8510 case IPPROTO_ROUTING: 8511 if (rthdr_cnt++) { 8512 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple rthdr")); 8513 REASON_SET(reason, PFRES_IPOPTIONS); 8514 return (PF_DROP); 8515 } 8516 if (!pf_pull_hdr(m, *off, &rthdr, sizeof(rthdr), NULL, 8517 reason, AF_INET6)) { 8518 /* fragments may be short */ 8519 if (*fragoff != 0) { 8520 *off = *fragoff; 8521 *nxt = IPPROTO_FRAGMENT; 8522 return (PF_PASS); 8523 } 8524 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short rthdr")); 8525 return (PF_DROP); 8526 } 8527 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8528 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 rthdr0")); 8529 REASON_SET(reason, PFRES_IPOPTIONS); 8530 return (PF_DROP); 8531 } 8532 /* FALLTHROUGH */ 8533 case IPPROTO_AH: 8534 case IPPROTO_HOPOPTS: 8535 case IPPROTO_DSTOPTS: 8536 if (!pf_pull_hdr(m, *off, &ext, sizeof(ext), NULL, 8537 reason, AF_INET6)) { 8538 /* fragments may be short */ 8539 if (*fragoff != 0) { 8540 *off = *fragoff; 8541 *nxt = IPPROTO_FRAGMENT; 8542 return (PF_PASS); 8543 } 8544 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short exthdr")); 8545 return (PF_DROP); 8546 } 8547 /* reassembly needs the ext header before the frag */ 8548 if (*fragoff == 0) 8549 *extoff = *off; 8550 if (*nxt == IPPROTO_HOPOPTS && *fragoff == 0) { 8551 if (pf_walk_option6(m, *off + sizeof(ext), 8552 *off + (ext.ip6e_len + 1) * 8, jumbolen, 8553 reason) != PF_PASS) 8554 return (PF_DROP); 8555 if (ntohs(h->ip6_plen) == 0 && *jumbolen != 0) { 8556 DPFPRINTF(PF_DEBUG_MISC, 8557 ("IPv6 missing jumbo")); 8558 REASON_SET(reason, PFRES_IPOPTIONS); 8559 return (PF_DROP); 8560 } 8561 } 8562 if (*nxt == IPPROTO_AH) 8563 *off += (ext.ip6e_len + 2) * 4; 8564 else 8565 *off += (ext.ip6e_len + 1) * 8; 8566 *nxt = ext.ip6e_nxt; 8567 break; 8568 case IPPROTO_TCP: 8569 case IPPROTO_UDP: 8570 case IPPROTO_SCTP: 8571 case IPPROTO_ICMPV6: 8572 /* fragments may be short, ignore inner header then */ 8573 if (*fragoff != 0 && ntohs(h->ip6_plen) < *off + 8574 (*nxt == IPPROTO_TCP ? sizeof(struct tcphdr) : 8575 *nxt == IPPROTO_UDP ? sizeof(struct udphdr) : 8576 *nxt == IPPROTO_SCTP ? sizeof(struct sctphdr) : 8577 sizeof(struct icmp6_hdr))) { 8578 *off = *fragoff; 8579 *nxt = IPPROTO_FRAGMENT; 8580 } 8581 /* FALLTHROUGH */ 8582 default: 8583 return (PF_PASS); 8584 } 8585 } 8586 } 8587 #endif 8588 8589 static void 8590 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 8591 { 8592 memset(pd, 0, sizeof(*pd)); 8593 pd->pf_mtag = pf_find_mtag(m); 8594 pd->m = m; 8595 } 8596 8597 static int 8598 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 8599 u_short *action, u_short *reason, struct pfi_kkif *kif, 8600 struct pf_rule_actions *default_actions) 8601 { 8602 pd->af = af; 8603 pd->dir = dir; 8604 pd->kif = kif; 8605 pd->m = *m0; 8606 pd->sidx = (dir == PF_IN) ? 0 : 1; 8607 pd->didx = (dir == PF_IN) ? 