1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40 #include <sys/cdefs.h> 41 #include "opt_bpf.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_pf.h" 45 #include "opt_sctp.h" 46 47 #include <sys/param.h> 48 #include <sys/bus.h> 49 #include <sys/endian.h> 50 #include <sys/gsb_crc32.h> 51 #include <sys/hash.h> 52 #include <sys/interrupt.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/limits.h> 56 #include <sys/mbuf.h> 57 #include <sys/md5.h> 58 #include <sys/random.h> 59 #include <sys/refcount.h> 60 #include <sys/sdt.h> 61 #include <sys/socket.h> 62 #include <sys/sysctl.h> 63 #include <sys/taskqueue.h> 64 #include <sys/ucred.h> 65 66 #include <net/if.h> 67 #include <net/if_var.h> 68 #include <net/if_private.h> 69 #include <net/if_types.h> 70 #include <net/if_vlan_var.h> 71 #include <net/route.h> 72 #include <net/route/nhop.h> 73 #include <net/vnet.h> 74 75 #include <net/pfil.h> 76 #include <net/pfvar.h> 77 #include <net/if_pflog.h> 78 #include <net/if_pfsync.h> 79 80 #include <netinet/in_pcb.h> 81 #include <netinet/in_var.h> 82 #include <netinet/in_fib.h> 83 #include <netinet/ip.h> 84 #include <netinet/ip_fw.h> 85 #include <netinet/ip_icmp.h> 86 #include <netinet/icmp_var.h> 87 #include <netinet/ip_var.h> 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/udp.h> 94 #include <netinet/udp_var.h> 95 96 /* dummynet */ 97 #include <netinet/ip_dummynet.h> 98 #include <netinet/ip_fw.h> 99 #include <netpfil/ipfw/dn_heap.h> 100 #include <netpfil/ipfw/ip_fw_private.h> 101 #include <netpfil/ipfw/ip_dn_private.h> 102 103 #ifdef INET6 104 #include <netinet/ip6.h> 105 #include <netinet/icmp6.h> 106 #include <netinet6/nd6.h> 107 #include <netinet6/ip6_var.h> 108 #include <netinet6/in6_pcb.h> 109 #include <netinet6/in6_fib.h> 110 #include <netinet6/scope6_var.h> 111 #endif /* INET6 */ 112 113 #include <netinet/sctp_header.h> 114 #include <netinet/sctp_crc32.h> 115 116 #include <machine/in_cksum.h> 117 #include <security/mac/mac_framework.h> 118 119 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 120 121 SDT_PROVIDER_DEFINE(pf); 122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 123 "struct pf_kstate *"); 124 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 125 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 126 "struct pf_kstate *"); 127 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 128 "struct pfi_kkif *"); 129 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 130 "struct pf_krule *", "struct mbuf *", "int"); 131 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 132 "struct pf_sctp_source *"); 133 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 134 "struct pf_kstate *", "struct pf_sctp_source *"); 135 136 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 137 "struct mbuf *"); 138 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 139 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 140 "int", "struct pf_keth_rule *", "char *"); 141 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 142 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 143 "int", "struct pf_keth_rule *"); 144 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 145 146 /* 147 * Global variables 148 */ 149 150 /* state tables */ 151 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 152 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 153 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 154 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 155 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 156 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 157 VNET_DEFINE(struct pf_kstatus, pf_status); 158 159 VNET_DEFINE(u_int32_t, ticket_altqs_active); 160 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 161 VNET_DEFINE(int, altqs_inactive_open); 162 VNET_DEFINE(u_int32_t, ticket_pabuf); 163 164 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 165 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 166 VNET_DEFINE(u_char, pf_tcp_secret[16]); 167 #define V_pf_tcp_secret VNET(pf_tcp_secret) 168 VNET_DEFINE(int, pf_tcp_secret_init); 169 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 170 VNET_DEFINE(int, pf_tcp_iss_off); 171 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 172 VNET_DECLARE(int, pf_vnet_active); 173 #define V_pf_vnet_active VNET(pf_vnet_active) 174 175 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 176 #define V_pf_purge_idx VNET(pf_purge_idx) 177 178 #ifdef PF_WANT_32_TO_64_COUNTER 179 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 180 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 181 182 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 183 VNET_DEFINE(size_t, pf_allrulecount); 184 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 185 #endif 186 187 struct pf_sctp_endpoint; 188 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 189 struct pf_sctp_source { 190 sa_family_t af; 191 struct pf_addr addr; 192 TAILQ_ENTRY(pf_sctp_source) entry; 193 }; 194 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 195 struct pf_sctp_endpoint 196 { 197 uint32_t v_tag; 198 struct pf_sctp_sources sources; 199 RB_ENTRY(pf_sctp_endpoint) entry; 200 }; 201 static int 202 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 203 { 204 return (a->v_tag - b->v_tag); 205 } 206 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 207 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 208 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 209 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 210 static struct mtx_padalign pf_sctp_endpoints_mtx; 211 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 212 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 213 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 214 215 /* 216 * Queue for pf_intr() sends. 217 */ 218 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 219 struct pf_send_entry { 220 STAILQ_ENTRY(pf_send_entry) pfse_next; 221 struct mbuf *pfse_m; 222 enum { 223 PFSE_IP, 224 PFSE_IP6, 225 PFSE_ICMP, 226 PFSE_ICMP6, 227 } pfse_type; 228 struct { 229 int type; 230 int code; 231 int mtu; 232 } icmpopts; 233 }; 234 235 STAILQ_HEAD(pf_send_head, pf_send_entry); 236 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 237 #define V_pf_sendqueue VNET(pf_sendqueue) 238 239 static struct mtx_padalign pf_sendqueue_mtx; 240 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 241 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 242 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 243 244 /* 245 * Queue for pf_overload_task() tasks. 246 */ 247 struct pf_overload_entry { 248 SLIST_ENTRY(pf_overload_entry) next; 249 struct pf_addr addr; 250 sa_family_t af; 251 uint8_t dir; 252 struct pf_krule *rule; 253 }; 254 255 SLIST_HEAD(pf_overload_head, pf_overload_entry); 256 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 257 #define V_pf_overloadqueue VNET(pf_overloadqueue) 258 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 259 #define V_pf_overloadtask VNET(pf_overloadtask) 260 261 static struct mtx_padalign pf_overloadqueue_mtx; 262 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 263 "pf overload/flush queue", MTX_DEF); 264 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 265 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 266 267 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 268 struct mtx_padalign pf_unlnkdrules_mtx; 269 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 270 MTX_DEF); 271 272 struct sx pf_config_lock; 273 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 274 275 struct mtx_padalign pf_table_stats_lock; 276 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 277 MTX_DEF); 278 279 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 280 #define V_pf_sources_z VNET(pf_sources_z) 281 uma_zone_t pf_mtag_z; 282 VNET_DEFINE(uma_zone_t, pf_state_z); 283 VNET_DEFINE(uma_zone_t, pf_state_key_z); 284 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 285 286 VNET_DEFINE(struct unrhdr64, pf_stateid); 287 288 static void pf_src_tree_remove_state(struct pf_kstate *); 289 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 290 u_int32_t); 291 static void pf_add_threshold(struct pf_threshold *); 292 static int pf_check_threshold(struct pf_threshold *); 293 294 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 295 u_int16_t *, u_int16_t *, struct pf_addr *, 296 u_int16_t, u_int8_t, sa_family_t); 297 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 298 struct tcphdr *, struct pf_state_peer *); 299 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 300 int *, u_int16_t *, u_int16_t *); 301 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 302 struct pf_addr *, struct pf_addr *, u_int16_t, 303 u_int16_t *, u_int16_t *, u_int16_t *, 304 u_int16_t *, u_int8_t, sa_family_t); 305 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 306 sa_family_t, struct pf_krule *, int); 307 static void pf_detach_state(struct pf_kstate *); 308 static int pf_state_key_attach(struct pf_state_key *, 309 struct pf_state_key *, struct pf_kstate *); 310 static void pf_state_key_detach(struct pf_kstate *, int); 311 static int pf_state_key_ctor(void *, int, void *, int); 312 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 313 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 314 struct pf_mtag *pf_mtag); 315 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 316 struct pf_krule *, struct mbuf **); 317 static int pf_dummynet_route(struct pf_pdesc *, 318 struct pf_kstate *, struct pf_krule *, 319 struct ifnet *, struct sockaddr *, struct mbuf **); 320 static int pf_test_eth_rule(int, struct pfi_kkif *, 321 struct mbuf **); 322 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 323 struct pfi_kkif *, struct mbuf *, int, 324 struct pf_pdesc *, struct pf_krule **, 325 struct pf_kruleset **, struct inpcb *, int); 326 static int pf_create_state(struct pf_krule *, struct pf_krule *, 327 struct pf_krule *, struct pf_pdesc *, 328 struct pf_ksrc_node *, struct pf_state_key *, 329 struct pf_state_key *, struct mbuf *, int, 330 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 331 struct pf_kstate **, int, u_int16_t, u_int16_t, 332 int, struct pf_krule_slist *, struct pf_udp_mapping *); 333 static int pf_state_key_addr_setup(struct pf_pdesc *, struct mbuf *, 334 int, struct pf_state_key_cmp *, int, struct pf_addr *, 335 int, struct pf_addr *, int); 336 static int pf_tcp_track_full(struct pf_kstate **, 337 struct pfi_kkif *, struct mbuf *, int, 338 struct pf_pdesc *, u_short *, int *); 339 static int pf_tcp_track_sloppy(struct pf_kstate **, 340 struct pf_pdesc *, u_short *); 341 static int pf_test_state_tcp(struct pf_kstate **, 342 struct pfi_kkif *, struct mbuf *, int, 343 struct pf_pdesc *, u_short *); 344 static int pf_test_state_udp(struct pf_kstate **, 345 struct pfi_kkif *, struct mbuf *, int, 346 struct pf_pdesc *); 347 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 348 struct pf_pdesc *, struct pf_kstate **, struct mbuf *, 349 int, int, struct pfi_kkif *, u_int16_t, u_int16_t, 350 int, int *, int, int); 351 static int pf_test_state_icmp(struct pf_kstate **, 352 struct pfi_kkif *, struct mbuf *, int, 353 struct pf_pdesc *, u_short *); 354 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 355 static void pf_sctp_multihome_delayed(struct pf_pdesc *, int, 356 struct pfi_kkif *, struct pf_kstate *, int); 357 static int pf_test_state_sctp(struct pf_kstate **, 358 struct pfi_kkif *, struct mbuf *, int, 359 struct pf_pdesc *, u_short *); 360 static int pf_test_state_other(struct pf_kstate **, 361 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 362 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 363 int, u_int16_t); 364 static int pf_check_proto_cksum(struct mbuf *, int, int, 365 u_int8_t, sa_family_t); 366 static int pf_walk_option6(struct mbuf *, int, int, uint32_t *, 367 u_short *); 368 static void pf_print_state_parts(struct pf_kstate *, 369 struct pf_state_key *, struct pf_state_key *); 370 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 371 bool, u_int8_t); 372 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 373 const struct pf_state_key_cmp *, u_int); 374 static int pf_src_connlimit(struct pf_kstate **); 375 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 376 static void pf_counters_inc(int, 377 struct pf_pdesc *, struct pfi_kkif *, 378 struct pf_kstate *, struct pf_krule *, 379 struct pf_krule *); 380 static void pf_overload_task(void *v, int pending); 381 static u_short pf_insert_src_node(struct pf_ksrc_node **, 382 struct pf_krule *, struct pf_addr *, sa_family_t); 383 static u_int pf_purge_expired_states(u_int, int); 384 static void pf_purge_unlinked_rules(void); 385 static int pf_mtag_uminit(void *, int, int); 386 static void pf_mtag_free(struct m_tag *); 387 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 388 int, struct pf_state_key *); 389 #ifdef INET 390 static void pf_route(struct mbuf **, struct pf_krule *, 391 struct ifnet *, struct pf_kstate *, 392 struct pf_pdesc *, struct inpcb *); 393 #endif /* INET */ 394 #ifdef INET6 395 static void pf_change_a6(struct pf_addr *, u_int16_t *, 396 struct pf_addr *, u_int8_t); 397 static void pf_route6(struct mbuf **, struct pf_krule *, 398 struct ifnet *, struct pf_kstate *, 399 struct pf_pdesc *, struct inpcb *); 400 #endif /* INET6 */ 401 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 402 403 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 404 405 extern int pf_end_threads; 406 extern struct proc *pf_purge_proc; 407 408 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 409 410 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK }; 411 412 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 413 do { \ 414 struct pf_state_key *nk; \ 415 if ((pd->dir) == PF_OUT) \ 416 nk = (_s)->key[PF_SK_STACK]; \ 417 else \ 418 nk = (_s)->key[PF_SK_WIRE]; \ 419 pf_packet_rework_nat(_m, _pd, _off, nk); \ 420 } while (0) 421 422 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 423 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 424 425 #define STATE_LOOKUP(i, k, s, pd) \ 426 do { \ 427 (s) = pf_find_state((i), (k), (pd->dir)); \ 428 SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s)); \ 429 if ((s) == NULL) \ 430 return (PF_DROP); \ 431 if (PACKET_LOOPED(pd)) \ 432 return (PF_PASS); \ 433 } while (0) 434 435 static struct pfi_kkif * 436 BOUND_IFACE(struct pf_kstate *st, struct pfi_kkif *k) 437 { 438 SDT_PROBE2(pf, ip, , bound_iface, st, k); 439 440 /* Floating unless otherwise specified. */ 441 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 442 return (V_pfi_all); 443 444 /* 445 * Initially set to all, because we don't know what interface we'll be 446 * sending this out when we create the state. 447 */ 448 if (st->rule->rt == PF_REPLYTO) 449 return (V_pfi_all); 450 451 /* Don't overrule the interface for states created on incoming packets. */ 452 if (st->direction == PF_IN) 453 return (k); 454 455 /* No route-to, so don't overrule. */ 456 if (st->rt != PF_ROUTETO) 457 return (k); 458 459 /* Bind to the route-to interface. */ 460 return (st->rt_kif); 461 } 462 463 #define STATE_INC_COUNTERS(s) \ 464 do { \ 465 struct pf_krule_item *mrm; \ 466 counter_u64_add(s->rule->states_cur, 1); \ 467 counter_u64_add(s->rule->states_tot, 1); \ 468 if (s->anchor != NULL) { \ 469 counter_u64_add(s->anchor->states_cur, 1); \ 470 counter_u64_add(s->anchor->states_tot, 1); \ 471 } \ 472 if (s->nat_rule != NULL) { \ 473 counter_u64_add(s->nat_rule->states_cur, 1);\ 474 counter_u64_add(s->nat_rule->states_tot, 1);\ 475 } \ 476 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 477 counter_u64_add(mrm->r->states_cur, 1); \ 478 counter_u64_add(mrm->r->states_tot, 1); \ 479 } \ 480 } while (0) 481 482 #define STATE_DEC_COUNTERS(s) \ 483 do { \ 484 struct pf_krule_item *mrm; \ 485 if (s->nat_rule != NULL) \ 486 counter_u64_add(s->nat_rule->states_cur, -1);\ 487 if (s->anchor != NULL) \ 488 counter_u64_add(s->anchor->states_cur, -1); \ 489 counter_u64_add(s->rule->states_cur, -1); \ 490 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 491 counter_u64_add(mrm->r->states_cur, -1); \ 492 } while (0) 493 494 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 495 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 496 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 497 VNET_DEFINE(struct pf_idhash *, pf_idhash); 498 VNET_DEFINE(struct pf_srchash *, pf_srchash); 499 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 500 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 501 502 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 503 "pf(4)"); 504 505 VNET_DEFINE(u_long, pf_hashmask); 506 VNET_DEFINE(u_long, pf_srchashmask); 507 VNET_DEFINE(u_long, pf_udpendpointhashmask); 508 VNET_DEFINE_STATIC(u_long, pf_hashsize); 509 #define V_pf_hashsize VNET(pf_hashsize) 510 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 511 #define V_pf_srchashsize VNET(pf_srchashsize) 512 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 513 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 514 u_long pf_ioctl_maxcount = 65535; 515 516 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 517 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 518 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 519 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 520 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 521 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 522 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 523 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 524 525 VNET_DEFINE(void *, pf_swi_cookie); 526 VNET_DEFINE(struct intr_event *, pf_swi_ie); 527 528 VNET_DEFINE(uint32_t, pf_hashseed); 529 #define V_pf_hashseed VNET(pf_hashseed) 530 531 static void 532 pf_sctp_checksum(struct mbuf *m, int off) 533 { 534 uint32_t sum = 0; 535 536 /* Zero out the checksum, to enable recalculation. */ 537 m_copyback(m, off + offsetof(struct sctphdr, checksum), 538 sizeof(sum), (caddr_t)&sum); 539 540 sum = sctp_calculate_cksum(m, off); 541 542 m_copyback(m, off + offsetof(struct sctphdr, checksum), 543 sizeof(sum), (caddr_t)&sum); 544 } 545 546 int 547 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 548 { 549 550 switch (af) { 551 #ifdef INET 552 case AF_INET: 553 if (a->addr32[0] > b->addr32[0]) 554 return (1); 555 if (a->addr32[0] < b->addr32[0]) 556 return (-1); 557 break; 558 #endif /* INET */ 559 #ifdef INET6 560 case AF_INET6: 561 if (a->addr32[3] > b->addr32[3]) 562 return (1); 563 if (a->addr32[3] < b->addr32[3]) 564 return (-1); 565 if (a->addr32[2] > b->addr32[2]) 566 return (1); 567 if (a->addr32[2] < b->addr32[2]) 568 return (-1); 569 if (a->addr32[1] > b->addr32[1]) 570 return (1); 571 if (a->addr32[1] < b->addr32[1]) 572 return (-1); 573 if (a->addr32[0] > b->addr32[0]) 574 return (1); 575 if (a->addr32[0] < b->addr32[0]) 576 return (-1); 577 break; 578 #endif /* INET6 */ 579 default: 580 panic("%s: unknown address family %u", __func__, af); 581 } 582 return (0); 583 } 584 585 static bool 586 pf_is_loopback(sa_family_t af, struct pf_addr *addr) 587 { 588 switch (af) { 589 case AF_INET: 590 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 591 case AF_INET6: 592 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 593 default: 594 panic("Unknown af %d", af); 595 } 596 } 597 598 static void 599 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 600 struct pf_state_key *nk) 601 { 602 603 switch (pd->proto) { 604 case IPPROTO_TCP: { 605 struct tcphdr *th = &pd->hdr.tcp; 606 607 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 608 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 609 &th->th_sum, &nk->addr[pd->sidx], 610 nk->port[pd->sidx], 0, pd->af); 611 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 612 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 613 &th->th_sum, &nk->addr[pd->didx], 614 nk->port[pd->didx], 0, pd->af); 615 m_copyback(m, off, sizeof(*th), (caddr_t)th); 616 break; 617 } 618 case IPPROTO_UDP: { 619 struct udphdr *uh = &pd->hdr.udp; 620 621 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 622 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 623 &uh->uh_sum, &nk->addr[pd->sidx], 624 nk->port[pd->sidx], 1, pd->af); 625 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 626 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 627 &uh->uh_sum, &nk->addr[pd->didx], 628 nk->port[pd->didx], 1, pd->af); 629 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 630 break; 631 } 632 case IPPROTO_SCTP: { 633 struct sctphdr *sh = &pd->hdr.sctp; 634 uint16_t checksum = 0; 635 636 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 637 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 638 &checksum, &nk->addr[pd->sidx], 639 nk->port[pd->sidx], 1, pd->af); 640 } 641 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 642 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 643 &checksum, &nk->addr[pd->didx], 644 nk->port[pd->didx], 1, pd->af); 645 } 646 647 break; 648 } 649 case IPPROTO_ICMP: { 650 struct icmp *ih = &pd->hdr.icmp; 651 652 if (nk->port[pd->sidx] != ih->icmp_id) { 653 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 654 ih->icmp_cksum, ih->icmp_id, 655 nk->port[pd->sidx], 0); 656 ih->icmp_id = nk->port[pd->sidx]; 657 pd->sport = &ih->icmp_id; 658 659 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 660 } 661 /* FALLTHROUGH */ 662 } 663 default: 664 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 665 switch (pd->af) { 666 case AF_INET: 667 pf_change_a(&pd->src->v4.s_addr, 668 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 669 0); 670 break; 671 case AF_INET6: 672 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 673 break; 674 } 675 } 676 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 677 switch (pd->af) { 678 case AF_INET: 679 pf_change_a(&pd->dst->v4.s_addr, 680 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 681 0); 682 break; 683 case AF_INET6: 684 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 685 break; 686 } 687 } 688 break; 689 } 690 } 691 692 static __inline uint32_t 693 pf_hashkey(const struct pf_state_key *sk) 694 { 695 uint32_t h; 696 697 h = murmur3_32_hash32((const uint32_t *)sk, 698 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 699 V_pf_hashseed); 700 701 return (h & V_pf_hashmask); 702 } 703 704 static __inline uint32_t 705 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 706 { 707 uint32_t h; 708 709 switch (af) { 710 case AF_INET: 711 h = murmur3_32_hash32((uint32_t *)&addr->v4, 712 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 713 break; 714 case AF_INET6: 715 h = murmur3_32_hash32((uint32_t *)&addr->v6, 716 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 717 break; 718 default: 719 panic("%s: unknown address family %u", __func__, af); 720 } 721 722 return (h & V_pf_srchashmask); 723 } 724 725 static inline uint32_t 726 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 727 { 728 uint32_t h; 729 730 h = murmur3_32_hash32((uint32_t *)endpoint, 731 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 732 V_pf_hashseed); 733 return (h & V_pf_udpendpointhashmask); 734 } 735 736 #ifdef ALTQ 737 static int 738 pf_state_hash(struct pf_kstate *s) 739 { 740 u_int32_t hv = (intptr_t)s / sizeof(*s); 741 742 hv ^= crc32(&s->src, sizeof(s->src)); 743 hv ^= crc32(&s->dst, sizeof(s->dst)); 744 if (hv == 0) 745 hv = 1; 746 return (hv); 747 } 748 #endif 749 750 static __inline void 751 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 752 { 753 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 754 s->dst.state = newstate; 755 if (which == PF_PEER_DST) 756 return; 757 if (s->src.state == newstate) 758 return; 759 if (s->creatorid == V_pf_status.hostid && 760 s->key[PF_SK_STACK] != NULL && 761 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 762 !(TCPS_HAVEESTABLISHED(s->src.state) || 763 s->src.state == TCPS_CLOSED) && 764 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 765 atomic_add_32(&V_pf_status.states_halfopen, -1); 766 767 s->src.state = newstate; 768 } 769 770 #ifdef INET6 771 void 772 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 773 { 774 switch (af) { 775 #ifdef INET 776 case AF_INET: 777 memcpy(&dst->v4, &src->v4, sizeof(dst->v4)); 778 break; 779 #endif /* INET */ 780 case AF_INET6: 781 memcpy(&dst->v6, &src->v6, sizeof(dst->v6)); 782 break; 783 } 784 } 785 #endif /* INET6 */ 786 787 static void 788 pf_init_threshold(struct pf_threshold *threshold, 789 u_int32_t limit, u_int32_t seconds) 790 { 791 threshold->limit = limit * PF_THRESHOLD_MULT; 792 threshold->seconds = seconds; 793 threshold->count = 0; 794 threshold->last = time_uptime; 795 } 796 797 static void 798 pf_add_threshold(struct pf_threshold *threshold) 799 { 800 u_int32_t t = time_uptime, diff = t - threshold->last; 801 802 if (diff >= threshold->seconds) 803 threshold->count = 0; 804 else 805 threshold->count -= threshold->count * diff / 806 threshold->seconds; 807 threshold->count += PF_THRESHOLD_MULT; 808 threshold->last = t; 809 } 810 811 static int 812 pf_check_threshold(struct pf_threshold *threshold) 813 { 814 return (threshold->count > threshold->limit); 815 } 816 817 static int 818 pf_src_connlimit(struct pf_kstate **state) 819 { 820 struct pf_overload_entry *pfoe; 821 int bad = 0; 822 823 PF_STATE_LOCK_ASSERT(*state); 824 /* 825 * XXXKS: The src node is accessed unlocked! 826 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 827 */ 828 829 (*state)->src_node->conn++; 830 (*state)->src.tcp_est = 1; 831 pf_add_threshold(&(*state)->src_node->conn_rate); 832 833 if ((*state)->rule->max_src_conn && 834 (*state)->rule->max_src_conn < 835 (*state)->src_node->conn) { 836 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 837 bad++; 838 } 839 840 if ((*state)->rule->max_src_conn_rate.limit && 841 pf_check_threshold(&(*state)->src_node->conn_rate)) { 842 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 843 bad++; 844 } 845 846 if (!bad) 847 return (0); 848 849 /* Kill this state. */ 850 (*state)->timeout = PFTM_PURGE; 851 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 852 853 if ((*state)->rule->overload_tbl == NULL) 854 return (1); 855 856 /* Schedule overloading and flushing task. */ 857 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 858 if (pfoe == NULL) 859 return (1); /* too bad :( */ 860 861 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 862 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 863 pfoe->rule = (*state)->rule; 864 pfoe->dir = (*state)->direction; 865 PF_OVERLOADQ_LOCK(); 866 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 867 PF_OVERLOADQ_UNLOCK(); 868 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 869 870 return (1); 871 } 872 873 static void 874 pf_overload_task(void *v, int pending) 875 { 876 struct pf_overload_head queue; 877 struct pfr_addr p; 878 struct pf_overload_entry *pfoe, *pfoe1; 879 uint32_t killed = 0; 880 881 CURVNET_SET((struct vnet *)v); 882 883 PF_OVERLOADQ_LOCK(); 884 queue = V_pf_overloadqueue; 885 SLIST_INIT(&V_pf_overloadqueue); 886 PF_OVERLOADQ_UNLOCK(); 887 888 bzero(&p, sizeof(p)); 889 SLIST_FOREACH(pfoe, &queue, next) { 890 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 891 if (V_pf_status.debug >= PF_DEBUG_MISC) { 892 printf("%s: blocking address ", __func__); 893 pf_print_host(&pfoe->addr, 0, pfoe->af); 894 printf("\n"); 895 } 896 897 p.pfra_af = pfoe->af; 898 switch (pfoe->af) { 899 #ifdef INET 900 case AF_INET: 901 p.pfra_net = 32; 902 p.pfra_ip4addr = pfoe->addr.v4; 903 break; 904 #endif 905 #ifdef INET6 906 case AF_INET6: 907 p.pfra_net = 128; 908 p.pfra_ip6addr = pfoe->addr.v6; 909 break; 910 #endif 911 } 912 913 PF_RULES_WLOCK(); 914 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 915 PF_RULES_WUNLOCK(); 916 } 917 918 /* 919 * Remove those entries, that don't need flushing. 920 */ 921 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 922 if (pfoe->rule->flush == 0) { 923 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 924 free(pfoe, M_PFTEMP); 925 } else 926 counter_u64_add( 927 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 928 929 /* If nothing to flush, return. */ 930 if (SLIST_EMPTY(&queue)) { 931 CURVNET_RESTORE(); 932 return; 933 } 934 935 for (int i = 0; i <= V_pf_hashmask; i++) { 936 struct pf_idhash *ih = &V_pf_idhash[i]; 937 struct pf_state_key *sk; 938 struct pf_kstate *s; 939 940 PF_HASHROW_LOCK(ih); 941 LIST_FOREACH(s, &ih->states, entry) { 942 sk = s->key[PF_SK_WIRE]; 943 SLIST_FOREACH(pfoe, &queue, next) 944 if (sk->af == pfoe->af && 945 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 946 pfoe->rule == s->rule) && 947 ((pfoe->dir == PF_OUT && 948 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 949 (pfoe->dir == PF_IN && 950 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 951 s->timeout = PFTM_PURGE; 952 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 953 killed++; 954 } 955 } 956 PF_HASHROW_UNLOCK(ih); 957 } 958 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 959 free(pfoe, M_PFTEMP); 960 if (V_pf_status.debug >= PF_DEBUG_MISC) 961 printf("%s: %u states killed", __func__, killed); 962 963 CURVNET_RESTORE(); 964 } 965 966 /* 967 * Can return locked on failure, so that we can consistently 968 * allocate and insert a new one. 969 */ 970 struct pf_ksrc_node * 971 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 972 struct pf_srchash **sh, bool returnlocked) 973 { 974 struct pf_ksrc_node *n; 975 976 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 977 978 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 979 PF_HASHROW_LOCK(*sh); 980 LIST_FOREACH(n, &(*sh)->nodes, entry) 981 if (n->rule == rule && n->af == af && 982 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 983 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 984 break; 985 986 if (n != NULL) { 987 n->states++; 988 PF_HASHROW_UNLOCK(*sh); 989 } else if (returnlocked == false) 990 PF_HASHROW_UNLOCK(*sh); 991 992 return (n); 993 } 994 995 static void 996 pf_free_src_node(struct pf_ksrc_node *sn) 997 { 998 999 for (int i = 0; i < 2; i++) { 1000 counter_u64_free(sn->bytes[i]); 1001 counter_u64_free(sn->packets[i]); 1002 } 1003 uma_zfree(V_pf_sources_z, sn); 1004 } 1005 1006 static u_short 1007 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 1008 struct pf_addr *src, sa_family_t af) 1009 { 1010 u_short reason = 0; 1011 struct pf_srchash *sh = NULL; 1012 1013 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 1014 rule->rpool.opts & PF_POOL_STICKYADDR), 1015 ("%s for non-tracking rule %p", __func__, rule)); 1016 1017 if (*sn == NULL) 1018 *sn = pf_find_src_node(src, rule, af, &sh, true); 1019 1020 if (*sn == NULL) { 1021 PF_HASHROW_ASSERT(sh); 1022 1023 if (rule->max_src_nodes && 1024 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 1025 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1026 PF_HASHROW_UNLOCK(sh); 1027 reason = PFRES_SRCLIMIT; 1028 goto done; 1029 } 1030 1031 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1032 if ((*sn) == NULL) { 1033 PF_HASHROW_UNLOCK(sh); 1034 reason = PFRES_MEMORY; 1035 goto done; 1036 } 1037 1038 for (int i = 0; i < 2; i++) { 1039 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1040 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1041 1042 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1043 pf_free_src_node(*sn); 1044 PF_HASHROW_UNLOCK(sh); 1045 reason = PFRES_MEMORY; 1046 goto done; 1047 } 1048 } 1049 1050 pf_init_threshold(&(*sn)->conn_rate, 1051 rule->max_src_conn_rate.limit, 1052 rule->max_src_conn_rate.seconds); 1053 1054 MPASS((*sn)->lock == NULL); 1055 (*sn)->lock = &sh->lock; 1056 1057 (*sn)->af = af; 1058 (*sn)->rule = rule; 1059 PF_ACPY(&(*sn)->addr, src, af); 1060 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 1061 (*sn)->creation = time_uptime; 1062 (*sn)->ruletype = rule->action; 1063 (*sn)->states = 1; 1064 if ((*sn)->rule != NULL) 1065 counter_u64_add((*sn)->rule->src_nodes, 1); 1066 PF_HASHROW_UNLOCK(sh); 1067 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1068 } else { 1069 if (rule->max_src_states && 1070 (*sn)->states >= rule->max_src_states) { 1071 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1072 1); 1073 reason = PFRES_SRCLIMIT; 1074 goto done; 1075 } 1076 } 1077 done: 1078 return (reason); 1079 } 1080 1081 void 1082 pf_unlink_src_node(struct pf_ksrc_node *src) 1083 { 1084 PF_SRC_NODE_LOCK_ASSERT(src); 1085 1086 LIST_REMOVE(src, entry); 1087 if (src->rule) 1088 counter_u64_add(src->rule->src_nodes, -1); 1089 } 1090 1091 u_int 1092 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1093 { 1094 struct pf_ksrc_node *sn, *tmp; 1095 u_int count = 0; 1096 1097 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1098 pf_free_src_node(sn); 1099 count++; 1100 } 1101 1102 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1103 1104 return (count); 1105 } 1106 1107 void 1108 pf_mtag_initialize(void) 1109 { 1110 1111 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1112 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1113 UMA_ALIGN_PTR, 0); 1114 } 1115 1116 /* Per-vnet data storage structures initialization. */ 1117 void 1118 pf_initialize(void) 1119 { 1120 struct pf_keyhash *kh; 1121 struct pf_idhash *ih; 1122 struct pf_srchash *sh; 1123 struct pf_udpendpointhash *uh; 1124 u_int i; 1125 1126 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1127 V_pf_hashsize = PF_HASHSIZ; 1128 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1129 V_pf_srchashsize = PF_SRCHASHSIZ; 1130 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1131 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1132 1133 V_pf_hashseed = arc4random(); 1134 1135 /* States and state keys storage. */ 1136 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1137 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1138 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1139 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1140 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1141 1142 V_pf_state_key_z = uma_zcreate("pf state keys", 1143 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1144 UMA_ALIGN_PTR, 0); 1145 1146 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1147 M_PFHASH, M_NOWAIT | M_ZERO); 1148 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1149 M_PFHASH, M_NOWAIT | M_ZERO); 1150 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1151 printf("pf: Unable to allocate memory for " 1152 "state_hashsize %lu.