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_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *", 125 "struct pf_kstate *"); 126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 127 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 128 "struct pf_kstate *"); 129 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 130 "struct pfi_kkif *"); 131 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 132 "struct pf_krule *", "struct mbuf *", "int"); 133 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 134 "struct pf_sctp_source *"); 135 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 136 "struct pf_kstate *", "struct pf_sctp_source *"); 137 138 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 139 "struct mbuf *"); 140 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 141 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 142 "int", "struct pf_keth_rule *", "char *"); 143 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 144 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 145 "int", "struct pf_keth_rule *"); 146 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 147 148 /* 149 * Global variables 150 */ 151 152 /* state tables */ 153 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 154 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 155 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 156 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 157 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 158 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 159 VNET_DEFINE(struct pf_kstatus, pf_status); 160 161 VNET_DEFINE(u_int32_t, ticket_altqs_active); 162 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 163 VNET_DEFINE(int, altqs_inactive_open); 164 VNET_DEFINE(u_int32_t, ticket_pabuf); 165 166 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 167 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 168 VNET_DEFINE(u_char, pf_tcp_secret[16]); 169 #define V_pf_tcp_secret VNET(pf_tcp_secret) 170 VNET_DEFINE(int, pf_tcp_secret_init); 171 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 172 VNET_DEFINE(int, pf_tcp_iss_off); 173 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 174 VNET_DECLARE(int, pf_vnet_active); 175 #define V_pf_vnet_active VNET(pf_vnet_active) 176 177 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 178 #define V_pf_purge_idx VNET(pf_purge_idx) 179 180 #ifdef PF_WANT_32_TO_64_COUNTER 181 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 182 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 183 184 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 185 VNET_DEFINE(size_t, pf_allrulecount); 186 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 187 #endif 188 189 struct pf_sctp_endpoint; 190 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 191 struct pf_sctp_source { 192 sa_family_t af; 193 struct pf_addr addr; 194 TAILQ_ENTRY(pf_sctp_source) entry; 195 }; 196 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 197 struct pf_sctp_endpoint 198 { 199 uint32_t v_tag; 200 struct pf_sctp_sources sources; 201 RB_ENTRY(pf_sctp_endpoint) entry; 202 }; 203 static int 204 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 205 { 206 return (a->v_tag - b->v_tag); 207 } 208 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 209 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 210 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 211 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 212 static struct mtx_padalign pf_sctp_endpoints_mtx; 213 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 214 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 215 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 216 217 /* 218 * Queue for pf_intr() sends. 219 */ 220 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 221 struct pf_send_entry { 222 STAILQ_ENTRY(pf_send_entry) pfse_next; 223 struct mbuf *pfse_m; 224 enum { 225 PFSE_IP, 226 PFSE_IP6, 227 PFSE_ICMP, 228 PFSE_ICMP6, 229 } pfse_type; 230 struct { 231 int type; 232 int code; 233 int mtu; 234 } icmpopts; 235 }; 236 237 STAILQ_HEAD(pf_send_head, pf_send_entry); 238 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 239 #define V_pf_sendqueue VNET(pf_sendqueue) 240 241 static struct mtx_padalign pf_sendqueue_mtx; 242 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 243 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 244 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 245 246 /* 247 * Queue for pf_overload_task() tasks. 248 */ 249 struct pf_overload_entry { 250 SLIST_ENTRY(pf_overload_entry) next; 251 struct pf_addr addr; 252 sa_family_t af; 253 uint8_t dir; 254 struct pf_krule *rule; 255 }; 256 257 SLIST_HEAD(pf_overload_head, pf_overload_entry); 258 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 259 #define V_pf_overloadqueue VNET(pf_overloadqueue) 260 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 261 #define V_pf_overloadtask VNET(pf_overloadtask) 262 263 static struct mtx_padalign pf_overloadqueue_mtx; 264 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 265 "pf overload/flush queue", MTX_DEF); 266 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 267 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 268 269 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 270 struct mtx_padalign pf_unlnkdrules_mtx; 271 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 272 MTX_DEF); 273 274 struct sx pf_config_lock; 275 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 276 277 struct mtx_padalign pf_table_stats_lock; 278 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 279 MTX_DEF); 280 281 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 282 #define V_pf_sources_z VNET(pf_sources_z) 283 uma_zone_t pf_mtag_z; 284 VNET_DEFINE(uma_zone_t, pf_state_z); 285 VNET_DEFINE(uma_zone_t, pf_state_key_z); 286 287 VNET_DEFINE(struct unrhdr64, pf_stateid); 288 289 static void pf_src_tree_remove_state(struct pf_kstate *); 290 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 291 u_int32_t); 292 static void pf_add_threshold(struct pf_threshold *); 293 static int pf_check_threshold(struct pf_threshold *); 294 295 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 296 u_int16_t *, u_int16_t *, struct pf_addr *, 297 u_int16_t, u_int8_t, sa_family_t); 298 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 299 struct tcphdr *, struct pf_state_peer *); 300 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 301 int *, u_int16_t *, u_int16_t *); 302 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 303 struct pf_addr *, struct pf_addr *, u_int16_t, 304 u_int16_t *, u_int16_t *, u_int16_t *, 305 u_int16_t *, u_int8_t, sa_family_t); 306 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 307 sa_family_t, struct pf_krule *, int); 308 static void pf_detach_state(struct pf_kstate *); 309 static int pf_state_key_attach(struct pf_state_key *, 310 struct pf_state_key *, struct pf_kstate *); 311 static void pf_state_key_detach(struct pf_kstate *, int); 312 static int pf_state_key_ctor(void *, int, void *, int); 313 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 314 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 315 struct pf_mtag *pf_mtag); 316 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 317 struct pf_krule *, struct mbuf **); 318 static int pf_dummynet_route(struct pf_pdesc *, 319 struct pf_kstate *, struct pf_krule *, 320 struct ifnet *, struct sockaddr *, struct mbuf **); 321 static int pf_test_eth_rule(int, struct pfi_kkif *, 322 struct mbuf **); 323 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 324 struct pfi_kkif *, struct mbuf *, int, 325 struct pf_pdesc *, struct pf_krule **, 326 struct pf_kruleset **, struct inpcb *); 327 static int pf_create_state(struct pf_krule *, struct pf_krule *, 328 struct pf_krule *, struct pf_pdesc *, 329 struct pf_ksrc_node *, struct pf_state_key *, 330 struct pf_state_key *, struct mbuf *, int, 331 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 332 struct pf_kstate **, int, u_int16_t, u_int16_t, 333 int, struct pf_krule_slist *); 334 static int pf_test_fragment(struct pf_krule **, struct pfi_kkif *, 335 struct mbuf *, void *, struct pf_pdesc *, 336 struct pf_krule **, struct pf_kruleset **); 337 static int pf_tcp_track_full(struct pf_kstate **, 338 struct pfi_kkif *, struct mbuf *, int, 339 struct pf_pdesc *, u_short *, int *); 340 static int pf_tcp_track_sloppy(struct pf_kstate **, 341 struct pf_pdesc *, u_short *); 342 static int pf_test_state_tcp(struct pf_kstate **, 343 struct pfi_kkif *, struct mbuf *, int, 344 void *, struct pf_pdesc *, u_short *); 345 static int pf_test_state_udp(struct pf_kstate **, 346 struct pfi_kkif *, struct mbuf *, int, 347 void *, struct pf_pdesc *); 348 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 349 struct pf_pdesc *, struct pf_kstate **, struct mbuf *, 350 int, struct pfi_kkif *, u_int16_t, u_int16_t, 351 int, int *, int, int); 352 static int pf_test_state_icmp(struct pf_kstate **, 353 struct pfi_kkif *, struct mbuf *, int, 354 void *, struct pf_pdesc *, u_short *); 355 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 356 static void pf_sctp_multihome_delayed(struct pf_pdesc *, int, 357 struct pfi_kkif *, struct pf_kstate *, int); 358 static int pf_test_state_sctp(struct pf_kstate **, 359 struct pfi_kkif *, struct mbuf *, int, 360 void *, struct pf_pdesc *, u_short *); 361 static int pf_test_state_other(struct pf_kstate **, 362 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 363 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 364 int, u_int16_t); 365 static int pf_check_proto_cksum(struct mbuf *, int, int, 366 u_int8_t, sa_family_t); 367 static void pf_print_state_parts(struct pf_kstate *, 368 struct pf_state_key *, struct pf_state_key *); 369 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 370 bool, u_int8_t); 371 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 372 const struct pf_state_key_cmp *, u_int); 373 static int pf_src_connlimit(struct pf_kstate **); 374 static void pf_overload_task(void *v, int pending); 375 static u_short pf_insert_src_node(struct pf_ksrc_node **, 376 struct pf_krule *, struct pf_addr *, sa_family_t); 377 static u_int pf_purge_expired_states(u_int, int); 378 static void pf_purge_unlinked_rules(void); 379 static int pf_mtag_uminit(void *, int, int); 380 static void pf_mtag_free(struct m_tag *); 381 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 382 int, struct pf_state_key *); 383 #ifdef INET 384 static void pf_route(struct mbuf **, struct pf_krule *, 385 struct ifnet *, struct pf_kstate *, 386 struct pf_pdesc *, struct inpcb *); 387 #endif /* INET */ 388 #ifdef INET6 389 static void pf_change_a6(struct pf_addr *, u_int16_t *, 390 struct pf_addr *, u_int8_t); 391 static void pf_route6(struct mbuf **, struct pf_krule *, 392 struct ifnet *, struct pf_kstate *, 393 struct pf_pdesc *, struct inpcb *); 394 #endif /* INET6 */ 395 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 396 397 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 398 399 extern int pf_end_threads; 400 extern struct proc *pf_purge_proc; 401 402 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 403 404 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_SOLICITED, PF_ICMP_MULTI_LINK }; 405 406 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 407 do { \ 408 struct pf_state_key *nk; \ 409 if ((pd->dir) == PF_OUT) \ 410 nk = (_s)->key[PF_SK_STACK]; \ 411 else \ 412 nk = (_s)->key[PF_SK_WIRE]; \ 413 pf_packet_rework_nat(_m, _pd, _off, nk); \ 414 } while (0) 415 416 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 417 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 418 419 #define STATE_LOOKUP(i, k, s, pd) \ 420 do { \ 421 (s) = pf_find_state((i), (k), (pd->dir)); \ 422 SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s)); \ 423 if ((s) == NULL) \ 424 return (PF_DROP); \ 425 if (PACKET_LOOPED(pd)) \ 426 return (PF_PASS); \ 427 } while (0) 428 429 static struct pfi_kkif * 430 BOUND_IFACE(struct pf_kstate *st, struct pfi_kkif *k) 431 { 432 SDT_PROBE2(pf, ip, , bound_iface, st, k); 433 434 /* Floating unless otherwise specified. */ 435 if (! (st->rule.ptr->rule_flag & PFRULE_IFBOUND)) 436 return (V_pfi_all); 437 438 /* 439 * Initially set to all, because we don't know what interface we'll be 440 * sending this out when we create the state. 441 */ 442 if (st->rule.ptr->rt == PF_REPLYTO) 443 return (V_pfi_all); 444 445 /* Don't overrule the interface for states created on incoming packets. */ 446 if (st->direction == PF_IN) 447 return (k); 448 449 /* No route-to, so don't overrule. */ 450 if (st->rt != PF_ROUTETO) 451 return (k); 452 453 /* Bind to the route-to interface. */ 454 return (st->rt_kif); 455 } 456 457 #define STATE_INC_COUNTERS(s) \ 458 do { \ 459 struct pf_krule_item *mrm; \ 460 counter_u64_add(s->rule.ptr->states_cur, 1); \ 461 counter_u64_add(s->rule.ptr->states_tot, 1); \ 462 if (s->anchor.ptr != NULL) { \ 463 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 464 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 465 } \ 466 if (s->nat_rule.ptr != NULL) { \ 467 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 468 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 469 } \ 470 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 471 counter_u64_add(mrm->r->states_cur, 1); \ 472 counter_u64_add(mrm->r->states_tot, 1); \ 473 } \ 474 } while (0) 475 476 #define STATE_DEC_COUNTERS(s) \ 477 do { \ 478 struct pf_krule_item *mrm; \ 479 if (s->nat_rule.ptr != NULL) \ 480 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 481 if (s->anchor.ptr != NULL) \ 482 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 483 counter_u64_add(s->rule.ptr->states_cur, -1); \ 484 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 485 counter_u64_add(mrm->r->states_cur, -1); \ 486 } while (0) 487 488 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 489 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 490 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 491 VNET_DEFINE(struct pf_idhash *, pf_idhash); 492 VNET_DEFINE(struct pf_srchash *, pf_srchash); 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_STATIC(u_long, pf_hashsize); 500 #define V_pf_hashsize VNET(pf_hashsize) 501 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 502 #define V_pf_srchashsize VNET(pf_srchashsize) 503 u_long pf_ioctl_maxcount = 65535; 504 505 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 506 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 507 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 508 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 509 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 510 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 511 512 VNET_DEFINE(void *, pf_swi_cookie); 513 VNET_DEFINE(struct intr_event *, pf_swi_ie); 514 515 VNET_DEFINE(uint32_t, pf_hashseed); 516 #define V_pf_hashseed VNET(pf_hashseed) 517 518 static void 519 pf_sctp_checksum(struct mbuf *m, int off) 520 { 521 uint32_t sum = 0; 522 523 /* Zero out the checksum, to enable recalculation. */ 524 m_copyback(m, off + offsetof(struct sctphdr, checksum), 525 sizeof(sum), (caddr_t)&sum); 526 527 sum = sctp_calculate_cksum(m, off); 528 529 m_copyback(m, off + offsetof(struct sctphdr, checksum), 530 sizeof(sum), (caddr_t)&sum); 531 } 532 533 int 534 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 535 { 536 537 switch (af) { 538 #ifdef INET 539 case AF_INET: 540 if (a->addr32[0] > b->addr32[0]) 541 return (1); 542 if (a->addr32[0] < b->addr32[0]) 543 return (-1); 544 break; 545 #endif /* INET */ 546 #ifdef INET6 547 case AF_INET6: 548 if (a->addr32[3] > b->addr32[3]) 549 return (1); 550 if (a->addr32[3] < b->addr32[3]) 551 return (-1); 552 if (a->addr32[2] > b->addr32[2]) 553 return (1); 554 if (a->addr32[2] < b->addr32[2]) 555 return (-1); 556 if (a->addr32[1] > b->addr32[1]) 557 return (1); 558 if (a->addr32[1] < b->addr32[1]) 559 return (-1); 560 if (a->addr32[0] > b->addr32[0]) 561 return (1); 562 if (a->addr32[0] < b->addr32[0]) 563 return (-1); 564 break; 565 #endif /* INET6 */ 566 default: 567 panic("%s: unknown address family %u", __func__, af); 568 } 569 return (0); 570 } 571 572 static void 573 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 574 struct pf_state_key *nk) 575 { 576 577 switch (pd->proto) { 578 case IPPROTO_TCP: { 579 struct tcphdr *th = &pd->hdr.tcp; 580 581 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 582 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 583 &th->th_sum, &nk->addr[pd->sidx], 584 nk->port[pd->sidx], 0, pd->af); 585 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 586 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 587 &th->th_sum, &nk->addr[pd->didx], 588 nk->port[pd->didx], 0, pd->af); 589 m_copyback(m, off, sizeof(*th), (caddr_t)th); 590 break; 591 } 592 case IPPROTO_UDP: { 593 struct udphdr *uh = &pd->hdr.udp; 594 595 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 596 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 597 &uh->uh_sum, &nk->addr[pd->sidx], 598 nk->port[pd->sidx], 1, pd->af); 599 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 600 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 601 &uh->uh_sum, &nk->addr[pd->didx], 602 nk->port[pd->didx], 1, pd->af); 603 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 604 break; 605 } 606 case IPPROTO_SCTP: { 607 struct sctphdr *sh = &pd->hdr.sctp; 608 uint16_t checksum = 0; 609 610 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 611 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 612 &checksum, &nk->addr[pd->sidx], 613 nk->port[pd->sidx], 1, pd->af); 614 } 615 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 616 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 617 &checksum, &nk->addr[pd->didx], 618 nk->port[pd->didx], 1, pd->af); 619 } 620 621 break; 622 } 623 case IPPROTO_ICMP: { 624 struct icmp *ih = &pd->hdr.icmp; 625 626 if (nk->port[pd->sidx] != ih->icmp_id) { 627 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 628 ih->icmp_cksum, ih->icmp_id, 629 nk->port[pd->sidx], 0); 630 ih->icmp_id = nk->port[pd->sidx]; 631 pd->sport = &ih->icmp_id; 632 633 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 634 } 635 /* FALLTHROUGH */ 636 } 637 default: 638 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 639 switch (pd->af) { 640 case AF_INET: 641 pf_change_a(&pd->src->v4.s_addr, 642 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 643 0); 644 break; 645 case AF_INET6: 646 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 647 break; 648 } 649 } 650 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 651 switch (pd->af) { 652 case AF_INET: 653 pf_change_a(&pd->dst->v4.s_addr, 654 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 655 0); 656 break; 657 case AF_INET6: 658 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 659 break; 660 } 661 } 662 break; 663 } 664 } 665 666 static __inline uint32_t 667 pf_hashkey(const struct pf_state_key *sk) 668 { 669 uint32_t h; 670 671 h = murmur3_32_hash32((const uint32_t *)sk, 672 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 673 V_pf_hashseed); 674 675 return (h & V_pf_hashmask); 676 } 677 678 static __inline uint32_t 679 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 680 { 681 uint32_t h; 682 683 switch (af) { 684 case AF_INET: 685 h = murmur3_32_hash32((uint32_t *)&addr->v4, 686 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 687 break; 688 case AF_INET6: 689 h = murmur3_32_hash32((uint32_t *)&addr->v6, 690 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 691 break; 692 default: 693 panic("%s: unknown address family %u", __func__, af); 694 } 695 696 return (h & V_pf_srchashmask); 697 } 698 699 #ifdef ALTQ 700 static int 701 pf_state_hash(struct pf_kstate *s) 702 { 703 u_int32_t hv = (intptr_t)s / sizeof(*s); 704 705 hv ^= crc32(&s->src, sizeof(s->src)); 706 hv ^= crc32(&s->dst, sizeof(s->dst)); 707 if (hv == 0) 708 hv = 1; 709 return (hv); 710 } 711 #endif 712 713 static __inline void 714 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 715 { 716 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 717 s->dst.state = newstate; 718 if (which == PF_PEER_DST) 719 return; 720 if (s->src.state == newstate) 721 return; 722 if (s->creatorid == V_pf_status.hostid && 723 s->key[PF_SK_STACK] != NULL && 724 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 725 !(TCPS_HAVEESTABLISHED(s->src.state) || 726 s->src.state == TCPS_CLOSED) && 727 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 728 atomic_add_32(&V_pf_status.states_halfopen, -1); 729 730 s->src.state = newstate; 731 } 732 733 #ifdef INET6 734 void 735 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 736 { 737 switch (af) { 738 #ifdef INET 739 case AF_INET: 740 memcpy(&dst->v4, &src->v4, sizeof(dst->v4)); 741 break; 742 #endif /* INET */ 743 case AF_INET6: 744 memcpy(&dst->v6, &src->v6, sizeof(dst->v6)); 745 break; 746 } 747 } 748 #endif /* INET6 */ 749 750 static void 751 pf_init_threshold(struct pf_threshold *threshold, 752 u_int32_t limit, u_int32_t seconds) 753 { 754 threshold->limit = limit * PF_THRESHOLD_MULT; 755 threshold->seconds = seconds; 756 threshold->count = 0; 757 threshold->last = time_uptime; 758 } 759 760 static void 761 pf_add_threshold(struct pf_threshold *threshold) 762 { 763 u_int32_t t = time_uptime, diff = t - threshold->last; 764 765 if (diff >= threshold->seconds) 766 threshold->count = 0; 767 else 768 threshold->count -= threshold->count * diff / 769 threshold->seconds; 770 threshold->count += PF_THRESHOLD_MULT; 771 threshold->last = t; 772 } 773 774 static int 775 pf_check_threshold(struct pf_threshold *threshold) 776 { 777 return (threshold->count > threshold->limit); 778 } 779 780 static int 781 pf_src_connlimit(struct pf_kstate **state) 782 { 783 struct pf_overload_entry *pfoe; 784 int bad = 0; 785 786 PF_STATE_LOCK_ASSERT(*state); 787 /* 788 * XXXKS: The src node is accessed unlocked! 789 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 790 */ 791 792 (*state)->src_node->conn++; 793 (*state)->src.tcp_est = 1; 794 pf_add_threshold(&(*state)->src_node->conn_rate); 795 796 if ((*state)->rule.ptr->max_src_conn && 797 (*state)->rule.ptr->max_src_conn < 798 (*state)->src_node->conn) { 799 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 800 bad++; 801 } 802 803 if ((*state)->rule.ptr->max_src_conn_rate.limit && 804 pf_check_threshold(&(*state)->src_node->conn_rate)) { 805 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 806 bad++; 807 } 808 809 if (!bad) 810 return (0); 811 812 /* Kill this state. */ 813 (*state)->timeout = PFTM_PURGE; 814 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 815 816 if ((*state)->rule.ptr->overload_tbl == NULL) 817 return (1); 818 819 /* Schedule overloading and flushing task. */ 820 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 821 if (pfoe == NULL) 822 return (1); /* too bad :( */ 823 824 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 825 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 826 pfoe->rule = (*state)->rule.ptr; 827 pfoe->dir = (*state)->direction; 828 PF_OVERLOADQ_LOCK(); 829 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 830 PF_OVERLOADQ_UNLOCK(); 831 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 832 833 return (1); 834 } 835 836 static void 837 pf_overload_task(void *v, int pending) 838 { 839 struct pf_overload_head queue; 840 struct pfr_addr p; 841 struct pf_overload_entry *pfoe, *pfoe1; 842 uint32_t killed = 0; 843 844 CURVNET_SET((struct vnet *)v); 845 846 PF_OVERLOADQ_LOCK(); 847 queue = V_pf_overloadqueue; 848 SLIST_INIT(&V_pf_overloadqueue); 849 PF_OVERLOADQ_UNLOCK(); 850 851 bzero(&p, sizeof(p)); 852 SLIST_FOREACH(pfoe, &queue, next) { 853 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 854 if (V_pf_status.debug >= PF_DEBUG_MISC) { 855 printf("%s: blocking address ", __func__); 856 pf_print_host(&pfoe->addr, 0, pfoe->af); 857 printf("\n"); 858 } 859 860 p.pfra_af = pfoe->af; 861 switch (pfoe->af) { 862 #ifdef INET 863 case AF_INET: 864 p.pfra_net = 32; 865 p.pfra_ip4addr = pfoe->addr.v4; 866 break; 867 #endif 868 #ifdef INET6 869 case AF_INET6: 870 p.pfra_net = 128; 871 p.pfra_ip6addr = pfoe->addr.v6; 872 break; 873 #endif 874 } 875 876 PF_RULES_WLOCK(); 877 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 878 PF_RULES_WUNLOCK(); 879 } 880 881 /* 882 * Remove those entries, that don't need flushing. 883 */ 884 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 885 if (pfoe->rule->flush == 0) { 886 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 887 free(pfoe, M_PFTEMP); 888 } else 889 counter_u64_add( 890 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 891 892 /* If nothing to flush, return. */ 893 if (SLIST_EMPTY(&queue)) { 894 CURVNET_RESTORE(); 895 return; 896 } 897 898 for (int i = 0; i <= V_pf_hashmask; i++) { 899 struct pf_idhash *ih = &V_pf_idhash[i]; 900 struct pf_state_key *sk; 901 struct pf_kstate *s; 902 903 PF_HASHROW_LOCK(ih); 904 LIST_FOREACH(s, &ih->states, entry) { 905 sk = s->key[PF_SK_WIRE]; 906 SLIST_FOREACH(pfoe, &queue, next) 907 if (sk->af == pfoe->af && 908 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 909 pfoe->rule == s->rule.ptr) && 910 ((pfoe->dir == PF_OUT && 911 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 912 (pfoe->dir == PF_IN && 913 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 914 s->timeout = PFTM_PURGE; 915 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 916 killed++; 917 } 918 } 919 PF_HASHROW_UNLOCK(ih); 920 } 921 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 922 free(pfoe, M_PFTEMP); 923 if (V_pf_status.debug >= PF_DEBUG_MISC) 924 printf("%s: %u states killed", __func__, killed); 925 926 CURVNET_RESTORE(); 927 } 928 929 /* 930 * Can return locked on failure, so that we can consistently 931 * allocate and insert a new one. 932 */ 933 struct pf_ksrc_node * 934 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 935 struct pf_srchash **sh, bool returnlocked) 936 { 937 struct pf_ksrc_node *n; 938 939 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 940 941 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 942 PF_HASHROW_LOCK(*sh); 943 LIST_FOREACH(n, &(*sh)->nodes, entry) 944 if (n->rule.ptr == rule && n->af == af && 945 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 946 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 947 break; 948 949 if (n != NULL) { 950 n->states++; 951 PF_HASHROW_UNLOCK(*sh); 952 } else if (returnlocked == false) 953 PF_HASHROW_UNLOCK(*sh); 954 955 return (n); 956 } 957 958 static void 959 pf_free_src_node(struct pf_ksrc_node *sn) 960 { 961 962 for (int i = 0; i < 2; i++) { 963 counter_u64_free(sn->bytes[i]); 964 counter_u64_free(sn->packets[i]); 965 } 966 uma_zfree(V_pf_sources_z, sn); 967 } 968 969 static u_short 970 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 971 struct pf_addr *src, sa_family_t af) 972 { 973 u_short reason = 0; 974 struct pf_srchash *sh = NULL; 975 976 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 977 rule->rpool.opts & PF_POOL_STICKYADDR), 978 ("%s for non-tracking rule %p", __func__, rule)); 979 980 if (*sn == NULL) 981 *sn = pf_find_src_node(src, rule, af, &sh, true); 982 983 if (*sn == NULL) { 984 PF_HASHROW_ASSERT(sh); 985 986 if (rule->max_src_nodes && 987 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 988 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 989 PF_HASHROW_UNLOCK(sh); 990 reason = PFRES_SRCLIMIT; 991 goto done; 992 } 993 994 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 995 if ((*sn) == NULL) { 996 PF_HASHROW_UNLOCK(sh); 997 reason = PFRES_MEMORY; 998 goto done; 999 } 1000 1001 for (int i = 0; i < 2; i++) { 1002 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1003 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1004 1005 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1006 pf_free_src_node(*sn); 1007 PF_HASHROW_UNLOCK(sh); 1008 reason = PFRES_MEMORY; 1009 goto done; 1010 } 1011 } 1012 1013 pf_init_threshold(&(*sn)->conn_rate, 1014 rule->max_src_conn_rate.limit, 1015 rule->max_src_conn_rate.seconds); 1016 1017 MPASS((*sn)->lock == NULL); 1018 (*sn)->lock = &sh->lock; 1019 1020 (*sn)->af = af; 1021 (*sn)->rule.ptr = rule; 1022 PF_ACPY(&(*sn)->addr, src, af); 1023 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 1024 (*sn)->creation = time_uptime; 1025 (*sn)->ruletype = rule->action; 1026 (*sn)->states = 1; 1027 if ((*sn)->rule.ptr != NULL) 1028 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 1029 PF_HASHROW_UNLOCK(sh); 1030 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1031 } else { 1032 if (rule->max_src_states && 1033 (*sn)->states >= rule->max_src_states) { 1034 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1035 1); 1036 reason = PFRES_SRCLIMIT; 1037 goto done; 1038 } 1039 } 1040 done: 1041 return (reason); 1042 } 1043 1044 void 1045 pf_unlink_src_node(struct pf_ksrc_node *src) 1046 { 1047 PF_SRC_NODE_LOCK_ASSERT(src); 1048 1049 LIST_REMOVE(src, entry); 1050 if (src->rule.ptr) 1051 counter_u64_add(src->rule.ptr->src_nodes, -1); 1052 } 1053 1054 u_int 1055 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1056 { 1057 struct pf_ksrc_node *sn, *tmp; 1058 u_int count = 0; 1059 1060 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1061 pf_free_src_node(sn); 1062 count++; 1063 } 1064 1065 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1066 1067 return (count); 1068 } 1069 1070 void 1071 pf_mtag_initialize(void) 1072 { 1073 1074 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1075 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1076 UMA_ALIGN_PTR, 0); 1077 } 1078 1079 /* Per-vnet data storage structures initialization. */ 1080 void 1081 pf_initialize(void) 1082 { 1083 struct pf_keyhash *kh; 1084 struct pf_idhash *ih; 1085 struct pf_srchash *sh; 1086 u_int i; 1087 1088 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1089 V_pf_hashsize = PF_HASHSIZ; 1090 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1091 V_pf_srchashsize = PF_SRCHASHSIZ; 1092 1093 V_pf_hashseed = arc4random(); 1094 1095 /* States and state keys storage. */ 1096 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1097 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1098 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1099 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1100 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1101 1102 V_pf_state_key_z = uma_zcreate("pf state keys", 1103 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1104 UMA_ALIGN_PTR, 0); 1105 1106 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1107 M_PFHASH, M_NOWAIT | M_ZERO); 1108 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1109 M_PFHASH, M_NOWAIT | M_ZERO); 1110 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1111 printf("pf: Unable to allocate memory for " 1112 "state_hashsize %lu.\n", V_pf_hashsize); 1113 1114 free(V_pf_keyhash, M_PFHASH); 1115 free(V_pf_idhash, M_PFHASH); 1116 1117 V_pf_hashsize = PF_HASHSIZ; 1118 V_pf_keyhash = mallocarray(V_pf_hashsize, 1119 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1120 V_pf_idhash = mallocarray(V_pf_hashsize, 1121 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1122 } 1123 1124 V_pf_hashmask = V_pf_hashsize - 1; 1125 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1126 i++, kh++, ih++) { 1127 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1128 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1129 } 1130 1131 /* Source nodes. */ 1132 V_pf_sources_z = uma_zcreate("pf source nodes", 1133 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1134 0); 1135 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1136 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1137 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1138 1139 V_pf_srchash = mallocarray(V_pf_srchashsize, 1140 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1141 if (V_pf_srchash == NULL) { 1142 printf("pf: Unable to allocate memory for " 1143 "source_hashsize %lu.