1 : 0; 8608 8609 TAILQ_INIT(&pd->sctp_multihome_jobs); 8610 if (default_actions != NULL) 8611 memcpy(&pd->act, default_actions, sizeof(pd->act)); 8612 8613 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 8614 pd->act.dnpipe = pd->pf_mtag->dnpipe; 8615 pd->act.flags = pd->pf_mtag->dnflags; 8616 } 8617 8618 switch (af) { 8619 #ifdef INET 8620 case AF_INET: { 8621 struct ip *h; 8622 8623 if (__predict_false((*m0)->m_len < sizeof(struct ip)) && 8624 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 8625 DPFPRINTF(PF_DEBUG_URGENT, 8626 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 8627 *action = PF_DROP; 8628 REASON_SET(reason, PFRES_SHORT); 8629 return (-1); 8630 } 8631 8632 if (pf_normalize_ip(m0, reason, pd) != PF_PASS) { 8633 /* We do IP header normalization and packet reassembly here */ 8634 *action = PF_DROP; 8635 return (-1); 8636 } 8637 pd->m = *m0; 8638 8639 h = mtod(pd->m, struct ip *); 8640 pd->off = h->ip_hl << 2; 8641 if (pd->off < (int)sizeof(*h)) { 8642 *action = PF_DROP; 8643 REASON_SET(reason, PFRES_SHORT); 8644 return (-1); 8645 } 8646 pd->src = (struct pf_addr *)&h->ip_src; 8647 pd->dst = (struct pf_addr *)&h->ip_dst; 8648 pd->ip_sum = &h->ip_sum; 8649 pd->proto_sum = NULL; 8650 pd->virtual_proto = pd->proto = h->ip_p; 8651 pd->tos = h->ip_tos; 8652 pd->ttl = h->ip_ttl; 8653 pd->tot_len = ntohs(h->ip_len); 8654 pd->act.rtableid = -1; 8655 8656 if (h->ip_hl > 5) /* has options */ 8657 pd->badopts++; 8658 8659 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) 8660 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8661 8662 break; 8663 } 8664 #endif 8665 #ifdef INET6 8666 case AF_INET6: { 8667 struct ip6_hdr *h; 8668 int fragoff; 8669 uint32_t jumbolen; 8670 uint8_t nxt; 8671 8672 if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) && 8673 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 8674 DPFPRINTF(PF_DEBUG_URGENT, 8675 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 8676 ", pullup failed\n")); 8677 *action = PF_DROP; 8678 REASON_SET(reason, PFRES_SHORT); 8679 return (-1); 8680 } 8681 8682 h = mtod(pd->m, struct ip6_hdr *); 8683 pd->off = 0; 8684 if (pf_walk_header6(pd->m, h, &pd->off, &pd->extoff, &fragoff, &nxt, 8685 &jumbolen, reason) != PF_PASS) { 8686 *action = PF_DROP; 8687 return (-1); 8688 } 8689 8690 h = mtod(pd->m, struct ip6_hdr *); 8691 pd->src = (struct pf_addr *)&h->ip6_src; 8692 pd->dst = (struct pf_addr *)&h->ip6_dst; 8693 pd->ip_sum = NULL; 8694 pd->proto_sum = NULL; 8695 pd->tos = IPV6_DSCP(h); 8696 pd->ttl = h->ip6_hlim; 8697 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8698 pd->virtual_proto = pd->proto = h->ip6_nxt; 8699 pd->act.rtableid = -1; 8700 8701 if (fragoff != 0) 8702 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8703 8704 /* 8705 * we do not support jumbogram. if we keep going, zero ip6_plen 8706 * will do something bad, so drop the packet for now. 8707 */ 8708 if (htons(h->ip6_plen) == 0) { 8709 *action = PF_DROP; 8710 return (-1); 8711 } 8712 8713 /* We do IP header normalization and packet reassembly here */ 8714 if (pf_normalize_ip6(m0, fragoff, reason, pd) != 8715 PF_PASS) { 8716 *action = PF_DROP; 8717 return (-1); 8718 } 8719 pd->m = *m0; 8720 if (pd->m == NULL) { 8721 /* packet sits in reassembly queue, no error */ 8722 *action = PF_PASS; 8723 return (-1); 8724 } 8725 8726 /* Update pointers into the packet. */ 8727 h = mtod(pd->m, struct ip6_hdr *); 8728 pd->src = (struct pf_addr *)&h->ip6_src; 8729 pd->dst = (struct pf_addr *)&h->ip6_dst; 8730 8731 /* 8732 * Reassembly may have changed the next protocol from fragment 8733 * to something else, so update. 