\n", V_pf_hashsize); 1153 1154 free(V_pf_keyhash, M_PFHASH); 1155 free(V_pf_idhash, M_PFHASH); 1156 1157 V_pf_hashsize = PF_HASHSIZ; 1158 V_pf_keyhash = mallocarray(V_pf_hashsize, 1159 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1160 V_pf_idhash = mallocarray(V_pf_hashsize, 1161 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1162 } 1163 1164 V_pf_hashmask = V_pf_hashsize - 1; 1165 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1166 i++, kh++, ih++) { 1167 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1168 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1169 } 1170 1171 /* Source nodes. */ 1172 V_pf_sources_z = uma_zcreate("pf source nodes", 1173 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1174 0); 1175 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1176 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1177 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1178 1179 V_pf_srchash = mallocarray(V_pf_srchashsize, 1180 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1181 if (V_pf_srchash == NULL) { 1182 printf("pf: Unable to allocate memory for " 1183 "source_hashsize %lu.\n", V_pf_srchashsize); 1184 1185 V_pf_srchashsize = PF_SRCHASHSIZ; 1186 V_pf_srchash = mallocarray(V_pf_srchashsize, 1187 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1188 } 1189 1190 V_pf_srchashmask = V_pf_srchashsize - 1; 1191 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1192 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1193 1194 1195 /* UDP endpoint mappings. */ 1196 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1197 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1198 UMA_ALIGN_PTR, 0); 1199 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1200 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1201 if (V_pf_udpendpointhash == NULL) { 1202 printf("pf: Unable to allocate memory for " 1203 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1204 1205 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1206 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1207 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1208 } 1209 1210 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1211 for (i = 0, uh = V_pf_udpendpointhash; 1212 i <= V_pf_udpendpointhashmask; 1213 i++, uh++) { 1214 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1215 MTX_DEF | MTX_DUPOK); 1216 } 1217 1218 /* ALTQ */ 1219 TAILQ_INIT(&V_pf_altqs[0]); 1220 TAILQ_INIT(&V_pf_altqs[1]); 1221 TAILQ_INIT(&V_pf_altqs[2]); 1222 TAILQ_INIT(&V_pf_altqs[3]); 1223 TAILQ_INIT(&V_pf_pabuf); 1224 V_pf_altqs_active = &V_pf_altqs[0]; 1225 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1226 V_pf_altqs_inactive = &V_pf_altqs[2]; 1227 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1228 1229 /* Send & overload+flush queues. */ 1230 STAILQ_INIT(&V_pf_sendqueue); 1231 SLIST_INIT(&V_pf_overloadqueue); 1232 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1233 1234 /* Unlinked, but may be referenced rules. */ 1235 TAILQ_INIT(&V_pf_unlinked_rules); 1236 } 1237 1238 void 1239 pf_mtag_cleanup(void) 1240 { 1241 1242 uma_zdestroy(pf_mtag_z); 1243 } 1244 1245 void 1246 pf_cleanup(void) 1247 { 1248 struct pf_keyhash *kh; 1249 struct pf_idhash *ih; 1250 struct pf_srchash *sh; 1251 struct pf_udpendpointhash *uh; 1252 struct pf_send_entry *pfse, *next; 1253 u_int i; 1254 1255 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1256 i <= V_pf_hashmask; 1257 i++, kh++, ih++) { 1258 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1259 __func__)); 1260 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1261 __func__)); 1262 mtx_destroy(&kh->lock); 1263 mtx_destroy(&ih->lock); 1264 } 1265 free(V_pf_keyhash, M_PFHASH); 1266 free(V_pf_idhash, M_PFHASH); 1267 1268 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1269 KASSERT(LIST_EMPTY(&sh->nodes), 1270 ("%s: source node hash not empty", __func__)); 1271 mtx_destroy(&sh->lock); 1272 } 1273 free(V_pf_srchash, M_PFHASH); 1274 1275 for (i = 0, uh = V_pf_udpendpointhash; 1276 i <= V_pf_udpendpointhashmask; 1277 i++, uh++) { 1278 KASSERT(LIST_EMPTY(&uh->endpoints), 1279 ("%s: udp endpoint hash not empty", __func__)); 1280 mtx_destroy(&uh->lock); 1281 } 1282 free(V_pf_udpendpointhash, M_PFHASH); 1283 1284 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1285 m_freem(pfse->pfse_m); 1286 free(pfse, M_PFTEMP); 1287 } 1288 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1289 1290 uma_zdestroy(V_pf_sources_z); 1291 uma_zdestroy(V_pf_state_z); 1292 uma_zdestroy(V_pf_state_key_z); 1293 uma_zdestroy(V_pf_udp_mapping_z); 1294 } 1295 1296 static int 1297 pf_mtag_uminit(void *mem, int size, int how) 1298 { 1299 struct m_tag *t; 1300 1301 t = (struct m_tag *)mem; 1302 t->m_tag_cookie = MTAG_ABI_COMPAT; 1303 t->m_tag_id = PACKET_TAG_PF; 1304 t->m_tag_len = sizeof(struct pf_mtag); 1305 t->m_tag_free = pf_mtag_free; 1306 1307 return (0); 1308 } 1309 1310 static void 1311 pf_mtag_free(struct m_tag *t) 1312 { 1313 1314 uma_zfree(pf_mtag_z, t); 1315 } 1316 1317 struct pf_mtag * 1318 pf_get_mtag(struct mbuf *m) 1319 { 1320 struct m_tag *mtag; 1321 1322 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1323 return ((struct pf_mtag *)(mtag + 1)); 1324 1325 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1326 if (mtag == NULL) 1327 return (NULL); 1328 bzero(mtag + 1, sizeof(struct pf_mtag)); 1329 m_tag_prepend(m, mtag); 1330 1331 return ((struct pf_mtag *)(mtag + 1)); 1332 } 1333 1334 static int 1335 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1336 struct pf_kstate *s) 1337 { 1338 struct pf_keyhash *khs, *khw, *kh; 1339 struct pf_state_key *sk, *cur; 1340 struct pf_kstate *si, *olds = NULL; 1341 int idx; 1342 1343 NET_EPOCH_ASSERT(); 1344 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1345 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1346 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1347 1348 /* 1349 * We need to lock hash slots of both keys. To avoid deadlock 1350 * we always lock the slot with lower address first. Unlock order 1351 * isn't important. 1352 * 1353 * We also need to lock ID hash slot before dropping key 1354 * locks. On success we return with ID hash slot locked. 1355 */ 1356 1357 if (skw == sks) { 1358 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1359 PF_HASHROW_LOCK(khs); 1360 } else { 1361 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1362 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1363 if (khs == khw) { 1364 PF_HASHROW_LOCK(khs); 1365 } else if (khs < khw) { 1366 PF_HASHROW_LOCK(khs); 1367 PF_HASHROW_LOCK(khw); 1368 } else { 1369 PF_HASHROW_LOCK(khw); 1370 PF_HASHROW_LOCK(khs); 1371 } 1372 } 1373 1374 #define KEYS_UNLOCK() do { \ 1375 if (khs != khw) { \ 1376 PF_HASHROW_UNLOCK(khs); \ 1377 PF_HASHROW_UNLOCK(khw); \ 1378 } else \ 1379 PF_HASHROW_UNLOCK(khs); \ 1380 } while (0) 1381 1382 /* 1383 * First run: start with wire key. 1384 */ 1385 sk = skw; 1386 kh = khw; 1387 idx = PF_SK_WIRE; 1388 1389 MPASS(s->lock == NULL); 1390 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1391 1392 keyattach: 1393 LIST_FOREACH(cur, &kh->keys, entry) 1394 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1395 break; 1396 1397 if (cur != NULL) { 1398 /* Key exists. Check for same kif, if none, add to key. */ 1399 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1400 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1401 1402 PF_HASHROW_LOCK(ih); 1403 if (si->kif == s->kif && 1404 si->direction == s->direction) { 1405 if (sk->proto == IPPROTO_TCP && 1406 si->src.state >= TCPS_FIN_WAIT_2 && 1407 si->dst.state >= TCPS_FIN_WAIT_2) { 1408 /* 1409 * New state matches an old >FIN_WAIT_2 1410 * state. We can't drop key hash locks, 1411 * thus we can't unlink it properly. 1412 * 1413 * As a workaround we drop it into 1414 * TCPS_CLOSED state, schedule purge 1415 * ASAP and push it into the very end 1416 * of the slot TAILQ, so that it won't 1417 * conflict with our new state. 1418 */ 1419 pf_set_protostate(si, PF_PEER_BOTH, 1420 TCPS_CLOSED); 1421 si->timeout = PFTM_PURGE; 1422 olds = si; 1423 } else { 1424 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1425 printf("pf: %s key attach " 1426 "failed on %s: ", 1427 (idx == PF_SK_WIRE) ? 1428 "wire" : "stack", 1429 s->kif->pfik_name); 1430 pf_print_state_parts(s, 1431 (idx == PF_SK_WIRE) ? 1432 sk : NULL, 1433 (idx == PF_SK_STACK) ? 1434 sk : NULL); 1435 printf(", existing: "); 1436 pf_print_state_parts(si, 1437 (idx == PF_SK_WIRE) ? 1438 sk : NULL, 1439 (idx == PF_SK_STACK) ? 1440 sk : NULL); 1441 printf("\n"); 1442 } 1443 s->timeout = PFTM_UNLINKED; 1444 PF_HASHROW_UNLOCK(ih); 1445 KEYS_UNLOCK(); 1446 uma_zfree(V_pf_state_key_z, sk); 1447 if (idx == PF_SK_STACK) 1448 pf_detach_state(s); 1449 return (EEXIST); /* collision! */ 1450 } 1451 } 1452 PF_HASHROW_UNLOCK(ih); 1453 } 1454 uma_zfree(V_pf_state_key_z, sk); 1455 s->key[idx] = cur; 1456 } else { 1457 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1458 s->key[idx] = sk; 1459 } 1460 1461 stateattach: 1462 /* List is sorted, if-bound states before floating. */ 1463 if (s->kif == V_pfi_all) 1464 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1465 else 1466 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1467 1468 if (olds) { 1469 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1470 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1471 key_list[idx]); 1472 olds = NULL; 1473 } 1474 1475 /* 1476 * Attach done. See how should we (or should not?) 1477 * attach a second key. 1478 */ 1479 if (sks == skw) { 1480 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1481 idx = PF_SK_STACK; 1482 sks = NULL; 1483 goto stateattach; 1484 } else if (sks != NULL) { 1485 /* 1486 * Continue attaching with stack key. 1487 */ 1488 sk = sks; 1489 kh = khs; 1490 idx = PF_SK_STACK; 1491 sks = NULL; 1492 goto keyattach; 1493 } 1494 1495 PF_STATE_LOCK(s); 1496 KEYS_UNLOCK(); 1497 1498 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1499 ("%s failure", __func__)); 1500 1501 return (0); 1502 #undef KEYS_UNLOCK 1503 } 1504 1505 static void 1506 pf_detach_state(struct pf_kstate *s) 1507 { 1508 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1509 struct pf_keyhash *kh; 1510 1511 NET_EPOCH_ASSERT(); 1512 MPASS(s->timeout >= PFTM_MAX); 1513 1514 pf_sctp_multihome_detach_addr(s); 1515 1516 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1517 V_pflow_export_state_ptr(s); 1518 1519 if (sks != NULL) { 1520 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1521 PF_HASHROW_LOCK(kh); 1522 if (s->key[PF_SK_STACK] != NULL) 1523 pf_state_key_detach(s, PF_SK_STACK); 1524 /* 1525 * If both point to same key, then we are done. 1526 */ 1527 if (sks == s->key[PF_SK_WIRE]) { 1528 pf_state_key_detach(s, PF_SK_WIRE); 1529 PF_HASHROW_UNLOCK(kh); 1530 return; 1531 } 1532 PF_HASHROW_UNLOCK(kh); 1533 } 1534 1535 if (s->key[PF_SK_WIRE] != NULL) { 1536 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1537 PF_HASHROW_LOCK(kh); 1538 if (s->key[PF_SK_WIRE] != NULL) 1539 pf_state_key_detach(s, PF_SK_WIRE); 1540 PF_HASHROW_UNLOCK(kh); 1541 } 1542 } 1543 1544 static void 1545 pf_state_key_detach(struct pf_kstate *s, int idx) 1546 { 1547 struct pf_state_key *sk = s->key[idx]; 1548 #ifdef INVARIANTS 1549 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1550 1551 PF_HASHROW_ASSERT(kh); 1552 #endif 1553 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1554 s->key[idx] = NULL; 1555 1556 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1557 LIST_REMOVE(sk, entry); 1558 uma_zfree(V_pf_state_key_z, sk); 1559 } 1560 } 1561 1562 static int 1563 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1564 { 1565 struct pf_state_key *sk = mem; 1566 1567 bzero(sk, sizeof(struct pf_state_key_cmp)); 1568 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1569 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1570 1571 return (0); 1572 } 1573 1574 static int 1575 pf_state_key_addr_setup(struct pf_pdesc *pd, struct mbuf *m, int off, 1576 struct pf_state_key_cmp *key, int sidx, struct pf_addr *saddr, 1577 int didx, struct pf_addr *daddr, int multi) 1578 { 1579 #ifdef INET6 1580 struct nd_neighbor_solicit nd; 1581 struct pf_addr *target; 1582 u_short action, reason; 1583 1584 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1585 goto copy; 1586 1587 switch (pd->hdr.icmp6.icmp6_type) { 1588 case ND_NEIGHBOR_SOLICIT: 1589 if (multi) 1590 return (-1); 1591 if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af)) 1592 return (-1); 1593 target = (struct pf_addr *)&nd.nd_ns_target; 1594 daddr = target; 1595 break; 1596 case ND_NEIGHBOR_ADVERT: 1597 if (multi) 1598 return (-1); 1599 if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af)) 1600 return (-1); 1601 target = (struct pf_addr *)&nd.nd_ns_target; 1602 saddr = target; 1603 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1604 key->addr[didx].addr32[0] = 0; 1605 key->addr[didx].addr32[1] = 0; 1606 key->addr[didx].addr32[2] = 0; 1607 key->addr[didx].addr32[3] = 0; 1608 daddr = NULL; /* overwritten */ 1609 } 1610 break; 1611 default: 1612 if (multi == PF_ICMP_MULTI_LINK) { 1613 key->addr[sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1614 key->addr[sidx].addr32[1] = 0; 1615 key->addr[sidx].addr32[2] = 0; 1616 key->addr[sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1617 saddr = NULL; /* overwritten */ 1618 } 1619 } 1620 copy: 1621 #endif 1622 if (saddr) 1623 PF_ACPY(&key->addr[sidx], saddr, pd->af); 1624 if (daddr) 1625 PF_ACPY(&key->addr[didx], daddr, pd->af); 1626 1627 return (0); 1628 } 1629 1630 struct pf_state_key * 1631 pf_state_key_setup(struct pf_pdesc *pd, struct mbuf *m, int off, 1632 struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport, 1633 u_int16_t dport) 1634 { 1635 struct pf_state_key *sk; 1636 1637 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1638 if (sk == NULL) 1639 return (NULL); 1640 1641 if (pf_state_key_addr_setup(pd, m, off, (struct pf_state_key_cmp *)sk, 1642 pd->sidx, pd->src, pd->didx, pd->dst, 0)) { 1643 uma_zfree(V_pf_state_key_z, sk); 1644 return (NULL); 1645 } 1646 1647 sk->port[pd->sidx] = sport; 1648 sk->port[pd->didx] = dport; 1649 sk->proto = pd->proto; 1650 sk->af = pd->af; 1651 1652 return (sk); 1653 } 1654 1655 struct pf_state_key * 1656 pf_state_key_clone(const struct pf_state_key *orig) 1657 { 1658 struct pf_state_key *sk; 1659 1660 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1661 if (sk == NULL) 1662 return (NULL); 1663 1664 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1665 1666 return (sk); 1667 } 1668 1669 int 1670 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1671 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1672 { 1673 struct pf_idhash *ih; 1674 struct pf_kstate *cur; 1675 int error; 1676 1677 NET_EPOCH_ASSERT(); 1678 1679 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1680 ("%s: sks not pristine", __func__)); 1681 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1682 ("%s: skw not pristine", __func__)); 1683 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1684 1685 s->kif = kif; 1686 s->orig_kif = orig_kif; 1687 1688 if (s->id == 0 && s->creatorid == 0) { 1689 s->id = alloc_unr64(&V_pf_stateid); 1690 s->id = htobe64(s->id); 1691 s->creatorid = V_pf_status.hostid; 1692 } 1693 1694 /* Returns with ID locked on success. */ 1695 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1696 return (error); 1697 1698 ih = &V_pf_idhash[PF_IDHASH(s)]; 1699 PF_HASHROW_ASSERT(ih); 1700 LIST_FOREACH(cur, &ih->states, entry) 1701 if (cur->id == s->id && cur->creatorid == s->creatorid) 1702 break; 1703 1704 if (cur != NULL) { 1705 s->timeout = PFTM_UNLINKED; 1706 PF_HASHROW_UNLOCK(ih); 1707 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1708 printf("pf: state ID collision: " 1709 "id: %016llx creatorid: %08x\n", 1710 (unsigned long long)be64toh(s->id), 1711 ntohl(s->creatorid)); 1712 } 1713 pf_detach_state(s); 1714 return (EEXIST); 1715 } 1716 LIST_INSERT_HEAD(&ih->states, s, entry); 1717 /* One for keys, one for ID hash. */ 1718 refcount_init(&s->refs, 2); 1719 1720 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1721 if (V_pfsync_insert_state_ptr != NULL) 1722 V_pfsync_insert_state_ptr(s); 1723 1724 /* Returns locked. */ 1725 return (0); 1726 } 1727 1728 /* 1729 * Find state by ID: returns with locked row on success. 1730 */ 1731 struct pf_kstate * 1732 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1733 { 1734 struct pf_idhash *ih; 1735 struct pf_kstate *s; 1736 1737 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1738 1739 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1740 1741 PF_HASHROW_LOCK(ih); 1742 LIST_FOREACH(s, &ih->states, entry) 1743 if (s->id == id && s->creatorid == creatorid) 1744 break; 1745 1746 if (s == NULL) 1747 PF_HASHROW_UNLOCK(ih); 1748 1749 return (s); 1750 } 1751 1752 /* 1753 * Find state by key. 1754 * Returns with ID hash slot locked on success. 1755 */ 1756 static struct pf_kstate * 1757 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1758 u_int dir) 1759 { 1760 struct pf_keyhash *kh; 1761 struct pf_state_key *sk; 1762 struct pf_kstate *s; 1763 int idx; 1764 1765 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1766 1767 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1768 1769 PF_HASHROW_LOCK(kh); 1770 LIST_FOREACH(sk, &kh->keys, entry) 1771 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1772 break; 1773 if (sk == NULL) { 1774 PF_HASHROW_UNLOCK(kh); 1775 return (NULL); 1776 } 1777 1778 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1779 1780 /* List is sorted, if-bound states before floating ones. */ 1781 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1782 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1783 PF_STATE_LOCK(s); 1784 PF_HASHROW_UNLOCK(kh); 1785 if (__predict_false(s->timeout >= PFTM_MAX)) { 1786 /* 1787 * State is either being processed by 1788 * pf_unlink_state() in an other thread, or 1789 * is scheduled for immediate expiry. 1790 */ 1791 PF_STATE_UNLOCK(s); 1792 return (NULL); 1793 } 1794 return (s); 1795 } 1796 PF_HASHROW_UNLOCK(kh); 1797 1798 return (NULL); 1799 } 1800 1801 /* 1802 * Returns with ID hash slot locked on success. 1803 */ 1804 struct pf_kstate * 1805 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1806 { 1807 struct pf_keyhash *kh; 1808 struct pf_state_key *sk; 1809 struct pf_kstate *s, *ret = NULL; 1810 int idx, inout = 0; 1811 1812 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1813 1814 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1815 1816 PF_HASHROW_LOCK(kh); 1817 LIST_FOREACH(sk, &kh->keys, entry) 1818 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1819 break; 1820 if (sk == NULL) { 1821 PF_HASHROW_UNLOCK(kh); 1822 return (NULL); 1823 } 1824 switch (dir) { 1825 case PF_IN: 1826 idx = PF_SK_WIRE; 1827 break; 1828 case PF_OUT: 1829 idx = PF_SK_STACK; 1830 break; 1831 case PF_INOUT: 1832 idx = PF_SK_WIRE; 1833 inout = 1; 1834 break; 1835 default: 1836 panic("%s: dir %u", __func__, dir); 1837 } 1838 second_run: 1839 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1840 if (more == NULL) { 1841 PF_STATE_LOCK(s); 1842 PF_HASHROW_UNLOCK(kh); 1843 return (s); 1844 } 1845 1846 if (ret) 1847 (*more)++; 1848 else { 1849 ret = s; 1850 PF_STATE_LOCK(s); 1851 } 1852 } 1853 if (inout == 1) { 1854 inout = 0; 1855 idx = PF_SK_STACK; 1856 goto second_run; 1857 } 1858 PF_HASHROW_UNLOCK(kh); 1859 1860 return (ret); 1861 } 1862 1863 /* 1864 * FIXME 1865 * This routine is inefficient -- locks the state only to unlock immediately on 1866 * return. 1867 * It is racy -- after the state is unlocked nothing stops other threads from 1868 * removing it. 1869 */ 1870 bool 1871 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 1872 { 1873 struct pf_kstate *s; 1874 1875 s = pf_find_state_all(key, dir, NULL); 1876 if (s != NULL) { 1877 PF_STATE_UNLOCK(s); 1878 return (true); 1879 } 1880 return (false); 1881 } 1882 1883 struct pf_udp_mapping * 1884 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 1885 struct pf_addr *nat_addr, uint16_t nat_port) 1886 { 1887 struct pf_udp_mapping *mapping; 1888 1889 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 1890 if (mapping == NULL) 1891 return (NULL); 1892 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 1893 mapping->endpoints[0].port = src_port; 1894 mapping->endpoints[0].af = af; 1895 mapping->endpoints[0].mapping = mapping; 1896 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 1897 mapping->endpoints[1].port = nat_port; 1898 mapping->endpoints[1].af = af; 1899 mapping->endpoints[1].mapping = mapping; 1900 refcount_init(&mapping->refs, 1); 1901 return (mapping); 1902 } 1903 1904 int 1905 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 1906 { 1907 struct pf_udpendpointhash *h0, *h1; 1908 struct pf_udp_endpoint *endpoint; 1909 int ret = EEXIST; 1910 1911 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1912 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1913 if (h0 == h1) { 1914 PF_HASHROW_LOCK(h0); 1915 } else if (h0 < h1) { 1916 PF_HASHROW_LOCK(h0); 1917 PF_HASHROW_LOCK(h1); 1918 } else { 1919 PF_HASHROW_LOCK(h1); 1920 PF_HASHROW_LOCK(h0); 1921 } 1922 1923 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 1924 if (bcmp(endpoint, &mapping->endpoints[0], 1925 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1926 break; 1927 } 1928 if (endpoint != NULL) 1929 goto cleanup; 1930 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 1931 if (bcmp(endpoint, &mapping->endpoints[1], 1932 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1933 break; 1934 } 1935 if (endpoint != NULL) 1936 goto cleanup; 1937 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 1938 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 1939 ret = 0; 1940 1941 cleanup: 1942 if (h0 != h1) { 1943 PF_HASHROW_UNLOCK(h0); 1944 PF_HASHROW_UNLOCK(h1); 1945 } else { 1946 PF_HASHROW_UNLOCK(h0); 1947 } 1948 return (ret); 1949 } 1950 1951 void 1952 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 1953 { 1954 /* refcount is synchronized on the source endpoint's row lock */ 1955 struct pf_udpendpointhash *h0, *h1; 1956 1957 if (mapping == NULL) 1958 return; 1959 1960 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1961 PF_HASHROW_LOCK(h0); 1962 if (refcount_release(&mapping->refs)) { 1963 LIST_REMOVE(&mapping->endpoints[0], entry); 1964 PF_HASHROW_UNLOCK(h0); 1965 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1966 PF_HASHROW_LOCK(h1); 1967 LIST_REMOVE(&mapping->endpoints[1], entry); 1968 PF_HASHROW_UNLOCK(h1); 1969 1970 uma_zfree(V_pf_udp_mapping_z, mapping); 1971 } else { 1972 PF_HASHROW_UNLOCK(h0); 1973 } 1974 } 1975 1976 1977 struct pf_udp_mapping * 1978 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 1979 { 1980 struct pf_udpendpointhash *uh; 1981 struct pf_udp_endpoint *endpoint; 1982 1983 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 1984 1985 PF_HASHROW_LOCK(uh); 1986 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 1987 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 1988 bcmp(endpoint, &endpoint->mapping->endpoints[0], 1989 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1990 break; 1991 } 1992 if (endpoint == NULL) { 1993 PF_HASHROW_UNLOCK(uh); 1994 return (NULL); 1995 } 1996 refcount_acquire(&endpoint->mapping->refs); 1997 PF_HASHROW_UNLOCK(uh); 1998 return (endpoint->mapping); 1999 } 2000 /* END state table stuff */ 2001 2002 static void 2003 pf_send(struct pf_send_entry *pfse) 2004 { 2005 2006 PF_SENDQ_LOCK(); 2007 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2008 PF_SENDQ_UNLOCK(); 2009 swi_sched(V_pf_swi_cookie, 0); 2010 } 2011 2012 static bool 2013 pf_isforlocal(struct mbuf *m, int af) 2014 { 2015 switch (af) { 2016 #ifdef INET 2017 case AF_INET: { 2018 struct ip *ip = mtod(m, struct ip *); 2019 2020 return (in_localip(ip->ip_dst)); 2021 } 2022 #endif 2023 #ifdef INET6 2024 case AF_INET6: { 2025 struct ip6_hdr *ip6; 2026 struct in6_ifaddr *ia; 2027 ip6 = mtod(m, struct ip6_hdr *); 2028 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2029 if (ia == NULL) 2030 return (false); 2031 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2032 } 2033 #endif 2034 default: 2035 panic("Unsupported af %d", af); 2036 } 2037 2038 return (false); 2039 } 2040 2041 int 2042 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2043 int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type) 2044 { 2045 /* 2046 * ICMP types marked with PF_OUT are typically responses to 2047 * PF_IN, and will match states in the opposite direction. 2048 * PF_IN ICMP types need to match a state with that type. 2049 */ 2050 *icmp_dir = PF_OUT; 2051 *multi = PF_ICMP_MULTI_LINK; 2052 /* Queries (and responses) */ 2053 switch (pd->af) { 2054 #ifdef INET 2055 case AF_INET: 2056 switch (type) { 2057 case ICMP_ECHO: 2058 *icmp_dir = PF_IN; 2059 case ICMP_ECHOREPLY: 2060 *virtual_type = ICMP_ECHO; 2061 *virtual_id = pd->hdr.icmp.icmp_id; 2062 break; 2063 2064 case ICMP_TSTAMP: 2065 *icmp_dir = PF_IN; 2066 case ICMP_TSTAMPREPLY: 2067 *virtual_type = ICMP_TSTAMP; 2068 *virtual_id = pd->hdr.icmp.icmp_id; 2069 break; 2070 2071 case ICMP_IREQ: 2072 *icmp_dir = PF_IN; 2073 case ICMP_IREQREPLY: 2074 *virtual_type = ICMP_IREQ; 2075 *virtual_id = pd->hdr.icmp.icmp_id; 2076 break; 2077 2078 case ICMP_MASKREQ: 2079 *icmp_dir = PF_IN; 2080 case ICMP_MASKREPLY: 2081 *virtual_type = ICMP_MASKREQ; 2082 *virtual_id = pd->hdr.icmp.icmp_id; 2083 break; 2084 2085 case ICMP_IPV6_WHEREAREYOU: 2086 *icmp_dir = PF_IN; 2087 case ICMP_IPV6_IAMHERE: 2088 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2089 *virtual_id = 0; /* Nothing sane to match on! */ 2090 break; 2091 2092 case ICMP_MOBILE_REGREQUEST: 2093 *icmp_dir = PF_IN; 2094 case ICMP_MOBILE_REGREPLY: 2095 *virtual_type = ICMP_MOBILE_REGREQUEST; 2096 *virtual_id = 0; /* Nothing sane to match on! */ 2097 break; 2098 2099 case ICMP_ROUTERSOLICIT: 2100 *icmp_dir = PF_IN; 2101 case ICMP_ROUTERADVERT: 2102 *virtual_type = ICMP_ROUTERSOLICIT; 2103 *virtual_id = 0; /* Nothing sane to match on! */ 2104 break; 2105 2106 /* These ICMP types map to other connections */ 2107 case ICMP_UNREACH: 2108 case ICMP_SOURCEQUENCH: 2109 case ICMP_REDIRECT: 2110 case ICMP_TIMXCEED: 2111 case ICMP_PARAMPROB: 2112 /* These will not be used, but set them anyway */ 2113 *icmp_dir = PF_IN; 2114 *virtual_type = type; 2115 *virtual_id = 0; 2116 HTONS(*virtual_type); 2117 return (1); /* These types match to another state */ 2118 2119 /* 2120 * All remaining ICMP types get their own states, 2121 * and will only match in one direction. 2122 */ 2123 default: 2124 *icmp_dir = PF_IN; 2125 *virtual_type = type; 2126 *virtual_id = 0; 2127 break; 2128 } 2129 break; 2130 #endif /* INET */ 2131 #ifdef INET6 2132 case AF_INET6: 2133 switch (type) { 2134 case ICMP6_ECHO_REQUEST: 2135 *icmp_dir = PF_IN; 2136 case ICMP6_ECHO_REPLY: 2137 *virtual_type = ICMP6_ECHO_REQUEST; 2138 *virtual_id = pd->hdr.icmp6.icmp6_id; 2139 break; 2140 2141 case MLD_LISTENER_QUERY: 2142 case MLD_LISTENER_REPORT: { 2143 /* 2144 * Listener Report can be sent by clients 2145 * without an associated Listener Query. 2146 * In addition to that, when Report is sent as a 2147 * reply to a Query its source and destination 2148 * address are different. 2149 */ 2150 *icmp_dir = PF_IN; 2151 *virtual_type = MLD_LISTENER_QUERY; 2152 *virtual_id = 0; 2153 break; 2154 } 2155 case MLD_MTRACE: 2156 *icmp_dir = PF_IN; 2157 case MLD_MTRACE_RESP: 2158 *virtual_type = MLD_MTRACE; 2159 *virtual_id = 0; /* Nothing sane to match on! */ 2160 break; 2161 2162 case ND_NEIGHBOR_SOLICIT: 2163 *icmp_dir = PF_IN; 2164 case ND_NEIGHBOR_ADVERT: { 2165 *virtual_type = ND_NEIGHBOR_SOLICIT; 2166 *virtual_id = 0; 2167 break; 2168 } 2169 2170 /* 2171 * These ICMP types map to other connections. 2172 * ND_REDIRECT can't be in this list because the triggering 2173 * packet header is optional. 2174 */ 2175 case ICMP6_DST_UNREACH: 2176 case ICMP6_PACKET_TOO_BIG: 2177 case ICMP6_TIME_EXCEEDED: 2178 case ICMP6_PARAM_PROB: 2179 /* These will not be used, but set them anyway */ 2180 *icmp_dir = PF_IN; 2181 *virtual_type = type; 2182 *virtual_id = 0; 2183 HTONS(*virtual_type); 2184 return (1); /* These types match to another state */ 2185 /* 2186 * All remaining ICMP6 types get their own states, 2187 * and will only match in one direction. 2188 */ 2189 default: 2190 *icmp_dir = PF_IN; 2191 *virtual_type = type; 2192 *virtual_id = 0; 2193 break; 2194 } 2195 break; 2196 #endif /* INET6 */ 2197 default: 2198 *icmp_dir = PF_IN; 2199 *virtual_type = type; 2200 *virtual_id = 0; 2201 break; 2202 } 2203 HTONS(*virtual_type); 2204 return (0); /* These types match to their own state */ 2205 } 2206 2207 void 2208 pf_intr(void *v) 2209 { 2210 struct epoch_tracker et; 2211 struct pf_send_head queue; 2212 struct pf_send_entry *pfse, *next; 2213 2214 CURVNET_SET((struct vnet *)v); 2215 2216 PF_SENDQ_LOCK(); 2217 queue = V_pf_sendqueue; 2218 STAILQ_INIT(&V_pf_sendqueue); 2219 PF_SENDQ_UNLOCK(); 2220 2221 NET_EPOCH_ENTER(et); 2222 2223 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2224 switch (pfse->pfse_type) { 2225 #ifdef INET 2226 case PFSE_IP: { 2227 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2228 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2229 pfse->pfse_m->m_pkthdr.csum_flags |= 2230 CSUM_IP_VALID | CSUM_IP_CHECKED; 2231 ip_input(pfse->pfse_m); 2232 } else { 2233 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2234 NULL); 2235 } 2236 break; 2237 } 2238 case PFSE_ICMP: 2239 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2240 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2241 break; 2242 #endif /* INET */ 2243 #ifdef INET6 2244 case PFSE_IP6: 2245 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2246 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2247 ip6_input(pfse->pfse_m); 2248 } else { 2249 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2250 NULL, NULL); 2251 } 2252 break; 2253 case PFSE_ICMP6: 2254 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2255 pfse->icmpopts.code, pfse->icmpopts.mtu); 2256 break; 2257 #endif /* INET6 */ 2258 default: 2259 panic("%s: unknown type", __func__); 2260 } 2261 free(pfse, M_PFTEMP); 2262 } 2263 NET_EPOCH_EXIT(et); 2264 CURVNET_RESTORE(); 2265 } 2266 2267 #define pf_purge_thread_period (hz / 10) 2268 2269 #ifdef PF_WANT_32_TO_64_COUNTER 2270 static void 2271 pf_status_counter_u64_periodic(void) 2272 { 2273 2274 PF_RULES_RASSERT(); 2275 2276 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2277 return; 2278 } 2279 2280 for (int i = 0; i < FCNT_MAX; i++) { 2281 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2282 } 2283 } 2284 2285 static void 2286 pf_kif_counter_u64_periodic(void) 2287 { 2288 struct pfi_kkif *kif; 2289 size_t r, run; 2290 2291 PF_RULES_RASSERT(); 2292 2293 if (__predict_false(V_pf_allkifcount == 0)) { 2294 return; 2295 } 2296 2297 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2298 return; 2299 } 2300 2301 run = V_pf_allkifcount / 10; 2302 if (run < 5) 2303 run = 5; 2304 2305 for (r = 0; r < run; r++) { 2306 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2307 if (kif == NULL) { 2308 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2309 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2310 break; 2311 } 2312 2313 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2314 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2315 2316 for (int i = 0; i < 2; i++) { 2317 for (int j = 0; j < 2; j++) { 2318 for (int k = 0; k < 2; k++) { 2319 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2320 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2321 } 2322 } 2323 } 2324 } 2325 } 2326 2327 static void 2328 pf_rule_counter_u64_periodic(void) 2329 { 2330 struct pf_krule *rule; 2331 size_t r, run; 2332 2333 PF_RULES_RASSERT(); 2334 2335 if (__predict_false(V_pf_allrulecount == 0)) { 2336 return; 2337 } 2338 2339 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2340 return; 2341 } 2342 2343 run = V_pf_allrulecount / 10; 2344 if (run < 5) 2345 run = 5; 2346 2347 for (r = 0; r < run; r++) { 2348 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2349 if (rule == NULL) { 2350 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2351 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2352 break; 2353 } 2354 2355 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2356 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2357 2358 pf_counter_u64_periodic(&rule->evaluations); 2359 for (int i = 0; i < 2; i++) { 2360 pf_counter_u64_periodic(&rule->packets[i]); 2361 pf_counter_u64_periodic(&rule->bytes[i]); 2362 } 2363 } 2364 } 2365 2366 static void 2367 pf_counter_u64_periodic_main(void) 2368 { 2369 PF_RULES_RLOCK_TRACKER; 2370 2371 V_pf_counter_periodic_iter++; 2372 2373 PF_RULES_RLOCK(); 2374 pf_counter_u64_critical_enter(); 2375 pf_status_counter_u64_periodic(); 2376 pf_kif_counter_u64_periodic(); 2377 pf_rule_counter_u64_periodic(); 2378 pf_counter_u64_critical_exit(); 2379 PF_RULES_RUNLOCK(); 2380 } 2381 #else 2382 #define pf_counter_u64_periodic_main() do { } while (0) 2383 #endif 2384 2385 void 2386 pf_purge_thread(void *unused __unused) 2387 { 2388 struct epoch_tracker et; 2389 2390 VNET_ITERATOR_DECL(vnet_iter); 2391 2392 sx_xlock(&pf_end_lock); 2393 while (pf_end_threads == 0) { 2394 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2395 2396 VNET_LIST_RLOCK(); 2397 NET_EPOCH_ENTER(et); 2398 VNET_FOREACH(vnet_iter) { 2399 CURVNET_SET(vnet_iter); 2400 2401 /* Wait until V_pf_default_rule is initialized. */ 2402 if (V_pf_vnet_active == 0) { 2403 CURVNET_RESTORE(); 2404 continue; 2405 } 2406 2407 pf_counter_u64_periodic_main(); 2408 2409 /* 2410 * Process 1/interval fraction of the state 2411 * table every run. 2412 */ 2413 V_pf_purge_idx = 2414 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2415 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2416 2417 /* 2418 * Purge other expired types every 2419 * PFTM_INTERVAL seconds. 