\n", V_pf_srchashsize); 1144 1145 V_pf_srchashsize = PF_SRCHASHSIZ; 1146 V_pf_srchash = mallocarray(V_pf_srchashsize, 1147 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1148 } 1149 1150 V_pf_srchashmask = V_pf_srchashsize - 1; 1151 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1152 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1153 1154 /* ALTQ */ 1155 TAILQ_INIT(&V_pf_altqs[0]); 1156 TAILQ_INIT(&V_pf_altqs[1]); 1157 TAILQ_INIT(&V_pf_altqs[2]); 1158 TAILQ_INIT(&V_pf_altqs[3]); 1159 TAILQ_INIT(&V_pf_pabuf); 1160 V_pf_altqs_active = &V_pf_altqs[0]; 1161 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1162 V_pf_altqs_inactive = &V_pf_altqs[2]; 1163 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1164 1165 /* Send & overload+flush queues. */ 1166 STAILQ_INIT(&V_pf_sendqueue); 1167 SLIST_INIT(&V_pf_overloadqueue); 1168 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1169 1170 /* Unlinked, but may be referenced rules. */ 1171 TAILQ_INIT(&V_pf_unlinked_rules); 1172 } 1173 1174 void 1175 pf_mtag_cleanup(void) 1176 { 1177 1178 uma_zdestroy(pf_mtag_z); 1179 } 1180 1181 void 1182 pf_cleanup(void) 1183 { 1184 struct pf_keyhash *kh; 1185 struct pf_idhash *ih; 1186 struct pf_srchash *sh; 1187 struct pf_send_entry *pfse, *next; 1188 u_int i; 1189 1190 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1191 i++, kh++, ih++) { 1192 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1193 __func__)); 1194 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1195 __func__)); 1196 mtx_destroy(&kh->lock); 1197 mtx_destroy(&ih->lock); 1198 } 1199 free(V_pf_keyhash, M_PFHASH); 1200 free(V_pf_idhash, M_PFHASH); 1201 1202 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1203 KASSERT(LIST_EMPTY(&sh->nodes), 1204 ("%s: source node hash not empty", __func__)); 1205 mtx_destroy(&sh->lock); 1206 } 1207 free(V_pf_srchash, M_PFHASH); 1208 1209 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1210 m_freem(pfse->pfse_m); 1211 free(pfse, M_PFTEMP); 1212 } 1213 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1214 1215 uma_zdestroy(V_pf_sources_z); 1216 uma_zdestroy(V_pf_state_z); 1217 uma_zdestroy(V_pf_state_key_z); 1218 } 1219 1220 static int 1221 pf_mtag_uminit(void *mem, int size, int how) 1222 { 1223 struct m_tag *t; 1224 1225 t = (struct m_tag *)mem; 1226 t->m_tag_cookie = MTAG_ABI_COMPAT; 1227 t->m_tag_id = PACKET_TAG_PF; 1228 t->m_tag_len = sizeof(struct pf_mtag); 1229 t->m_tag_free = pf_mtag_free; 1230 1231 return (0); 1232 } 1233 1234 static void 1235 pf_mtag_free(struct m_tag *t) 1236 { 1237 1238 uma_zfree(pf_mtag_z, t); 1239 } 1240 1241 struct pf_mtag * 1242 pf_get_mtag(struct mbuf *m) 1243 { 1244 struct m_tag *mtag; 1245 1246 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1247 return ((struct pf_mtag *)(mtag + 1)); 1248 1249 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1250 if (mtag == NULL) 1251 return (NULL); 1252 bzero(mtag + 1, sizeof(struct pf_mtag)); 1253 m_tag_prepend(m, mtag); 1254 1255 return ((struct pf_mtag *)(mtag + 1)); 1256 } 1257 1258 static int 1259 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1260 struct pf_kstate *s) 1261 { 1262 struct pf_keyhash *khs, *khw, *kh; 1263 struct pf_state_key *sk, *cur; 1264 struct pf_kstate *si, *olds = NULL; 1265 int idx; 1266 1267 NET_EPOCH_ASSERT(); 1268 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1269 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1270 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1271 1272 /* 1273 * We need to lock hash slots of both keys. To avoid deadlock 1274 * we always lock the slot with lower address first. Unlock order 1275 * isn't important. 1276 * 1277 * We also need to lock ID hash slot before dropping key 1278 * locks. On success we return with ID hash slot locked. 1279 */ 1280 1281 if (skw == sks) { 1282 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1283 PF_HASHROW_LOCK(khs); 1284 } else { 1285 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1286 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1287 if (khs == khw) { 1288 PF_HASHROW_LOCK(khs); 1289 } else if (khs < khw) { 1290 PF_HASHROW_LOCK(khs); 1291 PF_HASHROW_LOCK(khw); 1292 } else { 1293 PF_HASHROW_LOCK(khw); 1294 PF_HASHROW_LOCK(khs); 1295 } 1296 } 1297 1298 #define KEYS_UNLOCK() do { \ 1299 if (khs != khw) { \ 1300 PF_HASHROW_UNLOCK(khs); \ 1301 PF_HASHROW_UNLOCK(khw); \ 1302 } else \ 1303 PF_HASHROW_UNLOCK(khs); \ 1304 } while (0) 1305 1306 /* 1307 * First run: start with wire key. 1308 */ 1309 sk = skw; 1310 kh = khw; 1311 idx = PF_SK_WIRE; 1312 1313 MPASS(s->lock == NULL); 1314 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1315 1316 keyattach: 1317 LIST_FOREACH(cur, &kh->keys, entry) 1318 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1319 break; 1320 1321 if (cur != NULL) { 1322 /* Key exists. Check for same kif, if none, add to key. */ 1323 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1324 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1325 1326 PF_HASHROW_LOCK(ih); 1327 if (si->kif == s->kif && 1328 si->direction == s->direction) { 1329 if (sk->proto == IPPROTO_TCP && 1330 si->src.state >= TCPS_FIN_WAIT_2 && 1331 si->dst.state >= TCPS_FIN_WAIT_2) { 1332 /* 1333 * New state matches an old >FIN_WAIT_2 1334 * state. We can't drop key hash locks, 1335 * thus we can't unlink it properly. 1336 * 1337 * As a workaround we drop it into 1338 * TCPS_CLOSED state, schedule purge 1339 * ASAP and push it into the very end 1340 * of the slot TAILQ, so that it won't 1341 * conflict with our new state. 1342 */ 1343 pf_set_protostate(si, PF_PEER_BOTH, 1344 TCPS_CLOSED); 1345 si->timeout = PFTM_PURGE; 1346 olds = si; 1347 } else { 1348 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1349 printf("pf: %s key attach " 1350 "failed on %s: ", 1351 (idx == PF_SK_WIRE) ? 1352 "wire" : "stack", 1353 s->kif->pfik_name); 1354 pf_print_state_parts(s, 1355 (idx == PF_SK_WIRE) ? 1356 sk : NULL, 1357 (idx == PF_SK_STACK) ? 1358 sk : NULL); 1359 printf(", existing: "); 1360 pf_print_state_parts(si, 1361 (idx == PF_SK_WIRE) ? 1362 sk : NULL, 1363 (idx == PF_SK_STACK) ? 1364 sk : NULL); 1365 printf("\n"); 1366 } 1367 s->timeout = PFTM_UNLINKED; 1368 PF_HASHROW_UNLOCK(ih); 1369 KEYS_UNLOCK(); 1370 uma_zfree(V_pf_state_key_z, sk); 1371 if (idx == PF_SK_STACK) 1372 pf_detach_state(s); 1373 return (EEXIST); /* collision! */ 1374 } 1375 } 1376 PF_HASHROW_UNLOCK(ih); 1377 } 1378 uma_zfree(V_pf_state_key_z, sk); 1379 s->key[idx] = cur; 1380 } else { 1381 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1382 s->key[idx] = sk; 1383 } 1384 1385 stateattach: 1386 /* List is sorted, if-bound states before floating. */ 1387 if (s->kif == V_pfi_all) 1388 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1389 else 1390 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1391 1392 if (olds) { 1393 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1394 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1395 key_list[idx]); 1396 olds = NULL; 1397 } 1398 1399 /* 1400 * Attach done. See how should we (or should not?) 1401 * attach a second key. 1402 */ 1403 if (sks == skw) { 1404 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1405 idx = PF_SK_STACK; 1406 sks = NULL; 1407 goto stateattach; 1408 } else if (sks != NULL) { 1409 /* 1410 * Continue attaching with stack key. 1411 */ 1412 sk = sks; 1413 kh = khs; 1414 idx = PF_SK_STACK; 1415 sks = NULL; 1416 goto keyattach; 1417 } 1418 1419 PF_STATE_LOCK(s); 1420 KEYS_UNLOCK(); 1421 1422 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1423 ("%s failure", __func__)); 1424 1425 return (0); 1426 #undef KEYS_UNLOCK 1427 } 1428 1429 static void 1430 pf_detach_state(struct pf_kstate *s) 1431 { 1432 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1433 struct pf_keyhash *kh; 1434 1435 NET_EPOCH_ASSERT(); 1436 MPASS(s->timeout >= PFTM_MAX); 1437 1438 pf_sctp_multihome_detach_addr(s); 1439 1440 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1441 V_pflow_export_state_ptr(s); 1442 1443 if (sks != NULL) { 1444 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1445 PF_HASHROW_LOCK(kh); 1446 if (s->key[PF_SK_STACK] != NULL) 1447 pf_state_key_detach(s, PF_SK_STACK); 1448 /* 1449 * If both point to same key, then we are done. 1450 */ 1451 if (sks == s->key[PF_SK_WIRE]) { 1452 pf_state_key_detach(s, PF_SK_WIRE); 1453 PF_HASHROW_UNLOCK(kh); 1454 return; 1455 } 1456 PF_HASHROW_UNLOCK(kh); 1457 } 1458 1459 if (s->key[PF_SK_WIRE] != NULL) { 1460 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1461 PF_HASHROW_LOCK(kh); 1462 if (s->key[PF_SK_WIRE] != NULL) 1463 pf_state_key_detach(s, PF_SK_WIRE); 1464 PF_HASHROW_UNLOCK(kh); 1465 } 1466 } 1467 1468 static void 1469 pf_state_key_detach(struct pf_kstate *s, int idx) 1470 { 1471 struct pf_state_key *sk = s->key[idx]; 1472 #ifdef INVARIANTS 1473 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1474 1475 PF_HASHROW_ASSERT(kh); 1476 #endif 1477 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1478 s->key[idx] = NULL; 1479 1480 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1481 LIST_REMOVE(sk, entry); 1482 uma_zfree(V_pf_state_key_z, sk); 1483 } 1484 } 1485 1486 static int 1487 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1488 { 1489 struct pf_state_key *sk = mem; 1490 1491 bzero(sk, sizeof(struct pf_state_key_cmp)); 1492 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1493 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1494 1495 return (0); 1496 } 1497 1498 struct pf_state_key * 1499 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1500 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1501 { 1502 struct pf_state_key *sk; 1503 1504 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1505 if (sk == NULL) 1506 return (NULL); 1507 1508 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1509 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1510 sk->port[pd->sidx] = sport; 1511 sk->port[pd->didx] = dport; 1512 sk->proto = pd->proto; 1513 sk->af = pd->af; 1514 1515 return (sk); 1516 } 1517 1518 struct pf_state_key * 1519 pf_state_key_clone(const struct pf_state_key *orig) 1520 { 1521 struct pf_state_key *sk; 1522 1523 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1524 if (sk == NULL) 1525 return (NULL); 1526 1527 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1528 1529 return (sk); 1530 } 1531 1532 int 1533 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1534 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1535 { 1536 struct pf_idhash *ih; 1537 struct pf_kstate *cur; 1538 int error; 1539 1540 NET_EPOCH_ASSERT(); 1541 1542 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1543 ("%s: sks not pristine", __func__)); 1544 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1545 ("%s: skw not pristine", __func__)); 1546 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1547 1548 s->kif = kif; 1549 s->orig_kif = orig_kif; 1550 1551 if (s->id == 0 && s->creatorid == 0) { 1552 s->id = alloc_unr64(&V_pf_stateid); 1553 s->id = htobe64(s->id); 1554 s->creatorid = V_pf_status.hostid; 1555 } 1556 1557 /* Returns with ID locked on success. */ 1558 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1559 return (error); 1560 1561 ih = &V_pf_idhash[PF_IDHASH(s)]; 1562 PF_HASHROW_ASSERT(ih); 1563 LIST_FOREACH(cur, &ih->states, entry) 1564 if (cur->id == s->id && cur->creatorid == s->creatorid) 1565 break; 1566 1567 if (cur != NULL) { 1568 s->timeout = PFTM_UNLINKED; 1569 PF_HASHROW_UNLOCK(ih); 1570 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1571 printf("pf: state ID collision: " 1572 "id: %016llx creatorid: %08x\n", 1573 (unsigned long long)be64toh(s->id), 1574 ntohl(s->creatorid)); 1575 } 1576 pf_detach_state(s); 1577 return (EEXIST); 1578 } 1579 LIST_INSERT_HEAD(&ih->states, s, entry); 1580 /* One for keys, one for ID hash. */ 1581 refcount_init(&s->refs, 2); 1582 1583 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1584 if (V_pfsync_insert_state_ptr != NULL) 1585 V_pfsync_insert_state_ptr(s); 1586 1587 /* Returns locked. */ 1588 return (0); 1589 } 1590 1591 /* 1592 * Find state by ID: returns with locked row on success. 1593 */ 1594 struct pf_kstate * 1595 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1596 { 1597 struct pf_idhash *ih; 1598 struct pf_kstate *s; 1599 1600 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1601 1602 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1603 1604 PF_HASHROW_LOCK(ih); 1605 LIST_FOREACH(s, &ih->states, entry) 1606 if (s->id == id && s->creatorid == creatorid) 1607 break; 1608 1609 if (s == NULL) 1610 PF_HASHROW_UNLOCK(ih); 1611 1612 return (s); 1613 } 1614 1615 /* 1616 * Find state by key. 1617 * Returns with ID hash slot locked on success. 1618 */ 1619 static struct pf_kstate * 1620 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1621 u_int dir) 1622 { 1623 struct pf_keyhash *kh; 1624 struct pf_state_key *sk; 1625 struct pf_kstate *s; 1626 int idx; 1627 1628 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1629 1630 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1631 1632 PF_HASHROW_LOCK(kh); 1633 LIST_FOREACH(sk, &kh->keys, entry) 1634 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1635 break; 1636 if (sk == NULL) { 1637 PF_HASHROW_UNLOCK(kh); 1638 return (NULL); 1639 } 1640 1641 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1642 1643 /* List is sorted, if-bound states before floating ones. */ 1644 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1645 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1646 PF_STATE_LOCK(s); 1647 PF_HASHROW_UNLOCK(kh); 1648 if (__predict_false(s->timeout >= PFTM_MAX)) { 1649 /* 1650 * State is either being processed by 1651 * pf_unlink_state() in an other thread, or 1652 * is scheduled for immediate expiry. 1653 */ 1654 PF_STATE_UNLOCK(s); 1655 return (NULL); 1656 } 1657 return (s); 1658 } 1659 PF_HASHROW_UNLOCK(kh); 1660 1661 return (NULL); 1662 } 1663 1664 /* 1665 * Returns with ID hash slot locked on success. 1666 */ 1667 struct pf_kstate * 1668 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1669 { 1670 struct pf_keyhash *kh; 1671 struct pf_state_key *sk; 1672 struct pf_kstate *s, *ret = NULL; 1673 int idx, inout = 0; 1674 1675 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1676 1677 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1678 1679 PF_HASHROW_LOCK(kh); 1680 LIST_FOREACH(sk, &kh->keys, entry) 1681 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1682 break; 1683 if (sk == NULL) { 1684 PF_HASHROW_UNLOCK(kh); 1685 return (NULL); 1686 } 1687 switch (dir) { 1688 case PF_IN: 1689 idx = PF_SK_WIRE; 1690 break; 1691 case PF_OUT: 1692 idx = PF_SK_STACK; 1693 break; 1694 case PF_INOUT: 1695 idx = PF_SK_WIRE; 1696 inout = 1; 1697 break; 1698 default: 1699 panic("%s: dir %u", __func__, dir); 1700 } 1701 second_run: 1702 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1703 if (more == NULL) { 1704 PF_STATE_LOCK(s); 1705 PF_HASHROW_UNLOCK(kh); 1706 return (s); 1707 } 1708 1709 if (ret) 1710 (*more)++; 1711 else { 1712 ret = s; 1713 PF_STATE_LOCK(s); 1714 } 1715 } 1716 if (inout == 1) { 1717 inout = 0; 1718 idx = PF_SK_STACK; 1719 goto second_run; 1720 } 1721 PF_HASHROW_UNLOCK(kh); 1722 1723 return (ret); 1724 } 1725 1726 /* 1727 * FIXME 1728 * This routine is inefficient -- locks the state only to unlock immediately on 1729 * return. 1730 * It is racy -- after the state is unlocked nothing stops other threads from 1731 * removing it. 1732 */ 1733 bool 1734 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 1735 { 1736 struct pf_kstate *s; 1737 1738 s = pf_find_state_all(key, dir, NULL); 1739 if (s != NULL) { 1740 PF_STATE_UNLOCK(s); 1741 return (true); 1742 } 1743 return (false); 1744 } 1745 1746 /* END state table stuff */ 1747 1748 static void 1749 pf_send(struct pf_send_entry *pfse) 1750 { 1751 1752 PF_SENDQ_LOCK(); 1753 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1754 PF_SENDQ_UNLOCK(); 1755 swi_sched(V_pf_swi_cookie, 0); 1756 } 1757 1758 static bool 1759 pf_isforlocal(struct mbuf *m, int af) 1760 { 1761 switch (af) { 1762 #ifdef INET 1763 case AF_INET: { 1764 struct ip *ip = mtod(m, struct ip *); 1765 1766 return (in_localip(ip->ip_dst)); 1767 } 1768 #endif 1769 #ifdef INET6 1770 case AF_INET6: { 1771 struct ip6_hdr *ip6; 1772 struct in6_ifaddr *ia; 1773 ip6 = mtod(m, struct ip6_hdr *); 1774 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1775 if (ia == NULL) 1776 return (false); 1777 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 1778 } 1779 #endif 1780 default: 1781 panic("Unsupported af %d", af); 1782 } 1783 1784 return (false); 1785 } 1786 1787 int 1788 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 1789 int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type) 1790 { 1791 /* 1792 * ICMP types marked with PF_OUT are typically responses to 1793 * PF_IN, and will match states in the opposite direction. 1794 * PF_IN ICMP types need to match a state with that type. 1795 */ 1796 *icmp_dir = PF_OUT; 1797 *multi = PF_ICMP_MULTI_LINK; 1798 /* Queries (and responses) */ 1799 switch (pd->af) { 1800 #ifdef INET 1801 case AF_INET: 1802 switch (type) { 1803 case ICMP_ECHO: 1804 *icmp_dir = PF_IN; 1805 case ICMP_ECHOREPLY: 1806 *virtual_type = ICMP_ECHO; 1807 *virtual_id = pd->hdr.icmp.icmp_id; 1808 break; 1809 1810 case ICMP_TSTAMP: 1811 *icmp_dir = PF_IN; 1812 case ICMP_TSTAMPREPLY: 1813 *virtual_type = ICMP_TSTAMP; 1814 *virtual_id = pd->hdr.icmp.icmp_id; 1815 break; 1816 1817 case ICMP_IREQ: 1818 *icmp_dir = PF_IN; 1819 case ICMP_IREQREPLY: 1820 *virtual_type = ICMP_IREQ; 1821 *virtual_id = pd->hdr.icmp.icmp_id; 1822 break; 1823 1824 case ICMP_MASKREQ: 1825 *icmp_dir = PF_IN; 1826 case ICMP_MASKREPLY: 1827 *virtual_type = ICMP_MASKREQ; 1828 *virtual_id = pd->hdr.icmp.icmp_id; 1829 break; 1830 1831 case ICMP_IPV6_WHEREAREYOU: 1832 *icmp_dir = PF_IN; 1833 case ICMP_IPV6_IAMHERE: 1834 *virtual_type = ICMP_IPV6_WHEREAREYOU; 1835 *virtual_id = 0; /* Nothing sane to match on! */ 1836 break; 1837 1838 case ICMP_MOBILE_REGREQUEST: 1839 *icmp_dir = PF_IN; 1840 case ICMP_MOBILE_REGREPLY: 1841 *virtual_type = ICMP_MOBILE_REGREQUEST; 1842 *virtual_id = 0; /* Nothing sane to match on! */ 1843 break; 1844 1845 case ICMP_ROUTERSOLICIT: 1846 *icmp_dir = PF_IN; 1847 case ICMP_ROUTERADVERT: 1848 *virtual_type = ICMP_ROUTERSOLICIT; 1849 *virtual_id = 0; /* Nothing sane to match on! */ 1850 break; 1851 1852 /* These ICMP types map to other connections */ 1853 case ICMP_UNREACH: 1854 case ICMP_SOURCEQUENCH: 1855 case ICMP_REDIRECT: 1856 case ICMP_TIMXCEED: 1857 case ICMP_PARAMPROB: 1858 /* These will not be used, but set them anyway */ 1859 *icmp_dir = PF_IN; 1860 *virtual_type = type; 1861 *virtual_id = 0; 1862 HTONS(*virtual_type); 1863 return (1); /* These types match to another state */ 1864 1865 /* 1866 * All remaining ICMP types get their own states, 1867 * and will only match in one direction. 1868 */ 1869 default: 1870 *icmp_dir = PF_IN; 1871 *virtual_type = type; 1872 *virtual_id = 0; 1873 break; 1874 } 1875 break; 1876 #endif /* INET */ 1877 #ifdef INET6 1878 case AF_INET6: 1879 switch (type) { 1880 case ICMP6_ECHO_REQUEST: 1881 *icmp_dir = PF_IN; 1882 case ICMP6_ECHO_REPLY: 1883 *virtual_type = ICMP6_ECHO_REQUEST; 1884 *virtual_id = pd->hdr.icmp6.icmp6_id; 1885 break; 1886 1887 case MLD_LISTENER_QUERY: 1888 case MLD_LISTENER_REPORT: { 1889 /* 1890 * Listener Report can be sent by clients 1891 * without an associated Listener Query. 1892 * In addition to that, when Report is sent as a 1893 * reply to a Query its source and destination 1894 * address are different. 1895 */ 1896 *icmp_dir = PF_IN; 1897 *virtual_type = MLD_LISTENER_QUERY; 1898 *virtual_id = 0; 1899 break; 1900 } 1901 case MLD_MTRACE: 1902 *icmp_dir = PF_IN; 1903 case MLD_MTRACE_RESP: 1904 *virtual_type = MLD_MTRACE; 1905 *virtual_id = 0; /* Nothing sane to match on! */ 1906 break; 1907 1908 case ND_NEIGHBOR_SOLICIT: 1909 *icmp_dir = PF_IN; 1910 case ND_NEIGHBOR_ADVERT: { 1911 *virtual_type = ND_NEIGHBOR_SOLICIT; 1912 *virtual_id = 0; 1913 break; 1914 } 1915 1916 /* 1917 * These ICMP types map to other connections. 1918 * ND_REDIRECT can't be in this list because the triggering 1919 * packet header is optional. 1920 */ 1921 case ICMP6_DST_UNREACH: 1922 case ICMP6_PACKET_TOO_BIG: 1923 case ICMP6_TIME_EXCEEDED: 1924 case ICMP6_PARAM_PROB: 1925 /* These will not be used, but set them anyway */ 1926 *icmp_dir = PF_IN; 1927 *virtual_type = type; 1928 *virtual_id = 0; 1929 HTONS(*virtual_type); 1930 return (1); /* These types match to another state */ 1931 /* 1932 * All remaining ICMP6 types get their own states, 1933 * and will only match in one direction. 1934 */ 1935 default: 1936 *icmp_dir = PF_IN; 1937 *virtual_type = type; 1938 *virtual_id = 0; 1939 break; 1940 } 1941 break; 1942 #endif /* INET6 */ 1943 default: 1944 *icmp_dir = PF_IN; 1945 *virtual_type = type; 1946 *virtual_id = 0; 1947 break; 1948 } 1949 HTONS(*virtual_type); 1950 return (0); /* These types match to their own state */ 1951 } 1952 1953 void 1954 pf_intr(void *v) 1955 { 1956 struct epoch_tracker et; 1957 struct pf_send_head queue; 1958 struct pf_send_entry *pfse, *next; 1959 1960 CURVNET_SET((struct vnet *)v); 1961 1962 PF_SENDQ_LOCK(); 1963 queue = V_pf_sendqueue; 1964 STAILQ_INIT(&V_pf_sendqueue); 1965 PF_SENDQ_UNLOCK(); 1966 1967 NET_EPOCH_ENTER(et); 1968 1969 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1970 switch (pfse->pfse_type) { 1971 #ifdef INET 1972 case PFSE_IP: { 1973 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 1974 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1975 pfse->pfse_m->m_pkthdr.csum_flags |= 1976 CSUM_IP_VALID | CSUM_IP_CHECKED; 1977 ip_input(pfse->pfse_m); 1978 } else { 1979 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1980 NULL); 1981 } 1982 break; 1983 } 1984 case PFSE_ICMP: 1985 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1986 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1987 break; 1988 #endif /* INET */ 1989 #ifdef INET6 1990 case PFSE_IP6: 1991 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 1992 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1993 ip6_input(pfse->pfse_m); 1994 } else { 1995 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1996 NULL, NULL); 1997 } 1998 break; 1999 case PFSE_ICMP6: 2000 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2001 pfse->icmpopts.code, pfse->icmpopts.mtu); 2002 break; 2003 #endif /* INET6 */ 2004 default: 2005 panic("%s: unknown type", __func__); 2006 } 2007 free(pfse, M_PFTEMP); 2008 } 2009 NET_EPOCH_EXIT(et); 2010 CURVNET_RESTORE(); 2011 } 2012 2013 #define pf_purge_thread_period (hz / 10) 2014 2015 #ifdef PF_WANT_32_TO_64_COUNTER 2016 static void 2017 pf_status_counter_u64_periodic(void) 2018 { 2019 2020 PF_RULES_RASSERT(); 2021 2022 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2023 return; 2024 } 2025 2026 for (int i = 0; i < FCNT_MAX; i++) { 2027 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2028 } 2029 } 2030 2031 static void 2032 pf_kif_counter_u64_periodic(void) 2033 { 2034 struct pfi_kkif *kif; 2035 size_t r, run; 2036 2037 PF_RULES_RASSERT(); 2038 2039 if (__predict_false(V_pf_allkifcount == 0)) { 2040 return; 2041 } 2042 2043 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2044 return; 2045 } 2046 2047 run = V_pf_allkifcount / 10; 2048 if (run < 5) 2049 run = 5; 2050 2051 for (r = 0; r < run; r++) { 2052 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2053 if (kif == NULL) { 2054 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2055 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2056 break; 2057 } 2058 2059 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2060 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2061 2062 for (int i = 0; i < 2; i++) { 2063 for (int j = 0; j < 2; j++) { 2064 for (int k = 0; k < 2; k++) { 2065 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2066 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2067 } 2068 } 2069 } 2070 } 2071 } 2072 2073 static void 2074 pf_rule_counter_u64_periodic(void) 2075 { 2076 struct pf_krule *rule; 2077 size_t r, run; 2078 2079 PF_RULES_RASSERT(); 2080 2081 if (__predict_false(V_pf_allrulecount == 0)) { 2082 return; 2083 } 2084 2085 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2086 return; 2087 } 2088 2089 run = V_pf_allrulecount / 10; 2090 if (run < 5) 2091 run = 5; 2092 2093 for (r = 0; r < run; r++) { 2094 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2095 if (rule == NULL) { 2096 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2097 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2098 break; 2099 } 2100 2101 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2102 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2103 2104 pf_counter_u64_periodic(&rule->evaluations); 2105 for (int i = 0; i < 2; i++) { 2106 pf_counter_u64_periodic(&rule->packets[i]); 2107 pf_counter_u64_periodic(&rule->bytes[i]); 2108 } 2109 } 2110 } 2111 2112 static void 2113 pf_counter_u64_periodic_main(void) 2114 { 2115 PF_RULES_RLOCK_TRACKER; 2116 2117 V_pf_counter_periodic_iter++; 2118 2119 PF_RULES_RLOCK(); 2120 pf_counter_u64_critical_enter(); 2121 pf_status_counter_u64_periodic(); 2122 pf_kif_counter_u64_periodic(); 2123 pf_rule_counter_u64_periodic(); 2124 pf_counter_u64_critical_exit(); 2125 PF_RULES_RUNLOCK(); 2126 } 2127 #else 2128 #define pf_counter_u64_periodic_main() do { } while (0) 2129 #endif 2130 2131 void 2132 pf_purge_thread(void *unused __unused) 2133 { 2134 struct epoch_tracker et; 2135 2136 VNET_ITERATOR_DECL(vnet_iter); 2137 2138 sx_xlock(&pf_end_lock); 2139 while (pf_end_threads == 0) { 2140 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2141 2142 VNET_LIST_RLOCK(); 2143 NET_EPOCH_ENTER(et); 2144 VNET_FOREACH(vnet_iter) { 2145 CURVNET_SET(vnet_iter); 2146 2147 /* Wait until V_pf_default_rule is initialized. */ 2148 if (V_pf_vnet_active == 0) { 2149 CURVNET_RESTORE(); 2150 continue; 2151 } 2152 2153 pf_counter_u64_periodic_main(); 2154 2155 /* 2156 * Process 1/interval fraction of the state 2157 * table every run. 2158 */ 2159 V_pf_purge_idx = 2160 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2161 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2162 2163 /* 2164 * Purge other expired types every 2165 * PFTM_INTERVAL seconds. 2166 */ 2167 if (V_pf_purge_idx == 0) { 2168 /* 2169 * Order is important: 2170 * - states and src nodes reference rules 2171 * - states and rules reference kifs 2172 */ 2173 pf_purge_expired_fragments(); 2174 pf_purge_expired_src_nodes(); 2175 pf_purge_unlinked_rules(); 2176 pfi_kkif_purge(); 2177 } 2178 CURVNET_RESTORE(); 2179 } 2180 NET_EPOCH_EXIT(et); 2181 VNET_LIST_RUNLOCK(); 2182 } 2183 2184 pf_end_threads++; 2185 sx_xunlock(&pf_end_lock); 2186 kproc_exit(0); 2187 } 2188 2189 void 2190 pf_unload_vnet_purge(void) 2191 { 2192 2193 /* 2194 * To cleanse up all kifs and rules we need 2195 * two runs: first one clears reference flags, 2196 * then pf_purge_expired_states() doesn't 2197 * raise them, and then second run frees. 2198 */ 2199 pf_purge_unlinked_rules(); 2200 pfi_kkif_purge(); 2201 2202 /* 2203 * Now purge everything. 2204 */ 2205 pf_purge_expired_states(0, V_pf_hashmask); 2206 pf_purge_fragments(UINT_MAX); 2207 pf_purge_expired_src_nodes(); 2208 2209 /* 2210 * Now all kifs & rules should be unreferenced, 2211 * thus should be successfully freed. 2212 */ 2213 pf_purge_unlinked_rules(); 2214 pfi_kkif_purge(); 2215 } 2216 2217 u_int32_t 2218 pf_state_expires(const struct pf_kstate *state) 2219 { 2220 u_int32_t timeout; 2221 u_int32_t start; 2222 u_int32_t end; 2223 u_int32_t states; 2224 2225 /* handle all PFTM_* > PFTM_MAX here */ 2226 if (state->timeout == PFTM_PURGE) 2227 return (time_uptime); 2228 KASSERT(state->timeout != PFTM_UNLINKED, 2229 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2230 KASSERT((state->timeout < PFTM_MAX), 2231 ("pf_state_expires: timeout > PFTM_MAX")); 2232 timeout = state->rule.ptr->timeout[state->timeout]; 2233 if (!timeout) 2234 timeout = V_pf_default_rule.timeout[state->timeout]; 2235 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 2236 if (start && state->rule.ptr != &V_pf_default_rule) { 2237 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 2238 states = counter_u64_fetch(state->rule.ptr->states_cur); 2239 } else { 2240 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2241 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2242 states = V_pf_status.states; 2243 } 2244 if (end && states > start && start < end) { 2245 if (states < end) { 2246 timeout = (u_int64_t)timeout * (end - states) / 2247 (end - start); 2248 return ((state->expire / 1000) + timeout); 2249 } 2250 else 2251 return (time_uptime); 2252 } 2253 return ((state->expire / 1000) + timeout); 2254 } 2255 2256 void 2257 pf_purge_expired_src_nodes(void) 2258 { 2259 struct pf_ksrc_node_list freelist; 2260 struct pf_srchash *sh; 2261 struct pf_ksrc_node *cur, *next; 2262 int i; 2263 2264 LIST_INIT(&freelist); 2265 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2266 PF_HASHROW_LOCK(sh); 2267 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2268 if (cur->states == 0 && cur->expire <= time_uptime) { 2269 pf_unlink_src_node(cur); 2270 LIST_INSERT_HEAD(&freelist, cur, entry); 2271 } else if (cur->rule.ptr != NULL) 2272 cur->rule.ptr->rule_ref |= PFRULE_REFS; 2273 PF_HASHROW_UNLOCK(sh); 2274 } 2275 2276 pf_free_src_nodes(&freelist); 2277 2278 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2279 } 2280 2281 static void 2282 pf_src_tree_remove_state(struct pf_kstate *s) 2283 { 2284 struct pf_ksrc_node *sn; 2285 uint32_t timeout; 2286 2287 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 2288 s->rule.ptr->timeout[PFTM_SRC_NODE] : 2289 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2290 2291 if (s->src_node != NULL) { 2292 sn = s->src_node; 2293 PF_SRC_NODE_LOCK(sn); 2294 if (s->src.tcp_est) 2295 --sn->conn; 2296 if (--sn->states == 0) 2297 sn->expire = time_uptime + timeout; 2298 PF_SRC_NODE_UNLOCK(sn); 2299 } 2300 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2301 sn = s->nat_src_node; 2302 PF_SRC_NODE_LOCK(sn); 2303 if (--sn->states == 0) 2304 sn->expire = time_uptime + timeout; 2305 PF_SRC_NODE_UNLOCK(sn); 2306 } 2307 s->src_node = s->nat_src_node = NULL; 2308 } 2309 2310 /* 2311 * Unlink and potentilly free a state. Function may be 2312 * called with ID hash row locked, but always returns 2313 * unlocked, since it needs to go through key hash locking. 2314 */ 2315 int 2316 pf_unlink_state(struct pf_kstate *s) 2317 { 2318 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2319 2320 NET_EPOCH_ASSERT(); 2321 PF_HASHROW_ASSERT(ih); 2322 2323 if (s->timeout == PFTM_UNLINKED) { 2324 /* 2325 * State is being processed 2326 * by pf_unlink_state() in 2327 * an other thread. 2328 */ 2329 PF_HASHROW_UNLOCK(ih); 2330 return (0); /* XXXGL: undefined actually */ 2331 } 2332 2333 if (s->src.state == PF_TCPS_PROXY_DST) { 2334 /* XXX wire key the right one? */ 2335 pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af, 2336 &s->key[PF_SK_WIRE]->addr[1], 2337 &s->key[PF_SK_WIRE]->addr[0], 2338 s->key[PF_SK_WIRE]->port[1], 2339 s->key[PF_SK_WIRE]->port[0], 2340 s->src.seqhi, s->src.seqlo + 1, 2341 TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid); 2342 } 2343 2344 LIST_REMOVE(s, entry); 2345 pf_src_tree_remove_state(s); 2346 2347 if (V_pfsync_delete_state_ptr != NULL) 2348 V_pfsync_delete_state_ptr(s); 2349 2350 STATE_DEC_COUNTERS(s); 2351 2352 s->timeout = PFTM_UNLINKED; 2353 2354 /* Ensure we remove it from the list of halfopen states, if needed. */ 2355 if (s->key[PF_SK_STACK] != NULL && 2356 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2357 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2358 2359 PF_HASHROW_UNLOCK(ih); 2360 2361 pf_detach_state(s); 2362 /* pf_state_insert() initialises refs to 2 */ 2363 return (pf_release_staten(s, 2)); 2364 } 2365 2366 struct pf_kstate * 2367 pf_alloc_state(int flags) 2368 { 2369 2370 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2371 } 2372 2373 void 2374 pf_free_state(struct pf_kstate *cur) 2375 { 2376 struct pf_krule_item *ri; 2377 2378 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2379 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2380 cur->timeout)); 2381 2382 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2383 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2384 free(ri, M_PF_RULE_ITEM); 2385 } 2386 2387 pf_normalize_tcp_cleanup(cur); 2388 uma_zfree(V_pf_state_z, cur); 2389 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2390 } 2391 2392 /* 2393 * Called only from pf_purge_thread(), thus serialized. 2394 */ 2395 static u_int 2396 pf_purge_expired_states(u_int i, int maxcheck) 2397 { 2398 struct pf_idhash *ih; 2399 struct pf_kstate *s; 2400 struct pf_krule_item *mrm; 2401 size_t count __unused; 2402 2403 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2404 2405 /* 2406 * Go through hash and unlink states that expire now. 2407 */ 2408 while (maxcheck > 0) { 2409 count = 0; 2410 ih = &V_pf_idhash[i]; 2411 2412 /* only take the lock if we expect to do work */ 2413 if (!LIST_EMPTY(&ih->states)) { 2414 relock: 2415 PF_HASHROW_LOCK(ih); 2416 LIST_FOREACH(s, &ih->states, entry) { 2417 if (pf_state_expires(s) <= time_uptime) { 2418 V_pf_status.states -= 2419 pf_unlink_state(s); 2420 goto relock; 2421 } 2422 s->rule.ptr->rule_ref |= PFRULE_REFS; 2423 if (s->nat_rule.ptr != NULL) 2424 s->nat_rule.ptr->rule_ref |= PFRULE_REFS; 2425 if (s->anchor.ptr != NULL) 2426 s->anchor.ptr->rule_ref |= PFRULE_REFS; 2427 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2428 SLIST_FOREACH(mrm, &s->match_rules, entry) 2429 mrm->r->rule_ref |= PFRULE_REFS; 2430 if (s->rt_kif) 2431 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2432 count++; 2433 } 2434 PF_HASHROW_UNLOCK(ih); 2435 } 2436 2437 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2438 2439 /* Return when we hit end of hash. */ 2440 if (++i > V_pf_hashmask) { 2441 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2442 return (0); 2443 } 2444 2445 maxcheck--; 2446 } 2447 2448 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2449 2450 return (i); 2451 } 2452 2453 static void 2454 pf_purge_unlinked_rules(void) 2455 { 2456 struct pf_krulequeue tmpq; 2457 struct pf_krule *r, *r1; 2458 2459 /* 2460 * If we have overloading task pending, then we'd 2461 * better skip purging this time. There is a tiny 2462 * probability that overloading task references 2463 * an already unlinked rule. 2464 */ 2465 PF_OVERLOADQ_LOCK(); 2466 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2467 PF_OVERLOADQ_UNLOCK(); 2468 return; 2469 } 2470 PF_OVERLOADQ_UNLOCK(); 2471 2472 /* 2473 * Do naive mark-and-sweep garbage collecting of old rules. 