8734 */ 8735 pd->virtual_proto = pd->proto = h->ip6_nxt; 8736 pd->off = 0; 8737 8738 if (pf_walk_header6(pd->m, h, &pd->off, &pd->extoff, &fragoff, &nxt, 8739 &jumbolen, reason) != PF_PASS) { 8740 *action = PF_DROP; 8741 return (-1); 8742 } 8743 8744 if (fragoff != 0) 8745 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8746 8747 break; 8748 } 8749 #endif 8750 default: 8751 panic("pf_setup_pdesc called with illegal af %u", af); 8752 } 8753 8754 switch (pd->virtual_proto) { 8755 case IPPROTO_TCP: { 8756 struct tcphdr *th = &pd->hdr.tcp; 8757 8758 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), action, 8759 reason, af)) { 8760 *action = PF_DROP; 8761 REASON_SET(reason, PFRES_SHORT); 8762 return (-1); 8763 } 8764 pd->hdrlen = sizeof(*th); 8765 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 8766 pd->sport = &th->th_sport; 8767 pd->dport = &th->th_dport; 8768 break; 8769 } 8770 case IPPROTO_UDP: { 8771 struct udphdr *uh = &pd->hdr.udp; 8772 8773 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), action, 8774 reason, af)) { 8775 *action = PF_DROP; 8776 REASON_SET(reason, PFRES_SHORT); 8777 return (-1); 8778 } 8779 pd->hdrlen = sizeof(*uh); 8780 if (uh->uh_dport == 0 || 8781 ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off || 8782 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 8783 *action = PF_DROP; 8784 REASON_SET(reason, PFRES_SHORT); 8785 return (-1); 8786 } 8787 pd->sport = &uh->uh_sport; 8788 pd->dport = &uh->uh_dport; 8789 break; 8790 } 8791 case IPPROTO_SCTP: { 8792 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 8793 action, reason, af)) { 8794 *action = PF_DROP; 8795 REASON_SET(reason, PFRES_SHORT); 8796 return (-1); 8797 } 8798 pd->hdrlen = sizeof(pd->hdr.sctp); 8799 pd->p_len = pd->tot_len - pd->off; 8800 8801 pd->sport = &pd->hdr.sctp.src_port; 8802 pd->dport = &pd->hdr.sctp.dest_port; 8803 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 8804 *action = PF_DROP; 8805 REASON_SET(reason, PFRES_SHORT); 8806 return (-1); 8807 } 8808 if (pf_scan_sctp(pd) != PF_PASS) { 8809 *action = PF_DROP; 8810 REASON_SET(reason, PFRES_SHORT); 8811 return (-1); 8812 } 8813 break; 8814 } 8815 case IPPROTO_ICMP: { 8816 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 8817 action, reason, af)) { 8818 *action = PF_DROP; 8819 REASON_SET(reason, PFRES_SHORT); 8820 return (-1); 8821 } 8822 pd->hdrlen = ICMP_MINLEN; 8823 break; 8824 } 8825 #ifdef INET6 8826 case IPPROTO_ICMPV6: { 8827 size_t icmp_hlen = sizeof(struct icmp6_hdr); 8828 8829 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 8830 action, reason, af)) { 8831 *action = PF_DROP; 8832 REASON_SET(reason, PFRES_SHORT); 8833 return (-1); 8834 } 8835 /* ICMP headers we look further into to match state */ 8836 switch (pd->hdr.icmp6.icmp6_type) { 8837 case MLD_LISTENER_QUERY: 8838 case MLD_LISTENER_REPORT: 8839 icmp_hlen = sizeof(struct mld_hdr); 8840 break; 8841 case ND_NEIGHBOR_SOLICIT: 8842 case ND_NEIGHBOR_ADVERT: 8843 icmp_hlen = sizeof(struct nd_neighbor_solicit); 8844 break; 8845 } 8846 if (icmp_hlen > sizeof(struct icmp6_hdr) && 8847 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 8848 action, reason, af)) { 8849 *action = PF_DROP; 8850 REASON_SET(reason, PFRES_SHORT); 8851 return (-1); 8852 } 8853 pd->hdrlen = icmp_hlen; 8854 break; 8855 } 8856 #endif 8857 } 8858 return (0); 8859 } 8860 8861 