2420 */ 2421 if (V_pf_purge_idx == 0) { 2422 /* 2423 * Order is important: 2424 * - states and src nodes reference rules 2425 * - states and rules reference kifs 2426 */ 2427 pf_purge_expired_fragments(); 2428 pf_purge_expired_src_nodes(); 2429 pf_purge_unlinked_rules(); 2430 pfi_kkif_purge(); 2431 } 2432 CURVNET_RESTORE(); 2433 } 2434 NET_EPOCH_EXIT(et); 2435 VNET_LIST_RUNLOCK(); 2436 } 2437 2438 pf_end_threads++; 2439 sx_xunlock(&pf_end_lock); 2440 kproc_exit(0); 2441 } 2442 2443 void 2444 pf_unload_vnet_purge(void) 2445 { 2446 2447 /* 2448 * To cleanse up all kifs and rules we need 2449 * two runs: first one clears reference flags, 2450 * then pf_purge_expired_states() doesn't 2451 * raise them, and then second run frees. 2452 */ 2453 pf_purge_unlinked_rules(); 2454 pfi_kkif_purge(); 2455 2456 /* 2457 * Now purge everything. 2458 */ 2459 pf_purge_expired_states(0, V_pf_hashmask); 2460 pf_purge_fragments(UINT_MAX); 2461 pf_purge_expired_src_nodes(); 2462 2463 /* 2464 * Now all kifs & rules should be unreferenced, 2465 * thus should be successfully freed. 2466 */ 2467 pf_purge_unlinked_rules(); 2468 pfi_kkif_purge(); 2469 } 2470 2471 u_int32_t 2472 pf_state_expires(const struct pf_kstate *state) 2473 { 2474 u_int32_t timeout; 2475 u_int32_t start; 2476 u_int32_t end; 2477 u_int32_t states; 2478 2479 /* handle all PFTM_* > PFTM_MAX here */ 2480 if (state->timeout == PFTM_PURGE) 2481 return (time_uptime); 2482 KASSERT(state->timeout != PFTM_UNLINKED, 2483 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2484 KASSERT((state->timeout < PFTM_MAX), 2485 ("pf_state_expires: timeout > PFTM_MAX")); 2486 timeout = state->rule->timeout[state->timeout]; 2487 if (!timeout) 2488 timeout = V_pf_default_rule.timeout[state->timeout]; 2489 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2490 if (start && state->rule != &V_pf_default_rule) { 2491 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2492 states = counter_u64_fetch(state->rule->states_cur); 2493 } else { 2494 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2495 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2496 states = V_pf_status.states; 2497 } 2498 if (end && states > start && start < end) { 2499 if (states < end) { 2500 timeout = (u_int64_t)timeout * (end - states) / 2501 (end - start); 2502 return ((state->expire / 1000) + timeout); 2503 } 2504 else 2505 return (time_uptime); 2506 } 2507 return ((state->expire / 1000) + timeout); 2508 } 2509 2510 void 2511 pf_purge_expired_src_nodes(void) 2512 { 2513 struct pf_ksrc_node_list freelist; 2514 struct pf_srchash *sh; 2515 struct pf_ksrc_node *cur, *next; 2516 int i; 2517 2518 LIST_INIT(&freelist); 2519 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2520 PF_HASHROW_LOCK(sh); 2521 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2522 if (cur->states == 0 && cur->expire <= time_uptime) { 2523 pf_unlink_src_node(cur); 2524 LIST_INSERT_HEAD(&freelist, cur, entry); 2525 } else if (cur->rule != NULL) 2526 cur->rule->rule_ref |= PFRULE_REFS; 2527 PF_HASHROW_UNLOCK(sh); 2528 } 2529 2530 pf_free_src_nodes(&freelist); 2531 2532 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2533 } 2534 2535 static void 2536 pf_src_tree_remove_state(struct pf_kstate *s) 2537 { 2538 struct pf_ksrc_node *sn; 2539 uint32_t timeout; 2540 2541 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2542 s->rule->timeout[PFTM_SRC_NODE] : 2543 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2544 2545 if (s->src_node != NULL) { 2546 sn = s->src_node; 2547 PF_SRC_NODE_LOCK(sn); 2548 if (s->src.tcp_est) 2549 --sn->conn; 2550 if (--sn->states == 0) 2551 sn->expire = time_uptime + timeout; 2552 PF_SRC_NODE_UNLOCK(sn); 2553 } 2554 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2555 sn = s->nat_src_node; 2556 PF_SRC_NODE_LOCK(sn); 2557 if (--sn->states == 0) 2558 sn->expire = time_uptime + timeout; 2559 PF_SRC_NODE_UNLOCK(sn); 2560 } 2561 s->src_node = s->nat_src_node = NULL; 2562 } 2563 2564 /* 2565 * Unlink and potentilly free a state. Function may be 2566 * called with ID hash row locked, but always returns 2567 * unlocked, since it needs to go through key hash locking. 2568 */ 2569 int 2570 pf_unlink_state(struct pf_kstate *s) 2571 { 2572 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2573 2574 NET_EPOCH_ASSERT(); 2575 PF_HASHROW_ASSERT(ih); 2576 2577 if (s->timeout == PFTM_UNLINKED) { 2578 /* 2579 * State is being processed 2580 * by pf_unlink_state() in 2581 * an other thread. 2582 */ 2583 PF_HASHROW_UNLOCK(ih); 2584 return (0); /* XXXGL: undefined actually */ 2585 } 2586 2587 if (s->src.state == PF_TCPS_PROXY_DST) { 2588 /* XXX wire key the right one? */ 2589 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2590 &s->key[PF_SK_WIRE]->addr[1], 2591 &s->key[PF_SK_WIRE]->addr[0], 2592 s->key[PF_SK_WIRE]->port[1], 2593 s->key[PF_SK_WIRE]->port[0], 2594 s->src.seqhi, s->src.seqlo + 1, 2595 TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid); 2596 } 2597 2598 LIST_REMOVE(s, entry); 2599 pf_src_tree_remove_state(s); 2600 2601 if (V_pfsync_delete_state_ptr != NULL) 2602 V_pfsync_delete_state_ptr(s); 2603 2604 STATE_DEC_COUNTERS(s); 2605 2606 s->timeout = PFTM_UNLINKED; 2607 2608 /* Ensure we remove it from the list of halfopen states, if needed. */ 2609 if (s->key[PF_SK_STACK] != NULL && 2610 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2611 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2612 2613 PF_HASHROW_UNLOCK(ih); 2614 2615 pf_detach_state(s); 2616 2617 pf_udp_mapping_release(s->udp_mapping); 2618 2619 /* pf_state_insert() initialises refs to 2 */ 2620 return (pf_release_staten(s, 2)); 2621 } 2622 2623 struct pf_kstate * 2624 pf_alloc_state(int flags) 2625 { 2626 2627 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2628 } 2629 2630 void 2631 pf_free_state(struct pf_kstate *cur) 2632 { 2633 struct pf_krule_item *ri; 2634 2635 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2636 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2637 cur->timeout)); 2638 2639 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2640 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2641 free(ri, M_PF_RULE_ITEM); 2642 } 2643 2644 pf_normalize_tcp_cleanup(cur); 2645 uma_zfree(V_pf_state_z, cur); 2646 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2647 } 2648 2649 /* 2650 * Called only from pf_purge_thread(), thus serialized. 2651 */ 2652 static u_int 2653 pf_purge_expired_states(u_int i, int maxcheck) 2654 { 2655 struct pf_idhash *ih; 2656 struct pf_kstate *s; 2657 struct pf_krule_item *mrm; 2658 size_t count __unused; 2659 2660 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2661 2662 /* 2663 * Go through hash and unlink states that expire now. 2664 */ 2665 while (maxcheck > 0) { 2666 count = 0; 2667 ih = &V_pf_idhash[i]; 2668 2669 /* only take the lock if we expect to do work */ 2670 if (!LIST_EMPTY(&ih->states)) { 2671 relock: 2672 PF_HASHROW_LOCK(ih); 2673 LIST_FOREACH(s, &ih->states, entry) { 2674 if (pf_state_expires(s) <= time_uptime) { 2675 V_pf_status.states -= 2676 pf_unlink_state(s); 2677 goto relock; 2678 } 2679 s->rule->rule_ref |= PFRULE_REFS; 2680 if (s->nat_rule != NULL) 2681 s->nat_rule->rule_ref |= PFRULE_REFS; 2682 if (s->anchor != NULL) 2683 s->anchor->rule_ref |= PFRULE_REFS; 2684 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2685 SLIST_FOREACH(mrm, &s->match_rules, entry) 2686 mrm->r->rule_ref |= PFRULE_REFS; 2687 if (s->rt_kif) 2688 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2689 count++; 2690 } 2691 PF_HASHROW_UNLOCK(ih); 2692 } 2693 2694 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2695 2696 /* Return when we hit end of hash. */ 2697 if (++i > V_pf_hashmask) { 2698 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2699 return (0); 2700 } 2701 2702 maxcheck--; 2703 } 2704 2705 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2706 2707 return (i); 2708 } 2709 2710 static void 2711 pf_purge_unlinked_rules(void) 2712 { 2713 struct pf_krulequeue tmpq; 2714 struct pf_krule *r, *r1; 2715 2716 /* 2717 * If we have overloading task pending, then we'd 2718 * better skip purging this time. There is a tiny 2719 * probability that overloading task references 2720 * an already unlinked rule. 2721 */ 2722 PF_OVERLOADQ_LOCK(); 2723 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2724 PF_OVERLOADQ_UNLOCK(); 2725 return; 2726 } 2727 PF_OVERLOADQ_UNLOCK(); 2728 2729 /* 2730 * Do naive mark-and-sweep garbage collecting of old rules. 2731 * Reference flag is raised by pf_purge_expired_states() 2732 * and pf_purge_expired_src_nodes(). 2733 * 2734 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2735 * use a temporary queue. 2736 */ 2737 TAILQ_INIT(&tmpq); 2738 PF_UNLNKDRULES_LOCK(); 2739 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2740 if (!(r->rule_ref & PFRULE_REFS)) { 2741 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2742 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2743 } else 2744 r->rule_ref &= ~PFRULE_REFS; 2745 } 2746 PF_UNLNKDRULES_UNLOCK(); 2747 2748 if (!TAILQ_EMPTY(&tmpq)) { 2749 PF_CONFIG_LOCK(); 2750 PF_RULES_WLOCK(); 2751 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2752 TAILQ_REMOVE(&tmpq, r, entries); 2753 pf_free_rule(r); 2754 } 2755 PF_RULES_WUNLOCK(); 2756 PF_CONFIG_UNLOCK(); 2757 } 2758 } 2759 2760 void 2761 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2762 { 2763 switch (af) { 2764 #ifdef INET 2765 case AF_INET: { 2766 u_int32_t a = ntohl(addr->addr32[0]); 2767 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2768 (a>>8)&255, a&255); 2769 if (p) { 2770 p = ntohs(p); 2771 printf(":%u", p); 2772 } 2773 break; 2774 } 2775 #endif /* INET */ 2776 #ifdef INET6 2777 case AF_INET6: { 2778 u_int16_t b; 2779 u_int8_t i, curstart, curend, maxstart, maxend; 2780 curstart = curend = maxstart = maxend = 255; 2781 for (i = 0; i < 8; i++) { 2782 if (!addr->addr16[i]) { 2783 if (curstart == 255) 2784 curstart = i; 2785 curend = i; 2786 } else { 2787 if ((curend - curstart) > 2788 (maxend - maxstart)) { 2789 maxstart = curstart; 2790 maxend = curend; 2791 } 2792 curstart = curend = 255; 2793 } 2794 } 2795 if ((curend - curstart) > 2796 (maxend - maxstart)) { 2797 maxstart = curstart; 2798 maxend = curend; 2799 } 2800 for (i = 0; i < 8; i++) { 2801 if (i >= maxstart && i <= maxend) { 2802 if (i == 0) 2803 printf(":"); 2804 if (i == maxend) 2805 printf(":"); 2806 } else { 2807 b = ntohs(addr->addr16[i]); 2808 printf("%x", b); 2809 if (i < 7) 2810 printf(":"); 2811 } 2812 } 2813 if (p) { 2814 p = ntohs(p); 2815 printf("[%u]", p); 2816 } 2817 break; 2818 } 2819 #endif /* INET6 */ 2820 } 2821 } 2822 2823 void 2824 pf_print_state(struct pf_kstate *s) 2825 { 2826 pf_print_state_parts(s, NULL, NULL); 2827 } 2828 2829 static void 2830 pf_print_state_parts(struct pf_kstate *s, 2831 struct pf_state_key *skwp, struct pf_state_key *sksp) 2832 { 2833 struct pf_state_key *skw, *sks; 2834 u_int8_t proto, dir; 2835 2836 /* Do our best to fill these, but they're skipped if NULL */ 2837 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2838 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2839 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2840 dir = s ? s->direction : 0; 2841 2842 switch (proto) { 2843 case IPPROTO_IPV4: 2844 printf("IPv4"); 2845 break; 2846 case IPPROTO_IPV6: 2847 printf("IPv6"); 2848 break; 2849 case IPPROTO_TCP: 2850 printf("TCP"); 2851 break; 2852 case IPPROTO_UDP: 2853 printf("UDP"); 2854 break; 2855 case IPPROTO_ICMP: 2856 printf("ICMP"); 2857 break; 2858 case IPPROTO_ICMPV6: 2859 printf("ICMPv6"); 2860 break; 2861 default: 2862 printf("%u", proto); 2863 break; 2864 } 2865 switch (dir) { 2866 case PF_IN: 2867 printf(" in"); 2868 break; 2869 case PF_OUT: 2870 printf(" out"); 2871 break; 2872 } 2873 if (skw) { 2874 printf(" wire: "); 2875 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2876 printf(" "); 2877 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2878 } 2879 if (sks) { 2880 printf(" stack: "); 2881 if (sks != skw) { 2882 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2883 printf(" "); 2884 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2885 } else 2886 printf("-"); 2887 } 2888 if (s) { 2889 if (proto == IPPROTO_TCP) { 2890 printf(" [lo=%u high=%u win=%u modulator=%u", 2891 s->src.seqlo, s->src.seqhi, 2892 s->src.max_win, s->src.seqdiff); 2893 if (s->src.wscale && s->dst.wscale) 2894 printf(" wscale=%u", 2895 s->src.wscale & PF_WSCALE_MASK); 2896 printf("]"); 2897 printf(" [lo=%u high=%u win=%u modulator=%u", 2898 s->dst.seqlo, s->dst.seqhi, 2899 s->dst.max_win, s->dst.seqdiff); 2900 if (s->src.wscale && s->dst.wscale) 2901 printf(" wscale=%u", 2902 s->dst.wscale & PF_WSCALE_MASK); 2903 printf("]"); 2904 } 2905 printf(" %u:%u", s->src.state, s->dst.state); 2906 if (s->rule) 2907 printf(" @%d", s->rule->nr); 2908 } 2909 } 2910 2911 void 2912 pf_print_flags(u_int8_t f) 2913 { 2914 if (f) 2915 printf(" "); 2916 if (f & TH_FIN) 2917 printf("F"); 2918 if (f & TH_SYN) 2919 printf("S"); 2920 if (f & TH_RST) 2921 printf("R"); 2922 if (f & TH_PUSH) 2923 printf("P"); 2924 if (f & TH_ACK) 2925 printf("A"); 2926 if (f & TH_URG) 2927 printf("U"); 2928 if (f & TH_ECE) 2929 printf("E"); 2930 if (f & TH_CWR) 2931 printf("W"); 2932 } 2933 2934 #define PF_SET_SKIP_STEPS(i) \ 2935 do { \ 2936 while (head[i] != cur) { \ 2937 head[i]->skip[i] = cur; \ 2938 head[i] = TAILQ_NEXT(head[i], entries); \ 2939 } \ 2940 } while (0) 2941 2942 void 2943 pf_calc_skip_steps(struct pf_krulequeue *rules) 2944 { 2945 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2946 int i; 2947 2948 cur = TAILQ_FIRST(rules); 2949 prev = cur; 2950 for (i = 0; i < PF_SKIP_COUNT; ++i) 2951 head[i] = cur; 2952 while (cur != NULL) { 2953 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2954 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2955 if (cur->direction != prev->direction) 2956 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2957 if (cur->af != prev->af) 2958 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2959 if (cur->proto != prev->proto) 2960 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2961 if (cur->src.neg != prev->src.neg || 2962 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2963 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2964 if (cur->dst.neg != prev->dst.neg || 2965 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2966 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2967 if (cur->src.port[0] != prev->src.port[0] || 2968 cur->src.port[1] != prev->src.port[1] || 2969 cur->src.port_op != prev->src.port_op) 2970 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2971 if (cur->dst.port[0] != prev->dst.port[0] || 2972 cur->dst.port[1] != prev->dst.port[1] || 2973 cur->dst.port_op != prev->dst.port_op) 2974 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2975 2976 prev = cur; 2977 cur = TAILQ_NEXT(cur, entries); 2978 } 2979 for (i = 0; i < PF_SKIP_COUNT; ++i) 2980 PF_SET_SKIP_STEPS(i); 2981 } 2982 2983 int 2984 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2985 { 2986 if (aw1->type != aw2->type) 2987 return (1); 2988 switch (aw1->type) { 2989 case PF_ADDR_ADDRMASK: 2990 case PF_ADDR_RANGE: 2991 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2992 return (1); 2993 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2994 return (1); 2995 return (0); 2996 case PF_ADDR_DYNIFTL: 2997 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2998 case PF_ADDR_NOROUTE: 2999 case PF_ADDR_URPFFAILED: 3000 return (0); 3001 case PF_ADDR_TABLE: 3002 return (aw1->p.tbl != aw2->p.tbl); 3003 default: 3004 printf("invalid address type: %d\n", aw1->type); 3005 return (1); 3006 } 3007 } 3008 3009 /** 3010 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3011 * header isn't always a full checksum. In some cases (i.e. output) it's a 3012 * pseudo-header checksum, which is a partial checksum over src/dst IP 3013 * addresses, protocol number and length. 3014 * 3015 * That means we have the following cases: 3016 * * Input or forwarding: we don't have TSO, the checksum fields are full 3017 * checksums, we need to update the checksum whenever we change anything. 3018 * * Output (i.e. the checksum is a pseudo-header checksum): 3019 * x The field being updated is src/dst address or affects the length of 3020 * the packet. We need to update the pseudo-header checksum (note that this 3021 * checksum is not ones' complement). 3022 * x Some other field is being modified (e.g. src/dst port numbers): We 3023 * don't have to update anything. 3024 **/ 3025 u_int16_t 3026 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3027 { 3028 u_int32_t x; 3029 3030 x = cksum + old - new; 3031 x = (x + (x >> 16)) & 0xffff; 3032 3033 /* optimise: eliminate a branch when not udp */ 3034 if (udp && cksum == 0x0000) 3035 return cksum; 3036 if (udp && x == 0x0000) 3037 x = 0xffff; 3038 3039 return (u_int16_t)(x); 3040 } 3041 3042 static void 3043 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 3044 u_int8_t udp) 3045 { 3046 u_int16_t old = htons(hi ? (*f << 8) : *f); 3047 u_int16_t new = htons(hi ? ( v << 8) : v); 3048 3049 if (*f == v) 3050 return; 3051 3052 *f = v; 3053 3054 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3055 return; 3056 3057 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 3058 } 3059 3060 void 3061 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 3062 bool hi, u_int8_t udp) 3063 { 3064 u_int8_t *fb = (u_int8_t *)f; 3065 u_int8_t *vb = (u_int8_t *)&v; 3066 3067 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3068 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3069 } 3070 3071 void 3072 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 3073 bool hi, u_int8_t udp) 3074 { 3075 u_int8_t *fb = (u_int8_t *)f; 3076 u_int8_t *vb = (u_int8_t *)&v; 3077 3078 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3079 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3080 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3081 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3082 } 3083 3084 u_int16_t 3085 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3086 u_int16_t new, u_int8_t udp) 3087 { 3088 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3089 return (cksum); 3090 3091 return (pf_cksum_fixup(cksum, old, new, udp)); 3092 } 3093 3094 static void 3095 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 3096 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 3097 sa_family_t af) 3098 { 3099 struct pf_addr ao; 3100 u_int16_t po = *p; 3101 3102 PF_ACPY(&ao, a, af); 3103 PF_ACPY(a, an, af); 3104 3105 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3106 *pc = ~*pc; 3107 3108 *p = pn; 3109 3110 switch (af) { 3111 #ifdef INET 3112 case AF_INET: 3113 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3114 ao.addr16[0], an->addr16[0], 0), 3115 ao.addr16[1], an->addr16[1], 0); 3116 *p = pn; 3117 3118 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 3119 ao.addr16[0], an->addr16[0], u), 3120 ao.addr16[1], an->addr16[1], u); 3121 3122 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3123 break; 3124 #endif /* INET */ 3125 #ifdef INET6 3126 case AF_INET6: 3127 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3128 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3129 pf_cksum_fixup(pf_cksum_fixup(*pc, 3130 ao.addr16[0], an->addr16[0], u), 3131 ao.addr16[1], an->addr16[1], u), 3132 ao.addr16[2], an->addr16[2], u), 3133 ao.addr16[3], an->addr16[3], u), 3134 ao.addr16[4], an->addr16[4], u), 3135 ao.addr16[5], an->addr16[5], u), 3136 ao.addr16[6], an->addr16[6], u), 3137 ao.addr16[7], an->addr16[7], u); 3138 3139 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3140 break; 3141 #endif /* INET6 */ 3142 } 3143 3144 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3145 CSUM_DELAY_DATA_IPV6)) { 3146 *pc = ~*pc; 3147 if (! *pc) 3148 *pc = 0xffff; 3149 } 3150 } 3151 3152 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3153 void 3154 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3155 { 3156 u_int32_t ao; 3157 3158 memcpy(&ao, a, sizeof(ao)); 3159 memcpy(a, &an, sizeof(u_int32_t)); 3160 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3161 ao % 65536, an % 65536, u); 3162 } 3163 3164 void 3165 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3166 { 3167 u_int32_t ao; 3168 3169 memcpy(&ao, a, sizeof(ao)); 3170 memcpy(a, &an, sizeof(u_int32_t)); 3171 3172 *c = pf_proto_cksum_fixup(m, 3173 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3174 ao % 65536, an % 65536, udp); 3175 } 3176 3177 #ifdef INET6 3178 static void 3179 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3180 { 3181 struct pf_addr ao; 3182 3183 PF_ACPY(&ao, a, AF_INET6); 3184 PF_ACPY(a, an, AF_INET6); 3185 3186 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3187 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3188 pf_cksum_fixup(pf_cksum_fixup(*c, 3189 ao.addr16[0], an->addr16[0], u), 3190 ao.addr16[1], an->addr16[1], u), 3191 ao.addr16[2], an->addr16[2], u), 3192 ao.addr16[3], an->addr16[3], u), 3193 ao.addr16[4], an->addr16[4], u), 3194 ao.addr16[5], an->addr16[5], u), 3195 ao.addr16[6], an->addr16[6], u), 3196 ao.addr16[7], an->addr16[7], u); 3197 } 3198 #endif /* INET6 */ 3199 3200 static void 3201 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3202 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3203 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3204 { 3205 struct pf_addr oia, ooa; 3206 3207 PF_ACPY(&oia, ia, af); 3208 if (oa) 3209 PF_ACPY(&ooa, oa, af); 3210 3211 /* Change inner protocol port, fix inner protocol checksum. */ 3212 if (ip != NULL) { 3213 u_int16_t oip = *ip; 3214 u_int32_t opc; 3215 3216 if (pc != NULL) 3217 opc = *pc; 3218 *ip = np; 3219 if (pc != NULL) 3220 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3221 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3222 if (pc != NULL) 3223 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3224 } 3225 /* Change inner ip address, fix inner ip and icmp checksums. */ 3226 PF_ACPY(ia, na, af); 3227 switch (af) { 3228 #ifdef INET 3229 case AF_INET: { 3230 u_int32_t oh2c = *h2c; 3231 3232 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3233 oia.addr16[0], ia->addr16[0], 0), 3234 oia.addr16[1], ia->addr16[1], 0); 3235 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3236 oia.addr16[0], ia->addr16[0], 0), 3237 oia.addr16[1], ia->addr16[1], 0); 3238 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3239 break; 3240 } 3241 #endif /* INET */ 3242 #ifdef INET6 3243 case AF_INET6: 3244 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3245 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3246 pf_cksum_fixup(pf_cksum_fixup(*ic, 3247 oia.addr16[0], ia->addr16[0], u), 3248 oia.addr16[1], ia->addr16[1], u), 3249 oia.addr16[2], ia->addr16[2], u), 3250 oia.addr16[3], ia->addr16[3], u), 3251 oia.addr16[4], ia->addr16[4], u), 3252 oia.addr16[5], ia->addr16[5], u), 3253 oia.addr16[6], ia->addr16[6], u), 3254 oia.addr16[7], ia->addr16[7], u); 3255 break; 3256 #endif /* INET6 */ 3257 } 3258 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3259 if (oa) { 3260 PF_ACPY(oa, na, af); 3261 switch (af) { 3262 #ifdef INET 3263 case AF_INET: 3264 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3265 ooa.addr16[0], oa->addr16[0], 0), 3266 ooa.addr16[1], oa->addr16[1], 0); 3267 break; 3268 #endif /* INET */ 3269 #ifdef INET6 3270 case AF_INET6: 3271 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3272 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3273 pf_cksum_fixup(pf_cksum_fixup(*ic, 3274 ooa.addr16[0], oa->addr16[0], u), 3275 ooa.addr16[1], oa->addr16[1], u), 3276 ooa.addr16[2], oa->addr16[2], u), 3277 ooa.addr16[3], oa->addr16[3], u), 3278 ooa.addr16[4], oa->addr16[4], u), 3279 ooa.addr16[5], oa->addr16[5], u), 3280 ooa.addr16[6], oa->addr16[6], u), 3281 ooa.addr16[7], oa->addr16[7], u); 3282 break; 3283 #endif /* INET6 */ 3284 } 3285 } 3286 } 3287 3288 /* 3289 * Need to modulate the sequence numbers in the TCP SACK option 3290 * (credits to Krzysztof Pfaff for report and patch) 3291 */ 3292 static int 3293 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 3294 struct tcphdr *th, struct pf_state_peer *dst) 3295 { 3296 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3297 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3298 int copyback = 0, i, olen; 3299 struct sackblk sack; 3300 3301 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3302 if (hlen < TCPOLEN_SACKLEN || 3303 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3304 return 0; 3305 3306 while (hlen >= TCPOLEN_SACKLEN) { 3307 size_t startoff = opt - opts; 3308 olen = opt[1]; 3309 switch (*opt) { 3310 case TCPOPT_EOL: /* FALLTHROUGH */ 3311 case TCPOPT_NOP: 3312 opt++; 3313 hlen--; 3314 break; 3315 case TCPOPT_SACK: 3316 if (olen > hlen) 3317 olen = hlen; 3318 if (olen >= TCPOLEN_SACKLEN) { 3319 for (i = 2; i + TCPOLEN_SACK <= olen; 3320 i += TCPOLEN_SACK) { 3321 memcpy(&sack, &opt[i], sizeof(sack)); 3322 pf_patch_32_unaligned(m, 3323 &th->th_sum, &sack.start, 3324 htonl(ntohl(sack.start) - dst->seqdiff), 3325 PF_ALGNMNT(startoff), 3326 0); 3327 pf_patch_32_unaligned(m, &th->th_sum, 3328 &sack.end, 3329 htonl(ntohl(sack.end) - dst->seqdiff), 3330 PF_ALGNMNT(startoff), 3331 0); 3332 memcpy(&opt[i], &sack, sizeof(sack)); 3333 } 3334 copyback = 1; 3335 } 3336 /* FALLTHROUGH */ 3337 default: 3338 if (olen < 2) 3339 olen = 2; 3340 hlen -= olen; 3341 opt += olen; 3342 } 3343 } 3344 3345 if (copyback) 3346 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 3347 return (copyback); 3348 } 3349 3350 struct mbuf * 3351 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3352 const struct pf_addr *saddr, const struct pf_addr *daddr, 3353 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3354 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3355 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3356 { 3357 struct mbuf *m; 3358 int len, tlen; 3359 #ifdef INET 3360 struct ip *h = NULL; 3361 #endif /* INET */ 3362 #ifdef INET6 3363 struct ip6_hdr *h6 = NULL; 3364 #endif /* INET6 */ 3365 struct tcphdr *th; 3366 char *opt; 3367 struct pf_mtag *pf_mtag; 3368 3369 len = 0; 3370 th = NULL; 3371 3372 /* maximum segment size tcp option */ 3373 tlen = sizeof(struct tcphdr); 3374 if (mss) 3375 tlen += 4; 3376 3377 switch (af) { 3378 #ifdef INET 3379 case AF_INET: 3380 len = sizeof(struct ip) + tlen; 3381 break; 3382 #endif /* INET */ 3383 #ifdef INET6 3384 case AF_INET6: 3385 len = sizeof(struct ip6_hdr) + tlen; 3386 break; 3387 #endif /* INET6 */ 3388 default: 3389 panic("%s: unsupported af %d", __func__, af); 3390 } 3391 3392 m = m_gethdr(M_NOWAIT, MT_DATA); 3393 if (m == NULL) 3394 return (NULL); 3395 3396 #ifdef MAC 3397 mac_netinet_firewall_send(m); 3398 #endif 3399 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 3400 m_freem(m); 3401 return (NULL); 3402 } 3403 if (skip_firewall) 3404 m->m_flags |= M_SKIP_FIREWALL; 3405 pf_mtag->tag = mtag_tag; 3406 pf_mtag->flags = mtag_flags; 3407 3408 if (rtableid >= 0) 3409 M_SETFIB(m, rtableid); 3410 3411 #ifdef ALTQ 3412 if (r != NULL && r->qid) { 3413 pf_mtag->qid = r->qid; 3414 3415 /* add hints for ecn */ 3416 pf_mtag->hdr = mtod(m, struct ip *); 3417 } 3418 #endif /* ALTQ */ 3419 m->m_data += max_linkhdr; 3420 m->m_pkthdr.len = m->m_len = len; 3421 /* The rest of the stack assumes a rcvif, so provide one. 3422 * This is a locally generated packet, so .. close enough. */ 3423 m->m_pkthdr.rcvif = V_loif; 3424 bzero(m->m_data, len); 3425 switch (af) { 3426 #ifdef INET 3427 case AF_INET: 3428 h = mtod(m, struct ip *); 3429 3430 /* IP header fields included in the TCP checksum */ 3431 h->ip_p = IPPROTO_TCP; 3432 h->ip_len = htons(tlen); 3433 h->ip_src.s_addr = saddr->v4.s_addr; 3434 h->ip_dst.s_addr = daddr->v4.s_addr; 3435 3436 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3437 break; 3438 #endif /* INET */ 3439 #ifdef INET6 3440 case AF_INET6: 3441 h6 = mtod(m, struct ip6_hdr *); 3442 3443 /* IP header fields included in the TCP checksum */ 3444 h6->ip6_nxt = IPPROTO_TCP; 3445 h6->ip6_plen = htons(tlen); 3446 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3447 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3448 3449 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3450 break; 3451 #endif /* INET6 */ 3452 } 3453 3454 /* TCP header */ 3455 th->th_sport = sport; 3456 th->th_dport = dport; 3457 th->th_seq = htonl(seq); 3458 th->th_ack = htonl(ack); 3459 th->th_off = tlen >> 2; 3460 th->th_flags = tcp_flags; 3461 th->th_win = htons(win); 3462 3463 if (mss) { 3464 opt = (char *)(th + 1); 3465 opt[0] = TCPOPT_MAXSEG; 3466 opt[1] = 4; 3467 HTONS(mss); 3468 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 3469 } 3470 3471 switch (af) { 3472 #ifdef INET 3473 case AF_INET: 3474 /* TCP checksum */ 3475 th->th_sum = in_cksum(m, len); 3476 3477 /* Finish the IP header */ 3478 h->ip_v = 4; 3479 h->ip_hl = sizeof(*h) >> 2; 3480 h->ip_tos = IPTOS_LOWDELAY; 3481 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 3482 h->ip_len = htons(len); 3483 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3484 h->ip_sum = 0; 3485 break; 3486 #endif /* INET */ 3487 #ifdef INET6 3488 case AF_INET6: 3489 /* TCP checksum */ 3490 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3491 sizeof(struct ip6_hdr), tlen); 3492 3493 h6->ip6_vfc |= IPV6_VERSION; 3494 h6->ip6_hlim = IPV6_DEFHLIM; 3495 break; 3496 #endif /* INET6 */ 3497 } 3498 3499 return (m); 3500 } 3501 3502 static void 3503 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3504 uint8_t ttl, int rtableid) 3505 { 3506 struct mbuf *m; 3507 #ifdef INET 3508 struct ip *h = NULL; 3509 #endif /* INET */ 3510 #ifdef INET6 3511 struct ip6_hdr *h6 = NULL; 3512 #endif /* INET6 */ 3513 struct sctphdr *hdr; 3514 struct sctp_chunkhdr *chunk; 3515 struct pf_send_entry *pfse; 3516 int off = 0; 3517 3518 MPASS(af == pd->af); 3519 3520 m = m_gethdr(M_NOWAIT, MT_DATA); 3521 if (m == NULL) 3522 return; 3523 3524 m->m_data += max_linkhdr; 3525 m->m_flags |= M_SKIP_FIREWALL; 3526 /* The rest of the stack assumes a rcvif, so provide one. 3527 * This is a locally generated packet, so .. close enough. */ 3528 m->m_pkthdr.rcvif = V_loif; 3529 3530 /* IPv4|6 header */ 3531 switch (af) { 3532 #ifdef INET 3533 case AF_INET: 3534 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3535 3536 h = mtod(m, struct ip *); 3537 3538 /* IP header fields included in the TCP checksum */ 3539 3540 h->ip_p = IPPROTO_SCTP; 3541 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3542 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3543 h->ip_src = pd->dst->v4; 3544 h->ip_dst = pd->src->v4; 3545 3546 off += sizeof(struct ip); 3547 break; 3548 #endif /* INET */ 3549 #ifdef INET6 3550 case AF_INET6: 3551 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3552 3553 h6 = mtod(m, struct ip6_hdr *); 3554 3555 /* IP header fields included in the TCP checksum */ 3556 h6->ip6_vfc |= IPV6_VERSION; 3557 h6->ip6_nxt = IPPROTO_SCTP; 3558 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3559 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3560 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3561 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3562 3563 off += sizeof(struct ip6_hdr); 3564 break; 3565 #endif /* INET6 */ 3566 } 3567 3568 /* SCTP header */ 3569 hdr = mtodo(m, off); 3570 3571 hdr->src_port = pd->hdr.sctp.dest_port; 3572 hdr->dest_port = pd->hdr.sctp.src_port; 3573 hdr->v_tag = pd->sctp_initiate_tag; 3574 hdr->checksum = 0; 3575 3576 /* Abort chunk. */ 3577 off += sizeof(struct sctphdr); 3578 chunk = mtodo(m, off); 3579 3580 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3581 chunk->chunk_length = htons(sizeof(*chunk)); 3582 3583 /* SCTP checksum */ 3584 off += sizeof(*chunk); 3585 m->m_pkthdr.len = m->m_len = off; 3586 3587 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 3588 3589 if (rtableid >= 0) 3590 M_SETFIB(m, rtableid); 3591 3592 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3593 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3594 if (pfse == NULL) { 3595 m_freem(m); 3596 return; 3597 } 3598 3599 switch (af) { 3600 #ifdef INET 3601 case AF_INET: 3602 pfse->pfse_type = PFSE_IP; 3603 break; 3604 #endif /* INET */ 3605 #ifdef INET6 3606 case AF_INET6: 3607 pfse->pfse_type = PFSE_IP6; 3608 break; 3609 #endif /* INET6 */ 3610 } 3611 3612 pfse->pfse_m = m; 3613 pf_send(pfse); 3614 } 3615 3616 void 3617 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3618 const struct pf_addr *saddr, const struct pf_addr *daddr, 3619 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3620 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3621 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3622 { 3623 struct pf_send_entry *pfse; 3624 struct mbuf *m; 3625 3626 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3627 win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid); 3628 if (m == NULL) 3629 return; 3630 3631 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3632 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3633 if (pfse == NULL) { 3634 m_freem(m); 3635 return; 3636 } 3637 3638 switch (af) { 3639 #ifdef INET 3640 case AF_INET: 3641 pfse->pfse_type = PFSE_IP; 3642 break; 3643 #endif /* INET */ 3644 #ifdef INET6 3645 case AF_INET6: 3646 pfse->pfse_type = PFSE_IP6; 3647 break; 3648 #endif /* INET6 */ 3649 } 3650 3651 pfse->pfse_m = m; 3652 pf_send(pfse); 3653 } 3654 3655 static void 3656 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3657 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 3658 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 3659 u_short *reason, int rtableid) 3660 { 3661 struct pf_addr * const saddr = pd->src; 3662 struct pf_addr * const daddr = pd->dst; 3663 sa_family_t af = pd->af; 3664 3665 /* undo NAT changes, if they have taken place */ 3666 if (nr != NULL) { 3667 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3668 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3669 if (pd->sport) 3670 *pd->sport = sk->port[pd->sidx]; 3671 if (pd->dport) 3672 *pd->dport = sk->port[pd->didx]; 3673 if (pd->proto_sum) 3674 *pd->proto_sum = bproto_sum; 3675 if (pd->ip_sum) 3676 *pd->ip_sum = bip_sum; 3677 m_copyback(m, off, hdrlen, pd->hdr.any); 3678 } 3679 if (pd->proto == IPPROTO_TCP && 3680 ((r->rule_flag & PFRULE_RETURNRST) || 3681 (r->rule_flag & PFRULE_RETURN)) && 3682 !