2474 * Reference flag is raised by pf_purge_expired_states() 2475 * and pf_purge_expired_src_nodes(). 2476 * 2477 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2478 * use a temporary queue. 2479 */ 2480 TAILQ_INIT(&tmpq); 2481 PF_UNLNKDRULES_LOCK(); 2482 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2483 if (!(r->rule_ref & PFRULE_REFS)) { 2484 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2485 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2486 } else 2487 r->rule_ref &= ~PFRULE_REFS; 2488 } 2489 PF_UNLNKDRULES_UNLOCK(); 2490 2491 if (!TAILQ_EMPTY(&tmpq)) { 2492 PF_CONFIG_LOCK(); 2493 PF_RULES_WLOCK(); 2494 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2495 TAILQ_REMOVE(&tmpq, r, entries); 2496 pf_free_rule(r); 2497 } 2498 PF_RULES_WUNLOCK(); 2499 PF_CONFIG_UNLOCK(); 2500 } 2501 } 2502 2503 void 2504 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2505 { 2506 switch (af) { 2507 #ifdef INET 2508 case AF_INET: { 2509 u_int32_t a = ntohl(addr->addr32[0]); 2510 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2511 (a>>8)&255, a&255); 2512 if (p) { 2513 p = ntohs(p); 2514 printf(":%u", p); 2515 } 2516 break; 2517 } 2518 #endif /* INET */ 2519 #ifdef INET6 2520 case AF_INET6: { 2521 u_int16_t b; 2522 u_int8_t i, curstart, curend, maxstart, maxend; 2523 curstart = curend = maxstart = maxend = 255; 2524 for (i = 0; i < 8; i++) { 2525 if (!addr->addr16[i]) { 2526 if (curstart == 255) 2527 curstart = i; 2528 curend = i; 2529 } else { 2530 if ((curend - curstart) > 2531 (maxend - maxstart)) { 2532 maxstart = curstart; 2533 maxend = curend; 2534 } 2535 curstart = curend = 255; 2536 } 2537 } 2538 if ((curend - curstart) > 2539 (maxend - maxstart)) { 2540 maxstart = curstart; 2541 maxend = curend; 2542 } 2543 for (i = 0; i < 8; i++) { 2544 if (i >= maxstart && i <= maxend) { 2545 if (i == 0) 2546 printf(":"); 2547 if (i == maxend) 2548 printf(":"); 2549 } else { 2550 b = ntohs(addr->addr16[i]); 2551 printf("%x", b); 2552 if (i < 7) 2553 printf(":"); 2554 } 2555 } 2556 if (p) { 2557 p = ntohs(p); 2558 printf("[%u]", p); 2559 } 2560 break; 2561 } 2562 #endif /* INET6 */ 2563 } 2564 } 2565 2566 void 2567 pf_print_state(struct pf_kstate *s) 2568 { 2569 pf_print_state_parts(s, NULL, NULL); 2570 } 2571 2572 static void 2573 pf_print_state_parts(struct pf_kstate *s, 2574 struct pf_state_key *skwp, struct pf_state_key *sksp) 2575 { 2576 struct pf_state_key *skw, *sks; 2577 u_int8_t proto, dir; 2578 2579 /* Do our best to fill these, but they're skipped if NULL */ 2580 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2581 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2582 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2583 dir = s ? s->direction : 0; 2584 2585 switch (proto) { 2586 case IPPROTO_IPV4: 2587 printf("IPv4"); 2588 break; 2589 case IPPROTO_IPV6: 2590 printf("IPv6"); 2591 break; 2592 case IPPROTO_TCP: 2593 printf("TCP"); 2594 break; 2595 case IPPROTO_UDP: 2596 printf("UDP"); 2597 break; 2598 case IPPROTO_ICMP: 2599 printf("ICMP"); 2600 break; 2601 case IPPROTO_ICMPV6: 2602 printf("ICMPv6"); 2603 break; 2604 default: 2605 printf("%u", proto); 2606 break; 2607 } 2608 switch (dir) { 2609 case PF_IN: 2610 printf(" in"); 2611 break; 2612 case PF_OUT: 2613 printf(" out"); 2614 break; 2615 } 2616 if (skw) { 2617 printf(" wire: "); 2618 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2619 printf(" "); 2620 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2621 } 2622 if (sks) { 2623 printf(" stack: "); 2624 if (sks != skw) { 2625 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2626 printf(" "); 2627 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2628 } else 2629 printf("-"); 2630 } 2631 if (s) { 2632 if (proto == IPPROTO_TCP) { 2633 printf(" [lo=%u high=%u win=%u modulator=%u", 2634 s->src.seqlo, s->src.seqhi, 2635 s->src.max_win, s->src.seqdiff); 2636 if (s->src.wscale && s->dst.wscale) 2637 printf(" wscale=%u", 2638 s->src.wscale & PF_WSCALE_MASK); 2639 printf("]"); 2640 printf(" [lo=%u high=%u win=%u modulator=%u", 2641 s->dst.seqlo, s->dst.seqhi, 2642 s->dst.max_win, s->dst.seqdiff); 2643 if (s->src.wscale && s->dst.wscale) 2644 printf(" wscale=%u", 2645 s->dst.wscale & PF_WSCALE_MASK); 2646 printf("]"); 2647 } 2648 printf(" %u:%u", s->src.state, s->dst.state); 2649 } 2650 } 2651 2652 void 2653 pf_print_flags(u_int8_t f) 2654 { 2655 if (f) 2656 printf(" "); 2657 if (f & TH_FIN) 2658 printf("F"); 2659 if (f & TH_SYN) 2660 printf("S"); 2661 if (f & TH_RST) 2662 printf("R"); 2663 if (f & TH_PUSH) 2664 printf("P"); 2665 if (f & TH_ACK) 2666 printf("A"); 2667 if (f & TH_URG) 2668 printf("U"); 2669 if (f & TH_ECE) 2670 printf("E"); 2671 if (f & TH_CWR) 2672 printf("W"); 2673 } 2674 2675 #define PF_SET_SKIP_STEPS(i) \ 2676 do { \ 2677 while (head[i] != cur) { \ 2678 head[i]->skip[i].ptr = cur; \ 2679 head[i] = TAILQ_NEXT(head[i], entries); \ 2680 } \ 2681 } while (0) 2682 2683 void 2684 pf_calc_skip_steps(struct pf_krulequeue *rules) 2685 { 2686 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2687 int i; 2688 2689 cur = TAILQ_FIRST(rules); 2690 prev = cur; 2691 for (i = 0; i < PF_SKIP_COUNT; ++i) 2692 head[i] = cur; 2693 while (cur != NULL) { 2694 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2695 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2696 if (cur->direction != prev->direction) 2697 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2698 if (cur->af != prev->af) 2699 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2700 if (cur->proto != prev->proto) 2701 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2702 if (cur->src.neg != prev->src.neg || 2703 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2704 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2705 if (cur->src.port[0] != prev->src.port[0] || 2706 cur->src.port[1] != prev->src.port[1] || 2707 cur->src.port_op != prev->src.port_op) 2708 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2709 if (cur->dst.neg != prev->dst.neg || 2710 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2711 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2712 if (cur->dst.port[0] != prev->dst.port[0] || 2713 cur->dst.port[1] != prev->dst.port[1] || 2714 cur->dst.port_op != prev->dst.port_op) 2715 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2716 2717 prev = cur; 2718 cur = TAILQ_NEXT(cur, entries); 2719 } 2720 for (i = 0; i < PF_SKIP_COUNT; ++i) 2721 PF_SET_SKIP_STEPS(i); 2722 } 2723 2724 int 2725 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2726 { 2727 if (aw1->type != aw2->type) 2728 return (1); 2729 switch (aw1->type) { 2730 case PF_ADDR_ADDRMASK: 2731 case PF_ADDR_RANGE: 2732 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2733 return (1); 2734 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2735 return (1); 2736 return (0); 2737 case PF_ADDR_DYNIFTL: 2738 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2739 case PF_ADDR_NOROUTE: 2740 case PF_ADDR_URPFFAILED: 2741 return (0); 2742 case PF_ADDR_TABLE: 2743 return (aw1->p.tbl != aw2->p.tbl); 2744 default: 2745 printf("invalid address type: %d\n", aw1->type); 2746 return (1); 2747 } 2748 } 2749 2750 /** 2751 * Checksum updates are a little complicated because the checksum in the TCP/UDP 2752 * header isn't always a full checksum. In some cases (i.e. output) it's a 2753 * pseudo-header checksum, which is a partial checksum over src/dst IP 2754 * addresses, protocol number and length. 2755 * 2756 * That means we have the following cases: 2757 * * Input or forwarding: we don't have TSO, the checksum fields are full 2758 * checksums, we need to update the checksum whenever we change anything. 2759 * * Output (i.e. the checksum is a pseudo-header checksum): 2760 * x The field being updated is src/dst address or affects the length of 2761 * the packet. We need to update the pseudo-header checksum (note that this 2762 * checksum is not ones' complement). 2763 * x Some other field is being modified (e.g. src/dst port numbers): We 2764 * don't have to update anything. 2765 **/ 2766 u_int16_t 2767 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 2768 { 2769 u_int32_t x; 2770 2771 x = cksum + old - new; 2772 x = (x + (x >> 16)) & 0xffff; 2773 2774 /* optimise: eliminate a branch when not udp */ 2775 if (udp && cksum == 0x0000) 2776 return cksum; 2777 if (udp && x == 0x0000) 2778 x = 0xffff; 2779 2780 return (u_int16_t)(x); 2781 } 2782 2783 static void 2784 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 2785 u_int8_t udp) 2786 { 2787 u_int16_t old = htons(hi ? (*f << 8) : *f); 2788 u_int16_t new = htons(hi ? ( v << 8) : v); 2789 2790 if (*f == v) 2791 return; 2792 2793 *f = v; 2794 2795 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2796 return; 2797 2798 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 2799 } 2800 2801 void 2802 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 2803 bool hi, u_int8_t udp) 2804 { 2805 u_int8_t *fb = (u_int8_t *)f; 2806 u_int8_t *vb = (u_int8_t *)&v; 2807 2808 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2809 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2810 } 2811 2812 void 2813 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 2814 bool hi, u_int8_t udp) 2815 { 2816 u_int8_t *fb = (u_int8_t *)f; 2817 u_int8_t *vb = (u_int8_t *)&v; 2818 2819 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2820 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2821 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2822 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2823 } 2824 2825 u_int16_t 2826 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 2827 u_int16_t new, u_int8_t udp) 2828 { 2829 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2830 return (cksum); 2831 2832 return (pf_cksum_fixup(cksum, old, new, udp)); 2833 } 2834 2835 static void 2836 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 2837 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 2838 sa_family_t af) 2839 { 2840 struct pf_addr ao; 2841 u_int16_t po = *p; 2842 2843 PF_ACPY(&ao, a, af); 2844 PF_ACPY(a, an, af); 2845 2846 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2847 *pc = ~*pc; 2848 2849 *p = pn; 2850 2851 switch (af) { 2852 #ifdef INET 2853 case AF_INET: 2854 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2855 ao.addr16[0], an->addr16[0], 0), 2856 ao.addr16[1], an->addr16[1], 0); 2857 *p = pn; 2858 2859 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 2860 ao.addr16[0], an->addr16[0], u), 2861 ao.addr16[1], an->addr16[1], u); 2862 2863 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2864 break; 2865 #endif /* INET */ 2866 #ifdef INET6 2867 case AF_INET6: 2868 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2869 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2870 pf_cksum_fixup(pf_cksum_fixup(*pc, 2871 ao.addr16[0], an->addr16[0], u), 2872 ao.addr16[1], an->addr16[1], u), 2873 ao.addr16[2], an->addr16[2], u), 2874 ao.addr16[3], an->addr16[3], u), 2875 ao.addr16[4], an->addr16[4], u), 2876 ao.addr16[5], an->addr16[5], u), 2877 ao.addr16[6], an->addr16[6], u), 2878 ao.addr16[7], an->addr16[7], u); 2879 2880 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2881 break; 2882 #endif /* INET6 */ 2883 } 2884 2885 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 2886 CSUM_DELAY_DATA_IPV6)) { 2887 *pc = ~*pc; 2888 if (! *pc) 2889 *pc = 0xffff; 2890 } 2891 } 2892 2893 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2894 void 2895 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2896 { 2897 u_int32_t ao; 2898 2899 memcpy(&ao, a, sizeof(ao)); 2900 memcpy(a, &an, sizeof(u_int32_t)); 2901 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2902 ao % 65536, an % 65536, u); 2903 } 2904 2905 void 2906 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 2907 { 2908 u_int32_t ao; 2909 2910 memcpy(&ao, a, sizeof(ao)); 2911 memcpy(a, &an, sizeof(u_int32_t)); 2912 2913 *c = pf_proto_cksum_fixup(m, 2914 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 2915 ao % 65536, an % 65536, udp); 2916 } 2917 2918 #ifdef INET6 2919 static void 2920 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2921 { 2922 struct pf_addr ao; 2923 2924 PF_ACPY(&ao, a, AF_INET6); 2925 PF_ACPY(a, an, AF_INET6); 2926 2927 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2928 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2929 pf_cksum_fixup(pf_cksum_fixup(*c, 2930 ao.addr16[0], an->addr16[0], u), 2931 ao.addr16[1], an->addr16[1], u), 2932 ao.addr16[2], an->addr16[2], u), 2933 ao.addr16[3], an->addr16[3], u), 2934 ao.addr16[4], an->addr16[4], u), 2935 ao.addr16[5], an->addr16[5], u), 2936 ao.addr16[6], an->addr16[6], u), 2937 ao.addr16[7], an->addr16[7], u); 2938 } 2939 #endif /* INET6 */ 2940 2941 static void 2942 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2943 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2944 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2945 { 2946 struct pf_addr oia, ooa; 2947 2948 PF_ACPY(&oia, ia, af); 2949 if (oa) 2950 PF_ACPY(&ooa, oa, af); 2951 2952 /* Change inner protocol port, fix inner protocol checksum. */ 2953 if (ip != NULL) { 2954 u_int16_t oip = *ip; 2955 u_int32_t opc; 2956 2957 if (pc != NULL) 2958 opc = *pc; 2959 *ip = np; 2960 if (pc != NULL) 2961 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2962 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2963 if (pc != NULL) 2964 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2965 } 2966 /* Change inner ip address, fix inner ip and icmp checksums. */ 2967 PF_ACPY(ia, na, af); 2968 switch (af) { 2969 #ifdef INET 2970 case AF_INET: { 2971 u_int32_t oh2c = *h2c; 2972 2973 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2974 oia.addr16[0], ia->addr16[0], 0), 2975 oia.addr16[1], ia->addr16[1], 0); 2976 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2977 oia.addr16[0], ia->addr16[0], 0), 2978 oia.addr16[1], ia->addr16[1], 0); 2979 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2980 break; 2981 } 2982 #endif /* INET */ 2983 #ifdef INET6 2984 case AF_INET6: 2985 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2986 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2987 pf_cksum_fixup(pf_cksum_fixup(*ic, 2988 oia.addr16[0], ia->addr16[0], u), 2989 oia.addr16[1], ia->addr16[1], u), 2990 oia.addr16[2], ia->addr16[2], u), 2991 oia.addr16[3], ia->addr16[3], u), 2992 oia.addr16[4], ia->addr16[4], u), 2993 oia.addr16[5], ia->addr16[5], u), 2994 oia.addr16[6], ia->addr16[6], u), 2995 oia.addr16[7], ia->addr16[7], u); 2996 break; 2997 #endif /* INET6 */ 2998 } 2999 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3000 if (oa) { 3001 PF_ACPY(oa, na, af); 3002 switch (af) { 3003 #ifdef INET 3004 case AF_INET: 3005 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3006 ooa.addr16[0], oa->addr16[0], 0), 3007 ooa.addr16[1], oa->addr16[1], 0); 3008 break; 3009 #endif /* INET */ 3010 #ifdef INET6 3011 case AF_INET6: 3012 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3013 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3014 pf_cksum_fixup(pf_cksum_fixup(*ic, 3015 ooa.addr16[0], oa->addr16[0], u), 3016 ooa.addr16[1], oa->addr16[1], u), 3017 ooa.addr16[2], oa->addr16[2], u), 3018 ooa.addr16[3], oa->addr16[3], u), 3019 ooa.addr16[4], oa->addr16[4], u), 3020 ooa.addr16[5], oa->addr16[5], u), 3021 ooa.addr16[6], oa->addr16[6], u), 3022 ooa.addr16[7], oa->addr16[7], u); 3023 break; 3024 #endif /* INET6 */ 3025 } 3026 } 3027 } 3028 3029 /* 3030 * Need to modulate the sequence numbers in the TCP SACK option 3031 * (credits to Krzysztof Pfaff for report and patch) 3032 */ 3033 static int 3034 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 3035 struct tcphdr *th, struct pf_state_peer *dst) 3036 { 3037 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3038 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3039 int copyback = 0, i, olen; 3040 struct sackblk sack; 3041 3042 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3043 if (hlen < TCPOLEN_SACKLEN || 3044 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3045 return 0; 3046 3047 while (hlen >= TCPOLEN_SACKLEN) { 3048 size_t startoff = opt - opts; 3049 olen = opt[1]; 3050 switch (*opt) { 3051 case TCPOPT_EOL: /* FALLTHROUGH */ 3052 case TCPOPT_NOP: 3053 opt++; 3054 hlen--; 3055 break; 3056 case TCPOPT_SACK: 3057 if (olen > hlen) 3058 olen = hlen; 3059 if (olen >= TCPOLEN_SACKLEN) { 3060 for (i = 2; i + TCPOLEN_SACK <= olen; 3061 i += TCPOLEN_SACK) { 3062 memcpy(&sack, &opt[i], sizeof(sack)); 3063 pf_patch_32_unaligned(m, 3064 &th->th_sum, &sack.start, 3065 htonl(ntohl(sack.start) - dst->seqdiff), 3066 PF_ALGNMNT(startoff), 3067 0); 3068 pf_patch_32_unaligned(m, &th->th_sum, 3069 &sack.end, 3070 htonl(ntohl(sack.end) - dst->seqdiff), 3071 PF_ALGNMNT(startoff), 3072 0); 3073 memcpy(&opt[i], &sack, sizeof(sack)); 3074 } 3075 copyback = 1; 3076 } 3077 /* FALLTHROUGH */ 3078 default: 3079 if (olen < 2) 3080 olen = 2; 3081 hlen -= olen; 3082 opt += olen; 3083 } 3084 } 3085 3086 if (copyback) 3087 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 3088 return (copyback); 3089 } 3090 3091 struct mbuf * 3092 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3093 const struct pf_addr *saddr, const struct pf_addr *daddr, 3094 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3095 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3096 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3097 { 3098 struct mbuf *m; 3099 int len, tlen; 3100 #ifdef INET 3101 struct ip *h = NULL; 3102 #endif /* INET */ 3103 #ifdef INET6 3104 struct ip6_hdr *h6 = NULL; 3105 #endif /* INET6 */ 3106 struct tcphdr *th; 3107 char *opt; 3108 struct pf_mtag *pf_mtag; 3109 3110 len = 0; 3111 th = NULL; 3112 3113 /* maximum segment size tcp option */ 3114 tlen = sizeof(struct tcphdr); 3115 if (mss) 3116 tlen += 4; 3117 3118 switch (af) { 3119 #ifdef INET 3120 case AF_INET: 3121 len = sizeof(struct ip) + tlen; 3122 break; 3123 #endif /* INET */ 3124 #ifdef INET6 3125 case AF_INET6: 3126 len = sizeof(struct ip6_hdr) + tlen; 3127 break; 3128 #endif /* INET6 */ 3129 default: 3130 panic("%s: unsupported af %d", __func__, af); 3131 } 3132 3133 m = m_gethdr(M_NOWAIT, MT_DATA); 3134 if (m == NULL) 3135 return (NULL); 3136 3137 #ifdef MAC 3138 mac_netinet_firewall_send(m); 3139 #endif 3140 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 3141 m_freem(m); 3142 return (NULL); 3143 } 3144 if (skip_firewall) 3145 m->m_flags |= M_SKIP_FIREWALL; 3146 pf_mtag->tag = mtag_tag; 3147 pf_mtag->flags = mtag_flags; 3148 3149 if (rtableid >= 0) 3150 M_SETFIB(m, rtableid); 3151 3152 #ifdef ALTQ 3153 if (r != NULL && r->qid) { 3154 pf_mtag->qid = r->qid; 3155 3156 /* add hints for ecn */ 3157 pf_mtag->hdr = mtod(m, struct ip *); 3158 } 3159 #endif /* ALTQ */ 3160 m->m_data += max_linkhdr; 3161 m->m_pkthdr.len = m->m_len = len; 3162 /* The rest of the stack assumes a rcvif, so provide one. 3163 * This is a locally generated packet, so .. close enough. */ 3164 m->m_pkthdr.rcvif = V_loif; 3165 bzero(m->m_data, len); 3166 switch (af) { 3167 #ifdef INET 3168 case AF_INET: 3169 h = mtod(m, struct ip *); 3170 3171 /* IP header fields included in the TCP checksum */ 3172 h->ip_p = IPPROTO_TCP; 3173 h->ip_len = htons(tlen); 3174 h->ip_src.s_addr = saddr->v4.s_addr; 3175 h->ip_dst.s_addr = daddr->v4.s_addr; 3176 3177 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3178 break; 3179 #endif /* INET */ 3180 #ifdef INET6 3181 case AF_INET6: 3182 h6 = mtod(m, struct ip6_hdr *); 3183 3184 /* IP header fields included in the TCP checksum */ 3185 h6->ip6_nxt = IPPROTO_TCP; 3186 h6->ip6_plen = htons(tlen); 3187 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3188 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3189 3190 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3191 break; 3192 #endif /* INET6 */ 3193 } 3194 3195 /* TCP header */ 3196 th->th_sport = sport; 3197 th->th_dport = dport; 3198 th->th_seq = htonl(seq); 3199 th->th_ack = htonl(ack); 3200 th->th_off = tlen >> 2; 3201 th->th_flags = tcp_flags; 3202 th->th_win = htons(win); 3203 3204 if (mss) { 3205 opt = (char *)(th + 1); 3206 opt[0] = TCPOPT_MAXSEG; 3207 opt[1] = 4; 3208 HTONS(mss); 3209 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 3210 } 3211 3212 switch (af) { 3213 #ifdef INET 3214 case AF_INET: 3215 /* TCP checksum */ 3216 th->th_sum = in_cksum(m, len); 3217 3218 /* Finish the IP header */ 3219 h->ip_v = 4; 3220 h->ip_hl = sizeof(*h) >> 2; 3221 h->ip_tos = IPTOS_LOWDELAY; 3222 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 3223 h->ip_len = htons(len); 3224 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3225 h->ip_sum = 0; 3226 break; 3227 #endif /* INET */ 3228 #ifdef INET6 3229 case AF_INET6: 3230 /* TCP checksum */ 3231 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3232 sizeof(struct ip6_hdr), tlen); 3233 3234 h6->ip6_vfc |= IPV6_VERSION; 3235 h6->ip6_hlim = IPV6_DEFHLIM; 3236 break; 3237 #endif /* INET6 */ 3238 } 3239 3240 return (m); 3241 } 3242 3243 static void 3244 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3245 uint8_t ttl, int rtableid) 3246 { 3247 struct mbuf *m; 3248 #ifdef INET 3249 struct ip *h = NULL; 3250 #endif /* INET */ 3251 #ifdef INET6 3252 struct ip6_hdr *h6 = NULL; 3253 #endif /* INET6 */ 3254 struct sctphdr *hdr; 3255 struct sctp_chunkhdr *chunk; 3256 struct pf_send_entry *pfse; 3257 int off = 0; 3258 3259 MPASS(af == pd->af); 3260 3261 m = m_gethdr(M_NOWAIT, MT_DATA); 3262 if (m == NULL) 3263 return; 3264 3265 m->m_data += max_linkhdr; 3266 m->m_flags |= M_SKIP_FIREWALL; 3267 /* The rest of the stack assumes a rcvif, so provide one. 3268 * This is a locally generated packet, so .. close enough. */ 3269 m->m_pkthdr.rcvif = V_loif; 3270 3271 /* IPv4|6 header */ 3272 switch (af) { 3273 #ifdef INET 3274 case AF_INET: 3275 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3276 3277 h = mtod(m, struct ip *); 3278 3279 /* IP header fields included in the TCP checksum */ 3280 3281 h->ip_p = IPPROTO_SCTP; 3282 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3283 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3284 h->ip_src = pd->dst->v4; 3285 h->ip_dst = pd->src->v4; 3286 3287 off += sizeof(struct ip); 3288 break; 3289 #endif /* INET */ 3290 #ifdef INET6 3291 case AF_INET6: 3292 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3293 3294 h6 = mtod(m, struct ip6_hdr *); 3295 3296 /* IP header fields included in the TCP checksum */ 3297 h6->ip6_vfc |= IPV6_VERSION; 3298 h6->ip6_nxt = IPPROTO_SCTP; 3299 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3300 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3301 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3302 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3303 3304 off += sizeof(struct ip6_hdr); 3305 break; 3306 #endif /* INET6 */ 3307 } 3308 3309 /* SCTP header */ 3310 hdr = mtodo(m, off); 3311 3312 hdr->src_port = pd->hdr.sctp.dest_port; 3313 hdr->dest_port = pd->hdr.sctp.src_port; 3314 hdr->v_tag = pd->sctp_initiate_tag; 3315 hdr->checksum = 0; 3316 3317 /* Abort chunk. */ 3318 off += sizeof(struct sctphdr); 3319 chunk = mtodo(m, off); 3320 3321 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3322 chunk->chunk_length = htons(sizeof(*chunk)); 3323 3324 /* SCTP checksum */ 3325 off += sizeof(*chunk); 3326 m->m_pkthdr.len = m->m_len = off; 3327 3328 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 3329 3330 if (rtableid >= 0) 3331 M_SETFIB(m, rtableid); 3332 3333 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3334 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3335 if (pfse == NULL) { 3336 m_freem(m); 3337 return; 3338 } 3339 3340 switch (af) { 3341 #ifdef INET 3342 case AF_INET: 3343 pfse->pfse_type = PFSE_IP; 3344 break; 3345 #endif /* INET */ 3346 #ifdef INET6 3347 case AF_INET6: 3348 pfse->pfse_type = PFSE_IP6; 3349 break; 3350 #endif /* INET6 */ 3351 } 3352 3353 pfse->pfse_m = m; 3354 pf_send(pfse); 3355 } 3356 3357 void 3358 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3359 const struct pf_addr *saddr, const struct pf_addr *daddr, 3360 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3361 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3362 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3363 { 3364 struct pf_send_entry *pfse; 3365 struct mbuf *m; 3366 3367 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3368 win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid); 3369 if (m == NULL) 3370 return; 3371 3372 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3373 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3374 if (pfse == NULL) { 3375 m_freem(m); 3376 return; 3377 } 3378 3379 switch (af) { 3380 #ifdef INET 3381 case AF_INET: 3382 pfse->pfse_type = PFSE_IP; 3383 break; 3384 #endif /* INET */ 3385 #ifdef INET6 3386 case AF_INET6: 3387 pfse->pfse_type = PFSE_IP6; 3388 break; 3389 #endif /* INET6 */ 3390 } 3391 3392 pfse->pfse_m = m; 3393 pf_send(pfse); 3394 } 3395 3396 static void 3397 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3398 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 3399 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 3400 u_short *reason, int rtableid) 3401 { 3402 struct pf_addr * const saddr = pd->src; 3403 struct pf_addr * const daddr = pd->dst; 3404 sa_family_t af = pd->af; 3405 3406 /* undo NAT changes, if they have taken place */ 3407 if (nr != NULL) { 3408 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3409 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3410 if (pd->sport) 3411 *pd->sport = sk->port[pd->sidx]; 3412 if (pd->dport) 3413 *pd->dport = sk->port[pd->didx]; 3414 if (pd->proto_sum) 3415 *pd->proto_sum = bproto_sum; 3416 if (pd->ip_sum) 3417 *pd->ip_sum = bip_sum; 3418 m_copyback(m, off, hdrlen, pd->hdr.any); 3419 } 3420 if (pd->proto == IPPROTO_TCP && 3421 ((r->rule_flag & PFRULE_RETURNRST) || 3422 (r->rule_flag & PFRULE_RETURN)) && 3423 !(th->th_flags & TH_RST)) { 3424 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3425 int len = 0; 3426 #ifdef INET 3427 struct ip *h4; 3428 #endif 3429 #ifdef INET6 3430 struct ip6_hdr *h6; 3431 #endif 3432 3433 switch (af) { 3434 #ifdef INET 3435 case AF_INET: 3436 h4 = mtod(m, struct ip *); 3437 len = ntohs(h4->ip_len) - off; 3438 break; 3439 #endif 3440 #ifdef INET6 3441 case AF_INET6: 3442 h6 = mtod(m, struct ip6_hdr *); 3443 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3444 break; 3445 #endif 3446 } 3447 3448 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3449 REASON_SET(reason, PFRES_PROTCKSUM); 3450 else { 3451 if (th->th_flags & TH_SYN) 3452 ack++; 3453 if (th->th_flags & TH_FIN) 3454 ack++; 3455 pf_send_tcp(r, af, pd->dst, 3456 pd->src, th->th_dport, th->th_sport, 3457 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3458 r->return_ttl, true, 0, 0, rtableid); 3459 } 3460 } else if (pd->proto == IPPROTO_SCTP && 3461 (r->rule_flag & PFRULE_RETURN)) { 3462 pf_send_sctp_abort(af, pd, r->return_ttl, rtableid); 3463 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3464 r->return_icmp) 3465 pf_send_icmp(m, r->return_icmp >> 8, 3466 r->return_icmp & 255, af, r, rtableid); 3467 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3468 r->return_icmp6) 3469 pf_send_icmp(m, r->return_icmp6 >> 8, 3470 r->return_icmp6 & 255, af, r, rtableid); 3471 } 3472 3473 static int 3474 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3475 { 3476 struct m_tag *mtag; 3477 u_int8_t mpcp; 3478 3479 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3480 if (mtag == NULL) 3481 return (0); 3482 3483 if (prio == PF_PRIO_ZERO) 3484 prio = 0; 3485 3486 mpcp = *(uint8_t *)(mtag + 1); 3487 3488 return (mpcp == prio); 3489 } 3490 3491 static int 3492 pf_icmp_to_bandlim(uint8_t type) 3493 { 3494 switch (type) { 3495 case ICMP_ECHO: 3496 case ICMP_ECHOREPLY: 3497 return (BANDLIM_ICMP_ECHO); 3498 case ICMP_TSTAMP: 3499 case ICMP_TSTAMPREPLY: 3500 return (BANDLIM_ICMP_TSTAMP); 3501 case ICMP_UNREACH: 3502 default: 3503 return (BANDLIM_ICMP_UNREACH); 3504 } 3505 } 3506 3507 static void 3508 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3509 struct pf_krule *r, int rtableid) 3510 { 3511 struct pf_send_entry *pfse; 3512 struct mbuf *m0; 3513 struct pf_mtag *pf_mtag; 3514 3515 /* ICMP packet rate limitation. */ 3516 #ifdef INET6 3517 if (af == AF_INET6) { 3518 if (icmp6_ratelimit(NULL, type, code)) 3519 return; 3520 } 3521 #endif 3522 #ifdef INET 3523 if (af == AF_INET) { 3524 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3525 return; 3526 } 3527 #endif 3528 3529 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3530 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3531 if (pfse == NULL) 3532 return; 3533 3534 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3535 free(pfse, M_PFTEMP); 3536 return; 3537 } 3538 3539 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3540 free(pfse, M_PFTEMP); 3541 return; 3542 } 3543 /* XXX: revisit */ 3544 m0->m_flags |= M_SKIP_FIREWALL; 3545 3546 if (rtableid >= 0) 3547 M_SETFIB(m0, rtableid); 3548 3549 #ifdef ALTQ 3550 if (r->qid) { 3551 pf_mtag->qid = r->qid; 3552 /* add hints for ecn */ 3553 pf_mtag->hdr = mtod(m0, struct ip *); 3554 } 3555 #endif /* ALTQ */ 3556 3557 switch (af) { 3558 #ifdef INET 3559 case AF_INET: 3560 pfse->pfse_type = PFSE_ICMP; 3561 break; 3562 #endif /* INET */ 3563 #ifdef INET6 3564 case AF_INET6: 3565 pfse->pfse_type = PFSE_ICMP6; 3566 break; 3567 #endif /* INET6 */ 3568 } 3569 pfse->pfse_m = m0; 3570 pfse->icmpopts.type = type; 3571 pfse->icmpopts.code = code; 3572 pf_send(pfse); 3573 } 3574 3575 /* 3576 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3577 * If n is 0, they match if they are equal. If n is != 0, they match if they 3578 * are different. 3579 */ 3580 int 3581 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3582 struct pf_addr *b, sa_family_t af) 3583 { 3584 int match = 0; 3585 3586 switch (af) { 3587 #ifdef INET 3588 case AF_INET: 3589 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 3590 match++; 3591 break; 3592 #endif /* INET */ 3593 #ifdef INET6 3594 case AF_INET6: 3595 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 3596 match++; 3597 break; 3598 #endif /* INET6 */ 3599 } 3600 if (match) { 3601 if (n) 3602 return (0); 3603 else 3604 return (1); 3605 } else { 3606 if (n) 3607 return (1); 3608 else 3609 return (0); 3610 } 3611 } 3612 3613 /* 3614 * Return 1 if b <= a <= e, otherwise return 0. 3615 */ 3616 int 3617 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3618 struct pf_addr *a, sa_family_t af) 3619 { 3620 switch (af) { 3621 #ifdef INET 3622 case AF_INET: 3623 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3624 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3625 return (0); 3626 break; 3627 #endif /* INET */ 3628 #ifdef INET6 3629 case AF_INET6: { 3630 int i; 3631 3632 /* check a >= b */ 3633 for (i = 0; i < 4; ++i) 3634 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3635 break; 3636 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3637 return (0); 3638 /* check a <= e */ 3639 for (i = 0; i < 4; ++i) 3640 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3641 break; 3642 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3643 return (0); 3644 break; 3645 } 3646 #endif /* INET6 */ 3647 } 3648 return (1); 3649 } 3650 3651 static int 3652 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3653 { 3654 switch (op) { 3655 case PF_OP_IRG: 3656 return ((p > a1) && (p < a2)); 3657 case PF_OP_XRG: 3658 return ((p < a1) || (p > a2)); 3659 case PF_OP_RRG: 3660 return ((p >= a1) && (p <= a2)); 3661 case PF_OP_EQ: 3662 return (p == a1); 3663 case PF_OP_NE: 3664 return (p != a1); 3665 case PF_OP_LT: 3666 return (p < a1); 3667 case PF_OP_LE: 3668 return (p <= a1); 3669 case PF_OP_GT: 3670 return (p > a1); 3671 case PF_OP_GE: 3672 return (p >= a1); 3673 } 3674 return (0); /* never reached */ 3675 } 3676 3677 int 3678 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3679 { 3680 NTOHS(a1); 3681 NTOHS(a2); 3682 NTOHS(p); 3683 return (pf_match(op, a1, a2, p)); 3684 } 3685 3686 static int 3687 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3688 { 3689 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3690 return (0); 3691 return (pf_match(op, a1, a2, u)); 3692 } 3693 3694 static int 3695 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3696 { 3697 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3698 return (0); 3699 return (pf_match(op, a1, a2, g)); 3700 } 3701 3702 int 3703 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3704 { 3705 if (*tag == -1) 3706 *tag = mtag; 3707 3708 return ((!r->match_tag_not && r->match_tag == *tag) || 3709 (r->match_tag_not && r->match_tag != *tag)); 3710 } 3711 3712 int 3713 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3714 { 3715 3716 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3717 3718 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3719 return (ENOMEM); 3720 3721 pd->pf_mtag->tag = tag; 3722 3723 return (0); 3724 } 3725 3726 #define PF_ANCHOR_STACKSIZE 32 3727 struct pf_kanchor_stackframe { 3728 struct pf_kruleset *rs; 3729 struct pf_krule *r; /* XXX: + match bit */ 3730 struct pf_kanchor *child; 3731 }; 3732 3733 /* 3734 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3735 */ 3736 #define PF_ANCHORSTACK_MATCH 0x00000001 3737 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 3738 3739 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3740 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 3741 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3742 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3743 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3744 } while (0) 3745 3746 void 3747 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3748 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3749 int *match) 3750 { 3751 struct pf_kanchor_stackframe *f; 3752 3753 PF_RULES_RASSERT(); 3754 3755 if (match) 3756 *match = 0; 3757 if (*depth >= PF_ANCHOR_STACKSIZE) { 3758 printf("%s: anchor stack overflow on %s\n", 3759 __func__, (*r)->anchor->name); 3760 *r = TAILQ_NEXT(*r, entries); 3761 return; 3762 } else if (*depth == 0 && a != NULL) 3763 *a = *r; 3764 f = stack + (*depth)++; 3765 f->rs = *rs; 3766 f->r = *r; 3767 if ((*r)->anchor_wildcard) { 3768 struct pf_kanchor_node *parent = &(*r)->anchor->children; 3769 3770 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 3771 *r = NULL; 3772 return; 3773 } 3774 *rs = &f->child->ruleset; 3775 } else { 3776 f->child = NULL; 3777 *rs = &(*r)->anchor->ruleset; 3778 } 3779 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3780 } 3781 3782 int 3783 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3784 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3785 int *match) 3786 { 3787 struct pf_kanchor_stackframe *f; 3788 struct pf_krule *fr; 3789 int quick = 0; 3790 3791 PF_RULES_RASSERT(); 3792 3793 do { 3794 if (*depth <= 0) 3795 break; 3796 f = stack + *depth - 1; 3797 fr = PF_ANCHOR_RULE(f); 3798 if (f->child != NULL) { 3799 /* 3800 * This block traverses through 3801 * a wildcard anchor. 