static void 8862 pf_counters_inc(int action, struct pf_pdesc *pd, 8863 struct pf_kstate *s, struct pf_krule *r, struct pf_krule *a) 8864 { 8865 struct pf_krule *tr; 8866 int dir = pd->dir; 8867 int dirndx; 8868 8869 pf_counter_u64_critical_enter(); 8870 pf_counter_u64_add_protected( 8871 &pd->kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8872 pd->tot_len); 8873 pf_counter_u64_add_protected( 8874 &pd->kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8875 1); 8876 8877 if (action == PF_PASS || r->action == PF_DROP) { 8878 dirndx = (dir == PF_OUT); 8879 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8880 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 8881 pf_update_timestamp(r); 8882 8883 if (a != NULL) { 8884 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8885 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 8886 } 8887 if (s != NULL) { 8888 struct pf_krule_item *ri; 8889 8890 if (s->nat_rule != NULL) { 8891 pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx], 8892 1); 8893 pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx], 8894 pd->tot_len); 8895 } 8896 if (s->src_node != NULL) { 8897 counter_u64_add(s->src_node->packets[dirndx], 8898 1); 8899 counter_u64_add(s->src_node->bytes[dirndx], 8900 pd->tot_len); 8901 } 8902 if (s->nat_src_node != NULL) { 8903 counter_u64_add(s->nat_src_node->packets[dirndx], 8904 1); 8905 counter_u64_add(s->nat_src_node->bytes[dirndx], 8906 pd->tot_len); 8907 } 8908 dirndx = (dir == s->direction) ? 0 : 1; 8909 s->packets[dirndx]++; 8910 s->bytes[dirndx] += pd->tot_len; 8911 8912 SLIST_FOREACH(ri, &s->match_rules, entry) { 8913 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8914 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 8915 } 8916 } 8917 8918 tr = r; 8919 if (s != NULL && s->nat_rule != NULL && 8920 r == &V_pf_default_rule) 8921 tr = s->nat_rule; 8922 8923 if (tr->src.addr.type == PF_ADDR_TABLE) 8924 pfr_update_stats(tr->src.addr.p.tbl, 8925 (s == NULL) ? pd->src : 8926 &s->key[(s->direction == PF_IN)]-> 8927 addr[(s->direction == PF_OUT)], 8928 pd->af, pd->tot_len, dir == PF_OUT, 8929 r->action == PF_PASS, tr->src.neg); 8930 if (tr->dst.addr.type == PF_ADDR_TABLE) 8931 pfr_update_stats(tr->dst.addr.p.tbl, 8932 (s == NULL) ? pd->dst : 8933 &s->key[(s->direction == PF_IN)]-> 8934 addr[(s->direction == PF_IN)], 8935 pd->af, pd->tot_len, dir == PF_OUT, 8936 r->action == PF_PASS, tr->dst.neg); 8937 } 8938 pf_counter_u64_critical_exit(); 8939 } 8940 8941 #if defined(INET) || defined(INET6) 8942 int 8943 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8944 struct inpcb *inp, struct pf_rule_actions *default_actions) 8945 { 8946 struct pfi_kkif *kif; 8947 u_short action, reason = 0; 8948 struct m_tag *mtag; 8949 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 8950 struct pf_kstate *s = NULL; 8951 struct pf_kruleset *ruleset = NULL; 8952 struct pf_pdesc pd; 8953 int use_2nd_queue = 0; 8954 uint16_t tag; 8955 uint8_t rt; 8956 8957 PF_RULES_RLOCK_TRACKER; 8958 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8959 M_ASSERTPKTHDR(*m0); 8960 8961 if (!V_pf_status.running) 8962 return (PF_PASS); 8963 8964 PF_RULES_RLOCK(); 8965 8966 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8967 8968 if (__predict_false(kif == NULL)) { 8969 