(th->th_flags & TH_RST)) { 3683 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3684 int len = 0; 3685 #ifdef INET 3686 struct ip *h4; 3687 #endif 3688 #ifdef INET6 3689 struct ip6_hdr *h6; 3690 #endif 3691 3692 switch (af) { 3693 #ifdef INET 3694 case AF_INET: 3695 h4 = mtod(m, struct ip *); 3696 len = ntohs(h4->ip_len) - off; 3697 break; 3698 #endif 3699 #ifdef INET6 3700 case AF_INET6: 3701 h6 = mtod(m, struct ip6_hdr *); 3702 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3703 break; 3704 #endif 3705 } 3706 3707 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3708 REASON_SET(reason, PFRES_PROTCKSUM); 3709 else { 3710 if (th->th_flags & TH_SYN) 3711 ack++; 3712 if (th->th_flags & TH_FIN) 3713 ack++; 3714 pf_send_tcp(r, af, pd->dst, 3715 pd->src, th->th_dport, th->th_sport, 3716 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3717 r->return_ttl, true, 0, 0, rtableid); 3718 } 3719 } else if (pd->proto == IPPROTO_SCTP && 3720 (r->rule_flag & PFRULE_RETURN)) { 3721 pf_send_sctp_abort(af, pd, r->return_ttl, rtableid); 3722 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3723 r->return_icmp) 3724 pf_send_icmp(m, r->return_icmp >> 8, 3725 r->return_icmp & 255, af, r, rtableid); 3726 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3727 r->return_icmp6) 3728 pf_send_icmp(m, r->return_icmp6 >> 8, 3729 r->return_icmp6 & 255, af, r, rtableid); 3730 } 3731 3732 static int 3733 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3734 { 3735 struct m_tag *mtag; 3736 u_int8_t mpcp; 3737 3738 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3739 if (mtag == NULL) 3740 return (0); 3741 3742 if (prio == PF_PRIO_ZERO) 3743 prio = 0; 3744 3745 mpcp = *(uint8_t *)(mtag + 1); 3746 3747 return (mpcp == prio); 3748 } 3749 3750 static int 3751 pf_icmp_to_bandlim(uint8_t type) 3752 { 3753 switch (type) { 3754 case ICMP_ECHO: 3755 case ICMP_ECHOREPLY: 3756 return (BANDLIM_ICMP_ECHO); 3757 case ICMP_TSTAMP: 3758 case ICMP_TSTAMPREPLY: 3759 return (BANDLIM_ICMP_TSTAMP); 3760 case ICMP_UNREACH: 3761 default: 3762 return (BANDLIM_ICMP_UNREACH); 3763 } 3764 } 3765 3766 static void 3767 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3768 struct pf_krule *r, int rtableid) 3769 { 3770 struct pf_send_entry *pfse; 3771 struct mbuf *m0; 3772 struct pf_mtag *pf_mtag; 3773 3774 /* ICMP packet rate limitation. */ 3775 #ifdef INET6 3776 if (af == AF_INET6) { 3777 if (icmp6_ratelimit(NULL, type, code)) 3778 return; 3779 } 3780 #endif 3781 #ifdef INET 3782 if (af == AF_INET) { 3783 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3784 return; 3785 } 3786 #endif 3787 3788 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3789 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3790 if (pfse == NULL) 3791 return; 3792 3793 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3794 free(pfse, M_PFTEMP); 3795 return; 3796 } 3797 3798 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3799 free(pfse, M_PFTEMP); 3800 return; 3801 } 3802 /* XXX: revisit */ 3803 m0->m_flags |= M_SKIP_FIREWALL; 3804 3805 if (rtableid >= 0) 3806 M_SETFIB(m0, rtableid); 3807 3808 #ifdef ALTQ 3809 if (r->qid) { 3810 pf_mtag->qid = r->qid; 3811 /* add hints for ecn */ 3812 pf_mtag->hdr = mtod(m0, struct ip *); 3813 } 3814 #endif /* ALTQ */ 3815 3816 switch (af) { 3817 #ifdef INET 3818 case AF_INET: 3819 pfse->pfse_type = PFSE_ICMP; 3820 break; 3821 #endif /* INET */ 3822 #ifdef INET6 3823 case AF_INET6: 3824 pfse->pfse_type = PFSE_ICMP6; 3825 break; 3826 #endif /* INET6 */ 3827 } 3828 pfse->pfse_m = m0; 3829 pfse->icmpopts.type = type; 3830 pfse->icmpopts.code = code; 3831 pf_send(pfse); 3832 } 3833 3834 /* 3835 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3836 * If n is 0, they match if they are equal. If n is != 0, they match if they 3837 * are different. 3838 */ 3839 int 3840 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3841 struct pf_addr *b, sa_family_t af) 3842 { 3843 int match = 0; 3844 3845 switch (af) { 3846 #ifdef INET 3847 case AF_INET: 3848 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 3849 match++; 3850 break; 3851 #endif /* INET */ 3852 #ifdef INET6 3853 case AF_INET6: 3854 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 3855 match++; 3856 break; 3857 #endif /* INET6 */ 3858 } 3859 if (match) { 3860 if (n) 3861 return (0); 3862 else 3863 return (1); 3864 } else { 3865 if (n) 3866 return (1); 3867 else 3868 return (0); 3869 } 3870 } 3871 3872 /* 3873 * Return 1 if b <= a <= e, otherwise return 0. 3874 */ 3875 int 3876 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3877 struct pf_addr *a, sa_family_t af) 3878 { 3879 switch (af) { 3880 #ifdef INET 3881 case AF_INET: 3882 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3883 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3884 return (0); 3885 break; 3886 #endif /* INET */ 3887 #ifdef INET6 3888 case AF_INET6: { 3889 int i; 3890 3891 /* check a >= b */ 3892 for (i = 0; i < 4; ++i) 3893 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3894 break; 3895 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3896 return (0); 3897 /* check a <= e */ 3898 for (i = 0; i < 4; ++i) 3899 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3900 break; 3901 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3902 return (0); 3903 break; 3904 } 3905 #endif /* INET6 */ 3906 } 3907 return (1); 3908 } 3909 3910 static int 3911 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3912 { 3913 switch (op) { 3914 case PF_OP_IRG: 3915 return ((p > a1) && (p < a2)); 3916 case PF_OP_XRG: 3917 return ((p < a1) || (p > a2)); 3918 case PF_OP_RRG: 3919 return ((p >= a1) && (p <= a2)); 3920 case PF_OP_EQ: 3921 return (p == a1); 3922 case PF_OP_NE: 3923 return (p != a1); 3924 case PF_OP_LT: 3925 return (p < a1); 3926 case PF_OP_LE: 3927 return (p <= a1); 3928 case PF_OP_GT: 3929 return (p > a1); 3930 case PF_OP_GE: 3931 return (p >= a1); 3932 } 3933 return (0); /* never reached */ 3934 } 3935 3936 int 3937 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3938 { 3939 NTOHS(a1); 3940 NTOHS(a2); 3941 NTOHS(p); 3942 return (pf_match(op, a1, a2, p)); 3943 } 3944 3945 static int 3946 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3947 { 3948 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3949 return (0); 3950 return (pf_match(op, a1, a2, u)); 3951 } 3952 3953 static int 3954 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3955 { 3956 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3957 return (0); 3958 return (pf_match(op, a1, a2, g)); 3959 } 3960 3961 int 3962 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3963 { 3964 if (*tag == -1) 3965 *tag = mtag; 3966 3967 return ((!r->match_tag_not && r->match_tag == *tag) || 3968 (r->match_tag_not && r->match_tag != *tag)); 3969 } 3970 3971 static int 3972 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 3973 { 3974 struct ifnet *ifp = m->m_pkthdr.rcvif; 3975 struct pfi_kkif *kif; 3976 3977 if (ifp == NULL) 3978 return (0); 3979 3980 kif = (struct pfi_kkif *)ifp->if_pf_kif; 3981 3982 if (kif == NULL) { 3983 DPFPRINTF(PF_DEBUG_URGENT, 3984 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 3985 r->rcv_ifname)); 3986 return (0); 3987 } 3988 3989 return (pfi_kkif_match(r->rcv_kif, kif)); 3990 } 3991 3992 int 3993 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3994 { 3995 3996 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3997 3998 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3999 return (ENOMEM); 4000 4001 pd->pf_mtag->tag = tag; 4002 4003 return (0); 4004 } 4005 4006 #define PF_ANCHOR_STACKSIZE 32 4007 struct pf_kanchor_stackframe { 4008 struct pf_kruleset *rs; 4009 struct pf_krule *r; /* XXX: + match bit */ 4010 struct pf_kanchor *child; 4011 }; 4012 4013 /* 4014 * XXX: We rely on malloc(9) returning pointer aligned addresses. 4015 */ 4016 #define PF_ANCHORSTACK_MATCH 0x00000001 4017 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4018 4019 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4020 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4021 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4022 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4023 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4024 } while (0) 4025 4026 void 4027 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4028 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4029 int *match) 4030 { 4031 struct pf_kanchor_stackframe *f; 4032 4033 PF_RULES_RASSERT(); 4034 4035 if (match) 4036 *match = 0; 4037 if (*depth >= PF_ANCHOR_STACKSIZE) { 4038 printf("%s: anchor stack overflow on %s\n", 4039 __func__, (*r)->anchor->name); 4040 *r = TAILQ_NEXT(*r, entries); 4041 return; 4042 } else if (*depth == 0 && a != NULL) 4043 *a = *r; 4044 f = stack + (*depth)++; 4045 f->rs = *rs; 4046 f->r = *r; 4047 if ((*r)->anchor_wildcard) { 4048 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4049 4050 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4051 *r = NULL; 4052 return; 4053 } 4054 *rs = &f->child->ruleset; 4055 } else { 4056 f->child = NULL; 4057 *rs = &(*r)->anchor->ruleset; 4058 } 4059 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4060 } 4061 4062 int 4063 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4064 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4065 int *match) 4066 { 4067 struct pf_kanchor_stackframe *f; 4068 struct pf_krule *fr; 4069 int quick = 0; 4070 4071 PF_RULES_RASSERT(); 4072 4073 do { 4074 if (*depth <= 0) 4075 break; 4076 f = stack + *depth - 1; 4077 fr = PF_ANCHOR_RULE(f); 4078 if (f->child != NULL) { 4079 /* 4080 * This block traverses through 4081 * a wildcard anchor. 4082 */ 4083 if (match != NULL && *match) { 4084 /* 4085 * If any of "*" matched, then 4086 * "foo/ *" matched, mark frame 4087 * appropriately. 4088 */ 4089 PF_ANCHOR_SET_MATCH(f); 4090 *match = 0; 4091 } 4092 f->child = RB_NEXT(pf_kanchor_node, 4093 &fr->anchor->children, f->child); 4094 if (f->child != NULL) { 4095 *rs = &f->child->ruleset; 4096 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4097 if (*r == NULL) 4098 continue; 4099 else 4100 break; 4101 } 4102 } 4103 (*depth)--; 4104 if (*depth == 0 && a != NULL) 4105 *a = NULL; 4106 *rs = f->rs; 4107 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 4108 quick = fr->quick; 4109 *r = TAILQ_NEXT(fr, entries); 4110 } while (*r == NULL); 4111 4112 return (quick); 4113 } 4114 4115 struct pf_keth_anchor_stackframe { 4116 struct pf_keth_ruleset *rs; 4117 struct pf_keth_rule *r; /* XXX: + match bit */ 4118 struct pf_keth_anchor *child; 4119 }; 4120 4121 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4122 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4123 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4124 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4125 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4126 } while (0) 4127 4128 void 4129 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4130 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4131 struct pf_keth_rule **a, int *match) 4132 { 4133 struct pf_keth_anchor_stackframe *f; 4134 4135 NET_EPOCH_ASSERT(); 4136 4137 if (match) 4138 *match = 0; 4139 if (*depth >= PF_ANCHOR_STACKSIZE) { 4140 printf("%s: anchor stack overflow on %s\n", 4141 __func__, (*r)->anchor->name); 4142 *r = TAILQ_NEXT(*r, entries); 4143 return; 4144 } else if (*depth == 0 && a != NULL) 4145 *a = *r; 4146 f = stack + (*depth)++; 4147 f->rs = *rs; 4148 f->r = *r; 4149 if ((*r)->anchor_wildcard) { 4150 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4151 4152 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4153 *r = NULL; 4154 return; 4155 } 4156 *rs = &f->child->ruleset; 4157 } else { 4158 f->child = NULL; 4159 *rs = &(*r)->anchor->ruleset; 4160 } 4161 *r = TAILQ_FIRST((*rs)->active.rules); 4162 } 4163 4164 int 4165 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4166 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4167 struct pf_keth_rule **a, int *match) 4168 { 4169 struct pf_keth_anchor_stackframe *f; 4170 struct pf_keth_rule *fr; 4171 int quick = 0; 4172 4173 NET_EPOCH_ASSERT(); 4174 4175 do { 4176 if (*depth <= 0) 4177 break; 4178 f = stack + *depth - 1; 4179 fr = PF_ETH_ANCHOR_RULE(f); 4180 if (f->child != NULL) { 4181 /* 4182 * This block traverses through 4183 * a wildcard anchor. 4184 */ 4185 if (match != NULL && *match) { 4186 /* 4187 * If any of "*" matched, then 4188 * "foo/ *" matched, mark frame 4189 * appropriately. 4190 */ 4191 PF_ETH_ANCHOR_SET_MATCH(f); 4192 *match = 0; 4193 } 4194 f->child = RB_NEXT(pf_keth_anchor_node, 4195 &fr->anchor->children, f->child); 4196 if (f->child != NULL) { 4197 *rs = &f->child->ruleset; 4198 *r = TAILQ_FIRST((*rs)->active.rules); 4199 if (*r == NULL) 4200 continue; 4201 else 4202 break; 4203 } 4204 } 4205 (*depth)--; 4206 if (*depth == 0 && a != NULL) 4207 *a = NULL; 4208 *rs = f->rs; 4209 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4210 quick = fr->quick; 4211 *r = TAILQ_NEXT(fr, entries); 4212 } while (*r == NULL); 4213 4214 return (quick); 4215 } 4216 4217 #ifdef INET6 4218 void 4219 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4220 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4221 { 4222 switch (af) { 4223 #ifdef INET 4224 case AF_INET: 4225 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4226 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4227 break; 4228 #endif /* INET */ 4229 case AF_INET6: 4230 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4231 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4232 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4233 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4234 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4235 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4236 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4237 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4238 break; 4239 } 4240 } 4241 4242 void 4243 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4244 { 4245 switch (af) { 4246 #ifdef INET 4247 case AF_INET: 4248 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4249 break; 4250 #endif /* INET */ 4251 case AF_INET6: 4252 if (addr->addr32[3] == 0xffffffff) { 4253 addr->addr32[3] = 0; 4254 if (addr->addr32[2] == 0xffffffff) { 4255 addr->addr32[2] = 0; 4256 if (addr->addr32[1] == 0xffffffff) { 4257 addr->addr32[1] = 0; 4258 addr->addr32[0] = 4259 htonl(ntohl(addr->addr32[0]) + 1); 4260 } else 4261 addr->addr32[1] = 4262 htonl(ntohl(addr->addr32[1]) + 1); 4263 } else 4264 addr->addr32[2] = 4265 htonl(ntohl(addr->addr32[2]) + 1); 4266 } else 4267 addr->addr32[3] = 4268 htonl(ntohl(addr->addr32[3]) + 1); 4269 break; 4270 } 4271 } 4272 #endif /* INET6 */ 4273 4274 void 4275 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4276 { 4277 /* 4278 * Modern rules use the same flags in rules as they do in states. 4279 */ 4280 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4281 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4282 4283 /* 4284 * Old-style scrub rules have different flags which need to be translated. 4285 */ 4286 if (r->rule_flag & PFRULE_RANDOMID) 4287 a->flags |= PFSTATE_RANDOMID; 4288 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4289 a->flags |= PFSTATE_SETTOS; 4290 a->set_tos = r->set_tos; 4291 } 4292 4293 if (r->qid) 4294 a->qid = r->qid; 4295 if (r->pqid) 4296 a->pqid = r->pqid; 4297 if (r->rtableid >= 0) 4298 a->rtableid = r->rtableid; 4299 a->log |= r->log; 4300 if (r->min_ttl) 4301 a->min_ttl = r->min_ttl; 4302 if (r->max_mss) 4303 a->max_mss = r->max_mss; 4304 if (r->dnpipe) 4305 a->dnpipe = r->dnpipe; 4306 if (r->dnrpipe) 4307 a->dnrpipe = r->dnrpipe; 4308 if (r->dnpipe || r->dnrpipe) { 4309 if (r->free_flags & PFRULE_DN_IS_PIPE) 4310 a->flags |= PFSTATE_DN_IS_PIPE; 4311 else 4312 a->flags &= ~PFSTATE_DN_IS_PIPE; 4313 } 4314 if (r->scrub_flags & PFSTATE_SETPRIO) { 4315 a->set_prio[0] = r->set_prio[0]; 4316 a->set_prio[1] = r->set_prio[1]; 4317 } 4318 } 4319 4320 int 4321 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m) 4322 { 4323 struct pf_addr *saddr, *daddr; 4324 u_int16_t sport, dport; 4325 struct inpcbinfo *pi; 4326 struct inpcb *inp; 4327 4328 pd->lookup.uid = UID_MAX; 4329 pd->lookup.gid = GID_MAX; 4330 4331 switch (pd->proto) { 4332 case IPPROTO_TCP: 4333 sport = pd->hdr.tcp.th_sport; 4334 dport = pd->hdr.tcp.th_dport; 4335 pi = &V_tcbinfo; 4336 break; 4337 case IPPROTO_UDP: 4338 sport = pd->hdr.udp.uh_sport; 4339 dport = pd->hdr.udp.uh_dport; 4340 pi = &V_udbinfo; 4341 break; 4342 default: 4343 return (-1); 4344 } 4345 if (pd->dir == PF_IN) { 4346 saddr = pd->src; 4347 daddr = pd->dst; 4348 } else { 4349 u_int16_t p; 4350 4351 p = sport; 4352 sport = dport; 4353 dport = p; 4354 saddr = pd->dst; 4355 daddr = pd->src; 4356 } 4357 switch (pd->af) { 4358 #ifdef INET 4359 case AF_INET: 4360 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4361 dport, INPLOOKUP_RLOCKPCB, NULL, m); 4362 if (inp == NULL) { 4363 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4364 daddr->v4, dport, INPLOOKUP_WILDCARD | 4365 INPLOOKUP_RLOCKPCB, NULL, m); 4366 if (inp == NULL) 4367 return (-1); 4368 } 4369 break; 4370 #endif /* INET */ 4371 #ifdef INET6 4372 case AF_INET6: 4373 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4374 dport, INPLOOKUP_RLOCKPCB, NULL, m); 4375 if (inp == NULL) { 4376 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4377 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4378 INPLOOKUP_RLOCKPCB, NULL, m); 4379 if (inp == NULL) 4380 return (-1); 4381 } 4382 break; 4383 #endif /* INET6 */ 4384 4385 default: 4386 return (-1); 4387 } 4388 INP_RLOCK_ASSERT(inp); 4389 pd->lookup.uid = inp->inp_cred->cr_uid; 4390 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4391 INP_RUNLOCK(inp); 4392 4393 return (1); 4394 } 4395 4396 u_int8_t 4397 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 4398 { 4399 int hlen; 4400 u_int8_t hdr[60]; 4401 u_int8_t *opt, optlen; 4402 u_int8_t wscale = 0; 4403 4404 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 4405 if (hlen <= sizeof(struct tcphdr)) 4406 return (0); 4407 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 4408 return (0); 4409 opt = hdr + sizeof(struct tcphdr); 4410 hlen -= sizeof(struct tcphdr); 4411 while (hlen >= 3) { 4412 switch (*opt) { 4413 case TCPOPT_EOL: 4414 case TCPOPT_NOP: 4415 ++opt; 4416 --hlen; 4417 break; 4418 case TCPOPT_WINDOW: 4419 wscale = opt[2]; 4420 if (wscale > TCP_MAX_WINSHIFT) 4421 wscale = TCP_MAX_WINSHIFT; 4422 wscale |= PF_WSCALE_FLAG; 4423 /* FALLTHROUGH */ 4424 default: 4425 optlen = opt[1]; 4426 if (optlen < 2) 4427 optlen = 2; 4428 hlen -= optlen; 4429 opt += optlen; 4430 break; 4431 } 4432 } 4433 return (wscale); 4434 } 4435 4436 u_int16_t 4437 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 4438 { 4439 int hlen; 4440 u_int8_t hdr[60]; 4441 u_int8_t *opt, optlen; 4442 u_int16_t mss = V_tcp_mssdflt; 4443 4444 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 4445 if (hlen <= sizeof(struct tcphdr)) 4446 return (0); 4447 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 4448 return (0); 4449 opt = hdr + sizeof(struct tcphdr); 4450 hlen -= sizeof(struct tcphdr); 4451 while (hlen >= TCPOLEN_MAXSEG) { 4452 switch (*opt) { 4453 case TCPOPT_EOL: 4454 case TCPOPT_NOP: 4455 ++opt; 4456 --hlen; 4457 break; 4458 case TCPOPT_MAXSEG: 4459 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 4460 NTOHS(mss); 4461 /* FALLTHROUGH */ 4462 default: 4463 optlen = opt[1]; 4464 if (optlen < 2) 4465 optlen = 2; 4466 hlen -= optlen; 4467 opt += optlen; 4468 break; 4469 } 4470 } 4471 return (mss); 4472 } 4473 4474 static u_int16_t 4475 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 4476 { 4477 struct nhop_object *nh; 4478 #ifdef INET6 4479 struct in6_addr dst6; 4480 uint32_t scopeid; 4481 #endif /* INET6 */ 4482 int hlen = 0; 4483 uint16_t mss = 0; 4484 4485 NET_EPOCH_ASSERT(); 4486 4487 switch (af) { 4488 #ifdef INET 4489 case AF_INET: 4490 hlen = sizeof(struct ip); 4491 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 4492 if (nh != NULL) 4493 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4494 break; 4495 #endif /* INET */ 4496 #ifdef INET6 4497 case AF_INET6: 4498 hlen = sizeof(struct ip6_hdr); 4499 in6_splitscope(&addr->v6, &dst6, &scopeid); 4500 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4501 if (nh != NULL) 4502 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4503 break; 4504 #endif /* INET6 */ 4505 } 4506 4507 mss = max(V_tcp_mssdflt, mss); 4508 mss = min(mss, offer); 4509 mss = max(mss, 64); /* sanity - at least max opt space */ 4510 return (mss); 4511 } 4512 4513 static u_int32_t 4514 pf_tcp_iss(struct pf_pdesc *pd) 4515 { 4516 MD5_CTX ctx; 4517 u_int32_t digest[4]; 4518 4519 if (V_pf_tcp_secret_init == 0) { 4520 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4521 MD5Init(&V_pf_tcp_secret_ctx); 4522 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4523 sizeof(V_pf_tcp_secret)); 4524 V_pf_tcp_secret_init = 1; 4525 } 4526 4527 ctx = V_pf_tcp_secret_ctx; 4528 4529 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4530 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4531 if (pd->af == AF_INET6) { 4532 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4533 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4534 } else { 4535 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4536 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4537 } 4538 MD5Final((u_char *)digest, &ctx); 4539 V_pf_tcp_iss_off += 4096; 4540 #define ISN_RANDOM_INCREMENT (4096 - 1) 4541 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4542 V_pf_tcp_iss_off); 4543 #undef ISN_RANDOM_INCREMENT 4544 } 4545 4546 static bool 4547 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4548 { 4549 bool match = true; 4550 4551 /* Always matches if not set */ 4552 if (! r->isset) 4553 return (!r->neg); 4554 4555 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4556 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4557 match = false; 4558 break; 4559 } 4560 } 4561 4562 return (match ^ r->neg); 4563 } 4564 4565 static int 4566 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4567 { 4568 if (*tag == -1) 4569 *tag = mtag; 4570 4571 return ((!r->match_tag_not && r->match_tag == *tag) || 4572 (r->match_tag_not && r->match_tag != *tag)); 4573 } 4574 4575 static void 4576 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4577 { 4578 /* If we don't have the interface drop the packet. */ 4579 if (ifp == NULL) { 4580 m_freem(m); 4581 return; 4582 } 4583 4584 switch (ifp->if_type) { 4585 case IFT_ETHER: 4586 case IFT_XETHER: 4587 case IFT_L2VLAN: 4588 case IFT_BRIDGE: 4589 case IFT_IEEE8023ADLAG: 4590 break; 4591 default: 4592 m_freem(m); 4593 return; 4594 } 4595 4596 ifp->if_transmit(ifp, m); 4597 } 4598 4599 static int 4600 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4601 { 4602 #ifdef INET 4603 struct ip ip; 4604 #endif 4605 #ifdef INET6 4606 struct ip6_hdr ip6; 4607 #endif 4608 struct mbuf *m = *m0; 4609 struct ether_header *e; 4610 struct pf_keth_rule *r, *rm, *a = NULL; 4611 struct pf_keth_ruleset *ruleset = NULL; 4612 struct pf_mtag *mtag; 4613 struct pf_keth_ruleq *rules; 4614 struct pf_addr *src = NULL, *dst = NULL; 4615 struct pfi_kkif *bridge_to; 4616 sa_family_t af = 0; 4617 uint16_t proto; 4618 int asd = 0, match = 0; 4619 int tag = -1; 4620 uint8_t action; 4621 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4622 4623 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4624 NET_EPOCH_ASSERT(); 4625 4626 PF_RULES_RLOCK_TRACKER; 4627 4628 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4629 4630 mtag = pf_find_mtag(m); 4631 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4632 /* Dummynet re-injects packets after they've 4633 * completed their delay. We've already 4634 * processed them, so pass unconditionally. */ 4635 4636 /* But only once. We may see the packet multiple times (e.g. 4637 * PFIL_IN/PFIL_OUT). */ 4638 pf_dummynet_flag_remove(m, mtag); 4639 4640 return (PF_PASS); 4641 } 4642 4643 ruleset = V_pf_keth; 4644 rules = ck_pr_load_ptr(&ruleset->active.rules); 4645 r = TAILQ_FIRST(rules); 4646 rm = NULL; 4647 4648 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 4649 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 4650 DPFPRINTF(PF_DEBUG_URGENT, 4651 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 4652 ", pullup failed\n")); 4653 return (PF_DROP); 4654 } 4655 e = mtod(m, struct ether_header *); 4656 proto = ntohs(e->ether_type); 4657 4658 switch (proto) { 4659 #ifdef INET 4660 case ETHERTYPE_IP: { 4661 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4662 sizeof(ip))) 4663 return (PF_DROP); 4664 4665 af = AF_INET; 4666 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4667 (caddr_t)&ip); 4668 src = (struct pf_addr *)&ip.ip_src; 4669 dst = (struct pf_addr *)&ip.ip_dst; 4670 break; 4671 } 4672 #endif /* INET */ 4673 #ifdef INET6 4674 case ETHERTYPE_IPV6: { 4675 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4676 sizeof(ip6))) 4677 return (PF_DROP); 4678 4679 af = AF_INET6; 4680 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4681 (caddr_t)&ip6); 4682 src = (struct pf_addr *)&ip6.ip6_src; 4683 dst = (struct pf_addr *)&ip6.ip6_dst; 4684 break; 4685 } 4686 #endif /* INET6 */ 4687 } 4688 4689 PF_RULES_RLOCK(); 4690 4691 while (r != NULL) { 4692 counter_u64_add(r->evaluations, 1); 4693 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4694 4695 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4696 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4697 "kif"); 4698 r = r->skip[PFE_SKIP_IFP].ptr; 4699 } 4700 else if (r->direction && r->direction != dir) { 4701 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4702 "dir"); 4703 r = r->skip[PFE_SKIP_DIR].ptr; 4704 } 4705 else if (r->proto && r->proto != proto) { 4706 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4707 "proto"); 4708 r = r->skip[PFE_SKIP_PROTO].ptr; 4709 } 4710 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4711 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4712 "src"); 4713 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4714 } 4715 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4716 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4717 "dst"); 4718 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4719 } 4720 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4721 r->ipsrc.neg, kif, M_GETFIB(m))) { 4722 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4723 "ip_src"); 4724 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4725 } 4726 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4727 r->ipdst.neg, kif, M_GETFIB(m))) { 4728 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4729 "ip_dst"); 4730 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4731 } 4732 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4733 mtag ? mtag->tag : 0)) { 4734 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4735 "match_tag"); 4736 r = TAILQ_NEXT(r, entries); 4737 } 4738 else { 4739 if (r->tag) 4740 tag = r->tag; 4741 if (r->anchor == NULL) { 4742 /* Rule matches */ 4743 rm = r; 4744 4745 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4746 4747 if (r->quick) 4748 break; 4749 4750 r = TAILQ_NEXT(r, entries); 4751 } else { 4752 pf_step_into_keth_anchor(anchor_stack, &asd, 4753 &ruleset, &r, &a, &match); 4754 } 4755 } 4756 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4757 &ruleset, &r, &a, &match)) 4758 break; 4759 } 4760 4761 r = rm; 4762 4763 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4764 4765 /* Default to pass. */ 4766 if (r == NULL) { 4767 PF_RULES_RUNLOCK(); 4768 return (PF_PASS); 4769 } 4770 4771 /* Execute action. */ 4772 counter_u64_add(r->packets[dir == PF_OUT], 1); 4773 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4774 pf_update_timestamp(r); 4775 4776 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4777 if (r->action == PF_DROP) { 4778 PF_RULES_RUNLOCK(); 4779 return (PF_DROP); 4780 } 4781 4782 if (tag > 0) { 4783 if (mtag == NULL) 4784 mtag = pf_get_mtag(m); 4785 if (mtag == NULL) { 4786 PF_RULES_RUNLOCK(); 4787 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4788 return (PF_DROP); 4789 } 4790 mtag->tag = tag; 4791 } 4792 4793 if (r->qid != 0) { 4794 if (mtag == NULL) 4795 mtag = pf_get_mtag(m); 4796 if (mtag == NULL) { 4797 PF_RULES_RUNLOCK(); 4798 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4799 return (PF_DROP); 4800 } 4801 mtag->qid = r->qid; 4802 } 4803 4804 action = r->action; 4805 bridge_to = r->bridge_to; 4806 4807 /* Dummynet */ 4808 if (r->dnpipe) { 4809 struct ip_fw_args dnflow; 4810 4811 /* Drop packet if dummynet is not loaded. */ 4812 if (ip_dn_io_ptr == NULL) { 4813 PF_RULES_RUNLOCK(); 4814 m_freem(m); 4815 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4816 return (PF_DROP); 4817 } 4818 if (mtag == NULL) 4819 mtag = pf_get_mtag(m); 4820 if (mtag == NULL) { 4821 PF_RULES_RUNLOCK(); 4822 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4823 return (PF_DROP); 4824 } 4825 4826 bzero(&dnflow, sizeof(dnflow)); 4827 4828 /* We don't have port numbers here, so we set 0. That means 4829 * that we'll be somewhat limited in distinguishing flows (i.e. 4830 * only based on IP addresses, not based on port numbers), but 4831 * it's better than nothing. */ 4832 dnflow.f_id.dst_port = 0; 4833 dnflow.f_id.src_port = 0; 4834 dnflow.f_id.proto = 0; 4835 4836 dnflow.rule.info = r->dnpipe; 4837 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4838 if (r->dnflags & PFRULE_DN_IS_PIPE) 4839 dnflow.rule.info |= IPFW_IS_PIPE; 4840 4841 dnflow.f_id.extra = dnflow.rule.info; 4842 4843 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4844 dnflow.flags |= IPFW_ARGS_ETHER; 4845 dnflow.ifp = kif->pfik_ifp; 4846 4847 switch (af) { 4848 case AF_INET: 4849 dnflow.f_id.addr_type = 4; 4850 dnflow.f_id.src_ip = src->v4.s_addr; 4851 dnflow.f_id.dst_ip = dst->v4.s_addr; 4852 break; 4853 case AF_INET6: 4854 dnflow.flags |= IPFW_ARGS_IP6; 4855 dnflow.f_id.addr_type = 6; 4856 dnflow.f_id.src_ip6 = src->v6; 4857 dnflow.f_id.dst_ip6 = dst->v6; 4858 break; 4859 } 4860 4861 PF_RULES_RUNLOCK(); 4862 4863 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4864 ip_dn_io_ptr(m0, &dnflow); 4865 if (*m0 != NULL) 4866 pf_dummynet_flag_remove(m, mtag); 4867 } else { 4868 PF_RULES_RUNLOCK(); 4869 } 4870 4871 if (action == PF_PASS && bridge_to) { 4872 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4873 *m0 = NULL; /* We've eaten the packet. */ 4874 } 4875 4876 return (action); 4877 } 4878 4879 #define PF_TEST_ATTRIB(t, a)\ 4880 do { \ 4881 if (t) { \ 4882 r = a; \ 4883 goto nextrule; \ 4884 } \ 4885 } while (0) 4886 4887 static int 4888 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif, 4889 struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am, 4890 struct pf_kruleset **rsm, struct inpcb *inp, int hdrlen) 4891 { 4892 struct pf_krule *nr = NULL; 4893 struct pf_addr * const saddr = pd->src; 4894 struct pf_addr * const daddr = pd->dst; 4895 sa_family_t af = pd->af; 4896 struct pf_krule *r, *a = NULL; 4897 struct pf_kruleset *ruleset = NULL; 4898 struct pf_krule_slist match_rules; 4899 struct pf_krule_item *ri; 4900 struct pf_ksrc_node *nsn = NULL; 4901 struct tcphdr *th = &pd->hdr.tcp; 4902 struct pf_state_key *sk = NULL, *nk = NULL; 4903 u_short reason, transerror; 4904 int rewrite = 0; 4905 int tag = -1; 4906 int asd = 0; 4907 int match = 0; 4908 int state_icmp = 0, icmp_dir, multi; 4909 u_int16_t sport = 0, dport = 0, virtual_type, virtual_id; 4910 u_int16_t bproto_sum = 0, bip_sum = 0; 4911 u_int8_t icmptype = 0, icmpcode = 0; 4912 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4913 struct pf_udp_mapping *udp_mapping = NULL; 4914 4915 PF_RULES_RASSERT(); 4916 4917 SLIST_INIT(&match_rules); 4918 4919 if (inp != NULL) { 4920 INP_LOCK_ASSERT(inp); 4921 pd->lookup.uid = inp->inp_cred->cr_uid; 4922 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4923 pd->lookup.done = 1; 4924 } 4925 4926 switch (pd->virtual_proto) { 4927 case IPPROTO_TCP: 4928 sport = th->th_sport; 4929 dport = th->th_dport; 4930 break; 4931 case IPPROTO_UDP: 4932 sport = pd->hdr.udp.uh_sport; 4933 dport = pd->hdr.udp.uh_dport; 4934 break; 4935 case IPPROTO_SCTP: 4936 sport = pd->hdr.sctp.src_port; 4937 dport = pd->hdr.sctp.dest_port; 4938 break; 4939 #ifdef INET 4940 case IPPROTO_ICMP: 4941 MPASS(af == AF_INET); 4942 icmptype = pd->hdr.icmp.icmp_type; 4943 icmpcode = pd->hdr.icmp.icmp_code; 4944 state_icmp = pf_icmp_mapping(pd, icmptype, 4945 &icmp_dir, &multi, &virtual_id, &virtual_type); 4946 if (icmp_dir == PF_IN) { 4947 sport = virtual_id; 4948 dport = virtual_type; 4949 } else { 4950 sport = virtual_type; 4951 dport = virtual_id; 4952 } 4953 break; 4954 #endif /* INET */ 4955 #ifdef INET6 4956 case IPPROTO_ICMPV6: 4957 MPASS(af == AF_INET6); 4958 icmptype = pd->hdr.icmp6.icmp6_type; 4959 icmpcode = pd->hdr.icmp6.icmp6_code; 4960 state_icmp = pf_icmp_mapping(pd, icmptype, 4961 &icmp_dir, &multi, &virtual_id, &virtual_type); 4962 if (icmp_dir == PF_IN) { 4963 sport = virtual_id; 4964 dport = virtual_type; 4965 } else { 4966 sport = virtual_type; 4967 dport = virtual_id; 4968 } 4969 4970 break; 4971 #endif /* INET6 */ 4972 default: 4973 sport = dport = 0; 4974 break; 4975 } 4976 4977 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4978 4979 /* check packet for BINAT/NAT/RDR */ 4980 transerror = pf_get_translation(pd, m, off, kif, &nsn, &sk, 4981 &nk, saddr, daddr, sport, dport, anchor_stack, &nr, &udp_mapping); 4982 switch (transerror) { 4983 default: 4984 /* A translation error occurred. */ 4985 REASON_SET(&reason, transerror); 4986 goto cleanup; 4987 case PFRES_MAX: 4988 /* No match. */ 4989 break; 4990 case PFRES_MATCH: 4991 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4992 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4993 4994 if (nr->log) { 4995 PFLOG_PACKET(kif, m, PF_PASS, PFRES_MATCH, nr, a, 4996 ruleset, pd, 1); 4997 } 4998 4999 if (pd->ip_sum) 5000 bip_sum = *pd->ip_sum; 5001 5002 switch (pd->proto) { 5003 case IPPROTO_TCP: 5004 bproto_sum = th->th_sum; 5005 pd->proto_sum = &th->th_sum; 5006 5007 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 5008 nk->port[pd->sidx] != sport) { 5009 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 5010 &th->th_sum, &nk->addr[pd->sidx], 5011 nk->port[pd->sidx], 0, af); 5012 pd->sport = &th->th_sport; 5013 sport = th->th_sport; 5014 } 5015 5016 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 5017 nk->port[pd->didx] != dport) { 5018 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 5019 &th->th_sum, &nk->addr[pd->didx], 5020 nk->port[pd->didx], 0, af); 5021 dport = th->th_dport; 5022 pd->dport = &th->th_dport; 5023 } 5024 rewrite++; 5025 break; 5026 case IPPROTO_UDP: 5027 bproto_sum = pd->hdr.udp.uh_sum; 5028 pd->proto_sum = &pd->hdr.udp.uh_sum; 5029 5030 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 5031 nk->port[pd->sidx] != sport) { 5032 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 5033 pd->ip_sum, &pd->hdr.udp.uh_sum, 5034 &nk->addr[pd->sidx], 5035 nk->port[pd->sidx], 1, af); 5036 sport = pd->hdr.udp.uh_sport; 5037 pd->sport = &pd->hdr.udp.uh_sport; 5038 } 5039 5040 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 5041 nk->port[pd->didx] != dport) { 5042 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 5043 pd->ip_sum, &pd->hdr.udp.uh_sum, 5044 &nk->addr[pd->didx], 5045 nk->port[pd->didx], 1, af); 5046 dport = pd->hdr.udp.uh_dport; 5047 pd->dport = &pd->hdr.udp.uh_dport; 5048 } 5049 rewrite++; 5050 break; 5051 case IPPROTO_SCTP: { 5052 uint16_t checksum = 0; 5053 5054 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 5055 nk->port[pd->sidx] != sport) { 5056 pf_change_ap(m, saddr, &pd->hdr.sctp.src_port, 5057 pd->ip_sum, &checksum, 5058 &nk->addr[pd->sidx], 5059 nk->port[pd->sidx], 1, af); 5060 } 5061 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 5062 nk->port[pd->didx] != dport) { 5063 pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port, 5064 pd->ip_sum, &checksum, 5065 &nk->addr[pd->didx], 5066 nk->port[pd->didx], 1, af); 5067 } 5068 break; 5069 } 5070 #ifdef INET 5071 case IPPROTO_ICMP: 5072 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 5073 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 5074 nk->addr[pd->sidx].v4.s_addr, 0); 5075 5076 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 5077 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 5078 nk->addr[pd->didx].v4.s_addr, 0); 5079 5080 if (virtual_type == htons(ICMP_ECHO) && 5081 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5082 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5083 pd->hdr.icmp.icmp_cksum, sport, 5084 nk->port[pd->sidx], 0); 5085 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5086 pd->sport = &pd->hdr.icmp.icmp_id; 5087 } 5088 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5089 break; 5090 #endif /* INET */ 5091 #ifdef INET6 5092 case IPPROTO_ICMPV6: 5093 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 5094 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 5095 &nk->addr[pd->sidx], 0); 5096 5097 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 5098 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 5099 &nk->addr[pd->didx], 0); 5100 rewrite++; 5101 break; 5102 #endif /* INET */ 5103 default: 5104 switch (af) { 5105 #ifdef INET 5106 case AF_INET: 5107 if (PF_ANEQ(saddr, 5108 &nk->addr[pd->sidx], AF_INET)) 5109 pf_change_a(&saddr->v4.s_addr, 5110 pd->ip_sum, 5111 nk->addr[pd->sidx].v4.s_addr, 0); 5112 5113 if (PF_ANEQ(daddr, 5114 &nk->addr[pd->didx], AF_INET)) 5115 pf_change_a(&daddr->v4.s_addr, 5116 pd->ip_sum, 5117 nk->addr[pd->didx].v4.s_addr, 0); 5118 break; 5119 #endif /* INET */ 5120 #ifdef INET6 5121 case AF_INET6: 5122 if (PF_ANEQ(saddr, 5123 &nk->addr[pd->sidx], AF_INET6)) 5124 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 5125 5126 if (PF_ANEQ(daddr, 5127 &nk->addr[pd->didx], AF_INET6)) 5128 PF_ACPY(daddr, &nk->addr[pd->didx], af); 5129 break; 5130 #endif /* INET */ 5131 } 5132 break; 5133 } 5134 if (nr->natpass) 5135 r = NULL; 5136 } 5137 5138 while (r != NULL) { 5139 pf_counter_u64_add(&r->evaluations, 1); 5140 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, kif) == r->ifnot, 5141 r->skip[PF_SKIP_IFP]); 5142 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5143 r->skip[PF_SKIP_DIR]); 5144 PF_TEST_ATTRIB(r->af && r->af != af, 5145 r->skip[PF_SKIP_AF]); 5146 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5147 r->skip[PF_SKIP_PROTO]); 5148 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, saddr, af, 5149 r->src.neg, kif, M_GETFIB(m)), 5150 r->skip[PF_SKIP_SRC_ADDR]); 5151 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, daddr, af, 5152 r->dst.neg, NULL, M_GETFIB(m)), 5153 r->skip[PF_SKIP_DST_ADDR]); 5154 switch (pd->virtual_proto) { 5155 case PF_VPROTO_FRAGMENT: 5156 /* tcp/udp only. port_op always 0 in other cases */ 5157 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5158 TAILQ_NEXT(r, entries)); 5159 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5160 TAILQ_NEXT(r, entries)); 5161 /* icmp only. type/code always 0 in other cases */ 5162 PF_TEST_ATTRIB((r->type || r->code), 5163 TAILQ_NEXT(r, entries)); 5164 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5165 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5166 TAILQ_NEXT(r, entries)); 5167 break; 5168 5169 case IPPROTO_TCP: 5170 PF_TEST_ATTRIB((r->flagset & th->th_flags) != r->flags, 5171 TAILQ_NEXT(r, entries)); 5172 /* FALLTHROUGH */ 5173 case IPPROTO_SCTP: 5174 case IPPROTO_UDP: 5175 /* tcp/udp only. port_op always 0 in other cases */ 5176 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5177 r->src.port[0], r->src.port[1], sport), 5178 r->skip[PF_SKIP_SRC_PORT]); 5179 /* tcp/udp only. port_op always 0 in other cases */ 5180 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5181 r->dst.port[0], r->dst.port[1], dport), 5182 r->skip[PF_SKIP_DST_PORT]); 5183 /* tcp/udp only. uid.op always 0 in other cases */ 5184 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5185 pf_socket_lookup(pd, m), 1)) && 5186 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5187 pd->lookup.uid), 5188 TAILQ_NEXT(r, entries)); 5189 /* tcp/udp only. gid.op always 0 in other cases */ 5190 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5191 pf_socket_lookup(pd, m), 1)) && 5192 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5193 pd->lookup.gid), 5194 TAILQ_NEXT(r, entries)); 5195 break; 5196 5197 case IPPROTO_ICMP: 5198 case IPPROTO_ICMPV6: 5199 /* icmp only. type always 0 in other cases */ 5200 PF_TEST_ATTRIB(r->type && r->type != icmptype + 1, 5201 TAILQ_NEXT(r, entries)); 5202 /* icmp only. type always 0 in other cases */ 5203 PF_TEST_ATTRIB(r->code && r->code != icmpcode + 1, 5204 TAILQ_NEXT(r, entries)); 5205 break; 5206 5207 default: 5208 break; 5209 } 5210 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5211 TAILQ_NEXT(r, entries)); 5212 PF_TEST_ATTRIB(r->prio && 5213 !pf_match_ieee8021q_pcp(r->prio, m), 5214 TAILQ_NEXT(r, entries)); 5215 PF_TEST_ATTRIB(r->prob && 5216 r->prob <= arc4random(), 5217 TAILQ_NEXT(r, entries)); 5218 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(m, r, &tag, 5219 pd->pf_mtag ? pd->pf_mtag->tag : 0), 5220 TAILQ_NEXT(r, entries)); 5221 PF_TEST_ATTRIB(r->rcv_kif && !pf_match_rcvif(m, r), 5222 TAILQ_NEXT(r, entries)); 5223 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5224 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5225 TAILQ_NEXT(r, entries)); 5226 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5227 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5228 pf_osfp_fingerprint(pd, m, off, th), 5229 r->os_fingerprint)), 5230 TAILQ_NEXT(r, entries)); 5231 /* FALLTHROUGH */ 5232 if (r->tag) 5233 tag = r->tag; 5234 if (r->anchor == NULL) { 5235 if (r->action == PF_MATCH) { 5236 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5237 if (ri == NULL) { 5238 REASON_SET(&reason, PFRES_MEMORY); 5239 goto cleanup; 5240 } 5241 ri->r = r; 5242 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5243 pf_counter_u64_critical_enter(); 5244 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5245 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5246 pf_counter_u64_critical_exit(); 5247 pf_rule_to_actions(r, &pd->act); 5248 if (r->log || pd->act.log & PF_LOG_MATCHES) 5249 PFLOG_PACKET(kif, m, 5250 r->action, PFRES_MATCH, r, 5251 a, ruleset, pd, 1); 5252 } else { 5253 match = 1; 5254 *rm = r; 5255 *am = a; 5256 *rsm = ruleset; 5257 if (pd->act.log & PF_LOG_MATCHES) 5258 PFLOG_PACKET(kif, m, 5259 r->action, PFRES_MATCH, r, 5260 a, ruleset, pd, 1); 5261 } 5262 if ((*rm)->quick) 5263 break; 5264 r = TAILQ_NEXT(r, entries); 5265 } else 5266 pf_step_into_anchor(anchor_stack, &asd, 5267 &ruleset, PF_RULESET_FILTER, &r, &a, 5268 &match); 5269 nextrule: 5270 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5271 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5272 break; 5273 } 5274 r = *rm; 5275 a = *am; 5276 ruleset = *rsm; 5277 5278 REASON_SET(&reason, PFRES_MATCH); 5279 5280 /* apply actions for last matching pass/block rule */ 5281 pf_rule_to_actions(r, &pd->act); 5282 5283 if (r->log || pd->act.log & PF_LOG_MATCHES) { 5284 if (rewrite) 5285 m_copyback(m, off, hdrlen, pd->hdr.any); 5286 PFLOG_PACKET(kif, m, r->action, reason, r, a, ruleset, pd, 1); 5287 } 5288 5289 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5290 (r->action == PF_DROP) && 5291 ((r->rule_flag & PFRULE_RETURNRST) || 5292 (r->rule_flag & PFRULE_RETURNICMP) || 5293 (r->rule_flag & PFRULE_RETURN))) { 5294 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 5295 bip_sum, hdrlen, &reason, r->rtableid); 5296 } 5297 5298 if (r->action == PF_DROP) 5299 goto cleanup; 5300 5301 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5302 REASON_SET(&reason, PFRES_MEMORY); 5303 goto cleanup; 5304 } 5305 if (pd->act.rtableid >= 0) 5306 M_SETFIB(m, pd->act.rtableid); 5307 5308 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5309 (!state_icmp && (r->keep_state || nr != NULL || 5310 (pd->flags & PFDESC_TCP_NORM)))) { 5311 int action; 5312 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 5313 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 5314 hdrlen, &match_rules, udp_mapping); 5315 if (action != PF_PASS) { 5316 pf_udp_mapping_release(udp_mapping); 5317 if (action == PF_DROP && 5318 (r->rule_flag & PFRULE_RETURN)) 5319 pf_return(r, nr, pd, sk, off, m, th, kif, 5320 bproto_sum, bip_sum, hdrlen, &reason, 5321 pd->act.rtableid); 5322 return (action); 5323 } 5324 } else { 5325 while ((ri = SLIST_FIRST(&match_rules))) { 5326 SLIST_REMOVE_HEAD(&match_rules, entry); 5327 free(ri, M_PF_RULE_ITEM); 5328 } 5329 5330 uma_zfree(V_pf_state_key_z, sk); 5331 uma_zfree(V_pf_state_key_z, nk); 5332 pf_udp_mapping_release(udp_mapping); 5333 } 5334 5335 /* copy back packet headers if we performed NAT operations */ 5336 if (rewrite) 5337 m_copyback(m, off, hdrlen, pd->hdr.any); 5338 5339 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5340 pd->dir == PF_OUT && 5341 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 5342 /* 5343 * We want the state created, but we dont 5344 * want to send this in case a partner 5345 * firewall has to know about it to allow 5346 * replies through it. 5347 */ 5348 return (PF_DEFER); 5349 5350 return (PF_PASS); 5351 5352 cleanup: 5353 while ((ri = SLIST_FIRST(&match_rules))) { 5354 SLIST_REMOVE_HEAD(&match_rules, entry); 5355 free(ri, M_PF_RULE_ITEM); 5356 } 5357 5358 uma_zfree(V_pf_state_key_z, sk); 5359 uma_zfree(V_pf_state_key_z, nk); 5360 pf_udp_mapping_release(udp_mapping); 5361 5362 return (PF_DROP); 5363 } 5364 5365 static int 5366 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 5367 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 5368 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 5369 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 5370 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 5371 struct pf_krule_slist *match_rules, struct pf_udp_mapping *udp_mapping) 5372 { 5373 struct pf_kstate *s = NULL; 5374 struct pf_ksrc_node *sn = NULL; 5375 struct tcphdr *th = &pd->hdr.tcp; 5376 u_int16_t mss = V_tcp_mssdflt; 5377 u_short reason, sn_reason; 5378 struct pf_krule_item *ri; 5379 5380 /* check maximums */ 5381 if (r->max_states && 5382 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 5383 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 5384 REASON_SET(&reason, PFRES_MAXSTATES); 5385 goto csfailed; 5386 } 5387 /* src node for filter rule */ 5388 if ((r->rule_flag & PFRULE_SRCTRACK || 5389 r->rpool.opts & PF_POOL_STICKYADDR) && 5390 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 5391 REASON_SET(&reason, sn_reason); 5392 goto csfailed; 5393 } 5394 /* src node for translation rule */ 5395 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 5396 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 5397 pd->af)) != 0 ) { 5398 REASON_SET(&reason, sn_reason); 5399 goto csfailed; 5400 } 5401 s = pf_alloc_state(M_NOWAIT); 5402 if (s == NULL) { 5403 REASON_SET(&reason, PFRES_MEMORY); 5404 goto csfailed; 5405 } 5406 s->rule = r; 5407 s->nat_rule = nr; 5408 s->anchor = a; 5409 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 5410 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 5411 5412 STATE_INC_COUNTERS(s); 5413 if (r->allow_opts) 5414 s->state_flags |= PFSTATE_ALLOWOPTS; 5415 if (r->rule_flag & PFRULE_STATESLOPPY) 5416 s->state_flags |= PFSTATE_SLOPPY; 5417 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 5418 s->state_flags |= PFSTATE_SCRUB_TCP; 5419 if ((r->rule_flag & PFRULE_PFLOW) || 5420 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 5421 s->state_flags |= PFSTATE_PFLOW; 5422 5423 s->act.log = pd->act.log & PF_LOG_ALL; 5424 s->sync_state = PFSYNC_S_NONE; 5425 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 5426 5427 if (nr != NULL) 5428 s->act.log |= nr->log & PF_LOG_ALL; 5429 switch (pd->proto) { 5430 case IPPROTO_TCP: 5431 s->src.seqlo = ntohl(th->th_seq); 5432 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 5433 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 5434 r->keep_state == PF_STATE_MODULATE) { 5435 /* Generate sequence number modulator */ 5436 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 5437 0) 5438 s->src.seqdiff = 1; 5439 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 5440 htonl(s->src.seqlo + s->src.seqdiff), 0); 5441 *rewrite = 1; 5442 } else 5443 s->src.seqdiff = 0; 5444 if (th->th_flags & TH_SYN) { 5445 s->src.seqhi++; 5446 s->src.wscale = pf_get_wscale(m, off, 5447 th->th_off, pd->af); 5448 } 5449 s->src.max_win = MAX(ntohs(th->th_win), 1); 5450 if (s->src.wscale & PF_WSCALE_MASK) { 5451 /* Remove scale factor from initial window */ 5452 int win = s->src.max_win; 5453 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 5454 s->src.max_win = (win - 1) >> 5455 (s->src.wscale & PF_WSCALE_MASK); 5456 } 5457 if (th->th_flags & TH_FIN) 5458 s->src.seqhi++; 5459 s->dst.seqhi = 1; 5460 s->dst.max_win = 1; 5461 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 5462 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 5463 s->timeout = PFTM_TCP_FIRST_PACKET; 5464 atomic_add_32(&V_pf_status.states_halfopen, 1); 5465 break; 5466 case IPPROTO_UDP: 5467 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 5468 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 5469 s->timeout = PFTM_UDP_FIRST_PACKET; 5470 break; 5471 case IPPROTO_SCTP: 5472 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 5473 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 5474 s->timeout = PFTM_SCTP_FIRST_PACKET; 5475 break; 5476 case IPPROTO_ICMP: 5477 #ifdef INET6 5478 case IPPROTO_ICMPV6: 5479 #endif 5480 s->timeout = PFTM_ICMP_FIRST_PACKET; 5481 break; 5482 default: 5483 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 5484 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 5485 s->timeout = PFTM_OTHER_FIRST_PACKET; 5486 } 5487 5488 if (r->rt) { 5489 /* pf_map_addr increases the reason counters */ 5490 if ((reason = pf_map_addr_sn(pd->af, r, pd->src, &s->rt_addr, 5491 &s->rt_kif, NULL, &sn)) != 0) 5492 goto csfailed; 5493 s->rt = r->rt; 5494 } 5495 5496 s->creation = s->expire = pf_get_uptime(); 5497 5498 if (sn != NULL) 5499 s->src_node = sn; 5500 if (nsn != NULL) { 5501 /* XXX We only modify one side for now. */ 5502 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 5503 s->nat_src_node = nsn; 5504 } 5505 if (pd->proto == IPPROTO_TCP) { 5506 if (s->state_flags & PFSTATE_SCRUB_TCP && 5507 pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { 5508 REASON_SET(&reason, PFRES_MEMORY); 5509 goto csfailed; 5510 } 5511 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 5512 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 5513 &s->src, &s->dst, rewrite)) { 5514 /* This really shouldn't happen!!! */ 5515 DPFPRINTF(PF_DEBUG_URGENT, 5516 ("pf_normalize_tcp_stateful failed on first " 5517 "pkt\n")); 5518 goto csfailed; 5519 } 5520 } else if (pd->proto == IPPROTO_SCTP) { 5521 if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst)) 5522 goto csfailed; 5523 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 5524 goto csfailed; 5525 } 5526 s->direction = pd->dir; 5527 5528 /* 5529 * sk/nk could already been setup by pf_get_translation(). 5530 */ 5531 if (nr == NULL) { 5532 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 5533 __func__, nr, sk, nk)); 5534 sk = pf_state_key_setup(pd, m, off, pd->src, pd->dst, sport, dport); 5535 if (sk == NULL) 5536 goto csfailed; 5537 nk = sk; 5538 } else 5539 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 5540 __func__, nr, sk, nk)); 5541 5542 /* Swap sk/nk for PF_OUT. */ 5543 if (pf_state_insert(BOUND_IFACE(s, kif), kif, 5544 (pd->dir == PF_IN) ? sk : nk, 5545 (pd->dir == PF_IN) ? nk : sk, s)) { 5546 REASON_SET(&reason, PFRES_STATEINS); 5547 goto drop; 5548 } else 5549 *sm = s; 5550 5551 if (tag > 0) 5552 s->tag = tag; 5553 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 5554 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 5555 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 5556 /* undo NAT changes, if they have taken place */ 5557 if (nr != NULL) { 5558 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 5559 if (pd->dir == PF_OUT) 5560 skt = s->key[PF_SK_STACK]; 5561 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 5562 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 5563 if (pd->sport) 5564 *pd->sport = skt->port[pd->sidx]; 5565 if (pd->dport) 5566 *pd->dport = skt->port[pd->didx]; 5567 if (pd->proto_sum) 5568 *pd->proto_sum = bproto_sum; 5569 if (pd->ip_sum) 5570 *pd->ip_sum = bip_sum; 5571 m_copyback(m, off, hdrlen, pd->hdr.any); 5572 } 5573 s->src.seqhi = htonl(arc4random()); 5574 /* Find mss option */ 5575 int rtid = M_GETFIB(m); 5576 mss = pf_get_mss(m, off, th->th_off, pd->af); 5577 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5578 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5579 s->src.mss = mss; 5580 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5581 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5582 TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0, 5583 pd->act.rtableid); 5584 REASON_SET(&reason, PFRES_SYNPROXY); 5585 return (PF_SYNPROXY_DROP); 5586 } 5587 5588 s->udp_mapping = udp_mapping; 5589 5590 return (PF_PASS); 5591 5592 csfailed: 5593 while ((ri = SLIST_FIRST(match_rules))) { 5594 SLIST_REMOVE_HEAD(match_rules, entry); 5595 free(ri, M_PF_RULE_ITEM); 5596 } 5597 5598 uma_zfree(V_pf_state_key_z, sk); 5599 uma_zfree(V_pf_state_key_z, nk); 5600 5601 if (sn != NULL) { 5602 PF_SRC_NODE_LOCK(sn); 5603 if (--sn->states == 0 && sn->expire == 0) { 5604 pf_unlink_src_node(sn); 5605 uma_zfree(V_pf_sources_z, sn); 5606 counter_u64_add( 5607 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5608 } 5609 PF_SRC_NODE_UNLOCK(sn); 5610 } 5611 5612 if (nsn != sn && nsn != NULL) { 5613 PF_SRC_NODE_LOCK(nsn); 5614 if (--nsn->states == 0 && nsn->expire == 0) { 5615 pf_unlink_src_node(nsn); 5616 uma_zfree(V_pf_sources_z, nsn); 5617 counter_u64_add( 5618 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5619 } 5620 PF_SRC_NODE_UNLOCK(nsn); 5621 } 5622 5623 drop: 5624 if (s != NULL) { 5625 pf_src_tree_remove_state(s); 5626 s->timeout = PFTM_UNLINKED; 5627 STATE_DEC_COUNTERS(s); 5628 pf_free_state(s); 5629 } 5630 5631 return (PF_DROP); 5632 } 5633 5634 static int 5635 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 5636 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 5637 int *copyback) 5638 { 5639 struct tcphdr *th = &pd->hdr.tcp; 5640 struct pf_state_peer *src, *dst; 5641 u_int16_t win = ntohs(th->th_win); 5642 u_int32_t ack, end, data_end, seq, orig_seq; 5643 u_int8_t sws, dws, psrc, pdst; 5644 int ackskew; 5645 5646 if (pd->dir == (*state)->direction) { 5647 src = &(*state)->src; 5648 dst = &(*state)->dst; 5649 psrc = PF_PEER_SRC; 5650 pdst = PF_PEER_DST; 5651 } else { 5652 src = &(*state)->dst; 5653 dst = &(*state)->src; 5654 psrc = PF_PEER_DST; 5655 pdst = PF_PEER_SRC; 5656 } 5657 5658 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5659 sws = src->wscale & PF_WSCALE_MASK; 5660 dws = dst->wscale & PF_WSCALE_MASK; 5661 } else 5662 sws = dws = 0; 5663 5664 /* 5665 * Sequence tracking algorithm from Guido van Rooij's paper: 5666 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5667 * tcp_filtering.ps 5668 */ 5669 5670 orig_seq = seq = ntohl(th->th_seq); 5671 if (src->seqlo == 0) { 5672 /* First packet from this end. Set its state */ 5673 5674 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5675 src->scrub == NULL) { 5676 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 5677 REASON_SET(reason, PFRES_MEMORY); 5678 return (PF_DROP); 5679 } 5680 } 5681 5682 /* Deferred generation of sequence number modulator */ 5683 if (dst->seqdiff && !src->seqdiff) { 5684 /* use random iss for the TCP server */ 5685 while ((src->seqdiff = arc4random() - seq) == 0) 5686 ; 5687 ack = ntohl(th->th_ack) - dst->seqdiff; 5688 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5689 src->seqdiff), 0); 5690 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5691 *copyback = 1; 5692 } else { 5693 ack = ntohl(th->th_ack); 5694 } 5695 5696 end = seq + pd->p_len; 5697 if (th->th_flags & TH_SYN) { 5698 end++; 5699 if (dst->wscale & PF_WSCALE_FLAG) { 5700 src->wscale = pf_get_wscale(m, off, th->th_off, 5701 pd->af); 5702 if (src->wscale & PF_WSCALE_FLAG) { 5703 /* Remove scale factor from initial 5704 * window */ 5705 sws = src->wscale & PF_WSCALE_MASK; 5706 win = ((u_int32_t)win + (1 << sws) - 1) 5707 >> sws; 5708 dws = dst->wscale & PF_WSCALE_MASK; 5709 } else { 5710 /* fixup other window */ 5711 dst->max_win = MIN(TCP_MAXWIN, 5712 (u_int32_t)dst->max_win << 5713 (dst->wscale & PF_WSCALE_MASK)); 5714 /* in case of a retrans SYN|ACK */ 5715 dst->wscale = 0; 5716 } 5717 } 5718 } 5719 data_end = end; 5720 if (th->th_flags & TH_FIN) 5721 end++; 5722 5723 src->seqlo = seq; 5724 if (src->state < TCPS_SYN_SENT) 5725 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5726 5727 /* 5728 * May need to slide the window (seqhi may have been set by 5729 * the crappy stack check or if we picked up the connection 5730 * after establishment) 5731 */ 5732 if (src->seqhi == 1 || 5733 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5734 src->seqhi = end + MAX(1, dst->max_win << dws); 5735 if (win > src->max_win) 5736 src->max_win = win; 5737 5738 } else { 5739 ack = ntohl(th->th_ack) - dst->seqdiff; 5740 if (src->seqdiff) { 5741 /* Modulate sequence numbers */ 5742 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5743 src->seqdiff), 0); 5744 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5745 *copyback = 1; 5746 } 5747 end = seq + pd->p_len; 5748 if (th->th_flags & TH_SYN) 5749 end++; 5750 data_end = end; 5751 if (th->th_flags & TH_FIN) 5752 end++; 5753 } 5754 5755 if ((th->th_flags & TH_ACK) == 0) { 5756 /* Let it pass through the ack skew check */ 5757 ack = dst->seqlo; 5758 } else if ((ack == 0 && 5759 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5760 /* broken tcp stacks do not set ack */ 5761 (dst->state < TCPS_SYN_SENT)) { 5762 /* 5763 * Many stacks (ours included) will set the ACK number in an 5764 * FIN|ACK if the SYN times out -- no sequence to ACK. 5765 */ 5766 ack = dst->seqlo; 5767 } 5768 5769 if (seq == end) { 5770 /* Ease sequencing restrictions on no data packets */ 5771 seq = src->seqlo; 5772 data_end = end = seq; 5773 } 5774 5775 ackskew = dst->seqlo - ack; 5776 5777 /* 5778 * Need to demodulate the sequence numbers in any TCP SACK options 5779 * (Selective ACK). We could optionally validate the SACK values 5780 * against the current ACK window, either forwards or backwards, but 5781 * I'm not confident that SACK has been implemented properly 5782 * everywhere. It wouldn't surprise me if several stacks accidentally 5783 * SACK too far backwards of previously ACKed data. There really aren't 5784 * any security implications of bad SACKing unless the target stack 5785 * doesn't validate the option length correctly. Someone trying to 5786 * spoof into a TCP connection won't bother blindly sending SACK 5787 * options anyway. 5788 */ 5789 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5790 if (pf_modulate_sack(m, off, pd, th, dst)) 5791 *copyback = 1; 5792 } 5793 5794 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5795 if (SEQ_GEQ(src->seqhi, data_end) && 5796 /* Last octet inside other's window space */ 5797 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5798 /* Retrans: not more than one window back */ 5799 (ackskew >= -MAXACKWINDOW) && 5800 /* Acking not more than one reassembled fragment backwards */ 5801 (ackskew <= (MAXACKWINDOW << sws)) && 5802 /* Acking not more than one window forward */ 5803 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5804 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 5805 /* Require an exact/+1 sequence match on resets when possible */ 5806 5807 if (dst->scrub || src->scrub) { 5808 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5809 *state, src, dst, copyback)) 5810 return (PF_DROP); 5811 } 5812 5813 /* update max window */ 5814 if (src->max_win < win) 5815 src->max_win = win; 5816 /* synchronize sequencing */ 5817 if (SEQ_GT(end, src->seqlo)) 5818 src->seqlo = end; 5819 /* slide the window of what the other end can send */ 5820 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5821 dst->seqhi = ack + MAX((win << sws), 1); 5822 5823 /* update states */ 5824 if (th->th_flags & TH_SYN) 5825 if (src->state < TCPS_SYN_SENT) 5826 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5827 if (th->th_flags & TH_FIN) 5828 if (src->state < TCPS_CLOSING) 5829 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5830 if (th->th_flags & TH_ACK) { 5831 if (dst->state == TCPS_SYN_SENT) { 5832 pf_set_protostate(*state, pdst, 5833 TCPS_ESTABLISHED); 5834 if (src->state == TCPS_ESTABLISHED && 5835 (*state)->src_node != NULL && 5836 pf_src_connlimit(state)) { 5837 REASON_SET(reason, PFRES_SRCLIMIT); 5838 return (PF_DROP); 5839 } 5840 } else if (dst->state == TCPS_CLOSING) 5841 pf_set_protostate(*state, pdst, 5842 TCPS_FIN_WAIT_2); 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 /* Fall through to PASS packet */ 5865 5866 } else if ((dst->state < TCPS_SYN_SENT || 5867 dst->state >= TCPS_FIN_WAIT_2 || 5868 src->state >= TCPS_FIN_WAIT_2) && 5869 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 5870 /* Within a window forward of the originating packet */ 5871 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5872 /* Within a window backward of the originating packet */ 5873 5874 /* 5875 * This currently handles three situations: 5876 * 1) Stupid stacks will shotgun SYNs before their peer 5877 * replies. 5878 * 2) When PF catches an already established stream (the 5879 * firewall rebooted, the state table was flushed, routes 5880 * changed...) 5881 * 3) Packets get funky immediately after the connection 5882 * closes (this should catch Solaris spurious ACK|FINs 5883 * that web servers like to spew after a close) 5884 * 5885 * This must be a little more careful than the above code 5886 * since packet floods will also be caught here. We don't 5887 * update the TTL here to mitigate the damage of a packet 5888 * flood and so the same code can handle awkward establishment 5889 * and a loosened connection close. 5890 * In the establishment case, a correct peer response will 5891 * validate the connection, go through the normal state code 5892 * and keep updating the state TTL. 5893 */ 5894 5895 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5896 printf("pf: loose state match: "); 5897 pf_print_state(*state); 5898 pf_print_flags(th->th_flags); 5899 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5900 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5901 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5902 (unsigned long long)(*state)->packets[1], 5903 pd->dir == PF_IN ? "in" : "out", 5904 pd->dir == (*state)->direction ? "fwd" : "rev"); 5905 } 5906 5907 if (dst->scrub || src->scrub) { 5908 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5909 *state, src, dst, copyback)) 5910 return (PF_DROP); 5911 } 5912 5913 /* update max window */ 5914 if (src->max_win < win) 5915 src->max_win = win; 5916 /* synchronize sequencing */ 5917 if (SEQ_GT(end, src->seqlo)) 5918 src->seqlo = end; 5919 /* slide the window of what the other end can send */ 5920 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5921 dst->seqhi = ack + MAX((win << sws), 1); 5922 5923 /* 5924 * Cannot set dst->seqhi here since this could be a shotgunned 5925 * SYN and not an already established connection. 5926 */ 5927 5928 if (th->th_flags & TH_FIN) 5929 if (src->state < TCPS_CLOSING) 5930 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5931 if (th->th_flags & TH_RST) 5932 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5933 5934 /* Fall through to PASS packet */ 5935 5936 } else { 5937 if ((*state)->dst.state == TCPS_SYN_SENT && 5938 (*state)->src.state == TCPS_SYN_SENT) { 5939 /* Send RST for state mismatches during handshake */ 5940 if (!