3802 */ 3803 if (match != NULL && *match) { 3804 /* 3805 * If any of "*" matched, then 3806 * "foo/ *" matched, mark frame 3807 * appropriately. 3808 */ 3809 PF_ANCHOR_SET_MATCH(f); 3810 *match = 0; 3811 } 3812 f->child = RB_NEXT(pf_kanchor_node, 3813 &fr->anchor->children, f->child); 3814 if (f->child != NULL) { 3815 *rs = &f->child->ruleset; 3816 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3817 if (*r == NULL) 3818 continue; 3819 else 3820 break; 3821 } 3822 } 3823 (*depth)--; 3824 if (*depth == 0 && a != NULL) 3825 *a = NULL; 3826 *rs = f->rs; 3827 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 3828 quick = fr->quick; 3829 *r = TAILQ_NEXT(fr, entries); 3830 } while (*r == NULL); 3831 3832 return (quick); 3833 } 3834 3835 struct pf_keth_anchor_stackframe { 3836 struct pf_keth_ruleset *rs; 3837 struct pf_keth_rule *r; /* XXX: + match bit */ 3838 struct pf_keth_anchor *child; 3839 }; 3840 3841 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3842 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 3843 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3844 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3845 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3846 } while (0) 3847 3848 void 3849 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3850 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3851 struct pf_keth_rule **a, int *match) 3852 { 3853 struct pf_keth_anchor_stackframe *f; 3854 3855 NET_EPOCH_ASSERT(); 3856 3857 if (match) 3858 *match = 0; 3859 if (*depth >= PF_ANCHOR_STACKSIZE) { 3860 printf("%s: anchor stack overflow on %s\n", 3861 __func__, (*r)->anchor->name); 3862 *r = TAILQ_NEXT(*r, entries); 3863 return; 3864 } else if (*depth == 0 && a != NULL) 3865 *a = *r; 3866 f = stack + (*depth)++; 3867 f->rs = *rs; 3868 f->r = *r; 3869 if ((*r)->anchor_wildcard) { 3870 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 3871 3872 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 3873 *r = NULL; 3874 return; 3875 } 3876 *rs = &f->child->ruleset; 3877 } else { 3878 f->child = NULL; 3879 *rs = &(*r)->anchor->ruleset; 3880 } 3881 *r = TAILQ_FIRST((*rs)->active.rules); 3882 } 3883 3884 int 3885 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3886 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3887 struct pf_keth_rule **a, int *match) 3888 { 3889 struct pf_keth_anchor_stackframe *f; 3890 struct pf_keth_rule *fr; 3891 int quick = 0; 3892 3893 NET_EPOCH_ASSERT(); 3894 3895 do { 3896 if (*depth <= 0) 3897 break; 3898 f = stack + *depth - 1; 3899 fr = PF_ETH_ANCHOR_RULE(f); 3900 if (f->child != NULL) { 3901 /* 3902 * This block traverses through 3903 * a wildcard anchor. 3904 */ 3905 if (match != NULL && *match) { 3906 /* 3907 * If any of "*" matched, then 3908 * "foo/ *" matched, mark frame 3909 * appropriately. 3910 */ 3911 PF_ETH_ANCHOR_SET_MATCH(f); 3912 *match = 0; 3913 } 3914 f->child = RB_NEXT(pf_keth_anchor_node, 3915 &fr->anchor->children, f->child); 3916 if (f->child != NULL) { 3917 *rs = &f->child->ruleset; 3918 *r = TAILQ_FIRST((*rs)->active.rules); 3919 if (*r == NULL) 3920 continue; 3921 else 3922 break; 3923 } 3924 } 3925 (*depth)--; 3926 if (*depth == 0 && a != NULL) 3927 *a = NULL; 3928 *rs = f->rs; 3929 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 3930 quick = fr->quick; 3931 *r = TAILQ_NEXT(fr, entries); 3932 } while (*r == NULL); 3933 3934 return (quick); 3935 } 3936 3937 #ifdef INET6 3938 void 3939 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3940 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3941 { 3942 switch (af) { 3943 #ifdef INET 3944 case AF_INET: 3945 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3946 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3947 break; 3948 #endif /* INET */ 3949 case AF_INET6: 3950 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3951 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3952 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3953 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3954 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3955 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3956 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3957 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3958 break; 3959 } 3960 } 3961 3962 void 3963 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3964 { 3965 switch (af) { 3966 #ifdef INET 3967 case AF_INET: 3968 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3969 break; 3970 #endif /* INET */ 3971 case AF_INET6: 3972 if (addr->addr32[3] == 0xffffffff) { 3973 addr->addr32[3] = 0; 3974 if (addr->addr32[2] == 0xffffffff) { 3975 addr->addr32[2] = 0; 3976 if (addr->addr32[1] == 0xffffffff) { 3977 addr->addr32[1] = 0; 3978 addr->addr32[0] = 3979 htonl(ntohl(addr->addr32[0]) + 1); 3980 } else 3981 addr->addr32[1] = 3982 htonl(ntohl(addr->addr32[1]) + 1); 3983 } else 3984 addr->addr32[2] = 3985 htonl(ntohl(addr->addr32[2]) + 1); 3986 } else 3987 addr->addr32[3] = 3988 htonl(ntohl(addr->addr32[3]) + 1); 3989 break; 3990 } 3991 } 3992 #endif /* INET6 */ 3993 3994 void 3995 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 3996 { 3997 /* 3998 * Modern rules use the same flags in rules as they do in states. 3999 */ 4000 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4001 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4002 4003 /* 4004 * Old-style scrub rules have different flags which need to be translated. 4005 */ 4006 if (r->rule_flag & PFRULE_RANDOMID) 4007 a->flags |= PFSTATE_RANDOMID; 4008 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4009 a->flags |= PFSTATE_SETTOS; 4010 a->set_tos = r->set_tos; 4011 } 4012 4013 if (r->qid) 4014 a->qid = r->qid; 4015 if (r->pqid) 4016 a->pqid = r->pqid; 4017 if (r->rtableid >= 0) 4018 a->rtableid = r->rtableid; 4019 a->log |= r->log; 4020 if (r->min_ttl) 4021 a->min_ttl = r->min_ttl; 4022 if (r->max_mss) 4023 a->max_mss = r->max_mss; 4024 if (r->dnpipe) 4025 a->dnpipe = r->dnpipe; 4026 if (r->dnrpipe) 4027 a->dnrpipe = r->dnrpipe; 4028 if (r->dnpipe || r->dnrpipe) { 4029 if (r->free_flags & PFRULE_DN_IS_PIPE) 4030 a->flags |= PFSTATE_DN_IS_PIPE; 4031 else 4032 a->flags &= ~PFSTATE_DN_IS_PIPE; 4033 } 4034 if (r->scrub_flags & PFSTATE_SETPRIO) { 4035 a->set_prio[0] = r->set_prio[0]; 4036 a->set_prio[1] = r->set_prio[1]; 4037 } 4038 } 4039 4040 int 4041 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m) 4042 { 4043 struct pf_addr *saddr, *daddr; 4044 u_int16_t sport, dport; 4045 struct inpcbinfo *pi; 4046 struct inpcb *inp; 4047 4048 pd->lookup.uid = UID_MAX; 4049 pd->lookup.gid = GID_MAX; 4050 4051 switch (pd->proto) { 4052 case IPPROTO_TCP: 4053 sport = pd->hdr.tcp.th_sport; 4054 dport = pd->hdr.tcp.th_dport; 4055 pi = &V_tcbinfo; 4056 break; 4057 case IPPROTO_UDP: 4058 sport = pd->hdr.udp.uh_sport; 4059 dport = pd->hdr.udp.uh_dport; 4060 pi = &V_udbinfo; 4061 break; 4062 default: 4063 return (-1); 4064 } 4065 if (pd->dir == PF_IN) { 4066 saddr = pd->src; 4067 daddr = pd->dst; 4068 } else { 4069 u_int16_t p; 4070 4071 p = sport; 4072 sport = dport; 4073 dport = p; 4074 saddr = pd->dst; 4075 daddr = pd->src; 4076 } 4077 switch (pd->af) { 4078 #ifdef INET 4079 case AF_INET: 4080 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4081 dport, INPLOOKUP_RLOCKPCB, NULL, m); 4082 if (inp == NULL) { 4083 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4084 daddr->v4, dport, INPLOOKUP_WILDCARD | 4085 INPLOOKUP_RLOCKPCB, NULL, m); 4086 if (inp == NULL) 4087 return (-1); 4088 } 4089 break; 4090 #endif /* INET */ 4091 #ifdef INET6 4092 case AF_INET6: 4093 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4094 dport, INPLOOKUP_RLOCKPCB, NULL, m); 4095 if (inp == NULL) { 4096 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4097 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4098 INPLOOKUP_RLOCKPCB, NULL, m); 4099 if (inp == NULL) 4100 return (-1); 4101 } 4102 break; 4103 #endif /* INET6 */ 4104 4105 default: 4106 return (-1); 4107 } 4108 INP_RLOCK_ASSERT(inp); 4109 pd->lookup.uid = inp->inp_cred->cr_uid; 4110 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4111 INP_RUNLOCK(inp); 4112 4113 return (1); 4114 } 4115 4116 u_int8_t 4117 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 4118 { 4119 int hlen; 4120 u_int8_t hdr[60]; 4121 u_int8_t *opt, optlen; 4122 u_int8_t wscale = 0; 4123 4124 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 4125 if (hlen <= sizeof(struct tcphdr)) 4126 return (0); 4127 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 4128 return (0); 4129 opt = hdr + sizeof(struct tcphdr); 4130 hlen -= sizeof(struct tcphdr); 4131 while (hlen >= 3) { 4132 switch (*opt) { 4133 case TCPOPT_EOL: 4134 case TCPOPT_NOP: 4135 ++opt; 4136 --hlen; 4137 break; 4138 case TCPOPT_WINDOW: 4139 wscale = opt[2]; 4140 if (wscale > TCP_MAX_WINSHIFT) 4141 wscale = TCP_MAX_WINSHIFT; 4142 wscale |= PF_WSCALE_FLAG; 4143 /* FALLTHROUGH */ 4144 default: 4145 optlen = opt[1]; 4146 if (optlen < 2) 4147 optlen = 2; 4148 hlen -= optlen; 4149 opt += optlen; 4150 break; 4151 } 4152 } 4153 return (wscale); 4154 } 4155 4156 u_int16_t 4157 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 4158 { 4159 int hlen; 4160 u_int8_t hdr[60]; 4161 u_int8_t *opt, optlen; 4162 u_int16_t mss = V_tcp_mssdflt; 4163 4164 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 4165 if (hlen <= sizeof(struct tcphdr)) 4166 return (0); 4167 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 4168 return (0); 4169 opt = hdr + sizeof(struct tcphdr); 4170 hlen -= sizeof(struct tcphdr); 4171 while (hlen >= TCPOLEN_MAXSEG) { 4172 switch (*opt) { 4173 case TCPOPT_EOL: 4174 case TCPOPT_NOP: 4175 ++opt; 4176 --hlen; 4177 break; 4178 case TCPOPT_MAXSEG: 4179 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 4180 NTOHS(mss); 4181 /* FALLTHROUGH */ 4182 default: 4183 optlen = opt[1]; 4184 if (optlen < 2) 4185 optlen = 2; 4186 hlen -= optlen; 4187 opt += optlen; 4188 break; 4189 } 4190 } 4191 return (mss); 4192 } 4193 4194 static u_int16_t 4195 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 4196 { 4197 struct nhop_object *nh; 4198 #ifdef INET6 4199 struct in6_addr dst6; 4200 uint32_t scopeid; 4201 #endif /* INET6 */ 4202 int hlen = 0; 4203 uint16_t mss = 0; 4204 4205 NET_EPOCH_ASSERT(); 4206 4207 switch (af) { 4208 #ifdef INET 4209 case AF_INET: 4210 hlen = sizeof(struct ip); 4211 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 4212 if (nh != NULL) 4213 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4214 break; 4215 #endif /* INET */ 4216 #ifdef INET6 4217 case AF_INET6: 4218 hlen = sizeof(struct ip6_hdr); 4219 in6_splitscope(&addr->v6, &dst6, &scopeid); 4220 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4221 if (nh != NULL) 4222 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4223 break; 4224 #endif /* INET6 */ 4225 } 4226 4227 mss = max(V_tcp_mssdflt, mss); 4228 mss = min(mss, offer); 4229 mss = max(mss, 64); /* sanity - at least max opt space */ 4230 return (mss); 4231 } 4232 4233 static u_int32_t 4234 pf_tcp_iss(struct pf_pdesc *pd) 4235 { 4236 MD5_CTX ctx; 4237 u_int32_t digest[4]; 4238 4239 if (V_pf_tcp_secret_init == 0) { 4240 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4241 MD5Init(&V_pf_tcp_secret_ctx); 4242 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4243 sizeof(V_pf_tcp_secret)); 4244 V_pf_tcp_secret_init = 1; 4245 } 4246 4247 ctx = V_pf_tcp_secret_ctx; 4248 4249 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4250 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4251 if (pd->af == AF_INET6) { 4252 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4253 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4254 } else { 4255 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4256 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4257 } 4258 MD5Final((u_char *)digest, &ctx); 4259 V_pf_tcp_iss_off += 4096; 4260 #define ISN_RANDOM_INCREMENT (4096 - 1) 4261 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4262 V_pf_tcp_iss_off); 4263 #undef ISN_RANDOM_INCREMENT 4264 } 4265 4266 static bool 4267 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4268 { 4269 bool match = true; 4270 4271 /* Always matches if not set */ 4272 if (! r->isset) 4273 return (!r->neg); 4274 4275 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4276 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4277 match = false; 4278 break; 4279 } 4280 } 4281 4282 return (match ^ r->neg); 4283 } 4284 4285 static int 4286 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4287 { 4288 if (*tag == -1) 4289 *tag = mtag; 4290 4291 return ((!r->match_tag_not && r->match_tag == *tag) || 4292 (r->match_tag_not && r->match_tag != *tag)); 4293 } 4294 4295 static void 4296 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4297 { 4298 /* If we don't have the interface drop the packet. */ 4299 if (ifp == NULL) { 4300 m_freem(m); 4301 return; 4302 } 4303 4304 switch (ifp->if_type) { 4305 case IFT_ETHER: 4306 case IFT_XETHER: 4307 case IFT_L2VLAN: 4308 case IFT_BRIDGE: 4309 case IFT_IEEE8023ADLAG: 4310 break; 4311 default: 4312 m_freem(m); 4313 return; 4314 } 4315 4316 ifp->if_transmit(ifp, m); 4317 } 4318 4319 static int 4320 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4321 { 4322 #ifdef INET 4323 struct ip ip; 4324 #endif 4325 #ifdef INET6 4326 struct ip6_hdr ip6; 4327 #endif 4328 struct mbuf *m = *m0; 4329 struct ether_header *e; 4330 struct pf_keth_rule *r, *rm, *a = NULL; 4331 struct pf_keth_ruleset *ruleset = NULL; 4332 struct pf_mtag *mtag; 4333 struct pf_keth_ruleq *rules; 4334 struct pf_addr *src = NULL, *dst = NULL; 4335 struct pfi_kkif *bridge_to; 4336 sa_family_t af = 0; 4337 uint16_t proto; 4338 int asd = 0, match = 0; 4339 int tag = -1; 4340 uint8_t action; 4341 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4342 4343 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4344 NET_EPOCH_ASSERT(); 4345 4346 PF_RULES_RLOCK_TRACKER; 4347 4348 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4349 4350 mtag = pf_find_mtag(m); 4351 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4352 /* Dummynet re-injects packets after they've 4353 * completed their delay. We've already 4354 * processed them, so pass unconditionally. */ 4355 4356 /* But only once. We may see the packet multiple times (e.g. 4357 * PFIL_IN/PFIL_OUT). */ 4358 pf_dummynet_flag_remove(m, mtag); 4359 4360 return (PF_PASS); 4361 } 4362 4363 ruleset = V_pf_keth; 4364 rules = ck_pr_load_ptr(&ruleset->active.rules); 4365 r = TAILQ_FIRST(rules); 4366 rm = NULL; 4367 4368 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 4369 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 4370 DPFPRINTF(PF_DEBUG_URGENT, 4371 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 4372 ", pullup failed\n")); 4373 return (PF_DROP); 4374 } 4375 e = mtod(m, struct ether_header *); 4376 proto = ntohs(e->ether_type); 4377 4378 switch (proto) { 4379 #ifdef INET 4380 case ETHERTYPE_IP: { 4381 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4382 sizeof(ip))) 4383 return (PF_DROP); 4384 4385 af = AF_INET; 4386 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4387 (caddr_t)&ip); 4388 src = (struct pf_addr *)&ip.ip_src; 4389 dst = (struct pf_addr *)&ip.ip_dst; 4390 break; 4391 } 4392 #endif /* INET */ 4393 #ifdef INET6 4394 case ETHERTYPE_IPV6: { 4395 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4396 sizeof(ip6))) 4397 return (PF_DROP); 4398 4399 af = AF_INET6; 4400 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4401 (caddr_t)&ip6); 4402 src = (struct pf_addr *)&ip6.ip6_src; 4403 dst = (struct pf_addr *)&ip6.ip6_dst; 4404 break; 4405 } 4406 #endif /* INET6 */ 4407 } 4408 4409 PF_RULES_RLOCK(); 4410 4411 while (r != NULL) { 4412 counter_u64_add(r->evaluations, 1); 4413 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4414 4415 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4416 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4417 "kif"); 4418 r = r->skip[PFE_SKIP_IFP].ptr; 4419 } 4420 else if (r->direction && r->direction != dir) { 4421 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4422 "dir"); 4423 r = r->skip[PFE_SKIP_DIR].ptr; 4424 } 4425 else if (r->proto && r->proto != proto) { 4426 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4427 "proto"); 4428 r = r->skip[PFE_SKIP_PROTO].ptr; 4429 } 4430 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4431 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4432 "src"); 4433 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4434 } 4435 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4436 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4437 "dst"); 4438 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4439 } 4440 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4441 r->ipsrc.neg, kif, M_GETFIB(m))) { 4442 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4443 "ip_src"); 4444 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4445 } 4446 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4447 r->ipdst.neg, kif, M_GETFIB(m))) { 4448 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4449 "ip_dst"); 4450 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4451 } 4452 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4453 mtag ? mtag->tag : 0)) { 4454 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4455 "match_tag"); 4456 r = TAILQ_NEXT(r, entries); 4457 } 4458 else { 4459 if (r->tag) 4460 tag = r->tag; 4461 if (r->anchor == NULL) { 4462 /* Rule matches */ 4463 rm = r; 4464 4465 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4466 4467 if (r->quick) 4468 break; 4469 4470 r = TAILQ_NEXT(r, entries); 4471 } else { 4472 pf_step_into_keth_anchor(anchor_stack, &asd, 4473 &ruleset, &r, &a, &match); 4474 } 4475 } 4476 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4477 &ruleset, &r, &a, &match)) 4478 break; 4479 } 4480 4481 r = rm; 4482 4483 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4484 4485 /* Default to pass. */ 4486 if (r == NULL) { 4487 PF_RULES_RUNLOCK(); 4488 return (PF_PASS); 4489 } 4490 4491 /* Execute action. */ 4492 counter_u64_add(r->packets[dir == PF_OUT], 1); 4493 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4494 pf_update_timestamp(r); 4495 4496 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4497 if (r->action == PF_DROP) { 4498 PF_RULES_RUNLOCK(); 4499 return (PF_DROP); 4500 } 4501 4502 if (tag > 0) { 4503 if (mtag == NULL) 4504 mtag = pf_get_mtag(m); 4505 if (mtag == NULL) { 4506 PF_RULES_RUNLOCK(); 4507 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4508 return (PF_DROP); 4509 } 4510 mtag->tag = tag; 4511 } 4512 4513 if (r->qid != 0) { 4514 if (mtag == NULL) 4515 mtag = pf_get_mtag(m); 4516 if (mtag == NULL) { 4517 PF_RULES_RUNLOCK(); 4518 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4519 return (PF_DROP); 4520 } 4521 mtag->qid = r->qid; 4522 } 4523 4524 action = r->action; 4525 bridge_to = r->bridge_to; 4526 4527 /* Dummynet */ 4528 if (r->dnpipe) { 4529 struct ip_fw_args dnflow; 4530 4531 /* Drop packet if dummynet is not loaded. */ 4532 if (ip_dn_io_ptr == NULL) { 4533 PF_RULES_RUNLOCK(); 4534 m_freem(m); 4535 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4536 return (PF_DROP); 4537 } 4538 if (mtag == NULL) 4539 mtag = pf_get_mtag(m); 4540 if (mtag == NULL) { 4541 PF_RULES_RUNLOCK(); 4542 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4543 return (PF_DROP); 4544 } 4545 4546 bzero(&dnflow, sizeof(dnflow)); 4547 4548 /* We don't have port numbers here, so we set 0. That means 4549 * that we'll be somewhat limited in distinguishing flows (i.e. 4550 * only based on IP addresses, not based on port numbers), but 4551 * it's better than nothing. */ 4552 dnflow.f_id.dst_port = 0; 4553 dnflow.f_id.src_port = 0; 4554 dnflow.f_id.proto = 0; 4555 4556 dnflow.rule.info = r->dnpipe; 4557 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4558 if (r->dnflags & PFRULE_DN_IS_PIPE) 4559 dnflow.rule.info |= IPFW_IS_PIPE; 4560 4561 dnflow.f_id.extra = dnflow.rule.info; 4562 4563 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4564 dnflow.flags |= IPFW_ARGS_ETHER; 4565 dnflow.ifp = kif->pfik_ifp; 4566 4567 switch (af) { 4568 case AF_INET: 4569 dnflow.f_id.addr_type = 4; 4570 dnflow.f_id.src_ip = src->v4.s_addr; 4571 dnflow.f_id.dst_ip = dst->v4.s_addr; 4572 break; 4573 case AF_INET6: 4574 dnflow.flags |= IPFW_ARGS_IP6; 4575 dnflow.f_id.addr_type = 6; 4576 dnflow.f_id.src_ip6 = src->v6; 4577 dnflow.f_id.dst_ip6 = dst->v6; 4578 break; 4579 } 4580 4581 PF_RULES_RUNLOCK(); 4582 4583 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4584 ip_dn_io_ptr(m0, &dnflow); 4585 if (*m0 != NULL) 4586 pf_dummynet_flag_remove(m, mtag); 4587 } else { 4588 PF_RULES_RUNLOCK(); 4589 } 4590 4591 if (action == PF_PASS && bridge_to) { 4592 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4593 *m0 = NULL; /* We've eaten the packet. */ 4594 } 4595 4596 return (action); 4597 } 4598 4599 static int 4600 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif, 4601 struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am, 4602 struct pf_kruleset **rsm, struct inpcb *inp) 4603 { 4604 struct pf_krule *nr = NULL; 4605 struct pf_addr * const saddr = pd->src; 4606 struct pf_addr * const daddr = pd->dst; 4607 sa_family_t af = pd->af; 4608 struct pf_krule *r, *a = NULL; 4609 struct pf_kruleset *ruleset = NULL; 4610 struct pf_krule_slist match_rules; 4611 struct pf_krule_item *ri; 4612 struct pf_ksrc_node *nsn = NULL; 4613 struct tcphdr *th = &pd->hdr.tcp; 4614 struct pf_state_key *sk = NULL, *nk = NULL; 4615 u_short reason, transerror; 4616 int rewrite = 0, hdrlen = 0; 4617 int tag = -1; 4618 int asd = 0; 4619 int match = 0; 4620 int state_icmp = 0, icmp_dir, multi; 4621 u_int16_t sport = 0, dport = 0, virtual_type, virtual_id; 4622 u_int16_t bproto_sum = 0, bip_sum = 0; 4623 u_int8_t icmptype = 0, icmpcode = 0; 4624 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4625 4626 PF_RULES_RASSERT(); 4627 4628 SLIST_INIT(&match_rules); 4629 4630 if (inp != NULL) { 4631 INP_LOCK_ASSERT(inp); 4632 pd->lookup.uid = inp->inp_cred->cr_uid; 4633 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4634 pd->lookup.done = 1; 4635 } 4636 4637 switch (pd->proto) { 4638 case IPPROTO_TCP: 4639 sport = th->th_sport; 4640 dport = th->th_dport; 4641 hdrlen = sizeof(*th); 4642 break; 4643 case IPPROTO_UDP: 4644 sport = pd->hdr.udp.uh_sport; 4645 dport = pd->hdr.udp.uh_dport; 4646 hdrlen = sizeof(pd->hdr.udp); 4647 break; 4648 case IPPROTO_SCTP: 4649 sport = pd->hdr.sctp.src_port; 4650 dport = pd->hdr.sctp.dest_port; 4651 hdrlen = sizeof(pd->hdr.sctp); 4652 break; 4653 #ifdef INET 4654 case IPPROTO_ICMP: 4655 if (pd->af != AF_INET) 4656 break; 4657 hdrlen = sizeof(pd->hdr.icmp); 4658 icmptype = pd->hdr.icmp.icmp_type; 4659 icmpcode = pd->hdr.icmp.icmp_code; 4660 state_icmp = pf_icmp_mapping(pd, icmptype, 4661 &icmp_dir, &multi, &virtual_id, &virtual_type); 4662 if (icmp_dir == PF_IN) { 4663 sport = virtual_id; 4664 dport = virtual_type; 4665 } else { 4666 sport = virtual_type; 4667 dport = virtual_id; 4668 } 4669 break; 4670 #endif /* INET */ 4671 #ifdef INET6 4672 case IPPROTO_ICMPV6: 4673 if (af != AF_INET6) 4674 break; 4675 hdrlen = sizeof(pd->hdr.icmp6); 4676 icmptype = pd->hdr.icmp6.icmp6_type; 4677 icmpcode = pd->hdr.icmp6.icmp6_code; 4678 state_icmp = pf_icmp_mapping(pd, icmptype, 4679 &icmp_dir, &multi, &virtual_id, &virtual_type); 4680 if (icmp_dir == PF_IN) { 4681 sport = virtual_id; 4682 dport = virtual_type; 4683 } else { 4684 sport = virtual_type; 4685 dport = virtual_id; 4686 } 4687 4688 break; 4689 #endif /* INET6 */ 4690 default: 4691 sport = dport = hdrlen = 0; 4692 break; 4693 } 4694 4695 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4696 4697 /* check packet for BINAT/NAT/RDR */ 4698 transerror = pf_get_translation(pd, m, off, kif, &nsn, &sk, 4699 &nk, saddr, daddr, sport, dport, anchor_stack, &nr); 4700 switch (transerror) { 4701 default: 4702 /* A translation error occurred. */ 4703 REASON_SET(&reason, transerror); 4704 goto cleanup; 4705 case PFRES_MAX: 4706 /* No match. */ 4707 break; 4708 case PFRES_MATCH: 4709 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4710 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4711 4712 if (nr->log) { 4713 PFLOG_PACKET(kif, m, af, PF_PASS, PFRES_MATCH, nr, a, 4714 ruleset, pd, 1); 4715 } 4716 4717 if (pd->ip_sum) 4718 bip_sum = *pd->ip_sum; 4719 4720 switch (pd->proto) { 4721 case IPPROTO_TCP: 4722 bproto_sum = th->th_sum; 4723 pd->proto_sum = &th->th_sum; 4724 4725 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4726 nk->port[pd->sidx] != sport) { 4727 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 4728 &th->th_sum, &nk->addr[pd->sidx], 4729 nk->port[pd->sidx], 0, af); 4730 pd->sport = &th->th_sport; 4731 sport = th->th_sport; 4732 } 4733 4734 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4735 nk->port[pd->didx] != dport) { 4736 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 4737 &th->th_sum, &nk->addr[pd->didx], 4738 nk->port[pd->didx], 0, af); 4739 dport = th->th_dport; 4740 pd->dport = &th->th_dport; 4741 } 4742 rewrite++; 4743 break; 4744 case IPPROTO_UDP: 4745 bproto_sum = pd->hdr.udp.uh_sum; 4746 pd->proto_sum = &pd->hdr.udp.uh_sum; 4747 4748 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4749 nk->port[pd->sidx] != sport) { 4750 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 4751 pd->ip_sum, &pd->hdr.udp.uh_sum, 4752 &nk->addr[pd->sidx], 4753 nk->port[pd->sidx], 1, af); 4754 sport = pd->hdr.udp.uh_sport; 4755 pd->sport = &pd->hdr.udp.uh_sport; 4756 } 4757 4758 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4759 nk->port[pd->didx] != dport) { 4760 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 4761 pd->ip_sum, &pd->hdr.udp.uh_sum, 4762 &nk->addr[pd->didx], 4763 nk->port[pd->didx], 1, af); 4764 dport = pd->hdr.udp.uh_dport; 4765 pd->dport = &pd->hdr.udp.uh_dport; 4766 } 4767 rewrite++; 4768 break; 4769 case IPPROTO_SCTP: { 4770 uint16_t checksum = 0; 4771 4772 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4773 nk->port[pd->sidx] != sport) { 4774 pf_change_ap(m, saddr, &pd->hdr.sctp.src_port, 4775 pd->ip_sum, &checksum, 4776 &nk->addr[pd->sidx], 4777 nk->port[pd->sidx], 1, af); 4778 } 4779 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4780 nk->port[pd->didx] != dport) { 4781 pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port, 4782 pd->ip_sum, &checksum, 4783 &nk->addr[pd->didx], 4784 nk->port[pd->didx], 1, af); 4785 } 4786 break; 4787 } 4788 #ifdef INET 4789 case IPPROTO_ICMP: 4790 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 4791 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 4792 nk->addr[pd->sidx].v4.s_addr, 0); 4793 4794 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 4795 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 4796 nk->addr[pd->didx].v4.s_addr, 0); 4797 4798 if (virtual_type == htons(ICMP_ECHO) && 4799 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 4800 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 4801 pd->hdr.icmp.icmp_cksum, sport, 4802 nk->port[pd->sidx], 0); 4803 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 4804 pd->sport = &pd->hdr.icmp.icmp_id; 4805 } 4806 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 4807 break; 4808 #endif /* INET */ 4809 #ifdef INET6 4810 case IPPROTO_ICMPV6: 4811 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 4812 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 4813 &nk->addr[pd->sidx], 0); 4814 4815 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 4816 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 4817 &nk->addr[pd->didx], 0); 4818 rewrite++; 4819 break; 4820 #endif /* INET */ 4821 default: 4822 switch (af) { 4823 #ifdef INET 4824 case AF_INET: 4825 if (PF_ANEQ(saddr, 4826 &nk->addr[pd->sidx], AF_INET)) 4827 pf_change_a(&saddr->v4.s_addr, 4828 pd->ip_sum, 4829 nk->addr[pd->sidx].v4.s_addr, 0); 4830 4831 if (PF_ANEQ(daddr, 4832 &nk->addr[pd->didx], AF_INET)) 4833 pf_change_a(&daddr->v4.s_addr, 4834 pd->ip_sum, 4835 nk->addr[pd->didx].v4.s_addr, 0); 4836 break; 4837 #endif /* INET */ 4838 #ifdef INET6 4839 case AF_INET6: 4840 if (PF_ANEQ(saddr, 4841 &nk->addr[pd->sidx], AF_INET6)) 4842 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 4843 4844 if (PF_ANEQ(daddr, 4845 &nk->addr[pd->didx], AF_INET6)) 4846 PF_ACPY(daddr, &nk->addr[pd->didx], af); 4847 break; 4848 #endif /* INET */ 4849 } 4850 break; 4851 } 4852 if (nr->natpass) 4853 r = NULL; 4854 pd->nat_rule = nr; 4855 } 4856 4857 while (r != NULL) { 4858 pf_counter_u64_add(&r->evaluations, 1); 4859 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 4860 r = r->skip[PF_SKIP_IFP].ptr; 4861 else if (r->direction && r->direction != pd->dir) 4862 r = r->skip[PF_SKIP_DIR].ptr; 4863 else if (r->af && r->af != af) 4864 r = r->skip[PF_SKIP_AF].ptr; 4865 else if (r->proto && r->proto != pd->proto) 4866 r = r->skip[PF_SKIP_PROTO].ptr; 4867 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 4868 r->src.neg, kif, M_GETFIB(m))) 4869 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4870 /* tcp/udp only. port_op always 0 in other cases */ 4871 else if (r->src.port_op && !pf_match_port(r->src.port_op, 4872 r->src.port[0], r->src.port[1], sport)) 4873 r = r->skip[PF_SKIP_SRC_PORT].ptr; 4874 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 4875 r->dst.neg, NULL, M_GETFIB(m))) 4876 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4877 /* tcp/udp only. port_op always 0 in other cases */ 4878 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 4879 r->dst.port[0], r->dst.port[1], dport)) 4880 r = r->skip[PF_SKIP_DST_PORT].ptr; 4881 /* icmp only. type always 0 in other cases */ 4882 else if (r->type && r->type != icmptype + 1) 4883 r = TAILQ_NEXT(r, entries); 4884 /* icmp only. type always 0 in other cases */ 4885 else if (r->code && r->code != icmpcode + 1) 4886 r = TAILQ_NEXT(r, entries); 4887 else if (r->tos && !