DPFPRINTF(PF_DEBUG_URGENT, 8970 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 8971 PF_RULES_RUNLOCK(); 8972 return (PF_DROP); 8973 } 8974 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8975 PF_RULES_RUNLOCK(); 8976 return (PF_PASS); 8977 } 8978 8979 if ((*m0)->m_flags & M_SKIP_FIREWALL) { 8980 PF_RULES_RUNLOCK(); 8981 return (PF_PASS); 8982 } 8983 8984 #ifdef INET6 8985 /* 8986 * If we end up changing IP addresses (e.g. binat) the stack may get 8987 * confused and fail to send the icmp6 packet too big error. Just send 8988 * it here, before we do any NAT. 8989 */ 8990 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 8991 IN6_LINKMTU(ifp) < pf_max_frag_size(*m0)) { 8992 PF_RULES_RUNLOCK(); 8993 icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 8994 *m0 = NULL; 8995 return (PF_DROP); 8996 } 8997 #endif 8998 8999 if (__predict_false(! M_WRITABLE(*m0))) { 9000 *m0 = m_unshare(*m0, M_NOWAIT); 9001 if (*m0 == NULL) 9002 return (PF_DROP); 9003 } 9004 9005 pf_init_pdesc(&pd, *m0); 9006 9007 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 9008 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9009 9010 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 9011 pd.pf_mtag->if_idxgen); 9012 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 9013 PF_RULES_RUNLOCK(); 9014 m_freem(*m0); 9015 *m0 = NULL; 9016 return (PF_PASS); 9017 } 9018 PF_RULES_RUNLOCK(); 9019 (ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL); 9020 *m0 = NULL; 9021 return (PF_PASS); 9022 } 9023 9024 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 9025 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 9026 /* Dummynet re-injects packets after they've 9027 * completed their delay. We've already 9028 * processed them, so pass unconditionally. */ 9029 9030 /* But only once. We may see the packet multiple times (e.g. 9031 * PFIL_IN/PFIL_OUT). */ 9032 pf_dummynet_flag_remove(pd.m, pd.pf_mtag); 9033 PF_RULES_RUNLOCK(); 9034 9035 return (PF_PASS); 9036 } 9037 9038 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, 9039 kif, default_actions) == -1) { 9040 if (action != PF_PASS) 9041 pd.act.log |= PF_LOG_FORCE; 9042 goto done; 9043 } 9044 9045 if (__predict_false(ip_divert_ptr != NULL) && 9046 ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 9047 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 9048 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 9049 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 9050 if (pd.pf_mtag == NULL && 9051 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 9052 action = PF_DROP; 9053 goto done; 9054 } 9055 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 9056 } 9057 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 9058 pd.m->m_flags |= M_FASTFWD_OURS; 9059 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9060 } 9061 m_tag_delete(pd.m, mtag); 9062 9063 mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL); 9064 if (mtag != NULL) 9065 m_tag_delete(pd.m, mtag); 9066 } 9067 9068 switch (pd.virtual_proto) { 9069 case PF_VPROTO_FRAGMENT: 9070 /* 9071 * handle fragments that aren't reassembled by 9072 * normalization 9073 */ 9074 if (kif == NULL || r == NULL) /* pflog */ 9075 action = PF_DROP; 9076 else 9077 action = pf_test_rule(&r, &s, &pd, &a, 9078 &ruleset, inp); 