(th->th_flags & TH_RST)) 5941 pf_send_tcp((*state)->rule, pd->af, 5942 pd->dst, pd->src, th->th_dport, 5943 th->th_sport, ntohl(th->th_ack), 0, 5944 TH_RST, 0, 0, 5945 (*state)->rule->return_ttl, true, 0, 0, 5946 (*state)->act.rtableid); 5947 src->seqlo = 0; 5948 src->seqhi = 1; 5949 src->max_win = 1; 5950 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5951 printf("pf: BAD state: "); 5952 pf_print_state(*state); 5953 pf_print_flags(th->th_flags); 5954 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5955 "pkts=%llu:%llu dir=%s,%s\n", 5956 seq, orig_seq, ack, pd->p_len, ackskew, 5957 (unsigned long long)(*state)->packets[0], 5958 (unsigned long long)(*state)->packets[1], 5959 pd->dir == PF_IN ? "in" : "out", 5960 pd->dir == (*state)->direction ? "fwd" : "rev"); 5961 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5962 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 5963 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5964 ' ': '2', 5965 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5966 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5967 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 5968 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5969 } 5970 REASON_SET(reason, PFRES_BADSTATE); 5971 return (PF_DROP); 5972 } 5973 5974 return (PF_PASS); 5975 } 5976 5977 static int 5978 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5979 { 5980 struct tcphdr *th = &pd->hdr.tcp; 5981 struct pf_state_peer *src, *dst; 5982 u_int8_t psrc, pdst; 5983 5984 if (pd->dir == (*state)->direction) { 5985 src = &(*state)->src; 5986 dst = &(*state)->dst; 5987 psrc = PF_PEER_SRC; 5988 pdst = PF_PEER_DST; 5989 } else { 5990 src = &(*state)->dst; 5991 dst = &(*state)->src; 5992 psrc = PF_PEER_DST; 5993 pdst = PF_PEER_SRC; 5994 } 5995 5996 if (th->th_flags & TH_SYN) 5997 if (src->state < TCPS_SYN_SENT) 5998 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5999 if (th->th_flags & TH_FIN) 6000 if (src->state < TCPS_CLOSING) 6001 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6002 if (th->th_flags & TH_ACK) { 6003 if (dst->state == TCPS_SYN_SENT) { 6004 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 6005 if (src->state == TCPS_ESTABLISHED && 6006 (*state)->src_node != NULL && 6007 pf_src_connlimit(state)) { 6008 REASON_SET(reason, PFRES_SRCLIMIT); 6009 return (PF_DROP); 6010 } 6011 } else if (dst->state == TCPS_CLOSING) { 6012 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 6013 } else if (src->state == TCPS_SYN_SENT && 6014 dst->state < TCPS_SYN_SENT) { 6015 /* 6016 * Handle a special sloppy case where we only see one 6017 * half of the connection. If there is a ACK after 6018 * the initial SYN without ever seeing a packet from 6019 * the destination, set the connection to established. 6020 */ 6021 pf_set_protostate(*state, PF_PEER_BOTH, 6022 TCPS_ESTABLISHED); 6023 dst->state = src->state = TCPS_ESTABLISHED; 6024 if ((*state)->src_node != NULL && 6025 pf_src_connlimit(state)) { 6026 REASON_SET(reason, PFRES_SRCLIMIT); 6027 return (PF_DROP); 6028 } 6029 } else if (src->state == TCPS_CLOSING && 6030 dst->state == TCPS_ESTABLISHED && 6031 dst->seqlo == 0) { 6032 /* 6033 * Handle the closing of half connections where we 6034 * don't see the full bidirectional FIN/ACK+ACK 6035 * handshake. 6036 */ 6037 pf_set_protostate(*state, pdst, TCPS_CLOSING); 6038 } 6039 } 6040 if (th->th_flags & TH_RST) 6041 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6042 6043 /* update expire time */ 6044 (*state)->expire = pf_get_uptime(); 6045 if (src->state >= TCPS_FIN_WAIT_2 && 6046 dst->state >= TCPS_FIN_WAIT_2) 6047 (*state)->timeout = PFTM_TCP_CLOSED; 6048 else if (src->state >= TCPS_CLOSING && 6049 dst->state >= TCPS_CLOSING) 6050 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6051 else if (src->state < TCPS_ESTABLISHED || 6052 dst->state < TCPS_ESTABLISHED) 6053 (*state)->timeout = PFTM_TCP_OPENING; 6054 else if (src->state >= TCPS_CLOSING || 6055 dst->state >= TCPS_CLOSING) 6056 (*state)->timeout = PFTM_TCP_CLOSING; 6057 else 6058 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6059 6060 return (PF_PASS); 6061 } 6062 6063 static int 6064 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 6065 { 6066 struct pf_state_key *sk = (*state)->key[pd->didx]; 6067 struct tcphdr *th = &pd->hdr.tcp; 6068 6069 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 6070 if (pd->dir != (*state)->direction) { 6071 REASON_SET(reason, PFRES_SYNPROXY); 6072 return (PF_SYNPROXY_DROP); 6073 } 6074 if (th->th_flags & TH_SYN) { 6075 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 6076 REASON_SET(reason, PFRES_SYNPROXY); 6077 return (PF_DROP); 6078 } 6079 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6080 pd->src, th->th_dport, th->th_sport, 6081 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 6082 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0, 6083 (*state)->act.rtableid); 6084 REASON_SET(reason, PFRES_SYNPROXY); 6085 return (PF_SYNPROXY_DROP); 6086 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6087 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6088 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6089 REASON_SET(reason, PFRES_SYNPROXY); 6090 return (PF_DROP); 6091 } else if ((*state)->src_node != NULL && 6092 pf_src_connlimit(state)) { 6093 REASON_SET(reason, PFRES_SRCLIMIT); 6094 return (PF_DROP); 6095 } else 6096 pf_set_protostate(*state, PF_PEER_SRC, 6097 PF_TCPS_PROXY_DST); 6098 } 6099 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 6100 if (pd->dir == (*state)->direction) { 6101 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 6102 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6103 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6104 REASON_SET(reason, PFRES_SYNPROXY); 6105 return (PF_DROP); 6106 } 6107 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 6108 if ((*state)->dst.seqhi == 1) 6109 (*state)->dst.seqhi = htonl(arc4random()); 6110 pf_send_tcp((*state)->rule, pd->af, 6111 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6112 sk->port[pd->sidx], sk->port[pd->didx], 6113 (*state)->dst.seqhi, 0, TH_SYN, 0, 6114 (*state)->src.mss, 0, false, (*state)->tag, 0, 6115 (*state)->act.rtableid); 6116 REASON_SET(reason, PFRES_SYNPROXY); 6117 return (PF_SYNPROXY_DROP); 6118 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 6119 (TH_SYN|TH_ACK)) || 6120 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 6121 REASON_SET(reason, PFRES_SYNPROXY); 6122 return (PF_DROP); 6123 } else { 6124 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 6125 (*state)->dst.seqlo = ntohl(th->th_seq); 6126 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6127 pd->src, th->th_dport, th->th_sport, 6128 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6129 TH_ACK, (*state)->src.max_win, 0, 0, false, 6130 (*state)->tag, 0, (*state)->act.rtableid); 6131 pf_send_tcp((*state)->rule, pd->af, 6132 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6133 sk->port[pd->sidx], sk->port[pd->didx], 6134 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 6135 TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0, 6136 (*state)->act.rtableid); 6137 (*state)->src.seqdiff = (*state)->dst.seqhi - 6138 (*state)->src.seqlo; 6139 (*state)->dst.seqdiff = (*state)->src.seqhi - 6140 (*state)->dst.seqlo; 6141 (*state)->src.seqhi = (*state)->src.seqlo + 6142 (*state)->dst.max_win; 6143 (*state)->dst.seqhi = (*state)->dst.seqlo + 6144 (*state)->src.max_win; 6145 (*state)->src.wscale = (*state)->dst.wscale = 0; 6146 pf_set_protostate(*state, PF_PEER_BOTH, 6147 TCPS_ESTABLISHED); 6148 REASON_SET(reason, PFRES_SYNPROXY); 6149 return (PF_SYNPROXY_DROP); 6150 } 6151 } 6152 6153 return (PF_PASS); 6154 } 6155 6156 static int 6157 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif, 6158 struct mbuf *m, int off, struct pf_pdesc *pd, 6159 u_short *reason) 6160 { 6161 struct pf_state_key_cmp key; 6162 struct tcphdr *th = &pd->hdr.tcp; 6163 int copyback = 0; 6164 int action; 6165 struct pf_state_peer *src, *dst; 6166 6167 bzero(&key, sizeof(key)); 6168 key.af = pd->af; 6169 key.proto = IPPROTO_TCP; 6170 if (pd->dir == PF_IN) { /* wire side, straight */ 6171 PF_ACPY(&key.addr[0], pd->src, key.af); 6172 PF_ACPY(&key.addr[1], pd->dst, key.af); 6173 key.port[0] = th->th_sport; 6174 key.port[1] = th->th_dport; 6175 } else { /* stack side, reverse */ 6176 PF_ACPY(&key.addr[1], pd->src, key.af); 6177 PF_ACPY(&key.addr[0], pd->dst, key.af); 6178 key.port[1] = th->th_sport; 6179 key.port[0] = th->th_dport; 6180 } 6181 6182 STATE_LOOKUP(kif, &key, *state, pd); 6183 6184 if (pd->dir == (*state)->direction) { 6185 src = &(*state)->src; 6186 dst = &(*state)->dst; 6187 } else { 6188 src = &(*state)->dst; 6189 dst = &(*state)->src; 6190 } 6191 6192 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 6193 return (action); 6194 6195 if (dst->state >= TCPS_FIN_WAIT_2 && 6196 src->state >= TCPS_FIN_WAIT_2 && 6197 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 6198 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6199 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6200 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6201 printf("pf: state reuse "); 6202 pf_print_state(*state); 6203 pf_print_flags(th->th_flags); 6204 printf("\n"); 6205 } 6206 /* XXX make sure it's the same direction ?? */ 6207 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6208 pf_unlink_state(*state); 6209 *state = NULL; 6210 return (PF_DROP); 6211 } 6212 6213 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6214 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 6215 return (PF_DROP); 6216 } else { 6217 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 6218 ©back) == PF_DROP) 6219 return (PF_DROP); 6220 } 6221 6222 /* translate source/destination address, if necessary */ 6223 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6224 struct pf_state_key *nk = (*state)->key[pd->didx]; 6225 6226 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6227 nk->port[pd->sidx] != th->th_sport) 6228 pf_change_ap(m, pd->src, &th->th_sport, 6229 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 6230 nk->port[pd->sidx], 0, pd->af); 6231 6232 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6233 nk->port[pd->didx] != th->th_dport) 6234 pf_change_ap(m, pd->dst, &th->th_dport, 6235 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 6236 nk->port[pd->didx], 0, pd->af); 6237 copyback = 1; 6238 } 6239 6240 /* Copyback sequence modulation or stateful scrub changes if needed */ 6241 if (copyback) 6242 m_copyback(m, off, sizeof(*th), (caddr_t)th); 6243 6244 return (PF_PASS); 6245 } 6246 6247 static int 6248 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif, 6249 struct mbuf *m, int off, struct pf_pdesc *pd) 6250 { 6251 struct pf_state_peer *src, *dst; 6252 struct pf_state_key_cmp key; 6253 struct udphdr *uh = &pd->hdr.udp; 6254 uint8_t psrc, pdst; 6255 6256 bzero(&key, sizeof(key)); 6257 key.af = pd->af; 6258 key.proto = IPPROTO_UDP; 6259 if (pd->dir == PF_IN) { /* wire side, straight */ 6260 PF_ACPY(&key.addr[0], pd->src, key.af); 6261 PF_ACPY(&key.addr[1], pd->dst, key.af); 6262 key.port[0] = uh->uh_sport; 6263 key.port[1] = uh->uh_dport; 6264 } else { /* stack side, reverse */ 6265 PF_ACPY(&key.addr[1], pd->src, key.af); 6266 PF_ACPY(&key.addr[0], pd->dst, key.af); 6267 key.port[1] = uh->uh_sport; 6268 key.port[0] = uh->uh_dport; 6269 } 6270 6271 STATE_LOOKUP(kif, &key, *state, pd); 6272 6273 if (pd->dir == (*state)->direction) { 6274 src = &(*state)->src; 6275 dst = &(*state)->dst; 6276 psrc = PF_PEER_SRC; 6277 pdst = PF_PEER_DST; 6278 } else { 6279 src = &(*state)->dst; 6280 dst = &(*state)->src; 6281 psrc = PF_PEER_DST; 6282 pdst = PF_PEER_SRC; 6283 } 6284 6285 /* update states */ 6286 if (src->state < PFUDPS_SINGLE) 6287 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 6288 if (dst->state == PFUDPS_SINGLE) 6289 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 6290 6291 /* update expire time */ 6292 (*state)->expire = pf_get_uptime(); 6293 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 6294 (*state)->timeout = PFTM_UDP_MULTIPLE; 6295 else 6296 (*state)->timeout = PFTM_UDP_SINGLE; 6297 6298 /* translate source/destination address, if necessary */ 6299 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6300 struct pf_state_key *nk = (*state)->key[pd->didx]; 6301 6302 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6303 nk->port[pd->sidx] != uh->uh_sport) 6304 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 6305 &uh->uh_sum, &nk->addr[pd->sidx], 6306 nk->port[pd->sidx], 1, pd->af); 6307 6308 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6309 nk->port[pd->didx] != uh->uh_dport) 6310 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 6311 &uh->uh_sum, &nk->addr[pd->didx], 6312 nk->port[pd->didx], 1, pd->af); 6313 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 6314 } 6315 6316 return (PF_PASS); 6317 } 6318 6319 static int 6320 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif, 6321 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason) 6322 { 6323 struct pf_state_key_cmp key; 6324 struct pf_state_peer *src, *dst; 6325 struct sctphdr *sh = &pd->hdr.sctp; 6326 u_int8_t psrc; //, pdst; 6327 6328 bzero(&key, sizeof(key)); 6329 key.af = pd->af; 6330 key.proto = IPPROTO_SCTP; 6331 if (pd->dir == PF_IN) { /* wire side, straight */ 6332 PF_ACPY(&key.addr[0], pd->src, key.af); 6333 PF_ACPY(&key.addr[1], pd->dst, key.af); 6334 key.port[0] = sh->src_port; 6335 key.port[1] = sh->dest_port; 6336 } else { /* stack side, reverse */ 6337 PF_ACPY(&key.addr[1], pd->src, key.af); 6338 PF_ACPY(&key.addr[0], pd->dst, key.af); 6339 key.port[1] = sh->src_port; 6340 key.port[0] = sh->dest_port; 6341 } 6342 6343 STATE_LOOKUP(kif, &key, *state, pd); 6344 6345 if (pd->dir == (*state)->direction) { 6346 src = &(*state)->src; 6347 dst = &(*state)->dst; 6348 psrc = PF_PEER_SRC; 6349 } else { 6350 src = &(*state)->dst; 6351 dst = &(*state)->src; 6352 psrc = PF_PEER_DST; 6353 } 6354 6355 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 6356 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 6357 pd->sctp_flags & PFDESC_SCTP_INIT) { 6358 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 6359 pf_unlink_state(*state); 6360 *state = NULL; 6361 return (PF_DROP); 6362 } 6363 6364 /* Track state. */ 6365 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 6366 if (src->state < SCTP_COOKIE_WAIT) { 6367 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 6368 (*state)->timeout = PFTM_SCTP_OPENING; 6369 } 6370 } 6371 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 6372 MPASS(dst->scrub != NULL); 6373 if (dst->scrub->pfss_v_tag == 0) 6374 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 6375 } 6376 6377 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 6378 if (src->state < SCTP_ESTABLISHED) { 6379 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 6380 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 6381 } 6382 } 6383 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 6384 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6385 if (src->state < SCTP_SHUTDOWN_PENDING) { 6386 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 6387 (*state)->timeout = PFTM_SCTP_CLOSING; 6388 } 6389 } 6390 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6391 pf_set_protostate(*state, psrc, SCTP_CLOSED); 6392 (*state)->timeout = PFTM_SCTP_CLOSED; 6393 } 6394 6395 if (src->scrub != NULL) { 6396 if (src->scrub->pfss_v_tag == 0) { 6397 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 6398 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 6399 return (PF_DROP); 6400 } 6401 6402 (*state)->expire = pf_get_uptime(); 6403 6404 /* translate source/destination address, if necessary */ 6405 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6406 uint16_t checksum = 0; 6407 struct pf_state_key *nk = (*state)->key[pd->didx]; 6408 6409 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6410 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 6411 pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port, 6412 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 6413 nk->port[pd->sidx], 1, pd->af); 6414 } 6415 6416 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6417 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 6418 pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port, 6419 pd->ip_sum, &checksum, &nk->addr[pd->didx], 6420 nk->port[pd->didx], 1, pd->af); 6421 } 6422 } 6423 6424 return (PF_PASS); 6425 } 6426 6427 static void 6428 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 6429 { 6430 struct pf_sctp_endpoint key; 6431 struct pf_sctp_endpoint *ep; 6432 struct pf_state_key *sks = s->key[PF_SK_STACK]; 6433 struct pf_sctp_source *i, *tmp; 6434 6435 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 6436 return; 6437 6438 PF_SCTP_ENDPOINTS_LOCK(); 6439 6440 key.v_tag = s->dst.scrub->pfss_v_tag; 6441 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6442 if (ep != NULL) { 6443 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6444 if (pf_addr_cmp(&i->addr, 6445 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 6446 s->key[PF_SK_WIRE]->af) == 0) { 6447 SDT_PROBE3(pf, sctp, multihome, remove, 6448 key.v_tag, s, i); 6449 TAILQ_REMOVE(&ep->sources, i, entry); 6450 free(i, M_PFTEMP); 6451 break; 6452 } 6453 } 6454 6455 if (TAILQ_EMPTY(&ep->sources)) { 6456 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6457 free(ep, M_PFTEMP); 6458 } 6459 } 6460 6461 /* Other direction. */ 6462 key.v_tag = s->src.scrub->pfss_v_tag; 6463 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6464 if (ep != NULL) { 6465 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6466 if (pf_addr_cmp(&i->addr, 6467 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6468 s->key[PF_SK_WIRE]->af) == 0) { 6469 SDT_PROBE3(pf, sctp, multihome, remove, 6470 key.v_tag, s, i); 6471 TAILQ_REMOVE(&ep->sources, i, entry); 6472 free(i, M_PFTEMP); 6473 break; 6474 } 6475 } 6476 6477 if (TAILQ_EMPTY(&ep->sources)) { 6478 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6479 free(ep, M_PFTEMP); 6480 } 6481 } 6482 6483 PF_SCTP_ENDPOINTS_UNLOCK(); 6484 } 6485 6486 static void 6487 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6488 { 6489 struct pf_sctp_endpoint key = { 6490 .v_tag = v_tag, 6491 }; 6492 struct pf_sctp_source *i; 6493 struct pf_sctp_endpoint *ep; 6494 6495 PF_SCTP_ENDPOINTS_LOCK(); 6496 6497 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6498 if (ep == NULL) { 6499 ep = malloc(sizeof(struct pf_sctp_endpoint), 6500 M_PFTEMP, M_NOWAIT); 6501 if (ep == NULL) { 6502 PF_SCTP_ENDPOINTS_UNLOCK(); 6503 return; 6504 } 6505 6506 ep->v_tag = v_tag; 6507 TAILQ_INIT(&ep->sources); 6508 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6509 } 6510 6511 /* Avoid inserting duplicates. */ 6512 TAILQ_FOREACH(i, &ep->sources, entry) { 6513 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6514 PF_SCTP_ENDPOINTS_UNLOCK(); 6515 return; 6516 } 6517 } 6518 6519 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6520 if (i == NULL) { 6521 PF_SCTP_ENDPOINTS_UNLOCK(); 6522 return; 6523 } 6524 6525 i->af = pd->af; 6526 memcpy(&i->addr, a, sizeof(*a)); 6527 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6528 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 6529 6530 PF_SCTP_ENDPOINTS_UNLOCK(); 6531 } 6532 6533 static void 6534 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif, 6535 struct pf_kstate *s, int action) 6536 { 6537 struct pf_sctp_multihome_job *j, *tmp; 6538 struct pf_sctp_source *i; 6539 int ret __unused; 6540 struct pf_kstate *sm = NULL; 6541 struct pf_krule *ra = NULL; 6542 struct pf_krule *r = &V_pf_default_rule; 6543 struct pf_kruleset *rs = NULL; 6544 bool do_extra = true; 6545 6546 PF_RULES_RLOCK_TRACKER; 6547 6548 again: 6549 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6550 if (s == NULL || action != PF_PASS) 6551 goto free; 6552 6553 /* Confirm we don't recurse here. */ 6554 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6555 6556 switch (j->op) { 6557 case SCTP_ADD_IP_ADDRESS: { 6558 uint32_t v_tag = pd->sctp_initiate_tag; 6559 6560 if (v_tag == 0) { 6561 if (s->direction == pd->dir) 6562 v_tag = s->src.scrub->pfss_v_tag; 6563 else 6564 v_tag = s->dst.scrub->pfss_v_tag; 6565 } 6566 6567 /* 6568 * Avoid duplicating states. We'll already have 6569 * created a state based on the source address of 6570 * the packet, but SCTP endpoints may also list this 6571 * address again in the INIT(_ACK) parameters. 6572 */ 6573 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6574 break; 6575 } 6576 6577 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6578 PF_RULES_RLOCK(); 6579 sm = NULL; 6580 /* 6581 * New connections need to be floating, because 6582 * we cannot know what interfaces it will use. 6583 * That's why we pass V_pfi_all rather than kif. 6584 */ 6585 ret = pf_test_rule(&r, &sm, V_pfi_all, 6586 j->m, off, &j->pd, &ra, &rs, NULL, 6587 sizeof(j->pd.hdr.sctp)); 6588 PF_RULES_RUNLOCK(); 6589 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret); 6590 if (ret != PF_DROP && sm != NULL) { 6591 /* Inherit v_tag values. */ 6592 if (sm->direction == s->direction) { 6593 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6594 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6595 } else { 6596 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6597 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6598 } 6599 PF_STATE_UNLOCK(sm); 6600 } else { 6601 /* If we try duplicate inserts? */ 6602 break; 6603 } 6604 6605 /* Only add the address if we've actually allowed the state. */ 6606 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6607 6608 if (! do_extra) { 6609 break; 6610 } 6611 /* 6612 * We need to do this for each of our source addresses. 6613 * Find those based on the verification tag. 6614 */ 6615 struct pf_sctp_endpoint key = { 6616 .v_tag = pd->hdr.sctp.v_tag, 6617 }; 6618 struct pf_sctp_endpoint *ep; 6619 6620 PF_SCTP_ENDPOINTS_LOCK(); 6621 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6622 if (ep == NULL) { 6623 PF_SCTP_ENDPOINTS_UNLOCK(); 6624 break; 6625 } 6626 MPASS(ep != NULL); 6627 6628 TAILQ_FOREACH(i, &ep->sources, entry) { 6629 struct pf_sctp_multihome_job *nj; 6630 6631 /* SCTP can intermingle IPv4 and IPv6. */ 6632 if (i->af != pd->af) 6633 continue; 6634 6635 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6636 if (! nj) { 6637 continue; 6638 } 6639 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6640 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6641 nj->pd.src = &nj->src; 6642 // New destination address! 6643 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6644 nj->pd.dst = &nj->dst; 6645 nj->m = j->m; 6646 nj->op = j->op; 6647 6648 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6649 } 6650 PF_SCTP_ENDPOINTS_UNLOCK(); 6651 6652 break; 6653 } 6654 case SCTP_DEL_IP_ADDRESS: { 6655 struct pf_state_key_cmp key; 6656 uint8_t psrc; 6657 6658 bzero(&key, sizeof(key)); 6659 key.af = j->pd.af; 6660 key.proto = IPPROTO_SCTP; 6661 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6662 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6663 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6664 key.port[0] = j->pd.hdr.sctp.src_port; 6665 key.port[1] = j->pd.hdr.sctp.dest_port; 6666 } else { /* stack side, reverse */ 6667 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6668 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6669 key.port[1] = j->pd.hdr.sctp.src_port; 6670 key.port[0] = j->pd.hdr.sctp.dest_port; 6671 } 6672 6673 sm = pf_find_state(kif, &key, j->pd.dir); 6674 if (sm != NULL) { 6675 PF_STATE_LOCK_ASSERT(sm); 6676 if (j->pd.dir == sm->direction) { 6677 psrc = PF_PEER_SRC; 6678 } else { 6679 psrc = PF_PEER_DST; 6680 } 6681 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6682 sm->timeout = PFTM_SCTP_CLOSING; 6683 PF_STATE_UNLOCK(sm); 6684 } 6685 break; 6686 default: 6687 panic("Unknown op %#x", j->op); 6688 } 6689 } 6690 6691 free: 6692 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6693 free(j, M_PFTEMP); 6694 } 6695 6696 /* We may have inserted extra work while processing the list. */ 6697 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6698 do_extra = false; 6699 goto again; 6700 } 6701 } 6702 6703 static int 6704 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6705 struct pfi_kkif *kif, int op) 6706 { 6707 int off = 0; 6708 struct pf_sctp_multihome_job *job; 6709 6710 while (off < len) { 6711 struct sctp_paramhdr h; 6712 6713 if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL, 6714 pd->af)) 6715 return (PF_DROP); 6716 6717 /* Parameters are at least 4 bytes. */ 6718 if (ntohs(h.param_length) < 4) 6719 return (PF_DROP); 6720 6721 switch (ntohs(h.param_type)) { 6722 case SCTP_IPV4_ADDRESS: { 6723 struct in_addr t; 6724 6725 if (ntohs(h.param_length) != 6726 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6727 return (PF_DROP); 6728 6729 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6730 NULL, NULL, pd->af)) 6731 return (PF_DROP); 6732 6733 if (in_nullhost(t)) 6734 t.s_addr = pd->src->v4.s_addr; 6735 6736 /* 6737 * We hold the state lock (idhash) here, which means 6738 * that we can't acquire the keyhash, or we'll get a 6739 * LOR (and potentially double-lock things too). We also 6740 * can't release the state lock here, so instead we'll 6741 * enqueue this for async handling. 6742 * There's a relatively small race here, in that a 6743 * packet using the new addresses could arrive already, 6744 * but that's just though luck for it. 6745 */ 6746 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6747 if (! job) 6748 return (PF_DROP); 6749 6750 memcpy(&job->pd, pd, sizeof(*pd)); 6751 6752 // New source address! 6753 memcpy(&job->src, &t, sizeof(t)); 6754 job->pd.src = &job->src; 6755 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6756 job->pd.dst = &job->dst; 6757 job->m = m; 6758 job->op = op; 6759 6760 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6761 break; 6762 } 6763 #ifdef INET6 6764 case SCTP_IPV6_ADDRESS: { 6765 struct in6_addr t; 6766 6767 if (ntohs(h.param_length) != 6768 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6769 return (PF_DROP); 6770 6771 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6772 NULL, NULL, pd->af)) 6773 return (PF_DROP); 6774 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6775 break; 6776 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6777 memcpy(&t, &pd->src->v6, sizeof(t)); 6778 6779 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6780 if (! job) 6781 return (PF_DROP); 6782 6783 memcpy(&job->pd, pd, sizeof(*pd)); 6784 memcpy(&job->src, &t, sizeof(t)); 6785 job->pd.src = &job->src; 6786 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6787 job->pd.dst = &job->dst; 6788 job->m = m; 6789 job->op = op; 6790 6791 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6792 break; 6793 } 6794 #endif 6795 case SCTP_ADD_IP_ADDRESS: { 6796 int ret; 6797 struct sctp_asconf_paramhdr ah; 6798 6799 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6800 NULL, NULL, pd->af)) 6801 return (PF_DROP); 6802 6803 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6804 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6805 SCTP_ADD_IP_ADDRESS); 6806 if (ret != PF_PASS) 6807 return (ret); 6808 break; 6809 } 6810 case SCTP_DEL_IP_ADDRESS: { 6811 int ret; 6812 struct sctp_asconf_paramhdr ah; 6813 6814 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6815 NULL, NULL, pd->af)) 6816 return (PF_DROP); 6817 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6818 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6819 SCTP_DEL_IP_ADDRESS); 6820 if (ret != PF_PASS) 6821 return (ret); 6822 break; 6823 } 6824 default: 6825 break; 6826 } 6827 6828 off += roundup(ntohs(h.param_length), 4); 6829 } 6830 6831 return (PF_PASS); 6832 } 6833 int 6834 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6835 struct pfi_kkif *kif) 6836 { 6837 start += sizeof(struct sctp_init_chunk); 6838 len -= sizeof(struct sctp_init_chunk); 6839 6840 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6841 } 6842 6843 int 6844 pf_multihome_scan_asconf(struct mbuf *m, int start, int len, 6845 struct pf_pdesc *pd, struct pfi_kkif *kif) 6846 { 6847 start += sizeof(struct sctp_asconf_chunk); 6848 len -= sizeof(struct sctp_asconf_chunk); 6849 6850 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6851 } 6852 6853 int 6854 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 6855 struct pf_kstate **state, struct mbuf *m, int off, int direction, 6856 struct pfi_kkif *kif, u_int16_t icmpid, u_int16_t type, int icmp_dir, 6857 int *iidx, int multi, int inner) 6858 { 6859 key->af = pd->af; 6860 key->proto = pd->proto; 6861 if (icmp_dir == PF_IN) { 6862 *iidx = pd->sidx; 6863 key->port[pd->sidx] = icmpid; 6864 key->port[pd->didx] = type; 6865 } else { 6866 *iidx = pd->didx; 6867 key->port[pd->sidx] = type; 6868 key->port[pd->didx] = icmpid; 6869 } 6870 if (pf_state_key_addr_setup(pd, m, off, key, pd->sidx, pd->src, 6871 pd->didx, pd->dst, multi)) 6872 return (PF_DROP); 6873 6874 STATE_LOOKUP(kif, key, *state, pd); 6875 6876 if ((*state)->state_flags & PFSTATE_SLOPPY) 6877 return (-1); 6878 6879 /* Is this ICMP message flowing in right direction? */ 6880 if ((*state)->rule->type && 6881 (((!inner && (*state)->direction == direction) || 6882 (inner && (*state)->direction != direction)) ? 6883 PF_IN : PF_OUT) != icmp_dir) { 6884 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6885 printf("pf: icmp type %d in wrong direction (%d): ", 6886 ntohs(type), icmp_dir); 6887 pf_print_state(*state); 6888 printf("\n"); 6889 } 6890 PF_STATE_UNLOCK(*state); 6891 *state = NULL; 6892 return (PF_DROP); 6893 } 6894 return (-1); 6895 } 6896 6897 static int 6898 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif, 6899 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason) 6900 { 6901 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6902 u_int16_t *icmpsum, virtual_id, virtual_type; 6903 u_int8_t icmptype, icmpcode; 6904 int icmp_dir, iidx, ret, multi; 6905 struct pf_state_key_cmp key; 6906 #ifdef INET 6907 u_int16_t icmpid; 6908 #endif 6909 6910 MPASS(*state == NULL); 6911 6912 bzero(&key, sizeof(key)); 6913 switch (pd->proto) { 6914 #ifdef INET 6915 case IPPROTO_ICMP: 6916 icmptype = pd->hdr.icmp.icmp_type; 6917 icmpcode = pd->hdr.icmp.icmp_code; 6918 icmpid = pd->hdr.icmp.icmp_id; 6919 icmpsum = &pd->hdr.icmp.icmp_cksum; 6920 break; 6921 #endif /* INET */ 6922 #ifdef INET6 6923 case IPPROTO_ICMPV6: 6924 icmptype = pd->hdr.icmp6.icmp6_type; 6925 icmpcode = pd->hdr.icmp6.icmp6_code; 6926 #ifdef INET 6927 icmpid = pd->hdr.icmp6.icmp6_id; 6928 #endif 6929 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 6930 break; 6931 #endif /* INET6 */ 6932 } 6933 6934 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi, 6935 &virtual_id, &virtual_type) == 0) { 6936 /* 6937 * ICMP query/reply message not related to a TCP/UDP packet. 6938 * Search for an ICMP state. 6939 */ 6940 ret = pf_icmp_state_lookup(&key, pd, state, m, off, pd->dir, 6941 kif, virtual_id, virtual_type, icmp_dir, &iidx, 6942 PF_ICMP_MULTI_NONE, 0); 6943 if (ret >= 0) { 6944 MPASS(*state == NULL); 6945 if (ret == PF_DROP && pd->af == AF_INET6 && 6946 icmp_dir == PF_OUT) { 6947 ret = pf_icmp_state_lookup(&key, pd, state, m, off, 6948 pd->dir, kif, virtual_id, virtual_type, 6949 icmp_dir, &iidx, multi, 0); 6950 if (ret >= 0) { 6951 MPASS(*state == NULL); 6952 return (ret); 6953 } 6954 } else 6955 return (ret); 6956 } 6957 6958 (*state)->expire = pf_get_uptime(); 6959 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 6960 6961 /* translate source/destination address, if necessary */ 6962 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6963 struct pf_state_key *nk = (*state)->key[pd->didx]; 6964 6965 switch (pd->af) { 6966 #ifdef INET 6967 case AF_INET: 6968 if (PF_ANEQ(pd->src, 6969 &nk->addr[pd->sidx], AF_INET)) 6970 pf_change_a(&saddr->v4.s_addr, 6971 pd->ip_sum, 6972 nk->addr[pd->sidx].v4.s_addr, 0); 6973 6974 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 6975 AF_INET)) 6976 pf_change_a(&daddr->v4.s_addr, 6977 pd->ip_sum, 6978 nk->addr[pd->didx].v4.s_addr, 0); 6979 6980 if (nk->port[iidx] != 6981 pd->hdr.icmp.icmp_id) { 6982 pd->hdr.icmp.icmp_cksum = 6983 pf_cksum_fixup( 6984 pd->hdr.icmp.icmp_cksum, icmpid, 6985 nk->port[iidx], 0); 6986 pd->hdr.icmp.icmp_id = 6987 nk->port[iidx]; 6988 } 6989 6990 m_copyback(m, off, ICMP_MINLEN, 6991 (caddr_t )&pd->hdr.icmp); 6992 break; 6993 #endif /* INET */ 6994 #ifdef INET6 6995 case AF_INET6: 6996 if (PF_ANEQ(pd->src, 6997 &nk->addr[pd->sidx], AF_INET6)) 6998 pf_change_a6(saddr, 6999 &pd->hdr.icmp6.icmp6_cksum, 7000 &nk->addr[pd->sidx], 0); 7001 7002 if (PF_ANEQ(pd->dst, 7003 &nk->addr[pd->didx], AF_INET6)) 7004 pf_change_a6(daddr, 7005 &pd->hdr.icmp6.icmp6_cksum, 7006 &nk->addr[pd->didx], 0); 7007 7008 m_copyback(m, off, sizeof(struct icmp6_hdr), 7009 (caddr_t )&pd->hdr.icmp6); 7010 break; 7011 #endif /* INET6 */ 7012 } 7013 } 7014 return (PF_PASS); 7015 7016 } else { 7017 /* 7018 * ICMP error message in response to a TCP/UDP packet. 7019 * Extract the inner TCP/UDP header and search for that state. 