(r->tos == pd->tos)) 4888 r = TAILQ_NEXT(r, entries); 4889 else if (r->rule_flag & PFRULE_FRAGMENT) 4890 r = TAILQ_NEXT(r, entries); 4891 else if (pd->proto == IPPROTO_TCP && 4892 (r->flagset & th->th_flags) != r->flags) 4893 r = TAILQ_NEXT(r, entries); 4894 /* tcp/udp only. uid.op always 0 in other cases */ 4895 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 4896 pf_socket_lookup(pd, m), 1)) && 4897 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 4898 pd->lookup.uid)) 4899 r = TAILQ_NEXT(r, entries); 4900 /* tcp/udp only. gid.op always 0 in other cases */ 4901 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 4902 pf_socket_lookup(pd, m), 1)) && 4903 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 4904 pd->lookup.gid)) 4905 r = TAILQ_NEXT(r, entries); 4906 else if (r->prio && 4907 !pf_match_ieee8021q_pcp(r->prio, m)) 4908 r = TAILQ_NEXT(r, entries); 4909 else if (r->prob && 4910 r->prob <= arc4random()) 4911 r = TAILQ_NEXT(r, entries); 4912 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4913 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4914 r = TAILQ_NEXT(r, entries); 4915 else if (r->os_fingerprint != PF_OSFP_ANY && 4916 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 4917 pf_osfp_fingerprint(pd, m, off, th), 4918 r->os_fingerprint))) 4919 r = TAILQ_NEXT(r, entries); 4920 else { 4921 if (r->tag) 4922 tag = r->tag; 4923 if (r->anchor == NULL) { 4924 if (r->action == PF_MATCH) { 4925 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 4926 if (ri == NULL) { 4927 REASON_SET(&reason, PFRES_MEMORY); 4928 goto cleanup; 4929 } 4930 ri->r = r; 4931 SLIST_INSERT_HEAD(&match_rules, ri, entry); 4932 pf_counter_u64_critical_enter(); 4933 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 4934 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 4935 pf_counter_u64_critical_exit(); 4936 pf_rule_to_actions(r, &pd->act); 4937 if (r->log) 4938 PFLOG_PACKET(kif, m, af, 4939 r->action, PFRES_MATCH, r, 4940 a, ruleset, pd, 1); 4941 } else { 4942 match = 1; 4943 *rm = r; 4944 *am = a; 4945 *rsm = ruleset; 4946 } 4947 if ((*rm)->quick) 4948 break; 4949 r = TAILQ_NEXT(r, entries); 4950 } else 4951 pf_step_into_anchor(anchor_stack, &asd, 4952 &ruleset, PF_RULESET_FILTER, &r, &a, 4953 &match); 4954 } 4955 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4956 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4957 break; 4958 } 4959 r = *rm; 4960 a = *am; 4961 ruleset = *rsm; 4962 4963 REASON_SET(&reason, PFRES_MATCH); 4964 4965 /* apply actions for last matching pass/block rule */ 4966 pf_rule_to_actions(r, &pd->act); 4967 4968 if (r->log) { 4969 if (rewrite) 4970 m_copyback(m, off, hdrlen, pd->hdr.any); 4971 PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1); 4972 } 4973 4974 if ((r->action == PF_DROP) && 4975 ((r->rule_flag & PFRULE_RETURNRST) || 4976 (r->rule_flag & PFRULE_RETURNICMP) || 4977 (r->rule_flag & PFRULE_RETURN))) { 4978 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 4979 bip_sum, hdrlen, &reason, r->rtableid); 4980 } 4981 4982 if (r->action == PF_DROP) 4983 goto cleanup; 4984 4985 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4986 REASON_SET(&reason, PFRES_MEMORY); 4987 goto cleanup; 4988 } 4989 if (pd->act.rtableid >= 0) 4990 M_SETFIB(m, pd->act.rtableid); 4991 4992 if (!state_icmp && (r->keep_state || nr != NULL || 4993 (pd->flags & PFDESC_TCP_NORM))) { 4994 int action; 4995 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 4996 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 4997 hdrlen, &match_rules); 4998 if (action != PF_PASS) { 4999 if (action == PF_DROP && 5000 (r->rule_flag & PFRULE_RETURN)) 5001 pf_return(r, nr, pd, sk, off, m, th, kif, 5002 bproto_sum, bip_sum, hdrlen, &reason, 5003 pd->act.rtableid); 5004 return (action); 5005 } 5006 } else { 5007 while ((ri = SLIST_FIRST(&match_rules))) { 5008 SLIST_REMOVE_HEAD(&match_rules, entry); 5009 free(ri, M_PF_RULE_ITEM); 5010 } 5011 5012 uma_zfree(V_pf_state_key_z, sk); 5013 uma_zfree(V_pf_state_key_z, nk); 5014 } 5015 5016 /* copy back packet headers if we performed NAT operations */ 5017 if (rewrite) 5018 m_copyback(m, off, hdrlen, pd->hdr.any); 5019 5020 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5021 pd->dir == PF_OUT && 5022 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 5023 /* 5024 * We want the state created, but we dont 5025 * want to send this in case a partner 5026 * firewall has to know about it to allow 5027 * replies through it. 5028 */ 5029 return (PF_DEFER); 5030 5031 return (PF_PASS); 5032 5033 cleanup: 5034 while ((ri = SLIST_FIRST(&match_rules))) { 5035 SLIST_REMOVE_HEAD(&match_rules, entry); 5036 free(ri, M_PF_RULE_ITEM); 5037 } 5038 5039 uma_zfree(V_pf_state_key_z, sk); 5040 uma_zfree(V_pf_state_key_z, nk); 5041 return (PF_DROP); 5042 } 5043 5044 static int 5045 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 5046 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 5047 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 5048 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 5049 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 5050 struct pf_krule_slist *match_rules) 5051 { 5052 struct pf_kstate *s = NULL; 5053 struct pf_ksrc_node *sn = NULL; 5054 struct tcphdr *th = &pd->hdr.tcp; 5055 u_int16_t mss = V_tcp_mssdflt; 5056 u_short reason, sn_reason; 5057 struct pf_krule_item *ri; 5058 5059 /* check maximums */ 5060 if (r->max_states && 5061 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 5062 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 5063 REASON_SET(&reason, PFRES_MAXSTATES); 5064 goto csfailed; 5065 } 5066 /* src node for filter rule */ 5067 if ((r->rule_flag & PFRULE_SRCTRACK || 5068 r->rpool.opts & PF_POOL_STICKYADDR) && 5069 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 5070 REASON_SET(&reason, sn_reason); 5071 goto csfailed; 5072 } 5073 /* src node for translation rule */ 5074 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 5075 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 5076 pd->af)) != 0 ) { 5077 REASON_SET(&reason, sn_reason); 5078 goto csfailed; 5079 } 5080 s = pf_alloc_state(M_NOWAIT); 5081 if (s == NULL) { 5082 REASON_SET(&reason, PFRES_MEMORY); 5083 goto csfailed; 5084 } 5085 s->rule.ptr = r; 5086 s->nat_rule.ptr = nr; 5087 s->anchor.ptr = a; 5088 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 5089 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 5090 5091 STATE_INC_COUNTERS(s); 5092 if (r->allow_opts) 5093 s->state_flags |= PFSTATE_ALLOWOPTS; 5094 if (r->rule_flag & PFRULE_STATESLOPPY) 5095 s->state_flags |= PFSTATE_SLOPPY; 5096 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 5097 s->state_flags |= PFSTATE_SCRUB_TCP; 5098 if ((r->rule_flag & PFRULE_PFLOW) || 5099 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 5100 s->state_flags |= PFSTATE_PFLOW; 5101 5102 s->act.log = pd->act.log & PF_LOG_ALL; 5103 s->sync_state = PFSYNC_S_NONE; 5104 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 5105 5106 if (nr != NULL) 5107 s->act.log |= nr->log & PF_LOG_ALL; 5108 switch (pd->proto) { 5109 case IPPROTO_TCP: 5110 s->src.seqlo = ntohl(th->th_seq); 5111 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 5112 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 5113 r->keep_state == PF_STATE_MODULATE) { 5114 /* Generate sequence number modulator */ 5115 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 5116 0) 5117 s->src.seqdiff = 1; 5118 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 5119 htonl(s->src.seqlo + s->src.seqdiff), 0); 5120 *rewrite = 1; 5121 } else 5122 s->src.seqdiff = 0; 5123 if (th->th_flags & TH_SYN) { 5124 s->src.seqhi++; 5125 s->src.wscale = pf_get_wscale(m, off, 5126 th->th_off, pd->af); 5127 } 5128 s->src.max_win = MAX(ntohs(th->th_win), 1); 5129 if (s->src.wscale & PF_WSCALE_MASK) { 5130 /* Remove scale factor from initial window */ 5131 int win = s->src.max_win; 5132 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 5133 s->src.max_win = (win - 1) >> 5134 (s->src.wscale & PF_WSCALE_MASK); 5135 } 5136 if (th->th_flags & TH_FIN) 5137 s->src.seqhi++; 5138 s->dst.seqhi = 1; 5139 s->dst.max_win = 1; 5140 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 5141 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 5142 s->timeout = PFTM_TCP_FIRST_PACKET; 5143 atomic_add_32(&V_pf_status.states_halfopen, 1); 5144 break; 5145 case IPPROTO_UDP: 5146 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 5147 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 5148 s->timeout = PFTM_UDP_FIRST_PACKET; 5149 break; 5150 case IPPROTO_SCTP: 5151 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 5152 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 5153 s->timeout = PFTM_SCTP_FIRST_PACKET; 5154 break; 5155 case IPPROTO_ICMP: 5156 #ifdef INET6 5157 case IPPROTO_ICMPV6: 5158 #endif 5159 s->timeout = PFTM_ICMP_FIRST_PACKET; 5160 break; 5161 default: 5162 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 5163 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 5164 s->timeout = PFTM_OTHER_FIRST_PACKET; 5165 } 5166 5167 if (r->rt) { 5168 /* pf_map_addr increases the reason counters */ 5169 if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr, 5170 &s->rt_kif, NULL, &sn)) != 0) 5171 goto csfailed; 5172 s->rt = r->rt; 5173 } 5174 5175 s->creation = s->expire = pf_get_uptime(); 5176 5177 if (sn != NULL) 5178 s->src_node = sn; 5179 if (nsn != NULL) { 5180 /* XXX We only modify one side for now. */ 5181 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 5182 s->nat_src_node = nsn; 5183 } 5184 if (pd->proto == IPPROTO_TCP) { 5185 if (s->state_flags & PFSTATE_SCRUB_TCP && 5186 pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { 5187 REASON_SET(&reason, PFRES_MEMORY); 5188 goto drop; 5189 } 5190 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 5191 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 5192 &s->src, &s->dst, rewrite)) { 5193 /* This really shouldn't happen!!! */ 5194 DPFPRINTF(PF_DEBUG_URGENT, 5195 ("pf_normalize_tcp_stateful failed on first " 5196 "pkt\n")); 5197 goto drop; 5198 } 5199 } else if (pd->proto == IPPROTO_SCTP) { 5200 if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst)) 5201 goto drop; 5202 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 5203 goto drop; 5204 } 5205 s->direction = pd->dir; 5206 5207 /* 5208 * sk/nk could already been setup by pf_get_translation(). 5209 */ 5210 if (nr == NULL) { 5211 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 5212 __func__, nr, sk, nk)); 5213 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 5214 if (sk == NULL) 5215 goto csfailed; 5216 nk = sk; 5217 } else 5218 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 5219 __func__, nr, sk, nk)); 5220 5221 /* Swap sk/nk for PF_OUT. */ 5222 if (pf_state_insert(BOUND_IFACE(s, kif), kif, 5223 (pd->dir == PF_IN) ? sk : nk, 5224 (pd->dir == PF_IN) ? nk : sk, s)) { 5225 REASON_SET(&reason, PFRES_STATEINS); 5226 goto drop; 5227 } else 5228 *sm = s; 5229 5230 if (tag > 0) 5231 s->tag = tag; 5232 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 5233 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 5234 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 5235 /* undo NAT changes, if they have taken place */ 5236 if (nr != NULL) { 5237 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 5238 if (pd->dir == PF_OUT) 5239 skt = s->key[PF_SK_STACK]; 5240 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 5241 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 5242 if (pd->sport) 5243 *pd->sport = skt->port[pd->sidx]; 5244 if (pd->dport) 5245 *pd->dport = skt->port[pd->didx]; 5246 if (pd->proto_sum) 5247 *pd->proto_sum = bproto_sum; 5248 if (pd->ip_sum) 5249 *pd->ip_sum = bip_sum; 5250 m_copyback(m, off, hdrlen, pd->hdr.any); 5251 } 5252 s->src.seqhi = htonl(arc4random()); 5253 /* Find mss option */ 5254 int rtid = M_GETFIB(m); 5255 mss = pf_get_mss(m, off, th->th_off, pd->af); 5256 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5257 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5258 s->src.mss = mss; 5259 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5260 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5261 TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0, 5262 pd->act.rtableid); 5263 REASON_SET(&reason, PFRES_SYNPROXY); 5264 return (PF_SYNPROXY_DROP); 5265 } 5266 5267 return (PF_PASS); 5268 5269 csfailed: 5270 while ((ri = SLIST_FIRST(match_rules))) { 5271 SLIST_REMOVE_HEAD(match_rules, entry); 5272 free(ri, M_PF_RULE_ITEM); 5273 } 5274 5275 uma_zfree(V_pf_state_key_z, sk); 5276 uma_zfree(V_pf_state_key_z, nk); 5277 5278 if (sn != NULL) { 5279 PF_SRC_NODE_LOCK(sn); 5280 if (--sn->states == 0 && sn->expire == 0) { 5281 pf_unlink_src_node(sn); 5282 uma_zfree(V_pf_sources_z, sn); 5283 counter_u64_add( 5284 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5285 } 5286 PF_SRC_NODE_UNLOCK(sn); 5287 } 5288 5289 if (nsn != sn && nsn != NULL) { 5290 PF_SRC_NODE_LOCK(nsn); 5291 if (--nsn->states == 0 && nsn->expire == 0) { 5292 pf_unlink_src_node(nsn); 5293 uma_zfree(V_pf_sources_z, nsn); 5294 counter_u64_add( 5295 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5296 } 5297 PF_SRC_NODE_UNLOCK(nsn); 5298 } 5299 5300 drop: 5301 if (s != NULL) { 5302 pf_src_tree_remove_state(s); 5303 s->timeout = PFTM_UNLINKED; 5304 STATE_DEC_COUNTERS(s); 5305 pf_free_state(s); 5306 } 5307 5308 return (PF_DROP); 5309 } 5310 5311 static int 5312 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif, 5313 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 5314 struct pf_kruleset **rsm) 5315 { 5316 struct pf_krule *r, *a = NULL; 5317 struct pf_kruleset *ruleset = NULL; 5318 struct pf_krule_slist match_rules; 5319 struct pf_krule_item *ri; 5320 sa_family_t af = pd->af; 5321 u_short reason; 5322 int tag = -1; 5323 int asd = 0; 5324 int match = 0; 5325 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5326 5327 PF_RULES_RASSERT(); 5328 5329 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5330 SLIST_INIT(&match_rules); 5331 while (r != NULL) { 5332 pf_counter_u64_add(&r->evaluations, 1); 5333 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 5334 r = r->skip[PF_SKIP_IFP].ptr; 5335 else if (r->direction && r->direction != pd->dir) 5336 r = r->skip[PF_SKIP_DIR].ptr; 5337 else if (r->af && r->af != af) 5338 r = r->skip[PF_SKIP_AF].ptr; 5339 else if (r->proto && r->proto != pd->proto) 5340 r = r->skip[PF_SKIP_PROTO].ptr; 5341 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 5342 r->src.neg, kif, M_GETFIB(m))) 5343 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 5344 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 5345 r->dst.neg, NULL, M_GETFIB(m))) 5346 r = r->skip[PF_SKIP_DST_ADDR].ptr; 5347 else if (r->tos && !(r->tos == pd->tos)) 5348 r = TAILQ_NEXT(r, entries); 5349 else if (r->os_fingerprint != PF_OSFP_ANY) 5350 r = TAILQ_NEXT(r, entries); 5351 else if (pd->proto == IPPROTO_UDP && 5352 (r->src.port_op || r->dst.port_op)) 5353 r = TAILQ_NEXT(r, entries); 5354 else if (pd->proto == IPPROTO_TCP && 5355 (r->src.port_op || r->dst.port_op || r->flagset)) 5356 r = TAILQ_NEXT(r, entries); 5357 else if ((pd->proto == IPPROTO_ICMP || 5358 pd->proto == IPPROTO_ICMPV6) && 5359 (r->type || r->code)) 5360 r = TAILQ_NEXT(r, entries); 5361 else if (r->prio && 5362 !pf_match_ieee8021q_pcp(r->prio, m)) 5363 r = TAILQ_NEXT(r, entries); 5364 else if (r->prob && r->prob <= 5365 (arc4random() % (UINT_MAX - 1) + 1)) 5366 r = TAILQ_NEXT(r, entries); 5367 else if (r->match_tag && !pf_match_tag(m, r, &tag, 5368 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 5369 r = TAILQ_NEXT(r, entries); 5370 else { 5371 if (r->anchor == NULL) { 5372 if (r->action == PF_MATCH) { 5373 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5374 if (ri == NULL) { 5375 REASON_SET(&reason, PFRES_MEMORY); 5376 goto cleanup; 5377 } 5378 ri->r = r; 5379 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5380 pf_counter_u64_critical_enter(); 5381 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5382 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5383 pf_counter_u64_critical_exit(); 5384 pf_rule_to_actions(r, &pd->act); 5385 if (r->log) 5386 PFLOG_PACKET(kif, m, af, 5387 r->action, PFRES_MATCH, r, 5388 a, ruleset, pd, 1); 5389 } else { 5390 match = 1; 5391 *rm = r; 5392 *am = a; 5393 *rsm = ruleset; 5394 } 5395 if ((*rm)->quick) 5396 break; 5397 r = TAILQ_NEXT(r, entries); 5398 } else 5399 pf_step_into_anchor(anchor_stack, &asd, 5400 &ruleset, PF_RULESET_FILTER, &r, &a, 5401 &match); 5402 } 5403 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5404 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5405 break; 5406 } 5407 r = *rm; 5408 a = *am; 5409 ruleset = *rsm; 5410 5411 REASON_SET(&reason, PFRES_MATCH); 5412 5413 /* apply actions for last matching pass/block rule */ 5414 pf_rule_to_actions(r, &pd->act); 5415 5416 if (r->log) 5417 PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1); 5418 5419 if (r->action != PF_PASS) 5420 return (PF_DROP); 5421 5422 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5423 REASON_SET(&reason, PFRES_MEMORY); 5424 goto cleanup; 5425 } 5426 5427 return (PF_PASS); 5428 5429 cleanup: 5430 while ((ri = SLIST_FIRST(&match_rules))) { 5431 SLIST_REMOVE_HEAD(&match_rules, entry); 5432 free(ri, M_PF_RULE_ITEM); 5433 } 5434 5435 return (PF_DROP); 5436 } 5437 5438 static int 5439 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 5440 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 5441 int *copyback) 5442 { 5443 struct tcphdr *th = &pd->hdr.tcp; 5444 struct pf_state_peer *src, *dst; 5445 u_int16_t win = ntohs(th->th_win); 5446 u_int32_t ack, end, data_end, seq, orig_seq; 5447 u_int8_t sws, dws, psrc, pdst; 5448 int ackskew; 5449 5450 if (pd->dir == (*state)->direction) { 5451 src = &(*state)->src; 5452 dst = &(*state)->dst; 5453 psrc = PF_PEER_SRC; 5454 pdst = PF_PEER_DST; 5455 } else { 5456 src = &(*state)->dst; 5457 dst = &(*state)->src; 5458 psrc = PF_PEER_DST; 5459 pdst = PF_PEER_SRC; 5460 } 5461 5462 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5463 sws = src->wscale & PF_WSCALE_MASK; 5464 dws = dst->wscale & PF_WSCALE_MASK; 5465 } else 5466 sws = dws = 0; 5467 5468 /* 5469 * Sequence tracking algorithm from Guido van Rooij's paper: 5470 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5471 * tcp_filtering.ps 5472 */ 5473 5474 orig_seq = seq = ntohl(th->th_seq); 5475 if (src->seqlo == 0) { 5476 /* First packet from this end. Set its state */ 5477 5478 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5479 src->scrub == NULL) { 5480 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 5481 REASON_SET(reason, PFRES_MEMORY); 5482 return (PF_DROP); 5483 } 5484 } 5485 5486 /* Deferred generation of sequence number modulator */ 5487 if (dst->seqdiff && !src->seqdiff) { 5488 /* use random iss for the TCP server */ 5489 while ((src->seqdiff = arc4random() - seq) == 0) 5490 ; 5491 ack = ntohl(th->th_ack) - dst->seqdiff; 5492 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5493 src->seqdiff), 0); 5494 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5495 *copyback = 1; 5496 } else { 5497 ack = ntohl(th->th_ack); 5498 } 5499 5500 end = seq + pd->p_len; 5501 if (th->th_flags & TH_SYN) { 5502 end++; 5503 if (dst->wscale & PF_WSCALE_FLAG) { 5504 src->wscale = pf_get_wscale(m, off, th->th_off, 5505 pd->af); 5506 if (src->wscale & PF_WSCALE_FLAG) { 5507 /* Remove scale factor from initial 5508 * window */ 5509 sws = src->wscale & PF_WSCALE_MASK; 5510 win = ((u_int32_t)win + (1 << sws) - 1) 5511 >> sws; 5512 dws = dst->wscale & PF_WSCALE_MASK; 5513 } else { 5514 /* fixup other window */ 5515 dst->max_win = MIN(TCP_MAXWIN, 5516 (u_int32_t)dst->max_win << 5517 (dst->wscale & PF_WSCALE_MASK)); 5518 /* in case of a retrans SYN|ACK */ 5519 dst->wscale = 0; 5520 } 5521 } 5522 } 5523 data_end = end; 5524 if (th->th_flags & TH_FIN) 5525 end++; 5526 5527 src->seqlo = seq; 5528 if (src->state < TCPS_SYN_SENT) 5529 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5530 5531 /* 5532 * May need to slide the window (seqhi may have been set by 5533 * the crappy stack check or if we picked up the connection 5534 * after establishment) 5535 */ 5536 if (src->seqhi == 1 || 5537 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5538 src->seqhi = end + MAX(1, dst->max_win << dws); 5539 if (win > src->max_win) 5540 src->max_win = win; 5541 5542 } else { 5543 ack = ntohl(th->th_ack) - dst->seqdiff; 5544 if (src->seqdiff) { 5545 /* Modulate sequence numbers */ 5546 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5547 src->seqdiff), 0); 5548 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5549 *copyback = 1; 5550 } 5551 end = seq + pd->p_len; 5552 if (th->th_flags & TH_SYN) 5553 end++; 5554 data_end = end; 5555 if (th->th_flags & TH_FIN) 5556 end++; 5557 } 5558 5559 if ((th->th_flags & TH_ACK) == 0) { 5560 /* Let it pass through the ack skew check */ 5561 ack = dst->seqlo; 5562 } else if ((ack == 0 && 5563 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5564 /* broken tcp stacks do not set ack */ 5565 (dst->state < TCPS_SYN_SENT)) { 5566 /* 5567 * Many stacks (ours included) will set the ACK number in an 5568 * FIN|ACK if the SYN times out -- no sequence to ACK. 5569 */ 5570 ack = dst->seqlo; 5571 } 5572 5573 if (seq == end) { 5574 /* Ease sequencing restrictions on no data packets */ 5575 seq = src->seqlo; 5576 data_end = end = seq; 5577 } 5578 5579 ackskew = dst->seqlo - ack; 5580 5581 /* 5582 * Need to demodulate the sequence numbers in any TCP SACK options 5583 * (Selective ACK). We could optionally validate the SACK values 5584 * against the current ACK window, either forwards or backwards, but 5585 * I'm not confident that SACK has been implemented properly 5586 * everywhere. It wouldn't surprise me if several stacks accidentally 5587 * SACK too far backwards of previously ACKed data. There really aren't 5588 * any security implications of bad SACKing unless the target stack 5589 * doesn't validate the option length correctly. Someone trying to 5590 * spoof into a TCP connection won't bother blindly sending SACK 5591 * options anyway. 5592 */ 5593 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5594 if (pf_modulate_sack(m, off, pd, th, dst)) 5595 *copyback = 1; 5596 } 5597 5598 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5599 if (SEQ_GEQ(src->seqhi, data_end) && 5600 /* Last octet inside other's window space */ 5601 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5602 /* Retrans: not more than one window back */ 5603 (ackskew >= -MAXACKWINDOW) && 5604 /* Acking not more than one reassembled fragment backwards */ 5605 (ackskew <= (MAXACKWINDOW << sws)) && 5606 /* Acking not more than one window forward */ 5607 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5608 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 5609 /* Require an exact/+1 sequence match on resets when possible */ 5610 5611 if (dst->scrub || src->scrub) { 5612 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5613 *state, src, dst, copyback)) 5614 return (PF_DROP); 5615 } 5616 5617 /* update max window */ 5618 if (src->max_win < win) 5619 src->max_win = win; 5620 /* synchronize sequencing */ 5621 if (SEQ_GT(end, src->seqlo)) 5622 src->seqlo = end; 5623 /* slide the window of what the other end can send */ 5624 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5625 dst->seqhi = ack + MAX((win << sws), 1); 5626 5627 /* update states */ 5628 if (th->th_flags & TH_SYN) 5629 if (src->state < TCPS_SYN_SENT) 5630 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5631 if (th->th_flags & TH_FIN) 5632 if (src->state < TCPS_CLOSING) 5633 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5634 if (th->th_flags & TH_ACK) { 5635 if (dst->state == TCPS_SYN_SENT) { 5636 pf_set_protostate(*state, pdst, 5637 TCPS_ESTABLISHED); 5638 if (src->state == TCPS_ESTABLISHED && 5639 (*state)->src_node != NULL && 5640 pf_src_connlimit(state)) { 5641 REASON_SET(reason, PFRES_SRCLIMIT); 5642 return (PF_DROP); 5643 } 5644 } else if (dst->state == TCPS_CLOSING) 5645 pf_set_protostate(*state, pdst, 5646 TCPS_FIN_WAIT_2); 5647 } 5648 if (th->th_flags & TH_RST) 5649 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5650 5651 /* update expire time */ 5652 (*state)->expire = pf_get_uptime(); 5653 if (src->state >= TCPS_FIN_WAIT_2 && 5654 dst->state >= TCPS_FIN_WAIT_2) 5655 (*state)->timeout = PFTM_TCP_CLOSED; 5656 else if (src->state >= TCPS_CLOSING && 5657 dst->state >= TCPS_CLOSING) 5658 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5659 else if (src->state < TCPS_ESTABLISHED || 5660 dst->state < TCPS_ESTABLISHED) 5661 (*state)->timeout = PFTM_TCP_OPENING; 5662 else if (src->state >= TCPS_CLOSING || 5663 dst->state >= TCPS_CLOSING) 5664 (*state)->timeout = PFTM_TCP_CLOSING; 5665 else 5666 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5667 5668 /* Fall through to PASS packet */ 5669 5670 } else if ((dst->state < TCPS_SYN_SENT || 5671 dst->state >= TCPS_FIN_WAIT_2 || 5672 src->state >= TCPS_FIN_WAIT_2) && 5673 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 5674 /* Within a window forward of the originating packet */ 5675 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5676 /* Within a window backward of the originating packet */ 5677 5678 /* 5679 * This currently handles three situations: 5680 * 1) Stupid stacks will shotgun SYNs before their peer 5681 * replies. 5682 * 2) When PF catches an already established stream (the 5683 * firewall rebooted, the state table was flushed, routes 5684 * changed...) 5685 * 3) Packets get funky immediately after the connection 5686 * closes (this should catch Solaris spurious ACK|FINs 5687 * that web servers like to spew after a close) 5688 * 5689 * This must be a little more careful than the above code 5690 * since packet floods will also be caught here. We don't 5691 * update the TTL here to mitigate the damage of a packet 5692 * flood and so the same code can handle awkward establishment 5693 * and a loosened connection close. 5694 * In the establishment case, a correct peer response will 5695 * validate the connection, go through the normal state code 5696 * and keep updating the state TTL. 5697 */ 5698 5699 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5700 printf("pf: loose state match: "); 5701 pf_print_state(*state); 5702 pf_print_flags(th->th_flags); 5703 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5704 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5705 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5706 (unsigned long long)(*state)->packets[1], 5707 pd->dir == PF_IN ? "in" : "out", 5708 pd->dir == (*state)->direction ? "fwd" : "rev"); 5709 } 5710 5711 if (dst->scrub || src->scrub) { 5712 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5713 *state, src, dst, copyback)) 5714 return (PF_DROP); 5715 } 5716 5717 /* update max window */ 5718 if (src->max_win < win) 5719 src->max_win = win; 5720 /* synchronize sequencing */ 5721 if (SEQ_GT(end, src->seqlo)) 5722 src->seqlo = end; 5723 /* slide the window of what the other end can send */ 5724 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5725 dst->seqhi = ack + MAX((win << sws), 1); 5726 5727 /* 5728 * Cannot set dst->seqhi here since this could be a shotgunned 5729 * SYN and not an already established connection. 5730 */ 5731 5732 if (th->th_flags & TH_FIN) 5733 if (src->state < TCPS_CLOSING) 5734 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5735 if (th->th_flags & TH_RST) 5736 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5737 5738 /* Fall through to PASS packet */ 5739 5740 } else { 5741 if ((*state)->dst.state == TCPS_SYN_SENT && 5742 (*state)->src.state == TCPS_SYN_SENT) { 5743 /* Send RST for state mismatches during handshake */ 5744 if (!(th->th_flags & TH_RST)) 5745 pf_send_tcp((*state)->rule.ptr, pd->af, 5746 pd->dst, pd->src, th->th_dport, 5747 th->th_sport, ntohl(th->th_ack), 0, 5748 TH_RST, 0, 0, 5749 (*state)->rule.ptr->return_ttl, true, 0, 0, 5750 (*state)->act.rtableid); 5751 src->seqlo = 0; 5752 src->seqhi = 1; 5753 src->max_win = 1; 5754 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5755 printf("pf: BAD state: "); 5756 pf_print_state(*state); 5757 pf_print_flags(th->th_flags); 5758 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5759 "pkts=%llu:%llu dir=%s,%s\n", 5760 seq, orig_seq, ack, pd->p_len, ackskew, 5761 (unsigned long long)(*state)->packets[0], 5762 (unsigned long long)(*state)->packets[1], 5763 pd->dir == PF_IN ? "in" : "out", 5764 pd->dir == (*state)->direction ? "fwd" : "rev"); 5765 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5766 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 5767 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5768 ' ': '2', 5769 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5770 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5771 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 5772 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5773 } 5774 REASON_SET(reason, PFRES_BADSTATE); 5775 return (PF_DROP); 5776 } 5777 5778 return (PF_PASS); 5779 } 5780 5781 static int 5782 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5783 { 5784 struct tcphdr *th = &pd->hdr.tcp; 5785 struct pf_state_peer *src, *dst; 5786 u_int8_t psrc, pdst; 5787 5788 if (pd->dir == (*state)->direction) { 5789 src = &(*state)->src; 5790 dst = &(*state)->dst; 5791 psrc = PF_PEER_SRC; 5792 pdst = PF_PEER_DST; 5793 } else { 5794 src = &(*state)->dst; 5795 dst = &(*state)->src; 5796 psrc = PF_PEER_DST; 5797 pdst = PF_PEER_SRC; 5798 } 5799 5800 if (th->th_flags & TH_SYN) 5801 if (src->state < TCPS_SYN_SENT) 5802 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5803 if (th->th_flags & TH_FIN) 5804 if (src->state < TCPS_CLOSING) 5805 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5806 if (th->th_flags & TH_ACK) { 5807 if (dst->state == TCPS_SYN_SENT) { 5808 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 5809 if (src->state == TCPS_ESTABLISHED && 5810 (*state)->src_node != NULL && 5811 pf_src_connlimit(state)) { 5812 REASON_SET(reason, PFRES_SRCLIMIT); 5813 return (PF_DROP); 5814 } 5815 } else if (dst->state == TCPS_CLOSING) { 5816 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 5817 } else if (src->state == TCPS_SYN_SENT && 5818 dst->state < TCPS_SYN_SENT) { 5819 /* 5820 * Handle a special sloppy case where we only see one 5821 * half of the connection. If there is a ACK after 5822 * the initial SYN without ever seeing a packet from 5823 * the destination, set the connection to established. 5824 */ 5825 pf_set_protostate(*state, PF_PEER_BOTH, 5826 TCPS_ESTABLISHED); 5827 dst->state = src->state = TCPS_ESTABLISHED; 5828 if ((*state)->src_node != NULL && 5829 pf_src_connlimit(state)) { 5830 REASON_SET(reason, PFRES_SRCLIMIT); 5831 return (PF_DROP); 5832 } 5833 } else if (src->state == TCPS_CLOSING && 5834 dst->state == TCPS_ESTABLISHED && 5835 dst->seqlo == 0) { 5836 /* 5837 * Handle the closing of half connections where we 5838 * don't see the full bidirectional FIN/ACK+ACK 5839 * handshake. 