9079 if (action != PF_PASS) 9080 REASON_SET(&reason, PFRES_FRAG); 9081 break; 9082 9083 case IPPROTO_TCP: { 9084 /* Respond to SYN with a syncookie. */ 9085 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 9086 pd.dir == PF_IN && pf_synflood_check(&pd)) { 9087 pf_syncookie_send(&pd); 9088 action = PF_DROP; 9089 break; 9090 } 9091 9092 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 9093 use_2nd_queue = 1; 9094 action = pf_normalize_tcp(&pd); 9095 if (action == PF_DROP) 9096 goto done; 9097 action = pf_test_state_tcp(&s, &pd, &reason); 9098 if (action == PF_PASS) { 9099 if (V_pfsync_update_state_ptr != NULL) 9100 V_pfsync_update_state_ptr(s); 9101 r = s->rule; 9102 a = s->anchor; 9103 } else if (s == NULL) { 9104 /* Validate remote SYN|ACK, re-create original SYN if 9105 * valid. */ 9106 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 9107 TH_ACK && pf_syncookie_validate(&pd) && 9108 pd.dir == PF_IN) { 9109 struct mbuf *msyn; 9110 9111 msyn = pf_syncookie_recreate_syn(&pd); 9112 if (msyn == NULL) { 9113 action = PF_DROP; 9114 break; 9115 } 9116 9117 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 9118 &pd.act); 9119 m_freem(msyn); 9120 if (action != PF_PASS) 9121 break; 9122 9123 action = pf_test_state_tcp(&s, &pd, &reason); 9124 if (action != PF_PASS || s == NULL) { 9125 action = PF_DROP; 9126 break; 9127 } 9128 9129 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 9130 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 9131 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 9132 action = pf_synproxy(&pd, &s, &reason); 9133 break; 9134 } else { 9135 action = pf_test_rule(&r, &s, &pd, 9136 &a, &ruleset, inp); 9137 } 9138 } 9139 break; 9140 } 9141 9142 case IPPROTO_UDP: { 9143 action = pf_test_state_udp(&s, &pd); 9144 if (action == PF_PASS) { 9145 if (V_pfsync_update_state_ptr != NULL) 9146 V_pfsync_update_state_ptr(s); 9147 r = s->rule; 9148 a = s->anchor; 9149 } else if (s == NULL) 9150 action = pf_test_rule(&r, &s, &pd, 9151 &a, &ruleset, inp); 9152 break; 9153 } 9154 9155 case IPPROTO_SCTP: { 9156 action = pf_normalize_sctp(&pd); 9157 if (action == PF_DROP) 9158 goto done; 9159 action = pf_test_state_sctp(&s, &pd, &reason); 9160 if (action == PF_PASS) { 9161 if (V_pfsync_update_state_ptr != NULL) 9162 V_pfsync_update_state_ptr(s); 9163 r = s->rule; 9164 a = s->anchor; 9165 } else if (s == NULL) { 9166 action = pf_test_rule(&r, &s, 9167 &pd, &a, &ruleset, inp); 9168 } 9169 break; 9170 } 9171 9172 case IPPROTO_ICMP: { 9173 if (af != AF_INET) { 9174 action = PF_DROP; 9175 REASON_SET(&reason, PFRES_NORM); 9176 DPFPRINTF(PF_DEBUG_MISC, 9177 ("dropping IPv6 packet with ICMPv4 payload")); 9178 goto done; 9179 } 9180 action = pf_test_state_icmp(&s, &pd, &reason); 9181 if (action == PF_PASS) { 9182 if (V_pfsync_update_state_ptr != NULL) 9183 V_pfsync_update_state_ptr(s); 9184 r = s->rule; 9185 a = s->anchor; 9186 } else if (s == NULL) 9187 action = pf_test_rule(&r, &s, &pd, 9188 &a, &ruleset, inp); 9189 break; 9190 } 9191 9192 case IPPROTO_ICMPV6: { 9193 if (af != AF_INET6) { 9194 action = PF_DROP; 9195 REASON_SET(&reason, PFRES_NORM); 9196 DPFPRINTF(PF_DEBUG_MISC, 9197 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 9198 goto done; 9199 } 9200 action = pf_test_state_icmp(&s, &pd, &reason); 9201 