7020 */ 7021 7022 struct pf_pdesc pd2; 7023 bzero(&pd2, sizeof pd2); 7024 #ifdef INET 7025 struct ip h2; 7026 #endif /* INET */ 7027 #ifdef INET6 7028 struct ip6_hdr h2_6; 7029 int terminal = 0; 7030 #endif /* INET6 */ 7031 int ipoff2 = 0; 7032 int off2 = 0; 7033 7034 pd2.af = pd->af; 7035 pd2.dir = pd->dir; 7036 /* Payload packet is from the opposite direction. */ 7037 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7038 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7039 switch (pd->af) { 7040 #ifdef INET 7041 case AF_INET: 7042 /* offset of h2 in mbuf chain */ 7043 ipoff2 = off + ICMP_MINLEN; 7044 7045 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 7046 NULL, reason, pd2.af)) { 7047 DPFPRINTF(PF_DEBUG_MISC, 7048 ("pf: ICMP error message too short " 7049 "(ip)\n")); 7050 return (PF_DROP); 7051 } 7052 /* 7053 * ICMP error messages don't refer to non-first 7054 * fragments 7055 */ 7056 if (h2.ip_off & htons(IP_OFFMASK)) { 7057 REASON_SET(reason, PFRES_FRAG); 7058 return (PF_DROP); 7059 } 7060 7061 /* offset of protocol header that follows h2 */ 7062 off2 = ipoff2 + (h2.ip_hl << 2); 7063 7064 pd2.proto = h2.ip_p; 7065 pd2.src = (struct pf_addr *)&h2.ip_src; 7066 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7067 pd2.ip_sum = &h2.ip_sum; 7068 break; 7069 #endif /* INET */ 7070 #ifdef INET6 7071 case AF_INET6: 7072 ipoff2 = off + sizeof(struct icmp6_hdr); 7073 7074 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 7075 NULL, reason, pd2.af)) { 7076 DPFPRINTF(PF_DEBUG_MISC, 7077 ("pf: ICMP error message too short " 7078 "(ip6)\n")); 7079 return (PF_DROP); 7080 } 7081 pd2.proto = h2_6.ip6_nxt; 7082 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7083 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7084 pd2.ip_sum = NULL; 7085 off2 = ipoff2 + sizeof(h2_6); 7086 do { 7087 switch (pd2.proto) { 7088 case IPPROTO_FRAGMENT: 7089 /* 7090 * ICMPv6 error messages for 7091 * non-first fragments 7092 */ 7093 REASON_SET(reason, PFRES_FRAG); 7094 return (PF_DROP); 7095 case IPPROTO_AH: 7096 case IPPROTO_HOPOPTS: 7097 case IPPROTO_ROUTING: 7098 case IPPROTO_DSTOPTS: { 7099 /* get next header and header length */ 7100 struct ip6_ext opt6; 7101 7102 if (!pf_pull_hdr(m, off2, &opt6, 7103 sizeof(opt6), NULL, reason, 7104 pd2.af)) { 7105 DPFPRINTF(PF_DEBUG_MISC, 7106 ("pf: ICMPv6 short opt\n")); 7107 return (PF_DROP); 7108 } 7109 if (pd2.proto == IPPROTO_AH) 7110 off2 += (opt6.ip6e_len + 2) * 4; 7111 else 7112 off2 += (opt6.ip6e_len + 1) * 8; 7113 pd2.proto = opt6.ip6e_nxt; 7114 /* goto the next header */ 7115 break; 7116 } 7117 default: 7118 terminal++; 7119 break; 7120 } 7121 } while (!terminal); 7122 break; 7123 #endif /* INET6 */ 7124 } 7125 7126 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 7127 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7128 printf("pf: BAD ICMP %d:%d outer dst: ", 7129 icmptype, icmpcode); 7130 pf_print_host(pd->src, 0, pd->af); 7131 printf(" -> "); 7132 pf_print_host(pd->dst, 0, pd->af); 7133 printf(" inner src: "); 7134 pf_print_host(pd2.src, 0, pd2.af); 7135 printf(" -> "); 7136 pf_print_host(pd2.dst, 0, pd2.af); 7137 printf("\n"); 7138 } 7139 REASON_SET(reason, PFRES_BADSTATE); 7140 return (PF_DROP); 7141 } 7142 7143 switch (pd2.proto) { 7144 case IPPROTO_TCP: { 7145 struct tcphdr th; 7146 u_int32_t seq; 7147 struct pf_state_peer *src, *dst; 7148 u_int8_t dws; 7149 int copyback = 0; 7150 7151 /* 7152 * Only the first 8 bytes of the TCP header can be 7153 * expected. Don't access any TCP header fields after 7154 * th_seq, an ackskew test is not possible. 7155 */ 7156 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 7157 pd2.af)) { 7158 DPFPRINTF(PF_DEBUG_MISC, 7159 ("pf: ICMP error message too short " 7160 "(tcp)\n")); 7161 return (PF_DROP); 7162 } 7163 7164 key.af = pd2.af; 7165 key.proto = IPPROTO_TCP; 7166 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7167 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7168 key.port[pd2.sidx] = th.th_sport; 7169 key.port[pd2.didx] = th.th_dport; 7170 7171 STATE_LOOKUP(kif, &key, *state, pd); 7172 7173 if (pd->dir == (*state)->direction) { 7174 src = &(*state)->dst; 7175 dst = &(*state)->src; 7176 } else { 7177 src = &(*state)->src; 7178 dst = &(*state)->dst; 7179 } 7180 7181 if (src->wscale && dst->wscale) 7182 dws = dst->wscale & PF_WSCALE_MASK; 7183 else 7184 dws = 0; 7185 7186 /* Demodulate sequence number */ 7187 seq = ntohl(th.th_seq) - src->seqdiff; 7188 if (src->seqdiff) { 7189 pf_change_a(&th.th_seq, icmpsum, 7190 htonl(seq), 0); 7191 copyback = 1; 7192 } 7193 7194 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7195 (!SEQ_GEQ(src->seqhi, seq) || 7196 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7197 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7198 printf("pf: BAD ICMP %d:%d ", 7199 icmptype, icmpcode); 7200 pf_print_host(pd->src, 0, pd->af); 7201 printf(" -> "); 7202 pf_print_host(pd->dst, 0, pd->af); 7203 printf(" state: "); 7204 pf_print_state(*state); 7205 printf(" seq=%u\n", seq); 7206 } 7207 REASON_SET(reason, PFRES_BADSTATE); 7208 return (PF_DROP); 7209 } else { 7210 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7211 printf("pf: OK ICMP %d:%d ", 7212 icmptype, icmpcode); 7213 pf_print_host(pd->src, 0, pd->af); 7214 printf(" -> "); 7215 pf_print_host(pd->dst, 0, pd->af); 7216 printf(" state: "); 7217 pf_print_state(*state); 7218 printf(" seq=%u\n", seq); 7219 } 7220 } 7221 7222 /* translate source/destination address, if necessary */ 7223 if ((*state)->key[PF_SK_WIRE] != 7224 (*state)->key[PF_SK_STACK]) { 7225 struct pf_state_key *nk = 7226 (*state)->key[pd->didx]; 7227 7228 if (PF_ANEQ(pd2.src, 7229 &nk->addr[pd2.sidx], pd2.af) || 7230 nk->port[pd2.sidx] != th.th_sport) 7231 pf_change_icmp(pd2.src, &th.th_sport, 7232 daddr, &nk->addr[pd2.sidx], 7233 nk->port[pd2.sidx], NULL, 7234 pd2.ip_sum, icmpsum, 7235 pd->ip_sum, 0, pd2.af); 7236 7237 if (PF_ANEQ(pd2.dst, 7238 &nk->addr[pd2.didx], pd2.af) || 7239 nk->port[pd2.didx] != th.th_dport) 7240 pf_change_icmp(pd2.dst, &th.th_dport, 7241 saddr, &nk->addr[pd2.didx], 7242 nk->port[pd2.didx], NULL, 7243 pd2.ip_sum, icmpsum, 7244 pd->ip_sum, 0, pd2.af); 7245 copyback = 1; 7246 } 7247 7248 if (copyback) { 7249 switch (pd2.af) { 7250 #ifdef INET 7251 case AF_INET: 7252 m_copyback(m, off, ICMP_MINLEN, 7253 (caddr_t )&pd->hdr.icmp); 7254 m_copyback(m, ipoff2, sizeof(h2), 7255 (caddr_t )&h2); 7256 break; 7257 #endif /* INET */ 7258 #ifdef INET6 7259 case AF_INET6: 7260 m_copyback(m, off, 7261 sizeof(struct icmp6_hdr), 7262 (caddr_t )&pd->hdr.icmp6); 7263 m_copyback(m, ipoff2, sizeof(h2_6), 7264 (caddr_t )&h2_6); 7265 break; 7266 #endif /* INET6 */ 7267 } 7268 m_copyback(m, off2, 8, (caddr_t)&th); 7269 } 7270 7271 return (PF_PASS); 7272 break; 7273 } 7274 case IPPROTO_UDP: { 7275 struct udphdr uh; 7276 7277 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 7278 NULL, reason, pd2.af)) { 7279 DPFPRINTF(PF_DEBUG_MISC, 7280 ("pf: ICMP error message too short " 7281 "(udp)\n")); 7282 return (PF_DROP); 7283 } 7284 7285 key.af = pd2.af; 7286 key.proto = IPPROTO_UDP; 7287 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7288 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7289 key.port[pd2.sidx] = uh.uh_sport; 7290 key.port[pd2.didx] = uh.uh_dport; 7291 7292 STATE_LOOKUP(kif, &key, *state, pd); 7293 7294 /* translate source/destination address, if necessary */ 7295 if ((*state)->key[PF_SK_WIRE] != 7296 (*state)->key[PF_SK_STACK]) { 7297 struct pf_state_key *nk = 7298 (*state)->key[pd->didx]; 7299 7300 if (PF_ANEQ(pd2.src, 7301 &nk->addr[pd2.sidx], pd2.af) || 7302 nk->port[pd2.sidx] != uh.uh_sport) 7303 pf_change_icmp(pd2.src, &uh.uh_sport, 7304 daddr, &nk->addr[pd2.sidx], 7305 nk->port[pd2.sidx], &uh.uh_sum, 7306 pd2.ip_sum, icmpsum, 7307 pd->ip_sum, 1, pd2.af); 7308 7309 if (PF_ANEQ(pd2.dst, 7310 &nk->addr[pd2.didx], pd2.af) || 7311 nk->port[pd2.didx] != uh.uh_dport) 7312 pf_change_icmp(pd2.dst, &uh.uh_dport, 7313 saddr, &nk->addr[pd2.didx], 7314 nk->port[pd2.didx], &uh.uh_sum, 7315 pd2.ip_sum, icmpsum, 7316 pd->ip_sum, 1, pd2.af); 7317 7318 switch (pd2.af) { 7319 #ifdef INET 7320 case AF_INET: 7321 m_copyback(m, off, ICMP_MINLEN, 7322 (caddr_t )&pd->hdr.icmp); 7323 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7324 break; 7325 #endif /* INET */ 7326 #ifdef INET6 7327 case AF_INET6: 7328 m_copyback(m, off, 7329 sizeof(struct icmp6_hdr), 7330 (caddr_t )&pd->hdr.icmp6); 7331 m_copyback(m, ipoff2, sizeof(h2_6), 7332 (caddr_t )&h2_6); 7333 break; 7334 #endif /* INET6 */ 7335 } 7336 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 7337 } 7338 return (PF_PASS); 7339 break; 7340 } 7341 #ifdef INET 7342 case IPPROTO_ICMP: { 7343 struct icmp *iih = &pd2.hdr.icmp; 7344 7345 if (!pf_pull_hdr(m, off2, iih, ICMP_MINLEN, 7346 NULL, reason, pd2.af)) { 7347 DPFPRINTF(PF_DEBUG_MISC, 7348 ("pf: ICMP error message too short i" 7349 "(icmp)\n")); 7350 return (PF_DROP); 7351 } 7352 7353 icmpid = iih->icmp_id; 7354 pf_icmp_mapping(&pd2, iih->icmp_type, 7355 &icmp_dir, &multi, &virtual_id, &virtual_type); 7356 7357 ret = pf_icmp_state_lookup(&key, &pd2, state, m, off, 7358 pd2.dir, kif, virtual_id, virtual_type, 7359 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7360 if (ret >= 0) { 7361 MPASS(*state == NULL); 7362 return (ret); 7363 } 7364 7365 /* translate source/destination address, if necessary */ 7366 if ((*state)->key[PF_SK_WIRE] != 7367 (*state)->key[PF_SK_STACK]) { 7368 struct pf_state_key *nk = 7369 (*state)->key[pd->didx]; 7370 7371 if (PF_ANEQ(pd2.src, 7372 &nk->addr[pd2.sidx], pd2.af) || 7373 (virtual_type == htons(ICMP_ECHO) && 7374 nk->port[iidx] != iih->icmp_id)) 7375 pf_change_icmp(pd2.src, 7376 (virtual_type == htons(ICMP_ECHO)) ? 7377 &iih->icmp_id : NULL, 7378 daddr, &nk->addr[pd2.sidx], 7379 (virtual_type == htons(ICMP_ECHO)) ? 7380 nk->port[iidx] : 0, NULL, 7381 pd2.ip_sum, icmpsum, 7382 pd->ip_sum, 0, AF_INET); 7383 7384 if (PF_ANEQ(pd2.dst, 7385 &nk->addr[pd2.didx], pd2.af)) 7386 pf_change_icmp(pd2.dst, NULL, NULL, 7387 &nk->addr[pd2.didx], 0, NULL, 7388 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 7389 AF_INET); 7390 7391 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 7392 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7393 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)iih); 7394 } 7395 return (PF_PASS); 7396 break; 7397 } 7398 #endif /* INET */ 7399 #ifdef INET6 7400 case IPPROTO_ICMPV6: { 7401 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 7402 7403 if (!pf_pull_hdr(m, off2, iih, 7404 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 7405 DPFPRINTF(PF_DEBUG_MISC, 7406 ("pf: ICMP error message too short " 7407 "(icmp6)\n")); 7408 return (PF_DROP); 7409 } 7410 7411 pf_icmp_mapping(&pd2, iih->icmp6_type, 7412 &icmp_dir, &multi, &virtual_id, &virtual_type); 7413 7414 ret = pf_icmp_state_lookup(&key, &pd2, state, m, off, 7415 pd->dir, kif, virtual_id, virtual_type, 7416 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7417 if (ret >= 0) { 7418 MPASS(*state == NULL); 7419 if (ret == PF_DROP && pd2.af == AF_INET6 && 7420 icmp_dir == PF_OUT) { 7421 ret = pf_icmp_state_lookup(&key, &pd2, 7422 state, m, off, pd->dir, kif, 7423 virtual_id, virtual_type, 7424 icmp_dir, &iidx, multi, 1); 7425 if (ret >= 0) { 7426 MPASS(*state == NULL); 7427 return (ret); 7428 } 7429 } else 7430 return (ret); 7431 } 7432 7433 /* translate source/destination address, if necessary */ 7434 if ((*state)->key[PF_SK_WIRE] != 7435 (*state)->key[PF_SK_STACK]) { 7436 struct pf_state_key *nk = 7437 (*state)->key[pd->didx]; 7438 7439 if (PF_ANEQ(pd2.src, 7440 &nk->addr[pd2.sidx], pd2.af) || 7441 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 7442 nk->port[pd2.sidx] != iih->icmp6_id)) 7443 pf_change_icmp(pd2.src, 7444 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7445 ? &iih->icmp6_id : NULL, 7446 daddr, &nk->addr[pd2.sidx], 7447 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7448 ? nk->port[iidx] : 0, NULL, 7449 pd2.ip_sum, icmpsum, 7450 pd->ip_sum, 0, AF_INET6); 7451 7452 if (PF_ANEQ(pd2.dst, 7453 &nk->addr[pd2.didx], pd2.af)) 7454 pf_change_icmp(pd2.dst, NULL, NULL, 7455 &nk->addr[pd2.didx], 0, NULL, 7456 pd2.ip_sum, icmpsum, 7457 pd->ip_sum, 0, AF_INET6); 7458 7459 m_copyback(m, off, sizeof(struct icmp6_hdr), 7460 (caddr_t)&pd->hdr.icmp6); 7461 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 7462 m_copyback(m, off2, sizeof(struct icmp6_hdr), 7463 (caddr_t)iih); 7464 } 7465 return (PF_PASS); 7466 break; 7467 } 7468 #endif /* INET6 */ 7469 default: { 7470 key.af = pd2.af; 7471 key.proto = pd2.proto; 7472 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7473 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7474 key.port[0] = key.port[1] = 0; 7475 7476 STATE_LOOKUP(kif, &key, *state, pd); 7477 7478 /* translate source/destination address, if necessary */ 7479 if ((*state)->key[PF_SK_WIRE] != 7480 (*state)->key[PF_SK_STACK]) { 7481 struct pf_state_key *nk = 7482 (*state)->key[pd->didx]; 7483 7484 if (PF_ANEQ(pd2.src, 7485 &nk->addr[pd2.sidx], pd2.af)) 7486 pf_change_icmp(pd2.src, NULL, daddr, 7487 &nk->addr[pd2.sidx], 0, NULL, 7488 pd2.ip_sum, icmpsum, 7489 pd->ip_sum, 0, pd2.af); 7490 7491 if (PF_ANEQ(pd2.dst, 7492 &nk->addr[pd2.didx], pd2.af)) 7493 pf_change_icmp(pd2.dst, NULL, saddr, 7494 &nk->addr[pd2.didx], 0, NULL, 7495 pd2.ip_sum, icmpsum, 7496 pd->ip_sum, 0, pd2.af); 7497 7498 switch (pd2.af) { 7499 #ifdef INET 7500 case AF_INET: 7501 m_copyback(m, off, ICMP_MINLEN, 7502 (caddr_t)&pd->hdr.icmp); 7503 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7504 break; 7505 #endif /* INET */ 7506 #ifdef INET6 7507 case AF_INET6: 7508 m_copyback(m, off, 7509 sizeof(struct icmp6_hdr), 7510 (caddr_t )&pd->hdr.icmp6); 7511 m_copyback(m, ipoff2, sizeof(h2_6), 7512 (caddr_t )&h2_6); 7513 break; 7514 #endif /* INET6 */ 7515 } 7516 } 7517 return (PF_PASS); 7518 break; 7519 } 7520 } 7521 } 7522 } 7523 7524 static int 7525 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif, 7526 struct mbuf *m, struct pf_pdesc *pd) 7527 { 7528 struct pf_state_peer *src, *dst; 7529 struct pf_state_key_cmp key; 7530 uint8_t psrc, pdst; 7531 7532 bzero(&key, sizeof(key)); 7533 key.af = pd->af; 7534 key.proto = pd->proto; 7535 if (pd->dir == PF_IN) { 7536 PF_ACPY(&key.addr[0], pd->src, key.af); 7537 PF_ACPY(&key.addr[1], pd->dst, key.af); 7538 key.port[0] = key.port[1] = 0; 7539 } else { 7540 PF_ACPY(&key.addr[1], pd->src, key.af); 7541 PF_ACPY(&key.addr[0], pd->dst, key.af); 7542 key.port[1] = key.port[0] = 0; 7543 } 7544 7545 STATE_LOOKUP(kif, &key, *state, pd); 7546 7547 if (pd->dir == (*state)->direction) { 7548 src = &(*state)->src; 7549 dst = &(*state)->dst; 7550 psrc = PF_PEER_SRC; 7551 pdst = PF_PEER_DST; 7552 } else { 7553 src = &(*state)->dst; 7554 dst = &(*state)->src; 7555 psrc = PF_PEER_DST; 7556 pdst = PF_PEER_SRC; 7557 } 7558 7559 /* update states */ 7560 if (src->state < PFOTHERS_SINGLE) 7561 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7562 if (dst->state == PFOTHERS_SINGLE) 7563 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7564 7565 /* update expire time */ 7566 (*state)->expire = pf_get_uptime(); 7567 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7568 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7569 else 7570 (*state)->timeout = PFTM_OTHER_SINGLE; 7571 7572 /* translate source/destination address, if necessary */ 7573 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7574 struct pf_state_key *nk = (*state)->key[pd->didx]; 7575 7576 KASSERT(nk, ("%s: nk is null", __func__)); 7577 KASSERT(pd, ("%s: pd is null", __func__)); 7578 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7579 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7580 switch (pd->af) { 7581 #ifdef INET 7582 case AF_INET: 7583 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7584 pf_change_a(&pd->src->v4.s_addr, 7585 pd->ip_sum, 7586 nk->addr[pd->sidx].v4.s_addr, 7587 0); 7588 7589 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7590 pf_change_a(&pd->dst->v4.s_addr, 7591 pd->ip_sum, 7592 nk->addr[pd->didx].v4.s_addr, 7593 0); 7594 7595 break; 7596 #endif /* INET */ 7597 #ifdef INET6 7598 case AF_INET6: 7599 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6)) 7600 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7601 7602 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6)) 7603 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7604 #endif /* INET6 */ 7605 } 7606 } 7607 return (PF_PASS); 7608 } 7609 7610 /* 7611 * ipoff and off are measured from the start of the mbuf chain. 7612 * h must be at "ipoff" on the mbuf chain. 7613 */ 7614 void * 7615 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 7616 u_short *actionp, u_short *reasonp, sa_family_t af) 7617 { 7618 switch (af) { 7619 #ifdef INET 7620 case AF_INET: { 7621 const struct ip *h = mtod(m, struct ip *); 7622 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7623 7624 if (fragoff) { 7625 if (fragoff >= len) 7626 ACTION_SET(actionp, PF_PASS); 7627 else { 7628 ACTION_SET(actionp, PF_DROP); 7629 REASON_SET(reasonp, PFRES_FRAG); 7630 } 7631 return (NULL); 7632 } 7633 if (m->m_pkthdr.len < off + len || 7634 ntohs(h->ip_len) < off + len) { 7635 ACTION_SET(actionp, PF_DROP); 7636 REASON_SET(reasonp, PFRES_SHORT); 7637 return (NULL); 7638 } 7639 break; 7640 } 7641 #endif /* INET */ 7642 #ifdef INET6 7643 case AF_INET6: { 7644 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7645 7646 if (m->m_pkthdr.len < off + len || 7647 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7648 (unsigned)(off + len)) { 7649 ACTION_SET(actionp, PF_DROP); 7650 REASON_SET(reasonp, PFRES_SHORT); 7651 return (NULL); 7652 } 7653 break; 7654 } 7655 #endif /* INET6 */ 7656 } 7657 m_copydata(m, off, len, p); 7658 return (p); 7659 } 7660 7661 int 7662 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7663 int rtableid) 7664 { 7665 struct ifnet *ifp; 7666 7667 /* 7668 * Skip check for addresses with embedded interface scope, 7669 * as they would always match anyway. 7670 */ 7671 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7672 return (1); 7673 7674 if (af != AF_INET && af != AF_INET6) 7675 return (0); 7676 7677 if (kif == V_pfi_all) 7678 return (1); 7679 7680 /* Skip checks for ipsec interfaces */ 7681 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7682 return (1); 7683 7684 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7685 7686 switch (af) { 7687 #ifdef INET6 7688 case AF_INET6: 7689 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7690 ifp)); 7691 #endif 7692 #ifdef INET 7693 case AF_INET: 7694 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7695 ifp)); 7696 #endif 7697 } 7698 7699 return (0); 7700 } 7701 7702 #ifdef INET 7703 static void 7704 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7705 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7706 { 7707 struct mbuf *m0, *m1, *md; 7708 struct sockaddr_in dst; 7709 struct ip *ip; 7710 struct pfi_kkif *nkif = NULL; 7711 struct ifnet *ifp = NULL; 7712 struct pf_addr naddr; 7713 int error = 0; 7714 uint16_t ip_len, ip_off; 7715 uint16_t tmp; 7716 int r_rt, r_dir; 7717 7718 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7719 7720 if (s) { 7721 r_rt = s->rt; 7722 r_dir = s->direction; 7723 } else { 7724 r_rt = r->rt; 7725 r_dir = r->direction; 7726 } 7727 7728 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7729 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7730 __func__)); 7731 7732 if ((pd->pf_mtag == NULL && 7733 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7734 pd->pf_mtag->routed++ > 3) { 7735 m0 = *m; 7736 *m = NULL; 7737 goto bad_locked; 7738 } 7739 7740 if (r_rt == PF_DUPTO) { 7741 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7742 if (s == NULL) { 7743 ifp = r->rpool.cur->kif ? 7744 r->rpool.cur->kif->pfik_ifp : NULL; 7745 } else { 7746 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7747 /* If pfsync'd */ 7748 if (ifp == NULL && r->rpool.cur != NULL) 7749 ifp = r->rpool.cur->kif ? 7750 r->rpool.cur->kif->pfik_ifp : NULL; 7751 PF_STATE_UNLOCK(s); 7752 } 7753 if (ifp == oifp) { 7754 /* When the 2nd interface is not skipped */ 7755 return; 7756 } else { 7757 m0 = *m; 7758 *m = NULL; 7759 goto bad; 7760 } 7761 } else { 7762 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7763 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7764 if (s) 7765 PF_STATE_UNLOCK(s); 7766 return; 7767 } 7768 } 7769 } else { 7770 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7771 pf_dummynet(pd, s, r, m); 7772 if (s) 7773 PF_STATE_UNLOCK(s); 7774 return; 7775 } 7776 m0 = *m; 7777 } 7778 7779 ip = mtod(m0, struct ip *); 7780 7781 bzero(&dst, sizeof(dst)); 7782 dst.sin_family = AF_INET; 7783 dst.sin_len = sizeof(dst); 7784 dst.sin_addr = ip->ip_dst; 7785 7786 bzero(&naddr, sizeof(naddr)); 7787 7788 if (s == NULL) { 7789 if (TAILQ_EMPTY(&r->rpool.list)) { 7790 DPFPRINTF(PF_DEBUG_URGENT, 7791 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7792 goto bad_locked; 7793 } 7794 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7795 &naddr, &nkif, NULL); 7796 if (!PF_AZERO(&naddr, AF_INET)) 7797 dst.sin_addr.s_addr = naddr.v4.s_addr; 7798 ifp = nkif ? nkif->pfik_ifp : NULL; 7799 } else { 7800 struct pfi_kkif *kif; 7801 7802 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7803 dst.sin_addr.s_addr = 7804 s->rt_addr.v4.s_addr; 7805 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7806 kif = s->rt_kif; 7807 /* If pfsync'd */ 7808 if (ifp == NULL && r->rpool.cur != NULL) { 7809 ifp = r->rpool.cur->kif ? 7810 r->rpool.cur->kif->pfik_ifp : NULL; 7811 kif = r->rpool.cur->kif; 7812 } 7813 if (ifp != NULL && kif != NULL && 7814 r->rule_flag & PFRULE_IFBOUND && 7815 r->rt == PF_REPLYTO && 7816 s->kif == V_pfi_all) { 7817 s->kif = kif; 7818 s->orig_kif = oifp->if_pf_kif; 7819 } 7820 7821 PF_STATE_UNLOCK(s); 7822 } 7823 7824 if (ifp == NULL) 7825 goto bad; 7826 7827 if (pd->dir == PF_IN) { 7828 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 7829 &pd->act) != PF_PASS) 7830 goto bad; 7831 else if (m0 == NULL) 7832 goto done; 7833 if (m0->m_len < sizeof(struct ip)) { 7834 DPFPRINTF(PF_DEBUG_URGENT, 7835 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7836 goto bad; 7837 } 7838 ip = mtod(m0, struct ip *); 7839 } 7840 7841 if (ifp->if_flags & IFF_LOOPBACK) 7842 m0->m_flags |= M_SKIP_FIREWALL; 7843 7844 ip_len = ntohs(ip->ip_len); 7845 ip_off = ntohs(ip->ip_off); 7846 7847 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7848 m0->m_pkthdr.csum_flags |= CSUM_IP; 7849 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7850 in_delayed_cksum(m0); 7851 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7852 } 7853 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7854 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7855 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7856 } 7857 7858 if (pd->dir == PF_IN) { 7859 /* 7860 * Make sure dummynet gets the correct direction, in case it needs to 7861 * re-inject later. 7862 */ 7863 pd->dir = PF_OUT; 7864 7865 /* 7866 * The following processing is actually the rest of the inbound processing, even 7867 * though we've marked it as outbound (so we don't look through dummynet) and it 7868 * happens after the outbound processing (pf_test(PF_OUT) above). 7869 * Swap the dummynet pipe numbers, because it's going to come to the wrong 7870 * conclusion about what direction it's processing, and we can't fix it or it 7871 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 7872 * decision will pick the right pipe, and everything will mostly work as expected. 7873 */ 7874 tmp = pd->act.dnrpipe; 7875 pd->act.dnrpipe = pd->act.dnpipe; 7876 pd->act.dnpipe = tmp; 7877 } 7878 7879 /* 7880 * If small enough for interface, or the interface will take 7881 * care of the fragmentation for us, we can just send directly. 7882 */ 7883 if (ip_len <= ifp->if_mtu || 7884 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7885 ip->ip_sum = 0; 7886 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7887 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7888 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7889 } 7890 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7891 7892 md = m0; 7893 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7894 if (md != NULL) 7895 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7896 goto done; 7897 } 7898 7899 /* Balk when DF bit is set or the interface didn't support TSO. */ 7900 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7901 error = EMSGSIZE; 7902 KMOD_IPSTAT_INC(ips_cantfrag); 7903 if (r_rt != PF_DUPTO) { 7904 if (s && s->nat_rule != NULL) 7905 PACKET_UNDO_NAT(m0, pd, 7906 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7907 s); 7908 7909 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7910 ifp->if_mtu); 7911 goto done; 7912 } else 7913 goto bad; 7914 } 7915 7916 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7917 if (error) 7918 goto bad; 7919 7920 for (; m0; m0 = m1) { 7921 m1 = m0->m_nextpkt; 7922 m0->m_nextpkt = NULL; 7923 if (error == 0) { 7924 m_clrprotoflags(m0); 7925 md = m0; 7926 pd->pf_mtag = pf_find_mtag(md); 7927 error = pf_dummynet_route(pd, s, r, ifp, 7928 sintosa(&dst), &md); 7929 if (md != NULL) 7930 error = (*ifp->if_output)(ifp, md, 7931 sintosa(&dst), NULL); 7932 } else 7933 m_freem(m0); 7934 } 7935 7936 if (error == 0) 7937 KMOD_IPSTAT_INC(ips_fragmented); 7938 7939 done: 7940 if (r_rt != PF_DUPTO) 7941 *m = NULL; 7942 return; 7943 7944 bad_locked: 7945 if (s) 7946 PF_STATE_UNLOCK(s); 7947 bad: 7948 m_freem(m0); 7949 goto done; 7950 } 7951 #endif /* INET */ 7952 7953 #ifdef INET6 7954 static void 7955 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7956 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7957 { 7958 struct mbuf *m0, *md; 7959 struct sockaddr_in6 dst; 7960 struct ip6_hdr *ip6; 7961 struct pfi_kkif *nkif = NULL; 7962 struct ifnet *ifp = NULL; 7963 struct pf_addr naddr; 7964 int r_rt, r_dir; 7965 7966 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7967 7968 if (s) { 7969 r_rt = s->rt; 7970 r_dir = s->direction; 7971 } else { 7972 r_rt = r->rt; 7973 r_dir = r->direction; 7974 } 7975 7976 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7977 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7978 __func__)); 7979 7980 if ((pd->pf_mtag == NULL && 7981 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7982 pd->pf_mtag->routed++ > 3) { 7983 m0 = *m; 7984 *m = NULL; 7985 goto bad_locked; 7986 } 7987 7988 if (r_rt == PF_DUPTO) { 7989 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7990 if (s == NULL) { 7991 ifp = r->rpool.cur->kif ? 7992 r->rpool.cur->kif->pfik_ifp : NULL; 7993 } else { 7994 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7995 /* If pfsync'd */ 7996 if (ifp == NULL && r->rpool.cur != NULL) 7997 ifp = r->rpool.cur->kif ? 7998 r->rpool.cur->kif->pfik_ifp : NULL; 7999 PF_STATE_UNLOCK(s); 8000 } 8001 if (ifp == oifp) { 8002 /* When the 2nd interface is not skipped */ 8003 return; 8004 } else { 8005 m0 = *m; 8006 *m = NULL; 8007 goto bad; 8008 } 8009 } else { 8010 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 8011 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 8012 if (s) 8013 PF_STATE_UNLOCK(s); 8014 return; 8015 } 8016 } 8017 } else { 8018 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 8019 pf_dummynet(pd, s, r, m); 8020 if (s) 8021 PF_STATE_UNLOCK(s); 8022 return; 8023 } 8024 m0 = *m; 8025 } 8026 8027 ip6 = mtod(m0, struct ip6_hdr *); 8028 8029 bzero(&dst, sizeof(dst)); 8030 dst.sin6_family = AF_INET6; 8031 dst.sin6_len = sizeof(dst); 8032 dst.sin6_addr = ip6->ip6_dst; 8033 8034 bzero(&naddr, sizeof(naddr)); 8035 8036 if (s == NULL) { 8037 if (TAILQ_EMPTY(&r->rpool.list)) { 8038 DPFPRINTF(PF_DEBUG_URGENT, 8039 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 8040 goto bad_locked; 8041 } 8042 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 8043 &naddr, &nkif, NULL); 8044 if (!PF_AZERO(&naddr, AF_INET6)) 8045 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 8046 &naddr, AF_INET6); 8047 ifp = nkif ? nkif->pfik_ifp : NULL; 8048 } else { 8049 struct pfi_kkif *kif; 8050 8051 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 8052 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 8053 &s->rt_addr, AF_INET6); 8054 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 8055 kif = s->rt_kif; 8056 /* If pfsync'd */ 8057 if (ifp == NULL && r->rpool.cur != NULL) { 8058 ifp = r->rpool.cur->kif ? 8059 r->rpool.cur->kif->pfik_ifp : NULL; 8060 kif = r->rpool.cur->kif; 8061 } 8062 if (ifp != NULL && kif != NULL && 8063 r->rule_flag & PFRULE_IFBOUND && 8064 r->rt == PF_REPLYTO && 8065 s->kif == V_pfi_all) { 8066 s->kif = kif; 8067 s->orig_kif = oifp->if_pf_kif; 8068 } 8069 } 8070 8071 if (s) 8072 PF_STATE_UNLOCK(s); 8073 8074 if (ifp == NULL) 8075 goto bad; 8076 8077 if (pd->dir == PF_IN) { 8078 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD, ifp, &m0, inp, 8079 &pd->act) != PF_PASS) 8080 goto bad; 8081 else if (m0 == NULL) 8082 goto done; 8083 if (m0->m_len < sizeof(struct ip6_hdr)) { 8084 DPFPRINTF(PF_DEBUG_URGENT, 8085 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 8086 __func__)); 8087 goto bad; 8088 } 8089 ip6 = mtod(m0, struct ip6_hdr *); 8090 } 8091 8092 if (ifp->if_flags & IFF_LOOPBACK) 8093 m0->m_flags |= M_SKIP_FIREWALL; 8094 8095 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 8096 ~ifp->if_hwassist) { 8097 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 8098 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 8099 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 8100 } 8101 8102 /* 8103 * If the packet is too large for the outgoing interface, 8104 * send back an icmp6 error. 8105 */ 8106 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 8107 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 8108 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 8109 md = m0; 8110 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 8111 if (md != NULL) 8112 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 8113 } 8114 else { 8115 in6_ifstat_inc(ifp, ifs6_in_toobig); 8116 if (r_rt != PF_DUPTO) { 8117 if (s && s->nat_rule != NULL) 8118 PACKET_UNDO_NAT(m0, pd, 8119 ((caddr_t)ip6 - m0->m_data) + 8120 sizeof(struct ip6_hdr), s); 8121 8122 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 8123 } else 8124 goto bad; 8125 } 8126 8127 done: 8128 if (r_rt != PF_DUPTO) 8129 *m = NULL; 8130 return; 8131 8132 bad_locked: 8133 if (s) 8134 PF_STATE_UNLOCK(s); 8135 bad: 8136 m_freem(m0); 8137 goto done; 8138 } 8139 #endif /* INET6 */ 8140 8141 /* 8142 * FreeBSD supports cksum offloads for the following drivers. 8143 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 8144 * 8145 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 8146 * network driver performed cksum including pseudo header, need to verify 8147 * csum_data 8148 * CSUM_DATA_VALID : 8149 * network driver performed cksum, needs to additional pseudo header 8150 * cksum computation with partial csum_data(i.e. lack of H/W support for 8151 * pseudo header, for instance sk(4) and possibly gem(4)) 8152 * 8153 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 8154 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 8155 * TCP/UDP layer. 8156 * Also, set csum_data to 0xffff to force cksum validation. 