5840 */ 5841 pf_set_protostate(*state, pdst, TCPS_CLOSING); 5842 } 5843 } 5844 if (th->th_flags & TH_RST) 5845 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5846 5847 /* update expire time */ 5848 (*state)->expire = pf_get_uptime(); 5849 if (src->state >= TCPS_FIN_WAIT_2 && 5850 dst->state >= TCPS_FIN_WAIT_2) 5851 (*state)->timeout = PFTM_TCP_CLOSED; 5852 else if (src->state >= TCPS_CLOSING && 5853 dst->state >= TCPS_CLOSING) 5854 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5855 else if (src->state < TCPS_ESTABLISHED || 5856 dst->state < TCPS_ESTABLISHED) 5857 (*state)->timeout = PFTM_TCP_OPENING; 5858 else if (src->state >= TCPS_CLOSING || 5859 dst->state >= TCPS_CLOSING) 5860 (*state)->timeout = PFTM_TCP_CLOSING; 5861 else 5862 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5863 5864 return (PF_PASS); 5865 } 5866 5867 static int 5868 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 5869 { 5870 struct pf_state_key *sk = (*state)->key[pd->didx]; 5871 struct tcphdr *th = &pd->hdr.tcp; 5872 5873 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 5874 if (pd->dir != (*state)->direction) { 5875 REASON_SET(reason, PFRES_SYNPROXY); 5876 return (PF_SYNPROXY_DROP); 5877 } 5878 if (th->th_flags & TH_SYN) { 5879 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 5880 REASON_SET(reason, PFRES_SYNPROXY); 5881 return (PF_DROP); 5882 } 5883 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5884 pd->src, th->th_dport, th->th_sport, 5885 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 5886 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0, 5887 (*state)->act.rtableid); 5888 REASON_SET(reason, PFRES_SYNPROXY); 5889 return (PF_SYNPROXY_DROP); 5890 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 5891 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5892 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5893 REASON_SET(reason, PFRES_SYNPROXY); 5894 return (PF_DROP); 5895 } else if ((*state)->src_node != NULL && 5896 pf_src_connlimit(state)) { 5897 REASON_SET(reason, PFRES_SRCLIMIT); 5898 return (PF_DROP); 5899 } else 5900 pf_set_protostate(*state, PF_PEER_SRC, 5901 PF_TCPS_PROXY_DST); 5902 } 5903 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 5904 if (pd->dir == (*state)->direction) { 5905 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 5906 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5907 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5908 REASON_SET(reason, PFRES_SYNPROXY); 5909 return (PF_DROP); 5910 } 5911 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 5912 if ((*state)->dst.seqhi == 1) 5913 (*state)->dst.seqhi = htonl(arc4random()); 5914 pf_send_tcp((*state)->rule.ptr, pd->af, 5915 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5916 sk->port[pd->sidx], sk->port[pd->didx], 5917 (*state)->dst.seqhi, 0, TH_SYN, 0, 5918 (*state)->src.mss, 0, false, (*state)->tag, 0, 5919 (*state)->act.rtableid); 5920 REASON_SET(reason, PFRES_SYNPROXY); 5921 return (PF_SYNPROXY_DROP); 5922 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 5923 (TH_SYN|TH_ACK)) || 5924 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 5925 REASON_SET(reason, PFRES_SYNPROXY); 5926 return (PF_DROP); 5927 } else { 5928 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 5929 (*state)->dst.seqlo = ntohl(th->th_seq); 5930 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5931 pd->src, th->th_dport, th->th_sport, 5932 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 5933 TH_ACK, (*state)->src.max_win, 0, 0, false, 5934 (*state)->tag, 0, (*state)->act.rtableid); 5935 pf_send_tcp((*state)->rule.ptr, pd->af, 5936 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5937 sk->port[pd->sidx], sk->port[pd->didx], 5938 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 5939 TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0, 5940 (*state)->act.rtableid); 5941 (*state)->src.seqdiff = (*state)->dst.seqhi - 5942 (*state)->src.seqlo; 5943 (*state)->dst.seqdiff = (*state)->src.seqhi - 5944 (*state)->dst.seqlo; 5945 (*state)->src.seqhi = (*state)->src.seqlo + 5946 (*state)->dst.max_win; 5947 (*state)->dst.seqhi = (*state)->dst.seqlo + 5948 (*state)->src.max_win; 5949 (*state)->src.wscale = (*state)->dst.wscale = 0; 5950 pf_set_protostate(*state, PF_PEER_BOTH, 5951 TCPS_ESTABLISHED); 5952 REASON_SET(reason, PFRES_SYNPROXY); 5953 return (PF_SYNPROXY_DROP); 5954 } 5955 } 5956 5957 return (PF_PASS); 5958 } 5959 5960 static int 5961 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif, 5962 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 5963 u_short *reason) 5964 { 5965 struct pf_state_key_cmp key; 5966 struct tcphdr *th = &pd->hdr.tcp; 5967 int copyback = 0; 5968 int action; 5969 struct pf_state_peer *src, *dst; 5970 5971 bzero(&key, sizeof(key)); 5972 key.af = pd->af; 5973 key.proto = IPPROTO_TCP; 5974 if (pd->dir == PF_IN) { /* wire side, straight */ 5975 PF_ACPY(&key.addr[0], pd->src, key.af); 5976 PF_ACPY(&key.addr[1], pd->dst, key.af); 5977 key.port[0] = th->th_sport; 5978 key.port[1] = th->th_dport; 5979 } else { /* stack side, reverse */ 5980 PF_ACPY(&key.addr[1], pd->src, key.af); 5981 PF_ACPY(&key.addr[0], pd->dst, key.af); 5982 key.port[1] = th->th_sport; 5983 key.port[0] = th->th_dport; 5984 } 5985 5986 STATE_LOOKUP(kif, &key, *state, pd); 5987 5988 if (pd->dir == (*state)->direction) { 5989 src = &(*state)->src; 5990 dst = &(*state)->dst; 5991 } else { 5992 src = &(*state)->dst; 5993 dst = &(*state)->src; 5994 } 5995 5996 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 5997 return (action); 5998 5999 if (dst->state >= TCPS_FIN_WAIT_2 && 6000 src->state >= TCPS_FIN_WAIT_2 && 6001 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 6002 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6003 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6004 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6005 printf("pf: state reuse "); 6006 pf_print_state(*state); 6007 pf_print_flags(th->th_flags); 6008 printf("\n"); 6009 } 6010 /* XXX make sure it's the same direction ?? */ 6011 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6012 pf_unlink_state(*state); 6013 *state = NULL; 6014 return (PF_DROP); 6015 } 6016 6017 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6018 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 6019 return (PF_DROP); 6020 } else { 6021 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 6022 ©back) == PF_DROP) 6023 return (PF_DROP); 6024 } 6025 6026 /* translate source/destination address, if necessary */ 6027 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6028 struct pf_state_key *nk = (*state)->key[pd->didx]; 6029 6030 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6031 nk->port[pd->sidx] != th->th_sport) 6032 pf_change_ap(m, pd->src, &th->th_sport, 6033 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 6034 nk->port[pd->sidx], 0, pd->af); 6035 6036 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6037 nk->port[pd->didx] != th->th_dport) 6038 pf_change_ap(m, pd->dst, &th->th_dport, 6039 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 6040 nk->port[pd->didx], 0, pd->af); 6041 copyback = 1; 6042 } 6043 6044 /* Copyback sequence modulation or stateful scrub changes if needed */ 6045 if (copyback) 6046 m_copyback(m, off, sizeof(*th), (caddr_t)th); 6047 6048 return (PF_PASS); 6049 } 6050 6051 static int 6052 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif, 6053 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 6054 { 6055 struct pf_state_peer *src, *dst; 6056 struct pf_state_key_cmp key; 6057 struct udphdr *uh = &pd->hdr.udp; 6058 uint8_t psrc, pdst; 6059 6060 bzero(&key, sizeof(key)); 6061 key.af = pd->af; 6062 key.proto = IPPROTO_UDP; 6063 if (pd->dir == PF_IN) { /* wire side, straight */ 6064 PF_ACPY(&key.addr[0], pd->src, key.af); 6065 PF_ACPY(&key.addr[1], pd->dst, key.af); 6066 key.port[0] = uh->uh_sport; 6067 key.port[1] = uh->uh_dport; 6068 } else { /* stack side, reverse */ 6069 PF_ACPY(&key.addr[1], pd->src, key.af); 6070 PF_ACPY(&key.addr[0], pd->dst, key.af); 6071 key.port[1] = uh->uh_sport; 6072 key.port[0] = uh->uh_dport; 6073 } 6074 6075 STATE_LOOKUP(kif, &key, *state, pd); 6076 6077 if (pd->dir == (*state)->direction) { 6078 src = &(*state)->src; 6079 dst = &(*state)->dst; 6080 psrc = PF_PEER_SRC; 6081 pdst = PF_PEER_DST; 6082 } else { 6083 src = &(*state)->dst; 6084 dst = &(*state)->src; 6085 psrc = PF_PEER_DST; 6086 pdst = PF_PEER_SRC; 6087 } 6088 6089 /* update states */ 6090 if (src->state < PFUDPS_SINGLE) 6091 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 6092 if (dst->state == PFUDPS_SINGLE) 6093 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 6094 6095 /* update expire time */ 6096 (*state)->expire = pf_get_uptime(); 6097 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 6098 (*state)->timeout = PFTM_UDP_MULTIPLE; 6099 else 6100 (*state)->timeout = PFTM_UDP_SINGLE; 6101 6102 /* translate source/destination address, if necessary */ 6103 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6104 struct pf_state_key *nk = (*state)->key[pd->didx]; 6105 6106 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6107 nk->port[pd->sidx] != uh->uh_sport) 6108 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 6109 &uh->uh_sum, &nk->addr[pd->sidx], 6110 nk->port[pd->sidx], 1, pd->af); 6111 6112 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6113 nk->port[pd->didx] != uh->uh_dport) 6114 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 6115 &uh->uh_sum, &nk->addr[pd->didx], 6116 nk->port[pd->didx], 1, pd->af); 6117 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 6118 } 6119 6120 return (PF_PASS); 6121 } 6122 6123 static int 6124 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif, 6125 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 6126 { 6127 struct pf_state_key_cmp key; 6128 struct pf_state_peer *src, *dst; 6129 struct sctphdr *sh = &pd->hdr.sctp; 6130 u_int8_t psrc; //, pdst; 6131 6132 bzero(&key, sizeof(key)); 6133 key.af = pd->af; 6134 key.proto = IPPROTO_SCTP; 6135 if (pd->dir == PF_IN) { /* wire side, straight */ 6136 PF_ACPY(&key.addr[0], pd->src, key.af); 6137 PF_ACPY(&key.addr[1], pd->dst, key.af); 6138 key.port[0] = sh->src_port; 6139 key.port[1] = sh->dest_port; 6140 } else { /* stack side, reverse */ 6141 PF_ACPY(&key.addr[1], pd->src, key.af); 6142 PF_ACPY(&key.addr[0], pd->dst, key.af); 6143 key.port[1] = sh->src_port; 6144 key.port[0] = sh->dest_port; 6145 } 6146 6147 STATE_LOOKUP(kif, &key, *state, pd); 6148 6149 if (pd->dir == (*state)->direction) { 6150 src = &(*state)->src; 6151 dst = &(*state)->dst; 6152 psrc = PF_PEER_SRC; 6153 } else { 6154 src = &(*state)->dst; 6155 dst = &(*state)->src; 6156 psrc = PF_PEER_DST; 6157 } 6158 6159 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 6160 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 6161 pd->sctp_flags & PFDESC_SCTP_INIT) { 6162 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 6163 pf_unlink_state(*state); 6164 *state = NULL; 6165 return (PF_DROP); 6166 } 6167 6168 /* Track state. */ 6169 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 6170 if (src->state < SCTP_COOKIE_WAIT) { 6171 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 6172 (*state)->timeout = PFTM_SCTP_OPENING; 6173 } 6174 } 6175 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 6176 MPASS(dst->scrub != NULL); 6177 if (dst->scrub->pfss_v_tag == 0) 6178 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 6179 } 6180 6181 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 6182 if (src->state < SCTP_ESTABLISHED) { 6183 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 6184 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 6185 } 6186 } 6187 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 6188 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6189 if (src->state < SCTP_SHUTDOWN_PENDING) { 6190 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 6191 (*state)->timeout = PFTM_SCTP_CLOSING; 6192 } 6193 } 6194 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6195 pf_set_protostate(*state, psrc, SCTP_CLOSED); 6196 (*state)->timeout = PFTM_SCTP_CLOSED; 6197 } 6198 6199 if (src->scrub != NULL) { 6200 if (src->scrub->pfss_v_tag == 0) { 6201 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 6202 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 6203 return (PF_DROP); 6204 } 6205 6206 (*state)->expire = pf_get_uptime(); 6207 6208 /* translate source/destination address, if necessary */ 6209 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6210 uint16_t checksum = 0; 6211 struct pf_state_key *nk = (*state)->key[pd->didx]; 6212 6213 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6214 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 6215 pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port, 6216 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 6217 nk->port[pd->sidx], 1, pd->af); 6218 } 6219 6220 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6221 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 6222 pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port, 6223 pd->ip_sum, &checksum, &nk->addr[pd->didx], 6224 nk->port[pd->didx], 1, pd->af); 6225 } 6226 } 6227 6228 return (PF_PASS); 6229 } 6230 6231 static void 6232 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 6233 { 6234 struct pf_sctp_endpoint key; 6235 struct pf_sctp_endpoint *ep; 6236 struct pf_state_key *sks = s->key[PF_SK_STACK]; 6237 struct pf_sctp_source *i, *tmp; 6238 6239 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 6240 return; 6241 6242 PF_SCTP_ENDPOINTS_LOCK(); 6243 6244 key.v_tag = s->dst.scrub->pfss_v_tag; 6245 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6246 if (ep != NULL) { 6247 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6248 if (pf_addr_cmp(&i->addr, 6249 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 6250 s->key[PF_SK_WIRE]->af) == 0) { 6251 SDT_PROBE3(pf, sctp, multihome, remove, 6252 key.v_tag, s, i); 6253 TAILQ_REMOVE(&ep->sources, i, entry); 6254 free(i, M_PFTEMP); 6255 break; 6256 } 6257 } 6258 6259 if (TAILQ_EMPTY(&ep->sources)) { 6260 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6261 free(ep, M_PFTEMP); 6262 } 6263 } 6264 6265 /* Other direction. */ 6266 key.v_tag = s->src.scrub->pfss_v_tag; 6267 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6268 if (ep != NULL) { 6269 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6270 if (pf_addr_cmp(&i->addr, 6271 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6272 s->key[PF_SK_WIRE]->af) == 0) { 6273 SDT_PROBE3(pf, sctp, multihome, remove, 6274 key.v_tag, s, i); 6275 TAILQ_REMOVE(&ep->sources, i, entry); 6276 free(i, M_PFTEMP); 6277 break; 6278 } 6279 } 6280 6281 if (TAILQ_EMPTY(&ep->sources)) { 6282 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6283 free(ep, M_PFTEMP); 6284 } 6285 } 6286 6287 PF_SCTP_ENDPOINTS_UNLOCK(); 6288 } 6289 6290 static void 6291 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6292 { 6293 struct pf_sctp_endpoint key = { 6294 .v_tag = v_tag, 6295 }; 6296 struct pf_sctp_source *i; 6297 struct pf_sctp_endpoint *ep; 6298 6299 PF_SCTP_ENDPOINTS_LOCK(); 6300 6301 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6302 if (ep == NULL) { 6303 ep = malloc(sizeof(struct pf_sctp_endpoint), 6304 M_PFTEMP, M_NOWAIT); 6305 if (ep == NULL) { 6306 PF_SCTP_ENDPOINTS_UNLOCK(); 6307 return; 6308 } 6309 6310 ep->v_tag = v_tag; 6311 TAILQ_INIT(&ep->sources); 6312 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6313 } 6314 6315 /* Avoid inserting duplicates. */ 6316 TAILQ_FOREACH(i, &ep->sources, entry) { 6317 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6318 PF_SCTP_ENDPOINTS_UNLOCK(); 6319 return; 6320 } 6321 } 6322 6323 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6324 if (i == NULL) { 6325 PF_SCTP_ENDPOINTS_UNLOCK(); 6326 return; 6327 } 6328 6329 i->af = pd->af; 6330 memcpy(&i->addr, a, sizeof(*a)); 6331 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6332 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 6333 6334 PF_SCTP_ENDPOINTS_UNLOCK(); 6335 } 6336 6337 static void 6338 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif, 6339 struct pf_kstate *s, int action) 6340 { 6341 struct pf_sctp_multihome_job *j, *tmp; 6342 struct pf_sctp_source *i; 6343 int ret __unused; 6344 struct pf_kstate *sm = NULL; 6345 struct pf_krule *ra = NULL; 6346 struct pf_krule *r = &V_pf_default_rule; 6347 struct pf_kruleset *rs = NULL; 6348 bool do_extra = true; 6349 6350 PF_RULES_RLOCK_TRACKER; 6351 6352 again: 6353 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6354 if (s == NULL || action != PF_PASS) 6355 goto free; 6356 6357 /* Confirm we don't recurse here. */ 6358 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6359 6360 switch (j->op) { 6361 case SCTP_ADD_IP_ADDRESS: { 6362 uint32_t v_tag = pd->sctp_initiate_tag; 6363 6364 if (v_tag == 0) { 6365 if (s->direction == pd->dir) 6366 v_tag = s->src.scrub->pfss_v_tag; 6367 else 6368 v_tag = s->dst.scrub->pfss_v_tag; 6369 } 6370 6371 /* 6372 * Avoid duplicating states. We'll already have 6373 * created a state based on the source address of 6374 * the packet, but SCTP endpoints may also list this 6375 * address again in the INIT(_ACK) parameters. 6376 */ 6377 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6378 break; 6379 } 6380 6381 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6382 PF_RULES_RLOCK(); 6383 sm = NULL; 6384 /* 6385 * New connections need to be floating, because 6386 * we cannot know what interfaces it will use. 6387 * That's why we pass V_pfi_all rather than kif. 6388 */ 6389 ret = pf_test_rule(&r, &sm, V_pfi_all, 6390 j->m, off, &j->pd, &ra, &rs, NULL); 6391 PF_RULES_RUNLOCK(); 6392 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret); 6393 if (ret != PF_DROP && sm != NULL) { 6394 /* Inherit v_tag values. */ 6395 if (sm->direction == s->direction) { 6396 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6397 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6398 } else { 6399 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6400 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6401 } 6402 PF_STATE_UNLOCK(sm); 6403 } else { 6404 /* If we try duplicate inserts? */ 6405 break; 6406 } 6407 6408 /* Only add the address if we've actually allowed the state. */ 6409 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6410 6411 if (! do_extra) { 6412 break; 6413 } 6414 /* 6415 * We need to do this for each of our source addresses. 6416 * Find those based on the verification tag. 6417 */ 6418 struct pf_sctp_endpoint key = { 6419 .v_tag = pd->hdr.sctp.v_tag, 6420 }; 6421 struct pf_sctp_endpoint *ep; 6422 6423 PF_SCTP_ENDPOINTS_LOCK(); 6424 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6425 if (ep == NULL) { 6426 PF_SCTP_ENDPOINTS_UNLOCK(); 6427 break; 6428 } 6429 MPASS(ep != NULL); 6430 6431 TAILQ_FOREACH(i, &ep->sources, entry) { 6432 struct pf_sctp_multihome_job *nj; 6433 6434 /* SCTP can intermingle IPv4 and IPv6. */ 6435 if (i->af != pd->af) 6436 continue; 6437 6438 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6439 if (! nj) { 6440 continue; 6441 } 6442 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6443 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6444 nj->pd.src = &nj->src; 6445 // New destination address! 6446 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6447 nj->pd.dst = &nj->dst; 6448 nj->m = j->m; 6449 nj->op = j->op; 6450 6451 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6452 } 6453 PF_SCTP_ENDPOINTS_UNLOCK(); 6454 6455 break; 6456 } 6457 case SCTP_DEL_IP_ADDRESS: { 6458 struct pf_state_key_cmp key; 6459 uint8_t psrc; 6460 6461 bzero(&key, sizeof(key)); 6462 key.af = j->pd.af; 6463 key.proto = IPPROTO_SCTP; 6464 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6465 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6466 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6467 key.port[0] = j->pd.hdr.sctp.src_port; 6468 key.port[1] = j->pd.hdr.sctp.dest_port; 6469 } else { /* stack side, reverse */ 6470 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6471 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6472 key.port[1] = j->pd.hdr.sctp.src_port; 6473 key.port[0] = j->pd.hdr.sctp.dest_port; 6474 } 6475 6476 sm = pf_find_state(kif, &key, j->pd.dir); 6477 if (sm != NULL) { 6478 PF_STATE_LOCK_ASSERT(sm); 6479 if (j->pd.dir == sm->direction) { 6480 psrc = PF_PEER_SRC; 6481 } else { 6482 psrc = PF_PEER_DST; 6483 } 6484 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6485 sm->timeout = PFTM_SCTP_CLOSING; 6486 PF_STATE_UNLOCK(sm); 6487 } 6488 break; 6489 default: 6490 panic("Unknown op %#x", j->op); 6491 } 6492 } 6493 6494 free: 6495 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6496 free(j, M_PFTEMP); 6497 } 6498 6499 /* We may have inserted extra work while processing the list. */ 6500 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6501 do_extra = false; 6502 goto again; 6503 } 6504 } 6505 6506 static int 6507 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6508 struct pfi_kkif *kif, int op) 6509 { 6510 int off = 0; 6511 struct pf_sctp_multihome_job *job; 6512 6513 while (off < len) { 6514 struct sctp_paramhdr h; 6515 6516 if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL, 6517 pd->af)) 6518 return (PF_DROP); 6519 6520 /* Parameters are at least 4 bytes. */ 6521 if (ntohs(h.param_length) < 4) 6522 return (PF_DROP); 6523 6524 switch (ntohs(h.param_type)) { 6525 case SCTP_IPV4_ADDRESS: { 6526 struct in_addr t; 6527 6528 if (ntohs(h.param_length) != 6529 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6530 return (PF_DROP); 6531 6532 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6533 NULL, NULL, pd->af)) 6534 return (PF_DROP); 6535 6536 if (in_nullhost(t)) 6537 t.s_addr = pd->src->v4.s_addr; 6538 6539 /* 6540 * We hold the state lock (idhash) here, which means 6541 * that we can't acquire the keyhash, or we'll get a 6542 * LOR (and potentially double-lock things too). We also 6543 * can't release the state lock here, so instead we'll 6544 * enqueue this for async handling. 6545 * There's a relatively small race here, in that a 6546 * packet using the new addresses could arrive already, 6547 * but that's just though luck for it. 6548 */ 6549 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6550 if (! job) 6551 return (PF_DROP); 6552 6553 memcpy(&job->pd, pd, sizeof(*pd)); 6554 6555 // New source address! 6556 memcpy(&job->src, &t, sizeof(t)); 6557 job->pd.src = &job->src; 6558 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6559 job->pd.dst = &job->dst; 6560 job->m = m; 6561 job->op = op; 6562 6563 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6564 break; 6565 } 6566 #ifdef INET6 6567 case SCTP_IPV6_ADDRESS: { 6568 struct in6_addr t; 6569 6570 if (ntohs(h.param_length) != 6571 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6572 return (PF_DROP); 6573 6574 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6575 NULL, NULL, pd->af)) 6576 return (PF_DROP); 6577 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6578 break; 6579 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6580 memcpy(&t, &pd->src->v6, sizeof(t)); 6581 6582 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6583 if (! job) 6584 return (PF_DROP); 6585 6586 memcpy(&job->pd, pd, sizeof(*pd)); 6587 memcpy(&job->src, &t, sizeof(t)); 6588 job->pd.src = &job->src; 6589 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6590 job->pd.dst = &job->dst; 6591 job->m = m; 6592 job->op = op; 6593 6594 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6595 break; 6596 } 6597 #endif 6598 case SCTP_ADD_IP_ADDRESS: { 6599 int ret; 6600 struct sctp_asconf_paramhdr ah; 6601 6602 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6603 NULL, NULL, pd->af)) 6604 return (PF_DROP); 6605 6606 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6607 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6608 SCTP_ADD_IP_ADDRESS); 6609 if (ret != PF_PASS) 6610 return (ret); 6611 break; 6612 } 6613 case SCTP_DEL_IP_ADDRESS: { 6614 int ret; 6615 struct sctp_asconf_paramhdr ah; 6616 6617 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6618 NULL, NULL, pd->af)) 6619 return (PF_DROP); 6620 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6621 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6622 SCTP_DEL_IP_ADDRESS); 6623 if (ret != PF_PASS) 6624 return (ret); 6625 break; 6626 } 6627 default: 6628 break; 6629 } 6630 6631 off += roundup(ntohs(h.param_length), 4); 6632 } 6633 6634 return (PF_PASS); 6635 } 6636 int 6637 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6638 struct pfi_kkif *kif) 6639 { 6640 start += sizeof(struct sctp_init_chunk); 6641 len -= sizeof(struct sctp_init_chunk); 6642 6643 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6644 } 6645 6646 int 6647 pf_multihome_scan_asconf(struct mbuf *m, int start, int len, 6648 struct pf_pdesc *pd, struct pfi_kkif *kif) 6649 { 6650 start += sizeof(struct sctp_asconf_chunk); 6651 len -= sizeof(struct sctp_asconf_chunk); 6652 6653 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6654 } 6655 6656 int 6657 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 6658 struct pf_kstate **state, struct mbuf *m, int direction, struct pfi_kkif *kif, 6659 u_int16_t icmpid, u_int16_t type, int icmp_dir, int *iidx, int multi, 6660 int inner) 6661 { 6662 key->af = pd->af; 6663 key->proto = pd->proto; 6664 if (icmp_dir == PF_IN) { 6665 *iidx = pd->sidx; 6666 key->port[pd->sidx] = icmpid; 6667 key->port[pd->didx] = type; 6668 } else { 6669 *iidx = pd->didx; 6670 key->port[pd->sidx] = type; 6671 key->port[pd->didx] = icmpid; 6672 } 6673 if (pd->af == AF_INET6 && multi != PF_ICMP_MULTI_NONE) { 6674 switch (multi) { 6675 case PF_ICMP_MULTI_SOLICITED: 6676 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 6677 key->addr[pd->sidx].addr32[1] = 0; 6678 key->addr[pd->sidx].addr32[2] = IPV6_ADDR_INT32_ONE; 6679 key->addr[pd->sidx].addr32[3] = pd->src->addr32[3]; 6680 key->addr[pd->sidx].addr8[12] = 0xff; 6681 break; 6682 case PF_ICMP_MULTI_LINK: 6683 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 6684 key->addr[pd->sidx].addr32[1] = 0; 6685 key->addr[pd->sidx].addr32[2] = 0; 6686 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 6687 break; 6688 } 6689 } else 6690 PF_ACPY(&key->addr[pd->sidx], pd->src, key->af); 6691 PF_ACPY(&key->addr[pd->didx], pd->dst, key->af); 6692 6693 STATE_LOOKUP(kif, key, *state, pd); 6694 6695 /* Is this ICMP message flowing in right direction? */ 6696 if ((*state)->rule.ptr->type && 6697 (((!inner && (*state)->direction == direction) || 6698 (inner && (*state)->direction != direction)) ? 6699 PF_IN : PF_OUT) != icmp_dir) { 6700 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6701 printf("pf: icmp type %d in wrong direction (%d): ", 6702 icmp_dir, pd->dir); 6703 pf_print_state(*state); 6704 printf("\n"); 6705 } 6706 return (PF_DROP); 6707 } 6708 return (-1); 6709 } 6710 6711 static int 6712 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif, 6713 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 6714 { 6715 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6716 u_int16_t *icmpsum, virtual_id, virtual_type; 6717 u_int8_t icmptype, icmpcode; 6718 int icmp_dir, iidx, ret, multi; 6719 struct pf_state_key_cmp key; 6720 #ifdef INET 6721 u_int16_t icmpid; 6722 #endif 6723 6724 MPASS(*state == NULL); 6725 6726 bzero(&key, sizeof(key)); 6727 switch (pd->proto) { 6728 #ifdef INET 6729 case IPPROTO_ICMP: 6730 icmptype = pd->hdr.icmp.icmp_type; 6731 icmpcode = pd->hdr.icmp.icmp_code; 6732 icmpid = pd->hdr.icmp.icmp_id; 6733 icmpsum = &pd->hdr.icmp.icmp_cksum; 6734 break; 6735 #endif /* INET */ 6736 #ifdef INET6 6737 case IPPROTO_ICMPV6: 6738 icmptype = pd->hdr.icmp6.icmp6_type; 6739 icmpcode = pd->hdr.icmp6.icmp6_code; 6740 #ifdef INET 6741 icmpid = pd->hdr.icmp6.icmp6_id; 6742 #endif 6743 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 6744 break; 6745 #endif /* INET6 */ 6746 } 6747 6748 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi, 6749 &virtual_id, &virtual_type) == 0) { 6750 /* 6751 * ICMP query/reply message not related to a TCP/UDP packet. 6752 * Search for an ICMP state. 6753 */ 6754 ret = pf_icmp_state_lookup(&key, pd, state, m, pd->dir, 6755 kif, virtual_id, virtual_type, icmp_dir, &iidx, 6756 PF_ICMP_MULTI_NONE, 0); 6757 if (ret >= 0) { 6758 if (ret == PF_DROP && pd->af == AF_INET6 && 6759 icmp_dir == PF_OUT) { 6760 if (*state != NULL) 6761 PF_STATE_UNLOCK((*state)); 6762 ret = pf_icmp_state_lookup(&key, pd, state, m, 6763 pd->dir, kif, virtual_id, virtual_type, 6764 icmp_dir, &iidx, multi, 0); 6765 if (ret >= 0) 6766 return (ret); 6767 } else 6768 return (ret); 6769 } 6770 6771 (*state)->expire = pf_get_uptime(); 6772 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 6773 6774 /* translate source/destination address, if necessary */ 6775 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6776 struct pf_state_key *nk = (*state)->key[pd->didx]; 6777 6778 switch (pd->af) { 6779 #ifdef INET 6780 case AF_INET: 6781 if (PF_ANEQ(pd->src, 6782 &nk->addr[pd->sidx], AF_INET)) 6783 pf_change_a(&saddr->v4.s_addr, 6784 pd->ip_sum, 6785 nk->addr[pd->sidx].v4.s_addr, 0); 6786 6787 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 6788 AF_INET)) 6789 pf_change_a(&daddr->v4.s_addr, 6790 pd->ip_sum, 6791 nk->addr[pd->didx].v4.s_addr, 0); 6792 6793 if (nk->port[iidx] != 6794 pd->hdr.icmp.icmp_id) { 6795 pd->hdr.icmp.icmp_cksum = 6796 pf_cksum_fixup( 6797 pd->hdr.icmp.icmp_cksum, icmpid, 6798 nk->port[iidx], 0); 6799 pd->hdr.icmp.icmp_id = 6800 nk->port[iidx]; 6801 } 6802 6803 m_copyback(m, off, ICMP_MINLEN, 6804 (caddr_t )&pd->hdr.icmp); 6805 break; 6806 #endif /* INET */ 6807 #ifdef INET6 6808 case AF_INET6: 6809 if (PF_ANEQ(pd->src, 6810 &nk->addr[pd->sidx], AF_INET6)) 6811 pf_change_a6(saddr, 6812 &pd->hdr.icmp6.icmp6_cksum, 6813 &nk->addr[pd->sidx], 0); 6814 6815 if (PF_ANEQ(pd->dst, 6816 &nk->addr[pd->didx], AF_INET6)) 6817 pf_change_a6(daddr, 6818 &pd->hdr.icmp6.icmp6_cksum, 6819 &nk->addr[pd->didx], 0); 6820 6821 m_copyback(m, off, sizeof(struct icmp6_hdr), 6822 (caddr_t )&pd->hdr.icmp6); 6823 break; 6824 #endif /* INET6 */ 6825 } 6826 } 6827 return (PF_PASS); 6828 6829 } else { 6830 /* 6831 * ICMP error message in response to a TCP/UDP packet. 6832 * Extract the inner TCP/UDP header and search for that state. 6833 */ 6834 6835 struct pf_pdesc pd2; 6836 bzero(&pd2, sizeof pd2); 6837 #ifdef INET 6838 struct ip h2; 6839 #endif /* INET */ 6840 #ifdef INET6 6841 struct ip6_hdr h2_6; 6842 int terminal = 0; 6843 #endif /* INET6 */ 6844 int ipoff2 = 0; 6845 int off2 = 0; 6846 6847 pd2.af = pd->af; 6848 pd2.dir = pd->dir; 6849 /* Payload packet is from the opposite direction. */ 6850 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 6851 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 6852 switch (pd->af) { 6853 #ifdef INET 6854 case AF_INET: 6855 /* offset of h2 in mbuf chain */ 6856 ipoff2 = off + ICMP_MINLEN; 6857 6858 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 6859 NULL, reason, pd2.af)) { 6860 DPFPRINTF(PF_DEBUG_MISC, 6861 ("pf: ICMP error message too short " 6862 "(ip)\n")); 6863 return (PF_DROP); 6864 } 6865 /* 6866 * ICMP error messages don't refer to non-first 6867 * fragments 6868 */ 6869 if (h2.ip_off & htons(IP_OFFMASK)) { 6870 REASON_SET(reason, PFRES_FRAG); 6871 return (PF_DROP); 6872 } 6873 6874 /* offset of protocol header that follows h2 */ 6875 off2 = ipoff2 + (h2.ip_hl << 2); 6876 6877 pd2.proto = h2.ip_p; 6878 pd2.src = (struct pf_addr *)&h2.ip_src; 6879 pd2.dst = (struct pf_addr *)&h2.ip_dst; 6880 pd2.ip_sum = &h2.ip_sum; 6881 break; 6882 #endif /* INET */ 6883 #ifdef INET6 6884 case AF_INET6: 6885 ipoff2 = off + sizeof(struct icmp6_hdr); 6886 6887 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 6888 NULL, reason, pd2.af)) { 6889 DPFPRINTF(PF_DEBUG_MISC, 6890 ("pf: ICMP error message too short " 6891 "(ip6)\n")); 6892 return (PF_DROP); 6893 } 6894 pd2.proto = h2_6.ip6_nxt; 6895 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 6896 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 6897 pd2.ip_sum = NULL; 6898 off2 = ipoff2 + sizeof(h2_6); 6899 do { 6900 switch (pd2.proto) { 6901 case IPPROTO_FRAGMENT: 6902 /* 6903 * ICMPv6 error messages for 6904 * non-first fragments 6905 */ 6906 REASON_SET(reason, PFRES_FRAG); 6907 return (PF_DROP); 6908 case IPPROTO_AH: 6909 case IPPROTO_HOPOPTS: 6910 case IPPROTO_ROUTING: 6911 case IPPROTO_DSTOPTS: { 6912 /* get next header and header length */ 6913 struct ip6_ext opt6; 6914 6915 if (!pf_pull_hdr(m, off2, &opt6, 6916 sizeof(opt6), NULL, reason, 6917 pd2.af)) { 6918 DPFPRINTF(PF_DEBUG_MISC, 6919 ("pf: ICMPv6 short opt\n")); 6920 return (PF_DROP); 6921 } 6922 if (pd2.proto == IPPROTO_AH) 6923 off2 += (opt6.ip6e_len + 2) * 4; 6924 else 6925 off2 += (opt6.ip6e_len + 1) * 8; 6926 pd2.proto = opt6.ip6e_nxt; 6927 /* goto the next header */ 6928 break; 6929 } 6930 default: 6931 terminal++; 6932 break; 6933 } 6934 } while (!terminal); 6935 break; 6936 #endif /* INET6 */ 6937 } 6938 6939 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 6940 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6941 printf("pf: BAD ICMP %d:%d outer dst: ", 6942 icmptype, icmpcode); 6943 pf_print_host(pd->src, 0, pd->af); 6944 printf(" -> "); 6945 pf_print_host(pd->dst, 0, pd->af); 6946 printf(" inner src: "); 6947 pf_print_host(pd2.src, 0, pd2.af); 6948 printf(" -> "); 6949 pf_print_host(pd2.dst, 0, pd2.af); 6950 printf("\n"); 6951 } 6952 REASON_SET(reason, PFRES_BADSTATE); 6953 return (PF_DROP); 6954 } 6955 6956 switch (pd2.proto) { 6957 case IPPROTO_TCP: { 6958 struct tcphdr th; 6959 u_int32_t seq; 6960 struct pf_state_peer *src, *dst; 6961 u_int8_t dws; 6962 int copyback = 0; 6963 6964 /* 6965 * Only the first 8 bytes of the TCP header can be 6966 * expected. Don't access any TCP header fields after 6967 * th_seq, an ackskew test is not possible. 