if (action == PF_PASS) { 9202 if (V_pfsync_update_state_ptr != NULL) 9203 V_pfsync_update_state_ptr(s); 9204 r = s->rule; 9205 a = s->anchor; 9206 } else if (s == NULL) 9207 action = pf_test_rule(&r, &s, &pd, 9208 &a, &ruleset, inp); 9209 break; 9210 } 9211 9212 default: 9213 action = pf_test_state_other(&s, &pd); 9214 if (action == PF_PASS) { 9215 if (V_pfsync_update_state_ptr != NULL) 9216 V_pfsync_update_state_ptr(s); 9217 r = s->rule; 9218 a = s->anchor; 9219 } else if (s == NULL) 9220 action = pf_test_rule(&r, &s, &pd, 9221 &a, &ruleset, inp); 9222 break; 9223 } 9224 9225 done: 9226 PF_RULES_RUNLOCK(); 9227 9228 if (pd.m == NULL) 9229 goto eat_pkt; 9230 9231 if (action == PF_PASS && pd.badopts && 9232 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 9233 action = PF_DROP; 9234 REASON_SET(&reason, PFRES_IPOPTIONS); 9235 pd.act.log = PF_LOG_FORCE; 9236 DPFPRINTF(PF_DEBUG_MISC, 9237 ("pf: dropping packet with dangerous headers\n")); 9238 } 9239 9240 if (s) { 9241 uint8_t log = pd.act.log; 9242 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 9243 pd.act.log |= log; 9244 tag = s->tag; 9245 rt = s->rt; 9246 } else { 9247 tag = r->tag; 9248 rt = r->rt; 9249 } 9250 9251 if (tag > 0 && pf_tag_packet(&pd, tag)) { 9252 action = PF_DROP; 9253 REASON_SET(&reason, PFRES_MEMORY); 9254 } 9255 9256 pf_scrub(&pd); 9257 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 9258 pf_normalize_mss(&pd); 9259 9260 if (pd.act.rtableid >= 0) 9261 M_SETFIB(pd.m, pd.act.rtableid); 9262 9263 if (pd.act.flags & PFSTATE_SETPRIO) { 9264 if (pd.tos & IPTOS_LOWDELAY) 9265 use_2nd_queue = 1; 9266 if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) { 9267 action = PF_DROP; 9268 REASON_SET(&reason, PFRES_MEMORY); 9269 pd.act.log = PF_LOG_FORCE; 9270 DPFPRINTF(PF_DEBUG_MISC, 9271 ("pf: failed to allocate 802.1q mtag\n")); 9272 } 9273 } 9274 9275 #ifdef ALTQ 9276 if (action == PF_PASS && pd.act.qid) { 9277 if (pd.pf_mtag == NULL && 9278 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 9279 action = PF_DROP; 9280 REASON_SET(&reason, PFRES_MEMORY); 9281 } else { 9282 if (s != NULL) 9283 pd.pf_mtag->qid_hash = pf_state_hash(s); 9284 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 9285 pd.pf_mtag->qid = pd.act.pqid; 9286 else 9287 pd.pf_mtag->qid = pd.act.qid; 9288 /* Add hints for ecn. */ 9289 pd.pf_mtag->hdr = mtod(pd.m, void *); 9290 } 9291 } 9292 #endif /* ALTQ */ 9293 9294 /* 9295 * connections redirected to loopback should not match sockets 9296 * bound specifically to loopback due to security implications, 9297 * see tcp_input() and in_pcblookup_listen(). 9298 */ 9299 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9300 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 9301 (s->nat_rule->action == PF_RDR || 9302 s->nat_rule->action == PF_BINAT) && 9303 pf_is_loopback(af, pd.dst)) 9304 pd.m->m_flags |= M_SKIP_FIREWALL; 9305 9306 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 9307 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 9308 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 9309 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 9310 if (mtag != NULL) { 9311 ((struct pf_divert_mtag *)(mtag+1))->port = 9312 ntohs(r->divert.port); 9313 ((struct pf_divert_mtag *)(mtag+1))->idir = 9314 