8157 */ 8158 static int 8159 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 8160 { 8161 u_int16_t sum = 0; 8162 int hw_assist = 0; 8163 struct ip *ip; 8164 8165 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 8166 return (1); 8167 if (m->m_pkthdr.len < off + len) 8168 return (1); 8169 8170 switch (p) { 8171 case IPPROTO_TCP: 8172 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8173 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8174 sum = m->m_pkthdr.csum_data; 8175 } else { 8176 ip = mtod(m, struct ip *); 8177 sum = in_pseudo(ip->ip_src.s_addr, 8178 ip->ip_dst.s_addr, htonl((u_short)len + 8179 m->m_pkthdr.csum_data + IPPROTO_TCP)); 8180 } 8181 sum ^= 0xffff; 8182 ++hw_assist; 8183 } 8184 break; 8185 case IPPROTO_UDP: 8186 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8187 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8188 sum = m->m_pkthdr.csum_data; 8189 } else { 8190 ip = mtod(m, struct ip *); 8191 sum = in_pseudo(ip->ip_src.s_addr, 8192 ip->ip_dst.s_addr, htonl((u_short)len + 8193 m->m_pkthdr.csum_data + IPPROTO_UDP)); 8194 } 8195 sum ^= 0xffff; 8196 ++hw_assist; 8197 } 8198 break; 8199 case IPPROTO_ICMP: 8200 #ifdef INET6 8201 case IPPROTO_ICMPV6: 8202 #endif /* INET6 */ 8203 break; 8204 default: 8205 return (1); 8206 } 8207 8208 if (!hw_assist) { 8209 switch (af) { 8210 case AF_INET: 8211 if (p == IPPROTO_ICMP) { 8212 if (m->m_len < off) 8213 return (1); 8214 m->m_data += off; 8215 m->m_len -= off; 8216 sum = in_cksum(m, len); 8217 m->m_data -= off; 8218 m->m_len += off; 8219 } else { 8220 if (m->m_len < sizeof(struct ip)) 8221 return (1); 8222 sum = in4_cksum(m, p, off, len); 8223 } 8224 break; 8225 #ifdef INET6 8226 case AF_INET6: 8227 if (m->m_len < sizeof(struct ip6_hdr)) 8228 return (1); 8229 sum = in6_cksum(m, p, off, len); 8230 break; 8231 #endif /* INET6 */ 8232 default: 8233 return (1); 8234 } 8235 } 8236 if (sum) { 8237 switch (p) { 8238 case IPPROTO_TCP: 8239 { 8240 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 8241 break; 8242 } 8243 case IPPROTO_UDP: 8244 { 8245 KMOD_UDPSTAT_INC(udps_badsum); 8246 break; 8247 } 8248 #ifdef INET 8249 case IPPROTO_ICMP: 8250 { 8251 KMOD_ICMPSTAT_INC(icps_checksum); 8252 break; 8253 } 8254 #endif 8255 #ifdef INET6 8256 case IPPROTO_ICMPV6: 8257 { 8258 KMOD_ICMP6STAT_INC(icp6s_checksum); 8259 break; 8260 } 8261 #endif /* INET6 */ 8262 } 8263 return (1); 8264 } else { 8265 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 8266 m->m_pkthdr.csum_flags |= 8267 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 8268 m->m_pkthdr.csum_data = 0xffff; 8269 } 8270 } 8271 return (0); 8272 } 8273 8274 static bool 8275 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 8276 const struct pf_kstate *s, struct ip_fw_args *dnflow) 8277 { 8278 int dndir = r->direction; 8279 8280 if (s && dndir == PF_INOUT) { 8281 dndir = s->direction; 8282 } else if (dndir == PF_INOUT) { 8283 /* Assume primary direction. Happens when we've set dnpipe in 8284 * the ethernet level code. */ 8285 dndir = pd->dir; 8286 } 8287 8288 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 8289 return (false); 8290 8291 memset(dnflow, 0, sizeof(*dnflow)); 8292 8293 if (pd->dport != NULL) 8294 dnflow->f_id.dst_port = ntohs(*pd->dport); 8295 if (pd->sport != NULL) 8296 dnflow->f_id.src_port = ntohs(*pd->sport); 8297 8298 if (pd->dir == PF_IN) 8299 dnflow->flags |= IPFW_ARGS_IN; 8300 else 8301 dnflow->flags |= IPFW_ARGS_OUT; 8302 8303 if (pd->dir != dndir && pd->act.dnrpipe) { 8304 dnflow->rule.info = pd->act.dnrpipe; 8305 } 8306 else if (pd->dir == dndir && pd->act.dnpipe) { 8307 dnflow->rule.info = pd->act.dnpipe; 8308 } 8309 else { 8310 return (false); 8311 } 8312 8313 dnflow->rule.info |= IPFW_IS_DUMMYNET; 8314 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 8315 dnflow->rule.info |= IPFW_IS_PIPE; 8316 8317 dnflow->f_id.proto = pd->proto; 8318 dnflow->f_id.extra = dnflow->rule.info; 8319 switch (pd->af) { 8320 case AF_INET: 8321 dnflow->f_id.addr_type = 4; 8322 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 8323 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 8324 break; 8325 case AF_INET6: 8326 dnflow->flags |= IPFW_ARGS_IP6; 8327 dnflow->f_id.addr_type = 6; 8328 dnflow->f_id.src_ip6 = pd->src->v6; 8329 dnflow->f_id.dst_ip6 = pd->dst->v6; 8330 break; 8331 default: 8332 panic("Invalid AF"); 8333 break; 8334 } 8335 8336 return (true); 8337 } 8338 8339 int 8340 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8341 struct inpcb *inp) 8342 { 8343 struct pfi_kkif *kif; 8344 struct mbuf *m = *m0; 8345 8346 M_ASSERTPKTHDR(m); 8347 MPASS(ifp->if_vnet == curvnet); 8348 NET_EPOCH_ASSERT(); 8349 8350 if (!V_pf_status.running) 8351 return (PF_PASS); 8352 8353 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8354 8355 if (kif == NULL) { 8356 DPFPRINTF(PF_DEBUG_URGENT, 8357 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 8358 return (PF_DROP); 8359 } 8360 if (kif->pfik_flags & PFI_IFLAG_SKIP) 8361 return (PF_PASS); 8362 8363 if (m->m_flags & M_SKIP_FIREWALL) 8364 return (PF_PASS); 8365 8366 if (__predict_false(! M_WRITABLE(*m0))) { 8367 m = *m0 = m_unshare(*m0, M_NOWAIT); 8368 if (*m0 == NULL) 8369 return (PF_DROP); 8370 } 8371 8372 /* Stateless! */ 8373 return (pf_test_eth_rule(dir, kif, m0)); 8374 } 8375 8376 static __inline void 8377 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 8378 { 8379 struct m_tag *mtag; 8380 8381 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 8382 8383 /* dummynet adds this tag, but pf does not need it, 8384 * and keeping it creates unexpected behavior, 8385 * e.g. in case of divert(4) usage right after dummynet. */ 8386 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8387 if (mtag != NULL) 8388 m_tag_delete(m, mtag); 8389 } 8390 8391 static int 8392 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 8393 struct pf_krule *r, struct mbuf **m0) 8394 { 8395 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 8396 } 8397 8398 static int 8399 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 8400 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 8401 struct mbuf **m0) 8402 { 8403 NET_EPOCH_ASSERT(); 8404 8405 if (pd->act.dnpipe || pd->act.dnrpipe) { 8406 struct ip_fw_args dnflow; 8407 if (ip_dn_io_ptr == NULL) { 8408 m_freem(*m0); 8409 *m0 = NULL; 8410 return (ENOMEM); 8411 } 8412 8413 if (pd->pf_mtag == NULL && 8414 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 8415 m_freem(*m0); 8416 *m0 = NULL; 8417 return (ENOMEM); 8418 } 8419 8420 if (ifp != NULL) { 8421 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 8422 8423 pd->pf_mtag->if_index = ifp->if_index; 8424 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 8425 8426 MPASS(sa != NULL); 8427 8428 if (pd->af == AF_INET) 8429 memcpy(&pd->pf_mtag->dst, sa, 8430 sizeof(struct sockaddr_in)); 8431 else 8432 memcpy(&pd->pf_mtag->dst, sa, 8433 sizeof(struct sockaddr_in6)); 8434 } 8435 8436 if (s != NULL && s->nat_rule != NULL && 8437 s->nat_rule->action == PF_RDR && 8438 ( 8439 #ifdef INET 8440 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 8441 #endif 8442 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 8443 /* 8444 * If we're redirecting to loopback mark this packet 8445 * as being local. Otherwise it might get dropped 8446 * if dummynet re-injects. 8447 */ 8448 (*m0)->m_pkthdr.rcvif = V_loif; 8449 } 8450 8451 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 8452 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 8453 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 8454 ip_dn_io_ptr(m0, &dnflow); 8455 if (*m0 != NULL) { 8456 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8457 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 8458 } 8459 } 8460 } 8461 8462 return (0); 8463 } 8464 8465 #ifdef INET6 8466 static int 8467 pf_walk_option6(struct mbuf *m, int off, int end, uint32_t *jumbolen, 8468 u_short *reason) 8469 { 8470 struct ip6_opt opt; 8471 struct ip6_opt_jumbo jumbo; 8472 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 8473 8474 while (off < end) { 8475 if (!pf_pull_hdr(m, off, &opt.ip6o_type, sizeof(opt.ip6o_type), 8476 NULL, reason, AF_INET6)) { 8477 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt type")); 8478 return (PF_DROP); 8479 } 8480 if (opt.ip6o_type == IP6OPT_PAD1) { 8481 off++; 8482 continue; 8483 } 8484 if (!pf_pull_hdr(m, off, &opt, sizeof(opt), NULL, reason, 8485 AF_INET6)) { 8486 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt")); 8487 return (PF_DROP); 8488 } 8489 if (off + sizeof(opt) + opt.ip6o_len > end) { 8490 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 long opt")); 8491 REASON_SET(reason, PFRES_IPOPTIONS); 8492 return (PF_DROP); 8493 } 8494 switch (opt.ip6o_type) { 8495 case IP6OPT_JUMBO: 8496 if (*jumbolen != 0) { 8497 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple jumbo")); 8498 REASON_SET(reason, PFRES_IPOPTIONS); 8499 return (PF_DROP); 8500 } 8501 if (ntohs(h->ip6_plen) != 0) { 8502 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 bad jumbo plen")); 8503 REASON_SET(reason, PFRES_IPOPTIONS); 8504 return (PF_DROP); 8505 } 8506 if (!pf_pull_hdr(m, off, &jumbo, sizeof(jumbo), NULL, 8507 reason, AF_INET6)) { 8508 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbo")); 8509 return (PF_DROP); 8510 } 8511 memcpy(jumbolen, jumbo.ip6oj_jumbo_len, 8512 sizeof(*jumbolen)); 8513 *jumbolen = ntohl(*jumbolen); 8514 if (*jumbolen < IPV6_MAXPACKET) { 8515 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbolen")); 8516 REASON_SET(reason, PFRES_IPOPTIONS); 8517 return (PF_DROP); 8518 } 8519 break; 8520 default: 8521 break; 8522 } 8523 off += sizeof(opt) + opt.ip6o_len; 8524 } 8525 8526 return (PF_PASS); 8527 } 8528 8529 int 8530 pf_walk_header6(struct mbuf *m, uint8_t *nxt, int *off, int *extoff, 8531 uint32_t *jumbolen, u_short *reason) 8532 { 8533 struct ip6_ext ext; 8534 struct ip6_rthdr rthdr; 8535 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 8536 int rthdr_cnt = 0; 8537 8538 *nxt = h->ip6_nxt; 8539 *off = sizeof(struct ip6_hdr); 8540 *extoff = 0; 8541 *jumbolen = 0; 8542 for (;;) { 8543 switch (*nxt) { 8544 case IPPROTO_FRAGMENT: 8545 /* jumbo payload packets cannot be fragmented */ 8546 if (*jumbolen != 0) { 8547 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 fragmented jumbo")); 8548 REASON_SET(reason, PFRES_FRAG); 8549 return (PF_DROP); 8550 } 8551 return (PF_PASS); 8552 case IPPROTO_ROUTING: 8553 if (rthdr_cnt++) { 8554 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple rthdr")); 8555 REASON_SET(reason, PFRES_IPOPTIONS); 8556 return (PF_DROP); 8557 } 8558 if (!pf_pull_hdr(m, *off, &rthdr, sizeof(rthdr), NULL, 8559 reason, AF_INET6)) { 8560 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short rthdr")); 8561 return (PF_DROP); 8562 } 8563 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8564 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 rthdr0")); 8565 REASON_SET(reason, PFRES_IPOPTIONS); 8566 return (PF_DROP); 8567 } 8568 /* FALLTHROUGH */ 8569 case IPPROTO_AH: 8570 case IPPROTO_HOPOPTS: 8571 case IPPROTO_DSTOPTS: 8572 if (!pf_pull_hdr(m, *off, &ext, sizeof(ext), NULL, 8573 reason, AF_INET6)) { 8574 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short exthdr")); 8575 return (PF_DROP); 8576 } 8577 *extoff = *off; 8578 if (*nxt == IPPROTO_HOPOPTS) { 8579 if (pf_walk_option6(m, *off + sizeof(ext), 8580 *off + (ext.ip6e_len + 1) * 8, jumbolen, 8581 reason) != PF_PASS) 8582 return (PF_DROP); 8583 if (ntohs(h->ip6_plen) == 0 && *jumbolen != 0) { 8584 DPFPRINTF(PF_DEBUG_MISC, 8585 ("IPv6 missing jumbo")); 8586 REASON_SET(reason, PFRES_IPOPTIONS); 8587 return (PF_DROP); 8588 } 8589 } 8590 if (*nxt == IPPROTO_AH) 8591 *off += (ext.ip6e_len + 2) * 4; 8592 else 8593 *off += (ext.ip6e_len + 1) * 8; 8594 *nxt = ext.ip6e_nxt; 8595 break; 8596 default: 8597 return (PF_PASS); 8598 } 8599 } 8600 } 8601 #endif 8602 8603 static void 8604 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 8605 { 8606 memset(pd, 0, sizeof(*pd)); 8607 pd->pf_mtag = pf_find_mtag(m); 8608 } 8609 8610 static int 8611 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 8612 u_short *action, u_short *reason, struct pfi_kkif *kif, struct pf_krule **a, 8613 struct pf_krule **r, struct pf_kstate **s, struct pf_kruleset **ruleset, 8614 int *off, int *hdrlen, struct inpcb *inp, 8615 struct pf_rule_actions *default_actions) 8616 { 8617 struct mbuf *m = *m0; 8618 8619 pd->af = af; 8620 pd->dir = dir; 8621 8622 TAILQ_INIT(&pd->sctp_multihome_jobs); 8623 if (default_actions != NULL) 8624 memcpy(&pd->act, default_actions, sizeof(pd->act)); 8625 8626 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 8627 pd->act.dnpipe = pd->pf_mtag->dnpipe; 8628 pd->act.flags = pd->pf_mtag->dnflags; 8629 } 8630 8631 switch (af) { 8632 #ifdef INET 8633 case AF_INET: { 8634 struct ip *h; 8635 8636 if (__predict_false(m->m_len < sizeof(struct ip)) && 8637 (m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 8638 DPFPRINTF(PF_DEBUG_URGENT, 8639 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 8640 *action = PF_DROP; 8641 REASON_SET(reason, PFRES_SHORT); 8642 return (-1); 8643 } 8644 8645 if (pf_normalize_ip(m0, kif, reason, pd) != PF_PASS) { 8646 /* We do IP header normalization and packet reassembly here */ 8647 *action = PF_DROP; 8648 return (-1); 8649 } 8650 m = *m0; 8651 8652 h = mtod(m, struct ip *); 8653 *off = h->ip_hl << 2; 8654 if (*off < (int)sizeof(*h)) { 8655 *action = PF_DROP; 8656 REASON_SET(reason, PFRES_SHORT); 8657 return (-1); 8658 } 8659 pd->src = (struct pf_addr *)&h->ip_src; 8660 pd->dst = (struct pf_addr *)&h->ip_dst; 8661 pd->sport = pd->dport = NULL; 8662 pd->ip_sum = &h->ip_sum; 8663 pd->proto_sum = NULL; 8664 pd->virtual_proto = pd->proto = h->ip_p; 8665 pd->dir = dir; 8666 pd->sidx = (dir == PF_IN) ? 0 : 1; 8667 pd->didx = (dir == PF_IN) ? 1 : 0; 8668 pd->tos = h->ip_tos; 8669 pd->tot_len = ntohs(h->ip_len); 8670 pd->act.rtableid = -1; 8671 8672 if (h->ip_hl > 5) /* has options */ 8673 pd->badopts++; 8674 8675 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 8676 /* 8677 * handle fragments that aren't reassembled by 8678 * normalization 8679 */ 8680 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8681 if (kif == NULL || r == NULL) /* pflog */ 8682 *action = PF_DROP; 8683 else 8684 *action = pf_test_rule(r, s, kif, m, *off, 8685 pd, a, ruleset, inp, *hdrlen); 8686 if (*action != PF_PASS) 8687 REASON_SET(reason, PFRES_FRAG); 8688 return (-1); 8689 } 8690 8691 break; 8692 } 8693 #endif 8694 #ifdef INET6 8695 case AF_INET6: { 8696 struct ip6_hdr *h; 8697 int extoff; 8698 uint32_t jumbolen; 8699 uint8_t nxt; 8700 8701 if (__predict_false(m->m_len < sizeof(struct ip6_hdr)) && 8702 (m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 8703 DPFPRINTF(PF_DEBUG_URGENT, 8704 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 8705 ", pullup failed\n")); 8706 *action = PF_DROP; 8707 REASON_SET(reason, PFRES_SHORT); 8708 return (-1); 8709 } 8710 8711 if (pf_walk_header6(m, &nxt, off, &extoff, &jumbolen, reason) 8712 != PF_PASS) { 8713 *action = PF_DROP; 8714 return (-1); 8715 } 8716 8717 h = mtod(m, struct ip6_hdr *); 8718 pd->src = (struct pf_addr *)&h->ip6_src; 8719 pd->dst = (struct pf_addr *)&h->ip6_dst; 8720 pd->sport = pd->dport = NULL; 8721 pd->ip_sum = NULL; 8722 pd->proto_sum = NULL; 8723 pd->dir = dir; 8724 pd->sidx = (dir == PF_IN) ? 0 : 1; 8725 pd->didx = (dir == PF_IN) ? 1 : 0; 8726 pd->tos = IPV6_DSCP(h); 8727 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8728 pd->virtual_proto = pd->proto = h->ip6_nxt; 8729 pd->act.rtableid = -1; 8730 8731 /* We do IP header normalization and packet reassembly here */ 8732 if (pf_normalize_ip6(m0, kif, *off, reason, pd) != 8733 PF_PASS) { 8734 *action = PF_DROP; 8735 return (-1); 8736 } 8737 m = *m0; 8738 if (m == NULL) { 8739 /* packet sits in reassembly queue, no error */ 8740 *action = PF_PASS; 8741 return (-1); 8742 } 8743 8744 /* 8745 * Reassembly may have changed the next protocol from fragment 8746 * to something else, so update. 8747 */ 8748 h = mtod(m, struct ip6_hdr *); 8749 pd->virtual_proto = pd->proto = h->ip6_nxt; 8750 8751 /* recalc offset, refetch header, then update pd */ 8752 if (pf_walk_header6(m, &nxt, off, &extoff, &jumbolen, reason) != 8753 PF_PASS) { 8754 *action = PF_DROP; 8755 return (-1); 8756 } 8757 8758 if (pd->proto == IPPROTO_FRAGMENT) { 8759 /* 8760 * handle fragments that aren't reassembled by 8761 * normalization 8762 */ 8763 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8764 if (kif == NULL || r == NULL) /* pflog */ 8765 *action = PF_DROP; 8766 else 8767 *action = pf_test_rule(r, s, kif, m, *off, 8768 pd, a, ruleset, NULL /* XXX TODO */, *hdrlen); 8769 if (*action != PF_PASS) 8770 REASON_SET(reason, PFRES_FRAG); 8771 return (-1); 8772 } 8773 8774 break; 8775 } 8776 #endif 8777 default: 8778 panic("pf_setup_pdesc called with illegal af %u", af); 8779 } 8780 8781 switch (pd->proto) { 8782 case IPPROTO_TCP: { 8783 struct tcphdr *th = &pd->hdr.tcp; 8784 8785 if (!pf_pull_hdr(m, *off, th, sizeof(*th), action, 8786 reason, af)) { 8787 *action = PF_DROP; 8788 REASON_SET(reason, PFRES_SHORT); 8789 return (-1); 8790 } 8791 *hdrlen = sizeof(*th); 8792 pd->p_len = pd->tot_len - *off - (th->th_off << 2); 8793 pd->sport = &th->th_sport; 8794 pd->dport = &th->th_dport; 8795 break; 8796 } 8797 case IPPROTO_UDP: { 8798 struct udphdr *uh = &pd->hdr.udp; 8799 8800 if (!pf_pull_hdr(m, *off, uh, sizeof(*uh), action, 8801 reason, af)) { 8802 *action = PF_DROP; 8803 REASON_SET(reason, PFRES_SHORT); 8804 return (-1); 8805 } 8806 *hdrlen = sizeof(*uh); 8807 if (uh->uh_dport == 0 || 8808 ntohs(uh->uh_ulen) > m->m_pkthdr.len - *off || 8809 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 8810 *action = PF_DROP; 8811 REASON_SET(reason, PFRES_SHORT); 8812 return (-1); 8813 } 8814 pd->sport = &uh->uh_sport; 8815 pd->dport = &uh->uh_dport; 8816 break; 8817 } 8818 case IPPROTO_SCTP: { 8819 if (!pf_pull_hdr(m, *off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 8820 action, reason, af)) { 8821 *action = PF_DROP; 8822 REASON_SET(reason, PFRES_SHORT); 8823 return (-1); 8824 } 8825 *hdrlen = sizeof(pd->hdr.sctp); 8826 pd->p_len = pd->tot_len - *off; 8827 8828 pd->sport = &pd->hdr.sctp.src_port; 8829 pd->dport = &pd->hdr.sctp.dest_port; 8830 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 8831 *action = PF_DROP; 8832 REASON_SET(reason, PFRES_SHORT); 8833 return (-1); 8834 } 8835 if (pf_scan_sctp(m, *off, pd, kif) != PF_PASS) { 8836 *action = PF_DROP; 8837 REASON_SET(reason, PFRES_SHORT); 8838 return (-1); 8839 } 8840 break; 8841 } 8842 case IPPROTO_ICMP: { 8843 if (!pf_pull_hdr(m, *off, &pd->hdr.icmp, ICMP_MINLEN, 8844 action, reason, af)) { 8845 *action = PF_DROP; 8846 REASON_SET(reason, PFRES_SHORT); 8847 return (-1); 8848 } 8849 *hdrlen = ICMP_MINLEN; 8850 break; 8851 } 8852 #ifdef INET6 8853 case IPPROTO_ICMPV6: { 8854 size_t icmp_hlen = sizeof(struct icmp6_hdr); 8855 8856 if (!pf_pull_hdr(m, *off, &pd->hdr.icmp6, icmp_hlen, 8857 action, reason, af)) { 8858 *action = PF_DROP; 8859 REASON_SET(reason, PFRES_SHORT); 8860 return (-1); 8861 } 8862 /* ICMP headers we look further into to match state */ 8863 switch (pd->hdr.icmp6.icmp6_type) { 8864 case MLD_LISTENER_QUERY: 8865 case MLD_LISTENER_REPORT: 8866 icmp_hlen = sizeof(struct mld_hdr); 8867 break; 8868 case ND_NEIGHBOR_SOLICIT: 8869 case ND_NEIGHBOR_ADVERT: 8870 icmp_hlen = sizeof(struct nd_neighbor_solicit); 8871 break; 8872 } 8873 if (icmp_hlen > sizeof(struct icmp6_hdr) && 8874 !pf_pull_hdr(m, *off, &pd->hdr.icmp6, icmp_hlen, 8875 action, reason, af)) { 8876 *action = PF_DROP; 8877 REASON_SET(reason, PFRES_SHORT); 8878 return (-1); 8879 } 8880 *hdrlen = icmp_hlen; 8881 break; 8882 } 8883 #endif 8884 } 8885 return (0); 8886 } 8887 8888 static void 8889 pf_counters_inc(int action, struct pf_pdesc *pd, 8890 struct pfi_kkif *kif, struct pf_kstate *s, 8891 struct pf_krule *r, struct pf_krule *a) 8892 { 8893 struct pf_krule *tr; 8894 int dir = pd->dir; 8895 int dirndx; 8896 8897 pf_counter_u64_critical_enter(); 8898 pf_counter_u64_add_protected( 8899 &kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8900 pd->tot_len); 8901 pf_counter_u64_add_protected( 8902 &kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8903 1); 8904 8905 if (action == PF_PASS || r->action == PF_DROP) { 8906 dirndx = (dir == PF_OUT); 8907 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8908 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 8909 pf_update_timestamp(r); 8910 8911 if (a != NULL) { 8912 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8913 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 8914 } 8915 if (s != NULL) { 8916 struct pf_krule_item *ri; 8917 8918 if (s->nat_rule != NULL) { 8919 pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx], 8920 1); 8921 pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx], 8922 pd->tot_len); 8923 } 8924 if (s->src_node != NULL) { 8925 counter_u64_add(s->src_node->packets[dirndx], 8926 1); 8927 counter_u64_add(s->src_node->bytes[dirndx], 8928 pd->tot_len); 8929 } 8930 if (s->nat_src_node != NULL) { 8931 counter_u64_add(s->nat_src_node->packets[dirndx], 8932 1); 8933 counter_u64_add(s->nat_src_node->bytes[dirndx], 8934 pd->tot_len); 8935 } 8936 dirndx = (dir == s->direction) ? 0 : 1; 8937 s->packets[dirndx]++; 8938 s->bytes[dirndx] += pd->tot_len; 8939 8940 SLIST_FOREACH(ri, &s->match_rules, entry) { 8941 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8942 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 8943 } 8944 } 8945 8946 tr = r; 8947 if (s != NULL && s->nat_rule != NULL && 8948 r == &V_pf_default_rule) 8949 tr = s->nat_rule; 8950 8951 if (tr->src.addr.type == PF_ADDR_TABLE) 8952 pfr_update_stats(tr->src.addr.p.tbl, 8953 (s == NULL) ? pd->src : 8954 &s->key[(s->direction == PF_IN)]-> 8955 addr[(s->direction == PF_OUT)], 8956 pd->af, pd->tot_len, dir == PF_OUT, 8957 r->action == PF_PASS, tr->src.neg); 8958 if (tr->dst.addr.type == PF_ADDR_TABLE) 8959 pfr_update_stats(tr->dst.addr.p.tbl, 8960 (s == NULL) ? pd->dst : 8961 &s->key[(s->direction == PF_IN)]-> 8962 addr[(s->direction == PF_IN)], 8963 pd->af, pd->tot_len, dir == PF_OUT, 8964 r->action == PF_PASS, tr->dst.neg); 8965 } 8966 pf_counter_u64_critical_exit(); 8967 } 8968 8969 #if defined(INET) || defined(INET6) 8970 int 8971 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8972 struct inpcb *inp, struct pf_rule_actions *default_actions) 8973 { 8974 struct pfi_kkif *kif; 8975 u_short action, reason = 0; 8976 struct mbuf *m = *m0; 8977 #ifdef INET 8978 struct ip *h = NULL; 8979 #endif 8980 #ifdef INET6 8981 struct ip6_hdr *h6 = NULL; 8982 #endif 8983 struct m_tag *mtag; 8984 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 8985 struct pf_kstate *s = NULL; 8986 struct pf_kruleset *ruleset = NULL; 8987 struct pf_pdesc pd; 8988 int off, hdrlen, use_2nd_queue = 0; 8989 uint16_t tag; 8990 uint8_t rt; 8991 uint8_t ttl; 8992 8993 PF_RULES_RLOCK_TRACKER; 8994 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8995 M_ASSERTPKTHDR(m); 8996 8997 if (!V_pf_status.running) 8998 return (PF_PASS); 8999 9000 PF_RULES_RLOCK(); 9001 9002 kif = (struct pfi_kkif *)ifp->if_pf_kif; 9003 9004 if (__predict_false(kif == NULL)) { 9005 DPFPRINTF(PF_DEBUG_URGENT, 9006 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 9007 PF_RULES_RUNLOCK(); 9008 return (PF_DROP); 9009 } 9010 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 9011 PF_RULES_RUNLOCK(); 9012 return (PF_PASS); 9013 } 9014 9015 if (m->m_flags & M_SKIP_FIREWALL) { 9016 PF_RULES_RUNLOCK(); 9017 return (PF_PASS); 9018 } 9019 9020 #ifdef INET6 9021 /* 9022 * If we end up changing IP addresses (e.g. binat) the stack may get 9023 * confused and fail to send the icmp6 packet too big error. Just send 9024 * it here, before we do any NAT. 9025 */ 9026 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 9027 IN6_LINKMTU(ifp) < pf_max_frag_size(m)) { 9028 PF_RULES_RUNLOCK(); 9029 *m0 = NULL; 9030 icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 9031 return (PF_DROP); 9032 } 9033 #endif 9034 9035 if (__predict_false(! M_WRITABLE(*m0))) { 9036 m = *m0 = m_unshare(*m0, M_NOWAIT); 9037 if (*m0 == NULL) 9038 return (PF_DROP); 9039 } 9040 9041 pf_init_pdesc(&pd, m); 9042 9043 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 9044 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9045 9046 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 9047 pd.pf_mtag->if_idxgen); 9048 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 9049 PF_RULES_RUNLOCK(); 9050 m_freem(*m0); 9051 *m0 = NULL; 9052 return (PF_PASS); 9053 } 9054 PF_RULES_RUNLOCK(); 9055 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 9056 *m0 = NULL; 9057 return (PF_PASS); 9058 } 9059 9060 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 9061 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 9062 /* Dummynet re-injects packets after they've 9063 * completed their delay. We've already 9064 * processed them, so pass unconditionally. */ 9065 9066 /* But only once. We may see the packet multiple times (e.g. 9067 * PFIL_IN/PFIL_OUT). */ 9068 pf_dummynet_flag_remove(m, pd.pf_mtag); 9069 PF_RULES_RUNLOCK(); 9070 9071 return (PF_PASS); 9072 } 9073 9074 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, kif, &a, &r, 9075 &s, &ruleset, &off, &hdrlen, inp, default_actions) == -1) { 9076 if (action != PF_PASS) 9077 pd.act.log |= PF_LOG_FORCE; 9078 goto done; 9079 } 9080 m = *m0; 9081 9082 switch (af) { 9083 #ifdef INET 9084 case AF_INET: 9085 h = mtod(m, struct ip *); 9086 ttl = h->ip_ttl; 9087 break; 9088 #endif 9089 #ifdef INET6 9090 case AF_INET6: 9091 h6 = mtod(m, struct ip6_hdr *); 9092 ttl = h6->ip6_hlim; 9093 break; 9094 #endif 9095 default: 9096 panic("Unknown af %d", af); 9097 } 9098 9099 if (__predict_false(ip_divert_ptr != NULL) && 9100 ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 9101 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 9102 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 9103 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 9104 if (pd.pf_mtag == NULL && 9105 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9106 action = PF_DROP; 9107 goto done; 9108 } 9109 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 9110 } 9111 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 9112 m->m_flags |= M_FASTFWD_OURS; 9113 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9114 } 9115 m_tag_delete(m, mtag); 9116 9117 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 9118 if (mtag != NULL) 9119 m_tag_delete(m, mtag); 9120 } 9121 9122 #ifdef INET6 9123 /* 9124 * we do not support jumbogram. if we keep going, zero ip6_plen 9125 * will do something bad, so drop the packet for now. 9126 */ 9127 if (af == AF_INET6 && htons(h6->ip6_plen) == 0) { 9128 action = PF_DROP; 9129 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 9130 goto done; 9131 } 9132 #endif 9133 9134 switch (pd.proto) { 9135 case IPPROTO_TCP: { 9136 /* Respond to SYN with a syncookie. */ 9137 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 9138 pd.dir == PF_IN && pf_synflood_check(&pd)) { 9139 pf_syncookie_send(m, off, &pd); 9140 action = PF_DROP; 9141 break; 9142 } 9143 9144 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 9145 use_2nd_queue = 1; 9146 action = pf_normalize_tcp(kif, m, 0, off, &pd); 9147 if (action == PF_DROP) 9148 goto done; 9149 action = pf_test_state_tcp(&s, kif, m, off, &pd, &reason); 9150 if (action == PF_PASS) { 9151 if (V_pfsync_update_state_ptr != NULL) 9152 V_pfsync_update_state_ptr(s); 9153 r = s->rule; 9154 a = s->anchor; 9155 } else if (s == NULL) { 9156 /* Validate remote SYN|ACK, re-create original SYN if 9157 * valid. */ 9158 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 9159 TH_ACK && pf_syncookie_validate(&pd) && 9160 pd.dir == PF_IN) { 9161 struct mbuf *msyn; 9162 9163 msyn = pf_syncookie_recreate_syn(ttl, off, 9164 &pd); 9165 if (msyn == NULL) { 9166 action = PF_DROP; 9167 break; 9168 } 9169 9170 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 9171 &pd.act); 9172 m_freem(msyn); 9173 if (action != PF_PASS) 9174 break; 9175 9176 action = pf_test_state_tcp(&s, kif, m, off, 9177 &pd, &reason); 9178 if (action != PF_PASS || s == NULL) { 9179 action = PF_DROP; 9180 break; 9181 } 9182 9183 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 9184 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 9185 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 9186 action = pf_synproxy(&pd, &s, &reason); 9187 break; 9188 } else { 9189 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9190 &a, &ruleset, inp, hdrlen); 9191 } 9192 } 9193 break; 9194 } 9195 9196 case IPPROTO_UDP: { 9197 action = pf_test_state_udp(&s, kif, m, off, &pd); 9198 if (action == PF_PASS) { 9199 if (V_pfsync_update_state_ptr != NULL) 9200 V_pfsync_update_state_ptr(s); 9201 r = s->rule; 9202 a = s->anchor; 9203 } else if (s == NULL) 9204 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9205 &a, &ruleset, inp, hdrlen); 9206 break; 9207 } 9208 9209 case IPPROTO_SCTP: { 9210 action = pf_normalize_sctp(dir, kif, m, 0, off, &pd); 9211 if (action == PF_DROP) 9212 goto done; 9213 action = pf_test_state_sctp(&s, kif, m, off, &pd, 9214 &reason); 9215 if (action == PF_PASS) { 9216 if (V_pfsync_update_state_ptr != NULL) 9217 V_pfsync_update_state_ptr(s); 9218 r = s->rule; 9219 a = s->anchor; 9220 } else if (s == NULL) { 9221 action = pf_test_rule(&r, &s, kif, m, off, 9222 &pd, &a, &ruleset, inp, hdrlen); 9223 } 9224 break; 9225 } 9226 9227 case IPPROTO_ICMP: { 9228 if (af != AF_INET) { 9229 action = PF_DROP; 9230 DPFPRINTF(PF_DEBUG_MISC, 9231 ("dropping IPv6 packet with ICMPv4 payload")); 9232 goto done; 9233 } 9234 action = pf_test_state_icmp(&s, kif, m, off, &pd, &reason); 9235 if (action == PF_PASS) { 9236 if (V_pfsync_update_state_ptr != NULL) 9237 V_pfsync_update_state_ptr(s); 9238 r = s->rule; 9239 a = s->anchor; 9240 } else if (s == NULL) 9241 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9242 &a, &ruleset, inp, hdrlen); 9243 break; 9244 } 9245 9246 case IPPROTO_ICMPV6: { 9247 if (af != AF_INET6) { 9248 action = PF_DROP; 9249 DPFPRINTF(PF_DEBUG_MISC, 9250 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 9251 goto done; 9252 } 9253 action = pf_test_state_icmp(&s, kif, m, off, &pd, &reason); 9254 if (action == PF_PASS) { 9255 if (V_pfsync_update_state_ptr != NULL) 9256 V_pfsync_update_state_ptr(s); 9257 r = s->rule; 9258 a = s->anchor; 9259 } else if (s == NULL) 9260 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9261 &a, &ruleset, inp, hdrlen); 9262 break; 9263 } 9264 9265 default: 9266 action = pf_test_state_other(&s, kif, m, &pd); 9267 if (action == PF_PASS) { 9268 if (V_pfsync_update_state_ptr != NULL) 9269 V_pfsync_update_state_ptr(s); 9270 r = s->rule; 9271 a = s->anchor; 9272 } else if (s == NULL) 9273 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9274 &a, &ruleset, inp, hdrlen); 9275 break; 9276 } 9277 9278 done: 9279 m = *m0; 9280 PF_RULES_RUNLOCK(); 9281 9282 if (m == NULL) 9283 goto eat_pkt; 9284 9285 if (action == PF_PASS && pd.badopts && 9286 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 9287 action = PF_DROP; 9288 REASON_SET(&reason, PFRES_IPOPTIONS); 9289 pd.act.log = PF_LOG_FORCE; 9290 DPFPRINTF(PF_DEBUG_MISC, 9291 ("pf: dropping packet with dangerous headers\n")); 9292 } 9293 9294 if (s) { 9295 uint8_t log = pd.act.log; 9296 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 9297 pd.act.log |= log; 9298 tag = s->tag; 9299 rt = s->rt; 9300 } else { 9301 tag = r->tag; 9302 rt = r->rt; 9303 } 9304 9305 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 9306 action = PF_DROP; 9307 REASON_SET(&reason, PFRES_MEMORY); 9308 } 9309 9310 pf_scrub(m, &pd); 9311 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 9312 pf_normalize_mss(m, off, &pd); 9313 9314 if (pd.act.rtableid >= 0) 9315 M_SETFIB(m, pd.act.rtableid); 9316 9317 if (pd.act.flags & PFSTATE_SETPRIO) { 9318 if (pd.tos & IPTOS_LOWDELAY) 9319 use_2nd_queue = 1; 9320 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 9321 action = PF_DROP; 9322 REASON_SET(&reason, PFRES_MEMORY); 9323 pd.act.log = PF_LOG_FORCE; 9324 DPFPRINTF(PF_DEBUG_MISC, 9325 ("pf: failed to allocate 802.1q mtag\n")); 9326 } 9327 } 9328 9329 #ifdef ALTQ 9330 if (action == PF_PASS && pd.act.qid) { 9331 if (pd.pf_mtag == NULL && 9332 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9333 action = PF_DROP; 9334 REASON_SET(&reason, PFRES_MEMORY); 9335 } else { 9336 if (s != NULL) 9337 pd.pf_mtag->qid_hash = pf_state_hash(s); 9338 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 9339 pd.pf_mtag->qid = pd.act.pqid; 9340 else 9341 pd.pf_mtag->qid = pd.act.qid; 9342 /* Add hints for ecn. */ 9343 #ifdef INET 9344 if (af == AF_INET) 9345 pd.pf_mtag->hdr = h; 9346 #endif 9347 #ifdef INET6 9348 if (af == AF_INET6) 9349 pd.pf_mtag->hdr = h6; 9350 #endif 9351 } 9352 } 9353 #endif /* ALTQ */ 9354 9355 /* 9356 * connections redirected to loopback should not match sockets 9357 * bound specifically to loopback due to security implications, 9358 * see tcp_input() and in_pcblookup_listen(). 9359 */ 9360 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9361 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 9362 (s->nat_rule->action == PF_RDR || 9363 s->nat_rule->action == PF_BINAT) && 9364 pf_is_loopback(af, pd.dst)) 9365 m->m_flags |= M_SKIP_FIREWALL; 9366 9367 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 9368 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 9369 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 9370 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 9371 if (mtag != NULL) { 9372 ((struct pf_divert_mtag *)(mtag+1))->port = 9373 ntohs(r->divert.port); 9374 ((struct pf_divert_mtag *)(mtag+1))->idir = 9375 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 9376 PF_DIVERT_MTAG_DIR_OUT; 9377 9378 if (s) 9379 PF_STATE_UNLOCK(s); 9380 9381 m_tag_prepend(m, mtag); 9382 if (m->m_flags & M_FASTFWD_OURS) { 9383 if (pd.pf_mtag == NULL && 9384 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9385 action = PF_DROP; 9386 REASON_SET(&reason, PFRES_MEMORY); 9387 pd.act.log = PF_LOG_FORCE; 9388 DPFPRINTF(PF_DEBUG_MISC, 9389 ("pf: failed to allocate tag\n")); 9390 } else { 9391 pd.pf_mtag->flags |= 9392 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9393 m->m_flags &= ~M_FASTFWD_OURS; 9394 } 9395 } 9396 ip_divert_ptr(*m0, dir == PF_IN); 9397 *m0 = NULL; 9398 9399 return (action); 9400 } else { 9401 /* XXX: ipfw has the same behaviour! */ 9402 action = PF_DROP; 9403 REASON_SET(&reason, PFRES_MEMORY); 9404 pd.act.log = PF_LOG_FORCE; 9405 DPFPRINTF(PF_DEBUG_MISC, 9406 ("pf: failed to allocate divert tag\n")); 9407 } 9408 } 9409 /* XXX: Anybody working on it?! */ 9410 if (af == AF_INET6 && r->divert.port) 9411 printf("pf: divert(9) is not supported for IPv6\n"); 9412 9413 /* this flag will need revising if the pkt is forwarded */ 9414 if (pd.pf_mtag) 9415 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 9416 9417 if (pd.act.log) { 9418 struct pf_krule *lr; 9419 struct pf_krule_item *ri; 9420 9421 if (s != NULL && s->nat_rule != NULL && 9422 s->nat_rule->log & PF_LOG_ALL) 9423 lr = s->nat_rule; 9424 else 9425 lr = r; 9426 9427 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9428 PFLOG_PACKET(kif, m, action, reason, lr, a, 9429 ruleset, &pd, (s == NULL)); 9430 if (s) { 9431 SLIST_FOREACH(ri, &s->match_rules, entry) 9432 if (ri->r->log & PF_LOG_ALL) 9433 PFLOG_PACKET(kif, m, action, 9434 reason, ri->r, a, ruleset, &pd, 0); 9435 } 9436 } 9437 9438 pf_counters_inc(action, &pd, kif, s, r, a); 9439 9440 switch (action) { 9441 case PF_SYNPROXY_DROP: 9442 m_freem(*m0); 9443 case PF_DEFER: 9444 *m0 = NULL; 9445 action = PF_PASS; 9446 break; 9447 case PF_DROP: 9448 m_freem(*m0); 9449 *m0 = NULL; 9450 break; 9451 default: 9452 if (rt) { 9453 switch (af) { 9454 #ifdef INET 9455 case AF_INET: 9456 /* pf_route() returns unlocked. */ 9457 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 9458 break; 9459 #endif 9460 #ifdef INET6 9461 case AF_INET6: 9462 /* pf_route6() returns unlocked. */ 9463 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 9464 break; 9465 #endif 9466 default: 9467 panic("Unknown af %d", af); 9468 } 9469 goto out; 9470 } 9471 if (pf_dummynet(&pd, s, r, m0) != 0) { 9472 action = PF_DROP; 9473 REASON_SET(&reason, PFRES_MEMORY); 9474 } 9475 break; 9476 } 9477 9478 eat_pkt: 9479 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 9480 9481 if (s && action != PF_DROP) { 9482 if (!s->if_index_in && dir == PF_IN) 9483 s->if_index_in = ifp->if_index; 9484 else if (!s->if_index_out && dir == PF_OUT) 9485 s->if_index_out = ifp->if_index; 9486 } 9487 9488 if (s) 9489 PF_STATE_UNLOCK(s); 9490 9491 #ifdef INET6 9492 /* If reassembled packet passed, create new fragments. */ 9493 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 9494 (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 9495 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 9496 #endif 9497 9498 out: 9499 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 9500 9501 return (action); 9502 } 9503 #endif /* INET || INET6 */ 9504