6968 */ 6969 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 6970 pd2.af)) { 6971 DPFPRINTF(PF_DEBUG_MISC, 6972 ("pf: ICMP error message too short " 6973 "(tcp)\n")); 6974 return (PF_DROP); 6975 } 6976 6977 key.af = pd2.af; 6978 key.proto = IPPROTO_TCP; 6979 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6980 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6981 key.port[pd2.sidx] = th.th_sport; 6982 key.port[pd2.didx] = th.th_dport; 6983 6984 STATE_LOOKUP(kif, &key, *state, pd); 6985 6986 if (pd->dir == (*state)->direction) { 6987 src = &(*state)->dst; 6988 dst = &(*state)->src; 6989 } else { 6990 src = &(*state)->src; 6991 dst = &(*state)->dst; 6992 } 6993 6994 if (src->wscale && dst->wscale) 6995 dws = dst->wscale & PF_WSCALE_MASK; 6996 else 6997 dws = 0; 6998 6999 /* Demodulate sequence number */ 7000 seq = ntohl(th.th_seq) - src->seqdiff; 7001 if (src->seqdiff) { 7002 pf_change_a(&th.th_seq, icmpsum, 7003 htonl(seq), 0); 7004 copyback = 1; 7005 } 7006 7007 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7008 (!SEQ_GEQ(src->seqhi, seq) || 7009 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7010 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7011 printf("pf: BAD ICMP %d:%d ", 7012 icmptype, icmpcode); 7013 pf_print_host(pd->src, 0, pd->af); 7014 printf(" -> "); 7015 pf_print_host(pd->dst, 0, pd->af); 7016 printf(" state: "); 7017 pf_print_state(*state); 7018 printf(" seq=%u\n", seq); 7019 } 7020 REASON_SET(reason, PFRES_BADSTATE); 7021 return (PF_DROP); 7022 } else { 7023 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7024 printf("pf: OK ICMP %d:%d ", 7025 icmptype, icmpcode); 7026 pf_print_host(pd->src, 0, pd->af); 7027 printf(" -> "); 7028 pf_print_host(pd->dst, 0, pd->af); 7029 printf(" state: "); 7030 pf_print_state(*state); 7031 printf(" seq=%u\n", seq); 7032 } 7033 } 7034 7035 /* translate source/destination address, if necessary */ 7036 if ((*state)->key[PF_SK_WIRE] != 7037 (*state)->key[PF_SK_STACK]) { 7038 struct pf_state_key *nk = 7039 (*state)->key[pd->didx]; 7040 7041 if (PF_ANEQ(pd2.src, 7042 &nk->addr[pd2.sidx], pd2.af) || 7043 nk->port[pd2.sidx] != th.th_sport) 7044 pf_change_icmp(pd2.src, &th.th_sport, 7045 daddr, &nk->addr[pd2.sidx], 7046 nk->port[pd2.sidx], NULL, 7047 pd2.ip_sum, icmpsum, 7048 pd->ip_sum, 0, pd2.af); 7049 7050 if (PF_ANEQ(pd2.dst, 7051 &nk->addr[pd2.didx], pd2.af) || 7052 nk->port[pd2.didx] != th.th_dport) 7053 pf_change_icmp(pd2.dst, &th.th_dport, 7054 saddr, &nk->addr[pd2.didx], 7055 nk->port[pd2.didx], NULL, 7056 pd2.ip_sum, icmpsum, 7057 pd->ip_sum, 0, pd2.af); 7058 copyback = 1; 7059 } 7060 7061 if (copyback) { 7062 switch (pd2.af) { 7063 #ifdef INET 7064 case AF_INET: 7065 m_copyback(m, off, ICMP_MINLEN, 7066 (caddr_t )&pd->hdr.icmp); 7067 m_copyback(m, ipoff2, sizeof(h2), 7068 (caddr_t )&h2); 7069 break; 7070 #endif /* INET */ 7071 #ifdef INET6 7072 case AF_INET6: 7073 m_copyback(m, off, 7074 sizeof(struct icmp6_hdr), 7075 (caddr_t )&pd->hdr.icmp6); 7076 m_copyback(m, ipoff2, sizeof(h2_6), 7077 (caddr_t )&h2_6); 7078 break; 7079 #endif /* INET6 */ 7080 } 7081 m_copyback(m, off2, 8, (caddr_t)&th); 7082 } 7083 7084 return (PF_PASS); 7085 break; 7086 } 7087 case IPPROTO_UDP: { 7088 struct udphdr uh; 7089 7090 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 7091 NULL, reason, pd2.af)) { 7092 DPFPRINTF(PF_DEBUG_MISC, 7093 ("pf: ICMP error message too short " 7094 "(udp)\n")); 7095 return (PF_DROP); 7096 } 7097 7098 key.af = pd2.af; 7099 key.proto = IPPROTO_UDP; 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] = uh.uh_sport; 7103 key.port[pd2.didx] = uh.uh_dport; 7104 7105 STATE_LOOKUP(kif, &key, *state, pd); 7106 7107 /* translate source/destination address, if necessary */ 7108 if ((*state)->key[PF_SK_WIRE] != 7109 (*state)->key[PF_SK_STACK]) { 7110 struct pf_state_key *nk = 7111 (*state)->key[pd->didx]; 7112 7113 if (PF_ANEQ(pd2.src, 7114 &nk->addr[pd2.sidx], pd2.af) || 7115 nk->port[pd2.sidx] != uh.uh_sport) 7116 pf_change_icmp(pd2.src, &uh.uh_sport, 7117 daddr, &nk->addr[pd2.sidx], 7118 nk->port[pd2.sidx], &uh.uh_sum, 7119 pd2.ip_sum, icmpsum, 7120 pd->ip_sum, 1, pd2.af); 7121 7122 if (PF_ANEQ(pd2.dst, 7123 &nk->addr[pd2.didx], pd2.af) || 7124 nk->port[pd2.didx] != uh.uh_dport) 7125 pf_change_icmp(pd2.dst, &uh.uh_dport, 7126 saddr, &nk->addr[pd2.didx], 7127 nk->port[pd2.didx], &uh.uh_sum, 7128 pd2.ip_sum, icmpsum, 7129 pd->ip_sum, 1, pd2.af); 7130 7131 switch (pd2.af) { 7132 #ifdef INET 7133 case AF_INET: 7134 m_copyback(m, off, ICMP_MINLEN, 7135 (caddr_t )&pd->hdr.icmp); 7136 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7137 break; 7138 #endif /* INET */ 7139 #ifdef INET6 7140 case AF_INET6: 7141 m_copyback(m, off, 7142 sizeof(struct icmp6_hdr), 7143 (caddr_t )&pd->hdr.icmp6); 7144 m_copyback(m, ipoff2, sizeof(h2_6), 7145 (caddr_t )&h2_6); 7146 break; 7147 #endif /* INET6 */ 7148 } 7149 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 7150 } 7151 return (PF_PASS); 7152 break; 7153 } 7154 #ifdef INET 7155 case IPPROTO_ICMP: { 7156 struct icmp *iih = &pd2.hdr.icmp; 7157 7158 if (!pf_pull_hdr(m, off2, iih, ICMP_MINLEN, 7159 NULL, reason, pd2.af)) { 7160 DPFPRINTF(PF_DEBUG_MISC, 7161 ("pf: ICMP error message too short i" 7162 "(icmp)\n")); 7163 return (PF_DROP); 7164 } 7165 7166 icmpid = iih->icmp_id; 7167 pf_icmp_mapping(&pd2, iih->icmp_type, 7168 &icmp_dir, &multi, &virtual_id, &virtual_type); 7169 7170 ret = pf_icmp_state_lookup(&key, &pd2, state, m, 7171 pd2.dir, kif, virtual_id, virtual_type, 7172 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7173 if (ret >= 0) 7174 return (ret); 7175 7176 /* translate source/destination address, if necessary */ 7177 if ((*state)->key[PF_SK_WIRE] != 7178 (*state)->key[PF_SK_STACK]) { 7179 struct pf_state_key *nk = 7180 (*state)->key[pd->didx]; 7181 7182 if (PF_ANEQ(pd2.src, 7183 &nk->addr[pd2.sidx], pd2.af) || 7184 (virtual_type == htons(ICMP_ECHO) && 7185 nk->port[iidx] != iih->icmp_id)) 7186 pf_change_icmp(pd2.src, 7187 (virtual_type == htons(ICMP_ECHO)) ? 7188 &iih->icmp_id : NULL, 7189 daddr, &nk->addr[pd2.sidx], 7190 (virtual_type == htons(ICMP_ECHO)) ? 7191 nk->port[iidx] : 0, NULL, 7192 pd2.ip_sum, icmpsum, 7193 pd->ip_sum, 0, AF_INET); 7194 7195 if (PF_ANEQ(pd2.dst, 7196 &nk->addr[pd2.didx], pd2.af)) 7197 pf_change_icmp(pd2.dst, NULL, NULL, 7198 &nk->addr[pd2.didx], 0, NULL, 7199 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 7200 AF_INET); 7201 7202 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 7203 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7204 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)iih); 7205 } 7206 return (PF_PASS); 7207 break; 7208 } 7209 #endif /* INET */ 7210 #ifdef INET6 7211 case IPPROTO_ICMPV6: { 7212 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 7213 7214 if (!pf_pull_hdr(m, off2, iih, 7215 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 7216 DPFPRINTF(PF_DEBUG_MISC, 7217 ("pf: ICMP error message too short " 7218 "(icmp6)\n")); 7219 return (PF_DROP); 7220 } 7221 7222 pf_icmp_mapping(&pd2, iih->icmp6_type, 7223 &icmp_dir, &multi, &virtual_id, &virtual_type); 7224 7225 ret = pf_icmp_state_lookup(&key, &pd2, state, m, 7226 pd->dir, kif, virtual_id, virtual_type, 7227 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7228 if (ret >= 0) { 7229 if (ret == PF_DROP && pd->af == AF_INET6 && 7230 icmp_dir == PF_OUT) { 7231 if (*state != NULL) 7232 PF_STATE_UNLOCK((*state)); 7233 ret = pf_icmp_state_lookup(&key, pd, 7234 state, m, pd->dir, kif, 7235 virtual_id, virtual_type, 7236 icmp_dir, &iidx, multi, 1); 7237 if (ret >= 0) 7238 return (ret); 7239 } else 7240 return (ret); 7241 } 7242 7243 /* translate source/destination address, if necessary */ 7244 if ((*state)->key[PF_SK_WIRE] != 7245 (*state)->key[PF_SK_STACK]) { 7246 struct pf_state_key *nk = 7247 (*state)->key[pd->didx]; 7248 7249 if (PF_ANEQ(pd2.src, 7250 &nk->addr[pd2.sidx], pd2.af) || 7251 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 7252 nk->port[pd2.sidx] != iih->icmp6_id)) 7253 pf_change_icmp(pd2.src, 7254 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7255 ? &iih->icmp6_id : NULL, 7256 daddr, &nk->addr[pd2.sidx], 7257 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7258 ? nk->port[iidx] : 0, NULL, 7259 pd2.ip_sum, icmpsum, 7260 pd->ip_sum, 0, AF_INET6); 7261 7262 if (PF_ANEQ(pd2.dst, 7263 &nk->addr[pd2.didx], pd2.af)) 7264 pf_change_icmp(pd2.dst, NULL, NULL, 7265 &nk->addr[pd2.didx], 0, NULL, 7266 pd2.ip_sum, icmpsum, 7267 pd->ip_sum, 0, AF_INET6); 7268 7269 m_copyback(m, off, sizeof(struct icmp6_hdr), 7270 (caddr_t)&pd->hdr.icmp6); 7271 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 7272 m_copyback(m, off2, sizeof(struct icmp6_hdr), 7273 (caddr_t)iih); 7274 } 7275 return (PF_PASS); 7276 break; 7277 } 7278 #endif /* INET6 */ 7279 default: { 7280 key.af = pd2.af; 7281 key.proto = pd2.proto; 7282 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7283 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7284 key.port[0] = key.port[1] = 0; 7285 7286 STATE_LOOKUP(kif, &key, *state, pd); 7287 7288 /* translate source/destination address, if necessary */ 7289 if ((*state)->key[PF_SK_WIRE] != 7290 (*state)->key[PF_SK_STACK]) { 7291 struct pf_state_key *nk = 7292 (*state)->key[pd->didx]; 7293 7294 if (PF_ANEQ(pd2.src, 7295 &nk->addr[pd2.sidx], pd2.af)) 7296 pf_change_icmp(pd2.src, NULL, daddr, 7297 &nk->addr[pd2.sidx], 0, NULL, 7298 pd2.ip_sum, icmpsum, 7299 pd->ip_sum, 0, pd2.af); 7300 7301 if (PF_ANEQ(pd2.dst, 7302 &nk->addr[pd2.didx], pd2.af)) 7303 pf_change_icmp(pd2.dst, NULL, saddr, 7304 &nk->addr[pd2.didx], 0, NULL, 7305 pd2.ip_sum, icmpsum, 7306 pd->ip_sum, 0, pd2.af); 7307 7308 switch (pd2.af) { 7309 #ifdef INET 7310 case AF_INET: 7311 m_copyback(m, off, ICMP_MINLEN, 7312 (caddr_t)&pd->hdr.icmp); 7313 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7314 break; 7315 #endif /* INET */ 7316 #ifdef INET6 7317 case AF_INET6: 7318 m_copyback(m, off, 7319 sizeof(struct icmp6_hdr), 7320 (caddr_t )&pd->hdr.icmp6); 7321 m_copyback(m, ipoff2, sizeof(h2_6), 7322 (caddr_t )&h2_6); 7323 break; 7324 #endif /* INET6 */ 7325 } 7326 } 7327 return (PF_PASS); 7328 break; 7329 } 7330 } 7331 } 7332 } 7333 7334 static int 7335 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif, 7336 struct mbuf *m, struct pf_pdesc *pd) 7337 { 7338 struct pf_state_peer *src, *dst; 7339 struct pf_state_key_cmp key; 7340 uint8_t psrc, pdst; 7341 7342 bzero(&key, sizeof(key)); 7343 key.af = pd->af; 7344 key.proto = pd->proto; 7345 if (pd->dir == PF_IN) { 7346 PF_ACPY(&key.addr[0], pd->src, key.af); 7347 PF_ACPY(&key.addr[1], pd->dst, key.af); 7348 key.port[0] = key.port[1] = 0; 7349 } else { 7350 PF_ACPY(&key.addr[1], pd->src, key.af); 7351 PF_ACPY(&key.addr[0], pd->dst, key.af); 7352 key.port[1] = key.port[0] = 0; 7353 } 7354 7355 STATE_LOOKUP(kif, &key, *state, pd); 7356 7357 if (pd->dir == (*state)->direction) { 7358 src = &(*state)->src; 7359 dst = &(*state)->dst; 7360 psrc = PF_PEER_SRC; 7361 pdst = PF_PEER_DST; 7362 } else { 7363 src = &(*state)->dst; 7364 dst = &(*state)->src; 7365 psrc = PF_PEER_DST; 7366 pdst = PF_PEER_SRC; 7367 } 7368 7369 /* update states */ 7370 if (src->state < PFOTHERS_SINGLE) 7371 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7372 if (dst->state == PFOTHERS_SINGLE) 7373 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7374 7375 /* update expire time */ 7376 (*state)->expire = pf_get_uptime(); 7377 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7378 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7379 else 7380 (*state)->timeout = PFTM_OTHER_SINGLE; 7381 7382 /* translate source/destination address, if necessary */ 7383 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7384 struct pf_state_key *nk = (*state)->key[pd->didx]; 7385 7386 KASSERT(nk, ("%s: nk is null", __func__)); 7387 KASSERT(pd, ("%s: pd is null", __func__)); 7388 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7389 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7390 switch (pd->af) { 7391 #ifdef INET 7392 case AF_INET: 7393 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7394 pf_change_a(&pd->src->v4.s_addr, 7395 pd->ip_sum, 7396 nk->addr[pd->sidx].v4.s_addr, 7397 0); 7398 7399 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7400 pf_change_a(&pd->dst->v4.s_addr, 7401 pd->ip_sum, 7402 nk->addr[pd->didx].v4.s_addr, 7403 0); 7404 7405 break; 7406 #endif /* INET */ 7407 #ifdef INET6 7408 case AF_INET6: 7409 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7410 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7411 7412 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7413 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7414 #endif /* INET6 */ 7415 } 7416 } 7417 return (PF_PASS); 7418 } 7419 7420 /* 7421 * ipoff and off are measured from the start of the mbuf chain. 7422 * h must be at "ipoff" on the mbuf chain. 7423 */ 7424 void * 7425 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 7426 u_short *actionp, u_short *reasonp, sa_family_t af) 7427 { 7428 switch (af) { 7429 #ifdef INET 7430 case AF_INET: { 7431 const struct ip *h = mtod(m, struct ip *); 7432 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7433 7434 if (fragoff) { 7435 if (fragoff >= len) 7436 ACTION_SET(actionp, PF_PASS); 7437 else { 7438 ACTION_SET(actionp, PF_DROP); 7439 REASON_SET(reasonp, PFRES_FRAG); 7440 } 7441 return (NULL); 7442 } 7443 if (m->m_pkthdr.len < off + len || 7444 ntohs(h->ip_len) < off + len) { 7445 ACTION_SET(actionp, PF_DROP); 7446 REASON_SET(reasonp, PFRES_SHORT); 7447 return (NULL); 7448 } 7449 break; 7450 } 7451 #endif /* INET */ 7452 #ifdef INET6 7453 case AF_INET6: { 7454 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7455 7456 if (m->m_pkthdr.len < off + len || 7457 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7458 (unsigned)(off + len)) { 7459 ACTION_SET(actionp, PF_DROP); 7460 REASON_SET(reasonp, PFRES_SHORT); 7461 return (NULL); 7462 } 7463 break; 7464 } 7465 #endif /* INET6 */ 7466 } 7467 m_copydata(m, off, len, p); 7468 return (p); 7469 } 7470 7471 int 7472 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7473 int rtableid) 7474 { 7475 struct ifnet *ifp; 7476 7477 /* 7478 * Skip check for addresses with embedded interface scope, 7479 * as they would always match anyway. 7480 */ 7481 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7482 return (1); 7483 7484 if (af != AF_INET && af != AF_INET6) 7485 return (0); 7486 7487 if (kif == V_pfi_all) 7488 return (1); 7489 7490 /* Skip checks for ipsec interfaces */ 7491 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7492 return (1); 7493 7494 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7495 7496 switch (af) { 7497 #ifdef INET6 7498 case AF_INET6: 7499 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7500 ifp)); 7501 #endif 7502 #ifdef INET 7503 case AF_INET: 7504 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7505 ifp)); 7506 #endif 7507 } 7508 7509 return (0); 7510 } 7511 7512 #ifdef INET 7513 static void 7514 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7515 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7516 { 7517 struct mbuf *m0, *m1, *md; 7518 struct sockaddr_in dst; 7519 struct ip *ip; 7520 struct pfi_kkif *nkif = NULL; 7521 struct ifnet *ifp = NULL; 7522 struct pf_addr naddr; 7523 struct pf_ksrc_node *sn = NULL; 7524 int error = 0; 7525 uint16_t ip_len, ip_off; 7526 uint16_t tmp; 7527 int r_rt, r_dir; 7528 7529 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7530 7531 if (s) { 7532 r_rt = s->rt; 7533 r_dir = s->direction; 7534 } else { 7535 r_rt = r->rt; 7536 r_dir = r->direction; 7537 } 7538 7539 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7540 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7541 __func__)); 7542 7543 if ((pd->pf_mtag == NULL && 7544 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7545 pd->pf_mtag->routed++ > 3) { 7546 m0 = *m; 7547 *m = NULL; 7548 goto bad_locked; 7549 } 7550 7551 if (r_rt == PF_DUPTO) { 7552 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7553 if (s == NULL) { 7554 ifp = r->rpool.cur->kif ? 7555 r->rpool.cur->kif->pfik_ifp : NULL; 7556 } else { 7557 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7558 /* If pfsync'd */ 7559 if (ifp == NULL && r->rpool.cur != NULL) 7560 ifp = r->rpool.cur->kif ? 7561 r->rpool.cur->kif->pfik_ifp : NULL; 7562 PF_STATE_UNLOCK(s); 7563 } 7564 if (ifp == oifp) { 7565 /* When the 2nd interface is not skipped */ 7566 return; 7567 } else { 7568 m0 = *m; 7569 *m = NULL; 7570 goto bad; 7571 } 7572 } else { 7573 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7574 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7575 if (s) 7576 PF_STATE_UNLOCK(s); 7577 return; 7578 } 7579 } 7580 } else { 7581 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7582 pf_dummynet(pd, s, r, m); 7583 if (s) 7584 PF_STATE_UNLOCK(s); 7585 return; 7586 } 7587 m0 = *m; 7588 } 7589 7590 ip = mtod(m0, struct ip *); 7591 7592 bzero(&dst, sizeof(dst)); 7593 dst.sin_family = AF_INET; 7594 dst.sin_len = sizeof(dst); 7595 dst.sin_addr = ip->ip_dst; 7596 7597 bzero(&naddr, sizeof(naddr)); 7598 7599 if (s == NULL) { 7600 if (TAILQ_EMPTY(&r->rpool.list)) { 7601 DPFPRINTF(PF_DEBUG_URGENT, 7602 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7603 goto bad_locked; 7604 } 7605 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7606 &naddr, &nkif, NULL, &sn); 7607 if (!PF_AZERO(&naddr, AF_INET)) 7608 dst.sin_addr.s_addr = naddr.v4.s_addr; 7609 ifp = nkif ? nkif->pfik_ifp : NULL; 7610 } else { 7611 struct pfi_kkif *kif; 7612 7613 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7614 dst.sin_addr.s_addr = 7615 s->rt_addr.v4.s_addr; 7616 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7617 kif = s->rt_kif; 7618 /* If pfsync'd */ 7619 if (ifp == NULL && r->rpool.cur != NULL) { 7620 ifp = r->rpool.cur->kif ? 7621 r->rpool.cur->kif->pfik_ifp : NULL; 7622 kif = r->rpool.cur->kif; 7623 } 7624 if (ifp != NULL && kif != NULL && 7625 r->rule_flag & PFRULE_IFBOUND && 7626 r->rt == PF_REPLYTO && 7627 s->kif == V_pfi_all) { 7628 s->kif = kif; 7629 s->orig_kif = oifp->if_pf_kif; 7630 } 7631 7632 PF_STATE_UNLOCK(s); 7633 } 7634 7635 if (ifp == NULL) 7636 goto bad; 7637 7638 if (pd->dir == PF_IN) { 7639 if (pf_test(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS) 7640 goto bad; 7641 else if (m0 == NULL) 7642 goto done; 7643 if (m0->m_len < sizeof(struct ip)) { 7644 DPFPRINTF(PF_DEBUG_URGENT, 7645 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7646 goto bad; 7647 } 7648 ip = mtod(m0, struct ip *); 7649 } 7650 7651 if (ifp->if_flags & IFF_LOOPBACK) 7652 m0->m_flags |= M_SKIP_FIREWALL; 7653 7654 ip_len = ntohs(ip->ip_len); 7655 ip_off = ntohs(ip->ip_off); 7656 7657 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7658 m0->m_pkthdr.csum_flags |= CSUM_IP; 7659 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7660 in_delayed_cksum(m0); 7661 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7662 } 7663 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7664 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7665 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7666 } 7667 7668 if (pd->dir == PF_IN) { 7669 /* 7670 * Make sure dummynet gets the correct direction, in case it needs to 7671 * re-inject later. 7672 */ 7673 pd->dir = PF_OUT; 7674 7675 /* 7676 * The following processing is actually the rest of the inbound processing, even 7677 * though we've marked it as outbound (so we don't look through dummynet) and it 7678 * happens after the outbound processing (pf_test(PF_OUT) above). 7679 * Swap the dummynet pipe numbers, because it's going to come to the wrong 7680 * conclusion about what direction it's processing, and we can't fix it or it 7681 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 7682 * decision will pick the right pipe, and everything will mostly work as expected. 7683 */ 7684 tmp = pd->act.dnrpipe; 7685 pd->act.dnrpipe = pd->act.dnpipe; 7686 pd->act.dnpipe = tmp; 7687 } 7688 7689 /* 7690 * If small enough for interface, or the interface will take 7691 * care of the fragmentation for us, we can just send directly. 7692 */ 7693 if (ip_len <= ifp->if_mtu || 7694 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7695 ip->ip_sum = 0; 7696 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7697 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7698 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7699 } 7700 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7701 7702 md = m0; 7703 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7704 if (md != NULL) 7705 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7706 goto done; 7707 } 7708 7709 /* Balk when DF bit is set or the interface didn't support TSO. */ 7710 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7711 error = EMSGSIZE; 7712 KMOD_IPSTAT_INC(ips_cantfrag); 7713 if (r_rt != PF_DUPTO) { 7714 if (s && pd->nat_rule != NULL) 7715 PACKET_UNDO_NAT(m0, pd, 7716 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7717 s); 7718 7719 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7720 ifp->if_mtu); 7721 goto done; 7722 } else 7723 goto bad; 7724 } 7725 7726 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7727 if (error) 7728 goto bad; 7729 7730 for (; m0; m0 = m1) { 7731 m1 = m0->m_nextpkt; 7732 m0->m_nextpkt = NULL; 7733 if (error == 0) { 7734 m_clrprotoflags(m0); 7735 md = m0; 7736 pd->pf_mtag = pf_find_mtag(md); 7737 error = pf_dummynet_route(pd, s, r, ifp, 7738 sintosa(&dst), &md); 7739 if (md != NULL) 7740 error = (*ifp->if_output)(ifp, md, 7741 sintosa(&dst), NULL); 7742 } else 7743 m_freem(m0); 7744 } 7745 7746 if (error == 0) 7747 KMOD_IPSTAT_INC(ips_fragmented); 7748 7749 done: 7750 if (r_rt != PF_DUPTO) 7751 *m = NULL; 7752 return; 7753 7754 bad_locked: 7755 if (s) 7756 PF_STATE_UNLOCK(s); 7757 bad: 7758 m_freem(m0); 7759 goto done; 7760 } 7761 #endif /* INET */ 7762 7763 #ifdef INET6 7764 static void 7765 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7766 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7767 { 7768 struct mbuf *m0, *md; 7769 struct sockaddr_in6 dst; 7770 struct ip6_hdr *ip6; 7771 struct pfi_kkif *nkif = NULL; 7772 struct ifnet *ifp = NULL; 7773 struct pf_addr naddr; 7774 struct pf_ksrc_node *sn = NULL; 7775 int r_rt, r_dir; 7776 7777 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7778 7779 if (s) { 7780 r_rt = s->rt; 7781 r_dir = s->direction; 7782 } else { 7783 r_rt = r->rt; 7784 r_dir = r->direction; 7785 } 7786 7787 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7788 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7789 __func__)); 7790 7791 if ((pd->pf_mtag == NULL && 7792 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7793 pd->pf_mtag->routed++ > 3) { 7794 m0 = *m; 7795 *m = NULL; 7796 goto bad_locked; 7797 } 7798 7799 if (r_rt == PF_DUPTO) { 7800 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7801 if (s == NULL) { 7802 ifp = r->rpool.cur->kif ? 7803 r->rpool.cur->kif->pfik_ifp : NULL; 7804 } else { 7805 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7806 /* If pfsync'd */ 7807 if (ifp == NULL && r->rpool.cur != NULL) 7808 ifp = r->rpool.cur->kif ? 7809 r->rpool.cur->kif->pfik_ifp : NULL; 7810 PF_STATE_UNLOCK(s); 7811 } 7812 if (ifp == oifp) { 7813 /* When the 2nd interface is not skipped */ 7814 return; 7815 } else { 7816 m0 = *m; 7817 *m = NULL; 7818 goto bad; 7819 } 7820 } else { 7821 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7822 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7823 if (s) 7824 PF_STATE_UNLOCK(s); 7825 return; 7826 } 7827 } 7828 } else { 7829 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7830 pf_dummynet(pd, s, r, m); 7831 if (s) 7832 PF_STATE_UNLOCK(s); 7833 return; 7834 } 7835 m0 = *m; 7836 } 7837 7838 ip6 = mtod(m0, struct ip6_hdr *); 7839 7840 bzero(&dst, sizeof(dst)); 7841 dst.sin6_family = AF_INET6; 7842 dst.sin6_len = sizeof(dst); 7843 dst.sin6_addr = ip6->ip6_dst; 7844 7845 bzero(&naddr, sizeof(naddr)); 7846 7847 if (s == NULL) { 7848 if (TAILQ_EMPTY(&r->rpool.list)) { 7849 DPFPRINTF(PF_DEBUG_URGENT, 7850 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7851 goto bad_locked; 7852 } 7853 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 7854 &naddr, &nkif, NULL, &sn); 7855 if (!PF_AZERO(&naddr, AF_INET6)) 7856 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7857 &naddr, AF_INET6); 7858 ifp = nkif ? nkif->pfik_ifp : NULL; 7859 } else { 7860 struct pfi_kkif *kif; 7861 7862 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 7863 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7864 &s->rt_addr, AF_INET6); 7865 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7866 kif = s->rt_kif; 7867 /* If pfsync'd */ 7868 if (ifp == NULL && r->rpool.cur != NULL) { 7869 ifp = r->rpool.cur->kif ? 7870 r->rpool.cur->kif->pfik_ifp : NULL; 7871 kif = r->rpool.cur->kif; 7872 } 7873 if (ifp != NULL && kif != NULL && 7874 r->rule_flag & PFRULE_IFBOUND && 7875 r->rt == PF_REPLYTO && 7876 s->kif == V_pfi_all) { 7877 s->kif = kif; 7878 s->orig_kif = oifp->if_pf_kif; 7879 } 7880 } 7881 7882 if (s) 7883 PF_STATE_UNLOCK(s); 7884 7885 if (ifp == NULL) 7886 goto bad; 7887 7888 if (pd->dir == PF_IN) { 7889 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS) 7890 goto bad; 7891 else if (m0 == NULL) 7892 goto done; 7893 if (m0->m_len < sizeof(struct ip6_hdr)) { 7894 DPFPRINTF(PF_DEBUG_URGENT, 7895 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 7896 __func__)); 7897 goto bad; 7898 } 7899 ip6 = mtod(m0, struct ip6_hdr *); 7900 } 7901 7902 if (ifp->if_flags & IFF_LOOPBACK) 7903 m0->m_flags |= M_SKIP_FIREWALL; 7904 7905 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 7906 ~ifp->if_hwassist) { 7907 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 7908 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 7909 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 7910 } 7911 7912 /* 7913 * If the packet is too large for the outgoing interface, 7914 * send back an icmp6 error. 7915 */ 7916 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 7917 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 7918 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 7919 md = m0; 7920 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7921 if (md != NULL) 7922 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 7923 } 7924 else { 7925 in6_ifstat_inc(ifp, ifs6_in_toobig); 7926 if (r_rt != PF_DUPTO) { 7927 if (s && pd->nat_rule != NULL) 7928 PACKET_UNDO_NAT(m0, pd, 7929 ((caddr_t)ip6 - m0->m_data) + 7930 sizeof(struct ip6_hdr), s); 7931 7932 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 7933 } else 7934 goto bad; 7935 } 7936 7937 done: 7938 if (r_rt != PF_DUPTO) 7939 *m = NULL; 7940 return; 7941 7942 bad_locked: 7943 if (s) 7944 PF_STATE_UNLOCK(s); 7945 bad: 7946 m_freem(m0); 7947 goto done; 7948 } 7949 #endif /* INET6 */ 7950 7951 /* 7952 * FreeBSD supports cksum offloads for the following drivers. 7953 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 7954 * 7955 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 7956 * network driver performed cksum including pseudo header, need to verify 7957 * csum_data 7958 * CSUM_DATA_VALID : 7959 * network driver performed cksum, needs to additional pseudo header 7960 * cksum computation with partial csum_data(i.e. lack of H/W support for 7961 * pseudo header, for instance sk(4) and possibly gem(4)) 7962 * 7963 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 7964 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 7965 * TCP/UDP layer. 7966 * Also, set csum_data to 0xffff to force cksum validation. 7967 */ 7968 static int 7969 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 7970 { 7971 u_int16_t sum = 0; 7972 int hw_assist = 0; 7973 struct ip *ip; 7974 7975 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 7976 return (1); 7977 if (m->m_pkthdr.len < off + len) 7978 return (1); 7979 7980 switch (p) { 7981 case IPPROTO_TCP: 7982 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 7983 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 7984 sum = m->m_pkthdr.csum_data; 7985 } else { 7986 ip = mtod(m, struct ip *); 7987 sum = in_pseudo(ip->ip_src.s_addr, 7988 ip->ip_dst.s_addr, htonl((u_short)len + 7989 m->m_pkthdr.csum_data + IPPROTO_TCP)); 7990 } 7991 sum ^= 0xffff; 7992 ++hw_assist; 7993 } 7994 break; 7995 case IPPROTO_UDP: 7996 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 7997 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 7998 sum = m->m_pkthdr.csum_data; 7999 } else { 8000 ip = mtod(m, struct ip *); 8001 sum = in_pseudo(ip->ip_src.s_addr, 8002 ip->ip_dst.s_addr, htonl((u_short)len + 8003 m->m_pkthdr.csum_data + IPPROTO_UDP)); 8004 } 8005 sum ^= 0xffff; 8006 ++hw_assist; 8007 } 8008 break; 8009 case IPPROTO_ICMP: 8010 #ifdef INET6 8011 case IPPROTO_ICMPV6: 8012 #endif /* INET6 */ 8013 break; 8014 default: 8015 return (1); 8016 } 8017 8018 if (!hw_assist) { 8019 switch (af) { 8020 case AF_INET: 8021 if (p == IPPROTO_ICMP) { 8022 if (m->m_len < off) 8023 return (1); 8024 m->m_data += off; 8025 m->m_len -= off; 8026 sum = in_cksum(m, len); 8027 m->m_data -= off; 8028 m->m_len += off; 8029 } else { 8030 if (m->m_len < sizeof(struct ip)) 8031 return (1); 8032 sum = in4_cksum(m, p, off, len); 8033 } 8034 break; 8035 #ifdef INET6 8036 case AF_INET6: 8037 if (m->m_len < sizeof(struct ip6_hdr)) 8038 return (1); 8039 sum = in6_cksum(m, p, off, len); 8040 break; 8041 #endif /* INET6 */ 8042 default: 8043 return (1); 8044 } 8045 } 8046 if (sum) { 8047 switch (p) { 8048 case IPPROTO_TCP: 8049 { 8050 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 8051 break; 8052 } 8053 case IPPROTO_UDP: 8054 { 8055 KMOD_UDPSTAT_INC(udps_badsum); 8056 break; 8057 } 8058 #ifdef INET 8059 case IPPROTO_ICMP: 8060 { 8061 KMOD_ICMPSTAT_INC(icps_checksum); 8062 break; 8063 } 8064 #endif 8065 #ifdef INET6 8066 case IPPROTO_ICMPV6: 8067 { 8068 KMOD_ICMP6STAT_INC(icp6s_checksum); 8069 break; 8070 } 8071 #endif /* INET6 */ 8072 } 8073 return (1); 8074 } else { 8075 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 8076 m->m_pkthdr.csum_flags |= 8077 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 8078 m->m_pkthdr.csum_data = 0xffff; 8079 } 8080 } 8081 return (0); 8082 } 8083 8084 static bool 8085 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 8086 const struct pf_kstate *s, struct ip_fw_args *dnflow) 8087 { 8088 int dndir = r->direction; 8089 8090 if (s && dndir == PF_INOUT) { 8091 dndir = s->direction; 8092 } else if (dndir == PF_INOUT) { 8093 /* Assume primary direction. Happens when we've set dnpipe in 8094 * the ethernet level code. */ 8095 dndir = pd->dir; 8096 } 8097 8098 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 8099 return (false); 8100 8101 memset(dnflow, 0, sizeof(*dnflow)); 8102 8103 if (pd->dport != NULL) 8104 dnflow->f_id.dst_port = ntohs(*pd->dport); 8105 if (pd->sport != NULL) 8106 dnflow->f_id.src_port = ntohs(*pd->sport); 8107 8108 if (pd->dir == PF_IN) 8109 dnflow->flags |= IPFW_ARGS_IN; 8110 else 8111 dnflow->flags |= IPFW_ARGS_OUT; 8112 8113 if (pd->dir != dndir && pd->act.dnrpipe) { 8114 dnflow->rule.info = pd->act.dnrpipe; 8115 } 8116 else if (pd->dir == dndir && pd->act.dnpipe) { 8117 dnflow->rule.info = pd->act.dnpipe; 8118 } 8119 else { 8120 return (false); 8121 } 8122 8123 dnflow->rule.info |= IPFW_IS_DUMMYNET; 8124 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 8125 dnflow->rule.info |= IPFW_IS_PIPE; 8126 8127 dnflow->f_id.proto = pd->proto; 8128 dnflow->f_id.extra = dnflow->rule.info; 8129 switch (pd->af) { 8130 case AF_INET: 8131 dnflow->f_id.addr_type = 4; 8132 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 8133 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 8134 break; 8135 case AF_INET6: 8136 dnflow->flags |= IPFW_ARGS_IP6; 8137 dnflow->f_id.addr_type = 6; 8138 dnflow->f_id.src_ip6 = pd->src->v6; 8139 dnflow->f_id.dst_ip6 = pd->dst->v6; 8140 break; 8141 default: 8142 panic("Invalid AF"); 8143 break; 8144 } 8145 8146 return (true); 8147 } 8148 8149 int 8150 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8151 struct inpcb *inp) 8152 { 8153 struct pfi_kkif *kif; 8154 struct mbuf *m = *m0; 8155 8156 M_ASSERTPKTHDR(m); 8157 MPASS(ifp->if_vnet == curvnet); 8158 NET_EPOCH_ASSERT(); 8159 8160 if (!V_pf_status.running) 8161 return (PF_PASS); 8162 8163 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8164 8165 if (kif == NULL) { 8166 DPFPRINTF(PF_DEBUG_URGENT, 8167 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 8168 return (PF_DROP); 8169 } 8170 if (kif->pfik_flags & PFI_IFLAG_SKIP) 8171 return (PF_PASS); 8172 8173 if (m->m_flags & M_SKIP_FIREWALL) 8174 return (PF_PASS); 8175 8176 /* Stateless! */ 8177 return (pf_test_eth_rule(dir, kif, m0)); 8178 } 8179 8180 static __inline void 8181 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 8182 { 8183 struct m_tag *mtag; 8184 8185 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 8186 8187 /* dummynet adds this tag, but pf does not need it, 8188 * and keeping it creates unexpected behavior, 8189 * e.g. in case of divert(4) usage right after dummynet. */ 8190 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8191 if (mtag != NULL) 8192 m_tag_delete(m, mtag); 8193 } 8194 8195 static int 8196 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 8197 struct pf_krule *r, struct mbuf **m0) 8198 { 8199 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 8200 } 8201 8202 static int 8203 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 8204 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 8205 struct mbuf **m0) 8206 { 8207 NET_EPOCH_ASSERT(); 8208 8209 if (pd->act.dnpipe || pd->act.dnrpipe) { 8210 struct ip_fw_args dnflow; 8211 if (ip_dn_io_ptr == NULL) { 8212 m_freem(*m0); 8213 *m0 = NULL; 8214 return (ENOMEM); 8215 } 8216 8217 if (pd->pf_mtag == NULL && 8218 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 8219 m_freem(*m0); 8220 *m0 = NULL; 8221 return (ENOMEM); 8222 } 8223 8224 if (ifp != NULL) { 8225 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 8226 8227 pd->pf_mtag->if_index = ifp->if_index; 8228 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 8229 8230 MPASS(sa != NULL); 8231 8232 if (pd->af == AF_INET) 8233 memcpy(&pd->pf_mtag->dst, sa, 8234 sizeof(struct sockaddr_in)); 8235 else 8236 memcpy(&pd->pf_mtag->dst, sa, 8237 sizeof(struct sockaddr_in6)); 8238 } 8239 8240 if (s != NULL && s->nat_rule.ptr != NULL && 8241 s->nat_rule.ptr->action == PF_RDR && 8242 ( 8243 #ifdef INET 8244 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 8245 #endif 8246 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 8247 /* 8248 * If we're redirecting to loopback mark this packet 8249 * as being local. Otherwise it might get dropped 8250 * if dummynet re-injects. 