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 9315 PF_DIVERT_MTAG_DIR_OUT; 9316 9317 if (s) 9318 PF_STATE_UNLOCK(s); 9319 9320 m_tag_prepend(pd.m, mtag); 9321 if (pd.m->m_flags & M_FASTFWD_OURS) { 9322 if (pd.pf_mtag == NULL && 9323 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 9324 action = PF_DROP; 9325 REASON_SET(&reason, PFRES_MEMORY); 9326 pd.act.log = PF_LOG_FORCE; 9327 DPFPRINTF(PF_DEBUG_MISC, 9328 ("pf: failed to allocate tag\n")); 9329 } else { 9330 pd.pf_mtag->flags |= 9331 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9332 pd.m->m_flags &= ~M_FASTFWD_OURS; 9333 } 9334 } 9335 ip_divert_ptr(*m0, dir == PF_IN); 9336 *m0 = NULL; 9337 9338 return (action); 9339 } else { 9340 /* XXX: ipfw has the same behaviour! */ 9341 action = PF_DROP; 9342 REASON_SET(&reason, PFRES_MEMORY); 9343 pd.act.log = PF_LOG_FORCE; 9344 DPFPRINTF(PF_DEBUG_MISC, 9345 ("pf: failed to allocate divert tag\n")); 9346 } 9347 } 9348 /* XXX: Anybody working on it?! */ 9349 if (af == AF_INET6 && r->divert.port) 9350 printf("pf: divert(9) is not supported for IPv6\n"); 9351 9352 /* this flag will need revising if the pkt is forwarded */ 9353 if (pd.pf_mtag) 9354 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 9355 9356 if (pd.act.log) { 9357 struct pf_krule *lr; 9358 struct pf_krule_item *ri; 9359 9360 if (s != NULL && s->nat_rule != NULL && 9361 s->nat_rule->log & PF_LOG_ALL) 9362 lr = s->nat_rule; 9363 else 9364 lr = r; 9365 9366 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9367 PFLOG_PACKET(action, reason, lr, a, 9368 ruleset, &pd, (s == NULL)); 9369 if (s) { 9370 SLIST_FOREACH(ri, &s->match_rules, entry) 9371 if (ri->r->log & PF_LOG_ALL) 9372 PFLOG_PACKET(action, 9373 reason, ri->r, a, ruleset, &pd, 0); 9374 } 9375 } 9376 9377 pf_counters_inc(action, &pd, s, r, a); 9378 9379 switch (action) { 9380 case PF_SYNPROXY_DROP: 9381 m_freem(*m0); 9382 case PF_DEFER: 9383 *m0 = NULL; 9384 action = PF_PASS; 9385 break; 9386 case PF_DROP: 9387 m_freem(*m0); 9388 *m0 = NULL; 9389 break; 9390 default: 9391 if (rt) { 9392 switch (af) { 9393 #ifdef INET 9394 case AF_INET: 9395 /* pf_route() returns unlocked. */ 9396 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 9397 break; 9398 #endif 9399 #ifdef INET6 9400 case AF_INET6: 9401 /* pf_route6() returns unlocked. */ 9402 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 9403 break; 9404 #endif 9405 } 9406 goto out; 9407 } 9408 if (pf_dummynet(&pd, s, r, m0) != 0) { 9409 action = PF_DROP; 9410 REASON_SET(&reason, PFRES_MEMORY); 9411 } 9412 break; 9413 } 9414 9415 eat_pkt: 9416 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 9417 9418 if (s && action != PF_DROP) { 9419 if (!s->if_index_in && dir == PF_IN) 9420 s->if_index_in = ifp->if_index; 9421 else if (!s->if_index_out && dir == PF_OUT) 9422 s->if_index_out = ifp->if_index; 9423 } 9424 9425 if (s) 9426 PF_STATE_UNLOCK(s); 9427 9428 #ifdef INET6 9429 /* If reassembled packet passed, create new fragments. */ 9430 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 9431 (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 9432 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 9433 #endif 9434 9435 out: 9436 pf_sctp_multihome_delayed(&pd, kif, s, action); 9437 9438 return (action); 9439 } 9440 #endif /* INET || INET6 */ 9441