8251 */ 8252 (*m0)->m_pkthdr.rcvif = V_loif; 8253 } 8254 8255 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 8256 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 8257 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 8258 ip_dn_io_ptr(m0, &dnflow); 8259 if (*m0 != NULL) { 8260 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8261 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 8262 } 8263 } 8264 } 8265 8266 return (0); 8267 } 8268 8269 #ifdef INET 8270 int 8271 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8272 struct inpcb *inp, struct pf_rule_actions *default_actions) 8273 { 8274 struct pfi_kkif *kif; 8275 u_short action, reason = 0; 8276 struct mbuf *m = *m0; 8277 struct ip *h = NULL; 8278 struct m_tag *mtag; 8279 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 8280 struct pf_kstate *s = NULL; 8281 struct pf_kruleset *ruleset = NULL; 8282 struct pf_pdesc pd; 8283 int off, dirndx, use_2nd_queue = 0; 8284 uint16_t tag; 8285 uint8_t rt; 8286 8287 PF_RULES_RLOCK_TRACKER; 8288 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8289 M_ASSERTPKTHDR(m); 8290 8291 if (!V_pf_status.running) 8292 return (PF_PASS); 8293 8294 PF_RULES_RLOCK(); 8295 8296 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8297 8298 if (__predict_false(kif == NULL)) { 8299 DPFPRINTF(PF_DEBUG_URGENT, 8300 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 8301 PF_RULES_RUNLOCK(); 8302 return (PF_DROP); 8303 } 8304 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8305 PF_RULES_RUNLOCK(); 8306 return (PF_PASS); 8307 } 8308 8309 if (m->m_flags & M_SKIP_FIREWALL) { 8310 PF_RULES_RUNLOCK(); 8311 return (PF_PASS); 8312 } 8313 8314 memset(&pd, 0, sizeof(pd)); 8315 TAILQ_INIT(&pd.sctp_multihome_jobs); 8316 if (default_actions != NULL) 8317 memcpy(&pd.act, default_actions, sizeof(pd.act)); 8318 pd.pf_mtag = pf_find_mtag(m); 8319 8320 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 8321 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8322 8323 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 8324 pd.pf_mtag->if_idxgen); 8325 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 8326 PF_RULES_RUNLOCK(); 8327 m_freem(*m0); 8328 *m0 = NULL; 8329 return (PF_PASS); 8330 } 8331 PF_RULES_RUNLOCK(); 8332 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 8333 *m0 = NULL; 8334 return (PF_PASS); 8335 } 8336 8337 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 8338 pd.act.dnpipe = pd.pf_mtag->dnpipe; 8339 pd.act.flags = pd.pf_mtag->dnflags; 8340 } 8341 8342 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 8343 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 8344 /* Dummynet re-injects packets after they've 8345 * completed their delay. We've already 8346 * processed them, so pass unconditionally. */ 8347 8348 /* But only once. We may see the packet multiple times (e.g. 8349 * PFIL_IN/PFIL_OUT). */ 8350 pf_dummynet_flag_remove(m, pd.pf_mtag); 8351 PF_RULES_RUNLOCK(); 8352 8353 return (PF_PASS); 8354 } 8355 8356 pd.sport = pd.dport = NULL; 8357 pd.proto_sum = NULL; 8358 pd.dir = dir; 8359 pd.sidx = (dir == PF_IN) ? 0 : 1; 8360 pd.didx = (dir == PF_IN) ? 1 : 0; 8361 pd.af = AF_INET; 8362 pd.act.rtableid = -1; 8363 8364 if (__predict_false(m->m_len < sizeof(struct ip)) && 8365 (m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 8366 DPFPRINTF(PF_DEBUG_URGENT, 8367 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 8368 PF_RULES_RUNLOCK(); 8369 return (PF_DROP); 8370 } 8371 h = mtod(m, struct ip *); 8372 off = h->ip_hl << 2; 8373 8374 if (__predict_false(ip_divert_ptr != NULL) && 8375 ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 8376 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 8377 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 8378 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 8379 if (pd.pf_mtag == NULL && 8380 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8381 action = PF_DROP; 8382 goto done; 8383 } 8384 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 8385 } 8386 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 8387 m->m_flags |= M_FASTFWD_OURS; 8388 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 8389 } 8390 m_tag_delete(m, mtag); 8391 8392 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8393 if (mtag != NULL) 8394 m_tag_delete(m, mtag); 8395 } else if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) { 8396 /* We do IP header normalization and packet reassembly here */ 8397 action = PF_DROP; 8398 goto done; 8399 } 8400 m = *m0; /* pf_normalize messes with m0 */ 8401 h = mtod(m, struct ip *); 8402 8403 off = h->ip_hl << 2; 8404 if (off < (int)sizeof(struct ip)) { 8405 action = PF_DROP; 8406 REASON_SET(&reason, PFRES_SHORT); 8407 pd.act.log = PF_LOG_FORCE; 8408 goto done; 8409 } 8410 8411 pd.src = (struct pf_addr *)&h->ip_src; 8412 pd.dst = (struct pf_addr *)&h->ip_dst; 8413 pd.ip_sum = &h->ip_sum; 8414 pd.proto = h->ip_p; 8415 pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; 8416 pd.tot_len = ntohs(h->ip_len); 8417 8418 /* handle fragments that didn't get reassembled by normalization */ 8419 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 8420 action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset); 8421 goto done; 8422 } 8423 8424 switch (h->ip_p) { 8425 case IPPROTO_TCP: { 8426 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 8427 &action, &reason, AF_INET)) { 8428 if (action != PF_PASS) 8429 pd.act.log = PF_LOG_FORCE; 8430 goto done; 8431 } 8432 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 8433 8434 pd.sport = &pd.hdr.tcp.th_sport; 8435 pd.dport = &pd.hdr.tcp.th_dport; 8436 8437 /* Respond to SYN with a syncookie. */ 8438 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 8439 pd.dir == PF_IN && pf_synflood_check(&pd)) { 8440 pf_syncookie_send(m, off, &pd); 8441 action = PF_DROP; 8442 break; 8443 } 8444 8445 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 8446 use_2nd_queue = 1; 8447 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 8448 if (action == PF_DROP) 8449 goto done; 8450 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 8451 if (action == PF_PASS) { 8452 if (V_pfsync_update_state_ptr != NULL) 8453 V_pfsync_update_state_ptr(s); 8454 r = s->rule.ptr; 8455 a = s->anchor.ptr; 8456 } else if (s == NULL) { 8457 /* Validate remote SYN|ACK, re-create original SYN if 8458 * valid. */ 8459 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 8460 TH_ACK && pf_syncookie_validate(&pd) && 8461 pd.dir == PF_IN) { 8462 struct mbuf *msyn; 8463 8464 msyn = pf_syncookie_recreate_syn(h->ip_ttl, off, 8465 &pd); 8466 if (msyn == NULL) { 8467 action = PF_DROP; 8468 break; 8469 } 8470 8471 action = pf_test(dir, pflags, ifp, &msyn, inp, 8472 &pd.act); 8473 m_freem(msyn); 8474 if (action != PF_PASS) 8475 break; 8476 8477 action = pf_test_state_tcp(&s, kif, m, off, h, 8478 &pd, &reason); 8479 if (action != PF_PASS || s == NULL) { 8480 action = PF_DROP; 8481 break; 8482 } 8483 8484 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 8485 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 8486 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 8487 action = pf_synproxy(&pd, &s, &reason); 8488 break; 8489 } else { 8490 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8491 &a, &ruleset, inp); 8492 } 8493 } 8494 break; 8495 } 8496 8497 case IPPROTO_UDP: { 8498 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 8499 &action, &reason, AF_INET)) { 8500 if (action != PF_PASS) 8501 pd.act.log = PF_LOG_FORCE; 8502 goto done; 8503 } 8504 pd.sport = &pd.hdr.udp.uh_sport; 8505 pd.dport = &pd.hdr.udp.uh_dport; 8506 if (pd.hdr.udp.uh_dport == 0 || 8507 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 8508 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 8509 action = PF_DROP; 8510 REASON_SET(&reason, PFRES_SHORT); 8511 goto done; 8512 } 8513 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 8514 if (action == PF_PASS) { 8515 if (V_pfsync_update_state_ptr != NULL) 8516 V_pfsync_update_state_ptr(s); 8517 r = s->rule.ptr; 8518 a = s->anchor.ptr; 8519 } else if (s == NULL) 8520 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8521 &a, &ruleset, inp); 8522 break; 8523 } 8524 8525 case IPPROTO_SCTP: { 8526 if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp), 8527 &action, &reason, AF_INET)) { 8528 if (action != PF_PASS) 8529 pd.act.log |= PF_LOG_FORCE; 8530 goto done; 8531 } 8532 pd.p_len = pd.tot_len - off; 8533 8534 pd.sport = &pd.hdr.sctp.src_port; 8535 pd.dport = &pd.hdr.sctp.dest_port; 8536 if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) { 8537 action = PF_DROP; 8538 REASON_SET(&reason, PFRES_SHORT); 8539 goto done; 8540 } 8541 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 8542 if (action == PF_DROP) 8543 goto done; 8544 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 8545 &reason); 8546 if (action == PF_PASS) { 8547 if (V_pfsync_update_state_ptr != NULL) 8548 V_pfsync_update_state_ptr(s); 8549 r = s->rule.ptr; 8550 a = s->anchor.ptr; 8551 } else if (s == NULL) { 8552 action = pf_test_rule(&r, &s, kif, m, off, 8553 &pd, &a, &ruleset, inp); 8554 } 8555 break; 8556 } 8557 8558 case IPPROTO_ICMP: { 8559 if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN, 8560 &action, &reason, AF_INET)) { 8561 if (action != PF_PASS) 8562 pd.act.log = PF_LOG_FORCE; 8563 goto done; 8564 } 8565 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 8566 if (action == PF_PASS) { 8567 if (V_pfsync_update_state_ptr != NULL) 8568 V_pfsync_update_state_ptr(s); 8569 r = s->rule.ptr; 8570 a = s->anchor.ptr; 8571 } else if (s == NULL) 8572 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8573 &a, &ruleset, inp); 8574 break; 8575 } 8576 8577 #ifdef INET6 8578 case IPPROTO_ICMPV6: { 8579 action = PF_DROP; 8580 DPFPRINTF(PF_DEBUG_MISC, 8581 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 8582 goto done; 8583 } 8584 #endif 8585 8586 default: 8587 action = pf_test_state_other(&s, kif, m, &pd); 8588 if (action == PF_PASS) { 8589 if (V_pfsync_update_state_ptr != NULL) 8590 V_pfsync_update_state_ptr(s); 8591 r = s->rule.ptr; 8592 a = s->anchor.ptr; 8593 } else if (s == NULL) 8594 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8595 &a, &ruleset, inp); 8596 break; 8597 } 8598 8599 done: 8600 PF_RULES_RUNLOCK(); 8601 if (action == PF_PASS && h->ip_hl > 5 && 8602 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 8603 action = PF_DROP; 8604 REASON_SET(&reason, PFRES_IPOPTIONS); 8605 pd.act.log = PF_LOG_FORCE; 8606 DPFPRINTF(PF_DEBUG_MISC, 8607 ("pf: dropping packet with ip options\n")); 8608 } 8609 8610 if (s) { 8611 uint8_t log = pd.act.log; 8612 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 8613 pd.act.log |= log; 8614 tag = s->tag; 8615 rt = s->rt; 8616 } else { 8617 tag = r->tag; 8618 rt = r->rt; 8619 } 8620 8621 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 8622 action = PF_DROP; 8623 REASON_SET(&reason, PFRES_MEMORY); 8624 } 8625 8626 pf_scrub_ip(&m, &pd); 8627 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 8628 pf_normalize_mss(m, off, &pd); 8629 8630 if (pd.act.rtableid >= 0) 8631 M_SETFIB(m, pd.act.rtableid); 8632 8633 if (pd.act.flags & PFSTATE_SETPRIO) { 8634 if (pd.tos & IPTOS_LOWDELAY) 8635 use_2nd_queue = 1; 8636 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 8637 action = PF_DROP; 8638 REASON_SET(&reason, PFRES_MEMORY); 8639 pd.act.log = PF_LOG_FORCE; 8640 DPFPRINTF(PF_DEBUG_MISC, 8641 ("pf: failed to allocate 802.1q mtag\n")); 8642 } 8643 } 8644 8645 #ifdef ALTQ 8646 if (action == PF_PASS && pd.act.qid) { 8647 if (pd.pf_mtag == NULL && 8648 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8649 action = PF_DROP; 8650 REASON_SET(&reason, PFRES_MEMORY); 8651 } else { 8652 if (s != NULL) 8653 pd.pf_mtag->qid_hash = pf_state_hash(s); 8654 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 8655 pd.pf_mtag->qid = pd.act.pqid; 8656 else 8657 pd.pf_mtag->qid = pd.act.qid; 8658 /* Add hints for ecn. */ 8659 pd.pf_mtag->hdr = h; 8660 } 8661 } 8662 #endif /* ALTQ */ 8663 8664 /* 8665 * connections redirected to loopback should not match sockets 8666 * bound specifically to loopback due to security implications, 8667 * see tcp_input() and in_pcblookup_listen(). 8668 */ 8669 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 8670 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 8671 (s->nat_rule.ptr->action == PF_RDR || 8672 s->nat_rule.ptr->action == PF_BINAT) && 8673 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 8674 m->m_flags |= M_SKIP_FIREWALL; 8675 8676 if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS && 8677 r->divert.port && !PACKET_LOOPED(&pd)) { 8678 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 8679 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 8680 if (mtag != NULL) { 8681 ((struct pf_divert_mtag *)(mtag+1))->port = 8682 ntohs(r->divert.port); 8683 ((struct pf_divert_mtag *)(mtag+1))->idir = 8684 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 8685 PF_DIVERT_MTAG_DIR_OUT; 8686 8687 if (s) 8688 PF_STATE_UNLOCK(s); 8689 8690 m_tag_prepend(m, mtag); 8691 if (m->m_flags & M_FASTFWD_OURS) { 8692 if (pd.pf_mtag == NULL && 8693 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8694 action = PF_DROP; 8695 REASON_SET(&reason, PFRES_MEMORY); 8696 pd.act.log = PF_LOG_FORCE; 8697 DPFPRINTF(PF_DEBUG_MISC, 8698 ("pf: failed to allocate tag\n")); 8699 } else { 8700 pd.pf_mtag->flags |= 8701 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 8702 m->m_flags &= ~M_FASTFWD_OURS; 8703 } 8704 } 8705 ip_divert_ptr(*m0, dir == PF_IN); 8706 *m0 = NULL; 8707 8708 return (action); 8709 } else { 8710 /* XXX: ipfw has the same behaviour! */ 8711 action = PF_DROP; 8712 REASON_SET(&reason, PFRES_MEMORY); 8713 pd.act.log = PF_LOG_FORCE; 8714 DPFPRINTF(PF_DEBUG_MISC, 8715 ("pf: failed to allocate divert tag\n")); 8716 } 8717 } 8718 /* this flag will need revising if the pkt is forwarded */ 8719 if (pd.pf_mtag) 8720 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 8721 8722 if (pd.act.log) { 8723 struct pf_krule *lr; 8724 struct pf_krule_item *ri; 8725 8726 if (s != NULL && s->nat_rule.ptr != NULL && 8727 s->nat_rule.ptr->log & PF_LOG_ALL) 8728 lr = s->nat_rule.ptr; 8729 else 8730 lr = r; 8731 8732 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 8733 PFLOG_PACKET(kif, m, AF_INET, action, reason, lr, a, 8734 ruleset, &pd, (s == NULL)); 8735 if (s) { 8736 SLIST_FOREACH(ri, &s->match_rules, entry) 8737 if (ri->r->log & PF_LOG_ALL) 8738 PFLOG_PACKET(kif, m, AF_INET, action, 8739 reason, ri->r, a, ruleset, &pd, 0); 8740 } 8741 } 8742 8743 pf_counter_u64_critical_enter(); 8744 pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS], 8745 pd.tot_len); 8746 pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS], 8747 1); 8748 8749 if (action == PF_PASS || r->action == PF_DROP) { 8750 dirndx = (dir == PF_OUT); 8751 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8752 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 8753 pf_update_timestamp(r); 8754 8755 if (a != NULL) { 8756 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8757 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 8758 } 8759 if (s != NULL) { 8760 struct pf_krule_item *ri; 8761 8762 if (s->nat_rule.ptr != NULL) { 8763 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 8764 1); 8765 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 8766 pd.tot_len); 8767 } 8768 if (s->src_node != NULL) { 8769 counter_u64_add(s->src_node->packets[dirndx], 8770 1); 8771 counter_u64_add(s->src_node->bytes[dirndx], 8772 pd.tot_len); 8773 } 8774 if (s->nat_src_node != NULL) { 8775 counter_u64_add(s->nat_src_node->packets[dirndx], 8776 1); 8777 counter_u64_add(s->nat_src_node->bytes[dirndx], 8778 pd.tot_len); 8779 } 8780 dirndx = (dir == s->direction) ? 0 : 1; 8781 s->packets[dirndx]++; 8782 s->bytes[dirndx] += pd.tot_len; 8783 SLIST_FOREACH(ri, &s->match_rules, entry) { 8784 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8785 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len); 8786 } 8787 } 8788 tr = r; 8789 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 8790 if (nr != NULL && r == &V_pf_default_rule) 8791 tr = nr; 8792 if (tr->src.addr.type == PF_ADDR_TABLE) 8793 pfr_update_stats(tr->src.addr.p.tbl, 8794 (s == NULL) ? pd.src : 8795 &s->key[(s->direction == PF_IN)]-> 8796 addr[(s->direction == PF_OUT)], 8797 pd.af, pd.tot_len, dir == PF_OUT, 8798 r->action == PF_PASS, tr->src.neg); 8799 if (tr->dst.addr.type == PF_ADDR_TABLE) 8800 pfr_update_stats(tr->dst.addr.p.tbl, 8801 (s == NULL) ? pd.dst : 8802 &s->key[(s->direction == PF_IN)]-> 8803 addr[(s->direction == PF_IN)], 8804 pd.af, pd.tot_len, dir == PF_OUT, 8805 r->action == PF_PASS, tr->dst.neg); 8806 } 8807 pf_counter_u64_critical_exit(); 8808 8809 switch (action) { 8810 case PF_SYNPROXY_DROP: 8811 m_freem(*m0); 8812 case PF_DEFER: 8813 *m0 = NULL; 8814 action = PF_PASS; 8815 break; 8816 case PF_DROP: 8817 m_freem(*m0); 8818 *m0 = NULL; 8819 break; 8820 default: 8821 /* pf_route() returns unlocked. */ 8822 if (rt) { 8823 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 8824 goto out; 8825 } 8826 if (pf_dummynet(&pd, s, r, m0) != 0) { 8827 action = PF_DROP; 8828 REASON_SET(&reason, PFRES_MEMORY); 8829 } 8830 break; 8831 } 8832 8833 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 8834 8835 if (s && action != PF_DROP) { 8836 if (!s->if_index_in && dir == PF_IN) 8837 s->if_index_in = ifp->if_index; 8838 else if (!s->if_index_out && dir == PF_OUT) 8839 s->if_index_out = ifp->if_index; 8840 } 8841 8842 if (s) 8843 PF_STATE_UNLOCK(s); 8844 8845 out: 8846 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 8847 8848 return (action); 8849 } 8850 #endif /* INET */ 8851 8852 #ifdef INET6 8853 int 8854 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp, 8855 struct pf_rule_actions *default_actions) 8856 { 8857 struct pfi_kkif *kif; 8858 u_short action, reason = 0; 8859 struct mbuf *m = *m0, *n = NULL; 8860 struct m_tag *mtag; 8861 struct ip6_hdr *h = NULL; 8862 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 8863 struct pf_kstate *s = NULL; 8864 struct pf_kruleset *ruleset = NULL; 8865 struct pf_pdesc pd; 8866 int off, terminal = 0, dirndx, rh_cnt = 0, use_2nd_queue = 0; 8867 uint16_t tag; 8868 uint8_t rt; 8869 8870 PF_RULES_RLOCK_TRACKER; 8871 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8872 M_ASSERTPKTHDR(m); 8873 8874 if (!V_pf_status.running) 8875 return (PF_PASS); 8876 8877 PF_RULES_RLOCK(); 8878 8879 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8880 if (__predict_false(kif == NULL)) { 8881 DPFPRINTF(PF_DEBUG_URGENT, 8882 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 8883 PF_RULES_RUNLOCK(); 8884 return (PF_DROP); 8885 } 8886 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8887 PF_RULES_RUNLOCK(); 8888 return (PF_PASS); 8889 } 8890 8891 if (m->m_flags & M_SKIP_FIREWALL) { 8892 PF_RULES_RUNLOCK(); 8893 return (PF_PASS); 8894 } 8895 8896 /* 8897 * If we end up changing IP addresses (e.g. binat) the stack may get 8898 * confused and fail to send the icmp6 packet too big error. Just send 8899 * it here, before we do any NAT. 8900 */ 8901 if (dir == PF_OUT && pflags & PFIL_FWD && IN6_LINKMTU(ifp) < pf_max_frag_size(m)) { 8902 PF_RULES_RUNLOCK(); 8903 *m0 = NULL; 8904 icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 8905 return (PF_DROP); 8906 } 8907 8908 memset(&pd, 0, sizeof(pd)); 8909 TAILQ_INIT(&pd.sctp_multihome_jobs); 8910 if (default_actions != NULL) 8911 memcpy(&pd.act, default_actions, sizeof(pd.act)); 8912 pd.pf_mtag = pf_find_mtag(m); 8913 8914 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 8915 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8916 8917 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 8918 pd.pf_mtag->if_idxgen); 8919 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 8920 PF_RULES_RUNLOCK(); 8921 m_freem(*m0); 8922 *m0 = NULL; 8923 return (PF_PASS); 8924 } 8925 PF_RULES_RUNLOCK(); 8926 nd6_output_ifp(ifp, ifp, m, 8927 (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL); 8928 *m0 = NULL; 8929 return (PF_PASS); 8930 } 8931 8932 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 8933 pd.act.dnpipe = pd.pf_mtag->dnpipe; 8934 pd.act.flags = pd.pf_mtag->dnflags; 8935 } 8936 8937 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 8938 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 8939 pf_dummynet_flag_remove(m, pd.pf_mtag); 8940 /* Dummynet re-injects packets after they've 8941 * completed their delay. We've already 8942 * processed them, so pass unconditionally. */ 8943 PF_RULES_RUNLOCK(); 8944 return (PF_PASS); 8945 } 8946 8947 pd.sport = pd.dport = NULL; 8948 pd.ip_sum = NULL; 8949 pd.proto_sum = NULL; 8950 pd.dir = dir; 8951 pd.sidx = (dir == PF_IN) ? 0 : 1; 8952 pd.didx = (dir == PF_IN) ? 1 : 0; 8953 pd.af = AF_INET6; 8954 pd.act.rtableid = -1; 8955 8956 if (__predict_false(m->m_len < sizeof(struct ip6_hdr)) && 8957 (m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 8958 DPFPRINTF(PF_DEBUG_URGENT, 8959 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 8960 ", pullup failed\n")); 8961 PF_RULES_RUNLOCK(); 8962 return (PF_DROP); 8963 } 8964 h = mtod(m, struct ip6_hdr *); 8965 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8966 8967 /* We do IP header normalization and packet reassembly here */ 8968 if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) { 8969 action = PF_DROP; 8970 goto done; 8971 } 8972 m = *m0; /* pf_normalize messes with m0 */ 8973 h = mtod(m, struct ip6_hdr *); 8974 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8975 8976 /* 8977 * we do not support jumbogram. if we keep going, zero ip6_plen 8978 * will do something bad, so drop the packet for now. 8979 */ 8980 if (htons(h->ip6_plen) == 0) { 8981 action = PF_DROP; 8982 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 8983 goto done; 8984 } 8985 8986 pd.src = (struct pf_addr *)&h->ip6_src; 8987 pd.dst = (struct pf_addr *)&h->ip6_dst; 8988 pd.tos = IPV6_DSCP(h); 8989 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8990 8991 pd.proto = h->ip6_nxt; 8992 do { 8993 switch (pd.proto) { 8994 case IPPROTO_FRAGMENT: 8995 action = pf_test_fragment(&r, kif, m, h, &pd, &a, 8996 &ruleset); 8997 if (action == PF_DROP) 8998 REASON_SET(&reason, PFRES_FRAG); 8999 goto done; 9000 case IPPROTO_ROUTING: { 9001 struct ip6_rthdr rthdr; 9002 9003 if (rh_cnt++) { 9004 DPFPRINTF(PF_DEBUG_MISC, 9005 ("pf: IPv6 more than one rthdr\n")); 9006 action = PF_DROP; 9007 REASON_SET(&reason, PFRES_IPOPTIONS); 9008 pd.act.log = PF_LOG_FORCE; 9009 goto done; 9010 } 9011 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 9012 &reason, pd.af)) { 9013 DPFPRINTF(PF_DEBUG_MISC, 9014 ("pf: IPv6 short rthdr\n")); 9015 action = PF_DROP; 9016 REASON_SET(&reason, PFRES_SHORT); 9017 pd.act.log = PF_LOG_FORCE; 9018 goto done; 9019 } 9020 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 9021 DPFPRINTF(PF_DEBUG_MISC, 9022 ("pf: IPv6 rthdr0\n")); 9023 action = PF_DROP; 9024 REASON_SET(&reason, PFRES_IPOPTIONS); 9025 pd.act.log = PF_LOG_FORCE; 9026 goto done; 9027 } 9028 /* FALLTHROUGH */ 9029 } 9030 case IPPROTO_AH: 9031 case IPPROTO_HOPOPTS: 9032 case IPPROTO_DSTOPTS: { 9033 /* get next header and header length */ 9034 struct ip6_ext opt6; 9035 9036 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 9037 NULL, &reason, pd.af)) { 9038 DPFPRINTF(PF_DEBUG_MISC, 9039 ("pf: IPv6 short opt\n")); 9040 action = PF_DROP; 9041 pd.act.log = PF_LOG_FORCE; 9042 goto done; 9043 } 9044 if (pd.proto == IPPROTO_AH) 9045 off += (opt6.ip6e_len + 2) * 4; 9046 else 9047 off += (opt6.ip6e_len + 1) * 8; 9048 pd.proto = opt6.ip6e_nxt; 9049 /* goto the next header */ 9050 break; 9051 } 9052 default: 9053 terminal++; 9054 break; 9055 } 9056 } while (!terminal); 9057 9058 /* if there's no routing header, use unmodified mbuf for checksumming */ 9059 if (!n) 9060 n = m; 9061 9062 switch (pd.proto) { 9063 case IPPROTO_TCP: { 9064 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 9065 &action, &reason, AF_INET6)) { 9066 if (action != PF_PASS) 9067 pd.act.log |= PF_LOG_FORCE; 9068 goto done; 9069 } 9070 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 9071 pd.sport = &pd.hdr.tcp.th_sport; 9072 pd.dport = &pd.hdr.tcp.th_dport; 9073 9074 /* Respond to SYN with a syncookie. */ 9075 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 9076 pd.dir == PF_IN && pf_synflood_check(&pd)) { 9077 pf_syncookie_send(m, off, &pd); 9078 action = PF_DROP; 9079 break; 9080 } 9081 9082 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 9083 if (action == PF_DROP) 9084 goto done; 9085 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 9086 if (action == PF_PASS) { 9087 if (V_pfsync_update_state_ptr != NULL) 9088 V_pfsync_update_state_ptr(s); 9089 r = s->rule.ptr; 9090 a = s->anchor.ptr; 9091 } else if (s == NULL) { 9092 /* Validate remote SYN|ACK, re-create original SYN if 9093 * valid. */ 9094 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 9095 TH_ACK && pf_syncookie_validate(&pd) && 9096 pd.dir == PF_IN) { 9097 struct mbuf *msyn; 9098 9099 msyn = pf_syncookie_recreate_syn(h->ip6_hlim, 9100 off, &pd); 9101 if (msyn == NULL) { 9102 action = PF_DROP; 9103 break; 9104 } 9105 9106 action = pf_test6(dir, pflags, ifp, &msyn, inp, 9107 &pd.act); 9108 m_freem(msyn); 9109 if (action != PF_PASS) 9110 break; 9111 9112 action = pf_test_state_tcp(&s, kif, m, off, h, 9113 &pd, &reason); 9114 if (action != PF_PASS || s == NULL) { 9115 action = PF_DROP; 9116 break; 9117 } 9118 9119 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 9120 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 9121 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 9122 9123 action = pf_synproxy(&pd, &s, &reason); 9124 break; 9125 } else { 9126 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9127 &a, &ruleset, inp); 9128 } 9129 } 9130 break; 9131 } 9132 9133 case IPPROTO_UDP: { 9134 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 9135 &action, &reason, AF_INET6)) { 9136 if (action != PF_PASS) 9137 pd.act.log |= PF_LOG_FORCE; 9138 goto done; 9139 } 9140 pd.sport = &pd.hdr.udp.uh_sport; 9141 pd.dport = &pd.hdr.udp.uh_dport; 9142 if (pd.hdr.udp.uh_dport == 0 || 9143 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 9144 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 9145 action = PF_DROP; 9146 REASON_SET(&reason, PFRES_SHORT); 9147 goto done; 9148 } 9149 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 9150 if (action == PF_PASS) { 9151 if (V_pfsync_update_state_ptr != NULL) 9152 V_pfsync_update_state_ptr(s); 9153 r = s->rule.ptr; 9154 a = s->anchor.ptr; 9155 } else if (s == NULL) 9156 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9157 &a, &ruleset, inp); 9158 break; 9159 } 9160 9161 case IPPROTO_SCTP: { 9162 if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp), 9163 &action, &reason, AF_INET6)) { 9164 if (action != PF_PASS) 9165 pd.act.log |= PF_LOG_FORCE; 9166 goto done; 9167 } 9168 pd.sport = &pd.hdr.sctp.src_port; 9169 pd.dport = &pd.hdr.sctp.dest_port; 9170 if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) { 9171 action = PF_DROP; 9172 REASON_SET(&reason, PFRES_SHORT); 9173 goto done; 9174 } 9175 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 9176 if (action == PF_DROP) 9177 goto done; 9178 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 9179 &reason); 9180 if (action == PF_PASS) { 9181 if (V_pfsync_update_state_ptr != NULL) 9182 V_pfsync_update_state_ptr(s); 9183 r = s->rule.ptr; 9184 a = s->anchor.ptr; 9185 } else if (s == NULL) { 9186 action = pf_test_rule(&r, &s, kif, m, off, 9187 &pd, &a, &ruleset, inp); 9188 } 9189 break; 9190 } 9191 9192 case IPPROTO_ICMP: { 9193 action = PF_DROP; 9194 DPFPRINTF(PF_DEBUG_MISC, 9195 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 9196 goto done; 9197 } 9198 9199 case IPPROTO_ICMPV6: { 9200 if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6), 9201 &action, &reason, AF_INET6)) { 9202 if (action != PF_PASS) 9203 pd.act.log |= PF_LOG_FORCE; 9204 goto done; 9205 } 9206 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 9207 if (action == PF_PASS) { 9208 if (V_pfsync_update_state_ptr != NULL) 9209 V_pfsync_update_state_ptr(s); 9210 r = s->rule.ptr; 9211 a = s->anchor.ptr; 9212 } else if (s == NULL) 9213 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9214 &a, &ruleset, inp); 9215 break; 9216 } 9217 9218 default: 9219 action = pf_test_state_other(&s, kif, m, &pd); 9220 if (action == PF_PASS) { 9221 if (V_pfsync_update_state_ptr != NULL) 9222 V_pfsync_update_state_ptr(s); 9223 r = s->rule.ptr; 9224 a = s->anchor.ptr; 9225 } else if (s == NULL) 9226 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9227 &a, &ruleset, inp); 9228 break; 9229 } 9230 9231 done: 9232 PF_RULES_RUNLOCK(); 9233 if (n != m) { 9234 m_freem(n); 9235 n = NULL; 9236 } 9237 9238 /* handle dangerous IPv6 extension headers. */ 9239 if (action == PF_PASS && rh_cnt && 9240 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 9241 action = PF_DROP; 9242 REASON_SET(&reason, PFRES_IPOPTIONS); 9243 pd.act.log = r->log; 9244 DPFPRINTF(PF_DEBUG_MISC, 9245 ("pf: dropping packet with dangerous v6 headers\n")); 9246 } 9247 9248 if (s) { 9249 uint8_t log = pd.act.log; 9250 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 9251 pd.act.log |= log; 9252 tag = s->tag; 9253 rt = s->rt; 9254 } else { 9255 tag = r->tag; 9256 rt = r->rt; 9257 } 9258 9259 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 9260 action = PF_DROP; 9261 REASON_SET(&reason, PFRES_MEMORY); 9262 } 9263 9264 pf_scrub_ip6(&m, &pd); 9265 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 9266 pf_normalize_mss(m, off, &pd); 9267 9268 if (pd.act.rtableid >= 0) 9269 M_SETFIB(m, pd.act.rtableid); 9270 9271 if (pd.act.flags & PFSTATE_SETPRIO) { 9272 if (pd.tos & IPTOS_LOWDELAY) 9273 use_2nd_queue = 1; 9274 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 9275 action = PF_DROP; 9276 REASON_SET(&reason, PFRES_MEMORY); 9277 pd.act.log = PF_LOG_FORCE; 9278 DPFPRINTF(PF_DEBUG_MISC, 9279 ("pf: failed to allocate 802.1q mtag\n")); 9280 } 9281 } 9282 9283 #ifdef ALTQ 9284 if (action == PF_PASS && pd.act.qid) { 9285 if (pd.pf_mtag == NULL && 9286 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9287 action = PF_DROP; 9288 REASON_SET(&reason, PFRES_MEMORY); 9289 } else { 9290 if (s != NULL) 9291 pd.pf_mtag->qid_hash = pf_state_hash(s); 9292 if (pd.tos & IPTOS_LOWDELAY) 9293 pd.pf_mtag->qid = pd.act.pqid; 9294 else 9295 pd.pf_mtag->qid = pd.act.qid; 9296 /* Add hints for ecn. */ 9297 pd.pf_mtag->hdr = h; 9298 } 9299 } 9300 #endif /* ALTQ */ 9301 9302 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9303 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 9304 (s->nat_rule.ptr->action == PF_RDR || 9305 s->nat_rule.ptr->action == PF_BINAT) && 9306 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 9307 m->m_flags |= M_SKIP_FIREWALL; 9308 9309 /* XXX: Anybody working on it?! */ 9310 if (r->divert.port) 9311 printf("pf: divert(9) is not supported for IPv6\n"); 9312 9313 if (pd.act.log) { 9314 struct pf_krule *lr; 9315 struct pf_krule_item *ri; 9316 9317 if (s != NULL && s->nat_rule.ptr != NULL && 9318 s->nat_rule.ptr->log & PF_LOG_ALL) 9319 lr = s->nat_rule.ptr; 9320 else 9321 lr = r; 9322 9323 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9324 PFLOG_PACKET(kif, m, AF_INET6, action, reason, lr, a, ruleset, 9325 &pd, (s == NULL)); 9326 if (s) { 9327 SLIST_FOREACH(ri, &s->match_rules, entry) 9328 if (ri->r->log & PF_LOG_ALL) 9329 PFLOG_PACKET(kif, m, AF_INET6, action, reason, 9330 ri->r, a, ruleset, &pd, 0); 9331 } 9332 } 9333 9334 pf_counter_u64_critical_enter(); 9335 pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS], 9336 pd.tot_len); 9337 pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS], 9338 1); 9339 9340 if (action == PF_PASS || r->action == PF_DROP) { 9341 dirndx = (dir == PF_OUT); 9342 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 9343 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 9344 if (a != NULL) { 9345 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 9346 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 9347 } 9348 if (s != NULL) { 9349 if (s->nat_rule.ptr != NULL) { 9350 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 9351 1); 9352 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 9353 pd.tot_len); 9354 } 9355 if (s->src_node != NULL) { 9356 counter_u64_add(s->src_node->packets[dirndx], 9357 1); 9358 counter_u64_add(s->src_node->bytes[dirndx], 9359 pd.tot_len); 9360 } 9361 if (s->nat_src_node != NULL) { 9362 counter_u64_add(s->nat_src_node->packets[dirndx], 9363 1); 9364 counter_u64_add(s->nat_src_node->bytes[dirndx], 9365 pd.tot_len); 9366 } 9367 dirndx = (dir == s->direction) ? 0 : 1; 9368 s->packets[dirndx]++; 9369 s->bytes[dirndx] += pd.tot_len; 9370 } 9371 tr = r; 9372 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 9373 if (nr != NULL && r == &V_pf_default_rule) 9374 tr = nr; 9375 if (tr->src.addr.type == PF_ADDR_TABLE) 9376 pfr_update_stats(tr->src.addr.p.tbl, 9377 (s == NULL) ? pd.src : 9378 &s->key[(s->direction == PF_IN)]->addr[0], 9379 pd.af, pd.tot_len, dir == PF_OUT, 9380 r->action == PF_PASS, tr->src.neg); 9381 if (tr->dst.addr.type == PF_ADDR_TABLE) 9382 pfr_update_stats(tr->dst.addr.p.tbl, 9383 (s == NULL) ? pd.dst : 9384 &s->key[(s->direction == PF_IN)]->addr[1], 9385 pd.af, pd.tot_len, dir == PF_OUT, 9386 r->action == PF_PASS, tr->dst.neg); 9387 } 9388 pf_counter_u64_critical_exit(); 9389 9390 switch (action) { 9391 case PF_SYNPROXY_DROP: 9392 m_freem(*m0); 9393 case PF_DEFER: 9394 *m0 = NULL; 9395 action = PF_PASS; 9396 break; 9397 case PF_DROP: 9398 m_freem(*m0); 9399 *m0 = NULL; 9400 break; 9401 default: 9402 /* pf_route6() returns unlocked. */ 9403 if (rt) { 9404 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 9405 goto out; 9406 } 9407 if (pf_dummynet(&pd, s, r, m0) != 0) { 9408 action = PF_DROP; 9409 REASON_SET(&reason, PFRES_MEMORY); 9410 } 9411 break; 9412 } 9413 9414 if (s && action != PF_DROP) { 9415 if (!s->if_index_in && dir == PF_IN) 9416 s->if_index_in = ifp->if_index; 9417 else if (!s->if_index_out && dir == PF_OUT) 9418 s->if_index_out = ifp->if_index; 9419 } 9420 9421 if (s) 9422 PF_STATE_UNLOCK(s); 9423 9424 /* If reassembled packet passed, create new fragments. */ 9425 if (action == PF_PASS && *m0 && dir == PF_OUT && 9426 (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 9427 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 9428 9429 out: 9430 SDT_PROBE4(pf, ip, test6, done, action, reason, r, s); 9431 9432 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 9433 9434 return (action); 9435 } 9436 #endif /* INET6 */ 9437